ip_flow.c revision 1.68
1 /* $NetBSD: ip_flow.c,v 1.68 2016/06/13 08:04:44 knakahara Exp $ */ 2 3 /*- 4 * Copyright (c) 1998 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by the 3am Software Foundry ("3am"). It was developed by Matt Thomas. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __KERNEL_RCSID(0, "$NetBSD: ip_flow.c,v 1.68 2016/06/13 08:04:44 knakahara Exp $"); 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/malloc.h> 38 #include <sys/mbuf.h> 39 #include <sys/domain.h> 40 #include <sys/protosw.h> 41 #include <sys/socket.h> 42 #include <sys/socketvar.h> 43 #include <sys/errno.h> 44 #include <sys/time.h> 45 #include <sys/kernel.h> 46 #include <sys/pool.h> 47 #include <sys/sysctl.h> 48 49 #include <net/if.h> 50 #include <net/if_dl.h> 51 #include <net/route.h> 52 #include <net/pfil.h> 53 54 #include <netinet/in.h> 55 #include <netinet/in_systm.h> 56 #include <netinet/ip.h> 57 #include <netinet/in_pcb.h> 58 #include <netinet/in_var.h> 59 #include <netinet/ip_var.h> 60 #include <netinet/ip_private.h> 61 62 /* 63 * Similar code is very well commented in netinet6/ip6_flow.c 64 */ 65 66 #define IPFLOW_HASHBITS 6 /* should not be a multiple of 8 */ 67 68 static struct pool ipflow_pool; 69 70 LIST_HEAD(ipflowhead, ipflow); 71 72 #define IPFLOW_TIMER (5 * PR_SLOWHZ) 73 #define IPFLOW_DEFAULT_HASHSIZE (1 << IPFLOW_HASHBITS) 74 75 static struct ipflowhead *ipflowtable = NULL; 76 static struct ipflowhead ipflowlist; 77 static int ipflow_inuse; 78 79 #define IPFLOW_INSERT(bucket, ipf) \ 80 do { \ 81 LIST_INSERT_HEAD((bucket), (ipf), ipf_hash); \ 82 LIST_INSERT_HEAD(&ipflowlist, (ipf), ipf_list); \ 83 } while (/*CONSTCOND*/ 0) 84 85 #define IPFLOW_REMOVE(ipf) \ 86 do { \ 87 LIST_REMOVE((ipf), ipf_hash); \ 88 LIST_REMOVE((ipf), ipf_list); \ 89 } while (/*CONSTCOND*/ 0) 90 91 #ifndef IPFLOW_MAX 92 #define IPFLOW_MAX 256 93 #endif 94 static int ip_maxflows = IPFLOW_MAX; 95 static int ip_hashsize = IPFLOW_DEFAULT_HASHSIZE; 96 97 static void ipflow_sysctl_init(struct sysctllog **); 98 99 static size_t 100 ipflow_hash(const struct ip *ip) 101 { 102 size_t hash = ip->ip_tos; 103 size_t idx; 104 105 for (idx = 0; idx < 32; idx += IPFLOW_HASHBITS) { 106 hash += (ip->ip_dst.s_addr >> (32 - idx)) + 107 (ip->ip_src.s_addr >> idx); 108 } 109 110 return hash & (ip_hashsize-1); 111 } 112 113 static struct ipflow * 114 ipflow_lookup(const struct ip *ip) 115 { 116 size_t hash; 117 struct ipflow *ipf; 118 119 hash = ipflow_hash(ip); 120 121 LIST_FOREACH(ipf, &ipflowtable[hash], ipf_hash) { 122 if (ip->ip_dst.s_addr == ipf->ipf_dst.s_addr 123 && ip->ip_src.s_addr == ipf->ipf_src.s_addr 124 && ip->ip_tos == ipf->ipf_tos) 125 break; 126 } 127 return ipf; 128 } 129 130 void 131 ipflow_poolinit(void) 132 { 133 134 pool_init(&ipflow_pool, sizeof(struct ipflow), 0, 0, 0, "ipflowpl", 135 NULL, IPL_NET); 136 } 