xref: /freebsd/sys/netpfil/pf/pf_syncookies.c (revision bc7512cc58af2e8bbe5bbf5ca0059b1daa1da897)
1 /*	$OpenBSD: pf_syncookies.c,v 1.7 2018/09/10 15:54:28 henning Exp $ */
2 
3 /* Copyright (c) 2016,2017 Henning Brauer <henning@openbsd.org>
4  * Copyright (c) 2016 Alexandr Nedvedicky <sashan@openbsd.org>
5  *
6  * syncookie parts based on FreeBSD sys/netinet/tcp_syncache.c
7  *
8  * Copyright (c) 2001 McAfee, Inc.
9  * Copyright (c) 2006,2013 Andre Oppermann, Internet Business Solutions AG
10  * All rights reserved.
11  *
12  * This software was developed for the FreeBSD Project by Jonathan Lemon
13  * and McAfee Research, the Security Research Division of McAfee, Inc. under
14  * DARPA/SPAWAR contract N66001-01-C-8035 ("CBOSS"), as part of the
15  * DARPA CHATS research program. [2001 McAfee, Inc.]
16  *
17  * Redistribution and use in source and binary forms, with or without
18  * modification, are permitted provided that the following conditions
19  * are met:
20  * 1. Redistributions of source code must retain the above copyright
21  *    notice, this list of conditions and the following disclaimer.
22  * 2. Redistributions in binary form must reproduce the above copyright
23  *    notice, this list of conditions and the following disclaimer in the
24  *    documentation and/or other materials provided with the distribution.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 /*
40  * when we're under synflood, we use syncookies to prevent state table
41  * exhaustion. Trigger for the synflood mode is the number of half-open
42  * connections in the state table.
43  * We leave synflood mode when the number of half-open states - including
44  * in-flight syncookies - drops far enough again
45  */
46 
47 /*
48  * syncookie enabled Initial Sequence Number:
49  *  24 bit MAC
50  *   3 bit WSCALE index
51  *   3 bit MSS index
52  *   1 bit SACK permitted
53  *   1 bit odd/even secret
54  *
55  * References:
56  *  RFC4987 TCP SYN Flooding Attacks and Common Mitigations
57  *  http://cr.yp.to/syncookies.html    (overview)
58  *  http://cr.yp.to/syncookies/archive (details)
59  */
60 
61 //#include "pflog.h"
62 
63 #include <sys/param.h>
64 #include <sys/systm.h>
65 #include <sys/mbuf.h>
66 #include <sys/filio.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/kernel.h>
70 #include <sys/time.h>
71 #include <sys/proc.h>
72 #include <sys/rwlock.h>
73 #include <sys/syslog.h>
74 
75 #include <crypto/siphash/siphash.h>
76 
77 #include <net/if.h>
78 #include <net/if_var.h>
79 #include <net/if_types.h>
80 #include <net/route.h>
81 
82 #include <netinet/in.h>
83 #include <netinet/ip.h>
84 #include <netinet/tcp.h>
85 #include <netinet/tcp_var.h>
86 
87 #include <net/pfvar.h>
88 #include <netpfil/pf/pf_nv.h>
89 
90 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
91 
92 union pf_syncookie {
93 	uint8_t		cookie;
94 	struct {
95 		uint8_t	oddeven:1,
96 			sack_ok:1,
97 			wscale_idx:3,
98 			mss_idx:3;
99 	} flags;
100 };
101 
102 #define	PF_SYNCOOKIE_SECRET_SIZE	SIPHASH_KEY_LENGTH
103 #define	PF_SYNCOOKIE_SECRET_LIFETIME	15 /* seconds */
104 
105 /* Protected by PF_RULES_xLOCK. */
106 struct pf_syncookie_status {
107 	struct callout	keytimeout;
108 	uint8_t		oddeven;
109 	uint8_t		key[2][SIPHASH_KEY_LENGTH];
110 	uint32_t	hiwat;	/* absolute; # of states */
111 	uint32_t	lowat;
112 };
113 VNET_DEFINE_STATIC(struct pf_syncookie_status, pf_syncookie_status);
114 #define V_pf_syncookie_status	VNET(pf_syncookie_status)
115 
116 static int	pf_syncookies_setmode(u_int8_t);
117 void		pf_syncookie_rotate(void *);
118 void		pf_syncookie_newkey(void);
119 uint32_t	pf_syncookie_mac(struct pf_pdesc *, union pf_syncookie,
120 		    uint32_t);
121 uint32_t	pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *,
122 		    uint16_t);
123 
124 void
125 pf_syncookies_init(void)
