projects / deliverable / linux.git / blob
? search:


5b5ec0b688de34b9031cedaea2956f4833842bfe
[deliverable/linux.git] / net / netfilter / ipvs / ip_vs_xmit.c
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
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
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
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 *
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
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 */
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
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
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
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);
213 }
214
215 static inline bool ensure_mtu_is_adequate(int skb_af, int rt_mode,
216 struct ip_vs_iphdr *ipvsh,
217 struct sk_buff *skb, int mtu)
218 {
219 #ifdef CONFIG_IP_VS_IPV6
220 if (skb_af == AF_INET6) {
221 struct net *net = dev_net(skb_dst(skb)->dev);
222
223 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
224 if (!skb->dev)
225 skb->dev = net->loopback_dev;
226 /* only send ICMP too big on first fragment */
227 if (!ipvsh->fragoffs && !ip_vs_iph_icmp(ipvsh))
228 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
229 IP_VS_DBG(1, "frag needed for %pI6c\n",
230 &ipv6_hdr(skb)->saddr);
231 return false;
232 }
233 } else
234 #endif
235 {
236 struct netns_ipvs *ipvs = net_ipvs(skb_net(skb));
237
238 /* If we're going to tunnel the packet and pmtu discovery
239 * is disabled, we'll just fragment it anyway
240 */
241 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
242 return true;
243
244 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
245 skb->len > mtu && !skb_is_gso(skb) &&
246 !ip_vs_iph_icmp(ipvsh))) {
247 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
248 htonl(mtu));
249 IP_VS_DBG(1, "frag needed for %pI4\n",
250 &ip_hdr(skb)->saddr);
251 return false;
252 }
253 }
254
255 return true;
256 }
257
258 /* Get route to destination or remote server */
259 static int
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)
263 {
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 */
267 int mtu;
268 int local, noref = 1;
269
270 if (dest) {
271 dest_dst = __ip_vs_dst_check(dest);
272 if (likely(dest_dst))
273 rt = (struct rtable *) dest_dst->dst_cache;
274 else {
275 dest_dst = ip_vs_dest_dst_alloc();
276 spin_lock_bh(&dest->dst_lock);
277 if (!dest_dst) {
278 __ip_vs_dst_set(dest, NULL, NULL, 0);
279 spin_unlock_bh(&dest->dst_lock);
280 goto err_unreach;
281 }
282 rt = do_output_route4(net, dest->addr.ip, rt_mode,
283 &dest_dst->dst_saddr.ip);
284 if (!rt) {
285 __ip_vs_dst_set(dest, NULL, NULL, 0);
286 spin_unlock_bh(&dest->dst_lock);
287 ip_vs_dest_dst_free(dest_dst);
288 goto err_unreach;
289 }
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));
295 }
296 if (ret_saddr)
297 *ret_saddr = dest_dst->dst_saddr.ip;
298 } else {
299 __be32 saddr = htonl(INADDR_ANY);
300
301 noref = 0;
302
303 /* For such unconfigured boxes avoid many route lookups
304 * for performance reasons because we do not remember saddr
305 */
306 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
307 rt = do_output_route4(net, daddr, rt_mode, &saddr);
308 if (!rt)
309 goto err_unreach;
310 if (ret_saddr)
311 *ret_saddr = saddr;
312 }
313
314 local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
315 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
316 local))) {
317 IP_VS_DBG_RL("We are crossing local and non-local addresses"
318 " daddr=%pI4\n", &daddr);
319 goto err_put;
320 }
321
322 if (unlikely(local)) {
323 /* skb to local stack, preserve old route */
324 if (!noref)
325 ip_rt_put(rt);
326 return local;
327 }
328
329 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
330 mtu = dst_mtu(&rt->dst);
331 } else {
332 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
333 if (mtu < 68) {
334 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
335 goto err_put;
336 }
337 maybe_update_pmtu(skb_af, skb, mtu);
338 }
339
340 if (!ensure_mtu_is_adequate(skb_af, rt_mode, ipvsh, skb, mtu))
341 goto err_put;
342
343 skb_dst_drop(skb);
344 if (noref) {
345 if (!local)
346 skb_dst_set_noref(skb, &rt->dst);
347 else
