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 static VNET_DEFINE(struct inpcbhead, divcb); 115 static VNET_DEFINE(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, int 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; 195 196 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 197 if (mtag == NULL) { 198 m_freem(m); 199 return; 200 } 201 /* Assure header */ 202 if (m->m_len < sizeof(struct ip) && 203 (m = m_pullup(m, sizeof(struct ip))) == NULL) 204 return; 205 ip = mtod(m, struct ip *); 206 207 /* Delayed checksums are currently not compatible with divert. */ 208 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 209 in_delayed_cksum(m); 210 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 211 } 212 #ifdef SCTP 213 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 214 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 215 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 216 } 217 #endif 218 bzero(&divsrc, sizeof(divsrc)); 219 divsrc.sin_len = sizeof(divsrc); 220 divsrc.sin_family = AF_INET; 221 /* record matching rule, in host format */ 222 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 223 /* 224 * Record receive interface address, if any. 225 * But only for incoming packets. 226 */ 227 if (incoming) { 228 struct ifaddr *ifa; 229 struct ifnet *ifp; 230 231 /* Sanity check */ 232 M_ASSERTPKTHDR(m); 233 234 /* Find IP address for receive interface */ 235 ifp = m->m_pkthdr.rcvif; 236 if_addr_rlock(ifp); 237 CK_STAILQ_FOREACH(ifa, &ifp->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 if_addr_runlock(ifp); 245 } 246 /* 247 * Record the incoming interface name whenever we have one. 248 */ 249 if (m->m_pkthdr.rcvif) { 250 /* 251 * Hide the actual interface name in there in the 252 * sin_zero array. XXX This needs to be moved to a 253 * different sockaddr type for divert, e.g. 254 * sockaddr_div with multiple fields like 255 * sockaddr_dl. Presently we have only 7 bytes 256 * but that will do for now as most interfaces 257 * are 4 or less + 2 or less bytes for unit. 258 * There is probably a faster way of doing this, 259 * possibly taking it from the sockaddr_dl on the iface. 260 * This solves the problem of a P2P link and a LAN interface 261 * having the same address, which can result in the wrong 262 * interface being assigned to the packet when fed back 263 * into the divert socket. Theoretically if the daemon saves 264 * and re-uses the sockaddr_in as suggested in the man pages, 265 * this iface name will come along for the ride. 266 * (see div_output for the other half of this.) 267 */ 268 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 269 sizeof(divsrc.sin_zero)); 270 } 271 272 /* Put packet on socket queue, if any */ 273 sa = NULL; 274 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 275 INP_INFO_RLOCK(&V_divcbinfo); 276 LIST_FOREACH(inp, &V_divcb, inp_list) { 277 /* XXX why does only one socket match? */ 278 if (inp->inp_lport == nport) { 279 INP_RLOCK(inp); 280 sa = inp->inp_socket; 281 SOCKBUF_LOCK(&sa->so_rcv); 282 if (sbappendaddr_locked(&sa->so_rcv, 283 (struct sockaddr *)&divsrc, m, 284 (struct mbuf *)0) == 0) { 285 SOCKBUF_UNLOCK(&sa->so_rcv); 286 sa = NULL; /* force mbuf reclaim below */ 287 } else 288 sorwakeup_locked(sa); 289 INP_RUNLOCK(inp); 290 break; 291 } 292 } 293 INP_INFO_RUNLOCK(&V_divcbinfo); 294 if (sa == NULL) { 295 m_freem(m); 296 KMOD_IPSTAT_INC(ips_noproto); 297 KMOD_IPSTAT_DEC(ips_delivered); 298 } 299 } 300 301 /* 302 * Deliver packet back into the IP processing machinery. 