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 #ifdef INET
  90 #include <netinet/in_var.h>
  91 #include <netinet/in_fib.h>
  92 #endif
  93 #include <netinet/ip_var.h>
  94 #include <netinet/tcp_var.h>
  95 #ifdef TCPHPTS
  96 #include <netinet/tcp_hpts.h>
  97 #endif
  98 #include <netinet/udp.h>
  99 #include <netinet/udp_var.h>
 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 #include <net/route/nhop.h>
 107 #endif
 108
 109 #include <netipsec/ipsec_support.h>
 110
 111 #include <security/mac/mac_framework.h>
 112
 113 #define INPCBLBGROUP_SIZMIN     8
 114 #define INPCBLBGROUP_SIZMAX     256
 115
 116 static struct callout   ipport_tick_callout;
 117
 118 /*
 119  * These configure the range of local port addresses assigned to
 120  * "unspecified" outgoing connections/packets/whatever.
 121  */
 122 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;    /* 1023 */
 123 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;    /* 600 */
 124 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;     /* 10000 */
 125 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;       /* 65535 */
 126 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;      /* 49152 */
 127 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;        /* 65535 */
 128
 129 /*
 130  * Reserved ports accessible only to root. There are significant
 131  * security considerations that must be accounted for when changing these,
 132  * but the security benefits can be great. Please be careful.
 133  */
 134 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;    /* 1023 */
 135 VNET_DEFINE(int, ipport_reservedlow);
 136
 137 /* Variables dealing with random ephemeral port allocation. */
 138 VNET_DEFINE(int, ipport_randomized) = 1;        /* user controlled via sysctl */
 139 VNET_DEFINE(int, ipport_randomcps) = 10;        /* user controlled via sysctl */
 140 VNET_DEFINE(int, ipport_randomtime) = 45;       /* user controlled via sysctl */
 141 VNET_DEFINE(int, ipport_stoprandom);            /* toggled by ipport_tick */
 142 VNET_DEFINE(int, ipport_tcpallocs);
 143 VNET_DEFINE_STATIC(int, ipport_tcplastcount);
 144
 145 #define V_ipport_tcplastcount           VNET(ipport_tcplastcount)
 146
 147 static void     in_pcbremlists(struct inpcb *inp);
 148 #ifdef INET
 149 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 150                             struct in_addr faddr, u_int fport_arg,
 151                             struct in_addr laddr, u_int lport_arg,
 152                             int lookupflags, struct ifnet *ifp);
 153
 154 #define RANGECHK(var, min, max) \
 155         if ((var) < (min)) { (var) = (min); } \
 156         else if ((var) > (max)) { (var) = (max); }
 157
 158 static int
 159 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 160 {
 161         int error;
 162
 163         error = sysctl_handle_int(oidp, arg1, arg2, req);
 164         if (error == 0) {
 165                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 166                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 167                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 168                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 169                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 170                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 171         }
 172         return (error);
 173 }
 174
 175 #undef RANGECHK
 176
 177 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
 178     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 179     "IP Ports");
 180
 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 182     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 183     &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
 184     "");
 185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 186     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 187     &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
 188     "");
 189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 190     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 191     &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
 192     "");
 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 194     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 195     &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
 196     "");
 197 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 198     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 199     &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
 200     "");
 201 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 202     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 203     &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
 204     "");
 205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 206         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 207         &VNET_NAME(ipport_reservedhigh), 0, "");
 208 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 209         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 210 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 211         CTLFLAG_VNET | CTLFLAG_RW,
 212         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 213 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
 214         CTLFLAG_VNET | CTLFLAG_RW,
 215         &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
 216         "allocations before switching to a sequental one");
 217 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
 218         CTLFLAG_VNET | CTLFLAG_RW,
 219         &VNET_NAME(ipport_randomtime), 0,
 220         "Minimum time to keep sequental port "
 221         "allocation before switching to a random one");
 222
 223 #ifdef RATELIMIT
 224 counter_u64_t rate_limit_active;
 225 counter_u64_t rate_limit_alloc_fail;
 226 counter_u64_t rate_limit_set_ok;
 227
 228 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 229     "IP Rate Limiting");
 230 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
 231     &rate_limit_active, "Active rate limited connections");
 232 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
 233    &rate_limit_alloc_fail, "Rate limited connection failures");
 234 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
 235    &rate_limit_set_ok, "Rate limited setting succeeded");
 236 #endif /* RATELIMIT */
 237
 238 #endif /* INET */
 239
 240 /*
 241  * in_pcb.c: manage the Protocol Control Blocks.
 242  *
 243  * NOTE: It is assumed that most of these functions will be called with
 244  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
 245  * functions often modify hash chains or addresses in pcbs.
 246  */
 247
 248 static struct inpcblbgroup *
 249 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, u_char vflag,
 250     uint16_t port, const union in_dependaddr *addr, int size)
 251 {
 252         struct inpcblbgroup *grp;
 253         size_t bytes;
 254
 255         bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
 256         grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
 257         if (!grp)
 258                 return (NULL);
 259         grp->il_vflag = vflag;
 260         grp->il_lport = port;
 261         grp->il_dependladdr = *addr;
 262         grp->il_inpsiz = size;
 263         CK_LIST_INSERT_HEAD(hdr, grp, il_list);
 264         return (grp);
 265 }
 266
 267 static void
 268 in_pcblbgroup_free_deferred(epoch_context_t ctx)
 269 {
 270         struct inpcblbgroup *grp;
 271
 272         grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
 273         free(grp, M_PCB);
 274 }
 275
 276 static void
 277 in_pcblbgroup_free(struct inpcblbgroup *grp)
 278 {
 279
 280         CK_LIST_REMOVE(grp, il_list);
 281         NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
 282 }
 283
 284 static struct inpcblbgroup *
 285 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
 286     struct inpcblbgroup *old_grp, int size)
 287 {
 288         struct inpcblbgroup *grp;
 289         int i;
 290
 291         grp = in_pcblbgroup_alloc(hdr, old_grp->il_vflag,
 292             old_grp->il_lport, &old_grp->il_dependladdr, size);
 293         if (grp == NULL)
 294                 return (NULL);
 295
 296         KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
 297             ("invalid new local group size %d and old local group count %d",
 298              grp->il_inpsiz, old_grp->il_inpcnt));
 299
 300         for (i = 0; i < old_grp->il_inpcnt; ++i)
 301                 grp->il_inp[i] = old_grp->il_inp[i];
 302         grp->il_inpcnt = old_grp->il_inpcnt;
 303         in_pcblbgroup_free(old_grp);
 304         return (grp);
 305 }
 306
 307 /*
 308  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
 309  * and shrink group if possible.
