1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 #include "opt_sctp.h" 38 #ifndef INET 39 #error "IPDIVERT requires INET" 40 #endif 41 42 #include <sys/param.h> 43 #include <sys/eventhandler.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/module.h> 49 #include <sys/kernel.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/sysctl.h> 56 #include <net/vnet.h> 57 58 #include <net/if.h> 59 #include <net/if_var.h> 60 #include <net/netisr.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_pcb.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip_var.h> 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #include <netinet6/ip6_var.h> 71 #endif 72 #if defined(SCTP) || defined(SCTP_SUPPORT) 73 #include <netinet/sctp_crc32.h> 74 #endif 75 76 #include <security/mac/mac_framework.h> 77 /* 78 * Divert sockets 79 */ 80 81 /* 82 * Allocate enough space to hold a full IP packet 83 */ 84 #define DIVSNDQ (65536 + 100) 85 #define DIVRCVQ (65536 + 100) 86 87 /* 88 * Divert sockets work in conjunction with ipfw or other packet filters, 89 * see the divert(4) manpage for features. 90 * Packets are selected by the packet filter and tagged with an 91 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by 92 * the packet filter) and information on the matching filter rule for 93 * subsequent reinjection. The divert_port is used to put the packet 94 * on the corresponding divert socket, while the rule number is passed 95 * up (at least partially) as the sin_port in the struct sockaddr. 96 * 97 * Packets written to the divert socket carry in sin_addr a 98 * destination address, and in sin_port the number of the filter rule 99 * after which to continue processing. 100 * If the destination address is INADDR_ANY, the packet is treated as 101 * as outgoing and sent to ip_output(); otherwise it is treated as 102 * incoming and sent to ip_input(). 103 * Further, sin_zero carries some information on the interface, 104 * which can be used in the reinject -- see comments in the code. 105 * 106 * On reinjection, processing in ip_input() and ip_output() 107 * will be exactly the same as for the original packet, except that 108 * packet filter processing will start at the rule number after the one 109 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0 110 * will apply the entire ruleset to the packet). 111 */ 112 113 /* Internal variables. */ 114 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo); 115 #define V_divcbinfo VNET(divcbinfo) 116 117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 119 120 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m, 121 struct sockaddr_in *sin); 122 static int div_output_outbound(int family, struct socket *so, struct mbuf *m); 123 124 /* 125 * Initialize divert connection block queue. 126 */ 127 INPCBSTORAGE_DEFINE(divcbstor, "divinp", "divcb", "div", "divhash"); 128 129 static void 130 div_init(void *arg __unused) 131 { 132 133 /* 134 * XXX We don't use the hash list for divert IP, but it's easier to 135 * allocate one-entry hash lists than it is to check all over the 136 * place for hashbase == NULL. 137 */ 138 in_pcbinfo_init(&V_divcbinfo, &divcbstor, 1, 1); 139 } 140 VNET_SYSINIT(div_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_init, NULL); 141 142 static void 143 div_destroy(void *unused __unused) 144 { 145 146 in_pcbinfo_destroy(&V_divcbinfo); 147 } 148 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, div_destroy, NULL); 149 150 /* 151 * IPPROTO_DIVERT is not in the real IP protocol number space; this 152 * function should never be called. Just in case, drop any packets. 