ip_divert.c revision 122331
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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * $FreeBSD: head/sys/netinet/ip_divert.c 122331 2003-11-08 23:09:42Z sam $ 34 */ 35 36#include "opt_inet.h" 37#include "opt_ipfw.h" 38#include "opt_ipdivert.h" 39#include "opt_ipsec.h" 40#include "opt_mac.h" 41 42#ifndef INET 43#error "IPDIVERT requires INET." 44#endif 45 46#include <sys/param.h> 47#include <sys/kernel.h> 48#include <sys/lock.h> 49#include <sys/malloc.h> 50#include <sys/mac.h> 51#include <sys/mbuf.h> 52#include <sys/proc.h> 53#include <sys/protosw.h> 54#include <sys/signalvar.h> 55#include <sys/socket.h> 56#include <sys/socketvar.h> 57#include <sys/sx.h> 58#include <sys/sysctl.h> 59#include <sys/systm.h> 60 61#include <vm/uma.h> 62 63#include <net/if.h> 64#include <net/route.h> 65 66#include <netinet/in.h> 67#include <netinet/in_pcb.h> 68#include <netinet/in_systm.h> 69#include <netinet/in_var.h> 70#include <netinet/ip.h> 71#include <netinet/ip_var.h> 72 73/* 74 * Divert sockets 75 */ 76 77/* 78 * Allocate enough space to hold a full IP packet 79 */ 80#define DIVSNDQ (65536 + 100) 81#define DIVRCVQ (65536 + 100) 82 83/* 84 * Divert sockets work in conjunction with ipfw, see the divert(4) 85 * manpage for features. 86 * Internally, packets selected by ipfw in ip_input() or ip_output(), 87 * and never diverted before, are passed to the input queue of the 88 * divert socket with a given 'divert_port' number (as specified in 89 * the matching ipfw rule), and they are tagged with a 16 bit cookie 90 * (representing the rule number of the matching ipfw rule), which 91 * is passed to process reading from the socket. 92 * 93 * Packets written to the divert socket are again tagged with a cookie 94 * (usually the same as above) and a destination address. 95 * If the destination address is INADDR_ANY then the packet is 96 * treated as outgoing and sent to ip_output(), otherwise it is 97 * treated as incoming and sent to ip_input(). 98 * In both cases, the packet is tagged with the cookie. 99 * 100 * On reinjection, processing in ip_input() and ip_output() 101 * will be exactly the same as for the original packet, except that 102 * ipfw processing will start at the rule number after the one 103 * written in the cookie (so, tagging a packet with a cookie of 0 104 * will cause it to be effectively considered as a standard packet). 105 */ 106 107/* Internal variables */ 108static struct inpcbhead divcb; 109static struct inpcbinfo divcbinfo; 110 111static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 112static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 113 114/* 115 * Initialize divert connection block queue. 116 */ 117void 118div_init(void) 119{ 120 INP_INFO_LOCK_INIT(&divcbinfo, "div"); 121 LIST_INIT(&divcb); 122 divcbinfo.listhead = &divcb; 123 /* 124 * XXX We don't use the hash list for divert IP, but it's easier 125 * to allocate a one entry hash list than it is to check all 126 * over the place for hashbase == NULL. 127 */ 128 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); 129 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask); 130 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 131 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 132 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 133} 134 135/* 136 * IPPROTO_DIVERT is not in the real IP protocol number space; this 137 * function should never be called. Just in case, drop any packets. 