sys/netinet/ip_output.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)ip_output.c 8.3 (Berkeley) 1/21/94
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include "opt_inet.h"
  38 #include "opt_ipsec.h"
  39 #include "opt_kern_tls.h"
  40 #include "opt_mbuf_stress_test.h"
  41 #include "opt_ratelimit.h"
  42 #include "opt_route.h"
  43 #include "opt_rss.h"
  44 #include "opt_sctp.h"
  45
  46 #include <sys/param.h>
  47 #include <sys/systm.h>
  48 #include <sys/kernel.h>
  49 #include <sys/ktls.h>
  50 #include <sys/lock.h>
  51 #include <sys/malloc.h>
  52 #include <sys/mbuf.h>
  53 #include <sys/priv.h>
  54 #include <sys/proc.h>
  55 #include <sys/protosw.h>
  56 #include <sys/sdt.h>
  57 #include <sys/socket.h>
  58 #include <sys/socketvar.h>
  59 #include <sys/sysctl.h>
  60 #include <sys/ucred.h>
  61
  62 #include <net/if.h>
  63 #include <net/if_var.h>
  64 #include <net/if_vlan_var.h>
  65 #include <net/if_llatbl.h>
  66 #include <net/ethernet.h>
  67 #include <net/netisr.h>
  68 #include <net/pfil.h>
  69 #include <net/route.h>
  70 #include <net/route/nhop.h>
  71 #include <net/rss_config.h>
  72 #include <net/vnet.h>
  73
  74 #include <netinet/in.h>
  75 #include <netinet/in_fib.h>
  76 #include <netinet/in_kdtrace.h>
  77 #include <netinet/in_systm.h>
  78 #include <netinet/ip.h>
  79 #include <netinet/in_fib.h>
  80 #include <netinet/in_pcb.h>
  81 #include <netinet/in_rss.h>
  82 #include <netinet/in_var.h>
  83 #include <netinet/ip_var.h>
  84 #include <netinet/ip_options.h>
  85 #include <netinet/ip_mroute.h>
  86
  87 #include <netinet/udp.h>
  88 #include <netinet/udp_var.h>
  89
  90 #if defined(SCTP) || defined(SCTP_SUPPORT)
  91 #include <netinet/sctp.h>
  92 #include <netinet/sctp_crc32.h>
  93 #endif
  94
  95 #include <netipsec/ipsec_support.h>
  96
  97 #include <machine/in_cksum.h>
  98
  99 #include <security/mac/mac_framework.h>
 100
 101 #ifdef MBUF_STRESS_TEST
 102 static int mbuf_frag_size = 0;
 103 SYSCTL_INT(_net_inet_ip, OID_AUTO, mbuf_frag_size, CTLFLAG_RW,
 104         &mbuf_frag_size, 0, "Fragment outgoing mbufs to this size");
 105 #endif
 106
 107 static void     ip_mloopback(struct ifnet *, const struct mbuf *, int);
 108
 109 extern int in_mcast_loop;
 110 extern  struct protosw inetsw[];
 111
 112 static inline int
 113 ip_output_pfil(struct mbuf **mp, struct ifnet *ifp, int flags,
 114     struct inpcb *inp, struct sockaddr_in *dst, int *fibnum, int *error)
 115 {
 116         struct m_tag *fwd_tag = NULL;
 117         struct mbuf *m;
 118         struct in_addr odst;
 119         struct ip *ip;
 120         int pflags = PFIL_OUT;
 121
 122         if (flags & IP_FORWARDING)
 123                 pflags |= PFIL_FWD;
 124
 125         m = *mp;
 126         ip = mtod(m, struct ip *);
 127
 128         /* Run through list of hooks for output packets. */
 129         odst.s_addr = ip->ip_dst.s_addr;
 130         switch (pfil_run_hooks(V_inet_pfil_head, mp, ifp, pflags, inp)) {
 131         case PFIL_DROPPED:
 132                 *error = EACCES;
 133                 /* FALLTHROUGH */
 134         case PFIL_CONSUMED:
 135                 return 1; /* Finished */
 136         case PFIL_PASS:
 137                 *error = 0;
 138         }
 139         m = *mp;
 140         ip = mtod(m, struct ip *);
 141
 142         /* See if destination IP address was changed by packet filter. */
 143         if (odst.s_addr != ip->ip_dst.s_addr) {
 144                 m->m_flags |= M_SKIP_FIREWALL;
 145                 /* If destination is now ourself drop to ip_input(). */
 146                 if (in_localip(ip->ip_dst)) {
 147                         m->m_flags |= M_FASTFWD_OURS;
 148                         if (m->m_pkthdr.rcvif == NULL)
 149                                 m->m_pkthdr.rcvif = V_loif;
 150                         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 151                                 m->m_pkthdr.csum_flags |=
 152                                         CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 153                                 m->m_pkthdr.csum_data = 0xffff;
 154                         }
 155                         m->m_pkthdr.csum_flags |=
 156                                 CSUM_IP_CHECKED | CSUM_IP_VALID;
 157 #if defined(SCTP) || defined(SCTP_SUPPORT)
 158                         if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 159                                 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 160 #endif
 161                         *error = netisr_queue(NETISR_IP, m);
 162                         return 1; /* Finished */
 163                 }
 164
 165                 bzero(dst, sizeof(*dst));
 166                 dst->sin_family = AF_INET;
 167                 dst->sin_len = sizeof(*dst);
 168                 dst->sin_addr = ip->ip_dst;
 169
 170                 return -1; /* Reloop */
 171         }
 172         /* See if fib was changed by packet filter. */
 173         if ((*fibnum) != M_GETFIB(m)) {
 174                 m->m_flags |= M_SKIP_FIREWALL;
 175                 *fibnum = M_GETFIB(m);
 176                 return -1; /* Reloop for FIB change */
 177         }
 178
 179         /* See if local, if yes, send it to netisr with IP_FASTFWD_OURS. */
 180         if (m->m_flags & M_FASTFWD_OURS) {
 181                 if (m->m_pkthdr.rcvif == NULL)
 182                         m->m_pkthdr.rcvif = V_loif;
 183                 if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 184                         m->m_pkthdr.csum_flags |=
 185                                 CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
 186                         m->m_pkthdr.csum_data = 0xffff;
 187                 }
 188 #if defined(SCTP) || defined(SCTP_SUPPORT)
 189                 if (m->m_pkthdr.csum_flags & CSUM_SCTP)
 190                         m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID;
 191 #endif
 192                 m->m_pkthdr.csum_flags |=
 193                         CSUM_IP_CHECKED | CSUM_IP_VALID;
 194
 195                 *error = netisr_queue(NETISR_IP, m);
 196                 return 1; /* Finished */
 197         }
 198         /* Or forward to some other address? */
 199         if ((m->m_flags & M_IP_NEXTHOP) &&
 200             ((fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL)) != NULL)) {
 201                 bcopy((fwd_tag+1), dst, sizeof(struct sockaddr_in));
 202                 m->m_flags |= M_SKIP_FIREWALL;
 203                 m->m_flags &= ~M_IP_NEXTHOP;
 204                 m_tag_delete(m, fwd_tag);
 205
 206                 return -1; /* Reloop for CHANGE of dst */
 207         }
 208
 209         return 0;
 210 }
 211
 212 static int
 213 ip_output_send(struct inpcb *inp, struct ifnet *ifp, struct mbuf *m,
 214     const struct sockaddr *gw, struct route *ro, bool stamp_tag)
 215 {
 216 #ifdef KERN_TLS
 217         struct ktls_session *tls = NULL;
 218 #endif
 219         struct m_snd_tag *mst;
 220         int error;
 221
 222         MPASS((m->m_pkthdr.csum_flags & CSUM_SND_TAG) == 0);
 223         mst = NULL;
 224
 225 #ifdef KERN_TLS
 226         /*
 227          * If this is an unencrypted TLS record, save a reference to
 228          * the record.  This local reference is used to call
 229          * ktls_output_eagain after the mbuf has been freed (thus
 230          * dropping the mbuf's reference) in if_output.
