ip_divert.c revision 183982
1128612Sphk/*- 2128612Sphk * Copyright (c) 1982, 1986, 1988, 1993 3128612Sphk * The Regents of the University of California. All rights reserved. 4139103Sru * 5128612Sphk * Redistribution and use in source and binary forms, with or without 6128612Sphk * modification, are permitted provided that the following conditions 7128612Sphk * 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: head/sys/netinet/ip_divert.c 183982 2008-10-17 16:26:16Z bz $"); 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#include <sys/vimage.h> 62 63#include <vm/uma.h> 64 65#include <net/if.h> 66#include <net/netisr.h> 67#include <net/route.h> 68 69#include <netinet/in.h> 70#include <netinet/in_pcb.h> 71#include <netinet/in_systm.h> 72#include <netinet/in_var.h> 73#include <netinet/ip.h> 74#include <netinet/ip_divert.h> 75#include <netinet/ip_var.h> 76#include <netinet/ip_fw.h> 77 78#include <security/mac/mac_framework.h> 79 80/* 81 * Divert sockets 82 */ 83 84/* 85 * Allocate enough space to hold a full IP packet 86 */ 87#define DIVSNDQ (65536 + 100) 88#define DIVRCVQ (65536 + 100) 89 90/* 91 * Divert sockets work in conjunction with ipfw, see the divert(4) 92 * manpage for features. 93 * Internally, packets selected by ipfw in ip_input() or ip_output(), 94 * and never diverted before, are passed to the input queue of the 95 * divert socket with a given 'divert_port' number (as specified in 96 * the matching ipfw rule), and they are tagged with a 16 bit cookie 97 * (representing the rule number of the matching ipfw rule), which 98 * is passed to process reading from the socket. 99 * 100 * Packets written to the divert socket are again tagged with a cookie 101 * (usually the same as above) and a destination address. 102 * If the destination address is INADDR_ANY then the packet is 103 * treated as outgoing and sent to ip_output(), otherwise it is 104 * treated as incoming and sent to ip_input(). 105 * In both cases, the packet is tagged with the cookie. 106 * 107 * On reinjection, processing in ip_input() and ip_output() 108 * will be exactly the same as for the original packet, except that 109 * ipfw processing will start at the rule number after the one 110 * written in the cookie (so, tagging a packet with a cookie of 0 111 * will cause it to be effectively considered as a standard packet). 112 */ 113 114/* Internal variables. */ 115static struct inpcbhead divcb; 116static struct inpcbinfo divcbinfo; 117 118static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 119static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 120 121/* 122 * Initialize divert connection block queue. 123 */ 124static void 125div_zone_change(void *tag) 126{ 127 128 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 129} 130 131static int 132div_inpcb_init(void *mem, int size, int flags) 133{ 134 struct inpcb *inp = mem; 135 136 INP_LOCK_INIT(inp, "inp", "divinp"); 137 return (0); 138} 139 140static void 141div_inpcb_fini(void *mem, int size) 142{ 143 struct inpcb *inp = mem; 144 145 INP_LOCK_DESTROY(inp); 146} 147 148void 149div_init(void) 150{ 151 INIT_VNET_INET(curvnet); 152 153 INP_INFO_LOCK_INIT(&V_divcbinfo, "div"); 154 LIST_INIT(&V_divcb); 155 V_divcbinfo.ipi_listhead = &V_divcb; 156 /* 157 * XXX We don't use the hash list for divert IP, but it's easier 158 * to allocate a one entry hash list than it is to check all 159 * over the place for hashbase == NULL. 160 */ 161 V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask); 162 V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB, 163 &V_divcbinfo.ipi_porthashmask); 164 V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 165 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, 166 UMA_ZONE_NOFREE); 167 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 168 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, 169 NULL, EVENTHANDLER_PRI_ANY); 170} 171 172/* 173 * IPPROTO_DIVERT is not in the real IP protocol number space; this 174 * function should never be called. Just in case, drop any packets. 