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 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include "opt_sctp.h" 36 #ifndef INET 37 #error "IPDIVERT requires INET" 38 #endif 39 40 #include <sys/param.h> 41 #include <sys/kernel.h> 42 #include <sys/lock.h> 43 #include <sys/malloc.h> 44 #include <sys/mbuf.h> 45 #include <sys/module.h> 46 #include <sys/kernel.h> 47 #include <sys/priv.h> 48 #include <sys/proc.h> 49 #include <sys/protosw.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/sysctl.h> 53 #include <net/vnet.h> 54 55 #include <net/if.h> 56 #include <net/netisr.h> 57 58 #include <netinet/in.h> 59 #include <netinet/in_pcb.h> 60 #include <netinet/in_systm.h> 61 #include <netinet/in_var.h> 62 #include <netinet/ip.h> 63 #include <netinet/ip_var.h> 64 #ifdef INET6 65 #include <netinet/ip6.h> 66 #include <netinet6/ip6_var.h> 67 #endif 68 #ifdef SCTP 69 #include <netinet/sctp_crc32.h> 70 #endif 71 72 #include <security/mac/mac_framework.h> 73 74 /* 75 * Divert sockets 76 */ 77 78 /* 79 * Allocate enough space to hold a full IP packet 80 */ 81 #define DIVSNDQ (65536 + 100) 82 #define DIVRCVQ (65536 + 100) 83 84 /* 85 * Divert sockets work in conjunction with ipfw or other packet filters, 86 * see the divert(4) manpage for features. 87 * Packets are selected by the packet filter and tagged with an 88 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by 89 * the packet filter) and information on the matching filter rule for 90 * subsequent reinjection. The divert_port is used to put the packet 91 * on the corresponding divert socket, while the rule number is passed 92 * up (at least partially) as the sin_port in the struct sockaddr. 93 * 94 * Packets written to the divert socket carry in sin_addr a 95 * destination address, and in sin_port the number of the filter rule 96 * after which to continue processing. 97 * If the destination address is INADDR_ANY, the packet is treated as 98 * as outgoing and sent to ip_output(); otherwise it is treated as 99 * incoming and sent to ip_input(). 100 * Further, sin_zero carries some information on the interface, 101 * which can be used in the reinject -- see comments in the code. 102 * 103 * On reinjection, processing in ip_input() and ip_output() 104 * will be exactly the same as for the original packet, except that 105 * packet filter processing will start at the rule number after the one 106 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0 107 * will apply the entire ruleset to the packet). 108 */ 109 110 /* Internal variables. */ 111 static VNET_DEFINE(struct inpcbhead, divcb); 112 static VNET_DEFINE(struct inpcbinfo, divcbinfo); 113 114 #define V_divcb VNET(divcb) 115 #define V_divcbinfo VNET(divcbinfo) 116 117 static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 118 static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 119 120 static eventhandler_tag ip_divert_event_tag; 121 122 /* 123 * Initialize divert connection block queue. 124 */ 125 static void 126 div_zone_change(void *tag) 127 { 128 129 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 130 } 131 132 static int 133 div_inpcb_init(void *mem, int size, int flags) 134 { 135 struct inpcb *inp = mem; 136 137 INP_LOCK_INIT(inp, "inp", "divinp"); 138 return (0); 139 } 140 141 static void 142 div_inpcb_fini(void *mem, int size) 143 { 144 struct inpcb *inp = mem; 145 146 INP_LOCK_DESTROY(inp); 147 } 148 149 static void 150 div_init(void) 151 { 152 153 /* 154 * XXX We don't use the hash list for divert IP, but it's easier to 155 * allocate one-entry hash lists than it is to check all over the 156 * place for hashbase == NULL. 157 */ 158 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb", 159 div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE, 160 IPI_HASHFIELDS_NONE); 161 } 162 163 static void 164 div_destroy(void) 165 { 166 167 in_pcbinfo_destroy(&V_divcbinfo); 168 } 169 170 /* 171 * IPPROTO_DIVERT is not in the real IP protocol number space; this 172 * function should never be called. Just in case, drop any packets. 