ip_divert.c revision 223593
1230479Snetchild/*- 2230479Snetchild * Copyright (c) 1982, 1986, 1988, 1993 3230479Snetchild * The Regents of the University of California. All rights reserved. 4230479Snetchild * 5230479Snetchild * Redistribution and use in source and binary forms, with or without 6230479Snetchild * modification, are permitted provided that the following conditions 7230479Snetchild * are met: 8230479Snetchild * 1. Redistributions of source code must retain the above copyright 9230479Snetchild * notice, this list of conditions and the following disclaimer. 10230479Snetchild * 2. Redistributions in binary form must reproduce the above copyright 11230479Snetchild * notice, this list of conditions and the following disclaimer in the 12230479Snetchild * documentation and/or other materials provided with the distribution. 13230479Snetchild * 4. Neither the name of the University nor the names of its contributors 14230479Snetchild * may be used to endorse or promote products derived from this software 15230479Snetchild * without specific prior written permission. 16230479Snetchild * 17230479Snetchild * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18230479Snetchild * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19230479Snetchild * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20230479Snetchild * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21230479Snetchild * 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: head/sys/netinet/ip_divert.c 223593 2011-06-27 12:21:11Z glebius $"); 32 33#if !defined(KLD_MODULE) 34#include "opt_inet.h" 35#include "opt_sctp.h" 36#ifndef INET 37#error "IPDIVERT requires INET." 38#endif 39#endif 40#include "opt_inet6.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/lock.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/module.h> 48#include <sys/kernel.h> 49#include <sys/priv.h> 50#include <sys/proc.h> 51#include <sys/protosw.h> 52#include <sys/socket.h> 53#include <sys/socketvar.h> 54#include <sys/sysctl.h> 55#include <net/vnet.h> 56 57#include <net/if.h> 58#include <net/netisr.h> 59 60#include <netinet/in.h> 61#include <netinet/in_pcb.h> 62#include <netinet/in_systm.h> 63#include <netinet/in_var.h> 64#include <netinet/ip.h> 65#include <netinet/ip_var.h> 66#ifdef INET6 67#include <netinet/ip6.h> 68#include <netinet6/ip6_var.h> 69#endif 70#ifdef SCTP 71#include <netinet/sctp_crc32.h> 72#endif 73 74#include <security/mac/mac_framework.h> 75 76/* 77 * Divert sockets 78 */ 79 80/* 81 * Allocate enough space to hold a full IP packet 82 */ 83#define DIVSNDQ (65536 + 100) 84#define DIVRCVQ (65536 + 100) 85 86/* 87 * Divert sockets work in conjunction with ipfw or other packet filters, 88 * see the divert(4) manpage for features. 89 * Packets are selected by the packet filter and tagged with an 90 * MTAG_IPFW_RULE tag carrying the 'divert port' number (as set by 91 * the packet filter) and information on the matching filter rule for 92 * subsequent reinjection. The divert_port is used to put the packet 93 * on the corresponding divert socket, while the rule number is passed 94 * up (at least partially) as the sin_port in the struct sockaddr. 95 * 96 * Packets written to the divert socket carry in sin_addr a 97 * destination address, and in sin_port the number of the filter rule 98 * after which to continue processing. 99 * If the destination address is INADDR_ANY, the packet is treated as 100 * as outgoing and sent to ip_output(); otherwise it is treated as 101 * incoming and sent to ip_input(). 102 * Further, sin_zero carries some information on the interface, 103 * which can be used in the reinject -- see comments in the code. 104 * 105 * On reinjection, processing in ip_input() and ip_output() 106 * will be exactly the same as for the original packet, except that 107 * packet filter processing will start at the rule number after the one 108 * written in the sin_port (ipfw does not allow a rule #0, so sin_port=0 109 * will apply the entire ruleset to the packet). 110 */ 111 112/* Internal variables. */ 113static VNET_DEFINE(struct inpcbhead, divcb); 114static VNET_DEFINE(struct inpcbinfo, divcbinfo); 115 116#define V_divcb VNET(divcb) 117#define V_divcbinfo VNET(divcbinfo) 118 119static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 120static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 121 122static eventhandler_tag ip_divert_event_tag; 123 124/* 125 * Initialize divert connection block queue. 