1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 #include "opt_sctp.h" 38 #ifndef INET 39 #error "IPDIVERT requires INET" 40 #endif 41 42 #include <sys/param.h> 43 #include <sys/eventhandler.h> 44 #include <sys/kernel.h> 45 #include <sys/lock.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/module.h> 49 #include <sys/kernel.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/sysctl.h> 56 #include <net/vnet.h> 57 58 #include <net/if.h> 59 #include <net/if_var.h> 60 #include <net/netisr.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_pcb.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip_var.h> 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #include <netinet6/ip6_var.h> 71 #endif 72 #ifdef SCTP 73 #include <netinet/sctp_crc32.h> 74 #endif 75 76 #include <security/mac/mac_framework.h> 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 or other packet filters, 89 * see the divert(4) manpage for features. 90 * Packets are selected by the packet filter and tagged with an 91 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by 92 * the packet filter) and information on the matching filter rule for 93 * subsequent reinjection. The divert_port is used to put the packet 94 * on the corresponding divert socket, while the rule number is passed 95 * up (at least partially) as the sin_port in the struct sockaddr. 96 * 97 * Packets written to the divert socket carry in sin_addr a 98 * destination address, and in sin_port the number of the filter rule 99 * after which to continue processing. 100 * If the destination address is INADDR_ANY, the packet is treated as 101 * as outgoing and sent to ip_output(); otherwise it is treated as 102 * incoming and sent to ip_input(). 103 * Further, sin_zero carries some information on the interface, 104 * which can be used in the reinject -- see comments in the code. 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 * packet filter processing will start at the rule number after the one 109 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0 110 * will apply the entire ruleset to the packet). 111 */ 112 113 /* Internal variables. */ 114 VNET_DEFINE_STATIC(struct inpcbhead, divcb); 115 VNET_DEFINE_STATIC(struct inpcbinfo, divcbinfo); 116 117 #define V_divcb VNET(divcb) 118 #define V_divcbinfo VNET(divcbinfo) 119 120 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 121 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 122 123 static eventhandler_tag ip_divert_event_tag; 124 125 /* 126 * Initialize divert connection block queue. 127 */ 128 static void 129 div_zone_change(void *tag) 130 { 131 132 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 133 } 134 135 static int 136 div_inpcb_init(void *mem, int size, int flags) 137 { 138 struct inpcb *inp = mem; 139 140 INP_LOCK_INIT(inp, "inp", "divinp"); 141 return (0); 142 } 143 144 static void 145 div_init(void) 146 { 147 148 /* 149 * XXX We don't use the hash list for divert IP, but it's easier to 150 * allocate one-entry hash lists than it is to check all over the 151 * place for hashbase == NULL. 152 */ 153 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb", 154 div_inpcb_init, IPI_HASHFIELDS_NONE); 155 } 156 157 static void 158 div_destroy(void *unused __unused) 159 { 160 161 in_pcbinfo_destroy(&V_divcbinfo); 162 } 163 VNET_SYSUNINIT(divert, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, 164 div_destroy, NULL); 165 166 /* 167 * IPPROTO_DIVERT is not in the real IP protocol number space; this 168 * function should never be called. Just in case, drop any packets. 