173 */ 174 static void 175 div_input(struct mbuf *m, int off) 176 { 177 178 KMOD_IPSTAT_INC(ips_noproto); 179 m_freem(m); 180 } 181 182 /* 183 * Divert a packet by passing it up to the divert socket at port 'port'. 184 * 185 * Setup generic address and protocol structures for div_input routine, 186 * then pass them along with mbuf chain. 187 */ 188 static void 189 divert_packet(struct mbuf *m, int incoming) 190 { 191 struct ip *ip; 192 struct inpcb *inp; 193 struct socket *sa; 194 u_int16_t nport; 195 struct sockaddr_in divsrc; 196 struct m_tag *mtag; 197 198 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 199 if (mtag == NULL) { 200 m_freem(m); 201 return; 202 } 203 /* Assure header */ 204 if (m->m_len < sizeof(struct ip) && 205 (m = m_pullup(m, sizeof(struct ip))) == 0) 206 return; 207 ip = mtod(m, struct ip *); 208 209 /* Delayed checksums are currently not compatible with divert. */ 210 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 211 ip->ip_len = ntohs(ip->ip_len); 212 in_delayed_cksum(m); 213 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 214 ip->ip_len = htons(ip->ip_len); 215 } 216 #ifdef SCTP 217 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 218 ip->ip_len = ntohs(ip->ip_len); 219 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 220 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 221 ip->ip_len = htons(ip->ip_len); 222 } 223 #endif 224 bzero(&divsrc, sizeof(divsrc)); 225 divsrc.sin_len = sizeof(divsrc); 226 divsrc.sin_family = AF_INET; 227 /* record matching rule, in host format */ 228 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 229 /* 230 * Record receive interface address, if any. 231 * But only for incoming packets. 232 */ 233 if (incoming) { 234 struct ifaddr *ifa; 235 struct ifnet *ifp; 236 237 /* Sanity check */ 238 M_ASSERTPKTHDR(m); 239 240 /* Find IP address for receive interface */ 241 ifp = m->m_pkthdr.rcvif; 242 if_addr_rlock(ifp); 243 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 244 if (ifa->ifa_addr->sa_family != AF_INET) 245 continue; 246 divsrc.sin_addr = 247 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 248 break; 249 } 250 if_addr_runlock(ifp); 251 } 252 /* 253 * Record the incoming interface name whenever we have one. 254 */ 255 if (m->m_pkthdr.rcvif) { 256 /* 257 * Hide the actual interface name in there in the 258 * sin_zero array. XXX This needs to be moved to a 259 * different sockaddr type for divert, e.g. 260 * sockaddr_div with multiple fields like 261 * sockaddr_dl. Presently we have only 7 bytes 262 * but that will do for now as most interfaces 263 * are 4 or less + 2 or less bytes for unit. 264 * There is probably a faster way of doing this, 265 * possibly taking it from the sockaddr_dl on the iface. 