1 /*
2 * ip_vs_xmit.c: various packet transmitters for IPVS
3 *
4 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
5 * Julian Anastasov <ja@ssi.bg>
6 *
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.
11 *
12 * Changes:
13 *
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
22 * LOCAL_OUT rules:
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
26 */
27
28 #define KMSG_COMPONENT "IPVS"
29 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30
31 #include <linux/kernel.h>
32 #include <linux/slab.h>
33 #include <linux/tcp.h> /* for tcphdr */
34 #include <net/ip.h>
35 #include <net/tcp.h> /* for csum_tcpudp_magic */
36 #include <net/udp.h>
37 #include <net/icmp.h> /* for icmp_send */
38 #include <net/route.h> /* for ip_route_output */
39 #include <net/ipv6.h>
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>
46
47 #include <net/ip_vs.h>
48
49 enum {
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
53 * local
54 */
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 */
58 };
59
ip_vs_dest_dst_alloc(void)60 static inline struct ip_vs_dest_dst *ip_vs_dest_dst_alloc(void)
61 {
62 return kmalloc(sizeof(struct ip_vs_dest_dst), GFP_ATOMIC);
63 }
64
ip_vs_dest_dst_free(struct ip_vs_dest_dst * dest_dst)65 static inline void ip_vs_dest_dst_free(struct ip_vs_dest_dst *dest_dst)
66 {
67 kfree(dest_dst);
68 }
69
70 /*
71 * Destination cache to speed up outgoing route lookup
72 */
73 static inline void
__ip_vs_dst_set(struct ip_vs_dest * dest,struct ip_vs_dest_dst * dest_dst,struct dst_entry * dst,u32 dst_cookie)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)
76 {
77 struct ip_vs_dest_dst *old;
78
79 old = rcu_dereference_protected(dest->dest_dst,
80 lockdep_is_held(&dest->dst_lock));
81
82 if (dest_dst) {
83 dest_dst->dst_cache = dst;
84 dest_dst->dst_cookie = dst_cookie;
85 }
86 rcu_assign_pointer(dest->dest_dst, dest_dst);
87
88 if (old)
89 call_rcu(&old->rcu_head, ip_vs_dest_dst_rcu_free);
90 }
91
92 static inline struct ip_vs_dest_dst *
__ip_vs_dst_check(struct ip_vs_dest * dest)93 __ip_vs_dst_check(struct ip_vs_dest *dest)
94 {
95 struct ip_vs_dest_dst *dest_dst = rcu_dereference(dest->dest_dst);
96 struct dst_entry *dst;
97
98 if (!dest_dst)
99 return NULL;
100 dst = dest_dst->dst_cache;
101 if (dst->obsolete &&
102 dst->ops->check(dst, dest_dst->dst_cookie) == NULL)
103 return NULL;
104 return dest_dst;
105 }
106
107 static inline bool
__mtu_check_toobig_v6(const struct sk_buff * skb,u32 mtu)108 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
109 {
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.
113 */
114 if (IP6CB(skb)->frag_max_size > mtu)
115 return true; /* largest fragment violate MTU */
116 }
117 else if (skb->len > mtu && !skb_is_gso(skb)) {
118 return true; /* Packet size violate MTU size */
119 }
120 return false;
121 }
122
123 /* Get route to daddr, update *saddr, optionally bind route to saddr */
do_output_route4(struct net * net,__be32 daddr,int rt_mode,__be32 * saddr)124 static struct rtable *do_output_route4(struct net *net, __be32 daddr,
125 int rt_mode, __be32 *saddr)
126 {
127 struct flowi4 fl4;
128 struct rtable *rt;
129 int loop = 0;
130
131 memset(&fl4, 0, sizeof(fl4));
132 fl4.daddr = daddr;
133 fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
134 FLOWI_FLAG_KNOWN_NH : 0;
135
136 retry:
137 rt = ip_route_output_key(net, &fl4);
138 if (IS_ERR(rt)) {
139 /* Invalid saddr ? */
140 if (PTR_ERR(rt) == -EINVAL && *saddr &&
141 rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
142 *saddr = 0;
143 flowi4_update_output(&fl4, 0, 0, daddr, 0);
144 goto retry;
145 }
146 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
147 return NULL;
148 } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
149 ip_rt_put(rt);
150 *saddr = fl4.saddr;
151 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
152 loop++;
153 goto retry;
154 }
155 *saddr = fl4.saddr;
156 return rt;
157 }
158
159 #ifdef CONFIG_IP_VS_IPV6
__ip_vs_is_local_route6(struct rt6_info * rt)160 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
161 {
162 return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
163 }
164 #endif
165
crosses_local_route_boundary(int skb_af,struct sk_buff * skb,int rt_mode,bool new_rt_is_local)166 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
167 int rt_mode,
168 bool new_rt_is_local)
169 {
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;
175
176 #ifdef CONFIG_IP_VS_IPV6
177 if (skb_af == AF_INET6) {
178 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
179
180 source_is_loopback =
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));
185 } else
186 #endif
187 {
188 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
189 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
190 }
191
192 if (unlikely(new_rt_is_local)) {
193 if (!rt_mode_allow_local)
194 return true;
195 if (!rt_mode_allow_redirect && !old_rt_is_local)
