1 /* 2 * Copyright (c) 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of The DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 36 * The Regents of the University of California. All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. Neither the name of the University nor the names of its contributors 47 * may be used to endorse or promote products derived from this software 48 * without specific prior written permission. 49 * 50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 60 * SUCH DAMAGE. 61 * 62 * @(#)udp_usrreq.c 8.6 (Berkeley) 5/23/95 63 * $FreeBSD: src/sys/netinet/udp_usrreq.c,v 1.64.2.18 2003/01/24 05:11:34 sam Exp $ 64 */ 65 66 #include "opt_ipsec.h" 67 #include "opt_inet6.h" 68 69 #include <sys/param.h> 70 #include <sys/systm.h> 71 #include <sys/kernel.h> 72 #include <sys/malloc.h> 73 #include <sys/mbuf.h> 74 #include <sys/domain.h> 75 #include <sys/proc.h> 76 #include <sys/priv.h> 77 #include <sys/protosw.h> 78 #include <sys/socket.h> 79 #include <sys/socketvar.h> 80 #include <sys/sysctl.h> 81 #include <sys/syslog.h> 82 #include <sys/in_cksum.h> 83 #include <sys/ktr.h> 84 85 #include <sys/thread2.h> 86 #include <sys/socketvar2.h> 87 #include <sys/serialize.h> 88 89 #include <machine/stdarg.h> 90 91 #include <net/if.h> 92 #include <net/route.h> 93 #include <net/netmsg2.h> 94 #include <net/netisr2.h> 95 96 #include <netinet/in.h> 97 #include <netinet/in_systm.h> 98 #include <netinet/ip.h> 99 #ifdef INET6 100 #include <netinet/ip6.h> 101 #endif 102 #include <netinet/in_pcb.h> 103 #include <netinet/in_var.h> 104 #include <netinet/ip_var.h> 105 #ifdef INET6 106 #include <netinet6/ip6_var.h> 107 #endif 108 #include <netinet/ip_icmp.h> 109 #include <netinet/icmp_var.h> 110 #include <netinet/udp.h> 111 #include <netinet/udp_var.h> 112 113 #ifdef FAST_IPSEC 114 #include <netproto/ipsec/ipsec.h> 115 #endif 116 117 #ifdef IPSEC 118 #include <netinet6/ipsec.h> 119 #endif 120 121 #define MSGF_UDP_SEND MSGF_PROTO1 122 123 #define INP_DIRECT_DETACH INP_FLAG_PROTO2 124 125 #define UDP_KTR_STRING "inp=%p" 126 #define UDP_KTR_ARGS struct inpcb *inp 127 128 #ifndef KTR_UDP 129 #define KTR_UDP KTR_ALL 130 #endif 131 132 KTR_INFO_MASTER(udp); 133 KTR_INFO(KTR_UDP, udp, send_beg, 0, UDP_KTR_STRING, UDP_KTR_ARGS); 134 KTR_INFO(KTR_UDP, udp, send_end, 1, UDP_KTR_STRING, UDP_KTR_ARGS); 135 KTR_INFO(KTR_UDP, udp, send_ipout, 2, UDP_KTR_STRING, UDP_KTR_ARGS); 136 KTR_INFO(KTR_UDP, udp, redisp_ipout_beg, 3, UDP_KTR_STRING, UDP_KTR_ARGS); 137 KTR_INFO(KTR_UDP, udp, redisp_ipout_end, 4, UDP_KTR_STRING, UDP_KTR_ARGS); 138 KTR_INFO(KTR_UDP, udp, send_redisp, 5, UDP_KTR_STRING, UDP_KTR_ARGS); 139 KTR_INFO(KTR_UDP, udp, send_inswildcard, 6, UDP_KTR_STRING, UDP_KTR_ARGS); 140 141 #define logudp(name, inp) KTR_LOG(udp_##name, inp) 142 143 /* 144 * UDP protocol implementation. 145 * Per RFC 768, August, 1980. 146 */ 147 #ifndef COMPAT_42 148 static int udpcksum = 1; 149 #else 150 static int udpcksum = 0; /* XXX */ 151 #endif 152 SYSCTL_INT(_net_inet_udp, UDPCTL_CHECKSUM, checksum, CTLFLAG_RW, 153 &udpcksum, 0, "Enable checksumming of UDP packets"); 154 155 int log_in_vain = 0; 156 SYSCTL_INT(_net_inet_udp, OID_AUTO, log_in_vain, CTLFLAG_RW, 157 &log_in_vain, 0, "Log all incoming UDP packets"); 158 159 static int blackhole = 0; 160 SYSCTL_INT(_net_inet_udp, OID_AUTO, blackhole, CTLFLAG_RW, 161 &blackhole, 0, "Do not send port unreachables for refused connects"); 162 163 static int strict_mcast_mship = 1; 164 SYSCTL_INT(_net_inet_udp, OID_AUTO, strict_mcast_mship, CTLFLAG_RW, 165 &strict_mcast_mship, 0, "Only send multicast to member sockets"); 166 167 int udp_sosend_async = 1; 168 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_async, CTLFLAG_RW, 169 &udp_sosend_async, 0, "UDP asynchronized pru_send"); 170 171 int udp_sosend_prepend = 1; 172 SYSCTL_INT(_net_inet_udp, OID_AUTO, sosend_prepend, CTLFLAG_RW, 173 &udp_sosend_prepend, 0, 174 "Prepend enough space for proto and link header in pru_send"); 175 176 static int udp_reuseport_ext = 1; 177 SYSCTL_INT(_net_inet_udp, OID_AUTO, reuseport_ext, CTLFLAG_RW, 178 &udp_reuseport_ext, 0, "SO_REUSEPORT extension"); 179 180 struct inpcbinfo udbinfo[MAXCPU]; 181 182 #ifndef UDBHASHSIZE 183 #define UDBHASHSIZE 16 184 #endif 185 186 struct udpstat udpstat_percpu[MAXCPU] __cachealign; 187 188 static void udp_append(struct inpcb *last, struct ip *ip, 189 struct mbuf *n, int off, struct sockaddr_in *udp_in); 190 191 static int udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin, 192 struct sockaddr_in *if_sin, uint16_t hash); 193 194 static boolean_t udp_inswildcardhash(struct inpcb *inp, 195 struct netmsg_base *msg, int error); 196 static void udp_remwildcardhash(struct inpcb *inp); 197 198 static __inline int 199 udp_lportcpu(short lport) 200 { 201 return (ntohs(lport) % netisr_ncpus); 202 } 203 204 void 205 udp_init(void) 206 { 207 struct inpcbportinfo *portinfo; 208 int cpu; 209 210 portinfo = kmalloc_cachealign(sizeof(*portinfo) * netisr_ncpus, M_PCB, 211 M_WAITOK); 212 213 for (cpu = 0; cpu < netisr_ncpus; cpu++) { 214 struct inpcbinfo *uicb = &udbinfo[cpu]; 215 216 /* 217 * NOTE: 218 * UDP pcb list, wildcard hash table and localgroup hash 219 * table are shared. 220 */ 221 in_pcbinfo_init(uicb, cpu, TRUE); 222 uicb->hashbase = hashinit(UDBHASHSIZE, M_PCB, &uicb->hashmask); 223 224 in_pcbportinfo_init(&portinfo[cpu], UDBHASHSIZE, cpu); 225 in_pcbportinfo_set(uicb, portinfo, netisr_ncpus); 226 227 uicb->wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB, 228 &uicb->wildcardhashmask); 229 uicb->localgrphashbase = hashinit(UDBHASHSIZE, M_PCB, 230 &uicb->localgrphashmask); 231 232 uicb->ipi_size = sizeof(struct inpcb); 233 } 234 235 /* 236 * Initialize UDP statistics counters for each CPU. 237 */ 238 for (cpu = 0; cpu < netisr_ncpus; ++cpu) 239 bzero(&udpstat_percpu[cpu], sizeof(struct udpstat)); 240 } 241 242 static int 243 sysctl_udpstat(SYSCTL_HANDLER_ARGS) 244 { 245 int cpu, error = 0; 246 247 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 248 if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu], 249 sizeof(struct udpstat)))) 250 break; 251 if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu], 252 sizeof(struct udpstat)))) 253 break; 254 } 255 256 return (error); 257 } 258 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 259 0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics"); 260 261 void 262 udp_ctloutput(netmsg_t msg) 263 { 264 struct socket *so = msg->base.nm_so; 265 struct sockopt *sopt = msg->ctloutput.nm_sopt; 266 struct inpcb *inp = so->so_pcb; 267 268 if (inp == NULL) { 269 lwkt_replymsg(&msg->lmsg, EINVAL); 270 return; 271 } 272 273 if (sopt->sopt_level == IPPROTO_IP && sopt->sopt_dir == SOPT_SET) { 274 switch (sopt->sopt_name) { 275 case IP_MULTICAST_IF: 276 case IP_MULTICAST_VIF: 277 case IP_MULTICAST_TTL: 278 case IP_MULTICAST_LOOP: 279 case IP_ADD_MEMBERSHIP: 280 case IP_DROP_MEMBERSHIP: 281 /* 282 * This pr_ctloutput msg will be forwarded 283 * to netisr0 to run; we can't do direct 284 * detaching anymore. 