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.saddr = (rt_mode & IP_VS_RT_MODE_CONNECT) ? *saddr : 0;
134 fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
135 FLOWI_FLAG_KNOWN_NH : 0;
138 rt = ip_route_output_key(net, &fl4);
140 /* Invalid saddr ? */
141 if (PTR_ERR(rt) == -EINVAL && *saddr &&
142 rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
144 flowi4_update_output(&fl4, 0, 0, daddr, 0);
147 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
149 } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
152 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
160 #ifdef CONFIG_IP_VS_IPV6
161 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
163 return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
167 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
169 bool new_rt_is_local)
171 bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
172 bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
173 bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
174 bool source_is_loopback;
175 bool old_rt_is_local;
177 #ifdef CONFIG_IP_VS_IPV6
178 if (skb_af == AF_INET6) {
179 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
182 (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
183 (addr_type & IPV6_ADDR_LOOPBACK);
184 old_rt_is_local = __ip_vs_is_local_route6(
185 (struct rt6_info *)skb_dst(skb));
189 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
190 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
193 if (unlikely(new_rt_is_local)) {
194 if (!rt_mode_allow_local)
196 if (!rt_mode_allow_redirect && !old_rt_is_local)
199 if (!rt_mode_allow_non_local)
201 if (source_is_loopback)
207 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
209 struct sock *sk = skb->sk;
210 struct rtable *ort = skb_rtable(skb);
212 if (!skb->dev && sk && sk_fullsock(sk))
213 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu);
216 static inline bool ensure_mtu_is_adequate(int skb_af, int rt_mode,
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 = dev_net(skb_dst(skb)->dev);
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)
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 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
239 /* If we're going to tunnel the packet and pmtu discovery
240 * is disabled, we'll just fragment it anyway
242 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
245 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
246 skb->len > mtu && !skb_is_gso(skb))) {
247 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
249 IP_VS_DBG(1, "frag needed for %pI4\n",
250 &ip_hdr(skb)->saddr);
258 /* Get route to destination or remote server */
260 __ip_vs_get_out_rt(int skb_af, struct sk_buff *skb, struct ip_vs_dest *dest,
261 __be32 daddr, int rt_mode, __be32 *ret_saddr,
262 struct ip_vs_iphdr *ipvsh)
264 struct net *net = dev_net(skb_dst(skb)->dev);
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(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)
369 struct dst_entry *dst;
370 struct flowi6 fl6 = {
374 dst = ip6_route_output(net, NULL, &fl6);
379 if (ipv6_addr_any(&fl6.saddr) &&
380 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
381 &fl6.daddr, 0, &fl6.saddr) < 0)
384 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
390 *ret_saddr = fl6.saddr;
395 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
400 * Get route to destination or remote server
403 __ip_vs_get_out_rt_v6(int skb_af, struct sk_buff *skb, struct ip_vs_dest *dest,
404 struct in6_addr *daddr, struct in6_addr *ret_saddr,
405 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
407 struct net *net = dev_net(skb_dst(skb)->dev);
408 struct ip_vs_dest_dst *dest_dst;
409 struct rt6_info *rt; /* Route to the other host */
410 struct dst_entry *dst;
412 int local, noref = 1;
415 dest_dst = __ip_vs_dst_check(dest);
416 if (likely(dest_dst))
417 rt = (struct rt6_info *) dest_dst->dst_cache;
421 dest_dst = ip_vs_dest_dst_alloc();
422 spin_lock_bh(&dest->dst_lock);
424 __ip_vs_dst_set(dest, NULL, NULL, 0);
425 spin_unlock_bh(&dest->dst_lock);
428 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
429 &dest_dst->dst_saddr.in6,
432 __ip_vs_dst_set(dest, NULL, NULL, 0);
433 spin_unlock_bh(&dest->dst_lock);
434 ip_vs_dest_dst_free(dest_dst);
437 rt = (struct rt6_info *) dst;
438 cookie = rt->rt6i_node ? rt->rt6i_node->fn_sernum : 0;
439 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
440 spin_unlock_bh(&dest->dst_lock);
441 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
442 &dest->addr.in6, &dest_dst->dst_saddr.in6,
