sys/netinet/in.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1991, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  * Copyright (C) 2001 WIDE Project.  All rights reserved.
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 4. Neither the name of the University nor the names of its contributors
  15  *    may be used to endorse or promote products derived from this software
  16  *    without specific prior written permission.
  17  *
  18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  28  * SUCH DAMAGE.
  29  *
  30  *      @(#)in.c        8.4 (Berkeley) 1/9/95
  31  */
  32
  33 #include <sys/cdefs.h>
  34 __FBSDID("$FreeBSD$");
  35
  36 #include "opt_mpath.h"
  37
  38 #include <sys/param.h>
  39 #include <sys/systm.h>
  40 #include <sys/sockio.h>
  41 #include <sys/malloc.h>
  42 #include <sys/priv.h>
  43 #include <sys/socket.h>
  44 #include <sys/jail.h>
  45 #include <sys/kernel.h>
  46 #include <sys/proc.h>
  47 #include <sys/sysctl.h>
  48 #include <sys/syslog.h>
  49
  50 #include <net/if.h>
  51 #include <net/if_var.h>
  52 #include <net/if_arp.h>
  53 #include <net/if_dl.h>
  54 #include <net/if_llatbl.h>
  55 #include <net/if_types.h>
  56 #include <net/route.h>
  57 #include <net/vnet.h>
  58
  59 #include <netinet/in.h>
  60 #include <netinet/in_var.h>
  61 #include <netinet/in_pcb.h>
  62 #include <netinet/ip_var.h>
  63 #include <netinet/igmp_var.h>
  64 #include <netinet/udp.h>
  65 #include <netinet/udp_var.h>
  66
  67 static int in_mask2len(struct in_addr *);
  68 static void in_len2mask(struct in_addr *, int);
  69 static int in_lifaddr_ioctl(struct socket *, u_long, caddr_t,
  70         struct ifnet *, struct thread *);
  71
  72 static int      in_addprefix(struct in_ifaddr *, int);
  73 static int      in_scrubprefix(struct in_ifaddr *, u_int);
  74 static void     in_socktrim(struct sockaddr_in *);
  75 static int      in_ifinit(struct ifnet *,
  76             struct in_ifaddr *, struct sockaddr_in *, int);
  77 static void     in_purgemaddrs(struct ifnet *);
  78
  79 static VNET_DEFINE(int, sameprefixcarponly);
  80 #define V_sameprefixcarponly            VNET(sameprefixcarponly)
  81 SYSCTL_VNET_INT(_net_inet_ip, OID_AUTO, same_prefix_carp_only, CTLFLAG_RW,
  82         &VNET_NAME(sameprefixcarponly), 0,
  83         "Refuse to create same prefixes on different interfaces");
  84
  85 VNET_DECLARE(struct inpcbinfo, ripcbinfo);
  86 #define V_ripcbinfo                     VNET(ripcbinfo)
  87
  88 VNET_DECLARE(struct arpstat, arpstat);  /* ARP statistics, see if_arp.h */
  89 #define V_arpstat               VNET(arpstat)
  90
  91 /*
  92  * Return 1 if an internet address is for a ``local'' host
  93  * (one to which we have a connection).
  94  */
  95 int
  96 in_localaddr(struct in_addr in)
  97 {
  98         register u_long i = ntohl(in.s_addr);
  99         register struct in_ifaddr *ia;
 100
 101         IN_IFADDR_RLOCK();
 102         TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
 103                 if ((i & ia->ia_subnetmask) == ia->ia_subnet) {
 104                         IN_IFADDR_RUNLOCK();
 105                         return (1);
 106                 }
 107         }
 108         IN_IFADDR_RUNLOCK();
 109         return (0);
 110 }
 111
 112 /*
 113  * Return 1 if an internet address is for the local host and configured
 114  * on one of its interfaces.
 115  */
 116 int
 117 in_localip(struct in_addr in)
 118 {
 119         struct in_ifaddr *ia;
 120
 121         IN_IFADDR_RLOCK();
 122         LIST_FOREACH(ia, INADDR_HASH(in.s_addr), ia_hash) {
 123                 if (IA_SIN(ia)->sin_addr.s_addr == in.s_addr) {
 124                         IN_IFADDR_RUNLOCK();
 125                         return (1);
 126                 }
 127         }
 128         IN_IFADDR_RUNLOCK();
 129         return (0);
 130 }
 131
 132 /*
 133  * Determine whether an IP address is in a reserved set of addresses
 134  * that may not be forwarded, or whether datagrams to that destination
 135  * may be forwarded.
 136  */
 137 int
 138 in_canforward(struct in_addr in)
 139 {
 140         register u_long i = ntohl(in.s_addr);
 141         register u_long net;
 142
 143         if (IN_EXPERIMENTAL(i) || IN_MULTICAST(i) || IN_LINKLOCAL(i))
 144                 return (0);
 145         if (IN_CLASSA(i)) {
 146                 net = i & IN_CLASSA_NET;
 147                 if (net == 0 || net == (IN_LOOPBACKNET << IN_CLASSA_NSHIFT))
 148                         return (0);
 149         }
 150         return (1);
 151 }
 152
 153 /*
 154  * Trim a mask in a sockaddr
 155  */
 156 static void
 157 in_socktrim(struct sockaddr_in *ap)
 158 {
 159     register char *cplim = (char *) &ap->sin_addr;
 160     register char *cp = (char *) (&ap->sin_addr + 1);
 161
 162     ap->sin_len = 0;
 163     while (--cp >= cplim)
 164         if (*cp) {
 165             (ap)->sin_len = cp - (char *) (ap) + 1;
 166             break;
 167         }
 168 }
 169
 170 static int
 171 in_mask2len(mask)
 172         struct in_addr *mask;
 173 {
 174         int x, y;
 175         u_char *p;
 176
 177         p = (u_char *)mask;
 178         for (x = 0; x < sizeof(*mask); x++) {
 179                 if (p[x] != 0xff)
 180                         break;
 181         }
 182         y = 0;
 183         if (x < sizeof(*mask)) {
 184                 for (y = 0; y < 8; y++) {
 185                         if ((p[x] & (0x80 >> y)) == 0)
 186                                 break;
 187                 }
 188         }
 189         return (x * 8 + y);
 190 }
 191
 192 static void
 193 in_len2mask(struct in_addr *mask, int len)
 194 {
 195         int i;
 196         u_char *p;
 197
 198         p = (u_char *)mask;
 199         bzero(mask, sizeof(*mask));
 200         for (i = 0; i < len / 8; i++)
 201                 p[i] = 0xff;
 202         if (len % 8)
 203                 p[i] = (0xff00 >> (len % 8)) & 0xff;
 204 }
 205
 206 /*
 207  * Generic internet control operations (ioctl's).
 208  *
 209  * ifp is NULL if not an interface-specific ioctl.
 210  */
 211 /* ARGSUSED */
 212 int
 213 in_control(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp,
 214     struct thread *td)
 215 {
 216         register struct ifreq *ifr = (struct ifreq *)data;
 217         register struct in_ifaddr *ia, *iap;
 218         register struct ifaddr *ifa;
 219         struct in_addr allhosts_addr;
 220         struct in_addr dst;
 221         struct in_ifinfo *ii;
 222         struct in_aliasreq *ifra = (struct in_aliasreq *)data;
 223         struct sockaddr_in oldaddr;
 224         int error, hostIsNew, iaIsNew, maskIsNew;
 225         int iaIsFirst;
 226
 227         ia = NULL;
 228         iaIsFirst = 0;
 229         iaIsNew = 0;
 230         allhosts_addr.s_addr = htonl(INADDR_ALLHOSTS_GROUP);
 231
 232         /*
 233          * Filter out ioctls we implement directly; forward the rest on to
 234          * in_lifaddr_ioctl() and ifp->if_ioctl().
