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

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

Version: ~ [ linux-5.15-rc5 ] ~ [ linux-5.14.11 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.72 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.152 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.210 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.250 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.286 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.288 ] ~ [ 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_NON_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 static inline bool decrement_ttl(struct netns_ipvs *ipvs,
258                                  int skb_af,
259                                  struct sk_buff *skb)
260 {
261         struct net *net = ipvs->net;
262 
263 #ifdef CONFIG_IP_VS_IPV6
264         if (skb_af == AF_INET6) {
265                 struct dst_entry *dst = skb_dst(skb);
266 
267                 /* check and decrement ttl */
268                 if (ipv6_hdr(skb)->hop_limit <= 1) {
269                         struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
270 
271                         /* Force OUTPUT device used as source address */
272                         skb->dev = dst->dev;
273                         icmpv6_send(skb, ICMPV6_TIME_EXCEED,
274                                     ICMPV6_EXC_HOPLIMIT, 0);
275                         __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
276 
277                         return false;
278                 }
279 
280                 /* don't propagate ttl change to cloned packets */
281                 if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
282                         return false;
283 
284                 ipv6_hdr(skb)->hop_limit--;
285         } else
286 #endif
287         {
288                 if (ip_hdr(skb)->ttl <= 1) {
289                         /* Tell the sender its packet died... */
290                         __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
291                         icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
292                         return false;
293                 }
294 
295                 /* don't propagate ttl change to cloned packets */
296                 if (!skb_make_writable(skb, sizeof(struct iphdr)))
297                         return false;
298 
299                 /* Decrease ttl */
300                 ip_decrease_ttl(ip_hdr(skb));
301         }
302 
303         return true;
304 }
305 
306 /* Get route to destination or remote server */
307 static int
308 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
309                    struct ip_vs_dest *dest,
310                    __be32 daddr, int rt_mode, __be32 *ret_saddr,
311                    struct ip_vs_iphdr *ipvsh)
312 {
313         struct net *net = ipvs->net;
314         struct ip_vs_dest_dst *dest_dst;
315         struct rtable *rt;                      /* Route to the other host */
316         int mtu;
317         int local, noref = 1;
318 
319         if (dest) {
320                 dest_dst = __ip_vs_dst_check(dest);
321                 if (likely(dest_dst))
322                         rt = (struct rtable *) dest_dst->dst_cache;
323                 else {
324                         dest_dst = ip_vs_dest_dst_alloc();
325                         spin_lock_bh(&dest->dst_lock);
326                         if (!dest_dst) {
327                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
328                                 spin_unlock_bh(&dest->dst_lock);
329                                 goto err_unreach;
330                         }
331                         rt = do_output_route4(net, dest->addr.ip, rt_mode,
332                                               &dest_dst->dst_saddr.ip);
333                         if (!rt) {
334                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
335                                 spin_unlock_bh(&dest->dst_lock);
336                                 ip_vs_dest_dst_free(dest_dst);
337                                 goto err_unreach;
338                         }
339                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
340                         spin_unlock_bh(&dest->dst_lock);
341                         IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
342                                   &dest->addr.ip, &dest_dst->dst_saddr.ip,
343                                   atomic_read(&rt->dst.__refcnt));
344                 }
345                 if (ret_saddr)
346                         *ret_saddr = dest_dst->dst_saddr.ip;
347         } else {
348                 __be32 saddr = htonl(INADDR_ANY);
349 
350                 noref = 0;
351 
352                 /* For such unconfigured boxes avoid many route lookups
353                  * for performance reasons because we do not remember saddr
354                  */
355                 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
356                 rt = do_output_route4(net, daddr, rt_mode, &saddr);
357                 if (!rt)
358                         goto err_unreach;
359                 if (ret_saddr)
360                         *ret_saddr = saddr;
361         }
362 
363         local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
364         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
365                                                   local))) {
366                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
367                              " daddr=%pI4\n", &daddr);
368                 goto err_put;
369         }
370 
371         if (unlikely(local)) {
372                 /* skb to local stack, preserve old route */
373                 if (!noref)
374                         ip_rt_put(rt);
375                 return local;
376         }
377 
378         if (!decrement_ttl(ipvs, skb_af, skb))
379                 goto err_put;
380 
381         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
382                 mtu = dst_mtu(&rt->dst);
383         } else {
384                 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
385                 if (mtu < 68) {
386                         IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
387                         goto err_put;
388                 }
389                 maybe_update_pmtu(skb_af, skb, mtu);
390         }
391 
392         if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
393                 goto err_put;
394 
395         skb_dst_drop(skb);
396         if (noref) {
