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
21 * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
22 * - skb->pkt_type is not set yet
23 * - the only place where we can see skb->sk != NULL
26 #define KMSG_COMPONENT "IPVS"
27 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
29 #include <linux/kernel.h>
30 #include <linux/slab.h>
31 #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/addrconf.h>
40 #include <linux/icmpv6.h>
41 #include <linux/netfilter.h>
42 #include <linux/netfilter_ipv4.h>
44 #include <net/ip_vs.h>
48 * Destination cache to speed up outgoing route lookup
51 __ip_vs_dst_set(struct ip_vs_dest
*dest
, u32 rtos
, struct dst_entry
*dst
,
54 struct dst_entry
*old_dst
;
56 old_dst
= dest
->dst_cache
;
57 dest
->dst_cache
= dst
;
58 dest
->dst_rtos
= rtos
;
59 dest
->dst_cookie
= dst_cookie
;
63 static inline struct dst_entry
*
64 __ip_vs_dst_check(struct ip_vs_dest
*dest
, u32 rtos
)
66 struct dst_entry
*dst
= dest
->dst_cache
;
70 if ((dst
->obsolete
|| rtos
!= dest
->dst_rtos
) &&
71 dst
->ops
->check(dst
, dest
->dst_cookie
) == NULL
) {
72 dest
->dst_cache
= NULL
;
81 * Get route to destination or remote server
82 * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest,
83 * &4=Allow redirect from remote daddr to local
85 static struct rtable
*
86 __ip_vs_get_out_rt(struct sk_buff
*skb
, struct ip_vs_dest
*dest
,
87 __be32 daddr
, u32 rtos
, int rt_mode
)
89 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
90 struct rtable
*rt
; /* Route to the other host */
91 struct rtable
*ort
; /* Original route */
95 spin_lock(&dest
->dst_lock
);
96 if (!(rt
= (struct rtable
*)
97 __ip_vs_dst_check(dest
, rtos
))) {
99 .fl4_dst
= dest
->addr
.ip
,
103 if (ip_route_output_key(net
, &rt
, &fl
)) {
104 spin_unlock(&dest
->dst_lock
);
105 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n",
109 __ip_vs_dst_set(dest
, rtos
, dst_clone(&rt
->dst
), 0);
110 IP_VS_DBG(10, "new dst %pI4, refcnt=%d, rtos=%X\n",
112 atomic_read(&rt
->dst
.__refcnt
), rtos
);
114 spin_unlock(&dest
->dst_lock
);
121 if (ip_route_output_key(net
, &rt
, &fl
)) {
122 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n",
128 local
= rt
->rt_flags
& RTCF_LOCAL
;
129 if (!((local
? 1 : 2) & rt_mode
)) {
130 IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI4\n",
131 (rt
->rt_flags
& RTCF_LOCAL
) ?
132 "local":"non-local", &rt
->rt_dst
);
136 if (local
&& !(rt_mode
& 4) && !((ort
= skb_rtable(skb
)) &&
137 ort
->rt_flags
& RTCF_LOCAL
)) {
138 IP_VS_DBG_RL("Redirect from non-local address %pI4 to local "
139 "requires NAT method, dest: %pI4\n",
140 &ip_hdr(skb
)->daddr
, &rt
->rt_dst
);
144 if (unlikely(!local
&& ipv4_is_loopback(ip_hdr(skb
)->saddr
))) {
145 IP_VS_DBG_RL("Stopping traffic from loopback address %pI4 "
146 "to non-local address, dest: %pI4\n",
147 &ip_hdr(skb
)->saddr
, &rt
->rt_dst
);
155 /* Reroute packet to local IPv4 stack after DNAT */
157 __ip_vs_reroute_locally(struct sk_buff
*skb
)
159 struct rtable
*rt
= skb_rtable(skb
);
160 struct net_device
*dev
= rt
->dst
.dev
;
161 struct net
*net
= dev_net(dev
);
162 struct iphdr
*iph
= ip_hdr(skb
);
164 if (rt_is_input_route(rt