348 skb_dst_set(skb, dst_clone(&rt->dst));
349 } else
350 skb_dst_set(skb, &rt->dst);
351
352 return local;
353
354 err_put:
355 if (!noref)
356 ip_rt_put(rt);
357 return -1;
358
359 err_unreach:
360 dst_link_failure(skb);
361 return -1;
362 }
363
364 #ifdef CONFIG_IP_VS_IPV6
365 static struct dst_entry *
366 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
367 struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
368 {
369 struct dst_entry *dst;
370 struct flowi6 fl6 = {
371 .daddr = *daddr,
372 };
373
374 if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
375 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
376
377 dst = ip6_route_output(net, NULL, &fl6);
378 if (dst->error)
379 goto out_err;
380 if (!ret_saddr)
381 return dst;
382 if (ipv6_addr_any(&fl6.saddr) &&
383 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
384 &fl6.daddr, 0, &fl6.saddr) < 0)
385 goto out_err;
386 if (do_xfrm) {
387 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
388 if (IS_ERR(dst)) {
389 dst = NULL;
390 goto out_err;
391 }
392 }
393 *ret_saddr = fl6.saddr;
394 return dst;
395
396 out_err:
397 dst_release(dst);
398 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
399 return NULL;
400 }
401
402 /*
403 * Get route to destination or remote server
404 */
405 static int
406 __ip_vs_get_out_rt_v6(int skb_af, struct sk_buff *skb, struct ip_vs_dest *dest,
407 struct in6_addr *daddr, struct in6_addr *ret_saddr,
408 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
409 {
410 struct net *net = dev_net(skb_dst(skb)->dev);
411 struct ip_vs_dest_dst *dest_dst;
412 struct rt6_info *rt; /* Route to the other host */
413 struct dst_entry *dst;
414 int mtu;
415 int local, noref = 1;
416
417 if (dest) {
418 dest_dst = __ip_vs_dst_check(dest);
419 if (likely(dest_dst))
420 rt = (struct rt6_info *) dest_dst->dst_cache;
421 else {
422 u32 cookie;
423
424 dest_dst = ip_vs_dest_dst_alloc();
425 spin_lock_bh(&dest->dst_lock);
426 if (!dest_dst) {
427 __ip_vs_dst_set(dest, NULL, NULL, 0);
428 spin_unlock_bh(&dest->dst_lock);
429 goto err_unreach;
430 }
431 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
432 &dest_dst->dst_saddr.in6,
433 do_xfrm, rt_mode);
434 if (!dst) {
435 __ip_vs_dst_set(dest, NULL, NULL, 0);
436 spin_unlock_bh(&dest->dst_lock);
437 ip_vs_dest_dst_free(dest_dst);
438 goto err_unreach;
439 }
440 rt = (struct rt6_info *) dst;
441 cookie = rt6_get_cookie(rt);
442 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
443 spin_unlock_bh(&dest->dst_lock);
444 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
445 &dest->addr.in6, &dest_dst->dst_saddr.in6,
446 atomic_read(&rt->dst.__refcnt));
447 }
448 if (ret_saddr)
449 *ret_saddr = dest_dst->dst_saddr.in6;
450 } else {
451 noref = 0;
452 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
453 rt_mode);
454 if (!dst)
455 goto err_unreach;
456 rt = (struct rt6_info *) dst;
457 }
458
459 local = __ip_vs_is_local_route6(rt);
460
461 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
462 local))) {
463 IP_VS_DBG_RL("We are crossing local and non-local addresses"
464 " daddr=%pI6\n", daddr);
465 goto err_put;
466 }
467
468 if (unlikely(local)) {
469 /* skb to local stack, preserve old route */
470 if (!noref)
471 dst_release(&rt->dst);
472 return local;
473 }
474
475 /* MTU checking */
476 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
477 mtu = dst_mtu(&rt->dst);
478 else {
479 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
480 if (mtu < IPV6_MIN_MTU) {
481 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
482 IPV6_MIN_MTU);
483 goto err_put;
484 }
485 maybe_update_pmtu(skb_af, skb, mtu);
486 }
487
488 if (!ensure_mtu_is_adequate(skb_af, rt_mode, ipvsh, skb, mtu))
489 goto err_put;
490
491 skb_dst_drop(skb);
492 if (noref) {
493 if (!local)
494 skb_dst_set_noref(skb, &rt->dst);
495 else
496 skb_dst_set(skb, dst_clone(&rt->dst));
497 } else
498 skb_dst_set(skb, &rt->dst);
499
500 return local;
501
502 err_put:
503 if (!noref)
504 dst_release(&rt->dst);
505 return -1;
506
507 err_unreach:
508 /* The ip6_link_failure function requires the dev field to be set
509 * in order to get the net (further for the sake of fwmark
510 * reflection).