397                 if (!local)
398                         skb_dst_set_noref(skb, &rt->dst);
399                 else
400                         skb_dst_set(skb, dst_clone(&rt->dst));
401         } else
402                 skb_dst_set(skb, &rt->dst);
403 
404         return local;
405 
406 err_put:
407         if (!noref)
408                 ip_rt_put(rt);
409         return -1;
410 
411 err_unreach:
412         dst_link_failure(skb);
413         return -1;
414 }
415 
416 #ifdef CONFIG_IP_VS_IPV6
417 static struct dst_entry *
418 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
419                         struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
420 {
421         struct dst_entry *dst;
422         struct flowi6 fl6 = {
423                 .daddr = *daddr,
424         };
425 
426         if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
427                 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
428 
429         dst = ip6_route_output(net, NULL, &fl6);
430         if (dst->error)
431                 goto out_err;
432         if (!ret_saddr)
433                 return dst;
434         if (ipv6_addr_any(&fl6.saddr) &&
435             ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
436                                &fl6.daddr, 0, &fl6.saddr) < 0)
437                 goto out_err;
438         if (do_xfrm) {
439                 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
440                 if (IS_ERR(dst)) {
441                         dst = NULL;
442                         goto out_err;
443                 }
444         }
445         *ret_saddr = fl6.saddr;
446         return dst;
447 
448 out_err:
449         dst_release(dst);
450         IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
451         return NULL;
452 }
453 
454 /*
455  * Get route to destination or remote server
456  */
457 static int
458 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
459                       struct ip_vs_dest *dest,
460                       struct in6_addr *daddr, struct in6_addr *ret_saddr,
461                       struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
462 {
463         struct net *net = ipvs->net;
464         struct ip_vs_dest_dst *dest_dst;
465         struct rt6_info *rt;                    /* Route to the other host */
466         struct dst_entry *dst;
467         int mtu;
468         int local, noref = 1;
469 
470         if (dest) {
471                 dest_dst = __ip_vs_dst_check(dest);
472                 if (likely(dest_dst))
473                         rt = (struct rt6_info *) dest_dst->dst_cache;
474                 else {
475                         u32 cookie;
476 
477                         dest_dst = ip_vs_dest_dst_alloc();
478                         spin_lock_bh(&dest->dst_lock);
479                         if (!dest_dst) {
480                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
481                                 spin_unlock_bh(&dest->dst_lock);
482                                 goto err_unreach;
483                         }
484                         dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
485                                                       &dest_dst->dst_saddr.in6,
486                                                       do_xfrm, rt_mode);
487                         if (!dst) {
488                                 __ip_vs_dst_set(dest, NULL, NULL, 0);
489                                 spin_unlock_bh(&dest->dst_lock);
490                                 ip_vs_dest_dst_free(dest_dst);
491                                 goto err_unreach;
492                         }
493                         rt = (struct rt6_info *) dst;
494                         cookie = rt6_get_cookie(rt);
495                         __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
496                         spin_unlock_bh(&dest->dst_lock);
497                         IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
498                                   &dest->addr.in6, &dest_dst->dst_saddr.in6,
499                                   atomic_read(&rt->dst.__refcnt));
500                 }
501                 if (ret_saddr)
502                         *ret_saddr = dest_dst->dst_saddr.in6;
503         } else {
504                 noref = 0;
505                 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
506                                               rt_mode);
507                 if (!dst)
508                         goto err_unreach;
509                 rt = (struct rt6_info *) dst;
510         }
511 
512         local = __ip_vs_is_local_route6(rt);
513 
514         if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
515                                                   local))) {
516                 IP_VS_DBG_RL("We are crossing local and non-local addresses"
517                              " daddr=%pI6\n", daddr);
518                 goto err_put;
519         }
520 
521         if (unlikely(local)) {
522                 /* skb to local stack, preserve old route */
523                 if (!noref)
524                         dst_release(&rt->dst);
525                 return local;
526         }
527 
528         if (!decrement_ttl(ipvs, skb_af, skb))
529                 goto err_put;
530 
531         /* MTU checking */
532         if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
533                 mtu = dst_mtu(&rt->dst);
534         else {
535                 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
536                 if (mtu < IPV6_MIN_MTU) {
537                         IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
538                                      IPV6_MIN_MTU);
539                         goto err_put;
540                 }
541                 maybe_update_pmtu(skb_af, skb, mtu);
542         }
543 
544         if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
545                 goto err_put;
546 
547         skb_dst_drop(skb);
548         if (noref) {
549                 if (!local)
550                         skb_dst_set_noref(skb, &rt->dst);
551                 else
552                         skb_dst_set(skb, dst_clone(&rt->dst));
553         } else
554                 skb_dst_set(skb, &rt->dst);
555 
556         return local;
557 
558 err_put:
559         if (!noref)
560                 dst_release(&rt->dst);
561         return -1;
562 
563 err_unreach:
564         /* The ip6_link_failure function requires the dev field to be set
565          * in order to get the net (further for the sake of fwmark
566          * reflection).