 310  */
 311 static void
 312 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
 313     int i)
 314 {
 315         struct inpcblbgroup *grp, *new_grp;
 316
 317         grp = *grpp;
 318         for (; i + 1 < grp->il_inpcnt; ++i)
 319                 grp->il_inp[i] = grp->il_inp[i + 1];
 320         grp->il_inpcnt--;
 321
 322         if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
 323             grp->il_inpcnt <= grp->il_inpsiz / 4) {
 324                 /* Shrink this group. */
 325                 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
 326                 if (new_grp != NULL)
 327                         *grpp = new_grp;
 328         }
 329 }
 330
 331 /*
 332  * Add PCB to load balance group for SO_REUSEPORT_LB option.
 333  */
 334 static int
 335 in_pcbinslbgrouphash(struct inpcb *inp)
 336 {
 337         const static struct timeval interval = { 60, 0 };
 338         static struct timeval lastprint;
 339         struct inpcbinfo *pcbinfo;
 340         struct inpcblbgrouphead *hdr;
 341         struct inpcblbgroup *grp;
 342         uint32_t idx;
 343
 344         pcbinfo = inp->inp_pcbinfo;
 345
 346         INP_WLOCK_ASSERT(inp);
 347         INP_HASH_WLOCK_ASSERT(pcbinfo);
 348
 349         /*
 350          * Don't allow jailed socket to join local group.
 351          */
 352         if (inp->inp_socket != NULL && jailed(inp->inp_socket->so_cred))
 353                 return (0);
 354
 355 #ifdef INET6
 356         /*
 357          * Don't allow IPv4 mapped INET6 wild socket.
 358          */
 359         if ((inp->inp_vflag & INP_IPV4) &&
 360             inp->inp_laddr.s_addr == INADDR_ANY &&
 361             INP_CHECK_SOCKAF(inp->inp_socket, AF_INET6)) {
 362                 return (0);
 363         }
 364 #endif
 365
 366         idx = INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask);
 367         hdr = &pcbinfo->ipi_lbgrouphashbase[idx];
 368         CK_LIST_FOREACH(grp, hdr, il_list) {
 369                 if (grp->il_vflag == inp->inp_vflag &&
 370                     grp->il_lport == inp->inp_lport &&
 371                     memcmp(&grp->il_dependladdr,
 372                     &inp->inp_inc.inc_ie.ie_dependladdr,
 373                     sizeof(grp->il_dependladdr)) == 0)
 374                         break;
 375         }
 376         if (grp == NULL) {
 377                 /* Create new load balance group. */
 378                 grp = in_pcblbgroup_alloc(hdr, inp->inp_vflag,
 379                     inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
 380                     INPCBLBGROUP_SIZMIN);
 381                 if (grp == NULL)
 382                         return (ENOBUFS);
 383         } else if (grp->il_inpcnt == grp->il_inpsiz) {
 384                 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
 385                         if (ratecheck(&lastprint, &interval))
 386                                 printf("lb group port %d, limit reached\n",
 387                                     ntohs(grp->il_lport));
 388                         return (0);
 389                 }
 390
 391                 /* Expand this local group. */
 392                 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
 393                 if (grp == NULL)
 394                         return (ENOBUFS);
 395         }
 396
 397         KASSERT(grp->il_inpcnt < grp->il_inpsiz,
 398             ("invalid local group size %d and count %d", grp->il_inpsiz,
 399             grp->il_inpcnt));
 400
 401         grp->il_inp[grp->il_inpcnt] = inp;
 402         grp->il_inpcnt++;
 403         return (0);
 404 }
 405
 406 /*
 407  * Remove PCB from load balance group.
 408  */
 409 static void
 410 in_pcbremlbgrouphash(struct inpcb *inp)
 411 {
 412         struct inpcbinfo *pcbinfo;
 413         struct inpcblbgrouphead *hdr;
 414         struct inpcblbgroup *grp;
 415         int i;
 416
 417         pcbinfo = inp->inp_pcbinfo;
 418
 419         INP_WLOCK_ASSERT(inp);
 420         INP_HASH_WLOCK_ASSERT(pcbinfo);
 421
 422         hdr = &pcbinfo->ipi_lbgrouphashbase[
 423             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
 424         CK_LIST_FOREACH(grp, hdr, il_list) {
 425                 for (i = 0; i < grp->il_inpcnt; ++i) {
 426                         if (grp->il_inp[i] != inp)
 427                                 continue;
 428
 429                         if (grp->il_inpcnt == 1) {
 430                                 /* We are the last, free this local group. */
 431                                 in_pcblbgroup_free(grp);
 432                         } else {
 433                                 /* Pull up inpcbs, shrink group if possible. */
 434                                 in_pcblbgroup_reorder(hdr, &grp, i);
 435                         }
 436                         return;
 437                 }
 438         }
 439 }
 440
 441 /*
 442  * Different protocols initialize their inpcbs differently - giving
 443  * different name to the lock.  But they all are disposed the same.
 444  */
 445 static void
 446 inpcb_fini(void *mem, int size)
 447 {
 448         struct inpcb *inp = mem;
 449
 450         INP_LOCK_DESTROY(inp);
 451 }
 452
 453 /*
 454  * Initialize an inpcbinfo -- we should be able to reduce the number of
 455  * arguments in time.
 456  */
 457 void
 458 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
 459     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
 460     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
 461 {
 462
 463         porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
 464
 465         INP_INFO_LOCK_INIT(pcbinfo, name);
 466         INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");     /* XXXRW: argument? */
 467         INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
 468 #ifdef VIMAGE
 469         pcbinfo->ipi_vnet = curvnet;
 470 #endif
 471         pcbinfo->ipi_listhead = listhead;
 472         CK_LIST_INIT(pcbinfo->ipi_listhead);
 473         pcbinfo->ipi_count = 0;
 474         pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
 475             &pcbinfo->ipi_hashmask);
 476         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 477             &pcbinfo->ipi_porthashmask);
 478         pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
 479             &pcbinfo->ipi_lbgrouphashmask);
 480 #ifdef PCBGROUP
 481         in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
 482 #endif
 483         pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
 484             NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
 485         uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
 486         uma_zone_set_warning(pcbinfo->ipi_zone,
 487             "kern.ipc.maxsockets limit reached");
 488 }
 489
 490 /*
 491  * Destroy an inpcbinfo.
 492  */
 493 void
 494 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 495 {
 496
 497         KASSERT(pcbinfo->ipi_count == 0,
 498             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 499
 500         hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
 501         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 502             pcbinfo->ipi_porthashmask);
 503         hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
 504             pcbinfo->ipi_lbgrouphashmask);
 505 #ifdef PCBGROUP
 506         in_pcbgroup_destroy(pcbinfo);
 507 #endif
 508         uma_zdestroy(pcbinfo->ipi_zone);
 509         INP_LIST_LOCK_DESTROY(pcbinfo);
 510         INP_HASH_LOCK_DESTROY(pcbinfo);
 511         INP_INFO_LOCK_DESTROY(pcbinfo);
 512 }
 513
 514 /*
 515  * Allocate a PCB and associate it with the socket.