137 138 static int 139 ipflow_reinit(int table_size) 140 { 141 struct ipflowhead *new_table; 142 size_t i; 143 144 new_table = (struct ipflowhead *)malloc(sizeof(struct ipflowhead) * 145 table_size, M_RTABLE, M_NOWAIT); 146 147 if (new_table == NULL) 148 return 1; 149 150 if (ipflowtable != NULL) 151 free(ipflowtable, M_RTABLE); 152 153 ipflowtable = new_table; 154 ip_hashsize = table_size; 155 156 LIST_INIT(&ipflowlist); 157 for (i = 0; i < ip_hashsize; i++) 158 LIST_INIT(&ipflowtable[i]); 159 160 return 0; 161 } 162 163 void 164 ipflow_init(void) 165 { 166 (void)ipflow_reinit(ip_hashsize); 167 ipflow_sysctl_init(NULL); 168 } 169 170 int 171 ipflow_fastforward(struct mbuf *m) 172 { 173 struct ip *ip; 174 struct ip ip_store; 175 struct ipflow *ipf; 176 struct rtentry *rt; 177 const struct sockaddr *dst; 178 int error; 179 int iplen; 180 struct ifnet *ifp; 181 int s; 182 183 /* 184 * Are we forwarding packets? Big enough for an IP packet? 185 */ 186 if (!ipforwarding || ipflow_inuse == 0 || m->m_len < sizeof(struct ip)) 187 return 0; 188 189 /* 190 * Was packet received as a link-level multicast or broadcast? 191 * If so, don't try to fast forward.. 192 */ 193 if ((m->m_flags & (M_BCAST|M_MCAST)) != 0) 194 return 0; 195 196 /* 197 * IP header with no option and valid version and length 198 */ 199 if (IP_HDR_ALIGNED_P(mtod(m, const void *))) 200 ip = mtod(m, struct ip *); 201 else { 202 memcpy(&ip_store, mtod(m, const void *), sizeof(ip_store)); 203 ip = &ip_store; 204 } 205 iplen = ntohs(ip->ip_len); 206 if (ip->ip_v != IPVERSION || ip->ip_hl != (sizeof(struct ip) >> 2) || 207 iplen < sizeof(struct ip) || iplen > m->m_pkthdr.len) 208 return 0; 209 /* 210 * Find a flow. 211 */ 212 if ((ipf = ipflow_lookup(ip)) == NULL) 213 return 0; 214 215 ifp = m_get_rcvif(m, &s); 216 /* 217 * Verify the IP header checksum. 218 */ 219 switch (m->m_pkthdr.csum_flags & 220 ((ifp->if_csum_flags_rx & M_CSUM_IPv4) | 221 M_CSUM_IPv4_BAD)) { 222 case M_CSUM_IPv4|M_CSUM_IPv4_BAD: 223 m_put_rcvif(ifp, &s); 224 return 0; 225 226 case M_CSUM_IPv4: 227 /* Checksum was okay. */ 228 break; 229 230 default: 231 /* Must compute it ourselves. */ 232 if (in_cksum(m, sizeof(struct ip)) != 0) { 233 m_put_rcvif(ifp, &s); 234 return 0; 235 } 236 break; 237 } 238 m_put_rcvif(ifp, &s); 239 240 /* 241 * Route and interface still up? 242 */ 243 if ((rt = rtcache_validate(&ipf->ipf_ro)) == NULL || 244 (rt->rt_ifp->if_flags & IFF_UP) == 0 || 245 (rt->rt_flags & (RTF_BLACKHOLE | RTF_BROADCAST)) != 0) 246 return 0; 247 248 /* 249 * Packet size OK? TTL? 250 */ 251 if (m->m_pkthdr.len > rt->rt_ifp->if_mtu || ip->ip_ttl <= IPTTLDEC) 252 return 0; 253 254 /* 255 * Clear any in-bound checksum flags for this packet. 256 */ 257 m->m_pkthdr.csum_flags = 0; 258 259 /* 260 * Everything checks out and so we can forward this packet. 