285 * 286 * NOTE: 287 * Don't optimize for the sockets whose 288 * current so_port is netisr0's msgport. 289 * These sockets could be connect(2)'ed 290 * later and the so_port will be changed. 291 */ 292 inp->inp_flags &= ~INP_DIRECT_DETACH; 293 break; 294 } 295 } 296 return ip_ctloutput(msg); 297 } 298 299 /* 300 * Check multicast packets to make sure they are only sent to sockets with 301 * multicast memberships for the packet's destination address and arrival 302 * interface. Multicast packets to multicast-unaware sockets are also 303 * disallowed. 304 * 305 * Returns 0 if the packet is acceptable, -1 if it is not. 306 */ 307 static __inline int 308 check_multicast_membership(const struct ip *ip, const struct inpcb *inp, 309 const struct mbuf *m) 310 { 311 const struct ip_moptions *mopt; 312 int mshipno; 313 314 if (strict_mcast_mship == 0 || 315 !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 316 return (0); 317 } 318 319 ASSERT_NETISR0; 320 321 mopt = inp->inp_moptions; 322 if (mopt == NULL) 323 return (-1); 324 for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) { 325 const struct in_multi *maddr = mopt->imo_membership[mshipno]; 326 327 if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr && 328 m->m_pkthdr.rcvif == maddr->inm_ifp) { 329 return (0); 330 } 331 } 332 return (-1); 333 } 334 335 struct udp_mcast_arg { 336 struct inpcb *inp; 337 struct inpcb *last; 338 struct ip *ip; 339 struct mbuf *m; 340 int iphlen; 341 struct sockaddr_in *udp_in; 342 }; 343 344 static int 345 udp_mcast_input(struct udp_mcast_arg *arg) 346 { 347 struct inpcb *inp = arg->inp; 348 struct inpcb *last = arg->last; 349 struct ip *ip = arg->ip; 350 struct mbuf *m = arg->m; 351 352 if (check_multicast_membership(ip, inp, m) < 0) 353 return ERESTART; /* caller continue */ 354 355 if (last != NULL) { 356 struct mbuf *n; 357 358 #ifdef IPSEC 359 /* check AH/ESP integrity. */ 360 if (ipsec4_in_reject_so(m, last->inp_socket)) 361 ipsecstat.in_polvio++; 362 /* do not inject data to pcb */ 363 else 364 #endif /*IPSEC*/ 365 #ifdef FAST_IPSEC 366 /* check AH/ESP integrity. */ 367 if (ipsec4_in_reject(m, last)) 368 ; 369 else 370 #endif /*FAST_IPSEC*/ 371 if ((n = m_copypacket(m, M_NOWAIT)) != NULL) 372 udp_append(last, ip, n, 373 arg->iphlen + sizeof(struct udphdr), 374 arg->udp_in); 375 } 376 arg->last = last = inp; 377 378 /* 379 * Don't look for additional matches if this one does 380 * not have either the SO_REUSEPORT or SO_REUSEADDR 381 * socket options set. This heuristic avoids searching 382 * through all pcbs in the common case of a non-shared 383 * port. It * assumes that an application will never 384 * clear these options after setting them. 385 */ 386 if (!(last->inp_socket->so_options & 387 (SO_REUSEPORT | SO_REUSEADDR))) 388 return EJUSTRETURN; /* caller stop */ 389 return 0; 390 } 391 392 int 393 udp_input(struct mbuf **mp, int *offp, int proto) 394 { 395 struct sockaddr_in udp_in = { sizeof udp_in, AF_INET }; 396 int iphlen; 397 struct ip *ip; 398 struct udphdr *uh; 399 struct inpcb *inp; 400 struct mbuf *m; 401 struct mbuf *opts = NULL; 402 int len, off; 403 struct ip save_ip; 404 struct inpcbinfo *pcbinfo = &udbinfo[mycpuid]; 405 406 off = *offp; 407 m = *mp; 408 *mp = NULL; 409 410 iphlen = off; 411 udp_stat.udps_ipackets++; 412 413 /* 414 * Strip IP options, if any; should skip this, 415 * make available to user, and use on returned packets, 416 * but we don't yet have a way to check the checksum 417 * with options still present. 418 */ 419 if (iphlen > sizeof(struct ip)) { 420 ip_stripoptions(m); 421 iphlen = sizeof(struct ip); 422 } 423 424 /* 425 * IP and UDP headers are together in first mbuf. 426 * Already checked and pulled up in ip_demux(). 427 */ 428 KASSERT(m->m_len >= iphlen + sizeof(struct udphdr), 429 ("UDP header not in one mbuf")); 430 431 ip = mtod(m, struct ip *); 432 uh = (struct udphdr *)((caddr_t)ip + iphlen); 433 434 /* destination port of 0 is illegal, based on RFC768. */ 435 if (uh->uh_dport == 0) 436 goto bad; 437 438 /* 439 * Make mbuf data length reflect UDP length. 440 * If not enough data to reflect UDP length, drop. 441 */ 442 len = ntohs((u_short)uh->uh_ulen); 443 if (ip->ip_len != len) { 444 if (len > ip->ip_len || len < sizeof(struct udphdr)) { 445 udp_stat.udps_badlen++; 446 goto bad; 447 } 448 m_adj(m, len - ip->ip_len); 449 /* ip->ip_len = len; */ 450 } 451 /* 452 * Save a copy of the IP header in case we want restore it 453 * for sending an ICMP error message in response. 454 */ 455 save_ip = *ip; 456 457 /* 458 * Checksum extended UDP header and data. 459 */ 460 if (uh->uh_sum) { 461 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 462 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 463 uh->uh_sum = m->m_pkthdr.csum_data; 464 else 465 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, 466 ip->ip_dst.s_addr, htonl((u_short)len + 467 m->m_pkthdr.csum_data + IPPROTO_UDP)); 468 uh->uh_sum ^= 0xffff; 469 } else { 470 char b[9]; 471 472 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 473 bzero(((struct ipovly *)ip)->ih_x1, 9); 474 ((struct ipovly *)ip)->ih_len = uh->uh_ulen; 475 uh->uh_sum = in_cksum(m, len + sizeof(struct ip)); 476 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 477 } 478 if (uh->uh_sum) { 479 udp_stat.udps_badsum++; 480 m_freem(m); 481 return(IPPROTO_DONE); 482 } 483 } else 484 udp_stat.udps_nosum++; 485 486 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 487 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 488 struct inpcbhead *connhead; 489 struct inpcontainer *ic, *ic_marker; 490 struct inpcontainerhead *ichead; 491 struct udp_mcast_arg arg; 492 struct inpcb *last; 493 int error; 494 495 /* 496 * Deliver a multicast or broadcast datagram to *all* sockets 497 * for which the local and remote addresses and ports match 498 * those of the incoming datagram. This allows more than 499 * one process to receive multi/broadcasts on the same port. 500 * (This really ought to be done for unicast datagrams as 501 * well, but that would cause problems with existing 502 * applications that open both address-specific sockets and 503 * a wildcard socket listening to the same port -- they would 504 * end up receiving duplicates of every unicast datagram. 