~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/netfilter/ipvs/ip_vs_xmit.c

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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(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 
257 /* Get route to destination or remote server */
258 static int
259 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
260                    struct ip_vs_dest *dest,
261                    __be32 daddr, int rt_mode, __be32 *ret_saddr,
262                    struct ip_vs_iphdr *ipvsh)
263 {
264         struct net *net = ipvs->net;
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(ipvs, 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(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
407                       struct ip_vs_dest *dest,
408                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
409                       struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
410 {
411         struct net *net = ipvs->net;
412         struct ip_vs_dest_dst *dest_dst;
413         struct rt6_info *rt;                    /* Route to the other host */
414         struct dst_entry *dst;
415         int mtu;
416         int local, noref = 1;
417 
418         if (dest) {
419                 dest_dst = __ip_vs_dst_check(dest);
420                 if (likely(dest_dst))
421                         rt = (struct rt6_info *) dest_dst->dst_cache;
422                 else {
423                         u32 cookie;
424 
425                         dest_dst = ip_vs_dest_dst_alloc();
426                         spin_lock_bh(&dest->dst_lock);
427                         if (!dest_dst) {
428                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
429                                 spin_unlock_bh(&dest->dst_lock);
430                                 goto err_unreach;
431                         }
432                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
433                                                       &dest_dst->dst_saddr.in6,
434                                                       do_xfrm, rt_mode);
435                         if (!dst) {
436                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
437                                 spin_unlock_bh(&dest->dst_lock);
438                                 ip_vs_dest_dst_free(dest_dst);
439                                 goto err_unreach;
440                         }
441                         rt = (struct rt6_info *) dst;
442                         cookie = rt6_get_cookie(rt);
443                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
444                         spin_unlock_bh(&dest->dst_lock);
445                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
446                                   &dest->addr.in6, &dest_dst->dst_saddr.in6,
447                                   atomic_read(&rt->dst.__refcnt));
448                 }
449                 if (ret_saddr)
450                         *ret_saddr = dest_dst->dst_saddr.in6;
451         } else {
452                 noref = 0;
453                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
454                                               rt_mode);
455                 if (!dst)
456                         goto err_unreach;
457                 rt = (struct rt6_info *) dst;
458         }
459 
460         local = __ip_vs_is_local_route6(rt);
461 
462         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
463                                                   local))) {
464                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
465                              " daddr=%pI6\n", daddr);
466                 goto err_put;
467         }
468 
469         if (unlikely(local)) {
470                 /* skb to local stack, preserve old route */
471                 if (!noref)
472                         dst_release(&rt->dst);
473                 return local;
474         }
475 
476         /* MTU checking */
477         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
478                 mtu = dst_mtu(&rt->dst);
479         else {
480                 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
481                 if (mtu < IPV6_MIN_MTU) {
482                         IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
483                                      IPV6_MIN_MTU);
484                         goto err_put;
485                 }
486                 maybe_update_pmtu(skb_af, skb, mtu);
487         }
488 
489         if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
490                 goto err_put;
491 
492         skb_dst_drop(skb);
493         if (noref) {
494                 if (!local)
495                         skb_dst_set_noref(skb, &rt->dst);
496                 else
497                         skb_dst_set(skb, dst_clone(&rt->dst));
498         } else
499                 skb_dst_set(skb, &rt->dst);
500 
501         return local;
502 
503 err_put:
504         if (!noref)
505                 dst_release(&rt->dst);
506         return -1;
507 
508 err_unreach:
509         /* The ip6_link_failure function requires the dev field to be set
510          * in order to get the net (further for the sake of fwmark
511          * reflection).
512          */
513         if (!skb->dev)
514                 skb->dev = skb_dst(skb)->dev;
515 
516         dst_link_failure(skb);
517         return -1;
518 }
519 #endif
520 
521 
522 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
523 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
524                                             struct ip_vs_conn *cp)
525 {
526         int ret = NF_ACCEPT;
527 
528         skb->ipvs_property = 1;
529         if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
530                 ret = ip_vs_confirm_conntrack(skb);
531         if (ret == NF_ACCEPT) {
532                 nf_reset(skb);
533                 skb_forward_csum(skb);
534         }
535         return ret;
536 }
537 
538 /* In the event of a remote destination, it's possible that we would have
539  * matches against an old socket (particularly a TIME-WAIT socket). This
540  * causes havoc down the line (ip_local_out et. al. expect regular sockets
541  * and invalid memory accesses will happen) so simply drop the association
542  * in this case.
