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 * 4. 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 #if !defined(KLD_MODULE) 34 #include "opt_inet.h" 35 #include "opt_ipfw.h" 36 #include "opt_mac.h" 37 #ifndef INET 38 #error "IPDIVERT requires INET." 39 #endif 40 #ifndef IPFIREWALL 41 #error "IPDIVERT requires IPFIREWALL" 42 #endif 43 #endif 44 45 #include <sys/param.h> 46 #include <sys/kernel.h> 47 #include <sys/lock.h> 48 #include <sys/malloc.h> 49 #include <sys/mbuf.h> 50 #include <sys/module.h> 51 #include <sys/kernel.h> 52 #include <sys/priv.h> 53 #include <sys/proc.h> 54 #include <sys/protosw.h> 55 #include <sys/signalvar.h> 56 #include <sys/socket.h> 57 #include <sys/socketvar.h> 58 #include <sys/sx.h> 59 #include <sys/sysctl.h> 60 #include <sys/systm.h> 61 62 #include <vm/uma.h> 63 64 #include <net/if.h> 65 #include <net/netisr.h> 66 #include <net/route.h> 67 68 #include <netinet/in.h> 69 #include <netinet/in_pcb.h> 70 #include <netinet/in_systm.h> 71 #include <netinet/in_var.h> 72 #include <netinet/ip.h> 73 #include <netinet/ip_divert.h> 74 #include <netinet/ip_var.h> 75 #include <netinet/ip_fw.h> 76 77 #include <security/mac/mac_framework.h> 78 79 /* 80 * Divert sockets 81 */ 82 83 /* 84 * Allocate enough space to hold a full IP packet 85 */ 86 #define DIVSNDQ (65536 + 100) 87 #define DIVRCVQ (65536 + 100) 88 89 /* 90 * Divert sockets work in conjunction with ipfw, see the divert(4) 91 * manpage for features. 92 * Internally, packets selected by ipfw in ip_input() or ip_output(), 93 * and never diverted before, are passed to the input queue of the 94 * divert socket with a given 'divert_port' number (as specified in 95 * the matching ipfw rule), and they are tagged with a 16 bit cookie 96 * (representing the rule number of the matching ipfw rule), which 97 * is passed to process reading from the socket. 98 * 99 * Packets written to the divert socket are again tagged with a cookie 100 * (usually the same as above) and a destination address. 101 * If the destination address is INADDR_ANY then the packet is 102 * treated as outgoing and sent to ip_output(), otherwise it is 103 * treated as incoming and sent to ip_input(). 104 * In both cases, the packet is tagged with the cookie. 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 * ipfw processing will start at the rule number after the one 109 * written in the cookie (so, tagging a packet with a cookie of 0 110 * will cause it to be effectively considered as a standard packet). 111 */ 112 113 /* Internal variables. */ 114 static struct inpcbhead divcb; 115 static struct inpcbinfo divcbinfo; 116 117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 119 120 /* 121 * Initialize divert connection block queue. 122 */ 123 static void 124 div_zone_change(void *tag) 125 { 126 127 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 128 } 129 130 static int 131 div_inpcb_init(void *mem, int size, int flags) 132 { 133 struct inpcb *inp = mem; 134 135 INP_LOCK_INIT(inp, "inp", "divinp"); 136 return (0); 137 } 138 139 static void 140 div_inpcb_fini(void *mem, int size) 141 { 142 struct inpcb *inp = mem; 143 144 INP_LOCK_DESTROY(inp); 145 } 146 147 void 148 div_init(void) 149 { 150 151 INP_INFO_LOCK_INIT(&divcbinfo, "div"); 152 LIST_INIT(&divcb); 153 divcbinfo.ipi_listhead = &divcb; 154 /* 155 * XXX We don't use the hash list for divert IP, but it's easier 156 * to allocate a one entry hash list than it is to check all 157 * over the place for hashbase == NULL. 158 */ 159 divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &divcbinfo.ipi_hashmask); 160 divcbinfo.ipi_porthashbase = hashinit(1, M_PCB, 161 &divcbinfo.ipi_porthashmask); 162 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 163 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, 164 UMA_ZONE_NOFREE); 165 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 166 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, 167 NULL, EVENTHANDLER_PRI_ANY); 168 } 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 void 175 div_input(struct mbuf *m, int off) 176 { 177 ipstat.ips_noproto++; 178 m_freem(m); 179 } 180 181 /* 182 * Divert a packet by passing it up to the divert socket at port 'port'. 