126 */ 127static void 128div_zone_change(void *tag) 129{ 130 131 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 132} 133 134static int 135div_inpcb_init(void *mem, int size, int flags) 136{ 137 struct inpcb *inp = mem; 138 139 INP_LOCK_INIT(inp, "inp", "divinp"); 140 return (0); 141} 142 143static void 144div_inpcb_fini(void *mem, int size) 145{ 146 struct inpcb *inp = mem; 147 148 INP_LOCK_DESTROY(inp); 149} 150 151static void 152div_init(void) 153{ 154 155 /* 156 * XXX We don't use the hash list for divert IP, but it's easier to 157 * allocate one-entry hash lists than it is to check all over the 158 * place for hashbase == NULL. 159 */ 160 in_pcbinfo_init(&V_divcbinfo, "div", &V_divcb, 1, 1, "divcb", 161 div_inpcb_init, div_inpcb_fini, UMA_ZONE_NOFREE, 162 IPI_HASHFIELDS_NONE); 163} 164 165static void 166div_destroy(void) 167{ 168 169 in_pcbinfo_destroy(&V_divcbinfo); 170} 171 172/* 173 * IPPROTO_DIVERT is not in the real IP protocol number space; this 174 * function should never be called. Just in case, drop any packets. 175 */ 176static void 177div_input(struct mbuf *m, int off) 178{ 179 180 KMOD_IPSTAT_INC(ips_noproto); 181 m_freem(m); 182} 183 184/* 185 * Divert a packet by passing it up to the divert socket at port 'port'. 186 * 187 * Setup generic address and protocol structures for div_input routine, 188 * then pass them along with mbuf chain. 189 */ 190static void 191divert_packet(struct mbuf *m, int incoming) 192{ 193 struct ip *ip; 194 struct inpcb *inp; 195 struct socket *sa; 196 u_int16_t nport; 197 struct sockaddr_in divsrc; 198 struct m_tag *mtag; 199 200 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 201 if (mtag == NULL) { 202 m_freem(m); 203 return; 204 } 205 /* Assure header */ 206 if (m->m_len < sizeof(struct ip) && 207 (m = m_pullup(m, sizeof(struct ip))) == 0) 208 return; 209 ip = mtod(m, struct ip *); 210 211 /* Delayed checksums are currently not compatible with divert. */ 212 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 213 ip->ip_len = ntohs(ip->ip_len); 214 in_delayed_cksum(m); 215 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 216 ip->ip_len = htons(ip->ip_len); 217 } 218#ifdef SCTP 219 if (m->m_pkthdr.csum_flags & CSUM_SCTP) { 220 ip->ip_len = ntohs(ip->ip_len); 221 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 222 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 223 ip->ip_len = htons(ip->ip_len); 224 } 225#endif 226 bzero(&divsrc, sizeof(divsrc)); 227 divsrc.sin_len = sizeof(divsrc); 228 divsrc.sin_family = AF_INET; 229 /* record matching rule, in host format */ 230 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 231 /* 232 * Record receive interface address, if any. 233 * But only for incoming packets. 234 */ 235 if (incoming) { 236 struct ifaddr *ifa; 237 struct ifnet *ifp; 238 239 /* Sanity check */ 240 M_ASSERTPKTHDR(m); 241 242 /* Find IP address for receive interface */ 243 ifp = m->m_pkthdr.rcvif; 244 if_addr_rlock(ifp); 245 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 246 if (ifa->ifa_addr->sa_family != AF_INET) 247 continue; 248 divsrc.sin_addr = 249 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 250 break; 251 } 252 if_addr_runlock(ifp); 253 } 254 /* 255 * Record the incoming interface name whenever we have one. 256 */ 257 if (m->m_pkthdr.rcvif) { 258 /* 259 * Hide the actual interface name in there in the 260 * sin_zero array. XXX This needs to be moved to a 261 * different sockaddr type for divert, e.g. 262 * sockaddr_div with multiple fields like 263 * sockaddr_dl. Presently we have only 7 bytes 264 * but that will do for now as most interfaces 265 * are 4 or less + 2 or less bytes for unit. 266 * There is probably a faster way of doing this, 267 * possibly taking it from the sockaddr_dl on the iface. 