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) & ncpus2_mask); 202 } 203 204 void 205 udp_init(void) 206 { 207 struct inpcbportinfo *portinfo; 208 int cpu; 209 210 portinfo = kmalloc_cachealign(sizeof(*portinfo) * ncpus2, M_PCB, 211 M_WAITOK); 212 213 for (cpu = 0; cpu < ncpus2; 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 uicb->portinfo = portinfo; 226 uicb->portinfo_mask = ncpus2_mask; 227 228 uicb->wildcardhashbase = hashinit(UDBHASHSIZE, M_PCB, 229 &uicb->wildcardhashmask); 230 uicb->localgrphashbase = hashinit(UDBHASHSIZE, M_PCB, 231 &uicb->localgrphashmask); 232 233 uicb->ipi_size = sizeof(struct inpcb); 234 } 235 236 /* 237 * Initialize UDP statistics counters for each CPU. 238 */ 239 for (cpu = 0; cpu < ncpus; ++cpu) 240 bzero(&udpstat_percpu[cpu], sizeof(struct udpstat)); 241 } 242 243 static int 244 sysctl_udpstat(SYSCTL_HANDLER_ARGS) 245 { 246 int cpu, error = 0; 247 248 for (cpu = 0; cpu < ncpus; ++cpu) { 249 if ((error = SYSCTL_OUT(req, &udpstat_percpu[cpu], 250 sizeof(struct udpstat)))) 251 break; 252 if ((error = SYSCTL_IN(req, &udpstat_percpu[cpu], 253 sizeof(struct udpstat)))) 254 break; 255 } 256 257 return (error); 258 } 259 SYSCTL_PROC(_net_inet_udp, UDPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW), 260 0, 0, sysctl_udpstat, "S,udpstat", "UDP statistics"); 261 262 void 263 udp_ctloutput(netmsg_t msg) 264 { 265 struct socket *so = msg->base.nm_so; 266 struct sockopt *sopt = msg->ctloutput.nm_sopt; 267 struct inpcb *inp = so->so_pcb; 268 269 if (inp == NULL) { 270 lwkt_replymsg(&msg->lmsg, EINVAL); 271 return; 272 } 273 274 if (sopt->sopt_level == IPPROTO_IP && sopt->sopt_dir == SOPT_SET) { 275 switch (sopt->sopt_name) { 276 case IP_MULTICAST_IF: 277 case IP_MULTICAST_VIF: 278 case IP_MULTICAST_TTL: 279 case IP_MULTICAST_LOOP: 280 case IP_ADD_MEMBERSHIP: 281 case IP_DROP_MEMBERSHIP: 282 /* 283 * This pr_ctloutput msg will be forwarded 284 * to netisr0 to run; we can't do direct 285 * detaching anymore. 286 * 287 * NOTE: 288 * Don't optimize for the sockets whose 289 * current so_port is netisr0's msgport. 290 * These sockets could be connect(2)'ed 291 * later and the so_port will be changed. 292 */ 293 inp->inp_flags &= ~INP_DIRECT_DETACH; 294 break; 295 } 296 } 297 return ip_ctloutput(msg); 298 } 299 300 /* 301 * Check multicast packets to make sure they are only sent to sockets with 302 * multicast memberships for the packet's destination address and arrival 303 * interface. Multicast packets to multicast-unaware sockets are also 304 * disallowed. 305 * 306 * Returns 0 if the packet is acceptable, -1 if it is not. 307 */ 308 static __inline int 309 check_multicast_membership(const struct ip *ip, const struct inpcb *inp, 310 const struct mbuf *m) 311 { 312 const struct ip_moptions *mopt; 313 int mshipno; 314 315 if (strict_mcast_mship == 0 || 316 !IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) { 317 return (0); 318 } 319 320 ASSERT_IN_NETISR(0); 321 322 mopt = inp->inp_moptions; 323 if (mopt == NULL) 324 return (-1); 325 for (mshipno = 0; mshipno < mopt->imo_num_memberships; ++mshipno) { 326 const struct in_multi *maddr = mopt->imo_membership[mshipno]; 327 328 if (ip->ip_dst.s_addr == maddr->inm_addr.s_addr && 329 m->m_pkthdr.rcvif == maddr->inm_ifp) { 330 return (0); 331 } 332 } 333 return (-1); 334 } 335 336 struct udp_mcast_arg { 337 struct inpcb *inp; 338 struct inpcb *last; 339 struct ip *ip; 340 struct mbuf *m; 341 int iphlen; 342 struct sockaddr_in *udp_in; 343 }; 344 345 static int 346 udp_mcast_input(struct udp_mcast_arg *arg) 347 { 348 struct inpcb *inp = arg->inp; 349 struct inpcb *last = arg->last; 350 struct ip *ip = arg->ip; 351 struct mbuf *m = arg->m; 352 353 if (check_multicast_membership(ip, inp, m) < 0) 354 return ERESTART; /* caller continue */ 355 356 if (last != NULL) { 357 struct mbuf *n; 358 359 #ifdef IPSEC 360 /* check AH/ESP integrity. */ 361 if (ipsec4_in_reject_so(m, last->inp_socket)) 362 ipsecstat.in_polvio++; 363 /* do not inject data to pcb */ 364 else 365 #endif /*IPSEC*/ 366 #ifdef FAST_IPSEC 367 /* check AH/ESP integrity. */ 368 if (ipsec4_in_reject(m, last)) 369 ; 370 else 371 #endif /*FAST_IPSEC*/ 372 if ((n = m_copypacket(m, M_NOWAIT)) != NULL) 373 udp_append(last, ip, n, 374 arg->iphlen + sizeof(struct udphdr), 375 arg->udp_in); 376 } 377 arg->last = last = inp; 378 379 /* 380 * Don't look for additional matches if this one does 381 * not have either the SO_REUSEPORT or SO_REUSEADDR 382 * socket options set. This heuristic avoids searching 383 * through all pcbs in the common case of a non-shared 384 * port. It * assumes that an application will never 385 * clear these options after setting them. 386 */ 387 if (!