 235          */
 236         switch (cmd) {
 237         case SIOCAIFADDR:
 238         case SIOCDIFADDR:
 239         case SIOCGIFADDR:
 240         case SIOCGIFBRDADDR:
 241         case SIOCGIFDSTADDR:
 242         case SIOCGIFNETMASK:
 243         case SIOCSIFADDR:
 244         case SIOCSIFBRDADDR:
 245         case SIOCSIFDSTADDR:
 246         case SIOCSIFNETMASK:
 247                 break;
 248
 249         case SIOCALIFADDR:
 250                 if (td != NULL) {
 251                         error = priv_check(td, PRIV_NET_ADDIFADDR);
 252                         if (error)
 253                                 return (error);
 254                 }
 255                 if (ifp == NULL)
 256                         return (EINVAL);
 257                 return in_lifaddr_ioctl(so, cmd, data, ifp, td);
 258
 259         case SIOCDLIFADDR:
 260                 if (td != NULL) {
 261                         error = priv_check(td, PRIV_NET_DELIFADDR);
 262                         if (error)
 263                                 return (error);
 264                 }
 265                 if (ifp == NULL)
 266                         return (EINVAL);
 267                 return in_lifaddr_ioctl(so, cmd, data, ifp, td);
 268
 269         case SIOCGLIFADDR:
 270                 if (ifp == NULL)
 271                         return (EINVAL);
 272                 return in_lifaddr_ioctl(so, cmd, data, ifp, td);
 273
 274         default:
 275                 if (ifp == NULL || ifp->if_ioctl == NULL)
 276                         return (EOPNOTSUPP);
 277                 return ((*ifp->if_ioctl)(ifp, cmd, data));
 278         }
 279
 280         if (ifp == NULL)
 281                 return (EADDRNOTAVAIL);
 282
 283         /*
 284          * Security checks before we get involved in any work.
 285          */
 286         switch (cmd) {
 287         case SIOCAIFADDR:
 288         case SIOCSIFADDR:
 289         case SIOCSIFBRDADDR:
 290         case SIOCSIFNETMASK:
 291         case SIOCSIFDSTADDR:
 292                 if (td != NULL) {
 293                         error = priv_check(td, PRIV_NET_ADDIFADDR);
 294                         if (error)
 295                                 return (error);
 296                 }
 297                 break;
 298
 299         case SIOCDIFADDR:
 300                 if (td != NULL) {
 301                         error = priv_check(td, PRIV_NET_DELIFADDR);
 302                         if (error)
 303                                 return (error);
 304                 }
 305                 break;
 306         }
 307
 308         /*
 309          * Find address for this interface, if it exists.
 310          *
 311          * If an alias address was specified, find that one instead of the
 312          * first one on the interface, if possible.
 313          */
 314         dst = ((struct sockaddr_in *)&ifr->ifr_addr)->sin_addr;
 315         IN_IFADDR_RLOCK();
 316         LIST_FOREACH(iap, INADDR_HASH(dst.s_addr), ia_hash) {
 317                 if (iap->ia_ifp == ifp &&
 318                     iap->ia_addr.sin_addr.s_addr == dst.s_addr) {
 319                         if (td == NULL || prison_check_ip4(td->td_ucred,
 320                             &dst) == 0)
 321                                 ia = iap;
 322                         break;
 323                 }
 324         }
 325         if (ia != NULL)
 326                 ifa_ref(&ia->ia_ifa);
 327         IN_IFADDR_RUNLOCK();
 328         if (ia == NULL) {
 329                 IF_ADDR_RLOCK(ifp);
 330                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 331                         iap = ifatoia(ifa);
 332                         if (iap->ia_addr.sin_family == AF_INET) {
 333                                 if (td != NULL &&
 334                                     prison_check_ip4(td->td_ucred,
 335                                     &iap->ia_addr.sin_addr) != 0)
 336                                         continue;
 337                                 ia = iap;
 338                                 break;
 339                         }
 340                 }
 341                 if (ia != NULL)
 342                         ifa_ref(&ia->ia_ifa);
 343                 IF_ADDR_RUNLOCK(ifp);
 344         }
 345         if (ia == NULL)
 346                 iaIsFirst = 1;
 347
 348         error = 0;
 349         switch (cmd) {
 350         case SIOCAIFADDR:
 351         case SIOCDIFADDR:
 352                 if (ifra->ifra_addr.sin_family == AF_INET) {
 353                         struct in_ifaddr *oia;
 354
 355                         IN_IFADDR_RLOCK();
 356                         for (oia = ia; ia; ia = TAILQ_NEXT(ia, ia_link)) {
 357                                 if (ia->ia_ifp == ifp  &&
 358                                     ia->ia_addr.sin_addr.s_addr ==
 359                                     ifra->ifra_addr.sin_addr.s_addr)
 360                                         break;
 361                         }
 362                         if (ia != NULL && ia != oia)
 363                                 ifa_ref(&ia->ia_ifa);
 364                         if (oia != NULL && ia != oia)
 365                                 ifa_free(&oia->ia_ifa);
 366                         IN_IFADDR_RUNLOCK();
 367                         if ((ifp->if_flags & IFF_POINTOPOINT)
 368                             && (cmd == SIOCAIFADDR)
 369                             && (ifra->ifra_dstaddr.sin_addr.s_addr
 370                                 == INADDR_ANY)) {
 371                                 error = EDESTADDRREQ;
 372                                 goto out;
 373                         }
 374                 }
 375                 if (cmd == SIOCDIFADDR && ia == NULL) {
 376                         error = EADDRNOTAVAIL;
 377                         goto out;
 378                 }
 379                 /* FALLTHROUGH */
 380         case SIOCSIFADDR:
 381         case SIOCSIFNETMASK:
 382         case SIOCSIFDSTADDR:
 383                 if (ia == NULL) {
 384                         ia = (struct in_ifaddr *)
 385                                 malloc(sizeof *ia, M_IFADDR, M_NOWAIT |
 386                                     M_ZERO);
 387                         if (ia == NULL) {
 388                                 error = ENOBUFS;
 389                                 goto out;
 390                         }
 391
 392                         ifa = &ia->ia_ifa;
 393                         ifa_init(ifa);
 394                         ifa->ifa_addr = (struct sockaddr *)&ia->ia_addr;
 395                         ifa->ifa_dstaddr = (struct sockaddr *)&ia->ia_dstaddr;
 396                         ifa->ifa_netmask = (struct sockaddr *)&ia->ia_sockmask;
 397
 398                         ia->ia_sockmask.sin_len = 8;
 399                         ia->ia_sockmask.sin_family = AF_INET;
 400                         if (ifp->if_flags & IFF_BROADCAST) {
 401                                 ia->ia_broadaddr.sin_len = sizeof(ia->ia_addr);
 402                                 ia->ia_broadaddr.sin_family = AF_INET;
 403                         }
 404                         ia->ia_ifp = ifp;
 405
 406                         ifa_ref(ifa);                   /* if_addrhead */
 407                         IF_ADDR_WLOCK(ifp);
 408                         TAILQ_INSERT_TAIL(&ifp->if_addrhead, ifa, ifa_link);
 409                         IF_ADDR_WUNLOCK(ifp);
 410                         ifa_ref(ifa);                   /* in_ifaddrhead */
 411                         IN_IFADDR_WLOCK();
 412                         TAILQ_INSERT_TAIL(&V_in_ifaddrhead, ia, ia_link);
 413                         IN_IFADDR_WUNLOCK();
 414                         iaIsNew = 1;
 415                 }
 416                 break;
 417
 418         case SIOCSIFBRDADDR:
 419         case SIOCGIFADDR:
 420         case SIOCGIFNETMASK:
 421         case SIOCGIFDSTADDR:
 422         case SIOCGIFBRDADDR:
 423                 if (ia == NULL) {
 424                         error = EADDRNOTAVAIL;
 425                         goto out;
 426                 }
 427                 break;
 428         }
 429
 430         /*
 431          * Most paths in this switch return directly or via out.  Only paths
 432          * that remove the address break in order to hit common removal code.