443 atomic_read(&rt->dst.__refcnt));
446 *ret_saddr = dest_dst->dst_saddr.in6;
449 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm);
452 rt = (struct rt6_info *) dst;
455 local = __ip_vs_is_local_route6(rt);
457 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
459 IP_VS_DBG_RL("We are crossing local and non-local addresses"
460 " daddr=%pI6\n", daddr);
464 if (unlikely(local)) {
465 /* skb to local stack, preserve old route */
467 dst_release(&rt->dst);
472 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
473 mtu = dst_mtu(&rt->dst);
475 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
476 if (mtu < IPV6_MIN_MTU) {
477 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
481 maybe_update_pmtu(skb_af, skb, mtu);
484 if (!ensure_mtu_is_adequate(skb_af, rt_mode, ipvsh, skb, mtu))
490 skb_dst_set_noref(skb, &rt->dst);
492 skb_dst_set(skb, dst_clone(&rt->dst));
494 skb_dst_set(skb, &rt->dst);
500 dst_release(&rt->dst);
504 dst_link_failure(skb);
510 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
511 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
512 struct ip_vs_conn *cp)
516 skb->ipvs_property = 1;
517 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
518 ret = ip_vs_confirm_conntrack(skb);
519 if (ret == NF_ACCEPT) {
521 skb_forward_csum(skb);
526 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
527 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
528 struct ip_vs_conn *cp, int local)
532 skb->ipvs_property = 1;
533 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
536 ip_vs_update_conntrack(skb, cp, 1);
538 skb_forward_csum(skb);
539 NF_HOOK(pf, NF_INET_LOCAL_OUT, NULL, skb,
540 NULL, skb_dst(skb)->dev, dst_output_sk);
546 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
547 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
548 struct ip_vs_conn *cp, int local)
552 skb->ipvs_property = 1;
553 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
556 skb_forward_csum(skb);
557 NF_HOOK(pf, NF_INET_LOCAL_OUT, NULL, skb,
558 NULL, skb_dst(skb)->dev, dst_output_sk);
566 * NULL transmitter (do nothing except return NF_ACCEPT)
569 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
570 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
572 /* we do not touch skb and do not need pskb ptr */
573 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
579 * Let packets bypass the destination when the destination is not
580 * available, it may be only used in transparent cache cluster.
583 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
584 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
586 struct iphdr *iph = ip_hdr(skb);
591 if (__ip_vs_get_out_rt(cp->af, skb, NULL, iph->daddr,
592 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
597 /* Another hack: avoid icmp_send in ip_fragment */
600 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
613 #ifdef CONFIG_IP_VS_IPV6
615 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
616 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
621 if (__ip_vs_get_out_rt_v6(cp->af, skb, NULL, &ipvsh->daddr.in6, NULL,
622 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
625 /* Another hack: avoid icmp_send in ip_fragment */
628 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
643 * NAT transmitter (only for outside-to-inside nat forwarding)
644 * Not used for related ICMP
647 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
648 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
650 struct rtable *rt; /* Route to the other host */
651 int local, rc, was_input;
656 /* check if it is a connection of no-client-port */
657 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
660 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
663 ip_vs_conn_fill_cport(cp, *p);
664 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
667 was_input = rt_is_input_route(skb_rtable(skb));
668 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
669 IP_VS_RT_MODE_LOCAL |
670 IP_VS_RT_MODE_NON_LOCAL |
671 IP_VS_RT_MODE_RDR, NULL, ipvsh);
674 rt = skb_rtable(skb);
676 * Avoid duplicate tuple in reply direction for NAT traffic
677 * to local address when connection is sync-ed
679 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
680 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
681 enum ip_conntrack_info ctinfo;
682 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
684 if (ct && !nf_ct_is_untracked(ct)) {
685 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, 0,
687 "stopping DNAT to local address");
693 /* From world but DNAT to loopback address? */
