xref: /freebsd/sys/netpfil/pf/pf_syncookies.c (revision c07d6445eb89d9dd3950361b065b7bd110e3a043)
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/in_pcb.h>
84 #include <netinet/ip.h>
85 #include <netinet/tcp.h>
86 #include <netinet/tcp_var.h>
87 
88 #include <net/pfvar.h>
89 #include <netpfil/pf/pf_nv.h>
90 
91 #define	DPFPRINTF(n, x)	if (V_pf_status.debug >= (n)) printf x
92 
93 union pf_syncookie {
94 	uint8_t		cookie;
95 	struct {
96 		uint8_t	oddeven:1,
97 			sack_ok:1,
98 			wscale_idx:3,
99 			mss_idx:3;
100 	} flags;
101 };
102 
103 #define	PF_SYNCOOKIE_SECRET_SIZE	SIPHASH_KEY_LENGTH
104 #define	PF_SYNCOOKIE_SECRET_LIFETIME	15 /* seconds */
105 
106 /* Protected by PF_RULES_xLOCK. */
107 struct pf_syncookie_status {
108 	struct callout	keytimeout;
109 	uint8_t		oddeven;
110 	uint8_t		key[2][SIPHASH_KEY_LENGTH];
111 	uint32_t	hiwat;	/* absolute; # of states */
112 	uint32_t	lowat;
113 };
114 VNET_DEFINE_STATIC(struct pf_syncookie_status, pf_syncookie_status);
115 #define V_pf_syncookie_status	VNET(pf_syncookie_status)
116 
117 static int	pf_syncookies_setmode(u_int8_t);
118 void		pf_syncookie_rotate(void *);
119 void		pf_syncookie_newkey(void);
120 uint32_t	pf_syncookie_mac(struct pf_pdesc *, union pf_syncookie,
121 		    uint32_t);
122 uint32_t	pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *,
123 		    uint16_t);
124 
125 void
126 pf_syncookies_init(void)
127 {
128 	callout_init(&V_pf_syncookie_status.keytimeout, 1);
129 	PF_RULES_WLOCK();
130 	pf_syncookies_setmode(PF_SYNCOOKIES_NEVER);
131 	PF_RULES_WUNLOCK();
132 }
133 
134 void
135 pf_syncookies_cleanup(void)
136 {
137 	callout_stop(&V_pf_syncookie_status.keytimeout);
138 }
139 
140 int
141 pf_get_syncookies(struct pfioc_nv *nv)
142 {
143 	nvlist_t	*nvl = NULL;
144 	void		*nvlpacked = NULL;
145 	int		 error;
146 
147 #define ERROUT(x)	ERROUT_FUNCTION(errout, x)
148 
149 	nvl = nvlist_create(0);
150 	if (nvl == NULL)
151 		ERROUT(ENOMEM);
152 
153 	nvlist_add_bool(nvl, "enabled",
154 	    V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER);
155 	nvlist_add_bool(nvl, "adaptive",
156 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE);
157 	nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat);
158 	nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat);
159 
160 	nvlpacked = nvlist_pack(nvl, &nv->len);
161 	if (nvlpacked == NULL)
162 		ERROUT(ENOMEM);
163 
164 	if (nv->size == 0) {
165 		ERROUT(0);
166 	} else if (nv->size < nv->len) {
167 		ERROUT(ENOSPC);
168 	}
169 
170 	error = copyout(nvlpacked, nv->data, nv->len);
171 
172 #undef ERROUT
173 errout:
174 	nvlist_destroy(nvl);
175 	free(nvlpacked, M_NVLIST);
176 
177 	return (error);
178 }
179 
180 int
181 pf_set_syncookies(struct pfioc_nv *nv)
182 {
183 	nvlist_t	*nvl = NULL;
184 	void		*nvlpacked = NULL;
185 	int		 error;
186 	bool		 enabled, adaptive;
187 	uint32_t	 hiwat, lowat;
188 	uint8_t		 newmode;
189 
190 #define ERROUT(x)	ERROUT_FUNCTION(errout, x)
191 
192 	if (nv->len > pf_ioctl_maxcount)
193 		return (ENOMEM);
194 
195 	nvlpacked = malloc(nv->len, M_NVLIST, M_WAITOK);
196 	if (nvlpacked == NULL)
197 		return (ENOMEM);
198 
199 	error = copyin(nv->data, nvlpacked, nv->len);
200 	if (error)
201 		ERROUT(error);
202 
203 	nvl = nvlist_unpack(nvlpacked, nv->len, 0);
204 	if (nvl == NULL)
205 		ERROUT(EBADMSG);
206 
207 	if (! nvlist_exists_bool(nvl, "enabled")
208 	    || ! nvlist_exists_bool(nvl, "adaptive"))
209 		ERROUT(EBADMSG);
210 
211 	enabled = nvlist_get_bool(nvl, "enabled");
212 	adaptive = nvlist_get_bool(nvl, "adaptive");
213 	PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat,
214 	    V_pf_syncookie_status.hiwat));
215 	PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat,
216 	    V_pf_syncookie_status.lowat));
217 
218 	if (lowat >= hiwat)
219 		ERROUT(EINVAL);
220 
221 	newmode = PF_SYNCOOKIES_NEVER;
222 	if (enabled)
223 		newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS;
224 
225 	PF_RULES_WLOCK();
226 	error = pf_syncookies_setmode(newmode);
227 
228 	V_pf_syncookie_status.lowat = lowat;
229 	V_pf_syncookie_status.hiwat = hiwat;
230 
231 	PF_RULES_WUNLOCK();
232 
233 #undef ERROUT
234 errout:
235 	nvlist_destroy(nvl);
236 	free(nvlpacked, M_NVLIST);
237 
238 	return (error);
239 }
240 
241 static int
242 pf_syncookies_setmode(u_int8_t mode)
243 {
244 	if (mode > PF_SYNCOOKIES_MODE_MAX)
245 		return (EINVAL);
246 
247 	if (V_pf_status.syncookies_mode == mode)
248 		return (0);
249 
250 	V_pf_status.syncookies_mode = mode;
251 	if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) {
