2 * ip_vs_xmit.c: various packet transmitters for IPVS
4 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
5 * Julian Anastasov <ja@ssi.bg>
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License
9 * as published by the Free Software Foundation; either version
10 * 2 of the License, or (at your option) any later version.
14 * Description of forwarding methods:
15 * - all transmitters are called from LOCAL_IN (remote clients) and
16 * LOCAL_OUT (local clients) but for ICMP can be called from FORWARD
17 * - not all connections have destination server, for example,
18 * connections in backup server when fwmark is used
19 * - bypass connections use daddr from packet
20 * - we can use dst without ref while sending in RCU section, we use
21 * ref when returning NF_ACCEPT for NAT-ed packet via loopback
23 * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
24 * - skb->pkt_type is not set yet
25 * - the only place where we can see skb->sk != NULL
28 #define KMSG_COMPONENT "IPVS"
29 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
33 #include <linux/tcp.h> /* for tcphdr */
35 #include <net/tcp.h> /* for csum_tcpudp_magic */
37 #include <net/icmp.h> /* for icmp_send */
38 #include <net/route.h> /* for ip_route_output */
40 #include <net/ip6_route.h>
41 #include <net/ip_tunnels.h>
42 #include <net/addrconf.h>
43 #include <linux/icmpv6.h>
44 #include <linux/netfilter.h>
45 #include <linux/netfilter_ipv4.h>
47 #include <net/ip_vs.h>
50 IP_VS_RT_MODE_LOCAL = 1, /* Allow local dest */
51 IP_VS_RT_MODE_NON_LOCAL = 2, /* Allow non-local dest */
52 IP_VS_RT_MODE_RDR = 4, /* Allow redirect from remote daddr to
55 IP_VS_RT_MODE_CONNECT = 8, /* Always bind route to saddr */
56 IP_VS_RT_MODE_KNOWN_NH = 16,/* Route via remote addr */
57 IP_VS_RT_MODE_TUNNEL = 32,/* Tunnel mode */
60 static inline struct ip_vs_dest_dst *ip_vs_dest_dst_alloc(void)
62 return kmalloc(sizeof(struct ip_vs_dest_dst), GFP_ATOMIC);
65 static inline void ip_vs_dest_dst_free(struct ip_vs_dest_dst *dest_dst)
71 * Destination cache to speed up outgoing route lookup
74 __ip_vs_dst_set(struct ip_vs_dest *dest, struct ip_vs_dest_dst *dest_dst,
75 struct dst_entry *dst, u32 dst_cookie)
77 struct ip_vs_dest_dst *old;
79 old = rcu_dereference_protected(dest->dest_dst,
80 lockdep_is_held(&dest->dst_lock));
83 dest_dst->dst_cache = dst;
84 dest_dst->dst_cookie = dst_cookie;
86 rcu_assign_pointer(dest->dest_dst, dest_dst);
89 call_rcu(&old->rcu_head, ip_vs_dest_dst_rcu_free);
92 static inline struct ip_vs_dest_dst *
93 __ip_vs_dst_check(struct ip_vs_dest *dest)
95 struct ip_vs_dest_dst *dest_dst = rcu_dereference(dest->dest_dst);
96 struct dst_entry *dst;
100 dst = dest_dst->dst_cache;
102 dst->ops->check(dst, dest_dst->dst_cookie) == NULL)
108 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
110 if (IP6CB(skb)->frag_max_size) {
111 /* frag_max_size tell us that, this packet have been
112 * defragmented by netfilter IPv6 conntrack module.
