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 #include <sys/vimage.h> 63 64 #include <vm/uma.h> 65 66 #include <net/if.h> 67 #include <net/netisr.h> 68 #include <net/route.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 #include <netinet/vinet.h> 79 #ifdef SCTP 80 #include <netinet/sctp_crc32.h> 81 #endif 82 83 #include <security/mac/mac_framework.h> 84 85 /* 86 * Divert sockets 87 */ 88 89 /* 90 * Allocate enough space to hold a full IP packet 91 */ 92 #define DIVSNDQ (65536 + 100) 93 #define DIVRCVQ (65536 + 100) 94 95 /* 96 * Divert sockets work in conjunction with ipfw, see the divert(4) 97 * manpage for features. 98 * Internally, packets selected by ipfw in ip_input() or ip_output(), 99 * and never diverted before, are passed to the input queue of the 100 * divert socket with a given 'divert_port' number (as specified in 101 * the matching ipfw rule), and they are tagged with a 16 bit cookie 102 * (representing the rule number of the matching ipfw rule), which 103 * is passed to process reading from the socket. 104 * 105 * Packets written to the divert socket are again tagged with a cookie 106 * (usually the same as above) and a destination address. 107 * If the destination address is INADDR_ANY then the packet is 108 * treated as outgoing and sent to ip_output(), otherwise it is 109 * treated as incoming and sent to ip_input(). 110 * In both cases, the packet is tagged with the cookie. 111 * 112 * On reinjection, processing in ip_input() and ip_output() 113 * will be exactly the same as for the original packet, except that 114 * ipfw processing will start at the rule number after the one 115 * written in the cookie (so, tagging a packet with a cookie of 0 116 * will cause it to be effectively considered as a standard packet). 117 */ 118 119 /* Internal variables. */ 120 #ifdef VIMAGE_GLOBALS 121 static struct inpcbhead divcb; 122 static struct inpcbinfo divcbinfo; 123 #endif 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 INIT_VNET_INET(curvnet); 135 136 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 137 } 138 139 static int 140 div_inpcb_init(void *mem, int size, int flags) 141 { 142 struct inpcb *inp = mem; 143 144 INP_LOCK_INIT(inp, "inp", "divinp"); 145 return (0); 146 } 147 148 static void 149 div_inpcb_fini(void *mem, int size) 150 { 151 struct inpcb *inp = mem; 152 153 INP_LOCK_DESTROY(inp); 154 } 155 156 void 157 div_init(void) 158 { 159 INIT_VNET_INET(curvnet); 160 161 INP_INFO_LOCK_INIT(&V_divcbinfo, "div"); 162 LIST_INIT(&V_divcb); 163 V_divcbinfo.ipi_listhead = &V_divcb; 164 #ifdef VIMAGE 165 V_divcbinfo.ipi_vnet = curvnet; 166 #endif 167 /* 168 * XXX We don't use the hash list for divert IP, but it's easier 169 * to allocate a one entry hash list than it is to check all 170 * over the place for hashbase == NULL. 171 */ 172 V_divcbinfo.ipi_hashbase = hashinit(1, M_PCB, &V_divcbinfo.ipi_hashmask); 173 V_divcbinfo.ipi_porthashbase = hashinit(1, M_PCB, 174 &V_divcbinfo.ipi_porthashmask); 175 V_divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 176 NULL, NULL, div_inpcb_init, div_inpcb_fini, UMA_ALIGN_PTR, 177 UMA_ZONE_NOFREE); 178 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 179 EVENTHANDLER_REGISTER(maxsockets_change, div_zone_change, 180 NULL, EVENTHANDLER_PRI_ANY); 181 } 182 183 /* 184 * IPPROTO_DIVERT is not in the real IP protocol number space; this 185 * function should never be called. Just in case, drop any packets. 