ip_divert.c revision 82884
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 82884 2001-09-03 20:03:55Z julian $ 34 */ 35 36#include "opt_inet.h" 37#include "opt_ipfw.h" 38#include "opt_ipdivert.h" 39#include "opt_ipsec.h" 40 41#ifndef INET 42#error "IPDIVERT requires INET." 43#endif 44 45#include <sys/param.h> 46#include <sys/kernel.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/socket.h> 50#include <sys/protosw.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/systm.h> 54 55#include <vm/vm_zone.h> 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_var.h> 65#include <netinet/ip_var.h> 66 67/* 68 * Divert sockets 69 */ 70 71/* 72 * Allocate enough space to hold a full IP packet 73 */ 74#define DIVSNDQ (65536 + 100) 75#define DIVRCVQ (65536 + 100) 76 77/* 78 * A 16 bit cookie is passed to and from the user process. 79 * The user process can send it back to help the caller know 80 * something about where the packet originally came from. 81 * 82 * In the case of ipfw, then the cookie is the rule that sent 83 * us here. On reinjection is is the rule after which processing 84 * should continue. Leaving it the same will make processing start 85 * at the rule number after that which sent it here. Setting it to 86 * 0 will restart processing at the beginning. 87 * 88 * For divert_packet(), ip_divert_cookie is an input value only. 89 * For div_output(), ip_divert_cookie is an output value only. 90 */ 91u_int16_t ip_divert_cookie; 92 93/* Internal variables */ 94static struct inpcbhead divcb; 95static struct inpcbinfo divcbinfo; 96 97static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 98static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 99 100/* Optimization: have this preinitialized */ 101static struct sockaddr_in divsrc = { sizeof(divsrc), AF_INET }; 102 103/* Internal functions */ 104static int div_output(struct socket *so, 105 struct mbuf *m, struct sockaddr *addr, struct mbuf *control); 106 107/* 108 * Initialize divert connection block queue. 109 */ 110void 111div_init(void) 112{ 113 LIST_INIT(&divcb); 114 divcbinfo.listhead = &divcb; 115 /* 116 * XXX We don't use the hash list for divert IP, but it's easier 117 * to allocate a one entry hash list than it is to check all 118 * over the place for hashbase == NULL. 119 */ 120 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); 121 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask); 122 divcbinfo.ipi_zone = zinit("divcb", sizeof(struct inpcb), 123 maxsockets, ZONE_INTERRUPT, 0); 124} 125 126/* 127 * IPPROTO_DIVERT is not a real IP protocol; don't allow any packets 128 * with that protocol number to enter the system from the outside. 129 */ 130void 131div_input(struct mbuf *m, int off) 132{ 133 ipstat.ips_noproto++; 134 m_freem(m); 135} 136 137/* 138 * Divert a packet by passing it up to the divert socket at port 'port'. 139 * 140 * Setup generic address and protocol structures for div_input routine, 141 * then pass them along with mbuf chain. 142 */ 143void 144divert_packet(struct mbuf *m, int incoming, int port) 145{ 146 struct ip *ip; 147 struct inpcb *inp; 148 struct socket *sa; 149 u_int16_t nport; 150 151 /* Sanity check */ 152 KASSERT(port != 0, ("%s: port=0", __FUNCTION__)); 153 154 /* Record and reset divert cookie */ 155 divsrc.sin_port = ip_divert_cookie; 156 ip_divert_cookie = 0; 157 158 /* Assure header */ 159 if (m->m_len < sizeof(struct ip) && 160 (m = m_pullup(m, sizeof(struct ip))) == 0) { 161 return; 162 } 163 ip = mtod(m, struct ip *); 164 165 /* 166 * Record receive interface address, if any. 167 * But only for incoming packets. 