 433          */
 434         switch (cmd) {
 435         case SIOCGIFADDR:
 436                 *((struct sockaddr_in *)&ifr->ifr_addr) = ia->ia_addr;
 437                 goto out;
 438
 439         case SIOCGIFBRDADDR:
 440                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 441                         error = EINVAL;
 442                         goto out;
 443                 }
 444                 *((struct sockaddr_in *)&ifr->ifr_dstaddr) = ia->ia_broadaddr;
 445                 goto out;
 446
 447         case SIOCGIFDSTADDR:
 448                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
 449                         error = EINVAL;
 450                         goto out;
 451                 }
 452                 *((struct sockaddr_in *)&ifr->ifr_dstaddr) = ia->ia_dstaddr;
 453                 goto out;
 454
 455         case SIOCGIFNETMASK:
 456                 *((struct sockaddr_in *)&ifr->ifr_addr) = ia->ia_sockmask;
 457                 goto out;
 458
 459         case SIOCSIFDSTADDR:
 460                 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) {
 461                         error = EINVAL;
 462                         goto out;
 463                 }
 464                 oldaddr = ia->ia_dstaddr;
 465                 ia->ia_dstaddr = *(struct sockaddr_in *)&ifr->ifr_dstaddr;
 466                 if (ifp->if_ioctl != NULL) {
 467                         error = (*ifp->if_ioctl)(ifp, SIOCSIFDSTADDR,
 468                             (caddr_t)ia);
 469                         if (error) {
 470                                 ia->ia_dstaddr = oldaddr;
 471                                 goto out;
 472                         }
 473                 }
 474                 if (ia->ia_flags & IFA_ROUTE) {
 475                         ia->ia_ifa.ifa_dstaddr = (struct sockaddr *)&oldaddr;
 476                         rtinit(&(ia->ia_ifa), (int)RTM_DELETE, RTF_HOST);
 477                         ia->ia_ifa.ifa_dstaddr =
 478                                         (struct sockaddr *)&ia->ia_dstaddr;
 479                         rtinit(&(ia->ia_ifa), (int)RTM_ADD, RTF_HOST|RTF_UP);
 480                 }
 481                 goto out;
 482
 483         case SIOCSIFBRDADDR:
 484                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 485                         error = EINVAL;
 486                         goto out;
 487                 }
 488                 ia->ia_broadaddr = *(struct sockaddr_in *)&ifr->ifr_broadaddr;
 489                 goto out;
 490
 491         case SIOCSIFADDR:
 492                 error = in_ifinit(ifp, ia,
 493                     (struct sockaddr_in *) &ifr->ifr_addr, 1);
 494                 if (error != 0 && iaIsNew)
 495                         break;
 496                 if (error == 0) {
 497                         ii = ((struct in_ifinfo *)ifp->if_afdata[AF_INET]);
 498                         if (iaIsFirst &&
 499                             (ifp->if_flags & IFF_MULTICAST) != 0) {
 500                                 error = in_joingroup(ifp, &allhosts_addr,
 501                                     NULL, &ii->ii_allhosts);
 502                         }
 503                         EVENTHANDLER_INVOKE(ifaddr_event, ifp);
 504                 }
 505                 error = 0;
 506                 goto out;
 507
 508         case SIOCSIFNETMASK:
 509                 ia->ia_sockmask.sin_addr = ifra->ifra_addr.sin_addr;
 510                 ia->ia_subnetmask = ntohl(ia->ia_sockmask.sin_addr.s_addr);
 511                 goto out;
 512
 513         case SIOCAIFADDR:
 514                 maskIsNew = 0;
 515                 hostIsNew = 1;
 516                 error = 0;
 517                 if (ia->ia_addr.sin_family == AF_INET) {
 518                         if (ifra->ifra_addr.sin_len == 0) {
 519                                 ifra->ifra_addr = ia->ia_addr;
 520                                 hostIsNew = 0;
 521                         } else if (ifra->ifra_addr.sin_addr.s_addr ==
 522                                                ia->ia_addr.sin_addr.s_addr)
 523                                 hostIsNew = 0;
 524                 }
 525                 if (ifra->ifra_mask.sin_len) {
 526                         /*
 527                          * QL: XXX
 528                          * Need to scrub the prefix here in case
 529                          * the issued command is SIOCAIFADDR with
 530                          * the same address, but with a different
 531                          * prefix length. And if the prefix length
 532                          * is the same as before, then the call is
 533                          * un-necessarily executed here.
 534                          */
 535                         in_ifscrub(ifp, ia, LLE_STATIC);
 536                         ia->ia_sockmask = ifra->ifra_mask;
 537                         ia->ia_sockmask.sin_family = AF_INET;
 538                         ia->ia_subnetmask =
 539                             ntohl(ia->ia_sockmask.sin_addr.s_addr);
 540                         maskIsNew = 1;
 541                 }
 542                 if ((ifp->if_flags & IFF_POINTOPOINT) &&
 543                     (ifra->ifra_dstaddr.sin_family == AF_INET)) {
 544                         in_ifscrub(ifp, ia, LLE_STATIC);
 545                         ia->ia_dstaddr = ifra->ifra_dstaddr;
 546                         maskIsNew  = 1; /* We lie; but the effect's the same */
 547                 }
 548                 if (ifra->ifra_addr.sin_family == AF_INET &&
 549                     (hostIsNew || maskIsNew))
 550                         error = in_ifinit(ifp, ia, &ifra->ifra_addr, 0);
 551                 if (error != 0 && iaIsNew)
 552                         break;
 553
 554                 if ((ifp->if_flags & IFF_BROADCAST) &&
 555                     (ifra->ifra_broadaddr.sin_family == AF_INET))
 556                         ia->ia_broadaddr = ifra->ifra_broadaddr;
 557                 if (error == 0) {
 558                         ii = ((struct in_ifinfo *)ifp->if_afdata[AF_INET]);
 559                         if (iaIsFirst &&
 560                             (ifp->if_flags & IFF_MULTICAST) != 0) {
 561                                 error = in_joingroup(ifp, &allhosts_addr,
 562                                     NULL, &ii->ii_allhosts);
 563                         }
 564                         EVENTHANDLER_INVOKE(ifaddr_event, ifp);
 565                 }
 566                 goto out;
 567
 568         case SIOCDIFADDR:
 569                 /*
 570                  * in_ifscrub kills the interface route.
 571                  */
 572                 in_ifscrub(ifp, ia, LLE_STATIC);
 573
 574                 /*
 575                  * in_ifadown gets rid of all the rest of
 576                  * the routes.  This is not quite the right
 577                  * thing to do, but at least if we are running
 578                  * a routing process they will come back.
 579                  */
 580                 in_ifadown(&ia->ia_ifa, 1);
 581                 EVENTHANDLER_INVOKE(ifaddr_event, ifp);
 582                 error = 0;
 583                 break;
 584
 585         default:
 586                 panic("in_control: unsupported ioctl");
 587         }
 588
 589         IF_ADDR_WLOCK(ifp);
 590         /* Re-check that ia is still part of the list. */
 591         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 592                 if (ifa == &ia->ia_ifa)
 593                         break;
 594         }
 595         if (ifa == NULL) {
 596                 /*
 597                  * If we lost the race with another thread, there is no need to
 598                  * try it again for the next loop as there is no other exit
 599                  * path between here and out.
 600                  */
 601                 IF_ADDR_WUNLOCK(ifp);
 602                 error = EADDRNOTAVAIL;
 603                 goto out;
 604         }
 605         TAILQ_REMOVE(&ifp->if_addrhead, &ia->ia_ifa, ifa_link);
 606         IF_ADDR_WUNLOCK(ifp);
 607         ifa_free(&ia->ia_ifa);                          /* if_addrhead */
 608
 609         IN_IFADDR_WLOCK();
 610         TAILQ_REMOVE(&V_in_ifaddrhead, ia, ia_link);
 611         if (ia->ia_addr.sin_family == AF_INET) {
 612                 struct in_ifaddr *if_ia;
 613
 614                 LIST_REMOVE(ia, ia_hash);
 615                 IN_IFADDR_WUNLOCK();
 616                 /*
 617                  * If this is the last IPv4 address configured on this
 618                  * interface, leave the all-hosts group.
 619                  * No state-change report need be transmitted.
 620                  */
 621                 if_ia = NULL;
 622                 IFP_TO_IA(ifp, if_ia);
 623                 if (if_ia == NULL) {
 624                         ii = ((struct in_ifinfo *)ifp->if_afdata[AF_INET]);
 625                         IN_MULTI_LOCK();
 626                         if (ii->ii_allhosts) {
 627                                 (void)in_leavegroup_locked(ii->ii_allhosts,
 628                                     NULL);
 629                                 ii->ii_allhosts = NULL;
 630                         }
 631                         IN_MULTI_UNLOCK();
 632                 } else
 633                         ifa_free(&if_ia->ia_ifa);
 634         } else
 635                 IN_IFADDR_WUNLOCK();
 636         ifa_free(&ia->ia_ifa);                          /* in_ifaddrhead */
 637 out:
 638         if (ia != NULL)
 639                 ifa_free(&ia->ia_ifa);
 640         return (error);
 641 }
 642
 643 /*
 644  * SIOC[GAD]LIFADDR.
 645  *      SIOCGLIFADDR: get first address. (?!?)
 646  *      SIOCGLIFADDR with IFLR_PREFIX:
 647  *              get first address that matches the specified prefix.
 648  *      SIOCALIFADDR: add the specified address.
 649  *      SIOCALIFADDR with IFLR_PREFIX:
 650  *              EINVAL since we can't deduce hostid part of the address.
 651  *      SIOCDLIFADDR: delete the specified address.
 652  *      SIOCDLIFADDR with IFLR_PREFIX:
 653  *              delete the first address that matches the specified prefix.