303 * 304 * If no address specified, or address is 0.0.0.0, send to ip_output(); 305 * otherwise, send to ip_input() and mark as having been received on 306 * the interface with that address. 307 */ 308 static int 309 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 310 struct mbuf *control) 311 { 312 struct ip *const ip = mtod(m, struct ip *); 313 struct m_tag *mtag; 314 struct ipfw_rule_ref *dt; 315 int error = 0; 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 M_SETFIB(m, so->so_fibnum); 324 325 if (control) 326 m_freem(control); /* XXX */ 327 328 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 329 if (mtag == NULL) { 330 /* this should be normal */ 331 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 332 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 333 if (mtag == NULL) { 334 error = ENOBUFS; 335 goto cantsend; 336 } 337 m_tag_prepend(m, mtag); 338 } 339 dt = (struct ipfw_rule_ref *)(mtag+1); 340 341 /* Loopback avoidance and state recovery */ 342 if (sin) { 343 int i; 344 345 /* set the starting point. We provide a non-zero slot, 346 * but a non_matching chain_id to skip that info and use 347 * the rulenum/rule_id. 348 */ 349 dt->slot = 1; /* dummy, chain_id is invalid */ 350 dt->chain_id = 0; 351 dt->rulenum = sin->sin_port+1; /* host format ? */ 352 dt->rule_id = 0; 353 /* 354 * Find receive interface with the given name, stuffed 355 * (if it exists) in the sin_zero[] field. 356 * The name is user supplied data so don't trust its size 357 * or that it is zero terminated. 358 */ 359 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 360 ; 361 if ( i > 0 && i < sizeof(sin->sin_zero)) 362 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 363 } 364 365 /* Reinject packet into the system as incoming or outgoing */ 366 if (!sin || sin->sin_addr.s_addr == 0) { 367 struct mbuf *options = NULL; 368 struct inpcb *inp; 369 370 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 371 inp = sotoinpcb(so); 372 INP_RLOCK(inp); 373 switch (ip->ip_v) { 374 case IPVERSION: 375 /* 376 * Don't allow both user specified and setsockopt 377 * options, and don't allow packet length sizes that 378 * will crash. 379 */ 380 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 381 inp->inp_options != NULL) || 382 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 383 error = EINVAL; 384 INP_RUNLOCK(inp); 385 goto cantsend; 386 } 387 break; 388 #ifdef INET6 389 case IPV6_VERSION >> 4: 390 { 391 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 392 393 /* Don't allow packet length sizes that will crash */ 394 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 395 error = EINVAL; 396 INP_RUNLOCK(inp); 397 goto cantsend; 398 } 399 break; 400 } 401 #endif 402 default: 403 error = EINVAL; 404 INP_RUNLOCK(inp); 405 goto cantsend; 406 } 407 408 /* Send packet to output processing */ 409 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 410 411 #ifdef MAC 412 mac_inpcb_create_mbuf(inp, m); 413 #endif 414 /* 415 * Get ready to inject the packet into ip_output(). 416 * Just in case socket options were specified on the 417 * divert socket, we duplicate them. This is done 418 * to avoid having to hold the PCB locks over the call 419 * to ip_output(), as doing this results in a number of 420 * lock ordering complexities. 421 * 422 * Note that we set the multicast options argument for 423 * ip_output() to NULL since it should be invariant that 424 * they are not present. 