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 #define INPCBLBGROUP_SIZMIN     8
 112 #define INPCBLBGROUP_SIZMAX     256
 113
 114 static struct callout   ipport_tick_callout;
 115
 116 /*
 117  * These configure the range of local port addresses assigned to
 118  * "unspecified" outgoing connections/packets/whatever.
 119  */
 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;    /* 1023 */
 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;    /* 600 */
 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;     /* 10000 */
 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;       /* 65535 */
 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;      /* 49152 */
 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;        /* 65535 */
 126
 127 /*
 128  * Reserved ports accessible only to root. There are significant
 129  * security considerations that must be accounted for when changing these,
 130  * but the security benefits can be great. Please be careful.
 131  */
 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;    /* 1023 */
 133 VNET_DEFINE(int, ipport_reservedlow);
 134
 135 /* Variables dealing with random ephemeral port allocation. */
 136 VNET_DEFINE(int, ipport_randomized) = 1;        /* user controlled via sysctl */
 137 VNET_DEFINE(int, ipport_randomcps) = 10;        /* user controlled via sysctl */
 138 VNET_DEFINE(int, ipport_randomtime) = 45;       /* user controlled via sysctl */
 139 VNET_DEFINE(int, ipport_stoprandom);            /* toggled by ipport_tick */
 140 VNET_DEFINE(int, ipport_tcpallocs);
 141 static VNET_DEFINE(int, ipport_tcplastcount);
 142
 143 #define V_ipport_tcplastcount           VNET(ipport_tcplastcount)
 144
 145 static void     in_pcbremlists(struct inpcb *inp);
 146 #ifdef INET
 147 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 148                             struct in_addr faddr, u_int fport_arg,
 149                             struct in_addr laddr, u_int lport_arg,
 150                             int lookupflags, struct ifnet *ifp);
 151
 152 #define RANGECHK(var, min, max) \
 153         if ((var) < (min)) { (var) = (min); } \
 154         else if ((var) > (max)) { (var) = (max); }
 155
 156 static int
 157 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 158 {
 159         int error;
 160
 161         error = sysctl_handle_int(oidp, arg1, arg2, req);
 162         if (error == 0) {
 163                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 164                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 165                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 166                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 167                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 168                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 169         }
 170         return (error);
 171 }
 172
 173 #undef RANGECHK
 174
 175 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
 176     "IP Ports");
 177
 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 179         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 180         &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 182         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 183         &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 185         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 186         &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 188         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 189         &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 191         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 192         &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 194         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 195         &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 197         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 198         &VNET_NAME(ipport_reservedhigh), 0, "");
 199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 200         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 202         CTLFLAG_VNET | CTLFLAG_RW,
 203         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 204 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
 205         CTLFLAG_VNET | CTLFLAG_RW,
 206         &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
 207         "allocations before switching to a sequental one");
 208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
 209         CTLFLAG_VNET | CTLFLAG_RW,
 210         &VNET_NAME(ipport_randomtime), 0,
 211         "Minimum time to keep sequental port "
 212         "allocation before switching to a random one");
 213 #endif /* INET */
 214
 215 /*
 216  * in_pcb.c: manage the Protocol Control Blocks.
 217  *
 218  * NOTE: It is assumed that most of these functions will be called with
 219  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
 220  * functions often modify hash chains or addresses in pcbs.
 221  */
 222
 223 static struct inpcblbgroup *
 224 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
 225     uint16_t port, const union in_dependaddr *addr, int size)
 226 {
 227         struct inpcblbgroup *grp;
 228         size_t bytes;
 229
 230         bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
 231         grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
 232         if (!grp)
 233                 return (NULL);
 234         grp->il_vflag = vflag;
 235         grp->il_lport = port;
 236         grp->il_dependladdr = *addr;
 237         grp->il_inpsiz = size;
 238         LIST_INSERT_HEAD(hdr, grp, il_list);
 239         return (grp);
 240 }
 241
 242 static void
 243 in_pcblbgroup_free(struct inpcblbgroup *grp)
 244 {
 245
 246         LIST_REMOVE(grp, il_list);
 247         free(grp, M_TEMP);
 248 }
 249
 250 static struct inpcblbgroup *
 251 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
 252     struct inpcblbgroup *old_grp, int size)
 253 {
 254         struct inpcblbgroup *grp;
 255         int i;
 256
 257         grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
 258             old_grp->il_lport, &old_grp->il_dependladdr, size);
 259         if (!grp)
 260                 return (NULL);
 261
 262         KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
 263             ("invalid new local group size %d and old local group count %d",
 264              grp->il_inpsiz, old_grp->il_inpcnt));
 265
 266         for (i = 0; i < old_grp->il_inpcnt; ++i)
 267                 grp->il_inp[i] = old_grp->il_inp[i];
 268         grp->il_inpcnt = old_grp->il_inpcnt;
 269         in_pcblbgroup_free(old_grp);
 270         return (grp);
 271 }
 272
 273 /*
 274  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
 275  * and shrink group if possible.
 276  */
 277 static void
 278 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
 279     int i)
 280 {
 281         struct inpcblbgroup *grp = *grpp;
 282
 283         for (; i + 1 < grp->il_inpcnt; ++i)
 284                 grp->il_inp[i] = grp->il_inp[i + 1];
 285         grp->il_inpcnt--;
 286
 287         if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
 288             grp->il_inpcnt <= (grp->il_inpsiz / 4)) {
 289                 /* Shrink this group. */
 290                 struct inpcblbgroup *new_grp =
 291                         in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
 292                 if (new_grp)
 293                         *grpp = new_grp;
 294         }
 295         return;
 296 }
 297
 298 /*
 299  * Add PCB to load balance group for SO_REUSEPORT_LB option.
 300  */
 301 static int
 302 in_pcbinslbgrouphash(struct inpcb *inp)
 303 {
 304         struct inpcbinfo *pcbinfo;
 305         struct inpcblbgrouphead *hdr;
 306         struct inpcblbgroup *grp;
 307         uint16_t hashmask, lport;
 308         uint32_t group_index;
 309         struct ucred *cred;
 310         static int limit_logged = 0;
 311
 312         pcbinfo = inp->inp_pcbinfo;
 313
 314         INP_WLOCK_ASSERT(inp);
 315         INP_HASH_WLOCK_ASSERT(pcbinfo);
 316
 317         if (pcbinfo->ipi_lbgrouphashbase == NULL)
 318                 return (0);
 319
 320         hashmask = pcbinfo->ipi_lbgrouphashmask;
 321         lport = inp->inp_lport;
 322         group_index = INP_PCBLBGROUP_PORTHASH(lport, hashmask);
 323         hdr = &pcbinfo->ipi_lbgrouphashbase[group_index];
 324
 325         /*
 326          * Don't allow jailed socket to join local group.
 327          */
 328         if (inp->inp_socket != NULL)
 329                 cred = inp->inp_socket->so_cred;
 330         else
 331                 cred = NULL;
 332         if (cred != NULL && jailed(cred))
 333                 return (0);
 334
 335 #ifdef INET6
 336         /*
 337          * Don't allow IPv4 mapped INET6 wild socket.
 338          */
 339         if ((inp->inp_vflag & INP_IPV4) &&
 340             inp->inp_laddr.s_addr == INADDR_ANY &&
 341             INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
 342                 return (0);
 343         }
 344 #endif
 345
 346         hdr = &pcbinfo->ipi_lbgrouphashbase[
 347             INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
 348                 pcbinfo->ipi_lbgrouphashmask)];
 349         LIST_FOREACH(grp, hdr, il_list) {
 350                 if (grp->il_vflag == inp->inp_vflag &&
 351                     grp->il_lport == inp->inp_lport &&
 352                     memcmp(&grp->il_dependladdr,
 353                         &inp->inp_inc.inc_ie.ie_dependladdr,
 354                         sizeof(grp->il_dependladdr)) == 0) {
 355                         break;
 356                 }
 357         }
 358         if (grp == NULL) {
 359                 /* Create new load balance group. */
 360                 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
 361                     inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
 362                     INPCBLBGROUP_SIZMIN);
 363                 if (!grp)
 364                         return (ENOBUFS);
 365         } else if (grp->il_inpcnt == grp->il_inpsiz) {
 366                 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
 367                         if (!limit_logged) {
 368                                 limit_logged = 1;
 369                                 printf("lb group port %d, limit reached\n",
 370                                     ntohs(grp->il_lport));
 371                         }
 372                         return (0);
 373                 }
 374
 375                 /* Expand this local group. */
 376                 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
 377                 if (!grp)
 378                         return (ENOBUFS);
 379         }
 380
 381         KASSERT(grp->il_inpcnt < grp->il_inpsiz,
 382                         ("invalid local group size %d and count %d",
 383                          grp->il_inpsiz, grp->il_inpcnt));
 384
 385         grp->il_inp[grp->il_inpcnt] = inp;
 386         grp->il_inpcnt++;
 387         return (0);
 388 }
 389
 390 /*
 391  * Remove PCB from load balance group.
 392  */
 393 static void
 394 in_pcbremlbgrouphash(struct inpcb *inp)
 395 {
 396         struct inpcbinfo *pcbinfo;
 397         struct inpcblbgrouphead *hdr;
 398         struct inpcblbgroup *grp;
 399         int i;
 400
 401         pcbinfo = inp->inp_pcbinfo;
 402
 403         INP_WLOCK_ASSERT(inp);
 404         INP_HASH_WLOCK_ASSERT(pcbinfo);
 405
 406         if (pcbinfo->ipi_lbgrouphashbase == NULL)
 407                 return;
 408
 409         hdr = &pcbinfo->ipi_lbgrouphashbase[
 410             INP_PCBLBGROUP_PORTHASH(inp->inp_lport,
 411                 pcbinfo->ipi_lbgrouphashmask)];
 412
 413         LIST_FOREACH(grp, hdr, il_list) {
 414                 for (i = 0; i < grp->il_inpcnt; ++i) {
 415                         if (grp->il_inp[i] != inp)
 416                                 continue;
 417
 418                         if (grp->il_inpcnt == 1) {
 419                                 /* We are the last, free this local group. */
 420                                 in_pcblbgroup_free(grp);
 421                         } else {
 422                                 /* Pull up inpcbs, shrink group if possible. */
 423                                 in_pcblbgroup_reorder(hdr, &grp, i);
 424                         }
 425                         return;
 426                 }
 427         }
 428 }
 429
 430 /*
 431  * Different protocols initialize their inpcbs differently - giving
 432  * different name to the lock.  But they all are disposed the same.
 433  */
 434 static void
 435 inpcb_fini(void *mem, int size)
 436 {
 437         struct inpcb *inp = mem;
 438
 439         INP_LOCK_DESTROY(inp);
 440 }
 441
 442 /*
 443  * Initialize an inpcbinfo -- we should be able to reduce the number of
 444  * arguments in time.
 445  */
 446 void
 447 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
 448     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
 449     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
 450 {
 451
 452         INP_INFO_LOCK_INIT(pcbinfo, name);
 453         INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");     /* XXXRW: argument? */
 454         INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
 455 #ifdef VIMAGE
 456         pcbinfo->ipi_vnet = curvnet;
 457 #endif
 458         pcbinfo->ipi_listhead = listhead;
 459         LIST_INIT(pcbinfo->ipi_listhead);
 460         pcbinfo->ipi_count = 0;
 461         pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
 462             &pcbinfo->ipi_hashmask);
 463         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 464             &pcbinfo->ipi_porthashmask);
 465         pcbinfo->ipi_lbgrouphashbase = hashinit(hash_nelements, M_PCB,
 466             &pcbinfo->ipi_lbgrouphashmask);
 467 #ifdef PCBGROUP
 468         in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
 469 #endif
 470         pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
 471             NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
 472         uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
 473         uma_zone_set_warning(pcbinfo->ipi_zone,
 474             "kern.ipc.maxsockets limit reached");
 475 }
 476
 477 /*
 478  * Destroy an inpcbinfo.
 479  */
 480 void
 481 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 482 {
 483
 484         KASSERT(pcbinfo->ipi_count == 0,
 485             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 486
 487         hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
 488         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 489             pcbinfo->ipi_porthashmask);
 490         hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
 491             pcbinfo->ipi_lbgrouphashmask);
 492 #ifdef PCBGROUP
 493         in_pcbgroup_destroy(pcbinfo);
 494 #endif
 495         uma_zdestroy(pcbinfo->ipi_zone);
 496         INP_LIST_LOCK_DESTROY(pcbinfo);
 497         INP_HASH_LOCK_DESTROY(pcbinfo);
 498         INP_INFO_LOCK_DESTROY(pcbinfo);
 499 }
 500
 501 /*
 502  * Allocate a PCB and associate it with the socket.
 503  * On success return with the PCB locked.