268 * This solves the problem of a P2P link and a LAN interface 269 * having the same address, which can result in the wrong 270 * interface being assigned to the packet when fed back 271 * into the divert socket. Theoretically if the daemon saves 272 * and re-uses the sockaddr_in as suggested in the man pages, 273 * this iface name will come along for the ride. 274 * (see div_output for the other half of this.) 275 */ 276 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 277 sizeof(divsrc.sin_zero)); 278 } 279 280 /* Put packet on socket queue, if any */ 281 sa = NULL; 282 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 283 INP_INFO_RLOCK(&V_divcbinfo); 284 LIST_FOREACH(inp, &V_divcb, inp_list) { 285 /* XXX why does only one socket match? */ 286 if (inp->inp_lport == nport) { 287 INP_RLOCK(inp); 288 sa = inp->inp_socket; 289 SOCKBUF_LOCK(&sa->so_rcv); 290 if (sbappendaddr_locked(&sa->so_rcv, 291 (struct sockaddr *)&divsrc, m, 292 (struct mbuf *)0) == 0) { 293 SOCKBUF_UNLOCK(&sa->so_rcv); 294 sa = NULL; /* force mbuf reclaim below */ 295 } else 296 sorwakeup_locked(sa); 297 INP_RUNLOCK(inp); 298 break; 299 } 300 } 301 INP_INFO_RUNLOCK(&V_divcbinfo); 302 if (sa == NULL) { 303 m_freem(m); 304 KMOD_IPSTAT_INC(ips_noproto); 305 KMOD_IPSTAT_DEC(ips_delivered); 306 } 307} 308 309/* 310 * Deliver packet back into the IP processing machinery. 311 * 312 * If no address specified, or address is 0.0.0.0, send to ip_output(); 313 * otherwise, send to ip_input() and mark as having been received on 314 * the interface with that address. 315 */ 316static int 317div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 318 struct mbuf *control) 319{ 320 struct ip *const ip = mtod(m, struct ip *); 321 struct m_tag *mtag; 322 struct ipfw_rule_ref *dt; 323 int error = 0; 324 325 /* 326 * An mbuf may hasn't come from userland, but we pretend 327 * that it has. 328 */ 329 m->m_pkthdr.rcvif = NULL; 330 m->m_nextpkt = NULL; 331 M_SETFIB(m, so->so_fibnum); 332 333 if (control) 334 m_freem(control); /* XXX */ 335 336 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 337 if (mtag == NULL) { 338 /* this should be normal */ 339 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 340 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 341 if (mtag == NULL) { 342 error = ENOBUFS; 343 goto cantsend; 344 } 345 m_tag_prepend(m, mtag); 346 } 347 dt = (struct ipfw_rule_ref *)(mtag+1); 348 349 /* Loopback avoidance and state recovery */ 350 if (sin) { 351 int i; 352 353 /* set the starting point. We provide a non-zero slot, 354 * but a non_matching chain_id to skip that info and use 355 * the rulenum/rule_id. 356 */ 357 dt->slot = 1; /* dummy, chain_id is invalid */ 358 dt->chain_id = 0; 359 dt->rulenum = sin->sin_port+1; /* host format ? */ 360 dt->rule_id = 0; 361 /* 362 * Find receive interface with the given name, stuffed 363 * (if it exists) in the sin_zero[] field. 364 * The name is user supplied data so don't trust its size 365 * or that it is zero terminated. 366 */ 367 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 368 ; 369 if ( i > 0 && i < sizeof(sin->sin_zero)) 370 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 371 } 372 373 /* Reinject packet into the system as incoming or outgoing */ 374 if (!sin || sin->sin_addr.s_addr == 0) { 375 struct mbuf *options = NULL; 376 struct inpcb *inp; 377 378 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 379 inp = sotoinpcb(so); 380 INP_RLOCK(inp); 381 switch (ip->ip_v) { 382 case IPVERSION: 383 /* 384 * Don't allow both user specified and setsockopt 385 * options, and don't allow packet length sizes that 386 * will crash. 387 */ 388 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 389 inp->inp_options != NULL) || 390 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 391 error = EINVAL; 392 INP_RUNLOCK(inp); 393 goto cantsend; 394 } 395 396 /* Convert fields to host order for ip_output() */ 397 ip->ip_len = ntohs(ip->ip_len); 398 ip->ip_off = ntohs(ip->ip_off); 399 break; 400#ifdef INET6 401 case IPV6_VERSION >> 4: 402 { 403 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 404 405 /* Don't allow packet length sizes that will crash */ 406 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 407 error = EINVAL; 408 INP_RUNLOCK(inp); 409 goto cantsend; 410 } 411 412 ip6->ip6_plen = ntohs(ip6->ip6_plen); 413 } 414#endif 415 default: 416 error = EINVAL; 417 INP_RUNLOCK(inp); 418 goto cantsend; 419 } 420 421 /* Send packet to output processing */ 422 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 423 424#ifdef MAC 425 mac_inpcb_create_mbuf(inp, m); 426#endif 427 /* 428 * Get ready to inject the packet into ip_output(). 