(last->inp_socket->so_options & 388 (SO_REUSEPORT | SO_REUSEADDR))) 389 return EJUSTRETURN; /* caller stop */ 390 return 0; 391 } 392 393 int 394 udp_input(struct mbuf **mp, int *offp, int proto) 395 { 396 struct sockaddr_in udp_in = { sizeof udp_in, AF_INET }; 397 int iphlen; 398 struct ip *ip; 399 struct udphdr *uh; 400 struct inpcb *inp; 401 struct mbuf *m; 402 struct mbuf *opts = NULL; 403 int len, off; 404 struct ip save_ip; 405 struct inpcbinfo *pcbinfo = &udbinfo[mycpuid]; 406 407 off = *offp; 408 m = *mp; 409 *mp = NULL; 410 411 iphlen = off; 412 udp_stat.udps_ipackets++; 413 414 /* 415 * Strip IP options, if any; should skip this, 416 * make available to user, and use on returned packets, 417 * but we don't yet have a way to check the checksum 418 * with options still present. 419 */ 420 if (iphlen > sizeof(struct ip)) { 421 ip_stripoptions(m); 422 iphlen = sizeof(struct ip); 423 } 424 425 /* 426 * IP and UDP headers are together in first mbuf. 427 * Already checked and pulled up in ip_demux(). 428 */ 429 KASSERT(m->m_len >= iphlen + sizeof(struct udphdr), 430 ("UDP header not in one mbuf")); 431 432 ip = mtod(m, struct ip *); 433 uh = (struct udphdr *)((caddr_t)ip + iphlen); 434 435 /* destination port of 0 is illegal, based on RFC768. */ 436 if (uh->uh_dport == 0) 437 goto bad; 438 439 /* 440 * Make mbuf data length reflect UDP length. 441 * If not enough data to reflect UDP length, drop. 442 */ 443 len = ntohs((u_short)uh->uh_ulen); 444 if (ip->ip_len != len) { 445 if (len > ip->ip_len || len < sizeof(struct udphdr)) { 446 udp_stat.udps_badlen++; 447 goto bad; 448 } 449 m_adj(m, len - ip->ip_len); 450 /* ip->ip_len = len; */ 451 } 452 /* 453 * Save a copy of the IP header in case we want restore it 454 * for sending an ICMP error message in response. 455 */ 456 save_ip = *ip; 457 458 /* 459 * Checksum extended UDP header and data. 460 */ 461 if (uh->uh_sum) { 462 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 463 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 464 uh->uh_sum = m->m_pkthdr.csum_data; 465 else 466 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, 467 ip->ip_dst.s_addr, htonl((u_short)len + 468 m->m_pkthdr.csum_data + IPPROTO_UDP)); 469 uh->uh_sum ^= 0xffff; 470 } else { 471 char b[9]; 472 473 bcopy(((struct ipovly *)ip)->ih_x1, b, 9); 474 bzero(((struct ipovly *)ip)->ih_x1, 9); 475 ((struct ipovly *)ip)->ih_len = uh->uh_ulen; 476 uh->uh_sum = in_cksum(m, len + sizeof(struct ip)); 477 bcopy(b, ((struct ipovly *)ip)->ih_x1, 9); 478 } 479 if (uh->uh_sum) { 480 udp_stat.udps_badsum++; 481 m_freem(m); 482 return(IPPROTO_DONE); 483 } 484 } else 485 udp_stat.udps_nosum++; 486 487 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 488 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 489 struct inpcbhead *connhead; 490 struct inpcontainer *ic, *ic_marker; 491 struct inpcontainerhead *ichead; 492 struct udp_mcast_arg arg; 493 struct inpcb *last; 494 int error; 495 496 /* 497 * Deliver a multicast or broadcast datagram to *all* sockets 498 * for which the local and remote addresses and ports match 499 * those of the incoming datagram. This allows more than 500 * one process to receive multi/broadcasts on the same port. 501 * (This really ought to be done for unicast datagrams as 502 * well, but that would cause problems with existing 503 * applications that open both address-specific sockets and 504 * a wildcard socket listening to the same port -- they would 505 * end up receiving duplicates of every unicast datagram. 