169 */ 170 static int 171 div_input(struct mbuf **mp, int *offp, int proto) 172 { 173 struct mbuf *m = *mp; 174 175 KMOD_IPSTAT_INC(ips_noproto); 176 m_freem(m); 177 return (IPPROTO_DONE); 178 } 179 180 /* 181 * Divert a packet by passing it up to the divert socket at port 'port'. 182 * 183 * Setup generic address and protocol structures for div_input routine, 184 * then pass them along with mbuf chain. 185 */ 186 static void 187 divert_packet(struct mbuf *m, bool incoming) 188 { 189 struct ip *ip; 190 struct inpcb *inp; 191 struct socket *sa; 192 u_int16_t nport; 193 struct sockaddr_in divsrc; 194 struct m_tag *mtag; 195 struct epoch_tracker et; 196 197 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 198 if (mtag == NULL) { 199 m_freem(m); 200 return; 201 } 202 /* Assure header */ 203 if (m->m_len < sizeof(struct ip) && 204 (m = m_pullup(m, sizeof(struct ip))) == NULL) 205 return; 206 ip = mtod(m, struct ip *); 207 208 /* Delayed checksums are currently not compatible with divert. */ 209 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 210 in_delayed_cksum(m); 211 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 212 } 213 #ifdef SCTP 214 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 215 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 216 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 217 } 218 #endif 219 bzero(&divsrc, sizeof(divsrc)); 220 divsrc.sin_len = sizeof(divsrc); 221 divsrc.sin_family = AF_INET; 222 /* record matching rule, in host format */ 223 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 224 /* 225 * Record receive interface address, if any. 226 * But only for incoming packets. 227 */ 228 if (incoming) { 229 struct ifaddr *ifa; 230 struct ifnet *ifp; 231 232 /* Sanity check */ 233 M_ASSERTPKTHDR(m); 234 235 /* Find IP address for receive interface */ 236 ifp = m->m_pkthdr.rcvif; 237 if_addr_rlock(ifp); 238 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 239 if (ifa->ifa_addr->sa_family != AF_INET) 240 continue; 241 divsrc.sin_addr = 242 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 243 break; 244 } 245 if_addr_runlock(ifp); 246 } 247 /* 248 * Record the incoming interface name whenever we have one. 249 */ 250 if (m->m_pkthdr.rcvif) { 251 /* 252 * Hide the actual interface name in there in the 253 * sin_zero array. XXX This needs to be moved to a 254 * different sockaddr type for divert, e.g. 255 * sockaddr_div with multiple fields like 256 * sockaddr_dl. Presently we have only 7 bytes 257 * but that will do for now as most interfaces 258 * are 4 or less + 2 or less bytes for unit. 259 * There is probably a faster way of doing this, 260 * possibly taking it from the sockaddr_dl on the iface. 261 * This solves the problem of a P2P link and a LAN interface 262 * having the same address, which can result in the wrong 263 * interface being assigned to the packet when fed back 264 * into the divert socket. Theoretically if the daemon saves 265 * and re-uses the sockaddr_in as suggested in the man pages, 266 * this iface name will come along for the ride. 267 * (see div_output for the other half of this.) 268 */ 269 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 270 sizeof(divsrc.sin_zero)); 271 } 272 273 /* Put packet on socket queue, if any */ 274 sa = NULL; 275 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 276 INP_INFO_RLOCK_ET(&V_divcbinfo, et); 277 CK_LIST_FOREACH(inp, &V_divcb, inp_list) { 278 /* XXX why does only one socket match? */ 279 if (inp->inp_lport == nport) { 280 INP_RLOCK(inp); 281 sa = inp->inp_socket; 282 SOCKBUF_LOCK(&sa->so_rcv); 283 if (sbappendaddr_locked(&sa->so_rcv, 284 (struct sockaddr *)&divsrc, m, 285 (struct mbuf *)0) == 0) { 286 SOCKBUF_UNLOCK(&sa->so_rcv); 287 sa = NULL; /* force mbuf reclaim below */ 288 } else 289 sorwakeup_locked(sa); 290 INP_RUNLOCK(inp); 291 break; 292 } 293 } 294 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); 295 if (sa == NULL) { 296 m_freem(m); 297 KMOD_IPSTAT_INC(ips_noproto); 298 KMOD_IPSTAT_DEC(ips_delivered); 299 } 300 } 301 302 /* 303 * Deliver packet back into the IP processing machinery. 