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 pf_pdesc *, uint16_t); 123 124 void 125 pf_syncookies_init(void) 126 { 127 callout_init(&V_pf_syncookie_status.keytimeout, 1); 128 PF_RULES_WLOCK(); 129 130 V_pf_syncookie_status.hiwat = PF_SYNCOOKIES_HIWATPCT * 131 V_pf_limits[PF_LIMIT_STATES].limit / 100; 132 V_pf_syncookie_status.lowat = PF_SYNCOOKIES_LOWATPCT * 133 V_pf_limits[PF_LIMIT_STATES].limit / 100; 134 pf_syncookies_setmode(PF_SYNCOOKIES_ADAPTIVE); 135 136 PF_RULES_WUNLOCK(); 137 } 138 139 void 140 pf_syncookies_cleanup(void) 141 { 142 callout_stop(&V_pf_syncookie_status.keytimeout); 143 } 144 145 int 146 pf_get_syncookies(struct pfioc_nv *nv) 147 { 148 nvlist_t *nvl = NULL; 149 void *nvlpacked = NULL; 150 int error; 151 152 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 153 154 nvl = nvlist_create(0); 155 if (nvl == NULL) 156 ERROUT(ENOMEM); 157 158 nvlist_add_bool(nvl, "enabled", 159 V_pf_status.syncookies_mode != PF_SYNCOOKIES_NEVER); 160 nvlist_add_bool(nvl, "adaptive", 161 V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE); 162 nvlist_add_number(nvl, "highwater", V_pf_syncookie_status.hiwat); 163 nvlist_add_number(nvl, "lowwater", V_pf_syncookie_status.lowat); 164 nvlist_add_number(nvl, "halfopen_states", 165 atomic_load_32(&V_pf_status.states_halfopen)); 166 167 nvlpacked = nvlist_pack(nvl, &nv->len); 168 if (nvlpacked == NULL) 169 ERROUT(ENOMEM); 170 171 if (nv->size == 0) { 172 ERROUT(0); 173 } else if (nv->size < nv->len) { 174 ERROUT(ENOSPC); 175 } 176 177 error = copyout(nvlpacked, nv->data, nv->len); 178 179 #undef ERROUT 180 errout: 181 nvlist_destroy(nvl); 182 free(nvlpacked, M_NVLIST); 183 184 return (error); 185 } 186 187 int 188 pf_set_syncookies(struct pfioc_nv *nv) 189 { 190 nvlist_t *nvl = NULL; 191 void *nvlpacked = NULL; 192 int error; 193 bool enabled, adaptive; 194 uint32_t hiwat, lowat; 195 uint8_t newmode; 196 197 #define ERROUT(x) ERROUT_FUNCTION(errout, x) 198 199 if (nv->len > pf_ioctl_maxcount) 200 return (ENOMEM); 201 202 nvlpacked = malloc(nv->len, M_NVLIST, M_WAITOK); 203 error = copyin(nv->data, nvlpacked, nv->len); 204 if (error) 205 ERROUT(error); 206 207 nvl = nvlist_unpack(nvlpacked, nv->len, 0); 208 if (nvl == NULL) 209 ERROUT(EBADMSG); 210 211 if (! nvlist_exists_bool(nvl, "enabled") 212 || ! nvlist_exists_bool(nvl, "adaptive")) 213 ERROUT(EBADMSG); 214 215 enabled = nvlist_get_bool(nvl, "enabled"); 216 adaptive = nvlist_get_bool(nvl, "adaptive"); 217 PFNV_CHK(pf_nvuint32_opt(nvl, "highwater", &hiwat, 218 V_pf_syncookie_status.hiwat)); 219 PFNV_CHK(pf_nvuint32_opt(nvl, "lowwater", &lowat, 220 V_pf_syncookie_status.lowat)); 221 222 if (lowat >= hiwat) 223 ERROUT(EINVAL); 224 225 newmode = PF_SYNCOOKIES_NEVER; 226 if (enabled) 227 newmode = adaptive ? PF_SYNCOOKIES_ADAPTIVE : PF_SYNCOOKIES_ALWAYS; 228 229 PF_RULES_WLOCK(); 230 error = pf_syncookies_setmode(newmode); 231 