ip_divert.c revision 164033
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 * $FreeBSD: head/sys/netinet/ip_divert.c 164033 2006-11-06 13:42:10Z rwatson $ 30 */ 31 32#if !defined(KLD_MODULE) 33#include "opt_inet.h" 34#include "opt_ipfw.h" 35#include "opt_mac.h" 36#ifndef INET 37#error "IPDIVERT requires INET." 38#endif 39#ifndef IPFIREWALL 40#error "IPDIVERT requires IPFIREWALL" 41#endif 42#endif 43 44#include <sys/param.h> 45#include <sys/kernel.h> 46#include <sys/lock.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/module.h> 50#include <sys/kernel.h> 51#include <sys/priv.h> 52#include <sys/proc.h> 53#include <sys/protosw.h> 54#include <sys/signalvar.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/sx.h> 58#include <sys/sysctl.h> 59#include <sys/systm.h> 60 61#include <vm/uma.h> 62 63#include <net/if.h> 64#include <net/route.h> 65 66#include <netinet/in.h> 67#include <netinet/in_pcb.h> 68#include <netinet/in_systm.h> 69#include <netinet/in_var.h> 70#include <netinet/ip.h> 71#include <netinet/ip_divert.h> 72#include <netinet/ip_var.h> 73#include <netinet/ip_fw.h> 74 75#include <security/mac/mac_framework.h> 76 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, see the divert(4) 89 * manpage for features. 90 * Internally, packets selected by ipfw in ip_input() or ip_output(), 91 * and never diverted before, are passed to the input queue of the 92 * divert socket with a given 'divert_port' number (as specified in 93 * the matching ipfw rule), and they are tagged with a 16 bit cookie 94 * (representing the rule number of the matching ipfw rule), which 95 * is passed to process reading from the socket. 96 * 97 * Packets written to the divert socket are again tagged with a cookie 98 * (usually the same as above) and a destination address. 99 * If the destination address is INADDR_ANY then the packet is 100 * treated as outgoing and sent to ip_output(), otherwise it is 101 * treated as incoming and sent to ip_input(). 102 * In both cases, the packet is tagged with the cookie. 103 * 104 * On reinjection, processing in ip_input() and ip_output() 105 * will be exactly the same as for the original packet, except that 106 * ipfw processing will start at the rule number after the one 107 * written in the cookie (so, tagging a packet with a cookie of 0 108 * will cause it to be effectively considered as a standard packet). 109 */ 110 111/* Internal variables. */ 112static struct inpcbhead divcb; 113static struct inpcbinfo divcbinfo; 114 115static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 116static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 117 118/* 119 * Initialize divert connection block queue. 120 */ 121static void 122div_zone_change(void *tag) 123{ 124 125 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 126} 127 128static int 129div_inpcb_init(void *mem, int size, int flags) 130{ 131 struct inpcb *inp = (struct inpcb *) mem; 132 INP_LOCK_INIT(inp, "inp", "divinp"); 133 return (0); 134} 135 136static void 137div_inpcb_fini(void *mem, int size) 138{ 139 struct inpcb *inp = (struct inpcb *) mem; 140 INP_LOCK_DESTROY(inp); 141} 142 143 144void 145div_init(void) 146{ 147 INP_INFO_LOCK_INIT(&divcbinfo, "div"); 148 LIST_INIT(&divcb); 149 divcbinfo.listhead = &divcb; 150 /* 151 * XXX We don't use the hash list for divert IP, but it's easier 152 * to allocate a one entry hash list than it is to check all 153 * over the place for hashbase == NULL. 154 */ 155 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); 156 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask); 157 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 158 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 159 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 160 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, 161 NULL, EVENTHANDLER_PRI_ANY); 162} 163 164/* 165 * IPPROTO_DIVERT is not in the real IP protocol number space; this 166 * function should never be called. Just in case, drop any packets. 