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 union pf_syncookie { 92 uint8_t cookie; 93 struct { 94 uint8_t oddeven:1, 95 sack_ok:1, 96 wscale_idx:3, 97 mss_idx:3; 98 } flags; 99 }; 100 101 #define PF_SYNCOOKIE_SECRET_SIZE SIPHASH_KEY_LENGTH 102 #define PF_SYNCOOKIE_SECRET_LIFETIME 15 /* seconds */ 103 104 /* Protected by PF_RULES_xLOCK. */ 105 struct pf_syncookie_status { 106 struct callout keytimeout; 107 uint8_t oddeven; 108 uint8_t key[2][SIPHASH_KEY_LENGTH]; 109 uint32_t hiwat; /* absolute; # of states */ 110 uint32_t lowat; 111 }; 112 VNET_DEFINE_STATIC(struct pf_syncookie_status, pf_syncookie_status); 113 #define V_pf_syncookie_status VNET(pf_syncookie_status) 114 115 static int pf_syncookies_setmode(u_int8_t); 116 void pf_syncookie_rotate(void *); 117 void pf_syncookie_newkey(void); 118 uint32_t pf_syncookie_mac(struct pf_pdesc *, union pf_syncookie, 119 uint32_t); 120 uint32_t pf_syncookie_generate(struct pf_pdesc *, uint16_t); 121 122 void 123 pf_syncookies_init(void) 124 { 125 callout_init(&V_pf_syncookie_status.keytimeout, 1); 126 PF_RULES_WLOCK(); 127 128 V_pf_syncookie_status.hiwat = PF_SYNCOOKIES_HIWATPCT * 129 V_pf_limits[PF_LIMIT_STATES].limit / 100; 130 V_pf_syncookie_status.lowat = PF_SYNCOOKIES_LOWATPCT * 131 V_pf_limits[PF_LIMIT_STATES].limit / 100; 132 pf_syncookies_setmode(PF_SYNCOOKIES_ADAPTIVE); 133 134 PF_RULES_WUNLOCK(); 135 } 136 137 void 138 pf_syncookies_cleanup(void) 139 { 140 callout_stop(&V_pf_syncookie_status.keytimeout); 141 } 142 143 int 144 pf_get_syncookies(struct pfioc_nv *nv) 145 { 146 nvlist_t *nvl = NULL; 147 void *nvlpacked = NULL; 148 int error; 149 150 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 151 152 nvl = nvlist_create(0); 153 if (nvl == NULL) 154 ERROUT(ENOMEM); 155 156 nvlist_add_bool(nvl, "enabled", 157 V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER); 158 nvlist_add_bool(nvl, "adaptive", 159 V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE); 160 nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat); 161 nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat); 162 nvlist_add_number(nvl, "halfopen_states", 163 atomic_load_32(&V_pf_status.states_halfopen)); 164 165 nvlpacked = nvlist_pack(nvl, &nv->len); 166 if (nvlpacked == NULL) 167 ERROUT(ENOMEM); 168 169 if (nv->size == 0) { 170 ERROUT(0); 171 } else if (nv->size < nv->len) { 172 ERROUT(ENOSPC); 173 } 174 175 error = copyout(nvlpacked, nv->data, nv->len); 176 177 #undef ERROUT 178 errout: 179 nvlist_destroy(nvl); 180 free(nvlpacked, M_NVLIST); 181 182 return (error); 183 } 184 185 int 186 pf_set_syncookies(struct pfioc_nv *nv) 187 { 188 nvlist_t *nvl = NULL; 189 void *nvlpacked = NULL; 190 int error; 191 bool enabled, adaptive; 192 uint32_t hiwat, lowat; 193 uint8_t newmode; 194 195 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 196 197 if (nv->len > pf_ioctl_maxcount) 198 return (ENOMEM); 199 200 nvlpacked = malloc(nv->len, M_NVLIST, M_WAITOK); 201 error = copyin(nv->data, nvlpacked, nv->len); 202 if (error) 203 ERROUT(error); 204 205 nvl = nvlist_unpack(nvlpacked, nv->len, 0); 206 if (nvl == NULL) 207 ERROUT(EBADMSG); 208 209 if (! nvlist_exists_bool(nvl, "enabled") 210 || ! nvlist_exists_bool(nvl, "adaptive")) 211 ERROUT(EBADMSG); 212 213 enabled = nvlist_get_bool(nvl, "enabled"); 214 adaptive = nvlist_get_bool(nvl, "adaptive"); 215 PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat, 216 V_pf_syncookie_status.hiwat)); 217 PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat, 218 V_pf_syncookie_status.lowat)); 219 220 if (lowat >= hiwat) 221 ERROUT(EINVAL); 222 223 newmode = PF_SYNCOOKIES_NEVER; 224 if (enabled) 225 newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS; 226 227 PF_RULES_WLOCK(); 228 error = pf_syncookies_setmode(newmode); 229 230 V_pf_syncookie_status.lowat = lowat; 231 V_pf_syncookie_status.hiwat = hiwat; 232 233 PF_RULES_WUNLOCK(); 234 235 #undef ERROUT 236 errout: 237 nvlist_destroy(nvl); 238 free(nvlpacked, M_NVLIST); 239 240 return (error); 241 } 242 243 static int 244 pf_syncookies_setmode(u_int8_t mode) 245 { 246 if (mode > PF_SYNCOOKIES_MODE_MAX) 247 