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); 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 int flags; 935 936 struct udpiphdr *ui; 937 int len = m->m_pkthdr.len; 938 struct sockaddr_in *sin; /* really is initialized before use */ 939 int error = 0, cpu; 940 941 KKASSERT(msg->send.nm_control == NULL); 942 943 logudp(send_beg, inp); 944 945 if (inp == NULL) { 946 error = EINVAL; 947 goto release; 948 } 949 950 if (len + sizeof(struct udpiphdr) > IP_MAXPACKET) { 951 error = EMSGSIZE; 952 goto release; 953 } 954 955 if (inp->inp_lport == 0) { /* unbound socket */ 956 boolean_t forwarded; 957 958 error = in_pcbbind(inp, NULL, td); 959 if (error) 960 goto release; 961 962 /* 963 * Need to call udp_send again, after this inpcb is 964 * inserted into wildcard hash table. 965 */ 966 msg->send.base.lmsg.ms_flags |= MSGF_UDP_SEND; 967 forwarded = udp_inswildcardhash(inp, &msg->send.base, 0); 968 if (forwarded) { 969 /* 970 * The message is further forwarded, so we are 971 * done here. 972 */ 973 logudp(send_inswildcard, inp); 974 return; 975 } 976 } 977 978 if (dstaddr != NULL) { /* destination address specified */ 979 if (inp->inp_faddr.s_addr != INADDR_ANY) { 980 /* already connected */ 981 error = EISCONN; 982 goto release; 983 } 984 sin = (struct sockaddr_in *)dstaddr; 985 if (!prison_remote_ip(td, (struct sockaddr *)&sin)) { 986 error = EAFNOSUPPORT; /* IPv6 only jail */ 987 goto release; 988 } 989 } else { 990 if (inp->inp_faddr.s_addr == INADDR_ANY) { 991 /* no destination specified and not already connected */ 992 error = ENOTCONN; 993 goto release; 994 } 995 sin = NULL; 996 } 997 998 /* 999 * Calculate data length and get a mbuf 1000 * for UDP and IP headers. 1001 */ 1002 M_PREPEND(m, sizeof(struct udpiphdr), M_NOWAIT); 1003 if (m == NULL) { 1004 error = ENOBUFS; 1005 goto release; 1006 } 1007 1008 /* 1009 * Fill in mbuf with extended UDP header 1010 * and addresses and length put into network format. 1011 */ 1012 ui = mtod(m, struct udpiphdr *); 1013 bzero(ui->ui_x1, sizeof ui->ui_x1); /* XXX still needed? */ 1014 ui->ui_pr = IPPROTO_UDP; 1015 1016 /* 1017 * Set destination address. 1018 */ 1019 if (dstaddr != NULL) { /* use specified destination */ 1020 ui->ui_dst = sin->sin_addr; 1021 ui->ui_dport = sin->sin_port; 1022 } else { /* use connected destination */ 1023 ui->ui_dst = inp->inp_faddr; 1024 ui->ui_dport = inp->inp_fport; 1025 } 1026 1027 /* 1028 * Set source address. 1029 */ 1030 if (inp->inp_laddr.s_addr == INADDR_ANY || 1031 IN_MULTICAST(ntohl(inp->inp_laddr.s_addr))) { 1032 struct sockaddr_in *if_sin; 1033 1034 if (dstaddr == NULL) { 1035 /* 1036 * connect() had (or should have) failed because 1037 * the interface had no IP address, but the 1038 * application proceeded to call send() anyways. 1039 */ 1040 error = ENOTCONN; 1041 goto release; 1042 } 1043 1044 /* Look up outgoing interface. */ 1045 error = in_pcbladdr_find(inp, dstaddr, &if_sin, td, 1); 1046 if (error) 1047 goto release; 1048 ui->ui_src = if_sin->sin_addr; /* use address of interface */ 1049 } else { 1050 ui->ui_src = inp->inp_laddr; /* use non-null bound address */ 1051 } 1052 ui->ui_sport = inp->inp_lport; 1053 KASSERT(inp->inp_lport != 0, ("inp lport should have been bound")); 1054 1055 /* 1056 * Release the original thread, since it is no longer used 1057 */ 1058 if (pru_flags & PRUS_HELDTD) { 1059 lwkt_rele(td); 1060 pru_flags &= ~PRUS_HELDTD; 1061 } 1062 /* 1063 * Free the dest address, since it is no longer needed 1064 */ 1065 if (pru_flags & PRUS_FREEADDR) { 1066 kfree(dstaddr, M_SONAME); 1067 pru_flags &= ~PRUS_FREEADDR; 1068 } 1069 1070 ui->ui_ulen = htons((u_short)len + sizeof(struct udphdr)); 1071 1072 /* 1073 * Set up checksum and output datagram. 