30 */ 31 32#if !defined(KLD_MODULE) 33#include "opt_inet.h" 34#include "opt_ipfw.h" 35#include "opt_mac.h" 36#ifndef INET 37#error "IPDIVERT requires INET." 38#endif 39#ifndef IPFIREWALL 40#error "IPDIVERT requires IPFIREWALL" 41#endif 42#endif 43 44#include <sys/param.h> 45#include <sys/kernel.h> 46#include <sys/lock.h> 47#include <sys/malloc.h> 48#include <sys/mac.h> 49#include <sys/mbuf.h> 50#include <sys/module.h> 51#include <sys/kernel.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_divert.h> 72#include <netinet/ip_var.h> 73#include <netinet/ip_fw.h> 74 75/* 76 * Divert sockets 77 */ 78 79/* 80 * Allocate enough space to hold a full IP packet 81 */ 82#define DIVSNDQ (65536 + 100) 83#define DIVRCVQ (65536 + 100) 84 85/* 86 * Divert sockets work in conjunction with ipfw, see the divert(4) 87 * manpage for features. 88 * Internally, packets selected by ipfw in ip_input() or ip_output(), 89 * and never diverted before, are passed to the input queue of the 90 * divert socket with a given 'divert_port' number (as specified in 91 * the matching ipfw rule), and they are tagged with a 16 bit cookie 92 * (representing the rule number of the matching ipfw rule), which 93 * is passed to process reading from the socket. 94 * 95 * Packets written to the divert socket are again tagged with a cookie 96 * (usually the same as above) and a destination address. 97 * If the destination address is INADDR_ANY then the packet is 98 * treated as outgoing and sent to ip_output(), otherwise it is 99 * treated as incoming and sent to ip_input(). 100 * In both cases, the packet is tagged with the cookie. 101 * 102 * On reinjection, processing in ip_input() and ip_output() 103 * will be exactly the same as for the original packet, except that 104 * ipfw processing will start at the rule number after the one 105 * written in the cookie (so, tagging a packet with a cookie of 0 106 * will cause it to be effectively considered as a standard packet). 107 */ 108 109/* Internal variables. */ 110static struct inpcbhead divcb; 111static struct inpcbinfo divcbinfo; 112 113static u_long div_sendspace = DIVSNDQ; /* XXX sysctl ? */ 114static u_long div_recvspace = DIVRCVQ; /* XXX sysctl ? */ 115 116/* 117 * Initialize divert connection block queue. 118 */ 119void 120div_init(void) 121{ 122 INP_INFO_LOCK_INIT(&divcbinfo, "div"); 123 LIST_INIT(&divcb); 124 divcbinfo.listhead = &divcb; 125 /* 126 * XXX We don't use the hash list for divert IP, but it's easier 127 * to allocate a one entry hash list than it is to check all 128 * over the place for hashbase == NULL. 129 */ 130 divcbinfo.hashbase = hashinit(1, M_PCB, &divcbinfo.hashmask); 131 divcbinfo.porthashbase = hashinit(1, M_PCB, &divcbinfo.porthashmask); 132 divcbinfo.ipi_zone = uma_zcreate("divcb", sizeof(struct inpcb), 133 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 134 uma_zone_set_max(divcbinfo.ipi_zone, maxsockets); 135} 136 137/* 138 * IPPROTO_DIVERT is not in the real IP protocol number space; this 139 * function should never be called. Just in case, drop any packets. 140 */ 141void 142div_input(struct mbuf *m, int off) 143{ 144 ipstat.ips_noproto++; 145 m_freem(m); 146} 147 148/* 149 * Divert a packet by passing it up to the divert socket at port 'port'. 150 * 151 * Setup generic address and protocol structures for div_input routine, 152 * then pass them along with mbuf chain. 