sys/netinet/in_pcb.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1991, 1993, 1995
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2007-2009 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Robert N. M. Watson under
  11  * contract to Juniper Networks, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      @(#)in_pcb.c    8.4 (Berkeley) 5/24/95
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_ddb.h"
  44 #include "opt_ipsec.h"
  45 #include "opt_inet.h"
  46 #include "opt_inet6.h"
  47 #include "opt_ratelimit.h"
  48 #include "opt_pcbgroup.h"
  49 #include "opt_rss.h"
  50
  51 #include <sys/param.h>
  52 #include <sys/systm.h>
  53 #include <sys/lock.h>
  54 #include <sys/malloc.h>
  55 #include <sys/mbuf.h>
  56 #include <sys/callout.h>
  57 #include <sys/eventhandler.h>
  58 #include <sys/domain.h>
  59 #include <sys/protosw.h>
  60 #include <sys/rmlock.h>
  61 #include <sys/smp.h>
  62 #include <sys/socket.h>
  63 #include <sys/socketvar.h>
  64 #include <sys/sockio.h>
  65 #include <sys/priv.h>
  66 #include <sys/proc.h>
  67 #include <sys/refcount.h>
  68 #include <sys/jail.h>
  69 #include <sys/kernel.h>
  70 #include <sys/sysctl.h>
  71
  72 #ifdef DDB
  73 #include <ddb/ddb.h>
  74 #endif
  75
  76 #include <vm/uma.h>
  77
  78 #include <net/if.h>
  79 #include <net/if_var.h>
  80 #include <net/if_types.h>
  81 #include <net/if_llatbl.h>
  82 #include <net/route.h>
  83 #include <net/rss_config.h>
  84 #include <net/vnet.h>
  85
  86 #if defined(INET) || defined(INET6)
  87 #include <netinet/in.h>
  88 #include <netinet/in_pcb.h>
  89 #include <netinet/ip_var.h>
  90 #include <netinet/tcp_var.h>
  91 #ifdef TCPHPTS
  92 #include <netinet/tcp_hpts.h>
  93 #endif
  94 #include <netinet/udp.h>
  95 #include <netinet/udp_var.h>
  96 #endif
  97 #ifdef INET
  98 #include <netinet/in_var.h>
  99 #endif
 100 #ifdef INET6
 101 #include <netinet/ip6.h>
 102 #include <netinet6/in6_pcb.h>
 103 #include <netinet6/in6_var.h>
 104 #include <netinet6/ip6_var.h>
 105 #endif /* INET6 */
 106
 107 #include <netipsec/ipsec_support.h>
 108
 109 #include <security/mac/mac_framework.h>
 110
 111 static struct callout   ipport_tick_callout;
 112
 113 /*
 114  * These configure the range of local port addresses assigned to
 115  * "unspecified" outgoing connections/packets/whatever.
 116  */
 117 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;    /* 1023 */
 118 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;    /* 600 */
 119 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;     /* 10000 */
 120 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;       /* 65535 */
 121 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;      /* 49152 */
 122 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;        /* 65535 */
 123
 124 /*
 125  * Reserved ports accessible only to root. There are significant
 126  * security considerations that must be accounted for when changing these,
 127  * but the security benefits can be great. Please be careful.
 128  */
 129 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;    /* 1023 */
 130 VNET_DEFINE(int, ipport_reservedlow);
 131
 132 /* Variables dealing with random ephemeral port allocation. */
 133 VNET_DEFINE(int, ipport_randomized) = 1;        /* user controlled via sysctl */
 134 VNET_DEFINE(int, ipport_randomcps) = 10;        /* user controlled via sysctl */
 135 VNET_DEFINE(int, ipport_randomtime) = 45;       /* user controlled via sysctl */
 136 VNET_DEFINE(int, ipport_stoprandom);            /* toggled by ipport_tick */
 137 VNET_DEFINE(int, ipport_tcpallocs);
 138 static VNET_DEFINE(int, ipport_tcplastcount);
 139
 140 #define V_ipport_tcplastcount           VNET(ipport_tcplastcount)
 141
 142 static void     in_pcbremlists(struct inpcb *inp);
 143 #ifdef INET
 144 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 145                             struct in_addr faddr, u_int fport_arg,
 146                             struct in_addr laddr, u_int lport_arg,
 147                             int lookupflags, struct ifnet *ifp);
 148
 149 #define RANGECHK(var, min, max) \
 150         if ((var) < (min)) { (var) = (min); } \
 151         else if ((var) > (max)) { (var) = (max); }
 152
 153 static int
 154 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 155 {
 156         int error;
 157
 158         error = sysctl_handle_int(oidp, arg1, arg2, req);
 159         if (error == 0) {
 160                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 161                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 162                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 163                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 164                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 165                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 166         }
 167         return (error);
 168 }
 169
 170 #undef RANGECHK
 171
 172 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
 173     "IP Ports");
 174
 175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 176         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 177         &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 179         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 180         &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 182         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 183         &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 185         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 186         &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 188         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 189         &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 191         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 192         &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
 193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 194         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 195         &VNET_NAME(ipport_reservedhigh), 0, "");
 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 197         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 199         CTLFLAG_VNET | CTLFLAG_RW,
 200         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
 202         CTLFLAG_VNET | CTLFLAG_RW,
 203         &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
 204         "allocations before switching to a sequental one");
 205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
 206         CTLFLAG_VNET | CTLFLAG_RW,
 207         &VNET_NAME(ipport_randomtime), 0,
 208         "Minimum time to keep sequental port "
 209         "allocation before switching to a random one");
 210 #endif /* INET */
 211
 212 /*
 213  * in_pcb.c: manage the Protocol Control Blocks.
 214  *
 215  * NOTE: It is assumed that most of these functions will be called with
 216  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
 217  * functions often modify hash chains or addresses in pcbs.
 218  */
 219
 220 /*
 221  * Different protocols initialize their inpcbs differently - giving
 222  * different name to the lock.  But they all are disposed the same.
 223  */
 224 static void
 225 inpcb_fini(void *mem, int size)
 226 {
 227         struct inpcb *inp = mem;
 228
 229         INP_LOCK_DESTROY(inp);
 230 }
 231
 232 /*
 233  * Initialize an inpcbinfo -- we should be able to reduce the number of
 234  * arguments in time.
 235  */
 236 void
 237 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
 238     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
 239     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
 240 {
 241
 242         INP_INFO_LOCK_INIT(pcbinfo, name);
 243         INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");     /* XXXRW: argument? */
 244         INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
 245 #ifdef VIMAGE
 246         pcbinfo->ipi_vnet = curvnet;
 247 #endif
 248         pcbinfo->ipi_listhead = listhead;
 249         LIST_INIT(pcbinfo->ipi_listhead);
 250         pcbinfo->ipi_count = 0;
 251         pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
 252             &pcbinfo->ipi_hashmask);
 253         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 254             &pcbinfo->ipi_porthashmask);
 255 #ifdef PCBGROUP
 256         in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
 257 #endif
 258         pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
 259             NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
 260         uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
 261         uma_zone_set_warning(pcbinfo->ipi_zone,
 262             "kern.ipc.maxsockets limit reached");
 263 }
 264
 265 /*
 266  * Destroy an inpcbinfo.
 267  */
 268 void
 269 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 270 {
 271
 272         KASSERT(pcbinfo->ipi_count == 0,
 273             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 274
 275         hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
 276         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 277             pcbinfo->ipi_porthashmask);
 278 #ifdef PCBGROUP
 279         in_pcbgroup_destroy(pcbinfo);
 280 #endif
 281         uma_zdestroy(pcbinfo->ipi_zone);
 282         INP_LIST_LOCK_DESTROY(pcbinfo);
 283         INP_HASH_LOCK_DESTROY(pcbinfo);
 284         INP_INFO_LOCK_DESTROY(pcbinfo);
 285 }
 286
 287 /*
 288  * Allocate a PCB and associate it with the socket.
 289  * On success return with the PCB locked.
 290  */
 291 int
 292 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 293 {
 294         struct inpcb *inp;
 295         int error;
 296
 297 #ifdef INVARIANTS
 298         if (pcbinfo == &V_tcbinfo) {
 299                 INP_INFO_RLOCK_ASSERT(pcbinfo);
 300         } else {
 301                 INP_INFO_WLOCK_ASSERT(pcbinfo);
 302         }
 303 #endif
 304
 305         error = 0;
 306         inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
 307         if (inp == NULL)
 308                 return (ENOBUFS);
 309         bzero(&inp->inp_start_zero, inp_zero_size);
 310         inp->inp_pcbinfo = pcbinfo;
 311         inp->inp_socket = so;
 312         inp->inp_cred = crhold(so->so_cred);
 313         inp->inp_inc.inc_fibnum = so->so_fibnum;
 314 #ifdef MAC
 315         error = mac_inpcb_init(inp, M_NOWAIT);
 316         if (error != 0)
 317                 goto out;
 318         mac_inpcb_create(so, inp);
 319 #endif
 320 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 321         error = ipsec_init_pcbpolicy(inp);
 322         if (error != 0) {
 323 #ifdef MAC
 324                 mac_inpcb_destroy(inp);
 325 #endif
 326                 goto out;
 327         }
 328 #endif /*IPSEC*/
 329 #ifdef INET6
 330         if (INP_SOCKAF(so) == AF_INET6) {
 331                 inp->inp_vflag |= INP_IPV6PROTO;
 332                 if (V_ip6_v6only)
 333                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 334         }
 335 #endif
 336         INP_WLOCK(inp);
 337         INP_LIST_WLOCK(pcbinfo);
 338         LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
 339         pcbinfo->ipi_count++;
 340         so->so_pcb = (caddr_t)inp;
 341 #ifdef INET6
 342         if (V_ip6_auto_flowlabel)
 343                 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 344 #endif
 345         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 346         refcount_init(&inp->inp_refcount, 1);   /* Reference from inpcbinfo */
 347
 348         /*
 349          * Routes in inpcb's can cache L2 as well; they are guaranteed
 350          * to be cleaned up.
 351          */
 352         inp->inp_route.ro_flags = RT_LLE_CACHE;
 353         INP_LIST_WUNLOCK(pcbinfo);
 354 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 355 out:
 356         if (error != 0) {
 357                 crfree(inp->inp_cred);
 358                 uma_zfree(pcbinfo->ipi_zone, inp);
 359         }
 360 #endif
 361         return (error);
 362 }
 363
 364 #ifdef INET
 365 int
 366 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 367 {
 368         int anonport, error;
 369
 370         INP_WLOCK_ASSERT(inp);
 371         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 372
 373         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 374                 return (EINVAL);
 375         anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
 376         error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
 377             &inp->inp_lport, cred);
 378         if (error)
 379                 return (error);
 380         if (in_pcbinshash(inp) != 0) {
 381                 inp->inp_laddr.s_addr = INADDR_ANY;
 382                 inp->inp_lport = 0;
 383                 return (EAGAIN);
 384         }
 385         if (anonport)
 386                 inp->inp_flags |= INP_ANONPORT;
 387         return (0);
 388 }
 389 #endif
 390
 391 /*
 392  * Select a local port (number) to use.
 393  */
 394 #if defined(INET) || defined(INET6)
 395 int
 396 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 397     struct ucred *cred, int lookupflags)
 398 {
 399         struct inpcbinfo *pcbinfo;
 400         struct inpcb *tmpinp;
 401         unsigned short *lastport;
 402         int count, dorandom, error;
 403         u_short aux, first, last, lport;
 404 #ifdef INET
 405         struct in_addr laddr;
 406 #endif
 407
 408         pcbinfo = inp->inp_pcbinfo;
 409
 410         /*
 411          * Because no actual state changes occur here, a global write lock on
 412          * the pcbinfo isn't required.
