1/*- 2 * Copyright (c) 2015 Patrick Kelsey 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27/* 28 * This is a server-side implementation of TCP Fast Open (TFO) [RFC7413]. 29 * 30 * This implementation is currently considered to be experimental and is not 31 * included in kernel builds by default. To include this code, add the 32 * following line to your kernel config: 33 * 34 * options TCP_RFC7413 35 * 36 * The generated TFO cookies are the 64-bit output of 37 * SipHash24(<16-byte-key><client-ip>). Multiple concurrent valid keys are 38 * supported so that time-based rolling cookie invalidation policies can be 39 * implemented in the system. The default number of concurrent keys is 2. 40 * This can be adjusted in the kernel config as follows: 41 * 42 * options TCP_RFC7413_MAX_KEYS=<num-keys> 43 * 44 * 45 * The following TFO-specific sysctls are defined: 46 * 47 * net.inet.tcp.fastopen.acceptany (RW, default 0) 48 * When non-zero, all client-supplied TFO cookies will be considered to 49 * be valid. 50 * 51 * net.inet.tcp.fastopen.autokey (RW, default 120) 52 * When this and net.inet.tcp.fastopen.enabled are non-zero, a new key 53 * will be automatically generated after this many seconds. 54 * 55 * net.inet.tcp.fastopen.enabled (RW, default 0) 56 * When zero, no new TFO connections can be created. On the transition 57 * from enabled to disabled, all installed keys are removed. On the 58 * transition from disabled to enabled, if net.inet.tcp.fastopen.autokey 59 * is non-zero and there are no keys installed, a new key will be 60 * generated immediately. The transition from enabled to disabled does 61 * not affect any TFO connections in progress; it only prevents new ones 62 * from being made. 63 * 64 * net.inet.tcp.fastopen.keylen (RO) 65 * The key length in bytes. 66 * 67 * net.inet.tcp.fastopen.maxkeys (RO) 68 * The maximum number of keys supported. 69 * 70 * net.inet.tcp.fastopen.numkeys (RO) 71 * The current number of keys installed. 72 * 73 * net.inet.tcp.fastopen.setkey (WO) 74 * Install a new key by writing net.inet.tcp.fastopen.keylen bytes to this 75 * sysctl. 76 * 77 * 78 * In order for TFO connections to be created via a listen socket, that 79 * socket must have the TCP_FASTOPEN socket option set on it. This option 80 * can be set on the socket either before or after the listen() is invoked. 81 * Clearing this option on a listen socket after it has been set has no 82 * effect on existing TFO connections or TFO connections in progress; it 83 * only prevents new TFO connections from being made. 84 * 85 * For passively-created sockets, the TCP_FASTOPEN socket option can be 86 * queried to determine whether the connection was established using TFO. 87 * Note that connections that are established via a TFO SYN, but that fall 88 * back to using a non-TFO SYN|ACK will have the TCP_FASTOPEN socket option 89 * set. 90 * 91 * Per the RFC, this implementation limits the number of TFO connections 92 * that can be in the SYN_RECEIVED state on a per listen-socket basis. 93 * Whenever this limit is exceeded, requests for new TFO connections are 94 * serviced as non-TFO requests. Without such a limit, given a valid TFO 95 * cookie, an attacker could keep the listen queue in an overflow condition 96 * using a TFO SYN flood. This implementation sets the limit at half the 97 * configured listen backlog. 