x86/smpboot: Init apic mapping before usage
[cascardo/linux.git] / 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, struct sk_buff *skb, void *hp,
109                       const u8 protocol,
110                       const __be32 saddr, const __be32 daddr,
111                       const __be16 sport, const __be16 dport,
112                       const struct net_device *in,
113                       const enum nf_tproxy_lookup_t lookup_type)
114 {
115         struct sock *sk;
116         struct tcphdr *tcph;
117
118         switch (protocol) {
119         case IPPROTO_TCP:
120                 switch (lookup_type) {
121                 case NFT_LOOKUP_LISTENER:
122                         tcph = hp;
123                         sk = inet_lookup_listener(net, &tcp_hashinfo, skb,
124                                                     ip_hdrlen(skb) +
125                                                       __tcp_hdrlen(tcph),
126                                                     saddr, sport,
127                                                     daddr, dport,
128                                                     in->ifindex);
129
130                         if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
131                                 sk = NULL;
132                         /* NOTE: we return listeners even if bound to
133                          * 0.0.0.0, those are filtered out in
134                          * xt_socket, since xt_TPROXY needs 0 bound
135                          * listeners too
136                          */
137                         break;
138                 case NFT_LOOKUP_ESTABLISHED:
139                         sk = inet_lookup_established(net, &tcp_hashinfo,
140                                                     saddr, sport, daddr, dport,
141                                                     in->ifindex);
142                         break;
143                 default:
144                         BUG();
145                 }
146                 break;
147         case IPPROTO_UDP:
148                 sk = udp4_lib_lookup(net, saddr, sport, daddr, dport,
149                                      in->ifindex);
150                 if (sk) {
151                         int connected = (sk->sk_state == TCP_ESTABLISHED);
152                         int wildcard = (inet_sk(sk)->inet_rcv_saddr == 0);
153
154                         /* NOTE: we return listeners even if bound to
155                          * 0.0.0.0, those are filtered out in
156                          * xt_socket, since xt_TPROXY needs 0 bound
157                          * listeners too
158                          */
159                         if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) ||
160                             (lookup_type == NFT_LOOKUP_LISTENER && connected)) {
161                                 sock_put(sk);
162                                 sk = NULL;
163                         }
164                 }
165                 break;
166         default:
167                 WARN_ON(1);
168                 sk = NULL;
169         }
170
171         pr_debug("tproxy socket lookup: proto %u %08x:%u -> %08x:%u, lookup type: %d, sock %p\n",
172                  protocol, ntohl(saddr), ntohs(sport), ntohl(daddr), ntohs(dport), lookup_type, sk);
173
174         return sk;
175 }
176
177 #ifdef XT_TPROXY_HAVE_IPV6
178 static inline struct sock *
179 nf_tproxy_get_sock_v6(struct net *net, struct sk_buff *skb, int thoff, void *hp,
180                       const u8 protocol,
181                       const struct in6_addr *saddr, const struct in6_addr *daddr,
182                       const __be16 sport, const __be16 dport,
183                       const struct net_device *in,
184                       const enum nf_tproxy_lookup_t lookup_type)
185 {
186         struct sock *sk;
187         struct tcphdr *tcph;
188
189         switch (protocol) {
190         case IPPROTO_TCP:
191                 switch (lookup_type) {
192                 case NFT_LOOKUP_LISTENER:
193                         tcph = hp;
194                         sk = inet6_lookup_listener(net, &tcp_hashinfo, skb,
195                                                    thoff + __tcp_hdrlen(tcph),
196                                                    saddr, sport,
197                                                    daddr, ntohs(dport),
198                                                    in->ifindex);
199
200                         if (sk && !atomic_inc_not_zero(&sk->sk_refcnt))