567          */
568         if (!skb->dev)
569                 skb->dev = skb_dst(skb)->dev;
570 
571         dst_link_failure(skb);
572         return -1;
573 }
574 #endif
575 
576 
577 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
578 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
579                                             struct ip_vs_conn *cp)
580 {
581         int ret = NF_ACCEPT;
582 
583         skb->ipvs_property = 1;
584         if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
585                 ret = ip_vs_confirm_conntrack(skb);
586         if (ret == NF_ACCEPT) {
587                 nf_reset(skb);
588                 skb_forward_csum(skb);
589         }
590         return ret;
591 }
592 
593 /* In the event of a remote destination, it's possible that we would have
594  * matches against an old socket (particularly a TIME-WAIT socket). This
595  * causes havoc down the line (ip_local_out et. al. expect regular sockets
596  * and invalid memory accesses will happen) so simply drop the association
597  * in this case.
598 */
599 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
600 {
601         /* If dev is set, the packet came from the LOCAL_IN callback and
602          * not from a local TCP socket.
603          */
604         if (skb->dev)
605                 skb_orphan(skb);
606 }
607 
608 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
609 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
610                                          struct ip_vs_conn *cp, int local)
611 {
612         int ret = NF_STOLEN;
613 
614         skb->ipvs_property = 1;
615         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
616                 ip_vs_notrack(skb);
617         else
618                 ip_vs_update_conntrack(skb, cp, 1);
619 
620         /* Remove the early_demux association unless it's bound for the
621          * exact same port and address on this host after translation.
622          */
623         if (!local || cp->vport != cp->dport ||
624             !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
625                 ip_vs_drop_early_demux_sk(skb);
626 
627         if (!local) {
628                 skb_forward_csum(skb);
629                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
630                         NULL, skb_dst(skb)->dev, dst_output);
631         } else
632                 ret = NF_ACCEPT;
633 
634         return ret;
635 }
636 
637 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
638 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
639                                      struct ip_vs_conn *cp, int local)
640 {
641         int ret = NF_STOLEN;
642 
643         skb->ipvs_property = 1;
644         if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
645                 ip_vs_notrack(skb);
646         if (!local) {
647                 ip_vs_drop_early_demux_sk(skb);
648                 skb_forward_csum(skb);
649                 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
650                         NULL, skb_dst(skb)->dev, dst_output);
651         } else
652                 ret = NF_ACCEPT;
653         return ret;
654 }
655 
656 
657 /*
658  *      NULL transmitter (do nothing except return NF_ACCEPT)
659  */
660 int
661 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
662                 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
663 {
664         /* we do not touch skb and do not need pskb ptr */
665         return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
666 }
667 
668 
669 /*
670  *      Bypass transmitter
671  *      Let packets bypass the destination when the destination is not
672  *      available, it may be only used in transparent cache cluster.