511 */
512 if (!skb->dev)
513 skb->dev = skb_dst(skb)->dev;
514
515 dst_link_failure(skb);
516 return -1;
517 }
518 #endif
519
520
521 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
522 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
523 struct ip_vs_conn *cp)
524 {
525 int ret = NF_ACCEPT;
526
527 skb->ipvs_property = 1;
528 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
529 ret = ip_vs_confirm_conntrack(skb);
530 if (ret == NF_ACCEPT) {
531 nf_reset(skb);
532 skb_forward_csum(skb);
533 if (!skb->sk)
534 skb_sender_cpu_clear(skb);
535 }
536 return ret;
537 }
538
539 /* In the event of a remote destination, it's possible that we would have
540 * matches against an old socket (particularly a TIME-WAIT socket). This
541 * causes havoc down the line (ip_local_out et. al. expect regular sockets
542 * and invalid memory accesses will happen) so simply drop the association
543 * in this case.
544 */
545 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
546 {
547 /* If dev is set, the packet came from the LOCAL_IN callback and
548 * not from a local TCP socket.
549 */
550 if (skb->dev)
551 skb_orphan(skb);
552 }
553
554 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
555 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
556 struct ip_vs_conn *cp, int local)
557 {
558 int ret = NF_STOLEN;
559
560 skb->ipvs_property = 1;
561 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
562 ip_vs_notrack(skb);
563 else
564 ip_vs_update_conntrack(skb, cp, 1);
565
566 /* Remove the early_demux association unless it's bound for the
567 * exact same port and address on this host after translation.
568 */
569 if (!local || cp->vport != cp->dport ||
570 !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
571 ip_vs_drop_early_demux_sk(skb);
572
573 if (!local) {
574 skb_forward_csum(skb);
575 if (!skb->sk)
576 skb_sender_cpu_clear(skb);
577 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
578 NULL, skb_dst(skb)->dev, dst_output_okfn);
579 } else
580 ret = NF_ACCEPT;
581
582 return ret;
583 }
584
585 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
586 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
587 struct ip_vs_conn *cp, int local)
588 {
589 int ret = NF_STOLEN;
590
591 skb->ipvs_property = 1;
592 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
593 ip_vs_notrack(skb);
594 if (!local) {
595 ip_vs_drop_early_demux_sk(skb);
596 skb_forward_csum(skb);
597 if (!skb->sk)
598 skb_sender_cpu_clear(skb);
599 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
600 NULL, skb_dst(skb)->dev, dst_output_okfn);
601 } else
602 ret = NF_ACCEPT;
603 return ret;
604 }
605
606
607 /*
608 * NULL transmitter (do nothing except return NF_ACCEPT)
609 */
610 int
611 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
612 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
613 {
614 /* we do not touch skb and do not need pskb ptr */
615 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
616 }
617
618
619 /*
620 * Bypass transmitter
621 * Let packets bypass the destination when the destination is not
622 * available, it may be only used in transparent cache cluster.
623 */
624 int
625 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
626 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
627 {
628 struct iphdr *iph = ip_hdr(skb);
629
630 EnterFunction(10);
631
632 rcu_read_lock();
633 if (__ip_vs_get_out_rt(cp->af, skb, NULL, iph->daddr,
634 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
635 goto tx_error;
636
637 ip_send_check(iph);
638
639 /* Another hack: avoid icmp_send in ip_fragment */
640 skb->ignore_df = 1;
641
642 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
643 rcu_read_unlock();
644
645 LeaveFunction(10);
646 return NF_STOLEN;
647
648 tx_error:
649 kfree_skb(skb);
650 rcu_read_unlock();
651 LeaveFunction(10);
652 return NF_STOLEN;
653 }
654
655 #ifdef CONFIG_IP_VS_IPV6
656 int
657 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
658 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
659 {
660 struct ipv6hdr *iph = ipv6_hdr(skb);
661
662 EnterFunction(10);
663
664 rcu_read_lock();
665 if (__ip_vs_get_out_rt_v6(cp->af, skb, NULL, &iph->daddr, NULL,
666 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
667 goto tx_error;
668
669 /* Another hack: avoid icmp_send in ip_fragment */