 413          */
 414         INP_LOCK_ASSERT(inp);
 415         INP_HASH_LOCK_ASSERT(pcbinfo);
 416
 417         if (inp->inp_flags & INP_HIGHPORT) {
 418                 first = V_ipport_hifirstauto;   /* sysctl */
 419                 last  = V_ipport_hilastauto;
 420                 lastport = &pcbinfo->ipi_lasthi;
 421         } else if (inp->inp_flags & INP_LOWPORT) {
 422                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
 423                 if (error)
 424                         return (error);
 425                 first = V_ipport_lowfirstauto;  /* 1023 */
 426                 last  = V_ipport_lowlastauto;   /* 600 */
 427                 lastport = &pcbinfo->ipi_lastlow;
 428         } else {
 429                 first = V_ipport_firstauto;     /* sysctl */
 430                 last  = V_ipport_lastauto;
 431                 lastport = &pcbinfo->ipi_lastport;
 432         }
 433         /*
 434          * For UDP(-Lite), use random port allocation as long as the user
 435          * allows it.  For TCP (and as of yet unknown) connections,
 436          * use random port allocation only if the user allows it AND
 437          * ipport_tick() allows it.
 438          */
 439         if (V_ipport_randomized &&
 440                 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
 441                 pcbinfo == &V_ulitecbinfo))
 442                 dorandom = 1;
 443         else
 444                 dorandom = 0;
 445         /*
 446          * It makes no sense to do random port allocation if
 447          * we have the only port available.
 448          */
 449         if (first == last)
 450                 dorandom = 0;
 451         /* Make sure to not include UDP(-Lite) packets in the count. */
 452         if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
 453                 V_ipport_tcpallocs++;
 454         /*
 455          * Instead of having two loops further down counting up or down
 456          * make sure that first is always <= last and go with only one
 457          * code path implementing all logic.
 458          */
 459         if (first > last) {
 460                 aux = first;
 461                 first = last;
 462                 last = aux;
 463         }
 464
 465 #ifdef INET
 466         /* Make the compiler happy. */
 467         laddr.s_addr = 0;
 468         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 469                 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
 470                     __func__, inp));
 471                 laddr = *laddrp;
 472         }
 473 #endif
 474         tmpinp = NULL;  /* Make compiler happy. */
 475         lport = *lportp;
 476
 477         if (dorandom)
 478                 *lastport = first + (arc4random() % (last - first));
 479
 480         count = last - first;
 481
 482         do {
 483                 if (count-- < 0)        /* completely used? */
 484                         return (EADDRNOTAVAIL);
 485                 ++*lastport;
 486                 if (*lastport < first || *lastport > last)
 487                         *lastport = first;
 488                 lport = htons(*lastport);
 489
 490 #ifdef INET6
 491                 if ((inp->inp_vflag & INP_IPV6) != 0)
 492                         tmpinp = in6_pcblookup_local(pcbinfo,
 493                             &inp->in6p_laddr, lport, lookupflags, cred);
 494 #endif
 495 #if defined(INET) && defined(INET6)
 496                 else
 497 #endif
 498 #ifdef INET
 499                         tmpinp = in_pcblookup_local(pcbinfo, laddr,
 500                             lport, lookupflags, cred);
 501 #endif
 502         } while (tmpinp != NULL);
 503
 504 #ifdef INET
 505         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
 506                 laddrp->s_addr = laddr.s_addr;
 507 #endif
 508         *lportp = lport;
 509
 510         return (0);
 511 }
 512
 513 /*
 514  * Return cached socket options.
 515  */
 516 short
 517 inp_so_options(const struct inpcb *inp)
 518 {
 519    short so_options;
 520
 521    so_options = 0;
 522
 523    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 524            so_options |= SO_REUSEPORT;
 525    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 526            so_options |= SO_REUSEADDR;
 527    return (so_options);
 528 }
 529 #endif /* INET || INET6 */
 530
 531 /*
 532  * Check if a new BINDMULTI socket is allowed to be created.
 533  *
 534  * ni points to the new inp.
 535  * oi points to the exisitng inp.
 536  *
 537  * This checks whether the existing inp also has BINDMULTI and
 538  * whether the credentials match.
 539  */
 540 int
 541 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
 542 {
 543         /* Check permissions match */
 544         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 545             (ni->inp_cred->cr_uid !=
 546             oi->inp_cred->cr_uid))
 547                 return (0);
 548
 549         /* Check the existing inp has BINDMULTI set */
 550         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 551             ((oi->inp_flags2 & INP_BINDMULTI) == 0))
 552                 return (0);
 553
 554         /*
 555          * We're okay - either INP_BINDMULTI isn't set on ni, or
 556          * it is and it matches the checks.
 557          */
 558         return (1);
 559 }
 560
 561 #ifdef INET
 562 /*
 563  * Set up a bind operation on a PCB, performing port allocation
 564  * as required, but do not actually modify the PCB. Callers can
 565  * either complete the bind by setting inp_laddr/inp_lport and
 566  * calling in_pcbinshash(), or they can just use the resulting
 567  * port and address to authorise the sending of a once-off packet.
 568  *
 569  * On error, the values of *laddrp and *lportp are not changed.
 570  */
 571 int
 572 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
 573     u_short *lportp, struct ucred *cred)
 574 {
 575         struct socket *so = inp->inp_socket;
 576         struct sockaddr_in *sin;
 577         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 578         struct in_addr laddr;
 579         u_short lport = 0;
 580         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 581         int error;
 582
 583         /*
 584          * No state changes, so read locks are sufficient here.
 585          */
 586         INP_LOCK_ASSERT(inp);
 587         INP_HASH_LOCK_ASSERT(pcbinfo);
 588
 589         if (TAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
 590                 return (EADDRNOTAVAIL);
 591         laddr.s_addr = *laddrp;
 592         if (nam != NULL && laddr.s_addr != INADDR_ANY)
 593                 return (EINVAL);
 594         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
 595                 lookupflags = INPLOOKUP_WILDCARD;
 596         if (nam == NULL) {
 597                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 598                         return (error);
 599         } else {
 600                 sin = (struct sockaddr_in *)nam;
 601                 if (nam->sa_len != sizeof (*sin))
 602                         return (EINVAL);
 603 #ifdef notdef
 604                 /*
 605                  * We should check the family, but old programs
 606                  * incorrectly fail to initialize it.
 607                  */
 608                 if (sin->sin_family != AF_INET)
 609                         return (EAFNOSUPPORT);
 610 #endif
 611                 error = prison_local_ip4(cred, &sin->sin_addr);
 612                 if (error)
 613                         return (error);
 614                 if (sin->sin_port != *lportp) {
 615                         /* Don't allow the port to change. */
 616                         if (*lportp != 0)
 617                                 return (EINVAL);
 618                         lport = sin->sin_port;
 619                 }
 620                 /* NB: lport is left as 0 if the port isn't being changed. */
 621                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 622                         /*
 623                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 624                          * allow complete duplication of binding if
 625                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 626                          * and a multicast address is bound on both
 627                          * new and duplicated sockets.
 628                          */
 629                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 630                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 631                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 632                         sin->sin_port = 0;              /* yech... */
 633                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 634                         /*
 635                          * Is the address a local IP address?
 636                          * If INP_BINDANY is set, then the socket may be bound
 637                          * to any endpoint address, local or not.
 638                          */
 639                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 640                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 641                                 return (EADDRNOTAVAIL);
 642                 }
 643                 laddr = sin->sin_addr;
 644                 if (lport) {
 645                         struct inpcb *t;
 646                         struct tcptw *tw;
 647
 648                         /* GROSS */
 649                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 650                             ntohs(lport) >= V_ipport_reservedlow &&
 651                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
 652                             0))
 653                                 return (EACCES);
 654                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 655                             priv_check_cred(inp->inp_cred,
 656                             PRIV_NETINET_REUSEPORT, 0) != 0) {
 657                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 658                                     lport, INPLOOKUP_WILDCARD, cred);
 659         /*
 660          * XXX
 661          * This entire block sorely needs a rewrite.
 662          */
 663                                 if (t &&
 664                                     ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 665                                     ((t->inp_flags & INP_TIMEWAIT) == 0) &&
 666                                     (so->so_type != SOCK_STREAM ||
 667                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
 668                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
 669                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
 670                                      (t->inp_flags2 & INP_REUSEPORT) == 0) &&
 671                                     (inp->inp_cred->cr_uid !=
 672                                      t->inp_cred->cr_uid))
 673                                         return (EADDRINUSE);
 674
 675                                 /*
 676                                  * If the socket is a BINDMULTI socket, then
 677                                  * the credentials need to match and the
 678                                  * original socket also has to have been bound
 679                                  * with BINDMULTI.
 680                                  */
 681                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 682                                         return (EADDRINUSE);
 683                         }
 684                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 685                             lport, lookupflags, cred);
 686                         if (t && (t->inp_flags & INP_TIMEWAIT)) {
 687                                 /*
 688                                  * XXXRW: If an incpb has had its timewait
 689                                  * state recycled, we treat the address as
 690                                  * being in use (for now).  This is better
 691                                  * than a panic, but not desirable.
 692                                  */
 693                                 tw = intotw(t);
 694                                 if (tw == NULL ||
 695                                     (reuseport & tw->tw_so_options) == 0)
 696                                         return (EADDRINUSE);
 697                         } else if (t &&
 698                             ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 699                             (reuseport & inp_so_options(t)) == 0) {
 700 #ifdef INET6
 701                                 if (ntohl(sin->sin_addr.s_addr) !=
 702                                     INADDR_ANY ||
 703                                     ntohl(t->inp_laddr.s_addr) !=
 704                                     INADDR_ANY ||
 705                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
 706                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
 707 #endif
 708                                 return (EADDRINUSE);
 709                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 710                                         return (EADDRINUSE);
 711                         }
 712                 }
 713         }
 714         if (*lportp != 0)
 715                 lport = *lportp;
 716         if (lport == 0) {
 717                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
 718                 if (error != 0)
 719                         return (error);
 720
 721         }
 722         *laddrp = laddr.s_addr;
 723         *lportp = lport;
 724         return (0);
 725 }
 726
 727 /*
 728  * Connect from a socket to a specified address.
 729  * Both address and port must be specified in argument sin.
 730  * If don't have a local address for this socket yet,
 731  * then pick one.
 732  */
 733 int
 734 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
 735     struct ucred *cred, struct mbuf *m)
 736 {
 737         u_short lport, fport;
 738         in_addr_t laddr, faddr;
 739         int anonport, error;
 740
 741         INP_WLOCK_ASSERT(inp);
 742         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 743
 744         lport = inp->inp_lport;
 745         laddr = inp->inp_laddr.s_addr;
 746         anonport = (lport == 0);
 747         error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
 748             NULL, cred);
 749         if (error)
 750                 return (error);
 751
 752         /* Do the initial binding of the local address if required. */
 753         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
 754                 inp->inp_lport = lport;
 755                 inp->inp_laddr.s_addr = laddr;
 756                 if (in_pcbinshash(inp) != 0) {
 757                         inp->inp_laddr.s_addr = INADDR_ANY;
 758                         inp->inp_lport = 0;
 759                         return (EAGAIN);
 760                 }
 761         }
 762
 763         /* Commit the remaining changes. */
 764         inp->inp_lport = lport;
 765         inp->inp_laddr.s_addr = laddr;
 766         inp->inp_faddr.s_addr = faddr;
 767         inp->inp_fport = fport;
 768         in_pcbrehash_mbuf(inp, m);
 769
 770         if (anonport)
 771                 inp->inp_flags |= INP_ANONPORT;
 772         return (0);
 773 }
 774
 775 int
 776 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 777 {
 778
 779         return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
 780 }
 781
 782 /*
 783  * Do proper source address selection on an unbound socket in case
 784  * of connect. Take jails into account as well.
 785  */
 786 int
 787 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
 788     struct ucred *cred)
 789 {
 790         struct ifaddr *ifa;
 791         struct sockaddr *sa;
 792         struct sockaddr_in *sin;
 793         struct route sro;
 794         int error;
 795
 796         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
 797
 798         /*
 799          * Bypass source address selection and use the primary jail IP
 800          * if requested.
