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("$FreeBSD$"); 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/socketvar.h> 112#include <sys/sysctl.h> 113#include <sys/systm.h> 114 115#include <crypto/siphash/siphash.h> 116 117#include <net/vnet.h> 118 119#include <netinet/in.h> 120#include <netinet/in_pcb.h> 121#include <netinet/tcp_fastopen.h> 122#include <netinet/tcp_var.h> 123 124 125#define TCP_FASTOPEN_KEY_LEN SIPHASH_KEY_LENGTH 126 127#if !defined(TCP_RFC7413_MAX_KEYS) || (TCP_RFC7413_MAX_KEYS < 1) 128#define TCP_FASTOPEN_MAX_KEYS 2 129#else 130#define TCP_FASTOPEN_MAX_KEYS TCP_RFC7413_MAX_KEYS 131#endif 132 133struct tcp_fastopen_keylist { 134 unsigned int newest; 135 uint8_t key[TCP_FASTOPEN_MAX_KEYS][TCP_FASTOPEN_KEY_LEN]; 136}; 137 138struct tcp_fastopen_callout { 139 struct callout c; 140 struct vnet *v; 141}; 142 143SYSCTL_NODE(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW, 0, "TCP Fast Open"); 144 145static VNET_DEFINE(int, tcp_fastopen_acceptany) = 0; 146#define V_tcp_fastopen_acceptany VNET(tcp_fastopen_acceptany) 147SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, acceptany, 148 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_fastopen_acceptany), 0, 149 "Accept any non-empty cookie"); 150 151static VNET_DEFINE(unsigned int, tcp_fastopen_autokey) = 120; 152#define V_tcp_fastopen_autokey VNET(tcp_fastopen_autokey) 153static int sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS); 154SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, autokey, 155 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, 156 &sysctl_net_inet_tcp_fastopen_autokey, "IU", 157 "Number of seconds between auto-generation of a new key; zero disables"); 158 159VNET_DEFINE(unsigned int, tcp_fastopen_enabled) = 0; 160static int sysctl_net_inet_tcp_fastopen_enabled(SYSCTL_HANDLER_ARGS); 161SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, enabled, 162 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, NULL, 0, 163 &sysctl_net_inet_tcp_fastopen_enabled, "IU", 164 "Enable/disable TCP Fast Open processing"); 165 166SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, keylen, 167 CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_KEY_LEN, 168 "Key length in bytes"); 169 170SYSCTL_INT(_net_inet_tcp_fastopen, OID_AUTO, maxkeys, 171 CTLFLAG_RD, SYSCTL_NULL_INT_PTR, TCP_FASTOPEN_MAX_KEYS, 172 "Maximum number of keys supported"); 173 174static VNET_DEFINE(unsigned int, tcp_fastopen_numkeys) = 0; 175#define V_tcp_fastopen_numkeys VNET(tcp_fastopen_numkeys) 176SYSCTL_UINT(_net_inet_tcp_fastopen, OID_AUTO, numkeys, 177 CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_fastopen_numkeys), 0, 178 "Number of keys installed"); 179 180static int sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS); 181SYSCTL_PROC(_net_inet_tcp_fastopen, OID_AUTO, setkey, 182 CTLFLAG_VNET | CTLTYPE_OPAQUE | CTLFLAG_WR, NULL, 0, 183 &sysctl_net_inet_tcp_fastopen_setkey, "", 184 "Install a new key"); 185 186static VNET_DEFINE(struct rmlock, tcp_fastopen_keylock); 187#define V_tcp_fastopen_keylock VNET(tcp_fastopen_keylock) 188 189#define TCP_FASTOPEN_KEYS_RLOCK(t) rm_rlock(&V_tcp_fastopen_keylock, (t)) 190#define TCP_FASTOPEN_KEYS_RUNLOCK(t) rm_runlock(&V_tcp_fastopen_keylock, (t)) 191#define TCP_FASTOPEN_KEYS_WLOCK() rm_wlock(&V_tcp_fastopen_keylock) 192#define