304 * 305 * If no address specified, or address is 0.0.0.0, send to ip_output(); 306 * otherwise, send to ip_input() and mark as having been received on 307 * the interface with that address. 308 */ 309 static int 310 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 311 struct mbuf *control) 312 { 313 struct ip *const ip = mtod(m, struct ip *); 314 struct m_tag *mtag; 315 struct ipfw_rule_ref *dt; 316 int error = 0; 317 318 /* 319 * An mbuf may hasn't come from userland, but we pretend 320 * that it has. 321 */ 322 m->m_pkthdr.rcvif = NULL; 323 m->m_nextpkt = NULL; 324 M_SETFIB(m, so->so_fibnum); 325 326 if (control) 327 m_freem(control); /* XXX */ 328 329 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 330 if (mtag == NULL) { 331 /* this should be normal */ 332 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 333 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 334 if (mtag == NULL) { 335 error = ENOBUFS; 336 goto cantsend; 337 } 338 m_tag_prepend(m, mtag); 339 } 340 dt = (struct ipfw_rule_ref *)(mtag+1); 341 342 /* Loopback avoidance and state recovery */ 343 if (sin) { 344 int i; 345 346 /* set the starting point. We provide a non-zero slot, 347 * but a non_matching chain_id to skip that info and use 348 * the rulenum/rule_id. 349 */ 350 dt->slot = 1; /* dummy, chain_id is invalid */ 351 dt->chain_id = 0; 352 dt->rulenum = sin->sin_port+1; /* host format ? */ 353 dt->rule_id = 0; 354 /* 355 * Find receive interface with the given name, stuffed 356 * (if it exists) in the sin_zero[] field. 357 * The name is user supplied data so don't trust its size 358 * or that it is zero terminated. 359 */ 360 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 361 ; 362 if ( i > 0 && i < sizeof(sin->sin_zero)) 363 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 364 } 365 366 /* Reinject packet into the system as incoming or outgoing */ 367 if (!sin || sin->sin_addr.s_addr == 0) { 368 struct mbuf *options = NULL; 369 struct inpcb *inp; 370 371 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 372 inp = sotoinpcb(so); 373 INP_RLOCK(inp); 374 switch (ip->ip_v) { 375 case IPVERSION: 376 /* 377 * Don't allow both user specified and setsockopt 378 * options, and don't allow packet length sizes that 379 * will crash. 380 */ 381 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 382 inp->inp_options != NULL) || 383 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 384 error = EINVAL; 385 INP_RUNLOCK(inp); 386 goto cantsend; 387 } 388 break; 389 #ifdef INET6 390 case IPV6_VERSION >> 4: 391 { 392 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 393 394 /* Don't allow packet length sizes that will crash */ 395 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 396 error = EINVAL; 397 INP_RUNLOCK(inp); 398 goto cantsend; 399 } 400 break; 401 } 402 #endif 403 default: 404 error = EINVAL; 405 INP_RUNLOCK(inp); 406 goto cantsend; 407 } 408 409 /* Send packet to output processing */ 410 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 411 412 #ifdef MAC 413 mac_inpcb_create_mbuf(inp, m); 414 #endif 415 /* 416 * Get ready to inject the packet into ip_output(). 417 * Just in case socket options were specified on the 418 * divert socket, we duplicate them. This is done 419 * to avoid having to hold the PCB locks over the call 420 * to ip_output(), as doing this results in a number of 421 * lock ordering complexities. 