return (EINVAL); 248 249 if (V_pf_status.syncookies_mode == mode) 250 return (0); 251 252 V_pf_status.syncookies_mode = mode; 253 if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) { 254 pf_syncookie_newkey(); 255 V_pf_status.syncookies_active = true; 256 } 257 return (0); 258 } 259 260 int 261 pf_synflood_check(struct pf_pdesc *pd) 262 { 263 MPASS(pd->proto == IPPROTO_TCP); 264 PF_RULES_RASSERT(); 265 266 if (pd->pf_mtag && (pd->pf_mtag->flags & PF_MTAG_FLAG_SYNCOOKIE_RECREATED)) 267 return (0); 268 269 if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE) 270 return (V_pf_status.syncookies_mode); 271 272 if (!V_pf_status.syncookies_active && 273 atomic_load_32(&V_pf_status.states_halfopen) > 274 V_pf_syncookie_status.hiwat) { 275 /* We'd want to 'pf_syncookie_newkey()' here, but that requires 276 * the rules write lock, which we can't get with the read lock 277 * held. */ 278 callout_reset(&V_pf_syncookie_status.keytimeout, 0, 279 pf_syncookie_rotate, curvnet); 280 V_pf_status.syncookies_active = true; 281 DPFPRINTF(LOG_WARNING, 282 "synflood detected, enabling syncookies"); 283 // XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++; 284 } 285 286 return (V_pf_status.syncookies_active); 287 } 288 289 void 290 pf_syncookie_send(struct pf_pdesc *pd) 291 { 292 uint16_t mss; 293 uint32_t iss; 294 295 mss = max(V_tcp_mssdflt, pf_get_mss(pd)); 296 iss = pf_syncookie_generate(pd, mss); 297 pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport, 298 iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss, 299 0, M_SKIP_FIREWALL | (pd->m->m_flags & M_LOOP), 0, 0, 300 pd->act.rtableid); 301 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1); 302 /* XXX Maybe only in adaptive mode? */ 303 atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 304 1); 305 } 306 307 bool 308 pf_syncookie_check(struct pf_pdesc *pd) 309 { 310 uint32_t hash, ack, seq; 311 union pf_syncookie cookie; 312 313 MPASS(pd->proto == IPPROTO_TCP); 314 PF_RULES_RASSERT(); 315 316 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 317 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 318 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 319 320 /* we don't know oddeven before setting the cookie (union) */ 321 if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven]) 322 == 0) 323 return (0); 324 325 hash = pf_syncookie_mac(pd, cookie, seq); 326 if ((ack & ~0xff) != (hash & ~0xff)) 327 return (false); 328 329 return (true); 330 } 331 332 uint8_t 333 pf_syncookie_validate(struct pf_pdesc *pd) 334 { 335 uint32_t ack; 336 union pf_syncookie cookie; 337 338 if (! pf_syncookie_check(pd)) 339 return (0); 340 341 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 342 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 343 344 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1); 345 atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1); 346 347 return (1); 348 } 349 350 /* 351 * all following functions private 352 */ 353 void 354 pf_syncookie_rotate(void *arg) 355 { 356 CURVNET_SET((struct vnet *)arg); 357 358 /* do we want to disable syncookies? */ 359 if (V_pf_status.syncookies_active && 360 ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE && 361 (atomic_load_32(&V_pf_status.states_halfopen) + 362 atomic_load_64(&V_pf_status.syncookies_inflight[0]) + 363 atomic_load_64(&V_pf_status.syncookies_inflight[1])) < 364 V_pf_syncookie_status.lowat) || 365 V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER) 366 ) { 367 V_pf_status.syncookies_active = false; 368 DPFPRINTF(PF_DEBUG_MISC, "syncookies disabled"); 369 } 370 371 /* nothing in flight any more? delete keys and return */ 372 if (!V_pf_status.syncookies_active && 373 atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 && 374 atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) { 375 memset(V_pf_syncookie_status.key[0], 0, 376 PF_SYNCOOKIE_SECRET_SIZE); 377 memset(V_pf_syncookie_status.key[1], 0, 378 PF_SYNCOOKIE_SECRET_SIZE); 379 CURVNET_RESTORE(); 380 return; 381 } 382 383 PF_RULES_WLOCK(); 384 /* new key, including timeout */ 385 pf_syncookie_newkey(); 386 PF_RULES_WUNLOCK(); 