543 */
544 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
545 {
546         /* If dev is set, the packet came from the LOCAL_IN callback and
547          * not from a local TCP socket.
548          */
549         if (skb->dev)
550                 skb_orphan(skb);
551 }
552 
553 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
554 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
555                                          struct ip_vs_conn *cp, int local)
556 {
557         int ret = NF_STOLEN;
558 
559         skb->ipvs_property = 1;
560         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
561                 ip_vs_notrack(skb);
562         else
563                 ip_vs_update_conntrack(skb, cp, 1);
564 
565         /* Remove the early_demux association unless it's bound for the
566          * exact same port and address on this host after translation.
567          */
568         if (!local || cp->vport != cp->dport ||
569             !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
570                 ip_vs_drop_early_demux_sk(skb);
571 
572         if (!local) {
573                 skb_forward_csum(skb);
574                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
575                         NULL, skb_dst(skb)->dev, dst_output);
576         } else
577                 ret = NF_ACCEPT;
578 
579         return ret;
580 }
581 
582 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
583 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
584                                      struct ip_vs_conn *cp, int local)
585 {
586         int ret = NF_STOLEN;
587 
588         skb->ipvs_property = 1;
589         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
590                 ip_vs_notrack(skb);
591         if (!local) {
592                 ip_vs_drop_early_demux_sk(skb);
593                 skb_forward_csum(skb);
594                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
595                         NULL, skb_dst(skb)->dev, dst_output);
596         } else
597                 ret = NF_ACCEPT;
598         return ret;
599 }
600 
601 
602 /*
603  *      NULL transmitter (do nothing except return NF_ACCEPT)
604  */
605 int
606 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
607                 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
608 {
609         /* we do not touch skb and do not need pskb ptr */
610         return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
611 }
612 
613 
614 /*
615  *      Bypass transmitter
616  *      Let packets bypass the destination when the destination is not
617  *      available, it may be only used in transparent cache cluster.
618  */
619 int
620 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
621                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
622 {
623         struct iphdr  *iph = ip_hdr(skb);
624 
625         EnterFunction(10);
626 
627         rcu_read_lock();
628         if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
629                                IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
630                 goto tx_error;
631 
632         ip_send_check(iph);
633 
634         /* Another hack: avoid icmp_send in ip_fragment */
635         skb->ignore_df = 1;
636 
637         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
638         rcu_read_unlock();
639 
640         LeaveFunction(10);
641         return NF_STOLEN;
642 
643  tx_error:
644         kfree_skb(skb);
645         rcu_read_unlock();
646         LeaveFunction(10);
647         return NF_STOLEN;
648 }
649 
650 #ifdef CONFIG_IP_VS_IPV6
651 int
652 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
653                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
654 {
655         struct ipv6hdr *iph = ipv6_hdr(skb);
656 
657         EnterFunction(10);
658 
659         rcu_read_lock();
660         if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
661                                   &iph->daddr, NULL,
662                                   ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
663                 goto tx_error;
664 
665         /* Another hack: avoid icmp_send in ip_fragment */
666         skb->ignore_df = 1;
667 
668         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
669         rcu_read_unlock();
670 
671         LeaveFunction(10);
672         return NF_STOLEN;
673 
674  tx_error:
675         kfree_skb(skb);
676         rcu_read_unlock();
677         LeaveFunction(10);
678         return NF_STOLEN;
679 }
680 #endif
681 
682 /*
683  *      NAT transmitter (only for outside-to-inside nat forwarding)
684  *      Not used for related ICMP
685  */
686 int
687 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
688                struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
689 {
690         struct rtable *rt;              /* Route to the other host */
691         int local, rc, was_input;
692 
693         EnterFunction(10);
694 
695         rcu_read_lock();
696         /* check if it is a connection of no-client-port */
697         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
698                 __be16 _pt, *p;
699 
700                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
701                 if (p == NULL)
702                         goto tx_error;
703                 ip_vs_conn_fill_cport(cp, *p);
704                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
705         }
706 
707         was_input = rt_is_input_route(skb_rtable(skb));
708         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
709                                    IP_VS_RT_MODE_LOCAL |
710                                    IP_VS_RT_MODE_NON_LOCAL |
