Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / net / netfilter / xt_TPROXY.c
1 /*
2  * Transparent proxy support for Linux/iptables
3  *
4  * Copyright (c) 2006-2010 BalaBit IT Ltd.
5  * Author: Balazs Scheidler, Krisztian Kovacs
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  *
11  */
12 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
13 #include <linux/module.h>
14 #include <linux/skbuff.h>
15 #include <linux/ip.h>
16 #include <net/checksum.h>
17 #include <net/udp.h>
18 #include <net/tcp.h>
19 #include <net/inet_sock.h>
20 #include <net/inet_hashtables.h>
21 #include <linux/inetdevice.h>
22 #include <linux/netfilter/x_tables.h>
23 #include <linux/netfilter_ipv4/ip_tables.h>
24
25 #include <net/netfilter/ipv4/nf_defrag_ipv4.h>
26
27 #if IS_ENABLED(CONFIG_IP6_NF_IPTABLES)
28 #define XT_TPROXY_HAVE_IPV6 1
29 #include <net/if_inet6.h>
30 #include <net/addrconf.h>
31 #include <net/inet6_hashtables.h>
32 #include <linux/netfilter_ipv6/ip6_tables.h>
33 #include <net/netfilter/ipv6/nf_defrag_ipv6.h>
34 #endif
35
36 #include <linux/netfilter/xt_TPROXY.h>
37
38 enum nf_tproxy_lookup_t {
39          NFT_LOOKUP_LISTENER,
40          NFT_LOOKUP_ESTABLISHED,
41 };
42
43 static bool tproxy_sk_is_transparent(struct sock *sk)
44 {
45         switch (sk->sk_state) {
46         case TCP_TIME_WAIT:
47                 if (inet_twsk(sk)->tw_transparent)
48                         return true;
49                 break;
50         case TCP_NEW_SYN_RECV:
51                 if (inet_rsk(inet_reqsk(sk))->no_srccheck)
52                         return true;
53                 break;
54         default:
55                 if (inet_sk(sk)->transparent)
56                         return true;
57         }
58
59         sock_gen_put(sk);
60         return false;
61 }
62
63 static inline __be32
64 tproxy_laddr4(struct sk_buff *skb, __be32 user_laddr, __be32 daddr)
65 {
66         struct in_device *indev;
67         __be32 laddr;
68
69         if (user_laddr)
70                 return user_laddr;
71
72         laddr = 0;
73         rcu_read_lock();
74         indev = __in_dev_get_rcu(skb->dev);
75         for_primary_ifa(indev) {
76                 laddr = ifa->ifa_local;
77                 break;
78         } endfor_ifa(indev);
79         rcu_read_unlock();
80
81         return laddr ? laddr : daddr;
82 }
83
84 /*
85  * This is used when the user wants to intercept a connection matching
86  * an explicit iptables rule. In this case the sockets are assumed
87  * matching in preference order:
88  *
89  *   - match: if there's a fully established connection matching the
90  *     _packet_ tuple, it is returned, assuming the redirection
91  *     already took place and we process a packet belonging to an
92  *     established connection
93  *
94  *   - match: if there's a listening socket matching the redirection
95  *     (e.g. on-port & on-ip of the connection), it is returned,
96  *     regardless if it was bound to 0.0.0.0 or an explicit
97  *     address. The reasoning is that if there's an explicit rule, it
98  *     does not really matter if the listener is bound to an interface
99  *     or to 0. The user already stated that he wants redirection
100  *     (since he added the rule).
101  *
102  * Please note that there's an overlap between what a TPROXY target
103  * and a socket match will match. Normally if you have both rules the
104  * "socket" match will be the first one, effectively all packets
105  * belonging to established connections going through that one.