114 if (IP6CB(skb)->frag_max_size > mtu)
115 return true; /* largest fragment violate MTU */
117 else if (skb->len > mtu && !skb_is_gso(skb)) {
118 return true; /* Packet size violate MTU size */
123 /* Get route to daddr, update *saddr, optionally bind route to saddr */
124 static struct rtable *do_output_route4(struct net *net, __be32 daddr,
125 int rt_mode, __be32 *saddr)
131 memset(&fl4, 0, sizeof(fl4));
133 fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
134 FLOWI_FLAG_KNOWN_NH : 0;
137 rt = ip_route_output_key(net, &fl4);
139 /* Invalid saddr ? */
140 if (PTR_ERR(rt) == -EINVAL && *saddr &&
141 rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
143 flowi4_update_output(&fl4, 0, 0, daddr, 0);
146 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
148 } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
151 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
159 #ifdef CONFIG_IP_VS_IPV6
160 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
162 return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
166 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
168 bool new_rt_is_local)
170 bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
171 bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
172 bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
173 bool source_is_loopback;
174 bool old_rt_is_local;
176 #ifdef CONFIG_IP_VS_IPV6
177 if (skb_af == AF_INET6) {
178 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
181 (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
182 (addr_type & IPV6_ADDR_LOOPBACK);
183 old_rt_is_local = __ip_vs_is_local_route6(
184 (struct rt6_info *)skb_dst(skb));
188 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
189 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
192 if (unlikely(new_rt_is_local)) {
193 if (!rt_mode_allow_local)
195 if (!rt_mode_allow_redirect && !old_rt_is_local)
198 if (!rt_mode_allow_non_local)
200 if (source_is_loopback)
206 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
208 struct sock *sk = skb->sk;
209 struct rtable *ort = skb_rtable(skb);
211 if (!skb->dev && sk && sk_fullsock(sk))
212 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu);
215 static inline bool ensure_mtu_is_adequate(struct netns_ipvs *ipvs, int skb_af,
217 struct ip_vs_iphdr *ipvsh,
218 struct sk_buff *skb, int mtu)
220 #ifdef CONFIG_IP_VS_IPV6
221 if (skb_af == AF_INET6) {
222 struct net *net = ipvs->net;
224 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
226 skb->dev = net->loopback_dev;
227 /* only send ICMP too big on first fragment */
228 if (!ipvsh->fragoffs && !ip_vs_iph_icmp(ipvsh))
229 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
230 IP_VS_DBG(1, "frag needed for %pI6c\n",
231 &ipv6_hdr(skb)->saddr);
237 /* If we're going to tunnel the packet and pmtu discovery
238 * is disabled, we'll just fragment it anyway
240 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
243 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
244 skb->len > mtu && !skb_is_gso(skb) &&
245 !ip_vs_iph_icmp(ipvsh))) {
246 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
248 IP_VS_DBG(1, "frag needed for %pI4\n",
249 &ip_hdr(skb)->saddr);
257 /* Get route to destination or remote server */
259 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
260 struct ip_vs_dest *dest,
261 __be32 daddr, int rt_mode, __be32 *ret_saddr,
262 struct ip_vs_iphdr *ipvsh)
264 struct net *net = ipvs->net;
265 struct ip_vs_dest_dst *dest_dst;
266 struct rtable *rt; /* Route to the other host */
268 int local, noref = 1;
271 dest_dst = __ip_vs_dst_check(dest);
272 if (likely(dest_dst))
273 rt = (struct rtable *) dest_dst->dst_cache;
275 dest_dst = ip_vs_dest_dst_alloc();
276 spin_lock_bh(&dest->dst_lock);
278 __ip_vs_dst_set(dest, NULL, NULL, 0);
279 spin_unlock_bh(&dest->dst_lock);
282 rt = do_output_route4(net, dest->addr.ip, rt_mode,
283 &dest_dst->dst_saddr.ip);
285 __ip_vs_dst_set(dest, NULL, NULL, 0);
286 spin_unlock_bh(&dest->dst_lock);
287 ip_vs_dest_dst_free(dest_dst);
290 __ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
291 spin_unlock_bh(&dest->dst_lock);
292 IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
293 &dest->addr.ip, &dest_dst->dst_saddr.ip,
294 atomic_read(&rt->dst.__refcnt));
297 *ret_saddr = dest_dst->dst_saddr.ip;
299 __be32 saddr = htonl(INADDR_ANY);
303 /* For such unconfigured boxes avoid many route lookups
304 * for performance reasons because we do not remember saddr
306 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
307 rt = do_output_route4(net, daddr, rt_mode, &saddr);