153 */ 154static void 155divert_packet(struct mbuf *m, int incoming) 156{ 157 struct ip *ip; 158 struct inpcb *inp; 159 struct socket *sa; 160 u_int16_t nport; 161 struct sockaddr_in divsrc; 162 struct m_tag *mtag; 163 164 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL); 165 if (mtag == NULL) { 166 printf("%s: no divert tag\n", __func__); 167 m_freem(m); 168 return; 169 } 170 /* Assure header */ 171 if (m->m_len < sizeof(struct ip) && 172 (m = m_pullup(m, sizeof(struct ip))) == 0) 173 return; 174 ip = mtod(m, struct ip *); 175 176 /* Delayed checksums are currently not compatible with divert. */ 177 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) { 178 ip->ip_len = ntohs(ip->ip_len); 179 in_delayed_cksum(m); 180 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA; 181 ip->ip_len = htons(ip->ip_len); 182 } 183 184 /* 185 * Record receive interface address, if any. 186 * But only for incoming packets. 187 */ 188 bzero(&divsrc, sizeof(divsrc)); 189 divsrc.sin_len = sizeof(divsrc); 190 divsrc.sin_family = AF_INET; 191 divsrc.sin_port = divert_cookie(mtag); /* record matching rule */ 192 if (incoming) { 193 struct ifaddr *ifa; 194 195 /* Sanity check */ 196 M_ASSERTPKTHDR(m); 197 198 /* Find IP address for receive interface */ 199 TAILQ_FOREACH(ifa, &m->m_pkthdr.rcvif->if_addrhead, ifa_link) { 200 if (ifa->ifa_addr == NULL) 201 continue; 202 if (ifa->ifa_addr->sa_family != AF_INET) 203 continue; 204 divsrc.sin_addr = 205 ((struct sockaddr_in *) ifa->ifa_addr)->sin_addr; 206 break; 207 } 208 } 209 /* 210 * Record the incoming interface name whenever we have one. 211 */ 212 if (m->m_pkthdr.rcvif) { 213 /* 214 * Hide the actual interface name in there in the 215 * sin_zero array. XXX This needs to be moved to a 216 * different sockaddr type for divert, e.g. 217 * sockaddr_div with multiple fields like 218 * sockaddr_dl. Presently we have only 7 bytes 219 * but that will do for now as most interfaces 220 * are 4 or less + 2 or less bytes for unit. 221 * There is probably a faster way of doing this, 222 * possibly taking it from the sockaddr_dl on the iface. 223 * This solves the problem of a P2P link and a LAN interface 224 * having the same address, which can result in the wrong 225 * interface being assigned to the packet when fed back 226 * into the divert socket. Theoretically if the daemon saves 227 * and re-uses the sockaddr_in as suggested in the man pages, 228 * this iface name will come along for the ride. 229 * (see div_output for the other half of this.) 230 */ 231 strlcpy(divsrc.sin_zero, m->m_pkthdr.rcvif->if_xname, 232 sizeof(divsrc.sin_zero)); 233 } 234 235 /* Put packet on socket queue, if any */ 236 sa = NULL; 237 nport = htons((u_int16_t)divert_info(mtag)); 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 SOCKBUF_LOCK(&sa->so_rcv); 245 if (sbappendaddr_locked(&sa->so_rcv, 246 (struct sockaddr *)&divsrc, m, 247 (struct mbuf *)0) == 0) { 248 SOCKBUF_UNLOCK(&sa->so_rcv); 249 sa = NULL; /* force mbuf reclaim below */ 250 } else 251 sorwakeup_locked(sa); 252 INP_UNLOCK(inp); 253 break; 254 } 255 INP_UNLOCK(inp); 256 } 257 INP_INFO_RUNLOCK(&divcbinfo); 258 if (sa == NULL) { 259 m_freem(m); 260 ipstat.ips_noproto++; 261 ipstat.ips_delivered--; 262 } 263} 264 265/* 266 * Deliver packet back into the IP processing machinery. 