ip_fastfwd.c revision 185088
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 185088 2008-11-19 09:39:34Z zec $"); 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#include <sys/vimage.h> 91 92#include <net/pfil.h> 93#include <net/if.h> 94#include <net/if_types.h> 95#include <net/if_var.h> 96#include <net/if_dl.h> 97#include <net/route.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> 108 109#ifdef VIMAGE_GLOBALS 110static int ipfastforward_active; 111#endif 112SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_ip, OID_AUTO, fastforwarding, 113 CTLFLAG_RW, 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 INIT_VNET_INET(curvnet); 119 struct sockaddr_in *dst; 120 struct rtentry *rt; 121 122 /* 123 * Find route to destination. 124 */ 125 bzero(ro, sizeof(*ro)); 126 dst = (struct sockaddr_in *)&ro->ro_dst; 127 dst->sin_family = AF_INET; 128 dst->sin_len = sizeof(*dst); 129 dst->sin_addr.s_addr = dest.s_addr; 130 in_rtalloc_ign(ro, RTF_CLONING, M_GETFIB(m)); 131 132 /* 133 * Route there and interface still up? 134 */ 135 rt = ro->ro_rt; 136 if (rt && (rt->rt_flags & RTF_UP) && 137 (rt->rt_ifp->if_flags & IFF_UP) && 138 (rt->rt_ifp->if_drv_flags & IFF_DRV_RUNNING)) { 139 if (rt->rt_flags & RTF_GATEWAY) 140 dst = (struct sockaddr_in *)rt->rt_gateway; 141 } else { 142 V_ipstat.ips_noroute++; 143 V_ipstat.ips_cantforward++; 144 if (rt) 145 RTFREE(rt); 146 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 147 return NULL; 148 } 149 return dst; 150} 151 152/* 153 * Try to forward a packet based on the destination address. 154 * This is a fast path optimized for the plain forwarding case. 155 * If the packet is handled (and consumed) here then we return 1; 156 * otherwise 0 is returned and the packet should be delivered 157 * to ip_input for full processing. 158 */ 159struct mbuf * 160ip_fastforward(struct mbuf *m) 161{ 162 INIT_VNET_INET(curvnet); 163 struct ip *ip; 164 struct mbuf *m0 = NULL; 165 struct route ro; 166 struct sockaddr_in *dst = NULL; 167 struct ifnet *ifp; 168 struct in_addr odest, dest; 169 u_short sum, ip_len; 170 int error = 0; 171 int hlen, mtu; 172#ifdef IPFIREWALL_FORWARD 173 struct m_tag *fwd_tag; 174#endif 175 176 /* 177 * Are we active and forwarding packets? 178 */ 179 if (!V_ipfastforward_active || !V_ipforwarding) 180 return m; 181 182 M_ASSERTVALID(m); 183 M_ASSERTPKTHDR(m); 184 185 ro.ro_rt = NULL; 186 187 /* 188 * Step 1: check for packet drop conditions (and sanity checks) 189 */ 190 191 /* 192 * Is entire packet big enough? 193 */ 194 if (m->m_pkthdr.len < sizeof(struct ip)) { 195 V_ipstat.ips_tooshort++; 196 goto drop; 197 } 198 199 /* 200 * Is first mbuf large enough for ip header and is header present? 201 */ 202 if (m->m_len < sizeof (struct ip) && 203 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 204 V_ipstat.ips_toosmall++; 205 return NULL; /* mbuf already free'd */ 206 } 207 208 ip = mtod(m, struct ip *); 209 210 /* 211 * Is it IPv4? 212 */ 213 if (ip->ip_v != IPVERSION) { 214 V_ipstat.ips_badvers++; 215 goto drop; 216 } 217 218 /* 219 * Is IP header length correct and is it in first mbuf? 220 */ 221 hlen = ip->ip_hl << 2; 222 if (hlen < sizeof(struct ip)) { /* minimum header length */ 223 V_ipstat.ips_badlen++; 224 goto drop; 225 } 226 if (hlen > m->m_len) { 227 if ((m = m_pullup(m, hlen)) == NULL) { 228 V_ipstat.ips_badhlen++; 229 return NULL; /* mbuf already free'd */ 230 } 231 ip = mtod(m, struct ip *); 232 } 233 234 /* 235 * Checksum correct? 