314 local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
315 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
317 IP_VS_DBG_RL("We are crossing local and non-local addresses"
318 " daddr=%pI4\n", &daddr);
322 if (unlikely(local)) {
323 /* skb to local stack, preserve old route */
329 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
330 mtu = dst_mtu(&rt->dst);
332 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
334 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
337 maybe_update_pmtu(skb_af, skb, mtu);
340 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
346 skb_dst_set_noref(skb, &rt->dst);
348 skb_dst_set(skb, dst_clone(&rt->dst));
350 skb_dst_set(skb, &rt->dst);
360 dst_link_failure(skb);
364 #ifdef CONFIG_IP_VS_IPV6
365 static struct dst_entry *
366 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
367 struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
369 struct dst_entry *dst;
370 struct flowi6 fl6 = {
374 if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
375 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
377 dst = ip6_route_output(net, NULL, &fl6);
382 if (ipv6_addr_any(&fl6.saddr) &&
383 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
384 &fl6.daddr, 0, &fl6.saddr) < 0)
387 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
393 *ret_saddr = fl6.saddr;
398 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
403 * Get route to destination or remote server
406 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
407 struct ip_vs_dest *dest,
408 struct in6_addr *daddr, struct in6_addr *ret_saddr,
409 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
411 struct net *net = ipvs->net;
412 struct ip_vs_dest_dst *dest_dst;
413 struct rt6_info *rt; /* Route to the other host */
414 struct dst_entry *dst;
416 int local, noref = 1;
419 dest_dst = __ip_vs_dst_check(dest);
420 if (likely(dest_dst))
421 rt = (struct rt6_info *) dest_dst->dst_cache;
425 dest_dst = ip_vs_dest_dst_alloc();
426 spin_lock_bh(&dest->dst_lock);
428 __ip_vs_dst_set(dest, NULL, NULL, 0);
429 spin_unlock_bh(&dest->dst_lock);
432 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
433 &dest_dst->dst_saddr.in6,
436 __ip_vs_dst_set(dest, NULL, NULL, 0);
437 spin_unlock_bh(&dest->dst_lock);
438 ip_vs_dest_dst_free(dest_dst);
441 rt = (struct rt6_info *) dst;
442 cookie = rt6_get_cookie(rt);
443 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
444 spin_unlock_bh(&dest->dst_lock);
445 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
446 &dest->addr.in6, &dest_dst->dst_saddr.in6,
447 atomic_read(&rt->dst.__refcnt));
450 *ret_saddr = dest_dst->dst_saddr.in6;
453 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
457 rt = (struct rt6_info *) dst;
460 local = __ip_vs_is_local_route6(rt);
462 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
464 IP_VS_DBG_RL("We are crossing local and non-local addresses"
465 " daddr=%pI6\n", daddr);
469 if (unlikely(local)) {
470 /* skb to local stack, preserve old route */
472 dst_release(&rt->dst);
477 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
478 mtu = dst_mtu(&rt->dst);
480 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
481 if (mtu < IPV6_MIN_MTU) {
482 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
486 maybe_update_pmtu(skb_af, skb, mtu);
489 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
495 skb_dst_set_noref(skb, &rt->dst);
497 skb_dst_set(skb, dst_clone(&rt->dst));
499 skb_dst_set(skb, &rt->dst);
505 dst_release(&rt->dst);
509 /* The ip6_link_failure function requires the dev field to be set
510 * in order to get the net (further for the sake of fwmark
514 skb->dev = skb_dst(skb)->dev;
516 dst_link_failure(skb);
522 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
523 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
524 struct ip_vs_conn *cp)
528 skb->ipvs_property = 1;
529 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
530 ret = ip_vs_confirm_conntrack(skb);
531 if (ret == NF_ACCEPT) {
533 skb_forward_csum(skb);
535 skb_sender_cpu_clear(skb);
540 /* In the event of a remote destination, it's possible that we would have
541 * matches against an old socket (particularly a TIME-WAIT socket). This
542 * causes havoc down the line (ip_local_out et. al. expect regular sockets
543 * and invalid memory accesses will happen) so simply drop the association
546 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
548 /* If dev is set, the packet came from the LOCAL_IN callback and
549 * not from a local TCP socket.