267 * 268 * If no address specified, or address is 0.0.0.0, send to ip_output(); 269 * otherwise, send to ip_input() and mark as having been received on 270 * the interface with that address. 271 */ 272static int 273div_output(struct socket *so, struct mbuf *m, 274 struct sockaddr_in *sin, struct mbuf *control) 275{ 276 struct m_tag *mtag; 277 struct divert_tag *dt; 278 int error = 0; 279 280 /* 281 * An mbuf may hasn't come from userland, but we pretend 282 * that it has. 283 */ 284 m->m_pkthdr.rcvif = NULL; 285 m->m_nextpkt = NULL; 286 287 if (control) 288 m_freem(control); /* XXX */ 289 290 if ((mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL)) == NULL) { 291 mtag = m_tag_get(PACKET_TAG_DIVERT, sizeof(struct divert_tag), 292 M_NOWAIT | M_ZERO); 293 if (mtag == NULL) { 294 error = ENOBUFS; 295 goto cantsend; 296 } 297 dt = (struct divert_tag *)(mtag+1); 298 m_tag_prepend(m, mtag); 299 } else 300 dt = (struct divert_tag *)(mtag+1); 301 302 /* Loopback avoidance and state recovery */ 303 if (sin) { 304 int i; 305 306 dt->cookie = sin->sin_port; 307 /* 308 * Find receive interface with the given name, stuffed 309 * (if it exists) in the sin_zero[] field. 310 * The name is user supplied data so don't trust its size 311 * or that it is zero terminated. 312 */ 313 for (i = 0; i < sizeof(sin->sin_zero) && sin->sin_zero[i]; i++) 314 ; 315 if ( i > 0 && i < sizeof(sin->sin_zero)) 316 m->m_pkthdr.rcvif = ifunit(sin->sin_zero); 317 } 318 319 /* Reinject packet into the system as incoming or outgoing */ 320 if (!sin || sin->sin_addr.s_addr == 0) { 321 struct ip *const ip = mtod(m, struct ip *); 322 struct inpcb *inp; 323 324 dt->info |= IP_FW_DIVERT_OUTPUT_FLAG; 325 INP_INFO_WLOCK(&divcbinfo); 326 inp = sotoinpcb(so); 327 INP_LOCK(inp); 328 /* 329 * Don't allow both user specified and setsockopt options, 330 * and don't allow packet length sizes that will crash 331 */ 332 if (((ip->ip_hl != (sizeof (*ip) >> 2)) && inp->inp_options) || 333 ((u_short)ntohs(ip->ip_len) > m->m_pkthdr.len)) { 334 error = EINVAL; 335 m_freem(m); 336 } else { 337 /* Convert fields to host order for ip_output() */ 338 ip->ip_len = ntohs(ip->ip_len); 339 ip->ip_off = ntohs(ip->ip_off); 340 341 /* Send packet to output processing */ 342 ipstat.ips_rawout++; /* XXX */ 343 344#ifdef MAC 345 mac_create_mbuf_from_inpcb(inp, m); 346#endif 347 error = ip_output(m, 348 inp->inp_options, NULL, 349 ((so->so_options & SO_DONTROUTE) ? 350 IP_ROUTETOIF : 0) | 351 IP_ALLOWBROADCAST | IP_RAWOUTPUT, 352 inp->inp_moptions, NULL); 353 } 354 INP_UNLOCK(inp); 355 INP_INFO_WUNLOCK(&divcbinfo); 356 } else { 357 dt->info |= IP_FW_DIVERT_LOOPBACK_FLAG; 358 if (m->m_pkthdr.rcvif == NULL) { 359 /* 360 * No luck with the name, check by IP address. 361 * Clear the port and the ifname to make sure 362 * there are no distractions for ifa_ifwithaddr. 363 */ 364 struct ifaddr *ifa; 365 366 bzero(sin->sin_zero, sizeof(sin->sin_zero)); 367 sin->sin_port = 0; 368 ifa = ifa_ifwithaddr((struct sockaddr *) sin); 369 if (ifa == NULL) { 370 error = EADDRNOTAVAIL; 371 goto cantsend; 372 } 373 m->m_pkthdr.rcvif = ifa->ifa_ifp; 374 } 375#ifdef MAC 376 SOCK_LOCK(so); 377 mac_create_mbuf_from_socket(so, m); 378 SOCK_UNLOCK(so); 379#endif 380 /* Send packet to input processing */ 381 ip_input(m); 382 } 383 384 return error; 385 386cantsend: 387 m_freem(m); 388 