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_sctp.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/rwlock.h> 56 #include <sys/signalvar.h> 57 #include <sys/socket.h> 58 #include <sys/socketvar.h> 59 #include <sys/sx.h> 60 #include <sys/sysctl.h> 61 #include <sys/systm.h> 62 63 #include <vm/uma.h> 64 65 #include <net/if.h> 66 #include <net/netisr.h> 67 #include <net/route.h> 68 #include <net/vnet.h> 69 70 #include <netinet/in.h> 71 #include <netinet/in_pcb.h> 72 #include <netinet/in_systm.h> 73 #include <netinet/in_var.h> 74 #include <netinet/ip.h> 75 #include <netinet/ip_divert.h> 76 #include <netinet/ip_var.h> 77 #include <netinet/ip_fw.h> 78 #ifdef SCTP 79 #include <netinet/sctp_crc32.h> 80 #endif 81 82 #include <security/mac/mac_framework.h> 83 84 /* 85 * Divert sockets 86 */ 87 88 /* 89 * Allocate enough space to hold a full IP packet 90 */ 91 #define DIVSNDQ (65536 + 100) 92 #define DIVRCVQ (65536 + 100) 93 94 /* 95 * Divert sockets work in conjunction with ipfw, see the divert(4) 96 * manpage for features. 97 * Internally, packets selected by ipfw in ip_input() or ip_output(), 98 * and never diverted before, are passed to the input queue of the 99 * divert socket with a given 'divert_port' number (as specified in 100 * the matching ipfw rule), and they are tagged with a 16 bit cookie 101 * (representing the rule number of the matching ipfw rule), which 102 * is passed to process reading from the socket. 103 * 104 * Packets written to the divert socket are again tagged with a cookie 105 * (usually the same as above) and a destination address. 106 * If the destination address is INADDR_ANY then the packet is 107 * treated as outgoing and sent to ip_output(), otherwise it is 108 * treated as incoming and sent to ip_input(). 109 * In both cases, the packet is tagged with the cookie. 110 * 111 * On reinjection, processing in ip_input() and ip_output() 112 * will be exactly the same as for the original packet, except that 113 * ipfw processing will start at the rule number after the one 114 * written in the cookie (so, tagging a packet with a cookie of 0 115 * will cause it to be effectively considered as a standard packet). 116 */ 117 118 /* Internal variables. */ 119 static VNET_DEFINE(struct inpcbhead, divcb); 120 static VNET_DEFINE(struct inpcbinfo, divcbinfo); 121 122 #define V_divcb VNET(divcb) 123 #define V_divcbinfo VNET(divcbinfo) 124 125 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 126 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 127 128 /* 129 * Initialize divert connection block queue. 130 */ 131 static void 132 div_zone_change(void *tag) 133 { 134 135 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 136 } 137 138 static int 139 div_inpcb_init(void *mem, int size, int flags) 140 { 141 struct inpcb *inp = mem; 142 143 INP_LOCK_INIT(inp, "inp", "divinp"); 144 return (0); 145 } 146 147 static void 148 div_inpcb_fini(void *mem, int size) 149 { 150 struct inpcb *inp = mem; 151 152 INP_LOCK_DESTROY(inp); 153 } 154 155 void 156 div_init(void) 157 { 158 159 INP_INFO_LOCK_INIT(&V_divcbinfo, "div"); 160 LIST_INIT(&V_divcb); 161 V_divcbinfo.ipi_listhead = &V_divcb; 162 #ifdef VIMAGE 163 V_divcbinfo.ipi_vnet = curvnet; 164 #endif 165 /* 166 * XXX We don't use the hash list for divert IP, but it's easier 167 * to allocate a one entry hash list than it is to check all 168 * over the place for hashbase == NULL. 169 */ 170 V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask); 