555 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
556 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
557 struct ip_vs_conn *cp, int local)
561 skb->ipvs_property = 1;
562 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
565 ip_vs_update_conntrack(skb, cp, 1);
567 /* Remove the early_demux association unless it's bound for the
568 * exact same port and address on this host after translation.
570 if (!local || cp->vport != cp->dport ||
571 !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
572 ip_vs_drop_early_demux_sk(skb);
575 skb_forward_csum(skb);
577 skb_sender_cpu_clear(skb);
578 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
579 NULL, skb_dst(skb)->dev, dst_output);
586 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
587 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
588 struct ip_vs_conn *cp, int local)
592 skb->ipvs_property = 1;
593 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
596 ip_vs_drop_early_demux_sk(skb);
597 skb_forward_csum(skb);
599 skb_sender_cpu_clear(skb);
600 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
601 NULL, skb_dst(skb)->dev, dst_output);
609 * NULL transmitter (do nothing except return NF_ACCEPT)
612 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
613 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
615 /* we do not touch skb and do not need pskb ptr */
616 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
622 * Let packets bypass the destination when the destination is not
623 * available, it may be only used in transparent cache cluster.
626 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
627 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
629 struct iphdr *iph = ip_hdr(skb);
634 if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
635 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
640 /* Another hack: avoid icmp_send in ip_fragment */
643 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
656 #ifdef CONFIG_IP_VS_IPV6
658 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
659 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
661 struct ipv6hdr *iph = ipv6_hdr(skb);
666 if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
668 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
671 /* Another hack: avoid icmp_send in ip_fragment */
674 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
689 * NAT transmitter (only for outside-to-inside nat forwarding)
690 * Not used for related ICMP
693 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
694 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
696 struct rtable *rt; /* Route to the other host */
697 int local, rc, was_input;
702 /* check if it is a connection of no-client-port */
703 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
706 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
709 ip_vs_conn_fill_cport(cp, *p);
710 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
713 was_input = rt_is_input_route(skb_rtable(skb));
714 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
715 IP_VS_RT_MODE_LOCAL |
716 IP_VS_RT_MODE_NON_LOCAL |
717 IP_VS_RT_MODE_RDR, NULL, ipvsh);
720 rt = skb_rtable(skb);
722 * Avoid duplicate tuple in reply direction for NAT traffic
723 * to local address when connection is sync-ed
725 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
726 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
727 enum ip_conntrack_info ctinfo;
728 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
730 if (ct && !nf_ct_is_untracked(ct)) {
731 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
733 "stopping DNAT to local address");
739 /* From world but DNAT to loopback address? */
740 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
741 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
742 "ip_vs_nat_xmit(): stopping DNAT to loopback "
747 /* copy-on-write the packet before mangling it */
748 if (!skb_make_writable(skb, sizeof(struct iphdr)))
751 if (skb_cow(skb, rt->dst.dev->hard_header_len))
754 /* mangle the packet */
755 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
757 ip_hdr(skb)->daddr = cp->daddr.ip;
758 ip_send_check(ip_hdr(skb));
760 IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
762 /* FIXME: when application helper enlarges the packet and the length
763 is larger than the MTU of outgoing device, there will be still
766 /* Another hack: avoid icmp_send in ip_fragment */