TCP_FASTOPEN_KEYS_WUNLOCK() rm_wunlock(&V_tcp_fastopen_keylock) 193 194static VNET_DEFINE(struct tcp_fastopen_keylist, tcp_fastopen_keys); 195#define V_tcp_fastopen_keys VNET(tcp_fastopen_keys) 196 197static VNET_DEFINE(struct tcp_fastopen_callout, tcp_fastopen_autokey_ctx); 198#define V_tcp_fastopen_autokey_ctx VNET(tcp_fastopen_autokey_ctx) 199 200static VNET_DEFINE(uma_zone_t, counter_zone); 201#define V_counter_zone VNET(counter_zone) 202 203void 204tcp_fastopen_init(void) 205{ 206 V_counter_zone = uma_zcreate("tfo", sizeof(unsigned int), 207 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 208 rm_init(&V_tcp_fastopen_keylock, "tfo_keylock"); 209 callout_init_rm(&V_tcp_fastopen_autokey_ctx.c, 210 &V_tcp_fastopen_keylock, 0); 211 V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; 212} 213 214void 215tcp_fastopen_destroy(void) 216{ 217 callout_drain(&V_tcp_fastopen_autokey_ctx.c); 218 rm_destroy(&V_tcp_fastopen_keylock); 219 uma_zdestroy(V_counter_zone); 220} 221 222unsigned int * 223tcp_fastopen_alloc_counter(void) 224{ 225 unsigned int *counter; 226 counter = uma_zalloc(V_counter_zone, M_NOWAIT); 227 if (counter) 228 *counter = 1; 229 return (counter); 230} 231 232void 233tcp_fastopen_decrement_counter(unsigned int *counter) 234{ 235 if (*counter == 1) 236 uma_zfree(V_counter_zone, counter); 237 else 238 atomic_subtract_int(counter, 1); 239} 240 241static void 242tcp_fastopen_addkey_locked(uint8_t *key) 243{ 244 245 V_tcp_fastopen_keys.newest++; 246 if (V_tcp_fastopen_keys.newest == TCP_FASTOPEN_MAX_KEYS) 247 V_tcp_fastopen_keys.newest = 0; 248 memcpy(V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], key, 249 TCP_FASTOPEN_KEY_LEN); 250 if (V_tcp_fastopen_numkeys < TCP_FASTOPEN_MAX_KEYS) 251 V_tcp_fastopen_numkeys++; 252} 253 254static void 255tcp_fastopen_autokey_locked(void) 256{ 257 uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; 258 259 arc4rand(newkey, TCP_FASTOPEN_KEY_LEN, 0); 260 tcp_fastopen_addkey_locked(newkey); 261} 262 263static void 264tcp_fastopen_autokey_callout(void *arg) 265{ 266 struct tcp_fastopen_callout *ctx = arg; 267 268 CURVNET_SET(ctx->v); 269 tcp_fastopen_autokey_locked(); 270 callout_reset(&ctx->c, V_tcp_fastopen_autokey * hz, 271 tcp_fastopen_autokey_callout, ctx); 272 CURVNET_RESTORE(); 273} 274 275 276static uint64_t 277tcp_fastopen_make_cookie(uint8_t key[SIPHASH_KEY_LENGTH], struct in_conninfo *inc) 278{ 279 SIPHASH_CTX ctx; 280 uint64_t siphash; 281 282 SipHash24_Init(&ctx); 283 SipHash_SetKey(&ctx, key); 284 switch (inc->inc_flags & INC_ISIPV6) { 285#ifdef INET 286 case 0: 287 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(inc->inc_faddr)); 288 break; 289#endif 290#ifdef INET6 291 case INC_ISIPV6: 292 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(inc->inc6_faddr)); 293 break; 294#endif 295 } 296 SipHash_Final((u_int8_t *)&siphash, &ctx); 297 298 return (siphash); 299} 300 301 302/* 303 * Return values: 304 * -1 the cookie is invalid and no valid cookie is available 305 * 0 the cookie is invalid and the latest cookie has been returned 306 * 1 the cookie is valid and the latest cookie has been returned 307 */ 308int 309tcp_fastopen_check_cookie(struct in_conninfo *inc, uint8_t *cookie, 310 unsigned int len, uint64_t *latest_cookie) 311{ 312 struct rm_priotracker tracker; 313 unsigned int i, key_index; 314 uint64_t cur_cookie; 315 316 if (V_tcp_fastopen_acceptany) { 317 *latest_cookie = 0; 318 return (1); 319 } 320 321 if (len != TCP_FASTOPEN_COOKIE_LEN) { 322 if (V_tcp_fastopen_numkeys > 0) { 323 *latest_cookie = 324 tcp_fastopen_make_cookie( 325 V_tcp_fastopen_keys.key[V_tcp_fastopen_keys.newest], 326 inc); 327 return (0); 328 } 329 return (-1); 330 } 331 332 /* 333 * Check against each available key, from newest to oldest. 