168 */ 169 divsrc.sin_addr.s_addr = 0; 170 if (incoming) { 171 struct ifaddr *ifa; 172 173 /* Sanity check */ 174 KASSERT((m->m_flags & M_PKTHDR), ("%s: !PKTHDR", __FUNCTION__)); 175 176 /* Find IP address for receive interface */ 177 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 178 if (ifa->ifa_addr == NULL) 179 continue; 180 if (ifa->ifa_addr->sa_family != AF_INET) 181 continue; 182 divsrc.sin_addr = 183 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 184 break; 185 } 186 } 187 /* 188 * Record the incoming interface name whenever we have one. 189 */ 190 bzero(&divsrc.sin_zero, sizeof(divsrc.sin_zero)); 191 if (m->m_pkthdr.rcvif) { 192 /* 193 * Hide the actual interface name in there in the 194 * sin_zero array. XXX This needs to be moved to a 195 * different sockaddr type for divert, e.g. 196 * sockaddr_div with multiple fields like 197 * sockaddr_dl. Presently we have only 7 bytes 198 * but that will do for now as most interfaces 199 * are 4 or less + 2 or less bytes for unit. 200 * There is probably a faster way of doing this, 201 * possibly taking it from the sockaddr_dl on the iface. 202 * This solves the problem of a P2P link and a LAN interface 203 * having the same address, which can result in the wrong 204 * interface being assigned to the packet when fed back 205 * into the divert socket. Theoretically if the daemon saves 206 * and re-uses the sockaddr_in as suggested in the man pages, 207 * this iface name will come along for the ride. 208 * (see div_output for the other half of this.) 209 */ 210 snprintf(divsrc.sin_zero, sizeof(divsrc.sin_zero), 211 "%s%d", m->m_pkthdr.rcvif->if_name, 212 m->m_pkthdr.rcvif->if_unit); 213 } 214 215 /* Put packet on socket queue, if any */ 216 sa = NULL; 217 nport = htons((u_int16_t)port); 218 LIST_FOREACH(inp, &divcb, inp_list) { 219 if (inp->inp_lport == nport) 220 sa = inp->inp_socket; 221 } 222 if (sa) { 223 if (sbappendaddr(&sa->so_rcv, (struct sockaddr *)&divsrc, 224 m, (struct mbuf *)0) == 0) 225 m_freem(m); 226 else 227 sorwakeup(sa); 228 } else { 229 m_freem(m); 230 ipstat.ips_noproto++; 231 ipstat.ips_delivered--; 232 } 233} 234 235/* 236 * Deliver packet back into the IP processing machinery. 237 * 238 * If no address specified, or address is 0.0.0.0, send to ip_output(); 239 * otherwise, send to ip_input() and mark as having been received on 240 * the interface with that address. 241 */ 242static int 243div_output(so, m, addr, control) 244 struct socket *so; 245 register struct mbuf *m; 246 struct sockaddr *addr; 247 struct mbuf *control; 248{ 249 register struct inpcb *const inp = sotoinpcb(so); 250 register struct ip *const ip = mtod(m, struct ip *); 251 struct sockaddr_in *sin = (struct sockaddr_in *)addr; 252 int error = 0; 253 254 if (control) 255 m_freem(control); /* XXX */ 256 257 /* Loopback avoidance and state recovery */ 258 if (sin) { 259 int len = 0; 260 char *c = sin->sin_zero; 261 262 ip_divert_cookie = sin->sin_port; 263 264 /* 265 * Find receive interface with the given name or IP address. 266 * The name is user supplied data so don't trust it's size or 267 * that it is zero terminated. The name has priority. 268 * We are presently assuming that the sockaddr_in 269 * has not been replaced by a sockaddr_div, so we limit it 270 * to 16 bytes in total. the name is stuffed (if it exists) 271 * in the sin_zero[] field. 