171 V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB, 172 &V_divcbinfo.ipi_porthashmask); 173 V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 174 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, 175 UMA_ZONE_NOFREE); 176 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 177 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, 178 NULL, EVENTHANDLER_PRI_ANY); 179 } 180 181 /* 182 * IPPROTO_DIVERT is not in the real IP protocol number space; this 183 * function should never be called. Just in case, drop any packets. 184 */ 185 void 186 div_input(struct mbuf *m, int off) 187 { 188 189 KMOD_IPSTAT_INC(ips_noproto); 190 m_freem(m); 191 } 192 193 /* 194 * Divert a packet by passing it up to the divert socket at port 'port'. 195 * 196 * Setup generic address and protocol structures for div_input routine, 197 * then pass them along with mbuf chain. 198 */ 199 static void 200 divert_packet(struct mbuf *m, int incoming) 201 { 202 struct ip *ip; 203 struct inpcb *inp; 204 struct socket *sa; 205 u_int16_t nport; 206 struct sockaddr_in divsrc; 207 struct m_tag *mtag; 208 209 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 210 if (mtag == NULL) { 211 printf("%s: no divert tag\n", __func__); 212 m_freem(m); 213 return; 214 } 215 /* Assure header */ 216 if (m->m_len < sizeof(struct ip) && 217 (m = m_pullup(m, sizeof(struct ip))) == 0) 218 return; 219 ip = mtod(m, struct ip *); 220 221 /* Delayed checksums are currently not compatible with divert. */ 222 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 223 ip->ip_len = ntohs(ip->ip_len); 224 in_delayed_cksum(m); 225 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 226 ip->ip_len = htons(ip->ip_len); 227 } 228 #ifdef SCTP 229 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 230 ip->ip_len = ntohs(ip->ip_len); 231 sctp_delayed_cksum(m); 232 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 233 ip->ip_len = htons(ip->ip_len); 234 } 235 #endif 236 /* 237 * Record receive interface address, if any. 238 * But only for incoming packets. 239 */ 240 bzero(&divsrc, sizeof(divsrc)); 241 divsrc.sin_len = sizeof(divsrc); 242 divsrc.sin_family = AF_INET; 243 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 244 if (incoming) { 245 struct ifaddr *ifa; 246 struct ifnet *ifp; 247 248 /* Sanity check */ 249 M_ASSERTPKTHDR(m); 250 251 /* Find IP address for receive interface */ 252 ifp = m->m_pkthdr.rcvif; 253 if_addr_rlock(ifp); 254 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 255 if (ifa->ifa_addr->sa_family != AF_INET) 256 continue; 257 divsrc.sin_addr = 258 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 259 break; 260 } 261 if_addr_runlock(ifp); 262 } 263 /* 264 * Record the incoming interface name whenever we have one. 265 */ 266 if (m->m_pkthdr.rcvif) { 267 /* 268 * Hide the actual interface name in there in the 269 * sin_zero array. XXX This needs to be moved to a 270 * different sockaddr type for divert, e.g. 271 * sockaddr_div with multiple fields like 272 * sockaddr_dl. Presently we have only 7 bytes 273 * but that will do for now as most interfaces 274 * are 4 or less + 2 or less bytes for unit. 275 * There is probably a faster way of doing this, 276 * possibly taking it from the sockaddr_dl on the iface. 