 516  * On success return with the PCB locked.
 517  */
 518 int
 519 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 520 {
 521         struct inpcb *inp;
 522         int error;
 523
 524         error = 0;
 525         inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
 526         if (inp == NULL)
 527                 return (ENOBUFS);
 528         bzero(&inp->inp_start_zero, inp_zero_size);
 529 #ifdef NUMA
 530         inp->inp_numa_domain = M_NODOM;
 531 #endif
 532         inp->inp_pcbinfo = pcbinfo;
 533         inp->inp_socket = so;
 534         inp->inp_cred = crhold(so->so_cred);
 535         inp->inp_inc.inc_fibnum = so->so_fibnum;
 536 #ifdef MAC
 537         error = mac_inpcb_init(inp, M_NOWAIT);
 538         if (error != 0)
 539                 goto out;
 540         mac_inpcb_create(so, inp);
 541 #endif
 542 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 543         error = ipsec_init_pcbpolicy(inp);
 544         if (error != 0) {
 545 #ifdef MAC
 546                 mac_inpcb_destroy(inp);
 547 #endif
 548                 goto out;
 549         }
 550 #endif /*IPSEC*/
 551 #ifdef INET6
 552         if (INP_SOCKAF(so) == AF_INET6) {
 553                 inp->inp_vflag |= INP_IPV6PROTO;
 554                 if (V_ip6_v6only)
 555                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 556         }
 557 #endif
 558         INP_WLOCK(inp);
 559         INP_LIST_WLOCK(pcbinfo);
 560         CK_LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
 561         pcbinfo->ipi_count++;
 562         so->so_pcb = (caddr_t)inp;
 563 #ifdef INET6
 564         if (V_ip6_auto_flowlabel)
 565                 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 566 #endif
 567         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 568         refcount_init(&inp->inp_refcount, 1);   /* Reference from inpcbinfo */
 569
 570         /*
 571          * Routes in inpcb's can cache L2 as well; they are guaranteed
 572          * to be cleaned up.
 573          */
 574         inp->inp_route.ro_flags = RT_LLE_CACHE;
 575         INP_LIST_WUNLOCK(pcbinfo);
 576 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 577 out:
 578         if (error != 0) {
 579                 crfree(inp->inp_cred);
 580                 uma_zfree(pcbinfo->ipi_zone, inp);
 581         }
 582 #endif
 583         return (error);
 584 }
 585
 586 #ifdef INET
 587 int
 588 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 589 {
 590         int anonport, error;
 591
 592         INP_WLOCK_ASSERT(inp);
 593         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 594
 595         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 596                 return (EINVAL);
 597         anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
 598         error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
 599             &inp->inp_lport, cred);
 600         if (error)
 601                 return (error);
 602         if (in_pcbinshash(inp) != 0) {
 603                 inp->inp_laddr.s_addr = INADDR_ANY;
 604                 inp->inp_lport = 0;
 605                 return (EAGAIN);
 606         }
 607         if (anonport)
 608                 inp->inp_flags |= INP_ANONPORT;
 609         return (0);
 610 }
 611 #endif
 612
 613 #if defined(INET) || defined(INET6)
 614 /*
 615  * Assign a local port like in_pcb_lport(), but also used with connect()
 616  * and a foreign address and port.  If fsa is non-NULL, choose a local port
 617  * that is unused with those, otherwise one that is completely unused.
 618  * lsa can be NULL for IPv6.
 619  */
 620 int
 621 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
 622     struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
 623 {
 624         struct inpcbinfo *pcbinfo;
 625         struct inpcb *tmpinp;
 626         unsigned short *lastport;
 627         int count, dorandom, error;
 628         u_short aux, first, last, lport;
 629 #ifdef INET
 630         struct in_addr laddr, faddr;
 631 #endif
 632 #ifdef INET6
 633         struct in6_addr *laddr6, *faddr6;
 634 #endif
 635
 636         pcbinfo = inp->inp_pcbinfo;
 637
 638         /*
 639          * Because no actual state changes occur here, a global write lock on
 640          * the pcbinfo isn't required.
 641          */
 642         INP_LOCK_ASSERT(inp);
 643         INP_HASH_LOCK_ASSERT(pcbinfo);
 644
 645         if (inp->inp_flags & INP_HIGHPORT) {
 646                 first = V_ipport_hifirstauto;   /* sysctl */
 647                 last  = V_ipport_hilastauto;
 648                 lastport = &pcbinfo->ipi_lasthi;
 649         } else if (inp->inp_flags & INP_LOWPORT) {
 650                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
 651                 if (error)
 652                         return (error);
 653                 first = V_ipport_lowfirstauto;  /* 1023 */
 654                 last  = V_ipport_lowlastauto;   /* 600 */
 655                 lastport = &pcbinfo->ipi_lastlow;
 656         } else {
 657                 first = V_ipport_firstauto;     /* sysctl */
 658                 last  = V_ipport_lastauto;
 659                 lastport = &pcbinfo->ipi_lastport;
 660         }
 661         /*
 662          * For UDP(-Lite), use random port allocation as long as the user
 663          * allows it.  For TCP (and as of yet unknown) connections,
 664          * use random port allocation only if the user allows it AND
 665          * ipport_tick() allows it.
 666          */
 667         if (V_ipport_randomized &&
 668                 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
 669                 pcbinfo == &V_ulitecbinfo))
 670                 dorandom = 1;
 671         else
 672                 dorandom = 0;
 673         /*
 674          * It makes no sense to do random port allocation if
 675          * we have the only port available.
 676          */
 677         if (first == last)
 678                 dorandom = 0;
 679         /* Make sure to not include UDP(-Lite) packets in the count. */
 680         if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
 681                 V_ipport_tcpallocs++;
 682         /*
 683          * Instead of having two loops further down counting up or down
 684          * make sure that first is always <= last and go with only one
 685          * code path implementing all logic.