 801          */
 802         if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
 803                 return (0);
 804
 805         error = 0;
 806         bzero(&sro, sizeof(sro));
 807
 808         sin = (struct sockaddr_in *)&sro.ro_dst;
 809         sin->sin_family = AF_INET;
 810         sin->sin_len = sizeof(struct sockaddr_in);
 811         sin->sin_addr.s_addr = faddr->s_addr;
 812
 813         /*
 814          * If route is known our src addr is taken from the i/f,
 815          * else punt.
 816          *
 817          * Find out route to destination.
 818          */
 819         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
 820                 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
 821
 822         /*
 823          * If we found a route, use the address corresponding to
 824          * the outgoing interface.
 825          *
 826          * Otherwise assume faddr is reachable on a directly connected
 827          * network and try to find a corresponding interface to take
 828          * the source address from.
 829          */
 830         if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
 831                 struct in_ifaddr *ia;
 832                 struct ifnet *ifp;
 833
 834                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
 835                                         inp->inp_socket->so_fibnum));
 836                 if (ia == NULL)
 837                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
 838                                                 inp->inp_socket->so_fibnum));
 839                 if (ia == NULL) {
 840                         error = ENETUNREACH;
 841                         goto done;
 842                 }
 843
 844                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 845                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 846                         ifa_free(&ia->ia_ifa);
 847                         goto done;
 848                 }
 849
 850                 ifp = ia->ia_ifp;
 851                 ifa_free(&ia->ia_ifa);
 852                 ia = NULL;
 853                 IF_ADDR_RLOCK(ifp);
 854                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 855
 856                         sa = ifa->ifa_addr;
 857                         if (sa->sa_family != AF_INET)
 858                                 continue;
 859                         sin = (struct sockaddr_in *)sa;
 860                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 861                                 ia = (struct in_ifaddr *)ifa;
 862                                 break;
 863                         }
 864                 }
 865                 if (ia != NULL) {
 866                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 867                         IF_ADDR_RUNLOCK(ifp);
 868                         goto done;
 869                 }
 870                 IF_ADDR_RUNLOCK(ifp);
 871
 872                 /* 3. As a last resort return the 'default' jail address. */
 873                 error = prison_get_ip4(cred, laddr);
 874                 goto done;
 875         }
 876
 877         /*
 878          * If the outgoing interface on the route found is not
 879          * a loopback interface, use the address from that interface.
 880          * In case of jails do those three steps:
 881          * 1. check if the interface address belongs to the jail. If so use it.
 882          * 2. check if we have any address on the outgoing interface
 883          *    belonging to this jail. If so use it.
 884          * 3. as a last resort return the 'default' jail address.
 885          */
 886         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
 887                 struct in_ifaddr *ia;
 888                 struct ifnet *ifp;
 889
 890                 /* If not jailed, use the default returned. */
 891                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 892                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
 893                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 894                         goto done;
 895                 }
 896
 897                 /* Jailed. */
 898                 /* 1. Check if the iface address belongs to the jail. */
 899                 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
 900                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 901                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
 902                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 903                         goto done;
 904                 }
 905
 906                 /*
 907                  * 2. Check if we have any address on the outgoing interface
 908                  *    belonging to this jail.
 909                  */
 910                 ia = NULL;
 911                 ifp = sro.ro_rt->rt_ifp;
 912                 IF_ADDR_RLOCK(ifp);
 913                 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 914                         sa = ifa->ifa_addr;
 915                         if (sa->sa_family != AF_INET)
 916                                 continue;
 917                         sin = (struct sockaddr_in *)sa;
 918                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 919                                 ia = (struct in_ifaddr *)ifa;
 920                                 break;
 921                         }
 922                 }
 923                 if (ia != NULL) {
 924                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 925                         IF_ADDR_RUNLOCK(ifp);
 926                         goto done;
 927                 }
 928                 IF_ADDR_RUNLOCK(ifp);
 929
 930                 /* 3. As a last resort return the 'default' jail address. */
 931                 error = prison_get_ip4(cred, laddr);
 932                 goto done;
 933         }
 934
 935         /*
 936          * The outgoing interface is marked with 'loopback net', so a route
 937          * to ourselves is here.
 938          * Try to find the interface of the destination address and then
 939          * take the address from there. That interface is not necessarily
 940          * a loopback interface.
 941          * In case of jails, check that it is an address of the jail
 942          * and if we cannot find, fall back to the 'default' jail address.
 943          */
 944         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
 945                 struct sockaddr_in sain;
 946                 struct in_ifaddr *ia;
 947
 948                 bzero(&sain, sizeof(struct sockaddr_in));
 949                 sain.sin_family = AF_INET;
 950                 sain.sin_len = sizeof(struct sockaddr_in);
 951                 sain.sin_addr.s_addr = faddr->s_addr;
 952
 953                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
 954                                         inp->inp_socket->so_fibnum));
 955                 if (ia == NULL)
 956                         ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
 957                                                 inp->inp_socket->so_fibnum));
 958                 if (ia == NULL)
 959                         ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
 960
 961                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 962                         if (ia == NULL) {
 963                                 error = ENETUNREACH;
 964                                 goto done;
 965                         }
 966                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 967                         ifa_free(&ia->ia_ifa);
 968                         goto done;
 969                 }
 970
 971                 /* Jailed. */
 972                 if (ia != NULL) {
 973                         struct ifnet *ifp;
 974
 975                         ifp = ia->ia_ifp;
 976                         ifa_free(&ia->ia_ifa);
 977                         ia = NULL;
 978                         IF_ADDR_RLOCK(ifp);
 979                         TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 980
 981                                 sa = ifa->ifa_addr;
 982                                 if (sa->sa_family != AF_INET)
 983                                         continue;
 984                                 sin = (struct sockaddr_in *)sa;
 985                                 if (prison_check_ip4(cred,
 986                                     &sin->sin_addr) == 0) {
 987                                         ia = (struct in_ifaddr *)ifa;
 988                                         break;
 989                                 }
 990                         }
 991                         if (ia != NULL) {
 992                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 993                                 IF_ADDR_RUNLOCK(ifp);
 994                                 goto done;
 995                         }
 996                         IF_ADDR_RUNLOCK(ifp);
 997                 }
 998
 999                 /* 3. As a last resort return the 'default' jail address. */
1000                 error = prison_get_ip4(cred, laddr);
1001                 goto done;
1002         }
1003
1004 done:
1005         if (sro.ro_rt != NULL)
1006                 RTFREE(sro.ro_rt);
1007         return (error);
1008 }
1009
1010 /*
1011  * Set up for a connect from a socket to the specified address.
1012  * On entry, *laddrp and *lportp should contain the current local
1013  * address and port for the PCB; these are updated to the values
1014  * that should be placed in inp_laddr and inp_lport to complete
1015  * the connect.
1016  *
1017  * On success, *faddrp and *fportp will be set to the remote address
1018  * and port. These are not updated in the error case.
1019  *
1020  * If the operation fails because the connection already exists,
1021  * *oinpp will be set to the PCB of that connection so that the
1022  * caller can decide to override it. In all other cases, *oinpp
1023  * is set to NULL.
1024  */
1025 int
1026 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1027     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1028     struct inpcb **oinpp, struct ucred *cred)
1029 {
1030         struct rm_priotracker in_ifa_tracker;
1031         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1032         struct in_ifaddr *ia;
1033         struct inpcb *oinp;
1034         struct in_addr laddr, faddr;
1035         u_short lport, fport;
1036         int error;
1037
1038         /*
1039          * Because a global state change doesn't actually occur here, a read
1040          * lock is sufficient.
1041          */
1042         INP_LOCK_ASSERT(inp);
1043         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1044
1045         if (oinpp != NULL)
1046                 *oinpp = NULL;
1047         if (nam->sa_len != sizeof (*sin))
1048                 return (EINVAL);
1049         if (sin->sin_family != AF_INET)
1050                 return (EAFNOSUPPORT);
1051         if (sin->sin_port == 0)
1052                 return (EADDRNOTAVAIL);
1053         laddr.s_addr = *laddrp;
1054         lport = *lportp;
1055         faddr = sin->sin_addr;
1056         fport = sin->sin_port;
1057
1058         if (!TAILQ_EMPTY(&V_in_ifaddrhead)) {
1059                 /*
1060                  * If the destination address is INADDR_ANY,
1061                  * use the primary local address.
1062                  * If the supplied address is INADDR_BROADCAST,
1063                  * and the primary interface supports broadcast,
1064                  * choose the broadcast address for that interface.
1065                  */
1066                 if (faddr.s_addr == INADDR_ANY) {
1067                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1068                         faddr =
1069                             IA_SIN(TAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1070                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1071                         if (cred != NULL &&
1072                             (error = prison_get_ip4(cred, &faddr)) != 0)
1073                                 return (error);
1074                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1075                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1076                         if (TAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1077                             IFF_BROADCAST)
1078                                 faddr = satosin(&TAILQ_FIRST(
1079                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1080                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1081                 }
1082         }
1083         if (laddr.s_addr == INADDR_ANY) {
1084                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1085                 /*
1086                  * If the destination address is multicast and an outgoing
1087                  * interface has been set as a multicast option, prefer the
1088                  * address of that interface as our source address.
1089                  */
1090                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1091                     inp->inp_moptions != NULL) {
1092                         struct ip_moptions *imo;
1093                         struct ifnet *ifp;
1094
1095                         imo = inp->inp_moptions;
1096                         if (imo->imo_multicast_ifp != NULL) {
1097                                 ifp = imo->imo_multicast_ifp;
1098                                 IN_IFADDR_RLOCK(&in_ifa_tracker);
1099                                 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1100                                         if ((ia->ia_ifp == ifp) &&
1101                                             (cred == NULL ||
1102                                             prison_check_ip4(cred,
1103                                             &ia->ia_addr.sin_addr) == 0))
1104                                                 break;
1105                                 }
1106                                 if (ia == NULL)
1107                                         error = EADDRNOTAVAIL;
1108                                 else {
1109                                         laddr = ia->ia_addr.sin_addr;
1110                                         error = 0;
1111                                 }
1112                                 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1113                         }
1114                 }
1115                 if (error)
1116                         return (error);
1117         }
1118         oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1119             laddr, lport, 0, NULL);
1120         if (oinp != NULL) {
1121                 if (oinpp != NULL)
1122                         *oinpp = oinp;
1123                 return (EADDRINUSE);
1124         }
1125         if (lport == 0) {
1126                 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1127                     cred);
1128                 if (error)
1129                         return (error);
1130         }
1131         *laddrp = laddr.s_addr;
1132         *lportp = lport;
1133         *faddrp = faddr.s_addr;
1134         *fportp = fport;
1135         return (0);
1136 }
1137
1138 void
1139 in_pcbdisconnect(struct inpcb *inp)
1140 {
1141
1142         INP_WLOCK_ASSERT(inp);
1143         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1144
1145         inp->inp_faddr.s_addr = INADDR_ANY;
1146         inp->inp_fport = 0;
1147         in_pcbrehash(inp);
1148 }
1149 #endif /* INET */
1150
1151 /*
1152  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1153  * For most protocols, this will be invoked immediately prior to calling
1154  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1155  * socket, in which case in_pcbfree() is deferred.
1156  */
1157 void
1158 in_pcbdetach(struct inpcb *inp)
1159 {
1160
1161         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1162
1163 #ifdef RATELIMIT
1164         if (inp->inp_snd_tag != NULL)
1165                 in_pcbdetach_txrtlmt(inp);
1166 #endif
1167         inp->inp_socket->so_pcb = NULL;
1168         inp->inp_socket = NULL;
1169 }
1170
1171 /*
1172  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1173  * stability of an inpcb pointer despite the inpcb lock being released.  This
1174  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1175  * but where the inpcb lock may already held, or when acquiring a reference
1176  * via a pcbgroup.