673  */
674 int
675 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
676                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
677 {
678         struct iphdr  *iph = ip_hdr(skb);
679 
680         EnterFunction(10);
681 
682         if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
683                                IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
684                 goto tx_error;
685 
686         ip_send_check(iph);
687 
688         /* Another hack: avoid icmp_send in ip_fragment */
689         skb->ignore_df = 1;
690 
691         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
692 
693         LeaveFunction(10);
694         return NF_STOLEN;
695 
696  tx_error:
697         kfree_skb(skb);
698         LeaveFunction(10);
699         return NF_STOLEN;
700 }
701 
702 #ifdef CONFIG_IP_VS_IPV6
703 int
704 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
705                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
706 {
707         struct ipv6hdr *iph = ipv6_hdr(skb);
708 
709         EnterFunction(10);
710 
711         if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
712                                   &iph->daddr, NULL,
713                                   ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
714                 goto tx_error;
715 
716         /* Another hack: avoid icmp_send in ip_fragment */
717         skb->ignore_df = 1;
718 
719         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
720 
721         LeaveFunction(10);
722         return NF_STOLEN;
723 
724  tx_error:
725         kfree_skb(skb);
726         LeaveFunction(10);
727         return NF_STOLEN;
728 }
729 #endif
730 
731 /*
732  *      NAT transmitter (only for outside-to-inside nat forwarding)
733  *      Not used for related ICMP
734  */
735 int
736 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
737                struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
738 {
739         struct rtable *rt;              /* Route to the other host */
740         int local, rc, was_input;
741 
742         EnterFunction(10);
743 
744         /* check if it is a connection of no-client-port */
745         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
746                 __be16 _pt, *p;
747 
748                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
749                 if (p == NULL)
750                         goto tx_error;
751                 ip_vs_conn_fill_cport(cp, *p);
752                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
753         }
754 
755         was_input = rt_is_input_route(skb_rtable(skb));
756         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
757                                    IP_VS_RT_MODE_LOCAL |
758                                    IP_VS_RT_MODE_NON_LOCAL |
759                                    IP_VS_RT_MODE_RDR, NULL, ipvsh);
760         if (local < 0)
761                 goto tx_error;
762         rt = skb_rtable(skb);
763         /*
764          * Avoid duplicate tuple in reply direction for NAT traffic
765          * to local address when connection is sync-ed
766          */
767 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
768         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
769                 enum ip_conntrack_info ctinfo;
770                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
771 
772                 if (ct) {
773                         IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
774                                          "ip_vs_nat_xmit(): "
775                                          "stopping DNAT to local address");
776                         goto tx_error;
777                 }
778         }
779 #endif
780 
781         /* From world but DNAT to loopback address? */
782         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
783                 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
784                                  "ip_vs_nat_xmit(): stopping DNAT to loopback "
785                                  "address");
786                 goto tx_error;
787         }
788 
789         /* copy-on-write the packet before mangling it */
790         if (!skb_make_writable(skb, sizeof(struct iphdr)))
791                 goto tx_error;
792 
793         if (skb_cow(skb, rt->dst.dev->hard_header_len))
794                 goto tx_error;
795 
796         /* mangle the packet */
797         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
798                 goto tx_error;
799         ip_hdr(skb)->daddr = cp->daddr.ip;
800         ip_send_check(ip_hdr(skb));
801 
802         IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
803 
804         /* FIXME: when application helper enlarges the packet and the length
805            is larger than the MTU of outgoing device, there will be still
806            MTU problem. */
807 
808         /* Another hack: avoid icmp_send in ip_fragment */
809         skb->ignore_df = 1;
810 
811         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
812 
813         LeaveFunction(10);
814         return rc;
815 
816   tx_error:
817         kfree_skb(skb);
818         LeaveFunction(10);
819         return NF_STOLEN;
820 }
821 
822 #ifdef CONFIG_IP_VS_IPV6
823 int
824 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
825                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
826 {
827         struct rt6_info *rt;            /* Route to the other host */
828         int local, rc;
829 
830         EnterFunction(10);
831 
832         /* check if it is a connection of no-client-port */
833         if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
834                 __be16 _pt, *p;
835                 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
836                 if (p == NULL)
837                         goto tx_error;
838                 ip_vs_conn_fill_cport(cp, *p);
839                 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
840         }
841 
842         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
843                                       &cp->daddr.in6,
844                                       NULL, ipvsh, 0,
845                                       IP_VS_RT_MODE_LOCAL |
846                                       IP_VS_RT_MODE_NON_LOCAL |