 504  */
 505 int
 506 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 507 {
 508         struct inpcb *inp;
 509         int error;
 510
 511 #ifdef INVARIANTS
 512         if (pcbinfo == &V_tcbinfo) {
 513                 INP_INFO_RLOCK_ASSERT(pcbinfo);
 514         } else {
 515                 INP_INFO_WLOCK_ASSERT(pcbinfo);
 516         }
 517 #endif
 518
 519         error = 0;
 520         inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
 521         if (inp == NULL)
 522                 return (ENOBUFS);
 523         bzero(&inp->inp_start_zero, inp_zero_size);
 524         inp->inp_pcbinfo = pcbinfo;
 525         inp->inp_socket = so;
 526         inp->inp_cred = crhold(so->so_cred);
 527         inp->inp_inc.inc_fibnum = so->so_fibnum;
 528 #ifdef MAC
 529         error = mac_inpcb_init(inp, M_NOWAIT);
 530         if (error != 0)
 531                 goto out;
 532         mac_inpcb_create(so, inp);
 533 #endif
 534 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 535         error = ipsec_init_pcbpolicy(inp);
 536         if (error != 0) {
 537 #ifdef MAC
 538                 mac_inpcb_destroy(inp);
 539 #endif
 540                 goto out;
 541         }
 542 #endif /*IPSEC*/
 543 #ifdef INET6
 544         if (INP_SOCKAF(so) == AF_INET6) {
 545                 inp->inp_vflag |= INP_IPV6PROTO;
 546                 if (V_ip6_v6only)
 547                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 548         }
 549 #endif
 550         INP_WLOCK(inp);
 551         INP_LIST_WLOCK(pcbinfo);
 552         LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
 553         pcbinfo->ipi_count++;
 554         so->so_pcb = (caddr_t)inp;
 555 #ifdef INET6
 556         if (V_ip6_auto_flowlabel)
 557                 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 558 #endif
 559         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 560         refcount_init(&inp->inp_refcount, 1);   /* Reference from inpcbinfo */
 561
 562         /*
 563          * Routes in inpcb's can cache L2 as well; they are guaranteed
 564          * to be cleaned up.
 565          */
 566         inp->inp_route.ro_flags = RT_LLE_CACHE;
 567         INP_LIST_WUNLOCK(pcbinfo);
 568 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 569 out:
 570         if (error != 0) {
 571                 crfree(inp->inp_cred);
 572                 uma_zfree(pcbinfo->ipi_zone, inp);
 573         }
 574 #endif
 575         return (error);
 576 }
 577
 578 #ifdef INET
 579 int
 580 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 581 {
 582         int anonport, error;
 583
 584         INP_WLOCK_ASSERT(inp);
 585         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 586
 587         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 588                 return (EINVAL);
 589         anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
 590         error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
 591             &inp->inp_lport, cred);
 592         if (error)
 593                 return (error);
 594         if (in_pcbinshash(inp) != 0) {
 595                 inp->inp_laddr.s_addr = INADDR_ANY;
 596                 inp->inp_lport = 0;
 597                 return (EAGAIN);
 598         }
 599         if (anonport)
 600                 inp->inp_flags |= INP_ANONPORT;
 601         return (0);
 602 }
 603 #endif
 604
 605 /*
 606  * Select a local port (number) to use.
 607  */
 608 #if defined(INET) || defined(INET6)
 609 int
 610 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 611     struct ucred *cred, int lookupflags)
 612 {
 613         struct inpcbinfo *pcbinfo;
 614         struct inpcb *tmpinp;
 615         unsigned short *lastport;
 616         int count, dorandom, error;
 617         u_short aux, first, last, lport;
 618 #ifdef INET
 619         struct in_addr laddr;
 620 #endif
 621
 622         pcbinfo = inp->inp_pcbinfo;
 623
 624         /*
 625          * Because no actual state changes occur here, a global write lock on
 626          * the pcbinfo isn't required.
 627          */
 628         INP_LOCK_ASSERT(inp);
 629         INP_HASH_LOCK_ASSERT(pcbinfo);
 630
 631         if (inp->inp_flags & INP_HIGHPORT) {
 632                 first = V_ipport_hifirstauto;   /* sysctl */
 633                 last  = V_ipport_hilastauto;
 634                 lastport = &pcbinfo->ipi_lasthi;
 635         } else if (inp->inp_flags & INP_LOWPORT) {
 636                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
 637                 if (error)
 638                         return (error);
 639                 first = V_ipport_lowfirstauto;  /* 1023 */
 640                 last  = V_ipport_lowlastauto;   /* 600 */
 641                 lastport = &pcbinfo->ipi_lastlow;
 642         } else {
 643                 first = V_ipport_firstauto;     /* sysctl */
 644                 last  = V_ipport_lastauto;
 645                 lastport = &pcbinfo->ipi_lastport;
 646         }
 647         /*
 648          * For UDP(-Lite), use random port allocation as long as the user
 649          * allows it.  For TCP (and as of yet unknown) connections,
 650          * use random port allocation only if the user allows it AND
 651          * ipport_tick() allows it.
 652          */
 653         if (V_ipport_randomized &&
 654                 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
 655                 pcbinfo == &V_ulitecbinfo))
 656                 dorandom = 1;
 657         else
 658                 dorandom = 0;
 659         /*
 660          * It makes no sense to do random port allocation if
 661          * we have the only port available.
 662          */
 663         if (first == last)
 664                 dorandom = 0;
 665         /* Make sure to not include UDP(-Lite) packets in the count. */
 666         if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
 667                 V_ipport_tcpallocs++;
 668         /*
 669          * Instead of having two loops further down counting up or down
 670          * make sure that first is always <= last and go with only one
 671          * code path implementing all logic.
 672          */
 673         if (first > last) {
 674                 aux = first;
 675                 first = last;
 676                 last = aux;
 677         }
 678
 679 #ifdef INET
 680         /* Make the compiler happy. */
 681         laddr.s_addr = 0;
 682         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 683                 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
 684                     __func__, inp));
 685                 laddr = *laddrp;
 686         }
 687 #endif
 688         tmpinp = NULL;  /* Make compiler happy. */
 689         lport = *lportp;
 690
 691         if (dorandom)
 692                 *lastport = first + (arc4random() % (last - first));
 693
 694         count = last - first;
 695
 696         do {
 697                 if (count-- < 0)        /* completely used? */
 698                         return (EADDRNOTAVAIL);
 699                 ++*lastport;
 700                 if (*lastport < first || *lastport > last)
 701                         *lastport = first;
 702                 lport = htons(*lastport);
 703
 704 #ifdef INET6
 705                 if ((inp->inp_vflag & INP_IPV6) != 0)
 706                         tmpinp = in6_pcblookup_local(pcbinfo,
 707                             &inp->in6p_laddr, lport, lookupflags, cred);
 708 #endif
 709 #if defined(INET) && defined(INET6)
 710                 else
 711 #endif
 712 #ifdef INET
 713                         tmpinp = in_pcblookup_local(pcbinfo, laddr,
 714                             lport, lookupflags, cred);
 715 #endif
 716         } while (tmpinp != NULL);
 717
 718 #ifdef INET
 719         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
 720                 laddrp->s_addr = laddr.s_addr;
 721 #endif
 722         *lportp = lport;
 723
 724         return (0);
 725 }
 726
 727 /*
 728  * Return cached socket options.
 729  */
 730 int
 731 inp_so_options(const struct inpcb *inp)
 732 {
 733         int so_options;
 734
 735         so_options = 0;
 736
 737         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
 738                 so_options |= SO_REUSEPORT_LB;
 739         if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 740                 so_options |= SO_REUSEPORT;
 741         if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 742                 so_options |= SO_REUSEADDR;
 743         return (so_options);
 744 }
 745 #endif /* INET || INET6 */
 746
 747 /*
 748  * Check if a new BINDMULTI socket is allowed to be created.
 749  *
 750  * ni points to the new inp.
 751  * oi points to the exisitng inp.
 752  *
 753  * This checks whether the existing inp also has BINDMULTI and
 754  * whether the credentials match.
 755  */
 756 int
 757 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
 758 {
 759         /* Check permissions match */
 760         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 761             (ni->inp_cred->cr_uid !=
 762             oi->inp_cred->cr_uid))
 763                 return (0);
 764
 765         /* Check the existing inp has BINDMULTI set */
 766         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 767             ((oi->inp_flags2 & INP_BINDMULTI) == 0))
 768                 return (0);
 769
 770         /*
 771          * We're okay - either INP_BINDMULTI isn't set on ni, or
 772          * it is and it matches the checks.
 773          */
 774         return (1);
 775 }
 776
 777 #ifdef INET
 778 /*
 779  * Set up a bind operation on a PCB, performing port allocation
 780  * as required, but do not actually modify the PCB. Callers can
 781  * either complete the bind by setting inp_laddr/inp_lport and
 782  * calling in_pcbinshash(), or they can just use the resulting
 783  * port and address to authorise the sending of a once-off packet.
 784  *
 785  * On error, the values of *laddrp and *lportp are not changed.
 786  */
 787 int
 788 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
 789     u_short *lportp, struct ucred *cred)
 790 {
 791         struct socket *so = inp->inp_socket;
 792         struct sockaddr_in *sin;
 793         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 794         struct in_addr laddr;
 795         u_short lport = 0;
 796         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 797         int error;
 798
 799         /*
 800          * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
 801          * so that we don't have to add to the (already messy) code below.
 802          */
 803         int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
 804
 805         /*
 806          * No state changes, so read locks are sufficient here.
 807          */
 808         INP_LOCK_ASSERT(inp);
 809         INP_HASH_LOCK_ASSERT(pcbinfo);
 810
 811         if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
 812                 return (EADDRNOTAVAIL);
 813         laddr.s_addr = *laddrp;
 814         if (nam != NULL && laddr.s_addr != INADDR_ANY)
 815                 return (EINVAL);
 816         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
 817                 lookupflags = INPLOOKUP_WILDCARD;
 818         if (nam == NULL) {
 819                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 820                         return (error);
 821         } else {
 822                 sin = (struct sockaddr_in *)nam;
 823                 if (nam->sa_len != sizeof (*sin))
 824                         return (EINVAL);
 825 #ifdef notdef
 826                 /*
 827                  * We should check the family, but old programs
 828                  * incorrectly fail to initialize it.
 829                  */
 830                 if (sin->sin_family != AF_INET)
 831                         return (EAFNOSUPPORT);
 832 #endif
 833                 error = prison_local_ip4(cred, &sin->sin_addr);
 834                 if (error)
 835                         return (error);
 836                 if (sin->sin_port != *lportp) {
 837                         /* Don't allow the port to change. */
 838                         if (*lportp != 0)
 839                                 return (EINVAL);
 840                         lport = sin->sin_port;
 841                 }
 842                 /* NB: lport is left as 0 if the port isn't being changed. */
 843                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 844                         /*
 845                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 846                          * allow complete duplication of binding if
 847                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 848                          * and a multicast address is bound on both
 849                          * new and duplicated sockets.
 850                          */
 851                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 852                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 853                         /*
 854                          * XXX: How to deal with SO_REUSEPORT_LB here?
 855                          * Treat same as SO_REUSEPORT for now.
 856                          */
 857                         if ((so->so_options &
 858                             (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
 859                                 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
 860                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 861                         sin->sin_port = 0;              /* yech... */
 862                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 863                         /*
 864                          * Is the address a local IP address?
 865                          * If INP_BINDANY is set, then the socket may be bound
 866                          * to any endpoint address, local or not.
 867                          */
 868                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 869                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 870                                 return (EADDRNOTAVAIL);
 871                 }
 872                 laddr = sin->sin_addr;
 873                 if (lport) {
 874                         struct inpcb *t;
 875                         struct tcptw *tw;
 876
 877                         /* GROSS */
 878                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 879                             ntohs(lport) >= V_ipport_reservedlow &&
 880                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
 881                             0))
 882                                 return (EACCES);
 883                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 884                             priv_check_cred(inp->inp_cred,
 885                             PRIV_NETINET_REUSEPORT, 0) != 0) {
 886                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 887                                     lport, INPLOOKUP_WILDCARD, cred);
 888         /*
 889          * XXX
 890          * This entire block sorely needs a rewrite.
 891          */
 892                                 if (t &&
 893                                     ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 894                                     ((t->inp_flags & INP_TIMEWAIT) == 0) &&
 895                                     (so->so_type != SOCK_STREAM ||
 896                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
 897                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
 898                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
 899                                      (t->inp_flags2 & INP_REUSEPORT) ||
 900                                      (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
 901                                     (inp->inp_cred->cr_uid !=
 902                                      t->inp_cred->cr_uid))
 903                                         return (EADDRINUSE);
 904
 905                                 /*
 906                                  * If the socket is a BINDMULTI socket, then
 907                                  * the credentials need to match and the
 908                                  * original socket also has to have been bound
 909                                  * with BINDMULTI.
 910                                  */
 911                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 912                                         return (EADDRINUSE);
 913                         }
 914                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 915                             lport, lookupflags, cred);
 916                         if (t && (t->inp_flags & INP_TIMEWAIT)) {
 917                                 /*
 918                                  * XXXRW: If an incpb has had its timewait
 919                                  * state recycled, we treat the address as
 920                                  * being in use (for now).  This is better
 921                                  * than a panic, but not desirable.