)) {
165 unsigned long orefdst
= skb
->_skb_refdst
;
167 if (ip_route_input(skb
, iph
->daddr
, iph
->saddr
,
170 refdst_drop(orefdst
);
173 .fl4_dst
= iph
->daddr
,
174 .fl4_src
= iph
->saddr
,
175 .fl4_tos
= RT_TOS(iph
->tos
),
179 if (ip_route_output_key(net
, &rt
, &fl
))
181 if (!(rt
->rt_flags
& RTCF_LOCAL
)) {
185 /* Drop old route. */
187 skb_dst_set(skb
, &rt
->dst
);
192 #ifdef CONFIG_IP_VS_IPV6
194 static inline int __ip_vs_is_local_route6(struct rt6_info
*rt
)
196 return rt
->rt6i_dev
&& rt
->rt6i_dev
->flags
& IFF_LOOPBACK
;
199 static struct dst_entry
*
200 __ip_vs_route_output_v6(struct net
*net
, struct in6_addr
*daddr
,
201 struct in6_addr
*ret_saddr
, int do_xfrm
)
203 struct dst_entry
*dst
;
208 dst
= ip6_route_output(net
, NULL
, &fl
);
213 if (ipv6_addr_any(&fl
.fl6_src
) &&
214 ipv6_dev_get_saddr(net
, ip6_dst_idev(dst
)->dev
,
215 &fl
.fl6_dst
, 0, &fl
.fl6_src
) < 0)
217 if (do_xfrm
&& xfrm_lookup(net
, &dst
, &fl
, NULL
, 0) < 0)
219 ipv6_addr_copy(ret_saddr
, &fl
.fl6_src
);
224 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr
);
229 * Get route to destination or remote server
230 * rt_mode: flags, &1=Allow local dest, &2=Allow non-local dest,
231 * &4=Allow redirect from remote daddr to local
233 static struct rt6_info
*
234 __ip_vs_get_out_rt_v6(struct sk_buff
*skb
, struct ip_vs_dest
*dest
,
235 struct in6_addr
*daddr
, struct in6_addr
*ret_saddr
,
236 int do_xfrm
, int rt_mode
)
238 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
239 struct rt6_info
*rt
; /* Route to the other host */
240 struct rt6_info
*ort
; /* Original route */
241 struct dst_entry
*dst
;
245 spin_lock(&dest
->dst_lock
);
246 rt
= (struct rt6_info
*)__ip_vs_dst_check(dest
, 0);
250 dst
= __ip_vs_route_output_v6(net
, &dest
->addr
.in6
,
254 spin_unlock(&dest
->dst_lock
);
257 rt
= (struct rt6_info
*) dst
;
258 cookie
= rt
->rt6i_node
? rt
->rt6i_node
->fn_sernum
: 0;
259 __ip_vs_dst_set(dest
, 0, dst_clone(&rt
->dst
), cookie
);
260 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
261 &dest
->addr
.in6
, &dest
->dst_saddr
,
262 atomic_read(&rt
->dst
.__refcnt
));
265 ipv6_addr_copy(ret_saddr
, &dest
->dst_saddr
);
266 spin_unlock(&dest
->dst_lock
);
268 dst
= __ip_vs_route_output_v6(net
, daddr
, ret_saddr
, do_xfrm
);
271 rt
= (struct rt6_info
*) dst
;
274 local
= __ip_vs_is_local_route6(rt
);
275 if (!((local
? 1 : 2) & rt_mode
)) {
276 IP_VS_DBG_RL("Stopping traffic to %s address, dest: %pI6\n",
277 local
? "local":"non-local", daddr
);
278 dst_release(&rt
->dst
);
281 if (local
&& !(rt_mode
& 4) &&
282 !((ort
= (struct rt6_info
*) skb_dst(skb
)) &&
283 __ip_vs_is_local_route6(ort
))) {
284 IP_VS_DBG_RL("Redirect from non-local address %pI6 to local "
285 "requires NAT method, dest: %pI6\n",
286 &ipv6_hdr(skb
)->daddr
, daddr
);
287 dst_release(&rt
->dst
);
290 if (unlikely(!local
&& (!skb
->dev
|| skb
->dev
->flags
& IFF_LOOPBACK
) &&
291 ipv6_addr_type(&ipv6_hdr(skb
)->saddr
) &
292 IPV6_ADDR_LOOPBACK
)) {
293 IP_VS_DBG_RL("Stopping traffic from loopback address %pI6 "
294 "to non-local address, dest: %pI6\n",
295 &ipv6_hdr(skb
)->saddr
, daddr
);
296 dst_release(&rt
->dst