252 		pf_syncookie_newkey();
253 		V_pf_status.syncookies_active = true;
254 	}
255 	return (0);
256 }
257 
258 int
259 pf_synflood_check(struct pf_pdesc *pd)
260 {
261 	MPASS(pd->proto == IPPROTO_TCP);
262 	PF_RULES_RASSERT();
263 
264 	if (pd->pf_mtag && (pd->pf_mtag->tag & PF_TAG_SYNCOOKIE_RECREATED))
265 		return (0);
266 
267 	if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE)
268 		return (V_pf_status.syncookies_mode);
269 
270 	if (!V_pf_status.syncookies_active &&
271 	    atomic_load_32(&V_pf_status.states_halfopen) >
272 	    V_pf_syncookie_status.hiwat) {
273 		/* We'd want to 'pf_syncookie_newkey()' here, but that requires
274 		 * the rules write lock, which we can't get with the read lock
275 		 * held. */
276 		callout_reset(&V_pf_syncookie_status.keytimeout, 0,
277 		    pf_syncookie_rotate, curvnet);
278 		V_pf_status.syncookies_active = true;
279 		DPFPRINTF(LOG_WARNING,
280 		    ("synflood detected, enabling syncookies\n"));
281 		// XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++;
282 	}
283 
284 	return (V_pf_status.syncookies_active);
285 }
286 
287 void
288 pf_syncookie_send(struct mbuf *m, int off, struct pf_pdesc *pd)
289 {
290 	uint16_t	mss;
291 	uint32_t	iss;
292 
293 	mss = max(V_tcp_mssdflt, pf_get_mss(m, off, pd->hdr.tcp.th_off, pd->af));
294 	iss = pf_syncookie_generate(m, off, pd, mss);
295 	pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport,
296 	    iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss,
297 	    0, 1, 0);
298 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1);
299 	/* XXX Maybe only in adaptive mode? */
300 	atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven],
301 	    1);
302 }
303 
304 uint8_t
305 pf_syncookie_validate(struct pf_pdesc *pd)
306 {
307 	uint32_t		 hash, ack, seq;
308 	union pf_syncookie	 cookie;
309 
310 	MPASS(pd->proto == IPPROTO_TCP);
311 	PF_RULES_RASSERT();
312 
313 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
314 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
315 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
316 
317 	/* we don't know oddeven before setting the cookie (union) */
318         if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven])
319 	    == 0)
320                 return (0);
321 
322 	hash = pf_syncookie_mac(pd, cookie, seq);
323 	if ((ack & ~0xff) != (hash & ~0xff))
324 		return (0);
325 
326 	counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1);
327 	atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1);
328 
329 	return (1);
330 }
331 
332 /*
333  * all following functions private
334  */
335 void
336 pf_syncookie_rotate(void *arg)
337 {
338 	CURVNET_SET((struct vnet *)arg);
339 
340 	/* do we want to disable syncookies? */
341 	if (V_pf_status.syncookies_active &&
342 	    ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE &&
343 	    (atomic_load_32(&V_pf_status.states_halfopen) +
344 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) +
345 	    atomic_load_64(&V_pf_status.syncookies_inflight[1])) <
346 	    V_pf_syncookie_status.lowat) ||
347 	    V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER)
348 			) {
349 		V_pf_status.syncookies_active = false;
350 		DPFPRINTF(PF_DEBUG_MISC, ("syncookies disabled\n"));
351 	}
352 
353 	/* nothing in flight any more? delete keys and return */
354 	if (!V_pf_status.syncookies_active &&
355 	    atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 &&
356 	    atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) {
357 		memset(V_pf_syncookie_status.key[0], 0,
358 		    PF_SYNCOOKIE_SECRET_SIZE);
359 		memset(V_pf_syncookie_status.key[1], 0,
360 		    PF_SYNCOOKIE_SECRET_SIZE);
361 		CURVNET_RESTORE();
362 		return;
363 	}
364 
365 	PF_RULES_WLOCK();
366 	/* new key, including timeout */
367 	pf_syncookie_newkey();
368 	PF_RULES_WUNLOCK();
369 
370 	CURVNET_RESTORE();
371 }
372 
373 void
374 pf_syncookie_newkey(void)
375 {
376 	PF_RULES_WASSERT();
377 
378 	MPASS(V_pf_syncookie_status.oddeven < 2);
379 	V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1;
380 	atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0);
381 	arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven],
382 	    PF_SYNCOOKIE_SECRET_SIZE);
383 	callout_reset(&V_pf_syncookie_status.keytimeout,
384 	    PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet);
385 }
386 
387 /*
388  * Distribution and probability of certain MSS values.  Those in between are
389  * rounded down to the next lower one.