425 */ 426 KASSERT(inp->inp_moptions == NULL, 427 ("multicast options set on a divert socket")); 428 /* 429 * XXXCSJP: It is unclear to me whether or not it makes 430 * sense for divert sockets to have options. However, 431 * for now we will duplicate them with the INP locks 432 * held so we can use them in ip_output() without 433 * requring a reference to the pcb. 434 */ 435 if (inp->inp_options != NULL) { 436 options = m_dup(inp->inp_options, M_NOWAIT); 437 if (options == NULL) { 438 INP_RUNLOCK(inp); 439 error = ENOBUFS; 440 goto cantsend; 441 } 442 } 443 INP_RUNLOCK(inp); 444 445 switch (ip->ip_v) { 446 case IPVERSION: 447 error = ip_output(m, options, NULL, 448 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 449 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 450 break; 451 #ifdef INET6 452 case IPV6_VERSION >> 4: 453 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 454 break; 455 #endif 456 } 457 if (options != NULL) 458 m_freem(options); 459 } else { 460 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 461 if (m->m_pkthdr.rcvif == NULL) { 462 /* 463 * No luck with the name, check by IP address. 464 * Clear the port and the ifname to make sure 465 * there are no distractions for ifa_ifwithaddr. 466 */ 467 struct ifaddr *ifa; 468 469 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 470 sin->sin_port = 0; 471 NET_EPOCH_ENTER(); 472 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 473 if (ifa == NULL) { 474 error = EADDRNOTAVAIL; 475 NET_EPOCH_EXIT(); 476 goto cantsend; 477 } 478 m->m_pkthdr.rcvif = ifa->ifa_ifp; 479 NET_EPOCH_EXIT(); 480 } 481 #ifdef MAC 482 mac_socket_create_mbuf(so, m); 483 #endif 484 /* Send packet to input processing via netisr */ 485 switch (ip->ip_v) { 486 case IPVERSION: 487 /* 488 * Restore M_BCAST flag when destination address is 489 * broadcast. It is expected by ip_tryforward(). 490 */ 491 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) 492 m->m_flags |= M_MCAST; 493 else if (in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 494 m->m_flags |= M_BCAST; 495 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 496 break; 497 #ifdef INET6 498 case IPV6_VERSION >> 4: 499 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); 500 break; 501 #endif 502 default: 503 error = EINVAL; 504 goto cantsend; 505 } 506 } 507 508 return (error); 509 510 cantsend: 511 m_freem(m); 512 return (error); 513 } 514 515 static int 516 div_attach(struct socket *so, int proto, struct thread *td) 517 { 518 struct inpcb *inp; 519 int error; 520 521 inp = sotoinpcb(so); 522 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 523 if (td != NULL) { 524 error = priv_check(td, PRIV_NETINET_DIVERT); 525 if (error) 526 return (error); 527 } 528 error = soreserve(so, div_sendspace, div_recvspace); 529 if (error) 530 return error; 531 INP_INFO_WLOCK(&V_divcbinfo); 532 error = in_pcballoc(so, &V_divcbinfo); 533 if (error) { 534 INP_INFO_WUNLOCK(&V_divcbinfo); 535 return error; 536 } 537 inp = (struct inpcb *)so->so_pcb; 538 INP_INFO_WUNLOCK(&V_divcbinfo); 539 inp->inp_ip_p = proto; 540 inp->inp_vflag |= INP_IPV4; 541 inp->inp_flags |= INP_HDRINCL; 542 INP_WUNLOCK(inp); 543 return 0; 544 } 545 546 static void 547 div_detach(struct socket *so) 548 { 549 struct inpcb *inp; 550 551 inp = sotoinpcb(so); 552 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 553 INP_INFO_WLOCK(&V_divcbinfo); 554 INP_WLOCK(inp); 555 /* XXX defer destruction to epoch_call */ 556 in_pcbdetach(inp); 557 in_pcbfree(inp); 558 INP_INFO_WUNLOCK(&V_divcbinfo); 559 } 560 561 static int 562 div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 563 { 564 struct inpcb *inp; 565 int error; 566 567 inp = sotoinpcb(so); 568 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 569 /* in_pcbbind assumes that nam is a sockaddr_in 570 * and in_pcbbind requires a valid address. Since divert 571 * sockets don't we need to make sure the address is 572 * filled in properly. 