138 */ 139void 140div_input(struct mbuf *m, int off) 141{ 142 ipstat.ips_noproto++; 143 m_freem(m); 144} 145 146/* 147 * Divert a packet by passing it up to the divert socket at port 'port'. 148 * 149 * Setup generic address and protocol structures for div_input routine, 150 * then pass them along with mbuf chain. 151 */ 152void 153divert_packet(struct mbuf *m, int incoming, int port, int rule) 154{ 155 struct ip *ip; 156 struct inpcb *inp; 157 struct socket *sa; 158 u_int16_t nport; 159 struct sockaddr_in divsrc; 160 161 /* Sanity check */ 162 KASSERT(port != 0, ("%s: port=0", __func__)); 163 164 /* Assure header */ 165 if (m->m_len < sizeof(struct ip) && 166 (m = m_pullup(m, sizeof(struct ip))) == 0) 167 return; 168 ip = mtod(m, struct ip *); 169 170 /* 171 * Record receive interface address, if any. 172 * But only for incoming packets. 173 */ 174 bzero(&divsrc, sizeof(divsrc)); 175 divsrc.sin_len = sizeof(divsrc); 176 divsrc.sin_family = AF_INET; 177 divsrc.sin_port = rule; /* record matching rule */ 178 if (incoming) { 179 struct ifaddr *ifa; 180 181 /* Sanity check */ 182 M_ASSERTPKTHDR(m); 183 184 /* Find IP address for receive interface */ 185 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 186 if (ifa->ifa_addr == NULL) 187 continue; 188 if (ifa->ifa_addr->sa_family != AF_INET) 189 continue; 190 divsrc.sin_addr = 191 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 192 break; 193 } 194 } 195 /* 196 * Record the incoming interface name whenever we have one. 197 */ 198 if (m->m_pkthdr.rcvif) { 199 /* 200 * Hide the actual interface name in there in the 201 * sin_zero array. XXX This needs to be moved to a 202 * different sockaddr type for divert, e.g. 203 * sockaddr_div with multiple fields like 204 * sockaddr_dl. Presently we have only 7 bytes 205 * but that will do for now as most interfaces 206 * are 4 or less + 2 or less bytes for unit. 207 * There is probably a faster way of doing this, 208 * possibly taking it from the sockaddr_dl on the iface. 209 * This solves the problem of a P2P link and a LAN interface 210 * having the same address, which can result in the wrong 211 * interface being assigned to the packet when fed back 212 * into the divert socket. Theoretically if the daemon saves 213 * and re-uses the sockaddr_in as suggested in the man pages, 214 * this iface name will come along for the ride. 215 * (see div_output for the other half of this.) 216 */ 217 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 218 sizeof(divsrc.sin_zero)); 219 } 220 221 /* 222 * XXX sbappendaddr must be protected by Giant until 223 * we have locking at the socket layer. When entered 224 * from below we come in w/o Giant and must take it 225 * here. Unfortunately we cannot tell whether we're 226 * entering from above (already holding Giant), 227 * below (potentially without Giant), or otherwise 228 * (e.g. from tcp_syncache through a timeout) so we 229 * have to grab it regardless. This causes a LOR with 230 * the tcp lock, at least, and possibly others. For 231 * the moment we're ignoring this. Once sockets are 232 * locked this cruft can be removed. 233 */ 234 mtx_lock(&Giant); 235 /* Put packet on socket queue, if any */ 236 sa = NULL; 237 nport = htons((u_int16_t)port); 238 INP_INFO_RLOCK(&divcbinfo); 239 LIST_FOREACH(inp, &divcb, inp_list) { 240 INP_LOCK(inp); 241 /* XXX why does only one socket match? */ 242 if (inp->inp_lport == nport) { 243 sa = inp->inp_socket; 244 if (sbappendaddr(&sa->so_rcv, 245 (struct sockaddr *)&divsrc, m, 246 (struct mbuf *)0) == 0) 247 sa = NULL; /* force mbuf reclaim below */ 248 else 249 sorwakeup(sa); 250 INP_UNLOCK(inp); 251 break; 252 } 253 INP_UNLOCK(inp); 254 } 255 INP_INFO_RUNLOCK(&divcbinfo); 256 mtx_unlock(&Giant); 257 if (sa == NULL) { 258 m_freem(m); 259 ipstat.ips_noproto++; 260 ipstat.ips_delivered--; 261 } 262} 263 264/* 265 * Deliver packet back into the IP processing machinery. 