1177  *
1178  * in_pcbref() should be used only to provide brief memory stability, and
1179  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1180  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1181  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1182  * lock and rele are the *only* safe operations that may be performed on the
1183  * inpcb.
1184  *
1185  * While the inpcb will not be freed, releasing the inpcb lock means that the
1186  * connection's state may change, so the caller should be careful to
1187  * revalidate any cached state on reacquiring the lock.  Drop the reference
1188  * using in_pcbrele().
1189  */
1190 void
1191 in_pcbref(struct inpcb *inp)
1192 {
1193
1194         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1195
1196         refcount_acquire(&inp->inp_refcount);
1197 }
1198
1199 /*
1200  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1201  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1202  * return a flag indicating whether or not the inpcb remains valid.  If it is
1203  * valid, we return with the inpcb lock held.
1204  *
1205  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1206  * reference on an inpcb.  Historically more work was done here (actually, in
1207  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1208  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1209  * about memory stability (and continued use of the write lock).
1210  */
1211 int
1212 in_pcbrele_rlocked(struct inpcb *inp)
1213 {
1214         struct inpcbinfo *pcbinfo;
1215
1216         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1217
1218         INP_RLOCK_ASSERT(inp);
1219
1220         if (refcount_release(&inp->inp_refcount) == 0) {
1221                 /*
1222                  * If the inpcb has been freed, let the caller know, even if
1223                  * this isn't the last reference.
1224                  */
1225                 if (inp->inp_flags2 & INP_FREED) {
1226                         INP_RUNLOCK(inp);
1227                         return (1);
1228                 }
1229                 return (0);
1230         }
1231
1232         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1233 #ifdef TCPHPTS
1234         if (inp->inp_in_hpts || inp->inp_in_input) {
1235                 struct tcp_hpts_entry *hpts;
1236                 /*
1237                  * We should not be on the hpts at
1238                  * this point in any form. we must
1239                  * get the lock to be sure.
1240                  */
1241                 hpts = tcp_hpts_lock(inp);
1242                 if (inp->inp_in_hpts)
1243                         panic("Hpts:%p inp:%p at free still on hpts",
1244                               hpts, inp);
1245                 mtx_unlock(&hpts->p_mtx);
1246                 hpts = tcp_input_lock(inp);
1247                 if (inp->inp_in_input)
1248                         panic("Hpts:%p inp:%p at free still on input hpts",
1249                               hpts, inp);
1250                 mtx_unlock(&hpts->p_mtx);
1251         }
1252 #endif
1253         INP_RUNLOCK(inp);
1254         pcbinfo = inp->inp_pcbinfo;
1255         uma_zfree(pcbinfo->ipi_zone, inp);
1256         return (1);
1257 }
1258
1259 int
1260 in_pcbrele_wlocked(struct inpcb *inp)
1261 {
1262         struct inpcbinfo *pcbinfo;
1263
1264         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1265
1266         INP_WLOCK_ASSERT(inp);
1267
1268         if (refcount_release(&inp->inp_refcount) == 0) {
1269                 /*
1270                  * If the inpcb has been freed, let the caller know, even if
1271                  * this isn't the last reference.
1272                  */
1273                 if (inp->inp_flags2 & INP_FREED) {
1274                         INP_WUNLOCK(inp);
1275                         return (1);
1276                 }
1277                 return (0);
1278         }
1279
1280         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1281 #ifdef TCPHPTS
1282         if (inp->inp_in_hpts || inp->inp_in_input) {
1283                 struct tcp_hpts_entry *hpts;
1284                 /*
1285                  * We should not be on the hpts at
1286                  * this point in any form. we must
1287                  * get the lock to be sure.
1288                  */
1289                 hpts = tcp_hpts_lock(inp);
1290                 if (inp->inp_in_hpts)
1291                         panic("Hpts:%p inp:%p at free still on hpts",
1292                               hpts, inp);
1293                 mtx_unlock(&hpts->p_mtx);
1294                 hpts = tcp_input_lock(inp);
1295                 if (inp->inp_in_input)
1296                         panic("Hpts:%p inp:%p at free still on input hpts",
1297                               hpts, inp);
1298                 mtx_unlock(&hpts->p_mtx);
1299         }
1300 #endif
1301         INP_WUNLOCK(inp);
1302         pcbinfo = inp->inp_pcbinfo;
1303         uma_zfree(pcbinfo->ipi_zone, inp);
1304         return (1);
1305 }
1306
1307 /*
1308  * Temporary wrapper.
1309  */
1310 int
1311 in_pcbrele(struct inpcb *inp)
1312 {
1313
1314         return (in_pcbrele_wlocked(inp));
1315 }
1316
1317 /*
1318  * Unconditionally schedule an inpcb to be freed by decrementing its
1319  * reference count, which should occur only after the inpcb has been detached
1320  * from its socket.  If another thread holds a temporary reference (acquired
1321  * using in_pcbref()) then the free is deferred until that reference is
1322  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1323  * work, including removal from global lists, is done in this context, where
1324  * the pcbinfo lock is held.
1325  */
1326 void
1327 in_pcbfree(struct inpcb *inp)
1328 {
1329         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1330
1331 #ifdef INET6
1332         struct ip6_moptions *im6o = NULL;
1333 #endif
1334 #ifdef INET
1335         struct ip_moptions *imo = NULL;
1336 #endif
1337         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1338
1339         KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1340             ("%s: called twice for pcb %p", __func__, inp));
1341         if (inp->inp_flags2 & INP_FREED) {
1342                 INP_WUNLOCK(inp);
1343                 return;
1344         }
1345
1346 #ifdef INVARIANTS
1347         if (pcbinfo == &V_tcbinfo) {
1348                 INP_INFO_LOCK_ASSERT(pcbinfo);
1349         } else {
1350                 INP_INFO_WLOCK_ASSERT(pcbinfo);
1351         }
1352 #endif
1353         INP_WLOCK_ASSERT(inp);
1354
1355 #ifdef INET
1356         imo = inp->inp_moptions;
1357         inp->inp_moptions = NULL;
1358 #endif
1359         /* XXXRW: Do as much as possible here. */
1360 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1361         if (inp->inp_sp != NULL)
1362                 ipsec_delete_pcbpolicy(inp);
1363 #endif
1364         INP_LIST_WLOCK(pcbinfo);
1365         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1366         in_pcbremlists(inp);
1367         INP_LIST_WUNLOCK(pcbinfo);
1368 #ifdef INET6
1369         if (inp->inp_vflag & INP_IPV6PROTO) {
1370                 ip6_freepcbopts(inp->in6p_outputopts);
1371                 im6o = inp->in6p_moptions;
1372                 inp->in6p_moptions = NULL;
1373         }
1374 #endif
1375         if (inp->inp_options)
1376                 (void)m_free(inp->inp_options);
1377         RO_INVALIDATE_CACHE(&inp->inp_route);
1378
1379         inp->inp_vflag = 0;
1380         inp->inp_flags2 |= INP_FREED;
1381         crfree(inp->inp_cred);
1382 #ifdef MAC
1383         mac_inpcb_destroy(inp);
1384 #endif
1385         if (!in_pcbrele_wlocked(inp))
1386                 INP_WUNLOCK(inp);
1387 #if defined(INET) && defined(INET6)
1388         if (imo == NULL && im6o == NULL)
1389                 return;
1390 #endif
1391         INP_INFO_WUNLOCK(pcbinfo);
1392 #ifdef INET6
1393         ip6_freemoptions(im6o);
1394 #endif
1395 #ifdef INET
1396         inp_freemoptions(imo);
1397 #endif
1398         INP_INFO_WLOCK(pcbinfo);
1399 }
1400
1401 /*
1402  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1403  * port reservation, and preventing it from being returned by inpcb lookups.
1404  *
1405  * It is used by TCP to mark an inpcb as unused and avoid future packet
1406  * delivery or event notification when a socket remains open but TCP has
1407  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1408  * or a RST on the wire, and allows the port binding to be reused while still
1409  * maintaining the invariant that so_pcb always points to a valid inpcb until
1410  * in_pcbdetach().
1411  *
1412  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1413  * in_pcbnotifyall() and in_pcbpurgeif0()?
1414  */
1415 void
1416 in_pcbdrop(struct inpcb *inp)
1417 {
1418
1419         INP_WLOCK_ASSERT(inp);
1420
1421         /*
1422          * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1423          * the hash lock...?
1424          */
1425         inp->inp_flags |= INP_DROPPED;
1426         if (inp->inp_flags & INP_INHASHLIST) {
1427                 struct inpcbport *phd = inp->inp_phd;
1428
1429                 INP_HASH_WLOCK(inp->inp_pcbinfo);
1430                 LIST_REMOVE(inp, inp_hash);
1431                 LIST_REMOVE(inp, inp_portlist);
1432                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1433                         LIST_REMOVE(phd, phd_hash);
1434                         free(phd, M_PCB);
1435                 }
1436                 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1437                 inp->inp_flags &= ~INP_INHASHLIST;
1438 #ifdef PCBGROUP
1439                 in_pcbgroup_remove(inp);
1440 #endif
1441         }
1442 }
1443
1444 #ifdef INET
1445 /*
1446  * Common routines to return the socket addresses associated with inpcbs.
1447  */
1448 struct sockaddr *
1449 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1450 {
1451         struct sockaddr_in *sin;
1452
1453         sin = malloc(sizeof *sin, M_SONAME,
1454                 M_WAITOK | M_ZERO);
1455         sin->sin_family = AF_INET;
1456         sin->sin_len = sizeof(*sin);
1457         sin->sin_addr = *addr_p;
1458         sin->sin_port = port;
1459
1460         return (struct sockaddr *)sin;
1461 }
1462
1463 int
1464 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1465 {
1466         struct inpcb *inp;
1467         struct in_addr addr;
1468         in_port_t port;
1469
1470         inp = sotoinpcb(so);
1471         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1472
1473         INP_RLOCK(inp);
1474         port = inp->inp_lport;
1475         addr = inp->inp_laddr;
1476         INP_RUNLOCK(inp);
1477
1478         *nam = in_sockaddr(port, &addr);
1479         return 0;
1480 }
1481
1482 int
1483 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1484 {
1485         struct inpcb *inp;
1486         struct in_addr addr;
1487         in_port_t port;
1488
1489         inp = sotoinpcb(so);
1490         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1491
1492         INP_RLOCK(inp);
1493         port = inp->inp_fport;
1494         addr = inp->inp_faddr;
1495         INP_RUNLOCK(inp);
1496
1497         *nam = in_sockaddr(port, &addr);
1498         return 0;
1499 }
1500
1501 void
1502 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1503     struct inpcb *(*notify)(struct inpcb *, int))
1504 {
1505         struct inpcb *inp, *inp_temp;
1506
1507         INP_INFO_WLOCK(pcbinfo);
1508         LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1509                 INP_WLOCK(inp);
1510 #ifdef INET6
1511                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1512                         INP_WUNLOCK(inp);
1513                         continue;
1514                 }
1515 #endif
1516                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1517                     inp->inp_socket == NULL) {
1518                         INP_WUNLOCK(inp);
1519                         continue;
1520                 }
1521                 if ((*notify)(inp, errno))
1522                         INP_WUNLOCK(inp);
1523         }
1524         INP_INFO_WUNLOCK(pcbinfo);
1525 }
1526
1527 void
1528 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1529 {
1530         struct inpcb *inp;
1531         struct ip_moptions *imo;
1532         int i, gap;
1533
1534         INP_INFO_WLOCK(pcbinfo);
1535         LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1536                 INP_WLOCK(inp);
1537                 imo = inp->inp_moptions;
1538                 if ((inp->inp_vflag & INP_IPV4) &&
1539                     imo != NULL) {
1540                         /*
1541                          * Unselect the outgoing interface if it is being
1542                          * detached.
