1 /* 2 * Copyright (c) 1982, 1986, 1988, 1993 3 * The Regents of the University of California. 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. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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 * $FreeBSD: src/sys/netinet/ip_divert.c,v 1.42.2.6 2003/01/23 21:06:45 sam Exp $ 30 */ 31 32 #define _IP_VHL 33 34 #include "opt_inet.h" 35 #include "opt_ipdivert.h" 36 37 #ifndef INET 38 #error "IPDIVERT requires INET." 39 #endif 40 41 #include <sys/param.h> 42 #include <sys/kernel.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/socket.h> 46 #include <sys/protosw.h> 47 #include <sys/socketvar.h> 48 #include <sys/socketvar2.h> 49 #include <sys/sysctl.h> 50 #include <sys/systm.h> 51 #include <sys/proc.h> 52 #include <sys/priv.h> 53 #include <sys/in_cksum.h> 54 #include <sys/lock.h> 55 #include <sys/msgport.h> 56 57 #include <net/if.h> 58 #include <net/route.h> 59 60 #include <net/netmsg2.h> 61 #include <net/netisr2.h> 62 #include <sys/mplock2.h> 63 64 #include <netinet/in.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/ip.h> 67 #include <netinet/in_pcb.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip_var.h> 70 #include <netinet/ip_divert.h> 71 72 /* 73 * Divert sockets 74 */ 75 76 /* 77 * Allocate enough space to hold a full IP packet 78 */ 79 #define DIVSNDQ (65536 + 100) 80 #define DIVRCVQ (65536 + 100) 81 82 #define DIV_IS_OUTPUT(sin) ((sin) == NULL || (sin)->sin_addr.s_addr == 0) 83 84 #define DIV_OUTPUT 0x10000 85 #define DIV_INPUT 0x20000 86 87 /* 88 * Divert sockets work in conjunction with ipfw, see the divert(4) 89 * manpage for features. 90 * Internally, packets selected by ipfw in ip_input() or ip_output(), 91 * and never diverted before, are passed to the input queue of the 92 * divert socket with a given 'divert_port' number (as specified in 93 * the matching ipfw rule), and they are tagged with a 16 bit cookie 94 * (representing the rule number of the matching ipfw rule), which 95 * is passed to process reading from the socket. 96 * 97 * Packets written to the divert socket are again tagged with a cookie 98 * (usually the same as above) and a destination address. 99 * If the destination address is INADDR_ANY then the packet is 100 * treated as outgoing and sent to ip_output(), otherwise it is 101 * treated as incoming and sent to ip_input(). 102 * In both cases, the packet is tagged with the cookie. 103 * 104 * On reinjection, processing in ip_input() and ip_output() 105 * will be exactly the same as for the original packet, except that 106 * ipfw processing will start at the rule number after the one 107 * written in the cookie (so, tagging a packet with a cookie of 0 108 * will cause it to be effectively considered as a standard packet). 109 */ 110 111 /* Internal variables */ 112 static struct inpcbinfo divcbinfo; 113 static struct inpcbportinfo divcbportinfo; 114 115 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 116 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 117 118 static struct mbuf *ip_divert(struct mbuf *, int, int); 119 120 /* 121 * Initialize divert connection block queue. 122 */ 123 void 124 div_init(void) 125 { 126 in_pcbinfo_init(&divcbinfo, 0, FALSE); 127 in_pcbportinfo_init(&divcbportinfo, 1, 0); 128 /* 129 * XXX We don't use the hash list for divert IP, but it's easier 130 * to allocate a one entry hash list than it is to check all 131 * over the place for hashbase == NULL. 132 */ 133 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); 134 in_pcbportinfo_set(&divcbinfo, &divcbportinfo, 1); 135 divcbinfo.wildcardhashbase = hashinit(1, M_PCB, 136 &divcbinfo.wildcardhashmask); 137 divcbinfo.ipi_size = sizeof(struct inpcb); 138 ip_divert_p = ip_divert; 139 } 140 141 /* 142 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets 143 * with that protocol number to enter the system from the outside. 