1074 */ 1075 if (udpcksum) { 1076 ui->ui_sum = in_pseudo(ui->ui_src.s_addr, ui->ui_dst.s_addr, 1077 htons((u_short)len + sizeof(struct udphdr) + IPPROTO_UDP)); 1078 m->m_pkthdr.csum_flags = CSUM_UDP; 1079 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 1080 m->m_pkthdr.csum_thlen = sizeof(struct udphdr); 1081 } else { 1082 ui->ui_sum = 0; 1083 } 1084 ((struct ip *)ui)->ip_len = sizeof(struct udpiphdr) + len; 1085 ((struct ip *)ui)->ip_ttl = inp->inp_ip_ttl; /* XXX */ 1086 ((struct ip *)ui)->ip_tos = inp->inp_ip_tos; /* XXX */ 1087 udp_stat.udps_opackets++; 1088 1089 flags = IP_DEBUGROUTE | 1090 (inp->inp_socket->so_options & (SO_DONTROUTE | SO_BROADCAST)); 1091 if (pru_flags & PRUS_DONTROUTE) 1092 flags |= SO_DONTROUTE; 1093 1094 if (inp->inp_flags & INP_CONNECTED) { 1095 /* 1096 * For connected socket, this datagram has already 1097 * been in the correct netisr; no need to rehash. 1098 */ 1099 goto sendit; 1100 } 1101 1102 cpu = udp_addrcpu(ui->ui_dst.s_addr, ui->ui_dport, 1103 ui->ui_src.s_addr, ui->ui_sport); 1104 if (cpu != mycpuid) { 1105 struct mbuf *m_opt = NULL; 1106 struct netmsg_pru_send *smsg; 1107 struct lwkt_port *port = netisr_cpuport(cpu); 1108 1109 /* 1110 * Not on the CPU that matches this UDP datagram hash; 1111 * redispatch to the correct CPU to do the ip_output(). 1112 */ 1113 if (inp->inp_options != NULL) { 1114 /* 1115 * If there are ip options, then save a copy, 1116 * since accessing inp_options on other CPUs' 1117 * is not safe. 1118 * 1119 * XXX optimize this? 1120 */ 1121 m_opt = m_copym(inp->inp_options, 0, M_COPYALL, 1122 M_WAITOK); 1123 } 1124 if ((pru_flags & PRUS_NOREPLY) == 0) { 1125 /* 1126 * Change some parts of the original netmsg and 1127 * forward it to the target netisr. 1128 * 1129 * NOTE: so_port MUST NOT be checked in the target 1130 * netisr. 1131 */ 1132 smsg = &msg->send; 1133 smsg->nm_priv = flags; /* ip_output flags */ 1134 smsg->nm_m = m; 1135 smsg->nm_control = m_opt; /* XXX save ipopt */ 1136 smsg->base.lmsg.ms_flags |= MSGF_IGNSOPORT; 1137 smsg->base.nm_dispatch = udp_send_redispatch; 1138 lwkt_forwardmsg(port, &smsg->base.lmsg); 1139 } else { 1140 /* 1141 * Recreate the netmsg, since the original mbuf 1142 * could have been changed. And send it to the 1143 * target netisr. 1144 * 1145 * NOTE: so_port MUST NOT be checked in the target 1146 * netisr. 1147 */ 1148 smsg = &m->m_hdr.mh_sndmsg; 1149 netmsg_init(&smsg->base, so, &netisr_apanic_rport, 1150 MSGF_IGNSOPORT, udp_send_redispatch); 1151 smsg->nm_priv = flags; /* ip_output flags */ 1152 smsg->nm_flags = pru_flags; 1153 smsg->nm_m = m; 1154 smsg->nm_control = m_opt; /* XXX save ipopt */ 1155 lwkt_sendmsg(port, &smsg->base.lmsg); 1156 } 1157 1158 /* This UDP datagram is redispatched; done */ 1159 logudp(send_redisp, inp); 1160 return; 1161 } 1162 1163 sendit: 1164 logudp(send_ipout, inp); 1165 error = ip_output(m, inp->inp_options, &inp->inp_route, flags, 1166 inp->inp_moptions, inp); 1167 m = NULL; 1168 1169 release: 1170 if (m != NULL) 1171 m_freem(m); 1172 1173 if (pru_flags & PRUS_HELDTD) 1174 lwkt_rele(td); 1175 if (pru_flags & PRUS_FREEADDR) 1176 kfree(dstaddr, M_SONAME); 1177 if ((pru_flags & PRUS_NOREPLY) == 0) 1178 lwkt_replymsg(&msg->send.base.lmsg, error); 1179 1180 logudp(send_end, inp); 1181 } 1182 1183 u_long udp_sendspace = 9216; /* really max datagram size */ 1184 /* 40 1K datagrams */ 1185 SYSCTL_INT(_net_inet_udp, UDPCTL_MAXDGRAM, maxdgram, CTLFLAG_RW, 1186 &udp_sendspace, 0, "Maximum outgoing UDP datagram size"); 1187 1188 u_long udp_recvspace = 40 * (1024 + 1189 #ifdef INET6 1190 sizeof(struct sockaddr_in6) 1191 #else 1192 sizeof(struct sockaddr_in) 1193 #endif 1194 ); 1195 SYSCTL_INT(_net_inet_udp, UDPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 1196 &udp_recvspace, 0, "Maximum incoming UDP datagram size"); 1197 1198 /* 1199 * This should never happen, since UDP socket does not support 1200 * connection acception (SO_ACCEPTCONN, i.e. listen(2)). 1201 */ 1202 static void 1203 udp_abort(netmsg_t msg __unused) 1204 { 1205 panic("udp_abort is called"); 1206 } 1207 1208 static int 1209 udp_preattach(struct socket *so, int proto __unused, struct pru_attach_info *ai) 1210 { 1211 return soreserve(so, udp_sendspace, udp_recvspace, ai->sb_rlimit); 1212 } 1213 1214 static void 1215 udp_attach(netmsg_t msg) 1216 { 1217 struct socket *so = msg->attach.base.nm_so; 1218 struct pru_attach_info *ai = msg->attach.nm_ai; 1219 struct inpcb *inp; 1220 int error; 1221 1222 KASSERT(so->so_pcb == NULL, ("udp socket attached")); 1223 1224 if (ai != NULL) { 1225 error = udp_preattach(so, 0 /* don't care */, ai); 1226 if (error) 1227 goto out; 1228 } else { 1229 /* Post attach; do nothing */ 1230 } 1231 1232 error = in_pcballoc(so, &udbinfo[mycpuid]); 1233 if (error) 1234 goto out; 1235 1236 inp = so->so_pcb; 1237 inp->inp_flags |= INP_DIRECT_DETACH; 1238 inp->inp_ip_ttl = ip_defttl; 1239 error = 0; 1240 out: 1241 lwkt_replymsg(&msg->attach.base.lmsg, error); 1242 } 1243 1244 static void 1245 udp_inswildcard_replymsg(netmsg_t msg) 1246 { 1247 lwkt_msg_t lmsg = &msg->lmsg; 1248 1249 if (lmsg->ms_flags & MSGF_UDP_SEND) { 1250 udp_send(msg); 1251 /* msg is replied by udp_send() */ 1252 } else { 1253 lwkt_replymsg(lmsg, lmsg->ms_error); 1254 } 1255 } 1256 1257 static void 1258 udp_soreuseport_dispatch(netmsg_t msg) 1259 { 1260 /* This inpcb has already been in the wildcard hash. */ 1261 in_pcblink_flags(msg->base.nm_so->so_pcb, &udbinfo[mycpuid], 0); 1262 udp_inswildcard_replymsg(msg); 1263 } 1264 1265 static void 1266 udp_sosetport(struct lwkt_msg *msg, lwkt_port_t port) 1267 { 1268 sosetport(((struct netmsg_base *)msg)->nm_so, port); 1269 } 1270 1271 static boolean_t 1272 udp_inswildcardhash_oncpu(struct inpcb *inp, struct netmsg_base *msg) 1273 { 1274 int cpu; 1275 1276 KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid], 1277 ("not on owner cpu")); 1278 1279 in_pcbinswildcardhash(inp); 1280 for (cpu = 0; cpu < ncpus2; ++cpu) { 1281 if (cpu == mycpuid) { 1282 /* 1283 * This inpcb has been inserted by the above 1284 * in_pcbinswildcardhash(). 1285 */ 1286 continue; 1287 } 1288 in_pcbinswildcardhash_oncpu(inp, &udbinfo[cpu]); 1289 } 1290 1291 if (inp->inp_socket->so_options & SO_REUSEPORT) { 1292 /* 1293 * For SO_REUSEPORT socket, redistribute it based on its 1294 * local group index. 