LoongArch: Parse MADT to get multi-processor information
[linux-2.6-microblaze.git] / net / netfilter / nf_synproxy_core.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
4  */
5
6 #include <linux/module.h>
7 #include <linux/skbuff.h>
8 #include <asm/unaligned.h>
9 #include <net/tcp.h>
10 #include <net/netns/generic.h>
11 #include <linux/proc_fs.h>
12
13 #include <linux/netfilter_ipv6.h>
14 #include <linux/netfilter/nf_synproxy.h>
15
16 #include <net/netfilter/nf_conntrack.h>
17 #include <net/netfilter/nf_conntrack_ecache.h>
18 #include <net/netfilter/nf_conntrack_extend.h>
19 #include <net/netfilter/nf_conntrack_seqadj.h>
20 #include <net/netfilter/nf_conntrack_synproxy.h>
21 #include <net/netfilter/nf_conntrack_zones.h>
22 #include <net/netfilter/nf_synproxy.h>
23
24 unsigned int synproxy_net_id;
25 EXPORT_SYMBOL_GPL(synproxy_net_id);
26
27 bool
28 synproxy_parse_options(const struct sk_buff *skb, unsigned int doff,
29                        const struct tcphdr *th, struct synproxy_options *opts)
30 {
31         int length = (th->doff * 4) - sizeof(*th);
32         u8 buf[40], *ptr;
33
34         if (unlikely(length < 0))
35                 return false;
36
37         ptr = skb_header_pointer(skb, doff + sizeof(*th), length, buf);
38         if (ptr == NULL)
39                 return false;
40
41         opts->options = 0;
42         while (length > 0) {
43                 int opcode = *ptr++;
44                 int opsize;
45
46                 switch (opcode) {
47                 case TCPOPT_EOL:
48                         return true;
49                 case TCPOPT_NOP:
50                         length--;
51                         continue;
52                 default:
53                         if (length < 2)
54                                 return true;
55                         opsize = *ptr++;
56                         if (opsize < 2)
57                                 return true;
58                         if (opsize > length)
59                                 return true;
60
61                         switch (opcode) {
62                         case TCPOPT_MSS:
63                                 if (opsize == TCPOLEN_MSS) {
64                                         opts->mss_option = get_unaligned_be16(ptr);
65                                         opts->options |= NF_SYNPROXY_OPT_MSS;
66                                 }
67                                 break;
68                         case TCPOPT_WINDOW:
69                                 if (opsize == TCPOLEN_WINDOW) {
70                                         opts->wscale = *ptr;
71                                         if (opts->wscale > TCP_MAX_WSCALE)
72                                                 opts->wscale = TCP_MAX_WSCALE;
73                                         opts->options |= NF_SYNPROXY_OPT_WSCALE;
74                                 }
75                                 break;
76                         case TCPOPT_TIMESTAMP:
77                                 if (opsize == TCPOLEN_TIMESTAMP) {
78                                         opts->tsval = get_unaligned_be32(ptr);
79                                         opts->tsecr = get_unaligned_be32(ptr + 4);
80                                         opts->options |= NF_SYNPROXY_OPT_TIMESTAMP;
81                                 }
82                                 break;
83                         case TCPOPT_SACK_PERM:
84                                 if (opsize == TCPOLEN_SACK_PERM)
85                                         opts->options |= NF_SYNPROXY_OPT_SACK_PERM;
86                                 break;
87                         }
88
89                         ptr += opsize - 2;
90                         length -= opsize;
91                 }
92         }
93         return true;
94 }
95 EXPORT_SYMBOL_GPL(synproxy_parse_options);
96
97 static unsigned int
98 synproxy_options_size(const struct synproxy_options *opts)
99 {
100         unsigned int size = 0;
101
102         if (opts->options & NF_SYNPROXY_OPT_MSS)
103                 size += TCPOLEN_MSS_ALIGNED;
104         if (opts->options & NF_SYNPROXY_OPT_TIMESTAMP)
105                 size += TCPOLEN_TSTAMP_ALIGNED;
106         else if (opts->options & NF_SYNPROXY_OPT_SACK_PERM)
107                 size += TCPOLEN_SACKPERM_ALIGNED;
108         if (opts->options & NF_SYNPROXY_OPT_WSCALE)
109                 size += TCPOLEN_WSCALE_ALIGNED;
110
111         return size;
112 }
113
114 static void
115 synproxy_build_options(struct tcphdr *th, const struct synproxy_options *opts)
116 {
117         __be32 *ptr = (__be32 *)(th + 1);
118         u8 options = opts->options;
119
120         if (options & NF_SYNPROXY_OPT_MSS)
121                 *ptr++ = htonl((TCPOPT_MSS << 24) |
122                                (TCPOLEN_MSS << 16) |
123                                opts->mss_option);
124
125         if (options & NF_SYNPROXY_OPT_TIMESTAMP) {
126                 if (options & NF_SYNPROXY_OPT_SACK_PERM)
127                         *ptr++ = htonl((TCPOPT_SACK_PERM << 24) |
128                                        (TCPOLEN_SACK_PERM << 16) |
129                                        (TCPOPT_TIMESTAMP << 8) |
130                                        TCPOLEN_TIMESTAMP);
131                 else
132                         *ptr++ = htonl((TCPOPT_NOP << 24) |
133                                        (TCPOPT_NOP << 16) |
134                                        (TCPOPT_TIMESTAMP << 8) |
135                                        TCPOLEN_TIMESTAMP);
136
137                 *ptr++ = htonl(opts->tsval);
138                 *ptr++ = htonl(opts->tsecr);
139         } else if (options & NF_SYNPROXY_OPT_SACK_PERM)
140                 *ptr++ = htonl((TCPOPT_NOP << 24) |
141                                (TCPOPT_NOP << 16) |