506 * Those applications open the multiple sockets to overcome an 507 * inadequacy of the UDP socket interface, but for backwards 508 * compatibility we avoid the problem here rather than 509 * fixing the interface. Maybe 4.5BSD will remedy this?) 510 */ 511 512 /* 513 * Construct sockaddr format source address. 514 */ 515 udp_in.sin_port = uh->uh_sport; 516 udp_in.sin_addr = ip->ip_src; 517 arg.udp_in = &udp_in; 518 /* 519 * Locate pcb(s) for datagram. 520 * (Algorithm copied from raw_intr().) 521 */ 522 last = NULL; 523 arg.iphlen = iphlen; 524 525 connhead = &pcbinfo->hashbase[ 526 INP_PCBCONNHASH(ip->ip_src.s_addr, uh->uh_sport, 527 ip->ip_dst.s_addr, uh->uh_dport, pcbinfo->hashmask)]; 528 LIST_FOREACH(inp, connhead, inp_hash) { 529 #ifdef INET6 530 if (!INP_ISIPV4(inp)) 531 continue; 532 #endif 533 if (!in_hosteq(inp->inp_faddr, ip->ip_src) || 534 !in_hosteq(inp->inp_laddr, ip->ip_dst) || 535 inp->inp_fport != uh->uh_sport || 536 inp->inp_lport != uh->uh_dport) 537 continue; 538 539 arg.inp = inp; 540 arg.last = last; 541 arg.ip = ip; 542 arg.m = m; 543 544 error = udp_mcast_input(&arg); 545 if (error == ERESTART) 546 continue; 547 last = arg.last; 548 549 if (error == EJUSTRETURN) 550 goto done; 551 } 552 553 ichead = &pcbinfo->wildcardhashbase[ 554 INP_PCBWILDCARDHASH(uh->uh_dport, 555 pcbinfo->wildcardhashmask)]; 556 ic_marker = in_pcbcontainer_marker(mycpuid); 557 558 GET_PCBINFO_TOKEN(pcbinfo); 559 LIST_INSERT_HEAD(ichead, ic_marker, ic_list); 560 while ((ic = LIST_NEXT(ic_marker, ic_list)) != NULL) { 561 LIST_REMOVE(ic_marker, ic_list); 562 LIST_INSERT_AFTER(ic, ic_marker, ic_list); 563 564 inp = ic->ic_inp; 565 if (inp->inp_flags & INP_PLACEMARKER) 566 continue; 567 #ifdef INET6 568 if (!INP_ISIPV4(inp)) 569 continue; 570 #endif 571 if (inp->inp_lport != uh->uh_dport) 572 continue; 573 if (inp->inp_laddr.s_addr != INADDR_ANY && 574 inp->inp_laddr.s_addr != ip->ip_dst.s_addr) 575 continue; 576 577 arg.inp = inp; 578 arg.last = last; 579 arg.ip = ip; 580 arg.m = m; 581 582 error = udp_mcast_input(&arg); 583 if (error == ERESTART) 584 continue; 585 last = arg.last; 586 587 if (error == EJUSTRETURN) 588 break; 589 } 590 LIST_REMOVE(ic_marker, ic_list); 591 REL_PCBINFO_TOKEN(pcbinfo); 592 done: 593 if (last == NULL) { 594 /* 595 * No matching pcb found; discard datagram. 596 * (No need to send an ICMP Port Unreachable 597 * for a broadcast or multicast datgram.) 598 */ 599 udp_stat.udps_noportbcast++; 600 goto bad; 601 } 602 #ifdef IPSEC 603 /* check AH/ESP integrity. */ 604 if (ipsec4_in_reject_so(m, last->inp_socket)) { 605 ipsecstat.in_polvio++; 606 goto bad; 607 } 608 #endif /*IPSEC*/ 609 #ifdef FAST_IPSEC 610 /* check AH/ESP integrity. */ 611 if (ipsec4_in_reject(m, last)) 612 goto bad; 613 #endif /*FAST_IPSEC*/ 614 udp_append(last, ip, m, iphlen + sizeof(struct udphdr), 615 &udp_in); 616 return(IPPROTO_DONE); 617 } 618 /* 619 * Locate pcb for datagram. 620 */ 621 inp = in_pcblookup_pkthash(pcbinfo, ip->ip_src, uh->uh_sport, 622 ip->ip_dst, uh->uh_dport, TRUE, m->m_pkthdr.rcvif, 623 udp_reuseport_ext ? m : NULL); 624 if (inp == NULL) { 625 if (log_in_vain) { 626 char buf[sizeof "aaa.bbb.ccc.ddd"]; 627 628 strcpy(buf, inet_ntoa(ip->ip_dst)); 629 log(LOG_INFO, 630 "Connection attempt to UDP %s:%d from %s:%d\n", 631 buf, ntohs(uh->uh_dport), inet_ntoa(ip->ip_src), 632 ntohs(uh->uh_sport)); 633 } 634 udp_stat.udps_noport++; 635 if (m->m_flags & (M_BCAST | M_MCAST)) { 636 udp_stat.udps_noportbcast++; 637 goto bad; 638 } 639 if (blackhole) 640 goto bad; 641 #ifdef ICMP_BANDLIM 642 if (badport_bandlim(BANDLIM_ICMP_UNREACH) < 0) 643 goto bad; 644 #endif 645 *ip = save_ip; 646 ip->ip_len += iphlen; 647 icmp_error(m, ICMP_UNREACH, ICMP_UNREACH_PORT, 0, 0); 648 return(IPPROTO_DONE); 649 } 650 KASSERT(INP_ISIPV4(inp), ("not inet inpcb")); 651 #ifdef IPSEC 652 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 653 ipsecstat.in_polvio++; 654 goto bad; 655 } 656 #endif /*IPSEC*/ 657 #ifdef FAST_IPSEC 658 if (ipsec4_in_reject(m, inp)) 659 goto bad; 660 #endif /*FAST_IPSEC*/ 661 /* 662 * Check the minimum TTL for socket. 663 */ 664 if (ip->ip_ttl < inp->inp_ip_minttl) 665 goto bad; 666 667 /* 668 * Construct sockaddr format source address. 669 * Stuff source address and datagram in user buffer. 