847                                       IP_VS_RT_MODE_RDR);
848         if (local < 0)
849                 goto tx_error;
850         rt = (struct rt6_info *) skb_dst(skb);
851         /*
852          * Avoid duplicate tuple in reply direction for NAT traffic
853          * to local address when connection is sync-ed
854          */
855 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
856         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
857                 enum ip_conntrack_info ctinfo;
858                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
859 
860                 if (ct) {
861                         IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
862                                          "ip_vs_nat_xmit_v6(): "
863                                          "stopping DNAT to local address");
864                         goto tx_error;
865                 }
866         }
867 #endif
868 
869         /* From world but DNAT to loopback address? */
870         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
871             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
872                 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
873                                  "ip_vs_nat_xmit_v6(): "
874                                  "stopping DNAT to loopback address");
875                 goto tx_error;
876         }
877 
878         /* copy-on-write the packet before mangling it */
879         if (!skb_make_writable(skb, sizeof(struct ipv6hdr)))
880                 goto tx_error;
881 
882         if (skb_cow(skb, rt->dst.dev->hard_header_len))
883                 goto tx_error;
884 
885         /* mangle the packet */
886         if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
887                 goto tx_error;
888         ipv6_hdr(skb)->daddr = cp->daddr.in6;
889 
890         IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
891 
892         /* FIXME: when application helper enlarges the packet and the length
893            is larger than the MTU of outgoing device, there will be still
894            MTU problem. */
895 
896         /* Another hack: avoid icmp_send in ip_fragment */
897         skb->ignore_df = 1;
898 
899         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
900 
901         LeaveFunction(10);
902         return rc;
903 
904 tx_error:
905         LeaveFunction(10);
906         kfree_skb(skb);
907         return NF_STOLEN;
908 }
909 #endif
910 
911 /* When forwarding a packet, we must ensure that we've got enough headroom
912  * for the encapsulation packet in the skb.  This also gives us an
913  * opportunity to figure out what the payload_len, dsfield, ttl, and df
914  * values should be, so that we won't need to look at the old ip header
915  * again
916  */
917 static struct sk_buff *
918 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
919                            unsigned int max_headroom, __u8 *next_protocol,
920                            __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
921                            __be16 *df)
922 {
923         struct sk_buff *new_skb = NULL;
924         struct iphdr *old_iph = NULL;
925         __u8 old_dsfield;
926 #ifdef CONFIG_IP_VS_IPV6
927         struct ipv6hdr *old_ipv6h = NULL;
928 #endif
929 
930         ip_vs_drop_early_demux_sk(skb);
931 
932         if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
933                 new_skb = skb_realloc_headroom(skb, max_headroom);
934                 if (!new_skb)
935                         goto error;
936                 if (skb->sk)
937                         skb_set_owner_w(new_skb, skb->sk);
938                 consume_skb(skb);
939                 skb = new_skb;
940         }
941 
942 #ifdef CONFIG_IP_VS_IPV6
943         if (skb_af == AF_INET6) {
944                 old_ipv6h = ipv6_hdr(skb);
945                 *next_protocol = IPPROTO_IPV6;
946                 if (payload_len)
947                         *payload_len =
948                                 ntohs(old_ipv6h->payload_len) +
949                                 sizeof(*old_ipv6h);
950                 old_dsfield = ipv6_get_dsfield(old_ipv6h);
951                 *ttl = old_ipv6h->hop_limit;
952                 if (df)
953                         *df = 0;
954         } else
955 #endif
956         {
957                 old_iph = ip_hdr(skb);
958                 /* Copy DF, reset fragment offset and MF */
959                 if (df)
960                         *df = (old_iph->frag_off & htons(IP_DF));
961                 *next_protocol = IPPROTO_IPIP;
962 
963                 /* fix old IP header checksum */
964                 ip_send_check(old_iph);
965                 old_dsfield = ipv4_get_dsfield(old_iph);
966                 *ttl = old_iph->ttl;
967                 if (payload_len)
968                         *payload_len = ntohs(old_iph->tot_len);
969         }
970 
971         /* Implement full-functionality option for ECN encapsulation */
972         *dsfield = INET_ECN_encapsulate(old_dsfield, old_dsfield);
973 
974         return skb;
975 error:
976         kfree_skb(skb);
977         return ERR_PTR(-ENOMEM);
978 }
979 
980 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
981 {
982         switch (encaps_af) {
983         case AF_INET:
984                 return SKB_GSO_IPXIP4;
985         case AF_INET6:
986                 return SKB_GSO_IPXIP6;
987         default:
988                 return 0;
989         }
990 }
991 
992 /*
993  *   IP Tunneling transmitter
994  *
995  *   This function encapsulates the packet in a new IP packet, its
996  *   destination will be set to cp->daddr. Most code of this function
997  *   is taken from ipip.c.
998  *
999  *   It is used in VS/TUN cluster. The load balancer selects a real
1000  *   server from a cluster based on a scheduling algorithm,
1001  *   encapsulates the request packet and forwards it to the selected
1002  *   server. For example, all real servers are configured with
1003  *   "ifconfig tunl0 <Virtual IP Address> up". When the server receives
1004  *   the encapsulated packet, it will decapsulate the packet, processe
1005  *   the request and return the response packets directly to the client
1006  *   without passing the load balancer. This can greatly increase the
1007  *   scalability of virtual server.