334 */ 335 TCP_FASTOPEN_KEYS_RLOCK(&tracker); 336 key_index = V_tcp_fastopen_keys.newest; 337 for (i = 0; i < V_tcp_fastopen_numkeys; i++) { 338 cur_cookie = 339 tcp_fastopen_make_cookie(V_tcp_fastopen_keys.key[key_index], 340 inc); 341 if (i == 0) 342 *latest_cookie = cur_cookie; 343 if (memcmp(cookie, &cur_cookie, TCP_FASTOPEN_COOKIE_LEN) == 0) { 344 TCP_FASTOPEN_KEYS_RUNLOCK(&tracker); 345 return (1); 346 } 347 if (key_index == 0) 348 key_index = TCP_FASTOPEN_MAX_KEYS - 1; 349 else 350 key_index--; 351 } 352 TCP_FASTOPEN_KEYS_RUNLOCK(&tracker); 353 354 return (0); 355} 356 357static int 358sysctl_net_inet_tcp_fastopen_autokey(SYSCTL_HANDLER_ARGS) 359{ 360 int error; 361 unsigned int new; 362 363 new = V_tcp_fastopen_autokey; 364 error = sysctl_handle_int(oidp, &new, 0, req); 365 if (error == 0 && req->newptr) { 366 if (new > (INT_MAX / hz)) 367 return (EINVAL); 368 369 TCP_FASTOPEN_KEYS_WLOCK(); 370 if (V_tcp_fastopen_enabled) { 371 if (V_tcp_fastopen_autokey && !new) 372 callout_stop(&V_tcp_fastopen_autokey_ctx.c); 373 else if (new) 374 callout_reset(&V_tcp_fastopen_autokey_ctx.c, 375 new * hz, tcp_fastopen_autokey_callout, 376 &V_tcp_fastopen_autokey_ctx); 377 } 378 V_tcp_fastopen_autokey = new; 379 TCP_FASTOPEN_KEYS_WUNLOCK(); 380 } 381 382 return (error); 383} 384 385static int 386sysctl_net_inet_tcp_fastopen_enabled(SYSCTL_HANDLER_ARGS) 387{ 388 int error; 389 unsigned int new; 390 391 new = V_tcp_fastopen_enabled; 392 error = sysctl_handle_int(oidp, &new, 0, req); 393 if (error == 0 && req->newptr) { 394 if (V_tcp_fastopen_enabled && !new) { 395 /* enabled -> disabled */ 396 TCP_FASTOPEN_KEYS_WLOCK(); 397 V_tcp_fastopen_numkeys = 0; 398 V_tcp_fastopen_keys.newest = TCP_FASTOPEN_MAX_KEYS - 1; 399 if (V_tcp_fastopen_autokey) 400 callout_stop(&V_tcp_fastopen_autokey_ctx.c); 401 V_tcp_fastopen_enabled = 0; 402 TCP_FASTOPEN_KEYS_WUNLOCK(); 403 } else if (!V_tcp_fastopen_enabled && new) { 404 /* disabled -> enabled */ 405 TCP_FASTOPEN_KEYS_WLOCK(); 406 if (V_tcp_fastopen_autokey && 407 (V_tcp_fastopen_numkeys == 0)) { 408 tcp_fastopen_autokey_locked(); 409 callout_reset(&V_tcp_fastopen_autokey_ctx.c, 410 V_tcp_fastopen_autokey * hz, 411 tcp_fastopen_autokey_callout, 412 &V_tcp_fastopen_autokey_ctx); 413 } 414 V_tcp_fastopen_enabled = 1; 415 TCP_FASTOPEN_KEYS_WUNLOCK(); 416 } 417 } 418 return (error); 419} 420 421static int 422sysctl_net_inet_tcp_fastopen_setkey(SYSCTL_HANDLER_ARGS) 423{ 424 int error; 425 uint8_t newkey[TCP_FASTOPEN_KEY_LEN]; 426 427 if (req->oldptr != NULL || req->oldlen != 0) 428 return (EINVAL); 429 if (req->newptr == NULL) 430 return (EPERM); 431 if (req->newlen != sizeof(newkey)) 432 return (EINVAL); 433 error = SYSCTL_IN(req, newkey, sizeof(newkey)); 434 if (error) 435 return (error); 436 437 TCP_FASTOPEN_KEYS_WLOCK(); 438 tcp_fastopen_addkey_locked(newkey); 439 TCP_FASTOPEN_KEYS_WUNLOCK(); 440 441 return (0); 442} 443