196 return true;
197 } else {
198 if (!rt_mode_allow_non_local)
199 return true;
200 if (source_is_loopback)
201 return true;
202 }
203 return false;
204 }
205
maybe_update_pmtu(int skb_af,struct sk_buff * skb,int mtu)206 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
207 {
208 struct sock *sk = skb->sk;
209 struct rtable *ort = skb_rtable(skb);
210
211 if (!skb->dev && sk && sk_fullsock(sk))
212 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu, true);
213 }
214
ensure_mtu_is_adequate(struct netns_ipvs * ipvs,int skb_af,int rt_mode,struct ip_vs_iphdr * ipvsh,struct sk_buff * skb,int mtu)215 static inline bool ensure_mtu_is_adequate(struct netns_ipvs *ipvs, int skb_af,
216 int rt_mode,
217 struct ip_vs_iphdr *ipvsh,
218 struct sk_buff *skb, int mtu)
219 {
220 #ifdef CONFIG_IP_VS_IPV6
221 if (skb_af == AF_INET6) {
222 struct net *net = ipvs->net;
223
224 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
225 if (!skb->dev)
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);
232 return false;
233 }
234 } else
235 #endif
236 {
237 /* If we're going to tunnel the packet and pmtu discovery
238 * is disabled, we'll just fragment it anyway
239 */
240 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
241 return true;
242
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,
247 htonl(mtu));
248 IP_VS_DBG(1, "frag needed for %pI4\n",
249 &ip_hdr(skb)->saddr);
250 return false;
251 }
252 }
253
254 return true;
255 }
256
decrement_ttl(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb)257 static inline bool decrement_ttl(struct netns_ipvs *ipvs,
258 int skb_af,
259 struct sk_buff *skb)
260 {
261 struct net *net = ipvs->net;
262
263 #ifdef CONFIG_IP_VS_IPV6
264 if (skb_af == AF_INET6) {
265 struct dst_entry *dst = skb_dst(skb);
266
267 /* check and decrement ttl */
268 if (ipv6_hdr(skb)->hop_limit <= 1) {
269 /* Force OUTPUT device used as source address */
270 skb->dev = dst->dev;
271 icmpv6_send(skb, ICMPV6_TIME_EXCEED,
272 ICMPV6_EXC_HOPLIMIT, 0);
273 __IP6_INC_STATS(net, ip6_dst_idev(dst),
274 IPSTATS_MIB_INHDRERRORS);
275
276 return false;
277 }
278
279 /* don't propagate ttl change to cloned packets */
280 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
281 return false;
282
283 ipv6_hdr(skb)->hop_limit--;
284 } else
285 #endif
286 {
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);
291 return false;
292 }
293
294 /* don't propagate ttl change to cloned packets */
295 if (!skb_make_writable(skb, sizeof(struct iphdr)))
296 return false;
297
298 /* Decrease ttl */
299 ip_decrease_ttl(ip_hdr(skb));
300 }
301
302 return true;
303 }
304
305 /* Get route to destination or remote server */
306 static int
__ip_vs_get_out_rt(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb,struct ip_vs_dest * dest,__be32 daddr,int rt_mode,__be32 * ret_saddr,struct ip_vs_iphdr * ipvsh)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)
311 {
312 struct net *net = ipvs->net;
313 struct ip_vs_dest_dst *dest_dst;
314 struct rtable *rt; /* Route to the other host */
315 int mtu;
316 int local, noref = 1;
317
318 if (dest) {
319 dest_dst = __ip_vs_dst_check(dest);
320 if (likely(dest_dst))
321 rt = (struct rtable *) dest_dst->dst_cache;
322 else {
323 dest_dst = ip_vs_dest_dst_alloc();
324 spin_lock_bh(&dest->dst_lock);
325 if (!dest_dst) {
326 __ip_vs_dst_set(dest, NULL, NULL, 0);
327 spin_unlock_bh(&dest->dst_lock);
328 goto err_unreach;
329 }
330 rt = do_output_route4(net, dest->addr.ip, rt_mode,
331 &dest_dst->dst_saddr.ip);
332 if (!rt) {
333 __ip_vs_dst_set(dest, NULL, NULL, 0);
334 spin_unlock_bh(&dest->dst_lock);
335 ip_vs_dest_dst_free(dest_dst);
336 goto err_unreach;
337 }
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));
343 }
344 if (ret_saddr)
345 *ret_saddr = dest_dst->dst_saddr.ip;
346 } else {
347 __be32 saddr = htonl(INADDR_ANY);
348
349 noref = 0;
350
351 /* For such unconfigured boxes avoid many route lookups
352 * for performance reasons because we do not remember saddr
353 */
354 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
355 rt = do_output_route4(net, daddr, rt_mode, &saddr);
356 if (!rt)
357 goto err_unreach;
358 if (ret_saddr)
359 *ret_saddr = saddr;
360 }
361
362 local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
363 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
364 local))) {
365 IP_VS_DBG_RL("We are crossing local and non-local addresses"
366 " daddr=%pI4\n", &daddr);
367 goto err_put;
368 }
369
370 if (unlikely(local)) {
371 /* skb to local stack, preserve old route */
372 if (!noref)
373 ip_rt_put(rt);
374 return local;
375 }
376
377 if (!decrement_ttl(ipvs, skb_af, skb))
378 goto err_put;
379
380 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
381 mtu = dst_mtu(&rt->dst);
382 } else {
383 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
384 if (mtu < 68) {