106  */
107 static inline struct sock *
108 nf_tproxy_get_sock_v4(struct net *net, const u8 protocol,
109                       const __be32 saddr, const __be32 daddr,
110                       const __be16 sport, const __be16 dport,
111                       const struct net_device *in,
112                       const enum nf_tproxy_lookup_t lookup_type)
113 {
114         struct sock *sk;
115
116         switch (protocol) {
117         case IPPROTO_TCP:
118                 switch (lookup_type) {
119                 case NFT_LOOKUP_LISTENER:
120                         sk = inet_lookup_listener(net, &tcp_hashinfo,
121                                                     saddr, sport,
122                                                     daddr, dport,
123                                                     in->ifindex);
124
125                         /* NOTE: we return listeners even if bound to
126                          * 0.0.0.0, those are filtered out in
127                          * xt_socket, since xt_TPROXY needs 0 bound
128                          * listeners too
129                          */
130                         break;
131                 case NFT_LOOKUP_ESTABLISHED:
132                         sk = inet_lookup_established(net, &tcp_hashinfo,
133                                                     saddr, sport, daddr, dport,
134                                                     in->ifindex);
135                         break;
136                 default:
137                         BUG();
138                 }
139                 break;
140         case IPPROTO_UDP:
141                 sk = udp4_lib_lookup(net, saddr, sport, daddr, dport,
142                                      in->ifindex);
143                 if (sk) {
144                         int connected = (sk->sk_state == TCP_ESTABLISHED);
145                         int wildcard = (inet_sk(sk)->inet_rcv_saddr == 0);
146
147                         /* NOTE: we return listeners even if bound to
148                          * 0.0.0.0, those are filtered out in
149                          * xt_socket, since xt_TPROXY needs 0 bound
150                          * listeners too
151                          */
152                         if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) ||
153                             (lookup_type == NFT_LOOKUP_LISTENER && connected)) {
154                                 sock_put(sk);
155                                 sk = NULL;
156                         }
157                 }
158                 break;
159         default:
160                 WARN_ON(1);
161                 sk = NULL;
162         }
163
164         pr_debug("tproxy socket lookup: proto %u %08x:%u -> %08x:%u, lookup type: %d, sock %p\n",
165                  protocol, ntohl(saddr), ntohs(sport), ntohl(daddr), ntohs(dport), lookup_type, sk);
166
167         return sk;
168 }
169
170 #ifdef XT_TPROXY_HAVE_IPV6
171 static inline struct sock *
172 nf_tproxy_get_sock_v6(struct net *net, const u8 protocol,
173                       const struct in6_addr *saddr, const struct in6_addr *daddr,
174                       const __be16 sport, const __be16 dport,
175                       const struct net_device *in,
176                       const enum nf_tproxy_lookup_t lookup_type)
177 {
178         struct sock *sk;
179
180         switch (protocol) {
181         case IPPROTO_TCP:
182                 switch (lookup_type) {
183                 case NFT_LOOKUP_LISTENER:
184                         sk = inet6_lookup_listener(net, &tcp_hashinfo,
185                                                    saddr, sport,
186                                                    daddr, ntohs(dport),
187                                                    in->ifindex);
188
189                         /* NOTE: we return listeners even if bound to
190                          * 0.0.0.0, those are filtered out in
191                          * xt_socket, since xt_TPROXY needs 0 bound
192                          * listeners too
193                          */
194                         break;
195                 case NFT_LOOKUP_ESTABLISHED:
196                         sk = __inet6_lookup_established(net, &tcp_hashinfo,
197                                                         saddr, sport, daddr, ntohs(dport),
198                                                         in->ifindex);
199                         break;
200                 default:
201                         BUG();
202                 }
203                 break;
204         case IPPROTO_UDP:
205                 sk = udp6_lib_lookup(net, saddr, sport, daddr, dport,
206                                      in->ifindex);
207                 if (sk) {
208                         int connected = (sk->sk_state == TCP_ESTABLISHED);
209                         int wildcard = ipv6_addr_any(&sk->sk_v6_rcv_saddr);
210
211                         /* NOTE: we return listeners even if bound to
212                          * 0.0.0.0, those are filtered out in
213                          * xt_socket, since xt_TPROXY needs 0 bound
214                          * listeners too
215                          */
216                         if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) ||
217                             (lookup_type == NFT_LOOKUP_LISTENER && connected)) {
218                                 sock_put(sk);
219                                 sk = NULL;
220                         }
221                 }
222                 break;
223         default:
224                 WARN_ON(1);
225                 sk = NULL;
226         }
227
228         pr_debug("tproxy socket lookup: proto %u %pI6:%u -> %pI6:%u, lookup type: %d, sock %p\n",
229                  protocol, saddr, ntohs(sport), daddr, ntohs(dport), lookup_type, sk);
230
231         return sk;
232 }
233 #endif
234
235 /**
236  * tproxy_handle_time_wait4 - handle IPv4 TCP TIME_WAIT reopen redirections
237  * @skb:        The skb being processed.
238  * @laddr:      IPv4 address to redirect to or zero.
239  * @lport:      TCP port to redirect to or zero.
240  * @sk:         The TIME_WAIT TCP socket found by the lookup.