266 * This solves the problem of a P2P link and a LAN interface 267 * having the same address, which can result in the wrong 268 * interface being assigned to the packet when fed back 269 * into the divert socket. Theoretically if the daemon saves 270 * and re-uses the sockaddr_in as suggested in the man pages, 271 * this iface name will come along for the ride. 272 * (see div_output for the other half of this.) 273 */ 274 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 275 sizeof(divsrc.sin_zero)); 276 } 277 278 /* Put packet on socket queue, if any */ 279 sa = NULL; 280 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 281 INP_INFO_RLOCK(&V_divcbinfo); 282 LIST_FOREACH(inp, &V_divcb, inp_list) { 283 /* XXX why does only one socket match? */ 284 if (inp->inp_lport == nport) { 285 INP_RLOCK(inp); 286 sa = inp->inp_socket; 287 SOCKBUF_LOCK(&sa->so_rcv); 288 if (sbappendaddr_locked(&sa->so_rcv, 289 (struct sockaddr *)&divsrc, m, 290 (struct mbuf *)0) == 0) { 291 SOCKBUF_UNLOCK(&sa->so_rcv); 292 sa = NULL; /* force mbuf reclaim below */ 293 } else 294 sorwakeup_locked(sa); 295 INP_RUNLOCK(inp); 296 break; 297 } 298 } 299 INP_INFO_RUNLOCK(&V_divcbinfo); 300 if (sa == NULL) { 301 m_freem(m); 302 KMOD_IPSTAT_INC(ips_noproto); 303 KMOD_IPSTAT_DEC(ips_delivered); 304 } 305 } 306 307 /* 308 * Deliver packet back into the IP processing machinery. 309 * 310 * If no address specified, or address is 0.0.0.0, send to ip_output(); 311 * otherwise, send to ip_input() and mark as having been received on 312 * the interface with that address. 313 */ 314 static int 315 div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 316 struct mbuf *control) 317 { 318 struct ip *const ip = mtod(m, struct ip *); 319 struct m_tag *mtag; 320 struct ipfw_rule_ref *dt; 321 int error = 0; 322 323 /* 324 * An mbuf may hasn't come from userland, but we pretend 325 * that it has. 326 */ 327 m->m_pkthdr.rcvif = NULL; 328 m->m_nextpkt = NULL; 329 M_SETFIB(m, so->so_fibnum); 330 331 if (control) 332 m_freem(control); /* XXX */ 333 334 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 335 if (mtag == NULL) { 336 /* this should be normal */ 337 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 338 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 339 if (mtag == NULL) { 340 error = ENOBUFS; 341 goto cantsend; 342 } 343 m_tag_prepend(m, mtag); 344 } 345 dt = (struct ipfw_rule_ref *)(mtag+1); 346 347 /* Loopback avoidance and state recovery */ 348 if (sin) { 349 int i; 350 351 /* set the starting point. We provide a non-zero slot, 352 * but a non_matching chain_id to skip that info and use 353 * the rulenum/rule_id. 354 */ 355 dt->slot = 1; /* dummy, chain_id is invalid */ 356 dt->chain_id = 0; 357 dt->rulenum = sin->sin_port+1; /* host format ? */ 358 dt->rule_id = 0; 359 /* 360 * Find receive interface with the given name, stuffed 361 * (if it exists) in the sin_zero[] field. 362 * The name is user supplied data so don't trust its size 363 * or that it is zero terminated. 364 */ 365 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 366 ; 367 if ( i > 0 && i < sizeof(sin->sin_zero)) 368 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 369 } 370 371 /* Reinject packet into the system as incoming or outgoing */ 372 if (!sin || sin->sin_addr.s_addr == 0) { 373 struct mbuf *options = NULL; 374 struct inpcb *inp; 375 376 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 377 inp = sotoinpcb(so); 378 INP_RLOCK(inp); 379 switch (ip->ip_v) { 380 case IPVERSION: 381 /* 382 * Don't allow both user specified and setsockopt 383 * options, and don't allow packet length sizes that 384 * will crash. 385 */ 386 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 387 inp->inp_options != NULL) || 388 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 389 error = EINVAL; 390 INP_RUNLOCK(inp); 391 goto cantsend; 392 } 393 394 /* 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 break; 398 #ifdef INET6 399 case IPV6_VERSION >> 4: 400 { 401 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 402 403 /* Don't allow packet length sizes that will crash */ 404 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 405 error = EINVAL; 406 INP_RUNLOCK(inp); 407 goto cantsend; 408 } 409 410 ip6->ip6_plen = ntohs(ip6->ip6_plen); 411 break; 412 } 413 #endif 414 default: 415 error = EINVAL; 416 INP_RUNLOCK(inp); 417 goto cantsend; 418 } 419 420 /* Send packet to output processing */ 421 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 422 423 #ifdef MAC 424 mac_inpcb_create_mbuf(inp, m); 425 #endif 426 /* 427 * Get ready to inject the packet into ip_output(). 