670 */ 671 udp_in.sin_port = uh->uh_sport; 672 udp_in.sin_addr = ip->ip_src; 673 if ((inp->inp_flags & INP_CONTROLOPTS) || 674 (inp->inp_socket->so_options & SO_TIMESTAMP)) 675 ip_savecontrol(inp, &opts, ip, m); 676 m_adj(m, iphlen + sizeof(struct udphdr)); 677 678 lwkt_gettoken(&inp->inp_socket->so_rcv.ssb_token); 679 if (ssb_appendaddr(&inp->inp_socket->so_rcv, 680 (struct sockaddr *)&udp_in, m, opts) == 0) { 681 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token); 682 udp_stat.udps_fullsock++; 683 goto bad; 684 } 685 lwkt_reltoken(&inp->inp_socket->so_rcv.ssb_token); 686 sorwakeup(inp->inp_socket); 687 return(IPPROTO_DONE); 688 bad: 689 m_freem(m); 690 if (opts) 691 m_freem(opts); 692 return(IPPROTO_DONE); 693 } 694 695 /* 696 * subroutine of udp_input(), mainly for source code readability. 697 * caller must properly init udp_ip6 and udp_in6 beforehand. 698 */ 699 static void 700 udp_append(struct inpcb *last, struct ip *ip, struct mbuf *n, int off, 701 struct sockaddr_in *udp_in) 702 { 703 struct mbuf *opts = NULL; 704 int ret; 705 706 KASSERT(INP_ISIPV4(last), ("not inet inpcb")); 707 708 if (last->inp_flags & INP_CONTROLOPTS || 709 last->inp_socket->so_options & SO_TIMESTAMP) 710 ip_savecontrol(last, &opts, ip, n); 711 m_adj(n, off); 712 713 lwkt_gettoken(&last->inp_socket->so_rcv.ssb_token); 714 ret = ssb_appendaddr(&last->inp_socket->so_rcv, 715 (struct sockaddr *)udp_in, n, opts); 716 lwkt_reltoken(&last->inp_socket->so_rcv.ssb_token); 717 if (ret == 0) { 718 m_freem(n); 719 if (opts) 720 m_freem(opts); 721 udp_stat.udps_fullsock++; 722 } else { 723 sorwakeup(last->inp_socket); 724 } 725 } 726 727 /* 728 * Notify a udp user of an asynchronous error; 729 * just wake up so that he can collect error status. 730 */ 731 void 732 udp_notify(struct inpcb *inp, int error) 733 { 734 inp->inp_socket->so_error = error; 735 sorwakeup(inp->inp_socket); 736 sowwakeup(inp->inp_socket); 737 } 738 739 struct netmsg_udp_notify { 740 struct netmsg_base base; 741 inp_notify_t nm_notify; 742 struct in_addr nm_faddr; 743 int nm_arg; 744 }; 745 746 static void 747 udp_notifyall_oncpu(netmsg_t msg) 748 { 749 struct netmsg_udp_notify *nm = (struct netmsg_udp_notify *)msg; 750 int nextcpu, cpu = mycpuid; 751 752 in_pcbnotifyall(&udbinfo[cpu], nm->nm_faddr, nm->nm_arg, nm->nm_notify); 753 754 nextcpu = cpu + 1; 755 if (nextcpu < ncpus2) 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 CPUs */ 785 *cpuid = 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 notify = udp_get_inpnotify(cmd, sa, &ip, &cpuid); 810 if (notify == NULL) 811 goto done; 812 813 faddr = ((struct sockaddr_in *)sa)->sin_addr; 814 if (ip) { 815 const struct udphdr *uh; 816 struct inpcb *inp; 817 818 if (cpuid != mycpuid) 819 goto done; 820 821 uh = (const struct udphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 822 inp = in_pcblookup_hash(&udbinfo[mycpuid], faddr, uh->uh_dport, 823 ip->ip_src, uh->uh_sport, 0, NULL); 824 if (inp != NULL && inp->inp_socket != NULL) 825 notify(inp, inetctlerrmap[cmd]); 826 } else if (msg->ctlinput.nm_direct) { 827 if (cpuid != ncpus && cpuid != mycpuid) 828 goto done; 829 if (mycpuid >= ncpus2) 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_IN_NETISR(0); 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_global_ncpus2, "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 < ncpus2; ++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 if (inp->inp_socket->so_options & SO_REUSEPORT) { 1298 /* 1299 * For SO_REUSEPORT socket, redistribute it based on its 1300 * local group index. 1301 */ 1302 cpu = inp->inp_lgrpindex & ncpus2_mask; 1303 if (cpu != mycpuid) { 1304 struct lwkt_port *port = netisr_cpuport(cpu); 1305 lwkt_msg_t lmsg = &msg->lmsg; 1306 1307 /* 1308 * We are moving the protocol processing port the 1309 * socket is on, we have to unlink here and re-link 1310 * on the target cpu (this inpcb is still left in 1311 * the wildcard hash). 