505 * Those applications open the multiple sockets to overcome an 506 * inadequacy of the UDP socket interface, but for backwards 507 * compatibility we avoid the problem here rather than 508 * fixing the interface. Maybe 4.5BSD will remedy this?) 509 */ 510 511 /* 512 * Construct sockaddr format source address. 513 */ 514 udp_in.sin_port = uh->uh_sport; 515 udp_in.sin_addr = ip->ip_src; 516 arg.udp_in = &udp_in; 517 /* 518 * Locate pcb(s) for datagram. 519 * (Algorithm copied from raw_intr().) 520 */ 521 last = NULL; 522 arg.iphlen = iphlen; 523 524 connhead = &pcbinfo->hashbase[ 525 INP_PCBCONNHASH(ip->ip_src.s_addr, uh->uh_sport, 526 ip->ip_dst.s_addr, uh->uh_dport, pcbinfo->hashmask)]; 527 LIST_FOREACH(inp, connhead, inp_hash) { 528 #ifdef INET6 529 if (!INP_ISIPV4(inp)) 530 continue; 531 #endif 532 if (!in_hosteq(inp->inp_faddr, ip->ip_src) || 533 !in_hosteq(inp->inp_laddr, ip->ip_dst) || 534 inp->inp_fport != uh->uh_sport || 535 inp->inp_lport != uh->uh_dport) 536 continue; 537 538 arg.inp = inp; 539 arg.last = last; 540 arg.ip = ip; 541 arg.m = m; 542 543 error = udp_mcast_input(&arg); 544 if (error == ERESTART) 545 continue; 546 last = arg.last; 547 548 if (error == EJUSTRETURN) 549 goto done; 550 } 551 552 ichead = &pcbinfo->wildcardhashbase[ 553 INP_PCBWILDCARDHASH(uh->uh_dport, 554 pcbinfo->wildcardhashmask)]; 555 ic_marker = in_pcbcontainer_marker(); 556 557 GET_PCBINFO_TOKEN(pcbinfo); 558 LIST_INSERT_HEAD(ichead, ic_marker, ic_list); 559 while ((ic = LIST_NEXT(ic_marker, ic_list)) != NULL) { 560 LIST_REMOVE(ic_marker, ic_list); 561 LIST_INSERT_AFTER(ic, ic_marker, ic_list); 562 563 inp = ic->ic_inp; 564 if (inp->inp_flags & INP_PLACEMARKER) 565 continue; 566 #ifdef INET6 567 if (!INP_ISIPV4(inp)) 568 continue; 569 #endif 570 if (inp->inp_lport != uh->uh_dport) 571 continue; 572 if (inp->inp_laddr.s_addr != INADDR_ANY && 573 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 574 continue; 575 576 arg.inp = inp; 577 arg.last = last; 578 arg.ip = ip; 579 arg.m = m; 580 581 error = udp_mcast_input(&arg); 582 if (error == ERESTART) 583 continue; 584 last = arg.last; 585 586 if (error == EJUSTRETURN) 587 break; 588 } 589 LIST_REMOVE(ic_marker, ic_list); 590 REL_PCBINFO_TOKEN(pcbinfo); 591 done: 592 if (last == NULL) { 593 /* 594 * No matching pcb found; discard datagram. 595 * (No need to send an ICMP Port Unreachable 596 * for a broadcast or multicast datgram.) 597 */ 598 udp_stat.udps_noportbcast++; 599 goto bad; 600 } 601 #ifdef IPSEC 602 /* check AH/ESP integrity. */ 603 if (ipsec4_in_reject_so(m, last->inp_socket)) { 604 ipsecstat.in_polvio++; 605 goto bad; 606 } 607 #endif /*IPSEC*/ 608 #ifdef FAST_IPSEC 609 /* check AH/ESP integrity. */ 610 if (ipsec4_in_reject(m, last)) 611 goto bad; 612 #endif /*FAST_IPSEC*/ 613 udp_append(last, ip, m, iphlen + sizeof(struct udphdr), 614 &udp_in); 615 return(IPPROTO_DONE); 616 } 617 /* 618 * Locate pcb for datagram. 619 */ 620 inp = in_pcblookup_pkthash(pcbinfo, ip->ip_src, uh->uh_sport, 621 ip->ip_dst, uh->uh_dport, TRUE, m->m_pkthdr.rcvif, 622 udp_reuseport_ext ? m : NULL); 623 if (inp == NULL) { 624 if (log_in_vain) { 625 char src[INET_ADDRSTRLEN], dst[INET_ADDRSTRLEN]; 626 627 log(LOG_INFO, 628 "Connection attempt to UDP %s:%d from %s:%d\n", 629 kinet_ntoa(ip->ip_dst, dst), ntohs(uh->uh_dport), 630 kinet_ntoa(ip->ip_src, src), ntohs(uh->uh_sport)); 631 } 632 udp_stat.udps_noport++; 633 if (m->m_flags & (M_BCAST | M_MCAST)) { 634 udp_stat.udps_noportbcast++; 635 goto bad; 636 } 637 if (blackhole) 638 goto bad; 639 #ifdef ICMP_BANDLIM 640 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 641 goto bad; 642 #endif 643 *ip = save_ip; 644 ip->ip_len += iphlen; 645 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 646 return(IPPROTO_DONE); 647 } 648 KASSERT(INP_ISIPV4(inp), ("not inet inpcb")); 649 #ifdef IPSEC 650 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 651 ipsecstat.in_polvio++; 652 goto bad; 653 } 654 #endif /*IPSEC*/ 655 #ifdef FAST_IPSEC 656 if (ipsec4_in_reject(m, inp)) 657 goto bad; 658 #endif /*FAST_IPSEC*/ 659 /* 660 * Check the minimum TTL for socket. 661 */ 662 if (ip->ip_ttl < inp->inp_ip_minttl) 663 goto bad; 664 665 /* 666 * Construct sockaddr format source address. 667 * Stuff source address and datagram in user buffer. 668 */ 669 udp_in.sin_port = uh->uh_sport; 670 udp_in.sin_addr = ip->ip_src; 671 if ((inp->inp_flags & INP_CONTROLOPTS) || 672 (inp->inp_socket->so_options & SO_TIMESTAMP)) 673 ip_savecontrol(inp, &opts, ip, m); 674 m_adj(m, iphlen + sizeof(struct udphdr)); 675 676 lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token); 677 if (ssb_appendaddr(&inp->inp_socket->so_rcv, 678 (struct sockaddr *)&udp_in, m, opts) == 0) { 679 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token); 680 udp_stat.udps_fullsock++; 681 goto bad; 682 } 683 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token); 684 sorwakeup(inp->inp_socket); 685 return(IPPROTO_DONE); 686 bad: 687 m_freem(m); 688 if (opts) 689 m_freem(opts); 690 return(IPPROTO_DONE); 691 } 692 693 /* 694 * subroutine of udp_input(), mainly for source code readability. 695 * caller must properly init udp_ip6 and udp_in6 beforehand. 696 */ 697 static void 698 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off, 699 struct sockaddr_in *udp_in) 700 { 701 struct mbuf *opts = NULL; 702 int ret; 703 704 KASSERT(INP_ISIPV4(last), ("not inet inpcb")); 705 706 if (last->inp_flags & INP_CONTROLOPTS || 707 last->inp_socket->so_options & SO_TIMESTAMP) 708 ip_savecontrol(last, &opts, ip, n); 709 m_adj(n, off); 710 711 lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token); 712 ret = ssb_appendaddr(&last->inp_socket->so_rcv, 713 (struct sockaddr *)udp_in, n, opts); 714 lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token); 715 if (ret == 0) { 716 m_freem(n); 717 if (opts) 718 m_freem(opts); 719 udp_stat.udps_fullsock++; 720 } else { 721 sorwakeup(last->inp_socket); 722 } 723 } 724 725 /* 726 * Notify a udp user of an asynchronous error; 727 * just wake up so that he can collect error status. 