232 V_pf_syncookie_status.lowat = lowat; 233 V_pf_syncookie_status.hiwat = hiwat; 234 235 PF_RULES_WUNLOCK(); 236 237 #undef ERROUT 238 errout: 239 nvlist_destroy(nvl); 240 free(nvlpacked, M_NVLIST); 241 242 return (error); 243 } 244 245 static int 246 pf_syncookies_setmode(u_int8_t mode) 247 { 248 if (mode > PF_SYNCOOKIES_MODE_MAX) 249 return (EINVAL); 250 251 if (V_pf_status.syncookies_mode == mode) 252 return (0); 253 254 V_pf_status.syncookies_mode = mode; 255 if (V_pf_status.syncookies_mode == PF_SYNCOOKIES_ALWAYS) { 256 pf_syncookie_newkey(); 257 V_pf_status.syncookies_active = true; 258 } 259 return (0); 260 } 261 262 int 263 pf_synflood_check(struct pf_pdesc *pd) 264 { 265 MPASS(pd->proto == IPPROTO_TCP); 266 PF_RULES_RASSERT(); 267 268 if (pd->pf_mtag && (pd->pf_mtag->flags & PF_MTAG_FLAG_SYNCOOKIE_RECREATED)) 269 return (0); 270 271 if (V_pf_status.syncookies_mode != PF_SYNCOOKIES_ADAPTIVE) 272 return (V_pf_status.syncookies_mode); 273 274 if (!V_pf_status.syncookies_active && 275 atomic_load_32(&V_pf_status.states_halfopen) > 276 V_pf_syncookie_status.hiwat) { 277 /* We'd want to 'pf_syncookie_newkey()' here, but that requires 278 * the rules write lock, which we can't get with the read lock 279 * held. */ 280 callout_reset(&V_pf_syncookie_status.keytimeout, 0, 281 pf_syncookie_rotate, curvnet); 282 V_pf_status.syncookies_active = true; 283 DPFPRINTF(LOG_WARNING, 284 ("synflood detected, enabling syncookies\n")); 285 // XXXTODO V_pf_status.lcounters[LCNT_SYNFLOODS]++; 286 } 287 288 return (V_pf_status.syncookies_active); 289 } 290 291 void 292 pf_syncookie_send(struct pf_pdesc *pd) 293 { 294 uint16_t mss; 295 uint32_t iss; 296 297 mss = max(V_tcp_mssdflt, pf_get_mss(pd)); 298 iss = pf_syncookie_generate(pd, mss); 299 pf_send_tcp(NULL, pd->af, pd->dst, pd->src, *pd->dport, *pd->sport, 300 iss, ntohl(pd->hdr.tcp.th_seq) + 1, TH_SYN|TH_ACK, 0, mss, 301 0, M_SKIP_FIREWALL | (pd->m->m_flags & M_LOOP), 0, 0, 302 pd->act.rtableid); 303 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_SENT], 1); 304 /* XXX Maybe only in adaptive mode? */ 305 atomic_add_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 306 1); 307 } 308 309 bool 310 pf_syncookie_check(struct pf_pdesc *pd) 311 { 312 uint32_t hash, ack, seq; 313 union pf_syncookie cookie; 314 315 MPASS(pd->proto == IPPROTO_TCP); 316 PF_RULES_RASSERT(); 317 318 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 319 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 320 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 321 322 /* we don't know oddeven before setting the cookie (union) */ 323 if (atomic_load_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven]) 324 == 0) 325 return (0); 326 327 hash = pf_syncookie_mac(pd, cookie, seq); 328 if ((ack & ~0xff) != (hash & ~0xff)) 329 return (false); 330 331 return (true); 332 } 333 334 uint8_t 335 pf_syncookie_validate(struct