261 * Modify the TTL and incrementally change the checksum. 262 * 263 * This method of adding the checksum works on either endian CPU. 264 * If htons() is inlined, all the arithmetic is folded; otherwise 265 * the htons()s are combined by CSE due to the const attribute. 266 * 267 * Don't bother using HW checksumming here -- the incremental 268 * update is pretty fast. 269 */ 270 ip->ip_ttl -= IPTTLDEC; 271 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8)) 272 ip->ip_sum -= ~htons(IPTTLDEC << 8); 273 else 274 ip->ip_sum += htons(IPTTLDEC << 8); 275 276 /* 277 * Done modifying the header; copy it back, if necessary. 278 * 279 * XXX Use m_copyback_cow(9) here? --dyoung 280 */ 281 if (IP_HDR_ALIGNED_P(mtod(m, void *)) == 0) 282 memcpy(mtod(m, void *), &ip_store, sizeof(ip_store)); 283 284 /* 285 * Trim the packet in case it's too long.. 286 */ 287 if (m->m_pkthdr.len > iplen) { 288 if (m->m_len == m->m_pkthdr.len) { 289 m->m_len = iplen; 290 m->m_pkthdr.len = iplen; 291 } else 292 m_adj(m, iplen - m->m_pkthdr.len); 293 } 294 295 /* 296 * Send the packet on its way. All we can get back is ENOBUFS 297 */ 298 ipf->ipf_uses++; 299 PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER); 300 301 if (rt->rt_flags & RTF_GATEWAY) 302 dst = rt->rt_gateway; 303 else 304 dst = rtcache_getdst(&ipf->ipf_ro); 305 306 KERNEL_LOCK(1, NULL); 307 if ((error = (*rt->rt_ifp->if_output)(rt->rt_ifp, m, dst, rt)) != 0) { 308 if (error == ENOBUFS) 309 ipf->ipf_dropped++; 310 else 311 ipf->ipf_errors++; 312 } 313 KERNEL_UNLOCK_ONE(NULL); 314 return 1; 315 } 316 317 static void 319 ipflow_addstats(struct ipflow *ipf) 320 { 321 struct rtentry *rt; 322 uint64_t *ips; 323 324 if ((rt = rtcache_validate(&ipf->ipf_ro)) != NULL) 325 rt->rt_use += ipf->ipf_uses; 326 327 ips = IP_STAT_GETREF(); 328 ips[IP_STAT_CANTFORWARD] += ipf->ipf_errors + ipf->ipf_dropped; 329 ips[IP_STAT_TOTAL] += ipf->ipf_uses; 330 ips[IP_STAT_FORWARD] += ipf->ipf_uses; 331 ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses; 332 IP_STAT_PUTREF(); 333 } 334 335 static void 336 ipflow_free(struct ipflow *ipf) 337 { 338 int s; 339 /* 340 * Remove the flow from the hash table (at elevated IPL). 341 * Once it's off the list, we can deal with it at normal 342 * network IPL. 343 */ 344 s = splnet(); 345 IPFLOW_REMOVE(ipf); 346 splx(s); 347 ipflow_addstats(ipf); 348 rtcache_free(&ipf->ipf_ro); 349 ipflow_inuse--; 350 pool_put(&ipflow_pool, ipf); 351 } 352 353 struct ipflow * 354 ipflow_reap(bool just_one) 355 { 356 while (just_one || ipflow_inuse > ip_maxflows) { 357 struct ipflow *ipf, *maybe_ipf = NULL; 358 int s; 359 360 ipf = LIST_FIRST(&ipflowlist); 361 while (ipf != NULL) { 362 /* 363 * If this no longer points to a valid route 364 * reclaim it. 365 */ 366 if (rtcache_validate(&ipf->ipf_ro) == NULL) 367 goto done; 368 /* 369 * choose the one that's been least recently 370 * used or has had the least uses in the 371 * last 1.5 intervals. 