return error; 389} 390 391static int 392div_attach(struct socket *so, int proto, struct thread *td) 393{ 394 struct inpcb *inp; 395 int error; 396 397 INP_INFO_WLOCK(&divcbinfo); 398 inp = sotoinpcb(so); 399 if (inp != 0) { 400 INP_INFO_WUNLOCK(&divcbinfo); 401 return EINVAL; 402 } 403 if (td && (error = suser(td)) != 0) { 404 INP_INFO_WUNLOCK(&divcbinfo); 405 return error; 406 } 407 error = soreserve(so, div_sendspace, div_recvspace); 408 if (error) { 409 INP_INFO_WUNLOCK(&divcbinfo); 410 return error; 411 } 412 error = in_pcballoc(so, &divcbinfo, "divinp"); 413 if (error) { 414 INP_INFO_WUNLOCK(&divcbinfo); 415 return error; 416 } 417 inp = (struct inpcb *)so->so_pcb; 418 INP_LOCK(inp); 419 INP_INFO_WUNLOCK(&divcbinfo); 420 inp->inp_ip_p = proto; 421 inp->inp_vflag |= INP_IPV4; 422 inp->inp_flags |= INP_HDRINCL; 423 INP_UNLOCK(inp); 424 return 0; 425} 426 427static int 428div_detach(struct socket *so) 429{ 430 struct inpcb *inp; 431 432 INP_INFO_WLOCK(&divcbinfo); 433 inp = sotoinpcb(so); 434 if (inp == 0) { 435 INP_INFO_WUNLOCK(&divcbinfo); 436 return EINVAL; 437 } 438 INP_LOCK(inp); 439 in_pcbdetach(inp); 440 INP_INFO_WUNLOCK(&divcbinfo); 441 return 0; 442} 443 444static int 445div_bind(struct socket *so, struct sockaddr *nam, struct thread *td) 446{ 447 struct inpcb *inp; 448 int error; 449 450 INP_INFO_WLOCK(&divcbinfo); 451 inp = sotoinpcb(so); 452 if (inp == 0) { 453 INP_INFO_WUNLOCK(&divcbinfo); 454 return EINVAL; 455 } 456 /* in_pcbbind assumes that nam is a sockaddr_in 457 * and in_pcbbind requires a valid address. Since divert 458 * sockets don't we need to make sure the address is 459 * filled in properly. 460 * XXX -- divert should not be abusing in_pcbind 461 * and should probably have its own family. 462 */ 463 if (nam->sa_family != AF_INET) 464 error = EAFNOSUPPORT; 465 else { 466 ((struct sockaddr_in *)nam)->sin_addr.s_addr = INADDR_ANY; 467 INP_LOCK(inp); 468 error = in_pcbbind(inp, nam, td->td_ucred); 469 INP_UNLOCK(inp); 470 } 471 INP_INFO_WUNLOCK(&divcbinfo); 472 return error; 473} 474 475static int 476div_shutdown(struct socket *so) 477{ 478 struct inpcb *inp; 479 480 INP_INFO_RLOCK(&divcbinfo); 481 inp = sotoinpcb(so); 482 if (inp == 0) { 483 INP_INFO_RUNLOCK(&divcbinfo); 484 return EINVAL; 485 } 486 INP_LOCK(inp); 487 INP_INFO_RUNLOCK(&divcbinfo); 488 socantsendmore(so); 489 INP_UNLOCK(inp); 490 return 0; 491} 492 493static int 494div_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam, 495 struct mbuf *control, struct thread *td) 496{ 497 /* Packet must have a header (but that's about it) */ 498 if (m->m_len < sizeof (struct ip) && 499 (m = m_pullup(m, sizeof (struct ip))) == 0) { 500 ipstat.ips_toosmall++; 501 m_freem(m); 502 return EINVAL; 503 } 504 505 /* Send packet */ 506 return div_output(so, m, (struct sockaddr_in *)nam, control); 507} 508 509void 510div_ctlinput(int cmd, struct sockaddr *sa, void *vip) 511{ 512 struct in_addr faddr; 513 514 faddr = ((struct sockaddr_in *)sa)->sin_addr; 515 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 516 return; 517 if (PRC_IS_REDIRECT(cmd)) 518 return; 519} 520 521static int 522div_pcblist(SYSCTL_HANDLER_ARGS) 523{ 524 int error, i, n; 525 struct inpcb *inp, **inp_list; 526 inp_gen_t gencnt; 527 struct xinpgen xig; 528 529 /* 530 * The process of preparing the TCB list is too time-consuming and 531 * resource-intensive to repeat twice on every request. 