186 */ 187 void 188 div_input(struct mbuf *m, int off) 189 { 190 INIT_VNET_INET(curvnet); 191 192 IPSTAT_INC(ips_noproto); 193 m_freem(m); 194 } 195 196 /* 197 * Divert a packet by passing it up to the divert socket at port 'port'. 198 * 199 * Setup generic address and protocol structures for div_input routine, 200 * then pass them along with mbuf chain. 201 */ 202 static void 203 divert_packet(struct mbuf *m, int incoming) 204 { 205 INIT_VNET_INET(curvnet); 206 struct ip *ip; 207 struct inpcb *inp; 208 struct socket *sa; 209 u_int16_t nport; 210 struct sockaddr_in divsrc; 211 struct m_tag *mtag; 212 213 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 214 if (mtag == NULL) { 215 printf("%s: no divert tag\n", __func__); 216 m_freem(m); 217 return; 218 } 219 /* Assure header */ 220 if (m->m_len < sizeof(struct ip) && 221 (m = m_pullup(m, sizeof(struct ip))) == 0) 222 return; 223 ip = mtod(m, struct ip *); 224 225 /* Delayed checksums are currently not compatible with divert. */ 226 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 227 ip->ip_len = ntohs(ip->ip_len); 228 in_delayed_cksum(m); 229 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 230 ip->ip_len = htons(ip->ip_len); 231 } 232 #ifdef SCTP 233 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 234 ip->ip_len = ntohs(ip->ip_len); 235 sctp_delayed_cksum(m); 236 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 237 ip->ip_len = htons(ip->ip_len); 238 } 239 #endif 240 /* 241 * Record receive interface address, if any. 242 * But only for incoming packets. 243 */ 244 bzero(&divsrc, sizeof(divsrc)); 245 divsrc.sin_len = sizeof(divsrc); 246 divsrc.sin_family = AF_INET; 247 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 248 if (incoming) { 249 struct ifaddr *ifa; 250 struct ifnet *ifp; 251 252 /* Sanity check */ 253 M_ASSERTPKTHDR(m); 254 255 /* Find IP address for receive interface */ 256 ifp = m->m_pkthdr.rcvif; 257 IF_ADDR_LOCK(ifp); 258 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 259 if (ifa->ifa_addr->sa_family != AF_INET) 260 continue; 261 divsrc.sin_addr = 262 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 263 break; 264 } 265 IF_ADDR_UNLOCK(ifp); 266 } 267 /* 268 * Record the incoming interface name whenever we have one. 269 */ 270 if (m->m_pkthdr.rcvif) { 271 /* 272 * Hide the actual interface name in there in the 273 * sin_zero array. XXX This needs to be moved to a 274 * different sockaddr type for divert, e.g. 275 * sockaddr_div with multiple fields like 276 * sockaddr_dl. Presently we have only 7 bytes 277 * but that will do for now as most interfaces 278 * are 4 or less + 2 or less bytes for unit. 279 * There is probably a faster way of doing this, 280 * possibly taking it from the sockaddr_dl on the iface. 281 * This solves the problem of a P2P link and a LAN interface 282 * having the same address, which can result in the wrong 283 * interface being assigned to the packet when fed back 284 * into the divert socket. Theoretically if the daemon saves 285 * and re-uses the sockaddr_in as suggested in the man pages, 286 * this iface name will come along for the ride. 287 * (see div_output for the other half of this.) 288 */ 289 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 290 sizeof(divsrc.sin_zero)); 291 } 292 293 /* Put packet on socket queue, if any */ 294 sa = NULL; 295 