277 * This solves the problem of a P2P link and a LAN interface 278 * having the same address, which can result in the wrong 279 * interface being assigned to the packet when fed back 280 * into the divert socket. Theoretically if the daemon saves 281 * and re-uses the sockaddr_in as suggested in the man pages, 282 * this iface name will come along for the ride. 283 * (see div_output for the other half of this.) 284 */ 285 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 286 sizeof(divsrc.sin_zero)); 287 } 288 289 /* Put packet on socket queue, if any */ 290 sa = NULL; 291 nport = htons((u_int16_t)divert_info(mtag)); 292 INP_INFO_RLOCK(&V_divcbinfo); 293 LIST_FOREACH(inp, &V_divcb, inp_list) { 294 /* XXX why does only one socket match? */ 295 if (inp->inp_lport == nport) { 296 INP_RLOCK(inp); 297 sa = inp->inp_socket; 298 SOCKBUF_LOCK(&sa->so_rcv); 299 if (sbappendaddr_locked(&sa->so_rcv, 300 (struct sockaddr *)&divsrc, m, 301 (struct mbuf *)0) == 0) { 302 SOCKBUF_UNLOCK(&sa->so_rcv); 303 sa = NULL; /* force mbuf reclaim below */ 304 } else 305 sorwakeup_locked(sa); 306 INP_RUNLOCK(inp); 307 break; 308 } 309 } 310 INP_INFO_RUNLOCK(&V_divcbinfo); 311 if (sa == NULL) { 312 m_freem(m); 313 KMOD_IPSTAT_INC(ips_noproto); 314 KMOD_IPSTAT_DEC(ips_delivered); 315 } 316 } 317 318 /* 319 * Deliver packet back into the IP processing machinery. 320 * 321 * If no address specified, or address is 0.0.0.0, send to ip_output(); 322 * otherwise, send to ip_input() and mark as having been received on 323 * the interface with that address. 324 */ 325 static int 326 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 327 struct mbuf *control) 328 { 329 struct m_tag *mtag; 330 struct divert_tag *dt; 331 int error = 0; 332 struct mbuf *options; 333 334 /* 335 * An mbuf may hasn't come from userland, but we pretend 336 * that it has. 337 */ 338 m->m_pkthdr.rcvif = NULL; 339 m->m_nextpkt = NULL; 340 M_SETFIB(m, so->so_fibnum); 341 342 if (control) 343 m_freem(control); /* XXX */ 344 345 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 346 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 347 M_NOWAIT | M_ZERO); 348 if (mtag == NULL) { 349 error = ENOBUFS; 350 goto cantsend; 351 } 352 dt = (struct divert_tag *)(mtag+1); 353 m_tag_prepend(m, mtag); 354 } else 355 dt = (struct divert_tag *)(mtag+1); 356 357 /* Loopback avoidance and state recovery */ 358 if (sin) { 359 int i; 360 361 dt->cookie = sin->sin_port; 362 /* 363 * Find receive interface with the given name, stuffed 364 * (if it exists) in the sin_zero[] field. 365 * The name is user supplied data so don't trust its size 366 * or that it is zero terminated. 367 */ 368 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 369 ; 370 if ( i > 0 && i < sizeof(sin->sin_zero)) 371 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 372 } 373 374 /* Reinject packet into the system as incoming or outgoing */ 375 if (!sin || sin->sin_addr.s_addr == 0) { 376 struct ip *const ip = mtod(m, struct ip *); 377 struct inpcb *inp; 378 379 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 380 INP_INFO_WLOCK(&V_divcbinfo); 381 inp = sotoinpcb(so); 382 INP_RLOCK(inp); 383 /* 384 * Don't allow both user specified and setsockopt options, 385 * and don't allow packet length sizes that will crash 386 */ 387 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 388 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 389 error = EINVAL; 390 INP_RUNLOCK(inp); 391 INP_INFO_WUNLOCK(&V_divcbinfo); 392 m_freem(m); 393 } else { 394 /* Convert fields to host order for ip_output() */ 395 ip->ip_len = ntohs(ip->ip_len); 396 ip->ip_off = ntohs(ip->ip_off); 397 398 /* Send packet to output processing */ 399 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 400 401 #ifdef MAC 402 mac_inpcb_create_mbuf(inp, m); 403 #endif 404 /* 405 * Get ready to inject the packet into ip_output(). 