241  *
242  * We have to handle SYN packets arriving to TIME_WAIT sockets
243  * differently: instead of reopening the connection we should rather
244  * redirect the new connection to the proxy if there's a listener
245  * socket present.
246  *
247  * tproxy_handle_time_wait4() consumes the socket reference passed in.
248  *
249  * Returns the listener socket if there's one, the TIME_WAIT socket if
250  * no such listener is found, or NULL if the TCP header is incomplete.
251  */
252 static struct sock *
253 tproxy_handle_time_wait4(struct sk_buff *skb, __be32 laddr, __be16 lport,
254                         struct sock *sk)
255 {
256         const struct iphdr *iph = ip_hdr(skb);
257         struct tcphdr _hdr, *hp;
258
259         hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
260         if (hp == NULL) {
261                 inet_twsk_put(inet_twsk(sk));
262                 return NULL;
263         }
264
265         if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
266                 /* SYN to a TIME_WAIT socket, we'd rather redirect it
267                  * to a listener socket if there's one */
268                 struct sock *sk2;
269
270                 sk2 = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
271                                             iph->saddr, laddr ? laddr : iph->daddr,
272                                             hp->source, lport ? lport : hp->dest,
273                                             skb->dev, NFT_LOOKUP_LISTENER);
274                 if (sk2) {
275                         inet_twsk_deschedule(inet_twsk(sk));
276                         inet_twsk_put(inet_twsk(sk));
277                         sk = sk2;
278                 }
279         }
280
281         return sk;
282 }
283
284 /* assign a socket to the skb -- consumes sk */
285 static void
286 nf_tproxy_assign_sock(struct sk_buff *skb, struct sock *sk)
287 {
288         skb_orphan(skb);
289         skb->sk = sk;
290         skb->destructor = sock_edemux;
291 }
292
293 static unsigned int
294 tproxy_tg4(struct sk_buff *skb, __be32 laddr, __be16 lport,
295            u_int32_t mark_mask, u_int32_t mark_value)
296 {
297         const struct iphdr *iph = ip_hdr(skb);
298         struct udphdr _hdr, *hp;
299         struct sock *sk;
300
301         hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
302         if (hp == NULL)
303                 return NF_DROP;
304
305         /* check if there's an ongoing connection on the packet
306          * addresses, this happens if the redirect already happened
307          * and the current packet belongs to an already established
308          * connection */
309         sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
310                                    iph->saddr, iph->daddr,
311                                    hp->source, hp->dest,
312                                    skb->dev, NFT_LOOKUP_ESTABLISHED);
313
314         laddr = tproxy_laddr4(skb, laddr, iph->daddr);
315         if (!lport)
316                 lport = hp->dest;
317
318         /* UDP has no TCP_TIME_WAIT state, so we never enter here */
319         if (sk && sk->sk_state == TCP_TIME_WAIT)
320                 /* reopening a TIME_WAIT connection needs special handling */
321                 sk = tproxy_handle_time_wait4(skb, laddr, lport, sk);
322         else if (!sk)
323                 /* no, there's no established connection, check if
324                  * there's a listener on the redirected addr/port */
325                 sk = nf_tproxy_get_sock_v4(dev_net(skb->dev), iph->protocol,
326                                            iph->saddr, laddr,
327                                            hp->source, lport,
328                                            skb->dev, NFT_LOOKUP_LISTENER);
329
330         /* NOTE: assign_sock consumes our sk reference */
331         if (sk && tproxy_sk_is_transparent(sk)) {
332                 /* This should be in a separate target, but we don't do multiple
333                    targets on the same rule yet */
334                 skb->mark = (skb->mark & ~mark_mask) ^ mark_value;
335
336                 pr_debug("redirecting: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
337                          iph->protocol, &iph->daddr, ntohs(hp->dest),
338                          &laddr, ntohs(lport), skb->mark);
339
340                 nf_tproxy_assign_sock(skb, sk);
341                 return NF_ACCEPT;
342         }
343
344         pr_debug("no socket, dropping: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
345                  iph->protocol, &iph->saddr, ntohs(hp->source),
346                  &iph->daddr, ntohs(hp->dest), skb->mark);
347         return NF_DROP;
348 }
349
350 static unsigned int
351 tproxy_tg4_v0(struct sk_buff *skb, const struct xt_action_param *par)
352 {
353         const struct xt_tproxy_target_info *tgi = par->targinfo;
354
355         return tproxy_tg4(skb, tgi->laddr, tgi->lport, tgi->mark_mask, tgi->mark_value);
356 }
357