175 */ 176void 177div_input(struct mbuf *m, int off) 178{ 179 INIT_VNET_INET(curvnet); 180 181 V_ipstat.ips_noproto++; 182 m_freem(m); 183} 184 185/* 186 * Divert a packet by passing it up to the divert socket at port 'port'. 187 * 188 * Setup generic address and protocol structures for div_input routine, 189 * then pass them along with mbuf chain. 190 */ 191static void 192divert_packet(struct mbuf *m, int incoming) 193{ 194 INIT_VNET_INET(curvnet); 195 struct ip *ip; 196 struct inpcb *inp; 197 struct socket *sa; 198 u_int16_t nport; 199 struct sockaddr_in divsrc; 200 struct m_tag *mtag; 201 202 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 203 if (mtag == NULL) { 204 printf("%s: no divert tag\n", __func__); 205 m_freem(m); 206 return; 207 } 208 /* Assure header */ 209 if (m->m_len < sizeof(struct ip) && 210 (m = m_pullup(m, sizeof(struct ip))) == 0) 211 return; 212 ip = mtod(m, struct ip *); 213 214 /* Delayed checksums are currently not compatible with divert. */ 215 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 216 ip->ip_len = ntohs(ip->ip_len); 217 in_delayed_cksum(m); 218 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 219 ip->ip_len = htons(ip->ip_len); 220 } 221 222 /* 223 * Record receive interface address, if any. 224 * But only for incoming packets. 225 */ 226 bzero(&divsrc, sizeof(divsrc)); 227 divsrc.sin_len = sizeof(divsrc); 228 divsrc.sin_family = AF_INET; 229 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 230 if (incoming) { 231 struct ifaddr *ifa; 232 233 /* Sanity check */ 234 M_ASSERTPKTHDR(m); 235 236 /* Find IP address for receive interface */ 237 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 238 if (ifa->ifa_addr->sa_family != AF_INET) 239 continue; 240 divsrc.sin_addr = 241 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 242 break; 243 } 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)divert_info(mtag)); 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 V_ipstat.ips_noproto++; 296 V_ipstat.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 */ 307static int 308div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 309 struct mbuf *control) 310{ 311 INIT_VNET_INET(curvnet); 312 struct m_tag *mtag; 313 struct divert_tag *dt; 314 int error = 0; 315 struct mbuf *options; 316 317 /* 318 * An mbuf may hasn't come from userland, but we pretend 319 * that it has. 320 */ 321 m->m_pkthdr.rcvif = NULL; 322 m->m_nextpkt = NULL; 323 324 if (control) 325 m_freem(control); /* XXX */ 326 327 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 328 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 329 M_NOWAIT | M_ZERO); 330 if (mtag == NULL) { 331 error = ENOBUFS; 332 goto cantsend; 333 } 334 dt = (struct divert_tag *)(mtag+1); 335 m_tag_prepend(m, mtag); 336 } else 337 dt = (struct divert_tag *)(mtag+1); 338 339 /* Loopback avoidance and state recovery */ 340 if (sin) { 341 int i; 342 343 dt->cookie = sin->sin_port; 344 /* 345 * Find receive interface with the given name, stuffed 346 * (if it exists) in the sin_zero[] field. 347 * The name is user supplied data so don't trust its size 348 * or that it is zero terminated. 