1295 */ 1296 cpu = inp->inp_lgrpindex & ncpus2_mask; 1297 if (cpu != mycpuid) { 1298 struct lwkt_port *port = netisr_cpuport(cpu); 1299 lwkt_msg_t lmsg = &msg->lmsg; 1300 1301 /* 1302 * We are moving the protocol processing port the 1303 * socket is on, we have to unlink here and re-link 1304 * on the target cpu (this inpcb is still left in 1305 * the wildcard hash). 1306 */ 1307 in_pcbunlink_flags(inp, &udbinfo[mycpuid], 0); 1308 msg->nm_dispatch = udp_soreuseport_dispatch; 1309 1310 /* 1311 * See the related comment in tcp_usrreq.c 1312 * tcp_connect() 1313 */ 1314 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1315 lwkt_forwardmsg(port, lmsg); 1316 return TRUE; /* forwarded */ 1317 } 1318 } 1319 return FALSE; 1320 } 1321 1322 static void 1323 udp_inswildcardhash_dispatch(netmsg_t msg) 1324 { 1325 struct inpcb *inp = msg->base.nm_so->so_pcb; 1326 boolean_t forwarded; 1327 1328 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1329 KASSERT(udp_lportcpu(inp->inp_lport) == mycpuid, ("not target cpu")); 1330 1331 in_pcblink(inp, &udbinfo[mycpuid]); 1332 1333 forwarded = udp_inswildcardhash_oncpu(inp, &msg->base); 1334 if (forwarded) { 1335 /* The message is further forwarded, so we are done here. */ 1336 return; 1337 } 1338 udp_inswildcard_replymsg(msg); 1339 } 1340 1341 static boolean_t 1342 udp_inswildcardhash(struct inpcb *inp, struct netmsg_base *msg, int error) 1343 { 1344 lwkt_msg_t lmsg = &msg->lmsg; 1345 int cpu; 1346 1347 ASSERT_INP_NOTINHASH(inp); 1348 1349 /* This inpcb could no longer be directly detached */ 1350 inp->inp_flags &= ~INP_DIRECT_DETACH; 1351 1352 /* 1353 * Always clear the route cache, so we don't need to 1354 * worry about any owner CPU changes later. 1355 */ 1356 in_pcbresetroute(inp); 1357 1358 KASSERT(inp->inp_lport != 0, ("local port not set yet")); 1359 cpu = udp_lportcpu(inp->inp_lport); 1360 1361 lmsg->ms_error = error; 1362 if (cpu != mycpuid) { 1363 struct lwkt_port *port = netisr_cpuport(cpu); 1364 1365 /* 1366 * We are moving the protocol processing port the socket 1367 * is on, we have to unlink here and re-link on the 1368 * target cpu. 1369 */ 1370 in_pcbunlink(inp, &udbinfo[mycpuid]); 1371 msg->nm_dispatch = udp_inswildcardhash_dispatch; 1372 1373 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1374 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1375 lwkt_forwardmsg(port, lmsg); 1376 return TRUE; /* forwarded */ 1377 } 1378 1379 return udp_inswildcardhash_oncpu(inp, msg); 1380 } 1381 1382 static void 1383 udp_bind(netmsg_t msg) 1384 { 1385 struct socket *so = msg->bind.base.nm_so; 1386 struct inpcb *inp; 1387 int error; 1388 1389 inp = so->so_pcb; 1390 if (inp) { 1391 struct sockaddr *nam = msg->bind.nm_nam; 1392 struct thread *td = msg->bind.nm_td; 1393 struct sockaddr_in *sin; 1394 lwkt_port_t port; 1395 int cpu; 1396 1397 /* 1398 * Check "already bound" here (in_pcbbind() does the same 1399 * check though), so we don't forward a connected/bound 1400 * socket randomly which would panic in the following 1401 * in_pcbunlink(). 1402 */ 1403 if (inp->inp_lport != 0 || 1404 inp->inp_laddr.s_addr != INADDR_ANY) { 1405 error = EINVAL; /* already bound */ 1406 goto done; 1407 } 1408 1409 if (nam->sa_len != sizeof(*sin)) { 1410 error = EINVAL; 1411 goto done; 1412 } 1413 sin = (struct sockaddr_in *)nam; 1414 1415 cpu = udp_lportcpu(sin->sin_port); 1416 port = netisr_cpuport(cpu); 1417 1418 /* 1419 * See the related comment in tcp_usrreq.c tcp_usr_bind(). 