358 static unsigned int
359 tproxy_tg4_v1(struct sk_buff *skb, const struct xt_action_param *par)
360 {
361         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
362
363         return tproxy_tg4(skb, tgi->laddr.ip, tgi->lport, tgi->mark_mask, tgi->mark_value);
364 }
365
366 #ifdef XT_TPROXY_HAVE_IPV6
367
368 static inline const struct in6_addr *
369 tproxy_laddr6(struct sk_buff *skb, const struct in6_addr *user_laddr,
370               const struct in6_addr *daddr)
371 {
372         struct inet6_dev *indev;
373         struct inet6_ifaddr *ifa;
374         struct in6_addr *laddr;
375
376         if (!ipv6_addr_any(user_laddr))
377                 return user_laddr;
378         laddr = NULL;
379
380         rcu_read_lock();
381         indev = __in6_dev_get(skb->dev);
382         if (indev)
383                 list_for_each_entry(ifa, &indev->addr_list, if_list) {
384                         if (ifa->flags & (IFA_F_TENTATIVE | IFA_F_DEPRECATED))
385                                 continue;
386
387                         laddr = &ifa->addr;
388                         break;
389                 }
390         rcu_read_unlock();
391
392         return laddr ? laddr : daddr;
393 }
394
395 /**
396  * tproxy_handle_time_wait6 - handle IPv6 TCP TIME_WAIT reopen redirections
397  * @skb:        The skb being processed.
398  * @tproto:     Transport protocol.
399  * @thoff:      Transport protocol header offset.
400  * @par:        Iptables target parameters.
401  * @sk:         The TIME_WAIT TCP socket found by the lookup.
402  *
403  * We have to handle SYN packets arriving to TIME_WAIT sockets
404  * differently: instead of reopening the connection we should rather
405  * redirect the new connection to the proxy if there's a listener
406  * socket present.
407  *
408  * tproxy_handle_time_wait6() consumes the socket reference passed in.
409  *
410  * Returns the listener socket if there's one, the TIME_WAIT socket if
411  * no such listener is found, or NULL if the TCP header is incomplete.
412  */
413 static struct sock *
414 tproxy_handle_time_wait6(struct sk_buff *skb, int tproto, int thoff,
415                          const struct xt_action_param *par,
416                          struct sock *sk)
417 {
418         const struct ipv6hdr *iph = ipv6_hdr(skb);
419         struct tcphdr _hdr, *hp;
420         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
421
422         hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
423         if (hp == NULL) {
424                 inet_twsk_put(inet_twsk(sk));
425                 return NULL;
426         }
427
428         if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
429                 /* SYN to a TIME_WAIT socket, we'd rather redirect it
430                  * to a listener socket if there's one */
431                 struct sock *sk2;
432
433                 sk2 = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
434                                             &iph->saddr,
435                                             tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr),
436                                             hp->source,
437                                             tgi->lport ? tgi->lport : hp->dest,
438                                             skb->dev, NFT_LOOKUP_LISTENER);
439                 if (sk2) {
440                         inet_twsk_deschedule(inet_twsk(sk));
441                         inet_twsk_put(inet_twsk(sk));
442                         sk = sk2;
443                 }
444         }
445
446         return sk;
447 }
448
449 static unsigned int
450 tproxy_tg6_v1(struct sk_buff *skb, const struct xt_action_param *par)
451 {
452         const struct ipv6hdr *iph = ipv6_hdr(skb);
453         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
454         struct udphdr _hdr, *hp;
455         struct sock *sk;
456         const struct in6_addr *laddr;
457         __be16 lport;
458         int thoff = 0;
459         int tproto;
460
461         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
462         if (tproto < 0) {
463                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
464                 return NF_DROP;
465         }
466
467         hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
468         if (hp == NULL) {
469                 pr_debug("unable to grab transport header contents in IPv6 packet, dropping\n");
470                 return NF_DROP;
471         }
472
473         /* check if there's an ongoing connection on the packet
474          * addresses, this happens if the redirect already happened
475          * and the current packet belongs to an already established
476          * connection */
477         sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
478                                    &iph->saddr, &iph->daddr,
479                                    hp->source, hp->dest,
480                                    par->in, NFT_LOOKUP_ESTABLISHED);
481
482         laddr = tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr);