 922                                  */
 923                                 tw = intotw(t);
 924                                 if (tw == NULL ||
 925                                     ((reuseport & tw->tw_so_options) == 0 &&
 926                                         (reuseport_lb &
 927                                             tw->tw_so_options) == 0)) {
 928                                         return (EADDRINUSE);
 929                                 }
 930                         } else if (t &&
 931                                    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 932                                    (reuseport & inp_so_options(t)) == 0 &&
 933                                    (reuseport_lb & inp_so_options(t)) == 0) {
 934 #ifdef INET6
 935                                 if (ntohl(sin->sin_addr.s_addr) !=
 936                                     INADDR_ANY ||
 937                                     ntohl(t->inp_laddr.s_addr) !=
 938                                     INADDR_ANY ||
 939                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
 940                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
 941 #endif
 942                                                 return (EADDRINUSE);
 943                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 944                                         return (EADDRINUSE);
 945                         }
 946                 }
 947         }
 948         if (*lportp != 0)
 949                 lport = *lportp;
 950         if (lport == 0) {
 951                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
 952                 if (error != 0)
 953                         return (error);
 954
 955         }
 956         *laddrp = laddr.s_addr;
 957         *lportp = lport;
 958         return (0);
 959 }
 960
 961 /*
 962  * Connect from a socket to a specified address.
 963  * Both address and port must be specified in argument sin.
 964  * If don't have a local address for this socket yet,
 965  * then pick one.
 966  */
 967 int
 968 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
 969     struct ucred *cred, struct mbuf *m)
 970 {
 971         u_short lport, fport;
 972         in_addr_t laddr, faddr;
 973         int anonport, error;
 974
 975         INP_WLOCK_ASSERT(inp);
 976         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 977
 978         lport = inp->inp_lport;
 979         laddr = inp->inp_laddr.s_addr;
 980         anonport = (lport == 0);
 981         error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
 982             NULL, cred);
 983         if (error)
 984                 return (error);
 985
 986         /* Do the initial binding of the local address if required. */
 987         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
 988                 inp->inp_lport = lport;
 989                 inp->inp_laddr.s_addr = laddr;
 990                 if (in_pcbinshash(inp) != 0) {
 991                         inp->inp_laddr.s_addr = INADDR_ANY;
 992                         inp->inp_lport = 0;
 993                         return (EAGAIN);
 994                 }
 995         }
 996
 997         /* Commit the remaining changes. */
 998         inp->inp_lport = lport;
 999         inp->inp_laddr.s_addr = laddr;
1000         inp->inp_faddr.s_addr = faddr;
1001         inp->inp_fport = fport;
1002         in_pcbrehash_mbuf(inp, m);
1003
1004         if (anonport)
1005                 inp->inp_flags |= INP_ANONPORT;
1006         return (0);
1007 }
1008
1009 int
1010 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1011 {
1012
1013         return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
1014 }
1015
1016 /*
1017  * Do proper source address selection on an unbound socket in case
1018  * of connect. Take jails into account as well.
1019  */
1020 int
1021 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1022     struct ucred *cred)
1023 {
1024         struct ifaddr *ifa;
1025         struct sockaddr *sa;
1026         struct sockaddr_in *sin;
1027         struct route sro;
1028         int error;
1029
1030         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1031         /*
1032          * Bypass source address selection and use the primary jail IP
1033          * if requested.
1034          */
1035         if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1036                 return (0);
1037
1038         error = 0;
1039         bzero(&sro, sizeof(sro));
1040
1041         sin = (struct sockaddr_in *)&sro.ro_dst;
1042         sin->sin_family = AF_INET;
1043         sin->sin_len = sizeof(struct sockaddr_in);
1044         sin->sin_addr.s_addr = faddr->s_addr;
1045
1046         /*
1047          * If route is known our src addr is taken from the i/f,
1048          * else punt.
1049          *
1050          * Find out route to destination.
1051          */
1052         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1053                 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
1054
1055         /*
1056          * If we found a route, use the address corresponding to
1057          * the outgoing interface.
1058          *
1059          * Otherwise assume faddr is reachable on a directly connected
1060          * network and try to find a corresponding interface to take
1061          * the source address from.
1062          */
1063         NET_EPOCH_ENTER();
1064         if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
1065                 struct in_ifaddr *ia;
1066                 struct ifnet *ifp;
1067
1068                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1069                                         inp->inp_socket->so_fibnum));
1070                 if (ia == NULL) {
1071                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1072                                                 inp->inp_socket->so_fibnum));
1073
1074                 }
1075                 if (ia == NULL) {
1076                         error = ENETUNREACH;
1077                         goto done;
1078                 }
1079
1080                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1081                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1082                         goto done;
1083                 }
1084
1085                 ifp = ia->ia_ifp;
1086                 ia = NULL;
1087                 IF_ADDR_RLOCK(ifp);
1088                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1089
1090                         sa = ifa->ifa_addr;
1091                         if (sa->sa_family != AF_INET)
1092                                 continue;
1093                         sin = (struct sockaddr_in *)sa;
1094                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1095                                 ia = (struct in_ifaddr *)ifa;
1096                                 break;
1097                         }
1098                 }
1099                 if (ia != NULL) {
1100                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1101                         IF_ADDR_RUNLOCK(ifp);
1102                         goto done;
1103                 }
1104                 IF_ADDR_RUNLOCK(ifp);
1105
1106                 /* 3. As a last resort return the 'default' jail address. */
1107                 error = prison_get_ip4(cred, laddr);
1108                 goto done;
1109         }
1110
1111         /*
1112          * If the outgoing interface on the route found is not
1113          * a loopback interface, use the address from that interface.
1114          * In case of jails do those three steps:
1115          * 1. check if the interface address belongs to the jail. If so use it.
1116          * 2. check if we have any address on the outgoing interface
1117          *    belonging to this jail. If so use it.
1118          * 3. as a last resort return the 'default' jail address.
1119          */
1120         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
1121                 struct in_ifaddr *ia;
1122                 struct ifnet *ifp;
1123
1124                 /* If not jailed, use the default returned. */
1125                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1126                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1127                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1128                         goto done;
1129                 }
1130
1131                 /* Jailed. */
1132                 /* 1. Check if the iface address belongs to the jail. */
1133                 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
1134                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1135                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
1136                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1137                         goto done;
1138                 }
1139
1140                 /*
1141                  * 2. Check if we have any address on the outgoing interface
1142                  *    belonging to this jail.
1143                  */
1144                 ia = NULL;
1145                 ifp = sro.ro_rt->rt_ifp;
1146                 IF_ADDR_RLOCK(ifp);
1147                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1148                         sa = ifa->ifa_addr;
1149                         if (sa->sa_family != AF_INET)
1150                                 continue;
1151                         sin = (struct sockaddr_in *)sa;
1152                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1153                                 ia = (struct in_ifaddr *)ifa;
1154                                 break;
1155                         }
1156                 }
1157                 if (ia != NULL) {
1158                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1159                         IF_ADDR_RUNLOCK(ifp);
1160                         goto done;
1161                 }
1162                 IF_ADDR_RUNLOCK(ifp);
1163
1164                 /* 3. As a last resort return the 'default' jail address. */
1165                 error = prison_get_ip4(cred, laddr);
1166                 goto done;
1167         }
1168
1169         /*
1170          * The outgoing interface is marked with 'loopback net', so a route
1171          * to ourselves is here.
1172          * Try to find the interface of the destination address and then
1173          * take the address from there. That interface is not necessarily
1174          * a loopback interface.
1175          * In case of jails, check that it is an address of the jail
1176          * and if we cannot find, fall back to the 'default' jail address.
1177          */
1178         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
1179                 struct sockaddr_in sain;
1180                 struct in_ifaddr *ia;
1181
1182                 bzero(&sain, sizeof(struct sockaddr_in));
1183                 sain.sin_family = AF_INET;
1184                 sain.sin_len = sizeof(struct sockaddr_in);
1185                 sain.sin_addr.s_addr = faddr->s_addr;
1186
1187                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
1188                                         inp->inp_socket->so_fibnum));
1189                 if (ia == NULL)
1190                         ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
1191                                                 inp->inp_socket->so_fibnum));
1192                 if (ia == NULL)
1193                         ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
1194
1195                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1196                         if (ia == NULL) {
1197                                 error = ENETUNREACH;
1198                                 goto done;
1199                         }
1200                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1201                         goto done;
1202                 }
1203
1204                 /* Jailed. */
1205                 if (ia != NULL) {
1206                         struct ifnet *ifp;
1207
1208                         ifp = ia->ia_ifp;
1209                         ia = NULL;
1210                         IF_ADDR_RLOCK(ifp);
1211                         CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1212
1213                                 sa = ifa->ifa_addr;
1214                                 if (sa->sa_family != AF_INET)
1215                                         continue;
1216                                 sin = (struct sockaddr_in *)sa;
1217                                 if (prison_check_ip4(cred,
1218                                     &sin->sin_addr) == 0) {
1219                                         ia = (struct in_ifaddr *)ifa;
1220                                         break;
1221                                 }
1222                         }
1223                         if (ia != NULL) {
1224                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1225                                 IF_ADDR_RUNLOCK(ifp);
1226                                 goto done;
1227                         }
1228                         IF_ADDR_RUNLOCK(ifp);
1229                 }
1230
1231                 /* 3. As a last resort return the 'default' jail address. */
1232                 error = prison_get_ip4(cred, laddr);
1233                 goto done;
1234         }
1235
1236 done:
1237         NET_EPOCH_EXIT();
1238         if (sro.ro_rt != NULL)
1239                 RTFREE(sro.ro_rt);
1240         return (error);
1241 }
1242
1243 /*
1244  * Set up for a connect from a socket to the specified address.
1245  * On entry, *laddrp and *lportp should contain the current local
1246  * address and port for the PCB; these are updated to the values
1247  * that should be placed in inp_laddr and inp_lport to complete
1248  * the connect.
1249  *
1250  * On success, *faddrp and *fportp will be set to the remote address
1251  * and port. These are not updated in the error case.
1252  *
1253  * If the operation fails because the connection already exists,
1254  * *oinpp will be set to the PCB of that connection so that the
1255  * caller can decide to override it. In all other cases, *oinpp
1256  * is set to NULL.
1257  */
1258 int
1259 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1260     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1261     struct inpcb **oinpp, struct ucred *cred)
1262 {
1263         struct rm_priotracker in_ifa_tracker;
1264         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1265         struct in_ifaddr *ia;
1266         struct inpcb *oinp;
1267         struct in_addr laddr, faddr;
1268         u_short lport, fport;
1269         int error;
1270
1271         /*
1272          * Because a global state change doesn't actually occur here, a read
1273          * lock is sufficient.
1274          */
1275         INP_LOCK_ASSERT(inp);
1276         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1277
1278         if (oinpp != NULL)
1279                 *oinpp = NULL;
1280         if (nam->sa_len != sizeof (*sin))
1281                 return (EINVAL);
1282         if (sin->sin_family != AF_INET)
1283                 return (EAFNOSUPPORT);
1284         if (sin->sin_port == 0)
1285                 return (EADDRNOTAVAIL);
1286         laddr.s_addr = *laddrp;
1287         lport = *lportp;
1288         faddr = sin->sin_addr;
1289         fport = sin->sin_port;
1290
1291         if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1292                 /*
1293                  * If the destination address is INADDR_ANY,
1294                  * use the primary local address.
1295                  * If the supplied address is INADDR_BROADCAST,
1296                  * and the primary interface supports broadcast,
1297                  * choose the broadcast address for that interface.
1298                  */
1299                 if (faddr.s_addr == INADDR_ANY) {
1300                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1301                         faddr =
1302                             IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1303                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1304                         if (cred != NULL &&
1305                             (error = prison_get_ip4(cred, &faddr)) != 0)
1306                                 return (error);
1307                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1308                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1309                         if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1310                             IFF_BROADCAST)
1311                                 faddr = satosin(&CK_STAILQ_FIRST(
1312                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1313                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1314                 }
1315         }
1316         if (laddr.s_addr == INADDR_ANY) {
1317                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1318                 /*
1319                  * If the destination address is multicast and an outgoing
1320                  * interface has been set as a multicast option, prefer the
1321                  * address of that interface as our source address.