142                                (TCPOPT_SACK_PERM << 8) |
143                                TCPOLEN_SACK_PERM);
144
145         if (options & NF_SYNPROXY_OPT_WSCALE)
146                 *ptr++ = htonl((TCPOPT_NOP << 24) |
147                                (TCPOPT_WINDOW << 16) |
148                                (TCPOLEN_WINDOW << 8) |
149                                opts->wscale);
150 }
151
152 void synproxy_init_timestamp_cookie(const struct nf_synproxy_info *info,
153                                     struct synproxy_options *opts)
154 {
155         opts->tsecr = opts->tsval;
156         opts->tsval = tcp_time_stamp_raw() & ~0x3f;
157
158         if (opts->options & NF_SYNPROXY_OPT_WSCALE) {
159                 opts->tsval |= opts->wscale;
160                 opts->wscale = info->wscale;
161         } else
162                 opts->tsval |= 0xf;
163
164         if (opts->options & NF_SYNPROXY_OPT_SACK_PERM)
165                 opts->tsval |= 1 << 4;
166
167         if (opts->options & NF_SYNPROXY_OPT_ECN)
168                 opts->tsval |= 1 << 5;
169 }
170 EXPORT_SYMBOL_GPL(synproxy_init_timestamp_cookie);
171
172 static void
173 synproxy_check_timestamp_cookie(struct synproxy_options *opts)
174 {
175         opts->wscale = opts->tsecr & 0xf;
176         if (opts->wscale != 0xf)
177                 opts->options |= NF_SYNPROXY_OPT_WSCALE;
178
179         opts->options |= opts->tsecr & (1 << 4) ? NF_SYNPROXY_OPT_SACK_PERM : 0;
180
181         opts->options |= opts->tsecr & (1 << 5) ? NF_SYNPROXY_OPT_ECN : 0;
182 }
183
184 static unsigned int
185 synproxy_tstamp_adjust(struct sk_buff *skb, unsigned int protoff,
186                        struct tcphdr *th, struct nf_conn *ct,
187                        enum ip_conntrack_info ctinfo,
188                        const struct nf_conn_synproxy *synproxy)
189 {
190         unsigned int optoff, optend;
191         __be32 *ptr, old;
192
193         if (synproxy->tsoff == 0)
194                 return 1;
195
196         optoff = protoff + sizeof(struct tcphdr);
197         optend = protoff + th->doff * 4;
198
199         if (skb_ensure_writable(skb, optend))
200                 return 0;
201
202         while (optoff < optend) {
203                 unsigned char *op = skb->data + optoff;
204
205                 switch (op[0]) {
206                 case TCPOPT_EOL:
207                         return 1;
208                 case TCPOPT_NOP:
209                         optoff++;
210                         continue;
211                 default:
212                         if (optoff + 1 == optend ||
213                             optoff + op[1] > optend ||
214                             op[1] < 2)
215                                 return 0;
216                         if (op[0] == TCPOPT_TIMESTAMP &&
217                             op[1] == TCPOLEN_TIMESTAMP) {
218                                 if (CTINFO2DIR(ctinfo) == IP_CT_DIR_REPLY) {
219                                         ptr = (__be32 *)&op[2];
220                                         old = *ptr;
221                                         *ptr = htonl(ntohl(*ptr) -
222                                                      synproxy->tsoff);
223                                 } else {
224                                         ptr = (__be32 *)&op[6];
225                                         old = *ptr;
226                                         *ptr = htonl(ntohl(*ptr) +
227                                                      synproxy->tsoff);
228                                 }
229                                 inet_proto_csum_replace4(&th->check, skb,
230                                                          old, *ptr, false);
231                                 return 1;
232                         }
233                         optoff += op[1];
234                 }
235         }
236         return 1;
237 }
238
239 #ifdef CONFIG_PROC_FS
240 static void *synproxy_cpu_seq_start(struct seq_file *seq, loff_t *pos)
241 {
242         struct synproxy_net *snet = synproxy_pernet(seq_file_net(seq));
243         int cpu;
244
245         if (*pos == 0)
246                 return SEQ_START_TOKEN;
247
248         for (cpu = *pos - 1; cpu < nr_cpu_ids; cpu++) {
249                 if (!cpu_possible(cpu))
250                         continue;
251                 *pos = cpu + 1;
252                 return per_cpu_ptr(snet->stats, cpu);
253         }
254
255         return NULL;
256 }
257
258 static void *synproxy_cpu_seq_next(struct seq_file *seq, void *v, loff_t *pos)
259 {
260         struct synproxy_net *snet = synproxy_pernet(seq_file_net(seq));
261         int cpu;
262
263         for (cpu = *pos; cpu < nr_cpu_ids; cpu++) {
264                 if (!cpu_possible(cpu))
265                         continue;
266                 *pos = cpu + 1;
267                 return per_cpu_ptr(snet->stats, cpu);
268         }
269         (*pos)++;
270         return NULL;
271 }
272
273 static void synproxy_cpu_seq_stop(struct seq_file *seq, void *v)
274 {
275         return;
276 }
277
278 static int synproxy_cpu_seq_show(struct seq_file *seq, void *v)
279 {
280         struct synproxy_stats *stats = v;
281
282         if (v == SEQ_START_TOKEN) {
283                 seq_puts(seq, "entries\t\tsyn_received\t"
284                               "cookie_invalid\tcookie_valid\t"
285                               "cookie_retrans\tconn_reopened\n");
286                 return 0;
287         }
288
289         seq_printf(seq, "%08x\t%08x\t%08x\t%08x\t%08x\t%08x\n", 0,
290                    stats->syn_received,
291                    stats->cookie_invalid,
292                    stats->cookie_valid,
293                    stats->cookie_retrans,
294                    stats->conn_reopened);
295
296         return 0;
297 }
298
299 static const struct seq_operations synproxy_cpu_seq_ops = {