1312 */ 1313 in_pcbunlink_flags(inp, &udbinfo[mycpuid], 0); 1314 msg->nm_dispatch = udp_soreuseport_dispatch; 1315 1316 /* 1317 * See the related comment in tcp_usrreq.c 1318 * tcp_connect() 1319 */ 1320 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1321 lwkt_forwardmsg(port, lmsg); 1322 return TRUE; /* forwarded */ 1323 } 1324 } 1325 return FALSE; 1326 } 1327 1328 static void 1329 udp_inswildcardhash_dispatch(netmsg_t msg) 1330 { 1331 struct inpcb *inp = msg->base.nm_so->so_pcb; 1332 boolean_t forwarded; 1333 1334 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1335 KASSERT(udp_lportcpu(inp->inp_lport) == mycpuid, ("not target cpu")); 1336 1337 in_pcblink(inp, &udbinfo[mycpuid]); 1338 1339 forwarded = udp_inswildcardhash_oncpu(inp, &msg->base); 1340 if (forwarded) { 1341 /* The message is further forwarded, so we are done here. */ 1342 return; 1343 } 1344 udp_inswildcard_replymsg(msg); 1345 } 1346 1347 static boolean_t 1348 udp_inswildcardhash(struct inpcb *inp, struct netmsg_base *msg, int error) 1349 { 1350 lwkt_msg_t lmsg = &msg->lmsg; 1351 int cpu; 1352 1353 ASSERT_INP_NOTINHASH(inp); 1354 1355 /* This inpcb could no longer be directly detached */ 1356 inp->inp_flags &= ~INP_DIRECT_DETACH; 1357 1358 /* 1359 * Always clear the route cache, so we don't need to 1360 * worry about any owner CPU changes later. 1361 */ 1362 in_pcbresetroute(inp); 1363 1364 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1365 cpu = udp_lportcpu(inp->inp_lport); 1366 1367 lmsg->ms_error = error; 1368 if (cpu != mycpuid) { 1369 struct lwkt_port *port = netisr_cpuport(cpu); 1370 1371 /* 1372 * We are moving the protocol processing port the socket 1373 * is on, we have to unlink here and re-link on the 1374 * target cpu. 1375 */ 1376 in_pcbunlink(inp, &udbinfo[mycpuid]); 1377 msg->nm_dispatch = udp_inswildcardhash_dispatch; 1378 1379 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1380 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1381 lwkt_forwardmsg(port, lmsg); 1382 return TRUE; /* forwarded */ 1383 } 1384 1385 return udp_inswildcardhash_oncpu(inp, msg); 1386 } 1387 1388 static void 1389 udp_bind(netmsg_t msg) 1390 { 1391 struct socket *so = msg->bind.base.nm_so; 1392 struct inpcb *inp; 1393 int error; 1394 1395 inp = so->so_pcb; 1396 if (inp) { 1397 struct sockaddr *nam = msg->bind.nm_nam; 1398 struct thread *td = msg->bind.nm_td; 1399 struct sockaddr_in *sin; 1400 lwkt_port_t port; 1401 int cpu; 1402 1403 /* 1404 * Check "already bound" here (in_pcbbind() does the same 1405 * check though), so we don't forward a connected/bound 1406 * socket randomly which would panic in the following 1407 * in_pcbunlink(). 1408 */ 1409 if (inp->inp_lport != 0 || 1410 inp->inp_laddr.s_addr != INADDR_ANY) { 1411 error = EINVAL; /* already bound */ 1412 goto done; 1413 } 1414 1415 if (nam->sa_len != sizeof(*sin)) { 1416 error = EINVAL; 1417 goto done; 1418 } 1419 sin = (struct sockaddr_in *)nam; 1420 1421 cpu = udp_lportcpu(sin->sin_port); 1422 port = netisr_cpuport(cpu); 1423 1424 /* 1425 * See the related comment in tcp_usrreq.c tcp_usr_bind(). 1426 * The exception is that we use local port based netisr 1427 * to serialize in_pcbbind(). 1428 */ 1429 if (&curthread->td_msgport != port) { 1430 lwkt_msg_t lmsg = &msg->bind.base.lmsg; 1431 1432 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0, 1433 ("already asked to relink")); 1434 1435 in_pcbunlink(so->so_pcb, &udbinfo[mycpuid]); 1436 msg->bind.nm_flags |= PRUB_RELINK; 1437 1438 /* 1439 * See the related comment in tcp_usrreq.c 1440 * tcp_connect(). 1441 */ 1442 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1443 lwkt_forwardmsg(port, lmsg); 1444 /* msg invalid now */ 1445 return; 1446 } 1447 KASSERT(so->so_port == port, ("so_port is not netisr%d", cpu)); 1448 1449 if (msg->bind.nm_flags & PRUB_RELINK) { 1450 msg->bind.nm_flags &= ~PRUB_RELINK; 1451 in_pcblink(so->so_pcb, &udbinfo[mycpuid]); 1452 } 1453 KASSERT(inp->inp_pcbinfo == &udbinfo[cpu], 1454 ("pcbinfo is not udbinfo%d", cpu)); 1455 1456 error = in_pcbbind(inp, nam, td); 1457 if (error == 0) { 1458 boolean_t forwarded; 1459 1460 if (sin->sin_addr.s_addr != INADDR_ANY) 1461 inp->inp_flags |= INP_WASBOUND_NOTANY; 1462 1463 forwarded = udp_inswildcardhash(inp, 1464 &msg->bind.base, 0); 1465 if (forwarded) { 1466 /* 1467 * The message is further forwarded, so 1468 * we are done here. 1469 */ 1470 return; 