385 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
386 goto err_put;
387 }
388 maybe_update_pmtu(skb_af, skb, mtu);
389 }
390
391 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
392 goto err_put;
393
394 skb_dst_drop(skb);
395 if (noref) {
396 if (!local)
397 skb_dst_set_noref(skb, &rt->dst);
398 else
399 skb_dst_set(skb, dst_clone(&rt->dst));
400 } else
401 skb_dst_set(skb, &rt->dst);
402
403 return local;
404
405 err_put:
406 if (!noref)
407 ip_rt_put(rt);
408 return -1;
409
410 err_unreach:
411 dst_link_failure(skb);
412 return -1;
413 }
414
415 #ifdef CONFIG_IP_VS_IPV6
416 static struct dst_entry *
__ip_vs_route_output_v6(struct net * net,struct in6_addr * daddr,struct in6_addr * ret_saddr,int do_xfrm,int rt_mode)417 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
418 struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
419 {
420 struct dst_entry *dst;
421 struct flowi6 fl6 = {
422 .daddr = *daddr,
423 };
424
425 if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
426 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
427
428 dst = ip6_route_output(net, NULL, &fl6);
429 if (dst->error)
430 goto out_err;
431 if (!ret_saddr)
432 return dst;
433 if (ipv6_addr_any(&fl6.saddr) &&
434 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
435 &fl6.daddr, 0, &fl6.saddr) < 0)
436 goto out_err;
437 if (do_xfrm) {
438 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
439 if (IS_ERR(dst)) {
440 dst = NULL;
441 goto out_err;
442 }
443 }
444 *ret_saddr = fl6.saddr;
445 return dst;
446
447 out_err:
448 dst_release(dst);
449 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
450 return NULL;
451 }
452
453 /*
454 * Get route to destination or remote server
455 */
456 static int
__ip_vs_get_out_rt_v6(struct netns_ipvs * ipvs,int skb_af,struct sk_buff * skb,struct ip_vs_dest * dest,struct in6_addr * daddr,struct in6_addr * ret_saddr,struct ip_vs_iphdr * ipvsh,int do_xfrm,int rt_mode)457 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
458 struct ip_vs_dest *dest,
459 struct in6_addr *daddr, struct in6_addr *ret_saddr,
460 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
461 {
462 struct net *net = ipvs->net;
463 struct ip_vs_dest_dst *dest_dst;
464 struct rt6_info *rt; /* Route to the other host */
465 struct dst_entry *dst;
466 int mtu;
467 int local, noref = 1;
468
469 if (dest) {
470 dest_dst = __ip_vs_dst_check(dest);
471 if (likely(dest_dst))
472 rt = (struct rt6_info *) dest_dst->dst_cache;
473 else {
474 u32 cookie;
475
476 dest_dst = ip_vs_dest_dst_alloc();
477 spin_lock_bh(&dest->dst_lock);
478 if (!dest_dst) {
479 __ip_vs_dst_set(dest, NULL, NULL, 0);
480 spin_unlock_bh(&dest->dst_lock);
481 goto err_unreach;
482 }
483 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
484 &dest_dst->dst_saddr.in6,
485 do_xfrm, rt_mode);
486 if (!dst) {
487 __ip_vs_dst_set(dest, NULL, NULL, 0);
488 spin_unlock_bh(&dest->dst_lock);
489 ip_vs_dest_dst_free(dest_dst);
490 goto err_unreach;
491 }
492 rt = (struct rt6_info *) dst;
493 cookie = rt6_get_cookie(rt);
494 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
495 spin_unlock_bh(&dest->dst_lock);
496 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
497 &dest->addr.in6, &dest_dst->dst_saddr.in6,
498 atomic_read(&rt->dst.__refcnt));
499 }
500 if (ret_saddr)
501 *ret_saddr = dest_dst->dst_saddr.in6;
502 } else {
503 noref = 0;
504 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
505 rt_mode);
506 if (!dst)
507 goto err_unreach;
508 rt = (struct rt6_info *) dst;
509 }
510
511 local = __ip_vs_is_local_route6(rt);
512
513 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
514 local))) {
515 IP_VS_DBG_RL("We are crossing local and non-local addresses"
516 " daddr=%pI6\n", daddr);
517 goto err_put;
518 }
519
520 if (unlikely(local)) {
521 /* skb to local stack, preserve old route */
522 if (!noref)
523 dst_release(&rt->dst);
524 return local;
525 }
526
527 if (!decrement_ttl(ipvs, skb_af, skb))
528 goto err_put;
529
530 /* MTU checking */
531 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
532 mtu = dst_mtu(&rt->dst);
533 else {
534 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
535 if (mtu < IPV6_MIN_MTU) {
536 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
537 IPV6_MIN_MTU);
538 goto err_put;
539 }
540 maybe_update_pmtu(skb_af, skb, mtu);
541 }
542
543 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
544 goto err_put;
545
546 skb_dst_drop(skb);
547 if (noref) {
548 if (!local)
549 skb_dst_set_noref(skb, &rt->dst);
550 else
551 skb_dst_set(skb, dst_clone(&rt->dst));
552 } else
553 skb_dst_set(skb, &rt->dst);
554
555 return local;
556
557 err_put:
558 if (!noref)
559 dst_release(&rt->dst);
560 return -1;
561
562 err_unreach:
563 /* The ip6_link_failure function requires the dev field to be set
564 * in order to get the net (further for the sake of fwmark
565 * reflection).
