ip_divert.c revision 241342
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: head/sys/netinet/ip_divert.c 241342 2012-10-08 07:21:32Z glebius $"); 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. */ 111static VNET_DEFINE(struct inpcbhead, divcb); 112static VNET_DEFINE(struct inpcbinfo, divcbinfo); 113 114#define V_divcb VNET(divcb) 115#define V_divcbinfo VNET(divcbinfo) 116 117static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 118static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 119 120static eventhandler_tag ip_divert_event_tag; 121 122/* 123 * Initialize divert connection block queue. 124 */ 125static void 126div_zone_change(void *tag) 127{ 128 129 uma_zone_set_max(V_divcbinfo.ipi_zone, maxsockets); 130} 131 132static int 133div_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 141static void 142div_inpcb_fini(void *mem, int size) 143{ 144 struct inpcb *inp = mem; 145 146 INP_LOCK_DESTROY(inp); 147} 148 149static void 150div_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 163static void 164div_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 */ 174static void 175div_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 */ 188static void 189divert_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 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2)); 219 m->m_pkthdr.csum_flags &= ~CSUM_SCTP; 220 } 221#endif 222 bzero(&divsrc, sizeof(divsrc)); 223 divsrc.sin_len = sizeof(divsrc); 224 divsrc.sin_family = AF_INET; 225 /* record matching rule, in host format */ 226 divsrc.sin_port = ((struct ipfw_rule_ref *)(mtag+1))->rulenum; 227 /* 228 * Record receive interface address, if any. 229 * But only for incoming packets. 230 */ 231 if (incoming) { 232 struct ifaddr *ifa; 233 struct ifnet *ifp; 234 235 /* Sanity check */ 236 M_ASSERTPKTHDR(m); 237 238 /* Find IP address for receive interface */ 239 ifp = m->m_pkthdr.rcvif; 240 if_addr_rlock(ifp); 241 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 242 if (ifa->ifa_addr->sa_family != AF_INET) 243 continue; 244 divsrc.sin_addr = 245 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 246 break; 247 } 248 if_addr_runlock(ifp); 249 } 250 /* 251 * Record the incoming interface name whenever we have one. 252 */ 253 if (m->m_pkthdr.rcvif) { 254 /* 255 * Hide the actual interface name in there in the 256 * sin_zero array. XXX This needs to be moved to a 257 * different sockaddr type for divert, e.g. 258 * sockaddr_div with multiple fields like 259 * sockaddr_dl. Presently we have only 7 bytes 260 * but that will do for now as most interfaces 261 * are 4 or less + 2 or less bytes for unit. 262 * There is probably a faster way of doing this, 263 * possibly taking it from the sockaddr_dl on the iface. 264 * This solves the problem of a P2P link and a LAN interface 265 * having the same address, which can result in the wrong 266 * interface being assigned to the packet when fed back 267 * into the divert socket. Theoretically if the daemon saves 268 * and re-uses the sockaddr_in as suggested in the man pages, 269 * this iface name will come along for the ride. 270 * (see div_output for the other half of this.) 271 */ 272 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 273 sizeof(divsrc.sin_zero)); 274 } 275 276 /* Put packet on socket queue, if any */ 277 sa = NULL; 278 nport = htons((u_int16_t)(((struct ipfw_rule_ref *)(mtag+1))->info)); 279 INP_INFO_RLOCK(&V_divcbinfo); 280 