201                                 sk = NULL;
202                         /* NOTE: we return listeners even if bound to
203                          * 0.0.0.0, those are filtered out in
204                          * xt_socket, since xt_TPROXY needs 0 bound
205                          * listeners too
206                          */
207                         break;
208                 case NFT_LOOKUP_ESTABLISHED:
209                         sk = __inet6_lookup_established(net, &tcp_hashinfo,
210                                                         saddr, sport, daddr, ntohs(dport),
211                                                         in->ifindex);
212                         break;
213                 default:
214                         BUG();
215                 }
216                 break;
217         case IPPROTO_UDP:
218                 sk = udp6_lib_lookup(net, saddr, sport, daddr, dport,
219                                      in->ifindex);
220                 if (sk) {
221                         int connected = (sk->sk_state == TCP_ESTABLISHED);
222                         int wildcard = ipv6_addr_any(&sk->sk_v6_rcv_saddr);
223
224                         /* NOTE: we return listeners even if bound to
225                          * 0.0.0.0, those are filtered out in
226                          * xt_socket, since xt_TPROXY needs 0 bound
227                          * listeners too
228                          */
229                         if ((lookup_type == NFT_LOOKUP_ESTABLISHED && (!connected || wildcard)) ||
230                             (lookup_type == NFT_LOOKUP_LISTENER && connected)) {
231                                 sock_put(sk);
232                                 sk = NULL;
233                         }
234                 }
235                 break;
236         default:
237                 WARN_ON(1);
238                 sk = NULL;
239         }
240
241         pr_debug("tproxy socket lookup: proto %u %pI6:%u -> %pI6:%u, lookup type: %d, sock %p\n",
242                  protocol, saddr, ntohs(sport), daddr, ntohs(dport), lookup_type, sk);
243
244         return sk;
245 }
246 #endif
247
248 /**
249  * tproxy_handle_time_wait4 - handle IPv4 TCP TIME_WAIT reopen redirections
250  * @skb:        The skb being processed.
251  * @laddr:      IPv4 address to redirect to or zero.
252  * @lport:      TCP port to redirect to or zero.
253  * @sk:         The TIME_WAIT TCP socket found by the lookup.
254  *
255  * We have to handle SYN packets arriving to TIME_WAIT sockets
256  * differently: instead of reopening the connection we should rather
257  * redirect the new connection to the proxy if there's a listener
258  * socket present.
259  *
260  * tproxy_handle_time_wait4() consumes the socket reference passed in.
261  *
262  * Returns the listener socket if there's one, the TIME_WAIT socket if
263  * no such listener is found, or NULL if the TCP header is incomplete.
264  */
265 static struct sock *
266 tproxy_handle_time_wait4(struct net *net, struct sk_buff *skb,
267                          __be32 laddr, __be16 lport, struct sock *sk)
268 {
269         const struct iphdr *iph = ip_hdr(skb);
270         struct tcphdr _hdr, *hp;
271
272         hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
273         if (hp == NULL) {
274                 inet_twsk_put(inet_twsk(sk));
275                 return NULL;
276         }
277
278         if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
279                 /* SYN to a TIME_WAIT socket, we'd rather redirect it
280                  * to a listener socket if there's one */
281                 struct sock *sk2;
282
283                 sk2 = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol,
284                                             iph->saddr, laddr ? laddr : iph->daddr,
285                                             hp->source, lport ? lport : hp->dest,
286                                             skb->dev, NFT_LOOKUP_LISTENER);
287                 if (sk2) {
288                         inet_twsk_deschedule_put(inet_twsk(sk));
289                         sk = sk2;
290                 }
291         }
292
293         return sk;
294 }
295
296 /* assign a socket to the skb -- consumes sk */
297 static void
298 nf_tproxy_assign_sock(struct sk_buff *skb, struct sock *sk)
299 {
300         skb_orphan(skb);
301         skb->sk = sk;
302         skb->destructor = sock_edemux;
303 }
304
305 static unsigned int