566 */
567 if (!skb->dev)
568 skb->dev = skb_dst(skb)->dev;
569
570 dst_link_failure(skb);
571 return -1;
572 }
573 #endif
574
575
576 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
ip_vs_tunnel_xmit_prepare(struct sk_buff * skb,struct ip_vs_conn * cp)577 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
578 struct ip_vs_conn *cp)
579 {
580 int ret = NF_ACCEPT;
581
582 skb->ipvs_property = 1;
583 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
584 ret = ip_vs_confirm_conntrack(skb);
585 if (ret == NF_ACCEPT) {
586 nf_reset(skb);
587 skb_forward_csum(skb);
588 }
589 return ret;
590 }
591
592 /* In the event of a remote destination, it's possible that we would have
593 * matches against an old socket (particularly a TIME-WAIT socket). This
594 * causes havoc down the line (ip_local_out et. al. expect regular sockets
595 * and invalid memory accesses will happen) so simply drop the association
596 * in this case.
597 */
ip_vs_drop_early_demux_sk(struct sk_buff * skb)598 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
599 {
600 /* If dev is set, the packet came from the LOCAL_IN callback and
601 * not from a local TCP socket.
602 */
603 if (skb->dev)
604 skb_orphan(skb);
605 }
606
607 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
ip_vs_nat_send_or_cont(int pf,struct sk_buff * skb,struct ip_vs_conn * cp,int local)608 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
609 struct ip_vs_conn *cp, int local)
610 {
611 int ret = NF_STOLEN;
612
613 skb->ipvs_property = 1;
614 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
615 ip_vs_notrack(skb);
616 else
617 ip_vs_update_conntrack(skb, cp, 1);
618
619 /* Remove the early_demux association unless it's bound for the
620 * exact same port and address on this host after translation.
621 */
622 if (!local || cp->vport != cp->dport ||
623 !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
624 ip_vs_drop_early_demux_sk(skb);
625
626 if (!local) {
627 skb_forward_csum(skb);
628 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
629 NULL, skb_dst(skb)->dev, dst_output);
630 } else
631 ret = NF_ACCEPT;
632
633 return ret;
634 }
635
636 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
ip_vs_send_or_cont(int pf,struct sk_buff * skb,struct ip_vs_conn * cp,int local)637 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
638 struct ip_vs_conn *cp, int local)
639 {
640 int ret = NF_STOLEN;
641
642 skb->ipvs_property = 1;
643 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
644 ip_vs_notrack(skb);
645 if (!local) {
646 ip_vs_drop_early_demux_sk(skb);
647 skb_forward_csum(skb);
648 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
649 NULL, skb_dst(skb)->dev, dst_output);
650 } else
651 ret = NF_ACCEPT;
652 return ret;
653 }
654
655
656 /*
657 * NULL transmitter (do nothing except return NF_ACCEPT)
658 */
659 int
ip_vs_null_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)660 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
661 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
662 {
663 /* we do not touch skb and do not need pskb ptr */
664 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
665 }
666
667
668 /*
669 * Bypass transmitter
670 * Let packets bypass the destination when the destination is not
671 * available, it may be only used in transparent cache cluster.
672 */
673 int
ip_vs_bypass_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)674 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
675 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
676 {
677 struct iphdr *iph = ip_hdr(skb);
678
679 EnterFunction(10);
680
681 if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
682 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
683 goto tx_error;
684
685 ip_send_check(iph);
686
687 /* Another hack: avoid icmp_send in ip_fragment */
688 skb->ignore_df = 1;
689
690 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
691
692 LeaveFunction(10);
693 return NF_STOLEN;
694
695 tx_error:
696 kfree_skb(skb);
697 LeaveFunction(10);
698 return NF_STOLEN;
699 }
700
701 #ifdef CONFIG_IP_VS_IPV6
702 int
ip_vs_bypass_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)703 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
704 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
705 {
706 struct ipv6hdr *iph = ipv6_hdr(skb);
707
708 EnterFunction(10);
709
710 if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
711 &iph->daddr, NULL,
712 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
713 goto tx_error;
714
715 /* Another hack: avoid icmp_send in ip_fragment */
716 skb->ignore_df = 1;
717
718 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
719
720 LeaveFunction(10);
721 return NF_STOLEN;
722
723 tx_error:
724 kfree_skb(skb);
725 LeaveFunction(10);
726 return NF_STOLEN;
727 }
728 #endif
729
730 /*
731 * NAT transmitter (only for outside-to-inside nat forwarding)
732 * Not used for related ICMP
733 */
734 int
ip_vs_nat_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)735 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
736 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