670 skb->ignore_df = 1;
671
672 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
673 rcu_read_unlock();
674
675 LeaveFunction(10);
676 return NF_STOLEN;
677
678 tx_error:
679 kfree_skb(skb);
680 rcu_read_unlock();
681 LeaveFunction(10);
682 return NF_STOLEN;
683 }
684 #endif
685
686 /*
687 * NAT transmitter (only for outside-to-inside nat forwarding)
688 * Not used for related ICMP
689 */
690 int
691 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
692 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
693 {
694 struct rtable *rt; /* Route to the other host */
695 int local, rc, was_input;
696
697 EnterFunction(10);
698
699 rcu_read_lock();
700 /* check if it is a connection of no-client-port */
701 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
702 __be16 _pt, *p;
703
704 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
705 if (p == NULL)
706 goto tx_error;
707 ip_vs_conn_fill_cport(cp, *p);
708 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
709 }
710
711 was_input = rt_is_input_route(skb_rtable(skb));
712 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
713 IP_VS_RT_MODE_LOCAL |
714 IP_VS_RT_MODE_NON_LOCAL |
715 IP_VS_RT_MODE_RDR, NULL, ipvsh);
716 if (local < 0)
717 goto tx_error;
718 rt = skb_rtable(skb);
719 /*
720 * Avoid duplicate tuple in reply direction for NAT traffic
721 * to local address when connection is sync-ed
722 */
723 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
724 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
725 enum ip_conntrack_info ctinfo;
726 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
727
728 if (ct && !nf_ct_is_untracked(ct)) {
729 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
730 "ip_vs_nat_xmit(): "
731 "stopping DNAT to local address");
732 goto tx_error;
733 }
734 }
735 #endif
736
737 /* From world but DNAT to loopback address? */
738 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
739 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
740 "ip_vs_nat_xmit(): stopping DNAT to loopback "
741 "address");
742 goto tx_error;
743 }
744
745 /* copy-on-write the packet before mangling it */
746 if (!skb_make_writable(skb, sizeof(struct iphdr)))
747 goto tx_error;
748
749 if (skb_cow(skb, rt->dst.dev->hard_header_len))
750 goto tx_error;
751
752 /* mangle the packet */
753 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
754 goto tx_error;
755 ip_hdr(skb)->daddr = cp->daddr.ip;
756 ip_send_check(ip_hdr(skb));
757
758 IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
759
760 /* FIXME: when application helper enlarges the packet and the length
761 is larger than the MTU of outgoing device, there will be still
762 MTU problem. */
763
764 /* Another hack: avoid icmp_send in ip_fragment */
765 skb->ignore_df = 1;
766
767 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
768 rcu_read_unlock();
769
770 LeaveFunction(10);
771 return rc;
772
773 tx_error:
774 kfree_skb(skb);
775 rcu_read_unlock();
776 LeaveFunction(10);
777 return NF_STOLEN;
778 }
779
780 #ifdef CONFIG_IP_VS_IPV6
781 int
782 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
783 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
784 {
785 struct rt6_info *rt; /* Route to the other host */
786 int local, rc;
787
788 EnterFunction(10);
789
790 rcu_read_lock();
791 /* check if it is a connection of no-client-port */
792 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
793 __be16 _pt, *p;
794 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
795 if (p == NULL)
796 goto tx_error;
797 ip_vs_conn_fill_cport(cp, *p);
798 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
799 }
800
801 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
802 NULL, ipvsh, 0,
803 IP_VS_RT_MODE_LOCAL |
804 IP_VS_RT_MODE_NON_LOCAL |
805 IP_VS_RT_MODE_RDR);
806 if (local < 0)
807 goto tx_error;
808 rt = (struct rt6_info *) skb_dst(skb);
809 /*
810 * Avoid duplicate tuple in reply direction for NAT traffic
811 * to local address when connection is sync-ed
812 */
813 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
814 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
815 enum ip_conntrack_info ctinfo;
816 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
817
818 if (ct && !nf_ct_is_untracked(ct)) {