183 * 184 * Setup generic address and protocol structures for div_input routine, 185 * then pass them along with mbuf chain. 186 */ 187 static void 188 divert_packet(struct mbuf *m, int incoming) 189 { 190 struct ip *ip; 191 struct inpcb *inp; 192 struct socket *sa; 193 u_int16_t nport; 194 struct sockaddr_in divsrc; 195 struct m_tag *mtag; 196 197 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 198 if (mtag == NULL) { 199 printf("%s: no divert tag\n", __func__); 200 m_freem(m); 201 return; 202 } 203 /* Assure header */ 204 if (m->m_len < sizeof(struct ip) && 205 (m = m_pullup(m, sizeof(struct ip))) == 0) 206 return; 207 ip = mtod(m, struct ip *); 208 209 /* Delayed checksums are currently not compatible with divert. */ 210 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 211 ip->ip_len = ntohs(ip->ip_len); 212 in_delayed_cksum(m); 213 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 214 ip->ip_len = htons(ip->ip_len); 215 } 216 217 /* 218 * Record receive interface address, if any. 219 * But only for incoming packets. 220 */ 221 bzero(&divsrc, sizeof(divsrc)); 222 divsrc.sin_len = sizeof(divsrc); 223 divsrc.sin_family = AF_INET; 224 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 225 if (incoming) { 226 struct ifaddr *ifa; 227 228 /* Sanity check */ 229 M_ASSERTPKTHDR(m); 230 231 /* Find IP address for receive interface */ 232 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 233 if (ifa->ifa_addr->sa_family != AF_INET) 234 continue; 235 divsrc.sin_addr = 236 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 237 break; 238 } 239 } 240 /* 241 * Record the incoming interface name whenever we have one. 242 */ 243 if (m->m_pkthdr.rcvif) { 244 /* 245 * Hide the actual interface name in there in the 246 * sin_zero array. XXX This needs to be moved to a 247 * different sockaddr type for divert, e.g. 248 * sockaddr_div with multiple fields like 249 * sockaddr_dl. Presently we have only 7 bytes 250 * but that will do for now as most interfaces 251 * are 4 or less + 2 or less bytes for unit. 252 * There is probably a faster way of doing this, 253 * possibly taking it from the sockaddr_dl on the iface. 254 * This solves the problem of a P2P link and a LAN interface 255 * having the same address, which can result in the wrong 256 * interface being assigned to the packet when fed back 257 * into the divert socket. Theoretically if the daemon saves 258 * and re-uses the sockaddr_in as suggested in the man pages, 259 * this iface name will come along for the ride. 260 * (see div_output for the other half of this.) 261 */ 262 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 263 sizeof(divsrc.sin_zero)); 264 } 265 266 /* Put packet on socket queue, if any */ 267 sa = NULL; 268 nport = htons((u_int16_t)divert_info(mtag)); 269 INP_INFO_RLOCK(&divcbinfo); 270 LIST_FOREACH(inp, &divcb, inp_list) { 271 INP_LOCK(inp); 272 /* XXX why does only one socket match? */ 273 if (inp->inp_lport == nport) { 274 sa = inp->inp_socket; 275 SOCKBUF_LOCK(&sa->so_rcv); 276 if (sbappendaddr_locked(&sa->so_rcv, 277 (struct sockaddr *)&divsrc, m, 278 (struct mbuf *)0) == 0) { 279 SOCKBUF_UNLOCK(&sa->so_rcv); 280 sa = NULL; /* force mbuf reclaim below */ 281 } else 282 sorwakeup_locked(sa); 283 INP_UNLOCK(inp); 284 break; 285 } 286 INP_UNLOCK(inp); 287 } 288 INP_INFO_RUNLOCK(&divcbinfo); 289 if (sa == NULL) { 290 m_freem(m); 291 ipstat.ips_noproto++; 292 ipstat.ips_delivered--; 293 } 294 } 295 296 /* 297 * Deliver packet back into the IP processing machinery. 298 * 299 * If no address specified, or address is 0.0.0.0, send to ip_output(); 300 * otherwise, send to ip_input() and mark as having been received on 301 * the interface with that address. 