422 * 423 * Note that we set the multicast options argument for 424 * ip_output() to NULL since it should be invariant that 425 * they are not present. 426 */ 427 KASSERT(inp->inp_moptions == NULL, 428 ("multicast options set on a divert socket")); 429 /* 430 * XXXCSJP: It is unclear to me whether or not it makes 431 * sense for divert sockets to have options. However, 432 * for now we will duplicate them with the INP locks 433 * held so we can use them in ip_output() without 434 * requring a reference to the pcb. 435 */ 436 if (inp->inp_options != NULL) { 437 options = m_dup(inp->inp_options, M_NOWAIT); 438 if (options == NULL) { 439 INP_RUNLOCK(inp); 440 error = ENOBUFS; 441 goto cantsend; 442 } 443 } 444 INP_RUNLOCK(inp); 445 446 switch (ip->ip_v) { 447 case IPVERSION: 448 error = ip_output(m, options, NULL, 449 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 450 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 451 break; 452 #ifdef INET6 453 case IPV6_VERSION >> 4: 454 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 455 break; 456 #endif 457 } 458 if (options != NULL) 459 m_freem(options); 460 } else { 461 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 462 if (m->m_pkthdr.rcvif == NULL) { 463 /* 464 * No luck with the name, check by IP address. 465 * Clear the port and the ifname to make sure 466 * there are no distractions for ifa_ifwithaddr. 467 */ 468 struct epoch_tracker et; 469 struct ifaddr *ifa; 470 471 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 472 sin->sin_port = 0; 473 NET_EPOCH_ENTER(et); 474 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 475 if (ifa == NULL) { 476 error = EADDRNOTAVAIL; 477 NET_EPOCH_EXIT(et); 478 goto cantsend; 479 } 480 m->m_pkthdr.rcvif = ifa->ifa_ifp; 481 NET_EPOCH_EXIT(et); 482 } 483 #ifdef MAC 484 mac_socket_create_mbuf(so, m); 485 #endif 486 /* Send packet to input processing via netisr */ 487 switch (ip->ip_v) { 488 case IPVERSION: 489 /* 490 * Restore M_BCAST flag when destination address is 491 * broadcast. It is expected by ip_tryforward(). 492 */ 493 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 494 m->m_flags |= M_MCAST; 495 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 496 m->m_flags |= M_BCAST; 497 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 498 break; 499 #ifdef INET6 500 case IPV6_VERSION >> 4: 501 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); 502 break; 503 #endif 504 default: 505 error = EINVAL; 506 goto cantsend; 507 } 508 } 509 510 return (error); 511 512 cantsend: 513 m_freem(m); 514 return (error); 515 } 516 517 static int 518 div_attach(struct socket *so, int proto, struct thread *td) 519 { 520 struct inpcb *inp; 521 int error; 522 523 inp = sotoinpcb(so); 524 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 525 if (td != NULL) { 526 error = priv_check(td, PRIV_NETINET_DIVERT); 527 if (error) 528 return (error); 529 } 530 error = soreserve(so, div_sendspace, div_recvspace); 531 if (error) 532 return error; 533 INP_INFO_WLOCK(&V_divcbinfo); 534 error = in_pcballoc(so, &V_divcbinfo); 535 if (error) { 536 INP_INFO_WUNLOCK(&V_divcbinfo); 537 return error; 538 } 539 inp = (struct inpcb *)so->so_pcb; 540 INP_INFO_WUNLOCK(&V_divcbinfo); 541 inp->inp_ip_p = proto; 542 inp->inp_vflag |= INP_IPV4; 543 inp->inp_flags |= INP_HDRINCL; 544 INP_WUNLOCK(inp); 545 return 0; 546 } 547 548 static void 549 div_detach(struct socket *so) 550 { 551 struct inpcb *inp; 552 553 inp = sotoinpcb(so); 554 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 555 INP_INFO_WLOCK(&V_divcbinfo); 556 INP_WLOCK(inp); 557 in_pcbdetach(inp); 558 in_pcbfree(inp); 559 INP_INFO_WUNLOCK(&V_divcbinfo); 560 } 561 562 static int 563 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 564 { 565 struct inpcb *inp; 566 int error; 567 568 inp = sotoinpcb(so); 569 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 570 /* in_pcbbind assumes that nam is a sockaddr_in 571 * and in_pcbbind requires a valid address. Since divert 572 * sockets don't we need to make sure the address is 573 * filled in properly. 