1/*
2 * Copyright (c) 2013 Patrick McHardy <kaber@trash.net>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 */
8
9#include <linux/module.h>
10#include <linux/skbuff.h>
11#include <net/tcp.h>
12
13#include <linux/netfilter_ipv4/ip_tables.h>
14#include <linux/netfilter/x_tables.h>
15#include <linux/netfilter/xt_SYNPROXY.h>
16#include <net/netfilter/nf_conntrack.h>
17#include <net/netfilter/nf_conntrack_seqadj.h>
18#include <net/netfilter/nf_conntrack_synproxy.h>
19
20static struct iphdr *
21synproxy_build_ip(struct sk_buff *skb, u32 saddr, u32 daddr)
22{
23	struct iphdr *iph;
24
25	skb_reset_network_header(skb);
26	iph = (struct iphdr *)skb_put(skb, sizeof(*iph));
27	iph->version	= 4;
28	iph->ihl	= sizeof(*iph) / 4;
29	iph->tos	= 0;
30	iph->id		= 0;
31	iph->frag_off	= htons(IP_DF);
32	iph->ttl	= sysctl_ip_default_ttl;
33	iph->protocol	= IPPROTO_TCP;
34	iph->check	= 0;
35	iph->saddr	= saddr;
36	iph->daddr	= daddr;
37
38	return iph;
39}
40
41static void
42synproxy_send_tcp(const struct sk_buff *skb, struct sk_buff *nskb,
43		  struct nf_conntrack *nfct, enum ip_conntrack_info ctinfo,
44		  struct iphdr *niph, struct tcphdr *nth,
45		  unsigned int tcp_hdr_size)
46{
47	nth->check = ~tcp_v4_check(tcp_hdr_size, niph->saddr, niph->daddr, 0);
48	nskb->ip_summed   = CHECKSUM_PARTIAL;
49	nskb->csum_start  = (unsigned char *)nth - nskb->head;
50	nskb->csum_offset = offsetof(struct tcphdr, check);
51
52	skb_dst_set_noref(nskb, skb_dst(skb));
53	nskb->protocol = htons(ETH_P_IP);
54	if (ip_route_me_harder(nskb, RTN_UNSPEC))
55		goto free_nskb;
56
57	if (nfct) {
58		nskb->nfct = nfct;
59		nskb->nfctinfo = ctinfo;
60		nf_conntrack_get(nfct);
61	}
62
63	ip_local_out(nskb);
64	return;
65
66free_nskb:
67	kfree_skb(nskb);
68}
69
70static void
71synproxy_send_client_synack(const struct sk_buff *skb, const struct tcphdr *th,
72			    const struct synproxy_options *opts)
73{
74	struct sk_buff *nskb;
75	struct iphdr *iph, *niph;
76	struct tcphdr *nth;
77	unsigned int tcp_hdr_size;
78	u16 mss = opts->mss;
79
80	iph = ip_hdr(skb);
81
82	tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
83	nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
84			 GFP_ATOMIC);
85	if (nskb == NULL)
86		return;
87	skb_reserve(nskb, MAX_TCP_HEADER);
88
89	niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
90
91	skb_reset_transport_header(nskb);
92	nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
93	nth->source	= th->dest;
94	nth->dest	= th->source;
95	nth->seq	= htonl(__cookie_v4_init_sequence(iph, th, &mss));
96	nth->ack_seq	= htonl(ntohl(th->seq) + 1);
97	tcp_flag_word(nth) = TCP_FLAG_SYN | TCP_FLAG_ACK;
98	if (opts->options & XT_SYNPROXY_OPT_ECN)
99		tcp_flag_word(nth) |= TCP_FLAG_ECE;
100	nth->doff	= tcp_hdr_size / 4;
101	nth->window	= 0;
102	nth->check	= 0;
103	nth->urg_ptr	= 0;
104
105	synproxy_build_options(nth, opts);
106
107	synproxy_send_tcp(skb, nskb, skb->nfct, IP_CT_ESTABLISHED_REPLY,
108			  niph, nth, tcp_hdr_size);
109}
110
111static void
112synproxy_send_server_syn(const struct synproxy_net *snet,
113			 const struct sk_buff *skb, const struct tcphdr *th,
114			 const struct synproxy_options *opts, u32 recv_seq)
115{
116	struct sk_buff *nskb;
117	struct iphdr *iph, *niph;
118	struct tcphdr *nth;
119	unsigned int tcp_hdr_size;
120
121	iph = ip_hdr(skb);
122
123	tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
124	nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
125			 GFP_ATOMIC);
126	if (nskb == NULL)
127		return;
128	skb_reserve(nskb, MAX_TCP_HEADER);
129
130	niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
131
132	skb_reset_transport_header(nskb);
133	nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
134	nth->source	= th->source;
135	nth->dest	= th->dest;
136	nth->seq	= htonl(recv_seq - 1);
137	/* ack_seq is used to relay our ISN to the synproxy hook to initialize
138	 * sequence number translation once a connection tracking entry exists.