1322                  */
1323                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1324                     inp->inp_moptions != NULL) {
1325                         struct ip_moptions *imo;
1326                         struct ifnet *ifp;
1327
1328                         imo = inp->inp_moptions;
1329                         if (imo->imo_multicast_ifp != NULL) {
1330                                 ifp = imo->imo_multicast_ifp;
1331                                 IN_IFADDR_RLOCK(&in_ifa_tracker);
1332                                 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1333                                         if ((ia->ia_ifp == ifp) &&
1334                                             (cred == NULL ||
1335                                             prison_check_ip4(cred,
1336                                             &ia->ia_addr.sin_addr) == 0))
1337                                                 break;
1338                                 }
1339                                 if (ia == NULL)
1340                                         error = EADDRNOTAVAIL;
1341                                 else {
1342                                         laddr = ia->ia_addr.sin_addr;
1343                                         error = 0;
1344                                 }
1345                                 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1346                         }
1347                 }
1348                 if (error)
1349                         return (error);
1350         }
1351         oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1352             laddr, lport, 0, NULL);
1353         if (oinp != NULL) {
1354                 if (oinpp != NULL)
1355                         *oinpp = oinp;
1356                 return (EADDRINUSE);
1357         }
1358         if (lport == 0) {
1359                 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1360                     cred);
1361                 if (error)
1362                         return (error);
1363         }
1364         *laddrp = laddr.s_addr;
1365         *lportp = lport;
1366         *faddrp = faddr.s_addr;
1367         *fportp = fport;
1368         return (0);
1369 }
1370
1371 void
1372 in_pcbdisconnect(struct inpcb *inp)
1373 {
1374
1375         INP_WLOCK_ASSERT(inp);
1376         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1377
1378         inp->inp_faddr.s_addr = INADDR_ANY;
1379         inp->inp_fport = 0;
1380         in_pcbrehash(inp);
1381 }
1382 #endif /* INET */
1383
1384 /*
1385  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1386  * For most protocols, this will be invoked immediately prior to calling
1387  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1388  * socket, in which case in_pcbfree() is deferred.
1389  */
1390 void
1391 in_pcbdetach(struct inpcb *inp)
1392 {
1393
1394         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1395
1396 #ifdef RATELIMIT
1397         if (inp->inp_snd_tag != NULL)
1398                 in_pcbdetach_txrtlmt(inp);
1399 #endif
1400         inp->inp_socket->so_pcb = NULL;
1401         inp->inp_socket = NULL;
1402 }
1403
1404 /*
1405  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1406  * stability of an inpcb pointer despite the inpcb lock being released.  This
1407  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1408  * but where the inpcb lock may already held, or when acquiring a reference
1409  * via a pcbgroup.
1410  *
1411  * in_pcbref() should be used only to provide brief memory stability, and
1412  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1413  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1414  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1415  * lock and rele are the *only* safe operations that may be performed on the
1416  * inpcb.
1417  *
1418  * While the inpcb will not be freed, releasing the inpcb lock means that the
1419  * connection's state may change, so the caller should be careful to
1420  * revalidate any cached state on reacquiring the lock.  Drop the reference
1421  * using in_pcbrele().
1422  */
1423 void
1424 in_pcbref(struct inpcb *inp)
1425 {
1426
1427         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1428
1429         refcount_acquire(&inp->inp_refcount);
1430 }
1431
1432 /*
1433  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1434  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1435  * return a flag indicating whether or not the inpcb remains valid.  If it is
1436  * valid, we return with the inpcb lock held.
1437  *
1438  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1439  * reference on an inpcb.  Historically more work was done here (actually, in
1440  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1441  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1442  * about memory stability (and continued use of the write lock).
1443  */
1444 int
1445 in_pcbrele_rlocked(struct inpcb *inp)
1446 {
1447         struct inpcbinfo *pcbinfo;
1448
1449         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1450
1451         INP_RLOCK_ASSERT(inp);
1452
1453         if (refcount_release(&inp->inp_refcount) == 0) {
1454                 /*
1455                  * If the inpcb has been freed, let the caller know, even if
1456                  * this isn't the last reference.
1457                  */
1458                 if (inp->inp_flags2 & INP_FREED) {
1459                         INP_RUNLOCK(inp);
1460                         return (1);
1461                 }
1462                 return (0);
1463         }
1464
1465         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1466 #ifdef TCPHPTS
1467         if (inp->inp_in_hpts || inp->inp_in_input) {
1468                 struct tcp_hpts_entry *hpts;
1469                 /*
1470                  * We should not be on the hpts at
1471                  * this point in any form. we must
1472                  * get the lock to be sure.
1473                  */
1474                 hpts = tcp_hpts_lock(inp);
1475                 if (inp->inp_in_hpts)
1476                         panic("Hpts:%p inp:%p at free still on hpts",
1477                               hpts, inp);
1478                 mtx_unlock(&hpts->p_mtx);
1479                 hpts = tcp_input_lock(inp);
1480                 if (inp->inp_in_input)
1481                         panic("Hpts:%p inp:%p at free still on input hpts",
1482                               hpts, inp);
1483                 mtx_unlock(&hpts->p_mtx);
1484         }
1485 #endif
1486         INP_RUNLOCK(inp);
1487         pcbinfo = inp->inp_pcbinfo;
1488         uma_zfree(pcbinfo->ipi_zone, inp);
1489         return (1);
1490 }
1491
1492 int
1493 in_pcbrele_wlocked(struct inpcb *inp)
1494 {
1495         struct inpcbinfo *pcbinfo;
1496
1497         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1498
1499         INP_WLOCK_ASSERT(inp);
1500
1501         if (refcount_release(&inp->inp_refcount) == 0) {
1502                 /*
1503                  * If the inpcb has been freed, let the caller know, even if
1504                  * this isn't the last reference.
1505                  */
1506                 if (inp->inp_flags2 & INP_FREED) {
1507                         INP_WUNLOCK(inp);
1508                         return (1);
1509                 }
1510                 return (0);
1511         }
1512
1513         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1514 #ifdef TCPHPTS
1515         if (inp->inp_in_hpts || inp->inp_in_input) {
1516                 struct tcp_hpts_entry *hpts;
1517                 /*
1518                  * We should not be on the hpts at
1519                  * this point in any form. we must
1520                  * get the lock to be sure.
1521                  */
1522                 hpts = tcp_hpts_lock(inp);
1523                 if (inp->inp_in_hpts)
1524                         panic("Hpts:%p inp:%p at free still on hpts",
1525                               hpts, inp);
1526                 mtx_unlock(&hpts->p_mtx);
1527                 hpts = tcp_input_lock(inp);
1528                 if (inp->inp_in_input)
1529                         panic("Hpts:%p inp:%p at free still on input hpts",
1530                               hpts, inp);
1531                 mtx_unlock(&hpts->p_mtx);
1532         }
1533 #endif
1534         INP_WUNLOCK(inp);
1535         pcbinfo = inp->inp_pcbinfo;
1536         uma_zfree(pcbinfo->ipi_zone, inp);
1537         return (1);
1538 }
1539
1540 /*
1541  * Temporary wrapper.
1542  */
1543 int
1544 in_pcbrele(struct inpcb *inp)
1545 {
1546
1547         return (in_pcbrele_wlocked(inp));
1548 }
1549
1550 void
1551 in_pcblist_rele_rlocked(epoch_context_t ctx)
1552 {
1553         struct in_pcblist *il;
1554         struct inpcb *inp;
1555         struct inpcbinfo *pcbinfo;
1556         int i, n;
1557
1558         il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1559         pcbinfo = il->il_pcbinfo;
1560         n = il->il_count;
1561         INP_INFO_WLOCK(pcbinfo);
1562         for (i = 0; i < n; i++) {
1563                 inp = il->il_inp_list[i];
1564                 INP_RLOCK(inp);
1565                 if (!in_pcbrele_rlocked(inp))
1566                         INP_RUNLOCK(inp);
1567         }
1568         INP_INFO_WUNLOCK(pcbinfo);
1569         free(il, M_TEMP);
1570 }
1571
1572 /*
1573  * Unconditionally schedule an inpcb to be freed by decrementing its
1574  * reference count, which should occur only after the inpcb has been detached
1575  * from its socket.  If another thread holds a temporary reference (acquired
1576  * using in_pcbref()) then the free is deferred until that reference is
1577  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1578  * work, including removal from global lists, is done in this context, where
1579  * the pcbinfo lock is held.
1580  */
1581 void
1582 in_pcbfree(struct inpcb *inp)
1583 {
1584         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1585
1586 #ifdef INET6
1587         struct ip6_moptions *im6o = NULL;
1588 #endif
1589 #ifdef INET
1590         struct ip_moptions *imo = NULL;
1591 #endif
1592         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1593
1594         KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1595             ("%s: called twice for pcb %p", __func__, inp));
1596         if (inp->inp_flags2 & INP_FREED) {
1597                 INP_WUNLOCK(inp);
1598                 return;
1599         }
1600
1601 #ifdef INVARIANTS
1602         if (pcbinfo == &V_tcbinfo) {
1603                 INP_INFO_LOCK_ASSERT(pcbinfo);
1604         } else {
1605                 INP_INFO_WLOCK_ASSERT(pcbinfo);
1606         }
1607 #endif
1608         INP_WLOCK_ASSERT(inp);
1609
1610 #ifdef INET
1611         imo = inp->inp_moptions;
1612         inp->inp_moptions = NULL;
1613 #endif
1614         /* XXXRW: Do as much as possible here. */
1615 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1616         if (inp->inp_sp != NULL)
1617                 ipsec_delete_pcbpolicy(inp);
1618 #endif
1619         INP_LIST_WLOCK(pcbinfo);
1620         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1621         in_pcbremlists(inp);
1622         INP_LIST_WUNLOCK(pcbinfo);
1623 #ifdef INET6
1624         if (inp->inp_vflag & INP_IPV6PROTO) {
1625                 ip6_freepcbopts(inp->in6p_outputopts);
1626                 im6o = inp->in6p_moptions;
1627                 inp->in6p_moptions = NULL;
1628         }
1629 #endif
1630         if (inp->inp_options)
1631                 (void)m_free(inp->inp_options);
1632         RO_INVALIDATE_CACHE(&inp->inp_route);
1633
1634         inp->inp_vflag = 0;
1635         inp->inp_flags2 |= INP_FREED;
1636         crfree(inp->inp_cred);
1637 #ifdef MAC
1638         mac_inpcb_destroy(inp);
1639 #endif
1640 #ifdef INET6
1641         ip6_freemoptions(im6o);
1642 #endif
1643 #ifdef INET
1644         inp_freemoptions(imo);
1645 #endif
1646         if (!in_pcbrele_wlocked(inp))
1647                 INP_WUNLOCK(inp);
1648 }
1649
1650 /*
1651  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1652  * port reservation, and preventing it from being returned by inpcb lookups.
1653  *
1654  * It is used by TCP to mark an inpcb as unused and avoid future packet
1655  * delivery or event notification when a socket remains open but TCP has
1656  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1657  * or a RST on the wire, and allows the port binding to be reused while still
1658  * maintaining the invariant that so_pcb always points to a valid inpcb until
1659  * in_pcbdetach().
1660  *
1661  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1662  * in_pcbnotifyall() and in_pcbpurgeif0()?
1663  */
1664 void
1665 in_pcbdrop(struct inpcb *inp)
1666 {
1667
1668         INP_WLOCK_ASSERT(inp);
1669
1670         /*
1671          * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1672          * the hash lock...?
1673          */
1674         inp->inp_flags |= INP_DROPPED;
1675         if (inp->inp_flags & INP_INHASHLIST) {
1676                 struct inpcbport *phd = inp->inp_phd;
1677
1678                 INP_HASH_WLOCK(inp->inp_pcbinfo);
1679                 in_pcbremlbgrouphash(inp);
1680                 LIST_REMOVE(inp, inp_hash);
1681                 LIST_REMOVE(inp, inp_portlist);
1682                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1683                         LIST_REMOVE(phd, phd_hash);
1684                         free(phd, M_PCB);
1685                 }
1686                 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1687                 inp->inp_flags &= ~INP_INHASHLIST;
1688 #ifdef PCBGROUP
1689                 in_pcbgroup_remove(inp);
1690 #endif
1691         }
1692 }
1693
1694 #ifdef INET
1695 /*
1696  * Common routines to return the socket addresses associated with inpcbs.