144 */ 145 int 146 div_input(struct mbuf **mp, int *offp, int proto) 147 { 148 struct mbuf *m = *mp; 149 150 ipstat.ips_noproto++; 151 m_freem(m); 152 return(IPPROTO_DONE); 153 } 154 155 /* 156 * Divert a packet by passing it up to the divert socket at port 'port'. 157 * 158 * Setup generic address and protocol structures for div_input routine, 159 * then pass them along with mbuf chain. 160 */ 161 static void 162 div_packet(struct mbuf *m, int incoming, int port) 163 { 164 struct sockaddr_in divsrc = { sizeof divsrc, AF_INET }; 165 struct inpcb *inp; 166 struct socket *sa; 167 struct m_tag *mtag; 168 struct divert_info *divinfo; 169 u_int16_t nport; 170 171 ASSERT_NETISR0; 172 173 /* Locate the divert info */ 174 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL); 175 divinfo = m_tag_data(mtag); 176 divsrc.sin_port = divinfo->skipto; 177 178 /* 179 * Record receive interface address, if any. 180 * But only for incoming packets. 181 */ 182 divsrc.sin_addr.s_addr = 0; 183 if (incoming) { 184 struct ifaddr_container *ifac; 185 186 /* Find IP address for receive interface */ 187 TAILQ_FOREACH(ifac, &m->m_pkthdr.rcvif->if_addrheads[mycpuid], 188 ifa_link) { 189 struct ifaddr *ifa = ifac->ifa; 190 191 if (ifa->ifa_addr == NULL) 192 continue; 193 if (ifa->ifa_addr->sa_family != AF_INET) 194 continue; 195 divsrc.sin_addr = 196 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 197 break; 198 } 199 } 200 /* 201 * Record the incoming interface name whenever we have one. 202 */ 203 if (m->m_pkthdr.rcvif) { 204 /* 205 * Hide the actual interface name in there in the 206 * sin_zero array. XXX This needs to be moved to a 207 * different sockaddr type for divert, e.g. 208 * sockaddr_div with multiple fields like 209 * sockaddr_dl. Presently we have only 7 bytes 210 * but that will do for now as most interfaces 211 * are 4 or less + 2 or less bytes for unit. 212 * There is probably a faster way of doing this, 213 * possibly taking it from the sockaddr_dl on the iface. 214 * This solves the problem of a P2P link and a LAN interface 215 * having the same address, which can result in the wrong 216 * interface being assigned to the packet when fed back 217 * into the divert socket. Theoretically if the daemon saves 218 * and re-uses the sockaddr_in as suggested in the man pages, 219 * this iface name will come along for the ride. 220 * (see div_output for the other half of this.) 221 */ 222 ksnprintf(divsrc.sin_zero, sizeof divsrc.sin_zero, "%s", 223 m->m_pkthdr.rcvif->if_xname); 224 } 225 226 /* Put packet on socket queue, if any */ 227 sa = NULL; 228 nport = htons((u_int16_t)port); 229 230 /* 231 * Following loop to locate the inpcb is MPSAFE since the inpcb 232 * insertion/removal happens on the same CPU (CPU0). 233 */ 234 LIST_FOREACH(inp, &divcbinfo.pcblisthead, inp_list) { 235 if (inp->inp_flags & INP_PLACEMARKER) 236 continue; 237 if (inp->inp_lport == nport) 238 sa = inp->inp_socket; 239 } 240 if (sa) { 241 lwkt_gettoken(&sa->so_rcv.ssb_token); 242 if (ssb_appendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, m, NULL) == 0) { 243 m_freem(m); 244 soroverflow(sa); 245 } else 246 sorwakeup(sa); 247 lwkt_reltoken(&sa->so_rcv.ssb_token); 248 } else { 249 m_freem(m); 250 ipstat.ips_noproto++; 251 ipstat.ips_delivered--; 252 } 253 } 254 255 static void 256 div_packet_handler(netmsg_t