349 */ 350 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 351 ; 352 if ( i > 0 && i < sizeof(sin->sin_zero)) 353 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 354 } 355 356 /* Reinject packet into the system as incoming or outgoing */ 357 if (!sin || sin->sin_addr.s_addr == 0) { 358 struct ip *const ip = mtod(m, struct ip *); 359 struct inpcb *inp; 360 361 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 362 INP_INFO_WLOCK(&V_divcbinfo); 363 inp = sotoinpcb(so); 364 INP_RLOCK(inp); 365 /* 366 * Don't allow both user specified and setsockopt options, 367 * and don't allow packet length sizes that will crash 368 */ 369 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 370 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 371 error = EINVAL; 372 INP_RUNLOCK(inp); 373 INP_INFO_WUNLOCK(&V_divcbinfo); 374 m_freem(m); 375 } else { 376 /* Convert fields to host order for ip_output() */ 377 ip->ip_len = ntohs(ip->ip_len); 378 ip->ip_off = ntohs(ip->ip_off); 379 380 /* Send packet to output processing */ 381 V_ipstat.ips_rawout++; /* XXX */ 382 383#ifdef MAC 384 mac_inpcb_create_mbuf(inp, m); 385#endif 386 /* 387 * Get ready to inject the packet into ip_output(). 388 * Just in case socket options were specified on the 389 * divert socket, we duplicate them. This is done 390 * to avoid having to hold the PCB locks over the call 391 * to ip_output(), as doing this results in a number of 392 * lock ordering complexities. 393 * 394 * Note that we set the multicast options argument for 395 * ip_output() to NULL since it should be invariant that 396 * they are not present. 397 */ 398 KASSERT(inp->inp_moptions == NULL, 399 ("multicast options set on a divert socket")); 400 options = NULL; 401 /* 402 * XXXCSJP: It is unclear to me whether or not it makes 403 * sense for divert sockets to have options. However, 404 * for now we will duplicate them with the INP locks 405 * held so we can use them in ip_output() without 406 * requring a reference to the pcb. 407 */ 408 if (inp->inp_options != NULL) { 409 options = m_dup(inp->inp_options, M_DONTWAIT); 410 if (options == NULL) 411 error = ENOBUFS; 412 } 413 INP_RUNLOCK(inp); 414 INP_INFO_WUNLOCK(&V_divcbinfo); 415 if (error == ENOBUFS) { 416 m_freem(m); 417 return (error); 418 } 419 error = ip_output(m, options, NULL, 420 ((so->so_options & SO_DONTROUTE) ? 421 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST | 422 IP_RAWOUTPUT, NULL, NULL); 423 if (options != NULL) 424 m_freem(options); 425 } 426 } else { 427 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 428 if (m->m_pkthdr.rcvif == NULL) { 429 /* 430 * No luck with the name, check by IP address. 431 * Clear the port and the ifname to make sure 432 * there are no distractions for ifa_ifwithaddr. 433 */ 434 struct ifaddr *ifa; 435 436 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 437 sin->sin_port = 0; 438 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 439 if (ifa == NULL) { 440 error = EADDRNOTAVAIL; 441 goto cantsend; 442 } 443 m->m_pkthdr.rcvif = ifa->ifa_ifp; 444 } 445#ifdef MAC 446 SOCK_LOCK(so); 447 mac_socket_create_mbuf(so, m); 448 SOCK_UNLOCK(so); 449#endif 450 /* Send packet to input processing via netisr */ 451 netisr_queue(NETISR_IP, m); 452 } 453 454 return error; 455 456cantsend: 457 m_freem(m); 458 return error; 459} 460 461static int 462div_attach(struct socket *so, int proto, struct thread *td) 463{ 464 INIT_VNET_INET(so->so_vnet); 465 struct inpcb *inp; 466 int error; 467 468 inp = sotoinpcb(so); 469 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 470 if (td != NULL) { 471 error = priv_check(td, PRIV_NETINET_DIVERT); 472 if (error) 473 return (error); 474 } 475 error = soreserve(so, div_sendspace, div_recvspace); 476 if (error) 477 return error; 478 INP_INFO_WLOCK(&V_divcbinfo); 479 error = in_pcballoc(so, &V_divcbinfo); 480 if (error) { 481 INP_INFO_WUNLOCK(&V_divcbinfo); 482 return error; 483 } 484 inp = (struct inpcb *)so->so_pcb; 485 