1471 } 1472 } 1473 } else { 1474 error = EINVAL; 1475 } 1476 done: 1477 lwkt_replymsg(&msg->bind.base.lmsg, error); 1478 } 1479 1480 static int 1481 udp_preconnect(struct socket *so, const struct sockaddr *nam __unused, 1482 struct thread *td __unused) 1483 { 1484 sosetstate(so, SS_ISCONNECTED); /* XXX */ 1485 return 0; 1486 } 1487 1488 static void 1489 udp_connect(netmsg_t msg) 1490 { 1491 struct socket *so = msg->connect.base.nm_so; 1492 struct sockaddr *nam = msg->connect.nm_nam; 1493 struct thread *td = msg->connect.nm_td; 1494 struct inpcb *inp; 1495 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1496 struct sockaddr_in *if_sin; 1497 struct lwkt_port *port; 1498 uint16_t hash; 1499 int error; 1500 1501 KKASSERT(msg->connect.nm_m == NULL); 1502 1503 inp = so->so_pcb; 1504 if (inp == NULL) { 1505 error = EINVAL; 1506 goto out; 1507 } 1508 1509 if (msg->connect.nm_flags & PRUC_RECONNECT) { 1510 msg->connect.nm_flags &= ~PRUC_RECONNECT; 1511 in_pcblink(inp, &udbinfo[mycpuid]); 1512 } 1513 1514 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1515 error = EISCONN; 1516 goto out; 1517 } 1518 error = 0; 1519 1520 /* 1521 * Bind if we have to 1522 */ 1523 if (inp->inp_lport == 0) { 1524 error = in_pcbbind(inp, NULL, td); 1525 if (error) 1526 goto out; 1527 } 1528 1529 /* 1530 * Calculate the correct protocol processing thread. The connect 1531 * operation must run there. 1532 */ 1533 error = in_pcbladdr(inp, nam, &if_sin, td); 1534 if (error) 1535 goto out; 1536 if (!prison_remote_ip(td, nam)) { 1537 error = EAFNOSUPPORT; /* IPv6 only jail */ 1538 goto out; 1539 } 1540 1541 hash = udp_addrhash(sin->sin_addr.s_addr, sin->sin_port, 1542 inp->inp_laddr.s_addr != INADDR_ANY ? 1543 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, inp->inp_lport); 1544 port = netisr_hashport(hash); 1545 if (port != &curthread->td_msgport) { 1546 lwkt_msg_t lmsg = &msg->connect.base.lmsg; 1547 int nm_flags = PRUC_RECONNECT; 1548 1549 /* 1550 * in_pcbladdr() may have allocated a route entry for us 1551 * on the current CPU, but we need a route entry on the 1552 * inpcb's owner CPU, so free it here. 1553 */ 1554 in_pcbresetroute(inp); 1555 1556 if (inp->inp_flags & INP_WILDCARD) { 1557 /* 1558 * Remove this inpcb from the wildcard hash before 1559 * the socket's msgport changes. 1560 */ 1561 udp_remwildcardhash(inp); 1562 } 1563 1564 if (so->so_orig_port == NULL) { 1565 /* 1566 * First time change protocol processing port. 1567 * Save the current port for synchronization upon 1568 * udp_detach. 1569 */ 1570 so->so_orig_port = &curthread->td_msgport; 1571 } else { 1572 /* 1573 * We have changed protocol processing port more 1574 * than once. We could not do direct detach 1575 * anymore, because we lose the track of the 1576 * original protocol processing ports to perform 1577 * synchronization upon udp_detach. This should 1578 * be rare though. 1579 */ 1580 inp->inp_flags &= ~INP_DIRECT_DETACH; 1581 } 1582 1583 /* 1584 * We are moving the protocol processing port the socket 1585 * is on, we have to unlink here and re-link on the 1586 * target cpu. 1587 */ 1588 in_pcbunlink(inp, &udbinfo[mycpuid]); 1589 msg->connect.nm_flags |= nm_flags; 1590 1591 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1592 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1593 lwkt_forwardmsg(port, lmsg); 1594 /* msg invalid now */ 1595 return; 1596 } 1597 error = udp_connect_oncpu(inp, sin, if_sin, hash); 1598 out: 1599 if (msg->connect.nm_flags & PRUC_HELDTD) 1600 lwkt_rele(td); 1601 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) { 1602 if (inp->inp_lport == 0) { 1603 /* 1604 * As long as we have the local port, it is fine 1605 * for connect to fail, e.g. disconnect. 1606 */ 1607 so->so_error = error; 1608 } 1609 soclrstate(so, SS_ISCONNECTED); 1610 /* 1611 * Wake up callers blocked on this socket to make sure 1612 * that they can see this error. 1613 * 1614 * NOTE: 1615 * sodisconnected() can't be used here, which bricks 1616 * sending and receiving. 