429 * Just in case socket options were specified on the 430 * divert socket, we duplicate them. This is done 431 * to avoid having to hold the PCB locks over the call 432 * to ip_output(), as doing this results in a number of 433 * lock ordering complexities. 434 * 435 * Note that we set the multicast options argument for 436 * ip_output() to NULL since it should be invariant that 437 * they are not present. 438 */ 439 KASSERT(inp->inp_moptions == NULL, 440 ("multicast options set on a divert socket")); 441 /* 442 * XXXCSJP: It is unclear to me whether or not it makes 443 * sense for divert sockets to have options. However, 444 * for now we will duplicate them with the INP locks 445 * held so we can use them in ip_output() without 446 * requring a reference to the pcb. 447 */ 448 if (inp->inp_options != NULL) { 449 options = m_dup(inp->inp_options, M_NOWAIT); 450 if (options == NULL) { 451 INP_RUNLOCK(inp); 452 error = ENOBUFS; 453 goto cantsend; 454 } 455 } 456 INP_RUNLOCK(inp); 457 458 switch (ip->ip_v) { 459 case IPVERSION: 460 error = ip_output(m, options, NULL, 461 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 462 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 463 break; 464#ifdef INET6 465 case IPV6_VERSION >> 4: 466 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 467 break; 468#endif 469 } 470 if (options != NULL) 471 m_freem(options); 472 } else { 473 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 474 if (m->m_pkthdr.rcvif == NULL) { 475 /* 476 * No luck with the name, check by IP address. 477 * Clear the port and the ifname to make sure 478 * there are no distractions for ifa_ifwithaddr. 479 */ 480 struct ifaddr *ifa; 481 482 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 483 sin->sin_port = 0; 484 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 485 if (ifa == NULL) { 486 error = EADDRNOTAVAIL; 487 goto cantsend; 488 } 489 m->m_pkthdr.rcvif = ifa->ifa_ifp; 490 ifa_free(ifa); 491 } 492#ifdef MAC 493 mac_socket_create_mbuf(so, m); 494#endif 495 /* Send packet to input processing via netisr */ 496 switch (ip->ip_v) { 497 case IPVERSION: 498 netisr_queue_src(NETISR_IP, (uintptr_t)so, m); 499 break; 500#ifdef INET6 501 case IPV6_VERSION >> 4: 502 netisr_queue_src(NETISR_IPV6, (uintptr_t)so, m); 503 break; 504#endif 505 default: 506 error = EINVAL; 507 goto cantsend; 508 } 509 } 510 511 return (error); 512 513cantsend: 514 m_freem(m); 515 return (error); 516} 517 518static int 519div_attach(struct socket *so, int proto, struct thread *td) 520{ 521 struct inpcb *inp; 522 int error; 523 524 inp = sotoinpcb(so); 525 KASSERT(inp == NULL, ("div_attach: inp != NULL")); 526 if (td != NULL) { 527 error = priv_check(td, PRIV_NETINET_DIVERT); 528 if (error) 529 return (error); 530 } 531 error = soreserve(so, div_sendspace, div_recvspace); 532 if (error) 533 return error; 534 INP_INFO_WLOCK(&V_divcbinfo); 535 error = in_pcballoc(so, &V_divcbinfo); 536 if (error) { 537 INP_INFO_WUNLOCK(&V_divcbinfo); 538 return error; 539 } 540 inp = (struct inpcb *)so->so_pcb; 541 INP_INFO_WUNLOCK(&V_divcbinfo); 542 inp->inp_ip_p = proto; 543 inp->inp_vflag |= INP_IPV4; 544 inp->inp_flags |= INP_HDRINCL; 545 INP_WUNLOCK(inp); 546 return 0; 547} 548 549static void 550div_detach(struct socket *so) 551{ 552 struct inpcb *inp; 553 554 inp = sotoinpcb(so); 555 KASSERT(inp != NULL, ("div_detach: inp == NULL")); 556 INP_INFO_WLOCK(&V_divcbinfo); 557 INP_WLOCK(inp); 558 in_pcbdetach(inp); 559 in_pcbfree(inp); 560 INP_INFO_WUNLOCK(&V_divcbinfo); 561} 562 563static int 564div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 565{ 566 struct inpcb *inp; 567 int error; 568 569 inp = sotoinpcb(so); 570 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 571 /* in_pcbbind assumes that nam is a sockaddr_in 572 * and in_pcbbind requires a valid address. Since divert 573 * sockets don't we need to make sure the address is 574 * filled in properly. 575 * XXX -- divert should not be abusing in_pcbind 576 * and should probably have its own family. 