1008  *
1009  *   Used for ANY protocol
1010  */
1011 int
1012 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1013                   struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1014 {
1015         struct netns_ipvs *ipvs = cp->ipvs;
1016         struct net *net = ipvs->net;
1017         struct rtable *rt;                      /* Route to the other host */
1018         __be32 saddr;                           /* Source for tunnel */
1019         struct net_device *tdev;                /* Device to other host */
1020         __u8 next_protocol = 0;
1021         __u8 dsfield = 0;
1022         __u8 ttl = 0;
1023         __be16 df = 0;
1024         __be16 *dfp = NULL;
1025         struct iphdr  *iph;                     /* Our new IP header */
1026         unsigned int max_headroom;              /* The extra header space needed */
1027         int ret, local;
1028 
1029         EnterFunction(10);
1030 
1031         local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1032                                    IP_VS_RT_MODE_LOCAL |
1033                                    IP_VS_RT_MODE_NON_LOCAL |
1034                                    IP_VS_RT_MODE_CONNECT |
1035                                    IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
1036         if (local < 0)
1037                 goto tx_error;
1038         if (local)
1039                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1040 
1041         rt = skb_rtable(skb);
1042         tdev = rt->dst.dev;
1043 
1044         /*
1045          * Okay, now see if we can stuff it in the buffer as-is.
1046          */
1047         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1048 
1049         /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1050         dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1051         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1052                                          &next_protocol, NULL, &dsfield,
1053                                          &ttl, dfp);
1054         if (IS_ERR(skb))
1055                 goto tx_error;
1056 
1057         if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET, cp->af)))
1058                 goto tx_error;
1059 
1060         skb->transport_header = skb->network_header;
1061 
1062         skb_push(skb, sizeof(struct iphdr));
1063         skb_reset_network_header(skb);
1064         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1065 
1066         /*
1067          *      Push down and install the IPIP header.
1068          */
1069         iph                     =       ip_hdr(skb);
1070         iph->version            =       4;
1071         iph->ihl                =       sizeof(struct iphdr)>>2;
1072         iph->frag_off           =       df;
1073         iph->protocol           =       next_protocol;
1074         iph->tos                =       dsfield;
1075         iph->daddr              =       cp->daddr.ip;
1076         iph->saddr              =       saddr;
1077         iph->ttl                =       ttl;
1078         ip_select_ident(net, skb, NULL);
1079 
1080         /* Another hack: avoid icmp_send in ip_fragment */
1081         skb->ignore_df = 1;
1082 
1083         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1084         if (ret == NF_ACCEPT)
1085                 ip_local_out(net, skb->sk, skb);
1086         else if (ret == NF_DROP)
1087                 kfree_skb(skb);
1088 
1089         LeaveFunction(10);
1090 
1091         return NF_STOLEN;
1092 
1093   tx_error:
1094         if (!IS_ERR(skb))
1095                 kfree_skb(skb);
1096         LeaveFunction(10);
1097         return NF_STOLEN;
1098 }
1099 
1100 #ifdef CONFIG_IP_VS_IPV6
1101 int
1102 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1103                      struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1104 {
1105         struct rt6_info *rt;            /* Route to the other host */
1106         struct in6_addr saddr;          /* Source for tunnel */
1107         struct net_device *tdev;        /* Device to other host */
1108         __u8 next_protocol = 0;
1109         __u32 payload_len = 0;
1110         __u8 dsfield = 0;
1111         __u8 ttl = 0;
1112         struct ipv6hdr  *iph;           /* Our new IP header */
1113         unsigned int max_headroom;      /* The extra header space needed */
1114         int ret, local;
1115 
1116         EnterFunction(10);
1117 
1118         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1119                                       &cp->daddr.in6,
1120                                       &saddr, ipvsh, 1,
1121                                       IP_VS_RT_MODE_LOCAL |
1122                                       IP_VS_RT_MODE_NON_LOCAL |
1123                                       IP_VS_RT_MODE_TUNNEL);
1124         if (local < 0)
1125                 goto tx_error;
1126         if (local)
1127                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1128 
1129         rt = (struct rt6_info *) skb_dst(skb);
1130         tdev = rt->dst.dev;
1131 
1132         /*
1133          * Okay, now see if we can stuff it in the buffer as-is.