372 */ 373 if (maybe_ipf == NULL || 374 ipf->ipf_timer < maybe_ipf->ipf_timer || 375 (ipf->ipf_timer == maybe_ipf->ipf_timer && 376 ipf->ipf_last_uses + ipf->ipf_uses < 377 maybe_ipf->ipf_last_uses + 378 maybe_ipf->ipf_uses)) 379 maybe_ipf = ipf; 380 ipf = LIST_NEXT(ipf, ipf_list); 381 } 382 ipf = maybe_ipf; 383 done: 384 /* 385 * Remove the entry from the flow table. 386 */ 387 s = splnet(); 388 IPFLOW_REMOVE(ipf); 389 splx(s); 390 ipflow_addstats(ipf); 391 rtcache_free(&ipf->ipf_ro); 392 if (just_one) 393 return ipf; 394 pool_put(&ipflow_pool, ipf); 395 ipflow_inuse--; 396 } 397 return NULL; 398 } 399 400 void 401 ipflow_slowtimo(void) 402 { 403 struct rtentry *rt; 404 struct ipflow *ipf, *next_ipf; 405 uint64_t *ips; 406 407 mutex_enter(softnet_lock); 408 KERNEL_LOCK(1, NULL); 409 for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) { 410 next_ipf = LIST_NEXT(ipf, ipf_list); 411 if (PRT_SLOW_ISEXPIRED(ipf->ipf_timer) || 412 (rt = rtcache_validate(&ipf->ipf_ro)) == NULL) { 413 ipflow_free(ipf); 414 } else { 415 ipf->ipf_last_uses = ipf->ipf_uses; 416 rt->rt_use += ipf->ipf_uses; 417 ips = IP_STAT_GETREF(); 418 ips[IP_STAT_TOTAL] += ipf->ipf_uses; 419 ips[IP_STAT_FORWARD] += ipf->ipf_uses; 420 ips[IP_STAT_FASTFORWARD] += ipf->ipf_uses; 421 IP_STAT_PUTREF(); 422 ipf->ipf_uses = 0; 423 } 424 } 425 KERNEL_UNLOCK_ONE(NULL); 426 mutex_exit(softnet_lock); 427 } 428 429 void 430 ipflow_create(const struct route *ro, struct mbuf *m) 431 { 432 const struct ip *const ip = mtod(m, const struct ip *); 433 struct ipflow *ipf; 434 size_t hash; 435 int s; 436 437 /* 438 * Don't create cache entries for ICMP messages. 439 */ 440 if (ip_maxflows == 0 || ip->ip_p == IPPROTO_ICMP) 441 return; 442 443 KERNEL_LOCK(1, NULL); 444 445 /* 446 * See if an existing flow struct exists. If so remove it from its 447 * list and free the old route. If not, try to malloc a new one 448 * (if we aren't at our limit). 449 */ 450 ipf = ipflow_lookup(ip); 451 if (ipf == NULL) { 452 if (ipflow_inuse >= ip_maxflows) { 453 ipf = ipflow_reap(true); 454 } else { 455 ipf = pool_get(&ipflow_pool, PR_NOWAIT); 456 if (ipf == NULL) 457 goto out; 458 ipflow_inuse++; 459 } 460 memset(ipf, 0, sizeof(*ipf)); 461 } else { 462 s = splnet(); 463 IPFLOW_REMOVE(ipf); 464 splx(s); 465 ipflow_addstats(ipf); 466 rtcache_free(&ipf->ipf_ro); 467 ipf->ipf_uses = ipf->ipf_last_uses = 0; 468 ipf->ipf_errors = ipf->ipf_dropped = 0; 469 } 470 471 /* 472 * Fill in the updated information. 473 */ 474 rtcache_copy(&ipf->ipf_ro, ro); 475 ipf->ipf_dst = ip->ip_dst; 476 ipf->ipf_src = ip->ip_src; 477 ipf->ipf_tos = ip->ip_tos; 478 PRT_SLOW_ARM(ipf->ipf_timer, IPFLOW_TIMER); 479 480 /* 481 * Insert into the approriate bucket of the flow table. 