);
306 * Release dest->dst_cache before a dest is removed
309 ip_vs_dst_reset(struct ip_vs_dest
*dest
)
311 struct dst_entry
*old_dst
;
313 old_dst
= dest
->dst_cache
;
314 dest
->dst_cache
= NULL
;
315 dst_release(old_dst
);
318 #define IP_VS_XMIT_TUNNEL(skb, cp) \
320 int __ret = NF_ACCEPT; \
322 (skb)->ipvs_property = 1; \
323 if (unlikely((cp)->flags & IP_VS_CONN_F_NFCT)) \
324 __ret = ip_vs_confirm_conntrack(skb, cp); \
325 if (__ret == NF_ACCEPT) { \
327 skb_forward_csum(skb); \
332 #define IP_VS_XMIT_NAT(pf, skb, cp, local) \
334 (skb)->ipvs_property = 1; \
335 if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
336 ip_vs_notrack(skb); \
338 ip_vs_update_conntrack(skb, cp, 1); \
341 skb_forward_csum(skb); \
342 NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \
343 skb_dst(skb)->dev, dst_output); \
346 #define IP_VS_XMIT(pf, skb, cp, local) \
348 (skb)->ipvs_property = 1; \
349 if (likely(!((cp)->flags & IP_VS_CONN_F_NFCT))) \
350 ip_vs_notrack(skb); \
353 skb_forward_csum(skb); \
354 NF_HOOK(pf, NF_INET_LOCAL_OUT, (skb), NULL, \
355 skb_dst(skb)->dev, dst_output); \
360 * NULL transmitter (do nothing except return NF_ACCEPT)
363 ip_vs_null_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
364 struct ip_vs_protocol
*pp
)
366 /* we do not touch skb and do not need pskb ptr */
367 IP_VS_XMIT(NFPROTO_IPV4
, skb
, cp
, 1);
373 * Let packets bypass the destination when the destination is not
374 * available, it may be only used in transparent cache cluster.
377 ip_vs_bypass_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
378 struct ip_vs_protocol
*pp
)
380 struct rtable
*rt
; /* Route to the other host */
381 struct iphdr
*iph
= ip_hdr(skb
);
386 if (!(rt
= __ip_vs_get_out_rt(skb
, NULL
, iph
->daddr
,
387 RT_TOS(iph
->tos
), 2)))
391 mtu
= dst_mtu(&rt
->dst
);
392 if ((skb
->len
> mtu
) && (iph
->frag_off
& htons(IP_DF
)) &&
395 icmp_send(skb
, ICMP_DEST_UNREACH
,ICMP_FRAG_NEEDED
, htonl(mtu
));
396 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
401 * Call ip_send_check because we are not sure it is called
402 * after ip_defrag. Is copy-on-write needed?
404 if (unlikely((skb
= skb_share_check(skb
, GFP_ATOMIC
)) == NULL
)) {
408 ip_send_check(ip_hdr(skb
));
412 skb_dst_set(skb
, &rt
->dst
);
414 /* Another hack: avoid icmp_send in ip_fragment */
417 IP_VS_XMIT(NFPROTO_IPV4
, skb
, cp
, 0);
423 dst_link_failure(skb
);
430 #ifdef CONFIG_IP_VS_IPV6
432 ip_vs_bypass_xmit_v6(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
433 struct ip_vs_protocol
*pp
)
435 struct rt6_info
*rt
; /* Route to the other host */
436 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
441 if (!(rt
= __ip_vs_get_out_rt_v6(skb
, NULL
, &iph
->daddr
, NULL
, 0, 2)))
445 mtu
= dst_mtu(&rt
->dst
);
446 if (skb
->len
> mtu
&& !skb_is_gso(skb
)) {
448 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
450 skb
->dev
= net
->loopback_dev
;
452 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
453 dst_release(&rt
->dst
);
454 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
459 * Call ip_send_check because we are not sure it is called
460 * after ip_defrag. Is copy-on-write needed?