573 * XXX -- divert should not be abusing in_pcbind 574 * and should probably have its own family. 575 */ 576 if (nam->sa_family != AF_INET) 577 return EAFNOSUPPORT; 578 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 579 INP_INFO_WLOCK(&V_divcbinfo); 580 INP_WLOCK(inp); 581 INP_HASH_WLOCK(&V_divcbinfo); 582 error = in_pcbbind(inp, nam, td->td_ucred); 583 INP_HASH_WUNLOCK(&V_divcbinfo); 584 INP_WUNLOCK(inp); 585 INP_INFO_WUNLOCK(&V_divcbinfo); 586 return error; 587 } 588 589 static int 590 div_shutdown(struct socket *so) 591 { 592 struct inpcb *inp; 593 594 inp = sotoinpcb(so); 595 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 596 INP_WLOCK(inp); 597 socantsendmore(so); 598 INP_WUNLOCK(inp); 599 return 0; 600 } 601 602 static int 603 div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 604 struct mbuf *control, struct thread *td) 605 { 606 607 /* Packet must have a header (but that's about it) */ 608 if (m->m_len < sizeof (struct ip) && 609 (m = m_pullup(m, sizeof (struct ip))) == NULL) { 610 KMOD_IPSTAT_INC(ips_toosmall); 611 m_freem(m); 612 return EINVAL; 613 } 614 615 /* Send packet */ 616 return div_output(so, m, (struct sockaddr_in *)nam, control); 617 } 618 619 static void 620 div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 621 { 622 struct in_addr faddr; 623 624 faddr = ((struct sockaddr_in *)sa)->sin_addr; 625 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 626 return; 627 if (PRC_IS_REDIRECT(cmd)) 628 return; 629 } 630 631 static int 632 div_pcblist(SYSCTL_HANDLER_ARGS) 633 { 634 int error, i, n; 635 struct in_pcblist *il; 636 struct inpcb *inp, **inp_list; 637 inp_gen_t gencnt; 638 struct xinpgen xig; 639 640 /* 641 * The process of preparing the TCB list is too time-consuming and 642 * resource-intensive to repeat twice on every request. 643 */ 644 if (req->oldptr == 0) { 645 n = V_divcbinfo.ipi_count; 646 n += imax(n / 8, 10); 647 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 648 return 0; 649 } 650 651 if (req->newptr != 0) 652 return EPERM; 653 654 /* 655 * OK, now we're committed to doing something. 656 */ 657 INP_INFO_RLOCK(&V_divcbinfo); 658 gencnt = V_divcbinfo.ipi_gencnt; 659 n = V_divcbinfo.ipi_count; 660 INP_INFO_RUNLOCK(&V_divcbinfo); 661 662 error = sysctl_wire_old_buffer(req, 663 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 664 if (error != 0) 665 return (error); 666 667 xig.xig_len = sizeof xig; 668 xig.xig_count = n; 669 xig.xig_gen = gencnt; 670 xig.xig_sogen = so_gencnt; 671 error = SYSCTL_OUT(req, &xig, sizeof xig); 672 if (error) 673 return error; 674 675 il = malloc(sizeof(struct in_pcblist) + n * sizeof(struct inpcb *), M_TEMP, M_WAITOK|M_ZERO_INVARIANTS); 676 inp_list = il->il_inp_list; 677 678 INP_INFO_RLOCK(&V_divcbinfo); 679 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 680 inp = LIST_NEXT(inp, inp_list)) { 681 INP_WLOCK(inp); 682 if (inp->inp_gencnt <= gencnt && 683 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 684 in_pcbref(inp); 685 inp_list[i++] = inp; 686 } 687 INP_WUNLOCK(inp); 688 } 689 INP_INFO_RUNLOCK(&V_divcbinfo); 690 n = i; 691 692 error = 0; 693 for (i = 0; i < n; i++) { 694 inp = inp_list[i]; 695 INP_RLOCK(inp); 696 if (inp->inp_gencnt <= gencnt) { 697 struct xinpcb xi; 698 699 in_pcbtoxinpcb(inp, &xi); 700 INP_RUNLOCK(inp); 701 error = SYSCTL_OUT(req, &xi, sizeof xi); 702 } else 703 INP_RUNLOCK(inp); 704 } 705 il->il_count = n; 706 il->il_pcbinfo = &V_divcbinfo; 707 epoch_call(net_epoch_preempt, &il->il_epoch_ctx, in_pcblist_rele_rlocked); 708 709 if (!error) { 710 /* 711 * Give the user an updated idea of our state. 