306 tproxy_tg4(struct net *net, struct sk_buff *skb, __be32 laddr, __be16 lport,
307            u_int32_t mark_mask, u_int32_t mark_value)
308 {
309         const struct iphdr *iph = ip_hdr(skb);
310         struct udphdr _hdr, *hp;
311         struct sock *sk;
312
313         hp = skb_header_pointer(skb, ip_hdrlen(skb), sizeof(_hdr), &_hdr);
314         if (hp == NULL)
315                 return NF_DROP;
316
317         /* check if there's an ongoing connection on the packet
318          * addresses, this happens if the redirect already happened
319          * and the current packet belongs to an already established
320          * connection */
321         sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol,
322                                    iph->saddr, iph->daddr,
323                                    hp->source, hp->dest,
324                                    skb->dev, NFT_LOOKUP_ESTABLISHED);
325
326         laddr = tproxy_laddr4(skb, laddr, iph->daddr);
327         if (!lport)
328                 lport = hp->dest;
329
330         /* UDP has no TCP_TIME_WAIT state, so we never enter here */
331         if (sk && sk->sk_state == TCP_TIME_WAIT)
332                 /* reopening a TIME_WAIT connection needs special handling */
333                 sk = tproxy_handle_time_wait4(net, skb, laddr, lport, sk);
334         else if (!sk)
335                 /* no, there's no established connection, check if
336                  * there's a listener on the redirected addr/port */
337                 sk = nf_tproxy_get_sock_v4(net, skb, hp, iph->protocol,
338                                            iph->saddr, laddr,
339                                            hp->source, lport,
340                                            skb->dev, NFT_LOOKUP_LISTENER);
341
342         /* NOTE: assign_sock consumes our sk reference */
343         if (sk && tproxy_sk_is_transparent(sk)) {
344                 /* This should be in a separate target, but we don't do multiple
345                    targets on the same rule yet */
346                 skb->mark = (skb->mark & ~mark_mask) ^ mark_value;
347
348                 pr_debug("redirecting: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
349                          iph->protocol, &iph->daddr, ntohs(hp->dest),
350                          &laddr, ntohs(lport), skb->mark);
351
352                 nf_tproxy_assign_sock(skb, sk);
353                 return NF_ACCEPT;
354         }
355
356         pr_debug("no socket, dropping: proto %hhu %pI4:%hu -> %pI4:%hu, mark: %x\n",
357                  iph->protocol, &iph->saddr, ntohs(hp->source),
358                  &iph->daddr, ntohs(hp->dest), skb->mark);
359         return NF_DROP;
360 }
361
362 static unsigned int
363 tproxy_tg4_v0(struct sk_buff *skb, const struct xt_action_param *par)
364 {
365         const struct xt_tproxy_target_info *tgi = par->targinfo;
366
367         return tproxy_tg4(par->net, skb, tgi->laddr, tgi->lport, tgi->mark_mask, tgi->mark_value);
368 }
369
370 static unsigned int
371 tproxy_tg4_v1(struct sk_buff *skb, const struct xt_action_param *par)
372 {
373         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
374
375         return tproxy_tg4(par->net, skb, tgi->laddr.ip, tgi->lport, tgi->mark_mask, tgi->mark_value);
376 }
377
378 #ifdef XT_TPROXY_HAVE_IPV6
379
380 static inline const struct in6_addr *
381 tproxy_laddr6(struct sk_buff *skb, const struct in6_addr *user_laddr,
382               const struct in6_addr *daddr)
383 {
384         struct inet6_dev *indev;
385         struct inet6_ifaddr *ifa;
386         struct in6_addr *laddr;
387
388         if (!ipv6_addr_any(user_laddr))
389                 return user_laddr;
390         laddr = NULL;
391
392         rcu_read_lock();
393         indev = __in6_dev_get(skb->dev);
394         if (indev)
395                 list_for_each_entry(ifa, &indev->addr_list, if_list) {
396                         if (ifa->flags & (IFA_F_TENTATIVE | IFA_F_DEPRECATED))
397                                 continue;
398
399                         laddr = &ifa->addr;
400                         break;
401                 }
402         rcu_read_unlock();
403
404         return laddr ? laddr : daddr;
405 }
406
407 /**
408  * tproxy_handle_time_wait6 - handle IPv6 TCP TIME_WAIT reopen redirections
409  * @skb:        The skb being processed.
410  * @tproto:     Transport protocol.
411  * @thoff:      Transport protocol header offset.
412  * @par:        Iptables target parameters.
413  * @sk:         The TIME_WAIT TCP socket found by the lookup.