387 388 CURVNET_RESTORE(); 389 } 390 391 void 392 pf_syncookie_newkey(void) 393 { 394 PF_RULES_WASSERT(); 395 396 MPASS(V_pf_syncookie_status.oddeven < 2); 397 V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1; 398 atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0); 399 arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven], 400 PF_SYNCOOKIE_SECRET_SIZE); 401 callout_reset(&V_pf_syncookie_status.keytimeout, 402 PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet); 403 } 404 405 /* 406 * Distribution and probability of certain MSS values. Those in between are 407 * rounded down to the next lower one. 408 * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011] 409 * .2% .3% 5% 7% 7% 20% 15% 45% 410 */ 411 static int pf_syncookie_msstab[] = 412 { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 }; 413 414 /* 415 * Distribution and probability of certain WSCALE values. 416 * The absence of the WSCALE option is encoded with index zero. 417 * [WSCALE values histograms, Allman, 2012] 418 * X 10 10 35 5 6 14 10% by host 419 * X 11 4 5 5 18 49 3% by connections 420 */ 421 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 }; 422 423 uint32_t 424 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq) 425 { 426 SIPHASH_CTX ctx; 427 uint32_t siphash[2]; 428 429 PF_RULES_RASSERT(); 430 MPASS(pd->proto == IPPROTO_TCP); 431 432 SipHash24_Init(&ctx); 433 SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]); 434 435 switch (pd->af) { 436 case AF_INET: 437 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4)); 438 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4)); 439 break; 440 case AF_INET6: 441 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6)); 442 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6)); 443 break; 444 default: 445 panic("unknown address family"); 446 } 447 448 SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport)); 449 SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport)); 450 SipHash_Update(&ctx, &seq, sizeof(seq)); 451 SipHash_Update(&ctx, &cookie, sizeof(cookie)); 452 SipHash_Final((uint8_t *)&siphash, &ctx); 453 454 return (siphash[0] ^ siphash[1]); 455 } 456 457 uint32_t 458 pf_syncookie_generate(struct pf_pdesc *pd, uint16_t mss) 459 { 460 uint8_t i, wscale; 461 uint32_t iss, hash; 462 union pf_syncookie cookie; 463 464 PF_RULES_RASSERT(); 465 466 cookie.cookie = 0; 467 468 /* map MSS */ 469 for (i = nitems(pf_syncookie_msstab) - 1; 470 pf_syncookie_msstab[i] > mss && i > 0; i--) 471 /* nada */; 472 cookie.flags.mss_idx = i; 473 474 /* map WSCALE */ 475 wscale = pf_get_wscale(pd); 476 for (i = nitems(pf_syncookie_wstab) - 1; 477 pf_syncookie_wstab[i] > wscale && i > 0; i--) 478 /* nada */; 479 cookie.flags.wscale_idx = i; 480 cookie.flags.sack_ok = 0; /* XXX */ 481 482 cookie.flags.oddeven = V_pf_syncookie_status.oddeven; 483 hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq)); 484 485 /* 486 * Put the flags into the hash and XOR them to get better ISS number 487 * variance. This doesn't enhance the cryptographic strength and is 488 * done to prevent the 8 cookie bits from showing up directly on the 489 * wire. 490 */ 491 iss = hash & ~0xff; 492 iss |= cookie.cookie ^ (hash >> 24); 493 494 return (iss); 495 } 496 497 struct mbuf * 498 pf_syncookie_recreate_syn(struct pf_pdesc *pd) 499 { 500 uint8_t wscale; 501 uint16_t mss; 502 uint32_t ack, seq; 503 union pf_syncookie cookie; 504 505 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 506 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 507 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 508 509 if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) || 510 cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab)) 511 return (NULL); 512 513 mss = pf_syncookie_msstab[cookie.flags.mss_idx]; 514 wscale = pf_syncookie_wstab[cookie.flags.wscale_idx]; 515 516 return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport, 517 *pd->dport, seq, 0, TH_SYN, wscale, mss, pd->ttl, 518 (pd->m->m_flags & M_LOOP), 0, PF_MTAG_FLAG_SYNCOOKIE_RECREATED, 519 cookie.flags.sack_ok, pd->act.rtableid)); 520 } 521