462 skb
= skb_share_check(skb
, GFP_ATOMIC
);
463 if (unlikely(skb
== NULL
)) {
464 dst_release(&rt
->dst
);
470 skb_dst_set(skb
, &rt
->dst
);
472 /* Another hack: avoid icmp_send in ip_fragment */
475 IP_VS_XMIT(NFPROTO_IPV6
, skb
, cp
, 0);
481 dst_link_failure(skb
);
490 * NAT transmitter (only for outside-to-inside nat forwarding)
491 * Not used for related ICMP
494 ip_vs_nat_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
495 struct ip_vs_protocol
*pp
)
497 struct rtable
*rt
; /* Route to the other host */
499 struct iphdr
*iph
= ip_hdr(skb
);
504 /* check if it is a connection of no-client-port */
505 if (unlikely(cp
->flags
& IP_VS_CONN_F_NO_CPORT
)) {
507 p
= skb_header_pointer(skb
, iph
->ihl
*4, sizeof(_pt
), &_pt
);
510 ip_vs_conn_fill_cport(cp
, *p
);
511 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p
));
514 if (!(rt
= __ip_vs_get_out_rt(skb
, cp
->dest
, cp
->daddr
.ip
,
515 RT_TOS(iph
->tos
), 1|2|4)))
517 local
= rt
->rt_flags
& RTCF_LOCAL
;
519 * Avoid duplicate tuple in reply direction for NAT traffic
520 * to local address when connection is sync-ed
522 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
523 if (cp
->flags
& IP_VS_CONN_F_SYNC
&& local
) {
524 enum ip_conntrack_info ctinfo
;
525 struct nf_conn
*ct
= ct
= nf_ct_get(skb
, &ctinfo
);
527 if (ct
&& !nf_ct_is_untracked(ct
)) {
528 IP_VS_DBG_RL_PKT(10, AF_INET
, pp
, skb
, 0,
530 "stopping DNAT to local address");
536 /* From world but DNAT to loopback address? */
537 if (local
&& ipv4_is_loopback(rt
->rt_dst
) &&
538 rt_is_input_route(skb_rtable(skb
))) {
539 IP_VS_DBG_RL_PKT(1, AF_INET
, pp
, skb
, 0, "ip_vs_nat_xmit(): "
540 "stopping DNAT to loopback address");
545 mtu
= dst_mtu(&rt
->dst
);
546 if ((skb
->len
> mtu
) && (iph
->frag_off
& htons(IP_DF
)) &&
548 icmp_send(skb
, ICMP_DEST_UNREACH
,ICMP_FRAG_NEEDED
, htonl(mtu
));
549 IP_VS_DBG_RL_PKT(0, AF_INET
, pp
, skb
, 0,
550 "ip_vs_nat_xmit(): frag needed for");
554 /* copy-on-write the packet before mangling it */
555 if (!skb_make_writable(skb
, sizeof(struct iphdr
)))
558 if (skb_cow(skb
, rt
->dst
.dev
->hard_header_len
))
561 /* mangle the packet */
562 if (pp
->dnat_handler
&& !pp
->dnat_handler(skb
, pp
, cp
))
564 ip_hdr(skb
)->daddr
= cp
->daddr
.ip
;
565 ip_send_check(ip_hdr(skb
));
570 skb_dst_set(skb
, &rt
->dst
);
574 * Some IPv4 replies get local address from routes,
575 * not from iph, so while we DNAT after routing
576 * we need this second input/output route.
578 if (!__ip_vs_reroute_locally(skb
))
582 IP_VS_DBG_PKT(10, AF_INET
, pp
, skb
, 0, "After DNAT");
584 /* FIXME: when application helper enlarges the packet and the length
585 is larger than the MTU of outgoing device, there will be still
588 /* Another hack: avoid icmp_send in ip_fragment */
591 IP_VS_XMIT_NAT(NFPROTO_IPV4
, skb
, cp
, local
);
597 dst_link_failure(skb
);
607 #ifdef CONFIG_IP_VS_IPV6
609 ip_vs_nat_xmit_v6(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
610 struct ip_vs_protocol
*pp
)
612 struct rt6_info
*rt
; /* Route to the other host */
618 /* check if it is a connection of no-client-port */
619 if (unlikely(cp
->flags
& IP_VS_CONN_F_NO_CPORT
)) {
621 p
= skb_header_pointer(skb
, sizeof(struct ipv6hdr
),
625 ip_vs_conn_fill_cport(cp
, *p
);
626 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p
));
629 if (!(rt
= __ip_vs_get_out_rt_v6(skb
, cp
->dest
, &cp
->daddr
.in6
, NULL
,
632 local
= __ip_vs_is_local_route6(rt
);
634 * Avoid duplicate tuple in reply direction for NAT traffic
635 * to local address when connection is sync-ed
637 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
638 if (cp
->flags
& IP_VS_CONN_F_SYNC
&& local
) {
639 enum ip_conntrack_info ctinfo
;
640 struct nf_conn
*ct
= ct
= nf_ct_get(skb
, &ctinfo
);
642 if (ct
&& !nf_ct_is_untracked(ct
)) {
643 IP_VS_DBG_RL_PKT(10, AF_INET6
, pp
, skb
, 0,
644 "ip_vs_nat_xmit_v6(): "
645 "stopping DNAT to local address");
651 /* From world but DNAT to loopback address? */
652 if (local
&& skb
->dev
&& !(skb
->dev
->flags