414  *
415  * We have to handle SYN packets arriving to TIME_WAIT sockets
416  * differently: instead of reopening the connection we should rather
417  * redirect the new connection to the proxy if there's a listener
418  * socket present.
419  *
420  * tproxy_handle_time_wait6() consumes the socket reference passed in.
421  *
422  * Returns the listener socket if there's one, the TIME_WAIT socket if
423  * no such listener is found, or NULL if the TCP header is incomplete.
424  */
425 static struct sock *
426 tproxy_handle_time_wait6(struct sk_buff *skb, int tproto, int thoff,
427                          const struct xt_action_param *par,
428                          struct sock *sk)
429 {
430         const struct ipv6hdr *iph = ipv6_hdr(skb);
431         struct tcphdr _hdr, *hp;
432         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
433
434         hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
435         if (hp == NULL) {
436                 inet_twsk_put(inet_twsk(sk));
437                 return NULL;
438         }
439
440         if (hp->syn && !hp->rst && !hp->ack && !hp->fin) {
441                 /* SYN to a TIME_WAIT socket, we'd rather redirect it
442                  * to a listener socket if there's one */
443                 struct sock *sk2;
444
445                 sk2 = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp, tproto,
446                                             &iph->saddr,
447                                             tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr),
448                                             hp->source,
449                                             tgi->lport ? tgi->lport : hp->dest,
450                                             skb->dev, NFT_LOOKUP_LISTENER);
451                 if (sk2) {
452                         inet_twsk_deschedule_put(inet_twsk(sk));
453                         sk = sk2;
454                 }
455         }
456
457         return sk;
458 }
459
460 static unsigned int
461 tproxy_tg6_v1(struct sk_buff *skb, const struct xt_action_param *par)
462 {
463         const struct ipv6hdr *iph = ipv6_hdr(skb);
464         const struct xt_tproxy_target_info_v1 *tgi = par->targinfo;
465         struct udphdr _hdr, *hp;
466         struct sock *sk;
467         const struct in6_addr *laddr;
468         __be16 lport;
469         int thoff = 0;
470         int tproto;
471
472         tproto = ipv6_find_hdr(skb, &thoff, -1, NULL, NULL);
473         if (tproto < 0) {
474                 pr_debug("unable to find transport header in IPv6 packet, dropping\n");
475                 return NF_DROP;
476         }
477
478         hp = skb_header_pointer(skb, thoff, sizeof(_hdr), &_hdr);
479         if (hp == NULL) {
480                 pr_debug("unable to grab transport header contents in IPv6 packet, dropping\n");
481                 return NF_DROP;
482         }
483
484         /* check if there's an ongoing connection on the packet
485          * addresses, this happens if the redirect already happened
486          * and the current packet belongs to an already established
487          * connection */
488         sk = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp, tproto,
489                                    &iph->saddr, &iph->daddr,
490                                    hp->source, hp->dest,
491                                    par->in, NFT_LOOKUP_ESTABLISHED);
492
493         laddr = tproxy_laddr6(skb, &tgi->laddr.in6, &iph->daddr);
494         lport = tgi->lport ? tgi->lport : hp->dest;
495
496         /* UDP has no TCP_TIME_WAIT state, so we never enter here */
497         if (sk && sk->sk_state == TCP_TIME_WAIT)
498                 /* reopening a TIME_WAIT connection needs special handling */
499                 sk = tproxy_handle_time_wait6(skb, tproto, thoff, par, sk);
500         else if (!sk)
501                 /* no there's no established connection, check if
502                  * there's a listener on the redirected addr/port */
503                 sk = nf_tproxy_get_sock_v6(par->net, skb, thoff, hp,
504                                            tproto, &iph->saddr, laddr,
505                                            hp->source, lport,
