xref: /freebsd/sys/netpfil/pf/pf_syncookies.c (revision 3332f1b444d4a73238e9f59cca27bfc95fe936bd)
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 
145 	nvl = nvlist_create(0);
146 	if (nvl == NULL)
147 		return (ENOMEM);
148 
149 	nvlist_add_bool(nvl, "enabled",
150 	    V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER);
151 	nvlist_add_bool(nvl, "adaptive",
152 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE);
153 	nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat);
154 	nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat);
155 
156 	nvlpacked = nvlist_pack(nvl, &nv->len);
157 	if (nvlpacked == NULL) {
158 		nvlist_destroy(nvl);
159 		return (ENOMEM);
160 	}
161 	if (nv->size == 0) {
162 		nvlist_destroy(nvl);
163 		free(nvlpacked, M_TEMP);
164 		return (0);
165 	} else if (nv->size < nv->len) {
166 		nvlist_destroy(nvl);
167 		free(nvlpacked, M_TEMP);
168 		return (ENOSPC);
169 	}
170 
171 	return (copyout(nvlpacked, nv->data, nv->len));
172 }
173 
174 int
175 pf_set_syncookies(struct pfioc_nv *nv)
176 {
177 	nvlist_t	*nvl = NULL;
178 	void		*nvlpacked = NULL;
179 	int		 error;
180 	bool		 enabled, adaptive;
181 	uint32_t	 hiwat, lowat;
182 	uint8_t		 newmode;
183 
184 #define ERROUT(x)	ERROUT_FUNCTION(errout, x)
185 
186 	if (nv->len > pf_ioctl_maxcount)
187 		return (ENOMEM);
188 
189 	nvlpacked = malloc(nv->len, M_TEMP, M_WAITOK);
190 	if (nvlpacked == NULL)
191 		return (ENOMEM);
192 
193 	error = copyin(nv->data, nvlpacked, nv->len);
194 	if (error)
195 		ERROUT(error);
196 
197 	nvl = nvlist_unpack(nvlpacked, nv->len, 0);
198 	if (nvl == NULL)
199 		ERROUT(EBADMSG);
200 
201 	if (! nvlist_exists_bool(nvl, "enabled")
202 	    || ! nvlist_exists_bool(nvl, "adaptive"))
203 		ERROUT(EBADMSG);
204 
205 	enabled = nvlist_get_bool(nvl, "enabled");
206 	adaptive = nvlist_get_bool(nvl, "adaptive");
207 	PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat,
208 	    V_pf_syncookie_status.hiwat));
209 	PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat,
210 	    V_pf_syncookie_status.lowat));
211 
212 	if (lowat >= hiwat)
213 		ERROUT(EINVAL);
214 
215 	newmode = PF_SYNCOOKIES_NEVER;
216 	if (enabled)
217 		newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS;
218 
219 	PF_RULES_WLOCK();
220 	error = pf_syncookies_setmode(newmode);
221 
222 	V_pf_syncookie_status.lowat = lowat;
223 	V_pf_syncookie_status.hiwat = hiwat;
224 
225 	PF_RULES_WUNLOCK();
226 
227 #undef ERROUT
228 errout:
229 	nvlist_destroy(nvl);
230 	free(nvlpacked, M_TEMP);
231 
232 	return (error);
233 }
234 
235 static int
236 pf_syncookies_setmode(u_int8_t mode)
237 {
238 	if (mode > PF_SYNCOOKIES_MODE_MAX)
239 		return (EINVAL);
240 
241 	if (V_pf_status.syncookies_mode == mode)
242 		return (0);
243 
244 	V_pf_status.syncookies_mode = mode;
245 	if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) {
246 		pf_syncookie_newkey();
247 		V_pf_status.syncookies_active = true;
248 	}
249 	return (0);
250 }
251 
252 int
253 pf_synflood_check(struct pf_pdesc *pd)
254 {
255 	MPASS(pd->proto == IPPROTO_TCP);
256 	PF_RULES_RASSERT();
257 
258 	if (pd->pf_mtag && (pd->pf_mtag->tag & PF_TAG_SYNCOOKIE_RECREATED))
259 		return (0);
260 
261 	if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE)
262 		return (V_pf_status.syncookies_mode);
263 
264 	if (!V_pf_status.syncookies_active &&
265 	    atomic_load_32(&V_pf_status.states_halfopen) >
266 	    V_pf_syncookie_status.hiwat) {
267 		/* We'd want to 'pf_syncookie_newkey()' here, but that requires
268 		 * the rules write lock, which we can't get with the read lock
269 		 * held. */
270 		callout_reset(&V_pf_syncookie_status.keytimeout, 0,