266 * 267 * If no address specified, or address is 0.0.0.0, send to ip_output(); 268 * otherwise, send to ip_input() and mark as having been received on 269 * the interface with that address. 270 */ 271static int 272div_output(struct socket *so, struct mbuf *m, 273 struct sockaddr_in *sin, struct mbuf *control) 274{ 275 int error = 0; 276 struct m_hdr divert_tag; 277 278 /* 279 * Prepare the tag for divert info. Note that a packet 280 * with a 0 tag in mh_data is effectively untagged, 281 * so we could optimize that case. 282 */ 283 divert_tag.mh_type = MT_TAG; 284 divert_tag.mh_flags = PACKET_TAG_DIVERT; 285 divert_tag.mh_next = m; 286 divert_tag.mh_data = 0; /* the matching rule # */ 287 m->m_pkthdr.rcvif = NULL; /* XXX is it necessary ? */ 288 289#ifdef MAC 290 mac_create_mbuf_from_socket(so, m); 291#endif 292 293 if (control) 294 m_freem(control); /* XXX */ 295 296 /* Loopback avoidance and state recovery */ 297 if (sin) { 298 int i; 299 300 divert_tag.mh_data = (caddr_t)(uintptr_t)sin->sin_port; 301 /* 302 * Find receive interface with the given name, stuffed 303 * (if it exists) in the sin_zero[] field. 304 * The name is user supplied data so don't trust its size 305 * or that it is zero terminated. 306 */ 307 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 308 ; 309 if ( i > 0 && i < sizeof(sin->sin_zero)) 310 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 311 } 312 313 /* Reinject packet into the system as incoming or outgoing */ 314 if (!sin || sin->sin_addr.s_addr == 0) { 315 struct ip *const ip = mtod(m, struct ip *); 316 struct inpcb *inp; 317 318 INP_INFO_WLOCK(&divcbinfo); 319 inp = sotoinpcb(so); 320 INP_LOCK(inp); 321 /* 322 * Don't allow both user specified and setsockopt options, 323 * and don't allow packet length sizes that will crash 324 */ 325 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 326 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 327 error = EINVAL; 328 m_freem(m); 329 } else { 330 /* Convert fields to host order for ip_output() */ 331 ip->ip_len = ntohs(ip->ip_len); 332 ip->ip_off = ntohs(ip->ip_off); 333 334 /* Send packet to output processing */ 335 ipstat.ips_rawout++; /* XXX */ 336 337 error = ip_output((struct mbuf *)&divert_tag, 338 inp->inp_options, &inp->inp_route, 339 (so->so_options & SO_DONTROUTE) | 340 IP_ALLOWBROADCAST | IP_RAWOUTPUT, 341 inp->inp_moptions, NULL); 342 } 343 INP_UNLOCK(inp); 344 INP_INFO_WUNLOCK(&divcbinfo); 345 } else { 346 if (m->m_pkthdr.rcvif == NULL) { 347 /* 348 * No luck with the name, check by IP address. 349 * Clear the port and the ifname to make sure 350 * there are no distractions for ifa_ifwithaddr. 351 */ 352 struct ifaddr *ifa; 353 354 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 355 sin->sin_port = 0; 356 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 357 if (ifa == NULL) { 358 error = EADDRNOTAVAIL; 359 goto cantsend; 360 } 361 m->m_pkthdr.rcvif = ifa->ifa_ifp; 362 } 363 /* Send packet to input processing */ 364 ip_input((struct mbuf *)&divert_tag); 365 } 366 367 return error; 368 369cantsend: 370 m_freem(m); 371 return error; 372} 373 374static int 375div_attach(struct socket *so, int proto, struct thread *td) 376{ 377 struct inpcb *inp; 378 int error; 379 