819 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
820 "ip_vs_nat_xmit_v6(): "
821 "stopping DNAT to local address");
822 goto tx_error;
823 }
824 }
825 #endif
826
827 /* From world but DNAT to loopback address? */
828 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
829 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
830 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
831 "ip_vs_nat_xmit_v6(): "
832 "stopping DNAT to loopback address");
833 goto tx_error;
834 }
835
836 /* copy-on-write the packet before mangling it */
837 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
838 goto tx_error;
839
840 if (skb_cow(skb, rt->dst.dev->hard_header_len))
841 goto tx_error;
842
843 /* mangle the packet */
844 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
845 goto tx_error;
846 ipv6_hdr(skb)->daddr = cp->daddr.in6;
847
848 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
849
850 /* FIXME: when application helper enlarges the packet and the length
851 is larger than the MTU of outgoing device, there will be still
852 MTU problem. */
853
854 /* Another hack: avoid icmp_send in ip_fragment */
855 skb->ignore_df = 1;
856
857 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
858 rcu_read_unlock();
859
860 LeaveFunction(10);
861 return rc;
862
863 tx_error:
864 LeaveFunction(10);
865 kfree_skb(skb);
866 rcu_read_unlock();
867 return NF_STOLEN;
868 }
869 #endif
870
871 /* When forwarding a packet, we must ensure that we've got enough headroom
872 * for the encapsulation packet in the skb. This also gives us an
873 * opportunity to figure out what the payload_len, dsfield, ttl, and df
874 * values should be, so that we won't need to look at the old ip header
875 * again
876 */
877 static struct sk_buff *
878 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
879 unsigned int max_headroom, __u8 *next_protocol,
880 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
881 __be16 *df)
882 {
883 struct sk_buff *new_skb = NULL;
884 struct iphdr *old_iph = NULL;
885 #ifdef CONFIG_IP_VS_IPV6
886 struct ipv6hdr *old_ipv6h = NULL;
887 #endif
888
889 ip_vs_drop_early_demux_sk(skb);
890
891 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
892 new_skb = skb_realloc_headroom(skb, max_headroom);
893 if (!new_skb)
894 goto error;
895 if (skb->sk)
896 skb_set_owner_w(new_skb, skb->sk);
897 consume_skb(skb);
898 skb = new_skb;
899 }
900
901 #ifdef CONFIG_IP_VS_IPV6
902 if (skb_af == AF_INET6) {
903 old_ipv6h = ipv6_hdr(skb);
904 *next_protocol = IPPROTO_IPV6;
905 if (payload_len)
906 *payload_len =
907 ntohs(old_ipv6h->payload_len) +
908 sizeof(*old_ipv6h);
909 *dsfield = ipv6_get_dsfield(old_ipv6h);
910 *ttl = old_ipv6h->hop_limit;
911 if (df)
912 *df = 0;
913 } else
914 #endif
915 {
916 old_iph = ip_hdr(skb);
917 /* Copy DF, reset fragment offset and MF */
918 if (df)
919 *df = (old_iph->frag_off & htons(IP_DF));
920 *next_protocol = IPPROTO_IPIP;
921
922 /* fix old IP header checksum */
923 ip_send_check(old_iph);
924 *dsfield = ipv4_get_dsfield(old_iph);
925 *ttl = old_iph->ttl;
926 if (payload_len)
927 *payload_len = ntohs(old_iph->tot_len);
928 }
929
930 return skb;
931 error:
932 kfree_skb(skb);
933 return ERR_PTR(-ENOMEM);
934 }
935
936 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
937 {
938 if (encaps_af == AF_INET) {
939 if (orig_af == AF_INET)
940 return SKB_GSO_IPIP;
941
942 return SKB_GSO_SIT;
943 }
944
945 /* GSO: we need to provide proper SKB_GSO_ value for IPv6:
946 * SKB_GSO_SIT/IPV6
947 */
948 return 0;
949 }
950
951 /*
952 * IP Tunneling transmitter
953 *
954 * This function encapsulates the packet in a new IP packet, its
955 * destination will be set to cp->daddr. Most code of this function
956 * is taken from ipip.c.
957 *
958 * It is used in VS/TUN cluster. The load balancer selects a real
959 * server from a cluster based on a scheduling algorithm,
960 * encapsulates the request packet and forwards it to the selected
961 * server. For example, all real servers are configured with
962 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
963 * the encapsulated packet, it will decapsulate the packet, processe
964 * the request and return the response packets directly to the client
965 * without passing the load balancer. This can greatly increase the
966 * scalability of virtual server.