428 * Just in case socket options were specified on the 429 * divert socket, we duplicate them. This is done 430 * to avoid having to hold the PCB locks over the call 431 * to ip_output(), as doing this results in a number of 432 * lock ordering complexities. 433 * 434 * Note that we set the multicast options argument for 435 * ip_output() to NULL since it should be invariant that 436 * they are not present. 437 */ 438 KASSERT(inp->inp_moptions == NULL, 439 ("multicast options set on a divert socket")); 440 /* 441 * XXXCSJP: It is unclear to me whether or not it makes 442 * sense for divert sockets to have options. However, 443 * for now we will duplicate them with the INP locks 444 * held so we can use them in ip_output() without 445 * requring a reference to the pcb. 446 */ 447 if (inp->inp_options != NULL) { 448 options = m_dup(inp->inp_options, M_NOWAIT); 449 if (options == NULL) { 450 INP_RUNLOCK(inp); 451 error = ENOBUFS; 452 goto cantsend; 453 } 454 } 455 INP_RUNLOCK(inp); 456 457 switch (ip->ip_v) { 458 case IPVERSION: 459 error = ip_output(m, options, NULL, 460 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 461 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 462 break; 463 #ifdef INET6 464 case IPV6_VERSION >> 4: 465 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 466 break; 467 #endif 468 } 469 if (options != NULL) 470 m_freem(options); 471 } else { 472 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 473 if (m->m_pkthdr.rcvif == NULL) { 474 /* 475 * No luck with the name, check by IP address. 476 * Clear the port and the ifname to make sure 477 * there are no distractions for ifa_ifwithaddr. 478 */ 479 struct ifaddr *ifa; 480 481 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 482 sin->sin_port = 0; 483 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 484 if (ifa == NULL) { 485 error = EADDRNOTAVAIL; 486 goto cantsend; 487 } 488 m->m_pkthdr.rcvif = ifa->ifa_ifp; 489 ifa_free(ifa); 490 } 491 #ifdef MAC 492 mac_socket_create_mbuf(so, m); 493 #endif 494 /* Send packet to input processing via netisr */ 495 switch (ip->ip_v) { 496 case IPVERSION: 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))) == 0) { 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; 639 640 /* 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 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 676 if (inp_list == 0) 677 return ENOMEM; 678 679 INP_INFO_RLOCK(&V_divcbinfo); 680 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 681 inp = LIST_NEXT(inp, inp_list)) { 682 INP_WLOCK(inp); 683 if (inp->inp_gencnt <= gencnt && 684 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 685 in_pcbref(inp); 686 inp_list[i++] = inp; 687 } 688 INP_WUNLOCK(inp); 689 } 690 INP_INFO_RUNLOCK(&V_divcbinfo); 691 n = i; 692 693 error = 0; 694 for (i = 0; i < n; i++) { 695 inp = inp_list[i]; 696 INP_RLOCK(inp); 697 if (inp->inp_gencnt <= gencnt) { 698 struct xinpcb xi; 699 bzero(&xi, sizeof(xi)); 700 xi.xi_len = sizeof xi; 701 /* XXX should avoid extra copy */ 702 bcopy(inp, &xi.xi_inp, sizeof *inp); 703 if (inp->inp_socket) 704 sotoxsocket(inp->inp_socket, &xi.xi_socket); 705 INP_RUNLOCK(inp); 706 error = SYSCTL_OUT(req, &xi, sizeof xi); 707 } else 708 INP_RUNLOCK(inp); 709 } 710 INP_INFO_WLOCK(&V_divcbinfo); 711 for (i = 0; i < n; i++) { 712 inp = inp_list[i]; 713 INP_RLOCK(inp); 714 if (!in_pcbrele_rlocked(inp)) 715 INP_RUNLOCK(inp); 716 } 717 INP_INFO_WUNLOCK(&V_divcbinfo); 718 719 if (!error) { 720 /* 721 * Give the user an updated idea of our state. 