711                                    IP_VS_RT_MODE_RDR, NULL, ipvsh);
712         if (local < 0)
713                 goto tx_error;
714         rt = skb_rtable(skb);
715         /*
716          * Avoid duplicate tuple in reply direction for NAT traffic
717          * to local address when connection is sync-ed
718          */
719 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
720         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
721                 enum ip_conntrack_info ctinfo;
722                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
723 
724                 if (ct && !nf_ct_is_untracked(ct)) {
725                         IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
726                                          "ip_vs_nat_xmit(): "
727                                          "stopping DNAT to local address");
728                         goto tx_error;
729                 }
730         }
731 #endif
732 
733         /* From world but DNAT to loopback address? */
734         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
735                 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
736                                  "ip_vs_nat_xmit(): stopping DNAT to loopback "
737                                  "address");
738                 goto tx_error;
739         }
740 
741         /* copy-on-write the packet before mangling it */
742         if (!skb_make_writable(skb, sizeof(struct iphdr)))
743                 goto tx_error;
744 
745         if (skb_cow(skb, rt->dst.dev->hard_header_len))
746                 goto tx_error;
747 
748         /* mangle the packet */
749         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
750                 goto tx_error;
751         ip_hdr(skb)->daddr = cp->daddr.ip;
752         ip_send_check(ip_hdr(skb));
753 
754         IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
755 
756         /* FIXME: when application helper enlarges the packet and the length
757            is larger than the MTU of outgoing device, there will be still
758            MTU problem. */
759 
760         /* Another hack: avoid icmp_send in ip_fragment */
761         skb->ignore_df = 1;
762 
763         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
764         rcu_read_unlock();
765 
766         LeaveFunction(10);
767         return rc;
768 
769   tx_error:
770         kfree_skb(skb);
771         rcu_read_unlock();
772         LeaveFunction(10);
773         return NF_STOLEN;
774 }
775 
776 #ifdef CONFIG_IP_VS_IPV6
777 int
778 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
779                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
780 {
781         struct rt6_info *rt;            /* Route to the other host */
782         int local, rc;
783 
784         EnterFunction(10);
785 
786         rcu_read_lock();
787         /* check if it is a connection of no-client-port */
788         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
789                 __be16 _pt, *p;
790                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
791                 if (p == NULL)
792                         goto tx_error;
793                 ip_vs_conn_fill_cport(cp, *p);
794                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
795         }
796 
797         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
798                                       &cp->daddr.in6,
799                                       NULL, ipvsh, 0,
800                                       IP_VS_RT_MODE_LOCAL |
801                                       IP_VS_RT_MODE_NON_LOCAL |
802                                       IP_VS_RT_MODE_RDR);
803         if (local < 0)
804                 goto tx_error;
805         rt = (struct rt6_info *) skb_dst(skb);
806         /*
807          * Avoid duplicate tuple in reply direction for NAT traffic
808          * to local address when connection is sync-ed
809          */
810 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
811         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
812                 enum ip_conntrack_info ctinfo;
813                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
814 
815                 if (ct && !nf_ct_is_untracked(ct)) {
816                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
817                                          "ip_vs_nat_xmit_v6(): "
818                                          "stopping DNAT to local address");
819                         goto tx_error;
820                 }
821         }
822 #endif
823 
824         /* From world but DNAT to loopback address? */
825         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
826             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
827                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
828                                  "ip_vs_nat_xmit_v6(): "
829                                  "stopping DNAT to loopback address");
830                 goto tx_error;
831         }
832 
833         /* copy-on-write the packet before mangling it */
834         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
835                 goto tx_error;
836 
837         if (skb_cow(skb, rt->dst.dev->hard_header_len))
838                 goto tx_error;
839 
840         /* mangle the packet */
841         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
842                 goto tx_error;
843         ipv6_hdr(skb)->daddr = cp->daddr.in6;
844 
845         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
846 
847         /* FIXME: when application helper enlarges the packet and the length
848            is larger than the MTU of outgoing device, there will be still
849            MTU problem. */