& IFF_LOOPBACK
) &&
653 ipv6_addr_type(&rt
->rt6i_dst
.addr
) & IPV6_ADDR_LOOPBACK
) {
654 IP_VS_DBG_RL_PKT(1, AF_INET6
, pp
, skb
, 0,
655 "ip_vs_nat_xmit_v6(): "
656 "stopping DNAT to loopback address");
661 mtu
= dst_mtu(&rt
->dst
);
662 if (skb
->len
> mtu
&& !skb_is_gso(skb
)) {
664 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
666 skb
->dev
= net
->loopback_dev
;
668 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
669 IP_VS_DBG_RL_PKT(0, AF_INET6
, pp
, skb
, 0,
670 "ip_vs_nat_xmit_v6(): frag needed for");
674 /* copy-on-write the packet before mangling it */
675 if (!skb_make_writable(skb
, sizeof(struct ipv6hdr
)))
678 if (skb_cow(skb
, rt
->dst
.dev
->hard_header_len
))
681 /* mangle the packet */
682 if (pp
->dnat_handler
&& !pp
->dnat_handler(skb
, pp
, cp
))
684 ipv6_addr_copy(&ipv6_hdr(skb
)->daddr
, &cp
->daddr
.in6
);
686 if (!local
|| !skb
->dev
) {
687 /* drop the old route when skb is not shared */
689 skb_dst_set(skb
, &rt
->dst
);
691 /* destined to loopback, do we need to change route? */
692 dst_release(&rt
->dst
);
695 IP_VS_DBG_PKT(10, AF_INET6
, pp
, skb
, 0, "After DNAT");
697 /* FIXME: when application helper enlarges the packet and the length
698 is larger than the MTU of outgoing device, there will be still
701 /* Another hack: avoid icmp_send in ip_fragment */
704 IP_VS_XMIT_NAT(NFPROTO_IPV6
, skb
, cp
, local
);
710 dst_link_failure(skb
);
716 dst_release(&rt
->dst
);
723 * IP Tunneling transmitter
725 * This function encapsulates the packet in a new IP packet, its
726 * destination will be set to cp->daddr. Most code of this function
727 * is taken from ipip.c.
729 * It is used in VS/TUN cluster. The load balancer selects a real
730 * server from a cluster based on a scheduling algorithm,
731 * encapsulates the request packet and forwards it to the selected
732 * server. For example, all real servers are configured with
733 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
734 * the encapsulated packet, it will decapsulate the packet, processe
735 * the request and return the response packets directly to the client
736 * without passing the load balancer. This can greatly increase the
737 * scalability of virtual server.
739 * Used for ANY protocol
742 ip_vs_tunnel_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
743 struct ip_vs_protocol
*pp
)
745 struct rtable
*rt
; /* Route to the other host */
746 struct net_device
*tdev
; /* Device to other host */
747 struct iphdr
*old_iph
= ip_hdr(skb
);
748 u8 tos
= old_iph
->tos
;
749 __be16 df
= old_iph
->frag_off
;
750 struct iphdr
*iph
; /* Our new IP header */
751 unsigned int max_headroom
; /* The extra header space needed */
757 if (!(rt
= __ip_vs_get_out_rt(skb
, cp
->dest
, cp
->daddr
.ip
,
760 if (rt
->rt_flags
& RTCF_LOCAL
) {
762 IP_VS_XMIT(NFPROTO_IPV4
, skb
, cp
, 1);
767 mtu
= dst_mtu(&rt
->dst
) - sizeof(struct iphdr
);
769 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__
);
773 skb_dst(skb
)->ops
->update_pmtu(skb_dst(skb
), mtu
);
775 df
|= (old_iph
->frag_off
& htons(IP_DF
));
777 if ((old_iph
->frag_off
& htons(IP_DF
) &&
778 mtu
< ntohs(old_iph
->tot_len
) && !skb_is_gso(skb
))) {
779 icmp_send(skb
, ICMP_DEST_UNREACH
,ICMP_FRAG_NEEDED
, htonl(mtu
));
780 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
785 * Okay, now see if we can stuff it in the buffer as-is.
787 max_headroom
= LL_RESERVED_SPACE(tdev
) + sizeof(struct iphdr
);
789 if (skb_headroom(skb
) < max_headroom
790 || skb_cloned(skb
) || skb_shared(skb
)) {
791 struct sk_buff
*new_skb
=
792 skb_realloc_headroom(skb
, max_headroom
);
796 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
801 old_iph
= ip_hdr(skb
);
804 skb
->transport_header
= skb
->network_header
;
806 /* fix old IP header checksum */
807 ip_send_check(old_iph
);
809 skb_push(skb
, sizeof(struct iphdr
));
810 skb_reset_network_header(skb
);
811 memset(&(IPCB(skb
)->opt
), 0, sizeof(IPCB(skb
)->opt
));
815 skb_dst_set(skb
, &rt
->dst
);
818 * Push down and install the IPIP header.