577 */ 578 if (nam->sa_family != AF_INET) 579 return EAFNOSUPPORT; 580 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 581 INP_INFO_WLOCK(&V_divcbinfo); 582 INP_WLOCK(inp); 583 INP_HASH_WLOCK(&V_divcbinfo); 584 error = in_pcbbind(inp, nam, td->td_ucred); 585 INP_HASH_WUNLOCK(&V_divcbinfo); 586 INP_WUNLOCK(inp); 587 INP_INFO_WUNLOCK(&V_divcbinfo); 588 return error; 589} 590 591static int 592div_shutdown(struct socket *so) 593{ 594 struct inpcb *inp; 595 596 inp = sotoinpcb(so); 597 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 598 INP_WLOCK(inp); 599 socantsendmore(so); 600 INP_WUNLOCK(inp); 601 return 0; 602} 603 604static int 605div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 606 struct mbuf *control, struct thread *td) 607{ 608 609 /* Packet must have a header (but that's about it) */ 610 if (m->m_len < sizeof (struct ip) && 611 (m = m_pullup(m, sizeof (struct ip))) == 0) { 612 KMOD_IPSTAT_INC(ips_toosmall); 613 m_freem(m); 614 return EINVAL; 615 } 616 617 /* Send packet */ 618 return div_output(so, m, (struct sockaddr_in *)nam, control); 619} 620 621static void 622div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 623{ 624 struct in_addr faddr; 625 626 faddr = ((struct sockaddr_in *)sa)->sin_addr; 627 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 628 return; 629 if (PRC_IS_REDIRECT(cmd)) 630 return; 631} 632 633static int 634div_pcblist(SYSCTL_HANDLER_ARGS) 635{ 636 int error, i, n; 637 struct inpcb *inp, **inp_list; 638 inp_gen_t gencnt; 639 struct xinpgen xig; 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_RLOCK(&V_divcbinfo); 659 gencnt = V_divcbinfo.ipi_gencnt; 660 n = V_divcbinfo.ipi_count; 661 INP_INFO_RUNLOCK(&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 xig.xig_len = sizeof xig; 669 xig.xig_count = n; 670 xig.xig_gen = gencnt; 671 xig.xig_sogen = so_gencnt; 672 error = SYSCTL_OUT(req, &xig, sizeof xig); 673 if (error) 674 return error; 675 676 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 677 if (inp_list == 0) 678 return ENOMEM; 679 680 INP_INFO_RLOCK(&V_divcbinfo); 681 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 682 inp = LIST_NEXT(inp, inp_list)) { 683 INP_WLOCK(inp); 684 if (inp->inp_gencnt <= gencnt && 685 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 686 in_pcbref(inp); 687 inp_list[i++] = inp; 688 } 689 INP_WUNLOCK(inp); 690 } 691 INP_INFO_RUNLOCK(&V_divcbinfo); 692 n = i; 693 694 error = 0; 695 for (i = 0; i < n; i++) { 696 inp = inp_list[i]; 697 INP_RLOCK(inp); 698 if (inp->inp_gencnt <= gencnt) { 699 struct xinpcb xi; 700 bzero(&xi, sizeof(xi)); 701 xi.xi_len = sizeof xi; 702 /* XXX should avoid extra copy */ 703 bcopy(inp, &xi.xi_inp, sizeof *inp); 704 if (inp->inp_socket) 705 sotoxsocket(inp->inp_socket, &xi.xi_socket); 706 INP_RUNLOCK(inp); 707 error = SYSCTL_OUT(req, &xi, sizeof xi); 708 } else 709 INP_RUNLOCK(inp); 710 } 711 INP_INFO_WLOCK(&V_divcbinfo); 712 for (i = 0; i < n; i++) { 713 inp = inp_list[i]; 714 INP_RLOCK(inp); 715 if (!in_pcbrele_rlocked(inp)) 716 INP_RUNLOCK(inp); 717 } 718 INP_INFO_WUNLOCK(&V_divcbinfo); 719 720 if (!error) { 721 /* 722 * Give the user an updated idea of our state. 723 * If the generation differs from what we told 724 * her before, she knows that something happened 725 * while we were processing this request, and it 726 * might be necessary to retry. 727 */ 728 INP_INFO_RLOCK(&V_divcbinfo); 729 xig.xig_gen = V_divcbinfo.ipi_gencnt; 730 xig.xig_sogen = so_gencnt; 731 xig.xig_count = V_divcbinfo.ipi_count; 732 INP_INFO_RUNLOCK(&V_divcbinfo); 733 error = SYSCTL_OUT(req, &xig, sizeof xig); 734 } 735 free(inp_list, M_TEMP); 736 return error; 737} 738 739#ifdef SYSCTL_NODE 740SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 741SYSCTL_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 745struct 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 757struct 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 771static int 772div_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 838static moduledata_t ipdivertmod = { 839 "ipdivert", 840 div_modevent, 841 0 842}; 843 844DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 845MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2); 846MODULE_VERSION(ipdivert, 1); 847