769 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
782 #ifdef CONFIG_IP_VS_IPV6
784 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
785 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
787 struct rt6_info *rt; /* Route to the other host */
793 /* check if it is a connection of no-client-port */
794 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
796 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
799 ip_vs_conn_fill_cport(cp, *p);
800 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
803 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
806 IP_VS_RT_MODE_LOCAL |
807 IP_VS_RT_MODE_NON_LOCAL |
811 rt = (struct rt6_info *) skb_dst(skb);
813 * Avoid duplicate tuple in reply direction for NAT traffic
814 * to local address when connection is sync-ed
816 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
817 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
818 enum ip_conntrack_info ctinfo;
819 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
821 if (ct && !nf_ct_is_untracked(ct)) {
822 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
823 "ip_vs_nat_xmit_v6(): "
824 "stopping DNAT to local address");
830 /* From world but DNAT to loopback address? */
831 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
832 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
833 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
834 "ip_vs_nat_xmit_v6(): "
835 "stopping DNAT to loopback address");
839 /* copy-on-write the packet before mangling it */
840 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
843 if (skb_cow(skb, rt->dst.dev->hard_header_len))
846 /* mangle the packet */
847 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
849 ipv6_hdr(skb)->daddr = cp->daddr.in6;
851 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
853 /* FIXME: when application helper enlarges the packet and the length
854 is larger than the MTU of outgoing device, there will be still
857 /* Another hack: avoid icmp_send in ip_fragment */
860 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
874 /* When forwarding a packet, we must ensure that we've got enough headroom
875 * for the encapsulation packet in the skb. This also gives us an
876 * opportunity to figure out what the payload_len, dsfield, ttl, and df
877 * values should be, so that we won't need to look at the old ip header
880 static struct sk_buff *
881 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
882 unsigned int max_headroom, __u8 *next_protocol,
883 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
886 struct sk_buff *new_skb = NULL;
887 struct iphdr *old_iph = NULL;
888 #ifdef CONFIG_IP_VS_IPV6
889 struct ipv6hdr *old_ipv6h = NULL;
892 ip_vs_drop_early_demux_sk(skb);
894 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
895 new_skb = skb_realloc_headroom(skb, max_headroom);
899 skb_set_owner_w(new_skb, skb->sk);
904 #ifdef CONFIG_IP_VS_IPV6
905 if (skb_af == AF_INET6) {
906 old_ipv6h = ipv6_hdr(skb);
907 *next_protocol = IPPROTO_IPV6;
910 ntohs(old_ipv6h->payload_len) +
912 *dsfield = ipv6_get_dsfield(old_ipv6h);
913 *ttl = old_ipv6h->hop_limit;
919 old_iph = ip_hdr(skb);
920 /* Copy DF, reset fragment offset and MF */
922 *df = (old_iph->frag_off & htons(IP_DF));
923 *next_protocol = IPPROTO_IPIP;
925 /* fix old IP header checksum */
926 ip_send_check(old_iph);
927 *dsfield = ipv4_get_dsfield(old_iph);
930 *payload_len = ntohs(old_iph->tot_len);
936 return ERR_PTR(-ENOMEM);
939 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
941 if (encaps_af == AF_INET) {
942 if (orig_af == AF_INET)
948 /* GSO: we need to provide proper SKB_GSO_ value for IPv6:
955 * IP Tunneling transmitter
957 * This function encapsulates the packet in a new IP packet, its
958 * destination will be set to cp->daddr. Most code of this function
959 * is taken from ipip.c.
961 * It is used in VS/TUN cluster. The load balancer selects a real
962 * server from a cluster based on a scheduling algorithm,
963 * encapsulates the request packet and forwards it to the selected
964 * server. For example, all real servers are configured with
965 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
966 * the encapsulated packet, it will decapsulate the packet, processe
967 * the request and return the response packets directly to the client
968 * without passing the load balancer. This can greatly increase the
969 * scalability of virtual server.