236 */ 237 if (m->m_pkthdr.csum_flags & CSUM_IP_CHECKED) 238 sum = !(m->m_pkthdr.csum_flags & CSUM_IP_VALID); 239 else { 240 if (hlen == sizeof(struct ip)) 241 sum = in_cksum_hdr(ip); 242 else 243 sum = in_cksum(m, hlen); 244 } 245 if (sum) { 246 V_ipstat.ips_badsum++; 247 goto drop; 248 } 249 250 /* 251 * Remember that we have checked the IP header and found it valid. 252 */ 253 m->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); 254 255 ip_len = ntohs(ip->ip_len); 256 257 /* 258 * Is IP length longer than packet we have got? 259 */ 260 if (m->m_pkthdr.len < ip_len) { 261 V_ipstat.ips_tooshort++; 262 goto drop; 263 } 264 265 /* 266 * Is packet longer than IP header tells us? If yes, truncate packet. 267 */ 268 if (m->m_pkthdr.len > ip_len) { 269 if (m->m_len == m->m_pkthdr.len) { 270 m->m_len = ip_len; 271 m->m_pkthdr.len = ip_len; 272 } else 273 m_adj(m, ip_len - m->m_pkthdr.len); 274 } 275 276 /* 277 * Is packet from or to 127/8? 278 */ 279 if ((ntohl(ip->ip_dst.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET || 280 (ntohl(ip->ip_src.s_addr) >> IN_CLASSA_NSHIFT) == IN_LOOPBACKNET) { 281 V_ipstat.ips_badaddr++; 282 goto drop; 283 } 284 285#ifdef ALTQ 286 /* 287 * Is packet dropped by traffic conditioner? 288 */ 289 if (altq_input != NULL && (*altq_input)(m, AF_INET) == 0) 290 goto drop; 291#endif 292 293 /* 294 * Step 2: fallback conditions to normal ip_input path processing 295 */ 296 297 /* 298 * Only IP packets without options 299 */ 300 if (ip->ip_hl != (sizeof(struct ip) >> 2)) { 301 if (ip_doopts == 1) 302 return m; 303 else if (ip_doopts == 2) { 304 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_FILTER_PROHIB, 305 0, 0); 306 return NULL; /* mbuf already free'd */ 307 } 308 /* else ignore IP options and continue */ 309 } 310 311 /* 312 * Only unicast IP, not from loopback, no L2 or IP broadcast, 313 * no multicast, no INADDR_ANY 314 * 315 * XXX: Probably some of these checks could be direct drop 316 * conditions. However it is not clear whether there are some 317 * hacks or obscure behaviours which make it neccessary to 318 * let ip_input handle it. We play safe here and let ip_input 319 * deal with it until it is proven that we can directly drop it. 320 */ 321 if ((m->m_flags & (M_BCAST|M_MCAST)) || 322 (m->m_pkthdr.rcvif->if_flags & IFF_LOOPBACK) || 323 ntohl(ip->ip_src.s_addr) == (u_long)INADDR_BROADCAST || 324 ntohl(ip->ip_dst.s_addr) == (u_long)INADDR_BROADCAST || 325 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 326 IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 327 IN_LINKLOCAL(ntohl(ip->ip_src.s_addr)) || 328 IN_LINKLOCAL(ntohl(ip->ip_dst.s_addr)) || 329 ip->ip_src.s_addr == INADDR_ANY || 330 ip->ip_dst.s_addr == INADDR_ANY ) 331 return m; 332 333 /* 334 * Is it for a local address on this host? 335 */ 336 if (in_localip(ip->ip_dst)) 337 return m; 338 339 V_ipstat.ips_total++; 340 341 /* 342 * Step 3: incoming packet firewall processing 343 */ 344 345 /* 346 * Convert to host representation 347 */ 348 ip->ip_len = ntohs(ip->ip_len); 349 ip->ip_off = ntohs(ip->ip_off); 350 351 odest.s_addr = dest.s_addr = ip->ip_dst.s_addr; 352 353 /* 354 * Run through list of ipfilter hooks for input packets 355 */ 356 if (!PFIL_HOOKED(&inet_pfil_hook)) 357 goto passin; 358 359 if (pfil_run_hooks(&inet_pfil_hook, &m, m->m_pkthdr.rcvif, PFIL_IN, NULL) || 360 m == NULL) 361 goto drop; 362 363 M_ASSERTVALID(m); 364 M_ASSERTPKTHDR(m); 365 366 ip = mtod(m, struct ip *); /* m may have changed by pfil hook */ 367 dest.s_addr = ip->ip_dst.s_addr; 368 369 /* 370 * Destination address changed? 