850 
851         /* Another hack: avoid icmp_send in ip_fragment */
852         skb->ignore_df = 1;
853 
854         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
855         rcu_read_unlock();
856 
857         LeaveFunction(10);
858         return rc;
859 
860 tx_error:
861         LeaveFunction(10);
862         kfree_skb(skb);
863         rcu_read_unlock();
864         return NF_STOLEN;
865 }
866 #endif
867 
868 /* When forwarding a packet, we must ensure that we've got enough headroom
869  * for the encapsulation packet in the skb.  This also gives us an
870  * opportunity to figure out what the payload_len, dsfield, ttl, and df
871  * values should be, so that we won't need to look at the old ip header
872  * again
873  */
874 static struct sk_buff *
875 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
876                            unsigned int max_headroom, __u8 *next_protocol,
877                            __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
878                            __be16 *df)
879 {
880         struct sk_buff *new_skb = NULL;
881         struct iphdr *old_iph = NULL;
882 #ifdef CONFIG_IP_VS_IPV6
883         struct ipv6hdr *old_ipv6h = NULL;
884 #endif
885 
886         ip_vs_drop_early_demux_sk(skb);
887 
888         if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
889                 new_skb = skb_realloc_headroom(skb, max_headroom);
890                 if (!new_skb)
891                         goto error;
892                 if (skb->sk)
893                         skb_set_owner_w(new_skb, skb->sk);
894                 consume_skb(skb);
895                 skb = new_skb;
896         }
897 
898 #ifdef CONFIG_IP_VS_IPV6
899         if (skb_af == AF_INET6) {
900                 old_ipv6h = ipv6_hdr(skb);
901                 *next_protocol = IPPROTO_IPV6;
902                 if (payload_len)
903                         *payload_len =
904                                 ntohs(old_ipv6h->payload_len) +
905                                 sizeof(*old_ipv6h);
906                 *dsfield = ipv6_get_dsfield(old_ipv6h);
907                 *ttl = old_ipv6h->hop_limit;
908                 if (df)
909                         *df = 0;
910         } else
911 #endif
912         {
913                 old_iph = ip_hdr(skb);
914                 /* Copy DF, reset fragment offset and MF */
915                 if (df)
916                         *df = (old_iph->frag_off & htons(IP_DF));
917                 *next_protocol = IPPROTO_IPIP;
918 
919                 /* fix old IP header checksum */
920                 ip_send_check(old_iph);
921                 *dsfield = ipv4_get_dsfield(old_iph);
922                 *ttl = old_iph->ttl;
923                 if (payload_len)
924                         *payload_len = ntohs(old_iph->tot_len);
925         }
926 
927         return skb;
928 error:
929         kfree_skb(skb);
930         return ERR_PTR(-ENOMEM);
931 }
932 
933 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
934 {
935         switch (encaps_af) {
936         case AF_INET:
937                 return SKB_GSO_IPXIP4;
938         case AF_INET6:
939                 return SKB_GSO_IPXIP6;
940         default:
941                 return 0;
942         }
943 }
944 
945 /*
946  *   IP Tunneling transmitter
947  *
948  *   This function encapsulates the packet in a new IP packet, its
949  *   destination will be set to cp->daddr. Most code of this function
950  *   is taken from ipip.c.
951  *
952  *   It is used in VS/TUN cluster. The load balancer selects a real
953  *   server from a cluster based on a scheduling algorithm,
954  *   encapsulates the request packet and forwards it to the selected
955  *   server. For example, all real servers are configured with
956  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
957  *   the encapsulated packet, it will decapsulate the packet, processe
958  *   the request and return the response packets directly to the client
959  *   without passing the load balancer. This can greatly increase the
960  *   scalability of virtual server.
961  *
962  *   Used for ANY protocol
963  */
964 int
965 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
966                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
967 {
968         struct netns_ipvs *ipvs = cp->ipvs;
969         struct net *net = ipvs->net;
970         struct rtable *rt;                      /* Route to the other host */
971         __be32 saddr;                           /* Source for tunnel */
972         struct net_device *tdev;                /* Device to other host */
973         __u8 next_protocol = 0;
974         __u8 dsfield = 0;
975         __u8 ttl = 0;
976         __be16 df = 0;
977         __be16 *dfp = NULL;
978         struct iphdr  *iph;                     /* Our new IP header */
979         unsigned int max_headroom;              /* The extra header space needed */
980         int ret, local;
981 
982         EnterFunction(10);
983 
984         rcu_read_lock();
985         local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
986                                    IP_VS_RT_MODE_LOCAL |
987                                    IP_VS_RT_MODE_NON_LOCAL |
988                                    IP_VS_RT_MODE_CONNECT |
989                                    IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
990         if (local < 0)
991                 goto tx_error;
992         if (local) {
993                 rcu_read_unlock();
994                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
995         }
996 
997         rt = skb_rtable(skb);
998         tdev = rt->dst.dev;
999 
1000         /*
1001          * Okay, now see if we can stuff it in the buffer as-is.