1697  */
1698 struct sockaddr *
1699 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1700 {
1701         struct sockaddr_in *sin;
1702
1703         sin = malloc(sizeof *sin, M_SONAME,
1704                 M_WAITOK | M_ZERO);
1705         sin->sin_family = AF_INET;
1706         sin->sin_len = sizeof(*sin);
1707         sin->sin_addr = *addr_p;
1708         sin->sin_port = port;
1709
1710         return (struct sockaddr *)sin;
1711 }
1712
1713 int
1714 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1715 {
1716         struct inpcb *inp;
1717         struct in_addr addr;
1718         in_port_t port;
1719
1720         inp = sotoinpcb(so);
1721         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1722
1723         INP_RLOCK(inp);
1724         port = inp->inp_lport;
1725         addr = inp->inp_laddr;
1726         INP_RUNLOCK(inp);
1727
1728         *nam = in_sockaddr(port, &addr);
1729         return 0;
1730 }
1731
1732 int
1733 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1734 {
1735         struct inpcb *inp;
1736         struct in_addr addr;
1737         in_port_t port;
1738
1739         inp = sotoinpcb(so);
1740         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1741
1742         INP_RLOCK(inp);
1743         port = inp->inp_fport;
1744         addr = inp->inp_faddr;
1745         INP_RUNLOCK(inp);
1746
1747         *nam = in_sockaddr(port, &addr);
1748         return 0;
1749 }
1750
1751 void
1752 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1753     struct inpcb *(*notify)(struct inpcb *, int))
1754 {
1755         struct inpcb *inp, *inp_temp;
1756
1757         INP_INFO_WLOCK(pcbinfo);
1758         LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1759                 INP_WLOCK(inp);
1760 #ifdef INET6
1761                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1762                         INP_WUNLOCK(inp);
1763                         continue;
1764                 }
1765 #endif
1766                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1767                     inp->inp_socket == NULL) {
1768                         INP_WUNLOCK(inp);
1769                         continue;
1770                 }
1771                 if ((*notify)(inp, errno))
1772                         INP_WUNLOCK(inp);
1773         }
1774         INP_INFO_WUNLOCK(pcbinfo);
1775 }
1776
1777 void
1778 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1779 {
1780         struct inpcb *inp;
1781         struct ip_moptions *imo;
1782         int i, gap;
1783
1784         INP_INFO_WLOCK(pcbinfo);
1785         LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1786                 INP_WLOCK(inp);
1787                 imo = inp->inp_moptions;
1788                 if ((inp->inp_vflag & INP_IPV4) &&
1789                     imo != NULL) {
1790                         /*
1791                          * Unselect the outgoing interface if it is being
1792                          * detached.
1793                          */
1794                         if (imo->imo_multicast_ifp == ifp)
1795                                 imo->imo_multicast_ifp = NULL;
1796
1797                         /*
1798                          * Drop multicast group membership if we joined
1799                          * through the interface being detached.
1800                          *
1801                          * XXX This can all be deferred to an epoch_call
1802                          */
1803                         for (i = 0, gap = 0; i < imo->imo_num_memberships;
1804                             i++) {
1805                                 if (imo->imo_membership[i]->inm_ifp == ifp) {
1806                                         IN_MULTI_LOCK_ASSERT();
1807                                         in_leavegroup_locked(imo->imo_membership[i], NULL);
1808                                         gap++;
1809                                 } else if (gap != 0)
1810                                         imo->imo_membership[i - gap] =
1811                                             imo->imo_membership[i];
1812                         }
1813                         imo->imo_num_memberships -= gap;
1814                 }
1815                 INP_WUNLOCK(inp);
1816         }
1817         INP_INFO_WUNLOCK(pcbinfo);
1818 }
1819
1820 /*
1821  * Lookup a PCB based on the local address and port.  Caller must hold the
1822  * hash lock.  No inpcb locks or references are acquired.
1823  */
1824 #define INP_LOOKUP_MAPPED_PCB_COST      3
1825 struct inpcb *
1826 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1827     u_short lport, int lookupflags, struct ucred *cred)
1828 {
1829         struct inpcb *inp;
1830 #ifdef INET6
1831         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1832 #else
1833         int matchwild = 3;
1834 #endif
1835         int wildcard;
1836
1837         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1838             ("%s: invalid lookup flags %d", __func__, lookupflags));
1839
1840         INP_HASH_LOCK_ASSERT(pcbinfo);
1841
1842         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1843                 struct inpcbhead *head;
1844                 /*
1845                  * Look for an unconnected (wildcard foreign addr) PCB that
1846                  * matches the local address and port we're looking for.
1847                  */
1848                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1849                     0, pcbinfo->ipi_hashmask)];
1850                 LIST_FOREACH(inp, head, inp_hash) {
1851 #ifdef INET6
1852                         /* XXX inp locking */
1853                         if ((inp->inp_vflag & INP_IPV4) == 0)
1854                                 continue;
1855 #endif
1856                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
1857                             inp->inp_laddr.s_addr == laddr.s_addr &&
1858                             inp->inp_lport == lport) {
1859                                 /*
1860                                  * Found?
1861                                  */
1862                                 if (cred == NULL ||
1863                                     prison_equal_ip4(cred->cr_prison,
1864                                         inp->inp_cred->cr_prison))
1865                                         return (inp);
1866                         }
1867                 }
1868                 /*
1869                  * Not found.
1870                  */
1871                 return (NULL);
1872         } else {
1873                 struct inpcbporthead *porthash;
1874                 struct inpcbport *phd;
1875                 struct inpcb *match = NULL;
1876                 /*
1877                  * Best fit PCB lookup.
1878                  *
1879                  * First see if this local port is in use by looking on the
1880                  * port hash list.
1881                  */
1882                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1883                     pcbinfo->ipi_porthashmask)];
1884                 LIST_FOREACH(phd, porthash, phd_hash) {
1885                         if (phd->phd_port == lport)
1886                                 break;
1887                 }
1888                 if (phd != NULL) {
1889                         /*
1890                          * Port is in use by one or more PCBs. Look for best
1891                          * fit.
1892                          */
1893                         LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1894                                 wildcard = 0;
1895                                 if (cred != NULL &&
1896                                     !prison_equal_ip4(inp->inp_cred->cr_prison,
1897                                         cred->cr_prison))
1898                                         continue;
1899 #ifdef INET6
1900                                 /* XXX inp locking */
1901                                 if ((inp->inp_vflag & INP_IPV4) == 0)
1902                                         continue;
1903                                 /*
1904                                  * We never select the PCB that has
1905                                  * INP_IPV6 flag and is bound to :: if
1906                                  * we have another PCB which is bound
1907                                  * to 0.0.0.0.  If a PCB has the
1908                                  * INP_IPV6 flag, then we set its cost
1909                                  * higher than IPv4 only PCBs.
1910                                  *
1911                                  * Note that the case only happens
1912                                  * when a socket is bound to ::, under
1913                                  * the condition that the use of the
1914                                  * mapped address is allowed.
1915                                  */
1916                                 if ((inp->inp_vflag & INP_IPV6) != 0)
1917                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1918 #endif
1919                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
1920                                         wildcard++;
1921                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1922                                         if (laddr.s_addr == INADDR_ANY)
1923                                                 wildcard++;
1924                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
1925                                                 continue;
1926                                 } else {
1927                                         if (laddr.s_addr != INADDR_ANY)
1928                                                 wildcard++;
1929                                 }
1930                                 if (wildcard < matchwild) {
1931                                         match = inp;
1932                                         matchwild = wildcard;
1933                                         if (matchwild == 0)
1934                                                 break;
1935                                 }
1936                         }
1937                 }
1938                 return (match);
1939         }
1940 }
1941 #undef INP_LOOKUP_MAPPED_PCB_COST
1942
1943 static struct inpcb *
1944 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
1945   const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
1946   uint16_t fport, int lookupflags)
1947 {
1948         struct inpcb *local_wild = NULL;
1949         const struct inpcblbgrouphead *hdr;
1950         struct inpcblbgroup *grp;
1951         struct inpcblbgroup *grp_local_wild;
1952
1953         INP_HASH_LOCK_ASSERT(pcbinfo);
1954
1955         hdr = &pcbinfo->ipi_lbgrouphashbase[
1956                   INP_PCBLBGROUP_PORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
1957
1958         /*
1959          * Order of socket selection:
1960          * 1. non-wild.
1961          * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
1962          *
1963          * NOTE:
1964          * - Load balanced group does not contain jailed sockets
1965          * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
1966          */
1967         LIST_FOREACH(grp, hdr, il_list) {
1968 #ifdef INET6
1969                 if (!(grp->il_vflag & INP_IPV4))
1970                         continue;
1971 #endif
1972
1973                 if (grp->il_lport == lport) {
1974
1975                         uint32_t idx = 0;
1976                         int pkt_hash = INP_PCBLBGROUP_PKTHASH(faddr->s_addr,
1977                             lport, fport);
1978
1979                         idx = pkt_hash % grp->il_inpcnt;
1980
1981                         if (grp->il_laddr.s_addr == laddr->s_addr) {
1982                                 return (grp->il_inp[idx]);
1983                         } else {
1984                                 if (grp->il_laddr.s_addr == INADDR_ANY &&
1985                                         (lookupflags & INPLOOKUP_WILDCARD)) {
1986                                         local_wild = grp->il_inp[idx];
1987                                         grp_local_wild = grp;
1988                                 }
1989                         }
1990                 }
1991         }
1992         if (local_wild != NULL) {
1993                 return (local_wild);
1994         }
1995         return (NULL);
1996 }
1997
1998 #ifdef PCBGROUP
1999 /*
2000  * Lookup PCB in hash list, using pcbgroup tables.
2001  */
2002 static struct inpcb *
2003 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2004     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2005     u_int lport_arg, int lookupflags, struct ifnet *ifp)
2006 {
2007         struct inpcbhead *head;
2008         struct inpcb *inp, *tmpinp;
2009         u_short fport = fport_arg, lport = lport_arg;
2010         bool locked;
2011
2012         /*
2013          * First look for an exact match.
2014          */
2015         tmpinp = NULL;
2016         INP_GROUP_LOCK(pcbgroup);
2017         head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2018             pcbgroup->ipg_hashmask)];
2019         LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2020 #ifdef INET6
2021                 /* XXX inp locking */
2022                 if ((inp->inp_vflag & INP_IPV4) == 0)
2023                         continue;
2024 #endif
2025                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2026                     inp->inp_laddr.s_addr == laddr.s_addr &&
2027                     inp->inp_fport == fport &&
2028                     inp->inp_lport == lport) {
2029                         /*
2030                          * XXX We should be able to directly return
2031                          * the inp here, without any checks.
2032                          * Well unless both bound with SO_REUSEPORT?
2033                          */
2034                         if (prison_flag(inp->inp_cred, PR_IP4))
2035                                 goto found;
2036                         if (tmpinp == NULL)
2037                                 tmpinp = inp;
2038                 }
2039         }
2040         if (tmpinp != NULL) {
2041                 inp = tmpinp;
2042                 goto found;
2043         }
2044
2045 #ifdef  RSS
2046         /*
2047          * For incoming connections, we may wish to do a wildcard
2048          * match for an RSS-local socket.
2049          */
2050         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2051                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2052 #ifdef INET6
2053                 struct inpcb *local_wild_mapped = NULL;
2054 #endif
2055                 struct inpcb *jail_wild = NULL;
2056                 struct inpcbhead *head;
2057                 int injail;
2058
2059                 /*
2060                  * Order of socket selection - we always prefer jails.
2061                  *      1. jailed, non-wild.
2062                  *      2. jailed, wild.
2063                  *      3. non-jailed, non-wild.
2064                  *      4. non-jailed, wild.
2065                  */
2066
2067                 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2068                     lport, 0, pcbgroup->ipg_hashmask)];
2069                 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2070 #ifdef INET6
2071                         /* XXX inp locking */
2072                         if ((inp->inp_vflag & INP_IPV4) == 0)
2073                                 continue;
2074 #endif
2075                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2076                             inp->inp_lport != lport)
2077                                 continue;
2078
2079                         injail = prison_flag(inp->inp_cred, PR_IP4);
2080                         if (injail) {
2081                                 if (prison_check_ip4(inp->inp_cred,
2082                                     &laddr) != 0)
2083                                         continue;
2084                         } else {
2085                                 if (local_exact != NULL)
2086                                         continue;
2087                         }
2088
2089                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2090                                 if (injail)
2091                                         goto found;
2092                                 else
2093                                         local_exact = inp;
2094                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2095 #ifdef INET6
2096                                 /* XXX inp locking, NULL check */
2097                                 if (inp->inp_vflag & INP_IPV6PROTO)
2098                                         local_wild_mapped = inp;
2099                                 else
2100 #endif
2101                                         if (injail)
2102                                                 jail_wild = inp;
2103                                         else
2104                                                 local_wild = inp;
2105                         }
2106                 } /* LIST_FOREACH */
2107
2108                 inp = jail_wild;
2109                 if (inp == NULL)
2110                         inp = local_exact;
2111                 if (inp == NULL)
2112                         inp = local_wild;
2113 #ifdef INET6
2114                 if (inp == NULL)
2115                         inp = local_wild_mapped;
2116 #endif
2117                 if (inp != NULL)
2118                         goto found;
2119         }
2120 #endif
2121
2122         /*
2123          * Then look for a wildcard match, if requested.
2124          */
2125         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2126                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2127 #ifdef INET6
2128                 struct inpcb *local_wild_mapped = NULL;
2129 #endif
2130                 struct inpcb *jail_wild = NULL;
2131                 struct inpcbhead *head;
2132                 int injail;
2133
2134                 /*
2135                  * Order of socket selection - we always prefer jails.
2136                  *      1. jailed, non-wild.
2137                  *      2. jailed, wild.
2138                  *      3. non-jailed, non-wild.
2139                  *      4. non-jailed, wild.