 231          */
 232         if (m->m_next != NULL && mbuf_has_tls_session(m->m_next)) {
 233                 tls = ktls_hold(m->m_next->m_epg_tls);
 234                 mst = tls->snd_tag;
 235
 236                 /*
 237                  * If a TLS session doesn't have a valid tag, it must
 238                  * have had an earlier ifp mismatch, so drop this
 239                  * packet.
 240                  */
 241                 if (mst == NULL) {
 242                         error = EAGAIN;
 243                         goto done;
 244                 }
 245                 /*
 246                  * Always stamp tags that include NIC ktls.
 247                  */
 248                 stamp_tag = true;
 249         }
 250 #endif
 251 #ifdef RATELIMIT
 252         if (inp != NULL && mst == NULL) {
 253                 if ((inp->inp_flags2 & INP_RATE_LIMIT_CHANGED) != 0 ||
 254                     (inp->inp_snd_tag != NULL &&
 255                     inp->inp_snd_tag->ifp != ifp))
 256                         in_pcboutput_txrtlmt(inp, ifp, m);
 257
 258                 if (inp->inp_snd_tag != NULL)
 259                         mst = inp->inp_snd_tag;
 260         }
 261 #endif
 262         if (stamp_tag && mst != NULL) {
 263                 KASSERT(m->m_pkthdr.rcvif == NULL,
 264                     ("trying to add a send tag to a forwarded packet"));
 265                 if (mst->ifp != ifp) {
 266                         error = EAGAIN;
 267                         goto done;
 268                 }
 269
 270                 /* stamp send tag on mbuf */
 271                 m->m_pkthdr.snd_tag = m_snd_tag_ref(mst);
 272                 m->m_pkthdr.csum_flags |= CSUM_SND_TAG;
 273         }
 274
 275         error = (*ifp->if_output)(ifp, m, gw, ro);
 276
 277 done:
 278         /* Check for route change invalidating send tags. */
 279 #ifdef KERN_TLS
 280         if (tls != NULL) {
 281                 if (error == EAGAIN)
 282                         error = ktls_output_eagain(inp, tls);
 283                 ktls_free(tls);
 284         }
 285 #endif
 286 #ifdef RATELIMIT
 287         if (error == EAGAIN)
 288                 in_pcboutput_eagain(inp);
 289 #endif
 290         return (error);
 291 }
 292
 293 /* rte<>ro_flags translation */
 294 static inline void
 295 rt_update_ro_flags(struct route *ro, const struct nhop_object *nh)
 296 {
 297         int nh_flags = nh->nh_flags;
 298
 299         ro->ro_flags &= ~ (RT_REJECT|RT_BLACKHOLE|RT_HAS_GW);
 300
 301         ro->ro_flags |= (nh_flags & NHF_REJECT) ? RT_REJECT : 0;
 302         ro->ro_flags |= (nh_flags & NHF_BLACKHOLE) ? RT_BLACKHOLE : 0;
 303         ro->ro_flags |= (nh_flags & NHF_GATEWAY) ? RT_HAS_GW : 0;
 304 }
 305
 306 /*
 307  * IP output.  The packet in mbuf chain m contains a skeletal IP
 308  * header (with len, off, ttl, proto, tos, src, dst).
 309  * The mbuf chain containing the packet will be freed.
 310  * The mbuf opt, if present, will not be freed.
 311  * If route ro is present and has ro_rt initialized, route lookup would be
 312  * skipped and ro->ro_rt would be used. If ro is present but ro->ro_rt is NULL,
 313  * then result of route lookup is stored in ro->ro_rt.
 314  *
 315  * In the IP forwarding case, the packet will arrive with options already
 316  * inserted, so must have a NULL opt pointer.
 317  */
 318 int
 319 ip_output(struct mbuf *m, struct mbuf *opt, struct route *ro, int flags,
 320     struct ip_moptions *imo, struct inpcb *inp)
 321 {
 322         MROUTER_RLOCK_TRACKER;
 323         struct ip *ip;
 324         struct ifnet *ifp = NULL;       /* keep compiler happy */
 325         struct mbuf *m0;
 326         int hlen = sizeof (struct ip);
 327         int mtu = 0;
 328         int error = 0;
 329         int vlan_pcp = -1;
 330         struct sockaddr_in *dst;
 331         const struct sockaddr *gw;
 332         struct in_ifaddr *ia = NULL;
 333         struct in_addr src;
 334         int isbroadcast;
 335         uint16_t ip_len, ip_off;
 336         struct route iproute;
 337         uint32_t fibnum;
 338 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 339         int no_route_but_check_spd = 0;
 340 #endif
 341
 342         M_ASSERTPKTHDR(m);
 343         NET_EPOCH_ASSERT();
 344
 345         if (inp != NULL) {
 346                 INP_LOCK_ASSERT(inp);
 347                 M_SETFIB(m, inp->inp_inc.inc_fibnum);
 348                 if ((flags & IP_NODEFAULTFLOWID) == 0) {
 349                         m->m_pkthdr.flowid = inp->inp_flowid;
 350                         M_HASHTYPE_SET(m, inp->inp_flowtype);
 351                 }
 352                 if ((inp->inp_flags2 & INP_2PCP_SET) != 0)
 353                         vlan_pcp = (inp->inp_flags2 & INP_2PCP_MASK) >>
 354                             INP_2PCP_SHIFT;
 355 #ifdef NUMA
 356                 m->m_pkthdr.numa_domain = inp->inp_numa_domain;
 357 #endif
 358         }
 359
 360         if (opt) {
 361                 int len = 0;
 362                 m = ip_insertoptions(m, opt, &len);
 363                 if (len != 0)
 364                         hlen = len; /* ip->ip_hl is updated above */
 365         }
 366         ip = mtod(m, struct ip *);
 367         ip_len = ntohs(ip->ip_len);
 368         ip_off = ntohs(ip->ip_off);
 369
 370         if ((flags & (IP_FORWARDING|IP_RAWOUTPUT)) == 0) {
 371                 ip->ip_v = IPVERSION;
 372                 ip->ip_hl = hlen >> 2;
 373                 ip_fillid(ip);
 374         } else {
 375                 /* Header already set, fetch hlen from there */
 376                 hlen = ip->ip_hl << 2;
 377         }
 378         if ((flags & IP_FORWARDING) == 0)
 379                 IPSTAT_INC(ips_localout);
 380
 381         /*
 382          * dst/gw handling:
 383          *
 384          * gw is readonly but can point either to dst OR rt_gateway,
 385          * therefore we need restore gw if we're redoing lookup.
 386          */
 387         fibnum = (inp != NULL) ? inp->inp_inc.inc_fibnum : M_GETFIB(m);
 388         if (ro == NULL) {
 389                 ro = &iproute;
 390                 bzero(ro, sizeof (*ro));
 391         }
 392         dst = (struct sockaddr_in *)&ro->ro_dst;
 393         if (ro->ro_nh == NULL) {
 394                 dst->sin_family = AF_INET;
 395                 dst->sin_len = sizeof(*dst);
 396                 dst->sin_addr = ip->ip_dst;
 397         }
 398         gw = (const struct sockaddr *)dst;
 399 again:
 400         /*
 401          * Validate route against routing table additions;
 402          * a better/more specific route might have been added.
 403          */
 404         if (inp != NULL && ro->ro_nh != NULL)
 405                 NH_VALIDATE(ro, &inp->inp_rt_cookie, fibnum);
 406         /*
 407          * If there is a cached route,
 408          * check that it is to the same destination
 409          * and is still up.  If not, free it and try again.
 410          * The address family should also be checked in case of sharing the
 411          * cache with IPv6.
 412          * Also check whether routing cache needs invalidation.
 413          */
 414         if (ro->ro_nh != NULL &&
 415             ((!NH_IS_VALID(ro->ro_nh)) || dst->sin_family != AF_INET ||
 416             dst->sin_addr.s_addr != ip->ip_dst.s_addr))
 417                 RO_INVALIDATE_CACHE(ro);
 418         ia = NULL;
 419         /*
 420          * If routing to interface only, short circuit routing lookup.