300         .start          = synproxy_cpu_seq_start,
301         .next           = synproxy_cpu_seq_next,
302         .stop           = synproxy_cpu_seq_stop,
303         .show           = synproxy_cpu_seq_show,
304 };
305
306 static int __net_init synproxy_proc_init(struct net *net)
307 {
308         if (!proc_create_net("synproxy", 0444, net->proc_net_stat,
309                         &synproxy_cpu_seq_ops, sizeof(struct seq_net_private)))
310                 return -ENOMEM;
311         return 0;
312 }
313
314 static void __net_exit synproxy_proc_exit(struct net *net)
315 {
316         remove_proc_entry("synproxy", net->proc_net_stat);
317 }
318 #else
319 static int __net_init synproxy_proc_init(struct net *net)
320 {
321         return 0;
322 }
323
324 static void __net_exit synproxy_proc_exit(struct net *net)
325 {
326         return;
327 }
328 #endif /* CONFIG_PROC_FS */
329
330 static int __net_init synproxy_net_init(struct net *net)
331 {
332         struct synproxy_net *snet = synproxy_pernet(net);
333         struct nf_conn *ct;
334         int err = -ENOMEM;
335
336         ct = nf_ct_tmpl_alloc(net, &nf_ct_zone_dflt, GFP_KERNEL);
337         if (!ct)
338                 goto err1;
339
340         if (!nfct_seqadj_ext_add(ct))
341                 goto err2;
342         if (!nfct_synproxy_ext_add(ct))
343                 goto err2;
344
345         __set_bit(IPS_CONFIRMED_BIT, &ct->status);
346         snet->tmpl = ct;
347
348         snet->stats = alloc_percpu(struct synproxy_stats);
349         if (snet->stats == NULL)
350                 goto err2;
351
352         err = synproxy_proc_init(net);
353         if (err < 0)
354                 goto err3;
355
356         return 0;
357
358 err3:
359         free_percpu(snet->stats);
360 err2:
361         nf_ct_tmpl_free(ct);
362 err1:
363         return err;
364 }
365
366 static void __net_exit synproxy_net_exit(struct net *net)
367 {
368         struct synproxy_net *snet = synproxy_pernet(net);
369
370         nf_ct_put(snet->tmpl);
371         synproxy_proc_exit(net);
372         free_percpu(snet->stats);
373 }
374
375 static struct pernet_operations synproxy_net_ops = {
376         .init           = synproxy_net_init,
377         .exit           = synproxy_net_exit,
378         .id             = &synproxy_net_id,
379         .size           = sizeof(struct synproxy_net),
380 };
381
382 static int __init synproxy_core_init(void)
383 {
384         return register_pernet_subsys(&synproxy_net_ops);
385 }
386
387 static void __exit synproxy_core_exit(void)
388 {
389         unregister_pernet_subsys(&synproxy_net_ops);
390 }
391
392 module_init(synproxy_core_init);
393 module_exit(synproxy_core_exit);
394
395 static struct iphdr *
396 synproxy_build_ip(struct net *net, struct sk_buff *skb, __be32 saddr,
397                   __be32 daddr)
398 {
399         struct iphdr *iph;
400
401         skb_reset_network_header(skb);
402         iph = skb_put(skb, sizeof(*iph));
403         iph->version    = 4;
404         iph->ihl        = sizeof(*iph) / 4;
405         iph->tos        = 0;
406         iph->id         = 0;
407         iph->frag_off   = htons(IP_DF);
408         iph->ttl        = READ_ONCE(net->ipv4.sysctl_ip_default_ttl);
409         iph->protocol   = IPPROTO_TCP;
410         iph->check      = 0;
411         iph->saddr      = saddr;
412         iph->daddr      = daddr;
413
414         return iph;
415 }
416
417 static void
418 synproxy_send_tcp(struct net *net,
419                   const struct sk_buff *skb, struct sk_buff *nskb,
420                   struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
421                   struct iphdr *niph, struct tcphdr *nth,
422                   unsigned int tcp_hdr_size)
423 {
424         nth->check = ~tcp_v4_check(tcp_hdr_size, niph->saddr, niph->daddr, 0);
425         nskb->ip_summed   = CHECKSUM_PARTIAL;
426         nskb->csum_start  = (unsigned char *)nth - nskb->head;
427         nskb->csum_offset = offsetof(struct tcphdr, check);
428
429         skb_dst_set_noref(nskb, skb_dst(skb));
430         nskb->protocol = htons(ETH_P_IP);
431         if (ip_route_me_harder(net, nskb->sk, nskb, RTN_UNSPEC))
432                 goto free_nskb;
433
434         if (nfct) {
435                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
436                 nf_conntrack_get(nfct);
437         }
438
439         ip_local_out(net, nskb->sk, nskb);
440         return;
441
442 free_nskb:
443         kfree_skb(nskb);
444 }
445
446 void
447 synproxy_send_client_synack(struct net *net,
448                             const struct sk_buff *skb, const struct tcphdr *th,
449                             const struct synproxy_options *opts)
450 {
451         struct sk_buff *nskb;
452         struct iphdr *iph, *niph;
453         struct tcphdr *nth;
454         unsigned int tcp_hdr_size;
455         u16 mss = opts->mss_encode;
456
457         iph = ip_hdr(skb);
458
459         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
460         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
461                          GFP_ATOMIC);
462         if (!nskb)
463                 return;
464         skb_reserve(nskb, MAX_TCP_HEADER);
465
466         niph = synproxy_build_ip(net, nskb, iph->daddr, iph->saddr);
467
468         skb_reset_transport_header(nskb);
469         nth = skb_put(nskb, tcp_hdr_size);
470         nth->source     = th->dest;
471         nth->dest       = th->source;
472         nth->seq        = htonl(__cookie_v4_init_sequence(iph, th, &mss));
473         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
474         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
475         if (opts->options & NF_SYNPROXY_OPT_ECN)
476                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
477         nth->doff       = tcp_hdr_size / 4;