 686          */
 687         if (first > last) {
 688                 aux = first;
 689                 first = last;
 690                 last = aux;
 691         }
 692
 693 #ifdef INET
 694         laddr.s_addr = INADDR_ANY;
 695         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 696                 if (lsa != NULL)
 697                         laddr = ((struct sockaddr_in *)lsa)->sin_addr;
 698                 if (fsa != NULL)
 699                         faddr = ((struct sockaddr_in *)fsa)->sin_addr;
 700         }
 701 #endif
 702 #ifdef INET6
 703         laddr6 = NULL;
 704         if ((inp->inp_vflag & INP_IPV6) != 0) {
 705                 if (lsa != NULL)
 706                         laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
 707                 if (fsa != NULL)
 708                         faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
 709         }
 710 #endif
 711
 712         tmpinp = NULL;
 713         lport = *lportp;
 714
 715         if (dorandom)
 716                 *lastport = first + (arc4random() % (last - first));
 717
 718         count = last - first;
 719
 720         do {
 721                 if (count-- < 0)        /* completely used? */
 722                         return (EADDRNOTAVAIL);
 723                 ++*lastport;
 724                 if (*lastport < first || *lastport > last)
 725                         *lastport = first;
 726                 lport = htons(*lastport);
 727
 728                 if (fsa != NULL) {
 729 #ifdef INET
 730                         if (lsa->sa_family == AF_INET) {
 731                                 tmpinp = in_pcblookup_hash_locked(pcbinfo,
 732                                     faddr, fport, laddr, lport, lookupflags,
 733                                     NULL);
 734                         }
 735 #endif
 736 #ifdef INET6
 737                         if (lsa->sa_family == AF_INET6) {
 738                                 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
 739                                     faddr6, fport, laddr6, lport, lookupflags,
 740                                     NULL);
 741                         }
 742 #endif
 743                 } else {
 744 #ifdef INET6
 745                         if ((inp->inp_vflag & INP_IPV6) != 0)
 746                                 tmpinp = in6_pcblookup_local(pcbinfo,
 747                                     &inp->in6p_laddr, lport, lookupflags, cred);
 748 #endif
 749 #if defined(INET) && defined(INET6)
 750                         else
 751 #endif
 752 #ifdef INET
 753                                 tmpinp = in_pcblookup_local(pcbinfo, laddr,
 754                                     lport, lookupflags, cred);
 755 #endif
 756                 }
 757         } while (tmpinp != NULL);
 758
 759         *lportp = lport;
 760
 761         return (0);
 762 }
 763
 764 /*
 765  * Select a local port (number) to use.
 766  */
 767 int
 768 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 769     struct ucred *cred, int lookupflags)
 770 {
 771         struct sockaddr_in laddr;
 772
 773         if (laddrp) {
 774                 bzero(&laddr, sizeof(laddr));
 775                 laddr.sin_family = AF_INET;
 776                 laddr.sin_addr = *laddrp;
 777         }
 778         return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
 779             NULL, lportp, NULL, 0, cred, lookupflags));
 780 }
 781
 782 /*
 783  * Return cached socket options.
 784  */
 785 int
 786 inp_so_options(const struct inpcb *inp)
 787 {
 788         int so_options;
 789
 790         so_options = 0;
 791
 792         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
 793                 so_options |= SO_REUSEPORT_LB;
 794         if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 795                 so_options |= SO_REUSEPORT;
 796         if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 797                 so_options |= SO_REUSEADDR;
 798         return (so_options);
 799 }
 800 #endif /* INET || INET6 */
 801
 802 /*
 803  * Check if a new BINDMULTI socket is allowed to be created.
 804  *
 805  * ni points to the new inp.
 806  * oi points to the exisitng inp.
 807  *
 808  * This checks whether the existing inp also has BINDMULTI and
 809  * whether the credentials match.
 810  */
 811 int
 812 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
 813 {
 814         /* Check permissions match */
 815         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 816             (ni->inp_cred->cr_uid !=
 817             oi->inp_cred->cr_uid))
 818                 return (0);
 819
 820         /* Check the existing inp has BINDMULTI set */
 821         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 822             ((oi->inp_flags2 & INP_BINDMULTI) == 0))
 823                 return (0);
 824
 825         /*
 826          * We're okay - either INP_BINDMULTI isn't set on ni, or
 827          * it is and it matches the checks.
 828          */
 829         return (1);
 830 }
 831
 832 #ifdef INET
 833 /*
 834  * Set up a bind operation on a PCB, performing port allocation
 835  * as required, but do not actually modify the PCB. Callers can
 836  * either complete the bind by setting inp_laddr/inp_lport and
 837  * calling in_pcbinshash(), or they can just use the resulting
 838  * port and address to authorise the sending of a once-off packet.
 839  *
 840  * On error, the values of *laddrp and *lportp are not changed.
 841  */
 842 int
 843 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
 844     u_short *lportp, struct ucred *cred)
 845 {
 846         struct socket *so = inp->inp_socket;
 847         struct sockaddr_in *sin;
 848         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 849         struct in_addr laddr;
 850         u_short lport = 0;
 851         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 852         int error;
 853
 854         /*
 855          * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
 856          * so that we don't have to add to the (already messy) code below.
 857          */
 858         int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
 859
 860         /*
 861          * No state changes, so read locks are sufficient here.
 862          */
 863         INP_LOCK_ASSERT(inp);
 864         INP_HASH_LOCK_ASSERT(pcbinfo);
 865
 866         if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
 867                 return (EADDRNOTAVAIL);
 868         laddr.s_addr = *laddrp;
 869         if (nam != NULL && laddr.s_addr != INADDR_ANY)
 870                 return (EINVAL);
 871         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
 872                 lookupflags = INPLOOKUP_WILDCARD;
 873         if (nam == NULL) {
 874                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 875                         return (error);
 876         } else {
 877                 sin = (struct sockaddr_in *)nam;
 878                 if (nam->sa_len != sizeof (*sin))
 879                         return (EINVAL);
 880 #ifdef notdef
 881                 /*
 882                  * We should check the family, but old programs
 883                  * incorrectly fail to initialize it.
 884                  */
 885                 if (sin->sin_family != AF_INET)
 886                         return (EAFNOSUPPORT);
 887 #endif
 888                 error = prison_local_ip4(cred, &sin->sin_addr);
 889                 if (error)
 890                         return (error);
 891                 if (sin->sin_port != *lportp) {
 892                         /* Don't allow the port to change. */
 893                         if (*lportp != 0)
 894                                 return (EINVAL);
 895                         lport = sin->sin_port;
 896                 }
 897                 /* NB: lport is left as 0 if the port isn't being changed. */
 898                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 899                         /*
 900                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 901                          * allow complete duplication of binding if
 902                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 903                          * and a multicast address is bound on both
 904                          * new and duplicated sockets.
 905                          */
 906                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 907                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 908                         /*
 909                          * XXX: How to deal with SO_REUSEPORT_LB here?
 910                          * Treat same as SO_REUSEPORT for now.
 911                          */
 912                         if ((so->so_options &
 913                             (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
 914                                 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
 915                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 916                         sin->sin_port = 0;              /* yech... */
 917                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 918                         /*
 919                          * Is the address a local IP address?
 920                          * If INP_BINDANY is set, then the socket may be bound
 921                          * to any endpoint address, local or not.