728 */ 729 void 730 udp_notify(struct inpcb *inp, int error) 731 { 732 inp->inp_socket->so_error = error; 733 sorwakeup(inp->inp_socket); 734 sowwakeup(inp->inp_socket); 735 } 736 737 struct netmsg_udp_notify { 738 struct netmsg_base base; 739 inp_notify_t nm_notify; 740 struct in_addr nm_faddr; 741 int nm_arg; 742 }; 743 744 static void 745 udp_notifyall_oncpu(netmsg_t msg) 746 { 747 struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg; 748 int nextcpu, cpu = mycpuid; 749 750 ASSERT_NETISR_NCPUS(cpu); 751 752 in_pcbnotifyall(&udbinfo[cpu], nm->nm_faddr, nm->nm_arg, nm->nm_notify); 753 754 nextcpu = cpu + 1; 755 if (nextcpu < netisr_ncpus) 756 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg); 757 else 758 lwkt_replymsg(&nm->base.lmsg, 0); 759 } 760 761 inp_notify_t 762 udp_get_inpnotify(int cmd, const struct sockaddr *sa, 763 struct ip **ip0, int *cpuid) 764 { 765 struct in_addr faddr; 766 struct ip *ip = *ip0; 767 inp_notify_t notify = udp_notify; 768 769 faddr = ((const struct sockaddr_in *)sa)->sin_addr; 770 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 771 return NULL; 772 773 if (PRC_IS_REDIRECT(cmd)) { 774 ip = NULL; 775 notify = in_rtchange; 776 } else if (cmd == PRC_HOSTDEAD) { 777 ip = NULL; 778 } else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) { 779 return NULL; 780 } 781 782 if (cpuid != NULL) { 783 if (ip == NULL) { 784 /* Go through all effective netisr CPUs. */ 785 *cpuid = netisr_ncpus; 786 } else { 787 const struct udphdr *uh; 788 789 uh = (const struct udphdr *) 790 ((caddr_t)ip + (ip->ip_hl << 2)); 791 *cpuid = udp_addrcpu(faddr.s_addr, uh->uh_dport, 792 ip->ip_src.s_addr, uh->uh_sport); 793 } 794 } 795 796 *ip0 = ip; 797 return notify; 798 } 799 800 void 801 udp_ctlinput(netmsg_t msg) 802 { 803 struct sockaddr *sa = msg->ctlinput.nm_arg; 804 struct ip *ip = msg->ctlinput.nm_extra; 805 int cmd = msg->ctlinput.nm_cmd, cpuid; 806 inp_notify_t notify; 807 struct in_addr faddr; 808 809 ASSERT_NETISR_NCPUS(mycpuid); 810 811 notify = udp_get_inpnotify(cmd, sa, &ip, &cpuid); 812 if (notify == NULL) 813 goto done; 814 815 faddr = ((struct sockaddr_in *)sa)->sin_addr; 816 if (ip) { 817 const struct udphdr *uh; 818 struct inpcb *inp; 819 820 if (cpuid != mycpuid) 821 goto done; 822 823 uh = (const struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 824 inp = in_pcblookup_hash(&udbinfo[mycpuid], faddr, uh->uh_dport, 825 ip->ip_src, uh->uh_sport, 0, NULL); 826 if (inp != NULL && inp->inp_socket != NULL) 827 notify(inp, inetctlerrmap[cmd]); 828 } else if (msg->ctlinput.nm_direct) { 829 if (cpuid != netisr_ncpus && cpuid != mycpuid) 830 goto done; 831 832 in_pcbnotifyall(&udbinfo[mycpuid], faddr, inetctlerrmap[cmd], 833 notify); 834 } else { 835 struct netmsg_udp_notify *nm; 836 837 ASSERT_NETISR0; 838 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT); 839 netmsg_init(&nm->base, NULL, &netisr_afree_rport, 840 0, udp_notifyall_oncpu); 841 nm->nm_faddr = faddr; 842 nm->nm_arg = inetctlerrmap[cmd]; 843 nm->nm_notify = notify; 844 lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg); 845 } 846 done: 847 lwkt_replymsg(&msg->lmsg, 0); 848 } 849 850 SYSCTL_PROC(_net_inet_udp, UDPCTL_PCBLIST, pcblist, CTLFLAG_RD, udbinfo, 0, 851 in_pcblist_ncpus, "S,xinpcb", "List of active UDP sockets"); 852 853 static int 854 udp_getcred(SYSCTL_HANDLER_ARGS) 855 { 856 struct sockaddr_in addrs[2]; 857 struct ucred cred0, *cred = NULL; 858 struct inpcb *inp; 859 int error, cpu, origcpu; 860 861 error = priv_check(req->td, PRIV_ROOT); 862 if (error) 863 return (error); 864 error = SYSCTL_IN(req, addrs, sizeof addrs); 865 if (error) 866 return (error); 867 868 origcpu = mycpuid; 869 cpu = udp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port, 870 addrs[0].sin_addr.s_addr, addrs[0].sin_port); 871 872 lwkt_migratecpu(cpu); 873 874 inp = in_pcblookup_hash(&udbinfo[cpu], 875 addrs[1].sin_addr, addrs[1].sin_port, 876 addrs[0].sin_addr, addrs[0].sin_port, TRUE, NULL); 877 if (inp == NULL || inp->inp_socket == NULL) { 878 error = ENOENT; 879 } else if (inp->inp_socket->so_cred != NULL) { 880 cred0 = *(inp->inp_socket->so_cred); 881 cred = &cred0; 882 } 883 884 lwkt_migratecpu(origcpu); 885 886 if (error) 887 return error; 888 889 return SYSCTL_OUT(req, cred, sizeof(struct ucred)); 890 } 891 SYSCTL_PROC(_net_inet_udp, OID_AUTO, getcred, CTLTYPE_OPAQUE|CTLFLAG_RW, 892 0, 0, udp_getcred, "S,ucred", "Get the ucred of a UDP connection"); 893 894 static void 895 udp_send_redispatch(netmsg_t msg) 896 { 897 struct mbuf *m = msg->send.nm_m; 898 int pru_flags = msg->send.nm_flags; 899 struct inpcb *inp = msg->send.base.nm_so->so_pcb; 900 struct mbuf *m_opt = msg->send.nm_control; /* XXX save ipopt */ 901 int flags = msg->send.nm_priv; /* ip_output flags */ 902 int error; 903 904 logudp(redisp_ipout_beg, inp); 905 906 /* 907 * - Don't use inp route cache. It should only be used in the 908 * inp owner netisr. 909 * - Access to inp_moptions should be safe, since multicast UDP 910 * datagrams are redispatched to netisr0 and inp_moptions is 911 * changed only in netisr0. 912 */ 913 error = ip_output(m, m_opt, NULL, flags, inp->inp_moptions, inp); 914 if ((pru_flags & PRUS_NOREPLY) == 0) 915 lwkt_replymsg(&msg->send.base.lmsg, error); 916 917 if (m_opt != NULL) { 918 /* Free saved ip options, if any */ 919 m_freem(m_opt); 920 } 921 922 logudp(redisp_ipout_end, inp); 923 } 924 925 static void 926 udp_send(netmsg_t msg) 927 { 928 struct socket *so = msg->send.base.nm_so; 929 struct mbuf *m = msg->send.nm_m; 930 struct sockaddr *dstaddr = msg->send.nm_addr; 931 int pru_flags = msg->send.nm_flags; 932 struct inpcb *inp = so->so_pcb; 933 struct thread *td = msg->send.nm_td; 934 uint16_t hash; 935 int flags; 936 937 struct udpiphdr *ui; 938 int len = m->m_pkthdr.len; 939 struct sockaddr_in *sin; /* really is initialized before use */ 940 int error = 0, cpu; 941 942 KKASSERT(msg->send.nm_control == NULL); 943 944 logudp(send_beg, inp); 945 946 if (inp == NULL) { 947 error = EINVAL; 948 goto release; 949 } 950 951 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 952 error = EMSGSIZE; 953 goto release; 954 } 955 956 if (inp->inp_lport == 0) { /* unbound socket */ 957 boolean_t forwarded; 958 959 error = in_pcbbind(inp, NULL, td); 960 if (error) 961 goto release; 962 963 /* 964 * Need to call udp_send again, after this inpcb is 965 * inserted into wildcard hash table. 966 */ 967 msg->send.base.lmsg.ms_flags |= MSGF_UDP_SEND; 968 forwarded = udp_inswildcardhash(inp, &msg->send.base, 0); 969 if (forwarded) { 970 /* 971 * The message is further forwarded, so we are 972 * done here. 973 */ 974 logudp(send_inswildcard, inp); 975 return; 976 } 977 } 978 979 if (dstaddr != NULL) { /* destination address specified */ 980 if (inp->inp_faddr.s_addr != INADDR_ANY) { 981 /* already connected */ 982 error = EISCONN; 983 goto release; 984 } 985 sin = (struct sockaddr_in *)dstaddr; 986 if (!prison_remote_ip(td, (struct sockaddr *)&sin)) { 987 error = EAFNOSUPPORT; /* IPv6 only jail */ 988 goto release; 989 } 990 } else { 991 if (inp->inp_faddr.s_addr == INADDR_ANY) { 992 /* no destination specified and not already connected */ 993 error = ENOTCONN; 994 goto release; 995 } 996 sin = NULL; 997 } 998 999 /* 1000 * Calculate data length and get a mbuf 1001 * for UDP and IP headers. 