380 INP_INFO_WLOCK(&divcbinfo); 381 inp = sotoinpcb(so); 382 if (inp != 0) { 383 INP_INFO_WUNLOCK(&divcbinfo); 384 return EINVAL; 385 } 386 if (td && (error = suser(td)) != 0) { 387 INP_INFO_WUNLOCK(&divcbinfo); 388 return error; 389 } 390 error = soreserve(so, div_sendspace, div_recvspace); 391 if (error) { 392 INP_INFO_WUNLOCK(&divcbinfo); 393 return error; 394 } 395 error = in_pcballoc(so, &divcbinfo, td); 396 if (error) { 397 INP_INFO_WUNLOCK(&divcbinfo); 398 return error; 399 } 400 inp = (struct inpcb *)so->so_pcb; 401 INP_LOCK(inp); 402 INP_INFO_WUNLOCK(&divcbinfo); 403 inp->inp_ip_p = proto; 404 inp->inp_vflag |= INP_IPV4; 405 inp->inp_flags |= INP_HDRINCL; 406 /* The socket is always "connected" because 407 we always know "where" to send the packet */ 408 INP_UNLOCK(inp); 409 so->so_state |= SS_ISCONNECTED; 410 return 0; 411} 412 413static int 414div_detach(struct socket *so) 415{ 416 struct inpcb *inp; 417 418 INP_INFO_WLOCK(&divcbinfo); 419 inp = sotoinpcb(so); 420 if (inp == 0) { 421 INP_INFO_WUNLOCK(&divcbinfo); 422 return EINVAL; 423 } 424 INP_LOCK(inp); 425 in_pcbdetach(inp); 426 INP_INFO_WUNLOCK(&divcbinfo); 427 return 0; 428} 429 430static int 431div_abort(struct socket *so) 432{ 433 struct inpcb *inp; 434 435 INP_INFO_WLOCK(&divcbinfo); 436 inp = sotoinpcb(so); 437 if (inp == 0) { 438 INP_INFO_WUNLOCK(&divcbinfo); 439 return EINVAL; /* ??? possible? panic instead? */ 440 } 441 INP_LOCK(inp); 442 soisdisconnected(so); 443 in_pcbdetach(inp); 444 INP_INFO_WUNLOCK(&divcbinfo); 445 return 0; 446} 447 448static int 449div_disconnect(struct socket *so) 450{ 451 if ((so->so_state & SS_ISCONNECTED) == 0) 452 return ENOTCONN; 453 return div_abort(so); 454} 455 456static int 457div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 458{ 459 struct inpcb *inp; 460 int error; 461 462 INP_INFO_WLOCK(&divcbinfo); 463 inp = sotoinpcb(so); 464 if (inp == 0) { 465 INP_INFO_WUNLOCK(&divcbinfo); 466 return EINVAL; 467 } 468 /* in_pcbbind assumes that nam is a sockaddr_in 469 * and in_pcbbind requires a valid address. Since divert 470 * sockets don't we need to make sure the address is 471 * filled in properly. 472 * XXX -- divert should not be abusing in_pcbind 473 * and should probably have its own family. 474 */ 475 if (nam->sa_family != AF_INET) 476 error = EAFNOSUPPORT; 477 else { 478 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 479 INP_LOCK(inp); 480 error = in_pcbbind(inp, nam, td); 481 INP_UNLOCK(inp); 482 } 483 INP_INFO_WUNLOCK(&divcbinfo); 484 return error; 485} 486 487static int 488div_shutdown(struct socket *so) 489{ 490 struct inpcb *inp; 491 492 INP_INFO_RLOCK(&divcbinfo); 493 inp = sotoinpcb(so); 494 if (inp == 0) { 495 INP_INFO_RUNLOCK(&divcbinfo); 496 return EINVAL; 497 } 498 INP_LOCK(inp); 499 INP_INFO_RUNLOCK(&divcbinfo); 500 socantsendmore(so); 501 INP_UNLOCK(inp); 502 return 0; 503} 504 505static int 506div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 507 struct mbuf *control, struct thread *td) 508{ 509 /* Packet must have a header (but that's about it) */ 510 if (m->m_len < sizeof (struct ip) && 511 (m = m_pullup(m, sizeof (struct ip))) == 0) { 512 ipstat.ips_toosmall++; 513 m_freem(m); 514 return EINVAL; 515 } 516 517 /* Send packet */ 518 return div_output(so, m, (struct sockaddr_in *)nam, control); 519} 520 521void 522div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 523{ 524 struct in_addr faddr; 525 526 faddr = ((struct sockaddr_in *)sa)->sin_addr; 527 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 528 return; 529 if (PRC_IS_REDIRECT(cmd)) { 530 /* flush held routes */ 531 in_pcbnotifyall(&divcbinfo, faddr, 532 inetctlerrmap[cmd], in_rtchange); 533 } 534} 535 536static int 537div_pcblist(SYSCTL_HANDLER_ARGS) 538{ 539 int error, i, n; 540 struct inpcb *inp, **inp_list; 541 inp_gen_t gencnt; 542 struct xinpgen xig; 543 544 /* 545 * The process of preparing the TCB list is too time-consuming and 546 * resource-intensive to repeat twice on every request. 