 421          * The use of an all-ones broadcast address implies this; an
 422          * interface is specified by the broadcast address of an interface,
 423          * or the destination address of a ptp interface.
 424          */
 425         if (flags & IP_SENDONES) {
 426                 if ((ia = ifatoia(ifa_ifwithbroadaddr(sintosa(dst),
 427                                                       M_GETFIB(m)))) == NULL &&
 428                     (ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 429                                                     M_GETFIB(m)))) == NULL) {
 430                         IPSTAT_INC(ips_noroute);
 431                         error = ENETUNREACH;
 432                         goto bad;
 433                 }
 434                 ip->ip_dst.s_addr = INADDR_BROADCAST;
 435                 dst->sin_addr = ip->ip_dst;
 436                 ifp = ia->ia_ifp;
 437                 mtu = ifp->if_mtu;
 438                 ip->ip_ttl = 1;
 439                 isbroadcast = 1;
 440                 src = IA_SIN(ia)->sin_addr;
 441         } else if (flags & IP_ROUTETOIF) {
 442                 if ((ia = ifatoia(ifa_ifwithdstaddr(sintosa(dst),
 443                                                     M_GETFIB(m)))) == NULL &&
 444                     (ia = ifatoia(ifa_ifwithnet(sintosa(dst), 0,
 445                                                 M_GETFIB(m)))) == NULL) {
 446                         IPSTAT_INC(ips_noroute);
 447                         error = ENETUNREACH;
 448                         goto bad;
 449                 }
 450                 ifp = ia->ia_ifp;
 451                 mtu = ifp->if_mtu;
 452                 ip->ip_ttl = 1;
 453                 isbroadcast = ifp->if_flags & IFF_BROADCAST ?
 454                     in_ifaddr_broadcast(dst->sin_addr, ia) : 0;
 455                 src = IA_SIN(ia)->sin_addr;
 456         } else if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) &&
 457             imo != NULL && imo->imo_multicast_ifp != NULL) {
 458                 /*
 459                  * Bypass the normal routing lookup for multicast
 460                  * packets if the interface is specified.
 461                  */
 462                 ifp = imo->imo_multicast_ifp;
 463                 mtu = ifp->if_mtu;
 464                 IFP_TO_IA(ifp, ia);
 465                 isbroadcast = 0;        /* fool gcc */
 466                 /* Interface may have no addresses. */
 467                 if (ia != NULL)
 468                         src = IA_SIN(ia)->sin_addr;
 469                 else
 470                         src.s_addr = INADDR_ANY;
 471         } else if (ro != &iproute) {
 472                 if (ro->ro_nh == NULL) {
 473                         /*
 474                          * We want to do any cloning requested by the link
 475                          * layer, as this is probably required in all cases
 476                          * for correct operation (as it is for ARP).
 477                          */
 478                         uint32_t flowid;
 479                         flowid = m->m_pkthdr.flowid;
 480                         ro->ro_nh = fib4_lookup(fibnum, dst->sin_addr, 0,
 481                             NHR_REF, flowid);
 482
 483                         if (ro->ro_nh == NULL || (!NH_IS_VALID(ro->ro_nh))) {
 484 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 485                                 /*
 486                                  * There is no route for this packet, but it is
 487                                  * possible that a matching SPD entry exists.
 488                                  */
 489                                 no_route_but_check_spd = 1;
 490                                 goto sendit;
 491 #endif
 492                                 IPSTAT_INC(ips_noroute);
 493                                 error = EHOSTUNREACH;
 494                                 goto bad;
 495                         }
 496                 }
 497                 struct nhop_object *nh = ro->ro_nh;
 498
 499                 ia = ifatoia(nh->nh_ifa);
 500                 ifp = nh->nh_ifp;
 501                 counter_u64_add(nh->nh_pksent, 1);
 502                 rt_update_ro_flags(ro, nh);
 503                 if (nh->nh_flags & NHF_GATEWAY)
 504                         gw = &nh->gw_sa;
 505                 if (nh->nh_flags & NHF_HOST)
 506                         isbroadcast = (nh->nh_flags & NHF_BROADCAST);
 507                 else if ((ifp->if_flags & IFF_BROADCAST) && (gw->sa_family == AF_INET))
 508                         isbroadcast = in_ifaddr_broadcast(((const struct sockaddr_in *)gw)->sin_addr, ia);
 509                 else
 510                         isbroadcast = 0;
 511                 mtu = nh->nh_mtu;
 512                 src = IA_SIN(ia)->sin_addr;
 513         } else {
 514                 struct nhop_object *nh;
 515
 516                 nh = fib4_lookup(M_GETFIB(m), ip->ip_dst, 0, NHR_NONE,
 517                     m->m_pkthdr.flowid);
 518                 if (nh == NULL) {
 519 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 520                         /*
 521                          * There is no route for this packet, but it is
 522                          * possible that a matching SPD entry exists.
 523                          */
 524                         no_route_but_check_spd = 1;
 525                         goto sendit;
 526 #endif
 527                         IPSTAT_INC(ips_noroute);
 528                         error = EHOSTUNREACH;
 529                         goto bad;
 530                 }
 531                 ifp = nh->nh_ifp;
 532                 mtu = nh->nh_mtu;
 533                 rt_update_ro_flags(ro, nh);
 534                 if (nh->nh_flags & NHF_GATEWAY)
 535                         gw = &nh->gw_sa;
 536                 ia = ifatoia(nh->nh_ifa);
 537                 src = IA_SIN(ia)->sin_addr;
 538                 isbroadcast = (((nh->nh_flags & (NHF_HOST | NHF_BROADCAST)) ==
 539                     (NHF_HOST | NHF_BROADCAST)) ||
 540                     ((ifp->if_flags & IFF_BROADCAST) &&
 541                     (gw->sa_family == AF_INET) &&
 542                     in_ifaddr_broadcast(((const struct sockaddr_in *)gw)->sin_addr, ia)));
 543         }
 544
 545         /* Catch a possible divide by zero later. */
 546         KASSERT(mtu > 0, ("%s: mtu %d <= 0, ro=%p (nh_flags=0x%08x) ifp=%p",
 547             __func__, mtu, ro,
 548             (ro != NULL && ro->ro_nh != NULL) ? ro->ro_nh->nh_flags : 0, ifp));
 549
 550         if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
 551                 m->m_flags |= M_MCAST;
 552                 /*
 553                  * IP destination address is multicast.  Make sure "gw"
 554                  * still points to the address in "ro".  (It may have been
 555                  * changed to point to a gateway address, above.)
 556                  */
 557                 gw = (const struct sockaddr *)dst;
 558                 /*
 559                  * See if the caller provided any multicast options
 560                  */
 561                 if (imo != NULL) {
 562                         ip->ip_ttl = imo->imo_multicast_ttl;
 563                         if (imo->imo_multicast_vif != -1)
 564                                 ip->ip_src.s_addr =
 565                                     ip_mcast_src ?
 566                                     ip_mcast_src(imo->imo_multicast_vif) :
 567                                     INADDR_ANY;
 568                 } else
 569                         ip->ip_ttl = IP_DEFAULT_MULTICAST_TTL;
 570                 /*
 571                  * Confirm that the outgoing interface supports multicast.
 572                  */
 573                 if ((imo == NULL) || (imo->imo_multicast_vif == -1)) {
 574                         if ((ifp->if_flags & IFF_MULTICAST) == 0) {
 575                                 IPSTAT_INC(ips_noroute);
 576                                 error = ENETUNREACH;
 577                                 goto bad;
 578                         }
 579                 }
 580                 /*
 581                  * If source address not specified yet, use address
 582                  * of outgoing interface.