1134          */
1135         max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1136 
1137         skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1138                                          &next_protocol, &payload_len,
1139                                          &dsfield, &ttl, NULL);
1140         if (IS_ERR(skb))
1141                 goto tx_error;
1142 
1143         if (iptunnel_handle_offloads(skb, __tun_gso_type_mask(AF_INET6, cp->af)))
1144                 goto tx_error;
1145 
1146         skb->transport_header = skb->network_header;
1147 
1148         skb_push(skb, sizeof(struct ipv6hdr));
1149         skb_reset_network_header(skb);
1150         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1151 
1152         /*
1153          *      Push down and install the IPIP header.
1154          */
1155         iph                     =       ipv6_hdr(skb);
1156         iph->version            =       6;
1157         iph->nexthdr            =       next_protocol;
1158         iph->payload_len        =       htons(payload_len);
1159         memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1160         ipv6_change_dsfield(iph, 0, dsfield);
1161         iph->daddr = cp->daddr.in6;
1162         iph->saddr = saddr;
1163         iph->hop_limit          =       ttl;
1164 
1165         /* Another hack: avoid icmp_send in ip_fragment */
1166         skb->ignore_df = 1;
1167 
1168         ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1169         if (ret == NF_ACCEPT)
1170                 ip6_local_out(cp->ipvs->net, skb->sk, skb);
1171         else if (ret == NF_DROP)
1172                 kfree_skb(skb);
1173 
1174         LeaveFunction(10);
1175 
1176         return NF_STOLEN;
1177 
1178 tx_error:
1179         if (!IS_ERR(skb))
1180                 kfree_skb(skb);
1181         LeaveFunction(10);
1182         return NF_STOLEN;
1183 }
1184 #endif
1185 
1186 
1187 /*
1188  *      Direct Routing transmitter
1189  *      Used for ANY protocol
1190  */
1191 int
1192 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1193               struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1194 {
1195         int local;
1196 
1197         EnterFunction(10);
1198 
1199         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1200                                    IP_VS_RT_MODE_LOCAL |
1201                                    IP_VS_RT_MODE_NON_LOCAL |
1202                                    IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1203         if (local < 0)
1204                 goto tx_error;
1205         if (local)
1206                 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1207 
1208         ip_send_check(ip_hdr(skb));
1209 
1210         /* Another hack: avoid icmp_send in ip_fragment */
1211         skb->ignore_df = 1;
1212 
1213         ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1214 
1215         LeaveFunction(10);
1216         return NF_STOLEN;
1217 
1218   tx_error:
1219         kfree_skb(skb);
1220         LeaveFunction(10);
1221         return NF_STOLEN;
1222 }
1223 
1224 #ifdef CONFIG_IP_VS_IPV6
1225 int
1226 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1227                  struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1228 {
1229         int local;
1230 
1231         EnterFunction(10);
1232 
1233         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1234                                       &cp->daddr.in6,
1235                                       NULL, ipvsh, 0,
1236                                       IP_VS_RT_MODE_LOCAL |
1237                                       IP_VS_RT_MODE_NON_LOCAL |
1238                                       IP_VS_RT_MODE_KNOWN_NH);
1239         if (local < 0)
1240                 goto tx_error;
1241         if (local)
1242                 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1243 
1244         /* Another hack: avoid icmp_send in ip_fragment */
1245         skb->ignore_df = 1;
1246 
1247         ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1248 
1249         LeaveFunction(10);
1250         return NF_STOLEN;
1251 
1252 tx_error:
1253         kfree_skb(skb);
1254         LeaveFunction(10);
1255         return NF_STOLEN;
1256 }
1257 #endif
1258 
1259 
1260 /*
1261  *      ICMP packet transmitter
1262  *      called by the ip_vs_in_icmp
1263  */
1264 int
1265 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1266                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1267                 struct ip_vs_iphdr *iph)
1268 {
1269         struct rtable   *rt;    /* Route to the other host */
1270         int rc;
1271         int local;
1272         int rt_mode, was_input;
1273 
1274         EnterFunction(10);
1275 
1276         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1277            forwarded directly here, because there is no need to
1278            translate address/port back */
1279         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1280                 if (cp->packet_xmit)
1281                         rc = cp->packet_xmit(skb, cp, pp, iph);
1282                 else
1283                         rc = NF_ACCEPT;
1284                 /* do not touch skb anymore */
1285                 atomic_inc(&cp->in_pkts);
1286                 goto out;
1287         }
1288 
1289         /*
1290          * mangle and send the packet here (only for VS/NAT)
1291          */
1292         was_input = rt_is_input_route(skb_rtable(skb));
1293 
1294         /* LOCALNODE from FORWARD hook is not supported */
1295         rt_mode = (hooknum != NF_INET_FORWARD) ?