574 * XXX -- divert should not be abusing in_pcbind 575 * and should probably have its own family. 576 */ 577 if (nam->sa_family != AF_INET) 578 return EAFNOSUPPORT; 579 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 580 INP_INFO_WLOCK(&V_divcbinfo); 581 INP_WLOCK(inp); 582 INP_HASH_WLOCK(&V_divcbinfo); 583 error = in_pcbbind(inp, nam, td->td_ucred); 584 INP_HASH_WUNLOCK(&V_divcbinfo); 585 INP_WUNLOCK(inp); 586 INP_INFO_WUNLOCK(&V_divcbinfo); 587 return error; 588 } 589 590 static int 591 div_shutdown(struct socket *so) 592 { 593 struct inpcb *inp; 594 595 inp = sotoinpcb(so); 596 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 597 INP_WLOCK(inp); 598 socantsendmore(so); 599 INP_WUNLOCK(inp); 600 return 0; 601 } 602 603 static int 604 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 605 struct mbuf *control, struct thread *td) 606 { 607 608 /* Packet must have a header (but that's about it) */ 609 if (m->m_len < sizeof (struct ip) && 610 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 611 KMOD_IPSTAT_INC(ips_toosmall); 612 m_freem(m); 613 return EINVAL; 614 } 615 616 /* Send packet */ 617 return div_output(so, m, (struct sockaddr_in *)nam, control); 618 } 619 620 static void 621 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 622 { 623 struct in_addr faddr; 624 625 faddr = ((struct sockaddr_in *)sa)->sin_addr; 626 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 627 return; 628 if (PRC_IS_REDIRECT(cmd)) 629 return; 630 } 631 632 static int 633 div_pcblist(SYSCTL_HANDLER_ARGS) 634 { 635 int error, i, n; 636 struct inpcb *inp, **inp_list; 637 inp_gen_t gencnt; 638 struct xinpgen xig; 639 struct epoch_tracker et; 640 641 /* 642 * The process of preparing the TCB list is too time-consuming and 643 * resource-intensive to repeat twice on every request. 644 */ 645 if (req->oldptr == 0) { 646 n = V_divcbinfo.ipi_count; 647 n += imax(n / 8, 10); 648 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 649 return 0; 650 } 651 652 if (req->newptr != 0) 653 return EPERM; 654 655 /* 656 * OK, now we're committed to doing something. 657 */ 658 INP_INFO_WLOCK(&V_divcbinfo); 659 gencnt = V_divcbinfo.ipi_gencnt; 660 n = V_divcbinfo.ipi_count; 661 INP_INFO_WUNLOCK(&V_divcbinfo); 662 663 error = sysctl_wire_old_buffer(req, 664 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 665 if (error != 0) 666 return (error); 667 668 bzero(&xig, sizeof(xig)); 669 xig.xig_len = sizeof xig; 670 xig.xig_count = n; 671 xig.xig_gen = gencnt; 672 xig.xig_sogen = so_gencnt; 673 error = SYSCTL_OUT(req, &xig, sizeof xig); 674 if (error) 675 return error; 676 677 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 678 if (inp_list == NULL) 679 return ENOMEM; 680 681 INP_INFO_RLOCK_ET(&V_divcbinfo, et); 682 for (inp = CK_LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 683 inp = CK_LIST_NEXT(inp, inp_list)) { 684 INP_WLOCK(inp); 685 if (inp->inp_gencnt <= gencnt && 686 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 687 in_pcbref(inp); 688 inp_list[i++] = inp; 689 } 690 INP_WUNLOCK(inp); 691 } 692 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); 693 n = i; 694 695 error = 0; 696 for (i = 0; i < n; i++) { 697 inp = inp_list[i]; 698 INP_RLOCK(inp); 699 if (inp->inp_gencnt <= gencnt) { 700 struct xinpcb xi; 701 702 in_pcbtoxinpcb(inp, &xi); 703 INP_RUNLOCK(inp); 704 error = SYSCTL_OUT(req, &xi, sizeof xi); 705 } else 706 INP_RUNLOCK(inp); 707 } 708 INP_INFO_WLOCK(&V_divcbinfo); 709 for (i = 0; i < n; i++) { 710 inp = inp_list[i]; 711 INP_RLOCK(inp); 712 if (!in_pcbrele_rlocked(inp)) 713 INP_RUNLOCK(inp); 714 } 715 INP_INFO_WUNLOCK(&V_divcbinfo); 716 717 if (!error) { 718 struct epoch_tracker et; 719 /* 720 * Give the user an updated idea of our state. 