2140                  */
2141                 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2142                     0, pcbinfo->ipi_wildmask)];
2143                 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2144 #ifdef INET6
2145                         /* XXX inp locking */
2146                         if ((inp->inp_vflag & INP_IPV4) == 0)
2147                                 continue;
2148 #endif
2149                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2150                             inp->inp_lport != lport)
2151                                 continue;
2152
2153                         injail = prison_flag(inp->inp_cred, PR_IP4);
2154                         if (injail) {
2155                                 if (prison_check_ip4(inp->inp_cred,
2156                                     &laddr) != 0)
2157                                         continue;
2158                         } else {
2159                                 if (local_exact != NULL)
2160                                         continue;
2161                         }
2162
2163                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2164                                 if (injail)
2165                                         goto found;
2166                                 else
2167                                         local_exact = inp;
2168                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2169 #ifdef INET6
2170                                 /* XXX inp locking, NULL check */
2171                                 if (inp->inp_vflag & INP_IPV6PROTO)
2172                                         local_wild_mapped = inp;
2173                                 else
2174 #endif
2175                                         if (injail)
2176                                                 jail_wild = inp;
2177                                         else
2178                                                 local_wild = inp;
2179                         }
2180                 } /* LIST_FOREACH */
2181                 inp = jail_wild;
2182                 if (inp == NULL)
2183                         inp = local_exact;
2184                 if (inp == NULL)
2185                         inp = local_wild;
2186 #ifdef INET6
2187                 if (inp == NULL)
2188                         inp = local_wild_mapped;
2189 #endif
2190                 if (inp != NULL)
2191                         goto found;
2192         } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2193         INP_GROUP_UNLOCK(pcbgroup);
2194         return (NULL);
2195
2196 found:
2197         if (lookupflags & INPLOOKUP_WLOCKPCB)
2198                 locked = INP_TRY_WLOCK(inp);
2199         else if (lookupflags & INPLOOKUP_RLOCKPCB)
2200                 locked = INP_TRY_RLOCK(inp);
2201         else
2202                 panic("%s: locking bug", __func__);
2203         if (!locked)
2204                 in_pcbref(inp);
2205         INP_GROUP_UNLOCK(pcbgroup);
2206         if (!locked) {
2207                 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2208                         INP_WLOCK(inp);
2209                         if (in_pcbrele_wlocked(inp))
2210                                 return (NULL);
2211                 } else {
2212                         INP_RLOCK(inp);
2213                         if (in_pcbrele_rlocked(inp))
2214                                 return (NULL);
2215                 }
2216         }
2217 #ifdef INVARIANTS
2218         if (lookupflags & INPLOOKUP_WLOCKPCB)
2219                 INP_WLOCK_ASSERT(inp);
2220         else
2221                 INP_RLOCK_ASSERT(inp);
2222 #endif
2223         return (inp);
2224 }
2225 #endif /* PCBGROUP */
2226
2227 /*
2228  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2229  * that the caller has locked the hash list, and will not perform any further
2230  * locking or reference operations on either the hash list or the connection.
2231  */
2232 static struct inpcb *
2233 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2234     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2235     struct ifnet *ifp)
2236 {
2237         struct inpcbhead *head;
2238         struct inpcb *inp, *tmpinp;
2239         u_short fport = fport_arg, lport = lport_arg;
2240
2241         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2242             ("%s: invalid lookup flags %d", __func__, lookupflags));
2243
2244         INP_HASH_LOCK_ASSERT(pcbinfo);
2245
2246         /*
2247          * First look for an exact match.
2248          */
2249         tmpinp = NULL;
2250         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2251             pcbinfo->ipi_hashmask)];
2252         LIST_FOREACH(inp, head, inp_hash) {
2253 #ifdef INET6
2254                 /* XXX inp locking */
2255                 if ((inp->inp_vflag & INP_IPV4) == 0)
2256                         continue;
2257 #endif
2258                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2259                     inp->inp_laddr.s_addr == laddr.s_addr &&
2260                     inp->inp_fport == fport &&
2261                     inp->inp_lport == lport) {
2262                         /*
2263                          * XXX We should be able to directly return
2264                          * the inp here, without any checks.
2265                          * Well unless both bound with SO_REUSEPORT?
2266                          */
2267                         if (prison_flag(inp->inp_cred, PR_IP4))
2268                                 return (inp);
2269                         if (tmpinp == NULL)
2270                                 tmpinp = inp;
2271                 }
2272         }
2273         if (tmpinp != NULL)
2274                 return (tmpinp);
2275
2276         /*
2277          * Then look in lb group (for wildcard match).
2278          */
2279         if (pcbinfo->ipi_lbgrouphashbase != NULL &&
2280                 (lookupflags & INPLOOKUP_WILDCARD)) {
2281                 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2282                     fport, lookupflags);
2283                 if (inp != NULL) {
2284                         return (inp);
2285                 }
2286         }
2287
2288         /*
2289          * Then look for a wildcard match, if requested.
2290          */
2291         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2292                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2293 #ifdef INET6
2294                 struct inpcb *local_wild_mapped = NULL;
2295 #endif
2296                 struct inpcb *jail_wild = NULL;
2297                 int injail;
2298
2299                 /*
2300                  * Order of socket selection - we always prefer jails.
2301                  *      1. jailed, non-wild.
2302                  *      2. jailed, wild.
2303                  *      3. non-jailed, non-wild.
2304                  *      4. non-jailed, wild.
2305                  */
2306
2307                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2308                     0, pcbinfo->ipi_hashmask)];
2309                 LIST_FOREACH(inp, head, inp_hash) {
2310 #ifdef INET6
2311                         /* XXX inp locking */
2312                         if ((inp->inp_vflag & INP_IPV4) == 0)
2313                                 continue;
2314 #endif
2315                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2316                             inp->inp_lport != lport)
2317                                 continue;
2318
2319                         injail = prison_flag(inp->inp_cred, PR_IP4);
2320                         if (injail) {
2321                                 if (prison_check_ip4(inp->inp_cred,
2322                                     &laddr) != 0)
2323                                         continue;
2324                         } else {
2325                                 if (local_exact != NULL)
2326                                         continue;
2327                         }
2328
2329                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2330                                 if (injail)
2331                                         return (inp);
2332                                 else
2333                                         local_exact = inp;
2334                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2335 #ifdef INET6
2336                                 /* XXX inp locking, NULL check */
2337                                 if (inp->inp_vflag & INP_IPV6PROTO)
2338                                         local_wild_mapped = inp;
2339                                 else
2340 #endif
2341                                         if (injail)
2342                                                 jail_wild = inp;
2343                                         else
2344                                                 local_wild = inp;
2345                         }
2346                 } /* LIST_FOREACH */
2347                 if (jail_wild != NULL)
2348                         return (jail_wild);
2349                 if (local_exact != NULL)
2350                         return (local_exact);
2351                 if (local_wild != NULL)
2352                         return (local_wild);
2353 #ifdef INET6
2354                 if (local_wild_mapped != NULL)
2355                         return (local_wild_mapped);
2356 #endif
2357         } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2358
2359         return (NULL);
2360 }
2361
2362 /*
2363  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2364  * hash list lock, and will return the inpcb locked (i.e., requires
2365  * INPLOOKUP_LOCKPCB).
2366  */
2367 static struct inpcb *
2368 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2369     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2370     struct ifnet *ifp)
2371 {
2372         struct inpcb *inp;
2373         bool locked;
2374
2375         INP_HASH_RLOCK(pcbinfo);
2376         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2377             (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2378         if (inp != NULL) {
2379                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2380                         locked = INP_TRY_WLOCK(inp);
2381                 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2382                         locked = INP_TRY_RLOCK(inp);
2383                 else
2384                         panic("%s: locking bug", __func__);
2385                 if (!locked)
2386                         in_pcbref(inp);
2387                 INP_HASH_RUNLOCK(pcbinfo);
2388                 if (!locked) {
2389                         if (lookupflags & INPLOOKUP_WLOCKPCB) {
2390                                 INP_WLOCK(inp);
2391                                 if (in_pcbrele_wlocked(inp))
2392                                         return (NULL);
2393                         } else {
2394                                 INP_RLOCK(inp);
2395                                 if (in_pcbrele_rlocked(inp))
2396                                         return (NULL);
2397                         }
2398                 }
2399 #ifdef INVARIANTS
2400                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2401                         INP_WLOCK_ASSERT(inp);
2402                 else
2403                         INP_RLOCK_ASSERT(inp);
2404 #endif
2405         } else
2406                 INP_HASH_RUNLOCK(pcbinfo);
2407         return (inp);
2408 }
2409
2410 /*
2411  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2412  * from which a pre-calculated hash value may be extracted.
2413  *
2414  * Possibly more of this logic should be in in_pcbgroup.c.
2415  */
2416 struct inpcb *
2417 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2418     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2419 {
2420 #if defined(PCBGROUP) && !defined(RSS)
2421         struct inpcbgroup *pcbgroup;
2422 #endif
2423
2424         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2425             ("%s: invalid lookup flags %d", __func__, lookupflags));
2426         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2427             ("%s: LOCKPCB not set", __func__));
2428
2429         /*
2430          * When not using RSS, use connection groups in preference to the
2431          * reservation table when looking up 4-tuples.  When using RSS, just
2432          * use the reservation table, due to the cost of the Toeplitz hash
2433          * in software.
2434          *
2435          * XXXRW: This policy belongs in the pcbgroup code, as in principle
2436          * we could be doing RSS with a non-Toeplitz hash that is affordable
2437          * in software.
2438          */
2439 #if defined(PCBGROUP) && !defined(RSS)
2440         if (in_pcbgroup_enabled(pcbinfo)) {
2441                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2442                     fport);
2443                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2444                     laddr, lport, lookupflags, ifp));
2445         }
2446 #endif
2447         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2448             lookupflags, ifp));
2449 }
2450
2451 struct inpcb *
2452 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2453     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2454     struct ifnet *ifp, struct mbuf *m)
2455 {
2456 #ifdef PCBGROUP
2457         struct inpcbgroup *pcbgroup;
2458 #endif
2459
2460         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2461             ("%s: invalid lookup flags %d", __func__, lookupflags));
2462         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2463             ("%s: LOCKPCB not set", __func__));
2464
2465 #ifdef PCBGROUP
2466         /*
2467          * If we can use a hardware-generated hash to look up the connection
2468          * group, use that connection group to find the inpcb.  Otherwise
2469          * fall back on a software hash -- or the reservation table if we're
2470          * using RSS.
2471          *
2472          * XXXRW: As above, that policy belongs in the pcbgroup code.
2473          */
2474         if (in_pcbgroup_enabled(pcbinfo) &&
2475             !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2476                 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2477                     m->m_pkthdr.flowid);
2478                 if (pcbgroup != NULL)
2479                         return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2480                             fport, laddr, lport, lookupflags, ifp));
2481 #ifndef RSS
2482                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2483                     fport);
2484                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2485                     laddr, lport, lookupflags, ifp));
2486 #endif
2487         }
2488 #endif
2489         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2490             lookupflags, ifp));
2491 }
2492 #endif /* INET */
2493
2494 /*
2495  * Insert PCB onto various hash lists.
2496  */
2497 static int
2498 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2499 {
2500         struct inpcbhead *pcbhash;
2501         struct inpcbporthead *pcbporthash;
2502         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2503         struct inpcbport *phd;
2504         u_int32_t hashkey_faddr;
2505         int so_options;
2506
2507         INP_WLOCK_ASSERT(inp);
2508         INP_HASH_WLOCK_ASSERT(pcbinfo);
2509
2510         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2511             ("in_pcbinshash: INP_INHASHLIST"));
2512
2513 #ifdef INET6
2514         if (inp->inp_vflag & INP_IPV6)
2515                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2516         else
2517 #endif
2518         hashkey_faddr = inp->inp_faddr.s_addr;
2519
2520         pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2521                  inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2522
2523         pcbporthash = &pcbinfo->ipi_porthashbase[
2524             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2525
2526         /*
2527          * Add entry to load balance group.
2528          * Only do this if SO_REUSEPORT_LB is set.
2529          */
2530         so_options = inp_so_options(inp);
2531         if (so_options & SO_REUSEPORT_LB) {
2532                 int ret = in_pcbinslbgrouphash(inp);
2533                 if (ret) {
2534                         /* pcb lb group malloc fail (ret=ENOBUFS). */
2535                         return (ret);
2536                 }
2537         }
2538
2539         /*
2540          * Go through port list and look for a head for this lport.
2541          */
2542         LIST_FOREACH(phd, pcbporthash, phd_hash) {
2543                 if (phd->phd_port == inp->inp_lport)
2544                         break;
2545         }
2546         /*
2547          * If none exists, malloc one and tack it on.