126 {
127 	callout_init(&V_pf_syncookie_status.keytimeout, 1);
128 	PF_RULES_WLOCK();
129 	pf_syncookies_setmode(PF_SYNCOOKIES_NEVER);
130 	PF_RULES_WUNLOCK();
131 }
132 
133 void
134 pf_syncookies_cleanup(void)
135 {
136 	callout_stop(&V_pf_syncookie_status.keytimeout);
137 }
138 
139 int
140 pf_get_syncookies(struct pfioc_nv *nv)
141 {
142 	nvlist_t	*nvl = NULL;
143 	void		*nvlpacked = NULL;
144 	int		 error;
145 
146 #define ERROUT(x)	ERROUT_FUNCTION(errout, x)
147 
148 	nvl = nvlist_create(0);
149 	if (nvl == NULL)
150 		ERROUT(ENOMEM);
151 
152 	nvlist_add_bool(nvl, "enabled",
153 	    V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER);
154 	nvlist_add_bool(nvl, "adaptive",
155 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE);
156 	nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat);
157 	nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat);
158 
159 	nvlpacked = nvlist_pack(nvl, &nv->len);
160 	if (nvlpacked == NULL)
161 		ERROUT(ENOMEM);
162 
163 	if (nv->size == 0) {
164 		ERROUT(0);
165 	} else if (nv->size < nv->len) {
166 		ERROUT(ENOSPC);
167 	}
168 
169 	error = copyout(nvlpacked, nv->data, nv->len);
170 
171 #undef ERROUT
172 errout:
173 	nvlist_destroy(nvl);
174 	free(nvlpacked, M_TEMP);
175 
176 	return (error);
177 }
178 
179 int
180 pf_set_syncookies(struct pfioc_nv *nv)
181 {
182 	nvlist_t	*nvl = NULL;
183 	void		*nvlpacked = NULL;
184 	int		 error;
185 	bool		 enabled, adaptive;
186 	uint32_t	 hiwat, lowat;
187 	uint8_t		 newmode;
188 
189 #define ERROUT(x)	ERROUT_FUNCTION(errout, x)
190 
191 	if (nv->len > pf_ioctl_maxcount)
192 		return (ENOMEM);
193 
194 	nvlpacked = malloc(nv->len, M_TEMP, M_WAITOK);
195 	if (nvlpacked == NULL)
196 		return (ENOMEM);
197 
198 	error = copyin(nv->data, nvlpacked, nv->len);
199 	if (error)
200 		ERROUT(error);
201 
202 	nvl = nvlist_unpack(nvlpacked, nv->len, 0);
203 	if (nvl == NULL)
204 		ERROUT(EBADMSG);
205 
206 	if (! nvlist_exists_bool(nvl, "enabled")
207 	    || ! nvlist_exists_bool(nvl, "adaptive"))
208 		ERROUT(EBADMSG);
209 
210 	enabled = nvlist_get_bool(nvl, "enabled");
211 	adaptive = nvlist_get_bool(nvl, "adaptive");
212 	PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat,
213 	    V_pf_syncookie_status.hiwat));
214 	PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat,
215 	    V_pf_syncookie_status.lowat));
216 
217 	if (lowat >= hiwat)
218 		ERROUT(EINVAL);
219 
220 	newmode = PF_SYNCOOKIES_NEVER;
221 	if (enabled)
222 		newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS;
223 
224 	PF_RULES_WLOCK();
225 	error = pf_syncookies_setmode(newmode);
226 
227 	V_pf_syncookie_status.lowat = lowat;
228 	V_pf_syncookie_status.hiwat = hiwat;
229 
230 	PF_RULES_WUNLOCK();
231 
232 #undef ERROUT
233 errout:
234 	nvlist_destroy(nvl);
235 	free(nvlpacked, M_TEMP);
236 
237 	return (error);
238 }
239 
240 static int
241 pf_syncookies_setmode(u_int8_t mode)
242 {
243 	if (mode > PF_SYNCOOKIES_MODE_MAX)
244 		return (EINVAL);
245 
246 	if (V_pf_status.syncookies_mode == mode)
247 		return (0);
248 
249 	V_pf_status.syncookies_mode = mode;
250 	if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) {
251 		pf_syncookie_newkey();
252 		V_pf_status.syncookies_active = true;
253 	}
254 	return (0);
255 }
256 
257 int
258 pf_synflood_check(struct pf_pdesc *pd)
259 {
260 	MPASS(pd->proto == IPPROTO_TCP);
261 	PF_RULES_RASSERT();
262 
263 	if (pd->pf_mtag && (pd->pf_mtag->tag & PF_TAG_SYNCOOKIE_RECREATED))
264 		return (0);
265 
266 	if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE)
267 		return (V_pf_status.syncookies_mode);
268 
269 	if (!V_pf_status.syncookies_active &&
270 	    atomic_load_32(&V_pf_status.states_halfopen) >
271 	    V_pf_syncookie_status.hiwat) {
272 		/* We'd want to 'pf_syncookie_newkey()' here, but that requires