nport = htons((u_int16_t)divert_info(mtag)); 296 INP_INFO_RLOCK(&V_divcbinfo); 297 LIST_FOREACH(inp, &V_divcb, inp_list) { 298 /* XXX why does only one socket match? */ 299 if (inp->inp_lport == nport) { 300 INP_RLOCK(inp); 301 sa = inp->inp_socket; 302 SOCKBUF_LOCK(&sa->so_rcv); 303 if (sbappendaddr_locked(&sa->so_rcv, 304 (struct sockaddr *)&divsrc, m, 305 (struct mbuf *)0) == 0) { 306 SOCKBUF_UNLOCK(&sa->so_rcv); 307 sa = NULL; /* force mbuf reclaim below */ 308 } else 309 sorwakeup_locked(sa); 310 INP_RUNLOCK(inp); 311 break; 312 } 313 } 314 INP_INFO_RUNLOCK(&V_divcbinfo); 315 if (sa == NULL) { 316 m_freem(m); 317 IPSTAT_INC(ips_noproto); 318 IPSTAT_DEC(ips_delivered); 319 } 320 } 321 322 /* 323 * Deliver packet back into the IP processing machinery. 324 * 325 * If no address specified, or address is 0.0.0.0, send to ip_output(); 326 * otherwise, send to ip_input() and mark as having been received on 327 * the interface with that address. 328 */ 329 static int 330 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 331 struct mbuf *control) 332 { 333 INIT_VNET_INET(curvnet); 334 struct m_tag *mtag; 335 struct divert_tag *dt; 336 int error = 0; 337 struct mbuf *options; 338 339 /* 340 * An mbuf may hasn't come from userland, but we pretend 341 * that it has. 342 */ 343 m->m_pkthdr.rcvif = NULL; 344 m->m_nextpkt = NULL; 345 M_SETFIB(m, so->so_fibnum); 346 347 if (control) 348 m_freem(control); /* XXX */ 349 350 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 351 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 352 M_NOWAIT | M_ZERO); 353 if (mtag == NULL) { 354 error = ENOBUFS; 355 goto cantsend; 356 } 357 dt = (struct divert_tag *)(mtag+1); 358 m_tag_prepend(m, mtag); 359 } else 360 dt = (struct divert_tag *)(mtag+1); 361 362 /* Loopback avoidance and state recovery */ 363 if (sin) { 364 int i; 365 366 dt->cookie = sin->sin_port; 367 /* 368 * Find receive interface with the given name, stuffed 369 * (if it exists) in the sin_zero[] field. 370 * The name is user supplied data so don't trust its size 371 * or that it is zero terminated. 372 */ 373 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 374 ; 375 if ( i > 0 && i < sizeof(sin->sin_zero)) 376 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 377 } 378 379 /* Reinject packet into the system as incoming or outgoing */ 380 if (!sin || sin->sin_addr.s_addr == 0) { 381 struct ip *const ip = mtod(m, struct ip *); 382 struct inpcb *inp; 383 384 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 385 INP_INFO_WLOCK(&V_divcbinfo); 386 inp = sotoinpcb(so); 387 INP_RLOCK(inp); 388 /* 389 * Don't allow both user specified and setsockopt options, 390 * and don't allow packet length sizes that will crash 391 */ 392 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 393 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 394 error = EINVAL; 395 INP_RUNLOCK(inp); 396 INP_INFO_WUNLOCK(&V_divcbinfo); 397 m_freem(m); 398 } else { 399 /* Convert fields to host order for ip_output() */ 400 ip->ip_len = ntohs(ip->ip_len); 401 ip->ip_off = ntohs(ip->ip_off); 402 403 /* Send packet to output processing */ 404 IPSTAT_INC(ips_rawout); /* XXX */ 405 406 #ifdef MAC 407 mac_inpcb_create_mbuf(inp, m); 408 #endif 409 /* 410 * Get ready to inject the packet into ip_output(). 