478         nth->window     = 0;
479         nth->check      = 0;
480         nth->urg_ptr    = 0;
481
482         synproxy_build_options(nth, opts);
483
484         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
485                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
486 }
487 EXPORT_SYMBOL_GPL(synproxy_send_client_synack);
488
489 static void
490 synproxy_send_server_syn(struct net *net,
491                          const struct sk_buff *skb, const struct tcphdr *th,
492                          const struct synproxy_options *opts, u32 recv_seq)
493 {
494         struct synproxy_net *snet = synproxy_pernet(net);
495         struct sk_buff *nskb;
496         struct iphdr *iph, *niph;
497         struct tcphdr *nth;
498         unsigned int tcp_hdr_size;
499
500         iph = ip_hdr(skb);
501
502         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
503         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
504                          GFP_ATOMIC);
505         if (!nskb)
506                 return;
507         skb_reserve(nskb, MAX_TCP_HEADER);
508
509         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
510
511         skb_reset_transport_header(nskb);
512         nth = skb_put(nskb, tcp_hdr_size);
513         nth->source     = th->source;
514         nth->dest       = th->dest;
515         nth->seq        = htonl(recv_seq - 1);
516         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
517          * sequence number translation once a connection tracking entry exists.
518          */
519         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
520         tcp_flag_word(nth) = TCP_FLAG_SYN;
521         if (opts->options & NF_SYNPROXY_OPT_ECN)
522                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
523         nth->doff       = tcp_hdr_size / 4;
524         nth->window     = th->window;
525         nth->check      = 0;
526         nth->urg_ptr    = 0;
527
528         synproxy_build_options(nth, opts);
529
530         synproxy_send_tcp(net, skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
531                           niph, nth, tcp_hdr_size);
532 }
533
534 static void
535 synproxy_send_server_ack(struct net *net,
536                          const struct ip_ct_tcp *state,
537                          const struct sk_buff *skb, const struct tcphdr *th,
538                          const struct synproxy_options *opts)
539 {
540         struct sk_buff *nskb;
541         struct iphdr *iph, *niph;
542         struct tcphdr *nth;
543         unsigned int tcp_hdr_size;
544
545         iph = ip_hdr(skb);
546
547         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
548         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
549                          GFP_ATOMIC);
550         if (!nskb)
551                 return;
552         skb_reserve(nskb, MAX_TCP_HEADER);
553
554         niph = synproxy_build_ip(net, nskb, iph->daddr, iph->saddr);
555
556         skb_reset_transport_header(nskb);
557         nth = skb_put(nskb, tcp_hdr_size);
558         nth->source     = th->dest;
559         nth->dest       = th->source;
560         nth->seq        = htonl(ntohl(th->ack_seq));
561         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
562         tcp_flag_word(nth) = TCP_FLAG_ACK;
563         nth->doff       = tcp_hdr_size / 4;
564         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
565         nth->check      = 0;
566         nth->urg_ptr    = 0;
567
568         synproxy_build_options(nth, opts);
569
570         synproxy_send_tcp(net, skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
571 }
572
573 static void
574 synproxy_send_client_ack(struct net *net,
575                          const struct sk_buff *skb, const struct tcphdr *th,
576                          const struct synproxy_options *opts)
577 {
578         struct sk_buff *nskb;
579         struct iphdr *iph, *niph;
580         struct tcphdr *nth;
581         unsigned int tcp_hdr_size;
582
583         iph = ip_hdr(skb);
584
585         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
586         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
587                          GFP_ATOMIC);
588         if (!nskb)
589                 return;
590         skb_reserve(nskb, MAX_TCP_HEADER);
591
592         niph = synproxy_build_ip(net, nskb, iph->saddr, iph->daddr);
593
594         skb_reset_transport_header(nskb);
595         nth = skb_put(nskb, tcp_hdr_size);
596         nth->source     = th->source;
597         nth->dest       = th->dest;
598         nth->seq        = htonl(ntohl(th->seq) + 1);
599         nth->ack_seq    = th->ack_seq;
600         tcp_flag_word(nth) = TCP_FLAG_ACK;
601         nth->doff       = tcp_hdr_size / 4;
602         nth->window     = htons(ntohs(th->window) >> opts->wscale);
603         nth->check      = 0;
604         nth->urg_ptr    = 0;
605
606         synproxy_build_options(nth, opts);
607
608         synproxy_send_tcp(net, skb, nskb, skb_nfct(skb),
609                           IP_CT_ESTABLISHED_REPLY, niph, nth, tcp_hdr_size);
610 }
611
612 bool
613 synproxy_recv_client_ack(struct net *net,
614                          const struct sk_buff *skb, const struct tcphdr *th,
615                          struct synproxy_options *opts, u32 recv_seq)
616 {
617         struct synproxy_net *snet = synproxy_pernet(net);
618         int mss;
619
620         mss = __cookie_v4_check(ip_hdr(skb), th, ntohl(th->ack_seq) - 1);
621         if (mss == 0) {
622                 this_cpu_inc(snet->stats->cookie_invalid);
623                 return false;
624         }
625
626         this_cpu_inc(snet->stats->cookie_valid);
627         opts->mss_option = mss;
628         opts->options |= NF_SYNPROXY_OPT_MSS;
629
630         if (opts->options & NF_SYNPROXY_OPT_TIMESTAMP)
