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