167 */ 168void 169div_input(struct mbuf *m, int off) 170{ 171 ipstat.ips_noproto++; 172 m_freem(m); 173} 174 175/* 176 * Divert a packet by passing it up to the divert socket at port 'port'. 177 * 178 * Setup generic address and protocol structures for div_input routine, 179 * then pass them along with mbuf chain. 180 */ 181static void 182divert_packet(struct mbuf *m, int incoming) 183{ 184 struct ip *ip; 185 struct inpcb *inp; 186 struct socket *sa; 187 u_int16_t nport; 188 struct sockaddr_in divsrc; 189 struct m_tag *mtag; 190 191 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 192 if (mtag == NULL) { 193 printf("%s: no divert tag\n", __func__); 194 m_freem(m); 195 return; 196 } 197 /* Assure header */ 198 if (m->m_len < sizeof(struct ip) && 199 (m = m_pullup(m, sizeof(struct ip))) == 0) 200 return; 201 ip = mtod(m, struct ip *); 202 203 /* Delayed checksums are currently not compatible with divert. */ 204 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 205 ip->ip_len = ntohs(ip->ip_len); 206 in_delayed_cksum(m); 207 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 208 ip->ip_len = htons(ip->ip_len); 209 } 210 211 /* 212 * Record receive interface address, if any. 213 * But only for incoming packets. 214 */ 215 bzero(&divsrc, sizeof(divsrc)); 216 divsrc.sin_len = sizeof(divsrc); 217 divsrc.sin_family = AF_INET; 218 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 219 if (incoming) { 220 struct ifaddr *ifa; 221 222 /* Sanity check */ 223 M_ASSERTPKTHDR(m); 224 225 /* Find IP address for receive interface */ 226 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 227 if (ifa->ifa_addr->sa_family != AF_INET) 228 continue; 229 divsrc.sin_addr = 230 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 231 break; 232 } 233 } 234 /* 235 * Record the incoming interface name whenever we have one. 236 */ 237 if (m->m_pkthdr.rcvif) { 238 /* 239 * Hide the actual interface name in there in the 240 * sin_zero array. XXX This needs to be moved to a 241 * different sockaddr type for divert, e.g. 242 * sockaddr_div with multiple fields like 243 * sockaddr_dl. Presently we have only 7 bytes 244 * but that will do for now as most interfaces 245 * are 4 or less + 2 or less bytes for unit. 246 * There is probably a faster way of doing this, 247 * possibly taking it from the sockaddr_dl on the iface. 248 * This solves the problem of a P2P link and a LAN interface 249 * having the same address, which can result in the wrong 250 * interface being assigned to the packet when fed back 251 * into the divert socket. Theoretically if the daemon saves 252 * and re-uses the sockaddr_in as suggested in the man pages, 253 * this iface name will come along for the ride. 254 * (see div_output for the other half of this.) 255 */ 256 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 257 sizeof(divsrc.sin_zero)); 258 } 259 260 /* Put packet on socket queue, if any */ 261 sa = NULL; 262 nport = htons((u_int16_t)divert_info(mtag)); 263 INP_INFO_RLOCK(&divcbinfo); 264 LIST_FOREACH(inp, &divcb, inp_list) { 265 INP_LOCK(inp); 266 /* XXX why does only one socket match? */ 267 if (inp->inp_lport == nport) { 268 sa = inp->inp_socket; 269 SOCKBUF_LOCK(&sa->so_rcv); 270 if (sbappendaddr_locked(&sa->so_rcv, 271 (struct sockaddr *)&divsrc, m, 272 (struct mbuf *)0) == 0) { 273 SOCKBUF_UNLOCK(&sa->so_rcv); 274 sa = NULL; /* force mbuf reclaim below */ 275 } else 276 sorwakeup_locked(sa); 277 INP_UNLOCK(inp); 278 break; 279 } 280 INP_UNLOCK(inp); 281 } 282 INP_INFO_RUNLOCK(&divcbinfo); 283 if (sa == NULL) { 284 m_freem(m); 285 ipstat.ips_noproto++; 286 ipstat.ips_delivered--; 287 } 288} 289 290/* 291 * Deliver packet back into the IP processing machinery. 292 * 293 * If no address specified, or address is 0.0.0.0, send to ip_output(); 294 * otherwise, send to ip_input() and mark as having been received on 295 * the interface with that address. 