967 *
968 * Used for ANY protocol
969 */
970 int
971 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
972 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
973 {
974 struct net *net = skb_net(skb);
975 struct netns_ipvs *ipvs = net_ipvs(net);
976 struct rtable *rt; /* Route to the other host */
977 __be32 saddr; /* Source for tunnel */
978 struct net_device *tdev; /* Device to other host */
979 __u8 next_protocol = 0;
980 __u8 dsfield = 0;
981 __u8 ttl = 0;
982 __be16 df = 0;
983 __be16 *dfp = NULL;
984 struct iphdr *iph; /* Our new IP header */
985 unsigned int max_headroom; /* The extra header space needed */
986 int ret, local;
987
988 EnterFunction(10);
989
990 rcu_read_lock();
991 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
992 IP_VS_RT_MODE_LOCAL |
993 IP_VS_RT_MODE_NON_LOCAL |
994 IP_VS_RT_MODE_CONNECT |
995 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
996 if (local < 0)
997 goto tx_error;
998 if (local) {
999 rcu_read_unlock();
1000 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1001 }
1002
1003 rt = skb_rtable(skb);
1004 tdev = rt->dst.dev;
1005
1006 /*
1007 * Okay, now see if we can stuff it in the buffer as-is.
1008 */
1009 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1010
1011 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1012 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1013 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1014 &next_protocol, NULL, &dsfield,
1015 &ttl, dfp);
1016 if (IS_ERR(skb))
1017 goto tx_error;
1018
1019 skb = iptunnel_handle_offloads(
1020 skb, false, __tun_gso_type_mask(AF_INET, cp->af));
1021 if (IS_ERR(skb))
1022 goto tx_error;
1023
1024 skb->transport_header = skb->network_header;
1025
1026 skb_push(skb, sizeof(struct iphdr));
1027 skb_reset_network_header(skb);
1028 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1029
1030 /*
1031 * Push down and install the IPIP header.
1032 */
1033 iph = ip_hdr(skb);
1034 iph->version = 4;
1035 iph->ihl = sizeof(struct iphdr)>>2;
1036 iph->frag_off = df;
1037 iph->protocol = next_protocol;
1038 iph->tos = dsfield;
1039 iph->daddr = cp->daddr.ip;
1040 iph->saddr = saddr;
1041 iph->ttl = ttl;
1042 ip_select_ident(net, skb, NULL);
1043
1044 /* Another hack: avoid icmp_send in ip_fragment */
1045 skb->ignore_df = 1;
1046
1047 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1048 if (ret == NF_ACCEPT)
1049 ip_local_out(skb);
1050 else if (ret == NF_DROP)
1051 kfree_skb(skb);
1052 rcu_read_unlock();
1053
1054 LeaveFunction(10);
1055
1056 return NF_STOLEN;
1057
1058 tx_error:
1059 if (!IS_ERR(skb))
1060 kfree_skb(skb);
1061 rcu_read_unlock();
1062 LeaveFunction(10);
1063 return NF_STOLEN;
1064 }
1065
1066 #ifdef CONFIG_IP_VS_IPV6
1067 int
1068 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1069 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1070 {
1071 struct rt6_info *rt; /* Route to the other host */
1072 struct in6_addr saddr; /* Source for tunnel */
1073 struct net_device *tdev; /* Device to other host */
1074 __u8 next_protocol = 0;
1075 __u32 payload_len = 0;
1076 __u8 dsfield = 0;
1077 __u8 ttl = 0;
1078 struct ipv6hdr *iph; /* Our new IP header */
1079 unsigned int max_headroom; /* The extra header space needed */
1080 int ret, local;
1081
1082 EnterFunction(10);
1083
1084 rcu_read_lock();
1085 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1086 &saddr, ipvsh, 1,
1087 IP_VS_RT_MODE_LOCAL |
1088 IP_VS_RT_MODE_NON_LOCAL |
1089 IP_VS_RT_MODE_TUNNEL);
1090 if (local < 0)
1091 goto tx_error;
1092 if (local) {
1093 rcu_read_unlock();
1094 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1095 }
1096
1097 rt = (struct rt6_info *) skb_dst(skb);
1098 tdev = rt->dst.dev;
1099
1100 /*
1101 * Okay, now see if we can stuff it in the buffer as-is.
1102 */
1103 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1104
1105 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1106 &next_protocol, &payload_len,
1107 &dsfield, &ttl, NULL);
1108 if (IS_ERR(skb))
1109 goto tx_error;
1110
1111 skb = iptunnel_handle_offloads(
1112 skb, false, __tun_gso_type_mask(AF_INET6, cp->af));
1113 if (IS_ERR(skb))
1114 goto tx_error;
1115
1116 skb->transport_header = skb->network_header;
1117
1118 skb_push(skb, sizeof(struct ipv6hdr));
1119 skb_reset_network_header(skb);
1120 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1121
1122 /*
1123 * Push down and install the IPIP header.
