ip_fastfwd.c revision 196881
1/*- 2 * Copyright (c) 2003 Andre Oppermann, Internet Business Solutions AG 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 * 3. The name of the author may not be used to endorse or promote 14 * products derived from this software without specific prior written 15 * permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30/* 31 * ip_fastforward gets its speed from processing the forwarded packet to 32 * completion (if_output on the other side) without any queues or netisr's. 33 * The receiving interface DMAs the packet into memory, the upper half of 34 * driver calls ip_fastforward, we do our routing table lookup and directly 35 * send it off to the outgoing interface, which DMAs the packet to the 36 * network card. The only part of the packet we touch with the CPU is the 37 * IP header (unless there are complex firewall rules touching other parts 38 * of the packet, but that is up to you). We are essentially limited by bus 39 * bandwidth and how fast the network card/driver can set up receives and 40 * transmits. 41 * 42 * We handle basic errors, IP header errors, checksum errors, 43 * destination unreachable, fragmentation and fragmentation needed and 44 * report them via ICMP to the sender. 45 * 46 * Else if something is not pure IPv4 unicast forwarding we fall back to 47 * the normal ip_input processing path. We should only be called from 48 * interfaces connected to the outside world. 49 * 50 * Firewalling is fully supported including divert, ipfw fwd and ipfilter 51 * ipnat and address rewrite. 52 * 53 * IPSEC is not supported if this host is a tunnel broker. IPSEC is 54 * supported for connections to/from local host. 55 * 56 * We try to do the least expensive (in CPU ops) checks and operations 57 * first to catch junk with as little overhead as possible. 58 * 59 * We take full advantage of hardware support for IP checksum and 60 * fragmentation offloading. 61 * 62 * We don't do ICMP redirect in the fast forwarding path. I have had my own 63 * cases where two core routers with Zebra routing suite would send millions 64 * ICMP redirects to connected hosts if the destination router was not the 65 * default gateway. In one case it was filling the routing table of a host 66 * with approximately 300.000 cloned redirect entries until it ran out of 67 * kernel memory. However the networking code proved very robust and it didn't 68 * crash or fail in other ways. 69 */ 70 71/* 72 * Many thanks to Matt Thomas of NetBSD for basic structure of ip_flow.c which 73 * is being followed here. 74 */ 75 76#include <sys/cdefs.h> 77__FBSDID("$FreeBSD: head/sys/netinet/ip_fastfwd.c 196881 2009-09-06 07:29:22Z pjd $"); 78 79#include "opt_ipfw.h" 80#include "opt_ipstealth.h" 81 82#include <sys/param.h> 83#include <sys/systm.h> 84#include <sys/kernel.h> 85#include <sys/malloc.h> 86#include <sys/mbuf.h> 87#include <sys/protosw.h> 88#include <sys/socket.h> 89#include <sys/sysctl.h> 90 91#include <net/pfil.h> 92#include <net/if.h> 93#include <net/if_types.h> 94#include <net/if_var.h> 95#include <net/if_dl.h> 96#include <net/route.h> 97#include <net/vnet.h> 98 99#include <netinet/in.h> 100#include <netinet/in_systm.h> 101#include <netinet/in_var.h> 102#include <netinet/ip.h> 103#include <netinet/ip_var.h> 104#include <netinet/ip_icmp.h> 105#include <netinet/ip_options.h> 106 107#include <machine/in_cksum.h> 108 109static VNET_DEFINE(int, ipfastforward_active); 110#define V_ipfastforward_active VNET(ipfastforward_active) 111 112SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, fastforwarding, CTLFLAG_RW, 113 &VNET_NAME(ipfastforward_active), 0, "Enable fast IP forwarding"); 114 115static struct sockaddr_in * 116ip_findroute(struct route *ro, struct in_addr dest, struct mbuf *m) 117{ 118 struct sockaddr_in *dst; 119 struct rtentry *rt; 120 121 /* 122 * Find route to destination. 