98 * 99 */ 100 101#include <sys/cdefs.h> 102__FBSDID("$FreeBSD: releng/11.0/sys/netinet/tcp_fastopen.c 304088 2016-08-14 18:30:30Z karels $"); 103 104#include "opt_inet.h" 105 106#include <sys/param.h> 107#include <sys/kernel.h> 108#include <sys/limits.h> 109#include <sys/lock.h> 110#include <sys/rmlock.h> 111#include <sys/socket.h> 112#include <sys/socketvar.h> 113#include <sys/sysctl.h> 114#include <sys/systm.h> 115 116#include <crypto/siphash/siphash.h> 117 118#include <net/vnet.h> 119 120#include <netinet/in.h> 121#include <netinet/in_pcb.h> 122#include <netinet/tcp_fastopen.h> 123#include <netinet/tcp_var.h> 124 125 126#define TCP_FASTOPEN_KEY_LEN SIPHASH_KEY_LENGTH 127 128#if !defined(TCP_RFC7413_MAX_KEYS) || (TCP_RFC7413_MAX_KEYS < 1) 129#define TCP_FASTOPEN_MAX_KEYS 2 130#else 131#define TCP_FASTOPEN_MAX_KEYS TCP_RFC7413_MAX_KEYS 132#endif 133 134struct tcp_fastopen_keylist { 135 unsigned int newest; 136 uint8_t key[TCP_FASTOPEN_MAX_KEYS][TCP_FASTOPEN_KEY_LEN]; 137}; 138 139struct tcp_fastopen_callout { 140 struct callout c; 141 struct vnet *v; 142}; 143 144SYSCTL_NODE(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW, 0, "TCP Fast Open"); 145 146static VNET_DEFINE(int, tcp_fastopen_acceptany) = 0; 147#define V_tcp_fastopen_acceptany VNET(tcp_fastopen_acceptany) 148SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, acceptany, 149 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_fastopen_acceptany), 0, 150 "Accept any non-empty cookie"); 151 152static VNET_DEFINE(unsigned int, tcp_fastopen_autokey) = 120; 153#define V_tcp_fastopen_autokey VNET(tcp_fastopen_autokey) 154static int sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS); 155SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, autokey, 156 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, 157 &sysctl_net_inet_tcp_fastopen_autokey, "IU", 158 "Number of seconds between auto-generation of a new key; zero disables"); 159 160VNET_DEFINE(unsigned int, tcp_fastopen_enabled) = 0; 161static int sysctl_net_inet_tcp_fastopen_enabled(SYSCTL_HANDLER_ARGS); 162SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, enabled, 163 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, 164 &sysctl_net_inet_tcp_fastopen_enabled, "IU", 165 "Enable/disable TCP Fast Open processing"); 166 167SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, keylen, 168 CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_KEY_LEN, 169 "Key length in bytes"); 170 171SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, maxkeys, 172 CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_MAX_KEYS, 173 "Maximum number of keys supported"); 174 175static VNET_DEFINE(unsigned int, tcp_fastopen_numkeys) = 0; 176#define V_tcp_fastopen_numkeys VNET(tcp_fastopen_numkeys) 177SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, numkeys, 178 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_fastopen_numkeys), 0, 179 "Number of keys installed"); 180 181static int sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS); 182SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, setkey, 183 CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, 184 &sysctl_net_inet_tcp_fastopen_setkey, "", 185 "Install a new key"); 186 187static VNET_DEFINE(struct rmlock, tcp_fastopen_keylock); 188#define V_tcp_fastopen_keylock VNET(tcp_fastopen_keylock) 189 190#define TCP_FASTOPEN_KEYS_RLOCK(t) rm_rlock(&V_tcp_fastopen_keylock, (t)) 191#define