 922                          */
 923                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 924                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 925                                 return (EADDRNOTAVAIL);
 926                 }
 927                 laddr = sin->sin_addr;
 928                 if (lport) {
 929                         struct inpcb *t;
 930                         struct tcptw *tw;
 931
 932                         /* GROSS */
 933                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 934                             ntohs(lport) >= V_ipport_reservedlow &&
 935                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
 936                                 return (EACCES);
 937                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 938                             priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
 939                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 940                                     lport, INPLOOKUP_WILDCARD, cred);
 941         /*
 942          * XXX
 943          * This entire block sorely needs a rewrite.
 944          */
 945                                 if (t &&
 946                                     ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 947                                     ((t->inp_flags & INP_TIMEWAIT) == 0) &&
 948                                     (so->so_type != SOCK_STREAM ||
 949                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
 950                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
 951                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
 952                                      (t->inp_flags2 & INP_REUSEPORT) ||
 953                                      (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
 954                                     (inp->inp_cred->cr_uid !=
 955                                      t->inp_cred->cr_uid))
 956                                         return (EADDRINUSE);
 957
 958                                 /*
 959                                  * If the socket is a BINDMULTI socket, then
 960                                  * the credentials need to match and the
 961                                  * original socket also has to have been bound
 962                                  * with BINDMULTI.
 963                                  */
 964                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 965                                         return (EADDRINUSE);
 966                         }
 967                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 968                             lport, lookupflags, cred);
 969                         if (t && (t->inp_flags & INP_TIMEWAIT)) {
 970                                 /*
 971                                  * XXXRW: If an incpb has had its timewait
 972                                  * state recycled, we treat the address as
 973                                  * being in use (for now).  This is better
 974                                  * than a panic, but not desirable.
 975                                  */
 976                                 tw = intotw(t);
 977                                 if (tw == NULL ||
 978                                     ((reuseport & tw->tw_so_options) == 0 &&
 979                                         (reuseport_lb &
 980                                             tw->tw_so_options) == 0)) {
 981                                         return (EADDRINUSE);
 982                                 }
 983                         } else if (t &&
 984                                    ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 985                                    (reuseport & inp_so_options(t)) == 0 &&
 986                                    (reuseport_lb & inp_so_options(t)) == 0) {
 987 #ifdef INET6
 988                                 if (ntohl(sin->sin_addr.s_addr) !=
 989                                     INADDR_ANY ||
 990                                     ntohl(t->inp_laddr.s_addr) !=
 991                                     INADDR_ANY ||
 992                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
 993                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
 994 #endif
 995                                                 return (EADDRINUSE);
 996                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 997                                         return (EADDRINUSE);
 998                         }
 999                 }
1000         }
1001         if (*lportp != 0)
1002                 lport = *lportp;
1003         if (lport == 0) {
1004                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1005                 if (error != 0)
1006                         return (error);
1007         }
1008         *laddrp = laddr.s_addr;
1009         *lportp = lport;
1010         return (0);
1011 }
1012
1013 /*
1014  * Connect from a socket to a specified address.
1015  * Both address and port must be specified in argument sin.
1016  * If don't have a local address for this socket yet,
1017  * then pick one.
1018  */
1019 int
1020 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
1021     struct ucred *cred, struct mbuf *m, bool rehash)
1022 {
1023         u_short lport, fport;
1024         in_addr_t laddr, faddr;
1025         int anonport, error;
1026
1027         INP_WLOCK_ASSERT(inp);
1028         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1029
1030         lport = inp->inp_lport;
1031         laddr = inp->inp_laddr.s_addr;
1032         anonport = (lport == 0);
1033         error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
1034             NULL, cred);
1035         if (error)
1036                 return (error);
1037
1038         /* Do the initial binding of the local address if required. */
1039         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1040                 KASSERT(rehash == true,
1041                     ("Rehashing required for unbound inps"));
1042                 inp->inp_lport = lport;
1043                 inp->inp_laddr.s_addr = laddr;
1044                 if (in_pcbinshash(inp) != 0) {
1045                         inp->inp_laddr.s_addr = INADDR_ANY;
1046                         inp->inp_lport = 0;
1047                         return (EAGAIN);
1048                 }
1049         }
1050
1051         /* Commit the remaining changes. */
1052         inp->inp_lport = lport;
1053         inp->inp_laddr.s_addr = laddr;
1054         inp->inp_faddr.s_addr = faddr;
1055         inp->inp_fport = fport;
1056         if (rehash) {
1057                 in_pcbrehash_mbuf(inp, m);
1058         } else {
1059                 in_pcbinshash_mbuf(inp, m);
1060         }
1061
1062         if (anonport)
1063                 inp->inp_flags |= INP_ANONPORT;
1064         return (0);
1065 }
1066
1067 int
1068 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
1069 {
1070
1071         return (in_pcbconnect_mbuf(inp, nam, cred, NULL, true));
1072 }
1073
1074 /*
1075  * Do proper source address selection on an unbound socket in case
1076  * of connect. Take jails into account as well.
1077  */
1078 int
1079 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1080     struct ucred *cred)
1081 {
1082         struct ifaddr *ifa;
1083         struct sockaddr *sa;
1084         struct sockaddr_in *sin, dst;
1085         struct nhop_object *nh;
1086         int error;
1087
1088         NET_EPOCH_ASSERT();
1089         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1090         /*
1091          * Bypass source address selection and use the primary jail IP
1092          * if requested.
1093          */
1094         if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
1095                 return (0);
1096
1097         error = 0;
1098
1099         nh = NULL;
1100         bzero(&dst, sizeof(dst));
1101         sin = &dst;
1102         sin->sin_family = AF_INET;
1103         sin->sin_len = sizeof(struct sockaddr_in);
1104         sin->sin_addr.s_addr = faddr->s_addr;
1105
1106         /*
1107          * If route is known our src addr is taken from the i/f,
1108          * else punt.
1109          *
1110          * Find out route to destination.
1111          */
1112         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1113                 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1114                     0, NHR_NONE, 0);
1115
1116         /*
1117          * If we found a route, use the address corresponding to
1118          * the outgoing interface.
1119          *
1120          * Otherwise assume faddr is reachable on a directly connected
1121          * network and try to find a corresponding interface to take
1122          * the source address from.