737 {
738 struct rtable *rt; /* Route to the other host */
739 int local, rc, was_input;
740
741 EnterFunction(10);
742
743 /* check if it is a connection of no-client-port */
744 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
745 __be16 _pt, *p;
746
747 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
748 if (p == NULL)
749 goto tx_error;
750 ip_vs_conn_fill_cport(cp, *p);
751 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
752 }
753
754 was_input = rt_is_input_route(skb_rtable(skb));
755 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
756 IP_VS_RT_MODE_LOCAL |
757 IP_VS_RT_MODE_NON_LOCAL |
758 IP_VS_RT_MODE_RDR, NULL, ipvsh);
759 if (local < 0)
760 goto tx_error;
761 rt = skb_rtable(skb);
762 /*
763 * Avoid duplicate tuple in reply direction for NAT traffic
764 * to local address when connection is sync-ed
765 */
766 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
767 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
768 enum ip_conntrack_info ctinfo;
769 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
770
771 if (ct) {
772 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
773 "ip_vs_nat_xmit(): "
774 "stopping DNAT to local address");
775 goto tx_error;
776 }
777 }
778 #endif
779
780 /* From world but DNAT to loopback address? */
781 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
782 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
783 "ip_vs_nat_xmit(): stopping DNAT to loopback "
784 "address");
785 goto tx_error;
786 }
787
788 /* copy-on-write the packet before mangling it */
789 if (!skb_make_writable(skb, sizeof(struct iphdr)))
790 goto tx_error;
791
792 if (skb_cow(skb, rt->dst.dev->hard_header_len))
793 goto tx_error;
794
795 /* mangle the packet */
796 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
797 goto tx_error;
798 ip_hdr(skb)->daddr = cp->daddr.ip;
799 ip_send_check(ip_hdr(skb));
800
801 IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
802
803 /* FIXME: when application helper enlarges the packet and the length
804 is larger than the MTU of outgoing device, there will be still
805 MTU problem. */
806
807 /* Another hack: avoid icmp_send in ip_fragment */
808 skb->ignore_df = 1;
809
810 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
811
812 LeaveFunction(10);
813 return rc;
814
815 tx_error:
816 kfree_skb(skb);
817 LeaveFunction(10);
818 return NF_STOLEN;
819 }
820
821 #ifdef CONFIG_IP_VS_IPV6
822 int
ip_vs_nat_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)823 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
824 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
825 {
826 struct rt6_info *rt; /* Route to the other host */
827 int local, rc;
828
829 EnterFunction(10);
830
831 /* check if it is a connection of no-client-port */
832 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
833 __be16 _pt, *p;
834 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
835 if (p == NULL)
836 goto tx_error;
837 ip_vs_conn_fill_cport(cp, *p);
838 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
839 }
840
841 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
842 &cp->daddr.in6,
843 NULL, ipvsh, 0,
844 IP_VS_RT_MODE_LOCAL |
845 IP_VS_RT_MODE_NON_LOCAL |
846 IP_VS_RT_MODE_RDR);
847 if (local < 0)
848 goto tx_error;
849 rt = (struct rt6_info *) skb_dst(skb);
850 /*
851 * Avoid duplicate tuple in reply direction for NAT traffic
852 * to local address when connection is sync-ed
853 */
854 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
855 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
856 enum ip_conntrack_info ctinfo;
857 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
858
859 if (ct) {
860 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
861 "ip_vs_nat_xmit_v6(): "
862 "stopping DNAT to local address");
863 goto tx_error;
864 }
865 }
866 #endif
867
868 /* From world but DNAT to loopback address? */
869 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
870 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
871 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
872 "ip_vs_nat_xmit_v6(): "
873 "stopping DNAT to loopback address");
874 goto tx_error;
875 }
876
877 /* copy-on-write the packet before mangling it */
878 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
879 goto tx_error;
880
881 if (skb_cow(skb, rt->dst.dev->hard_header_len))
882 goto tx_error;
883
884 /* mangle the packet */
885 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
886 goto tx_error;
887 ipv6_hdr(skb)->daddr = cp->daddr.in6;
888
889 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
890
891 /* FIXME: when application helper enlarges the packet and the length
892 is larger than the MTU of outgoing device, there will be still
893 MTU problem. */
894
895 /* Another hack: avoid icmp_send in ip_fragment */
896 skb->ignore_df = 1;
897
898 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
899
900 LeaveFunction(10);
901 return rc;
902
903 tx_error:
904 LeaveFunction(10);
905 kfree_skb(skb);
906 return NF_STOLEN;
907 }
908 #endif
909