411 * Just in case socket options were specified on the 412 * divert socket, we duplicate them. This is done 413 * to avoid having to hold the PCB locks over the call 414 * to ip_output(), as doing this results in a number of 415 * lock ordering complexities. 416 * 417 * Note that we set the multicast options argument for 418 * ip_output() to NULL since it should be invariant that 419 * they are not present. 420 */ 421 KASSERT(inp->inp_moptions == NULL, 422 ("multicast options set on a divert socket")); 423 options = NULL; 424 /* 425 * XXXCSJP: It is unclear to me whether or not it makes 426 * sense for divert sockets to have options. However, 427 * for now we will duplicate them with the INP locks 428 * held so we can use them in ip_output() without 429 * requring a reference to the pcb. 430 */ 431 if (inp->inp_options != NULL) { 432 options = m_dup(inp->inp_options, M_DONTWAIT); 433 if (options == NULL) 434 error = ENOBUFS; 435 } 436 INP_RUNLOCK(inp); 437 INP_INFO_WUNLOCK(&V_divcbinfo); 438 if (error == ENOBUFS) { 439 m_freem(m); 440 return (error); 441 } 442 error = ip_output(m, options, NULL, 443 ((so->so_options & SO_DONTROUTE) ? 444 IP_ROUTETOIF : 0) | IP_ALLOWBROADCAST | 445 IP_RAWOUTPUT, NULL, NULL); 446 if (options != NULL) 447 m_freem(options); 448 } 449 } else { 450 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 451 if (m->m_pkthdr.rcvif == NULL) { 452 /* 453 * No luck with the name, check by IP address. 454 * Clear the port and the ifname to make sure 455 * there are no distractions for ifa_ifwithaddr. 456 */ 457 struct ifaddr *ifa; 458 459 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 460 sin->sin_port = 0; 461 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 462 if (ifa == NULL) { 463 error = EADDRNOTAVAIL; 464 goto cantsend; 465 } 466 m->m_pkthdr.rcvif = ifa->ifa_ifp; 467 } 468 #ifdef MAC 469 mac_socket_create_mbuf(so, m); 470 #endif 471 /* Send packet to input processing via netisr */ 472 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 473 } 474 475 return error; 476 477 cantsend: 478 m_freem(m); 479 return error; 480 } 481 482 static int 483 div_attach(struct socket *so, int proto, struct thread *td) 484 { 485 INIT_VNET_INET(so->so_vnet); 486 struct inpcb *inp; 487 int error; 488 489 inp = sotoinpcb(so); 490 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 491 if (td != NULL) { 492 error = priv_check(td, PRIV_NETINET_DIVERT); 493 if (error) 494 return (error); 495 } 496 error = soreserve(so, div_sendspace, div_recvspace); 497 if (error) 498 return error; 499 INP_INFO_WLOCK(&V_divcbinfo); 500 error = in_pcballoc(so, &V_divcbinfo); 501 if (error) { 502 INP_INFO_WUNLOCK(&V_divcbinfo); 503 return error; 504 } 505 inp = (struct inpcb *)so->so_pcb; 506 INP_INFO_WUNLOCK(&V_divcbinfo); 507 inp->inp_ip_p = proto; 508 inp->inp_vflag |= INP_IPV4; 509 inp->inp_flags |= INP_HDRINCL; 510 INP_WUNLOCK(inp); 511 return 0; 512 } 513 514 static void 515 div_detach(struct socket *so) 516 { 517 INIT_VNET_INET(so->so_vnet); 518 struct inpcb *inp; 519 520 inp = sotoinpcb(so); 521 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 522 INP_INFO_WLOCK(&V_divcbinfo); 523 INP_WLOCK(inp); 524 in_pcbdetach(inp); 525 in_pcbfree(inp); 526 INP_INFO_WUNLOCK(&V_divcbinfo); 527 } 528 529 static int 530 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 531 { 532 INIT_VNET_INET(so->so_vnet); 533 struct inpcb *inp; 534 int error; 535 536 inp = sotoinpcb(so); 537 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 538 /* in_pcbbind assumes that nam is a sockaddr_in 539 * and in_pcbbind requires a valid address. Since divert 540 * sockets don't we need to make sure the address is 541 * filled in properly. 