1123          */
1124         if (nh == NULL || nh->nh_ifp == NULL) {
1125                 struct in_ifaddr *ia;
1126                 struct ifnet *ifp;
1127
1128                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1129                                         inp->inp_socket->so_fibnum));
1130                 if (ia == NULL) {
1131                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1132                                                 inp->inp_socket->so_fibnum));
1133                 }
1134                 if (ia == NULL) {
1135                         error = ENETUNREACH;
1136                         goto done;
1137                 }
1138
1139                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1140                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1141                         goto done;
1142                 }
1143
1144                 ifp = ia->ia_ifp;
1145                 ia = NULL;
1146                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1147                         sa = ifa->ifa_addr;
1148                         if (sa->sa_family != AF_INET)
1149                                 continue;
1150                         sin = (struct sockaddr_in *)sa;
1151                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1152                                 ia = (struct in_ifaddr *)ifa;
1153                                 break;
1154                         }
1155                 }
1156                 if (ia != NULL) {
1157                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1158                         goto done;
1159                 }
1160
1161                 /* 3. As a last resort return the 'default' jail address. */
1162                 error = prison_get_ip4(cred, laddr);
1163                 goto done;
1164         }
1165
1166         /*
1167          * If the outgoing interface on the route found is not
1168          * a loopback interface, use the address from that interface.
1169          * In case of jails do those three steps:
1170          * 1. check if the interface address belongs to the jail. If so use it.
1171          * 2. check if we have any address on the outgoing interface
1172          *    belonging to this jail. If so use it.
1173          * 3. as a last resort return the 'default' jail address.
1174          */
1175         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1176                 struct in_ifaddr *ia;
1177                 struct ifnet *ifp;
1178
1179                 /* If not jailed, use the default returned. */
1180                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1181                         ia = (struct in_ifaddr *)nh->nh_ifa;
1182                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1183                         goto done;
1184                 }
1185
1186                 /* Jailed. */
1187                 /* 1. Check if the iface address belongs to the jail. */
1188                 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1189                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1190                         ia = (struct in_ifaddr *)nh->nh_ifa;
1191                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1192                         goto done;
1193                 }
1194
1195                 /*
1196                  * 2. Check if we have any address on the outgoing interface
1197                  *    belonging to this jail.
1198                  */
1199                 ia = NULL;
1200                 ifp = nh->nh_ifp;
1201                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1202                         sa = ifa->ifa_addr;
1203                         if (sa->sa_family != AF_INET)
1204                                 continue;
1205                         sin = (struct sockaddr_in *)sa;
1206                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1207                                 ia = (struct in_ifaddr *)ifa;
1208                                 break;
1209                         }
1210                 }
1211                 if (ia != NULL) {
1212                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1213                         goto done;
1214                 }
1215
1216                 /* 3. As a last resort return the 'default' jail address. */
1217                 error = prison_get_ip4(cred, laddr);
1218                 goto done;
1219         }
1220
1221         /*
1222          * The outgoing interface is marked with 'loopback net', so a route
1223          * to ourselves is here.
1224          * Try to find the interface of the destination address and then
1225          * take the address from there. That interface is not necessarily
1226          * a loopback interface.
1227          * In case of jails, check that it is an address of the jail
1228          * and if we cannot find, fall back to the 'default' jail address.
1229          */
1230         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1231                 struct in_ifaddr *ia;
1232
1233                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1234                                         inp->inp_socket->so_fibnum));
1235                 if (ia == NULL)
1236                         ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1237                                                 inp->inp_socket->so_fibnum));
1238                 if (ia == NULL)
1239                         ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1240
1241                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
1242                         if (ia == NULL) {
1243                                 error = ENETUNREACH;
1244                                 goto done;
1245                         }
1246                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1247                         goto done;
1248                 }
1249
1250                 /* Jailed. */
1251                 if (ia != NULL) {
1252                         struct ifnet *ifp;
1253
1254                         ifp = ia->ia_ifp;
1255                         ia = NULL;
1256                         CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1257                                 sa = ifa->ifa_addr;
1258                                 if (sa->sa_family != AF_INET)
1259                                         continue;
1260                                 sin = (struct sockaddr_in *)sa;
1261                                 if (prison_check_ip4(cred,
1262                                     &sin->sin_addr) == 0) {
1263                                         ia = (struct in_ifaddr *)ifa;
1264                                         break;
1265                                 }
1266                         }
1267                         if (ia != NULL) {
1268                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1269                                 goto done;
1270                         }
1271                 }
1272
1273                 /* 3. As a last resort return the 'default' jail address. */
1274                 error = prison_get_ip4(cred, laddr);
1275                 goto done;
1276         }
1277
1278 done:
1279         return (error);
1280 }
1281
1282 /*
1283  * Set up for a connect from a socket to the specified address.
1284  * On entry, *laddrp and *lportp should contain the current local
1285  * address and port for the PCB; these are updated to the values
1286  * that should be placed in inp_laddr and inp_lport to complete
1287  * the connect.
1288  *
1289  * On success, *faddrp and *fportp will be set to the remote address
1290  * and port. These are not updated in the error case.
1291  *
1292  * If the operation fails because the connection already exists,
1293  * *oinpp will be set to the PCB of that connection so that the
1294  * caller can decide to override it. In all other cases, *oinpp
1295  * is set to NULL.
1296  */
1297 int
1298 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1299     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1300     struct inpcb **oinpp, struct ucred *cred)
1301 {
1302         struct rm_priotracker in_ifa_tracker;
1303         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1304         struct in_ifaddr *ia;
1305         struct inpcb *oinp;
1306         struct in_addr laddr, faddr;
1307         u_short lport, fport;
1308         int error;
1309
1310         /*
1311          * Because a global state change doesn't actually occur here, a read
1312          * lock is sufficient.
1313          */
1314         NET_EPOCH_ASSERT();
1315         INP_LOCK_ASSERT(inp);
1316         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1317
1318         if (oinpp != NULL)
1319                 *oinpp = NULL;
1320         if (nam->sa_len != sizeof (*sin))
1321                 return (EINVAL);
1322         if (sin->sin_family != AF_INET)
1323                 return (EAFNOSUPPORT);
1324         if (sin->sin_port == 0)
1325                 return (EADDRNOTAVAIL);
1326         laddr.s_addr = *laddrp;
1327         lport = *lportp;
1328         faddr = sin->sin_addr;
1329         fport = sin->sin_port;
1330
1331         if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1332                 /*
1333                  * If the destination address is INADDR_ANY,
1334                  * use the primary local address.
1335                  * If the supplied address is INADDR_BROADCAST,
1336                  * and the primary interface supports broadcast,
1337                  * choose the broadcast address for that interface.
1338                  */
1339                 if (faddr.s_addr == INADDR_ANY) {
1340                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1341                         faddr =
1342                             IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1343                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1344                         if (cred != NULL &&
1345                             (error = prison_get_ip4(cred, &faddr)) != 0)
1346                                 return (error);
1347                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1348                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1349                         if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1350                             IFF_BROADCAST)
1351                                 faddr = satosin(&CK_STAILQ_FIRST(
1352                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1353                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1354                 }
1355         }
1356         if (laddr.s_addr == INADDR_ANY) {
1357                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1358                 /*
1359                  * If the destination address is multicast and an outgoing
1360                  * interface has been set as a multicast option, prefer the
1361                  * address of that interface as our source address.