123 */ 124 bzero(ro, sizeof(*ro)); 125 dst = (struct sockaddr_in *)&ro->ro_dst; 126 dst->sin_family = AF_INET; 127 dst->sin_len = sizeof(*dst); 128 dst->sin_addr.s_addr = dest.s_addr; 129 in_rtalloc_ign(ro, 0, M_GETFIB(m)); 130 131 /* 132 * Route there and interface still up? 133 */ 134 rt = ro->ro_rt; 135 if (rt && (rt->rt_flags & RTF_UP) && 136 (rt->rt_ifp->if_flags & IFF_UP) && 137 (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) { 138 if (rt->rt_flags & RTF_GATEWAY) 139 dst = (struct sockaddr_in *)rt->rt_gateway; 140 } else { 141 IPSTAT_INC(ips_noroute); 142 IPSTAT_INC(ips_cantforward); 143 if (rt) 144 RTFREE(rt); 145 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 146 return NULL; 147 } 148 return dst; 149} 150 151/* 152 * Try to forward a packet based on the destination address. 153 * This is a fast path optimized for the plain forwarding case. 154 * If the packet is handled (and consumed) here then we return NULL; 155 * otherwise mbuf is returned and the packet should be delivered 156 * to ip_input for full processing. 157 */ 158struct mbuf * 159ip_fastforward(struct mbuf *m) 160{ 161 struct ip *ip; 162 struct mbuf *m0 = NULL; 163 struct route ro; 164 struct sockaddr_in *dst = NULL; 165 struct ifnet *ifp; 166 struct in_addr odest, dest; 167 u_short sum, ip_len; 168 int error = 0; 169 int hlen, mtu; 170#ifdef IPFIREWALL_FORWARD 171 struct m_tag *fwd_tag; 172#endif 173 174 /* 175 * Are we active and forwarding packets? 176 */ 177 if (!V_ipfastforward_active || !V_ipforwarding) 178 return m; 179 180 M_ASSERTVALID(m); 181 M_ASSERTPKTHDR(m); 182 183 bzero(&ro, sizeof(ro)); 184 185 /* 186 * Step 1: check for packet drop conditions (and sanity checks) 187 */ 188 189 /* 190 * Is entire packet big enough? 191 */ 192 if (m->m_pkthdr.len < sizeof(struct ip)) { 193 IPSTAT_INC(ips_tooshort); 194 goto drop; 195 } 196 197 /* 198 * Is first mbuf large enough for ip header and is header present? 199 */ 200 if (m->m_len < sizeof (struct ip) && 201 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 202 IPSTAT_INC(ips_toosmall); 203 return NULL; /* mbuf already free'd */ 204 } 205 206 ip = mtod(m, struct ip *); 207 208 /* 209 * Is it IPv4? 210 */ 211 if (ip->ip_v != IPVERSION) { 212 IPSTAT_INC(ips_badvers); 213 goto drop; 214 } 215 216 /* 217 * Is IP header length correct and is it in first mbuf? 218 */ 219 hlen = ip->ip_hl << 2; 220 if (hlen < sizeof(struct ip)) { /* minimum header length */ 221 IPSTAT_INC(ips_badlen); 222 goto drop; 223 } 224 if (hlen > m->m_len) { 225 if ((m = m_pullup(m, hlen)) == NULL) { 226 IPSTAT_INC(ips_badhlen); 227 return NULL; /* mbuf already free'd */ 228 } 229 ip = mtod(m, struct ip *); 230 } 231 232 /* 233 * Checksum correct? 234 */ 235 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) 236 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 237 else { 238 if (hlen == sizeof(struct ip)) 239 sum = in_cksum_hdr(ip); 240 else 241 sum = in_cksum(m, hlen); 242 } 243 if (sum) { 244 IPSTAT_INC(ips_badsum); 245 goto drop; 246 } 247 248 /* 249 * Remember that we have checked the IP header and found it valid. 250 */ 251 m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); 252 253 ip_len = ntohs(ip->ip_len); 254 255 /* 256 * Is IP length longer than packet we have got? 257 */ 258 if (m->m_pkthdr.len < ip_len) { 259 IPSTAT_INC(ips_tooshort); 260 goto drop; 261 } 262 263 /* 264 * Is packet longer than IP header tells us? If yes, truncate packet. 265 */ 266 if (m->m_pkthdr.len > ip_len) { 267 if (m->m_len == m->m_pkthdr.len) { 268 m->m_len = ip_len; 269 m->m_pkthdr.len = ip_len; 270 } else 271 m_adj(m, ip_len - m->m_pkthdr.len); 272 } 273 274 /* 275 * Is packet from or to 127/8? 276 */ 277 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 278 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 279 IPSTAT_INC(ips_badaddr); 280 goto drop; 281 } 282 283#ifdef ALTQ 284 /* 285 * Is packet dropped by traffic conditioner? 