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(); 130 131 V_pf_syncookie_status.hiwat = PF_SYNCOOKIES_HIWATPCT * 132 V_pf_limits[PF_LIMIT_STATES].limit / 100; 133 V_pf_syncookie_status.lowat = PF_SYNCOOKIES_LOWATPCT * 134 V_pf_limits[PF_LIMIT_STATES].limit / 100; 135 pf_syncookies_setmode(PF_SYNCOOKIES_ADAPTIVE); 136 137 PF_RULES_WUNLOCK(); 138 } 139 140 void 141 pf_syncookies_cleanup(void) 142 { 143 callout_stop(&V_pf_syncookie_status.keytimeout); 144 } 145 146 int 147 pf_get_syncookies(struct pfioc_nv *nv) 148 { 149 nvlist_t *nvl = NULL; 150 void *nvlpacked = NULL; 151 int error; 152 153 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 154 155 nvl = nvlist_create(0); 156 if (nvl == NULL) 157 ERROUT(ENOMEM); 158 159 nvlist_add_bool(nvl, "enabled", 160 V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER); 161 nvlist_add_bool(nvl, "adaptive", 162 V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE); 163 nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat); 164 nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat); 165 166 nvlpacked = nvlist_pack(nvl, &nv->len); 167 if (nvlpacked == NULL) 168 ERROUT(ENOMEM); 169 170 if (nv->size == 0) { 171 ERROUT(0); 172 } else if (nv->size < nv->len) { 173 ERROUT(ENOSPC); 174 } 175 176 error = copyout(nvlpacked, nv->data, nv->len); 177 178 #undef ERROUT 179 errout: 180 nvlist_destroy(nvl); 181 free(nvlpacked, M_NVLIST); 182 183 return (error); 184 } 185 186 int 187 pf_set_syncookies(struct pfioc_nv *nv) 188 { 189 nvlist_t *nvl = NULL; 190 void *nvlpacked = NULL; 191 int error; 192 bool enabled, adaptive; 193 uint32_t hiwat, lowat; 194 uint8_t newmode; 195 196 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 197 198 if (nv->len > pf_ioctl_maxcount) 199 return (ENOMEM); 200 201 nvlpacked = malloc(nv->len, M_NVLIST, M_WAITOK); 202 if (nvlpacked == NULL) 203 return (ENOMEM); 204 205 error = copyin(nv->data, nvlpacked, nv->len); 206 if (error) 207 ERROUT(error); 208 209 nvl = nvlist_unpack(nvlpacked, nv->len, 0); 210 if (nvl == NULL) 211 ERROUT(EBADMSG); 212 213 if (! nvlist_exists_bool(nvl, "enabled") 214 || ! nvlist_exists_bool(nvl, "adaptive")) 215 ERROUT(EBADMSG); 216 217 enabled = nvlist_get_bool(nvl, "enabled"); 218 adaptive = nvlist_get_bool(nvl, "adaptive"); 219 PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat, 220 V_pf_syncookie_status.hiwat)); 221 PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat, 222 V_pf_syncookie_status.lowat)); 223 224 if (lowat >= hiwat) 225 ERROUT(EINVAL); 226 227 newmode = PF_SYNCOOKIES_NEVER; 228 if (enabled) 229 newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS; 230 231 PF_RULES_WLOCK(); 232 error = pf_syncookies_setmode(newmode); 233 234 V_pf_syncookie_status.lowat = lowat; 235 V_pf_syncookie_status.hiwat = hiwat; 236 237 PF_RULES_WUNLOCK(); 238 239 #undef ERROUT 240 errout: 241 nvlist_destroy(nvl); 242 free(nvlpacked, M_NVLIST); 243 244 return (error); 245 } 246 247 static int 248 pf_syncookies_setmode(u_int8_t mode) 249 { 250 if (mode > PF_SYNCOOKIES_MODE_MAX) 251 return (EINVAL); 252 253 if (V_pf_status.syncookies_mode == mode) 254 return (0); 255 256 V_pf_status.syncookies_mode = mode; 257 if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) { 258 pf_syncookie_newkey(); 259 V_pf_status.syncookies_active = true; 260 } 261 return (0); 262 } 263 264 int 265 pf_synflood_check(struct pf_pdesc *pd) 266 { 267 MPASS(pd->proto == IPPROTO_TCP); 268 PF_RULES_RASSERT(); 269 270 if (pd->pf_mtag && (pd->pf_mtag->flags & PF_MTAG_FLAG_SYNCOOKIE_RECREATED)) 271 return (0); 272 273 if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE) 274 return (V_pf_status.syncookies_mode); 275 276 if (!V_pf_status.syncookies_active && 277 atomic_load_32(&V_pf_status.states_halfopen) > 278 V_pf_syncookie_status.hiwat) { 