TCP_FASTOPEN_KEYS_RUNLOCK(t) rm_runlock(&V_tcp_fastopen_keylock, (t)) 192#define TCP_FASTOPEN_KEYS_WLOCK() rm_wlock(&V_tcp_fastopen_keylock) 193#define TCP_FASTOPEN_KEYS_WUNLOCK() rm_wunlock(&V_tcp_fastopen_keylock) 194 195static VNET_DEFINE(struct tcp_fastopen_keylist, tcp_fastopen_keys); 196#define V_tcp_fastopen_keys VNET(tcp_fastopen_keys) 197 198static VNET_DEFINE(struct tcp_fastopen_callout, tcp_fastopen_autokey_ctx); 199#define V_tcp_fastopen_autokey_ctx VNET(tcp_fastopen_autokey_ctx) 200 201static VNET_DEFINE(uma_zone_t, counter_zone); 202#define V_counter_zone VNET(counter_zone) 203 204void 205tcp_fastopen_init(void) 206{ 207 V_counter_zone = uma_zcreate("tfo", sizeof(unsigned int), 208 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 209 rm_init(&V_tcp_fastopen_keylock, "tfo_keylock"); 210 callout_init_rm(&V_tcp_fastopen_autokey_ctx.c, 211 &V_tcp_fastopen_keylock, 0); 212 V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; 213} 214 215void 216tcp_fastopen_destroy(void) 217{ 218 callout_drain(&V_tcp_fastopen_autokey_ctx.c); 219 rm_destroy(&V_tcp_fastopen_keylock); 220 uma_zdestroy(V_counter_zone); 221} 222 223unsigned int * 224tcp_fastopen_alloc_counter(void) 225{ 226 unsigned int *counter; 227 counter = uma_zalloc(V_counter_zone, M_NOWAIT); 228 if (counter) 229 *counter = 1; 230 return (counter); 231} 232 233void 234tcp_fastopen_decrement_counter(unsigned int *counter) 235{ 236 if (*counter == 1) 237 uma_zfree(V_counter_zone, counter); 238 else 239 atomic_subtract_int(counter, 1); 240} 241 242static void 243tcp_fastopen_addkey_locked(uint8_t *key) 244{ 245 246 V_tcp_fastopen_keys.newest++; 247 if (V_tcp_fastopen_keys.newest == TCP_FASTOPEN_MAX_KEYS) 248 V_tcp_fastopen_keys.newest = 0; 249 memcpy(V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], key, 250 TCP_FASTOPEN_KEY_LEN); 251 if (V_tcp_fastopen_numkeys < TCP_FASTOPEN_MAX_KEYS) 252 V_tcp_fastopen_numkeys++; 253} 254 255static void 256tcp_fastopen_autokey_locked(void) 257{ 258 uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; 259 260 arc4rand(newkey, TCP_FASTOPEN_KEY_LEN, 0); 261 tcp_fastopen_addkey_locked(newkey); 262} 263 264static void 265tcp_fastopen_autokey_callout(void *arg) 266{ 267 struct tcp_fastopen_callout *ctx = arg; 268 269 CURVNET_SET(ctx->v); 270 tcp_fastopen_autokey_locked(); 271 callout_reset(&ctx->c, V_tcp_fastopen_autokey * hz, 272 tcp_fastopen_autokey_callout, ctx); 273 CURVNET_RESTORE(); 274} 275 276 277static uint64_t 278tcp_fastopen_make_cookie(uint8_t key[SIPHASH_KEY_LENGTH], struct in_conninfo *inc) 279{ 280 SIPHASH_CTX ctx; 281 uint64_t siphash; 282 283 SipHash24_Init(&ctx); 284 SipHash_SetKey(&ctx, key); 285 switch (inc->inc_flags & INC_ISIPV6) { 286#ifdef INET 287 case 0: 288 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(inc->inc_faddr)); 289 break; 290#endif 291#ifdef INET6 292 case INC_ISIPV6: 293 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(inc->inc6_faddr)); 294 break; 295#endif 296 } 297 SipHash_Final((u_int8_t *)&siphash, &ctx); 298 299 return (siphash); 300} 301 302 303/* 304 * Return values: 305 * -1 the cookie is invalid and no valid cookie is available 306 * 0 the cookie is invalid and the latest cookie has been returned 307 * 1 the cookie is valid and the latest cookie has been returned 308 */ 309int 310tcp_fastopen_check_cookie(struct in_conninfo *inc, uint8_t *cookie, 311 unsigned int len, uint64_t *latest_cookie) 312{ 313 struct rm_priotracker tracker; 314 unsigned int i, key_index; 315 uint64_t cur_cookie; 316 317 if (V_tcp_fastopen_acceptany) { 318 *latest_cookie = 0; 319 return (1); 320 } 321 322 if (len != TCP_FASTOPEN_COOKIE_LEN) { 323 if (V_tcp_fastopen_numkeys > 0) { 324 *latest_cookie = 325 tcp_fastopen_make_cookie( 326 V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], 327 inc); 328 return (0); 329 } 330 return (-1); 331 } 332 333 /* 334 * Check against each available key, from newest to oldest. 