371 */ 372 if (odest.s_addr != dest.s_addr) { 373 /* 374 * Is it now for a local address on this host? 375 */ 376 if (in_localip(dest)) 377 goto forwardlocal; 378 /* 379 * Go on with new destination address 380 */ 381 } 382#ifdef IPFIREWALL_FORWARD 383 if (m->m_flags & M_FASTFWD_OURS) { 384 /* 385 * ipfw changed it for a local address on this host. 386 */ 387 goto forwardlocal; 388 } 389#endif /* IPFIREWALL_FORWARD */ 390 391passin: 392 /* 393 * Step 4: decrement TTL and look up route 394 */ 395 396 /* 397 * Check TTL 398 */ 399#ifdef IPSTEALTH 400 if (!V_ipstealth) { 401#endif 402 if (ip->ip_ttl <= IPTTLDEC) { 403 icmp_error(m, ICMP_TIMXCEED, ICMP_TIMXCEED_INTRANS, 0, 0); 404 return NULL; /* mbuf already free'd */ 405 } 406 407 /* 408 * Decrement the TTL and incrementally change the IP header checksum. 409 * Don't bother doing this with hw checksum offloading, it's faster 410 * doing it right here. 411 */ 412 ip->ip_ttl -= IPTTLDEC; 413 if (ip->ip_sum >= (u_int16_t) ~htons(IPTTLDEC << 8)) 414 ip->ip_sum -= ~htons(IPTTLDEC << 8); 415 else 416 ip->ip_sum += htons(IPTTLDEC << 8); 417#ifdef IPSTEALTH 418 } 419#endif 420 421 /* 422 * Find route to destination. 423 */ 424 if ((dst = ip_findroute(&ro, dest, m)) == NULL) 425 return NULL; /* icmp unreach already sent */ 426 ifp = ro.ro_rt->rt_ifp; 427 428 /* 429 * Immediately drop blackholed traffic, and directed broadcasts 430 * for either the all-ones or all-zero subnet addresses on 431 * locally attached networks. 432 */ 433 if ((ro.ro_rt->rt_flags & (RTF_BLACKHOLE|RTF_BROADCAST)) != 0) 434 goto drop; 435 436 /* 437 * Step 5: outgoing firewall packet processing 438 */ 439 440 /* 441 * Run through list of hooks for output packets. 442 */ 443 if (!PFIL_HOOKED(&inet_pfil_hook)) 444 goto passout; 445 446 if (pfil_run_hooks(&inet_pfil_hook, &m, ifp, PFIL_OUT, NULL) || m == NULL) { 447 goto drop; 448 } 449 450 M_ASSERTVALID(m); 451 M_ASSERTPKTHDR(m); 452 453 ip = mtod(m, struct ip *); 454 dest.s_addr = ip->ip_dst.s_addr; 455 456 /* 457 * Destination address changed? 458 */ 459#ifndef IPFIREWALL_FORWARD 460 if (odest.s_addr != dest.s_addr) { 461#else 462 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 463 if (odest.s_addr != dest.s_addr || fwd_tag != NULL) { 464#endif /* IPFIREWALL_FORWARD */ 465 /* 466 * Is it now for a local address on this host? 467 */ 468#ifndef IPFIREWALL_FORWARD 469 if (in_localip(dest)) { 470#else 471 if (m->m_flags & M_FASTFWD_OURS || in_localip(dest)) { 472#endif /* IPFIREWALL_FORWARD */ 473forwardlocal: 474 /* 475 * Return packet for processing by ip_input(). 476 * Keep host byte order as expected at ip_input's 477 * "ours"-label. 