273 		 * the rules write lock, which we can't get with the read lock
274 		 * held. */
275 		callout_reset(&V_pf_syncookie_status.keytimeout, 0,
276 		    pf_syncookie_rotate, curvnet);
277 		V_pf_status.syncookies_active = true;
278 		DPFPRINTF(LOG_WARNING,
279 		    ("synflood detected, enabling syncookies\n"));
280 		// XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++;
281 	}
282 
283 	return (V_pf_status.syncookies_active);
284 }
285 
286 void
287 pf_syncookie_send(struct mbuf *m, int off, struct pf_pdesc *pd)
288 {
289 	uint16_t	mss;
290 	uint32_t	iss;
291 
292 	mss = max(V_tcp_mssdflt, pf_get_mss(m, off, pd->hdr.tcp.th_off, pd->af));
293 	iss = pf_syncookie_generate(m, off, pd, mss);
294 	pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport,
295 	    iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss,
296 	    0, 1, 0);
297 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1);
298 	/* XXX Maybe only in adaptive mode? */
299 	atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven],
300 	    1);
301 }
302 
303 uint8_t
304 pf_syncookie_validate(struct pf_pdesc *pd)
305 {
306 	uint32_t		 hash, ack, seq;
307 	union pf_syncookie	 cookie;
308 
309 	MPASS(pd->proto == IPPROTO_TCP);
310 	PF_RULES_RASSERT();
311 
312 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
313 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
314 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
315 
316 	/* we don't know oddeven before setting the cookie (union) */
317         if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven])
318 	    == 0)
319                 return (0);
320 
321 	hash = pf_syncookie_mac(pd, cookie, seq);
322 	if ((ack & ~0xff) != (hash & ~0xff))
323 		return (0);
324 
325 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1);
326 	atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1);
327 
328 	return (1);
329 }
330 
331 /*
332  * all following functions private
333  */
334 void
335 pf_syncookie_rotate(void *arg)
336 {
337 	CURVNET_SET((struct vnet *)arg);
338 
339 	/* do we want to disable syncookies? */
340 	if (V_pf_status.syncookies_active &&
341 	    ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE &&
342 	    (atomic_load_32(&V_pf_status.states_halfopen) +
343 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) +
344 	    atomic_load_64(&V_pf_status.syncookies_inflight[1])) <
345 	    V_pf_syncookie_status.lowat) ||
346 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER)
347 			) {
348 		V_pf_status.syncookies_active = false;
349 		DPFPRINTF(PF_DEBUG_MISC, ("syncookies disabled\n"));
350 	}
351 
352 	/* nothing in flight any more? delete keys and return */
353 	if (!V_pf_status.syncookies_active &&
354 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 &&
355 	    atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) {
356 		memset(V_pf_syncookie_status.key[0], 0,
357 		    PF_SYNCOOKIE_SECRET_SIZE);
358 		memset(V_pf_syncookie_status.key[1], 0,
359 		    PF_SYNCOOKIE_SECRET_SIZE);
360 		CURVNET_RESTORE();
361 		return;
362 	}
363 
364 	PF_RULES_WLOCK();
365 	/* new key, including timeout */
366 	pf_syncookie_newkey();
367 	PF_RULES_WUNLOCK();
368 
369 	CURVNET_RESTORE();
370 }
371 
372 void
373 pf_syncookie_newkey(void)
374 {
375 	PF_RULES_WASSERT();
376 
377 	MPASS(V_pf_syncookie_status.oddeven < 2);
378 	V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1;
379 	atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0);
380 	arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven],
381 	    PF_SYNCOOKIE_SECRET_SIZE);
382 	callout_reset(&V_pf_syncookie_status.keytimeout,
383 	    PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet);
384 }
385 
386 /*
387  * Distribution and probability of certain MSS values.  Those in between are
388  * rounded down to the next lower one.