971 * Used for ANY protocol
974 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
975 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
977 struct netns_ipvs *ipvs = cp->ipvs;
978 struct net *net = ipvs->net;
979 struct rtable *rt; /* Route to the other host */
980 __be32 saddr; /* Source for tunnel */
981 struct net_device *tdev; /* Device to other host */
982 __u8 next_protocol = 0;
987 struct iphdr *iph; /* Our new IP header */
988 unsigned int max_headroom; /* The extra header space needed */
994 local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
995 IP_VS_RT_MODE_LOCAL |
996 IP_VS_RT_MODE_NON_LOCAL |
997 IP_VS_RT_MODE_CONNECT |
998 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
1003 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1006 rt = skb_rtable(skb);
1010 * Okay, now see if we can stuff it in the buffer as-is.
1012 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1014 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1015 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1016 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1017 &next_protocol, NULL, &dsfield,
1022 skb = iptunnel_handle_offloads(
1023 skb, false, __tun_gso_type_mask(AF_INET, cp->af));
1027 skb->transport_header = skb->network_header;
1029 skb_push(skb, sizeof(struct iphdr));
1030 skb_reset_network_header(skb);
1031 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1034 * Push down and install the IPIP header.
1038 iph->ihl = sizeof(struct iphdr)>>2;
1040 iph->protocol = next_protocol;
1042 iph->daddr = cp->daddr.ip;
1045 ip_select_ident(net, skb, NULL);
1047 /* Another hack: avoid icmp_send in ip_fragment */
1050 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1051 if (ret == NF_ACCEPT)
1052 ip_local_out(net, skb->sk, skb);
1053 else if (ret == NF_DROP)
1069 #ifdef CONFIG_IP_VS_IPV6
1071 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1072 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1074 struct rt6_info *rt; /* Route to the other host */
1075 struct in6_addr saddr; /* Source for tunnel */
1076 struct net_device *tdev; /* Device to other host */
1077 __u8 next_protocol = 0;
1078 __u32 payload_len = 0;
1081 struct ipv6hdr *iph; /* Our new IP header */
1082 unsigned int max_headroom; /* The extra header space needed */
1088 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1091 IP_VS_RT_MODE_LOCAL |
1092 IP_VS_RT_MODE_NON_LOCAL |
1093 IP_VS_RT_MODE_TUNNEL);
1098 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1101 rt = (struct rt6_info *) skb_dst(skb);
1105 * Okay, now see if we can stuff it in the buffer as-is.
1107 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1109 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1110 &next_protocol, &payload_len,
1111 &dsfield, &ttl, NULL);
1115 skb = iptunnel_handle_offloads(
1116 skb, false, __tun_gso_type_mask(AF_INET6, cp->af));
1120 skb->transport_header = skb->network_header;
1122 skb_push(skb, sizeof(struct ipv6hdr));
1123 skb_reset_network_header(skb);
1124 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1127 * Push down and install the IPIP header.