153 */ 154 static int 155 div_input(struct mbuf **mp, int *offp, int proto) 156 { 157 struct mbuf *m = *mp; 158 159 KMOD_IPSTAT_INC(ips_noproto); 160 m_freem(m); 161 return (IPPROTO_DONE); 162 } 163 164 static bool 165 div_port_match(const struct inpcb *inp, void *v) 166 { 167 uint16_t nport = *(uint16_t *)v; 168 169 return (inp->inp_lport == nport); 170 } 171 172 /* 173 * Divert a packet by passing it up to the divert socket at port 'port'. 174 * 175 * Setup generic address and protocol structures for div_input routine, 176 * then pass them along with mbuf chain. 177 */ 178 static void 179 divert_packet(struct mbuf *m, bool incoming) 180 { 181 #if defined(SCTP) || defined(SCTP_SUPPORT) 182 struct ip *ip; 183 #endif 184 struct inpcb *inp; 185 struct socket *sa; 186 u_int16_t nport; 187 struct sockaddr_in divsrc; 188 struct inpcb_iterator inpi = INP_ITERATOR(&V_divcbinfo, 189 INPLOOKUP_RLOCKPCB, div_port_match, &nport); 190 struct m_tag *mtag; 191 192 NET_EPOCH_ASSERT(); 193 194 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 195 if (mtag == NULL) { 196 m_freem(m); 197 return; 198 } 199 /* Assure header */ 200 if (m->m_len < sizeof(struct ip) && 201 (m = m_pullup(m, sizeof(struct ip))) == NULL) 202 return; 203 204 /* Delayed checksums are currently not compatible with divert. */ 205 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 206 in_delayed_cksum(m); 207 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 208 } 209 #if defined(SCTP) || defined(SCTP_SUPPORT) 210 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 211 ip = mtod(m, struct ip *); 212 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 213 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 214 } 215 #endif 216 #ifdef INET6 217 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA_IPV6) { 218 in6_delayed_cksum(m, m->m_pkthdr.len - 219 sizeof(struct ip6_hdr), sizeof(struct ip6_hdr)); 220 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA_IPV6; 221 } 222 #if defined(SCTP) || defined(SCTP_SUPPORT) 223 if (m->m_pkthdr.csum_flags & CSUM_SCTP_IPV6) { 224 sctp_delayed_cksum(m, sizeof(struct ip6_hdr)); 225 m->m_pkthdr.csum_flags &= ~CSUM_SCTP_IPV6; 226 } 227 #endif 228 #endif /* INET6 */ 229 bzero(&divsrc, sizeof(divsrc)); 230 divsrc.sin_len = sizeof(divsrc); 231 divsrc.sin_family = AF_INET; 232 /* record matching rule, in host format */ 233 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 234 /* 235 * Record receive interface address, if any. 236 * But only for incoming packets. 237 */ 238 if (incoming) { 239 struct ifaddr *ifa; 240 struct ifnet *ifp; 241 242 /* Sanity check */ 243 M_ASSERTPKTHDR(m); 244 245 /* Find IP address for receive interface */ 246 ifp = m->m_pkthdr.rcvif; 247 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 248 if (ifa->ifa_addr->sa_family != AF_INET) 249 continue; 250 divsrc.sin_addr = 251 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 252 break; 253 } 254 } 255 /* 256 * Record the incoming interface name whenever we have one. 257 */ 258 if (m->m_pkthdr.rcvif) { 259 /* 260 * Hide the actual interface name in there in the 261 * sin_zero array. XXX This needs to be moved to a 262 * different sockaddr type for divert, e.g. 263 * sockaddr_div with multiple fields like 264 * sockaddr_dl. Presently we have only 7 bytes 265 * but that will do for now as most interfaces 266 * are 4 or less + 2 or less bytes for unit. 267 * There is probably a faster way of doing this, 268 * possibly taking it from the sockaddr_dl on the iface. 269 * This solves the problem of a P2P link and a LAN interface 270 * having the same address, which can result in the wrong 271 * interface being assigned to the packet when fed back 272 * into the divert socket. Theoretically if the daemon saves 273 * and re-uses the sockaddr_in as suggested in the man pages, 274 * this iface name will come along for the ride. 275 * (see div_output for the other half of this.) 276 */ 277 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 278 sizeof(divsrc.sin_zero)); 279 } 280 281 /* Put packet on socket queue, if any */ 282 sa = NULL; 283 /* nport is inp_next's context. */ 284 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 285 while ((inp = inp_next(&inpi)) != NULL) { 286 sa = inp->inp_socket; 287 SOCKBUF_LOCK(&sa->so_rcv); 288 if (sbappendaddr_locked(&sa->so_rcv, 289 (struct sockaddr *)&divsrc, m, NULL) == 0) { 290 soroverflow_locked(sa); 291 sa = NULL; /* force mbuf reclaim below */ 292 } else 293 sorwakeup_locked(sa); 294 /* XXX why does only one socket match? */ 295 INP_RUNLOCK(inp); 296 break; 297 } 298 if (sa == NULL) { 299 m_freem(m); 300 KMOD_IPSTAT_INC(ips_noproto); 301 KMOD_IPSTAT_DEC(ips_delivered); 302 } 303 } 304 305 /* 306 * Deliver packet back into the IP processing machinery. 307 * 308 * If no address specified, or address is 0.0.0.0, send to ip_output(); 309 * otherwise, send to ip_input() and mark as having been received on 310 * the interface with that address. 311 */ 312 static int 313 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 314 struct mbuf *control) 315 { 316 struct epoch_tracker et; 317 const struct ip *ip; 318 struct m_tag *mtag; 319 struct ipfw_rule_ref *dt; 320 int error, family; 321 322 if (control) { 323 m_freem(control); /* XXX */ 324 control = NULL; 325 } 326 327 if (sin != NULL) { 328 if (sin->sin_family != AF_INET) { 329 m_freem(m); 330 return (EAFNOSUPPORT); 331 } 332 if (sin->sin_len != sizeof(*sin)) { 333 m_freem(m); 334 return (EINVAL); 335 } 336 } 337 338 /* 339 * An mbuf may hasn't come from userland, but we pretend 340 * that it has. 341 */ 342 m->m_pkthdr.rcvif = NULL; 343 m->m_nextpkt = NULL; 344 M_SETFIB(m, so->so_fibnum); 345 346 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 347 if (mtag == NULL) { 348 /* this should be normal */ 349 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 350 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 351 if (mtag == NULL) { 352 m_freem(m); 353 return (ENOBUFS); 354 } 355 m_tag_prepend(m, mtag); 356 } 357 dt = (struct ipfw_rule_ref *)(mtag+1); 358 359 /* Loopback avoidance and state recovery */ 360 if (sin) { 361 int i; 362 363 /* set the starting point. We provide a non-zero slot, 364 * but a non_matching chain_id to skip that info and use 365 * the rulenum/rule_id. 366 */ 367 dt->slot = 1; /* dummy, chain_id is invalid */ 368 dt->chain_id = 0; 369 dt->rulenum = sin->sin_port+1; /* host format ? */ 370 dt->rule_id = 0; 371 /* XXX: broken for IPv6 */ 372 /* 373 * Find receive interface with the given name, stuffed 374 * (if it exists) in the sin_zero[] field. 375 * The name is user supplied data so don't trust its size 376 * or that it is zero terminated. 377 */ 378 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 379 ; 380 if ( i > 0 && i < sizeof(sin->sin_zero)) 381 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 382 } 383 384 ip = mtod(m, struct ip *); 385 switch (ip->ip_v) { 386 case IPVERSION: 387 family = AF_INET; 388 break; 389 #ifdef INET6 390 case IPV6_VERSION >> 4: 391 family = AF_INET6; 392 break; 393 #endif 394 default: 395 m_freem(m); 396 return (EAFNOSUPPORT); 397 } 398 399 /* Reinject packet into the system as incoming or outgoing */ 400 NET_EPOCH_ENTER(et); 401 if (!sin || sin->sin_addr.s_addr == 0) { 402 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 403 error = div_output_outbound(family, so, m); 404 } else { 405 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 406 error = div_output_inbound(family, so, m, sin); 407 } 408 NET_EPOCH_EXIT(et); 409 410 return (error); 411 } 412 413 /* 414 * Sends mbuf @m to the wire via ip[6]_output(). 415 * 416 * Returns 0 on success or an errno value on failure. @m is always consumed. 417 */ 418 static int 419 div_output_outbound(int family, struct socket *so, struct mbuf *m) 420 { 421 struct ip *const ip = mtod(m, struct ip *); 422 struct mbuf *options; 423 struct inpcb *inp; 424 int error; 425 426 inp = sotoinpcb(so); 427 INP_RLOCK(inp); 428 switch (family) { 429 case AF_INET: 430 /* 431 * Don't allow both user specified and setsockopt 432 * options, and don't allow packet length sizes that 433 * will crash. 434 */ 435 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 436 inp->inp_options != NULL) || 437 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 438 INP_RUNLOCK(inp); 439 m_freem(m); 440 return (EINVAL); 441 } 442 break; 443 #ifdef INET6 444 case AF_INET6: 445 { 446 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 447 448 /* Don't allow packet length sizes that will crash */ 449 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 450 INP_RUNLOCK(inp); 451 m_freem(m); 452 return (EINVAL); 453 } 454 break; 455 } 456 #endif 457 } 458 459 /* Send packet to output processing */ 460 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 461 462 #ifdef MAC 463 mac_inpcb_create_mbuf(inp, m); 464 #endif 465 /* 466 * Get ready to inject the packet into ip_output(). 467 * Just in case socket options were specified on the 468 * divert socket, we duplicate them. This is done 469 * to avoid having to hold the PCB locks over the call 470 * to ip_output(), as doing this results in a number of 471 * lock ordering complexities. 472 * 473 * Note that we set the multicast options argument for 474 * ip_output() to NULL since it should be invariant that 475 * they are not present. 476 */ 477 KASSERT(inp->inp_moptions == NULL, 478 ("multicast options set on a divert socket")); 479 /* 480 * XXXCSJP: It is unclear to me whether or not it makes 481 * sense for divert sockets to have options. However, 482 * for now we will duplicate them with the INP locks 483 * held so we can use them in ip_output() without 484 * requring a reference to the pcb. 485 */ 486 options = NULL; 487 if (inp->inp_options != NULL) { 488 options = m_dup(inp->inp_options, M_NOWAIT); 489 if (options == NULL) { 490 INP_RUNLOCK(inp); 491 m_freem(m); 492 return (ENOBUFS); 493 } 494 } 495 INP_RUNLOCK(inp); 496 497 error = 0; 498 switch (family) { 499 case AF_INET: 500 error = ip_output(m, options, NULL, 501 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 502 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 503 break; 504 #ifdef INET6 505 case AF_INET6: 506 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 507 break; 508 #endif 509 } 510 if (options != NULL) 511 m_freem(options); 512 513 return (error); 514 } 515 516 /* 517 * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue. 518 * 519 * Returns 0 on success or an errno value on failure. @m is always consumed. 520 */ 521 static int 522 div_output_inbound(int family, struct socket *so, struct mbuf *m, 523 struct sockaddr_in *sin) 524 { 525 const struct ip *ip; 526 struct ifaddr *ifa; 527 528 if (m->m_pkthdr.rcvif == NULL) { 529 /* 530 * No luck with the name, check by IP address. 531 * Clear the port and the ifname to make sure 532 * there are no distractions for ifa_ifwithaddr. 533 */ 534 535 /* XXX: broken for IPv6 */ 536 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 537 sin->sin_port = 0; 538 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 539 if (ifa == NULL) { 540 m_freem(m); 541 return (EADDRNOTAVAIL); 542 } 543 m->m_pkthdr.rcvif = ifa->ifa_ifp; 544 } 545 #ifdef MAC 546 mac_socket_create_mbuf(so, m); 547 #endif 548 /* Send packet to input processing via netisr */ 549 switch (family) { 550 case AF_INET: 551 ip = mtod(m, struct ip *); 552 /* 553 * Restore M_BCAST flag when destination address is 554 * broadcast. It is expected by ip_tryforward(). 555 */ 556 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 557 m->m_flags |= M_MCAST; 558 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 559 m->m_flags |= M_BCAST; 560 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 561 break; 562 #ifdef INET6 563 case AF_INET6: 564 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); 565 break; 566 #endif 567 default: 568 m_freem(m); 569 return (EINVAL); 570 } 571 572 return (0); 573 } 574 575 static int 576 div_attach(struct socket *so, int proto, struct thread *td) 577 { 578 struct inpcb *inp; 579 int error; 580 581 inp = sotoinpcb(so); 582 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 583 if (td != NULL) { 584 error = priv_check(td, PRIV_NETINET_DIVERT); 585 if (error) 586 return (error); 587 } 588 error = soreserve(so, div_sendspace, div_recvspace); 589 if (error) 590 return error; 591 error = in_pcballoc(so, &V_divcbinfo); 592 if (error) 593 return error; 594 inp = (struct inpcb *)so->so_pcb; 595 inp->inp_ip_p = proto; 596 inp->inp_vflag |= INP_IPV4; 597 inp->inp_flags |= INP_HDRINCL; 598 INP_WUNLOCK(inp); 599 return 0; 600 } 601 602 static void 603 div_detach(struct socket *so) 604 { 605 struct inpcb *inp; 606 607 inp = sotoinpcb(so); 608 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 609 INP_WLOCK(inp); 610 in_pcbdetach(inp); 611 in_pcbfree(inp); 612 } 613 614 static int 615 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 616 { 617 struct inpcb *inp; 618 int error; 619 620 inp = sotoinpcb(so); 621 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 622 /* in_pcbbind assumes that nam is a sockaddr_in 623 * and in_pcbbind requires a valid address. Since divert 624 * sockets don't we need to make sure the address is 625 * filled in properly. 626 * XXX -- divert should not be abusing in_pcbind 627 * and should probably have its own family. 628 */ 629 if (nam->sa_family != AF_INET) 630 return EAFNOSUPPORT; 631 if (nam->sa_len != sizeof(struct sockaddr_in)) 632 return EINVAL; 633 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 634 INP_WLOCK(inp); 635 INP_HASH_WLOCK(&V_divcbinfo); 636 error = in_pcbbind(inp, nam, td->td_ucred); 637 INP_HASH_WUNLOCK(&V_divcbinfo); 638 INP_WUNLOCK(inp); 639 return error; 640 } 641 642 static int 643 div_shutdown(struct socket *so) 644 { 645 struct inpcb *inp; 646 647 inp = sotoinpcb(so); 648 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 649 INP_WLOCK(inp); 650 socantsendmore(so); 651 INP_WUNLOCK(inp); 652 return 0; 653 } 654 655 static int 656 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 657 struct mbuf *control, struct thread *td) 658 { 659 660 /* Packet must have a header (but that's about it) */ 661 if (m->m_len < sizeof (struct ip) && 662 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 663 KMOD_IPSTAT_INC(ips_toosmall); 664 if (control != NULL) 665 m_freem(control); 666 m_freem(m); 667 return EINVAL; 668 } 669 670 /* Send packet */ 671 return div_output(so, m, (struct sockaddr_in *)nam, control); 672 } 673 674 static int 675 div_pcblist(SYSCTL_HANDLER_ARGS) 676 { 677 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_divcbinfo, 678 INPLOOKUP_RLOCKPCB); 679 struct xinpgen xig; 680 struct inpcb *inp; 681 int error; 682 683 if (req->newptr != 0) 684 return EPERM; 685 686 if (req->oldptr == 0) { 687 int n; 688 689 n = V_divcbinfo.ipi_count; 690 n += imax(n / 8, 10); 691 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 692 return 0; 693 } 694 695 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 696 return (error); 697 698 bzero(&xig, sizeof(xig)); 699 xig.xig_len = sizeof xig; 700 xig.xig_count = V_divcbinfo.ipi_count; 701 xig.xig_gen = V_divcbinfo.ipi_gencnt; 702 xig.xig_sogen = so_gencnt; 703 error = SYSCTL_OUT(req, &xig, sizeof xig); 704 if (error) 705 return error; 706 707 while ((inp = inp_next(&inpi)) != NULL) { 708 if (inp->inp_gencnt <= xig.xig_gen) { 709 struct xinpcb xi; 710 711 in_pcbtoxinpcb(inp, &xi); 712 error = SYSCTL_OUT(req, &xi, sizeof xi); 713 if (error) { 714 INP_RUNLOCK(inp); 715 break; 716 } 717 } 718 } 719 720 if (!error) { 721 /* 722 * Give the user an updated idea of our state. 723 * If the generation differs from what we told 724 * her before, she knows that something happened 725 * while we were processing this request, and it 726 * might be necessary to retry. 727 */ 728 xig.xig_gen = V_divcbinfo.ipi_gencnt; 729 xig.xig_sogen = so_gencnt; 730 xig.xig_count = V_divcbinfo.ipi_count; 731 error = SYSCTL_OUT(req, &xig, sizeof xig); 732 } 733 734 return (error); 735 } 736 737 #ifdef SYSCTL_NODE 738 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, 739 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 740 "IPDIVERT"); 741 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, 742 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 743 NULL, 0, div_pcblist, "S,xinpcb", 744 "List of active divert sockets"); 745 #endif 746 747 struct pr_usrreqs div_usrreqs = { 748 .pru_attach = div_attach, 749 .pru_bind = div_bind, 750 .pru_control = in_control, 751 .pru_detach = div_detach, 752 .pru_peeraddr = in_getpeeraddr, 753 .pru_send = div_send, 754 .pru_shutdown = div_shutdown, 755 .pru_sockaddr = in_getsockaddr, 756 .pru_sosetlabel = in_pcbsosetlabel 757 }; 758 759 struct protosw div_protosw = { 760 .pr_type = SOCK_RAW, 761 .pr_protocol = IPPROTO_DIVERT, 762 .pr_flags = PR_ATOMIC|PR_ADDR, 763 .pr_input = div_input, 764 .pr_usrreqs = &div_usrreqs 765 }; 766 767 static int 768 div_modevent(module_t mod, int type, void *unused) 769 { 770 int err = 0; 771 772 switch (type) { 773 case MOD_LOAD: 774 /* 775 * Protocol will be initialized by pf_proto_register(). 776 * We don't have to register ip_protox because we are not 777 * a true IP protocol that goes over the wire. 778 */ 779 err = pf_proto_register(PF_INET, &div_protosw); 780 if (err != 0) 781 return (err); 782 ip_divert_ptr = divert_packet; 783 break; 784 case MOD_QUIESCE: 785 /* 786 * IPDIVERT may normally not be unloaded because of the 787 * potential race conditions. Tell kldunload we can't be 788 * unloaded unless the unload is forced. 789 */ 790 err = EPERM; 791 break; 792 case MOD_UNLOAD: 793 /* 794 * Forced unload. 795 * 796 * Module ipdivert can only be unloaded if no sockets are 797 * connected. Maybe this can be changed later to forcefully 798 * disconnect any open sockets. 799 * 800 * XXXRW: Note that there is a slight race here, as a new 801 * socket open request could be spinning on the lock and then 802 * we destroy the lock. 803 */ 804 INP_INFO_WLOCK(&V_divcbinfo); 805 if (V_divcbinfo.ipi_count != 0) { 806 err = EBUSY; 807 INP_INFO_WUNLOCK(&V_divcbinfo); 808 break; 809 } 810 ip_divert_ptr = NULL; 811 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 812 INP_INFO_WUNLOCK(&V_divcbinfo); 813 #ifndef VIMAGE 814 div_destroy(NULL); 815 #endif 816 break; 817 default: 818 err = EOPNOTSUPP; 819 break; 820 } 821 return err; 822 } 823 824 static moduledata_t ipdivertmod = { 825 "ipdivert", 826 div_modevent, 827 0 828 }; 829 830 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 831 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3); 832 MODULE_VERSION(ipdivert, 1); 833