482 */ 483 hash = ipflow_hash(ip); 484 s = splnet(); 485 IPFLOW_INSERT(&ipflowtable[hash], ipf); 486 splx(s); 487 488 out: 489 KERNEL_UNLOCK_ONE(NULL); 490 } 491 492 int 493 ipflow_invalidate_all(int new_size) 494 { 495 struct ipflow *ipf, *next_ipf; 496 int s, error; 497 498 error = 0; 499 s = splnet(); 500 for (ipf = LIST_FIRST(&ipflowlist); ipf != NULL; ipf = next_ipf) { 501 next_ipf = LIST_NEXT(ipf, ipf_list); 502 ipflow_free(ipf); 503 } 504 505 if (new_size) 506 error = ipflow_reinit(new_size); 507 splx(s); 508 509 return error; 510 } 511 512 #ifdef GATEWAY 513 /* 514 * sysctl helper routine for net.inet.ip.maxflows. 515 */ 516 static int 517 sysctl_net_inet_ip_maxflows(SYSCTLFN_ARGS) 518 { 519 int error; 520 521 error = sysctl_lookup(SYSCTLFN_CALL(rnode)); 522 if (error || newp == NULL) 523 return (error); 524 525 mutex_enter(softnet_lock); 526 KERNEL_LOCK(1, NULL); 527 528 ipflow_reap(false); 529 530 KERNEL_UNLOCK_ONE(NULL); 531 mutex_exit(softnet_lock); 532 533 return (0); 534 } 535 536 static int 537 sysctl_net_inet_ip_hashsize(SYSCTLFN_ARGS) 538 { 539 int error, tmp; 540 struct sysctlnode node; 541 542 node = *rnode; 543 tmp = ip_hashsize; 544 node.sysctl_data = &tmp; 545 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 546 if (error || newp == NULL) 547 return (error); 548 549 if ((tmp & (tmp - 1)) == 0 && tmp != 0) { 550 /* 551 * Can only fail due to malloc() 552 */ 553 mutex_enter(softnet_lock); 554 KERNEL_LOCK(1, NULL); 555 556 error = ipflow_invalidate_all(tmp); 557 558 KERNEL_UNLOCK_ONE(NULL); 559 mutex_exit(softnet_lock); 560 561 } else { 562 /* 563 * EINVAL if not a power of 2 564 */ 565 error = EINVAL; 566 } 567 568 return error; 569 } 570 #endif /* GATEWAY */ 571 572 static void 573 ipflow_sysctl_init(struct sysctllog **clog) 574 { 575 sysctl_createv(clog, 0, NULL, NULL, 576 CTLFLAG_PERMANENT, 577 CTLTYPE_NODE, "inet", 578 SYSCTL_DESCR("PF_INET related settings"), 579 NULL, 0, NULL, 0, 580 CTL_NET, PF_INET, CTL_EOL); 581 sysctl_createv(clog, 0, NULL, NULL, 582 CTLFLAG_PERMANENT, 583 CTLTYPE_NODE, "ip", 584 SYSCTL_DESCR("IPv4 related settings"), 585 NULL, 0, NULL, 0, 586 CTL_NET, PF_INET, IPPROTO_IP, CTL_EOL); 587 588 #ifdef GATEWAY 589 sysctl_createv(clog, 0, NULL, NULL, 590 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 591 CTLTYPE_INT, "maxflows", 592 SYSCTL_DESCR("Number of flows for fast forwarding"), 593 sysctl_net_inet_ip_maxflows, 0, &ip_maxflows, 0, 594 CTL_NET, PF_INET, IPPROTO_IP, 595 IPCTL_MAXFLOWS, CTL_EOL); 596 sysctl_createv(clog, 0, NULL, NULL, 597 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 598 CTLTYPE_INT, "hashsize", 599 SYSCTL_DESCR("Size of hash table for fast forwarding (IPv4)"), 600 sysctl_net_inet_ip_hashsize, 0, &ip_hashsize, 0, 601 CTL_NET, PF_INET, IPPROTO_IP, 602 CTL_CREATE, CTL_EOL); 603 #endif /* GATEWAY */ 604 } 605
Indexes created Fri Oct 17 09:09:57 GMT 2025