279 /* We'd want to 'pf_syncookie_newkey()' here, but that requires 280 * the rules write lock, which we can't get with the read lock 281 * held. */ 282 callout_reset(&V_pf_syncookie_status.keytimeout, 0, 283 pf_syncookie_rotate, curvnet); 284 V_pf_status.syncookies_active = true; 285 DPFPRINTF(LOG_WARNING, 286 ("synflood detected, enabling syncookies\n")); 287 // XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++; 288 } 289 290 return (V_pf_status.syncookies_active); 291 } 292 293 void 294 pf_syncookie_send(struct mbuf *m, int off, struct pf_pdesc *pd) 295 { 296 uint16_t mss; 297 uint32_t iss; 298 299 mss = max(V_tcp_mssdflt, pf_get_mss(m, off, pd->hdr.tcp.th_off, pd->af)); 300 iss = pf_syncookie_generate(m, off, pd, mss); 301 pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport, 302 iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss, 303 0, true, 0, 0, pd->act.rtableid); 304 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1); 305 /* XXX Maybe only in adaptive mode? */ 306 atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 307 1); 308 } 309 310 bool 311 pf_syncookie_check(struct pf_pdesc *pd) 312 { 313 uint32_t hash, ack, seq; 314 union pf_syncookie cookie; 315 316 MPASS(pd->proto == IPPROTO_TCP); 317 PF_RULES_RASSERT(); 318 319 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 320 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 321 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 322 323 /* we don't know oddeven before setting the cookie (union) */ 324 if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven]) 325 == 0) 326 return (0); 327 328 hash = pf_syncookie_mac(pd, cookie, seq); 329 if ((ack & ~0xff) != (hash & ~0xff)) 330 return (false); 331 332 return (true); 333 } 334 335 uint8_t 336 pf_syncookie_validate(struct pf_pdesc *pd) 337 { 338 uint32_t ack; 339 union pf_syncookie cookie; 340 341 if (! pf_syncookie_check(pd)) 342 return (0); 343 344 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 345 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 346 347 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1); 348 atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1); 349 350 return (1); 351 } 352 353 /* 354 * all following functions private 355 */ 356 void 357 pf_syncookie_rotate(void *arg) 358 { 359 CURVNET_SET((struct vnet *)arg); 360 361 /* do we want to disable syncookies? */ 362 if (V_pf_status.syncookies_active && 363 ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE && 364 (atomic_load_32(&V_pf_status.states_halfopen) + 365 atomic_load_64(&V_pf_status.syncookies_inflight[0]) + 366 atomic_load_64(&V_pf_status.syncookies_inflight[1])) < 367 V_pf_syncookie_status.lowat) || 368 V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER) 369 ) { 370 V_pf_status.syncookies_active = false; 371 DPFPRINTF(PF_DEBUG_MISC, ("syncookies disabled\n")); 372 } 373 374 /* nothing in flight any more? delete keys and return */ 375 if (!V_pf_status.syncookies_active && 376 atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 && 377 atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) { 378 memset(V_pf_syncookie_status.key[0], 0, 379 PF_SYNCOOKIE_SECRET_SIZE); 380 memset(V_pf_syncookie_status.key[1], 0, 381 PF_SYNCOOKIE_SECRET_SIZE); 382 CURVNET_RESTORE(); 383 return; 384 } 385 386 PF_RULES_WLOCK(); 387 /* new key, including timeout */ 388 pf_syncookie_newkey(); 389 PF_RULES_WUNLOCK(); 390 391 CURVNET_RESTORE(); 392 } 393 394 void 395 pf_syncookie_newkey(void) 396 { 397 PF_RULES_WASSERT(); 398 399 MPASS(V_pf_syncookie_status.oddeven < 2); 400 V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1; 401 atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0); 402 arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven], 403 PF_SYNCOOKIE_SECRET_SIZE); 404 callout_reset(&V_pf_syncookie_status.keytimeout, 405 PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet); 406 } 407 408 /* 409 * Distribution and probability of certain MSS values. Those in between are 410 * rounded down to the next lower one. 411 * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011] 412 * .2% .3% 5% 7% 7% 20% 15% 45% 413 */ 414 static int pf_syncookie_msstab[] = 415 { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 }; 416 417 /* 418 * Distribution and probability of certain WSCALE values. 