2548          */
2549         if (phd == NULL) {
2550                 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2551                 if (phd == NULL) {
2552                         return (ENOBUFS); /* XXX */
2553                 }
2554                 phd->phd_port = inp->inp_lport;
2555                 LIST_INIT(&phd->phd_pcblist);
2556                 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2557         }
2558         inp->inp_phd = phd;
2559         LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2560         LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2561         inp->inp_flags |= INP_INHASHLIST;
2562 #ifdef PCBGROUP
2563         if (do_pcbgroup_update)
2564                 in_pcbgroup_update(inp);
2565 #endif
2566         return (0);
2567 }
2568
2569 /*
2570  * For now, there are two public interfaces to insert an inpcb into the hash
2571  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2572  * is used only in the TCP syncache, where in_pcbinshash is called before the
2573  * full 4-tuple is set for the inpcb, and we don't want to install in the
2574  * pcbgroup until later.
2575  *
2576  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2577  * connection groups, and partially initialised inpcbs should not be exposed
2578  * to either reservation hash tables or pcbgroups.
2579  */
2580 int
2581 in_pcbinshash(struct inpcb *inp)
2582 {
2583
2584         return (in_pcbinshash_internal(inp, 1));
2585 }
2586
2587 int
2588 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2589 {
2590
2591         return (in_pcbinshash_internal(inp, 0));
2592 }
2593
2594 /*
2595  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2596  * changed. NOTE: This does not handle the case of the lport changing (the
2597  * hashed port list would have to be updated as well), so the lport must
2598  * not change after in_pcbinshash() has been called.
2599  */
2600 void
2601 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2602 {
2603         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2604         struct inpcbhead *head;
2605         u_int32_t hashkey_faddr;
2606
2607         INP_WLOCK_ASSERT(inp);
2608         INP_HASH_WLOCK_ASSERT(pcbinfo);
2609
2610         KASSERT(inp->inp_flags & INP_INHASHLIST,
2611             ("in_pcbrehash: !INP_INHASHLIST"));
2612
2613 #ifdef INET6
2614         if (inp->inp_vflag & INP_IPV6)
2615                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2616         else
2617 #endif
2618         hashkey_faddr = inp->inp_faddr.s_addr;
2619
2620         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2621                 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2622
2623         LIST_REMOVE(inp, inp_hash);
2624         LIST_INSERT_HEAD(head, inp, inp_hash);
2625
2626 #ifdef PCBGROUP
2627         if (m != NULL)
2628                 in_pcbgroup_update_mbuf(inp, m);
2629         else
2630                 in_pcbgroup_update(inp);
2631 #endif
2632 }
2633
2634 void
2635 in_pcbrehash(struct inpcb *inp)
2636 {
2637
2638         in_pcbrehash_mbuf(inp, NULL);
2639 }
2640
2641 /*
2642  * Remove PCB from various lists.
2643  */
2644 static void
2645 in_pcbremlists(struct inpcb *inp)
2646 {
2647         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2648
2649 #ifdef INVARIANTS
2650         if (pcbinfo == &V_tcbinfo) {
2651                 INP_INFO_RLOCK_ASSERT(pcbinfo);
2652         } else {
2653                 INP_INFO_WLOCK_ASSERT(pcbinfo);
2654         }
2655 #endif
2656
2657         INP_WLOCK_ASSERT(inp);
2658         INP_LIST_WLOCK_ASSERT(pcbinfo);
2659
2660         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2661         if (inp->inp_flags & INP_INHASHLIST) {
2662                 struct inpcbport *phd = inp->inp_phd;
2663
2664                 INP_HASH_WLOCK(pcbinfo);
2665
2666                 /* XXX: Only do if SO_REUSEPORT_LB set? */
2667                 in_pcbremlbgrouphash(inp);
2668
2669                 LIST_REMOVE(inp, inp_hash);
2670                 LIST_REMOVE(inp, inp_portlist);
2671                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2672                         LIST_REMOVE(phd, phd_hash);
2673                         free(phd, M_PCB);
2674                 }
2675                 INP_HASH_WUNLOCK(pcbinfo);
2676                 inp->inp_flags &= ~INP_INHASHLIST;
2677         }
2678         LIST_REMOVE(inp, inp_list);
2679         pcbinfo->ipi_count--;
2680 #ifdef PCBGROUP
2681         in_pcbgroup_remove(inp);
2682 #endif
2683 }
2684
2685 /*
2686  * Check for alternatives when higher level complains
2687  * about service problems.  For now, invalidate cached
2688  * routing information.  If the route was created dynamically
2689  * (by a redirect), time to try a default gateway again.
2690  */
2691 void
2692 in_losing(struct inpcb *inp)
2693 {
2694
2695         RO_INVALIDATE_CACHE(&inp->inp_route);
2696         return;
2697 }
2698
2699 /*
2700  * A set label operation has occurred at the socket layer, propagate the
2701  * label change into the in_pcb for the socket.
2702  */
2703 void
2704 in_pcbsosetlabel(struct socket *so)
2705 {
2706 #ifdef MAC
2707         struct inpcb *inp;
2708
2709         inp = sotoinpcb(so);
2710         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2711
2712         INP_WLOCK(inp);
2713         SOCK_LOCK(so);
2714         mac_inpcb_sosetlabel(so, inp);
2715         SOCK_UNLOCK(so);
2716         INP_WUNLOCK(inp);
2717 #endif
2718 }
2719
2720 /*
2721  * ipport_tick runs once per second, determining if random port allocation
2722  * should be continued.  If more than ipport_randomcps ports have been
2723  * allocated in the last second, then we return to sequential port
2724  * allocation. We return to random allocation only once we drop below
2725  * ipport_randomcps for at least ipport_randomtime seconds.
2726  */
2727 static void
2728 ipport_tick(void *xtp)
2729 {
2730         VNET_ITERATOR_DECL(vnet_iter);
2731
2732         VNET_LIST_RLOCK_NOSLEEP();
2733         VNET_FOREACH(vnet_iter) {
2734                 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2735                 if (V_ipport_tcpallocs <=
2736                     V_ipport_tcplastcount + V_ipport_randomcps) {
2737                         if (V_ipport_stoprandom > 0)
2738                                 V_ipport_stoprandom--;
2739                 } else
2740                         V_ipport_stoprandom = V_ipport_randomtime;
2741                 V_ipport_tcplastcount = V_ipport_tcpallocs;
2742                 CURVNET_RESTORE();
2743         }
2744         VNET_LIST_RUNLOCK_NOSLEEP();
2745         callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2746 }
2747
2748 static void
2749 ip_fini(void *xtp)
2750 {
2751
2752         callout_stop(&ipport_tick_callout);
2753 }
2754
2755 /*
2756  * The ipport_callout should start running at about the time we attach the
2757  * inet or inet6 domains.
2758  */
2759 static void
2760 ipport_tick_init(const void *unused __unused)
2761 {
2762
2763         /* Start ipport_tick. */
2764         callout_init(&ipport_tick_callout, 1);
2765         callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2766         EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2767                 SHUTDOWN_PRI_DEFAULT);
2768 }
2769 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2770     ipport_tick_init, NULL);
2771
2772 void
2773 inp_wlock(struct inpcb *inp)
2774 {
2775
2776         INP_WLOCK(inp);
2777 }
2778
2779 void
2780 inp_wunlock(struct inpcb *inp)
2781 {
2782
2783         INP_WUNLOCK(inp);
2784 }
2785
2786 void
2787 inp_rlock(struct inpcb *inp)
2788 {
2789
2790         INP_RLOCK(inp);
2791 }
2792
2793 void
2794 inp_runlock(struct inpcb *inp)
2795 {
2796
2797         INP_RUNLOCK(inp);
2798 }
2799
2800 #ifdef INVARIANT_SUPPORT
2801 void
2802 inp_lock_assert(struct inpcb *inp)
2803 {
2804
2805         INP_WLOCK_ASSERT(inp);
2806 }
2807
2808 void
2809 inp_unlock_assert(struct inpcb *inp)
2810 {
2811
2812         INP_UNLOCK_ASSERT(inp);
2813 }
2814 #endif
2815
2816 void
2817 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2818 {
2819         struct inpcb *inp;
2820
2821         INP_INFO_WLOCK(&V_tcbinfo);
2822         LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2823                 INP_WLOCK(inp);
2824                 func(inp, arg);
2825                 INP_WUNLOCK(inp);
2826         }
2827         INP_INFO_WUNLOCK(&V_tcbinfo);
2828 }
2829
2830 struct socket *
2831 inp_inpcbtosocket(struct inpcb *inp)
2832 {
2833
2834         INP_WLOCK_ASSERT(inp);
2835         return (inp->inp_socket);
2836 }
2837
2838 struct tcpcb *
2839 inp_inpcbtotcpcb(struct inpcb *inp)
2840 {
2841
2842         INP_WLOCK_ASSERT(inp);
2843         return ((struct tcpcb *)inp->inp_ppcb);
2844 }
2845
2846 int
2847 inp_ip_tos_get(const struct inpcb *inp)
2848 {
2849
2850         return (inp->inp_ip_tos);
2851 }
2852
2853 void
2854 inp_ip_tos_set(struct inpcb *inp, int val)
2855 {
2856
2857         inp->inp_ip_tos = val;
2858 }
2859
2860 void
2861 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2862     uint32_t *faddr, uint16_t *fp)
2863 {
2864
2865         INP_LOCK_ASSERT(inp);
2866         *laddr = inp->inp_laddr.s_addr;
2867         *faddr = inp->inp_faddr.s_addr;
2868         *lp = inp->inp_lport;
2869         *fp = inp->inp_fport;
2870 }
2871
2872 struct inpcb *
2873 so_sotoinpcb(struct socket *so)
2874 {
2875
2876         return (sotoinpcb(so));
2877 }
2878
2879 struct tcpcb *
2880 so_sototcpcb(struct socket *so)
2881 {
2882
2883         return (sototcpcb(so));
2884 }
2885
2886 /*
2887  * Create an external-format (``xinpcb'') structure using the information in
2888  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2889  * reduce the spew of irrelevant information over this interface, to isolate
2890  * user code from changes in the kernel structure, and potentially to provide
2891  * information-hiding if we decide that some of this information should be
2892  * hidden from users.