1002          */
1003         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1004 
1005         /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1006         dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1007         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1008                                          &next_protocol, NULL, &dsfield,
1009                                          &ttl, dfp);
1010         if (IS_ERR(skb))
1011                 goto tx_error;
1012 
1013         if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET, cp->af)))
1014                 goto tx_error;
1015 
1016         skb->transport_header = skb->network_header;
1017 
1018         skb_push(skb, sizeof(struct iphdr));
1019         skb_reset_network_header(skb);
1020         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1021 
1022         /*
1023          *      Push down and install the IPIP header.
1024          */
1025         iph                     =       ip_hdr(skb);
1026         iph->version            =       4;
1027         iph->ihl                =       sizeof(struct iphdr)>>2;
1028         iph->frag_off           =       df;
1029         iph->protocol           =       next_protocol;
1030         iph->tos                =       dsfield;
1031         iph->daddr              =       cp->daddr.ip;
1032         iph->saddr              =       saddr;
1033         iph->ttl                =       ttl;
1034         ip_select_ident(net, skb, NULL);
1035 
1036         /* Another hack: avoid icmp_send in ip_fragment */
1037         skb->ignore_df = 1;
1038 
1039         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1040         if (ret == NF_ACCEPT)
1041                 ip_local_out(net, skb->sk, skb);
1042         else if (ret == NF_DROP)
1043                 kfree_skb(skb);
1044         rcu_read_unlock();
1045 
1046         LeaveFunction(10);
1047 
1048         return NF_STOLEN;
1049 
1050   tx_error:
1051         if (!IS_ERR(skb))
1052                 kfree_skb(skb);
1053         rcu_read_unlock();
1054         LeaveFunction(10);
1055         return NF_STOLEN;
1056 }
1057 
1058 #ifdef CONFIG_IP_VS_IPV6
1059 int
1060 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1061                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1062 {
1063         struct rt6_info *rt;            /* Route to the other host */
1064         struct in6_addr saddr;          /* Source for tunnel */
1065         struct net_device *tdev;        /* Device to other host */
1066         __u8 next_protocol = 0;
1067         __u32 payload_len = 0;
1068         __u8 dsfield = 0;
1069         __u8 ttl = 0;
1070         struct ipv6hdr  *iph;           /* Our new IP header */
1071         unsigned int max_headroom;      /* The extra header space needed */
1072         int ret, local;
1073 
1074         EnterFunction(10);
1075 
1076         rcu_read_lock();
1077         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1078                                       &cp->daddr.in6,
1079                                       &saddr, ipvsh, 1,
1080                                       IP_VS_RT_MODE_LOCAL |
1081                                       IP_VS_RT_MODE_NON_LOCAL |
1082                                       IP_VS_RT_MODE_TUNNEL);
1083         if (local < 0)
1084                 goto tx_error;
1085         if (local) {
1086                 rcu_read_unlock();
1087                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1088         }
1089 
1090         rt = (struct rt6_info *) skb_dst(skb);
1091         tdev = rt->dst.dev;
1092 
1093         /*
1094          * Okay, now see if we can stuff it in the buffer as-is.
1095          */
1096         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1097 
1098         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1099                                          &next_protocol, &payload_len,
1100                                          &dsfield, &ttl, NULL);
1101         if (IS_ERR(skb))
1102                 goto tx_error;
1103 
1104         if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET6, cp->af)))
1105                 goto tx_error;
1106 
1107         skb->transport_header = skb->network_header;
1108 
1109         skb_push(skb, sizeof(struct ipv6hdr));
1110         skb_reset_network_header(skb);
1111         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1112 
1113         /*
1114          *      Push down and install the IPIP header.