1543                          */
1544                         if (imo->imo_multicast_ifp == ifp)
1545                                 imo->imo_multicast_ifp = NULL;
1546
1547                         /*
1548                          * Drop multicast group membership if we joined
1549                          * through the interface being detached.
1550                          */
1551                         for (i = 0, gap = 0; i < imo->imo_num_memberships;
1552                             i++) {
1553                                 if (imo->imo_membership[i]->inm_ifp == ifp) {
1554                                         IN_MULTI_LOCK_ASSERT();
1555                                         in_leavegroup_locked(imo->imo_membership[i], NULL);
1556                                         gap++;
1557                                 } else if (gap != 0)
1558                                         imo->imo_membership[i - gap] =
1559                                             imo->imo_membership[i];
1560                         }
1561                         imo->imo_num_memberships -= gap;
1562                 }
1563                 INP_WUNLOCK(inp);
1564         }
1565         INP_INFO_WUNLOCK(pcbinfo);
1566 }
1567
1568 /*
1569  * Lookup a PCB based on the local address and port.  Caller must hold the
1570  * hash lock.  No inpcb locks or references are acquired.
1571  */
1572 #define INP_LOOKUP_MAPPED_PCB_COST      3
1573 struct inpcb *
1574 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1575     u_short lport, int lookupflags, struct ucred *cred)
1576 {
1577         struct inpcb *inp;
1578 #ifdef INET6
1579         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1580 #else
1581         int matchwild = 3;
1582 #endif
1583         int wildcard;
1584
1585         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1586             ("%s: invalid lookup flags %d", __func__, lookupflags));
1587
1588         INP_HASH_LOCK_ASSERT(pcbinfo);
1589
1590         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1591                 struct inpcbhead *head;
1592                 /*
1593                  * Look for an unconnected (wildcard foreign addr) PCB that
1594                  * matches the local address and port we're looking for.
1595                  */
1596                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1597                     0, pcbinfo->ipi_hashmask)];
1598                 LIST_FOREACH(inp, head, inp_hash) {
1599 #ifdef INET6
1600                         /* XXX inp locking */
1601                         if ((inp->inp_vflag & INP_IPV4) == 0)
1602                                 continue;
1603 #endif
1604                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
1605                             inp->inp_laddr.s_addr == laddr.s_addr &&
1606                             inp->inp_lport == lport) {
1607                                 /*
1608                                  * Found?
1609                                  */
1610                                 if (cred == NULL ||
1611                                     prison_equal_ip4(cred->cr_prison,
1612                                         inp->inp_cred->cr_prison))
1613                                         return (inp);
1614                         }
1615                 }
1616                 /*
1617                  * Not found.
1618                  */
1619                 return (NULL);
1620         } else {
1621                 struct inpcbporthead *porthash;
1622                 struct inpcbport *phd;
1623                 struct inpcb *match = NULL;
1624                 /*
1625                  * Best fit PCB lookup.
1626                  *
1627                  * First see if this local port is in use by looking on the
1628                  * port hash list.
1629                  */
1630                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1631                     pcbinfo->ipi_porthashmask)];
1632                 LIST_FOREACH(phd, porthash, phd_hash) {
1633                         if (phd->phd_port == lport)
1634                                 break;
1635                 }
1636                 if (phd != NULL) {
1637                         /*
1638                          * Port is in use by one or more PCBs. Look for best
1639                          * fit.
1640                          */
1641                         LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1642                                 wildcard = 0;
1643                                 if (cred != NULL &&
1644                                     !prison_equal_ip4(inp->inp_cred->cr_prison,
1645                                         cred->cr_prison))
1646                                         continue;
1647 #ifdef INET6
1648                                 /* XXX inp locking */
1649                                 if ((inp->inp_vflag & INP_IPV4) == 0)
1650                                         continue;
1651                                 /*
1652                                  * We never select the PCB that has
1653                                  * INP_IPV6 flag and is bound to :: if
1654                                  * we have another PCB which is bound
1655                                  * to 0.0.0.0.  If a PCB has the
1656                                  * INP_IPV6 flag, then we set its cost
1657                                  * higher than IPv4 only PCBs.
1658                                  *
1659                                  * Note that the case only happens
1660                                  * when a socket is bound to ::, under
1661                                  * the condition that the use of the
1662                                  * mapped address is allowed.
1663                                  */
1664                                 if ((inp->inp_vflag & INP_IPV6) != 0)
1665                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1666 #endif
1667                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
1668                                         wildcard++;
1669                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1670                                         if (laddr.s_addr == INADDR_ANY)
1671                                                 wildcard++;
1672                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
1673                                                 continue;
1674                                 } else {
1675                                         if (laddr.s_addr != INADDR_ANY)
1676                                                 wildcard++;
1677                                 }
1678                                 if (wildcard < matchwild) {
1679                                         match = inp;
1680                                         matchwild = wildcard;
1681                                         if (matchwild == 0)
1682                                                 break;
1683                                 }
1684                         }
1685                 }
1686                 return (match);
1687         }
1688 }
1689 #undef INP_LOOKUP_MAPPED_PCB_COST
1690
1691 #ifdef PCBGROUP
1692 /*
1693  * Lookup PCB in hash list, using pcbgroup tables.
1694  */
1695 static struct inpcb *
1696 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1697     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1698     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1699 {
1700         struct inpcbhead *head;
1701         struct inpcb *inp, *tmpinp;
1702         u_short fport = fport_arg, lport = lport_arg;
1703         bool locked;
1704
1705         /*
1706          * First look for an exact match.
1707          */
1708         tmpinp = NULL;
1709         INP_GROUP_LOCK(pcbgroup);
1710         head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1711             pcbgroup->ipg_hashmask)];
1712         LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1713 #ifdef INET6
1714                 /* XXX inp locking */
1715                 if ((inp->inp_vflag & INP_IPV4) == 0)
1716                         continue;
1717 #endif
1718                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1719                     inp->inp_laddr.s_addr == laddr.s_addr &&
1720                     inp->inp_fport == fport &&
1721                     inp->inp_lport == lport) {
1722                         /*
1723                          * XXX We should be able to directly return
1724                          * the inp here, without any checks.
1725                          * Well unless both bound with SO_REUSEPORT?
1726                          */
1727                         if (prison_flag(inp->inp_cred, PR_IP4))
1728                                 goto found;
1729                         if (tmpinp == NULL)
1730                                 tmpinp = inp;
1731                 }
1732         }
1733         if (tmpinp != NULL) {
1734                 inp = tmpinp;
1735                 goto found;
1736         }
1737
1738 #ifdef  RSS
1739         /*
1740          * For incoming connections, we may wish to do a wildcard
1741          * match for an RSS-local socket.
1742          */
1743         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1744                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1745 #ifdef INET6
1746                 struct inpcb *local_wild_mapped = NULL;
1747 #endif
1748                 struct inpcb *jail_wild = NULL;
1749                 struct inpcbhead *head;
1750                 int injail;
1751
1752                 /*
1753                  * Order of socket selection - we always prefer jails.
1754                  *      1. jailed, non-wild.
1755                  *      2. jailed, wild.
1756                  *      3. non-jailed, non-wild.
1757                  *      4. non-jailed, wild.
1758                  */
1759
1760                 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1761                     lport, 0, pcbgroup->ipg_hashmask)];
1762                 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1763 #ifdef INET6
1764                         /* XXX inp locking */
1765                         if ((inp->inp_vflag & INP_IPV4) == 0)
1766                                 continue;
1767 #endif
1768                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
1769                             inp->inp_lport != lport)
1770                                 continue;
1771
1772                         injail = prison_flag(inp->inp_cred, PR_IP4);
1773                         if (injail) {
1774                                 if (prison_check_ip4(inp->inp_cred,
1775                                     &laddr) != 0)
1776                                         continue;
1777                         } else {
1778                                 if (local_exact != NULL)
1779                                         continue;
1780                         }
1781
1782                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
1783                                 if (injail)
1784                                         goto found;
1785                                 else
1786                                         local_exact = inp;
1787                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1788 #ifdef INET6
1789                                 /* XXX inp locking, NULL check */
1790                                 if (inp->inp_vflag & INP_IPV6PROTO)
1791                                         local_wild_mapped = inp;
1792                                 else
1793 #endif
1794                                         if (injail)
1795                                                 jail_wild = inp;
1796                                         else
1797                                                 local_wild = inp;
1798                         }
1799                 } /* LIST_FOREACH */
1800
1801                 inp = jail_wild;
1802                 if (inp == NULL)
1803                         inp = local_exact;
1804                 if (inp == NULL)
1805                         inp = local_wild;
1806 #ifdef INET6
1807                 if (inp == NULL)
1808                         inp = local_wild_mapped;
1809 #endif
1810                 if (inp != NULL)
1811                         goto found;
1812         }
1813 #endif
1814
1815         /*
1816          * Then look for a wildcard match, if requested.
1817          */
1818         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1819                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1820 #ifdef INET6
1821                 struct inpcb *local_wild_mapped = NULL;
1822 #endif
1823                 struct inpcb *jail_wild = NULL;
1824                 struct inpcbhead *head;
1825                 int injail;
1826
1827                 /*
1828                  * Order of socket selection - we always prefer jails.
1829                  *      1. jailed, non-wild.
1830                  *      2. jailed, wild.
1831                  *      3. non-jailed, non-wild.
1832                  *      4. non-jailed, wild.
1833                  */
1834                 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1835                     0, pcbinfo->ipi_wildmask)];
1836                 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1837 #ifdef INET6
1838                         /* XXX inp locking */
1839                         if ((inp->inp_vflag & INP_IPV4) == 0)
1840                                 continue;
1841 #endif
1842                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
1843                             inp->inp_lport != lport)
1844                                 continue;
1845
1846                         injail = prison_flag(inp->inp_cred, PR_IP4);
1847                         if (injail) {
1848                                 if (prison_check_ip4(inp->inp_cred,
1849                                     &laddr) != 0)
1850                                         continue;
1851                         } else {
1852                                 if (local_exact != NULL)
1853                                         continue;
1854                         }
1855
1856                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
1857                                 if (injail)
1858                                         goto found;
1859                                 else
1860                                         local_exact = inp;
1861                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1862 #ifdef INET6
1863                                 /* XXX inp locking, NULL check */
1864                                 if (inp->inp_vflag & INP_IPV6PROTO)
1865                                         local_wild_mapped = inp;
1866                                 else
1867 #endif
1868                                         if (injail)
1869                                                 jail_wild = inp;
1870                                         else
1871                                                 local_wild = inp;
1872                         }
1873                 } /* LIST_FOREACH */
1874                 inp = jail_wild;
1875                 if (inp == NULL)
1876                         inp = local_exact;
1877                 if (inp == NULL)
1878                         inp = local_wild;
1879 #ifdef INET6
1880                 if (inp == NULL)
1881                         inp = local_wild_mapped;
1882 #endif
1883                 if (inp != NULL)
1884                         goto found;
1885         } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1886         INP_GROUP_UNLOCK(pcbgroup);
1887         return (NULL);
1888
1889 found:
1890         if (lookupflags & INPLOOKUP_WLOCKPCB)
1891                 locked = INP_TRY_WLOCK(inp);
1892         else if (lookupflags & INPLOOKUP_RLOCKPCB)
1893                 locked = INP_TRY_RLOCK(inp);
1894         else
1895                 panic("%s: locking bug", __func__);
1896         if (!locked)
1897                 in_pcbref(inp);
1898         INP_GROUP_UNLOCK(pcbgroup);
1899         if (!locked) {
1900                 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1901                         INP_WLOCK(inp);
1902                         if (in_pcbrele_wlocked(inp))
1903                                 return (NULL);
1904                 } else {
1905                         INP_RLOCK(inp);
1906                         if (in_pcbrele_rlocked(inp))
1907                                 return (NULL);
1908                 }
1909         }
1910 #ifdef INVARIANTS
1911         if (lookupflags & INPLOOKUP_WLOCKPCB)
1912                 INP_WLOCK_ASSERT(inp);
1913         else
1914                 INP_RLOCK_ASSERT(inp);
1915 #endif
1916         return (inp);
1917 }
1918 #endif /* PCBGROUP */
1919
1920 /*
1921  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1922  * that the caller has locked the hash list, and will not perform any further
1923  * locking or reference operations on either the hash list or the connection.