msg) 257 { 258 struct mbuf *m; 259 int port, incoming = 0; 260 261 m = msg->packet.nm_packet; 262 263 port = msg->lmsg.u.ms_result32 & 0xffff; 264 if (msg->lmsg.u.ms_result32 & DIV_INPUT) 265 incoming = 1; 266 div_packet(m, incoming, port); 267 /* no reply, msg embedded in mbuf */ 268 } 269 270 static void 271 divert_packet(struct mbuf *m, int incoming) 272 { 273 struct m_tag *mtag; 274 struct divert_info *divinfo; 275 int port; 276 277 M_ASSERTPKTHDR(m); 278 279 /* Assure header */ 280 if (m->m_len < sizeof(struct ip) && 281 (m = m_pullup(m, sizeof(struct ip))) == NULL) 282 return; 283 284 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL); 285 KASSERT(mtag != NULL, ("%s no divert tag!", __func__)); 286 divinfo = m_tag_data(mtag); 287 288 port = divinfo->port; 289 KASSERT(port != 0, ("%s: port=0", __func__)); 290 291 if (mycpuid != 0) { 292 struct netmsg_packet *nmp; 293 294 nmp = &m->m_hdr.mh_netmsg; 295 netmsg_init(&nmp->base, NULL, &netisr_apanic_rport, 296 0, div_packet_handler); 297 nmp->nm_packet = m; 298 299 nmp->base.lmsg.u.ms_result32 = port; /* port is 16bits */ 300 if (incoming) 301 nmp->base.lmsg.u.ms_result32 |= DIV_INPUT; 302 else 303 nmp->base.lmsg.u.ms_result32 |= DIV_OUTPUT; 304 305 lwkt_sendmsg(netisr_cpuport(0), &nmp->base.lmsg); 306 } else { 307 div_packet(m, incoming, port); 308 } 309 } 310 311 /* 312 * Deliver packet back into the IP processing machinery. 313 * 314 * If no address specified, or address is 0.0.0.0, send to ip_output(); 315 * otherwise, send to ip_input() and mark as having been received on 316 * the interface with that address. 317 */ 318 static int 319 div_output(struct socket *so, struct mbuf *m, 320 struct sockaddr_in *sin, struct mbuf *control) 321 { 322 int error = 0; 323 struct m_tag *mtag; 324 struct divert_info *divinfo; 325 326 ASSERT_NETISR0; 327 328 if (control) 329 m_freem(control); /* XXX */ 330 331 /* 332 * Prepare the tag for divert info. Note that a packet 333 * with a 0 tag in mh_data is effectively untagged, 334 * so we could optimize that case. 335 */ 336 mtag = m_tag_get(PACKET_TAG_IPFW_DIVERT, sizeof(*divinfo), M_NOWAIT); 337 if (mtag == NULL) { 338 error = ENOBUFS; 339 goto cantsend; 340 } 341 m_tag_prepend(m, mtag); 342 343 /* Loopback avoidance and state recovery */ 344 divinfo = m_tag_data(mtag); 345 if (sin) 346 divinfo->skipto = sin->sin_port; 347 else 348 divinfo->skipto = 0; 349 350 /* Reinject packet into the system as incoming or outgoing */ 351 if (DIV_IS_OUTPUT(sin)) { 352 struct ip *const ip = mtod(m, struct ip *); 353 354 /* Don't allow packet length sizes that will crash */ 355 if ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len) { 356 error = EINVAL; 357 goto cantsend; 358 } 359 360 /* Convert fields to host order for ip_output() */ 361 ip->ip_len = ntohs(ip->ip_len); 362 ip->ip_off = ntohs(ip->ip_off); 363 364 /* Send packet to output processing */ 365 ipstat.ips_rawout++; /* XXX */ 366 error = ip_output(m, NULL, NULL, 367 (so->so_options & SO_DONTROUTE) | 368 IP_ALLOWBROADCAST | IP_RAWOUTPUT, 369 NULL, NULL); 370 } else { 371 ip_input(m); 372 } 373 return error; 374 375 cantsend: 376 m_freem(m); 377 return error; 378 } 379 380 static void 381 div_attach(netmsg_t msg) 382 { 383 struct socket *so = msg->attach.base.nm_so; 384 int proto = msg->attach.nm_proto; 385 struct pru_attach_info *ai = msg->attach.nm_ai; 386 struct inpcb *inp; 387 int error; 388 389 ASSERT_NETISR0; 390 391 inp = so->so_pcb; 392 if (inp) 393 panic("div_attach"); 394 error = priv_check_cred(ai->p_ucred, PRIV_ROOT, NULL_CRED_OKAY); 395 if (error) 396 goto out; 397 398 error = soreserve(so, div_sendspace, div_recvspace, ai->sb_rlimit); 399 if (error) 400 goto out; 401 error = in_pcballoc(so, &divcbinfo); 402 if (error) 403 goto out; 404 inp = (struct inpcb *)so->so_pcb; 405 inp->inp_ip_p = proto; 406 inp->inp_flags |= INP_HDRINCL; 407 /* 408 * The socket is always "connected" because 409 * we always know "where" to send the packet. 