822 iph
->ihl
= sizeof(struct iphdr
)>>2;
824 iph
->protocol
= IPPROTO_IPIP
;
826 iph
->daddr
= rt
->rt_dst
;
827 iph
->saddr
= rt
->rt_src
;
828 iph
->ttl
= old_iph
->ttl
;
829 ip_select_ident(iph
, &rt
->dst
, NULL
);
831 /* Another hack: avoid icmp_send in ip_fragment */
834 ret
= IP_VS_XMIT_TUNNEL(skb
, cp
);
835 if (ret
== NF_ACCEPT
)
837 else if (ret
== NF_DROP
)
845 dst_link_failure(skb
);
855 #ifdef CONFIG_IP_VS_IPV6
857 ip_vs_tunnel_xmit_v6(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
858 struct ip_vs_protocol
*pp
)
860 struct rt6_info
*rt
; /* Route to the other host */
861 struct in6_addr saddr
; /* Source for tunnel */
862 struct net_device
*tdev
; /* Device to other host */
863 struct ipv6hdr
*old_iph
= ipv6_hdr(skb
);
864 struct ipv6hdr
*iph
; /* Our new IP header */
865 unsigned int max_headroom
; /* The extra header space needed */
871 if (!(rt
= __ip_vs_get_out_rt_v6(skb
, cp
->dest
, &cp
->daddr
.in6
,
874 if (__ip_vs_is_local_route6(rt
)) {
875 dst_release(&rt
->dst
);
876 IP_VS_XMIT(NFPROTO_IPV6
, skb
, cp
, 1);
881 mtu
= dst_mtu(&rt
->dst
) - sizeof(struct ipv6hdr
);
882 if (mtu
< IPV6_MIN_MTU
) {
883 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__
,
888 skb_dst(skb
)->ops
->update_pmtu(skb_dst(skb
), mtu
);
890 if (mtu
< ntohs(old_iph
->payload_len
) + sizeof(struct ipv6hdr
) &&
893 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
895 skb
->dev
= net
->loopback_dev
;
897 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
898 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
903 * Okay, now see if we can stuff it in the buffer as-is.
905 max_headroom
= LL_RESERVED_SPACE(tdev
) + sizeof(struct ipv6hdr
);
907 if (skb_headroom(skb
) < max_headroom
908 || skb_cloned(skb
) || skb_shared(skb
)) {
909 struct sk_buff
*new_skb
=
910 skb_realloc_headroom(skb
, max_headroom
);
912 dst_release(&rt
->dst
);
914 IP_VS_ERR_RL("%s(): no memory\n", __func__
);
919 old_iph
= ipv6_hdr(skb
);
922 skb
->transport_header
= skb
->network_header
;
924 skb_push(skb
, sizeof(struct ipv6hdr
));
925 skb_reset_network_header(skb
);
926 memset(&(IPCB(skb
)->opt
), 0, sizeof(IPCB(skb
)->opt
));
930 skb_dst_set(skb
, &rt
->dst
);
933 * Push down and install the IPIP header.
937 iph
->nexthdr
= IPPROTO_IPV6
;
938 iph
->payload_len
= old_iph
->payload_len
;
939 be16_add_cpu(&iph
->payload_len
, sizeof(*old_iph
));
940 iph
->priority
= old_iph
->priority
;
941 memset(&iph
->flow_lbl
, 0, sizeof(iph
->flow_lbl
));
942 ipv6_addr_copy(&iph
->daddr
, &cp
->daddr
.in6
);
943 ipv6_addr_copy(&iph
->saddr
, &saddr
);
944 iph
->hop_limit
= old_iph
->hop_limit
;
946 /* Another hack: avoid icmp_send in ip_fragment */
949 ret
= IP_VS_XMIT_TUNNEL(skb
, cp
);
950 if (ret
== NF_ACCEPT
)
952 else if (ret
== NF_DROP
)
960 dst_link_failure(skb
);
966 dst_release(&rt
->dst
);
973 * Direct Routing transmitter
974 * Used for ANY protocol
977 ip_vs_dr_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
978 struct ip_vs_protocol
*pp
)
980 struct rtable
*rt
; /* Route to the other host */
981 struct iphdr
*iph
= ip_hdr(skb
);
986 if (!(rt
= __ip_vs_get_out_rt(skb
, cp
->dest
, cp
->daddr
.ip
,
987 RT_TOS(iph
->tos
), 1|2)))
989 if (rt
->rt_flags
& RTCF_LOCAL
) {
991 IP_VS_XMIT(NFPROTO_IPV4
, skb
, cp
, 1);
995 mtu
= dst_mtu(&rt
->dst
);
996 if ((iph
->frag_off
& htons(IP_DF
)) && skb
->len
> mtu
&&
998 icmp_send(skb
, ICMP_DEST_UNREACH
,ICMP_FRAG_NEEDED
, htonl(mtu
));
1000 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
1005 * Call ip_send_check because we are not sure it is called
1006 * after ip_defrag. Is copy-on-write needed?