1115          */
1116         iph                     =       ipv6_hdr(skb);
1117         iph->version            =       6;
1118         iph->nexthdr            =       next_protocol;
1119         iph->payload_len        =       htons(payload_len);
1120         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1121         ipv6_change_dsfield(iph, 0, dsfield);
1122         iph->daddr = cp->daddr.in6;
1123         iph->saddr = saddr;
1124         iph->hop_limit          =       ttl;
1125 
1126         /* Another hack: avoid icmp_send in ip_fragment */
1127         skb->ignore_df = 1;
1128 
1129         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1130         if (ret == NF_ACCEPT)
1131                 ip6_local_out(cp->ipvs->net, skb->sk, skb);
1132         else if (ret == NF_DROP)
1133                 kfree_skb(skb);
1134         rcu_read_unlock();
1135 
1136         LeaveFunction(10);
1137 
1138         return NF_STOLEN;
1139 
1140 tx_error:
1141         if (!IS_ERR(skb))
1142                 kfree_skb(skb);
1143         rcu_read_unlock();
1144         LeaveFunction(10);
1145         return NF_STOLEN;
1146 }
1147 #endif
1148 
1149 
1150 /*
1151  *      Direct Routing transmitter
1152  *      Used for ANY protocol
1153  */
1154 int
1155 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1156               struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1157 {
1158         int local;
1159 
1160         EnterFunction(10);
1161 
1162         rcu_read_lock();
1163         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1164                                    IP_VS_RT_MODE_LOCAL |
1165                                    IP_VS_RT_MODE_NON_LOCAL |
1166                                    IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1167         if (local < 0)
1168                 goto tx_error;
1169         if (local) {
1170                 rcu_read_unlock();
1171                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1172         }
1173 
1174         ip_send_check(ip_hdr(skb));
1175 
1176         /* Another hack: avoid icmp_send in ip_fragment */
1177         skb->ignore_df = 1;
1178 
1179         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1180         rcu_read_unlock();
1181 
1182         LeaveFunction(10);
1183         return NF_STOLEN;
1184 
1185   tx_error:
1186         kfree_skb(skb);
1187         rcu_read_unlock();
1188         LeaveFunction(10);
1189         return NF_STOLEN;
1190 }
1191 
1192 #ifdef CONFIG_IP_VS_IPV6
1193 int
1194 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1195                  struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1196 {
1197         int local;
1198 
1199         EnterFunction(10);
1200 
1201         rcu_read_lock();
1202         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1203                                       &cp->daddr.in6,
1204                                       NULL, ipvsh, 0,
1205                                       IP_VS_RT_MODE_LOCAL |
1206                                       IP_VS_RT_MODE_NON_LOCAL |
1207                                       IP_VS_RT_MODE_KNOWN_NH);
1208         if (local < 0)
1209                 goto tx_error;
1210         if (local) {
1211                 rcu_read_unlock();
1212                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1213         }
1214 
1215         /* Another hack: avoid icmp_send in ip_fragment */
1216         skb->ignore_df = 1;
1217 
1218         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1219         rcu_read_unlock();
1220 
1221         LeaveFunction(10);
1222         return NF_STOLEN;
1223 
1224 tx_error:
1225         kfree_skb(skb);
1226         rcu_read_unlock();
1227         LeaveFunction(10);
1228         return NF_STOLEN;
1229 }
1230 #endif
1231 
1232 
1233 /*
1234  *      ICMP packet transmitter
1235  *      called by the ip_vs_in_icmp
1236  */
1237 int
1238 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1239                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1240                 struct ip_vs_iphdr *iph)
1241 {
1242         struct rtable   *rt;    /* Route to the other host */
1243         int rc;
1244         int local;
1245         int rt_mode, was_input;
1246 
1247         EnterFunction(10);
1248 
1249         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1250            forwarded directly here, because there is no need to
1251            translate address/port back */
1252         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1253                 if (cp->packet_xmit)
1254                         rc = cp->packet_xmit(skb, cp, pp, iph);
1255                 else
1256                         rc = NF_ACCEPT;
1257                 /* do not touch skb anymore */
1258                 atomic_inc(&cp->in_pkts);
1259                 goto out;
1260         }
1261 
1262         /*
1263          * mangle and send the packet here (only for VS/NAT)
1264          */
1265         was_input = rt_is_input_route(skb_rtable(skb));
1266 
1267         /* LOCALNODE from FORWARD hook is not supported */
1268         rt_mode = (hooknum != NF_INET_FORWARD) ?