 654  * return values:
 655  *      EINVAL on invalid parameters
 656  *      EADDRNOTAVAIL on prefix match failed/specified address not found
 657  *      other values may be returned from in_ioctl()
 658  */
 659 static int
 660 in_lifaddr_ioctl(struct socket *so, u_long cmd, caddr_t data,
 661     struct ifnet *ifp, struct thread *td)
 662 {
 663         struct if_laddrreq *iflr = (struct if_laddrreq *)data;
 664         struct ifaddr *ifa;
 665
 666         /* sanity checks */
 667         if (data == NULL || ifp == NULL) {
 668                 panic("invalid argument to in_lifaddr_ioctl");
 669                 /*NOTRECHED*/
 670         }
 671
 672         switch (cmd) {
 673         case SIOCGLIFADDR:
 674                 /* address must be specified on GET with IFLR_PREFIX */
 675                 if ((iflr->flags & IFLR_PREFIX) == 0)
 676                         break;
 677                 /*FALLTHROUGH*/
 678         case SIOCALIFADDR:
 679         case SIOCDLIFADDR:
 680                 /* address must be specified on ADD and DELETE */
 681                 if (iflr->addr.ss_family != AF_INET)
 682                         return (EINVAL);
 683                 if (iflr->addr.ss_len != sizeof(struct sockaddr_in))
 684                         return (EINVAL);
 685                 /* XXX need improvement */
 686                 if (iflr->dstaddr.ss_family
 687                  && iflr->dstaddr.ss_family != AF_INET)
 688                         return (EINVAL);
 689                 if (iflr->dstaddr.ss_family
 690                  && iflr->dstaddr.ss_len != sizeof(struct sockaddr_in))
 691                         return (EINVAL);
 692                 break;
 693         default: /*shouldn't happen*/
 694                 return (EOPNOTSUPP);
 695         }
 696         if (sizeof(struct in_addr) * 8 < iflr->prefixlen)
 697                 return (EINVAL);
 698
 699         switch (cmd) {
 700         case SIOCALIFADDR:
 701             {
 702                 struct in_aliasreq ifra;
 703
 704                 if (iflr->flags & IFLR_PREFIX)
 705                         return (EINVAL);
 706
 707                 /* copy args to in_aliasreq, perform ioctl(SIOCAIFADDR). */
 708                 bzero(&ifra, sizeof(ifra));
 709                 bcopy(iflr->iflr_name, ifra.ifra_name,
 710                         sizeof(ifra.ifra_name));
 711
 712                 bcopy(&iflr->addr, &ifra.ifra_addr, iflr->addr.ss_len);
 713
 714                 if (iflr->dstaddr.ss_family) {  /*XXX*/
 715                         bcopy(&iflr->dstaddr, &ifra.ifra_dstaddr,
 716                                 iflr->dstaddr.ss_len);
 717                 }
 718
 719                 ifra.ifra_mask.sin_family = AF_INET;
 720                 ifra.ifra_mask.sin_len = sizeof(struct sockaddr_in);
 721                 in_len2mask(&ifra.ifra_mask.sin_addr, iflr->prefixlen);
 722
 723                 return (in_control(so, SIOCAIFADDR, (caddr_t)&ifra, ifp, td));
 724             }
 725         case SIOCGLIFADDR:
 726         case SIOCDLIFADDR:
 727             {
 728                 struct in_ifaddr *ia;
 729                 struct in_addr mask, candidate, match;
 730                 struct sockaddr_in *sin;
 731
 732                 bzero(&mask, sizeof(mask));
 733                 bzero(&match, sizeof(match));
 734                 if (iflr->flags & IFLR_PREFIX) {
 735                         /* lookup a prefix rather than address. */
 736                         in_len2mask(&mask, iflr->prefixlen);
 737
 738                         sin = (struct sockaddr_in *)&iflr->addr;
 739                         match.s_addr = sin->sin_addr.s_addr;
 740                         match.s_addr &= mask.s_addr;
 741
 742                         /* if you set extra bits, that's wrong */
 743                         if (match.s_addr != sin->sin_addr.s_addr)
 744                                 return (EINVAL);
 745
 746                 } else {
 747                         /* on getting an address, take the 1st match */
 748                         /* on deleting an address, do exact match */
 749                         if (cmd != SIOCGLIFADDR) {
 750                                 in_len2mask(&mask, 32);
 751                                 sin = (struct sockaddr_in *)&iflr->addr;
 752                                 match.s_addr = sin->sin_addr.s_addr;
 753                         }
 754                 }
 755
 756                 IF_ADDR_RLOCK(ifp);
 757                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 758                         if (ifa->ifa_addr->sa_family != AF_INET)
 759                                 continue;
 760                         if (match.s_addr == 0)
 761                                 break;
 762                         candidate.s_addr = ((struct sockaddr_in *)&ifa->ifa_addr)->sin_addr.s_addr;
 763                         candidate.s_addr &= mask.s_addr;
 764                         if (candidate.s_addr == match.s_addr)
 765                                 break;
 766                 }
 767                 if (ifa != NULL)
 768                         ifa_ref(ifa);
 769                 IF_ADDR_RUNLOCK(ifp);
 770                 if (ifa == NULL)
 771                         return (EADDRNOTAVAIL);
 772                 ia = (struct in_ifaddr *)ifa;
 773
 774                 if (cmd == SIOCGLIFADDR) {
 775                         /* fill in the if_laddrreq structure */
 776                         bcopy(&ia->ia_addr, &iflr->addr, ia->ia_addr.sin_len);
 777
 778                         if ((ifp->if_flags & IFF_POINTOPOINT) != 0) {
 779                                 bcopy(&ia->ia_dstaddr, &iflr->dstaddr,
 780                                         ia->ia_dstaddr.sin_len);
 781                         } else
 782                                 bzero(&iflr->dstaddr, sizeof(iflr->dstaddr));
 783
 784                         iflr->prefixlen =
 785                                 in_mask2len(&ia->ia_sockmask.sin_addr);
 786
 787                         iflr->flags = 0;        /*XXX*/
 788                         ifa_free(ifa);
 789
 790                         return (0);
 791                 } else {
 792                         struct in_aliasreq ifra;
 793
 794                         /* fill in_aliasreq and do ioctl(SIOCDIFADDR) */
 795                         bzero(&ifra, sizeof(ifra));
 796                         bcopy(iflr->iflr_name, ifra.ifra_name,
 797                                 sizeof(ifra.ifra_name));
 798
 799                         bcopy(&ia->ia_addr, &ifra.ifra_addr,
 800                                 ia->ia_addr.sin_len);
 801                         if ((ifp->if_flags & IFF_POINTOPOINT) != 0) {
 802                                 bcopy(&ia->ia_dstaddr, &ifra.ifra_dstaddr,
 803                                         ia->ia_dstaddr.sin_len);
 804                         }
 805                         bcopy(&ia->ia_sockmask, &ifra.ifra_dstaddr,
 806                                 ia->ia_sockmask.sin_len);
 807                         ifa_free(ifa);
 808
 809                         return (in_control(so, SIOCDIFADDR, (caddr_t)&ifra,
 810                             ifp, td));
 811                 }
 812             }
 813         }
 814
 815         return (EOPNOTSUPP);    /*just for safety*/
 816 }
 817
 818 /*
 819  * Delete any existing route for an interface.
 820  */
 821 void
 822 in_ifscrub(struct ifnet *ifp, struct in_ifaddr *ia, u_int flags)
 823 {
 824
 825         in_scrubprefix(ia, flags);
 826 }
 827
 828 /*
 829  * Initialize an interface's internet address
 830  * and routing table entry.
 831  */
 832 static int
 833 in_ifinit(struct ifnet *ifp, struct in_ifaddr *ia, struct sockaddr_in *sin,
 834     int scrub)
 835 {
 836         register u_long i = ntohl(sin->sin_addr.s_addr);
 837         struct sockaddr_in oldaddr;
 838         int s = splimp(), flags = RTF_UP, error = 0;
 839
 840         oldaddr = ia->ia_addr;
 841         if (oldaddr.sin_family == AF_INET)
 842                 LIST_REMOVE(ia, ia_hash);
 843         ia->ia_addr = *sin;
 844         if (ia->ia_addr.sin_family == AF_INET) {
 845                 IN_IFADDR_WLOCK();
 846                 LIST_INSERT_HEAD(INADDR_HASH(ia->ia_addr.sin_addr.s_addr),
 847                     ia, ia_hash);
 848                 IN_IFADDR_WUNLOCK();
 849         }
 850         /*
 851          * Give the interface a chance to initialize
 852          * if this is its first address,
 853          * and to validate the address if necessary.