 583                  */
 584                 if (ip->ip_src.s_addr == INADDR_ANY)
 585                         ip->ip_src = src;
 586
 587                 if ((imo == NULL && in_mcast_loop) ||
 588                     (imo && imo->imo_multicast_loop)) {
 589                         /*
 590                          * Loop back multicast datagram if not expressly
 591                          * forbidden to do so, even if we are not a member
 592                          * of the group; ip_input() will filter it later,
 593                          * thus deferring a hash lookup and mutex acquisition
 594                          * at the expense of a cheap copy using m_copym().
 595                          */
 596                         ip_mloopback(ifp, m, hlen);
 597                 } else {
 598                         /*
 599                          * If we are acting as a multicast router, perform
 600                          * multicast forwarding as if the packet had just
 601                          * arrived on the interface to which we are about
 602                          * to send.  The multicast forwarding function
 603                          * recursively calls this function, using the
 604                          * IP_FORWARDING flag to prevent infinite recursion.
 605                          *
 606                          * Multicasts that are looped back by ip_mloopback(),
 607                          * above, will be forwarded by the ip_input() routine,
 608                          * if necessary.
 609                          */
 610                         MROUTER_RLOCK();
 611                         if (V_ip_mrouter && (flags & IP_FORWARDING) == 0) {
 612                                 /*
 613                                  * If rsvp daemon is not running, do not
 614                                  * set ip_moptions. This ensures that the packet
 615                                  * is multicast and not just sent down one link
 616                                  * as prescribed by rsvpd.
 617                                  */
 618                                 if (!V_rsvp_on)
 619                                         imo = NULL;
 620                                 if (ip_mforward &&
 621                                     ip_mforward(ip, ifp, m, imo) != 0) {
 622                                         MROUTER_RUNLOCK();
 623                                         m_freem(m);
 624                                         goto done;
 625                                 }
 626                         }
 627                         MROUTER_RUNLOCK();
 628                 }
 629
 630                 /*
 631                  * Multicasts with a time-to-live of zero may be looped-
 632                  * back, above, but must not be transmitted on a network.
 633                  * Also, multicasts addressed to the loopback interface
 634                  * are not sent -- the above call to ip_mloopback() will
 635                  * loop back a copy. ip_input() will drop the copy if
 636                  * this host does not belong to the destination group on
 637                  * the loopback interface.
 638                  */
 639                 if (ip->ip_ttl == 0 || ifp->if_flags & IFF_LOOPBACK) {
 640                         m_freem(m);
 641                         goto done;
 642                 }
 643
 644                 goto sendit;
 645         }
 646
 647         /*
 648          * If the source address is not specified yet, use the address
 649          * of the outoing interface.
 650          */
 651         if (ip->ip_src.s_addr == INADDR_ANY)
 652                 ip->ip_src = src;
 653
 654         /*
 655          * Look for broadcast address and
 656          * verify user is allowed to send
 657          * such a packet.
 658          */
 659         if (isbroadcast) {
 660                 if ((ifp->if_flags & IFF_BROADCAST) == 0) {
 661                         error = EADDRNOTAVAIL;
 662                         goto bad;
 663                 }
 664                 if ((flags & IP_ALLOWBROADCAST) == 0) {
 665                         error = EACCES;
 666                         goto bad;
 667                 }
 668                 /* don't allow broadcast messages to be fragmented */
 669                 if (ip_len > mtu) {
 670                         error = EMSGSIZE;
 671                         goto bad;
 672                 }
 673                 m->m_flags |= M_BCAST;
 674         } else {
 675                 m->m_flags &= ~M_BCAST;
 676         }
 677
 678 sendit:
 679 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 680         if (IPSEC_ENABLED(ipv4)) {
 681                 if ((error = IPSEC_OUTPUT(ipv4, m, inp)) != 0) {
 682                         if (error == EINPROGRESS)
 683                                 error = 0;
 684                         goto done;
 685                 }
 686         }
 687         /*
 688          * Check if there was a route for this packet; return error if not.
 689          */
 690         if (no_route_but_check_spd) {
 691                 IPSTAT_INC(ips_noroute);
 692                 error = EHOSTUNREACH;
 693                 goto bad;
 694         }
 695         /* Update variables that are affected by ipsec4_output(). */
 696         ip = mtod(m, struct ip *);
 697         hlen = ip->ip_hl << 2;
 698 #endif /* IPSEC */
 699
 700         /* Jump over all PFIL processing if hooks are not active. */
 701         if (PFIL_HOOKED_OUT(V_inet_pfil_head)) {
 702                 switch (ip_output_pfil(&m, ifp, flags, inp, dst, &fibnum,
 703                     &error)) {
 704                 case 1: /* Finished */
 705                         goto done;
 706
 707                 case 0: /* Continue normally */
 708                         ip = mtod(m, struct ip *);
 709                         break;
 710
 711                 case -1: /* Need to try again */
 712                         /* Reset everything for a new round */
 713                         if (ro != NULL) {
 714                                 RO_NHFREE(ro);
 715                                 ro->ro_prepend = NULL;
 716                         }
 717                         gw = (const struct sockaddr *)dst;
 718                         ip = mtod(m, struct ip *);
 719                         goto again;
 720                 }
 721         }
 722
 723         if (vlan_pcp > -1)
 724                 EVL_APPLY_PRI(m, vlan_pcp);
 725
 726         /* IN_LOOPBACK must not appear on the wire - RFC1122. */
 727         if (IN_LOOPBACK(ntohl(ip->ip_dst.s_addr)) ||
 728             IN_LOOPBACK(ntohl(ip->ip_src.s_addr))) {
 729                 if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
 730                         IPSTAT_INC(ips_badaddr);
 731                         error = EADDRNOTAVAIL;
 732                         goto bad;
 733                 }
 734         }
 735
 736         /* Ensure the packet data is mapped if the interface requires it. */
 737         if ((ifp->if_capenable & IFCAP_MEXTPG) == 0) {
 738                 m = mb_unmapped_to_ext(m);
 739                 if (m == NULL) {
 740                         IPSTAT_INC(ips_odropped);
 741                         error = ENOBUFS;
 742                         goto bad;
 743                 }
 744         }
 745
 746         m->m_pkthdr.csum_flags |= CSUM_IP;
 747         if (m->m_pkthdr.csum_flags & CSUM_DELAY_DATA & ~ifp->if_hwassist) {
 748                 in_delayed_cksum(m);
 749                 m->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 750         }
 751 #if defined(SCTP) || defined(SCTP_SUPPORT)
 752         if (m->m_pkthdr.csum_flags & CSUM_SCTP & ~ifp->if_hwassist) {
 753                 sctp_delayed_cksum(m, (uint32_t)(ip->ip_hl << 2));
 754                 m->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 755         }
 756 #endif
 757
 758         /*
 759          * If small enough for interface, or the interface will take
 760          * care of the fragmentation for us, we can just send directly.
 761          * Note that if_vxlan could have requested TSO even though the outer
 762          * frame is UDP.  It is correct to not fragment such datagrams and
 763          * instead just pass them on to the driver.
 764          */
 765         if (ip_len <= mtu ||
 766             (m->m_pkthdr.csum_flags & ifp->if_hwassist &
 767             (CSUM_TSO | CSUM_INNER_TSO)) != 0) {
 768                 ip->ip_sum = 0;
 769                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~ifp->if_hwassist) {
 770                         ip->ip_sum = in_cksum(m, hlen);
 771                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
 772                 }
 773
 774                 /*
 775                  * Record statistics for this interface address.
 776                  * With CSUM_TSO the byte/packet count will be slightly
 777                  * incorrect because we count the IP+TCP headers only
 778                  * once instead of for every generated packet.
 779                  */
 780                 if (!(flags & IP_FORWARDING) && ia) {
 781                         if (m->m_pkthdr.csum_flags &
 782                             (CSUM_TSO | CSUM_INNER_TSO))
 783                                 counter_u64_add(ia->ia_ifa.ifa_opackets,
 784                                     m->m_pkthdr.len / m->m_pkthdr.tso_segsz);
 785                         else
 786                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 787
 788                         counter_u64_add(ia->ia_ifa.ifa_obytes, m->m_pkthdr.len);
 789                 }
 790 #ifdef MBUF_STRESS_TEST
 791                 if (mbuf_frag_size && m->m_pkthdr.len > mbuf_frag_size)
 792                         m = m_fragment(m, M_NOWAIT, mbuf_frag_size);
 793 #endif
 794                 /*
 795                  * Reset layer specific mbuf flags
 796                  * to avoid confusing lower layers.