335 */ 336 TCP_FASTOPEN_KEYS_RLOCK(&tracker); 337 key_index = V_tcp_fastopen_keys.newest; 338 for (i = 0; i < V_tcp_fastopen_numkeys; i++) { 339 cur_cookie = 340 tcp_fastopen_make_cookie(V_tcp_fastopen_keys.key[key_index], 341 inc); 342 if (i == 0) 343 *latest_cookie = cur_cookie; 344 if (memcmp(cookie, &cur_cookie, TCP_FASTOPEN_COOKIE_LEN) == 0) { 345 TCP_FASTOPEN_KEYS_RUNLOCK(&tracker); 346 return (1); 347 } 348 if (key_index == 0) 349 key_index = TCP_FASTOPEN_MAX_KEYS - 1; 350 else 351 key_index--; 352 } 353 TCP_FASTOPEN_KEYS_RUNLOCK(&tracker); 354 355 return (0); 356} 357 358static int 359sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS) 360{ 361 int error; 362 unsigned int new; 363 364 new = V_tcp_fastopen_autokey; 365 error = sysctl_handle_int(oidp, &new, 0, req); 366 if (error == 0 && req->newptr) { 367 if (new > (INT_MAX / hz)) 368 return (EINVAL); 369 370 TCP_FASTOPEN_KEYS_WLOCK(); 371 if (V_tcp_fastopen_enabled) { 372 if (V_tcp_fastopen_autokey && !new) 373 callout_stop(&V_tcp_fastopen_autokey_ctx.c); 374 else if (new) 375 callout_reset(&V_tcp_fastopen_autokey_ctx.c, 376 new * hz, tcp_fastopen_autokey_callout, 377 &V_tcp_fastopen_autokey_ctx); 378 } 379 V_tcp_fastopen_autokey = new; 380 TCP_FASTOPEN_KEYS_WUNLOCK(); 381 } 382 383 return (error); 384} 385 386static int 387sysctl_net_inet_tcp_fastopen_enabled(SYSCTL_HANDLER_ARGS) 388{ 389 int error; 390 unsigned int new; 391 392 new = V_tcp_fastopen_enabled; 393 error = sysctl_handle_int(oidp, &new, 0, req); 394 if (error == 0 && req->newptr) { 395 if (V_tcp_fastopen_enabled && !new) { 396 /* enabled -> disabled */ 397 TCP_FASTOPEN_KEYS_WLOCK(); 398 V_tcp_fastopen_numkeys = 0; 399 V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; 400 if (V_tcp_fastopen_autokey) 401 callout_stop(&V_tcp_fastopen_autokey_ctx.c); 402 V_tcp_fastopen_enabled = 0; 403 TCP_FASTOPEN_KEYS_WUNLOCK(); 404 } else if (!V_tcp_fastopen_enabled && new) { 405 /* disabled -> enabled */ 406 TCP_FASTOPEN_KEYS_WLOCK(); 407 if (V_tcp_fastopen_autokey && 408 (V_tcp_fastopen_numkeys == 0)) { 409 tcp_fastopen_autokey_locked(); 410 callout_reset(&V_tcp_fastopen_autokey_ctx.c, 411 V_tcp_fastopen_autokey * hz, 412 tcp_fastopen_autokey_callout, 413 &V_tcp_fastopen_autokey_ctx); 414 } 415 V_tcp_fastopen_enabled = 1; 416 TCP_FASTOPEN_KEYS_WUNLOCK(); 417 } 418 } 419 return (error); 420} 421 422static int 423sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS) 424{ 425 int error; 426 uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; 427 428 if (req->oldptr != NULL || req->oldlen != 0) 429 return (EINVAL); 430 if (req->newptr == NULL) 431 return (EPERM); 432 if (req->newlen != sizeof(newkey)) 433 return (EINVAL); 434 error = SYSCTL_IN(req, newkey, sizeof(newkey)); 435 if (error) 436 return (error); 437 438 TCP_FASTOPEN_KEYS_WLOCK(); 439 tcp_fastopen_addkey_locked(newkey); 440 TCP_FASTOPEN_KEYS_WUNLOCK(); 441 442 return (0); 443} 444