1008 if (unlikely((skb
= skb_share_check(skb
, GFP_ATOMIC
)) == NULL
)) {
1012 ip_send_check(ip_hdr(skb
));
1014 /* drop old route */
1016 skb_dst_set(skb
, &rt
->dst
);
1018 /* Another hack: avoid icmp_send in ip_fragment */
1021 IP_VS_XMIT(NFPROTO_IPV4
, skb
, cp
, 0);
1027 dst_link_failure(skb
);
1034 #ifdef CONFIG_IP_VS_IPV6
1036 ip_vs_dr_xmit_v6(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
1037 struct ip_vs_protocol
*pp
)
1039 struct rt6_info
*rt
; /* Route to the other host */
1044 if (!(rt
= __ip_vs_get_out_rt_v6(skb
, cp
->dest
, &cp
->daddr
.in6
, NULL
,
1047 if (__ip_vs_is_local_route6(rt
)) {
1048 dst_release(&rt
->dst
);
1049 IP_VS_XMIT(NFPROTO_IPV6
, skb
, cp
, 1);
1053 mtu
= dst_mtu(&rt
->dst
);
1054 if (skb
->len
> mtu
) {
1056 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
1058 skb
->dev
= net
->loopback_dev
;
1060 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
1061 dst_release(&rt
->dst
);
1062 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
1067 * Call ip_send_check because we are not sure it is called
1068 * after ip_defrag. Is copy-on-write needed?
1070 skb
= skb_share_check(skb
, GFP_ATOMIC
);
1071 if (unlikely(skb
== NULL
)) {
1072 dst_release(&rt
->dst
);
1076 /* drop old route */
1078 skb_dst_set(skb
, &rt
->dst
);
1080 /* Another hack: avoid icmp_send in ip_fragment */
1083 IP_VS_XMIT(NFPROTO_IPV6
, skb
, cp
, 0);
1089 dst_link_failure(skb
);
1099 * ICMP packet transmitter
1100 * called by the ip_vs_in_icmp
1103 ip_vs_icmp_xmit(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
1104 struct ip_vs_protocol
*pp
, int offset
)
1106 struct rtable
*rt
; /* Route to the other host */
1113 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1114 forwarded directly here, because there is no need to
1115 translate address/port back */
1116 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
) {
1117 if (cp
->packet_xmit
)
1118 rc
= cp
->packet_xmit(skb
, cp
, pp
);
1121 /* do not touch skb anymore */
1122 atomic_inc(&cp
->in_pkts
);
1127 * mangle and send the packet here (only for VS/NAT)
1130 if (!(rt
= __ip_vs_get_out_rt(skb
, cp
->dest
, cp
->daddr
.ip
,
1131 RT_TOS(ip_hdr(skb
)->tos
), 1|2|4)))
1133 local
= rt
->rt_flags
& RTCF_LOCAL
;
1136 * Avoid duplicate tuple in reply direction for NAT traffic
1137 * to local address when connection is sync-ed
1139 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1140 if (cp
->flags
& IP_VS_CONN_F_SYNC
&& local
) {
1141 enum ip_conntrack_info ctinfo
;
1142 struct nf_conn
*ct
= ct
= nf_ct_get(skb
, &ctinfo
);
1144 if (ct
&& !nf_ct_is_untracked(ct
)) {
1145 IP_VS_DBG(10, "%s(): "
1146 "stopping DNAT to local address %pI4\n",
1147 __func__
, &cp
->daddr
.ip
);
1153 /* From world but DNAT to loopback address? */
1154 if (local
&& ipv4_is_loopback(rt
->rt_dst
) &&
1155 rt_is_input_route(skb_rtable(skb
))) {
1156 IP_VS_DBG(1, "%s(): "
1157 "stopping DNAT to loopback %pI4\n",
1158 __func__
, &cp
->daddr
.ip
);
1163 mtu
= dst_mtu(&rt
->dst
);
1164 if ((skb
->len
> mtu
) && (ip_hdr(skb
)->frag_off
& htons(IP_DF
)) &&
1166 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_FRAG_NEEDED
, htonl(mtu
));
1167 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
1171 /* copy-on-write the packet before mangling it */
1172 if (!skb_make_writable(skb
, offset
))
1175 if (skb_cow(skb
, rt
->dst
.dev
->hard_header_len
))
1178 ip_vs_nat_icmp(skb
, pp
, cp
, 0);
1181 /* drop the old route when skb is not shared */
1183 skb_dst_set(skb
, &rt
->dst
);
1187 * Some IPv4 replies get local address from routes,
1188 * not from iph, so while we DNAT after routing
1189 * we need this second input/output route.