296 */ 297static int 298div_output(struct socket *so, struct mbuf *m, 299 struct sockaddr_in *sin, struct mbuf *control) 300{ 301 struct m_tag *mtag; 302 struct divert_tag *dt; 303 int error = 0; 304 305 /* 306 * An mbuf may hasn't come from userland, but we pretend 307 * that it has. 308 */ 309 m->m_pkthdr.rcvif = NULL; 310 m->m_nextpkt = NULL; 311 312 if (control) 313 m_freem(control); /* XXX */ 314 315 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 316 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 317 M_NOWAIT | M_ZERO); 318 if (mtag == NULL) { 319 error = ENOBUFS; 320 goto cantsend; 321 } 322 dt = (struct divert_tag *)(mtag+1); 323 m_tag_prepend(m, mtag); 324 } else 325 dt = (struct divert_tag *)(mtag+1); 326 327 /* Loopback avoidance and state recovery */ 328 if (sin) { 329 int i; 330 331 dt->cookie = sin->sin_port; 332 /* 333 * Find receive interface with the given name, stuffed 334 * (if it exists) in the sin_zero[] field. 335 * The name is user supplied data so don't trust its size 336 * or that it is zero terminated. 337 */ 338 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 339 ; 340 if ( i > 0 && i < sizeof(sin->sin_zero)) 341 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 342 } 343 344 /* Reinject packet into the system as incoming or outgoing */ 345 if (!sin || sin->sin_addr.s_addr == 0) { 346 struct ip *const ip = mtod(m, struct ip *); 347 struct inpcb *inp; 348 349 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 350 INP_INFO_WLOCK(&divcbinfo); 351 inp = sotoinpcb(so); 352 INP_LOCK(inp); 353 /* 354 * Don't allow both user specified and setsockopt options, 355 * and don't allow packet length sizes that will crash 356 */ 357 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 358 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 359 error = EINVAL; 360 m_freem(m); 361 } else { 362 /* Convert fields to host order for ip_output() */ 363 ip->ip_len = ntohs(ip->ip_len); 364 ip->ip_off = ntohs(ip->ip_off); 365 366 /* Send packet to output processing */ 367 ipstat.ips_rawout++; /* XXX */ 368 369#ifdef MAC 370 mac_create_mbuf_from_inpcb(inp, m); 371#endif 372 error = ip_output(m, 373 inp->inp_options, NULL, 374 ((so->so_options & SO_DONTROUTE) ? 375 IP_ROUTETOIF : 0) | 376 IP_ALLOWBROADCAST | IP_RAWOUTPUT, 377 inp->inp_moptions, NULL); 378 } 379 INP_UNLOCK(inp); 380 INP_INFO_WUNLOCK(&divcbinfo); 381 } else { 382 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 383 if (m->m_pkthdr.rcvif == NULL) { 384 /* 385 * No luck with the name, check by IP address. 386 * Clear the port and the ifname to make sure 387 * there are no distractions for ifa_ifwithaddr. 388 */ 389 struct ifaddr *ifa; 390 391 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 392 sin->sin_port = 0; 393 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 394 if (ifa == NULL) { 395 error = EADDRNOTAVAIL; 396 goto cantsend; 397 } 398 m->m_pkthdr.rcvif = ifa->ifa_ifp; 399 } 400#ifdef MAC 401 SOCK_LOCK(so); 402 mac_create_mbuf_from_socket(so, m); 403 SOCK_UNLOCK(so); 404#endif 405 /* Send packet to input processing */ 406 ip_input(m); 407 } 408 409 return error; 410 411cantsend: 412 m_freem(m); 413 return error; 414} 415 416static int 417div_attach(struct socket *so, int proto, struct thread *td) 418{ 419 struct inpcb *inp; 420 int error; 421 422 inp = sotoinpcb(so); 423 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 424 if (td != NULL) { 425 error = priv_check(td, PRIV_NETINET_DIVERT); 426 if (error) 427 return (error); 428 } 429 error = soreserve(so, div_sendspace, div_recvspace); 430 if (error) 431 return error; 432 INP_INFO_WLOCK(&divcbinfo); 433 error = in_pcballoc(so, &divcbinfo); 434 if (error) { 435 INP_INFO_WUNLOCK(&divcbinfo); 436 return error; 437 } 438 inp = (struct inpcb *)so->so_pcb; 439 INP_INFO_WUNLOCK(&divcbinfo); 