542 * XXX -- divert should not be abusing in_pcbind 543 * and should probably have its own family. 544 */ 545 if (nam->sa_family != AF_INET) 546 return EAFNOSUPPORT; 547 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 548 INP_INFO_WLOCK(&V_divcbinfo); 549 INP_WLOCK(inp); 550 error = in_pcbbind(inp, nam, td->td_ucred); 551 INP_WUNLOCK(inp); 552 INP_INFO_WUNLOCK(&V_divcbinfo); 553 return error; 554 } 555 556 static int 557 div_shutdown(struct socket *so) 558 { 559 struct inpcb *inp; 560 561 inp = sotoinpcb(so); 562 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 563 INP_WLOCK(inp); 564 socantsendmore(so); 565 INP_WUNLOCK(inp); 566 return 0; 567 } 568 569 static int 570 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 571 struct mbuf *control, struct thread *td) 572 { 573 INIT_VNET_INET(so->so_vnet); 574 575 /* Packet must have a header (but that's about it) */ 576 if (m->m_len < sizeof (struct ip) && 577 (m = m_pullup(m, sizeof (struct ip))) == 0) { 578 IPSTAT_INC(ips_toosmall); 579 m_freem(m); 580 return EINVAL; 581 } 582 583 /* Send packet */ 584 return div_output(so, m, (struct sockaddr_in *)nam, control); 585 } 586 587 void 588 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 589 { 590 struct in_addr faddr; 591 592 faddr = ((struct sockaddr_in *)sa)->sin_addr; 593 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 594 return; 595 if (PRC_IS_REDIRECT(cmd)) 596 return; 597 } 598 599 static int 600 div_pcblist(SYSCTL_HANDLER_ARGS) 601 { 602 INIT_VNET_INET(curvnet); 603 int error, i, n; 604 struct inpcb *inp, **inp_list; 605 inp_gen_t gencnt; 606 struct xinpgen xig; 607 608 /* 609 * The process of preparing the TCB list is too time-consuming and 610 * resource-intensive to repeat twice on every request. 611 */ 612 if (req->oldptr == 0) { 613 n = V_divcbinfo.ipi_count; 614 req->oldidx = 2 * (sizeof xig) 615 + (n + n/8) * sizeof(struct xinpcb); 616 return 0; 617 } 618 619 if (req->newptr != 0) 620 return EPERM; 621 622 /* 623 * OK, now we're committed to doing something. 624 */ 625 INP_INFO_RLOCK(&V_divcbinfo); 626 gencnt = V_divcbinfo.ipi_gencnt; 627 n = V_divcbinfo.ipi_count; 628 INP_INFO_RUNLOCK(&V_divcbinfo); 629 630 error = sysctl_wire_old_buffer(req, 631 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 632 if (error != 0) 633 return (error); 634 635 xig.xig_len = sizeof xig; 636 xig.xig_count = n; 637 xig.xig_gen = gencnt; 638 xig.xig_sogen = so_gencnt; 639 error = SYSCTL_OUT(req, &xig, sizeof xig); 640 if (error) 641 return error; 642 643 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 644 if (inp_list == 0) 645 return ENOMEM; 646 647 INP_INFO_RLOCK(&V_divcbinfo); 648 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 649 inp = LIST_NEXT(inp, inp_list)) { 650 INP_RLOCK(inp); 651 if (inp->inp_gencnt <= gencnt && 652 cr_canseeinpcb(req->td->td_ucred, inp) == 0) 653 inp_list[i++] = inp; 654 INP_RUNLOCK(inp); 655 } 656 INP_INFO_RUNLOCK(&V_divcbinfo); 657 n = i; 658 659 error = 0; 660 for (i = 0; i < n; i++) { 661 inp = inp_list[i]; 662 INP_RLOCK(inp); 663 if (inp->inp_gencnt <= gencnt) { 664 struct xinpcb xi; 665 bzero(&xi, sizeof(xi)); 666 xi.xi_len = sizeof xi; 667 /* XXX should avoid extra copy */ 668 bcopy(inp, &xi.xi_inp, sizeof *inp); 669 if (inp->inp_socket) 670 sotoxsocket(inp->inp_socket, &xi.xi_socket); 671 INP_RUNLOCK(inp); 672 error = SYSCTL_OUT(req, &xi, sizeof xi); 673 } else 674 INP_RUNLOCK(inp); 675 } 676 if (!error) { 677 /* 678 * Give the user an updated idea of our state. 