1420 * The exception is that we use local port based netisr 1421 * to serialize in_pcbbind(). 1422 */ 1423 if (&curthread->td_msgport != port) { 1424 lwkt_msg_t lmsg = &msg->bind.base.lmsg; 1425 1426 KASSERT((msg->bind.nm_flags & PRUB_RELINK) == 0, 1427 ("already asked to relink")); 1428 1429 in_pcbunlink(so->so_pcb, &udbinfo[mycpuid]); 1430 msg->bind.nm_flags |= PRUB_RELINK; 1431 1432 /* 1433 * See the related comment in tcp_usrreq.c 1434 * tcp_connect(). 1435 */ 1436 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1437 lwkt_forwardmsg(port, lmsg); 1438 /* msg invalid now */ 1439 return; 1440 } 1441 KASSERT(so->so_port == port, ("so_port is not netisr%d", cpu)); 1442 1443 if (msg->bind.nm_flags & PRUB_RELINK) { 1444 msg->bind.nm_flags &= ~PRUB_RELINK; 1445 in_pcblink(so->so_pcb, &udbinfo[mycpuid]); 1446 } 1447 KASSERT(inp->inp_pcbinfo == &udbinfo[cpu], 1448 ("pcbinfo is not udbinfo%d", cpu)); 1449 1450 error = in_pcbbind(inp, nam, td); 1451 if (error == 0) { 1452 boolean_t forwarded; 1453 1454 if (sin->sin_addr.s_addr != INADDR_ANY) 1455 inp->inp_flags |= INP_WASBOUND_NOTANY; 1456 1457 forwarded = udp_inswildcardhash(inp, 1458 &msg->bind.base, 0); 1459 if (forwarded) { 1460 /* 1461 * The message is further forwarded, so 1462 * we are done here. 1463 */ 1464 return; 1465 } 1466 } 1467 } else { 1468 error = EINVAL; 1469 } 1470 done: 1471 lwkt_replymsg(&msg->bind.base.lmsg, error); 1472 } 1473 1474 static int 1475 udp_preconnect(struct socket *so, const struct sockaddr *nam __unused, 1476 struct thread *td __unused) 1477 { 1478 sosetstate(so, SS_ISCONNECTED); /* XXX */ 1479 return 0; 1480 } 1481 1482 static void 1483 udp_connect(netmsg_t msg) 1484 { 1485 struct socket *so = msg->connect.base.nm_so; 1486 struct sockaddr *nam = msg->connect.nm_nam; 1487 struct thread *td = msg->connect.nm_td; 1488 struct inpcb *inp; 1489 struct sockaddr_in *sin = (struct sockaddr_in *)nam; 1490 struct sockaddr_in *if_sin; 1491 struct lwkt_port *port; 1492 int error; 1493 1494 KKASSERT(msg->connect.nm_m == NULL); 1495 1496 inp = so->so_pcb; 1497 if (inp == NULL) { 1498 error = EINVAL; 1499 goto out; 1500 } 1501 1502 if (msg->connect.nm_flags & PRUC_RECONNECT) { 1503 msg->connect.nm_flags &= ~PRUC_RECONNECT; 1504 in_pcblink(inp, &udbinfo[mycpuid]); 1505 } 1506 1507 if (inp->inp_faddr.s_addr != INADDR_ANY) { 1508 error = EISCONN; 1509 goto out; 1510 } 1511 error = 0; 1512 1513 /* 1514 * Bind if we have to 1515 */ 1516 if (inp->inp_lport == 0) { 1517 error = in_pcbbind(inp, NULL, td); 1518 if (error) 1519 goto out; 1520 } 1521 1522 /* 1523 * Calculate the correct protocol processing thread. The connect 1524 * operation must run there. 1525 */ 1526 error = in_pcbladdr(inp, nam, &if_sin, td); 1527 if (error) 1528 goto out; 1529 if (!prison_remote_ip(td, nam)) { 1530 error = EAFNOSUPPORT; /* IPv6 only jail */ 1531 goto out; 1532 } 1533 1534 port = udp_addrport(sin->sin_addr.s_addr, sin->sin_port, 1535 inp->inp_laddr.s_addr != INADDR_ANY ? 1536 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr, inp->inp_lport); 1537 if (port != &curthread->td_msgport) { 1538 lwkt_msg_t lmsg = &msg->connect.base.lmsg; 1539 int nm_flags = PRUC_RECONNECT; 1540 1541 /* 1542 * in_pcbladdr() may have allocated a route entry for us 1543 * on the current CPU, but we need a route entry on the 1544 * inpcb's owner CPU, so free it here. 1545 */ 1546 in_pcbresetroute(inp); 1547 1548 if (inp->inp_flags & INP_WILDCARD) { 1549 /* 1550 * Remove this inpcb from the wildcard hash before 1551 * the socket's msgport changes. 