 854          */
 855         if (ifp->if_ioctl != NULL) {
 856                 error = (*ifp->if_ioctl)(ifp, SIOCSIFADDR, (caddr_t)ia);
 857                 if (error) {
 858                         splx(s);
 859                         /* LIST_REMOVE(ia, ia_hash) is done in in_control */
 860                         ia->ia_addr = oldaddr;
 861                         IN_IFADDR_WLOCK();
 862                         if (ia->ia_addr.sin_family == AF_INET)
 863                                 LIST_INSERT_HEAD(INADDR_HASH(
 864                                     ia->ia_addr.sin_addr.s_addr), ia, ia_hash);
 865                         else
 866                                 /*
 867                                  * If oldaddr family is not AF_INET (e.g.
 868                                  * interface has been just created) in_control
 869                                  * does not call LIST_REMOVE, and we end up
 870                                  * with bogus ia entries in hash
 871                                  */
 872                                 LIST_REMOVE(ia, ia_hash);
 873                         IN_IFADDR_WUNLOCK();
 874                         return (error);
 875                 }
 876         }
 877         splx(s);
 878         if (scrub) {
 879                 ia->ia_ifa.ifa_addr = (struct sockaddr *)&oldaddr;
 880                 in_ifscrub(ifp, ia, LLE_STATIC);
 881                 ia->ia_ifa.ifa_addr = (struct sockaddr *)&ia->ia_addr;
 882         }
 883         /*
 884          * Be compatible with network classes, if netmask isn't supplied,
 885          * guess it based on classes.
 886          */
 887         if (ia->ia_subnetmask == 0) {
 888                 if (IN_CLASSA(i))
 889                         ia->ia_subnetmask = IN_CLASSA_NET;
 890                 else if (IN_CLASSB(i))
 891                         ia->ia_subnetmask = IN_CLASSB_NET;
 892                 else
 893                         ia->ia_subnetmask = IN_CLASSC_NET;
 894                 ia->ia_sockmask.sin_addr.s_addr = htonl(ia->ia_subnetmask);
 895         }
 896         ia->ia_subnet = i & ia->ia_subnetmask;
 897         in_socktrim(&ia->ia_sockmask);
 898         /*
 899          * XXX: carp(4) does not have interface route
 900          */
 901         if (ifp->if_type == IFT_CARP)
 902                 return (0);
 903         /*
 904          * Add route for the network.
 905          */
 906         ia->ia_ifa.ifa_metric = ifp->if_metric;
 907         if (ifp->if_flags & IFF_BROADCAST) {
 908                 if (ia->ia_subnetmask == IN_RFC3021_MASK)
 909                         ia->ia_broadaddr.sin_addr.s_addr = INADDR_BROADCAST;
 910                 else
 911                         ia->ia_broadaddr.sin_addr.s_addr =
 912                             htonl(ia->ia_subnet | ~ia->ia_subnetmask);
 913         } else if (ifp->if_flags & IFF_LOOPBACK) {
 914                 ia->ia_dstaddr = ia->ia_addr;
 915                 flags |= RTF_HOST;
 916         } else if (ifp->if_flags & IFF_POINTOPOINT) {
 917                 if (ia->ia_dstaddr.sin_family != AF_INET)
 918                         return (0);
 919                 flags |= RTF_HOST;
 920         }
 921         if ((error = in_addprefix(ia, flags)) != 0)
 922                 return (error);
 923
 924         if (ia->ia_addr.sin_addr.s_addr == INADDR_ANY)
 925                 return (0);
 926
 927         if (ifp->if_flags & IFF_POINTOPOINT) {
 928                 if (ia->ia_dstaddr.sin_addr.s_addr == ia->ia_addr.sin_addr.s_addr)
 929                         return (0);
 930         }
 931
 932
 933         /*
 934          * add a loopback route to self
 935          */
 936         if (V_useloopback && !(ifp->if_flags & IFF_LOOPBACK)) {
 937                 struct route ia_ro;
 938
 939                 bzero(&ia_ro, sizeof(ia_ro));
 940                 *((struct sockaddr_in *)(&ia_ro.ro_dst)) = ia->ia_addr;
 941                 rtalloc_ign_fib(&ia_ro, 0, RT_DEFAULT_FIB);
 942                 if ((ia_ro.ro_rt != NULL) && (ia_ro.ro_rt->rt_ifp != NULL) &&
 943                     (ia_ro.ro_rt->rt_ifp == V_loif)) {
 944                         RT_LOCK(ia_ro.ro_rt);
 945                         RT_ADDREF(ia_ro.ro_rt);
 946                         RTFREE_LOCKED(ia_ro.ro_rt);
 947                 } else
 948                         error = ifa_add_loopback_route((struct ifaddr *)ia,
 949                             (struct sockaddr *)&ia->ia_addr);
 950                 if (error == 0)
 951                         ia->ia_flags |= IFA_RTSELF;
 952                 if (ia_ro.ro_rt != NULL)
 953                         RTFREE(ia_ro.ro_rt);
 954         }
 955
 956         return (error);
 957 }
 958
 959 #define rtinitflags(x) \
 960         ((((x)->ia_ifp->if_flags & (IFF_LOOPBACK | IFF_POINTOPOINT)) != 0) \
 961             ? RTF_HOST : 0)
 962
 963 /*
 964  * Generate a routing message when inserting or deleting
 965  * an interface address alias.
 966  */
 967 static void in_addralias_rtmsg(int cmd, struct in_addr *prefix,
 968     struct in_ifaddr *target)
 969 {
 970         struct route pfx_ro;
 971         struct sockaddr_in *pfx_addr;
 972         struct rtentry msg_rt;
 973
 974         /* QL: XXX
 975          * This is a bit questionable because there is no
 976          * additional route entry added/deleted for an address
 977          * alias. Therefore this route report is inaccurate.
 978          */
 979         bzero(&pfx_ro, sizeof(pfx_ro));
 980         pfx_addr = (struct sockaddr_in *)(&pfx_ro.ro_dst);
 981         pfx_addr->sin_len = sizeof(*pfx_addr);
 982         pfx_addr->sin_family = AF_INET;
 983         pfx_addr->sin_addr = *prefix;
 984         rtalloc_ign_fib(&pfx_ro, 0, 0);
 985         if (pfx_ro.ro_rt != NULL) {
 986                 msg_rt = *pfx_ro.ro_rt;
 987
 988                 /* QL: XXX
 989                  * Point the gateway to the new interface
 990                  * address as if a new prefix route entry has
 991                  * been added through the new address alias.
 992                  * All other parts of the rtentry is accurate,
 993                  * e.g., rt_key, rt_mask, rt_ifp etc.
 994                  */
 995                 msg_rt.rt_gateway = (struct sockaddr *)&target->ia_addr;
 996                 rt_newaddrmsg(cmd, (struct ifaddr *)target, 0, &msg_rt);
 997                 RTFREE(pfx_ro.ro_rt);
 998         }
 999         return;
1000 }
1001
1002 /*
1003  * Check if we have a route for the given prefix already or add one accordingly.
1004  */
1005 static int
1006 in_addprefix(struct in_ifaddr *target, int flags)
1007 {
1008         struct in_ifaddr *ia;
1009         struct in_addr prefix, mask, p, m;
1010         int error;
1011
1012         if ((flags & RTF_HOST) != 0) {
1013                 prefix = target->ia_dstaddr.sin_addr;
1014                 mask.s_addr = 0;
1015         } else {
1016                 prefix = target->ia_addr.sin_addr;
1017                 mask = target->ia_sockmask.sin_addr;
1018                 prefix.s_addr &= mask.s_addr;
1019         }
1020
1021         IN_IFADDR_RLOCK();
1022         TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1023                 if (rtinitflags(ia)) {
1024                         p = ia->ia_dstaddr.sin_addr;
1025
1026                         if (prefix.s_addr != p.s_addr)
1027                                 continue;
1028                 } else {
1029                         p = ia->ia_addr.sin_addr;
1030                         m = ia->ia_sockmask.sin_addr;
1031                         p.s_addr &= m.s_addr;
1032
1033                         if (prefix.s_addr != p.s_addr ||
1034                             mask.s_addr != m.s_addr)
1035                                 continue;
1036                 }
1037
1038                 /*
1039                  * If we got a matching prefix route inserted by other
1040                  * interface address, we are done here.
1041                  */
1042                 if (ia->ia_flags & IFA_ROUTE) {
1043 #ifdef RADIX_MPATH
1044                         if (ia->ia_addr.sin_addr.s_addr ==
1045                             target->ia_addr.sin_addr.s_addr) {
1046                                 IN_IFADDR_RUNLOCK();
1047                                 return (EEXIST);
1048                         } else
1049                                 break;
1050 #endif
1051                         if (V_sameprefixcarponly &&
1052                             target->ia_ifp->if_type != IFT_CARP &&
1053                             ia->ia_ifp->if_type != IFT_CARP) {
1054                                 IN_IFADDR_RUNLOCK();
1055                                 return (EEXIST);
1056                         } else {
1057                                 in_addralias_rtmsg(RTM_ADD, &prefix, target);
1058                                 IN_IFADDR_RUNLOCK();
1059                                 return (0);
1060                         }
1061                 }
1062         }
1063         IN_IFADDR_RUNLOCK();
1064
1065         /*
1066          * No-one seem to have this prefix route, so we try to insert it.