721 * If the generation differs from what we told 722 * her before, she knows that something happened 723 * while we were processing this request, and it 724 * might be necessary to retry. 725 */ 726 INP_INFO_RLOCK_ET(&V_divcbinfo, et); 727 xig.xig_gen = V_divcbinfo.ipi_gencnt; 728 xig.xig_sogen = so_gencnt; 729 xig.xig_count = V_divcbinfo.ipi_count; 730 INP_INFO_RUNLOCK_ET(&V_divcbinfo, et); 731 error = SYSCTL_OUT(req, &xig, sizeof xig); 732 } 733 free(inp_list, M_TEMP); 734 return error; 735 } 736 737 #ifdef SYSCTL_NODE 738 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, 739 "IPDIVERT"); 740 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 741 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets"); 742 #endif 743 744 struct pr_usrreqs div_usrreqs = { 745 .pru_attach = div_attach, 746 .pru_bind = div_bind, 747 .pru_control = in_control, 748 .pru_detach = div_detach, 749 .pru_peeraddr = in_getpeeraddr, 750 .pru_send = div_send, 751 .pru_shutdown = div_shutdown, 752 .pru_sockaddr = in_getsockaddr, 753 .pru_sosetlabel = in_pcbsosetlabel 754 }; 755 756 struct protosw div_protosw = { 757 .pr_type = SOCK_RAW, 758 .pr_protocol = IPPROTO_DIVERT, 759 .pr_flags = PR_ATOMIC|PR_ADDR, 760 .pr_input = div_input, 761 .pr_ctlinput = div_ctlinput, 762 .pr_ctloutput = ip_ctloutput, 763 .pr_init = div_init, 764 .pr_usrreqs = &div_usrreqs 765 }; 766 767 static int 768 div_modevent(module_t mod, int type, void *unused) 769 { 770 int err = 0; 771 772 switch (type) { 773 case MOD_LOAD: 774 /* 775 * Protocol will be initialized by pf_proto_register(). 776 * We don't have to register ip_protox because we are not 777 * a true IP protocol that goes over the wire. 778 */ 779 err = pf_proto_register(PF_INET, &div_protosw); 780 if (err != 0) 781 return (err); 782 ip_divert_ptr = divert_packet; 783 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, 784 div_zone_change, NULL, EVENTHANDLER_PRI_ANY); 785 break; 786 case MOD_QUIESCE: 787 /* 788 * IPDIVERT may normally not be unloaded because of the 789 * potential race conditions. Tell kldunload we can't be 790 * unloaded unless the unload is forced. 791 */ 792 err = EPERM; 793 break; 794 case MOD_UNLOAD: 795 /* 796 * Forced unload. 797 * 798 * Module ipdivert can only be unloaded if no sockets are 799 * connected. Maybe this can be changed later to forcefully 800 * disconnect any open sockets. 801 * 802 * XXXRW: Note that there is a slight race here, as a new 803 * socket open request could be spinning on the lock and then 804 * we destroy the lock. 805 */ 806 INP_INFO_WLOCK(&V_divcbinfo); 807 if (V_divcbinfo.ipi_count != 0) { 808 err = EBUSY; 809 INP_INFO_WUNLOCK(&V_divcbinfo); 810 break; 811 } 812 ip_divert_ptr = NULL; 813 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 814 INP_INFO_WUNLOCK(&V_divcbinfo); 815 #ifndef VIMAGE 816 div_destroy(NULL); 817 #endif 818 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); 819 break; 820 default: 821 err = EOPNOTSUPP; 822 break; 823 } 824 return err; 825 } 826 827 static moduledata_t ipdivertmod = { 828 "ipdivert", 829 div_modevent, 830 0 831 }; 832 833 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 834 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3); 835 MODULE_VERSION(ipdivert, 1); 836