286 */ 287 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 288 goto drop; 289#endif 290 291 /* 292 * Step 2: fallback conditions to normal ip_input path processing 293 */ 294 295 /* 296 * Only IP packets without options 297 */ 298 if (ip->ip_hl != (sizeof(struct ip) >> 2)) { 299 if (ip_doopts == 1) 300 return m; 301 else if (ip_doopts == 2) { 302 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB, 303 0, 0); 304 return NULL; /* mbuf already free'd */ 305 } 306 /* else ignore IP options and continue */ 307 } 308 309 /* 310 * Only unicast IP, not from loopback, no L2 or IP broadcast, 311 * no multicast, no INADDR_ANY 312 * 313 * XXX: Probably some of these checks could be direct drop 314 * conditions. However it is not clear whether there are some 315 * hacks or obscure behaviours which make it neccessary to 316 * let ip_input handle it. We play safe here and let ip_input 317 * deal with it until it is proven that we can directly drop it. 318 */ 319 if ((m->m_flags & (M_BCAST|M_MCAST)) || 320 (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) || 321 ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST || 322 ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST || 323 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 324 IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 325 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) || 326 IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) || 327 ip->ip_src.s_addr == INADDR_ANY || 328 ip->ip_dst.s_addr == INADDR_ANY ) 329 return m; 330 331 /* 332 * Is it for a local address on this host? 333 */ 334 if (in_localip(ip->ip_dst)) 335 return m; 336 337 IPSTAT_INC(ips_total); 338 339 /* 340 * Step 3: incoming packet firewall processing 341 */ 342 343 /* 344 * Convert to host representation 345 */ 346 ip->ip_len = ntohs(ip->ip_len); 347 ip->ip_off = ntohs(ip->ip_off); 348 349 odest.s_addr = dest.s_addr = ip->ip_dst.s_addr; 350 351 /* 352 * Run through list of ipfilter hooks for input packets 353 */ 354 if (!PFIL_HOOKED(&inet_pfil_hook)) 355 goto passin; 356 357 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) || 358 m == NULL) 359 goto drop; 360 361 M_ASSERTVALID(m); 362 M_ASSERTPKTHDR(m); 363 364 ip = mtod(m, struct ip *); /* m may have changed by pfil hook */ 365 dest.s_addr = ip->ip_dst.s_addr; 366 367 /* 368 * Destination address changed? 369 */ 370 if (odest.s_addr != dest.s_addr) { 371 /* 372 * Is it now for a local address on this host? 373 */ 374 if (in_localip(dest)) 375 goto forwardlocal; 376 /* 377 * Go on with new destination address 378 */ 379 } 380#ifdef IPFIREWALL_FORWARD 381 if (m->m_flags & M_FASTFWD_OURS) { 382 /* 383 * ipfw changed it for a local address on this host. 384 */ 385 goto forwardlocal; 386 } 387#endif /* IPFIREWALL_FORWARD */ 388 389passin: 390 /* 391 * Step 4: decrement TTL and look up route 392 */ 393 394 /* 395 * Check TTL 396 */ 397#ifdef IPSTEALTH 398 if (!V_ipstealth) { 399#endif 400 if (ip->ip_ttl <= IPTTLDEC) { 401 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0); 402 return NULL; /* mbuf already free'd */ 403 } 404 405 /* 406 * Decrement the TTL and incrementally change the IP header checksum. 407 * Don't bother doing this with hw checksum offloading, it's faster 408 * doing it right here. 409 */ 410 ip->ip_ttl -= IPTTLDEC; 411 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8)) 412 ip->ip_sum -= ~htons(IPTTLDEC << 8); 413 else 414 ip->ip_sum += htons(IPTTLDEC << 8); 415#ifdef IPSTEALTH 416 } 417#endif 418 419 /* 420 * Find route to destination. 421 */ 422 if ((dst = ip_findroute(&ro, dest, m)) == NULL) 423 return NULL; /* icmp unreach already sent */ 424 ifp = ro.ro_rt->rt_ifp; 425 426 /* 427 * Immediately drop blackholed traffic, and directed broadcasts 428 * for either the all-ones or all-zero subnet addresses on 429 * locally attached networks. 430 */ 431 if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0) 432 goto drop; 433 434 /* 435 * Step 5: outgoing firewall packet processing 436 */ 437 438 /* 439 * Run through list of hooks for output packets. 440 */ 441 if (!PFIL_HOOKED(&inet_pfil_hook)) 442 goto passout; 443 444 if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) { 445 goto drop; 446 } 447 448 M_ASSERTVALID(m); 449 M_ASSERTPKTHDR(m); 450 451 ip = mtod(m, struct ip *); 452 dest.s_addr = ip->ip_dst.s_addr; 453 454 /* 455 * Destination address changed? 