139	 */
140	nth->ack_seq	= htonl(ntohl(th->ack_seq) - 1);
141	tcp_flag_word(nth) = TCP_FLAG_SYN;
142	if (opts->options & XT_SYNPROXY_OPT_ECN)
143		tcp_flag_word(nth) |= TCP_FLAG_ECE | TCP_FLAG_CWR;
144	nth->doff	= tcp_hdr_size / 4;
145	nth->window	= th->window;
146	nth->check	= 0;
147	nth->urg_ptr	= 0;
148
149	synproxy_build_options(nth, opts);
150
151	synproxy_send_tcp(skb, nskb, &snet->tmpl->ct_general, IP_CT_NEW,
152			  niph, nth, tcp_hdr_size);
153}
154
155static void
156synproxy_send_server_ack(const struct synproxy_net *snet,
157			 const struct ip_ct_tcp *state,
158			 const struct sk_buff *skb, const struct tcphdr *th,
159			 const struct synproxy_options *opts)
160{
161	struct sk_buff *nskb;
162	struct iphdr *iph, *niph;
163	struct tcphdr *nth;
164	unsigned int tcp_hdr_size;
165
166	iph = ip_hdr(skb);
167
168	tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
169	nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
170			 GFP_ATOMIC);
171	if (nskb == NULL)
172		return;
173	skb_reserve(nskb, MAX_TCP_HEADER);
174
175	niph = synproxy_build_ip(nskb, iph->daddr, iph->saddr);
176
177	skb_reset_transport_header(nskb);
178	nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
179	nth->source	= th->dest;
180	nth->dest	= th->source;
181	nth->seq	= htonl(ntohl(th->ack_seq));
182	nth->ack_seq	= htonl(ntohl(th->seq) + 1);
183	tcp_flag_word(nth) = TCP_FLAG_ACK;
184	nth->doff	= tcp_hdr_size / 4;
185	nth->window	= htons(state->seen[IP_CT_DIR_ORIGINAL].td_maxwin);
186	nth->check	= 0;
187	nth->urg_ptr	= 0;
188
189	synproxy_build_options(nth, opts);
190
191	synproxy_send_tcp(skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
192}
193
194static void
195synproxy_send_client_ack(const struct synproxy_net *snet,
196			 const struct sk_buff *skb, const struct tcphdr *th,
197			 const struct synproxy_options *opts)
198{
199	struct sk_buff *nskb;
200	struct iphdr *iph, *niph;
201	struct tcphdr *nth;
202	unsigned int tcp_hdr_size;
203
204	iph = ip_hdr(skb);
205
206	tcp_hdr_size = sizeof(*nth) + synproxy_options_size(opts);
207	nskb = alloc_skb(sizeof(*niph) + tcp_hdr_size + MAX_TCP_HEADER,
208			 GFP_ATOMIC);
209	if (nskb == NULL)
210		return;
211	skb_reserve(nskb, MAX_TCP_HEADER);
212
213	niph = synproxy_build_ip(nskb, iph->saddr, iph->daddr);
214
215	skb_reset_transport_header(nskb);
216	nth = (struct tcphdr *)skb_put(nskb, tcp_hdr_size);
217	nth->source	= th->source;
218	nth->dest	= th->dest;
219	nth->seq	= htonl(ntohl(th->seq) + 1);
220	nth->ack_seq	= th->ack_seq;
221	tcp_flag_word(nth) = TCP_FLAG_ACK;
222	nth->doff	= tcp_hdr_size / 4;
223	nth->window	= ntohs(htons(th->window) >> opts->wscale);
224	nth->check	= 0;
225	nth->urg_ptr	= 0;
226
227	synproxy_build_options(nth, opts);
228
229	synproxy_send_tcp(skb, nskb, NULL, 0, niph, nth, tcp_hdr_size);
230}
231
232static bool
233synproxy_recv_client_ack(const struct synproxy_net *snet,
234			 const struct sk_buff *skb, const struct tcphdr *th,
235			 struct synproxy_options *opts, u32 recv_seq)
236{
237	int mss;
238
239	mss = __cookie_v4_check(ip_hdr(skb), th, ntohl(th->ack_seq) - 1);
240	if (mss == 0) {
241		this_cpu_inc(snet->stats->cookie_invalid);
242		return false;
243	}
244
245	this_cpu_inc(snet->stats->cookie_valid);
246	opts->mss = mss;
247	opts->options |= XT_SYNPROXY_OPT_MSS;
248
249	if (opts->options & XT_SYNPROXY_OPT_TIMESTAMP)
250		synproxy_check_timestamp_cookie(opts);
251