406 * Just in case socket options were specified on the 407 * divert socket, we duplicate them. This is done 408 * to avoid having to hold the PCB locks over the call 409 * to ip_output(), as doing this results in a number of 410 * lock ordering complexities. 411 * 412 * Note that we set the multicast options argument for 413 * ip_output() to NULL since it should be invariant that 414 * they are not present. 415 */ 416 KASSERT(inp->inp_moptions == NULL, 417 ("multicast options set on a divert socket")); 418 options = NULL; 419 /* 420 * XXXCSJP: It is unclear to me whether or not it makes 421 * sense for divert sockets to have options. However, 422 * for now we will duplicate them with the INP locks 423 * held so we can use them in ip_output() without 424 * requring a reference to the pcb. 425 */ 426 if (inp->inp_options != NULL) { 427 options = m_dup(inp->inp_options, M_DONTWAIT); 428 if (options == NULL) 429 error = ENOBUFS; 430 } 431 INP_RUNLOCK(inp); 432 INP_INFO_WUNLOCK(&V_divcbinfo); 433 if (error == ENOBUFS) { 434 m_freem(m); 435 return (error); 436 } 437 error = ip_output(m, options, NULL, 438 ((so->so_options & SO_DONTROUTE) ? 439 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST | 440 IP_RAWOUTPUT, NULL, NULL); 441 if (options != NULL) 442 m_freem(options); 443 } 444 } else { 445 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 446 if (m->m_pkthdr.rcvif == NULL) { 447 /* 448 * No luck with the name, check by IP address. 449 * Clear the port and the ifname to make sure 450 * there are no distractions for ifa_ifwithaddr. 451 */ 452 struct ifaddr *ifa; 453 454 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 455 sin->sin_port = 0; 456 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 457 if (ifa == NULL) { 458 error = EADDRNOTAVAIL; 459 goto cantsend; 460 } 461 m->m_pkthdr.rcvif = ifa->ifa_ifp; 462 ifa_free(ifa); 463 } 464 #ifdef MAC 465 mac_socket_create_mbuf(so, m); 466 #endif 467 /* Send packet to input processing via netisr */ 468 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 469 } 470 471 return error; 472 473 cantsend: 474 m_freem(m); 475 return error; 476 } 477 478 static int 479 div_attach(struct socket *so, int proto, struct thread *td) 480 { 481 struct inpcb *inp; 482 int error; 483 484 inp = sotoinpcb(so); 485 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 486 if (td != NULL) { 487 error = priv_check(td, PRIV_NETINET_DIVERT); 488 if (error) 489 return (error); 490 } 491 error = soreserve(so, div_sendspace, div_recvspace); 492 if (error) 493 return error; 494 INP_INFO_WLOCK(&V_divcbinfo); 495 error = in_pcballoc(so, &V_divcbinfo); 496 if (error) { 497 INP_INFO_WUNLOCK(&V_divcbinfo); 498 return error; 499 } 500 inp = (struct inpcb *)so->so_pcb; 501 INP_INFO_WUNLOCK(&V_divcbinfo); 502 inp->inp_ip_p = proto; 503 inp->inp_vflag |= INP_IPV4; 504 inp->inp_flags |= INP_HDRINCL; 505 INP_WUNLOCK(inp); 506 return 0; 507 } 508 509 static void 510 div_detach(struct socket *so) 511 { 512 struct inpcb *inp; 513 514 inp = sotoinpcb(so); 515 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 516 INP_INFO_WLOCK(&V_divcbinfo); 517 INP_WLOCK(inp); 518 in_pcbdetach(inp); 519 in_pcbfree(inp); 520 INP_INFO_WUNLOCK(&V_divcbinfo); 521 } 522 523 static int 524 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 525 { 526 struct inpcb *inp; 527 int error; 528 529 inp = sotoinpcb(so); 530 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 531 /* in_pcbbind assumes that nam is a sockaddr_in 532 * and in_pcbbind requires a valid address. Since divert 533 * sockets don't we need to make sure the address is 534 * filled in properly. 