1002 */ 1003 M_PREPEND(m, sizeof(struct udpiphdr), M_NOWAIT); 1004 if (m == NULL) { 1005 error = ENOBUFS; 1006 goto release; 1007 } 1008 1009 /* 1010 * Fill in mbuf with extended UDP header 1011 * and addresses and length put into network format. 1012 */ 1013 ui = mtod(m, struct udpiphdr *); 1014 bzero(ui->ui_x1, sizeof ui->ui_x1); /* XXX still needed? */ 1015 ui->ui_pr = IPPROTO_UDP; 1016 1017 /* 1018 * Set destination address. 1019 */ 1020 if (dstaddr != NULL) { /* use specified destination */ 1021 ui->ui_dst = sin->sin_addr; 1022 ui->ui_dport = sin->sin_port; 1023 } else { /* use connected destination */ 1024 ui->ui_dst = inp->inp_faddr; 1025 ui->ui_dport = inp->inp_fport; 1026 } 1027 1028 /* 1029 * Set source address. 1030 */ 1031 if (inp->inp_laddr.s_addr == INADDR_ANY || 1032 IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) { 1033 struct sockaddr_in *if_sin; 1034 1035 if (dstaddr == NULL) { 1036 /* 1037 * connect() had (or should have) failed because 1038 * the interface had no IP address, but the 1039 * application proceeded to call send() anyways. 1040 */ 1041 error = ENOTCONN; 1042 goto release; 1043 } 1044 1045 /* Look up outgoing interface. */ 1046 error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1); 1047 if (error) 1048 goto release; 1049 ui->ui_src = if_sin->sin_addr; /* use address of interface */ 1050 } else { 1051 ui->ui_src = inp->inp_laddr; /* use non-null bound address */ 1052 } 1053 ui->ui_sport = inp->inp_lport; 1054 KASSERT(inp->inp_lport != 0, ("inp lport should have been bound")); 1055 1056 /* 1057 * Release the original thread, since it is no longer used 1058 */ 1059 if (pru_flags & PRUS_HELDTD) { 1060 lwkt_rele(td); 1061 pru_flags &= ~PRUS_HELDTD; 1062 } 1063 /* 1064 * Free the dest address, since it is no longer needed 1065 */ 1066 if (pru_flags & PRUS_FREEADDR) { 1067 kfree(dstaddr, M_SONAME); 1068 pru_flags &= ~PRUS_FREEADDR; 1069 } 1070 1071 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1072 1073 /* 1074 * Set up checksum and output datagram. 1075 */ 1076 if (udpcksum) { 1077 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr, 1078 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP)); 1079 m->m_pkthdr.csum_flags = CSUM_UDP; 1080 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1081 m->m_pkthdr.csum_thlen = sizeof(struct udphdr); 1082 } else { 1083 ui->ui_sum = 0; 1084 } 1085 ((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len; 1086 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1087 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */ 1088 udp_stat.udps_opackets++; 1089 1090 flags = IP_DEBUGROUTE | 1091 (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)); 1092 if (pru_flags & PRUS_DONTROUTE) 1093 flags |= SO_DONTROUTE; 1094 1095 if (inp->inp_flags & INP_CONNECTED) { 1096 /* 1097 * For connected socket, this datagram has already 1098 * been in the correct netisr; no need to rehash. 1099 */ 1100 KASSERT(inp->inp_flags & INP_HASH, ("inpcb has no hash")); 1101 m_sethash(m, inp->inp_hashval); 1102 goto sendit; 1103 } 1104 1105 hash = udp_addrhash(ui->ui_dst.s_addr, ui->ui_dport, 1106 ui->ui_src.s_addr, ui->ui_sport); 1107 m_sethash(m, hash); 1108 1109 cpu = netisr_hashcpu(hash); 1110 if (cpu != mycpuid) { 1111 struct mbuf *m_opt = NULL; 1112 struct netmsg_pru_send *smsg; 1113 struct lwkt_port *port = netisr_cpuport(cpu); 1114 1115 /* 1116 * Not on the CPU that matches this UDP datagram hash; 1117 * redispatch to the correct CPU to do the ip_output(). 1118 */ 1119 if (inp->inp_options != NULL) { 1120 /* 1121 * If there are ip options, then save a copy, 1122 * since accessing inp_options on other CPUs' 1123 * is not safe. 1124 * 1125 * XXX optimize this? 1126 */ 1127 m_opt = m_copym(inp->inp_options, 0, M_COPYALL, 1128 M_WAITOK); 1129 } 1130 if ((pru_flags & PRUS_NOREPLY) == 0) { 1131 /* 1132 * Change some parts of the original netmsg and 1133 * forward it to the target netisr. 1134 * 1135 * NOTE: so_port MUST NOT be checked in the target 1136 * netisr. 1137 */ 1138 smsg = &msg->send; 1139 smsg->nm_priv = flags; /* ip_output flags */ 1140 smsg->nm_m = m; 1141 smsg->nm_control = m_opt; /* XXX save ipopt */ 1142 smsg->base.lmsg.ms_flags |= MSGF_IGNSOPORT; 1143 smsg->base.nm_dispatch = udp_send_redispatch; 1144 lwkt_forwardmsg(port, &smsg->base.lmsg); 1145 } else { 1146 /* 1147 * Recreate the netmsg, since the original mbuf 1148 * could have been changed. And send it to the 1149 * target netisr. 1150 * 1151 * NOTE: so_port MUST NOT be checked in the target 1152 * netisr. 1153 */ 1154 smsg = &m->m_hdr.mh_sndmsg; 1155 netmsg_init(&smsg->base, so, &netisr_apanic_rport, 1156 MSGF_IGNSOPORT, udp_send_redispatch); 1157 smsg->nm_priv = flags; /* ip_output flags */ 1158 smsg->nm_flags = pru_flags; 1159 smsg->nm_m = m; 1160 smsg->nm_control = m_opt; /* XXX save ipopt */ 1161 lwkt_sendmsg(port, &smsg->base.lmsg); 1162 } 1163 1164 /* This UDP datagram is redispatched; done */ 1165 logudp(send_redisp, inp); 1166 return; 1167 } 1168 1169 sendit: 1170 logudp(send_ipout, inp); 1171 error = ip_output(m, inp->inp_options, &inp->inp_route, flags, 1172 inp->inp_moptions, inp); 1173 m = NULL; 1174 1175 release: 1176 if (m != NULL) 1177 m_freem(m); 1178 1179 if (pru_flags & PRUS_HELDTD) 1180 lwkt_rele(td); 1181 if (pru_flags & PRUS_FREEADDR) 1182 kfree(dstaddr, M_SONAME); 1183 if ((pru_flags & PRUS_NOREPLY) == 0) 1184 lwkt_replymsg(&msg->send.base.lmsg, error); 1185 1186 logudp(send_end, inp); 1187 } 1188 1189 u_long udp_sendspace = 9216; /* really max datagram size */ 1190 /* 40 1K datagrams */ 1191 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 1192 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 1193 1194 u_long udp_recvspace = 40 * (1024 + 1195 #ifdef INET6 1196 sizeof(struct sockaddr_in6) 1197 #else 1198 sizeof(struct sockaddr_in) 1199 #endif 1200 ); 1201 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1202 &udp_recvspace, 0, "Maximum incoming UDP datagram size"); 1203 1204 /* 1205 * This should never happen, since UDP socket does not support 1206 * connection acception (SO_ACCEPTCONN, i.e. listen(2)). 