712 * If the generation differs from what we told 713 * her before, she knows that something happened 714 * while we were processing this request, and it 715 * might be necessary to retry. 716 */ 717 INP_INFO_RLOCK(&V_divcbinfo); 718 xig.xig_gen = V_divcbinfo.ipi_gencnt; 719 xig.xig_sogen = so_gencnt; 720 xig.xig_count = V_divcbinfo.ipi_count; 721 INP_INFO_RUNLOCK(&V_divcbinfo); 722 error = SYSCTL_OUT(req, &xig, sizeof xig); 723 } 724 return error; 725 } 726 727 #ifdef SYSCTL_NODE 728 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, 729 "IPDIVERT"); 730 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 731 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets"); 732 #endif 733 734 struct pr_usrreqs div_usrreqs = { 735 .pru_attach = div_attach, 736 .pru_bind = div_bind, 737 .pru_control = in_control, 738 .pru_detach = div_detach, 739 .pru_peeraddr = in_getpeeraddr, 740 .pru_send = div_send, 741 .pru_shutdown = div_shutdown, 742 .pru_sockaddr = in_getsockaddr, 743 .pru_sosetlabel = in_pcbsosetlabel 744 }; 745 746 struct protosw div_protosw = { 747 .pr_type = SOCK_RAW, 748 .pr_protocol = IPPROTO_DIVERT, 749 .pr_flags = PR_ATOMIC|PR_ADDR, 750 .pr_input = div_input, 751 .pr_ctlinput = div_ctlinput, 752 .pr_ctloutput = ip_ctloutput, 753 .pr_init = div_init, 754 .pr_usrreqs = &div_usrreqs 755 }; 756 757 static int 758 div_modevent(module_t mod, int type, void *unused) 759 { 760 int err = 0; 761 762 switch (type) { 763 case MOD_LOAD: 764 /* 765 * Protocol will be initialized by pf_proto_register(). 766 * We don't have to register ip_protox because we are not 767 * a true IP protocol that goes over the wire. 768 */ 769 err = pf_proto_register(PF_INET, &div_protosw); 770 if (err != 0) 771 return (err); 772 ip_divert_ptr = divert_packet; 773 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, 774 div_zone_change, NULL, EVENTHANDLER_PRI_ANY); 775 break; 776 case MOD_QUIESCE: 777 /* 778 * IPDIVERT may normally not be unloaded because of the 779 * potential race conditions. Tell kldunload we can't be 780 * unloaded unless the unload is forced. 781 */ 782 err = EPERM; 783 break; 784 case MOD_UNLOAD: 785 /* 786 * Forced unload. 787 * 788 * Module ipdivert can only be unloaded if no sockets are 789 * connected. Maybe this can be changed later to forcefully 790 * disconnect any open sockets. 791 * 792 * XXXRW: Note that there is a slight race here, as a new 793 * socket open request could be spinning on the lock and then 794 * we destroy the lock. 795 */ 796 INP_INFO_WLOCK(&V_divcbinfo); 797 if (V_divcbinfo.ipi_count != 0) { 798 err = EBUSY; 799 INP_INFO_WUNLOCK(&V_divcbinfo); 800 break; 801 } 802 ip_divert_ptr = NULL; 803 /* XXX defer to epoch_call ? */ 804 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 805 INP_INFO_WUNLOCK(&V_divcbinfo); 806 #ifndef VIMAGE 807 div_destroy(NULL); 808 #endif 809 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); 810 break; 811 default: 812 err = EOPNOTSUPP; 813 break; 814 } 815 return err; 816 } 817 818 static moduledata_t ipdivertmod = { 819 "ipdivert", 820 div_modevent, 821 0 822 }; 823 824 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_FIREWALL, SI_ORDER_ANY); 825 MODULE_DEPEND(ipdivert, ipfw, 3, 3, 3); 826 MODULE_VERSION(ipdivert, 1); 827