389  * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011]
390  *   .2%  .3%   5%    7%    7%    20%   15%   45%
391  */
392 static int pf_syncookie_msstab[] =
393     { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 };
394 
395 /*
396  * Distribution and probability of certain WSCALE values.
397  * The absence of the WSCALE option is encoded with index zero.
398  * [WSCALE values histograms, Allman, 2012]
399  *                                  X 10 10 35  5  6 14 10%   by host
400  *                                  X 11  4  5  5 18 49  3%   by connections
401  */
402 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 };
403 
404 uint32_t
405 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq)
406 {
407 	SIPHASH_CTX	ctx;
408 	uint32_t	siphash[2];
409 
410 	PF_RULES_RASSERT();
411 	MPASS(pd->proto == IPPROTO_TCP);
412 
413 	SipHash24_Init(&ctx);
414 	SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]);
415 
416 	switch (pd->af) {
417 	case AF_INET:
418 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4));
419 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4));
420 		break;
421 	case AF_INET6:
422 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6));
423 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6));
424 		break;
425 	default:
426 		panic("unknown address family");
427 	}
428 
429 	SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport));
430 	SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport));
431 	SipHash_Update(&ctx, &seq, sizeof(seq));
432 	SipHash_Update(&ctx, &cookie, sizeof(cookie));
433 	SipHash_Final((uint8_t *)&siphash, &ctx);
434 
435 	return (siphash[0] ^ siphash[1]);
436 }
437 
438 uint32_t
439 pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *pd,
440     uint16_t mss)
441 {
442 	uint8_t			 i, wscale;
443 	uint32_t		 iss, hash;
444 	union pf_syncookie	 cookie;
445 
446 	PF_RULES_RASSERT();
447 
448 	cookie.cookie = 0;
449 
450 	/* map MSS */
451 	for (i = nitems(pf_syncookie_msstab) - 1;
452 	    pf_syncookie_msstab[i] > mss && i > 0; i--)
453 		/* nada */;
454 	cookie.flags.mss_idx = i;
455 
456 	/* map WSCALE */
457 	wscale = pf_get_wscale(m, off, pd->hdr.tcp.th_off, pd->af);
458 	for (i = nitems(pf_syncookie_wstab) - 1;
459 	    pf_syncookie_wstab[i] > wscale && i > 0; i--)
460 		/* nada */;
461 	cookie.flags.wscale_idx = i;
462 	cookie.flags.sack_ok = 0;	/* XXX */
463 
464 	cookie.flags.oddeven = V_pf_syncookie_status.oddeven;
465 	hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq));
466 
467 	/*
468 	 * Put the flags into the hash and XOR them to get better ISS number
469 	 * variance.  This doesn't enhance the cryptographic strength and is
470 	 * done to prevent the 8 cookie bits from showing up directly on the
471 	 * wire.
472 	 */
473 	iss = hash & ~0xff;
474 	iss |= cookie.cookie ^ (hash >> 24);
475 
476 	return (iss);
477 }
478 
479 struct mbuf *
480 pf_syncookie_recreate_syn(uint8_t ttl, int off, struct pf_pdesc *pd)
481 {
482 	uint8_t			 wscale;
483 	uint16_t		 mss;
484 	uint32_t		 ack, seq;
485 	union pf_syncookie	 cookie;
486 
487 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
488 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
489 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
490 
491 	if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) ||
492 	    cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab))
493 		return (NULL);
494 
495 	mss = pf_syncookie_msstab[cookie.flags.mss_idx];
496 	wscale = pf_syncookie_wstab[cookie.flags.wscale_idx];
497 
498 	return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport,
499 	    *pd->dport, seq, 0, TH_SYN, wscale, mss, ttl, 0,
500 	    PF_TAG_SYNCOOKIE_RECREATED));
501 }
502