910 /* When forwarding a packet, we must ensure that we've got enough headroom
911 * for the encapsulation packet in the skb. This also gives us an
912 * opportunity to figure out what the payload_len, dsfield, ttl, and df
913 * values should be, so that we won't need to look at the old ip header
914 * again
915 */
916 static struct sk_buff *
ip_vs_prepare_tunneled_skb(struct sk_buff * skb,int skb_af,unsigned int max_headroom,__u8 * next_protocol,__u32 * payload_len,__u8 * dsfield,__u8 * ttl,__be16 * df)917 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
918 unsigned int max_headroom, __u8 *next_protocol,
919 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
920 __be16 *df)
921 {
922 struct sk_buff *new_skb = NULL;
923 struct iphdr *old_iph = NULL;
924 __u8 old_dsfield;
925 #ifdef CONFIG_IP_VS_IPV6
926 struct ipv6hdr *old_ipv6h = NULL;
927 #endif
928
929 ip_vs_drop_early_demux_sk(skb);
930
931 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
932 new_skb = skb_realloc_headroom(skb, max_headroom);
933 if (!new_skb)
934 goto error;
935 if (skb->sk)
936 skb_set_owner_w(new_skb, skb->sk);
937 consume_skb(skb);
938 skb = new_skb;
939 }
940
941 #ifdef CONFIG_IP_VS_IPV6
942 if (skb_af == AF_INET6) {
943 old_ipv6h = ipv6_hdr(skb);
944 *next_protocol = IPPROTO_IPV6;
945 if (payload_len)
946 *payload_len =
947 ntohs(old_ipv6h->payload_len) +
948 sizeof(*old_ipv6h);
949 old_dsfield = ipv6_get_dsfield(old_ipv6h);
950 *ttl = old_ipv6h->hop_limit;
951 if (df)
952 *df = 0;
953 } else
954 #endif
955 {
956 old_iph = ip_hdr(skb);
957 /* Copy DF, reset fragment offset and MF */
958 if (df)
959 *df = (old_iph->frag_off & htons(IP_DF));
960 *next_protocol = IPPROTO_IPIP;
961
962 /* fix old IP header checksum */
963 ip_send_check(old_iph);
964 old_dsfield = ipv4_get_dsfield(old_iph);
965 *ttl = old_iph->ttl;
966 if (payload_len)
967 *payload_len = ntohs(old_iph->tot_len);
968 }
969
970 /* Implement full-functionality option for ECN encapsulation */
971 *dsfield = INET_ECN_encapsulate(old_dsfield, old_dsfield);
972
973 return skb;
974 error:
975 kfree_skb(skb);
976 return ERR_PTR(-ENOMEM);
977 }
978
__tun_gso_type_mask(int encaps_af,int orig_af)979 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
980 {
981 switch (encaps_af) {
982 case AF_INET:
983 return SKB_GSO_IPXIP4;
984 case AF_INET6:
985 return SKB_GSO_IPXIP6;
986 default:
987 return 0;
988 }
989 }
990
991 /*
992 * IP Tunneling transmitter
993 *
994 * This function encapsulates the packet in a new IP packet, its
995 * destination will be set to cp->daddr. Most code of this function
996 * is taken from ipip.c.
997 *
998 * It is used in VS/TUN cluster. The load balancer selects a real
999 * server from a cluster based on a scheduling algorithm,
1000 * encapsulates the request packet and forwards it to the selected
1001 * server. For example, all real servers are configured with
1002 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
1003 * the encapsulated packet, it will decapsulate the packet, processe
1004 * the request and return the response packets directly to the client
1005 * without passing the load balancer. This can greatly increase the
1006 * scalability of virtual server.
1007 *
1008 * Used for ANY protocol
1009 */
1010 int
ip_vs_tunnel_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1011 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1012 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1013 {
1014 struct netns_ipvs *ipvs = cp->ipvs;
1015 struct net *net = ipvs->net;
1016 struct rtable *rt; /* Route to the other host */
1017 __be32 saddr; /* Source for tunnel */
1018 struct net_device *tdev; /* Device to other host */
1019 __u8 next_protocol = 0;
1020 __u8 dsfield = 0;
1021 __u8 ttl = 0;
1022 __be16 df = 0;
1023 __be16 *dfp = NULL;
1024 struct iphdr *iph; /* Our new IP header */
1025 unsigned int max_headroom; /* The extra header space needed */
1026 int ret, local;
1027
1028 EnterFunction(10);
1029
1030 local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1031 IP_VS_RT_MODE_LOCAL |
1032 IP_VS_RT_MODE_NON_LOCAL |
1033 IP_VS_RT_MODE_CONNECT |
1034 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
1035 if (local < 0)
1036 goto tx_error;
1037 if (local)
1038 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1039
1040 rt = skb_rtable(skb);
1041 tdev = rt->dst.dev;
1042
1043 /*
1044 * Okay, now see if we can stuff it in the buffer as-is.
1045 */
1046 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1047
1048 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1049 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1050 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1051 &next_protocol, NULL, &dsfield,
1052 &ttl, dfp);
1053 if (IS_ERR(skb))
1054 goto tx_error;
1055
1056 if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET, cp->af)))
1057 goto tx_error;
1058
1059 skb->transport_header = skb->network_header;
1060
1061 skb_push(skb, sizeof(struct iphdr));
1062 skb_reset_network_header(skb);
1063 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1064
1065 /*
1066 * Push down and install the IPIP header.