252	synproxy_send_server_syn(snet, skb, th, opts, recv_seq);
253	return true;
254}
255
256static unsigned int
257synproxy_tg4(struct sk_buff *skb, const struct xt_action_param *par)
258{
259	const struct xt_synproxy_info *info = par->targinfo;
260	struct synproxy_net *snet = synproxy_pernet(dev_net(par->in));
261	struct synproxy_options opts = {};
262	struct tcphdr *th, _th;
263
264	if (nf_ip_checksum(skb, par->hooknum, par->thoff, IPPROTO_TCP))
265		return NF_DROP;
266
267	th = skb_header_pointer(skb, par->thoff, sizeof(_th), &_th);
268	if (th == NULL)
269		return NF_DROP;
270
271	if (!synproxy_parse_options(skb, par->thoff, th, &opts))
272		return NF_DROP;
273
274	if (th->syn && !(th->ack || th->fin || th->rst)) {
275		/* Initial SYN from client */
276		this_cpu_inc(snet->stats->syn_received);
277
278		if (th->ece && th->cwr)
279			opts.options |= XT_SYNPROXY_OPT_ECN;
280
281		opts.options &= info->options;
282		if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
283			synproxy_init_timestamp_cookie(info, &opts);
284		else
285			opts.options &= ~(XT_SYNPROXY_OPT_WSCALE |
286					  XT_SYNPROXY_OPT_SACK_PERM |
287					  XT_SYNPROXY_OPT_ECN);
288
289		synproxy_send_client_synack(skb, th, &opts);
290		return NF_DROP;
291
292	} else if (th->ack && !(th->fin || th->rst || th->syn)) {
293		/* ACK from client */
294		synproxy_recv_client_ack(snet, skb, th, &opts, ntohl(th->seq));
295		return NF_DROP;
296	}
297
298	return XT_CONTINUE;
299}
300
301static unsigned int ipv4_synproxy_hook(const struct nf_hook_ops *ops,
302				       struct sk_buff *skb,
303				       const struct net_device *in,
304				       const struct net_device *out,
305				       int (*okfn)(struct sk_buff *))
306{
307	struct synproxy_net *snet = synproxy_pernet(dev_net(in ? : out));
308	enum ip_conntrack_info ctinfo;
309	struct nf_conn *ct;
310	struct nf_conn_synproxy *synproxy;
311	struct synproxy_options opts = {};
312	const struct ip_ct_tcp *state;
313	struct tcphdr *th, _th;
314	unsigned int thoff;
315
316	ct = nf_ct_get(skb, &ctinfo);
317	if (ct == NULL)
318		return NF_ACCEPT;
319
320	synproxy = nfct_synproxy(ct);
321	if (synproxy == NULL)
322		return NF_ACCEPT;
323
324	if (nf_is_loopback_packet(skb))
325		return NF_ACCEPT;
326
327	thoff = ip_hdrlen(skb);
328	th = skb_header_pointer(skb, thoff, sizeof(_th), &_th);
329	if (th == NULL)
330		return NF_DROP;
331
332	state = &ct->proto.tcp;
333	switch (state->state) {
334	case TCP_CONNTRACK_CLOSE:
335		if (th->rst && !test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
336			nf_ct_seqadj_init(ct, ctinfo, synproxy->isn -
337						      ntohl(th->seq) + 1);
338			break;
339		}
340
341		if (!th->syn || th->ack ||
342		    CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
343			break;
344
345		/* Reopened connection - reset the sequence number and timestamp
346		 * adjustments, they will get initialized once the connection is
347		 * reestablished.
348		 */
349		nf_ct_seqadj_init(ct, ctinfo, 0);
350		synproxy->tsoff = 0;
351		this_cpu_inc(snet->stats->conn_reopened);
352
353		/* fall through */
354	case TCP_CONNTRACK_SYN_SENT:
355		if (!synproxy_parse_options(skb, thoff, th, &opts))
356			return NF_DROP;
357
358		if (!th->syn && th->ack &&
359		    CTINFO2DIR(ctinfo) == IP_CT_DIR_ORIGINAL) {
360			/* Keep-Alives are sent with SEG.SEQ = SND.NXT-1,
361			 * therefore we need to add 1 to make the SYN sequence
362			 * number match the one of first SYN.