1362                  */
1363                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1364                     inp->inp_moptions != NULL) {
1365                         struct ip_moptions *imo;
1366                         struct ifnet *ifp;
1367
1368                         imo = inp->inp_moptions;
1369                         if (imo->imo_multicast_ifp != NULL) {
1370                                 ifp = imo->imo_multicast_ifp;
1371                                 IN_IFADDR_RLOCK(&in_ifa_tracker);
1372                                 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1373                                         if ((ia->ia_ifp == ifp) &&
1374                                             (cred == NULL ||
1375                                             prison_check_ip4(cred,
1376                                             &ia->ia_addr.sin_addr) == 0))
1377                                                 break;
1378                                 }
1379                                 if (ia == NULL)
1380                                         error = EADDRNOTAVAIL;
1381                                 else {
1382                                         laddr = ia->ia_addr.sin_addr;
1383                                         error = 0;
1384                                 }
1385                                 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1386                         }
1387                 }
1388                 if (error)
1389                         return (error);
1390         }
1391         if (lport != 0) {
1392                 oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1393                     fport, laddr, lport, 0, NULL);
1394                 if (oinp != NULL) {
1395                         if (oinpp != NULL)
1396                                 *oinpp = oinp;
1397                         return (EADDRINUSE);
1398                 }
1399         } else {
1400                 struct sockaddr_in lsin, fsin;
1401
1402                 bzero(&lsin, sizeof(lsin));
1403                 bzero(&fsin, sizeof(fsin));
1404                 lsin.sin_family = AF_INET;
1405                 lsin.sin_addr = laddr;
1406                 fsin.sin_family = AF_INET;
1407                 fsin.sin_addr = faddr;
1408                 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1409                     &lport, (struct sockaddr *)& fsin, fport, cred,
1410                     INPLOOKUP_WILDCARD);
1411                 if (error)
1412                         return (error);
1413         }
1414         *laddrp = laddr.s_addr;
1415         *lportp = lport;
1416         *faddrp = faddr.s_addr;
1417         *fportp = fport;
1418         return (0);
1419 }
1420
1421 void
1422 in_pcbdisconnect(struct inpcb *inp)
1423 {
1424
1425         INP_WLOCK_ASSERT(inp);
1426         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1427
1428         inp->inp_faddr.s_addr = INADDR_ANY;
1429         inp->inp_fport = 0;
1430         in_pcbrehash(inp);
1431 }
1432 #endif /* INET */
1433
1434 /*
1435  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1436  * For most protocols, this will be invoked immediately prior to calling
1437  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1438  * socket, in which case in_pcbfree() is deferred.
1439  */
1440 void
1441 in_pcbdetach(struct inpcb *inp)
1442 {
1443
1444         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1445
1446 #ifdef RATELIMIT
1447         if (inp->inp_snd_tag != NULL)
1448                 in_pcbdetach_txrtlmt(inp);
1449 #endif
1450         inp->inp_socket->so_pcb = NULL;
1451         inp->inp_socket = NULL;
1452 }
1453
1454 /*
1455  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1456  * stability of an inpcb pointer despite the inpcb lock being released.  This
1457  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1458  * but where the inpcb lock may already held, or when acquiring a reference
1459  * via a pcbgroup.
1460  *
1461  * in_pcbref() should be used only to provide brief memory stability, and
1462  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1463  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1464  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1465  * lock and rele are the *only* safe operations that may be performed on the
1466  * inpcb.
1467  *
1468  * While the inpcb will not be freed, releasing the inpcb lock means that the
1469  * connection's state may change, so the caller should be careful to
1470  * revalidate any cached state on reacquiring the lock.  Drop the reference
1471  * using in_pcbrele().
1472  */
1473 void
1474 in_pcbref(struct inpcb *inp)
1475 {
1476
1477         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1478
1479         refcount_acquire(&inp->inp_refcount);
1480 }
1481
1482 /*
1483  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1484  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1485  * return a flag indicating whether or not the inpcb remains valid.  If it is
1486  * valid, we return with the inpcb lock held.
1487  *
1488  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1489  * reference on an inpcb.  Historically more work was done here (actually, in
1490  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1491  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1492  * about memory stability (and continued use of the write lock).
1493  */
1494 int
1495 in_pcbrele_rlocked(struct inpcb *inp)
1496 {
1497         struct inpcbinfo *pcbinfo;
1498
1499         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1500
1501         INP_RLOCK_ASSERT(inp);
1502
1503         if (refcount_release(&inp->inp_refcount) == 0) {
1504                 /*
1505                  * If the inpcb has been freed, let the caller know, even if
1506                  * this isn't the last reference.
1507                  */
1508                 if (inp->inp_flags2 & INP_FREED) {
1509                         INP_RUNLOCK(inp);
1510                         return (1);
1511                 }
1512                 return (0);
1513         }
1514
1515         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1516 #ifdef TCPHPTS
1517         if (inp->inp_in_hpts || inp->inp_in_input) {
1518                 struct tcp_hpts_entry *hpts;
1519                 /*
1520                  * We should not be on the hpts at
1521                  * this point in any form. we must
1522                  * get the lock to be sure.
1523                  */
1524                 hpts = tcp_hpts_lock(inp);
1525                 if (inp->inp_in_hpts)
1526                         panic("Hpts:%p inp:%p at free still on hpts",
1527                               hpts, inp);
1528                 mtx_unlock(&hpts->p_mtx);
1529                 hpts = tcp_input_lock(inp);
1530                 if (inp->inp_in_input)
1531                         panic("Hpts:%p inp:%p at free still on input hpts",
1532                               hpts, inp);
1533                 mtx_unlock(&hpts->p_mtx);
1534         }
1535 #endif
1536         INP_RUNLOCK(inp);
1537         pcbinfo = inp->inp_pcbinfo;
1538         uma_zfree(pcbinfo->ipi_zone, inp);
1539         return (1);
1540 }
1541
1542 int
1543 in_pcbrele_wlocked(struct inpcb *inp)
1544 {
1545         struct inpcbinfo *pcbinfo;
1546
1547         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1548
1549         INP_WLOCK_ASSERT(inp);
1550
1551         if (refcount_release(&inp->inp_refcount) == 0) {
1552                 /*
1553                  * If the inpcb has been freed, let the caller know, even if
1554                  * this isn't the last reference.
1555                  */
1556                 if (inp->inp_flags2 & INP_FREED) {
1557                         INP_WUNLOCK(inp);
1558                         return (1);
1559                 }
1560                 return (0);
1561         }
1562
1563         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1564 #ifdef TCPHPTS
1565         if (inp->inp_in_hpts || inp->inp_in_input) {
1566                 struct tcp_hpts_entry *hpts;
1567                 /*
1568                  * We should not be on the hpts at
1569                  * this point in any form. we must
1570                  * get the lock to be sure.