440 inp->inp_ip_p = proto; 441 inp->inp_vflag |= INP_IPV4; 442 inp->inp_flags |= INP_HDRINCL; 443 INP_UNLOCK(inp); 444 return 0; 445} 446 447static void 448div_detach(struct socket *so) 449{ 450 struct inpcb *inp; 451 452 inp = sotoinpcb(so); 453 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 454 INP_INFO_WLOCK(&divcbinfo); 455 INP_LOCK(inp); 456 in_pcbdetach(inp); 457 in_pcbfree(inp); 458 INP_INFO_WUNLOCK(&divcbinfo); 459} 460 461static int 462div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 463{ 464 struct inpcb *inp; 465 int error; 466 467 inp = sotoinpcb(so); 468 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 469 /* in_pcbbind assumes that nam is a sockaddr_in 470 * and in_pcbbind requires a valid address. Since divert 471 * sockets don't we need to make sure the address is 472 * filled in properly. 473 * XXX -- divert should not be abusing in_pcbind 474 * and should probably have its own family. 475 */ 476 if (nam->sa_family != AF_INET) 477 return EAFNOSUPPORT; 478 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 479 INP_INFO_WLOCK(&divcbinfo); 480 INP_LOCK(inp); 481 error = in_pcbbind(inp, nam, td->td_ucred); 482 INP_UNLOCK(inp); 483 INP_INFO_WUNLOCK(&divcbinfo); 484 return error; 485} 486 487static int 488div_shutdown(struct socket *so) 489{ 490 struct inpcb *inp; 491 492 inp = sotoinpcb(so); 493 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 494 INP_LOCK(inp); 495 socantsendmore(so); 496 INP_UNLOCK(inp); 497 return 0; 498} 499 500static int 501div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 502 struct mbuf *control, struct thread *td) 503{ 504 /* Packet must have a header (but that's about it) */ 505 if (m->m_len < sizeof (struct ip) && 506 (m = m_pullup(m, sizeof (struct ip))) == 0) { 507 ipstat.ips_toosmall++; 508 m_freem(m); 509 return EINVAL; 510 } 511 512 /* Send packet */ 513 return div_output(so, m, (struct sockaddr_in *)nam, control); 514} 515 516void 517div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 518{ 519 struct in_addr faddr; 520 521 faddr = ((struct sockaddr_in *)sa)->sin_addr; 522 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 523 return; 524 if (PRC_IS_REDIRECT(cmd)) 525 return; 526} 527 528static int 529div_pcblist(SYSCTL_HANDLER_ARGS) 530{ 531 int error, i, n; 532 struct inpcb *inp, **inp_list; 533 inp_gen_t gencnt; 534 struct xinpgen xig; 535 536 /* 537 * The process of preparing the TCB list is too time-consuming and 538 * resource-intensive to repeat twice on every request. 539 */ 540 if (req->oldptr == 0) { 541 n = divcbinfo.ipi_count; 542 req->oldidx = 2 * (sizeof xig) 543 + (n + n/8) * sizeof(struct xinpcb); 544 return 0; 545 } 546 547 if (req->newptr != 0) 548 return EPERM; 549 550 /* 551 * OK, now we're committed to doing something. 552 */ 553 INP_INFO_RLOCK(&divcbinfo); 554 gencnt = divcbinfo.ipi_gencnt; 555 n = divcbinfo.ipi_count; 556 INP_INFO_RUNLOCK(&divcbinfo); 557 558 error = sysctl_wire_old_buffer(req, 559 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 560 if (error != 0) 561 return (error); 562 563 xig.xig_len = sizeof xig; 564 xig.xig_count = n; 565 xig.xig_gen = gencnt; 566 xig.xig_sogen = so_gencnt; 567 error = SYSCTL_OUT(req, &xig, sizeof xig); 568 if (error) 569 return error; 570 571 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 572 if (inp_list == 0) 573 return ENOMEM; 574 575 INP_INFO_RLOCK(&divcbinfo); 576 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n; 577 inp = LIST_NEXT(inp, inp_list)) { 578 INP_LOCK(inp); 579 if (inp->inp_gencnt <= gencnt && 580 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 581 inp_list[i++] = inp; 582 INP_UNLOCK(inp); 583 } 584 INP_INFO_RUNLOCK(&divcbinfo); 585 n = i; 586 587 error = 0; 588 for (i = 0; i < n; i++) { 589 inp = inp_list[i]; 590 INP_LOCK(inp); 591 if (inp->inp_gencnt <= gencnt) { 592 struct xinpcb xi; 593 bzero(&xi, sizeof(xi)); 594 xi.xi_len = sizeof xi; 595 /* XXX should avoid extra copy */ 596 bcopy(inp, &xi.xi_inp, sizeof *inp); 597 if (inp->inp_socket) 598 sotoxsocket(inp->inp_socket, &xi.xi_socket); 599 INP_UNLOCK(inp); 600 error = SYSCTL_OUT(req, &xi, sizeof xi); 601 } else 602 INP_UNLOCK(inp); 603 } 604 if (!error) { 605 /* 606 * Give the user an updated idea of our state. 