722 * If the generation differs from what we told 723 * her before, she knows that something happened 724 * while we were processing this request, and it 725 * might be necessary to retry. 726 */ 727 INP_INFO_RLOCK(&V_divcbinfo); 728 xig.xig_gen = V_divcbinfo.ipi_gencnt; 729 xig.xig_sogen = so_gencnt; 730 xig.xig_count = V_divcbinfo.ipi_count; 731 INP_INFO_RUNLOCK(&V_divcbinfo); 732 error = SYSCTL_OUT(req, &xig, sizeof xig); 733 } 734 free(inp_list, M_TEMP); 735 return error; 736 } 737 738 #ifdef SYSCTL_NODE 739 static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, 740 "IPDIVERT"); 741 SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 742 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets"); 743 #endif 744 745 struct pr_usrreqs div_usrreqs = { 746 .pru_attach = div_attach, 747 .pru_bind = div_bind, 748 .pru_control = in_control, 749 .pru_detach = div_detach, 750 .pru_peeraddr = in_getpeeraddr, 751 .pru_send = div_send, 752 .pru_shutdown = div_shutdown, 753 .pru_sockaddr = in_getsockaddr, 754 .pru_sosetlabel = in_pcbsosetlabel 755 }; 756 757 struct protosw div_protosw = { 758 .pr_type = SOCK_RAW, 759 .pr_protocol = IPPROTO_DIVERT, 760 .pr_flags = PR_ATOMIC|PR_ADDR, 761 .pr_input = div_input, 762 .pr_ctlinput = div_ctlinput, 763 .pr_ctloutput = ip_ctloutput, 764 .pr_init = div_init, 765 #ifdef VIMAGE 766 .pr_destroy = div_destroy, 767 #endif 768 .pr_usrreqs = &div_usrreqs 769 }; 770 771 static int 772 div_modevent(module_t mod, int type, void *unused) 773 { 774 int err = 0; 775 #ifndef VIMAGE 776 int n; 777 #endif 778 779 switch (type) { 780 case MOD_LOAD: 781 /* 782 * Protocol will be initialized by pf_proto_register(). 783 * We don't have to register ip_protox because we are not 784 * a true IP protocol that goes over the wire. 785 */ 786 err = pf_proto_register(PF_INET, &div_protosw); 787 if (err != 0) 788 return (err); 789 ip_divert_ptr = divert_packet; 790 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, 791 div_zone_change, NULL, EVENTHANDLER_PRI_ANY); 792 break; 793 case MOD_QUIESCE: 794 /* 795 * IPDIVERT may normally not be unloaded because of the 796 * potential race conditions. Tell kldunload we can't be 797 * unloaded unless the unload is forced. 798 */ 799 err = EPERM; 800 break; 801 case MOD_UNLOAD: 802 #ifdef VIMAGE 803 err = EPERM; 804 break; 805 #else 806 /* 807 * Forced unload. 808 * 809 * Module ipdivert can only be unloaded if no sockets are 810 * connected. Maybe this can be changed later to forcefully 811 * disconnect any open sockets. 812 * 813 * XXXRW: Note that there is a slight race here, as a new 814 * socket open request could be spinning on the lock and then 815 * we destroy the lock. 816 */ 817 INP_INFO_WLOCK(&V_divcbinfo); 818 n = V_divcbinfo.ipi_count; 819 if (n != 0) { 820 err = EBUSY; 821 INP_INFO_WUNLOCK(&V_divcbinfo); 822 break; 823 } 824 ip_divert_ptr = NULL; 825 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 826 INP_INFO_WUNLOCK(&V_divcbinfo); 827 div_destroy(); 828 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); 829 break; 830 #endif /* !VIMAGE */ 831 default: 832 err = EOPNOTSUPP; 833 break; 834 } 835 return err; 836 } 837 838 static moduledata_t ipdivertmod = { 839 "ipdivert", 840 div_modevent, 841 0 842 }; 843 844 DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 845 MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2); 846 MODULE_VERSION(ipdivert, 1); 847