419 * The absence of the WSCALE option is encoded with index zero. 420 * [WSCALE values histograms, Allman, 2012] 421 * X 10 10 35 5 6 14 10% by host 422 * X 11 4 5 5 18 49 3% by connections 423 */ 424 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 }; 425 426 uint32_t 427 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq) 428 { 429 SIPHASH_CTX ctx; 430 uint32_t siphash[2]; 431 432 PF_RULES_RASSERT(); 433 MPASS(pd->proto == IPPROTO_TCP); 434 435 SipHash24_Init(&ctx); 436 SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]); 437 438 switch (pd->af) { 439 case AF_INET: 440 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4)); 441 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4)); 442 break; 443 case AF_INET6: 444 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6)); 445 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6)); 446 break; 447 default: 448 panic("unknown address family"); 449 } 450 451 SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport)); 452 SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport)); 453 SipHash_Update(&ctx, &seq, sizeof(seq)); 454 SipHash_Update(&ctx, &cookie, sizeof(cookie)); 455 SipHash_Final((uint8_t *)&siphash, &ctx); 456 457 return (siphash[0] ^ siphash[1]); 458 } 459 460 uint32_t 461 pf_syncookie_generate(struct mbuf *m, int off, struct pf_pdesc *pd, 462 uint16_t mss) 463 { 464 uint8_t i, wscale; 465 uint32_t iss, hash; 466 union pf_syncookie cookie; 467 468 PF_RULES_RASSERT(); 469 470 cookie.cookie = 0; 471 472 /* map MSS */ 473 for (i = nitems(pf_syncookie_msstab) - 1; 474 pf_syncookie_msstab[i] > mss && i > 0; i--) 475 /* nada */; 476 cookie.flags.mss_idx = i; 477 478 /* map WSCALE */ 479 wscale = pf_get_wscale(m, off, pd->hdr.tcp.th_off, pd->af); 480 for (i = nitems(pf_syncookie_wstab) - 1; 481 pf_syncookie_wstab[i] > wscale && i > 0; i--) 482 /* nada */; 483 cookie.flags.wscale_idx = i; 484 cookie.flags.sack_ok = 0; /* XXX */ 485 486 cookie.flags.oddeven = V_pf_syncookie_status.oddeven; 487 hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq)); 488 489 /* 490 * Put the flags into the hash and XOR them to get better ISS number 491 * variance. This doesn't enhance the cryptographic strength and is 492 * done to prevent the 8 cookie bits from showing up directly on the 493 * wire. 494 */ 495 iss = hash & ~0xff; 496 iss |= cookie.cookie ^ (hash >> 24); 497 498 return (iss); 499 } 500 501 struct mbuf * 502 pf_syncookie_recreate_syn(uint8_t ttl, int off, struct pf_pdesc *pd) 503 { 504 uint8_t wscale; 505 uint16_t mss; 506 uint32_t ack, seq; 507 union pf_syncookie cookie; 508 509 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 510 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 511 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 512 513 if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) || 514 cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab)) 515 return (NULL); 516 517 mss = pf_syncookie_msstab[cookie.flags.mss_idx]; 518 wscale = pf_syncookie_wstab[cookie.flags.wscale_idx]; 519 520 return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport, 521 *pd->dport, seq, 0, TH_SYN, wscale, mss, ttl, false, 0, 522 PF_MTAG_FLAG_SYNCOOKIE_RECREATED, pd->act.rtableid)); 523 } 524