478 */ 479 m->m_flags |= M_FASTFWD_OURS; 480 if (ro.ro_rt) 481 RTFREE(ro.ro_rt); 482 return m; 483 } 484 /* 485 * Redo route lookup with new destination address 486 */ 487#ifdef IPFIREWALL_FORWARD 488 if (fwd_tag) { 489 dest.s_addr = ((struct sockaddr_in *) 490 (fwd_tag + 1))->sin_addr.s_addr; 491 m_tag_delete(m, fwd_tag); 492 } 493#endif /* IPFIREWALL_FORWARD */ 494 RTFREE(ro.ro_rt); 495 if ((dst = ip_findroute(&ro, dest, m)) == NULL) 496 return NULL; /* icmp unreach already sent */ 497 ifp = ro.ro_rt->rt_ifp; 498 } 499 500passout: 501 /* 502 * Step 6: send off the packet 503 */ 504 505 /* 506 * Check if route is dampned (when ARP is unable to resolve) 507 */ 508 if ((ro.ro_rt->rt_flags & RTF_REJECT) && 509 (ro.ro_rt->rt_rmx.rmx_expire == 0 || 510 time_uptime < ro.ro_rt->rt_rmx.rmx_expire)) { 511 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 512 goto consumed; 513 } 514 515#ifndef ALTQ 516 /* 517 * Check if there is enough space in the interface queue 518 */ 519 if ((ifp->if_snd.ifq_len + ip->ip_len / ifp->if_mtu + 1) >= 520 ifp->if_snd.ifq_maxlen) { 521 V_ipstat.ips_odropped++; 522 /* would send source quench here but that is depreciated */ 523 goto drop; 524 } 525#endif 526 527 /* 528 * Check if media link state of interface is not down 529 */ 530 if (ifp->if_link_state == LINK_STATE_DOWN) { 531 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_HOST, 0, 0); 532 goto consumed; 533 } 534 535 /* 536 * Check if packet fits MTU or if hardware will fragment for us 537 */ 538 if (ro.ro_rt->rt_rmx.rmx_mtu) 539 mtu = min(ro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu); 540 else 541 mtu = ifp->if_mtu; 542 543 if (ip->ip_len <= mtu || 544 (ifp->if_hwassist & CSUM_FRAGMENT && (ip->ip_off & IP_DF) == 0)) { 545 /* 546 * Restore packet header fields to original values 547 */ 548 ip->ip_len = htons(ip->ip_len); 549 ip->ip_off = htons(ip->ip_off); 550 /* 551 * Send off the packet via outgoing interface 552 */ 553 error = (*ifp->if_output)(ifp, m, 554 (struct sockaddr *)dst, ro.ro_rt); 555 } else { 556 /* 557 * Handle EMSGSIZE with icmp reply needfrag for TCP MTU discovery 558 */ 559 if (ip->ip_off & IP_DF) { 560 V_ipstat.ips_cantfrag++; 561 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_NEEDFRAG, 562 0, mtu); 563 goto consumed; 564 } else { 565 /* 566 * We have to fragment the packet 567 */ 568 m->m_pkthdr.csum_flags |= CSUM_IP; 569 /* 570 * ip_fragment expects ip_len and ip_off in host byte 571 * order but returns all packets in network byte order 572 */ 573 if (ip_fragment(ip, &m, mtu, ifp->if_hwassist, 574 (~ifp->if_hwassist & CSUM_DELAY_IP))) { 575 goto drop; 576 } 577 KASSERT(m != NULL, ("null mbuf and no error")); 578 /* 579 * Send off the fragments via outgoing interface 580 */ 581 error = 0; 582 do { 583 m0 = m->m_nextpkt; 584 m->m_nextpkt = NULL; 585 586 error = (*ifp->if_output)(ifp, m, 587 (struct sockaddr *)dst, ro.ro_rt); 588 if (error) 589 break; 590 } while ((m = m0) != NULL); 591 if (error) { 592 /* Reclaim remaining fragments */ 593 for (m = m0; m; m = m0) { 594 m0 = m->m_nextpkt; 595 m_freem(m); 596 } 597 } else 598 V_ipstat.ips_fragmented++; 599 } 600 } 601 602 if (error != 0) 603 V_ipstat.ips_odropped++; 604 else { 605 ro.ro_rt->rt_rmx.rmx_pksent++; 606 V_ipstat.ips_forward++; 607 V_ipstat.ips_fastforward++; 608 } 609consumed: 610 RTFREE(ro.ro_rt); 611 return NULL; 612drop: 613 if (m) 614 m_freem(m); 615 if (ro.ro_rt) 616 RTFREE(ro.ro_rt); 617 return NULL; 618} 619