272 */ 273 while (*c++ && (len++ < sizeof(sin->sin_zero))); 274 if ((len > 0) && (len < sizeof(sin->sin_zero))) 275 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 276 } else { 277 ip_divert_cookie = 0; 278 } 279 280 /* Reinject packet into the system as incoming or outgoing */ 281 if (!sin || sin->sin_addr.s_addr == 0) { 282 /* 283 * Don't allow both user specified and setsockopt options, 284 * and don't allow packet length sizes that will crash 285 */ 286 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 287 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 288 error = EINVAL; 289 goto cantsend; 290 } 291 292 /* Convert fields to host order for ip_output() */ 293 NTOHS(ip->ip_len); 294 NTOHS(ip->ip_off); 295 296 /* Send packet to output processing */ 297 ipstat.ips_rawout++; /* XXX */ 298 error = ip_output(m, inp->inp_options, &inp->inp_route, 299 (so->so_options & SO_DONTROUTE) | 300 IP_ALLOWBROADCAST | IP_RAWOUTPUT, 301 inp->inp_moptions); 302 } else { 303 struct ifaddr *ifa; 304 305 /* If no luck with the name above. check by IP address. */ 306 if (m->m_pkthdr.rcvif == NULL) { 307 /* 308 * Make sure there are no distractions 309 * for ifa_ifwithaddr. Clear the port and the ifname. 310 * Maybe zap all 8 bytes at once using a 64bit write? 311 */ 312 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 313 /* *((u_int64_t *)sin->sin_zero) = 0; */ /* XXX ?? */ 314 sin->sin_port = 0; 315 if (!(ifa = ifa_ifwithaddr((struct sockaddr *) sin))) { 316 error = EADDRNOTAVAIL; 317 goto cantsend; 318 } 319 m->m_pkthdr.rcvif = ifa->ifa_ifp; 320 } 321 322 /* Send packet to input processing */ 323 ip_input(m); 324 } 325 326 /* paranoid: Reset for next time (and other packets) */ 327 /* almost definitly already done in the ipfw filter but.. */ 328 ip_divert_cookie = 0; 329 return error; 330 331cantsend: 332 m_freem(m); 333 ip_divert_cookie = 0; 334 return error; 335} 336 337static int 338div_attach(struct socket *so, int proto, struct proc *p) 339{ 340 struct inpcb *inp; 341 int error, s; 342 343 inp = sotoinpcb(so); 344 if (inp) 345 panic("div_attach"); 346 if (p && (error = suser(p)) != 0) 347 return error; 348 349 error = soreserve(so, div_sendspace, div_recvspace); 350 if (error) 351 return error; 352 s = splnet(); 353 error = in_pcballoc(so, &divcbinfo, p); 354 splx(s); 355 if (error) 356 return error; 357 inp = (struct inpcb *)so->so_pcb; 358 inp->inp_ip_p = proto; 359 inp->inp_vflag |= INP_IPV4; 360 inp->inp_flags |= INP_HDRINCL; 361 /* The socket is always "connected" because 362 we always know "where" to send the packet */ 363 so->so_state |= SS_ISCONNECTED; 364 return 0; 365} 366 367static int 368div_detach(struct socket *so) 369{ 370 struct inpcb *inp; 371 372 inp = sotoinpcb(so); 373 if (inp == 0) 374 panic("div_detach"); 375 in_pcbdetach(inp); 376 return 0; 377} 378 379static int 380div_abort(struct socket *so) 381{ 382 soisdisconnected(so); 383 return div_detach(so); 384} 385 386static int 387div_disconnect(struct socket *so) 388{ 389 if ((so->so_state & SS_ISCONNECTED) == 0) 390 return ENOTCONN; 391 return div_abort(so); 392} 393 394static int 395div_bind(struct socket *so, struct sockaddr *nam, struct proc *p) 396{ 397 struct inpcb *inp; 398 int s; 399 int error; 400 401 s = splnet(); 402 inp = sotoinpcb(so); 403 /* in_pcbbind assumes that the socket is a sockaddr_in 404 * and in_pcbbind requires a valid address. Since divert 405 * sockets don't we need to make sure the address is 406 * filled in properly. 