1124 */
1125 iph = ipv6_hdr(skb);
1126 iph->version = 6;
1127 iph->nexthdr = next_protocol;
1128 iph->payload_len = htons(payload_len);
1129 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1130 ipv6_change_dsfield(iph, 0, dsfield);
1131 iph->daddr = cp->daddr.in6;
1132 iph->saddr = saddr;
1133 iph->hop_limit = ttl;
1134
1135 /* Another hack: avoid icmp_send in ip_fragment */
1136 skb->ignore_df = 1;
1137
1138 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1139 if (ret == NF_ACCEPT)
1140 ip6_local_out(skb);
1141 else if (ret == NF_DROP)
1142 kfree_skb(skb);
1143 rcu_read_unlock();
1144
1145 LeaveFunction(10);
1146
1147 return NF_STOLEN;
1148
1149 tx_error:
1150 if (!IS_ERR(skb))
1151 kfree_skb(skb);
1152 rcu_read_unlock();
1153 LeaveFunction(10);
1154 return NF_STOLEN;
1155 }
1156 #endif
1157
1158
1159 /*
1160 * Direct Routing transmitter
1161 * Used for ANY protocol
1162 */
1163 int
1164 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1165 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1166 {
1167 int local;
1168
1169 EnterFunction(10);
1170
1171 rcu_read_lock();
1172 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip,
1173 IP_VS_RT_MODE_LOCAL |
1174 IP_VS_RT_MODE_NON_LOCAL |
1175 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1176 if (local < 0)
1177 goto tx_error;
1178 if (local) {
1179 rcu_read_unlock();
1180 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1181 }
1182
1183 ip_send_check(ip_hdr(skb));
1184
1185 /* Another hack: avoid icmp_send in ip_fragment */
1186 skb->ignore_df = 1;
1187
1188 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1189 rcu_read_unlock();
1190
1191 LeaveFunction(10);
1192 return NF_STOLEN;
1193
1194 tx_error:
1195 kfree_skb(skb);
1196 rcu_read_unlock();
1197 LeaveFunction(10);
1198 return NF_STOLEN;
1199 }
1200
1201 #ifdef CONFIG_IP_VS_IPV6
1202 int
1203 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1204 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1205 {
1206 int local;
1207
1208 EnterFunction(10);
1209
1210 rcu_read_lock();
1211 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1212 NULL, ipvsh, 0,
1213 IP_VS_RT_MODE_LOCAL |
1214 IP_VS_RT_MODE_NON_LOCAL |
1215 IP_VS_RT_MODE_KNOWN_NH);
1216 if (local < 0)
1217 goto tx_error;
1218 if (local) {
1219 rcu_read_unlock();
1220 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1221 }
1222
1223 /* Another hack: avoid icmp_send in ip_fragment */
1224 skb->ignore_df = 1;
1225
1226 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1227 rcu_read_unlock();
1228
1229 LeaveFunction(10);
1230 return NF_STOLEN;
1231
1232 tx_error:
1233 kfree_skb(skb);
1234 rcu_read_unlock();
1235 LeaveFunction(10);
1236 return NF_STOLEN;
1237 }
1238 #endif
1239
1240
1241 /*
1242 * ICMP packet transmitter
1243 * called by the ip_vs_in_icmp
1244 */
1245 int
1246 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1247 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1248 struct ip_vs_iphdr *iph)
1249 {
1250 struct rtable *rt; /* Route to the other host */
1251 int rc;
1252 int local;
1253 int rt_mode, was_input;
1254
1255 EnterFunction(10);
1256
1257 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1258 forwarded directly here, because there is no need to
1259 translate address/port back */
1260 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1261 if (cp->packet_xmit)
1262 rc = cp->packet_xmit(skb, cp, pp, iph);
1263 else
1264 rc = NF_ACCEPT;
1265 /* do not touch skb anymore */
1266 atomic_inc(&cp->in_pkts);
1267 goto out;
1268 }
1269
1270 /*
1271 * mangle and send the packet here (only for VS/NAT)
1272 */
1273 was_input = rt_is_input_route(skb_rtable(skb));
1274
1275 /* LOCALNODE from FORWARD hook is not supported */
1276 rt_mode = (hooknum != NF_INET_FORWARD) ?