1552 */ 1553 udp_remwildcardhash(inp); 1554 } 1555 1556 if (so->so_orig_port == NULL) { 1557 /* 1558 * First time change protocol processing port. 1559 * Save the current port for synchronization upon 1560 * udp_detach. 1561 */ 1562 so->so_orig_port = &curthread->td_msgport; 1563 } else { 1564 /* 1565 * We have changed protocol processing port more 1566 * than once. We could not do direct detach 1567 * anymore, because we lose the track of the 1568 * original protocol processing ports to perform 1569 * synchronization upon udp_detach. This should 1570 * be rare though. 1571 */ 1572 inp->inp_flags &= ~INP_DIRECT_DETACH; 1573 } 1574 1575 /* 1576 * We are moving the protocol processing port the socket 1577 * is on, we have to unlink here and re-link on the 1578 * target cpu. 1579 */ 1580 in_pcbunlink(inp, &udbinfo[mycpuid]); 1581 msg->connect.nm_flags |= nm_flags; 1582 1583 /* See the related comment in tcp_usrreq.c tcp_connect() */ 1584 lwkt_setmsg_receipt(lmsg, udp_sosetport); 1585 lwkt_forwardmsg(port, lmsg); 1586 /* msg invalid now */ 1587 return; 1588 } 1589 error = udp_connect_oncpu(inp, sin, if_sin); 1590 out: 1591 if (msg->connect.nm_flags & PRUC_HELDTD) 1592 lwkt_rele(td); 1593 if (error && (msg->connect.nm_flags & PRUC_ASYNC)) { 1594 if (inp->inp_lport == 0) { 1595 /* 1596 * As long as we have the local port, it is fine 1597 * for connect to fail, e.g. disconnect. 1598 */ 1599 so->so_error = error; 1600 } 1601 soclrstate(so, SS_ISCONNECTED); 1602 /* 1603 * Wake up callers blocked on this socket to make sure 1604 * that they can see this error. 1605 * 1606 * NOTE: 1607 * sodisconnected() can't be used here, which bricks 1608 * sending and receiving. 1609 */ 1610 wakeup(&so->so_timeo); 1611 sowwakeup(so); 1612 sorwakeup(so); 1613 } 1614 if (error && inp != NULL && inp->inp_lport != 0 && 1615 (inp->inp_flags & INP_WILDCARD) == 0) { 1616 boolean_t forwarded; 1617 1618 /* Connect failed; put it to wildcard hash. */ 1619 forwarded = udp_inswildcardhash(inp, &msg->connect.base, 1620 error); 1621 if (forwarded) { 1622 /* 1623 * The message is further forwarded, so we are done 1624 * here. 1625 */ 1626 return; 1627 } 1628 } 1629 lwkt_replymsg(&msg->connect.base.lmsg, error); 1630 } 1631 1632 static void 1633 udp_remwildcardhash(struct inpcb *inp) 1634 { 1635 int cpu; 1636 1637 KASSERT(inp->inp_pcbinfo == &udbinfo[mycpuid], 1638 ("not on owner cpu")); 1639 1640 for (cpu = 0; cpu < ncpus2; ++cpu) { 1641 if (cpu == mycpuid) { 1642 /* 1643 * This inpcb will be removed by the later 1644 * in_pcbremwildcardhash(). 1645 */ 1646 continue; 1647 } 1648 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpu]); 1649 } 1650 in_pcbremwildcardhash(inp); 1651 } 1652 1653 static int 1654 udp_connect_oncpu(struct inpcb *inp, struct sockaddr_in *sin, 1655 struct sockaddr_in *if_sin) 1656 { 1657 struct socket *so = inp->inp_socket; 1658 struct inpcb *oinp; 1659 1660 oinp = in_pcblookup_hash(inp->inp_pcbinfo, 1661 sin->sin_addr, sin->sin_port, 1662 inp->inp_laddr.s_addr != INADDR_ANY ? 1663 inp->inp_laddr : if_sin->sin_addr, inp->inp_lport, FALSE, NULL); 1664 if (oinp != NULL) 1665 return EADDRINUSE; 1666 1667 /* 1668 * No more errors can occur, finish adjusting the socket 1669 * and change the processing port to reflect the connected 1670 * socket. Once set we can no longer safely mess with the 1671 * socket. 