407 * XXX -- divert should not be abusing in_pcbind 408 * and should probably have its own family. 409 */ 410 if (nam->sa_family != AF_INET) { 411 error = EAFNOSUPPORT; 412 } else { 413 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 414 error = in_pcbbind(inp, nam, p); 415 } 416 splx(s); 417 return error; 418} 419 420static int 421div_shutdown(struct socket *so) 422{ 423 socantsendmore(so); 424 return 0; 425} 426 427static int 428div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 429 struct mbuf *control, struct proc *p) 430{ 431 /* Packet must have a header (but that's about it) */ 432 if (m->m_len < sizeof (struct ip) && 433 (m = m_pullup(m, sizeof (struct ip))) == 0) { 434 ipstat.ips_toosmall++; 435 m_freem(m); 436 return EINVAL; 437 } 438 439 /* Send packet */ 440 return div_output(so, m, nam, control); 441} 442 443static int 444div_pcblist(SYSCTL_HANDLER_ARGS) 445{ 446 int error, i, n, s; 447 struct inpcb *inp, **inp_list; 448 inp_gen_t gencnt; 449 struct xinpgen xig; 450 451 /* 452 * The process of preparing the TCB list is too time-consuming and 453 * resource-intensive to repeat twice on every request. 454 */ 455 if (req->oldptr == 0) { 456 n = divcbinfo.ipi_count; 457 req->oldidx = 2 * (sizeof xig) 458 + (n + n/8) * sizeof(struct xinpcb); 459 return 0; 460 } 461 462 if (req->newptr != 0) 463 return EPERM; 464 465 /* 466 * OK, now we're committed to doing something. 467 */ 468 s = splnet(); 469 gencnt = divcbinfo.ipi_gencnt; 470 n = divcbinfo.ipi_count; 471 splx(s); 472 473 xig.xig_len = sizeof xig; 474 xig.xig_count = n; 475 xig.xig_gen = gencnt; 476 xig.xig_sogen = so_gencnt; 477 error = SYSCTL_OUT(req, &xig, sizeof xig); 478 if (error) 479 return error; 480 481 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 482 if (inp_list == 0) 483 return ENOMEM; 484 485 s = splnet(); 486 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n; 487 inp = LIST_NEXT(inp, inp_list)) { 488 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp)) 489 inp_list[i++] = inp; 490 } 491 splx(s); 492 n = i; 493 494 error = 0; 495 for (i = 0; i < n; i++) { 496 inp = inp_list[i]; 497 if (inp->inp_gencnt <= gencnt) { 498 struct xinpcb xi; 499 xi.xi_len = sizeof xi; 500 /* XXX should avoid extra copy */ 501 bcopy(inp, &xi.xi_inp, sizeof *inp); 502 if (inp->inp_socket) 503 sotoxsocket(inp->inp_socket, &xi.xi_socket); 504 error = SYSCTL_OUT(req, &xi, sizeof xi); 505 } 506 } 507 if (!error) { 508 /* 509 * Give the user an updated idea of our state. 510 * If the generation differs from what we told 511 * her before, she knows that something happened 512 * while we were processing this request, and it 513 * might be necessary to retry. 514 */ 515 s = splnet(); 516 xig.xig_gen = divcbinfo.ipi_gencnt; 517 xig.xig_sogen = so_gencnt; 518 xig.xig_count = divcbinfo.ipi_count; 519 splx(s); 520 error = SYSCTL_OUT(req, &xig, sizeof xig); 521 } 522 free(inp_list, M_TEMP); 523 return error; 524} 525 526SYSCTL_DECL(_net_inet_divert); 527SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 528 div_pcblist, "S,xinpcb", "List of active divert sockets"); 529 530struct pr_usrreqs div_usrreqs = { 531 div_abort, pru_accept_notsupp, div_attach, div_bind, 532 pru_connect_notsupp, pru_connect2_notsupp, in_control, div_detach, 533 div_disconnect, pru_listen_notsupp, in_setpeeraddr, pru_rcvd_notsupp, 534 pru_rcvoob_notsupp, div_send, pru_sense_null, div_shutdown, 535 in_setsockaddr, sosend, soreceive, sopoll 536}; 537