1067          */
1068         error = rtinit(&target->ia_ifa, (int)RTM_ADD, flags);
1069         if (!error)
1070                 target->ia_flags |= IFA_ROUTE;
1071         return (error);
1072 }
1073
1074 extern void arp_ifscrub(struct ifnet *ifp, uint32_t addr);
1075
1076 /*
1077  * If there is no other address in the system that can serve a route to the
1078  * same prefix, remove the route.  Hand over the route to the new address
1079  * otherwise.
1080  */
1081 static int
1082 in_scrubprefix(struct in_ifaddr *target, u_int flags)
1083 {
1084         struct in_ifaddr *ia;
1085         struct in_addr prefix, mask, p;
1086         int error = 0;
1087         struct sockaddr_in prefix0, mask0;
1088
1089         /*
1090          * Remove the loopback route to the interface address.
1091          * The "useloopback" setting is not consulted because if the
1092          * user configures an interface address, turns off this
1093          * setting, and then tries to delete that interface address,
1094          * checking the current setting of "useloopback" would leave
1095          * that interface address loopback route untouched, which
1096          * would be wrong. Therefore the interface address loopback route
1097          * deletion is unconditional.
1098          */
1099         if ((target->ia_addr.sin_addr.s_addr != INADDR_ANY) &&
1100             !(target->ia_ifp->if_flags & IFF_LOOPBACK) &&
1101             (target->ia_flags & IFA_RTSELF)) {
1102                 struct route ia_ro;
1103                 int freeit = 0;
1104
1105                 bzero(&ia_ro, sizeof(ia_ro));
1106                 *((struct sockaddr_in *)(&ia_ro.ro_dst)) = target->ia_addr;
1107                 rtalloc_ign_fib(&ia_ro, 0, 0);
1108                 if ((ia_ro.ro_rt != NULL) && (ia_ro.ro_rt->rt_ifp != NULL) &&
1109                     (ia_ro.ro_rt->rt_ifp == V_loif)) {
1110                         RT_LOCK(ia_ro.ro_rt);
1111                         if (ia_ro.ro_rt->rt_refcnt <= 1)
1112                                 freeit = 1;
1113                         else if (flags & LLE_STATIC) {
1114                                 RT_REMREF(ia_ro.ro_rt);
1115                                 target->ia_flags &= ~IFA_RTSELF;
1116                         }
1117                         RTFREE_LOCKED(ia_ro.ro_rt);
1118                 }
1119                 if (freeit && (flags & LLE_STATIC)) {
1120                         error = ifa_del_loopback_route((struct ifaddr *)target,
1121                             (struct sockaddr *)&target->ia_addr);
1122                         if (error == 0)
1123                                 target->ia_flags &= ~IFA_RTSELF;
1124                 }
1125                 if ((flags & LLE_STATIC) &&
1126                         !(target->ia_ifp->if_flags & IFF_NOARP))
1127                         /* remove arp cache */
1128                         arp_ifscrub(target->ia_ifp, IA_SIN(target)->sin_addr.s_addr);
1129         }
1130
1131         if (rtinitflags(target))
1132                 prefix = target->ia_dstaddr.sin_addr;
1133         else {
1134                 prefix = target->ia_addr.sin_addr;
1135                 mask = target->ia_sockmask.sin_addr;
1136                 prefix.s_addr &= mask.s_addr;
1137         }
1138
1139         if ((target->ia_flags & IFA_ROUTE) == 0) {
1140                 in_addralias_rtmsg(RTM_DELETE, &prefix, target);
1141                 return (0);
1142         }
1143
1144         IN_IFADDR_RLOCK();
1145         TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1146                 if (rtinitflags(ia))
1147                         p = ia->ia_dstaddr.sin_addr;
1148                 else {
1149                         p = ia->ia_addr.sin_addr;
1150                         p.s_addr &= ia->ia_sockmask.sin_addr.s_addr;
1151                 }
1152
1153                 if ((prefix.s_addr != p.s_addr) ||
1154                     !(ia->ia_ifp->if_flags & IFF_UP))
1155                         continue;
1156
1157                 /*
1158                  * If we got a matching prefix address, move IFA_ROUTE and
1159                  * the route itself to it.  Make sure that routing daemons
1160                  * get a heads-up.
1161                  *
1162                  * XXX: a special case for carp(4) interface - this should
1163                  *      be more generally specified as an interface that
1164                  *      doesn't support such action.
1165                  */
1166                 if ((ia->ia_flags & IFA_ROUTE) == 0
1167                     && (ia->ia_ifp->if_type != IFT_CARP)) {
1168                         ifa_ref(&ia->ia_ifa);
1169                         IN_IFADDR_RUNLOCK();
1170                         error = rtinit(&(target->ia_ifa), (int)RTM_DELETE,
1171                             rtinitflags(target));
1172                         if (error == 0)
1173                                 target->ia_flags &= ~IFA_ROUTE;
1174                         else
1175                                 log(LOG_INFO, "in_scrubprefix: err=%d, old prefix delete failed\n",
1176                                         error);
1177                         error = rtinit(&ia->ia_ifa, (int)RTM_ADD,
1178                             rtinitflags(ia) | RTF_UP);
1179                         if (error == 0)
1180                                 ia->ia_flags |= IFA_ROUTE;
1181                         else
1182                                 log(LOG_INFO, "in_scrubprefix: err=%d, new prefix add failed\n",
1183                                         error);
1184                         ifa_free(&ia->ia_ifa);
1185                         return (error);
1186                 }
1187         }
1188         IN_IFADDR_RUNLOCK();
1189
1190         /*
1191          * remove all L2 entries on the given prefix
1192          */
1193         bzero(&prefix0, sizeof(prefix0));
1194         prefix0.sin_len = sizeof(prefix0);
1195         prefix0.sin_family = AF_INET;
1196         prefix0.sin_addr.s_addr = target->ia_subnet;
1197         bzero(&mask0, sizeof(mask0));
1198         mask0.sin_len = sizeof(mask0);
1199         mask0.sin_family = AF_INET;
1200         mask0.sin_addr.s_addr = target->ia_subnetmask;
1201         lltable_prefix_free(AF_INET, (struct sockaddr *)&prefix0,
1202             (struct sockaddr *)&mask0, flags);
1203
1204         /*
1205          * As no-one seem to have this prefix, we can remove the route.
1206          */
1207         error = rtinit(&(target->ia_ifa), (int)RTM_DELETE, rtinitflags(target));
1208         if (error == 0)
1209                 target->ia_flags &= ~IFA_ROUTE;
1210         else
1211                 log(LOG_INFO, "in_scrubprefix: err=%d, prefix delete failed\n", error);
1212         return (error);
1213 }
1214
1215 #undef rtinitflags
1216
1217 /*
1218  * Return 1 if the address might be a local broadcast address.
1219  */
1220 int
1221 in_broadcast(struct in_addr in, struct ifnet *ifp)
1222 {
1223         register struct ifaddr *ifa;
1224         u_long t;
1225
1226         if (in.s_addr == INADDR_BROADCAST ||
1227             in.s_addr == INADDR_ANY)
1228                 return (1);
1229         if ((ifp->if_flags & IFF_BROADCAST) == 0)
1230                 return (0);
1231         t = ntohl(in.s_addr);
1232         /*
1233          * Look through the list of addresses for a match
1234          * with a broadcast address.
1235          */
1236 #define ia ((struct in_ifaddr *)ifa)
1237         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
1238                 if (ifa->ifa_addr->sa_family == AF_INET &&
1239                     (in.s_addr == ia->ia_broadaddr.sin_addr.s_addr ||
1240                      /*
1241                       * Check for old-style (host 0) broadcast, but
1242                       * taking into account that RFC 3021 obsoletes it.
1243                       */
1244                     (ia->ia_subnetmask != IN_RFC3021_MASK &&
1245                     t == ia->ia_subnet)) &&
1246                      /*
1247                       * Check for an all one subnetmask. These
1248                       * only exist when an interface gets a secondary
1249                       * address.
1250                       */
1251                     ia->ia_subnetmask != (u_long)0xffffffff)
1252                             return (1);
1253         return (0);
1254 #undef ia
1255 }
1256
1257 /*
1258  * On interface removal, clean up IPv4 data structures hung off of the ifnet.