483         lport = tgi->lport ? tgi->lport : hp->dest;
484
485         /* UDP has no TCP_TIME_WAIT state, so we never enter here */
486         if (sk && sk->sk_state == TCP_TIME_WAIT)
487                 /* reopening a TIME_WAIT connection needs special handling */
488                 sk = tproxy_handle_time_wait6(skb, tproto, thoff, par, sk);
489         else if (!sk)
490                 /* no there's no established connection, check if
491                  * there's a listener on the redirected addr/port */
492                 sk = nf_tproxy_get_sock_v6(dev_net(skb->dev), tproto,
493                                            &iph->saddr, laddr,
494                                            hp->source, lport,
495                                            par->in, NFT_LOOKUP_LISTENER);
496
497         /* NOTE: assign_sock consumes our sk reference */
498         if (sk && tproxy_sk_is_transparent(sk)) {
499                 /* This should be in a separate target, but we don't do multiple
500                    targets on the same rule yet */
501                 skb->mark = (skb->mark & ~tgi->mark_mask) ^ tgi->mark_value;
502
503                 pr_debug("redirecting: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n",
504                          tproto, &iph->saddr, ntohs(hp->source),
505                          laddr, ntohs(lport), skb->mark);
506
507                 nf_tproxy_assign_sock(skb, sk);
508                 return NF_ACCEPT;
509         }
510
511         pr_debug("no socket, dropping: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n",
512                  tproto, &iph->saddr, ntohs(hp->source),
513                  &iph->daddr, ntohs(hp->dest), skb->mark);
514
515         return NF_DROP;
516 }
517
518 static int tproxy_tg6_check(const struct xt_tgchk_param *par)
519 {
520         const struct ip6t_ip6 *i = par->entryinfo;
521
522         if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) &&
523             !(i->invflags & IP6T_INV_PROTO))
524                 return 0;
525
526         pr_info("Can be used only in combination with "
527                 "either -p tcp or -p udp\n");
528         return -EINVAL;
529 }
530 #endif
531
532 static int tproxy_tg4_check(const struct xt_tgchk_param *par)
533 {
534         const struct ipt_ip *i = par->entryinfo;
535
536         if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP)
537             && !(i->invflags & IPT_INV_PROTO))
538                 return 0;
539
540         pr_info("Can be used only in combination with "
541                 "either -p tcp or -p udp\n");
542         return -EINVAL;
543 }
544
545 static struct xt_target tproxy_tg_reg[] __read_mostly = {
546         {
547                 .name           = "TPROXY",
548                 .family         = NFPROTO_IPV4,
549                 .table          = "mangle",
550                 .target         = tproxy_tg4_v0,
551                 .revision       = 0,
552                 .targetsize     = sizeof(struct xt_tproxy_target_info),
553                 .checkentry     = tproxy_tg4_check,
554                 .hooks          = 1 << NF_INET_PRE_ROUTING,
555                 .me             = THIS_MODULE,
556         },
557         {
558                 .name           = "TPROXY",
559                 .family         = NFPROTO_IPV4,
560                 .table          = "mangle",
561                 .target         = tproxy_tg4_v1,
562                 .revision       = 1,
563                 .targetsize     = sizeof(struct xt_tproxy_target_info_v1),
564                 .checkentry     = tproxy_tg4_check,
565                 .hooks          = 1 << NF_INET_PRE_ROUTING,
566                 .me             = THIS_MODULE,
567         },
568 #ifdef XT_TPROXY_HAVE_IPV6
569         {
570                 .name           = "TPROXY",
571                 .family         = NFPROTO_IPV6,
572                 .table          = "mangle",
573                 .target         = tproxy_tg6_v1,
574                 .revision       = 1,
575                 .targetsize     = sizeof(struct xt_tproxy_target_info_v1),
576                 .checkentry     = tproxy_tg6_check,
577                 .hooks          = 1 << NF_INET_PRE_ROUTING,
578                 .me             = THIS_MODULE,
579         },
580 #endif
581
582 };
583
584 static int __init tproxy_tg_init(void)
585 {
586         nf_defrag_ipv4_enable();
587 #ifdef XT_TPROXY_HAVE_IPV6
588         nf_defrag_ipv6_enable();
589 #endif
590
591         return xt_register_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
592 }
593
594 static void __exit tproxy_tg_exit(void)
595 {
596         xt_unregister_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
597 }
598
599 module_init(tproxy_tg_init);
600 module_exit(tproxy_tg_exit);
601 MODULE_LICENSE("GPL");
602 MODULE_AUTHOR("Balazs Scheidler, Krisztian Kovacs");
603 MODULE_DESCRIPTION("Netfilter transparent proxy (TPROXY) target module.");
604 MODULE_ALIAS("ipt_TPROXY");
605 MODULE_ALIAS("ip6t_TPROXY");