410 */ 411 sosetstate(so, SS_ISCONNECTED); 412 error = 0; 413 out: 414 lwkt_replymsg(&msg->attach.base.lmsg, error); 415 } 416 417 static void 418 div_detach(netmsg_t msg) 419 { 420 struct socket *so = msg->detach.base.nm_so; 421 struct inpcb *inp; 422 423 ASSERT_NETISR0; 424 425 inp = so->so_pcb; 426 if (inp == NULL) 427 panic("div_detach"); 428 in_pcbdetach(inp); 429 lwkt_replymsg(&msg->detach.base.lmsg, 0); 430 } 431 432 static void 433 div_abort(netmsg_t msg) 434 { 435 /* 436 * Divert socket does not support listen(2), 437 * so this should never be called. 438 */ 439 panic("div_abort is called"); 440 } 441 442 static void 443 div_disconnect(netmsg_t msg) 444 { 445 struct socket *so = msg->disconnect.base.nm_so; 446 int error; 447 448 ASSERT_NETISR0; 449 450 if (so->so_state & SS_ISCONNECTED) { 451 soisdisconnected(so); 452 error = 0; 453 } else { 454 error = ENOTCONN; 455 } 456 lwkt_replymsg(&msg->disconnect.base.lmsg, error); 457 } 458 459 static void 460 div_bind(netmsg_t msg) 461 { 462 struct socket *so = msg->bind.base.nm_so; 463 struct sockaddr *nam = msg->bind.nm_nam; 464 int error; 465 466 ASSERT_NETISR0; 467 468 /* 469 * in_pcbbind assumes that nam is a sockaddr_in 470 * and in_pcbbind requires a valid address. Since divert 471 * sockets don't we need to make sure the address is 472 * filled in properly. 473 * XXX -- divert should not be abusing in_pcbind 474 * and should probably have its own family. 475 */ 476 if (nam->sa_family != AF_INET) { 477 error = EAFNOSUPPORT; 478 } else { 479 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 480 error = in_pcbbind(so->so_pcb, nam, msg->bind.nm_td); 481 } 482 lwkt_replymsg(&msg->bind.base.lmsg, error); 483 } 484 485 static void 486 div_shutdown(netmsg_t msg) 487 { 488 struct socket *so = msg->shutdown.base.nm_so; 489 490 ASSERT_NETISR0; 491 492 socantsendmore(so); 493 494 lwkt_replymsg(&msg->shutdown.base.lmsg, 0); 495 } 496 497 static void 498 div_send(netmsg_t msg) 499 { 500 struct socket *so = msg->send.base.nm_so; 501 struct mbuf *m = msg->send.nm_m; 502 struct sockaddr *nam = msg->send.nm_addr; 503 struct mbuf *control = msg->send.nm_control; 504 int error; 505 506 /* Length check already done in ip_hashfn() */ 507 KASSERT(m->m_len >= sizeof(struct ip), ("IP header not in one mbuf")); 508 509 /* Send packet */ 510 error = div_output(so, m, (struct sockaddr_in *)nam, control); 511 lwkt_replymsg(&msg->send.base.lmsg, error); 512 } 513 514 SYSCTL_DECL(_net_inet_divert); 515 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, &divcbinfo, 1, 516 in_pcblist_range, "S,xinpcb", "List of active divert sockets"); 517 518 struct pr_usrreqs div_usrreqs = { 519 .pru_abort = div_abort, 520 .pru_accept = pr_generic_notsupp, 521 .pru_attach = div_attach, 522 .pru_bind = div_bind, 523 .pru_connect = pr_generic_notsupp, 524 .pru_connect2 = pr_generic_notsupp, 525 .pru_control = in_control_dispatch, 526 .pru_detach = div_detach, 527 .pru_disconnect = div_disconnect, 528 .pru_listen = pr_generic_notsupp, 529 .pru_peeraddr = in_setpeeraddr_dispatch, 530 .pru_rcvd = pr_generic_notsupp, 531 .pru_rcvoob = pr_generic_notsupp, 532 .pru_send = div_send, 533 .pru_sense = pru_sense_null, 534 .pru_shutdown = div_shutdown, 535 .pru_sockaddr = in_setsockaddr_dispatch, 536 .pru_sosend = sosend, 537 .pru_soreceive = soreceive 538 }; 539 540 static struct mbuf * 541 ip_divert_out(struct mbuf *m, int tee) 542 { 543 struct mbuf *clone = NULL; 544 struct ip *ip = mtod(m, struct ip *); 545 546 /* Clone packet if we're doing a 'tee' */ 547 if (tee) 548 clone = m_dup(m, M_NOWAIT); 549 550 /* 551 * XXX 552 * delayed checksums are not currently compatible 553 * with divert sockets. 