302 */ 303 static int 304 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 305 struct mbuf *control) 306 { 307 struct m_tag *mtag; 308 struct divert_tag *dt; 309 int error = 0; 310 struct mbuf *options; 311 312 /* 313 * An mbuf may hasn't come from userland, but we pretend 314 * that it has. 315 */ 316 m->m_pkthdr.rcvif = NULL; 317 m->m_nextpkt = NULL; 318 319 if (control) 320 m_freem(control); /* XXX */ 321 322 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 323 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 324 M_NOWAIT | M_ZERO); 325 if (mtag == NULL) { 326 error = ENOBUFS; 327 goto cantsend; 328 } 329 dt = (struct divert_tag *)(mtag+1); 330 m_tag_prepend(m, mtag); 331 } else 332 dt = (struct divert_tag *)(mtag+1); 333 334 /* Loopback avoidance and state recovery */ 335 if (sin) { 336 int i; 337 338 dt->cookie = sin->sin_port; 339 /* 340 * Find receive interface with the given name, stuffed 341 * (if it exists) in the sin_zero[] field. 342 * The name is user supplied data so don't trust its size 343 * or that it is zero terminated. 344 */ 345 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 346 ; 347 if ( i > 0 && i < sizeof(sin->sin_zero)) 348 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 349 } 350 351 /* Reinject packet into the system as incoming or outgoing */ 352 if (!sin || sin->sin_addr.s_addr == 0) { 353 struct ip *const ip = mtod(m, struct ip *); 354 struct inpcb *inp; 355 356 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 357 INP_INFO_WLOCK(&divcbinfo); 358 inp = sotoinpcb(so); 359 INP_LOCK(inp); 360 /* 361 * Don't allow both user specified and setsockopt options, 362 * and don't allow packet length sizes that will crash 363 */ 364 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 365 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 366 error = EINVAL; 367 INP_UNLOCK(inp); 368 INP_INFO_WUNLOCK(&divcbinfo); 369 m_freem(m); 370 } else { 371 /* Convert fields to host order for ip_output() */ 372 ip->ip_len = ntohs(ip->ip_len); 373 ip->ip_off = ntohs(ip->ip_off); 374 375 /* Send packet to output processing */ 376 ipstat.ips_rawout++; /* XXX */ 377 378 #ifdef MAC 379 mac_inpcb_create_mbuf(inp, m); 380 #endif 381 /* 382 * Get ready to inject the packet into ip_output(). 383 * Just in case socket options were specified on the 384 * divert socket, we duplicate them. This is done 385 * to avoid having to hold the PCB locks over the call 386 * to ip_output(), as doing this results in a number of 387 * lock ordering complexities. 388 * 389 * Note that we set the multicast options argument for 390 * ip_output() to NULL since it should be invariant that 391 * they are not present. 392 */ 393 KASSERT(inp->inp_moptions == NULL, 394 ("multicast options set on a divert socket")); 395 options = NULL; 396 /* 397 * XXXCSJP: It is unclear to me whether or not it makes 398 * sense for divert sockets to have options. However, 399 * for now we will duplicate them with the INP locks 400 * held so we can use them in ip_output() without 401 * requring a reference to the pcb. 402 */ 403 if (inp->inp_options != NULL) { 404 options = m_dup(inp->inp_options, M_DONTWAIT); 405 if (options == NULL) 406 error = ENOBUFS; 407 } 408 INP_UNLOCK(inp); 409 INP_INFO_WUNLOCK(&divcbinfo); 410 if (error == ENOBUFS) { 411 m_freem(m); 412 return (error); 413 } 414 error = ip_output(m, options, NULL, 415 ((so->so_options & SO_DONTROUTE) ? 416 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST | 417 IP_RAWOUTPUT, NULL, NULL); 418 if (options != NULL) 419 m_freem(options); 420 } 421 } else { 422 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 423 if (m->m_pkthdr.rcvif == NULL) { 424 /* 425 * No luck with the name, check by IP address. 426 * Clear the port and the ifname to make sure 427 * there are no distractions for ifa_ifwithaddr. 