1617 */ 1618 wakeup(&so->so_timeo); 1619 sowwakeup(so); 1620 sorwakeup(so); 1621 } 1622 if (error && inp != NULL && inp->inp_lport != 0 && 1623 (inp->inp_flags & INP_WILDCARD) == 0) { 1624 boolean_t forwarded; 1625 1626 /* Connect failed; put it to wildcard hash. */ 1627 forwarded = udp_inswildcardhash(inp, &msg->connect.base, 1628 error); 1629 if (forwarded) { 1630 /* 1631 * The message is further forwarded, so we are done 1632 * here. 1633 */ 1634 return; 1635 } 1636 } 1637 lwkt_replymsg(&msg->connect.base.lmsg, error); 1638 } 1639 1640 static void 1641 udp_remwildcardhash(struct inpcb *inp) 1642 { 1643 int cpu; 1644 1645 KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid], 1646 ("not on owner cpu")); 1647 1648 for (cpu = 0; cpu < ncpus2; ++cpu) { 1649 if (cpu == mycpuid) { 1650 /* 1651 * This inpcb will be removed by the later 1652 * in_pcbremwildcardhash(). 1653 */ 1654 continue; 1655 } 1656 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpu]); 1657 } 1658 in_pcbremwildcardhash(inp); 1659 } 1660 1661 static int 1662 udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin, 1663 struct sockaddr_in *if_sin, uint16_t hash) 1664 { 1665 struct socket *so = inp->inp_socket; 1666 struct inpcb *oinp; 1667 1668 oinp = in_pcblookup_hash(inp->inp_pcbinfo, 1669 sin->sin_addr, sin->sin_port, 1670 inp->inp_laddr.s_addr != INADDR_ANY ? 1671 inp->inp_laddr : if_sin->sin_addr, inp->inp_lport, FALSE, NULL); 1672 if (oinp != NULL) 1673 return EADDRINUSE; 1674 1675 /* 1676 * No more errors can occur, finish adjusting the socket 1677 * and change the processing port to reflect the connected 1678 * socket. Once set we can no longer safely mess with the 1679 * socket. 1680 */ 1681 1682 if (inp->inp_flags & INP_WILDCARD) 1683 udp_remwildcardhash(inp); 1684 1685 if (inp->inp_laddr.s_addr == INADDR_ANY) 1686 inp->inp_laddr = if_sin->sin_addr; 1687 inp->inp_faddr = sin->sin_addr; 1688 inp->inp_fport = sin->sin_port; 1689 in_pcbinsconnhash(inp); 1690 1691 inp->inp_flags |= INP_HASH; 1692 inp->inp_hashval = hash; 1693 1694 soisconnected(so); 1695 1696 return 0; 1697 } 1698 1699 static void 1700 udp_detach2(struct socket *so) 1701 { 1702 in_pcbdetach(so->so_pcb); 1703 sodiscard(so); 1704 sofree(so); 1705 } 1706 1707 static void 1708 udp_detach_final_dispatch(netmsg_t msg) 1709 { 1710 udp_detach2(msg->base.nm_so); 1711 } 1712 1713 static void 1714 udp_detach_oncpu_dispatch(netmsg_t msg) 1715 { 1716 struct netmsg_base *clomsg = &msg->base; 1717 struct socket *so = clomsg->nm_so; 1718 struct inpcb *inp = so->so_pcb; 1719 struct thread *td = curthread; 1720 int nextcpu, cpuid = mycpuid; 1721 1722 KASSERT(td->td_type == TD_TYPE_NETISR, ("not in netisr")); 1723 1724 if (inp->inp_flags & INP_WILDCARD) { 1725 /* 1726 * This inp will be removed on the inp's 1727 * owner CPU later, so don't do it now. 1728 */ 1729 if (&td->td_msgport != so->so_port) 1730 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpuid]); 1731 } 1732 1733 if (cpuid == 0) { 1734 /* 1735 * Free and clear multicast socket option, 1736 * which is only accessed in netisr0. 1737 */ 1738 ip_freemoptions(inp->inp_moptions); 1739 inp->inp_moptions = NULL; 1740 } 1741 1742 nextcpu = cpuid + 1; 1743 if (nextcpu < ncpus2) { 1744 lwkt_forwardmsg(netisr_cpuport(nextcpu), &clomsg->lmsg); 1745 } else { 1746 /* 1747 * No one could see this inpcb now; destroy this 1748 * inpcb in its owner netisr. 1749 */ 1750 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1751 udp_detach_final_dispatch); 1752 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1753 } 1754 } 1755 1756 static void 1757 udp_detach_syncorig_dispatch(netmsg_t msg) 1758 { 1759 struct netmsg_base *clomsg = &msg->base; 1760 struct socket *so = clomsg->nm_so; 1761 1762 /* 1763 * Original protocol processing port is synchronized; 1764 * destroy this inpcb in its owner netisr. 1765 */ 1766 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1767 udp_detach_final_dispatch); 1768 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1769 } 1770 1771 static void 1772 udp_detach(netmsg_t msg) 1773 { 1774 struct socket *so = msg->detach.base.nm_so; 1775 struct netmsg_base *clomsg; 1776 struct inpcb *inp; 1777 1778 inp = so->so_pcb; 1779 if (inp == NULL) { 1780 lwkt_replymsg(&msg->detach.base.lmsg, EINVAL); 1781 return; 1782 } 1783 1784 /* 1785 * Reply EJUSTRETURN ASAP, we will call sodiscard() and 1786 * sofree() later. 