1207 */ 1208 static void 1209 udp_abort(netmsg_t msg __unused) 1210 { 1211 panic("udp_abort is called"); 1212 } 1213 1214 static int 1215 udp_preattach(struct socket *so, int proto __unused, struct pru_attach_info *ai) 1216 { 1217 return soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit); 1218 } 1219 1220 static void 1221 udp_attach(netmsg_t msg) 1222 { 1223 struct socket *so = msg->attach.base.nm_so; 1224 struct pru_attach_info *ai = msg->attach.nm_ai; 1225 struct inpcb *inp; 1226 int error; 1227 1228 KASSERT(so->so_pcb == NULL, ("udp socket attached")); 1229 1230 if (ai != NULL) { 1231 error = udp_preattach(so, 0 /* don't care */, ai); 1232 if (error) 1233 goto out; 1234 } else { 1235 /* Post attach; do nothing */ 1236 } 1237 1238 error = in_pcballoc(so, &udbinfo[mycpuid]); 1239 if (error) 1240 goto out; 1241 1242 inp = so->so_pcb; 1243 inp->inp_flags |= INP_DIRECT_DETACH; 1244 inp->inp_ip_ttl = ip_defttl; 1245 error = 0; 1246 out: 1247 lwkt_replymsg(&msg->attach.base.lmsg, error); 1248 } 1249 1250 static void 1251 udp_inswildcard_replymsg(netmsg_t msg) 1252 { 1253 lwkt_msg_t lmsg = &msg->lmsg; 1254 1255 if (lmsg->ms_flags & MSGF_UDP_SEND) { 1256 udp_send(msg); 1257 /* msg is replied by udp_send() */ 1258 } else { 1259 lwkt_replymsg(lmsg, lmsg->ms_error); 1260 } 1261 } 1262 1263 static void 1264 udp_soreuseport_dispatch(netmsg_t msg) 1265 { 1266 /* This inpcb has already been in the wildcard hash. */ 1267 in_pcblink_flags(msg->base.nm_so->so_pcb, &udbinfo[mycpuid], 0); 1268 udp_inswildcard_replymsg(msg); 1269 } 1270 1271 static void 1272 udp_sosetport(struct lwkt_msg *msg, lwkt_port_t port) 1273 { 1274 sosetport(((struct netmsg_base *)msg)->nm_so, port); 1275 } 1276 1277 static boolean_t 1278 udp_inswildcardhash_oncpu(struct inpcb *inp, struct netmsg_base *msg) 1279 { 1280 int cpu; 1281 1282 KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid], 1283 ("not on owner cpu")); 1284 1285 in_pcbinswildcardhash(inp); 1286 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 1287 if (cpu == mycpuid) { 1288 /* 1289 * This inpcb has been inserted by the above 1290 * in_pcbinswildcardhash(). 1291 */ 1292 continue; 1293 } 1294 in_pcbinswildcardhash_oncpu(inp, &udbinfo[cpu]); 1295 } 1296 1297 /* NOTE: inp_lgrpindex is _not_ assigned in jail. */ 1298 if ((inp->inp_socket->so_options & SO_REUSEPORT) && 1299 inp->inp_lgrpindex >= 0) { 1300 /* 1301 * For SO_REUSEPORT socket, redistribute it based on its 1302 * local group index. 1303 */ 1304 cpu = inp->inp_lgrpindex % netisr_ncpus; 1305 if (cpu != mycpuid) { 1306 struct lwkt_port *port = netisr_cpuport(cpu); 1307 lwkt_msg_t lmsg = &msg->lmsg; 1308 1309 /* 1310 * We are moving the protocol processing port the 1311 * socket is on, we have to unlink here and re-link 1312 * on the target cpu (this inpcb is still left in 1313 * the wildcard hash). 1314 */ 1315 in_pcbunlink_flags(inp, &udbinfo[mycpuid], 0); 1316 msg->nm_dispatch = udp_soreuseport_dispatch; 1317 1318 /* 1319 * See the related comment in tcp_usrreq.c 1320 * tcp_connect() 1321 */ 1322 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1323 lwkt_forwardmsg(port, lmsg); 1324 return TRUE; /* forwarded */ 1325 } 1326 } 1327 return FALSE; 1328 } 1329 1330 static void 1331 udp_inswildcardhash_dispatch(netmsg_t msg) 1332 { 1333 struct inpcb *inp = msg->base.nm_so->so_pcb; 1334 boolean_t forwarded; 1335 1336 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1337 KASSERT(udp_lportcpu(inp->inp_lport) == mycpuid, ("not target cpu")); 1338 1339 in_pcblink(inp, &udbinfo[mycpuid]); 1340 1341 forwarded = udp_inswildcardhash_oncpu(inp, &msg->base); 1342 if (forwarded) { 1343 /* The message is further forwarded, so we are done here. */ 1344 return; 1345 } 1346 udp_inswildcard_replymsg(msg); 1347 } 1348 1349 static boolean_t 1350 udp_inswildcardhash(struct inpcb *inp, struct netmsg_base *msg, int error) 1351 { 1352 lwkt_msg_t lmsg = &msg->lmsg; 1353 int cpu; 1354 1355 ASSERT_INP_NOTINHASH(inp); 1356 1357 /* This inpcb could no longer be directly detached */ 1358 inp->inp_flags &= ~INP_DIRECT_DETACH; 1359 1360 /* 1361 * Always clear the route cache, so we don't need to 1362 * worry about any owner CPU changes later. 1363 */ 1364 in_pcbresetroute(inp); 1365 1366 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1367 cpu = udp_lportcpu(inp->inp_lport); 1368 1369 lmsg->ms_error = error; 1370 if (cpu != mycpuid) { 1371 struct lwkt_port *port = netisr_cpuport(cpu); 1372 1373 /* 1374 * We are moving the protocol processing port the socket 1375 * is on, we have to unlink here and re-link on the 1376 * target cpu. 1377 */ 1378 in_pcbunlink(inp, &udbinfo[mycpuid]); 1379 msg->nm_dispatch = udp_inswildcardhash_dispatch; 1380 1381 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1382 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1383 lwkt_forwardmsg(port, lmsg); 1384 return TRUE; /* forwarded */ 1385 } 1386 1387 return udp_inswildcardhash_oncpu(inp, msg); 1388 } 1389 1390 static void 1391 udp_bind(netmsg_t msg) 1392 { 1393 struct socket *so = msg->bind.base.nm_so; 1394 struct inpcb *inp; 1395 int error; 1396 1397 inp = so->so_pcb; 1398 if (inp) { 1399 struct sockaddr *nam = msg->bind.nm_nam; 1400 struct thread *td = msg->bind.nm_td; 1401 struct sockaddr_in *sin; 1402 lwkt_port_t port; 1403 int cpu; 1404 1405 /* 1406 * Check "already bound" here (in_pcbbind() does the same 1407 * check though), so we don't forward a connected/bound 1408 * socket randomly which would panic in the following 1409 * in_pcbunlink(). 1410 */ 1411 if (inp->inp_lport != 0 || 1412 inp->inp_laddr.s_addr != INADDR_ANY) { 1413 error = EINVAL; /* already bound */ 1414 goto done; 1415 } 1416 1417 if (nam->sa_len != sizeof(*sin)) { 1418 error = EINVAL; 1419 goto done; 1420 } 1421 sin = (struct sockaddr_in *)nam; 1422 1423 cpu = udp_lportcpu(sin->sin_port); 1424 port = netisr_cpuport(cpu); 1425 1426 /* 1427 * See the related comment in tcp_usrreq.c tcp_usr_bind(). 1428 * The exception is that we use local port based netisr 1429 * to serialize in_pcbbind(). 1430 */ 1431 if (&curthread->td_msgport != port) { 1432 lwkt_msg_t lmsg = &msg->bind.base.lmsg; 1433 1434 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0, 1435 ("already asked to relink")); 1436 1437 in_pcbunlink(so->so_pcb, &udbinfo[mycpuid]); 1438 msg->bind.nm_flags |= PRUB_RELINK; 1439 1440 /* 1441 * See the related comment in tcp_usrreq.c 1442 * tcp_connect(). 