428 */ 429 struct ifaddr *ifa; 430 431 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 432 sin->sin_port = 0; 433 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 434 if (ifa == NULL) { 435 error = EADDRNOTAVAIL; 436 goto cantsend; 437 } 438 m->m_pkthdr.rcvif = ifa->ifa_ifp; 439 } 440 #ifdef MAC 441 SOCK_LOCK(so); 442 mac_socket_create_mbuf(so, m); 443 SOCK_UNLOCK(so); 444 #endif 445 /* Send packet to input processing via netisr */ 446 netisr_queue(NETISR_IP, m); 447 } 448 449 return error; 450 451 cantsend: 452 m_freem(m); 453 return error; 454 } 455 456 static int 457 div_attach(struct socket *so, int proto, struct thread *td) 458 { 459 struct inpcb *inp; 460 int error; 461 462 inp = sotoinpcb(so); 463 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 464 if (td != NULL) { 465 error = priv_check(td, PRIV_NETINET_DIVERT); 466 if (error) 467 return (error); 468 } 469 error = soreserve(so, div_sendspace, div_recvspace); 470 if (error) 471 return error; 472 INP_INFO_WLOCK(&divcbinfo); 473 error = in_pcballoc(so, &divcbinfo); 474 if (error) { 475 INP_INFO_WUNLOCK(&divcbinfo); 476 return error; 477 } 478 inp = (struct inpcb *)so->so_pcb; 479 INP_INFO_WUNLOCK(&divcbinfo); 480 inp->inp_ip_p = proto; 481 inp->inp_vflag |= INP_IPV4; 482 inp->inp_flags |= INP_HDRINCL; 483 INP_UNLOCK(inp); 484 return 0; 485 } 486 487 static void 488 div_detach(struct socket *so) 489 { 490 struct inpcb *inp; 491 492 inp = sotoinpcb(so); 493 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 494 INP_INFO_WLOCK(&divcbinfo); 495 INP_LOCK(inp); 496 in_pcbdetach(inp); 497 in_pcbfree(inp); 498 INP_INFO_WUNLOCK(&divcbinfo); 499 } 500 501 static int 502 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 503 { 504 struct inpcb *inp; 505 int error; 506 507 inp = sotoinpcb(so); 508 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 509 /* in_pcbbind assumes that nam is a sockaddr_in 510 * and in_pcbbind requires a valid address. Since divert 511 * sockets don't we need to make sure the address is 512 * filled in properly. 513 * XXX -- divert should not be abusing in_pcbind 514 * and should probably have its own family. 515 */ 516 if (nam->sa_family != AF_INET) 517 return EAFNOSUPPORT; 518 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 519 INP_INFO_WLOCK(&divcbinfo); 520 INP_LOCK(inp); 521 error = in_pcbbind(inp, nam, td->td_ucred); 522 INP_UNLOCK(inp); 523 INP_INFO_WUNLOCK(&divcbinfo); 524 return error; 525 } 526 527 static int 528 div_shutdown(struct socket *so) 529 { 530 struct inpcb *inp; 531 532 inp = sotoinpcb(so); 533 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 534 INP_LOCK(inp); 535 socantsendmore(so); 536 INP_UNLOCK(inp); 537 return 0; 538 } 539 540 static int 541 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 542 struct mbuf *control, struct thread *td) 543 { 544 /* Packet must have a header (but that's about it) */ 545 if (m->m_len < sizeof (struct ip) && 546 (m = m_pullup(m, sizeof (struct ip))) == 0) { 547 ipstat.ips_toosmall++; 548 m_freem(m); 549 return EINVAL; 550 } 551 552 /* Send packet */ 553 return div_output(so, m, (struct sockaddr_in *)nam, control); 554 } 555 556 void 557 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 558 { 559 struct in_addr faddr; 560 561 faddr = ((struct sockaddr_in *)sa)->sin_addr; 562 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 563 return; 564 if (PRC_IS_REDIRECT(cmd)) 565 return; 566 } 567 568 static int 569 div_pcblist(SYSCTL_HANDLER_ARGS) 570 { 571 int error, i, n; 572 struct inpcb *inp, **inp_list; 573 inp_gen_t gencnt; 574 struct xinpgen xig; 575 576 /* 577 * The process of preparing the TCB list is too time-consuming and 578 * resource-intensive to repeat twice on every request. 579 */ 580 if (req->oldptr == 0) { 581 n = divcbinfo.ipi_count; 582 req->oldidx = 2 * (sizeof xig) 583 + (n + n/8) * sizeof(struct xinpcb); 584 return 0; 585 } 586 587 if (req->newptr != 0) 588 return EPERM; 589 590 /* 591 * OK, now we're committed to doing something. 