694 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
695 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, 0, "ip_vs_nat_xmit(): "
696 "stopping DNAT to loopback address");
700 /* copy-on-write the packet before mangling it */
701 if (!skb_make_writable(skb, sizeof(struct iphdr)))
704 if (skb_cow(skb, rt->dst.dev->hard_header_len))
707 /* mangle the packet */
708 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
710 ip_hdr(skb)->daddr = cp->daddr.ip;
711 ip_send_check(ip_hdr(skb));
713 IP_VS_DBG_PKT(10, AF_INET, pp, skb, 0, "After DNAT");
715 /* FIXME: when application helper enlarges the packet and the length
716 is larger than the MTU of outgoing device, there will be still
719 /* Another hack: avoid icmp_send in ip_fragment */
722 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
735 #ifdef CONFIG_IP_VS_IPV6
737 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
738 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
740 struct rt6_info *rt; /* Route to the other host */
746 /* check if it is a connection of no-client-port */
747 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
749 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
752 ip_vs_conn_fill_cport(cp, *p);
753 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
756 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
758 IP_VS_RT_MODE_LOCAL |
759 IP_VS_RT_MODE_NON_LOCAL |
763 rt = (struct rt6_info *) skb_dst(skb);
765 * Avoid duplicate tuple in reply direction for NAT traffic
766 * to local address when connection is sync-ed
768 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
769 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
770 enum ip_conntrack_info ctinfo;
771 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
773 if (ct && !nf_ct_is_untracked(ct)) {
774 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, 0,
775 "ip_vs_nat_xmit_v6(): "
776 "stopping DNAT to local address");
782 /* From world but DNAT to loopback address? */
783 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
784 ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
785 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, 0,
786 "ip_vs_nat_xmit_v6(): "
787 "stopping DNAT to loopback address");
791 /* copy-on-write the packet before mangling it */
792 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
795 if (skb_cow(skb, rt->dst.dev->hard_header_len))
798 /* mangle the packet */
799 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
801 ipv6_hdr(skb)->daddr = cp->daddr.in6;
803 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, 0, "After DNAT");
805 /* FIXME: when application helper enlarges the packet and the length
806 is larger than the MTU of outgoing device, there will be still
809 /* Another hack: avoid icmp_send in ip_fragment */
812 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
826 /* When forwarding a packet, we must ensure that we've got enough headroom
827 * for the encapsulation packet in the skb. This also gives us an
828 * opportunity to figure out what the payload_len, dsfield, ttl, and df
829 * values should be, so that we won't need to look at the old ip header
832 static struct sk_buff *
833 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
834 unsigned int max_headroom, __u8 *next_protocol,
835 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
838 struct sk_buff *new_skb = NULL;
839 struct iphdr *old_iph = NULL;
840 #ifdef CONFIG_IP_VS_IPV6
841 struct ipv6hdr *old_ipv6h = NULL;
844 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
845 new_skb = skb_realloc_headroom(skb, max_headroom);
849 skb_set_owner_w(new_skb, skb->sk);
854 #ifdef CONFIG_IP_VS_IPV6
855 if (skb_af == AF_INET6) {
856 old_ipv6h = ipv6_hdr(skb);
857 *next_protocol = IPPROTO_IPV6;
860 ntohs(old_ipv6h->payload_len) +
862 *dsfield = ipv6_get_dsfield(old_ipv6h);
863 *ttl = old_ipv6h->hop_limit;
869 old_iph = ip_hdr(skb);
870 /* Copy DF, reset fragment offset and MF */
872 *df = (old_iph->frag_off & htons(IP_DF));
873 *next_protocol = IPPROTO_IPIP;
875 /* fix old IP header checksum */
876 ip_send_check(old_iph);
877 *dsfield = ipv4_get_dsfield(old_iph);
880 *payload_len = ntohs(old_iph->tot_len);
886 return ERR_PTR(-ENOMEM);
889 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
891 if (encaps_af == AF_INET) {
892 if (orig_af == AF_INET)
898 /* GSO: we need to provide proper SKB_GSO_ value for IPv6:
905 * IP Tunneling transmitter
907 * This function encapsulates the packet in a new IP packet, its
908 * destination will be set to cp->daddr. Most code of this function
909 * is taken from ipip.c.