1443 */ 1444 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1445 lwkt_forwardmsg(port, lmsg); 1446 /* msg invalid now */ 1447 return; 1448 } 1449 KASSERT(so->so_port == port, ("so_port is not netisr%d", cpu)); 1450 1451 if (msg->bind.nm_flags & PRUB_RELINK) { 1452 msg->bind.nm_flags &= ~PRUB_RELINK; 1453 in_pcblink(so->so_pcb, &udbinfo[mycpuid]); 1454 } 1455 KASSERT(inp->inp_pcbinfo == &udbinfo[cpu], 1456 ("pcbinfo is not udbinfo%d", cpu)); 1457 1458 error = in_pcbbind(inp, nam, td); 1459 if (error == 0) { 1460 boolean_t forwarded; 1461 1462 if (sin->sin_addr.s_addr != INADDR_ANY) 1463 inp->inp_flags |= INP_WASBOUND_NOTANY; 1464 1465 forwarded = udp_inswildcardhash(inp, 1466 &msg->bind.base, 0); 1467 if (forwarded) { 1468 /* 1469 * The message is further forwarded, so 1470 * we are done here. 1471 */ 1472 return; 1473 } 1474 } 1475 } else { 1476 error = EINVAL; 1477 } 1478 done: 1479 lwkt_replymsg(&msg->bind.base.lmsg, error); 1480 } 1481 1482 static int 1483 udp_preconnect(struct socket *so, const struct sockaddr *nam __unused, 1484 struct thread *td __unused) 1485 { 1486 sosetstate(so, SS_ISCONNECTED); /* XXX */ 1487 return 0; 1488 } 1489 1490 static void 1491 udp_connect(netmsg_t msg) 1492 { 1493 struct socket *so = msg->connect.base.nm_so; 1494 struct sockaddr *nam = msg->connect.nm_nam; 1495 struct thread *td = msg->connect.nm_td; 1496 struct inpcb *inp; 1497 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1498 struct sockaddr_in *if_sin; 1499 struct lwkt_port *port; 1500 uint16_t hash; 1501 int error; 1502 1503 KKASSERT(msg->connect.nm_m == NULL); 1504 1505 inp = so->so_pcb; 1506 if (inp == NULL) { 1507 error = EINVAL; 1508 goto out; 1509 } 1510 1511 if (msg->connect.nm_flags & PRUC_RECONNECT) { 1512 msg->connect.nm_flags &= ~PRUC_RECONNECT; 1513 in_pcblink(inp, &udbinfo[mycpuid]); 1514 } 1515 1516 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1517 error = EISCONN; 1518 goto out; 1519 } 1520 error = 0; 1521 1522 /* 1523 * Bind if we have to 1524 */ 1525 if (inp->inp_lport == 0) { 1526 error = in_pcbbind(inp, NULL, td); 1527 if (error) 1528 goto out; 1529 } 1530 1531 /* 1532 * Calculate the correct protocol processing thread. The connect 1533 * operation must run there. 1534 */ 1535 error = in_pcbladdr(inp, nam, &if_sin, td); 1536 if (error) 1537 goto out; 1538 if (!prison_remote_ip(td, nam)) { 1539 error = EAFNOSUPPORT; /* IPv6 only jail */ 1540 goto out; 1541 } 1542 1543 hash = udp_addrhash(sin->sin_addr.s_addr, sin->sin_port, 1544 inp->inp_laddr.s_addr != INADDR_ANY ? 1545 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, inp->inp_lport); 1546 port = netisr_hashport(hash); 1547 if (port != &curthread->td_msgport) { 1548 lwkt_msg_t lmsg = &msg->connect.base.lmsg; 1549 int nm_flags = PRUC_RECONNECT; 1550 1551 /* 1552 * in_pcbladdr() may have allocated a route entry for us 1553 * on the current CPU, but we need a route entry on the 1554 * inpcb's owner CPU, so free it here. 1555 */ 1556 in_pcbresetroute(inp); 1557 1558 if (inp->inp_flags & INP_WILDCARD) { 1559 /* 1560 * Remove this inpcb from the wildcard hash before 1561 * the socket's msgport changes. 1562 */ 1563 udp_remwildcardhash(inp); 1564 } 1565 1566 if (so->so_orig_port == NULL) { 1567 /* 1568 * First time change protocol processing port. 1569 * Save the current port for synchronization upon 1570 * udp_detach. 1571 */ 1572 so->so_orig_port = &curthread->td_msgport; 1573 } else { 1574 /* 1575 * We have changed protocol processing port more 1576 * than once. We could not do direct detach 1577 * anymore, because we lose the track of the 1578 * original protocol processing ports to perform 1579 * synchronization upon udp_detach. This should 1580 * be rare though. 1581 */ 1582 inp->inp_flags &= ~INP_DIRECT_DETACH; 1583 } 1584 1585 /* 1586 * We are moving the protocol processing port the socket 1587 * is on, we have to unlink here and re-link on the 1588 * target cpu. 1589 */ 1590 in_pcbunlink(inp, &udbinfo[mycpuid]); 1591 msg->connect.nm_flags |= nm_flags; 1592 1593 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1594 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1595 lwkt_forwardmsg(port, lmsg); 1596 /* msg invalid now */ 1597 return; 1598 } 1599 error = udp_connect_oncpu(inp, sin, if_sin, hash); 1600 out: 1601 if (msg->connect.nm_flags & PRUC_HELDTD) 1602 lwkt_rele(td); 1603 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) { 1604 if (inp->inp_lport == 0) { 1605 /* 1606 * As long as we have the local port, it is fine 1607 * for connect to fail, e.g. disconnect. 1608 */ 1609 so->so_error = error; 1610 } 1611 soclrstate(so, SS_ISCONNECTED); 1612 /* 1613 * Wake up callers blocked on this socket to make sure 1614 * that they can see this error. 1615 * 1616 * NOTE: 1617 * sodisconnected() can't be used here, which bricks 1618 * sending and receiving. 1619 */ 1620 wakeup(&so->so_timeo); 1621 sowwakeup(so); 1622 sorwakeup(so); 1623 } 1624 if (error && inp != NULL && inp->inp_lport != 0 && 1625 (inp->inp_flags & INP_WILDCARD) == 0) { 1626 boolean_t forwarded; 1627 1628 /* Connect failed; put it to wildcard hash. */ 1629 forwarded = udp_inswildcardhash(inp, &msg->connect.base, 1630 error); 1631 if (forwarded) { 1632 /* 1633 * The message is further forwarded, so we are done 1634 * here. 1635 */ 1636 return; 1637 } 1638 } 1639 lwkt_replymsg(&msg->connect.base.lmsg, error); 1640 } 1641 1642 static void 1643 udp_remwildcardhash(struct inpcb *inp) 1644 { 1645 int cpu; 1646 1647 KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid], 1648 ("not on owner cpu")); 1649 1650 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 1651 if (cpu == mycpuid) { 1652 /* 1653 * This inpcb will be removed by the later 1654 * in_pcbremwildcardhash(). 1655 */ 1656 continue; 1657 } 1658 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpu]); 1659 } 1660 in_pcbremwildcardhash(inp); 1661 } 1662 1663 static int 1664 udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin, 1665 struct sockaddr_in *if_sin, uint16_t hash) 1666 { 1667 struct socket *so = inp->inp_socket; 1668 struct inpcb *oinp; 1669 1670 oinp = in_pcblookup_hash(inp->inp_pcbinfo, 1671 sin->sin_addr, sin->sin_port, 1672 inp->inp_laddr.s_addr != INADDR_ANY ? 1673 inp->inp_laddr : if_sin->sin_addr, inp->inp_lport, FALSE, NULL); 1674 if (oinp != NULL) 1675 return EADDRINUSE; 1676 1677 /* 1678 * No more errors can occur, finish adjusting the socket 1679 * and change the processing port to reflect the connected 1680 * socket. Once set we can no longer safely mess with the 1681 * socket. 