 797                  */
 798                 m_clrprotoflags(m);
 799                 IP_PROBE(send, NULL, NULL, ip, ifp, ip, NULL);
 800                 error = ip_output_send(inp, ifp, m, gw, ro,
 801                     (flags & IP_NO_SND_TAG_RL) ? false : true);
 802                 goto done;
 803         }
 804
 805         /* Balk when DF bit is set or the interface didn't support TSO. */
 806         if ((ip_off & IP_DF) ||
 807             (m->m_pkthdr.csum_flags & (CSUM_TSO | CSUM_INNER_TSO))) {
 808                 error = EMSGSIZE;
 809                 IPSTAT_INC(ips_cantfrag);
 810                 goto bad;
 811         }
 812
 813         /*
 814          * Too large for interface; fragment if possible. If successful,
 815          * on return, m will point to a list of packets to be sent.
 816          */
 817         error = ip_fragment(ip, &m, mtu, ifp->if_hwassist);
 818         if (error)
 819                 goto bad;
 820         for (; m; m = m0) {
 821                 m0 = m->m_nextpkt;
 822                 m->m_nextpkt = 0;
 823                 if (error == 0) {
 824                         /* Record statistics for this interface address. */
 825                         if (ia != NULL) {
 826                                 counter_u64_add(ia->ia_ifa.ifa_opackets, 1);
 827                                 counter_u64_add(ia->ia_ifa.ifa_obytes,
 828                                     m->m_pkthdr.len);
 829                         }
 830                         /*
 831                          * Reset layer specific mbuf flags
 832                          * to avoid confusing upper layers.
 833                          */
 834                         m_clrprotoflags(m);
 835
 836                         IP_PROBE(send, NULL, NULL, mtod(m, struct ip *), ifp,
 837                             mtod(m, struct ip *), NULL);
 838                         error = ip_output_send(inp, ifp, m, gw, ro, true);
 839                 } else
 840                         m_freem(m);
 841         }
 842
 843         if (error == 0)
 844                 IPSTAT_INC(ips_fragmented);
 845
 846 done:
 847         return (error);
 848  bad:
 849         m_freem(m);
 850         goto done;
 851 }
 852
 853 /*
 854  * Create a chain of fragments which fit the given mtu. m_frag points to the
 855  * mbuf to be fragmented; on return it points to the chain with the fragments.
 856  * Return 0 if no error. If error, m_frag may contain a partially built
 857  * chain of fragments that should be freed by the caller.
 858  *
 859  * if_hwassist_flags is the hw offload capabilities (see if_data.ifi_hwassist)
 860  */
 861 int
 862 ip_fragment(struct ip *ip, struct mbuf **m_frag, int mtu,
 863     u_long if_hwassist_flags)
 864 {
 865         int error = 0;
 866         int hlen = ip->ip_hl << 2;
 867         int len = (mtu - hlen) & ~7;    /* size of payload in each fragment */
 868         int off;
 869         struct mbuf *m0 = *m_frag;      /* the original packet          */
 870         int firstlen;
 871         struct mbuf **mnext;
 872         int nfrags;
 873         uint16_t ip_len, ip_off;
 874
 875         ip_len = ntohs(ip->ip_len);
 876         ip_off = ntohs(ip->ip_off);
 877
 878         if (ip_off & IP_DF) {   /* Fragmentation not allowed */
 879                 IPSTAT_INC(ips_cantfrag);
 880                 return EMSGSIZE;
 881         }
 882
 883         /*
 884          * Must be able to put at least 8 bytes per fragment.
 885          */
 886         if (len < 8)
 887                 return EMSGSIZE;
 888
 889         /*
 890          * If the interface will not calculate checksums on
 891          * fragmented packets, then do it here.
 892          */
 893         if (m0->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
 894                 in_delayed_cksum(m0);
 895                 m0->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
 896         }
 897 #if defined(SCTP) || defined(SCTP_SUPPORT)
 898         if (m0->m_pkthdr.csum_flags & CSUM_SCTP) {
 899                 sctp_delayed_cksum(m0, hlen);
 900                 m0->m_pkthdr.csum_flags &= ~CSUM_SCTP;
 901         }
 902 #endif
 903         if (len > PAGE_SIZE) {
 904                 /*
 905                  * Fragment large datagrams such that each segment
 906                  * contains a multiple of PAGE_SIZE amount of data,
 907                  * plus headers. This enables a receiver to perform
 908                  * page-flipping zero-copy optimizations.
 909                  *
 910                  * XXX When does this help given that sender and receiver
 911                  * could have different page sizes, and also mtu could
 912                  * be less than the receiver's page size ?
 913                  */
 914                 int newlen;
 915
 916                 off = MIN(mtu, m0->m_pkthdr.len);
 917
 918                 /*
 919                  * firstlen (off - hlen) must be aligned on an
 920                  * 8-byte boundary
 921                  */
 922                 if (off < hlen)
 923                         goto smart_frag_failure;
 924                 off = ((off - hlen) & ~7) + hlen;
 925                 newlen = (~PAGE_MASK) & mtu;
 926                 if ((newlen + sizeof (struct ip)) > mtu) {
 927                         /* we failed, go back the default */
 928 smart_frag_failure:
 929                         newlen = len;
 930                         off = hlen + len;
 931                 }
 932                 len = newlen;
 933
 934         } else {
 935                 off = hlen + len;
 936         }
 937
 938         firstlen = off - hlen;
 939         mnext = &m0->m_nextpkt;         /* pointer to next packet */
 940
 941         /*
 942          * Loop through length of segment after first fragment,
 943          * make new header and copy data of each part and link onto chain.
 944          * Here, m0 is the original packet, m is the fragment being created.
 945          * The fragments are linked off the m_nextpkt of the original
 946          * packet, which after processing serves as the first fragment.
 947          */
 948         for (nfrags = 1; off < ip_len; off += len, nfrags++) {
 949                 struct ip *mhip;        /* ip header on the fragment */
 950                 struct mbuf *m;
 951                 int mhlen = sizeof (struct ip);
 952
 953                 m = m_gethdr(M_NOWAIT, MT_DATA);
 954                 if (m == NULL) {
 955                         error = ENOBUFS;
 956                         IPSTAT_INC(ips_odropped);
 957                         goto done;
 958                 }
 959                 /*
 960                  * Make sure the complete packet header gets copied
 961                  * from the originating mbuf to the newly created
 962                  * mbuf. This also ensures that existing firewall
 963                  * classification(s), VLAN tags and so on get copied
 964                  * to the resulting fragmented packet(s):
 965                  */
 966                 if (m_dup_pkthdr(m, m0, M_NOWAIT) == 0) {
 967                         m_free(m);
 968                         error = ENOBUFS;
 969                         IPSTAT_INC(ips_odropped);
 970                         goto done;
 971                 }
 972                 /*
 973                  * In the first mbuf, leave room for the link header, then
 974                  * copy the original IP header including options. The payload
 975                  * goes into an additional mbuf chain returned by m_copym().