547 */ 548 if (req->oldptr == 0) { 549 n = divcbinfo.ipi_count; 550 req->oldidx = 2 * (sizeof xig) 551 + (n + n/8) * sizeof(struct xinpcb); 552 return 0; 553 } 554 555 if (req->newptr != 0) 556 return EPERM; 557 558 /* 559 * OK, now we're committed to doing something. 560 */ 561 INP_INFO_RLOCK(&divcbinfo); 562 gencnt = divcbinfo.ipi_gencnt; 563 n = divcbinfo.ipi_count; 564 INP_INFO_RUNLOCK(&divcbinfo); 565 566 sysctl_wire_old_buffer(req, 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 567 568 xig.xig_len = sizeof xig; 569 xig.xig_count = n; 570 xig.xig_gen = gencnt; 571 xig.xig_sogen = so_gencnt; 572 error = SYSCTL_OUT(req, &xig, sizeof xig); 573 if (error) 574 return error; 575 576 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 577 if (inp_list == 0) 578 return ENOMEM; 579 580 INP_INFO_RLOCK(&divcbinfo); 581 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n; 582 inp = LIST_NEXT(inp, inp_list)) { 583 INP_LOCK(inp); 584 if (inp->inp_gencnt <= gencnt && 585 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 586 inp_list[i++] = inp; 587 INP_UNLOCK(inp); 588 } 589 INP_INFO_RUNLOCK(&divcbinfo); 590 n = i; 591 592 error = 0; 593 for (i = 0; i < n; i++) { 594 inp = inp_list[i]; 595 if (inp->inp_gencnt <= gencnt) { 596 struct xinpcb xi; 597 xi.xi_len = sizeof xi; 598 /* XXX should avoid extra copy */ 599 bcopy(inp, &xi.xi_inp, sizeof *inp); 600 if (inp->inp_socket) 601 sotoxsocket(inp->inp_socket, &xi.xi_socket); 602 error = SYSCTL_OUT(req, &xi, sizeof xi); 603 } 604 } 605 if (!error) { 606 /* 607 * Give the user an updated idea of our state. 608 * If the generation differs from what we told 609 * her before, she knows that something happened 610 * while we were processing this request, and it 611 * might be necessary to retry. 612 */ 613 INP_INFO_RLOCK(&divcbinfo); 614 xig.xig_gen = divcbinfo.ipi_gencnt; 615 xig.xig_sogen = so_gencnt; 616 xig.xig_count = divcbinfo.ipi_count; 617 INP_INFO_RUNLOCK(&divcbinfo); 618 error = SYSCTL_OUT(req, &xig, sizeof xig); 619 } 620 free(inp_list, M_TEMP); 621 return error; 622} 623 624/* 625 * This is the wrapper function for in_setsockaddr. We just pass down 626 * the pcbinfo for in_setpeeraddr to lock. 627 */ 628static int 629div_sockaddr(struct socket *so, struct sockaddr **nam) 630{ 631 return (in_setsockaddr(so, nam, &divcbinfo)); 632} 633 634/* 635 * This is the wrapper function for in_setpeeraddr. We just pass down 636 * the pcbinfo for in_setpeeraddr to lock. 637 */ 638static int 639div_peeraddr(struct socket *so, struct sockaddr **nam) 640{ 641 return (in_setpeeraddr(so, nam, &divcbinfo)); 642} 643 644 645SYSCTL_DECL(_net_inet_divert); 646SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 647 div_pcblist, "S,xinpcb", "List of active divert sockets"); 648 649struct pr_usrreqs div_usrreqs = { 650 div_abort, pru_accept_notsupp, div_attach, div_bind, 651 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach, 652 div_disconnect, pru_listen_notsupp, div_peeraddr, pru_rcvd_notsupp, 653 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown, 654 div_sockaddr, sosend, soreceive, sopoll 655}; 656