456 */ 457#ifndef IPFIREWALL_FORWARD 458 if (odest.s_addr != dest.s_addr) { 459#else 460 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 461 if (odest.s_addr != dest.s_addr || fwd_tag != NULL) { 462#endif /* IPFIREWALL_FORWARD */ 463 /* 464 * Is it now for a local address on this host? 465 */ 466#ifndef IPFIREWALL_FORWARD 467 if (in_localip(dest)) { 468#else 469 if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) { 470#endif /* IPFIREWALL_FORWARD */ 471forwardlocal: 472 /* 473 * Return packet for processing by ip_input(). 474 * Keep host byte order as expected at ip_input's 475 * "ours"-label. 476 */ 477 m->m_flags |= M_FASTFWD_OURS; 478 if (ro.ro_rt) 479 RTFREE(ro.ro_rt); 480 return m; 481 } 482 /* 483 * Redo route lookup with new destination address 484 */ 485#ifdef IPFIREWALL_FORWARD 486 if (fwd_tag) { 487 dest.s_addr = ((struct sockaddr_in *) 488 (fwd_tag + 1))->sin_addr.s_addr; 489 m_tag_delete(m, fwd_tag); 490 } 491#endif /* IPFIREWALL_FORWARD */ 492 RTFREE(ro.ro_rt); 493 if ((dst = ip_findroute(&ro, dest, m)) == NULL) 494 return NULL; /* icmp unreach already sent */ 495 ifp = ro.ro_rt->rt_ifp; 496 } 497 498passout: 499 /* 500 * Step 6: send off the packet 501 */ 502 503 /* 504 * Check if route is dampned (when ARP is unable to resolve) 505 */ 506 if ((ro.ro_rt->rt_flags & RTF_REJECT) && 507 (ro.ro_rt->rt_rmx.rmx_expire == 0 || 508 time_uptime < ro.ro_rt->rt_rmx.rmx_expire)) { 509 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 510 goto consumed; 511 } 512 513#ifndef ALTQ 514 /* 515 * Check if there is enough space in the interface queue 516 */ 517 if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >= 518 ifp->if_snd.ifq_maxlen) { 519 IPSTAT_INC(ips_odropped); 520 /* would send source quench here but that is depreciated */ 521 goto drop; 522 } 523#endif 524 525 /* 526 * Check if media link state of interface is not down 527 */ 528 if (ifp->if_link_state == LINK_STATE_DOWN) { 529 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 530 goto consumed; 531 } 532 533 /* 534 * Check if packet fits MTU or if hardware will fragment for us 535 */ 536 if (ro.ro_rt->rt_rmx.rmx_mtu) 537 mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 538 else 539 mtu = ifp->if_mtu; 540 541 if (ip->ip_len <= mtu || 542 (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) { 543 /* 544 * Restore packet header fields to original values 545 */ 546 ip->ip_len = htons(ip->ip_len); 547 ip->ip_off = htons(ip->ip_off); 548 /* 549 * Send off the packet via outgoing interface 550 */ 551 error = (*ifp->if_output)(ifp, m, 552 (struct sockaddr *)dst, &ro); 553 } else { 554 /* 555 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery 556 */ 557 if (ip->ip_off & IP_DF) { 558 IPSTAT_INC(ips_cantfrag); 559 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 560 0, mtu); 561 goto consumed; 562 } else { 563 /* 564 * We have to fragment the packet 565 */ 566 m->m_pkthdr.csum_flags |= CSUM_IP; 567 /* 568 * ip_fragment expects ip_len and ip_off in host byte 569 * order but returns all packets in network byte order 570 */ 571 if (ip_fragment(ip, &m, mtu, ifp->if_hwassist, 572 (~ifp->if_hwassist & CSUM_DELAY_IP))) { 573 goto drop; 574 } 575 KASSERT(m != NULL, ("null mbuf and no error")); 576 /* 577 * Send off the fragments via outgoing interface 578 */ 579 error = 0; 580 do { 581 m0 = m->m_nextpkt; 582 m->m_nextpkt = NULL; 583 584 error = (*ifp->if_output)(ifp, m, 585 (struct sockaddr *)dst, &ro); 586 if (error) 587 break; 588 } while ((m = m0) != NULL); 589 if (error) { 590 /* Reclaim remaining fragments */ 591 for (m = m0; m; m = m0) { 592 m0 = m->m_nextpkt; 593 m_freem(m); 594 } 595 } else 596 IPSTAT_INC(ips_fragmented); 597 } 598 } 599 600 if (error != 0) 601 IPSTAT_INC(ips_odropped); 602 else { 603 ro.ro_rt->rt_rmx.rmx_pksent++; 604 IPSTAT_INC(ips_forward); 605 IPSTAT_INC(ips_fastforward); 606 } 607consumed: 608 RTFREE(ro.ro_rt); 609 return NULL; 610drop: 611 if (m) 612 m_freem(m); 613 if (ro.ro_rt) 614 RTFREE(ro.ro_rt); 615 return NULL; 616} 617