1672 */ 1673 1674 if (inp->inp_flags & INP_WILDCARD) 1675 udp_remwildcardhash(inp); 1676 1677 if (inp->inp_laddr.s_addr == INADDR_ANY) 1678 inp->inp_laddr = if_sin->sin_addr; 1679 inp->inp_faddr = sin->sin_addr; 1680 inp->inp_fport = sin->sin_port; 1681 in_pcbinsconnhash(inp); 1682 1683 soisconnected(so); 1684 1685 return 0; 1686 } 1687 1688 static void 1689 udp_detach2(struct socket *so) 1690 { 1691 in_pcbdetach(so->so_pcb); 1692 sodiscard(so); 1693 sofree(so); 1694 } 1695 1696 static void 1697 udp_detach_final_dispatch(netmsg_t msg) 1698 { 1699 udp_detach2(msg->base.nm_so); 1700 } 1701 1702 static void 1703 udp_detach_oncpu_dispatch(netmsg_t msg) 1704 { 1705 struct netmsg_base *clomsg = &msg->base; 1706 struct socket *so = clomsg->nm_so; 1707 struct inpcb *inp = so->so_pcb; 1708 struct thread *td = curthread; 1709 int nextcpu, cpuid = mycpuid; 1710 1711 KASSERT(td->td_type == TD_TYPE_NETISR, ("not in netisr")); 1712 1713 if (inp->inp_flags & INP_WILDCARD) { 1714 /* 1715 * This inp will be removed on the inp's 1716 * owner CPU later, so don't do it now. 1717 */ 1718 if (&td->td_msgport != so->so_port) 1719 in_pcbremwildcardhash_oncpu(inp, &udbinfo[cpuid]); 1720 } 1721 1722 if (cpuid == 0) { 1723 /* 1724 * Free and clear multicast socket option, 1725 * which is only accessed in netisr0. 1726 */ 1727 ip_freemoptions(inp->inp_moptions); 1728 inp->inp_moptions = NULL; 1729 } 1730 1731 nextcpu = cpuid + 1; 1732 if (nextcpu < ncpus2) { 1733 lwkt_forwardmsg(netisr_cpuport(nextcpu), &clomsg->lmsg); 1734 } else { 1735 /* 1736 * No one could see this inpcb now; destroy this 1737 * inpcb in its owner netisr. 1738 */ 1739 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1740 udp_detach_final_dispatch); 1741 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1742 } 1743 } 1744 1745 static void 1746 udp_detach_syncorig_dispatch(netmsg_t msg) 1747 { 1748 struct netmsg_base *clomsg = &msg->base; 1749 struct socket *so = clomsg->nm_so; 1750 1751 /* 1752 * Original protocol processing port is synchronized; 1753 * destroy this inpcb in its owner netisr. 1754 */ 1755 netmsg_init(clomsg, so, &netisr_apanic_rport, 0, 1756 udp_detach_final_dispatch); 1757 lwkt_sendmsg(so->so_port, &clomsg->lmsg); 1758 } 1759 1760 static void 1761 udp_detach(netmsg_t msg) 1762 { 1763 struct socket *so = msg->detach.base.nm_so; 1764 struct netmsg_base *clomsg; 1765 struct inpcb *inp; 1766 1767 inp = so->so_pcb; 1768 if (inp == NULL) { 1769 lwkt_replymsg(&msg->detach.base.lmsg, EINVAL); 1770 return; 1771 } 1772 1773 /* 1774 * Reply EJUSTRETURN ASAP, we will call sodiscard() and 1775 * sofree() later. 1776 */ 1777 lwkt_replymsg(&msg->detach.base.lmsg, EJUSTRETURN); 1778 1779 if (ncpus2 == 1) { 1780 /* Only one CPU, detach the inpcb directly. */ 1781 udp_detach2(so); 1782 return; 1783 } 1784 1785 /* 1786 * Remove this inpcb from the inpcb list first, so that 1787 * no one could find this inpcb from the inpcb list. 1788 */ 1789 in_pcbofflist(inp); 1790 1791 /* 1792 * Remove this inpcb from the local port hash directly 1793 * here, so that its bound local port could be recycled 1794 * timely. 