1259  */
1260 void
1261 in_ifdetach(struct ifnet *ifp)
1262 {
1263
1264         in_pcbpurgeif0(&V_ripcbinfo, ifp);
1265         in_pcbpurgeif0(&V_udbinfo, ifp);
1266         in_purgemaddrs(ifp);
1267 }
1268
1269 /*
1270  * Delete all IPv4 multicast address records, and associated link-layer
1271  * multicast address records, associated with ifp.
1272  * XXX It looks like domifdetach runs AFTER the link layer cleanup.
1273  * XXX This should not race with ifma_protospec being set during
1274  * a new allocation, if it does, we have bigger problems.
1275  */
1276 static void
1277 in_purgemaddrs(struct ifnet *ifp)
1278 {
1279         LIST_HEAD(,in_multi) purgeinms;
1280         struct in_multi         *inm, *tinm;
1281         struct ifmultiaddr      *ifma;
1282
1283         LIST_INIT(&purgeinms);
1284         IN_MULTI_LOCK();
1285
1286         /*
1287          * Extract list of in_multi associated with the detaching ifp
1288          * which the PF_INET layer is about to release.
1289          * We need to do this as IF_ADDR_LOCK() may be re-acquired
1290          * by code further down.
1291          */
1292         IF_ADDR_RLOCK(ifp);
1293         TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1294                 if (ifma->ifma_addr->sa_family != AF_INET ||
1295                     ifma->ifma_protospec == NULL)
1296                         continue;
1297 #if 0
1298                 KASSERT(ifma->ifma_protospec != NULL,
1299                     ("%s: ifma_protospec is NULL", __func__));
1300 #endif
1301                 inm = (struct in_multi *)ifma->ifma_protospec;
1302                 LIST_INSERT_HEAD(&purgeinms, inm, inm_link);
1303         }
1304         IF_ADDR_RUNLOCK(ifp);
1305
1306         LIST_FOREACH_SAFE(inm, &purgeinms, inm_link, tinm) {
1307                 LIST_REMOVE(inm, inm_link);
1308                 inm_release_locked(inm);
1309         }
1310         igmp_ifdetach(ifp);
1311
1312         IN_MULTI_UNLOCK();
1313 }
1314
1315 #include <net/if_dl.h>
1316 #include <netinet/if_ether.h>
1317
1318 struct in_llentry {
1319         struct llentry          base;
1320         struct sockaddr_in      l3_addr4;
1321 };
1322
1323 /*
1324  * Deletes an address from the address table.
1325  * This function is called by the timer functions
1326  * such as arptimer() and nd6_llinfo_timer(), and
1327  * the caller does the locking.
1328  */
1329 static void
1330 in_lltable_free(struct lltable *llt, struct llentry *lle)
1331 {
1332         LLE_WUNLOCK(lle);
1333         LLE_LOCK_DESTROY(lle);
1334         free(lle, M_LLTABLE);
1335 }
1336
1337 static struct llentry *
1338 in_lltable_new(const struct sockaddr *l3addr, u_int flags)
1339 {
1340         struct in_llentry *lle;
1341
1342         lle = malloc(sizeof(struct in_llentry), M_LLTABLE, M_DONTWAIT | M_ZERO);
1343         if (lle == NULL)                /* NB: caller generates msg */
1344                 return NULL;
1345
1346         /*
1347          * For IPv4 this will trigger "arpresolve" to generate
1348          * an ARP request.
1349          */
1350         lle->base.la_expire = time_uptime; /* mark expired */
1351         lle->l3_addr4 = *(const struct sockaddr_in *)l3addr;
1352         lle->base.lle_refcnt = 1;
1353         lle->base.lle_free = in_lltable_free;
1354         LLE_LOCK_INIT(&lle->base);
1355         callout_init_rw(&lle->base.la_timer, &lle->base.lle_lock,
1356             CALLOUT_RETURNUNLOCKED);
1357
1358         return (&lle->base);
1359 }
1360
1361 #define IN_ARE_MASKED_ADDR_EQUAL(d, a, m)       (                       \
1362             (((ntohl((d)->sin_addr.s_addr) ^ (a)->sin_addr.s_addr) & (m)->sin_addr.s_addr)) == 0 )
1363
1364 static void
1365 in_lltable_prefix_free(struct lltable *llt, const struct sockaddr *prefix,
1366     const struct sockaddr *mask, u_int flags)
1367 {
1368         const struct sockaddr_in *pfx = (const struct sockaddr_in *)prefix;
1369         const struct sockaddr_in *msk = (const struct sockaddr_in *)mask;
1370         struct llentry *lle, *next;
1371         int i;
1372         size_t pkts_dropped;
1373
1374         IF_AFDATA_WLOCK(llt->llt_ifp);
1375         for (i = 0; i < LLTBL_HASHTBL_SIZE; i++) {
1376                 LIST_FOREACH_SAFE(lle, &llt->lle_head[i], lle_next, next) {
1377                         /*
1378                          * (flags & LLE_STATIC) means deleting all entries
1379                          * including static ARP entries.
1380                          */
1381                         if (IN_ARE_MASKED_ADDR_EQUAL(satosin(L3_ADDR(lle)),
1382                             pfx, msk) && ((flags & LLE_STATIC) ||
1383                             !(lle->la_flags & LLE_STATIC))) {
1384                                 LLE_WLOCK(lle);
1385                                 if (callout_stop(&lle->la_timer))
1386                                         LLE_REMREF(lle);
1387                                 pkts_dropped = llentry_free(lle);
1388                                 ARPSTAT_ADD(dropped, pkts_dropped);
1389                         }
1390                 }
1391         }
1392         IF_AFDATA_WUNLOCK(llt->llt_ifp);
1393 }
1394
1395
1396 static int
1397 in_lltable_rtcheck(struct ifnet *ifp, u_int flags, const struct sockaddr *l3addr)
1398 {
1399         struct rtentry *rt;
1400
1401         KASSERT(l3addr->sa_family == AF_INET,
1402             ("sin_family %d", l3addr->sa_family));
1403
1404         /* XXX rtalloc1 should take a const param */
1405         rt = rtalloc1(__DECONST(struct sockaddr *, l3addr), 0, 0);
1406
1407         if (rt == NULL)
1408                 return (EINVAL);
1409
1410         /*
1411          * If the gateway for an existing host route matches the target L3
1412          * address, which is a special route inserted by some implementation
1413          * such as MANET, and the interface is of the correct type, then
1414          * allow for ARP to proceed.
1415          */
1416         if (rt->rt_flags & RTF_GATEWAY) {
1417                 if (!(rt->rt_flags & RTF_HOST) || !rt->rt_ifp ||
1418                     rt->rt_ifp->if_type != IFT_ETHER ||
1419                     (rt->rt_ifp->if_flags & (IFF_NOARP | IFF_STATICARP)) != 0 ||
1420                     memcmp(rt->rt_gateway->sa_data, l3addr->sa_data,
1421                     sizeof(in_addr_t)) != 0) {
1422                         RTFREE_LOCKED(rt);
1423                         return (EINVAL);
1424                 }
1425         }
1426
1427         /*
1428          * Make sure that at least the destination address is covered
1429          * by the route. This is for handling the case where 2 or more
1430          * interfaces have the same prefix. An incoming packet arrives
1431          * on one interface and the corresponding outgoing packet leaves
1432          * another interface.
1433          */
1434         if (!(rt->rt_flags & RTF_HOST) && rt->rt_ifp != ifp) {
1435                 const char *sa, *mask, *addr, *lim;
1436                 int len;
1437
1438                 mask = (const char *)rt_mask(rt);
1439                 /*
1440                  * Just being extra cautious to avoid some custom
1441                  * code getting into trouble.
1442                  */
1443                 if (mask == NULL) {
1444                         RTFREE_LOCKED(rt);
1445                         return (EINVAL);
1446                 }
1447
1448                 sa = (const char *)rt_key(rt);
1449                 addr = (const char *)l3addr;
1450                 len = ((const struct sockaddr_in *)l3addr)->sin_len;
1451                 lim = addr + len;
1452
1453                 for ( ; addr < lim; sa++, mask++, addr++) {
1454                         if ((*sa ^ *addr) & *mask) {
1455 #ifdef DIAGNOSTIC
1456                                 log(LOG_INFO, "IPv4 address: \"%s\" is not on the network\n",
1457                                     inet_ntoa(((const struct sockaddr_in *)l3addr)->sin_addr));
1458 #endif
1459                                 RTFREE_LOCKED(rt);
1460                                 return (EINVAL);
1461                         }
1462                 }
1463         }
1464
1465         RTFREE_LOCKED(rt);
1466         return (0);
1467 }
1468
1469 /*
1470  * Return NULL if not found or marked for deletion.