1924  */
1925 static struct inpcb *
1926 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1927     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1928     struct ifnet *ifp)
1929 {
1930         struct inpcbhead *head;
1931         struct inpcb *inp, *tmpinp;
1932         u_short fport = fport_arg, lport = lport_arg;
1933
1934         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1935             ("%s: invalid lookup flags %d", __func__, lookupflags));
1936
1937         INP_HASH_LOCK_ASSERT(pcbinfo);
1938
1939         /*
1940          * First look for an exact match.
1941          */
1942         tmpinp = NULL;
1943         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1944             pcbinfo->ipi_hashmask)];
1945         LIST_FOREACH(inp, head, inp_hash) {
1946 #ifdef INET6
1947                 /* XXX inp locking */
1948                 if ((inp->inp_vflag & INP_IPV4) == 0)
1949                         continue;
1950 #endif
1951                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1952                     inp->inp_laddr.s_addr == laddr.s_addr &&
1953                     inp->inp_fport == fport &&
1954                     inp->inp_lport == lport) {
1955                         /*
1956                          * XXX We should be able to directly return
1957                          * the inp here, without any checks.
1958                          * Well unless both bound with SO_REUSEPORT?
1959                          */
1960                         if (prison_flag(inp->inp_cred, PR_IP4))
1961                                 return (inp);
1962                         if (tmpinp == NULL)
1963                                 tmpinp = inp;
1964                 }
1965         }
1966         if (tmpinp != NULL)
1967                 return (tmpinp);
1968
1969         /*
1970          * Then look for a wildcard match, if requested.
1971          */
1972         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1973                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1974 #ifdef INET6
1975                 struct inpcb *local_wild_mapped = NULL;
1976 #endif
1977                 struct inpcb *jail_wild = NULL;
1978                 int injail;
1979
1980                 /*
1981                  * Order of socket selection - we always prefer jails.
1982                  *      1. jailed, non-wild.
1983                  *      2. jailed, wild.
1984                  *      3. non-jailed, non-wild.
1985                  *      4. non-jailed, wild.
1986                  */
1987
1988                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1989                     0, pcbinfo->ipi_hashmask)];
1990                 LIST_FOREACH(inp, head, inp_hash) {
1991 #ifdef INET6
1992                         /* XXX inp locking */
1993                         if ((inp->inp_vflag & INP_IPV4) == 0)
1994                                 continue;
1995 #endif
1996                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
1997                             inp->inp_lport != lport)
1998                                 continue;
1999
2000                         injail = prison_flag(inp->inp_cred, PR_IP4);
2001                         if (injail) {
2002                                 if (prison_check_ip4(inp->inp_cred,
2003                                     &laddr) != 0)
2004                                         continue;
2005                         } else {
2006                                 if (local_exact != NULL)
2007                                         continue;
2008                         }
2009
2010                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2011                                 if (injail)
2012                                         return (inp);
2013                                 else
2014                                         local_exact = inp;
2015                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2016 #ifdef INET6
2017                                 /* XXX inp locking, NULL check */
2018                                 if (inp->inp_vflag & INP_IPV6PROTO)
2019                                         local_wild_mapped = inp;
2020                                 else
2021 #endif
2022                                         if (injail)
2023                                                 jail_wild = inp;
2024                                         else
2025                                                 local_wild = inp;
2026                         }
2027                 } /* LIST_FOREACH */
2028                 if (jail_wild != NULL)
2029                         return (jail_wild);
2030                 if (local_exact != NULL)
2031                         return (local_exact);
2032                 if (local_wild != NULL)
2033                         return (local_wild);
2034 #ifdef INET6
2035                 if (local_wild_mapped != NULL)
2036                         return (local_wild_mapped);
2037 #endif
2038         } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2039
2040         return (NULL);
2041 }
2042
2043 /*
2044  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2045  * hash list lock, and will return the inpcb locked (i.e., requires
2046  * INPLOOKUP_LOCKPCB).
2047  */
2048 static struct inpcb *
2049 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2050     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2051     struct ifnet *ifp)
2052 {
2053         struct inpcb *inp;
2054         bool locked;
2055
2056         INP_HASH_RLOCK(pcbinfo);
2057         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2058             (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2059         if (inp != NULL) {
2060                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2061                         locked = INP_TRY_WLOCK(inp);
2062                 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2063                         locked = INP_TRY_RLOCK(inp);
2064                 else
2065                         panic("%s: locking bug", __func__);
2066                 if (!locked)
2067                         in_pcbref(inp);
2068                 INP_HASH_RUNLOCK(pcbinfo);
2069                 if (!locked) {
2070                         if (lookupflags & INPLOOKUP_WLOCKPCB) {
2071                                 INP_WLOCK(inp);
2072                                 if (in_pcbrele_wlocked(inp))
2073                                         return (NULL);
2074                         } else {
2075                                 INP_RLOCK(inp);
2076                                 if (in_pcbrele_rlocked(inp))
2077                                         return (NULL);
2078                         }
2079                 }
2080 #ifdef INVARIANTS
2081                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2082                         INP_WLOCK_ASSERT(inp);
2083                 else
2084                         INP_RLOCK_ASSERT(inp);
2085 #endif
2086         } else
2087                 INP_HASH_RUNLOCK(pcbinfo);
2088         return (inp);
2089 }
2090
2091 /*
2092  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2093  * from which a pre-calculated hash value may be extracted.
2094  *
2095  * Possibly more of this logic should be in in_pcbgroup.c.
2096  */
2097 struct inpcb *
2098 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2099     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2100 {
2101 #if defined(PCBGROUP) && !defined(RSS)
2102         struct inpcbgroup *pcbgroup;
2103 #endif
2104
2105         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2106             ("%s: invalid lookup flags %d", __func__, lookupflags));
2107         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2108             ("%s: LOCKPCB not set", __func__));
2109
2110         /*
2111          * When not using RSS, use connection groups in preference to the
2112          * reservation table when looking up 4-tuples.  When using RSS, just
2113          * use the reservation table, due to the cost of the Toeplitz hash
2114          * in software.
2115          *
2116          * XXXRW: This policy belongs in the pcbgroup code, as in principle
2117          * we could be doing RSS with a non-Toeplitz hash that is affordable
2118          * in software.
2119          */
2120 #if defined(PCBGROUP) && !defined(RSS)
2121         if (in_pcbgroup_enabled(pcbinfo)) {
2122                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2123                     fport);
2124                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2125                     laddr, lport, lookupflags, ifp));
2126         }
2127 #endif
2128         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2129             lookupflags, ifp));
2130 }
2131
2132 struct inpcb *
2133 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2134     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2135     struct ifnet *ifp, struct mbuf *m)
2136 {
2137 #ifdef PCBGROUP
2138         struct inpcbgroup *pcbgroup;
2139 #endif
2140
2141         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2142             ("%s: invalid lookup flags %d", __func__, lookupflags));
2143         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2144             ("%s: LOCKPCB not set", __func__));
2145
2146 #ifdef PCBGROUP
2147         /*
2148          * If we can use a hardware-generated hash to look up the connection
2149          * group, use that connection group to find the inpcb.  Otherwise
2150          * fall back on a software hash -- or the reservation table if we're
2151          * using RSS.
2152          *
2153          * XXXRW: As above, that policy belongs in the pcbgroup code.
2154          */
2155         if (in_pcbgroup_enabled(pcbinfo) &&
2156             !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2157                 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2158                     m->m_pkthdr.flowid);
2159                 if (pcbgroup != NULL)
2160                         return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2161                             fport, laddr, lport, lookupflags, ifp));
2162 #ifndef RSS
2163                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2164                     fport);
2165                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2166                     laddr, lport, lookupflags, ifp));
2167 #endif
2168         }
2169 #endif
2170         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2171             lookupflags, ifp));
2172 }
2173 #endif /* INET */
2174
2175 /*
2176  * Insert PCB onto various hash lists.
2177  */
2178 static int
2179 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2180 {
2181         struct inpcbhead *pcbhash;
2182         struct inpcbporthead *pcbporthash;
2183         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2184         struct inpcbport *phd;
2185         u_int32_t hashkey_faddr;
2186
2187         INP_WLOCK_ASSERT(inp);
2188         INP_HASH_WLOCK_ASSERT(pcbinfo);
2189
2190         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2191             ("in_pcbinshash: INP_INHASHLIST"));
2192
2193 #ifdef INET6
2194         if (inp->inp_vflag & INP_IPV6)
2195                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2196         else
2197 #endif
2198         hashkey_faddr = inp->inp_faddr.s_addr;
2199
2200         pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2201                  inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2202
2203         pcbporthash = &pcbinfo->ipi_porthashbase[
2204             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2205
2206         /*
2207          * Go through port list and look for a head for this lport.
2208          */
2209         LIST_FOREACH(phd, pcbporthash, phd_hash) {
2210                 if (phd->phd_port == inp->inp_lport)
2211                         break;
2212         }
2213         /*
2214          * If none exists, malloc one and tack it on.
2215          */
2216         if (phd == NULL) {
2217                 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2218                 if (phd == NULL) {
2219                         return (ENOBUFS); /* XXX */
2220                 }
2221                 phd->phd_port = inp->inp_lport;
2222                 LIST_INIT(&phd->phd_pcblist);
2223                 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2224         }
2225         inp->inp_phd = phd;
2226         LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2227         LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2228         inp->inp_flags |= INP_INHASHLIST;
2229 #ifdef PCBGROUP
2230         if (do_pcbgroup_update)
2231                 in_pcbgroup_update(inp);
2232 #endif
2233         return (0);
2234 }
2235
2236 /*
2237  * For now, there are two public interfaces to insert an inpcb into the hash
2238  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2239  * is used only in the TCP syncache, where in_pcbinshash is called before the
2240  * full 4-tuple is set for the inpcb, and we don't want to install in the
2241  * pcbgroup until later.
2242  *
2243  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2244  * connection groups, and partially initialised inpcbs should not be exposed
2245  * to either reservation hash tables or pcbgroups.
2246  */
2247 int
2248 in_pcbinshash(struct inpcb *inp)
2249 {
2250
2251         return (in_pcbinshash_internal(inp, 1));
2252 }
2253
2254 int
2255 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2256 {
2257
2258         return (in_pcbinshash_internal(inp, 0));
2259 }
2260
2261 /*
2262  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2263  * changed. NOTE: This does not handle the case of the lport changing (the
2264  * hashed port list would have to be updated as well), so the lport must
2265  * not change after in_pcbinshash() has been called.
2266  */
2267 void
2268 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2269 {
2270         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2271         struct inpcbhead *head;
2272         u_int32_t hashkey_faddr;
2273
2274         INP_WLOCK_ASSERT(inp);
2275         INP_HASH_WLOCK_ASSERT(pcbinfo);
2276
2277         KASSERT(inp->inp_flags & INP_INHASHLIST,
2278             ("in_pcbrehash: !INP_INHASHLIST"));
2279
2280 #ifdef INET6
2281         if (inp->inp_vflag & INP_IPV6)
2282                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2283         else
2284 #endif
2285         hashkey_faddr = inp->inp_faddr.s_addr;
2286
2287         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2288                 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2289
2290         LIST_REMOVE(inp, inp_hash);
2291         LIST_INSERT_HEAD(head, inp, inp_hash);
2292
2293 #ifdef PCBGROUP
2294         if (m != NULL)
2295                 in_pcbgroup_update_mbuf(inp, m);
2296         else
2297                 in_pcbgroup_update(inp);
2298 #endif
2299 }
2300
2301 void
2302 in_pcbrehash(struct inpcb *inp)
2303 {
2304
2305         in_pcbrehash_mbuf(inp, NULL);
2306 }
2307
2308 /*
2309  * Remove PCB from various lists.