INP_INFO_WUNLOCK(&V_divcbinfo); 486 inp->inp_ip_p = proto; 487 inp->inp_vflag |= INP_IPV4; 488 inp->inp_flags |= INP_HDRINCL; 489 INP_WUNLOCK(inp); 490 return 0; 491} 492 493static void 494div_detach(struct socket *so) 495{ 496 INIT_VNET_INET(so->so_vnet); 497 struct inpcb *inp; 498 499 inp = sotoinpcb(so); 500 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 501 INP_INFO_WLOCK(&V_divcbinfo); 502 INP_WLOCK(inp); 503 in_pcbdetach(inp); 504 in_pcbfree(inp); 505 INP_INFO_WUNLOCK(&V_divcbinfo); 506} 507 508static int 509div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 510{ 511 INIT_VNET_INET(so->so_vnet); 512 struct inpcb *inp; 513 int error; 514 515 inp = sotoinpcb(so); 516 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 517 /* in_pcbbind assumes that nam is a sockaddr_in 518 * and in_pcbbind requires a valid address. Since divert 519 * sockets don't we need to make sure the address is 520 * filled in properly. 521 * XXX -- divert should not be abusing in_pcbind 522 * and should probably have its own family. 523 */ 524 if (nam->sa_family != AF_INET) 525 return EAFNOSUPPORT; 526 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 527 INP_INFO_WLOCK(&V_divcbinfo); 528 INP_WLOCK(inp); 529 error = in_pcbbind(inp, nam, td->td_ucred); 530 INP_WUNLOCK(inp); 531 INP_INFO_WUNLOCK(&V_divcbinfo); 532 return error; 533} 534 535static int 536div_shutdown(struct socket *so) 537{ 538 struct inpcb *inp; 539 540 inp = sotoinpcb(so); 541 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 542 INP_WLOCK(inp); 543 socantsendmore(so); 544 INP_WUNLOCK(inp); 545 return 0; 546} 547 548static int 549div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 550 struct mbuf *control, struct thread *td) 551{ 552 INIT_VNET_INET(so->so_vnet); 553 554 /* Packet must have a header (but that's about it) */ 555 if (m->m_len < sizeof (struct ip) && 556 (m = m_pullup(m, sizeof (struct ip))) == 0) { 557 V_ipstat.ips_toosmall++; 558 m_freem(m); 559 return EINVAL; 560 } 561 562 /* Send packet */ 563 return div_output(so, m, (struct sockaddr_in *)nam, control); 564} 565 566void 567div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 568{ 569 struct in_addr faddr; 570 571 faddr = ((struct sockaddr_in *)sa)->sin_addr; 572 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 573 return; 574 if (PRC_IS_REDIRECT(cmd)) 575 return; 576} 577 578static int 579div_pcblist(SYSCTL_HANDLER_ARGS) 580{ 581 int error, i, n; 582 struct inpcb *inp, **inp_list; 583 inp_gen_t gencnt; 584 struct xinpgen xig; 585 586 /* 587 * The process of preparing the TCB list is too time-consuming and 588 * resource-intensive to repeat twice on every request. 589 */ 590 if (req->oldptr == 0) { 591 n = V_divcbinfo.ipi_count; 592 req->oldidx = 2 * (sizeof xig) 593 + (n + n/8) * sizeof(struct xinpcb); 594 return 0; 595 } 596 597 if (req->newptr != 0) 598 return EPERM; 599 600 /* 601 * OK, now we're committed to doing something. 602 */ 603 INP_INFO_RLOCK(&V_divcbinfo); 604 gencnt = V_divcbinfo.ipi_gencnt; 605 n = V_divcbinfo.ipi_count; 606 INP_INFO_RUNLOCK(&V_divcbinfo); 607 608 error = sysctl_wire_old_buffer(req, 609 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 610 if (error != 0) 611 return (error); 612 613 xig.xig_len = sizeof xig; 614 xig.xig_count = n; 615 xig.xig_gen = gencnt; 616 xig.xig_sogen = so_gencnt; 617 error = SYSCTL_OUT(req, &xig, sizeof xig); 618 if (error) 619 return error; 620 621 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 622 if (inp_list == 0) 623 return ENOMEM; 624 625 INP_INFO_RLOCK(&V_divcbinfo); 626 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 627 inp = LIST_NEXT(inp, inp_list)) { 628 INP_RLOCK(inp); 629 if (inp->inp_gencnt <= gencnt && 630 