535 * XXX -- divert should not be abusing in_pcbind 536 * and should probably have its own family. 537 */ 538 if (nam->sa_family != AF_INET) 539 return EAFNOSUPPORT; 540 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 541 INP_INFO_WLOCK(&V_divcbinfo); 542 INP_WLOCK(inp); 543 error = in_pcbbind(inp, nam, td->td_ucred); 544 INP_WUNLOCK(inp); 545 INP_INFO_WUNLOCK(&V_divcbinfo); 546 return error; 547 } 548 549 static int 550 div_shutdown(struct socket *so) 551 { 552 struct inpcb *inp; 553 554 inp = sotoinpcb(so); 555 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 556 INP_WLOCK(inp); 557 socantsendmore(so); 558 INP_WUNLOCK(inp); 559 return 0; 560 } 561 562 static int 563 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 564 struct mbuf *control, struct thread *td) 565 { 566 567 /* Packet must have a header (but that's about it) */ 568 if (m->m_len < sizeof (struct ip) && 569 (m = m_pullup(m, sizeof (struct ip))) == 0) { 570 KMOD_IPSTAT_INC(ips_toosmall); 571 m_freem(m); 572 return EINVAL; 573 } 574 575 /* Send packet */ 576 return div_output(so, m, (struct sockaddr_in *)nam, control); 577 } 578 579 void 580 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 581 { 582 struct in_addr faddr; 583 584 faddr = ((struct sockaddr_in *)sa)->sin_addr; 585 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 586 return; 587 if (PRC_IS_REDIRECT(cmd)) 588 return; 589 } 590 591 static int 592 div_pcblist(SYSCTL_HANDLER_ARGS) 593 { 594 int error, i, n; 595 struct inpcb *inp, **inp_list; 596 inp_gen_t gencnt; 597 struct xinpgen xig; 598 599 /* 600 * The process of preparing the TCB list is too time-consuming and 601 * resource-intensive to repeat twice on every request. 602 */ 603 if (req->oldptr == 0) { 604 n = V_divcbinfo.ipi_count; 605 req->oldidx = 2 * (sizeof xig) 606 + (n + n/8) * sizeof(struct xinpcb); 607 return 0; 608 } 609 610 if (req->newptr != 0) 611 return EPERM; 612 613 /* 614 * OK, now we're committed to doing something. 615 */ 616 INP_INFO_RLOCK(&V_divcbinfo); 617 gencnt = V_divcbinfo.ipi_gencnt; 618 n = V_divcbinfo.ipi_count; 619 INP_INFO_RUNLOCK(&V_divcbinfo); 620 621 error = sysctl_wire_old_buffer(req, 622 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 623 if (error != 0) 624 return (error); 625 626 xig.xig_len = sizeof xig; 627 xig.xig_count = n; 628 xig.xig_gen = gencnt; 629 xig.xig_sogen = so_gencnt; 630 error = SYSCTL_OUT(req, &xig, sizeof xig); 631 if (error) 632 return error; 633 634 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 635 if (inp_list == 0) 636 return ENOMEM; 637 638 INP_INFO_RLOCK(&V_divcbinfo); 639 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 640 inp = LIST_NEXT(inp, inp_list)) { 641 INP_RLOCK(inp); 642 if (inp->inp_gencnt <= gencnt && 643 cr_canseeinpcb(req->td->td_ucred, inp) == 0) 644 inp_list[i++] = inp; 645 INP_RUNLOCK(inp); 646 } 647 INP_INFO_RUNLOCK(&V_divcbinfo); 648 n = i; 649 650 error = 0; 651 for (i = 0; i < n; i++) { 652 inp = inp_list[i]; 653 INP_RLOCK(inp); 654 if (inp->inp_gencnt <= gencnt) { 655 struct xinpcb xi; 656 bzero(&xi, sizeof(xi)); 657 xi.xi_len = sizeof xi; 658 /* XXX should avoid extra copy */ 659 bcopy(inp, &xi.xi_inp, sizeof *inp); 660 if (inp->inp_socket) 661 sotoxsocket(inp->inp_socket, &xi.xi_socket); 662 INP_RUNLOCK(inp); 663 error = SYSCTL_OUT(req, &xi, sizeof xi); 664 } else 665 INP_RUNLOCK(inp); 666 } 667 if (!error) { 668 /* 669 * Give the user an updated idea of our state. 