1682 */ 1683 1684 if (inp->inp_flags & INP_WILDCARD) 1685 udp_remwildcardhash(inp); 1686 1687 if (inp->inp_laddr.s_addr == INADDR_ANY) 1688 inp->inp_laddr = if_sin->sin_addr; 1689 inp->inp_faddr = sin->sin_addr; 1690 inp->inp_fport = sin->sin_port; 1691 in_pcbinsconnhash(inp); 1692 1693 inp->inp_flags |= INP_HASH; 1694 inp->inp_hashval = hash; 1695 1696 soisconnected(so); 1697 1698 return 0; 1699 } 1700 1701 static void 1702 udp_detach2(struct socket *so) 1703 { 1704 in_pcbdetach(so->so_pcb); 1705 sodiscard(so); 1706 sofree(so); 1707 } 1708 1709 static void 1710 udp_detach_final_dispatch(netmsg_t msg) 1711 { 1712 udp_detach2(msg->base.nm_so); 1713 } 1714 1715 static void 1716 udp_detach_oncpu_dispatch(netmsg_t msg) 1717 { 1718 struct netmsg_base *clomsg = &msg->base; 1719 struct socket *so = clomsg->nm_so; 1720 struct inpcb *inp = so->so_pcb; 1721 struct thread *td = curthread; 1722 int nextcpu, cpuid = mycpuid; 1723 1724 KASSERT(td->td_type == TD_TYPE_NETISR, ("not in netisr")); 1725 1726 if (inp->inp_flags & INP_WILDCARD) { 1727 /* 1728 * This inp will be removed on the inp's 1729 * owner CPU later, so don't do it now. 1730 */ 1731 if (&td->td_msgport != so->so_port) 1732 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpuid]); 1733 } 1734 1735 if (cpuid == 0) { 1736 /* 1737 * Free and clear multicast socket option, 1738 * which is only accessed in netisr0. 1739 */ 1740 ip_freemoptions(inp->inp_moptions); 1741 inp->inp_moptions = NULL; 1742 } 1743 1744 nextcpu = cpuid + 1; 1745 if (nextcpu < netisr_ncpus) { 1746 lwkt_forwardmsg(netisr_cpuport(nextcpu), &clomsg->lmsg); 1747 } else { 1748 /* 1749 * No one could see this inpcb now; destroy this 1750 * inpcb in its owner netisr. 1751 */ 1752 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1753 udp_detach_final_dispatch); 1754 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1755 } 1756 } 1757 1758 static void 1759 udp_detach_syncorig_dispatch(netmsg_t msg) 1760 { 1761 struct netmsg_base *clomsg = &msg->base; 1762 struct socket *so = clomsg->nm_so; 1763 1764 /* 1765 * Original protocol processing port is synchronized; 1766 * destroy this inpcb in its owner netisr. 1767 */ 1768 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1769 udp_detach_final_dispatch); 1770 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1771 } 1772 1773 static void 1774 udp_detach(netmsg_t msg) 1775 { 1776 struct socket *so = msg->detach.base.nm_so; 1777 struct netmsg_base *clomsg; 1778 struct inpcb *inp; 1779 1780 inp = so->so_pcb; 1781 if (inp == NULL) { 1782 lwkt_replymsg(&msg->detach.base.lmsg, EINVAL); 1783 return; 1784 } 1785 1786 /* 1787 * Reply EJUSTRETURN ASAP, we will call sodiscard() and 1788 * sofree() later. 1789 */ 1790 lwkt_replymsg(&msg->detach.base.lmsg, EJUSTRETURN); 1791 1792 if (netisr_ncpus == 1) { 1793 /* Only one CPU, detach the inpcb directly. */ 1794 udp_detach2(so); 1795 return; 1796 } 1797 1798 /* 1799 * Remove this inpcb from the inpcb list first, so that 1800 * no one could find this inpcb from the inpcb list. 1801 */ 1802 in_pcbofflist(inp); 1803 1804 /* 1805 * Remove this inpcb from the local port hash directly 1806 * here, so that its bound local port could be recycled 1807 * timely. 1808 */ 1809 in_pcbremporthash(inp); 1810 1811 if (inp->inp_flags & INP_DIRECT_DETACH) { 1812 /* 1813 * Direct detaching is allowed 1814 */ 1815 KASSERT((inp->inp_flags & INP_WILDCARD) == 0, 1816 ("in the wildcardhash")); 1817 KASSERT(inp->inp_moptions == NULL, ("has mcast options")); 1818 if (so->so_orig_port == NULL) { 1819 udp_detach2(so); 1820 } else { 1821 /* 1822 * Protocol processing port changed once, so 1823 * we need to make sure that there are nothing 1824 * left on the original protocol processing 1825 * port before we destroy this socket and inpcb. 1826 * This is more lightweight than going through 1827 * all UDP processing netisrs. 1828 */ 1829 clomsg = &so->so_clomsg; 1830 netmsg_init(clomsg, so, &netisr_apanic_rport, 1831 MSGF_IGNSOPORT, udp_detach_syncorig_dispatch); 1832 lwkt_sendmsg(so->so_orig_port, &clomsg->lmsg); 1833 } 1834 return; 1835 } 1836 1837 /* 1838 * Go through netisrs which process UDP to make sure 1839 * no one could find this inpcb anymore. 1840 */ 1841 clomsg = &so->so_clomsg; 1842 netmsg_init(clomsg, so, &netisr_apanic_rport, MSGF_IGNSOPORT, 1843 udp_detach_oncpu_dispatch); 1844 lwkt_sendmsg(netisr_cpuport(0), &clomsg->lmsg); 1845 } 1846 1847 static void 1848 udp_disconnect(netmsg_t msg) 1849 { 1850 struct socket *so = msg->disconnect.base.nm_so; 1851 struct inpcb *inp; 1852 boolean_t forwarded; 1853 int error = 0; 1854 1855 inp = so->so_pcb; 1856 if (inp == NULL) { 1857 error = EINVAL; 1858 goto out; 1859 } 1860 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1861 error = ENOTCONN; 1862 goto out; 1863 } 1864 1865 soclrstate(so, SS_ISCONNECTED); /* XXX */ 1866 1867 in_pcbdisconnect(inp); 1868 inp->inp_flags &= ~INP_HASH; 1869 1870 /* 1871 * Follow traditional BSD behavior and retain the local port 1872 * binding. But, fix the old misbehavior of overwriting any 1873 * previously bound local address. 1874 */ 1875 if (!(inp->inp_flags & INP_WASBOUND_NOTANY)) 1876 inp->inp_laddr.s_addr = INADDR_ANY; 1877 1878 if (so->so_state & SS_ISCLOSING) { 1879 /* 1880 * If this socket is being closed, there is no need 1881 * to put this socket back into wildcard hash table. 1882 */ 1883 error = 0; 1884 goto out; 1885 } 1886 1887 forwarded = udp_inswildcardhash(inp, &msg->disconnect.base, 0); 1888 if (forwarded) { 1889 /* 1890 * The message is further forwarded, so we are done 1891 * here. 1892 */ 1893 return; 1894 } 1895 out: 1896 lwkt_replymsg(&msg->disconnect.base.lmsg, error); 1897 } 1898 1899 void 1900 udp_shutdown(netmsg_t msg) 1901 { 1902 struct socket *so = msg->shutdown.base.nm_so; 1903 struct inpcb *inp; 1904 int error; 1905 1906 inp = so->so_pcb; 1907 if (inp) { 1908 socantsendmore(so); 1909 error = 0; 1910 } else { 1911 error = EINVAL; 1912 } 1913 lwkt_replymsg(&msg->shutdown.base.lmsg, error); 1914 } 1915 1916 struct pr_usrreqs udp_usrreqs = { 1917 .pru_abort = udp_abort, 1918 .pru_accept = pr_generic_notsupp, 1919 .pru_attach = udp_attach, 1920 .pru_bind = udp_bind, 1921 .pru_connect = udp_connect, 1922 .pru_connect2 = pr_generic_notsupp, 1923 .pru_control = in_control_dispatch, 1924 .pru_detach = udp_detach, 1925 .pru_disconnect = udp_disconnect, 1926 .pru_listen = pr_generic_notsupp, 1927 .pru_peeraddr = in_setpeeraddr_dispatch, 1928 .pru_rcvd = pr_generic_notsupp, 1929 .pru_rcvoob = pr_generic_notsupp, 1930 .pru_send = udp_send, 1931 .pru_sense = pru_sense_null, 1932 .pru_shutdown = udp_shutdown, 1933 .pru_sockaddr = in_setsockaddr_dispatch, 1934 .pru_sosend = sosendudp, 1935 .pru_soreceive = soreceive, 1936 .pru_preconnect = udp_preconnect, 1937 .pru_preattach = udp_preattach 1938 }; 1939