631                 synproxy_check_timestamp_cookie(opts);
632
633         synproxy_send_server_syn(net, skb, th, opts, recv_seq);
634         return true;
635 }
636 EXPORT_SYMBOL_GPL(synproxy_recv_client_ack);
637
638 unsigned int
639 ipv4_synproxy_hook(void *priv, struct sk_buff *skb,
640                    const struct nf_hook_state *nhs)
641 {
642         struct net *net = nhs->net;
643         struct synproxy_net *snet = synproxy_pernet(net);
644         enum ip_conntrack_info ctinfo;
645         struct nf_conn *ct;
646         struct nf_conn_synproxy *synproxy;
647         struct synproxy_options opts = {};
648         const struct ip_ct_tcp *state;
649         struct tcphdr *th, _th;
650         unsigned int thoff;
651
652         ct = nf_ct_get(skb, &ctinfo);
653         if (!ct)
654                 return NF_ACCEPT;
655
656         synproxy = nfct_synproxy(ct);
657         if (!synproxy)
658                 return NF_ACCEPT;
659
660         if (nf_is_loopback_packet(skb) ||
661             ip_hdr(skb)->protocol != IPPROTO_TCP)
662                 return NF_ACCEPT;
663
664         thoff = ip_hdrlen(skb);
665         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
666         if (!th)
667                 return NF_DROP;
668
669         state = &ct->proto.tcp;
670         switch (state->state) {
671         case TCP_CONNTRACK_CLOSE:
672                 if (th->rst && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
673                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
674                                                       ntohl(th->seq) + 1);
675                         break;
676                 }
677
678                 if (!th->syn || th->ack ||
679                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
680                         break;
681
682                 /* Reopened connection - reset the sequence number and timestamp
683                  * adjustments, they will get initialized once the connection is
684                  * reestablished.
685                  */
686                 nf_ct_seqadj_init(ct, ctinfo, 0);
687                 synproxy->tsoff = 0;
688                 this_cpu_inc(snet->stats->conn_reopened);
689                 fallthrough;
690         case TCP_CONNTRACK_SYN_SENT:
691                 if (!synproxy_parse_options(skb, thoff, th, &opts))
692                         return NF_DROP;
693
694                 if (!th->syn && th->ack &&
695                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
696                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
697                          * therefore we need to add 1 to make the SYN sequence
698                          * number match the one of first SYN.
699                          */
700                         if (synproxy_recv_client_ack(net, skb, th, &opts,
701                                                      ntohl(th->seq) + 1)) {
702                                 this_cpu_inc(snet->stats->cookie_retrans);
703                                 consume_skb(skb);
704                                 return NF_STOLEN;
705                         } else {
706                                 return NF_DROP;
707                         }
708                 }
709
710                 synproxy->isn = ntohl(th->ack_seq);
711                 if (opts.options & NF_SYNPROXY_OPT_TIMESTAMP)
712                         synproxy->its = opts.tsecr;
713
714                 nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
715                 break;
716         case TCP_CONNTRACK_SYN_RECV:
717                 if (!th->syn || !th->ack)
718                         break;
719
720                 if (!synproxy_parse_options(skb, thoff, th, &opts))
721                         return NF_DROP;
722
723                 if (opts.options & NF_SYNPROXY_OPT_TIMESTAMP) {
724                         synproxy->tsoff = opts.tsval - synproxy->its;
725                         nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
726                 }
727
728                 opts.options &= ~(NF_SYNPROXY_OPT_MSS |
729                                   NF_SYNPROXY_OPT_WSCALE |
730                                   NF_SYNPROXY_OPT_SACK_PERM);
731
732                 swap(opts.tsval, opts.tsecr);
733                 synproxy_send_server_ack(net, state, skb, th, &opts);
734
735                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
736                 nf_conntrack_event_cache(IPCT_SEQADJ, ct);
737
738                 swap(opts.tsval, opts.tsecr);
739                 synproxy_send_client_ack(net, skb, th, &opts);
740
741                 consume_skb(skb);
742                 return NF_STOLEN;
743         default:
744                 break;
745         }
746
747         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
748         return NF_ACCEPT;
749 }
750 EXPORT_SYMBOL_GPL(ipv4_synproxy_hook);
751
752 static const struct nf_hook_ops ipv4_synproxy_ops[] = {
753         {
754                 .hook           = ipv4_synproxy_hook,
755                 .pf             = NFPROTO_IPV4,
756                 .hooknum        = NF_INET_LOCAL_IN,
757                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
758         },
759         {
760                 .hook           = ipv4_synproxy_hook,
761                 .pf             = NFPROTO_IPV4,
762                 .hooknum        = NF_INET_POST_ROUTING,
763                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
764         },
765 };
766
767 int nf_synproxy_ipv4_init(struct synproxy_net *snet, struct net *net)
768 {
769         int err;
770
771         if (snet->hook_ref4 == 0) {
772                 err = nf_register_net_hooks(net, ipv4_synproxy_ops,
773                                             ARRAY_SIZE(ipv4_synproxy_ops));
774                 if (err)
775                         return err;