1067 */
1068 iph = ip_hdr(skb);
1069 iph->version = 4;
1070 iph->ihl = sizeof(struct iphdr)>>2;
1071 iph->frag_off = df;
1072 iph->protocol = next_protocol;
1073 iph->tos = dsfield;
1074 iph->daddr = cp->daddr.ip;
1075 iph->saddr = saddr;
1076 iph->ttl = ttl;
1077 ip_select_ident(net, skb, NULL);
1078
1079 /* Another hack: avoid icmp_send in ip_fragment */
1080 skb->ignore_df = 1;
1081
1082 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1083 if (ret == NF_ACCEPT)
1084 ip_local_out(net, skb->sk, skb);
1085 else if (ret == NF_DROP)
1086 kfree_skb(skb);
1087
1088 LeaveFunction(10);
1089
1090 return NF_STOLEN;
1091
1092 tx_error:
1093 if (!IS_ERR(skb))
1094 kfree_skb(skb);
1095 LeaveFunction(10);
1096 return NF_STOLEN;
1097 }
1098
1099 #ifdef CONFIG_IP_VS_IPV6
1100 int
ip_vs_tunnel_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1101 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1102 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1103 {
1104 struct rt6_info *rt; /* Route to the other host */
1105 struct in6_addr saddr; /* Source for tunnel */
1106 struct net_device *tdev; /* Device to other host */
1107 __u8 next_protocol = 0;
1108 __u32 payload_len = 0;
1109 __u8 dsfield = 0;
1110 __u8 ttl = 0;
1111 struct ipv6hdr *iph; /* Our new IP header */
1112 unsigned int max_headroom; /* The extra header space needed */
1113 int ret, local;
1114
1115 EnterFunction(10);
1116
1117 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1118 &cp->daddr.in6,
1119 &saddr, ipvsh, 1,
1120 IP_VS_RT_MODE_LOCAL |
1121 IP_VS_RT_MODE_NON_LOCAL |
1122 IP_VS_RT_MODE_TUNNEL);
1123 if (local < 0)
1124 goto tx_error;
1125 if (local)
1126 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1127
1128 rt = (struct rt6_info *) skb_dst(skb);
1129 tdev = rt->dst.dev;
1130
1131 /*
1132 * Okay, now see if we can stuff it in the buffer as-is.
1133 */
1134 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1135
1136 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1137 &next_protocol, &payload_len,
1138 &dsfield, &ttl, NULL);
1139 if (IS_ERR(skb))
1140 goto tx_error;
1141
1142 if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET6, cp->af)))
1143 goto tx_error;
1144
1145 skb->transport_header = skb->network_header;
1146
1147 skb_push(skb, sizeof(struct ipv6hdr));
1148 skb_reset_network_header(skb);
1149 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1150
1151 /*
1152 * Push down and install the IPIP header.
1153 */
1154 iph = ipv6_hdr(skb);
1155 iph->version = 6;
1156 iph->nexthdr = next_protocol;
1157 iph->payload_len = htons(payload_len);
1158 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1159 ipv6_change_dsfield(iph, 0, dsfield);
1160 iph->daddr = cp->daddr.in6;
1161 iph->saddr = saddr;
1162 iph->hop_limit = ttl;
1163
1164 /* Another hack: avoid icmp_send in ip_fragment */
1165 skb->ignore_df = 1;
1166
1167 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1168 if (ret == NF_ACCEPT)
1169 ip6_local_out(cp->ipvs->net, skb->sk, skb);
1170 else if (ret == NF_DROP)
1171 kfree_skb(skb);
1172
1173 LeaveFunction(10);
1174
1175 return NF_STOLEN;
1176
1177 tx_error:
1178 if (!IS_ERR(skb))
1179 kfree_skb(skb);
1180 LeaveFunction(10);
1181 return NF_STOLEN;
1182 }
1183 #endif
1184
1185
1186 /*
1187 * Direct Routing transmitter
1188 * Used for ANY protocol
1189 */
1190 int
ip_vs_dr_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1191 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1192 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1193 {
1194 int local;
1195
1196 EnterFunction(10);
1197
1198 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1199 IP_VS_RT_MODE_LOCAL |
1200 IP_VS_RT_MODE_NON_LOCAL |
1201 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1202 if (local < 0)
1203 goto tx_error;
1204 if (local)
1205 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1206
1207 ip_send_check(ip_hdr(skb));
1208
1209 /* Another hack: avoid icmp_send in ip_fragment */
1210 skb->ignore_df = 1;
1211
1212 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1213
1214 LeaveFunction(10);
1215 return NF_STOLEN;
1216
1217 tx_error:
1218 kfree_skb(skb);
1219 LeaveFunction(10);
1220 return NF_STOLEN;
1221 }
1222
1223 #ifdef CONFIG_IP_VS_IPV6
1224 int
ip_vs_dr_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,struct ip_vs_iphdr * ipvsh)1225 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1226 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1227 {
1228 int local;
1229
1230 EnterFunction(10);
1231
1232 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1233 &cp->daddr.in6,
1234 NULL, ipvsh, 0,
1235 IP_VS_RT_MODE_LOCAL |
1236 IP_VS_RT_MODE_NON_LOCAL |
1237 IP_VS_RT_MODE_KNOWN_NH);
1238 if (local < 0)
1239 goto tx_error;
1240 if (local)
1241 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1242
1243 /* Another hack: avoid icmp_send in ip_fragment */
1244 skb->ignore_df = 1;
1245
1246 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1247
1248 LeaveFunction(10);
1249 return NF_STOLEN;
1250
1251 tx_error:
1252 kfree_skb(skb);
1253 LeaveFunction(10);
1254 return NF_STOLEN;
1255 }
1256 #endif
1257
1258
1259 /*
1260 * ICMP packet transmitter
1261 * called by the ip_vs_in_icmp
1262 */
1263 int
ip_vs_icmp_xmit(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,int offset,unsigned int hooknum,struct ip_vs_iphdr * iph)1264 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1265 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1266 struct ip_vs_iphdr *iph)
1267 {
1268 struct rtable *rt; /* Route to the other host */
1269 int rc;
1270 int local;
1271 int rt_mode, was_input;
1272
1273 EnterFunction(10);
1274
1275 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1276 forwarded directly here, because there is no need to
1277 translate address/port back */
1278 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1279 if (cp->packet_xmit)
1280 rc = cp->packet_xmit(skb, cp, pp, iph);
1281 else
1282 rc = NF_ACCEPT;
1283 /* do not touch skb anymore */
1284 atomic_inc(&cp->in_pkts);
1285 goto out;
1286 }
1287
1288 /*
1289 * mangle and send the packet here (only for VS/NAT)
1290 */
1291 was_input = rt_is_input_route(skb_rtable(skb));
1292
1293 /* LOCALNODE from FORWARD hook is not supported */
1294 rt_mode = (hooknum != NF_INET_FORWARD) ?