LIST_FOREACH(inp, &V_divcb, inp_list) { 281 /* XXX why does only one socket match? */ 282 if (inp->inp_lport == nport) { 283 INP_RLOCK(inp); 284 sa = inp->inp_socket; 285 SOCKBUF_LOCK(&sa->so_rcv); 286 if (sbappendaddr_locked(&sa->so_rcv, 287 (struct sockaddr *)&divsrc, m, 288 (struct mbuf *)0) == 0) { 289 SOCKBUF_UNLOCK(&sa->so_rcv); 290 sa = NULL; /* force mbuf reclaim below */ 291 } else 292 sorwakeup_locked(sa); 293 INP_RUNLOCK(inp); 294 break; 295 } 296 } 297 INP_INFO_RUNLOCK(&V_divcbinfo); 298 if (sa == NULL) { 299 m_freem(m); 300 KMOD_IPSTAT_INC(ips_noproto); 301 KMOD_IPSTAT_DEC(ips_delivered); 302 } 303} 304 305/* 306 * Deliver packet back into the IP processing machinery. 307 * 308 * If no address specified, or address is 0.0.0.0, send to ip_output(); 309 * otherwise, send to ip_input() and mark as having been received on 310 * the interface with that address. 311 */ 312static int 313div_output(struct socket *so, struct mbuf *m, struct sockaddr_in *sin, 314 struct mbuf *control) 315{ 316 struct ip *const ip = mtod(m, struct ip *); 317 struct m_tag *mtag; 318 struct ipfw_rule_ref *dt; 319 int error = 0; 320 321 /* 322 * An mbuf may hasn't come from userland, but we pretend 323 * that it has. 324 */ 325 m->m_pkthdr.rcvif = NULL; 326 m->m_nextpkt = NULL; 327 M_SETFIB(m, so->so_fibnum); 328 329 if (control) 330 m_freem(control); /* XXX */ 331 332 mtag = m_tag_locate(m, MTAG_IPFW_RULE, 0, NULL); 333 if (mtag == NULL) { 334 /* this should be normal */ 335 mtag = m_tag_alloc(MTAG_IPFW_RULE, 0, 336 sizeof(struct ipfw_rule_ref), M_NOWAIT | M_ZERO); 337 if (mtag == NULL) { 338 error = ENOBUFS; 339 goto cantsend; 340 } 341 m_tag_prepend(m, mtag); 342 } 343 dt = (struct ipfw_rule_ref *)(mtag+1); 344 345 /* Loopback avoidance and state recovery */ 346 if (sin) { 347 int i; 348 349 /* set the starting point. We provide a non-zero slot, 350 * but a non_matching chain_id to skip that info and use 351 * the rulenum/rule_id. 352 */ 353 dt->slot = 1; /* dummy, chain_id is invalid */ 354 dt->chain_id = 0; 355 dt->rulenum = sin->sin_port+1; /* host format ? */ 356 dt->rule_id = 0; 357 /* 358 * Find receive interface with the given name, stuffed 359 * (if it exists) in the sin_zero[] field. 360 * The name is user supplied data so don't trust its size 361 * or that it is zero terminated. 362 */ 363 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 364 ; 365 if ( i > 0 && i < sizeof(sin->sin_zero)) 366 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 367 } 368 369 /* Reinject packet into the system as incoming or outgoing */ 370 if (!sin || sin->sin_addr.s_addr == 0) { 371 struct mbuf *options = NULL; 372 struct inpcb *inp; 373 374 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_OUT; 375 inp = sotoinpcb(so); 376 INP_RLOCK(inp); 377 switch (ip->ip_v) { 378 case IPVERSION: 379 /* 380 * Don't allow both user specified and setsockopt 381 * options, and don't allow packet length sizes that 382 * will crash. 