1269                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1270                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1271         rcu_read_lock();
1272         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1273                                    NULL, iph);
1274         if (local < 0)
1275                 goto tx_error;
1276         rt = skb_rtable(skb);
1277 
1278         /*
1279          * Avoid duplicate tuple in reply direction for NAT traffic
1280          * to local address when connection is sync-ed
1281          */
1282 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1283         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1284                 enum ip_conntrack_info ctinfo;
1285                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1286 
1287                 if (ct && !nf_ct_is_untracked(ct)) {
1288                         IP_VS_DBG(10, "%s(): "
1289                                   "stopping DNAT to local address %pI4\n",
1290                                   __func__, &cp->daddr.ip);
1291                         goto tx_error;
1292                 }
1293         }
1294 #endif
1295 
1296         /* From world but DNAT to loopback address? */
1297         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1298                 IP_VS_DBG(1, "%s(): "
1299                           "stopping DNAT to loopback %pI4\n",
1300                           __func__, &cp->daddr.ip);
1301                 goto tx_error;
1302         }
1303 
1304         /* copy-on-write the packet before mangling it */
1305         if (!skb_make_writable(skb, offset))
1306                 goto tx_error;
1307 
1308         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1309                 goto tx_error;
1310 
1311         ip_vs_nat_icmp(skb, pp, cp, 0);
1312 
1313         /* Another hack: avoid icmp_send in ip_fragment */
1314         skb->ignore_df = 1;
1315 
1316         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1317         rcu_read_unlock();
1318         goto out;
1319 
1320   tx_error:
1321         kfree_skb(skb);
1322         rcu_read_unlock();
1323         rc = NF_STOLEN;
1324   out:
1325         LeaveFunction(10);
1326         return rc;
1327 }
1328 
1329 #ifdef CONFIG_IP_VS_IPV6
1330 int
1331 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1332                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1333                 struct ip_vs_iphdr *ipvsh)
1334 {
1335         struct rt6_info *rt;    /* Route to the other host */
1336         int rc;
1337         int local;
1338         int rt_mode;
1339 
1340         EnterFunction(10);
1341 
1342         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1343            forwarded directly here, because there is no need to
1344            translate address/port back */
1345         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1346                 if (cp->packet_xmit)
1347                         rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1348                 else
1349                         rc = NF_ACCEPT;
1350                 /* do not touch skb anymore */
1351                 atomic_inc(&cp->in_pkts);
1352                 goto out;
1353         }
1354 
1355         /*
1356          * mangle and send the packet here (only for VS/NAT)
1357          */
1358 
1359         /* LOCALNODE from FORWARD hook is not supported */
1360         rt_mode = (hooknum != NF_INET_FORWARD) ?
1361                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1362                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1363         rcu_read_lock();
1364         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1365                                       &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1366         if (local < 0)
1367                 goto tx_error;
1368         rt = (struct rt6_info *) skb_dst(skb);
1369         /*
1370          * Avoid duplicate tuple in reply direction for NAT traffic
1371          * to local address when connection is sync-ed
1372          */
1373 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1374         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1375                 enum ip_conntrack_info ctinfo;
1376                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1377 
1378                 if (ct && !nf_ct_is_untracked(ct)) {
1379                         IP_VS_DBG(10, "%s(): "
1380                                   "stopping DNAT to local address %pI6\n",
1381                                   __func__, &cp->daddr.in6);
1382                         goto tx_error;
1383                 }
1384         }
1385 #endif
1386 
1387         /* From world but DNAT to loopback address? */
1388         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1389             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1390                 IP_VS_DBG(1, "%s(): "
1391                           "stopping DNAT to loopback %pI6\n",
1392                           __func__, &cp->daddr.in6);
1393                 goto tx_error;
1394         }
1395 
1396         /* copy-on-write the packet before mangling it */
1397         if (!skb_make_writable(skb, offset))
1398                 goto tx_error;
1399 
1400         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1401                 goto tx_error;
1402 
1403         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1404 
1405         /* Another hack: avoid icmp_send in ip_fragment */
1406         skb->ignore_df = 1;
1407 
1408         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1409         rcu_read_unlock();
1410         goto out;
1411 
1412 tx_error:
1413         kfree_skb(skb);
1414         rcu_read_unlock();
1415         rc = NF_STOLEN;
1416 out:
1417         LeaveFunction(10);
1418         return rc;
1419 }
1420 #endif
1421 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp