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 inpcbhead, divcb); 115 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo); 116 117 #define V_divcb VNET(divcb) 118 #define V_divcbinfo VNET(divcbinfo) 119 120 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 121 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 122 123 static eventhandler_tag ip_divert_event_tag; 124 125 static int div_output_inbound(int fmaily, struct socket *so, struct mbuf *m, 126 struct sockaddr_in *sin); 127 static int div_output_outbound(int family, struct socket *so, struct mbuf *m); 128 129 /* 130 * Initialize divert connection block queue. 131 */ 132 static void 133 div_zone_change(void *tag) 134 { 135 136 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 137 } 138 139 static int 140 div_inpcb_init(void *mem, int size, int flags) 141 { 142 struct inpcb *inp = mem; 143 144 INP_LOCK_INIT(inp, "inp", "divinp"); 145 return (0); 146 } 147 148 static void 149 div_init(void) 150 { 151 152 /* 153 * XXX We don't use the hash list for divert IP, but it's easier to 154 * allocate one-entry hash lists than it is to check all over the 155 * place for hashbase == NULL. 156 */ 157 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb", 158 div_inpcb_init, IPI_HASHFIELDS_NONE); 159 } 160 161 static void 162 div_destroy(void *unused __unused) 163 { 164 165 in_pcbinfo_destroy(&V_divcbinfo); 166 } 167 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 168 div_destroy, NULL); 169 170 /* 171 * IPPROTO_DIVERT is not in the real IP protocol number space; this 172 * function should never be called. Just in case, drop any packets. 173 */ 174 static int 175 div_input(struct mbuf **mp, int *offp, int proto) 176 { 177 struct mbuf *m = *mp; 178 179 KMOD_IPSTAT_INC(ips_noproto); 180 m_freem(m); 181 return (IPPROTO_DONE); 182 } 183 184 /* 185 * Divert a packet by passing it up to the divert socket at port 'port'. 186 * 187 * Setup generic address and protocol structures for div_input routine, 188 * then pass them along with mbuf chain. 189 */ 190 static void 191 divert_packet(struct mbuf *m, bool incoming) 192 { 193 struct ip *ip; 194 struct inpcb *inp; 195 struct socket *sa; 196 u_int16_t nport; 197 struct sockaddr_in divsrc; 198 struct m_tag *mtag; 199 200 NET_EPOCH_ASSERT(); 201 202 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 203 if (mtag == NULL) { 204 m_freem(m); 205 return; 206 } 207 /* Assure header */ 208 if (m->m_len < sizeof(struct ip) && 209 (m = m_pullup(m, sizeof(struct ip))) == NULL) 210 return; 211 ip = mtod(m, struct ip *); 212 213 /* Delayed checksums are currently not compatible with divert. */ 214 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 215 in_delayed_cksum(m); 216 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 217 } 218 #if defined(SCTP) || defined(SCTP_SUPPORT) 219 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 220 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 221 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 222 } 223 #endif 224 bzero(&divsrc, sizeof(divsrc)); 225 divsrc.sin_len = sizeof(divsrc); 226 divsrc.sin_family = AF_INET; 227 /* record matching rule, in host format */ 228 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 229 /* 230 * Record receive interface address, if any. 231 * But only for incoming packets. 232 */ 233 if (incoming) { 234 struct ifaddr *ifa; 235 struct ifnet *ifp; 236 237 /* Sanity check */ 238 M_ASSERTPKTHDR(m); 239 240 /* Find IP address for receive interface */ 241 ifp = m->m_pkthdr.rcvif; 242 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 243 if (ifa->ifa_addr->sa_family != AF_INET) 244 continue; 245 divsrc.sin_addr = 246 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 247 break; 248 } 249 } 250 /* 251 * Record the incoming interface name whenever we have one. 252 */ 253 if (m->m_pkthdr.rcvif) { 254 /* 255 * Hide the actual interface name in there in the 256 * sin_zero array. XXX This needs to be moved to a 257 * different sockaddr type for divert, e.g. 258 * sockaddr_div with multiple fields like 259 * sockaddr_dl. Presently we have only 7 bytes 260 * but that will do for now as most interfaces 261 * are 4 or less + 2 or less bytes for unit. 262 * There is probably a faster way of doing this, 263 * possibly taking it from the sockaddr_dl on the iface. 264 * This solves the problem of a P2P link and a LAN interface 265 * having the same address, which can result in the wrong 266 * interface being assigned to the packet when fed back 267 * into the divert socket. Theoretically if the daemon saves 268 * and re-uses the sockaddr_in as suggested in the man pages, 269 * this iface name will come along for the ride. 270 * (see div_output for the other half of this.) 271 */ 272 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 273 sizeof(divsrc.sin_zero)); 274 } 275 276 /* Put packet on socket queue, if any */ 277 sa = NULL; 278 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 279 CK_LIST_FOREACH(inp, &V_divcb, inp_list) { 280 /* XXX why does only one socket match? */ 281 if (inp->inp_lport == nport) { 282 INP_RLOCK(inp); 283 if (__predict_false(inp->inp_flags2 & INP_FREED)) { 284 INP_RUNLOCK(inp); 285 continue; 286 } 287 sa = inp->inp_socket; 288 SOCKBUF_LOCK(&sa->so_rcv); 289 if (sbappendaddr_locked(&sa->so_rcv, 290 (struct sockaddr *)&divsrc, m, 291 (struct mbuf *)0) == 0) { 292 SOCKBUF_UNLOCK(&sa->so_rcv); 293 sa = NULL; /* force mbuf reclaim below */ 294 } else 295 sorwakeup_locked(sa); 296 INP_RUNLOCK(inp); 297 break; 298 } 299 } 300 if (sa == NULL) { 301 m_freem(m); 302 KMOD_IPSTAT_INC(ips_noproto); 303 KMOD_IPSTAT_DEC(ips_delivered); 304 } 305 } 306 307 /* 308 * Deliver packet back into the IP processing machinery. 309 * 310 * If no address specified, or address is 0.0.0.0, send to ip_output(); 311 * otherwise, send to ip_input() and mark as having been received on 312 * the interface with that address. 313 */ 314 static int 315 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 316 struct mbuf *control) 317 { 318 struct epoch_tracker et; 319 const struct ip *ip; 320 struct m_tag *mtag; 321 struct ipfw_rule_ref *dt; 322 int error, family; 323 324 /* 325 * An mbuf may hasn't come from userland, but we pretend 326 * that it has. 327 */ 328 m->m_pkthdr.rcvif = NULL; 329 m->m_nextpkt = NULL; 330 M_SETFIB(m, so->so_fibnum); 331 332 if (control) 333 m_freem(control); /* XXX */ 334 335 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 336 if (mtag == NULL) { 337 /* this should be normal */ 338 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 339 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 340 if (mtag == NULL) { 341 m_freem(m); 342 return (ENOBUFS); 343 } 344 m_tag_prepend(m, mtag); 345 } 346 dt = (struct ipfw_rule_ref *)(mtag+1); 347 348 /* Loopback avoidance and state recovery */ 349 if (sin) { 350 int i; 351 352 /* set the starting point. We provide a non-zero slot, 353 * but a non_matching chain_id to skip that info and use 354 * the rulenum/rule_id. 355 */ 356 dt->slot = 1; /* dummy, chain_id is invalid */ 357 dt->chain_id = 0; 358 dt->rulenum = sin->sin_port+1; /* host format ? */ 359 dt->rule_id = 0; 360 /* XXX: broken for IPv6 */ 361 /* 362 * Find receive interface with the given name, stuffed 363 * (if it exists) in the sin_zero[] field. 364 * The name is user supplied data so don't trust its size 365 * or that it is zero terminated. 366 */ 367 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 368 ; 369 if ( i > 0 && i < sizeof(sin->sin_zero)) 370 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 371 } 372 373 ip = mtod(m, struct ip *); 374 switch (ip->ip_v) { 375 case IPVERSION: 376 family = AF_INET; 377 break; 378 #ifdef INET6 379 case IPV6_VERSION >> 4: 380 family = AF_INET6; 381 break; 382 #endif 383 default: 384 m_freem(m); 385 return (EAFNOSUPPORT); 386 } 387 388 /* Reinject packet into the system as incoming or outgoing */ 389 NET_EPOCH_ENTER(et); 390 if (!sin || sin->sin_addr.s_addr == 0) { 391 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 392 error = div_output_outbound(family, so, m); 393 } else { 394 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 395 error = div_output_inbound(family, so, m, sin); 396 } 397 NET_EPOCH_EXIT(et); 398 399 if (error != 0) 400 m_freem(m); 401 402 return (error); 403 } 404 405 /* 406 * Sends mbuf @m to the wire via ip[6]_output(). 407 * 408 * Returns 0 on success, @m is consumed. 409 * On failure, returns error code. It is caller responsibility to free @m. 410 */ 411 static int 412 div_output_outbound(int family, struct socket *so, struct mbuf *m) 413 { 414 struct ip *const ip = mtod(m, struct ip *); 415 struct mbuf *options; 416 struct inpcb *inp; 417 int error; 418 419 inp = sotoinpcb(so); 420 INP_RLOCK(inp); 421 switch (family) { 422 case AF_INET: 423 /* 424 * Don't allow both user specified and setsockopt 425 * options, and don't allow packet length sizes that 426 * will crash. 427 */ 428 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 429 inp->inp_options != NULL) || 430 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 431 INP_RUNLOCK(inp); 432 return (EINVAL); 433 } 434 break; 435 #ifdef INET6 436 case AF_INET6: 437 { 438 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 439 440 /* Don't allow packet length sizes that will crash */ 441 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 442 INP_RUNLOCK(inp); 443 return (EINVAL); 444 } 445 break; 446 } 447 #endif 448 } 449 450 /* Send packet to output processing */ 451 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 452 453 #ifdef MAC 454 mac_inpcb_create_mbuf(inp, m); 455 #endif 456 /* 457 * Get ready to inject the packet into ip_output(). 458 * Just in case socket options were specified on the 459 * divert socket, we duplicate them. This is done 460 * to avoid having to hold the PCB locks over the call 461 * to ip_output(), as doing this results in a number of 462 * lock ordering complexities. 463 * 464 * Note that we set the multicast options argument for 465 * ip_output() to NULL since it should be invariant that 466 * they are not present. 467 */ 468 KASSERT(inp->inp_moptions == NULL, 469 ("multicast options set on a divert socket")); 470 /* 471 * XXXCSJP: It is unclear to me whether or not it makes 472 * sense for divert sockets to have options. However, 473 * for now we will duplicate them with the INP locks 474 * held so we can use them in ip_output() without 475 * requring a reference to the pcb. 476 */ 477 options = NULL; 478 if (inp->inp_options != NULL) { 479 options = m_dup(inp->inp_options, M_NOWAIT); 480 if (options == NULL) { 481 INP_RUNLOCK(inp); 482 return (ENOBUFS); 483 } 484 } 485 INP_RUNLOCK(inp); 486 487 error = 0; 488 switch (family) { 489 case AF_INET: 490 error = ip_output(m, options, NULL, 491 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 492 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 493 break; 494 #ifdef INET6 495 case AF_INET6: 496 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 497 break; 498 #endif 499 } 500 if (options != NULL) 501 m_freem(options); 502 503 return (error); 504 } 505 506 /* 507 * Schedules mbuf @m for local processing via IPv4/IPv6 netisr queue. 508 * 509 * Returns 0 on success, @m is consumed. 510 * Returns error code on failure. It is caller responsibility to free @m. 511 */ 512 static int 513 div_output_inbound(int family, struct socket *so, struct mbuf *m, 514 struct sockaddr_in *sin) 515 { 516 const struct ip *ip; 517 struct ifaddr *ifa; 518 519 if (m->m_pkthdr.rcvif == NULL) { 520 /* 521 * No luck with the name, check by IP address. 522 * Clear the port and the ifname to make sure 523 * there are no distractions for ifa_ifwithaddr. 524 */ 525 526 /* XXX: broken for IPv6 */ 527 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 528 sin->sin_port = 0; 529 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 530 if (ifa == NULL) 531 return (EADDRNOTAVAIL); 532 m->m_pkthdr.rcvif = ifa->ifa_ifp; 533 } 534 #ifdef MAC 535 mac_socket_create_mbuf(so, m); 536 #endif 537 /* Send packet to input processing via netisr */ 538 switch (family) { 539 case AF_INET: 540 ip = mtod(m, struct ip *); 541 /* 542 * Restore M_BCAST flag when destination address is 543 * broadcast. It is expected by ip_tryforward(). 544 */ 545 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 546 m->m_flags |= M_MCAST; 547 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 548 m->m_flags |= M_BCAST; 549 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 550 break; 551 #ifdef INET6 552 case AF_INET6: 553 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); 554 break; 555 #endif 556 default: 557 return (EINVAL); 558 } 559 560 return (0); 561 } 562 563 static int 564 div_attach(struct socket *so, int proto, struct thread *td) 565 { 566 struct inpcb *inp; 567 int error; 568 569 inp = sotoinpcb(so); 570 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 571 if (td != NULL) { 572 error = priv_check(td, PRIV_NETINET_DIVERT); 573 if (error) 574 return (error); 575 } 576 error = soreserve(so, div_sendspace, div_recvspace); 577 if (error) 578 return error; 579 INP_INFO_WLOCK(&V_divcbinfo); 580 error = in_pcballoc(so, &V_divcbinfo); 581 if (error) { 582 INP_INFO_WUNLOCK(&V_divcbinfo); 583 return error; 584 } 585 inp = (struct inpcb *)so->so_pcb; 586 INP_INFO_WUNLOCK(&V_divcbinfo); 587 inp->inp_ip_p = proto; 588 inp->inp_vflag |= INP_IPV4; 589 inp->inp_flags |= INP_HDRINCL; 590 INP_WUNLOCK(inp); 591 return 0; 592 } 593 594 static void 595 div_detach(struct socket *so) 596 { 597 struct inpcb *inp; 598 599 inp = sotoinpcb(so); 600 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 601 INP_INFO_WLOCK(&V_divcbinfo); 602 INP_WLOCK(inp); 603 in_pcbdetach(inp); 604 in_pcbfree(inp); 605 INP_INFO_WUNLOCK(&V_divcbinfo); 606 } 607 608 static int 609 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 610 { 611 struct inpcb *inp; 612 int error; 613 614 inp = sotoinpcb(so); 615 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 616 /* in_pcbbind assumes that nam is a sockaddr_in 617 * and in_pcbbind requires a valid address. Since divert 618 * sockets don't we need to make sure the address is 619 * filled in properly. 620 * XXX -- divert should not be abusing in_pcbind 621 * and should probably have its own family. 622 */ 623 if (nam->sa_family != AF_INET) 624 return EAFNOSUPPORT; 625 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 626 INP_INFO_WLOCK(&V_divcbinfo); 627 INP_WLOCK(inp); 628 INP_HASH_WLOCK(&V_divcbinfo); 629 error = in_pcbbind(inp, nam, td->td_ucred); 630 INP_HASH_WUNLOCK(&V_divcbinfo); 631 INP_WUNLOCK(inp); 632 INP_INFO_WUNLOCK(&V_divcbinfo); 633 return error; 634 } 635 636 static int 637 div_shutdown(struct socket *so) 638 { 639 struct inpcb *inp; 640 641 inp = sotoinpcb(so); 642 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 643 INP_WLOCK(inp); 644 socantsendmore(so); 645 INP_WUNLOCK(inp); 646 return 0; 647 } 648 649 static int 650 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 651 struct mbuf *control, struct thread *td) 652 { 653 654 /* Packet must have a header (but that's about it) */ 655 if (m->m_len < sizeof (struct ip) && 656 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 657 KMOD_IPSTAT_INC(ips_toosmall); 658 m_freem(m); 659 return EINVAL; 660 } 661 662 /* Send packet */ 663 return div_output(so, m, (struct sockaddr_in *)nam, control); 664 } 665 666 static void 667 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 668 { 669 struct in_addr faddr; 670 671 faddr = ((struct sockaddr_in *)sa)->sin_addr; 672 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 673 return; 674 if (PRC_IS_REDIRECT(cmd)) 675 return; 676 } 677 678 static int 679 div_pcblist(SYSCTL_HANDLER_ARGS) 680 { 681 struct xinpgen xig; 682 struct epoch_tracker et; 683 struct inpcb *inp; 684 int error; 685 686 if (req->newptr != 0) 687 return EPERM; 688 689 if (req->oldptr == 0) { 690 int n; 691 692 n = V_divcbinfo.ipi_count; 693 n += imax(n / 8, 10); 694 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 695 return 0; 696 } 697 698 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 699 return (error); 700 701 bzero(&xig, sizeof(xig)); 702 xig.xig_len = sizeof xig; 703 xig.xig_count = V_divcbinfo.ipi_count; 704 xig.xig_gen = V_divcbinfo.ipi_gencnt; 705 xig.xig_sogen = so_gencnt; 706 error = SYSCTL_OUT(req, &xig, sizeof xig); 707 if (error) 708 return error; 709 710 NET_EPOCH_ENTER(et); 711 for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead); 712 inp != NULL; 713 inp = CK_LIST_NEXT(inp, inp_list)) { 714 INP_RLOCK(inp); 715 if (inp->inp_gencnt <= xig.xig_gen) { 716 struct xinpcb xi; 717 718 in_pcbtoxinpcb(inp, &xi); 719 INP_RUNLOCK(inp); 720 error = SYSCTL_OUT(req, &xi, sizeof xi); 721 } else 722 INP_RUNLOCK(inp); 723 } 724 NET_EPOCH_EXIT(et); 725 726 if (!error) { 727 /* 728 * Give the user an updated idea of our state. 729 * If the generation differs from what we told 730 * her before, she knows that something happened 731 * while we were processing this request, and it 732 * might be necessary to retry. 733 */ 734 xig.xig_gen = V_divcbinfo.ipi_gencnt; 735 xig.xig_sogen = so_gencnt; 736 xig.xig_count = V_divcbinfo.ipi_count; 737 error = SYSCTL_OUT(req, &xig, sizeof xig); 738 } 739 740 return (error); 741 } 742 743 #ifdef SYSCTL_NODE 744 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, 745 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 746 "IPDIVERT"); 747 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, 748 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 749 NULL, 0, div_pcblist, "S,xinpcb", 750 "List of active divert sockets"); 751 #endif 752 753 struct pr_usrreqs div_usrreqs = { 754 .pru_attach = div_attach, 755 .pru_bind = div_bind, 756 .pru_control = in_control, 757 .pru_detach = div_detach, 758 .pru_peeraddr = in_getpeeraddr, 759 .pru_send = div_send, 760 .pru_shutdown = div_shutdown, 761 .pru_sockaddr = in_getsockaddr, 762 .pru_sosetlabel = in_pcbsosetlabel 763 }; 764 765 struct protosw div_protosw = { 766 .pr_type = SOCK_RAW, 767 .pr_protocol = IPPROTO_DIVERT, 768 .pr_flags = PR_ATOMIC|PR_ADDR, 769 .pr_input = div_input, 770 .pr_ctlinput = div_ctlinput, 771 .pr_ctloutput = ip_ctloutput, 772 .pr_init = div_init, 773 .pr_usrreqs = &div_usrreqs 774 }; 775 776 static int 777 div_modevent(module_t mod, int type, void *unused) 778 { 779 int err = 0; 780 781 switch (type) { 782 case MOD_LOAD: 783 /* 784 * Protocol will be initialized by pf_proto_register(). 785 * We don't have to register ip_protox because we are not 786 * a true IP protocol that goes over the wire. 787 */ 788 err = pf_proto_register(PF_INET, &div_protosw); 789 if (err != 0) 790 return (err); 791 ip_divert_ptr = divert_packet; 792 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, 793 div_zone_change, NULL, EVENTHANDLER_PRI_ANY); 794 break; 795 case MOD_QUIESCE: 796 /* 797 * IPDIVERT may normally not be unloaded because of the 798 * potential race conditions. Tell kldunload we can't be 799 * unloaded unless the unload is forced. 800 */ 801 err = EPERM; 802 break; 803 case MOD_UNLOAD: 804 /* 805 * Forced unload. 806 * 807 * Module ipdivert can only be unloaded if no sockets are 808 * connected. Maybe this can be changed later to forcefully 809 * disconnect any open sockets. 810 * 811 * XXXRW: Note that there is a slight race here, as a new 812 * socket open request could be spinning on the lock and then 813 * we destroy the lock. 814 */ 815 INP_INFO_WLOCK(&V_divcbinfo); 816 if (V_divcbinfo.ipi_count != 0) { 817 err = EBUSY; 818 INP_INFO_WUNLOCK(&V_divcbinfo); 819 break; 820 } 821 ip_divert_ptr = NULL; 822 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 823 INP_INFO_WUNLOCK(&V_divcbinfo); 824 #ifndef VIMAGE 825 div_destroy(NULL); 826 #endif 827 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); 828 break; 829 default: 830 err = EOPNOTSUPP; 831 break; 832 } 833 return err; 834 } 835 836 static moduledata_t ipdivertmod = { 837 "ipdivert", 838 div_modevent, 839 0 840 }; 841 842 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 843 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3); 844 MODULE_VERSION(ipdivert, 1); 845