554 */ 555 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 556 in_delayed_cksum(m); 557 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 558 } 559 560 /* Restore packet header fields to original values */ 561 ip->ip_len = htons(ip->ip_len); 562 ip->ip_off = htons(ip->ip_off); 563 564 /* Deliver packet to divert input routine */ 565 divert_packet(m, 0); 566 567 /* If 'tee', continue with original packet */ 568 return clone; 569 } 570 571 static struct mbuf * 572 ip_divert_in(struct mbuf *m, int tee) 573 { 574 struct mbuf *clone = NULL; 575 struct ip *ip = mtod(m, struct ip *); 576 struct m_tag *mtag; 577 578 if (ip->ip_off & (IP_MF | IP_OFFMASK)) { 579 const struct divert_info *divinfo; 580 u_short frag_off; 581 int hlen; 582 583 /* 584 * Only trust divert info in the fragment 585 * at offset 0. 586 */ 587 frag_off = ip->ip_off << 3; 588 if (frag_off != 0) { 589 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL); 590 m_tag_delete(m, mtag); 591 } 592 593 /* 594 * Attempt reassembly; if it succeeds, proceed. 595 * ip_reass() will return a different mbuf. 596 */ 597 m = ip_reass(m); 598 if (m == NULL) 599 return NULL; 600 ip = mtod(m, struct ip *); 601 602 /* Caller need to redispatch the packet, if it is for us */ 603 m->m_pkthdr.fw_flags |= FW_MBUF_REDISPATCH; 604 605 /* 606 * Get the header length of the reassembled 607 * packet 608 */ 609 hlen = IP_VHL_HL(ip->ip_vhl) << 2; 610 611 /* 612 * Restore original checksum before diverting 613 * packet 614 */ 615 ip->ip_len += hlen; 616 ip->ip_len = htons(ip->ip_len); 617 ip->ip_off = htons(ip->ip_off); 618 ip->ip_sum = 0; 619 if (hlen == sizeof(struct ip)) 620 ip->ip_sum = in_cksum_hdr(ip); 621 else 622 ip->ip_sum = in_cksum(m, hlen); 623 ip->ip_off = ntohs(ip->ip_off); 624 ip->ip_len = ntohs(ip->ip_len); 625 626 /* 627 * Only use the saved divert info 628 */ 629 mtag = m_tag_find(m, PACKET_TAG_IPFW_DIVERT, NULL); 630 if (mtag == NULL) { 631 /* Wrongly configured ipfw */ 632 kprintf("ip_input no divert info\n"); 633 m_freem(m); 634 return NULL; 635 } 636 divinfo = m_tag_data(mtag); 637 tee = divinfo->tee; 638 } 639 640 /* 641 * Divert or tee packet to the divert protocol if 642 * required. 643 */ 644 645 /* Clone packet if we're doing a 'tee' */ 646 if (tee) 647 clone = m_dup(m, M_NOWAIT); 648 649 /* 650 * Restore packet header fields to original 651 * values 652 */ 653 ip->ip_len = htons(ip->ip_len); 654 ip->ip_off = htons(ip->ip_off); 655 656 /* Deliver packet to divert input routine */ 657 divert_packet(m, 1); 658 659 /* Catch invalid reference */ 660 m = NULL; 661 ip = NULL; 662 663 ipstat.ips_delivered++; 664 665 /* If 'tee', continue with original packet */ 666 if (clone != NULL) { 667 /* 668 * Complete processing of the packet. 669 * XXX Better safe than sorry, remove the DIVERT tag. 670 */ 671 mtag = m_tag_find(clone, PACKET_TAG_IPFW_DIVERT, NULL); 672 KKASSERT(mtag != NULL); 673 m_tag_delete(clone, mtag); 674 } 675 return clone; 676 } 677 678 static struct mbuf * 679 ip_divert(struct mbuf *m, int tee, int incoming) 680 { 681 struct mbuf *ret; 682 683 if (incoming) 684 ret = ip_divert_in(m, tee); 685 else 686 ret = ip_divert_out(m, tee); 687 return ret; 688 } 689