532 */ 533 if (req->oldptr == 0) { 534 n = divcbinfo.ipi_count; 535 req->oldidx = 2 * (sizeof xig) 536 + (n + n/8) * sizeof(struct xinpcb); 537 return 0; 538 } 539 540 if (req->newptr != 0) 541 return EPERM; 542 543 /* 544 * OK, now we're committed to doing something. 545 */ 546 INP_INFO_RLOCK(&divcbinfo); 547 gencnt = divcbinfo.ipi_gencnt; 548 n = divcbinfo.ipi_count; 549 INP_INFO_RUNLOCK(&divcbinfo); 550 551 error = sysctl_wire_old_buffer(req, 552 2 * sizeof(xig) + n*sizeof(struct xinpcb)); 553 if (error != 0) 554 return (error); 555 556 xig.xig_len = sizeof xig; 557 xig.xig_count = n; 558 xig.xig_gen = gencnt; 559 xig.xig_sogen = so_gencnt; 560 error = SYSCTL_OUT(req, &xig, sizeof xig); 561 if (error) 562 return error; 563 564 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 565 if (inp_list == 0) 566 return ENOMEM; 567 568 INP_INFO_RLOCK(&divcbinfo); 569 for (inp = LIST_FIRST(divcbinfo.listhead), i = 0; inp && i < n; 570 inp = LIST_NEXT(inp, inp_list)) { 571 INP_LOCK(inp); 572 if (inp->inp_gencnt <= gencnt && 573 cr_canseesocket(req->td->td_ucred, inp->inp_socket) == 0) 574 inp_list[i++] = inp; 575 INP_UNLOCK(inp); 576 } 577 INP_INFO_RUNLOCK(&divcbinfo); 578 n = i; 579 580 error = 0; 581 for (i = 0; i < n; i++) { 582 inp = inp_list[i]; 583 if (inp->inp_gencnt <= gencnt) { 584 struct xinpcb xi; 585 bzero(&xi, sizeof(xi)); 586 xi.xi_len = sizeof xi; 587 /* XXX should avoid extra copy */ 588 bcopy(inp, &xi.xi_inp, sizeof *inp); 589 if (inp->inp_socket) 590 sotoxsocket(inp->inp_socket, &xi.xi_socket); 591 error = SYSCTL_OUT(req, &xi, sizeof xi); 592 } 593 } 594 if (!error) { 595 /* 596 * Give the user an updated idea of our state. 597 * If the generation differs from what we told 598 * her before, she knows that something happened 599 * while we were processing this request, and it 600 * might be necessary to retry. 601 */ 602 INP_INFO_RLOCK(&divcbinfo); 603 xig.xig_gen = divcbinfo.ipi_gencnt; 604 xig.xig_sogen = so_gencnt; 605 xig.xig_count = divcbinfo.ipi_count; 606 INP_INFO_RUNLOCK(&divcbinfo); 607 error = SYSCTL_OUT(req, &xig, sizeof xig); 608 } 609 free(inp_list, M_TEMP); 610 return error; 611} 612 613/* 614 * This is the wrapper function for in_setsockaddr. We just pass down 615 * the pcbinfo for in_setpeeraddr to lock. 616 */ 617static int 618div_sockaddr(struct socket *so, struct sockaddr **nam) 619{ 620 return (in_setsockaddr(so, nam, &divcbinfo)); 621} 622 623/* 624 * This is the wrapper function for in_setpeeraddr. We just pass down 625 * the pcbinfo for in_setpeeraddr to lock. 626 */ 627static int 628div_peeraddr(struct socket *so, struct sockaddr **nam) 629{ 630 return (in_setpeeraddr(so, nam, &divcbinfo)); 631} 632 633#ifdef SYSCTL_NODE 634SYSCTL_NODE(_net_inet, IPPROTO_DIVERT, divert, CTLFLAG_RW, 0, "IPDIVERT"); 635SYSCTL_PROC(_net_inet_divert, OID_AUTO, pcblist, CTLFLAG_RD, 0, 0, 636 div_pcblist, "S,xinpcb", "List of active divert sockets"); 637#endif 638 639struct pr_usrreqs div_usrreqs = { 640 .pru_attach = div_attach, 641 .pru_bind = div_bind, 642 .pru_control = in_control, 643 .pru_detach = div_detach, 644 .pru_peeraddr = div_peeraddr, 645 .pru_send = div_send, 646 .pru_shutdown = div_shutdown, 647 .pru_sockaddr = div_sockaddr, 648 .pru_sosetlabel = in_pcbsosetlabel 649}; 650 651struct protosw div_protosw = {
|