383 */ 384 if ((((ip->ip_hl << 2) != sizeof(struct ip)) && 385 inp->inp_options != NULL) || 386 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 387 error = EINVAL; 388 INP_RUNLOCK(inp); 389 goto cantsend; 390 } 391 392 /* Convert fields to host order for ip_output() */ 393 ip->ip_len = ntohs(ip->ip_len); 394 ip->ip_off = ntohs(ip->ip_off); 395 break; 396#ifdef INET6 397 case IPV6_VERSION >> 4: 398 { 399 struct ip6_hdr *const ip6 = mtod(m, struct ip6_hdr *); 400 401 /* Don't allow packet length sizes that will crash */ 402 if (((u_short)ntohs(ip6->ip6_plen) > m->m_pkthdr.len)) { 403 error = EINVAL; 404 INP_RUNLOCK(inp); 405 goto cantsend; 406 } 407 408 ip6->ip6_plen = ntohs(ip6->ip6_plen); 409 break; 410 } 411#endif 412 default: 413 error = EINVAL; 414 INP_RUNLOCK(inp); 415 goto cantsend; 416 } 417 418 /* Send packet to output processing */ 419 KMOD_IPSTAT_INC(ips_rawout); /* XXX */ 420 421#ifdef MAC 422 mac_inpcb_create_mbuf(inp, m); 423#endif 424 /* 425 * Get ready to inject the packet into ip_output(). 426 * Just in case socket options were specified on the 427 * divert socket, we duplicate them. This is done 428 * to avoid having to hold the PCB locks over the call 429 * to ip_output(), as doing this results in a number of 430 * lock ordering complexities. 431 * 432 * Note that we set the multicast options argument for 433 * ip_output() to NULL since it should be invariant that 434 * they are not present. 435 */ 436 KASSERT(inp->inp_moptions == NULL, 437 ("multicast options set on a divert socket")); 438 /* 439 * XXXCSJP: It is unclear to me whether or not it makes 440 * sense for divert sockets to have options. However, 441 * for now we will duplicate them with the INP locks 442 * held so we can use them in ip_output() without 443 * requring a reference to the pcb. 444 */ 445 if (inp->inp_options != NULL) { 446 options = m_dup(inp->inp_options, M_NOWAIT); 447 if (options == NULL) { 448 INP_RUNLOCK(inp); 449 error = ENOBUFS; 450 goto cantsend; 451 } 452 } 453 INP_RUNLOCK(inp); 454 455 switch (ip->ip_v) { 456 case IPVERSION: 457 error = ip_output(m, options, NULL, 458 ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0) 459 | IP_ALLOWBROADCAST | IP_RAWOUTPUT, NULL, NULL); 460 break; 461#ifdef INET6 462 case IPV6_VERSION >> 4: 463 error = ip6_output(m, NULL, NULL, 0, NULL, NULL, NULL); 464 break; 465#endif 466 } 467 if (options != NULL) 468 m_freem(options); 469 } else { 470 dt->info |= IPFW_IS_DIVERT | IPFW_INFO_IN; 471 if (m->m_pkthdr.rcvif == NULL) { 472 /* 473 * No luck with the name, check by IP address. 474 * Clear the port and the ifname to make sure 475 * there are no distractions for ifa_ifwithaddr. 476 */ 477 struct ifaddr *ifa; 478 479 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 480 sin->sin_port = 0; 481 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 482 if (ifa == NULL) { 483 error = EADDRNOTAVAIL; 484 goto cantsend; 485 } 486 m->m_pkthdr.rcvif = ifa->ifa_ifp; 487 ifa_free(ifa); 488 } 489#ifdef MAC 490 mac_socket_create_mbuf(so, m); 491#endif 492 /* Send packet to input processing via netisr */ 493 switch (ip->ip_v) { 494 case IPVERSION: 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 510cantsend: 511 m_freem(m); 512 return (error); 513} 514 515static int 516div_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 546static void 547div_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 in_pcbdetach(inp); 556 in_pcbfree(inp); 557 INP_INFO_WUNLOCK(&V_divcbinfo); 558} 559 560static int 561div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 562{ 563 struct inpcb *inp; 564 int error; 565 566 inp = sotoinpcb(so); 567 KASSERT(inp != NULL, ("div_bind: inp == NULL")); 568 /* in_pcbbind assumes that nam is a sockaddr_in 569 * and in_pcbbind requires a valid address. Since divert 570 * sockets don't we need to make sure the address is 571 * filled in properly. 572 * XXX -- divert should not be abusing in_pcbind 573 * and should probably have its own family. 