679 * If the generation differs from what we told 680 * her before, she knows that something happened 681 * while we were processing this request, and it 682 * might be necessary to retry. 683 */ 684 INP_INFO_RLOCK(&V_divcbinfo); 685 xig.xig_gen = V_divcbinfo.ipi_gencnt; 686 xig.xig_sogen = so_gencnt; 687 xig.xig_count = V_divcbinfo.ipi_count; 688 INP_INFO_RUNLOCK(&V_divcbinfo); 689 error = SYSCTL_OUT(req, &xig, sizeof xig); 690 } 691 free(inp_list, M_TEMP); 692 return error; 693 } 694 695 #ifdef SYSCTL_NODE 696 SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 697 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 698 div_pcblist, "S,xinpcb", "List of active divert sockets"); 699 #endif 700 701 struct pr_usrreqs div_usrreqs = { 702 .pru_attach = div_attach, 703 .pru_bind = div_bind, 704 .pru_control = in_control, 705 .pru_detach = div_detach, 706 .pru_peeraddr = in_getpeeraddr, 707 .pru_send = div_send, 708 .pru_shutdown = div_shutdown, 709 .pru_sockaddr = in_getsockaddr, 710 .pru_sosetlabel = in_pcbsosetlabel 711 }; 712 713 struct protosw div_protosw = { 714 .pr_type = SOCK_RAW, 715 .pr_protocol = IPPROTO_DIVERT, 716 .pr_flags = PR_ATOMIC|PR_ADDR, 717 .pr_input = div_input, 718 .pr_ctlinput = div_ctlinput, 719 .pr_ctloutput = ip_ctloutput, 720 .pr_init = div_init, 721 .pr_usrreqs = &div_usrreqs 722 }; 723 724 static int 725 div_modevent(module_t mod, int type, void *unused) 726 { 727 INIT_VNET_INET(curvnet); /* XXX move to iattach - revisit!!! */ 728 int err = 0; 729 int n; 730 731 switch (type) { 732 case MOD_LOAD: 733 /* 734 * Protocol will be initialized by pf_proto_register(). 735 * We don't have to register ip_protox because we are not 736 * a true IP protocol that goes over the wire. 737 */ 738 err = pf_proto_register(PF_INET, &div_protosw); 739 ip_divert_ptr = divert_packet; 740 break; 741 case MOD_QUIESCE: 742 /* 743 * IPDIVERT may normally not be unloaded because of the 744 * potential race conditions. Tell kldunload we can't be 745 * unloaded unless the unload is forced. 746 */ 747 err = EPERM; 748 break; 749 case MOD_UNLOAD: 750 /* 751 * Forced unload. 752 * 753 * Module ipdivert can only be unloaded if no sockets are 754 * connected. Maybe this can be changed later to forcefully 755 * disconnect any open sockets. 756 * 757 * XXXRW: Note that there is a slight race here, as a new 758 * socket open request could be spinning on the lock and then 759 * we destroy the lock. 760 */ 761 INP_INFO_WLOCK(&V_divcbinfo); 762 n = V_divcbinfo.ipi_count; 763 if (n != 0) { 764 err = EBUSY; 765 INP_INFO_WUNLOCK(&V_divcbinfo); 766 break; 767 } 768 ip_divert_ptr = NULL; 769 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 770 INP_INFO_WUNLOCK(&V_divcbinfo); 771 INP_INFO_LOCK_DESTROY(&V_divcbinfo); 772 uma_zdestroy(V_divcbinfo.ipi_zone); 773 break; 774 default: 775 err = EOPNOTSUPP; 776 break; 777 } 778 return err; 779 } 780 781 static moduledata_t ipdivertmod = { 782 "ipdivert", 783 div_modevent, 784 0 785 }; 786 787 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 788 MODULE_DEPEND(dummynet, ipfw, 2, 2, 2); 789 MODULE_VERSION(ipdivert, 1); 790