592 */ 593 INP_INFO_RLOCK(&divcbinfo); 594 gencnt = divcbinfo.ipi_gencnt; 595 n = divcbinfo.ipi_count; 596 INP_INFO_RUNLOCK(&divcbinfo); 597 598 error = sysctl_wire_old_buffer(req, 599 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 600 if (error != 0) 601 return (error); 602 603 xig.xig_len = sizeof xig; 604 xig.xig_count = n; 605 xig.xig_gen = gencnt; 606 xig.xig_sogen = so_gencnt; 607 error = SYSCTL_OUT(req, &xig, sizeof xig); 608 if (error) 609 return error; 610 611 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 612 if (inp_list == 0) 613 return ENOMEM; 614 615 INP_INFO_RLOCK(&divcbinfo); 616 for (inp = LIST_FIRST(divcbinfo.ipi_listhead), i = 0; inp && i < n; 617 inp = LIST_NEXT(inp, inp_list)) { 618 INP_LOCK(inp); 619 if (inp->inp_gencnt <= gencnt && 620 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 621 inp_list[i++] = inp; 622 INP_UNLOCK(inp); 623 } 624 INP_INFO_RUNLOCK(&divcbinfo); 625 n = i; 626 627 error = 0; 628 for (i = 0; i < n; i++) { 629 inp = inp_list[i]; 630 INP_LOCK(inp); 631 if (inp->inp_gencnt <= gencnt) { 632 struct xinpcb xi; 633 bzero(&xi, sizeof(xi)); 634 xi.xi_len = sizeof xi; 635 /* XXX should avoid extra copy */ 636 bcopy(inp, &xi.xi_inp, sizeof *inp); 637 if (inp->inp_socket) 638 sotoxsocket(inp->inp_socket, &xi.xi_socket); 639 INP_UNLOCK(inp); 640 error = SYSCTL_OUT(req, &xi, sizeof xi); 641 } else 642 INP_UNLOCK(inp); 643 } 644 if (!error) { 645 /* 646 * Give the user an updated idea of our state. 647 * If the generation differs from what we told 648 * her before, she knows that something happened 649 * while we were processing this request, and it 650 * might be necessary to retry. 651 */ 652 INP_INFO_RLOCK(&divcbinfo); 653 xig.xig_gen = divcbinfo.ipi_gencnt; 654 xig.xig_sogen = so_gencnt; 655 xig.xig_count = divcbinfo.ipi_count; 656 INP_INFO_RUNLOCK(&divcbinfo); 657 error = SYSCTL_OUT(req, &xig, sizeof xig); 658 } 659 free(inp_list, M_TEMP); 660 return error; 661 } 662 663 #ifdef SYSCTL_NODE 664 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 665 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 666 div_pcblist, "S,xinpcb", "List of active divert sockets"); 667 #endif 668 669 struct pr_usrreqs div_usrreqs = { 670 .pru_attach = div_attach, 671 .pru_bind = div_bind, 672 .pru_control = in_control, 673 .pru_detach = div_detach, 674 .pru_peeraddr = in_getpeeraddr, 675 .pru_send = div_send, 676 .pru_shutdown = div_shutdown, 677 .pru_sockaddr = in_getsockaddr, 678 .pru_sosetlabel = in_pcbsosetlabel 679 }; 680 681 struct protosw div_protosw = { 682 .pr_type = SOCK_RAW, 683 .pr_protocol = IPPROTO_DIVERT, 684 .pr_flags = PR_ATOMIC|PR_ADDR, 685 .pr_input = div_input, 686 .pr_ctlinput = div_ctlinput, 687 .pr_ctloutput = ip_ctloutput, 688 .pr_init = div_init, 689 .pr_usrreqs = &div_usrreqs 690 }; 691 692 static int 693 div_modevent(module_t mod, int type, void *unused) 694 { 695 int err = 0; 696 int n; 697 698 switch (type) { 699 case MOD_LOAD: 700 /* 701 * Protocol will be initialized by pf_proto_register(). 702 * We don't have to register ip_protox because we are not 703 * a true IP protocol that goes over the wire. 704 */ 705 err = pf_proto_register(PF_INET, &div_protosw); 706 ip_divert_ptr = divert_packet; 707 break; 708 case MOD_QUIESCE: 709 /* 710 * IPDIVERT may normally not be unloaded because of the 711 * potential race conditions. Tell kldunload we can't be 712 * unloaded unless the unload is forced. 713 */ 714 err = EPERM; 715 break; 716 case MOD_UNLOAD: 717 /* 718 * Forced unload. 719 * 720 * Module ipdivert can only be unloaded if no sockets are 721 * connected. Maybe this can be changed later to forcefully 722 * disconnect any open sockets. 723 * 724 * XXXRW: Note that there is a slight race here, as a new 725 * socket open request could be spinning on the lock and then 726 * we destroy the lock. 727 */ 728 INP_INFO_WLOCK(&divcbinfo); 729 n = divcbinfo.ipi_count; 730 if (n != 0) { 731 err = EBUSY; 732 INP_INFO_WUNLOCK(&divcbinfo); 733 break; 734 } 735 ip_divert_ptr = NULL; 736 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 737 INP_INFO_WUNLOCK(&divcbinfo); 738 INP_INFO_LOCK_DESTROY(&divcbinfo); 739 uma_zdestroy(divcbinfo.ipi_zone); 740 break; 741 default: 742 err = EOPNOTSUPP; 743 break; 744 } 745 return err; 746 } 747 748 static moduledata_t ipdivertmod = { 749 "ipdivert", 750 div_modevent, 751 0 752 }; 753 754 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 755 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); 756 MODULE_VERSION(ipdivert, 1); 757