574 */ 575 if (nam->sa_family != AF_INET) 576 return EAFNOSUPPORT; 577 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 578 INP_INFO_WLOCK(&V_divcbinfo); 579 INP_WLOCK(inp); 580 INP_HASH_WLOCK(&V_divcbinfo); 581 error = in_pcbbind(inp, nam, td->td_ucred); 582 INP_HASH_WUNLOCK(&V_divcbinfo); 583 INP_WUNLOCK(inp); 584 INP_INFO_WUNLOCK(&V_divcbinfo); 585 return error; 586} 587 588static int 589div_shutdown(struct socket *so) 590{ 591 struct inpcb *inp; 592 593 inp = sotoinpcb(so); 594 KASSERT(inp != NULL, ("div_shutdown: inp == NULL")); 595 INP_WLOCK(inp); 596 socantsendmore(so); 597 INP_WUNLOCK(inp); 598 return 0; 599} 600 601static int 602div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 603 struct mbuf *control, struct thread *td) 604{ 605 606 /* Packet must have a header (but that's about it) */ 607 if (m->m_len < sizeof (struct ip) && 608 (m = m_pullup(m, sizeof (struct ip))) == 0) { 609 KMOD_IPSTAT_INC(ips_toosmall); 610 m_freem(m); 611 return EINVAL; 612 } 613 614 /* Send packet */ 615 return div_output(so, m, (struct sockaddr_in *)nam, control); 616} 617 618static void 619div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 620{ 621 struct in_addr faddr; 622 623 faddr = ((struct sockaddr_in *)sa)->sin_addr; 624 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 625 return; 626 if (PRC_IS_REDIRECT(cmd)) 627 return; 628} 629 630static int 631div_pcblist(SYSCTL_HANDLER_ARGS) 632{ 633 int error, i, n; 634 struct inpcb *inp, **inp_list; 635 inp_gen_t gencnt; 636 struct xinpgen xig; 637 638 /* 639 * The process of preparing the TCB list is too time-consuming and 640 * resource-intensive to repeat twice on every request. 641 */ 642 if (req->oldptr == 0) { 643 n = V_divcbinfo.ipi_count; 644 n += imax(n / 8, 10); 645 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xinpcb); 646 return 0; 647 } 648 649 if (req->newptr != 0) 650 return EPERM; 651 652 /* 653 * OK, now we're committed to doing something. 654 */ 655 INP_INFO_RLOCK(&V_divcbinfo); 656 gencnt = V_divcbinfo.ipi_gencnt; 657 n = V_divcbinfo.ipi_count; 658 INP_INFO_RUNLOCK(&V_divcbinfo); 659 660 error = sysctl_wire_old_buffer(req, 661 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 662 if (error != 0) 663 return (error); 664 665 xig.xig_len = sizeof xig; 666 xig.xig_count = n; 667 xig.xig_gen = gencnt; 668 xig.xig_sogen = so_gencnt; 669 error = SYSCTL_OUT(req, &xig, sizeof xig); 670 if (error) 671 return error; 672 673 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 674 if (inp_list == 0) 675 return ENOMEM; 676 677 INP_INFO_RLOCK(&V_divcbinfo); 678 for (inp = LIST_FIRST(V_divcbinfo.ipi_listhead), i = 0; inp && i < n; 679 inp = LIST_NEXT(inp, inp_list)) { 680 INP_WLOCK(inp); 681 if (inp->inp_gencnt <= gencnt && 682 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 683 in_pcbref(inp); 684 inp_list[i++] = inp; 685 } 686 INP_WUNLOCK(inp); 687 } 688 INP_INFO_RUNLOCK(&V_divcbinfo); 689 n = i; 690 691 error = 0; 692 for (i = 0; i < n; i++) { 693 inp = inp_list[i]; 694 INP_RLOCK(inp); 695 if (inp->inp_gencnt <= gencnt) { 696 struct xinpcb xi; 697 bzero(&xi, sizeof(xi)); 698 xi.xi_len = sizeof xi; 699 /* XXX should avoid extra copy */ 700 bcopy(inp, &xi.xi_inp, sizeof *inp); 701 if (inp->inp_socket) 702 sotoxsocket(inp->inp_socket, &xi.xi_socket); 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 /* 719 * Give the user an updated idea of our state. 