1787 */ 1788 lwkt_replymsg(&msg->detach.base.lmsg, EJUSTRETURN); 1789 1790 if (ncpus2 == 1) { 1791 /* Only one CPU, detach the inpcb directly. */ 1792 udp_detach2(so); 1793 return; 1794 } 1795 1796 /* 1797 * Remove this inpcb from the inpcb list first, so that 1798 * no one could find this inpcb from the inpcb list. 1799 */ 1800 in_pcbofflist(inp); 1801 1802 /* 1803 * Remove this inpcb from the local port hash directly 1804 * here, so that its bound local port could be recycled 1805 * timely. 1806 */ 1807 in_pcbremporthash(inp); 1808 1809 if (inp->inp_flags & INP_DIRECT_DETACH) { 1810 /* 1811 * Direct detaching is allowed 1812 */ 1813 KASSERT((inp->inp_flags & INP_WILDCARD) == 0, 1814 ("in the wildcardhash")); 1815 KASSERT(inp->inp_moptions == NULL, ("has mcast options")); 1816 if (so->so_orig_port == NULL) { 1817 udp_detach2(so); 1818 } else { 1819 /* 1820 * Protocol processing port changed once, so 1821 * we need to make sure that there are nothing 1822 * left on the original protocol processing 1823 * port before we destroy this socket and inpcb. 1824 * This is more lightweight than going through 1825 * all UDP processing netisrs. 1826 */ 1827 clomsg = &so->so_clomsg; 1828 netmsg_init(clomsg, so, &netisr_apanic_rport, 1829 MSGF_IGNSOPORT, udp_detach_syncorig_dispatch); 1830 lwkt_sendmsg(so->so_orig_port, &clomsg->lmsg); 1831 } 1832 return; 1833 } 1834 1835 /* 1836 * Go through netisrs which process UDP to make sure 1837 * no one could find this inpcb anymore. 1838 */ 1839 clomsg = &so->so_clomsg; 1840 netmsg_init(clomsg, so, &netisr_apanic_rport, MSGF_IGNSOPORT, 1841 udp_detach_oncpu_dispatch); 1842 lwkt_sendmsg(netisr_cpuport(0), &clomsg->lmsg); 1843 } 1844 1845 static void 1846 udp_disconnect(netmsg_t msg) 1847 { 1848 struct socket *so = msg->disconnect.base.nm_so; 1849 struct inpcb *inp; 1850 boolean_t forwarded; 1851 int error = 0; 1852 1853 inp = so->so_pcb; 1854 if (inp == NULL) { 1855 error = EINVAL; 1856 goto out; 1857 } 1858 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1859 error = ENOTCONN; 1860 goto out; 1861 } 1862 1863 soclrstate(so, SS_ISCONNECTED); /* XXX */ 1864 1865 in_pcbdisconnect(inp); 1866 inp->inp_flags &= ~INP_HASH; 1867 1868 /* 1869 * Follow traditional BSD behavior and retain the local port 1870 * binding. But, fix the old misbehavior of overwriting any 1871 * previously bound local address. 1872 */ 1873 if (!(inp->inp_flags & INP_WASBOUND_NOTANY)) 1874 inp->inp_laddr.s_addr = INADDR_ANY; 1875 1876 if (so->so_state & SS_ISCLOSING) { 1877 /* 1878 * If this socket is being closed, there is no need 1879 * to put this socket back into wildcard hash table. 1880 */ 1881 error = 0; 1882 goto out; 1883 } 1884 1885 forwarded = udp_inswildcardhash(inp, &msg->disconnect.base, 0); 1886 if (forwarded) { 1887 /* 1888 * The message is further forwarded, so we are done 1889 * here. 1890 */ 1891 return; 1892 } 1893 out: 1894 lwkt_replymsg(&msg->disconnect.base.lmsg, error); 1895 } 1896 1897 void 1898 udp_shutdown(netmsg_t msg) 1899 { 1900 struct socket *so = msg->shutdown.base.nm_so; 1901 struct inpcb *inp; 1902 int error; 1903 1904 inp = so->so_pcb; 1905 if (inp) { 1906 socantsendmore(so); 1907 error = 0; 1908 } else { 1909 error = EINVAL; 1910 } 1911 lwkt_replymsg(&msg->shutdown.base.lmsg, error); 1912 } 1913 1914 struct pr_usrreqs udp_usrreqs = { 1915 .pru_abort = udp_abort, 1916 .pru_accept = pr_generic_notsupp, 1917 .pru_attach = udp_attach, 1918 .pru_bind = udp_bind, 1919 .pru_connect = udp_connect, 1920 .pru_connect2 = pr_generic_notsupp, 1921 .pru_control = in_control_dispatch, 1922 .pru_detach = udp_detach, 1923 .pru_disconnect = udp_disconnect, 1924 .pru_listen = pr_generic_notsupp, 1925 .pru_peeraddr = in_setpeeraddr_dispatch, 1926 .pru_rcvd = pr_generic_notsupp, 1927 .pru_rcvoob = pr_generic_notsupp, 1928 .pru_send = udp_send, 1929 .pru_sense = pru_sense_null, 1930 .pru_shutdown = udp_shutdown, 1931 .pru_sockaddr = in_setsockaddr_dispatch, 1932 .pru_sosend = sosendudp, 1933 .pru_soreceive = soreceive, 1934 .pru_preconnect = udp_preconnect, 1935 .pru_preattach = udp_preattach 1936 }; 1937