2310  */
2311 static void
2312 in_pcbremlists(struct inpcb *inp)
2313 {
2314         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2315
2316 #ifdef INVARIANTS
2317         if (pcbinfo == &V_tcbinfo) {
2318                 INP_INFO_RLOCK_ASSERT(pcbinfo);
2319         } else {
2320                 INP_INFO_WLOCK_ASSERT(pcbinfo);
2321         }
2322 #endif
2323
2324         INP_WLOCK_ASSERT(inp);
2325         INP_LIST_WLOCK_ASSERT(pcbinfo);
2326
2327         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2328         if (inp->inp_flags & INP_INHASHLIST) {
2329                 struct inpcbport *phd = inp->inp_phd;
2330
2331                 INP_HASH_WLOCK(pcbinfo);
2332                 LIST_REMOVE(inp, inp_hash);
2333                 LIST_REMOVE(inp, inp_portlist);
2334                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2335                         LIST_REMOVE(phd, phd_hash);
2336                         free(phd, M_PCB);
2337                 }
2338                 INP_HASH_WUNLOCK(pcbinfo);
2339                 inp->inp_flags &= ~INP_INHASHLIST;
2340         }
2341         LIST_REMOVE(inp, inp_list);
2342         pcbinfo->ipi_count--;
2343 #ifdef PCBGROUP
2344         in_pcbgroup_remove(inp);
2345 #endif
2346 }
2347
2348 /*
2349  * Check for alternatives when higher level complains
2350  * about service problems.  For now, invalidate cached
2351  * routing information.  If the route was created dynamically
2352  * (by a redirect), time to try a default gateway again.
2353  */
2354 void
2355 in_losing(struct inpcb *inp)
2356 {
2357
2358         RO_INVALIDATE_CACHE(&inp->inp_route);
2359         return;
2360 }
2361
2362 /*
2363  * A set label operation has occurred at the socket layer, propagate the
2364  * label change into the in_pcb for the socket.
2365  */
2366 void
2367 in_pcbsosetlabel(struct socket *so)
2368 {
2369 #ifdef MAC
2370         struct inpcb *inp;
2371
2372         inp = sotoinpcb(so);
2373         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2374
2375         INP_WLOCK(inp);
2376         SOCK_LOCK(so);
2377         mac_inpcb_sosetlabel(so, inp);
2378         SOCK_UNLOCK(so);
2379         INP_WUNLOCK(inp);
2380 #endif
2381 }
2382
2383 /*
2384  * ipport_tick runs once per second, determining if random port allocation
2385  * should be continued.  If more than ipport_randomcps ports have been
2386  * allocated in the last second, then we return to sequential port
2387  * allocation. We return to random allocation only once we drop below
2388  * ipport_randomcps for at least ipport_randomtime seconds.
2389  */
2390 static void
2391 ipport_tick(void *xtp)
2392 {
2393         VNET_ITERATOR_DECL(vnet_iter);
2394
2395         VNET_LIST_RLOCK_NOSLEEP();
2396         VNET_FOREACH(vnet_iter) {
2397                 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2398                 if (V_ipport_tcpallocs <=
2399                     V_ipport_tcplastcount + V_ipport_randomcps) {
2400                         if (V_ipport_stoprandom > 0)
2401                                 V_ipport_stoprandom--;
2402                 } else
2403                         V_ipport_stoprandom = V_ipport_randomtime;
2404                 V_ipport_tcplastcount = V_ipport_tcpallocs;
2405                 CURVNET_RESTORE();
2406         }
2407         VNET_LIST_RUNLOCK_NOSLEEP();
2408         callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2409 }
2410
2411 static void
2412 ip_fini(void *xtp)
2413 {
2414
2415         callout_stop(&ipport_tick_callout);
2416 }
2417
2418 /*
2419  * The ipport_callout should start running at about the time we attach the
2420  * inet or inet6 domains.
2421  */
2422 static void
2423 ipport_tick_init(const void *unused __unused)
2424 {
2425
2426         /* Start ipport_tick. */
2427         callout_init(&ipport_tick_callout, 1);
2428         callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2429         EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2430                 SHUTDOWN_PRI_DEFAULT);
2431 }
2432 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2433     ipport_tick_init, NULL);
2434
2435 void
2436 inp_wlock(struct inpcb *inp)
2437 {
2438
2439         INP_WLOCK(inp);
2440 }
2441
2442 void
2443 inp_wunlock(struct inpcb *inp)
2444 {
2445
2446         INP_WUNLOCK(inp);
2447 }
2448
2449 void
2450 inp_rlock(struct inpcb *inp)
2451 {
2452
2453         INP_RLOCK(inp);
2454 }
2455
2456 void
2457 inp_runlock(struct inpcb *inp)
2458 {
2459
2460         INP_RUNLOCK(inp);
2461 }
2462
2463 #ifdef INVARIANT_SUPPORT
2464 void
2465 inp_lock_assert(struct inpcb *inp)
2466 {
2467
2468         INP_WLOCK_ASSERT(inp);
2469 }
2470
2471 void
2472 inp_unlock_assert(struct inpcb *inp)
2473 {
2474
2475         INP_UNLOCK_ASSERT(inp);
2476 }
2477 #endif
2478
2479 void
2480 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2481 {
2482         struct inpcb *inp;
2483
2484         INP_INFO_WLOCK(&V_tcbinfo);
2485         LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2486                 INP_WLOCK(inp);
2487                 func(inp, arg);
2488                 INP_WUNLOCK(inp);
2489         }
2490         INP_INFO_WUNLOCK(&V_tcbinfo);
2491 }
2492
2493 struct socket *
2494 inp_inpcbtosocket(struct inpcb *inp)
2495 {
2496
2497         INP_WLOCK_ASSERT(inp);
2498         return (inp->inp_socket);
2499 }
2500
2501 struct tcpcb *
2502 inp_inpcbtotcpcb(struct inpcb *inp)
2503 {
2504
2505         INP_WLOCK_ASSERT(inp);
2506         return ((struct tcpcb *)inp->inp_ppcb);
2507 }
2508
2509 int
2510 inp_ip_tos_get(const struct inpcb *inp)
2511 {
2512
2513         return (inp->inp_ip_tos);
2514 }
2515
2516 void
2517 inp_ip_tos_set(struct inpcb *inp, int val)
2518 {
2519
2520         inp->inp_ip_tos = val;
2521 }
2522
2523 void
2524 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2525     uint32_t *faddr, uint16_t *fp)
2526 {
2527
2528         INP_LOCK_ASSERT(inp);
2529         *laddr = inp->inp_laddr.s_addr;
2530         *faddr = inp->inp_faddr.s_addr;
2531         *lp = inp->inp_lport;
2532         *fp = inp->inp_fport;
2533 }
2534
2535 struct inpcb *
2536 so_sotoinpcb(struct socket *so)
2537 {
2538
2539         return (sotoinpcb(so));
2540 }
2541
2542 struct tcpcb *
2543 so_sototcpcb(struct socket *so)
2544 {
2545
2546         return (sototcpcb(so));
2547 }
2548
2549 /*
2550  * Create an external-format (``xinpcb'') structure using the information in
2551  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2552  * reduce the spew of irrelevant information over this interface, to isolate
2553  * user code from changes in the kernel structure, and potentially to provide
2554  * information-hiding if we decide that some of this information should be
2555  * hidden from users.
2556  */
2557 void
2558 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2559 {
2560
2561         xi->xi_len = sizeof(struct xinpcb);
2562         if (inp->inp_socket)
2563                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2564         else
2565                 bzero(&xi->xi_socket, sizeof(struct xsocket));
2566         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2567         xi->inp_gencnt = inp->inp_gencnt;
2568         xi->inp_ppcb = inp->inp_ppcb;
2569         xi->inp_flow = inp->inp_flow;
2570         xi->inp_flowid = inp->inp_flowid;
2571         xi->inp_flowtype = inp->inp_flowtype;
2572         xi->inp_flags = inp->inp_flags;
2573         xi->inp_flags2 = inp->inp_flags2;
2574         xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2575         xi->in6p_cksum = inp->in6p_cksum;
2576         xi->in6p_hops = inp->in6p_hops;
2577         xi->inp_ip_tos = inp->inp_ip_tos;
2578         xi->inp_vflag = inp->inp_vflag;
2579         xi->inp_ip_ttl = inp->inp_ip_ttl;
2580         xi->inp_ip_p = inp->inp_ip_p;
2581         xi->inp_ip_minttl = inp->inp_ip_minttl;
2582 }
2583
2584 #ifdef DDB
2585 static void
2586 db_print_indent(int indent)
2587 {
2588         int i;
2589
2590         for (i = 0; i < indent; i++)
2591                 db_printf(" ");
2592 }
2593
2594 static void
2595 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2596 {
2597         char faddr_str[48], laddr_str[48];
2598
2599         db_print_indent(indent);
2600         db_printf("%s at %p\n", name, inc);
2601
2602         indent += 2;
2603
2604 #ifdef INET6
2605         if (inc->inc_flags & INC_ISIPV6) {
2606                 /* IPv6. */
2607                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2608                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2609         } else
2610 #endif
2611         {
2612                 /* IPv4. */
2613                 inet_ntoa_r(inc->inc_laddr, laddr_str);
2614                 inet_ntoa_r(inc->inc_faddr, faddr_str);
2615         }
2616         db_print_indent(indent);
2617         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2618             ntohs(inc->inc_lport));
2619         db_print_indent(indent);
2620         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2621             ntohs(inc->inc_fport));
2622 }
2623
2624 static void
2625 db_print_inpflags(int inp_flags)
2626 {
2627         int comma;
2628
2629         comma = 0;
2630         if (inp_flags & INP_RECVOPTS) {
2631                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2632                 comma = 1;
2633         }
2634         if (inp_flags & INP_RECVRETOPTS) {
2635                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2636                 comma = 1;
2637         }
2638         if (inp_flags & INP_RECVDSTADDR) {
2639                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2640                 comma = 1;
2641         }
2642         if (inp_flags & INP_ORIGDSTADDR) {
2643                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2644                 comma = 1;
2645         }
2646         if (inp_flags & INP_HDRINCL) {
2647                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2648                 comma = 1;
2649         }
2650         if (inp_flags & INP_HIGHPORT) {
2651                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2652                 comma = 1;
2653         }
2654         if (inp_flags & INP_LOWPORT) {
2655                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2656                 comma = 1;
2657         }
2658         if (inp_flags & INP_ANONPORT) {
2659                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2660                 comma = 1;
2661         }
2662         if (inp_flags & INP_RECVIF) {
2663                 db_printf("%sINP_RECVIF", comma ? ", " : "");
2664                 comma = 1;
2665         }
2666         if (inp_flags & INP_MTUDISC) {
2667                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2668                 comma = 1;
2669         }
2670         if (inp_flags & INP_RECVTTL) {
2671                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2672                 comma = 1;
2673         }
2674         if (inp_flags & INP_DONTFRAG) {
2675                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2676                 comma = 1;
2677         }
2678         if (inp_flags & INP_RECVTOS) {
2679                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2680                 comma = 1;
2681         }
2682         if (inp_flags & IN6P_IPV6_V6ONLY) {
2683                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2684                 comma = 1;
2685         }
2686         if (inp_flags & IN6P_PKTINFO) {
2687                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2688                 comma = 1;
2689         }
2690         if (inp_flags & IN6P_HOPLIMIT) {
2691                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2692                 comma = 1;
2693         }
2694         if (inp_flags & IN6P_HOPOPTS) {
2695                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2696                 comma = 1;
2697         }
2698         if (inp_flags & IN6P_DSTOPTS) {
2699                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2700                 comma = 1;
2701         }
2702         if (inp_flags & IN6P_RTHDR) {
2703                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2704                 comma = 1;
2705         }
2706         if (inp_flags & IN6P_RTHDRDSTOPTS) {
2707                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2708                 comma = 1;
2709         }
2710         if (inp_flags & IN6P_TCLASS) {
2711                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2712                 comma = 1;
2713         }
2714         if (inp_flags & IN6P_AUTOFLOWLABEL) {
2715                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2716                 comma = 1;
2717         }
2718         if (inp_flags & INP_TIMEWAIT) {
2719                 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2720                 comma  = 1;
2721         }
2722         if (inp_flags & INP_ONESBCAST) {
2723                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2724                 comma  = 1;