911 * It is used in VS/TUN cluster. The load balancer selects a real
912 * server from a cluster based on a scheduling algorithm,
913 * encapsulates the request packet and forwards it to the selected
914 * server. For example, all real servers are configured with
915 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
916 * the encapsulated packet, it will decapsulate the packet, processe
917 * the request and return the response packets directly to the client
918 * without passing the load balancer. This can greatly increase the
919 * scalability of virtual server.
921 * Used for ANY protocol
924 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
925 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
927 struct net *net = skb_net(skb);
928 struct netns_ipvs *ipvs = net_ipvs(net);
929 struct rtable *rt; /* Route to the other host */
930 __be32 saddr; /* Source for tunnel */
931 struct net_device *tdev; /* Device to other host */
932 __u8 next_protocol = 0;
937 struct iphdr *iph; /* Our new IP header */
938 unsigned int max_headroom; /* The extra header space needed */
944 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
945 IP_VS_RT_MODE_LOCAL |
946 IP_VS_RT_MODE_NON_LOCAL |
947 IP_VS_RT_MODE_CONNECT |
948 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
953 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
956 rt = skb_rtable(skb);
960 * Okay, now see if we can stuff it in the buffer as-is.
962 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
964 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
965 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
966 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
967 &next_protocol, NULL, &dsfield,
972 skb = iptunnel_handle_offloads(
973 skb, false, __tun_gso_type_mask(AF_INET, cp->af));
977 skb->transport_header = skb->network_header;
979 skb_push(skb, sizeof(struct iphdr));
980 skb_reset_network_header(skb);
981 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
984 * Push down and install the IPIP header.
988 iph->ihl = sizeof(struct iphdr)>>2;
990 iph->protocol = next_protocol;
992 iph->daddr = cp->daddr.ip;
995 ip_select_ident(net, skb, NULL);
997 /* Another hack: avoid icmp_send in ip_fragment */
1000 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1001 if (ret == NF_ACCEPT)
1003 else if (ret == NF_DROP)
1019 #ifdef CONFIG_IP_VS_IPV6
1021 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1022 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1024 struct rt6_info *rt; /* Route to the other host */
1025 struct in6_addr saddr; /* Source for tunnel */
1026 struct net_device *tdev; /* Device to other host */
1027 __u8 next_protocol = 0;
1028 __u32 payload_len = 0;
1031 struct ipv6hdr *iph; /* Our new IP header */
1032 unsigned int max_headroom; /* The extra header space needed */
1038 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1040 IP_VS_RT_MODE_LOCAL |
1041 IP_VS_RT_MODE_NON_LOCAL |
1042 IP_VS_RT_MODE_TUNNEL);
1047 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1050 rt = (struct rt6_info *) skb_dst(skb);
1054 * Okay, now see if we can stuff it in the buffer as-is.
1056 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1058 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1059 &next_protocol, &payload_len,
1060 &dsfield, &ttl, NULL);
1064 skb = iptunnel_handle_offloads(
1065 skb, false, __tun_gso_type_mask(AF_INET6, cp->af));
1069 skb->transport_header = skb->network_header;
1071 skb_push(skb, sizeof(struct ipv6hdr));
1072 skb_reset_network_header(skb);
1073 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1076 * Push down and install the IPIP header.