776         }
777
778         snet->hook_ref4++;
779         return 0;
780 }
781 EXPORT_SYMBOL_GPL(nf_synproxy_ipv4_init);
782
783 void nf_synproxy_ipv4_fini(struct synproxy_net *snet, struct net *net)
784 {
785         snet->hook_ref4--;
786         if (snet->hook_ref4 == 0)
787                 nf_unregister_net_hooks(net, ipv4_synproxy_ops,
788                                         ARRAY_SIZE(ipv4_synproxy_ops));
789 }
790 EXPORT_SYMBOL_GPL(nf_synproxy_ipv4_fini);
791
792 #if IS_ENABLED(CONFIG_IPV6)
793 static struct ipv6hdr *
794 synproxy_build_ip_ipv6(struct net *net, struct sk_buff *skb,
795                        const struct in6_addr *saddr,
796                        const struct in6_addr *daddr)
797 {
798         struct ipv6hdr *iph;
799
800         skb_reset_network_header(skb);
801         iph = skb_put(skb, sizeof(*iph));
802         ip6_flow_hdr(iph, 0, 0);
803         iph->hop_limit  = net->ipv6.devconf_all->hop_limit;
804         iph->nexthdr    = IPPROTO_TCP;
805         iph->saddr      = *saddr;
806         iph->daddr      = *daddr;
807
808         return iph;
809 }
810
811 static void
812 synproxy_send_tcp_ipv6(struct net *net,
813                        const struct sk_buff *skb, struct sk_buff *nskb,
814                        struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
815                        struct ipv6hdr *niph, struct tcphdr *nth,
816                        unsigned int tcp_hdr_size)
817 {
818         struct dst_entry *dst;
819         struct flowi6 fl6;
820         int err;
821
822         nth->check = ~tcp_v6_check(tcp_hdr_size, &niph->saddr, &niph->daddr, 0);
823         nskb->ip_summed   = CHECKSUM_PARTIAL;
824         nskb->csum_start  = (unsigned char *)nth - nskb->head;
825         nskb->csum_offset = offsetof(struct tcphdr, check);
826
827         memset(&fl6, 0, sizeof(fl6));
828         fl6.flowi6_proto = IPPROTO_TCP;
829         fl6.saddr = niph->saddr;
830         fl6.daddr = niph->daddr;
831         fl6.fl6_sport = nth->source;
832         fl6.fl6_dport = nth->dest;
833         security_skb_classify_flow((struct sk_buff *)skb,
834                                    flowi6_to_flowi_common(&fl6));
835         err = nf_ip6_route(net, &dst, flowi6_to_flowi(&fl6), false);
836         if (err) {
837                 goto free_nskb;
838         }
839
840         dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
841         if (IS_ERR(dst))
842                 goto free_nskb;
843
844         skb_dst_set(nskb, dst);
845
846         if (nfct) {
847                 nf_ct_set(nskb, (struct nf_conn *)nfct, ctinfo);
848                 nf_conntrack_get(nfct);
849         }
850
851         ip6_local_out(net, nskb->sk, nskb);
852         return;
853
854 free_nskb:
855         kfree_skb(nskb);
856 }
857
858 void
859 synproxy_send_client_synack_ipv6(struct net *net,
860                                  const struct sk_buff *skb,
861                                  const struct tcphdr *th,
862                                  const struct synproxy_options *opts)
863 {
864         struct sk_buff *nskb;
865         struct ipv6hdr *iph, *niph;
866         struct tcphdr *nth;
867         unsigned int tcp_hdr_size;
868         u16 mss = opts->mss_encode;
869
870         iph = ipv6_hdr(skb);
871
872         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
873         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
874                          GFP_ATOMIC);
875         if (!nskb)
876                 return;
877         skb_reserve(nskb, MAX_TCP_HEADER);
878
879         niph = synproxy_build_ip_ipv6(net, nskb, &iph->daddr, &iph->saddr);
880
881         skb_reset_transport_header(nskb);
882         nth = skb_put(nskb, tcp_hdr_size);
883         nth->source     = th->dest;
884         nth->dest       = th->source;
885         nth->seq        = htonl(nf_ipv6_cookie_init_sequence(iph, th, &mss));
886         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
887         tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
888         if (opts->options & NF_SYNPROXY_OPT_ECN)
889                 tcp_flag_word(nth) |= TCP_FLAG_ECE;
890         nth->doff       = tcp_hdr_size / 4;
891         nth->window     = 0;
892         nth->check      = 0;
893         nth->urg_ptr    = 0;
894
895         synproxy_build_options(nth, opts);
896
897         synproxy_send_tcp_ipv6(net, skb, nskb, skb_nfct(skb),
898                                IP_CT_ESTABLISHED_REPLY, niph, nth,
899                                tcp_hdr_size);
900 }
901 EXPORT_SYMBOL_GPL(synproxy_send_client_synack_ipv6);
902
903 static void
904 synproxy_send_server_syn_ipv6(struct net *net, const struct sk_buff *skb,
905                               const struct tcphdr *th,
906                               const struct synproxy_options *opts, u32 recv_seq)
907 {
908         struct synproxy_net *snet = synproxy_pernet(net);
909         struct sk_buff *nskb;
910         struct ipv6hdr *iph, *niph;
911         struct tcphdr *nth;
912         unsigned int tcp_hdr_size;
913
914         iph = ipv6_hdr(skb);
915
916         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
917         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
918                          GFP_ATOMIC);
919         if (!nskb)
920                 return;
921         skb_reserve(nskb, MAX_TCP_HEADER);
922
923         niph = synproxy_build_ip_ipv6(net, nskb, &iph->saddr, &iph->daddr);
924
925         skb_reset_transport_header(nskb);
926         nth = skb_put(nskb, tcp_hdr_size);
927         nth->source     = th->source;
928         nth->dest       = th->dest;
929         nth->seq        = htonl(recv_seq - 1);
930         /* ack_seq is used to relay our ISN to the synproxy hook to initialize
931          * sequence number translation once a connection tracking entry exists.