2725         }
2726         if (inp_flags & INP_DROPPED) {
2727                 db_printf("%sINP_DROPPED", comma ? ", " : "");
2728                 comma  = 1;
2729         }
2730         if (inp_flags & INP_SOCKREF) {
2731                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2732                 comma  = 1;
2733         }
2734         if (inp_flags & IN6P_RFC2292) {
2735                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2736                 comma = 1;
2737         }
2738         if (inp_flags & IN6P_MTU) {
2739                 db_printf("IN6P_MTU%s", comma ? ", " : "");
2740                 comma = 1;
2741         }
2742 }
2743
2744 static void
2745 db_print_inpvflag(u_char inp_vflag)
2746 {
2747         int comma;
2748
2749         comma = 0;
2750         if (inp_vflag & INP_IPV4) {
2751                 db_printf("%sINP_IPV4", comma ? ", " : "");
2752                 comma  = 1;
2753         }
2754         if (inp_vflag & INP_IPV6) {
2755                 db_printf("%sINP_IPV6", comma ? ", " : "");
2756                 comma  = 1;
2757         }
2758         if (inp_vflag & INP_IPV6PROTO) {
2759                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2760                 comma  = 1;
2761         }
2762 }
2763
2764 static void
2765 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2766 {
2767
2768         db_print_indent(indent);
2769         db_printf("%s at %p\n", name, inp);
2770
2771         indent += 2;
2772
2773         db_print_indent(indent);
2774         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2775
2776         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2777
2778         db_print_indent(indent);
2779         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2780             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2781
2782         db_print_indent(indent);
2783         db_printf("inp_label: %p   inp_flags: 0x%x (",
2784            inp->inp_label, inp->inp_flags);
2785         db_print_inpflags(inp->inp_flags);
2786         db_printf(")\n");
2787
2788         db_print_indent(indent);
2789         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2790             inp->inp_vflag);
2791         db_print_inpvflag(inp->inp_vflag);
2792         db_printf(")\n");
2793
2794         db_print_indent(indent);
2795         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2796             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2797
2798         db_print_indent(indent);
2799 #ifdef INET6
2800         if (inp->inp_vflag & INP_IPV6) {
2801                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
2802                     "in6p_moptions: %p\n", inp->in6p_options,
2803                     inp->in6p_outputopts, inp->in6p_moptions);
2804                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2805                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2806                     inp->in6p_hops);
2807         } else
2808 #endif
2809         {
2810                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2811                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2812                     inp->inp_options, inp->inp_moptions);
2813         }
2814
2815         db_print_indent(indent);
2816         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2817             (uintmax_t)inp->inp_gencnt);
2818 }
2819
2820 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2821 {
2822         struct inpcb *inp;
2823
2824         if (!have_addr) {
2825                 db_printf("usage: show inpcb <addr>\n");
2826                 return;
2827         }
2828         inp = (struct inpcb *)addr;
2829
2830         db_print_inpcb(inp, "inpcb", 0);
2831 }
2832 #endif /* DDB */
2833
2834 #ifdef RATELIMIT
2835 /*
2836  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2837  * if any.
2838  */
2839 int
2840 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2841 {
2842         union if_snd_tag_modify_params params = {
2843                 .rate_limit.max_rate = max_pacing_rate,
2844         };
2845         struct m_snd_tag *mst;
2846         struct ifnet *ifp;
2847         int error;
2848
2849         mst = inp->inp_snd_tag;
2850         if (mst == NULL)
2851                 return (EINVAL);
2852
2853         ifp = mst->ifp;
2854         if (ifp == NULL)
2855                 return (EINVAL);
2856
2857         if (ifp->if_snd_tag_modify == NULL) {
2858                 error = EOPNOTSUPP;
2859         } else {
2860                 error = ifp->if_snd_tag_modify(mst, &params);
2861         }
2862         return (error);
2863 }
2864
2865 /*
2866  * Query existing TX rate limit based on the existing
2867  * "inp->inp_snd_tag", if any.
2868  */
2869 int
2870 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
2871 {
2872         union if_snd_tag_query_params params = { };
2873         struct m_snd_tag *mst;
2874         struct ifnet *ifp;
2875         int error;
2876
2877         mst = inp->inp_snd_tag;
2878         if (mst == NULL)
2879                 return (EINVAL);
2880
2881         ifp = mst->ifp;
2882         if (ifp == NULL)
2883                 return (EINVAL);
2884
2885         if (ifp->if_snd_tag_query == NULL) {
2886                 error = EOPNOTSUPP;
2887         } else {
2888                 error = ifp->if_snd_tag_query(mst, &params);
2889                 if (error == 0 &&  p_max_pacing_rate != NULL)
2890                         *p_max_pacing_rate = params.rate_limit.max_rate;
2891         }
2892         return (error);
2893 }
2894
2895 /*
2896  * Query existing TX queue level based on the existing
2897  * "inp->inp_snd_tag", if any.
2898  */
2899 int
2900 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
2901 {
2902         union if_snd_tag_query_params params = { };
2903         struct m_snd_tag *mst;
2904         struct ifnet *ifp;
2905         int error;
2906
2907         mst = inp->inp_snd_tag;
2908         if (mst == NULL)
2909                 return (EINVAL);
2910
2911         ifp = mst->ifp;
2912         if (ifp == NULL)
2913                 return (EINVAL);
2914
2915         if (ifp->if_snd_tag_query == NULL)
2916                 return (EOPNOTSUPP);
2917
2918         error = ifp->if_snd_tag_query(mst, &params);
2919         if (error == 0 &&  p_txqueue_level != NULL)
2920                 *p_txqueue_level = params.rate_limit.queue_level;
2921         return (error);
2922 }
2923
2924 /*
2925  * Allocate a new TX rate limit send tag from the network interface
2926  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
2927  */
2928 int
2929 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
2930     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
2931 {
2932         union if_snd_tag_alloc_params params = {
2933                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
2934                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
2935                 .rate_limit.hdr.flowid = flowid,
2936                 .rate_limit.hdr.flowtype = flowtype,
2937                 .rate_limit.max_rate = max_pacing_rate,
2938         };
2939         int error;
2940
2941         INP_WLOCK_ASSERT(inp);
2942
2943         if (inp->inp_snd_tag != NULL)
2944                 return (EINVAL);
2945
2946         if (ifp->if_snd_tag_alloc == NULL) {
2947                 error = EOPNOTSUPP;
2948         } else {
2949                 error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
2950
2951                 /*
2952                  * At success increment the refcount on
2953                  * the send tag's network interface:
2954                  */
2955                 if (error == 0)
2956                         if_ref(inp->inp_snd_tag->ifp);
2957         }
2958         return (error);
2959 }
2960
2961 /*
2962  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
2963  * if any:
2964  */
2965 void
2966 in_pcbdetach_txrtlmt(struct inpcb *inp)
2967 {
2968         struct m_snd_tag *mst;
2969         struct ifnet *ifp;
2970
2971         INP_WLOCK_ASSERT(inp);
2972
2973         mst = inp->inp_snd_tag;
2974         inp->inp_snd_tag = NULL;
2975
2976         if (mst == NULL)
2977                 return;
2978
2979         ifp = mst->ifp;
2980         if (ifp == NULL)
2981                 return;
2982
2983         /*
2984          * If the device was detached while we still had reference(s)
2985          * on the ifp, we assume if_snd_tag_free() was replaced with
2986          * stubs.
2987          */
2988         ifp->if_snd_tag_free(mst);
2989
2990         /* release reference count on network interface */
2991         if_rele(ifp);
2992 }
2993
2994 /*
2995  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
2996  * is set in the fast path and will attach/detach/modify the TX rate
2997  * limit send tag based on the socket's so_max_pacing_rate value.
2998  */
2999 void
3000 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3001 {
3002         struct socket *socket;
3003         uint32_t max_pacing_rate;
3004         bool did_upgrade;
3005         int error;
3006
3007         if (inp == NULL)
3008                 return;
3009
3010         socket = inp->inp_socket;
3011         if (socket == NULL)
3012                 return;
3013
3014         if (!INP_WLOCKED(inp)) {
3015                 /*
3016                  * NOTE: If the write locking fails, we need to bail
3017                  * out and use the non-ratelimited ring for the
3018                  * transmit until there is a new chance to get the
3019                  * write lock.
3020                  */
3021                 if (!INP_TRY_UPGRADE(inp))
3022                         return;
3023                 did_upgrade = 1;
3024         } else {
3025                 did_upgrade = 0;
3026         }
3027
3028         /*
3029          * NOTE: The so_max_pacing_rate value is read unlocked,
3030          * because atomic updates are not required since the variable
3031          * is checked at every mbuf we send. It is assumed that the
3032          * variable read itself will be atomic.
3033          */
3034         max_pacing_rate = socket->so_max_pacing_rate;
3035
3036         /*
3037          * NOTE: When attaching to a network interface a reference is
3038          * made to ensure the network interface doesn't go away until
3039          * all ratelimit connections are gone. The network interface
3040          * pointers compared below represent valid network interfaces,
3041          * except when comparing towards NULL.
3042          */
3043         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3044                 error = 0;
3045         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3046                 if (inp->inp_snd_tag != NULL)
3047                         in_pcbdetach_txrtlmt(inp);
3048                 error = 0;
3049         } else if (inp->inp_snd_tag == NULL) {
3050                 /*
3051                  * In order to utilize packet pacing with RSS, we need
3052                  * to wait until there is a valid RSS hash before we
3053                  * can proceed:
3054                  */
3055                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3056                         error = EAGAIN;
3057                 } else {
3058                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3059                             mb->m_pkthdr.flowid, max_pacing_rate);
3060                 }
3061         } else {
3062                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3063         }
3064         if (error == 0 || error == EOPNOTSUPP)
3065                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3066         if (did_upgrade)
3067                 INP_DOWNGRADE(inp);
3068 }
3069
3070 /*
3071  * Track route changes for TX rate limiting.
3072  */
3073 void
3074 in_pcboutput_eagain(struct inpcb *inp)
3075 {
3076         struct socket *socket;
3077         bool did_upgrade;
3078
3079         if (inp == NULL)
3080                 return;
3081
3082         socket = inp->inp_socket;
3083         if (socket == NULL)
3084                 return;
3085
3086         if (inp->inp_snd_tag == NULL)
3087                 return;
3088
3089         if (!INP_WLOCKED(inp)) {
3090                 /*
3091                  * NOTE: If the write locking fails, we need to bail
3092                  * out and use the non-ratelimited ring for the
3093                  * transmit until there is a new chance to get the
3094                  * write lock.
3095                  */
3096                 if (!INP_TRY_UPGRADE(inp))
3097                         return;
3098                 did_upgrade = 1;
3099         } else {
3100                 did_upgrade = 0;
3101         }
3102
3103         /* detach rate limiting */
3104         in_pcbdetach_txrtlmt(inp);
3105
3106         /* make sure new mbuf send tag allocation is made */
3107         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3108
3109         if (did_upgrade)
3110                 INP_DOWNGRADE(inp);
3111 }
3112 #endif /* RATELIMIT */