1191 if (!__ip_vs_reroute_locally(skb
))
1195 /* Another hack: avoid icmp_send in ip_fragment */
1198 IP_VS_XMIT_NAT(NFPROTO_IPV4
, skb
, cp
, local
);
1204 dst_link_failure(skb
);
1216 #ifdef CONFIG_IP_VS_IPV6
1218 ip_vs_icmp_xmit_v6(struct sk_buff
*skb
, struct ip_vs_conn
*cp
,
1219 struct ip_vs_protocol
*pp
, int offset
)
1221 struct rt6_info
*rt
; /* Route to the other host */
1228 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1229 forwarded directly here, because there is no need to
1230 translate address/port back */
1231 if (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
) {
1232 if (cp
->packet_xmit
)
1233 rc
= cp
->packet_xmit(skb
, cp
, pp
);
1236 /* do not touch skb anymore */
1237 atomic_inc(&cp
->in_pkts
);
1242 * mangle and send the packet here (only for VS/NAT)
1245 if (!(rt
= __ip_vs_get_out_rt_v6(skb
, cp
->dest
, &cp
->daddr
.in6
, NULL
,
1249 local
= __ip_vs_is_local_route6(rt
);
1251 * Avoid duplicate tuple in reply direction for NAT traffic
1252 * to local address when connection is sync-ed
1254 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1255 if (cp
->flags
& IP_VS_CONN_F_SYNC
&& local
) {
1256 enum ip_conntrack_info ctinfo
;
1257 struct nf_conn
*ct
= ct
= nf_ct_get(skb
, &ctinfo
);
1259 if (ct
&& !nf_ct_is_untracked(ct
)) {
1260 IP_VS_DBG(10, "%s(): "
1261 "stopping DNAT to local address %pI6\n",
1262 __func__
, &cp
->daddr
.in6
);
1268 /* From world but DNAT to loopback address? */
1269 if (local
&& skb
->dev
&& !(skb
->dev
->flags
& IFF_LOOPBACK
) &&
1270 ipv6_addr_type(&rt
->rt6i_dst
.addr
) & IPV6_ADDR_LOOPBACK
) {
1271 IP_VS_DBG(1, "%s(): "
1272 "stopping DNAT to loopback %pI6\n",
1273 __func__
, &cp
->daddr
.in6
);
1278 mtu
= dst_mtu(&rt
->dst
);
1279 if (skb
->len
> mtu
&& !skb_is_gso(skb
)) {
1281 struct net
*net
= dev_net(skb_dst(skb
)->dev
);
1283 skb
->dev
= net
->loopback_dev
;
1285 icmpv6_send(skb
, ICMPV6_PKT_TOOBIG
, 0, mtu
);
1286 IP_VS_DBG_RL("%s(): frag needed\n", __func__
);
1290 /* copy-on-write the packet before mangling it */
1291 if (!skb_make_writable(skb
, offset
))
1294 if (skb_cow(skb
, rt
->dst
.dev
->hard_header_len
))
1297 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 0);
1299 if (!local
|| !skb
->dev
) {
1300 /* drop the old route when skb is not shared */
1302 skb_dst_set(skb
, &rt
->dst
);
1304 /* destined to loopback, do we need to change route? */
1305 dst_release(&rt
->dst
);
1308 /* Another hack: avoid icmp_send in ip_fragment */
1311 IP_VS_XMIT_NAT(NFPROTO_IPV6
, skb
, cp
, local
);
1317 dst_link_failure(skb
);
1325 dst_release(&rt
->dst
);