1471  * If found return lle read locked.
1472  */
1473 static struct llentry *
1474 in_lltable_lookup(struct lltable *llt, u_int flags, const struct sockaddr *l3addr)
1475 {
1476         const struct sockaddr_in *sin = (const struct sockaddr_in *)l3addr;
1477         struct ifnet *ifp = llt->llt_ifp;
1478         struct llentry *lle;
1479         struct llentries *lleh;
1480         u_int hashkey;
1481
1482         IF_AFDATA_LOCK_ASSERT(ifp);
1483         KASSERT(l3addr->sa_family == AF_INET,
1484             ("sin_family %d", l3addr->sa_family));
1485
1486         hashkey = sin->sin_addr.s_addr;
1487         lleh = &llt->lle_head[LLATBL_HASH(hashkey, LLTBL_HASHMASK)];
1488         LIST_FOREACH(lle, lleh, lle_next) {
1489                 struct sockaddr_in *sa2 = satosin(L3_ADDR(lle));
1490                 if (lle->la_flags & LLE_DELETED)
1491                         continue;
1492                 if (sa2->sin_addr.s_addr == sin->sin_addr.s_addr)
1493                         break;
1494         }
1495         if (lle == NULL) {
1496 #ifdef DIAGNOSTIC
1497                 if (flags & LLE_DELETE)
1498                         log(LOG_INFO, "interface address is missing from cache = %p  in delete\n", lle);
1499 #endif
1500                 if (!(flags & LLE_CREATE))
1501                         return (NULL);
1502                 /*
1503                  * A route that covers the given address must have
1504                  * been installed 1st because we are doing a resolution,
1505                  * verify this.
1506                  */
1507                 if (!(flags & LLE_IFADDR) &&
1508                     in_lltable_rtcheck(ifp, flags, l3addr) != 0)
1509                         goto done;
1510
1511                 lle = in_lltable_new(l3addr, flags);
1512                 if (lle == NULL) {
1513                         log(LOG_INFO, "lla_lookup: new lle malloc failed\n");
1514                         goto done;
1515                 }
1516                 lle->la_flags = flags & ~LLE_CREATE;
1517                 if ((flags & (LLE_CREATE | LLE_IFADDR)) == (LLE_CREATE | LLE_IFADDR)) {
1518                         bcopy(IF_LLADDR(ifp), &lle->ll_addr, ifp->if_addrlen);
1519                         lle->la_flags |= (LLE_VALID | LLE_STATIC);
1520                 }
1521
1522                 lle->lle_tbl  = llt;
1523                 lle->lle_head = lleh;
1524                 lle->la_flags |= LLE_LINKED;
1525                 LIST_INSERT_HEAD(lleh, lle, lle_next);
1526         } else if (flags & LLE_DELETE) {
1527                 if (!(lle->la_flags & LLE_IFADDR) || (flags & LLE_IFADDR)) {
1528                         LLE_WLOCK(lle);
1529                         lle->la_flags |= LLE_DELETED;
1530                         EVENTHANDLER_INVOKE(lle_event, lle, LLENTRY_DELETED);
1531 #ifdef DIAGNOSTIC
1532                         log(LOG_INFO, "ifaddr cache = %p is deleted\n", lle);
1533 #endif
1534                         if ((lle->la_flags &
1535                             (LLE_STATIC | LLE_IFADDR)) == LLE_STATIC)
1536                                 llentry_free(lle);
1537                         else
1538                                 LLE_WUNLOCK(lle);
1539                 }
1540                 lle = (void *)-1;
1541
1542         }
1543         if (LLE_IS_VALID(lle)) {
1544                 if (flags & LLE_EXCLUSIVE)
1545                         LLE_WLOCK(lle);
1546                 else
1547                         LLE_RLOCK(lle);
1548         }
1549 done:
1550         return (lle);
1551 }
1552
1553 static int
1554 in_lltable_dump(struct lltable *llt, struct sysctl_req *wr)
1555 {
1556 #define SIN(lle)        ((struct sockaddr_in *) L3_ADDR(lle))
1557         struct ifnet *ifp = llt->llt_ifp;
1558         struct llentry *lle;
1559         /* XXX stack use */
1560         struct {
1561                 struct rt_msghdr        rtm;
1562                 struct sockaddr_inarp   sin;
1563                 struct sockaddr_dl      sdl;
1564         } arpc;
1565         int error, i;
1566
1567         LLTABLE_LOCK_ASSERT();
1568
1569         error = 0;
1570         for (i = 0; i < LLTBL_HASHTBL_SIZE; i++) {
1571                 LIST_FOREACH(lle, &llt->lle_head[i], lle_next) {
1572                         struct sockaddr_dl *sdl;
1573
1574                         /* skip deleted entries */
1575                         if ((lle->la_flags & LLE_DELETED) == LLE_DELETED)
1576                                 continue;
1577                         /* Skip if jailed and not a valid IP of the prison. */
1578                         if (prison_if(wr->td->td_ucred, L3_ADDR(lle)) != 0)
1579                                 continue;
1580                         /*
1581                          * produce a msg made of:
1582                          *  struct rt_msghdr;
1583                          *  struct sockaddr_inarp; (IPv4)
1584                          *  struct sockaddr_dl;
1585                          */
1586                         bzero(&arpc, sizeof(arpc));
1587                         arpc.rtm.rtm_msglen = sizeof(arpc);
1588                         arpc.rtm.rtm_version = RTM_VERSION;
1589                         arpc.rtm.rtm_type = RTM_GET;
1590                         arpc.rtm.rtm_flags = RTF_UP;
1591                         arpc.rtm.rtm_addrs = RTA_DST | RTA_GATEWAY;
1592                         arpc.sin.sin_family = AF_INET;
1593                         arpc.sin.sin_len = sizeof(arpc.sin);
1594                         arpc.sin.sin_addr.s_addr = SIN(lle)->sin_addr.s_addr;
1595
1596                         /* publish */
1597                         if (lle->la_flags & LLE_PUB) {
1598                                 arpc.rtm.rtm_flags |= RTF_ANNOUNCE;
1599                                 /* proxy only */
1600                                 if (lle->la_flags & LLE_PROXY)
1601                                         arpc.sin.sin_other = SIN_PROXY;
1602                         }
1603
1604                         sdl = &arpc.sdl;
1605                         sdl->sdl_family = AF_LINK;
1606                         sdl->sdl_len = sizeof(*sdl);
1607                         sdl->sdl_index = ifp->if_index;
1608                         sdl->sdl_type = ifp->if_type;
1609                         if ((lle->la_flags & LLE_VALID) == LLE_VALID) {
1610                                 sdl->sdl_alen = ifp->if_addrlen;
1611                                 bcopy(&lle->ll_addr, LLADDR(sdl), ifp->if_addrlen);
1612                         } else {
1613                                 sdl->sdl_alen = 0;
1614                                 bzero(LLADDR(sdl), ifp->if_addrlen);
1615                         }
1616
1617                         arpc.rtm.rtm_rmx.rmx_expire =
1618                             lle->la_flags & LLE_STATIC ? 0 : lle->la_expire;
1619                         arpc.rtm.rtm_flags |= (RTF_HOST | RTF_LLDATA);
1620                         if (lle->la_flags & LLE_STATIC)
1621                                 arpc.rtm.rtm_flags |= RTF_STATIC;
1622                         arpc.rtm.rtm_index = ifp->if_index;
1623                         error = SYSCTL_OUT(wr, &arpc, sizeof(arpc));
1624                         if (error)
1625                                 break;
1626                 }
1627         }
1628         return error;
1629 #undef SIN
1630 }
1631
1632 void *
1633 in_domifattach(struct ifnet *ifp)
1634 {
1635         struct in_ifinfo *ii;
1636         struct lltable *llt;
1637
1638         ii = malloc(sizeof(struct in_ifinfo), M_IFADDR, M_WAITOK|M_ZERO);
1639
1640         llt = lltable_init(ifp, AF_INET);
1641         if (llt != NULL) {
1642                 llt->llt_prefix_free = in_lltable_prefix_free;
1643                 llt->llt_lookup = in_lltable_lookup;
1644                 llt->llt_dump = in_lltable_dump;
1645         }
1646         ii->ii_llt = llt;
1647
1648         ii->ii_igmp = igmp_domifattach(ifp);
1649
1650         return ii;
1651 }
1652
1653 void
1654 in_domifdetach(struct ifnet *ifp, void *aux)
1655 {
1656         struct in_ifinfo *ii = (struct in_ifinfo *)aux;
1657
1658         igmp_domifdetach(ifp);
1659         lltable_free(ii->ii_llt);
1660         free(ii, M_IFADDR);
1661 }