363			 */
364			if (synproxy_recv_client_ack(snet, skb, th, &opts,
365						     ntohl(th->seq) + 1))
366				this_cpu_inc(snet->stats->cookie_retrans);
367
368			return NF_DROP;
369		}
370
371		synproxy->isn = ntohl(th->ack_seq);
372		if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
373			synproxy->its = opts.tsecr;
374		break;
375	case TCP_CONNTRACK_SYN_RECV:
376		if (!th->syn || !th->ack)
377			break;
378
379		if (!synproxy_parse_options(skb, thoff, th, &opts))
380			return NF_DROP;
381
382		if (opts.options & XT_SYNPROXY_OPT_TIMESTAMP)
383			synproxy->tsoff = opts.tsval - synproxy->its;
384
385		opts.options &= ~(XT_SYNPROXY_OPT_MSS |
386				  XT_SYNPROXY_OPT_WSCALE |
387				  XT_SYNPROXY_OPT_SACK_PERM);
388
389		swap(opts.tsval, opts.tsecr);
390		synproxy_send_server_ack(snet, state, skb, th, &opts);
391
392		nf_ct_seqadj_init(ct, ctinfo, synproxy->isn - ntohl(th->seq));
393
394		swap(opts.tsval, opts.tsecr);
395		synproxy_send_client_ack(snet, skb, th, &opts);
396
397		consume_skb(skb);
398		return NF_STOLEN;
399	default:
400		break;
401	}
402
403	synproxy_tstamp_adjust(skb, thoff, th, ct, ctinfo, synproxy);
404	return NF_ACCEPT;
405}
406
407static int synproxy_tg4_check(const struct xt_tgchk_param *par)
408{
409	const struct ipt_entry *e = par->entryinfo;
410
411	if (e->ip.proto != IPPROTO_TCP ||
412	    e->ip.invflags & XT_INV_PROTO)
413		return -EINVAL;
414
415	return nf_ct_l3proto_try_module_get(par->family);
416}
417
418static void synproxy_tg4_destroy(const struct xt_tgdtor_param *par)
419{
420	nf_ct_l3proto_module_put(par->family);
421}
422
423static struct xt_target synproxy_tg4_reg __read_mostly = {
424	.name		= "SYNPROXY",
425	.family		= NFPROTO_IPV4,
426	.hooks		= (1 << NF_INET_LOCAL_IN) | (1 << NF_INET_FORWARD),
427	.target		= synproxy_tg4,
428	.targetsize	= sizeof(struct xt_synproxy_info),
429	.checkentry	= synproxy_tg4_check,
430	.destroy	= synproxy_tg4_destroy,
431	.me		= THIS_MODULE,
432};
433
434static struct nf_hook_ops ipv4_synproxy_ops[] __read_mostly = {
435	{
436		.hook		= ipv4_synproxy_hook,
437		.owner		= THIS_MODULE,
438		.pf		= NFPROTO_IPV4,
439		.hooknum	= NF_INET_LOCAL_IN,
440		.priority	= NF_IP_PRI_CONNTRACK_CONFIRM - 1,
441	},
442	{
443		.hook		= ipv4_synproxy_hook,
444		.owner		= THIS_MODULE,
445		.pf		= NFPROTO_IPV4,
446		.hooknum	= NF_INET_POST_ROUTING,
447		.priority	= NF_IP_PRI_CONNTRACK_CONFIRM - 1,
448	},
449};
450
451static int __init synproxy_tg4_init(void)
452{
453	int err;
454
455	err = nf_register_hooks(ipv4_synproxy_ops,
456				ARRAY_SIZE(ipv4_synproxy_ops));
457	if (err < 0)
458		goto err1;
459
460	err = xt_register_target(&synproxy_tg4_reg);
461	if (err < 0)
462		goto err2;
463
464	return 0;
465
466err2:
467	nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
468err1:
469	return err;
470}
471
472static void __exit synproxy_tg4_exit(void)
473{
474	xt_unregister_target(&synproxy_tg4_reg);
475	nf_unregister_hooks(ipv4_synproxy_ops, ARRAY_SIZE(ipv4_synproxy_ops));
476}
477
478module_init(synproxy_tg4_init);
479module_exit(synproxy_tg4_exit);
480
481MODULE_LICENSE("GPL");
482MODULE_AUTHOR("Patrick McHardy <kaber@trash.net>");
483