607 * If the generation differs from what we told 608 * her before, she knows that something happened 609 * while we were processing this request, and it 610 * might be necessary to retry. 611 */ 612 INP_INFO_RLOCK(&divcbinfo); 613 xig.xig_gen = divcbinfo.ipi_gencnt; 614 xig.xig_sogen = so_gencnt; 615 xig.xig_count = divcbinfo.ipi_count; 616 INP_INFO_RUNLOCK(&divcbinfo); 617 error = SYSCTL_OUT(req, &xig, sizeof xig); 618 } 619 free(inp_list, M_TEMP); 620 return error; 621} 622 623/* 624 * This is the wrapper function for in_setsockaddr. We just pass down 625 * the pcbinfo for in_setpeeraddr to lock. 626 */ 627static int 628div_sockaddr(struct socket *so, struct sockaddr **nam) 629{ 630 return (in_setsockaddr(so, nam, &divcbinfo)); 631} 632 633/* 634 * This is the wrapper function for in_setpeeraddr. We just pass down 635 * the pcbinfo for in_setpeeraddr to lock. 636 */ 637static int 638div_peeraddr(struct socket *so, struct sockaddr **nam) 639{ 640 return (in_setpeeraddr(so, nam, &divcbinfo)); 641} 642 643#ifdef SYSCTL_NODE 644SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 645SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 646 div_pcblist, "S,xinpcb", "List of active divert sockets"); 647#endif 648 649struct pr_usrreqs div_usrreqs = { 650 .pru_attach = div_attach, 651 .pru_bind = div_bind, 652 .pru_control = in_control, 653 .pru_detach = div_detach, 654 .pru_peeraddr = div_peeraddr, 655 .pru_send = div_send, 656 .pru_shutdown = div_shutdown, 657 .pru_sockaddr = div_sockaddr, 658 .pru_sosetlabel = in_pcbsosetlabel 659}; 660 661struct protosw div_protosw = { 662 .pr_type = SOCK_RAW, 663 .pr_protocol = IPPROTO_DIVERT, 664 .pr_flags = PR_ATOMIC|PR_ADDR, 665 .pr_input = div_input, 666 .pr_ctlinput = div_ctlinput, 667 .pr_ctloutput = ip_ctloutput, 668 .pr_init = div_init, 669 .pr_usrreqs = &div_usrreqs 670}; 671 672static int 673div_modevent(module_t mod, int type, void *unused) 674{ 675 int err = 0; 676 int n; 677 678 switch (type) { 679 case MOD_LOAD: 680 /* 681 * Protocol will be initialized by pf_proto_register(). 682 * We don't have to register ip_protox because we are not 683 * a true IP protocol that goes over the wire. 684 */ 685 err = pf_proto_register(PF_INET, &div_protosw); 686 ip_divert_ptr = divert_packet; 687 break; 688 case MOD_QUIESCE: 689 /* 690 * IPDIVERT may normally not be unloaded because of the 691 * potential race conditions. Tell kldunload we can't be 692 * unloaded unless the unload is forced. 693 */ 694 err = EPERM; 695 break; 696 case MOD_UNLOAD: 697 /* 698 * Forced unload. 699 * 700 * Module ipdivert can only be unloaded if no sockets are 701 * connected. Maybe this can be changed later to forcefully 702 * disconnect any open sockets. 703 * 704 * XXXRW: Note that there is a slight race here, as a new 705 * socket open request could be spinning on the lock and then 706 * we destroy the lock. 707 */ 708 INP_INFO_WLOCK(&divcbinfo); 709 n = divcbinfo.ipi_count; 710 if (n != 0) { 711 err = EBUSY; 712 INP_INFO_WUNLOCK(&divcbinfo); 713 break; 714 } 715 ip_divert_ptr = NULL; 716 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 717 INP_INFO_WUNLOCK(&divcbinfo); 718 INP_INFO_LOCK_DESTROY(&divcbinfo); 719 uma_zdestroy(divcbinfo.ipi_zone); 720 break; 721 default: 722 err = EOPNOTSUPP; 723 break; 724 } 725 return err; 726} 727 728static moduledata_t ipdivertmod = { 729 "ipdivert", 730 div_modevent, 731 0 732}; 733 734DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 735MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); 736MODULE_VERSION(ipdivert, 1); 737