1795 */ 1796 in_pcbremporthash(inp); 1797 1798 if (inp->inp_flags & INP_DIRECT_DETACH) { 1799 /* 1800 * Direct detaching is allowed 1801 */ 1802 KASSERT((inp->inp_flags & INP_WILDCARD) == 0, 1803 ("in the wildcardhash")); 1804 KASSERT(inp->inp_moptions == NULL, ("has mcast options")); 1805 if (so->so_orig_port == NULL) { 1806 udp_detach2(so); 1807 } else { 1808 /* 1809 * Protocol processing port changed once, so 1810 * we need to make sure that there are nothing 1811 * left on the original protocol processing 1812 * port before we destroy this socket and inpcb. 1813 * This is more lightweight than going through 1814 * all UDP processing netisrs. 1815 */ 1816 clomsg = &so->so_clomsg; 1817 netmsg_init(clomsg, so, &netisr_apanic_rport, 1818 MSGF_IGNSOPORT, udp_detach_syncorig_dispatch); 1819 lwkt_sendmsg(so->so_orig_port, &clomsg->lmsg); 1820 } 1821 return; 1822 } 1823 1824 /* 1825 * Go through netisrs which process UDP to make sure 1826 * no one could find this inpcb anymore. 1827 */ 1828 clomsg = &so->so_clomsg; 1829 netmsg_init(clomsg, so, &netisr_apanic_rport, MSGF_IGNSOPORT, 1830 udp_detach_oncpu_dispatch); 1831 lwkt_sendmsg(netisr_cpuport(0), &clomsg->lmsg); 1832 } 1833 1834 static void 1835 udp_disconnect(netmsg_t msg) 1836 { 1837 struct socket *so = msg->disconnect.base.nm_so; 1838 struct inpcb *inp; 1839 boolean_t forwarded; 1840 int error = 0; 1841 1842 inp = so->so_pcb; 1843 if (inp == NULL) { 1844 error = EINVAL; 1845 goto out; 1846 } 1847 if (inp->inp_faddr.s_addr == INADDR_ANY) { 1848 error = ENOTCONN; 1849 goto out; 1850 } 1851 1852 soclrstate(so, SS_ISCONNECTED); /* XXX */ 1853 1854 in_pcbdisconnect(inp); 1855 1856 /* 1857 * Follow traditional BSD behavior and retain the local port 1858 * binding. But, fix the old misbehavior of overwriting any 1859 * previously bound local address. 1860 */ 1861 if (!(inp->inp_flags & INP_WASBOUND_NOTANY)) 1862 inp->inp_laddr.s_addr = INADDR_ANY; 1863 1864 if (so->so_state & SS_ISCLOSING) { 1865 /* 1866 * If this socket is being closed, there is no need 1867 * to put this socket back into wildcard hash table. 1868 */ 1869 error = 0; 1870 goto out; 1871 } 1872 1873 forwarded = udp_inswildcardhash(inp, &msg->disconnect.base, 0); 1874 if (forwarded) { 1875 /* 1876 * The message is further forwarded, so we are done 1877 * here. 1878 */ 1879 return; 1880 } 1881 out: 1882 lwkt_replymsg(&msg->disconnect.base.lmsg, error); 1883 } 1884 1885 void 1886 udp_shutdown(netmsg_t msg) 1887 { 1888 struct socket *so = msg->shutdown.base.nm_so; 1889 struct inpcb *inp; 1890 int error; 1891 1892 inp = so->so_pcb; 1893 if (inp) { 1894 socantsendmore(so); 1895 error = 0; 1896 } else { 1897 error = EINVAL; 1898 } 1899 lwkt_replymsg(&msg->shutdown.base.lmsg, error); 1900 } 1901 1902 struct pr_usrreqs udp_usrreqs = { 1903 .pru_abort = udp_abort, 1904 .pru_accept = pr_generic_notsupp, 1905 .pru_attach = udp_attach, 1906 .pru_bind = udp_bind, 1907 .pru_connect = udp_connect, 1908 .pru_connect2 = pr_generic_notsupp, 1909 .pru_control = in_control_dispatch, 1910 .pru_detach = udp_detach, 1911 .pru_disconnect = udp_disconnect, 1912 .pru_listen = pr_generic_notsupp, 1913 .pru_peeraddr = in_setpeeraddr_dispatch, 1914 .pru_rcvd = pr_generic_notsupp, 1915 .pru_rcvoob = pr_generic_notsupp, 1916 .pru_send = udp_send, 1917 .pru_sense = pru_sense_null, 1918 .pru_shutdown = udp_shutdown, 1919 .pru_sockaddr = in_setsockaddr_dispatch, 1920 .pru_sosend = sosendudp, 1921 .pru_soreceive = soreceive, 1922 .pru_preconnect = udp_preconnect, 1923 .pru_preattach = udp_preattach 1924 }; 1925