pf_pdesc *pd) 336 { 337 uint32_t ack; 338 union pf_syncookie cookie; 339 340 if (! pf_syncookie_check(pd)) 341 return (0); 342 343 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 344 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 345 346 counter_u64_add(V_pf_status.lcounters[KLCNT_SYNCOOKIES_VALID], 1); 347 atomic_add_64(&V_pf_status.syncookies_inflight[cookie.flags.oddeven], -1); 348 349 return (1); 350 } 351 352 /* 353 * all following functions private 354 */ 355 void 356 pf_syncookie_rotate(void *arg) 357 { 358 CURVNET_SET((struct vnet *)arg); 359 360 /* do we want to disable syncookies? */ 361 if (V_pf_status.syncookies_active && 362 ((V_pf_status.syncookies_mode == PF_SYNCOOKIES_ADAPTIVE && 363 (atomic_load_32(&V_pf_status.states_halfopen) + 364 atomic_load_64(&V_pf_status.syncookies_inflight[0]) + 365 atomic_load_64(&V_pf_status.syncookies_inflight[1])) < 366 V_pf_syncookie_status.lowat) || 367 V_pf_status.syncookies_mode == PF_SYNCOOKIES_NEVER) 368 ) { 369 V_pf_status.syncookies_active = false; 370 DPFPRINTF(PF_DEBUG_MISC, ("syncookies disabled\n")); 371 } 372 373 /* nothing in flight any more? delete keys and return */ 374 if (!V_pf_status.syncookies_active && 375 atomic_load_64(&V_pf_status.syncookies_inflight[0]) == 0 && 376 atomic_load_64(&V_pf_status.syncookies_inflight[1]) == 0) { 377 memset(V_pf_syncookie_status.key[0], 0, 378 PF_SYNCOOKIE_SECRET_SIZE); 379 memset(V_pf_syncookie_status.key[1], 0, 380 PF_SYNCOOKIE_SECRET_SIZE); 381 CURVNET_RESTORE(); 382 return; 383 } 384 385 PF_RULES_WLOCK(); 386 /* new key, including timeout */ 387 pf_syncookie_newkey(); 388 PF_RULES_WUNLOCK(); 389 390 CURVNET_RESTORE(); 391 } 392 393 void 394 pf_syncookie_newkey(void) 395 { 396 PF_RULES_WASSERT(); 397 398 MPASS(V_pf_syncookie_status.oddeven < 2); 399 V_pf_syncookie_status.oddeven = (V_pf_syncookie_status.oddeven + 1) & 0x1; 400 atomic_store_64(&V_pf_status.syncookies_inflight[V_pf_syncookie_status.oddeven], 0); 401 arc4random_buf(V_pf_syncookie_status.key[V_pf_syncookie_status.oddeven], 402 PF_SYNCOOKIE_SECRET_SIZE); 403 callout_reset(&V_pf_syncookie_status.keytimeout, 404 PF_SYNCOOKIE_SECRET_LIFETIME * hz, pf_syncookie_rotate, curvnet); 405 } 406 407 /* 408 * Distribution and probability of certain MSS values. Those in between are 409 * rounded down to the next lower one. 410 * [An Analysis of TCP Maximum Segment Sizes, S. Alcock and R. Nelson, 2011] 411 * .2% .3% 5% 7% 7% 20% 15% 45% 412 */ 413 static int pf_syncookie_msstab[] = 414 { 216, 536, 1200, 1360, 1400, 1440, 1452, 1460 }; 415 416 /* 417 * Distribution and probability of certain WSCALE values. 