2893  */
2894 void
2895 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2896 {
2897
2898         xi->xi_len = sizeof(struct xinpcb);
2899         if (inp->inp_socket)
2900                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2901         else
2902                 bzero(&xi->xi_socket, sizeof(struct xsocket));
2903         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2904         xi->inp_gencnt = inp->inp_gencnt;
2905         xi->inp_ppcb = inp->inp_ppcb;
2906         xi->inp_flow = inp->inp_flow;
2907         xi->inp_flowid = inp->inp_flowid;
2908         xi->inp_flowtype = inp->inp_flowtype;
2909         xi->inp_flags = inp->inp_flags;
2910         xi->inp_flags2 = inp->inp_flags2;
2911         xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2912         xi->in6p_cksum = inp->in6p_cksum;
2913         xi->in6p_hops = inp->in6p_hops;
2914         xi->inp_ip_tos = inp->inp_ip_tos;
2915         xi->inp_vflag = inp->inp_vflag;
2916         xi->inp_ip_ttl = inp->inp_ip_ttl;
2917         xi->inp_ip_p = inp->inp_ip_p;
2918         xi->inp_ip_minttl = inp->inp_ip_minttl;
2919 }
2920
2921 #ifdef DDB
2922 static void
2923 db_print_indent(int indent)
2924 {
2925         int i;
2926
2927         for (i = 0; i < indent; i++)
2928                 db_printf(" ");
2929 }
2930
2931 static void
2932 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2933 {
2934         char faddr_str[48], laddr_str[48];
2935
2936         db_print_indent(indent);
2937         db_printf("%s at %p\n", name, inc);
2938
2939         indent += 2;
2940
2941 #ifdef INET6
2942         if (inc->inc_flags & INC_ISIPV6) {
2943                 /* IPv6. */
2944                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2945                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2946         } else
2947 #endif
2948         {
2949                 /* IPv4. */
2950                 inet_ntoa_r(inc->inc_laddr, laddr_str);
2951                 inet_ntoa_r(inc->inc_faddr, faddr_str);
2952         }
2953         db_print_indent(indent);
2954         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2955             ntohs(inc->inc_lport));
2956         db_print_indent(indent);
2957         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2958             ntohs(inc->inc_fport));
2959 }
2960
2961 static void
2962 db_print_inpflags(int inp_flags)
2963 {
2964         int comma;
2965
2966         comma = 0;
2967         if (inp_flags & INP_RECVOPTS) {
2968                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2969                 comma = 1;
2970         }
2971         if (inp_flags & INP_RECVRETOPTS) {
2972                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2973                 comma = 1;
2974         }
2975         if (inp_flags & INP_RECVDSTADDR) {
2976                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2977                 comma = 1;
2978         }
2979         if (inp_flags & INP_ORIGDSTADDR) {
2980                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2981                 comma = 1;
2982         }
2983         if (inp_flags & INP_HDRINCL) {
2984                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2985                 comma = 1;
2986         }
2987         if (inp_flags & INP_HIGHPORT) {
2988                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2989                 comma = 1;
2990         }
2991         if (inp_flags & INP_LOWPORT) {
2992                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2993                 comma = 1;
2994         }
2995         if (inp_flags & INP_ANONPORT) {
2996                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2997                 comma = 1;
2998         }
2999         if (inp_flags & INP_RECVIF) {
3000                 db_printf("%sINP_RECVIF", comma ? ", " : "");
3001                 comma = 1;
3002         }
3003         if (inp_flags & INP_MTUDISC) {
3004                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3005                 comma = 1;
3006         }
3007         if (inp_flags & INP_RECVTTL) {
3008                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3009                 comma = 1;
3010         }
3011         if (inp_flags & INP_DONTFRAG) {
3012                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3013                 comma = 1;
3014         }
3015         if (inp_flags & INP_RECVTOS) {
3016                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3017                 comma = 1;
3018         }
3019         if (inp_flags & IN6P_IPV6_V6ONLY) {
3020                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3021                 comma = 1;
3022         }
3023         if (inp_flags & IN6P_PKTINFO) {
3024                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3025                 comma = 1;
3026         }
3027         if (inp_flags & IN6P_HOPLIMIT) {
3028                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3029                 comma = 1;
3030         }
3031         if (inp_flags & IN6P_HOPOPTS) {
3032                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3033                 comma = 1;
3034         }
3035         if (inp_flags & IN6P_DSTOPTS) {
3036                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3037                 comma = 1;
3038         }
3039         if (inp_flags & IN6P_RTHDR) {
3040                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3041                 comma = 1;
3042         }
3043         if (inp_flags & IN6P_RTHDRDSTOPTS) {
3044                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3045                 comma = 1;
3046         }
3047         if (inp_flags & IN6P_TCLASS) {
3048                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3049                 comma = 1;
3050         }
3051         if (inp_flags & IN6P_AUTOFLOWLABEL) {
3052                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3053                 comma = 1;
3054         }
3055         if (inp_flags & INP_TIMEWAIT) {
3056                 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3057                 comma  = 1;
3058         }
3059         if (inp_flags & INP_ONESBCAST) {
3060                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3061                 comma  = 1;
3062         }
3063         if (inp_flags & INP_DROPPED) {
3064                 db_printf("%sINP_DROPPED", comma ? ", " : "");
3065                 comma  = 1;
3066         }
3067         if (inp_flags & INP_SOCKREF) {
3068                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3069                 comma  = 1;
3070         }
3071         if (inp_flags & IN6P_RFC2292) {
3072                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3073                 comma = 1;
3074         }
3075         if (inp_flags & IN6P_MTU) {
3076                 db_printf("IN6P_MTU%s", comma ? ", " : "");
3077                 comma = 1;
3078         }
3079 }
3080
3081 static void
3082 db_print_inpvflag(u_char inp_vflag)
3083 {
3084         int comma;
3085
3086         comma = 0;
3087         if (inp_vflag & INP_IPV4) {
3088                 db_printf("%sINP_IPV4", comma ? ", " : "");
3089                 comma  = 1;
3090         }
3091         if (inp_vflag & INP_IPV6) {
3092                 db_printf("%sINP_IPV6", comma ? ", " : "");
3093                 comma  = 1;
3094         }
3095         if (inp_vflag & INP_IPV6PROTO) {
3096                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3097                 comma  = 1;
3098         }
3099 }
3100
3101 static void
3102 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3103 {
3104
3105         db_print_indent(indent);
3106         db_printf("%s at %p\n", name, inp);
3107
3108         indent += 2;
3109
3110         db_print_indent(indent);
3111         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3112
3113         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3114
3115         db_print_indent(indent);
3116         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3117             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3118
3119         db_print_indent(indent);
3120         db_printf("inp_label: %p   inp_flags: 0x%x (",
3121            inp->inp_label, inp->inp_flags);
3122         db_print_inpflags(inp->inp_flags);
3123         db_printf(")\n");
3124
3125         db_print_indent(indent);
3126         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3127             inp->inp_vflag);
3128         db_print_inpvflag(inp->inp_vflag);
3129         db_printf(")\n");
3130
3131         db_print_indent(indent);
3132         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3133             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3134
3135         db_print_indent(indent);
3136 #ifdef INET6
3137         if (inp->inp_vflag & INP_IPV6) {
3138                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
3139                     "in6p_moptions: %p\n", inp->in6p_options,
3140                     inp->in6p_outputopts, inp->in6p_moptions);
3141                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3142                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3143                     inp->in6p_hops);
3144         } else
3145 #endif
3146         {
3147                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3148                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3149                     inp->inp_options, inp->inp_moptions);
3150         }
3151
3152         db_print_indent(indent);
3153         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3154             (uintmax_t)inp->inp_gencnt);
3155 }
3156
3157 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3158 {
3159         struct inpcb *inp;
3160
3161         if (!have_addr) {
3162                 db_printf("usage: show inpcb <addr>\n");
3163                 return;
3164         }
3165         inp = (struct inpcb *)addr;
3166
3167         db_print_inpcb(inp, "inpcb", 0);
3168 }
3169 #endif /* DDB */
3170
3171 #ifdef RATELIMIT
3172 /*
3173  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3174  * if any.
3175  */
3176 int
3177 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3178 {
3179         union if_snd_tag_modify_params params = {
3180                 .rate_limit.max_rate = max_pacing_rate,
3181         };
3182         struct m_snd_tag *mst;
3183         struct ifnet *ifp;
3184         int error;
3185
3186         mst = inp->inp_snd_tag;
3187         if (mst == NULL)
3188                 return (EINVAL);
3189
3190         ifp = mst->ifp;
3191         if (ifp == NULL)
3192                 return (EINVAL);
3193
3194         if (ifp->if_snd_tag_modify == NULL) {
3195                 error = EOPNOTSUPP;
3196         } else {
3197                 error = ifp->if_snd_tag_modify(mst, &params);
3198         }
3199         return (error);
3200 }
3201
3202 /*
3203  * Query existing TX rate limit based on the existing
3204  * "inp->inp_snd_tag", if any.
3205  */
3206 int
3207 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3208 {
3209         union if_snd_tag_query_params params = { };
3210         struct m_snd_tag *mst;
3211         struct ifnet *ifp;
3212         int error;
3213
3214         mst = inp->inp_snd_tag;
3215         if (mst == NULL)
3216                 return (EINVAL);
3217
3218         ifp = mst->ifp;
3219         if (ifp == NULL)
3220                 return (EINVAL);
3221
3222         if (ifp->if_snd_tag_query == NULL) {
3223                 error = EOPNOTSUPP;
3224         } else {
3225                 error = ifp->if_snd_tag_query(mst, &params);
3226                 if (error == 0 &&  p_max_pacing_rate != NULL)
3227                         *p_max_pacing_rate = params.rate_limit.max_rate;
3228         }
3229         return (error);
3230 }
3231
3232 /*
3233  * Query existing TX queue level based on the existing
3234  * "inp->inp_snd_tag", if any.
3235  */
3236 int
3237 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3238 {
3239         union if_snd_tag_query_params params = { };
3240         struct m_snd_tag *mst;
3241         struct ifnet *ifp;
3242         int error;
3243
3244         mst = inp->inp_snd_tag;
3245         if (mst == NULL)
3246                 return (EINVAL);
3247
3248         ifp = mst->ifp;
3249         if (ifp == NULL)
3250                 return (EINVAL);
3251
3252         if (ifp->if_snd_tag_query == NULL)
3253                 return (EOPNOTSUPP);
3254
3255         error = ifp->if_snd_tag_query(mst, &params);
3256         if (error == 0 &&  p_txqueue_level != NULL)
3257                 *p_txqueue_level = params.rate_limit.queue_level;
3258         return (error);
3259 }
3260
3261 /*
3262  * Allocate a new TX rate limit send tag from the network interface
3263  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3264  */
3265 int
3266 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3267     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
3268 {
3269         union if_snd_tag_alloc_params params = {
3270                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3271                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3272                 .rate_limit.hdr.flowid = flowid,
3273                 .rate_limit.hdr.flowtype = flowtype,
3274                 .rate_limit.max_rate = max_pacing_rate,
3275         };
3276         int error;
3277
3278         INP_WLOCK_ASSERT(inp);
3279
3280         if (inp->inp_snd_tag != NULL)
3281                 return (EINVAL);
3282
3283         if (ifp->if_snd_tag_alloc == NULL) {
3284                 error = EOPNOTSUPP;
3285         } else {
3286                 error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3287
3288                 /*
3289                  * At success increment the refcount on
3290                  * the send tag's network interface:
3291                  */
3292                 if (error == 0)
3293                         if_ref(inp->inp_snd_tag->ifp);
3294         }
3295         return (error);
3296 }
3297
3298 /*
3299  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3300  * if any:
3301  */
3302 void
3303 in_pcbdetach_txrtlmt(struct inpcb *inp)
3304 {
3305         struct m_snd_tag *mst;
3306         struct ifnet *ifp;
3307
3308         INP_WLOCK_ASSERT(inp);
3309
3310         mst = inp->inp_snd_tag;
3311         inp->inp_snd_tag = NULL;
3312
3313         if (mst == NULL)
3314                 return;
3315
3316         ifp = mst->ifp;
3317         if (ifp == NULL)
3318                 return;
3319
3320         /*
3321          * If the device was detached while we still had reference(s)
3322          * on the ifp, we assume if_snd_tag_free() was replaced with
3323          * stubs.
3324          */
3325         ifp->if_snd_tag_free(mst);
3326
3327         /* release reference count on network interface */
3328         if_rele(ifp);
3329 }
3330
3331 /*
3332  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3333  * is set in the fast path and will attach/detach/modify the TX rate
3334  * limit send tag based on the socket's so_max_pacing_rate value.
3335  */
3336 void
3337 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3338 {
3339         struct socket *socket;
3340         uint32_t max_pacing_rate;
3341         bool did_upgrade;
3342         int error;
3343
3344         if (inp == NULL)
3345                 return;
3346
3347         socket = inp->inp_socket;
3348         if (socket == NULL)
3349                 return;
3350
3351         if (!INP_WLOCKED(inp)) {
3352                 /*
3353                  * NOTE: If the write locking fails, we need to bail
3354                  * out and use the non-ratelimited ring for the
3355                  * transmit until there is a new chance to get the
3356                  * write lock.
3357                  */
3358                 if (!INP_TRY_UPGRADE(inp))
3359                         return;
3360                 did_upgrade = 1;
3361         } else {
3362                 did_upgrade = 0;
3363         }
3364
3365         /*
3366          * NOTE: The so_max_pacing_rate value is read unlocked,
3367          * because atomic updates are not required since the variable
3368          * is checked at every mbuf we send. It is assumed that the
3369          * variable read itself will be atomic.
3370          */
3371         max_pacing_rate = socket->so_max_pacing_rate;
3372
3373         /*
3374          * NOTE: When attaching to a network interface a reference is
3375          * made to ensure the network interface doesn't go away until
3376          * all ratelimit connections are gone. The network interface
3377          * pointers compared below represent valid network interfaces,
3378          * except when comparing towards NULL.
3379          */
3380         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3381                 error = 0;
3382         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3383                 if (inp->inp_snd_tag != NULL)
3384                         in_pcbdetach_txrtlmt(inp);
3385                 error = 0;
3386         } else if (inp->inp_snd_tag == NULL) {
3387                 /*
3388                  * In order to utilize packet pacing with RSS, we need
3389                  * to wait until there is a valid RSS hash before we
3390                  * can proceed:
3391                  */
3392                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3393                         error = EAGAIN;
3394                 } else {
3395                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3396                             mb->m_pkthdr.flowid, max_pacing_rate);
3397                 }
3398         } else {
3399                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3400         }
3401         if (error == 0 || error == EOPNOTSUPP)
3402                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3403         if (did_upgrade)
3404                 INP_DOWNGRADE(inp);
3405 }
3406
3407 /*
3408  * Track route changes for TX rate limiting.
3409  */
3410 void
3411 in_pcboutput_eagain(struct inpcb *inp)
3412 {
3413         struct socket *socket;
3414         bool did_upgrade;
3415
3416         if (inp == NULL)
3417                 return;
3418
3419         socket = inp->inp_socket;
3420         if (socket == NULL)
3421                 return;
3422
3423         if (inp->inp_snd_tag == NULL)
3424                 return;
3425
3426         if (!INP_WLOCKED(inp)) {
3427                 /*
3428                  * NOTE: If the write locking fails, we need to bail
3429                  * out and use the non-ratelimited ring for the
3430                  * transmit until there is a new chance to get the
3431                  * write lock.
3432                  */
3433                 if (!INP_TRY_UPGRADE(inp))
3434                         return;
3435                 did_upgrade = 1;
3436         } else {
3437                 did_upgrade = 0;
3438         }
3439
3440         /* detach rate limiting */
3441         in_pcbdetach_txrtlmt(inp);
3442
3443         /* make sure new mbuf send tag allocation is made */
3444         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3445
3446         if (did_upgrade)
3447                 INP_DOWNGRADE(inp);
3448 }
3449 #endif /* RATELIMIT */