506                                            par->in, NFT_LOOKUP_LISTENER);
507
508         /* NOTE: assign_sock consumes our sk reference */
509         if (sk && tproxy_sk_is_transparent(sk)) {
510                 /* This should be in a separate target, but we don't do multiple
511                    targets on the same rule yet */
512                 skb->mark = (skb->mark & ~tgi->mark_mask) ^ tgi->mark_value;
513
514                 pr_debug("redirecting: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n",
515                          tproto, &iph->saddr, ntohs(hp->source),
516                          laddr, ntohs(lport), skb->mark);
517
518                 nf_tproxy_assign_sock(skb, sk);
519                 return NF_ACCEPT;
520         }
521
522         pr_debug("no socket, dropping: proto %hhu %pI6:%hu -> %pI6:%hu, mark: %x\n",
523                  tproto, &iph->saddr, ntohs(hp->source),
524                  &iph->daddr, ntohs(hp->dest), skb->mark);
525
526         return NF_DROP;
527 }
528
529 static int tproxy_tg6_check(const struct xt_tgchk_param *par)
530 {
531         const struct ip6t_ip6 *i = par->entryinfo;
532
533         if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP) &&
534             !(i->invflags & IP6T_INV_PROTO))
535                 return 0;
536
537         pr_info("Can be used only in combination with "
538                 "either -p tcp or -p udp\n");
539         return -EINVAL;
540 }
541 #endif
542
543 static int tproxy_tg4_check(const struct xt_tgchk_param *par)
544 {
545         const struct ipt_ip *i = par->entryinfo;
546
547         if ((i->proto == IPPROTO_TCP || i->proto == IPPROTO_UDP)
548             && !(i->invflags & IPT_INV_PROTO))
549                 return 0;
550
551         pr_info("Can be used only in combination with "
552                 "either -p tcp or -p udp\n");
553         return -EINVAL;
554 }
555
556 static struct xt_target tproxy_tg_reg[] __read_mostly = {
557         {
558                 .name           = "TPROXY",
559                 .family         = NFPROTO_IPV4,
560                 .table          = "mangle",
561                 .target         = tproxy_tg4_v0,
562                 .revision       = 0,
563                 .targetsize     = sizeof(struct xt_tproxy_target_info),
564                 .checkentry     = tproxy_tg4_check,
565                 .hooks          = 1 << NF_INET_PRE_ROUTING,
566                 .me             = THIS_MODULE,
567         },
568         {
569                 .name           = "TPROXY",
570                 .family         = NFPROTO_IPV4,
571                 .table          = "mangle",
572                 .target         = tproxy_tg4_v1,
573                 .revision       = 1,
574                 .targetsize     = sizeof(struct xt_tproxy_target_info_v1),
575                 .checkentry     = tproxy_tg4_check,
576                 .hooks          = 1 << NF_INET_PRE_ROUTING,
577                 .me             = THIS_MODULE,
578         },
579 #ifdef XT_TPROXY_HAVE_IPV6
580         {
581                 .name           = "TPROXY",
582                 .family         = NFPROTO_IPV6,
583                 .table          = "mangle",
584                 .target         = tproxy_tg6_v1,
585                 .revision       = 1,
586                 .targetsize     = sizeof(struct xt_tproxy_target_info_v1),
587                 .checkentry     = tproxy_tg6_check,
588                 .hooks          = 1 << NF_INET_PRE_ROUTING,
589                 .me             = THIS_MODULE,
590         },
591 #endif
592
593 };
594
595 static int __init tproxy_tg_init(void)
596 {
597         nf_defrag_ipv4_enable();
598 #ifdef XT_TPROXY_HAVE_IPV6
599         nf_defrag_ipv6_enable();
600 #endif
601
602         return xt_register_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
603 }
604
605 static void __exit tproxy_tg_exit(void)
606 {
607         xt_unregister_targets(tproxy_tg_reg, ARRAY_SIZE(tproxy_tg_reg));
608 }
609
610 module_init(tproxy_tg_init);
611 module_exit(tproxy_tg_exit);
612 MODULE_LICENSE("GPL");
613 MODULE_AUTHOR("Balazs Scheidler, Krisztian Kovacs");
614 MODULE_DESCRIPTION("Netfilter transparent proxy (TPROXY) target module.");
615 MODULE_ALIAS("ipt_TPROXY");
616 MODULE_ALIAS("ip6t_TPROXY");