418 * The absence of the WSCALE option is encoded with index zero. 419 * [WSCALE values histograms, Allman, 2012] 420 * X 10 10 35 5 6 14 10% by host 421 * X 11 4 5 5 18 49 3% by connections 422 */ 423 static int pf_syncookie_wstab[] = { 0, 0, 1, 2, 4, 6, 7, 8 }; 424 425 uint32_t 426 pf_syncookie_mac(struct pf_pdesc *pd, union pf_syncookie cookie, uint32_t seq) 427 { 428 SIPHASH_CTX ctx; 429 uint32_t siphash[2]; 430 431 PF_RULES_RASSERT(); 432 MPASS(pd->proto == IPPROTO_TCP); 433 434 SipHash24_Init(&ctx); 435 SipHash_SetKey(&ctx, V_pf_syncookie_status.key[cookie.flags.oddeven]); 436 437 switch (pd->af) { 438 case AF_INET: 439 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v4)); 440 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v4)); 441 break; 442 case AF_INET6: 443 SipHash_Update(&ctx, pd->src, sizeof(pd->src->v6)); 444 SipHash_Update(&ctx, pd->dst, sizeof(pd->dst->v6)); 445 break; 446 default: 447 panic("unknown address family"); 448 } 449 450 SipHash_Update(&ctx, pd->sport, sizeof(*pd->sport)); 451 SipHash_Update(&ctx, pd->dport, sizeof(*pd->dport)); 452 SipHash_Update(&ctx, &seq, sizeof(seq)); 453 SipHash_Update(&ctx, &cookie, sizeof(cookie)); 454 SipHash_Final((uint8_t *)&siphash, &ctx); 455 456 return (siphash[0] ^ siphash[1]); 457 } 458 459 uint32_t 460 pf_syncookie_generate(struct pf_pdesc *pd, uint16_t mss) 461 { 462 uint8_t i, wscale; 463 uint32_t iss, hash; 464 union pf_syncookie cookie; 465 466 PF_RULES_RASSERT(); 467 468 cookie.cookie = 0; 469 470 /* map MSS */ 471 for (i = nitems(pf_syncookie_msstab) - 1; 472 pf_syncookie_msstab[i] > mss && i > 0; i--) 473 /* nada */; 474 cookie.flags.mss_idx = i; 475 476 /* map WSCALE */ 477 wscale = pf_get_wscale(pd); 478 for (i = nitems(pf_syncookie_wstab) - 1; 479 pf_syncookie_wstab[i] > wscale && i > 0; i--) 480 /* nada */; 481 cookie.flags.wscale_idx = i; 482 cookie.flags.sack_ok = 0; /* XXX */ 483 484 cookie.flags.oddeven = V_pf_syncookie_status.oddeven; 485 hash = pf_syncookie_mac(pd, cookie, ntohl(pd->hdr.tcp.th_seq)); 486 487 /* 488 * Put the flags into the hash and XOR them to get better ISS number 489 * variance. This doesn't enhance the cryptographic strength and is 490 * done to prevent the 8 cookie bits from showing up directly on the 491 * wire. 492 */ 493 iss = hash & ~0xff; 494 iss |= cookie.cookie ^ (hash >> 24); 495 496 return (iss); 497 } 498 499 struct mbuf * 500 pf_syncookie_recreate_syn(struct pf_pdesc *pd) 501 { 502 uint8_t wscale; 503 uint16_t mss; 504 uint32_t ack, seq; 505 union pf_syncookie cookie; 506 507 seq = ntohl(pd->hdr.tcp.th_seq) - 1; 508 ack = ntohl(pd->hdr.tcp.th_ack) - 1; 509 cookie.cookie = (ack & 0xff) ^ (ack >> 24); 510 511 if (cookie.flags.mss_idx >= nitems(pf_syncookie_msstab) || 512 cookie.flags.wscale_idx >= nitems(pf_syncookie_wstab)) 513 return (NULL); 514 515 mss = pf_syncookie_msstab[cookie.flags.mss_idx]; 516 wscale = pf_syncookie_wstab[cookie.flags.wscale_idx]; 517 518 return (pf_build_tcp(NULL, pd->af, pd->src, pd->dst, *pd->sport, 519 *pd->dport, seq, 0, TH_SYN, wscale, mss, pd->ttl, 520 (pd->m->m_flags & M_LOOP), 0, PF_MTAG_FLAG_SYNCOOKIE_RECREATED, 521 pd->act.rtableid)); 522 } 523