932          */
933         nth->ack_seq    = htonl(ntohl(th->ack_seq) - 1);
934         tcp_flag_word(nth) = TCP_FLAG_SYN;
935         if (opts->options & NF_SYNPROXY_OPT_ECN)
936                 tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
937         nth->doff       = tcp_hdr_size / 4;
938         nth->window     = th->window;
939         nth->check      = 0;
940         nth->urg_ptr    = 0;
941
942         synproxy_build_options(nth, opts);
943
944         synproxy_send_tcp_ipv6(net, skb, nskb, &snet->tmpl->ct_general,
945                                IP_CT_NEW, niph, nth, tcp_hdr_size);
946 }
947
948 static void
949 synproxy_send_server_ack_ipv6(struct net *net, const struct ip_ct_tcp *state,
950                               const struct sk_buff *skb,
951                               const struct tcphdr *th,
952                               const struct synproxy_options *opts)
953 {
954         struct sk_buff *nskb;
955         struct ipv6hdr *iph, *niph;
956         struct tcphdr *nth;
957         unsigned int tcp_hdr_size;
958
959         iph = ipv6_hdr(skb);
960
961         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
962         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
963                          GFP_ATOMIC);
964         if (!nskb)
965                 return;
966         skb_reserve(nskb, MAX_TCP_HEADER);
967
968         niph = synproxy_build_ip_ipv6(net, nskb, &iph->daddr, &iph->saddr);
969
970         skb_reset_transport_header(nskb);
971         nth = skb_put(nskb, tcp_hdr_size);
972         nth->source     = th->dest;
973         nth->dest       = th->source;
974         nth->seq        = htonl(ntohl(th->ack_seq));
975         nth->ack_seq    = htonl(ntohl(th->seq) + 1);
976         tcp_flag_word(nth) = TCP_FLAG_ACK;
977         nth->doff       = tcp_hdr_size / 4;
978         nth->window     = htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
979         nth->check      = 0;
980         nth->urg_ptr    = 0;
981
982         synproxy_build_options(nth, opts);
983
984         synproxy_send_tcp_ipv6(net, skb, nskb, NULL, 0, niph, nth,
985                                tcp_hdr_size);
986 }
987
988 static void
989 synproxy_send_client_ack_ipv6(struct net *net, const struct sk_buff *skb,
990                               const struct tcphdr *th,
991                               const struct synproxy_options *opts)
992 {
993         struct sk_buff *nskb;
994         struct ipv6hdr *iph, *niph;
995         struct tcphdr *nth;
996         unsigned int tcp_hdr_size;
997
998         iph = ipv6_hdr(skb);
999
1000         tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
1001         nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
1002                          GFP_ATOMIC);
1003         if (!nskb)
1004                 return;
1005         skb_reserve(nskb, MAX_TCP_HEADER);
1006
1007         niph = synproxy_build_ip_ipv6(net, nskb, &iph->saddr, &iph->daddr);
1008
1009         skb_reset_transport_header(nskb);
1010         nth = skb_put(nskb, tcp_hdr_size);
1011         nth->source     = th->source;
1012         nth->dest       = th->dest;
1013         nth->seq        = htonl(ntohl(th->seq) + 1);
1014         nth->ack_seq    = th->ack_seq;
1015         tcp_flag_word(nth) = TCP_FLAG_ACK;
1016         nth->doff       = tcp_hdr_size / 4;
1017         nth->window     = htons(ntohs(th->window) >> opts->wscale);
1018         nth->check      = 0;
1019         nth->urg_ptr    = 0;
1020
1021         synproxy_build_options(nth, opts);
1022
1023         synproxy_send_tcp_ipv6(net, skb, nskb, skb_nfct(skb),
1024                                IP_CT_ESTABLISHED_REPLY, niph, nth,
1025                                tcp_hdr_size);
1026 }
1027
1028 bool
1029 synproxy_recv_client_ack_ipv6(struct net *net,
1030                               const struct sk_buff *skb,
1031                               const struct tcphdr *th,
1032                               struct synproxy_options *opts, u32 recv_seq)
1033 {
1034         struct synproxy_net *snet = synproxy_pernet(net);
1035         int mss;
1036
1037         mss = nf_cookie_v6_check(ipv6_hdr(skb), th, ntohl(th->ack_seq) - 1);
1038         if (mss == 0) {
1039                 this_cpu_inc(snet->stats->cookie_invalid);
1040                 return false;
1041         }
1042
1043         this_cpu_inc(snet->stats->cookie_valid);
1044         opts->mss_option = mss;
1045         opts->options |= NF_SYNPROXY_OPT_MSS;
1046
1047         if (opts->options & NF_SYNPROXY_OPT_TIMESTAMP)
1048                 synproxy_check_timestamp_cookie(opts);
1049
1050         synproxy_send_server_syn_ipv6(net, skb, th, opts, recv_seq);
1051         return true;
1052 }
1053 EXPORT_SYMBOL_GPL(synproxy_recv_client_ack_ipv6);
1054
1055 unsigned int
1056 ipv6_synproxy_hook(void *priv, struct sk_buff *skb,
1057                    const struct nf_hook_state *nhs)
1058 {
1059         struct net *net = nhs->net;
1060         struct synproxy_net *snet = synproxy_pernet(net);
1061         enum ip_conntrack_info ctinfo;
1062         struct nf_conn *ct;
1063         struct nf_conn_synproxy *synproxy;
1064         struct synproxy_options opts = {};
1065         const struct ip_ct_tcp *state;
1066         struct tcphdr *th, _th;
1067         __be16 frag_off;
1068         u8 nexthdr;
1069         int thoff;
1070
1071         ct = nf_ct_get(skb, &ctinfo);
1072         if (!ct)
1073                 return NF_ACCEPT;
1074
1075         synproxy = nfct_synproxy(ct);
1076         if (!synproxy)
1077                 return NF_ACCEPT;
1078
1079         if (nf_is_loopback_packet(skb))
1080                 return NF_ACCEPT;
1081
1082         nexthdr = ipv6_hdr(skb)->nexthdr;
1083         thoff = ipv6_skip_exthdr(skb, sizeof(struct ipv6hdr), &nexthdr,
1084                                  &frag_off);
1085         if (thoff < 0 || nexthdr != IPPROTO_TCP)
1086                 return NF_ACCEPT;
1087
1088         th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
1089         if (!th)
1090                 return NF_DROP;
1091
1092         state = &ct->proto.tcp;
1093         switch (state->state) {
1094         case TCP_CONNTRACK_CLOSE:
1095                 if (th->rst && CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL) {
1096                         nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
1097                                                       ntohl(th->seq) + 1);
1098                         break;
1099                 }
1100
1101                 if (!th->syn || th->ack ||
1102                     CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
1103                         break;
1104
1105                 /* Reopened connection - reset the sequence number and timestamp
1106                  * adjustments, they will get initialized once the connection is
1107                  * reestablished.