cr_canseeinpcb(req->td->td_ucred, inp) == 0) 631 inp_list[i++] = inp; 632 INP_RUNLOCK(inp); 633 } 634 INP_INFO_RUNLOCK(&V_divcbinfo); 635 n = i; 636 637 error = 0; 638 for (i = 0; i < n; i++) { 639 inp = inp_list[i]; 640 INP_RLOCK(inp); 641 if (inp->inp_gencnt <= gencnt) { 642 struct xinpcb xi; 643 bzero(&xi, sizeof(xi)); 644 xi.xi_len = sizeof xi; 645 /* XXX should avoid extra copy */ 646 bcopy(inp, &xi.xi_inp, sizeof *inp); 647 if (inp->inp_socket) 648 sotoxsocket(inp->inp_socket, &xi.xi_socket); 649 INP_RUNLOCK(inp); 650 error = SYSCTL_OUT(req, &xi, sizeof xi); 651 } else 652 INP_RUNLOCK(inp); 653 } 654 if (!error) { 655 /* 656 * Give the user an updated idea of our state. 657 * If the generation differs from what we told 658 * her before, she knows that something happened 659 * while we were processing this request, and it 660 * might be necessary to retry. 661 */ 662 INP_INFO_RLOCK(&V_divcbinfo); 663 xig.xig_gen = V_divcbinfo.ipi_gencnt; 664 xig.xig_sogen = so_gencnt; 665 xig.xig_count = V_divcbinfo.ipi_count; 666 INP_INFO_RUNLOCK(&V_divcbinfo); 667 error = SYSCTL_OUT(req, &xig, sizeof xig); 668 } 669 free(inp_list, M_TEMP); 670 return error; 671} 672 673#ifdef SYSCTL_NODE 674SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 675SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 676 div_pcblist, "S,xinpcb", "List of active divert sockets"); 677#endif 678 679struct pr_usrreqs div_usrreqs = { 680 .pru_attach = div_attach, 681 .pru_bind = div_bind, 682 .pru_control = in_control, 683 .pru_detach = div_detach, 684 .pru_peeraddr = in_getpeeraddr, 685 .pru_send = div_send, 686 .pru_shutdown = div_shutdown, 687 .pru_sockaddr = in_getsockaddr, 688 .pru_sosetlabel = in_pcbsosetlabel 689}; 690 691struct protosw div_protosw = { 692 .pr_type = SOCK_RAW, 693 .pr_protocol = IPPROTO_DIVERT, 694 .pr_flags = PR_ATOMIC|PR_ADDR, 695 .pr_input = div_input, 696 .pr_ctlinput = div_ctlinput, 697 .pr_ctloutput = ip_ctloutput, 698 .pr_init = div_init, 699 .pr_usrreqs = &div_usrreqs 700}; 701 702static int 703div_modevent(module_t mod, int type, void *unused) 704{ 705 int err = 0; 706 int n; 707 708 switch (type) { 709 case MOD_LOAD: 710 /* 711 * Protocol will be initialized by pf_proto_register(). 712 * We don't have to register ip_protox because we are not 713 * a true IP protocol that goes over the wire. 714 */ 715 err = pf_proto_register(PF_INET, &div_protosw); 716 ip_divert_ptr = divert_packet; 717 break; 718 case MOD_QUIESCE: 719 /* 720 * IPDIVERT may normally not be unloaded because of the 721 * potential race conditions. Tell kldunload we can't be 722 * unloaded unless the unload is forced. 723 */ 724 err = EPERM; 725 break; 726 case MOD_UNLOAD: 727 /* 728 * Forced unload. 729 * 730 * Module ipdivert can only be unloaded if no sockets are 731 * connected. Maybe this can be changed later to forcefully 732 * disconnect any open sockets. 733 * 734 * XXXRW: Note that there is a slight race here, as a new 735 * socket open request could be spinning on the lock and then 736 * we destroy the lock. 737 */ 738 INP_INFO_WLOCK(&V_divcbinfo); 739 n = V_divcbinfo.ipi_count; 740 if (n != 0) { 741 err = EBUSY; 742 INP_INFO_WUNLOCK(&V_divcbinfo); 743 break; 744 } 745 ip_divert_ptr = NULL; 746 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 747 INP_INFO_WUNLOCK(&V_divcbinfo); 748 INP_INFO_LOCK_DESTROY(&V_divcbinfo); 749 uma_zdestroy(V_divcbinfo.ipi_zone); 750 break; 751 default: 752 err = EOPNOTSUPP; 753 break; 754 } 755 return err; 756} 757 758static moduledata_t ipdivertmod = { 759 "ipdivert", 760 div_modevent, 761 0 762}; 763 764DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 765MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); 766MODULE_VERSION(ipdivert, 1); 767