 976                  */
 977                 m->m_data += max_linkhdr;
 978                 mhip = mtod(m, struct ip *);
 979                 *mhip = *ip;
 980                 if (hlen > sizeof (struct ip)) {
 981                         mhlen = ip_optcopy(ip, mhip) + sizeof (struct ip);
 982                         mhip->ip_v = IPVERSION;
 983                         mhip->ip_hl = mhlen >> 2;
 984                 }
 985                 m->m_len = mhlen;
 986                 /* XXX do we need to add ip_off below ? */
 987                 mhip->ip_off = ((off - hlen) >> 3) + ip_off;
 988                 if (off + len >= ip_len)
 989                         len = ip_len - off;
 990                 else
 991                         mhip->ip_off |= IP_MF;
 992                 mhip->ip_len = htons((u_short)(len + mhlen));
 993                 m->m_next = m_copym(m0, off, len, M_NOWAIT);
 994                 if (m->m_next == NULL) {        /* copy failed */
 995                         m_free(m);
 996                         error = ENOBUFS;        /* ??? */
 997                         IPSTAT_INC(ips_odropped);
 998                         goto done;
 999                 }
1000                 m->m_pkthdr.len = mhlen + len;
1001 #ifdef MAC
1002                 mac_netinet_fragment(m0, m);
1003 #endif
1004                 mhip->ip_off = htons(mhip->ip_off);
1005                 mhip->ip_sum = 0;
1006                 if (m->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
1007                         mhip->ip_sum = in_cksum(m, mhlen);
1008                         m->m_pkthdr.csum_flags &= ~CSUM_IP;
1009                 }
1010                 *mnext = m;
1011                 mnext = &m->m_nextpkt;
1012         }
1013         IPSTAT_ADD(ips_ofragments, nfrags);
1014
1015         /*
1016          * Update first fragment by trimming what's been copied out
1017          * and updating header.
1018          */
1019         m_adj(m0, hlen + firstlen - ip_len);
1020         m0->m_pkthdr.len = hlen + firstlen;
1021         ip->ip_len = htons((u_short)m0->m_pkthdr.len);
1022         ip->ip_off = htons(ip_off | IP_MF);
1023         ip->ip_sum = 0;
1024         if (m0->m_pkthdr.csum_flags & CSUM_IP & ~if_hwassist_flags) {
1025                 ip->ip_sum = in_cksum(m0, hlen);
1026                 m0->m_pkthdr.csum_flags &= ~CSUM_IP;
1027         }
1028
1029 done:
1030         *m_frag = m0;
1031         return error;
1032 }
1033
1034 void
1035 in_delayed_cksum(struct mbuf *m)
1036 {
1037         struct ip *ip;
1038         struct udphdr *uh;
1039         uint16_t cklen, csum, offset;
1040
1041         ip = mtod(m, struct ip *);
1042         offset = ip->ip_hl << 2 ;
1043
1044         if (m->m_pkthdr.csum_flags & CSUM_UDP) {
1045                 /* if udp header is not in the first mbuf copy udplen */
1046                 if (offset + sizeof(struct udphdr) > m->m_len) {
1047                         m_copydata(m, offset + offsetof(struct udphdr,
1048                             uh_ulen), sizeof(cklen), (caddr_t)&cklen);
1049                         cklen = ntohs(cklen);
1050                 } else {
1051                         uh = (struct udphdr *)mtodo(m, offset);
1052                         cklen = ntohs(uh->uh_ulen);
1053                 }
1054                 csum = in_cksum_skip(m, cklen + offset, offset);
1055                 if (csum == 0)
1056                         csum = 0xffff;
1057         } else {
1058                 cklen = ntohs(ip->ip_len);
1059                 csum = in_cksum_skip(m, cklen, offset);
1060         }
1061         offset += m->m_pkthdr.csum_data;        /* checksum offset */
1062
1063         if (offset + sizeof(csum) > m->m_len)
1064                 m_copyback(m, offset, sizeof(csum), (caddr_t)&csum);
1065         else
1066                 *(u_short *)mtodo(m, offset) = csum;
1067 }
1068
1069 /*
1070  * IP socket option processing.
1071  */
1072 int
1073 ip_ctloutput(struct socket *so, struct sockopt *sopt)
1074 {
1075         struct inpcb *inp = sotoinpcb(so);
1076         int     error, optval;
1077 #ifdef  RSS
1078         uint32_t rss_bucket;
1079         int retval;
1080 #endif
1081
1082         error = optval = 0;
1083         if (sopt->sopt_level != IPPROTO_IP) {
1084                 error = EINVAL;
1085
1086                 if (sopt->sopt_level == SOL_SOCKET &&
1087                     sopt->sopt_dir == SOPT_SET) {
1088                         switch (sopt->sopt_name) {
1089                         case SO_REUSEADDR:
1090                                 INP_WLOCK(inp);
1091                                 if ((so->so_options & SO_REUSEADDR) != 0)
1092                                         inp->inp_flags2 |= INP_REUSEADDR;
1093                                 else
1094                                         inp->inp_flags2 &= ~INP_REUSEADDR;
1095                                 INP_WUNLOCK(inp);
1096                                 error = 0;
1097                                 break;
1098                         case SO_REUSEPORT:
1099                                 INP_WLOCK(inp);
1100                                 if ((so->so_options & SO_REUSEPORT) != 0)
1101                                         inp->inp_flags2 |= INP_REUSEPORT;
1102                                 else
1103                                         inp->inp_flags2 &= ~INP_REUSEPORT;
1104                                 INP_WUNLOCK(inp);
1105                                 error = 0;
1106                                 break;
1107                         case SO_REUSEPORT_LB:
1108                                 INP_WLOCK(inp);
1109                                 if ((so->so_options & SO_REUSEPORT_LB) != 0)
1110                                         inp->inp_flags2 |= INP_REUSEPORT_LB;
1111                                 else
1112                                         inp->inp_flags2 &= ~INP_REUSEPORT_LB;
1113                                 INP_WUNLOCK(inp);
1114                                 error = 0;
1115                                 break;
1116                         case SO_SETFIB:
1117                                 INP_WLOCK(inp);
1118                                 inp->inp_inc.inc_fibnum = so->so_fibnum;
1119                                 INP_WUNLOCK(inp);
1120                                 error = 0;
1121                                 break;
1122                         case SO_MAX_PACING_RATE:
1123 #ifdef RATELIMIT
1124                                 INP_WLOCK(inp);
1125                                 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
1126                                 INP_WUNLOCK(inp);
1127                                 error = 0;
1128 #else
1129                                 error = EOPNOTSUPP;
1130 #endif
1131                                 break;
1132                         default:
1133                                 break;
1134                         }
1135                 }
1136                 return (error);
1137         }
1138
1139         switch (sopt->sopt_dir) {
1140         case SOPT_SET:
1141                 switch (sopt->sopt_name) {
1142                 case IP_OPTIONS:
1143 #ifdef notyet
1144                 case IP_RETOPTS:
1145 #endif
1146                 {
1147                         struct mbuf *m;
1148                         if (sopt->sopt_valsize > MLEN) {
1149                                 error = EMSGSIZE;
1150                                 break;
1151                         }
1152                         m = m_get(sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
1153                         if (m == NULL) {
1154                                 error = ENOBUFS;
1155                                 break;
1156                         }
1157                         m->m_len = sopt->sopt_valsize;
1158                         error = sooptcopyin(sopt, mtod(m, char *), m->m_len,
1159                                             m->m_len);
1160                         if (error) {
1161                                 m_free(m);
1162                                 break;
1163                         }
1164                         INP_WLOCK(inp);
1165                         error = ip_pcbopts(inp, sopt->sopt_name, m);
1166                         INP_WUNLOCK(inp);
1167                         return (error);
1168                 }
1169
1170                 case IP_BINDANY:
1171                         if (sopt->sopt_td != NULL) {
1172                                 error = priv_check(sopt->sopt_td,
1173                                     PRIV_NETINET_BINDANY);
1174                                 if (error)
1175                                         break;
1176                         }
1177                         /* FALLTHROUGH */
1178                 case IP_BINDMULTI:
1179 #ifdef  RSS
1180                 case IP_RSS_LISTEN_BUCKET:
1181 #endif
1182                 case IP_TOS:
1183                 case IP_TTL:
1184                 case IP_MINTTL:
1185                 case IP_RECVOPTS:
1186                 case IP_RECVRETOPTS:
1187                 case IP_ORIGDSTADDR:
1188                 case IP_RECVDSTADDR:
1189                 case IP_RECVTTL:
1190                 case IP_RECVIF:
1191                 case IP_ONESBCAST:
1192                 case IP_DONTFRAG:
1193                 case IP_RECVTOS:
1194                 case IP_RECVFLOWID:
1195 #ifdef  RSS
1196                 case IP_RECVRSSBUCKETID:
1197 #endif