1108                  */
1109                 nf_ct_seqadj_init(ct, ctinfo, 0);
1110                 synproxy->tsoff = 0;
1111                 this_cpu_inc(snet->stats->conn_reopened);
1112                 fallthrough;
1113         case TCP_CONNTRACK_SYN_SENT:
1114                 if (!synproxy_parse_options(skb, thoff, th, &opts))
1115                         return NF_DROP;
1116
1117                 if (!th->syn && th->ack &&
1118                     CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
1119                         /* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
1120                          * therefore we need to add 1 to make the SYN sequence
1121                          * number match the one of first SYN.
1122                          */
1123                         if (synproxy_recv_client_ack_ipv6(net, skb, th, &opts,
1124                                                           ntohl(th->seq) + 1)) {
1125                                 this_cpu_inc(snet->stats->cookie_retrans);
1126                                 consume_skb(skb);
1127                                 return NF_STOLEN;
1128                         } else {
1129                                 return NF_DROP;
1130                         }
1131                 }
1132
1133                 synproxy->isn = ntohl(th->ack_seq);
1134                 if (opts.options & NF_SYNPROXY_OPT_TIMESTAMP)
1135                         synproxy->its = opts.tsecr;
1136
1137                 nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
1138                 break;
1139         case TCP_CONNTRACK_SYN_RECV:
1140                 if (!th->syn || !th->ack)
1141                         break;
1142
1143                 if (!synproxy_parse_options(skb, thoff, th, &opts))
1144                         return NF_DROP;
1145
1146                 if (opts.options & NF_SYNPROXY_OPT_TIMESTAMP) {
1147                         synproxy->tsoff = opts.tsval - synproxy->its;
1148                         nf_conntrack_event_cache(IPCT_SYNPROXY, ct);
1149                 }
1150
1151                 opts.options &= ~(NF_SYNPROXY_OPT_MSS |
1152                                   NF_SYNPROXY_OPT_WSCALE |
1153                                   NF_SYNPROXY_OPT_SACK_PERM);
1154
1155                 swap(opts.tsval, opts.tsecr);
1156                 synproxy_send_server_ack_ipv6(net, state, skb, th, &opts);
1157
1158                 nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
1159                 nf_conntrack_event_cache(IPCT_SEQADJ, ct);
1160
1161                 swap(opts.tsval, opts.tsecr);
1162                 synproxy_send_client_ack_ipv6(net, skb, th, &opts);
1163
1164                 consume_skb(skb);
1165                 return NF_STOLEN;
1166         default:
1167                 break;
1168         }
1169
1170         synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
1171         return NF_ACCEPT;
1172 }
1173 EXPORT_SYMBOL_GPL(ipv6_synproxy_hook);
1174
1175 static const struct nf_hook_ops ipv6_synproxy_ops[] = {
1176         {
1177                 .hook           = ipv6_synproxy_hook,
1178                 .pf             = NFPROTO_IPV6,
1179                 .hooknum        = NF_INET_LOCAL_IN,
1180                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
1181         },
1182         {
1183                 .hook           = ipv6_synproxy_hook,
1184                 .pf             = NFPROTO_IPV6,
1185                 .hooknum        = NF_INET_POST_ROUTING,
1186                 .priority       = NF_IP_PRI_CONNTRACK_CONFIRM - 1,
1187         },
1188 };
1189
1190 int
1191 nf_synproxy_ipv6_init(struct synproxy_net *snet, struct net *net)
1192 {
1193         int err;
1194
1195         if (snet->hook_ref6 == 0) {
1196                 err = nf_register_net_hooks(net, ipv6_synproxy_ops,
1197                                             ARRAY_SIZE(ipv6_synproxy_ops));
1198                 if (err)
1199                         return err;
1200         }
1201
1202         snet->hook_ref6++;
1203         return 0;
1204 }
1205 EXPORT_SYMBOL_GPL(nf_synproxy_ipv6_init);
1206
1207 void
1208 nf_synproxy_ipv6_fini(struct synproxy_net *snet, struct net *net)
1209 {
1210         snet->hook_ref6--;
1211         if (snet->hook_ref6 == 0)
1212                 nf_unregister_net_hooks(net, ipv6_synproxy_ops,
1213                                         ARRAY_SIZE(ipv6_synproxy_ops));
1214 }
1215 EXPORT_SYMBOL_GPL(nf_synproxy_ipv6_fini);
1216 #endif /* CONFIG_IPV6 */
1217
1218 MODULE_LICENSE("GPL");
1219 MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
1220 MODULE_DESCRIPTION("nftables SYNPROXY expression support");