1295 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1296 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1297 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1298 NULL, iph);
1299 if (local < 0)
1300 goto tx_error;
1301 rt = skb_rtable(skb);
1302
1303 /*
1304 * Avoid duplicate tuple in reply direction for NAT traffic
1305 * to local address when connection is sync-ed
1306 */
1307 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1308 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1309 enum ip_conntrack_info ctinfo;
1310 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1311
1312 if (ct) {
1313 IP_VS_DBG(10, "%s(): "
1314 "stopping DNAT to local address %pI4\n",
1315 __func__, &cp->daddr.ip);
1316 goto tx_error;
1317 }
1318 }
1319 #endif
1320
1321 /* From world but DNAT to loopback address? */
1322 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1323 IP_VS_DBG(1, "%s(): "
1324 "stopping DNAT to loopback %pI4\n",
1325 __func__, &cp->daddr.ip);
1326 goto tx_error;
1327 }
1328
1329 /* copy-on-write the packet before mangling it */
1330 if (!skb_make_writable(skb, offset))
1331 goto tx_error;
1332
1333 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1334 goto tx_error;
1335
1336 ip_vs_nat_icmp(skb, pp, cp, 0);
1337
1338 /* Another hack: avoid icmp_send in ip_fragment */
1339 skb->ignore_df = 1;
1340
1341 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1342 goto out;
1343
1344 tx_error:
1345 kfree_skb(skb);
1346 rc = NF_STOLEN;
1347 out:
1348 LeaveFunction(10);
1349 return rc;
1350 }
1351
1352 #ifdef CONFIG_IP_VS_IPV6
1353 int
ip_vs_icmp_xmit_v6(struct sk_buff * skb,struct ip_vs_conn * cp,struct ip_vs_protocol * pp,int offset,unsigned int hooknum,struct ip_vs_iphdr * ipvsh)1354 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1355 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1356 struct ip_vs_iphdr *ipvsh)
1357 {
1358 struct rt6_info *rt; /* Route to the other host */
1359 int rc;
1360 int local;
1361 int rt_mode;
1362
1363 EnterFunction(10);
1364
1365 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1366 forwarded directly here, because there is no need to
1367 translate address/port back */
1368 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1369 if (cp->packet_xmit)
1370 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1371 else
1372 rc = NF_ACCEPT;
1373 /* do not touch skb anymore */
1374 atomic_inc(&cp->in_pkts);
1375 goto out;
1376 }
1377
1378 /*
1379 * mangle and send the packet here (only for VS/NAT)
1380 */
1381
1382 /* LOCALNODE from FORWARD hook is not supported */
1383 rt_mode = (hooknum != NF_INET_FORWARD) ?
1384 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1385 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1386 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1387 &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1388 if (local < 0)
1389 goto tx_error;
1390 rt = (struct rt6_info *) skb_dst(skb);
1391 /*
1392 * Avoid duplicate tuple in reply direction for NAT traffic
1393 * to local address when connection is sync-ed
1394 */
1395 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1396 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1397 enum ip_conntrack_info ctinfo;
1398 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1399
1400 if (ct) {
1401 IP_VS_DBG(10, "%s(): "
1402 "stopping DNAT to local address %pI6\n",
1403 __func__, &cp->daddr.in6);
1404 goto tx_error;
1405 }
1406 }
1407 #endif
1408
1409 /* From world but DNAT to loopback address? */
1410 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1411 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1412 IP_VS_DBG(1, "%s(): "
1413 "stopping DNAT to loopback %pI6\n",
1414 __func__, &cp->daddr.in6);
1415 goto tx_error;
1416 }
1417
1418 /* copy-on-write the packet before mangling it */
1419 if (!skb_make_writable(skb, offset))
1420 goto tx_error;
1421
1422 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1423 goto tx_error;
1424
1425 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1426
1427 /* Another hack: avoid icmp_send in ip_fragment */
1428 skb->ignore_df = 1;
1429
1430 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1431 goto out;
1432
1433 tx_error:
1434 kfree_skb(skb);
1435 rc = NF_STOLEN;
1436 out:
1437 LeaveFunction(10);
1438 return rc;
1439 }
1440 #endif
1441