1277 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1278 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1279 rcu_read_lock();
1280 local = __ip_vs_get_out_rt(cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1281 NULL, iph);
1282 if (local < 0)
1283 goto tx_error;
1284 rt = skb_rtable(skb);
1285
1286 /*
1287 * Avoid duplicate tuple in reply direction for NAT traffic
1288 * to local address when connection is sync-ed
1289 */
1290 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1291 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1292 enum ip_conntrack_info ctinfo;
1293 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1294
1295 if (ct && !nf_ct_is_untracked(ct)) {
1296 IP_VS_DBG(10, "%s(): "
1297 "stopping DNAT to local address %pI4\n",
1298 __func__, &cp->daddr.ip);
1299 goto tx_error;
1300 }
1301 }
1302 #endif
1303
1304 /* From world but DNAT to loopback address? */
1305 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1306 IP_VS_DBG(1, "%s(): "
1307 "stopping DNAT to loopback %pI4\n",
1308 __func__, &cp->daddr.ip);
1309 goto tx_error;
1310 }
1311
1312 /* copy-on-write the packet before mangling it */
1313 if (!skb_make_writable(skb, offset))
1314 goto tx_error;
1315
1316 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1317 goto tx_error;
1318
1319 ip_vs_nat_icmp(skb, pp, cp, 0);
1320
1321 /* Another hack: avoid icmp_send in ip_fragment */
1322 skb->ignore_df = 1;
1323
1324 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1325 rcu_read_unlock();
1326 goto out;
1327
1328 tx_error:
1329 kfree_skb(skb);
1330 rcu_read_unlock();
1331 rc = NF_STOLEN;
1332 out:
1333 LeaveFunction(10);
1334 return rc;
1335 }
1336
1337 #ifdef CONFIG_IP_VS_IPV6
1338 int
1339 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1340 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1341 struct ip_vs_iphdr *ipvsh)
1342 {
1343 struct rt6_info *rt; /* Route to the other host */
1344 int rc;
1345 int local;
1346 int rt_mode;
1347
1348 EnterFunction(10);
1349
1350 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1351 forwarded directly here, because there is no need to
1352 translate address/port back */
1353 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1354 if (cp->packet_xmit)
1355 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1356 else
1357 rc = NF_ACCEPT;
1358 /* do not touch skb anymore */
1359 atomic_inc(&cp->in_pkts);
1360 goto out;
1361 }
1362
1363 /*
1364 * mangle and send the packet here (only for VS/NAT)
1365 */
1366
1367 /* LOCALNODE from FORWARD hook is not supported */
1368 rt_mode = (hooknum != NF_INET_FORWARD) ?
1369 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1370 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1371 rcu_read_lock();
1372 local = __ip_vs_get_out_rt_v6(cp->af, skb, cp->dest, &cp->daddr.in6,
1373 NULL, ipvsh, 0, rt_mode);
1374 if (local < 0)
1375 goto tx_error;
1376 rt = (struct rt6_info *) skb_dst(skb);
1377 /*
1378 * Avoid duplicate tuple in reply direction for NAT traffic
1379 * to local address when connection is sync-ed
1380 */
1381 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1382 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1383 enum ip_conntrack_info ctinfo;
1384 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1385
1386 if (ct && !nf_ct_is_untracked(ct)) {
1387 IP_VS_DBG(10, "%s(): "
1388 "stopping DNAT to local address %pI6\n",
1389 __func__, &cp->daddr.in6);
1390 goto tx_error;
1391 }
1392 }
1393 #endif
1394
1395 /* From world but DNAT to loopback address? */
1396 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1397 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1398 IP_VS_DBG(1, "%s(): "
1399 "stopping DNAT to loopback %pI6\n",
1400 __func__, &cp->daddr.in6);
1401 goto tx_error;
1402 }
1403
1404 /* copy-on-write the packet before mangling it */
1405 if (!skb_make_writable(skb, offset))
1406 goto tx_error;
1407
1408 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1409 goto tx_error;
1410
1411 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1412
1413 /* Another hack: avoid icmp_send in ip_fragment */
1414 skb->ignore_df = 1;
1415
1416 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1417 rcu_read_unlock();
1418 goto out;
1419
1420 tx_error:
1421 kfree_skb(skb);
1422 rcu_read_unlock();
1423 rc = NF_STOLEN;
1424 out:
1425 LeaveFunction(10);
1426 return rc;
1427 }
1428 #endif
Linux kernel - Deliverables
This page took 0.0578 seconds and 4 git commands to generate.