720 * If the generation differs from what we told 721 * her before, she knows that something happened 722 * while we were processing this request, and it 723 * might be necessary to retry. 724 */ 725 INP_INFO_RLOCK(&V_divcbinfo); 726 xig.xig_gen = V_divcbinfo.ipi_gencnt; 727 xig.xig_sogen = so_gencnt; 728 xig.xig_count = V_divcbinfo.ipi_count; 729 INP_INFO_RUNLOCK(&V_divcbinfo); 730 error = SYSCTL_OUT(req, &xig, sizeof xig); 731 } 732 free(inp_list, M_TEMP); 733 return error; 734} 735 736#ifdef SYSCTL_NODE 737static SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, 738 "IPDIVERT"); 739SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD, 740 NULL, 0, div_pcblist, "S,xinpcb", "List of active divert sockets"); 741#endif 742 743struct pr_usrreqs div_usrreqs = { 744 .pru_attach = div_attach, 745 .pru_bind = div_bind, 746 .pru_control = in_control, 747 .pru_detach = div_detach, 748 .pru_peeraddr = in_getpeeraddr, 749 .pru_send = div_send, 750 .pru_shutdown = div_shutdown, 751 .pru_sockaddr = in_getsockaddr, 752 .pru_sosetlabel = in_pcbsosetlabel 753}; 754 755struct protosw div_protosw = { 756 .pr_type = SOCK_RAW, 757 .pr_protocol = IPPROTO_DIVERT, 758 .pr_flags = PR_ATOMIC|PR_ADDR, 759 .pr_input = div_input, 760 .pr_ctlinput = div_ctlinput, 761 .pr_ctloutput = ip_ctloutput, 762 .pr_init = div_init, 763#ifdef VIMAGE 764 .pr_destroy = div_destroy, 765#endif 766 .pr_usrreqs = &div_usrreqs 767}; 768 769static int 770div_modevent(module_t mod, int type, void *unused) 771{ 772 int err = 0; 773#ifndef VIMAGE 774 int n; 775#endif 776 777 switch (type) { 778 case MOD_LOAD: 779 /* 780 * Protocol will be initialized by pf_proto_register(). 781 * We don't have to register ip_protox because we are not 782 * a true IP protocol that goes over the wire. 783 */ 784 err = pf_proto_register(PF_INET, &div_protosw); 785 if (err != 0) 786 return (err); 787 ip_divert_ptr = divert_packet; 788 ip_divert_event_tag = EVENTHANDLER_REGISTER(maxsockets_change, 789 div_zone_change, NULL, EVENTHANDLER_PRI_ANY); 790 break; 791 case MOD_QUIESCE: 792 /* 793 * IPDIVERT may normally not be unloaded because of the 794 * potential race conditions. Tell kldunload we can't be 795 * unloaded unless the unload is forced. 796 */ 797 err = EPERM; 798 break; 799 case MOD_UNLOAD: 800#ifdef VIMAGE 801 err = EPERM; 802 break; 803#else 804 /* 805 * Forced unload. 806 * 807 * Module ipdivert can only be unloaded if no sockets are 808 * connected. Maybe this can be changed later to forcefully 809 * disconnect any open sockets. 810 * 811 * XXXRW: Note that there is a slight race here, as a new 812 * socket open request could be spinning on the lock and then 813 * we destroy the lock. 814 */ 815 INP_INFO_WLOCK(&V_divcbinfo); 816 n = V_divcbinfo.ipi_count; 817 if (n != 0) { 818 err = EBUSY; 819 INP_INFO_WUNLOCK(&V_divcbinfo); 820 break; 821 } 822 ip_divert_ptr = NULL; 823 err = pf_proto_unregister(PF_INET, IPPROTO_DIVERT, SOCK_RAW); 824 INP_INFO_WUNLOCK(&V_divcbinfo); 825 div_destroy(); 826 EVENTHANDLER_DEREGISTER(maxsockets_change, ip_divert_event_tag); 827 break; 828#endif /* !VIMAGE */ 829 default: 830 err = EOPNOTSUPP; 831 break; 832 } 833 return err; 834} 835 836static moduledata_t ipdivertmod = { 837 "ipdivert", 838 div_modevent, 839 0 840}; 841 842DECLARE_MODULE(ipdivert, ipdivertmod, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY); 843MODULE_DEPEND(ipdivert, ipfw, 2, 2, 2); 844MODULE_VERSION(ipdivert, 1); 845