670 * If the generation differs from what we told 671 * her before, she knows that something happened 672 * while we were processing this request, and it 673 * might be necessary to retry. 674 */ 675 INP_INFO_RLOCK(&V_divcbinfo); 676 xig.xig_gen = V_divcbinfo.ipi_gencnt; 677 xig.xig_sogen = so_gencnt; 678 xig.xig_count = V_divcbinfo.ipi_count; 679 INP_INFO_RUNLOCK(&V_divcbinfo); 680 error = SYSCTL_OUT(req, &xig, sizeof xig); 681 } 682 free(inp_list, M_TEMP); 683 return error; 684 } 685 686 #ifdef SYSCTL_NODE 687 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 688 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 689 div_pcblist, "S,xinpcb", "List of active divert sockets"); 690 #endif 691 692 struct pr_usrreqs div_usrreqs = { 693 .pru_attach = div_attach, 694 .pru_bind = div_bind, 695 .pru_control = in_control, 696 .pru_detach = div_detach, 697 .pru_peeraddr = in_getpeeraddr, 698 .pru_send = div_send, 699 .pru_shutdown = div_shutdown, 700 .pru_sockaddr = in_getsockaddr, 701 .pru_sosetlabel = in_pcbsosetlabel 702 }; 703 704 struct protosw div_protosw = { 705 .pr_type = SOCK_RAW, 706 .pr_protocol = IPPROTO_DIVERT, 707 .pr_flags = PR_ATOMIC|PR_ADDR, 708 .pr_input = div_input, 709 .pr_ctlinput = div_ctlinput, 710 .pr_ctloutput = ip_ctloutput, 711 .pr_init = div_init, 712 .pr_usrreqs = &div_usrreqs 713 }; 714 715 static int 716 div_modevent(module_t mod, int type, void *unused) 717 { 718 int err = 0; 719 int n; 720 721 switch (type) { 722 case MOD_LOAD: 723 /* 724 * Protocol will be initialized by pf_proto_register(). 725 * We don't have to register ip_protox because we are not 726 * a true IP protocol that goes over the wire. 727 */ 728 err = pf_proto_register(PF_INET, &div_protosw); 729 ip_divert_ptr = divert_packet; 730 break; 731 case MOD_QUIESCE: 732 /* 733 * IPDIVERT may normally not be unloaded because of the 734 * potential race conditions. Tell kldunload we can't be 735 * unloaded unless the unload is forced. 736 */ 737 err = EPERM; 738 break; 739 case MOD_UNLOAD: 740 /* 741 * Forced unload. 742 * 743 * Module ipdivert can only be unloaded if no sockets are 744 * connected. Maybe this can be changed later to forcefully 745 * disconnect any open sockets. 746 * 747 * XXXRW: Note that there is a slight race here, as a new 748 * socket open request could be spinning on the lock and then 749 * we destroy the lock. 750 */ 751 INP_INFO_WLOCK(&V_divcbinfo); 752 n = V_divcbinfo.ipi_count; 753 if (n != 0) { 754 err = EBUSY; 755 INP_INFO_WUNLOCK(&V_divcbinfo); 756 break; 757 } 758 ip_divert_ptr = NULL; 759 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 760 INP_INFO_WUNLOCK(&V_divcbinfo); 761 INP_INFO_LOCK_DESTROY(&V_divcbinfo); 762 uma_zdestroy(V_divcbinfo.ipi_zone); 763 break; 764 default: 765 err = EOPNOTSUPP; 766 break; 767 } 768 return err; 769 } 770 771 static moduledata_t ipdivertmod = { 772 "ipdivert", 773 div_modevent, 774 0 775 }; 776 777 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 778 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); 779 MODULE_VERSION(ipdivert, 1); 780