1198                 case IP_VLAN_PCP:
1199                         error = sooptcopyin(sopt, &optval, sizeof optval,
1200                                             sizeof optval);
1201                         if (error)
1202                                 break;
1203
1204                         switch (sopt->sopt_name) {
1205                         case IP_TOS:
1206                                 inp->inp_ip_tos = optval;
1207                                 break;
1208
1209                         case IP_TTL:
1210                                 inp->inp_ip_ttl = optval;
1211                                 break;
1212
1213                         case IP_MINTTL:
1214                                 if (optval >= 0 && optval <= MAXTTL)
1215                                         inp->inp_ip_minttl = optval;
1216                                 else
1217                                         error = EINVAL;
1218                                 break;
1219
1220 #define OPTSET(bit) do {                                                \
1221         INP_WLOCK(inp);                                                 \
1222         if (optval)                                                     \
1223                 inp->inp_flags |= bit;                                  \
1224         else                                                            \
1225                 inp->inp_flags &= ~bit;                                 \
1226         INP_WUNLOCK(inp);                                               \
1227 } while (0)
1228
1229 #define OPTSET2(bit, val) do {                                          \
1230         INP_WLOCK(inp);                                                 \
1231         if (val)                                                        \
1232                 inp->inp_flags2 |= bit;                                 \
1233         else                                                            \
1234                 inp->inp_flags2 &= ~bit;                                \
1235         INP_WUNLOCK(inp);                                               \
1236 } while (0)
1237
1238                         case IP_RECVOPTS:
1239                                 OPTSET(INP_RECVOPTS);
1240                                 break;
1241
1242                         case IP_RECVRETOPTS:
1243                                 OPTSET(INP_RECVRETOPTS);
1244                                 break;
1245
1246                         case IP_RECVDSTADDR:
1247                                 OPTSET(INP_RECVDSTADDR);
1248                                 break;
1249
1250                         case IP_ORIGDSTADDR:
1251                                 OPTSET2(INP_ORIGDSTADDR, optval);
1252                                 break;
1253
1254                         case IP_RECVTTL:
1255                                 OPTSET(INP_RECVTTL);
1256                                 break;
1257
1258                         case IP_RECVIF:
1259                                 OPTSET(INP_RECVIF);
1260                                 break;
1261
1262                         case IP_ONESBCAST:
1263                                 OPTSET(INP_ONESBCAST);
1264                                 break;
1265                         case IP_DONTFRAG:
1266                                 OPTSET(INP_DONTFRAG);
1267                                 break;
1268                         case IP_BINDANY:
1269                                 OPTSET(INP_BINDANY);
1270                                 break;
1271                         case IP_RECVTOS:
1272                                 OPTSET(INP_RECVTOS);
1273                                 break;
1274                         case IP_BINDMULTI:
1275                                 OPTSET2(INP_BINDMULTI, optval);
1276                                 break;
1277                         case IP_RECVFLOWID:
1278                                 OPTSET2(INP_RECVFLOWID, optval);
1279                                 break;
1280 #ifdef  RSS
1281                         case IP_RSS_LISTEN_BUCKET:
1282                                 if ((optval >= 0) &&
1283                                     (optval < rss_getnumbuckets())) {
1284                                         inp->inp_rss_listen_bucket = optval;
1285                                         OPTSET2(INP_RSS_BUCKET_SET, 1);
1286                                 } else {
1287                                         error = EINVAL;
1288                                 }
1289                                 break;
1290                         case IP_RECVRSSBUCKETID:
1291                                 OPTSET2(INP_RECVRSSBUCKETID, optval);
1292                                 break;
1293 #endif
1294                         case IP_VLAN_PCP:
1295                                 if ((optval >= -1) && (optval <=
1296                                     (INP_2PCP_MASK >> INP_2PCP_SHIFT))) {
1297                                         if (optval == -1) {
1298                                                 INP_WLOCK(inp);
1299                                                 inp->inp_flags2 &=
1300                                                     ~(INP_2PCP_SET |
1301                                                       INP_2PCP_MASK);
1302                                                 INP_WUNLOCK(inp);
1303                                         } else {
1304                                                 INP_WLOCK(inp);
1305                                                 inp->inp_flags2 |=
1306                                                     INP_2PCP_SET;
1307                                                 inp->inp_flags2 &=
1308                                                     ~INP_2PCP_MASK;
1309                                                 inp->inp_flags2 |=
1310                                                     optval << INP_2PCP_SHIFT;
1311                                                 INP_WUNLOCK(inp);
1312                                         }
1313                                 } else
1314                                         error = EINVAL;
1315                                 break;
1316                         }
1317                         break;
1318 #undef OPTSET
1319 #undef OPTSET2
1320
1321                 /*
1322                  * Multicast socket options are processed by the in_mcast
1323                  * module.
1324                  */
1325                 case IP_MULTICAST_IF:
1326                 case IP_MULTICAST_VIF:
1327                 case IP_MULTICAST_TTL:
1328                 case IP_MULTICAST_LOOP:
1329                 case IP_ADD_MEMBERSHIP:
1330                 case IP_DROP_MEMBERSHIP:
1331                 case IP_ADD_SOURCE_MEMBERSHIP:
1332                 case IP_DROP_SOURCE_MEMBERSHIP:
1333                 case IP_BLOCK_SOURCE:
1334                 case IP_UNBLOCK_SOURCE:
1335                 case IP_MSFILTER:
1336                 case MCAST_JOIN_GROUP:
1337                 case MCAST_LEAVE_GROUP:
1338                 case MCAST_JOIN_SOURCE_GROUP:
1339                 case MCAST_LEAVE_SOURCE_GROUP:
1340                 case MCAST_BLOCK_SOURCE:
1341                 case MCAST_UNBLOCK_SOURCE:
1342                         error = inp_setmoptions(inp, sopt);
1343                         break;
1344
1345                 case IP_PORTRANGE:
1346                         error = sooptcopyin(sopt, &optval, sizeof optval,
1347                                             sizeof optval);
1348                         if (error)
1349                                 break;
1350
1351                         INP_WLOCK(inp);
1352                         switch (optval) {
1353                         case IP_PORTRANGE_DEFAULT:
1354                                 inp->inp_flags &= ~(INP_LOWPORT);
1355                                 inp->inp_flags &= ~(INP_HIGHPORT);
1356                                 break;
1357
1358                         case IP_PORTRANGE_HIGH:
1359                                 inp->inp_flags &= ~(INP_LOWPORT);
1360                                 inp->inp_flags |= INP_HIGHPORT;
1361                                 break;
1362
1363                         case IP_PORTRANGE_LOW:
1364                                 inp->inp_flags &= ~(INP_HIGHPORT);
1365                                 inp->inp_flags |= INP_LOWPORT;
1366                                 break;
1367
1368                         default:
1369                                 error = EINVAL;
1370                                 break;
1371                         }
1372                         INP_WUNLOCK(inp);
1373                         break;
1374
1375 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1376                 case IP_IPSEC_POLICY:
1377                         if (IPSEC_ENABLED(ipv4)) {
1378                                 error = IPSEC_PCBCTL(ipv4, inp, sopt);
1379                                 break;
1380                         }
1381                         /* FALLTHROUGH */
1382 #endif /* IPSEC */
1383
1384                 default:
1385                         error = ENOPROTOOPT;
1386                         break;
1387                 }
1388                 break;
1389
1390         case SOPT_GET:
1391                 switch (sopt->sopt_name) {
1392                 case IP_OPTIONS:
1393                 case IP_RETOPTS:
1394                         INP_RLOCK(inp);
1395                         if (inp->inp_options) {