1296                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1297                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1298         local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1299                                    NULL, iph);
1300         if (local < 0)
1301                 goto tx_error;
1302         rt = skb_rtable(skb);
1303 
1304         /*
1305          * Avoid duplicate tuple in reply direction for NAT traffic
1306          * to local address when connection is sync-ed
1307          */
1308 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1309         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1310                 enum ip_conntrack_info ctinfo;
1311                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1312 
1313                 if (ct) {
1314                         IP_VS_DBG(10, "%s(): "
1315                                   "stopping DNAT to local address %pI4\n",
1316                                   __func__, &cp->daddr.ip);
1317                         goto tx_error;
1318                 }
1319         }
1320 #endif
1321 
1322         /* From world but DNAT to loopback address? */
1323         if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1324                 IP_VS_DBG(1, "%s(): "
1325                           "stopping DNAT to loopback %pI4\n",
1326                           __func__, &cp->daddr.ip);
1327                 goto tx_error;
1328         }
1329 
1330         /* copy-on-write the packet before mangling it */
1331         if (!skb_make_writable(skb, offset))
1332                 goto tx_error;
1333 
1334         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1335                 goto tx_error;
1336 
1337         ip_vs_nat_icmp(skb, pp, cp, 0);
1338 
1339         /* Another hack: avoid icmp_send in ip_fragment */
1340         skb->ignore_df = 1;
1341 
1342         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1343         goto out;
1344 
1345   tx_error:
1346         kfree_skb(skb);
1347         rc = NF_STOLEN;
1348   out:
1349         LeaveFunction(10);
1350         return rc;
1351 }
1352 
1353 #ifdef CONFIG_IP_VS_IPV6
1354 int
1355 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1356                 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1357                 struct ip_vs_iphdr *ipvsh)
1358 {
1359         struct rt6_info *rt;    /* Route to the other host */
1360         int rc;
1361         int local;
1362         int rt_mode;
1363 
1364         EnterFunction(10);
1365 
1366         /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1367            forwarded directly here, because there is no need to
1368            translate address/port back */
1369         if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1370                 if (cp->packet_xmit)
1371                         rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1372                 else
1373                         rc = NF_ACCEPT;
1374                 /* do not touch skb anymore */
1375                 atomic_inc(&cp->in_pkts);
1376                 goto out;
1377         }
1378 
1379         /*
1380          * mangle and send the packet here (only for VS/NAT)
1381          */
1382 
1383         /* LOCALNODE from FORWARD hook is not supported */
1384         rt_mode = (hooknum != NF_INET_FORWARD) ?
1385                   IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1386                   IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1387         local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1388                                       &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1389         if (local < 0)
1390                 goto tx_error;
1391         rt = (struct rt6_info *) skb_dst(skb);
1392         /*
1393          * Avoid duplicate tuple in reply direction for NAT traffic
1394          * to local address when connection is sync-ed
1395          */
1396 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1397         if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1398                 enum ip_conntrack_info ctinfo;
1399                 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1400 
1401                 if (ct) {
1402                         IP_VS_DBG(10, "%s(): "
1403                                   "stopping DNAT to local address %pI6\n",
1404                                   __func__, &cp->daddr.in6);
1405                         goto tx_error;
1406                 }
1407         }
1408 #endif
1409 
1410         /* From world but DNAT to loopback address? */
1411         if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1412             ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1413                 IP_VS_DBG(1, "%s(): "
1414                           "stopping DNAT to loopback %pI6\n",
1415                           __func__, &cp->daddr.in6);
1416                 goto tx_error;
1417         }
1418 
1419         /* copy-on-write the packet before mangling it */
1420         if (!skb_make_writable(skb, offset))
1421                 goto tx_error;
1422 
1423         if (skb_cow(skb, rt->dst.dev->hard_header_len))
1424                 goto tx_error;
1425 
1426         ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1427 
1428         /* Another hack: avoid icmp_send in ip_fragment */
1429         skb->ignore_df = 1;
1430 
1431         rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
1432         goto out;
1433 
1434 tx_error:
1435         kfree_skb(skb);
1436         rc = NF_STOLEN;
1437 out:
1438         LeaveFunction(10);
1439         return rc;
1440 }
1441 #endif
1442 

~ [ 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