271 		    pf_syncookie_rotate, curvnet);
272 		V_pf_status.syncookies_active = true;
273 		DPFPRINTF(LOG_WARNING,
274 		    ("synflood detected, enabling syncookies\n"));
275 		// XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++;
276 	}
277 
278 	return (V_pf_status.syncookies_active);
279 }
280 
281 void
282 pf_syncookie_send(struct mbuf *m, int off, struct pf_pdesc *pd)
283 {
284 	uint16_t	mss;
285 	uint32_t	iss;
286 
287 	mss = max(V_tcp_mssdflt, pf_get_mss(m, off, pd->hdr.tcp.th_off, pd->af));
288 	iss = pf_syncookie_generate(m, off, pd, mss);
289 	pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport,
290 	    iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss,
291 	    0, 1, 0);
292 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1);
293 	/* XXX Maybe only in adaptive mode? */
294 	atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven],
295 	    1);
296 }
297 
298 uint8_t
299 pf_syncookie_validate(struct pf_pdesc *pd)
300 {
301 	uint32_t		 hash, ack, seq;
302 	union pf_syncookie	 cookie;
303 
304 	MPASS(pd->proto == IPPROTO_TCP);
305 	PF_RULES_RASSERT();
306 
307 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
308 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
309 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
310 
311 	/* we don't know oddeven before setting the cookie (union) */
312         if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven])
313 	    == 0)
314                 return (0);
315 
316 	hash = pf_syncookie_mac(pd, cookie, seq);
317 	if ((ack & ~0xff) != (hash & ~0xff))
318 		return (0);
319 
320 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1);
321 	atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1);
322 
323 	return (1);
324 }
325 
326 /*
327  * all following functions private
328  */
329 void
330 pf_syncookie_rotate(void *arg)
331 {
332 	CURVNET_SET((struct vnet *)arg);
333 
334 	/* do we want to disable syncookies? */
335 	if (V_pf_status.syncookies_active &&
336 	    ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE &&
337 	    (atomic_load_32(&V_pf_status.states_halfopen) +
338 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) +
339 	    atomic_load_64(&V_pf_status.syncookies_inflight[1])) <
340 	    V_pf_syncookie_status.lowat) ||
341 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER)
342 			) {
343 		V_pf_status.syncookies_active = false;
344 		DPFPRINTF(PF_DEBUG_MISC, ("syncookies disabled\n"));
345 	}
346 
347 	/* nothing in flight any more? delete keys and return */
348 	if (!V_pf_status.syncookies_active &&
349 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 &&
350 	    atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) {
351 		memset(V_pf_syncookie_status.key[0], 0,
352 		    PF_SYNCOOKIE_SECRET_SIZE);
353 		memset(V_pf_syncookie_status.key[1], 0,
354 		    PF_SYNCOOKIE_SECRET_SIZE);
355 		CURVNET_RESTORE();
356 		return;
357 	}
358 
359 	PF_RULES_WLOCK();
360 	/* new key, including timeout */
361 	pf_syncookie_newkey();
362 	PF_RULES_WUNLOCK();
363 
364 	CURVNET_RESTORE();
365 }
366 
367 void
368 pf_syncookie_newkey(void)
369 {
370 	PF_RULES_WASSERT();
371 
372 	MPASS(V_pf_syncookie_status.oddeven < 2);
373 	V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1;
374 	atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0);
375 	arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven],
376 	    PF_SYNCOOKIE_SECRET_SIZE);
377 	callout_reset(&V_pf_syncookie_status.keytimeout,
378 	    PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet);
379 }
380 
381 /*
382  * Distribution and probability of certain MSS values.  Those in between are
383  * rounded down to the next lower one.