1078 iph = ipv6_hdr(skb);
1080 iph->nexthdr = next_protocol;
1081 iph->payload_len = htons(payload_len);
1082 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1083 ipv6_change_dsfield(iph, 0, dsfield);
1084 iph->daddr = cp->daddr.in6;
1086 iph->hop_limit = ttl;
1088 /* Another hack: avoid icmp_send in ip_fragment */
1091 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1092 if (ret == NF_ACCEPT)
1094 else if (ret == NF_DROP)
1113 * Direct Routing transmitter
1114 * Used for ANY protocol
1117 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1118 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1125 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
1126 IP_VS_RT_MODE_LOCAL |
1127 IP_VS_RT_MODE_NON_LOCAL |
1128 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1133 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1136 ip_send_check(ip_hdr(skb));
1138 /* Another hack: avoid icmp_send in ip_fragment */
1141 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1154 #ifdef CONFIG_IP_VS_IPV6
1156 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1157 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1164 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1166 IP_VS_RT_MODE_LOCAL |
1167 IP_VS_RT_MODE_NON_LOCAL);
1172 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1175 /* Another hack: avoid icmp_send in ip_fragment */
1178 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1194 * ICMP packet transmitter
1195 * called by the ip_vs_in_icmp
1198 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1199 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1200 struct ip_vs_iphdr *iph)
1202 struct rtable *rt; /* Route to the other host */
1205 int rt_mode, was_input;
1209 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1210 forwarded directly here, because there is no need to
1211 translate address/port back */
1212 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1213 if (cp->packet_xmit)
1214 rc = cp->packet_xmit(skb, cp, pp, iph);
1217 /* do not touch skb anymore */
1218 atomic_inc(&cp->in_pkts);
1223 * mangle and send the packet here (only for VS/NAT)
1225 was_input = rt_is_input_route(skb_rtable(skb));
1227 /* LOCALNODE from FORWARD hook is not supported */
1228 rt_mode = (hooknum != NF_INET_FORWARD) ?
1229 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1230 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1232 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1236 rt = skb_rtable(skb);
1239 * Avoid duplicate tuple in reply direction for NAT traffic
1240 * to local address when connection is sync-ed
1242 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1243 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1244 enum ip_conntrack_info ctinfo;
1245 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1247 if (ct && !nf_ct_is_untracked(ct)) {
1248 IP_VS_DBG(10, "%s(): "
1249 "stopping DNAT to local address %pI4\n",
1250 __func__, &cp->daddr.ip);
1256 /* From world but DNAT to loopback address? */
1257 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1258 IP_VS_DBG(1, "%s(): "
1259 "stopping DNAT to loopback %pI4\n",
1260 __func__, &cp->daddr.ip);
1264 /* copy-on-write the packet before mangling it */
1265 if (!skb_make_writable(skb, offset))
1268 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1271 ip_vs_nat_icmp(skb, pp, cp, 0);
1273 /* Another hack: avoid icmp_send in ip_fragment */
1276 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1289 #ifdef CONFIG_IP_VS_IPV6
1291 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1292 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1293 struct ip_vs_iphdr *ipvsh)
1295 struct rt6_info *rt; /* Route to the other host */
1302 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1303 forwarded directly here, because there is no need to
1304 translate address/port back */
1305 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1306 if (cp->packet_xmit)
1307 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1310 /* do not touch skb anymore */
1311 atomic_inc(&cp->in_pkts);
1316 * mangle and send the packet here (only for VS/NAT)
1319 /* LOCALNODE from FORWARD hook is not supported */
1320 rt_mode = (hooknum != NF_INET_FORWARD) ?
1321 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1322 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1324 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1325 NULL, ipvsh, 0, rt_mode);
1328 rt = (struct rt6_info *) skb_dst(skb);
1330 * Avoid duplicate tuple in reply direction for NAT traffic
1331 * to local address when connection is sync-ed
1333 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1334 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1335 enum ip_conntrack_info ctinfo;
1336 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1338 if (ct && !nf_ct_is_untracked(ct)) {
1339 IP_VS_DBG(10, "%s(): "
1340 "stopping DNAT to local address %pI6\n",
1341 __func__, &cp->daddr.in6);
1347 /* From world but DNAT to loopback address? */
1348 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1349 ipv6_addr_type(&rt->rt6i_dst.addr) & IPV6_ADDR_LOOPBACK) {
1350 IP_VS_DBG(1, "%s(): "
1351 "stopping DNAT to loopback %pI6\n",
1352 __func__, &cp->daddr.in6);
1356 /* copy-on-write the packet before mangling it */
1357 if (!skb_make_writable(skb, offset))
1360 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1363 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1365 /* Another hack: avoid icmp_send in ip_fragment */
1368 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);