390  * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011]
391  *   .2%  .3%   5%    7%    7%    20%   15%   45%
392  */
393 static int pf_syncookie_msstab[] =
394     { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 };
395 
396 /*
397  * Distribution and probability of certain WSCALE values.
398  * The absence of the WSCALE option is encoded with index zero.
399  * [WSCALE values histograms, Allman, 2012]
400  *                                  X 10 10 35  5  6 14 10%   by host
401  *                                  X 11  4  5  5 18 49  3%   by connections
402  */
403 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 };
404 
405 uint32_t
406 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq)
407 {
408 	SIPHASH_CTX	ctx;
409 	uint32_t	siphash[2];
410 
411 	PF_RULES_RASSERT();
412 	MPASS(pd->proto == IPPROTO_TCP);
413 
414 	SipHash24_Init(&ctx);
415 	SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]);
416 
417 	switch (pd->af) {
418 	case AF_INET:
419 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4));
420 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4));
421 		break;
422 	case AF_INET6:
423 		SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6));
424 		SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6));
425 		break;
426 	default:
427 		panic("unknown address family");
428 	}
429 
430 	SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport));
431 	SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport));
432 	SipHash_Update(&ctx, &seq, sizeof(seq));
433 	SipHash_Update(&ctx, &cookie, sizeof(cookie));
434 	SipHash_Final((uint8_t *)&siphash, &ctx);
435 
436 	return (siphash[0] ^ siphash[1]);
437 }
438 
439 uint32_t
440 pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *pd,
441     uint16_t mss)
442 {
443 	uint8_t			 i, wscale;
444 	uint32_t		 iss, hash;
445 	union pf_syncookie	 cookie;
446 
447 	PF_RULES_RASSERT();
448 
449 	cookie.cookie = 0;
450 
451 	/* map MSS */
452 	for (i = nitems(pf_syncookie_msstab) - 1;
453 	    pf_syncookie_msstab[i] > mss && i > 0; i--)
454 		/* nada */;
455 	cookie.flags.mss_idx = i;
456 
457 	/* map WSCALE */
458 	wscale = pf_get_wscale(m, off, pd->hdr.tcp.th_off, pd->af);
459 	for (i = nitems(pf_syncookie_wstab) - 1;
460 	    pf_syncookie_wstab[i] > wscale && i > 0; i--)
461 		/* nada */;
462 	cookie.flags.wscale_idx = i;
463 	cookie.flags.sack_ok = 0;	/* XXX */
464 
465 	cookie.flags.oddeven = V_pf_syncookie_status.oddeven;
466 	hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq));
467 
468 	/*
469 	 * Put the flags into the hash and XOR them to get better ISS number
470 	 * variance.  This doesn't enhance the cryptographic strength and is
471 	 * done to prevent the 8 cookie bits from showing up directly on the
472 	 * wire.
473 	 */
474 	iss = hash & ~0xff;
475 	iss |= cookie.cookie ^ (hash >> 24);
476 
477 	return (iss);
478 }
479 
480 struct mbuf *
481 pf_syncookie_recreate_syn(uint8_t ttl, int off, struct pf_pdesc *pd)
482 {
483 	uint8_t			 wscale;
484 	uint16_t		 mss;
485 	uint32_t		 ack, seq;
486 	union pf_syncookie	 cookie;
487 
488 	seq = ntohl(pd->hdr.tcp.th_seq) - 1;
489 	ack = ntohl(pd->hdr.tcp.th_ack) - 1;
490 	cookie.cookie = (ack & 0xff) ^ (ack >> 24);
491 
492 	if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) ||
493 	    cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab))
494 		return (NULL);
495 
496 	mss = pf_syncookie_msstab[cookie.flags.mss_idx];
497 	wscale = pf_syncookie_wstab[cookie.flags.wscale_idx];
498 
499 	return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport,
500 	    *pd->dport, seq, 0, TH_SYN, wscale, mss, ttl, 0,
501 	    PF_TAG_SYNCOOKIE_RECREATED));
502 }
503