384  * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011]
385  *   .2%  .3%   5%    7%    7%    20%   15%   45%
386  */
387 static int pf_syncookie_msstab[] =
388     { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 };
389 
390 /*
391  * Distribution and probability of certain WSCALE values.
392  * The absence of the WSCALE option is encoded with index zero.
393  * [WSCALE values histograms, Allman, 2012]
394  *                                  X 10 10 35  5  6 14 10%   by host
395  *                                  X 11  4  5  5 18 49  3%   by connections
396  */
397 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 };
398 
399 uint32_t
400 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq)
401 {
402 	SIPHASH_CTX	ctx;
403 	uint32_t	siphash[2];
404 
405 	PF_RULES_RASSERT();
406 	MPASS(pd->proto == IPPROTO_TCP);
407 
408 	SipHash24_Init(&ctx);
409 	SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]);
410 
411 	switch (pd->af) {
412 	case AF_INET:
413 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4));
414 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4));
415 		break;
416 	case AF_INET6:
417 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6));
418 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6));
419 		break;
420 	default:
421 		panic("unknown address family");
422 	}
423 
424 	SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport));
425 	SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport));
426 	SipHash_Update(&ctx, &seq, sizeof(seq));
427 	SipHash_Update(&ctx, &cookie, sizeof(cookie));
428 	SipHash_Final((uint8_t *)&siphash, &ctx);
429 
430 	return (siphash[0] ^ siphash[1]);
431 }
432 
433 uint32_t
434 pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *pd,
435     uint16_t mss)
436 {
437 	uint8_t			 i, wscale;
438 	uint32_t		 iss, hash;
439 	union pf_syncookie	 cookie;
440 
441 	PF_RULES_RASSERT();
442 
443 	cookie.cookie = 0;
444 
445 	/* map MSS */
446 	for (i = nitems(pf_syncookie_msstab) - 1;
447 	    pf_syncookie_msstab[i] > mss && i > 0; i--)
448 		/* nada */;
449 	cookie.flags.mss_idx = i;
450 
451 	/* map WSCALE */
452 	wscale = pf_get_wscale(m, off, pd->hdr.tcp.th_off, pd->af);
453 	for (i = nitems(pf_syncookie_wstab) - 1;
454 	    pf_syncookie_wstab[i] > wscale && i > 0; i--)
455 		/* nada */;
456 	cookie.flags.wscale_idx = i;
457 	cookie.flags.sack_ok = 0;	/* XXX */
458 
459 	cookie.flags.oddeven = V_pf_syncookie_status.oddeven;
460 	hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq));
461 
462 	/*
463 	 * Put the flags into the hash and XOR them to get better ISS number
464 	 * variance.  This doesn't enhance the cryptographic strength and is
465 	 * done to prevent the 8 cookie bits from showing up directly on the
466 	 * wire.
467 	 */
468 	iss = hash & ~0xff;
469 	iss |= cookie.cookie ^ (hash >> 24);
470 
471 	return (iss);
472 }
473 
474 struct mbuf *
475 pf_syncookie_recreate_syn(uint8_t ttl, int off, struct pf_pdesc *pd)
476 {
477 	uint8_t			 wscale;
478 	uint16_t		 mss;
479 	uint32_t		 ack, seq;
480 	union pf_syncookie	 cookie;
481 
482 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
483 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
484 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
485 
486 	if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) ||
487 	    cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab))
488 		return (NULL);
489 
490 	mss = pf_syncookie_msstab[cookie.flags.mss_idx];
491 	wscale = pf_syncookie_wstab[cookie.flags.wscale_idx];
492 
493 	return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport,
494 	    *pd->dport, seq, 0, TH_SYN, wscale, mss, ttl, 0,
495 	    PF_TAG_SYNCOOKIE_RECREATED));
496 }
497