1396                                 struct mbuf *options;
1397
1398                                 options = m_copym(inp->inp_options, 0,
1399                                     M_COPYALL, M_NOWAIT);
1400                                 INP_RUNLOCK(inp);
1401                                 if (options != NULL) {
1402                                         error = sooptcopyout(sopt,
1403                                                              mtod(options, char *),
1404                                                              options->m_len);
1405                                         m_freem(options);
1406                                 } else
1407                                         error = ENOMEM;
1408                         } else {
1409                                 INP_RUNLOCK(inp);
1410                                 sopt->sopt_valsize = 0;
1411                         }
1412                         break;
1413
1414                 case IP_TOS:
1415                 case IP_TTL:
1416                 case IP_MINTTL:
1417                 case IP_RECVOPTS:
1418                 case IP_RECVRETOPTS:
1419                 case IP_ORIGDSTADDR:
1420                 case IP_RECVDSTADDR:
1421                 case IP_RECVTTL:
1422                 case IP_RECVIF:
1423                 case IP_PORTRANGE:
1424                 case IP_ONESBCAST:
1425                 case IP_DONTFRAG:
1426                 case IP_BINDANY:
1427                 case IP_RECVTOS:
1428                 case IP_BINDMULTI:
1429                 case IP_FLOWID:
1430                 case IP_FLOWTYPE:
1431                 case IP_RECVFLOWID:
1432 #ifdef  RSS
1433                 case IP_RSSBUCKETID:
1434                 case IP_RECVRSSBUCKETID:
1435 #endif
1436                 case IP_VLAN_PCP:
1437                         switch (sopt->sopt_name) {
1438                         case IP_TOS:
1439                                 optval = inp->inp_ip_tos;
1440                                 break;
1441
1442                         case IP_TTL:
1443                                 optval = inp->inp_ip_ttl;
1444                                 break;
1445
1446                         case IP_MINTTL:
1447                                 optval = inp->inp_ip_minttl;
1448                                 break;
1449
1450 #define OPTBIT(bit)     (inp->inp_flags & bit ? 1 : 0)
1451 #define OPTBIT2(bit)    (inp->inp_flags2 & bit ? 1 : 0)
1452
1453                         case IP_RECVOPTS:
1454                                 optval = OPTBIT(INP_RECVOPTS);
1455                                 break;
1456
1457                         case IP_RECVRETOPTS:
1458                                 optval = OPTBIT(INP_RECVRETOPTS);
1459                                 break;
1460
1461                         case IP_RECVDSTADDR:
1462                                 optval = OPTBIT(INP_RECVDSTADDR);
1463                                 break;
1464
1465                         case IP_ORIGDSTADDR:
1466                                 optval = OPTBIT2(INP_ORIGDSTADDR);
1467                                 break;
1468
1469                         case IP_RECVTTL:
1470                                 optval = OPTBIT(INP_RECVTTL);
1471                                 break;
1472
1473                         case IP_RECVIF:
1474                                 optval = OPTBIT(INP_RECVIF);
1475                                 break;
1476
1477                         case IP_PORTRANGE:
1478                                 if (inp->inp_flags & INP_HIGHPORT)
1479                                         optval = IP_PORTRANGE_HIGH;
1480                                 else if (inp->inp_flags & INP_LOWPORT)
1481                                         optval = IP_PORTRANGE_LOW;
1482                                 else
1483                                         optval = 0;
1484                                 break;
1485
1486                         case IP_ONESBCAST:
1487                                 optval = OPTBIT(INP_ONESBCAST);
1488                                 break;
1489                         case IP_DONTFRAG:
1490                                 optval = OPTBIT(INP_DONTFRAG);
1491                                 break;
1492                         case IP_BINDANY:
1493                                 optval = OPTBIT(INP_BINDANY);
1494                                 break;
1495                         case IP_RECVTOS:
1496                                 optval = OPTBIT(INP_RECVTOS);
1497                                 break;
1498                         case IP_FLOWID:
1499                                 optval = inp->inp_flowid;
1500                                 break;
1501                         case IP_FLOWTYPE:
1502                                 optval = inp->inp_flowtype;
1503                                 break;
1504                         case IP_RECVFLOWID:
1505                                 optval = OPTBIT2(INP_RECVFLOWID);
1506                                 break;
1507 #ifdef  RSS
1508                         case IP_RSSBUCKETID:
1509                                 retval = rss_hash2bucket(inp->inp_flowid,
1510                                     inp->inp_flowtype,
1511                                     &rss_bucket);
1512                                 if (retval == 0)
1513                                         optval = rss_bucket;
1514                                 else
1515                                         error = EINVAL;
1516                                 break;
1517                         case IP_RECVRSSBUCKETID:
1518                                 optval = OPTBIT2(INP_RECVRSSBUCKETID);
1519                                 break;
1520 #endif
1521                         case IP_BINDMULTI:
1522                                 optval = OPTBIT2(INP_BINDMULTI);
1523                                 break;
1524                         case IP_VLAN_PCP:
1525                                 if (OPTBIT2(INP_2PCP_SET)) {
1526                                         optval = (inp->inp_flags2 &
1527                                             INP_2PCP_MASK) >> INP_2PCP_SHIFT;
1528                                 } else {
1529                                         optval = -1;
1530                                 }
1531                                 break;
1532                         }
1533                         error = sooptcopyout(sopt, &optval, sizeof optval);
1534                         break;
1535
1536                 /*
1537                  * Multicast socket options are processed by the in_mcast
1538                  * module.
1539                  */
1540                 case IP_MULTICAST_IF:
1541                 case IP_MULTICAST_VIF:
1542                 case IP_MULTICAST_TTL:
1543                 case IP_MULTICAST_LOOP:
1544                 case IP_MSFILTER:
1545                         error = inp_getmoptions(inp, sopt);
1546                         break;
1547
1548 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1549                 case IP_IPSEC_POLICY:
1550                         if (IPSEC_ENABLED(ipv4)) {
1551                                 error = IPSEC_PCBCTL(ipv4, inp, sopt);
1552                                 break;
1553                         }
1554                         /* FALLTHROUGH */
1555 #endif /* IPSEC */
1556
1557                 default:
1558                         error = ENOPROTOOPT;
1559                         break;
1560                 }
1561                 break;
1562         }
1563         return (error);
1564 }
1565
1566 /*
1567  * Routine called from ip_output() to loop back a copy of an IP multicast
1568  * packet to the input queue of a specified interface.  Note that this
1569  * calls the output routine of the loopback "driver", but with an interface
1570  * pointer that might NOT be a loopback interface -- evil, but easier than
1571  * replicating that code here.
1572  */
1573 static void
1574 ip_mloopback(struct ifnet *ifp, const struct mbuf *m, int hlen)
1575 {
1576         struct ip *ip;
1577         struct mbuf *copym;
1578
1579         /*
1580          * Make a deep copy of the packet because we're going to
1581          * modify the pack in order to generate checksums.
1582          */
1583         copym = m_dup(m, M_NOWAIT);
1584         if (copym != NULL && (!M_WRITABLE(copym) || copym->m_len < hlen))
1585                 copym = m_pullup(copym, hlen);
1586         if (copym != NULL) {
1587                 /* If needed, compute the checksum and mark it as valid. */
1588                 if (copym->m_pkthdr.csum_flags & CSUM_DELAY_DATA) {
1589                         in_delayed_cksum(copym);
1590                         copym->m_pkthdr.csum_flags &= ~CSUM_DELAY_DATA;
1591                         copym->m_pkthdr.csum_flags |=
1592                             CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1593                         copym->m_pkthdr.csum_data = 0xffff;
1594                 }
1595                 /*
1596                  * We don't bother to fragment if the IP length is greater
1597                  * than the interface's MTU.  Can this possibly matter?
1598                  */
1599                 ip = mtod(copym, struct ip *);
1600                 ip->ip_sum = 0;
1601                 ip->ip_sum = in_cksum(copym, hlen);
1602                 if_simloop(ifp, copym, AF_INET, 0);
1603         }
1604 }