1129 iph = ipv6_hdr(skb);
1131 iph->nexthdr = next_protocol;
1132 iph->payload_len = htons(payload_len);
1133 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1134 ipv6_change_dsfield(iph, 0, dsfield);
1135 iph->daddr = cp->daddr.in6;
1137 iph->hop_limit = ttl;
1139 /* Another hack: avoid icmp_send in ip_fragment */
1142 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1143 if (ret == NF_ACCEPT)
1144 ip6_local_out(cp->ipvs->net, skb->sk, skb);
1145 else if (ret == NF_DROP)
1164 * Direct Routing transmitter
1165 * Used for ANY protocol
1168 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1169 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1176 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1177 IP_VS_RT_MODE_LOCAL |
1178 IP_VS_RT_MODE_NON_LOCAL |
1179 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1184 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1187 ip_send_check(ip_hdr(skb));
1189 /* Another hack: avoid icmp_send in ip_fragment */
1192 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1205 #ifdef CONFIG_IP_VS_IPV6
1207 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1208 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1215 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1218 IP_VS_RT_MODE_LOCAL |
1219 IP_VS_RT_MODE_NON_LOCAL |
1220 IP_VS_RT_MODE_KNOWN_NH);
1225 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1228 /* Another hack: avoid icmp_send in ip_fragment */
1231 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1247 * ICMP packet transmitter
1248 * called by the ip_vs_in_icmp
1251 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1252 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1253 struct ip_vs_iphdr *iph)
1255 struct rtable *rt; /* Route to the other host */
1258 int rt_mode, was_input;
1262 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1263 forwarded directly here, because there is no need to
1264 translate address/port back */
1265 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1266 if (cp->packet_xmit)
1267 rc = cp->packet_xmit(skb, cp, pp, iph);
1270 /* do not touch skb anymore */
1271 atomic_inc(&cp->in_pkts);
1276 * mangle and send the packet here (only for VS/NAT)
1278 was_input = rt_is_input_route(skb_rtable(skb));
1280 /* LOCALNODE from FORWARD hook is not supported */
1281 rt_mode = (hooknum != NF_INET_FORWARD) ?
1282 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1283 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1285 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1289 rt = skb_rtable(skb);
1292 * Avoid duplicate tuple in reply direction for NAT traffic
1293 * to local address when connection is sync-ed
1295 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1296 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1297 enum ip_conntrack_info ctinfo;
1298 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1300 if (ct && !nf_ct_is_untracked(ct)) {
1301 IP_VS_DBG(10, "%s(): "
1302 "stopping DNAT to local address %pI4\n",
1303 __func__, &cp->daddr.ip);
1309 /* From world but DNAT to loopback address? */
1310 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1311 IP_VS_DBG(1, "%s(): "
1312 "stopping DNAT to loopback %pI4\n",
1313 __func__, &cp->daddr.ip);
1317 /* copy-on-write the packet before mangling it */
1318 if (!skb_make_writable(skb, offset))
1321 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1324 ip_vs_nat_icmp(skb, pp, cp, 0);
1326 /* Another hack: avoid icmp_send in ip_fragment */
1329 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1342 #ifdef CONFIG_IP_VS_IPV6
1344 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1345 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1346 struct ip_vs_iphdr *ipvsh)
1348 struct rt6_info *rt; /* Route to the other host */
1355 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1356 forwarded directly here, because there is no need to
1357 translate address/port back */
1358 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1359 if (cp->packet_xmit)
1360 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1363 /* do not touch skb anymore */
1364 atomic_inc(&cp->in_pkts);
1369 * mangle and send the packet here (only for VS/NAT)
1372 /* LOCALNODE from FORWARD hook is not supported */
1373 rt_mode = (hooknum != NF_INET_FORWARD) ?
1374 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1375 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1377 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1378 &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1381 rt = (struct rt6_info *) skb_dst(skb);
1383 * Avoid duplicate tuple in reply direction for NAT traffic
1384 * to local address when connection is sync-ed
1386 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1387 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1388 enum ip_conntrack_info ctinfo;
1389 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1391 if (ct && !nf_ct_is_untracked(ct)) {
1392 IP_VS_DBG(10, "%s(): "
1393 "stopping DNAT to local address %pI6\n",
1394 __func__, &cp->daddr.in6);
1400 /* From world but DNAT to loopback address? */
1401 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1402 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1403 IP_VS_DBG(1, "%s(): "
1404 "stopping DNAT to loopback %pI6\n",
1405 __func__, &cp->daddr.in6);
1409 /* copy-on-write the packet before mangling it */
1410 if (!skb_make_writable(skb, offset))
1413 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1416 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1418 /* Another hack: avoid icmp_send in ip_fragment */
1421 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);