1571                  */
1572                 hpts = tcp_hpts_lock(inp);
1573                 if (inp->inp_in_hpts)
1574                         panic("Hpts:%p inp:%p at free still on hpts",
1575                               hpts, inp);
1576                 mtx_unlock(&hpts->p_mtx);
1577                 hpts = tcp_input_lock(inp);
1578                 if (inp->inp_in_input)
1579                         panic("Hpts:%p inp:%p at free still on input hpts",
1580                               hpts, inp);
1581                 mtx_unlock(&hpts->p_mtx);
1582         }
1583 #endif
1584         INP_WUNLOCK(inp);
1585         pcbinfo = inp->inp_pcbinfo;
1586         uma_zfree(pcbinfo->ipi_zone, inp);
1587         return (1);
1588 }
1589
1590 /*
1591  * Temporary wrapper.
1592  */
1593 int
1594 in_pcbrele(struct inpcb *inp)
1595 {
1596
1597         return (in_pcbrele_wlocked(inp));
1598 }
1599
1600 void
1601 in_pcblist_rele_rlocked(epoch_context_t ctx)
1602 {
1603         struct in_pcblist *il;
1604         struct inpcb *inp;
1605         struct inpcbinfo *pcbinfo;
1606         int i, n;
1607
1608         il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1609         pcbinfo = il->il_pcbinfo;
1610         n = il->il_count;
1611         INP_INFO_WLOCK(pcbinfo);
1612         for (i = 0; i < n; i++) {
1613                 inp = il->il_inp_list[i];
1614                 INP_RLOCK(inp);
1615                 if (!in_pcbrele_rlocked(inp))
1616                         INP_RUNLOCK(inp);
1617         }
1618         INP_INFO_WUNLOCK(pcbinfo);
1619         free(il, M_TEMP);
1620 }
1621
1622 static void
1623 inpcbport_free(epoch_context_t ctx)
1624 {
1625         struct inpcbport *phd;
1626
1627         phd = __containerof(ctx, struct inpcbport, phd_epoch_ctx);
1628         free(phd, M_PCB);
1629 }
1630
1631 static void
1632 in_pcbfree_deferred(epoch_context_t ctx)
1633 {
1634         struct inpcb *inp;
1635         int released __unused;
1636
1637         inp = __containerof(ctx, struct inpcb, inp_epoch_ctx);
1638
1639         INP_WLOCK(inp);
1640         CURVNET_SET(inp->inp_vnet);
1641 #ifdef INET
1642         struct ip_moptions *imo = inp->inp_moptions;
1643         inp->inp_moptions = NULL;
1644 #endif
1645         /* XXXRW: Do as much as possible here. */
1646 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1647         if (inp->inp_sp != NULL)
1648                 ipsec_delete_pcbpolicy(inp);
1649 #endif
1650 #ifdef INET6
1651         struct ip6_moptions *im6o = NULL;
1652         if (inp->inp_vflag & INP_IPV6PROTO) {
1653                 ip6_freepcbopts(inp->in6p_outputopts);
1654                 im6o = inp->in6p_moptions;
1655                 inp->in6p_moptions = NULL;
1656         }
1657 #endif
1658         if (inp->inp_options)
1659                 (void)m_free(inp->inp_options);
1660         inp->inp_vflag = 0;
1661         crfree(inp->inp_cred);
1662 #ifdef MAC
1663         mac_inpcb_destroy(inp);
1664 #endif
1665         released = in_pcbrele_wlocked(inp);
1666         MPASS(released);
1667 #ifdef INET6
1668         ip6_freemoptions(im6o);
1669 #endif
1670 #ifdef INET
1671         inp_freemoptions(imo);
1672 #endif
1673         CURVNET_RESTORE();
1674 }
1675
1676 /*
1677  * Unconditionally schedule an inpcb to be freed by decrementing its
1678  * reference count, which should occur only after the inpcb has been detached
1679  * from its socket.  If another thread holds a temporary reference (acquired
1680  * using in_pcbref()) then the free is deferred until that reference is
1681  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1682  * work, including removal from global lists, is done in this context, where
1683  * the pcbinfo lock is held.
1684  */
1685 void
1686 in_pcbfree(struct inpcb *inp)
1687 {
1688         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1689
1690         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1691         KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1692             ("%s: called twice for pcb %p", __func__, inp));
1693         if (inp->inp_flags2 & INP_FREED) {
1694                 INP_WUNLOCK(inp);
1695                 return;
1696         }
1697
1698         INP_WLOCK_ASSERT(inp);
1699         INP_LIST_WLOCK(pcbinfo);
1700         in_pcbremlists(inp);
1701         INP_LIST_WUNLOCK(pcbinfo);
1702         RO_INVALIDATE_CACHE(&inp->inp_route);
1703         /* mark as destruction in progress */
1704         inp->inp_flags2 |= INP_FREED;
1705         INP_WUNLOCK(inp);
1706         NET_EPOCH_CALL(in_pcbfree_deferred, &inp->inp_epoch_ctx);
1707 }
1708
1709 /*
1710  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1711  * port reservation, and preventing it from being returned by inpcb lookups.
1712  *
1713  * It is used by TCP to mark an inpcb as unused and avoid future packet
1714  * delivery or event notification when a socket remains open but TCP has
1715  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1716  * or a RST on the wire, and allows the port binding to be reused while still
1717  * maintaining the invariant that so_pcb always points to a valid inpcb until
1718  * in_pcbdetach().
1719  *
1720  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1721  * in_pcbnotifyall() and in_pcbpurgeif0()?
1722  */
1723 void
1724 in_pcbdrop(struct inpcb *inp)
1725 {
1726
1727         INP_WLOCK_ASSERT(inp);
1728 #ifdef INVARIANTS
1729         if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1730                 MPASS(inp->inp_refcount > 1);
1731 #endif
1732
1733         /*
1734          * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1735          * the hash lock...?
1736          */
1737         inp->inp_flags |= INP_DROPPED;
1738         if (inp->inp_flags & INP_INHASHLIST) {
1739                 struct inpcbport *phd = inp->inp_phd;
1740
1741                 INP_HASH_WLOCK(inp->inp_pcbinfo);
1742                 in_pcbremlbgrouphash(inp);
1743                 CK_LIST_REMOVE(inp, inp_hash);
1744                 CK_LIST_REMOVE(inp, inp_portlist);
1745                 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
1746                         CK_LIST_REMOVE(phd, phd_hash);
1747                         NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
1748                 }
1749                 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1750                 inp->inp_flags &= ~INP_INHASHLIST;
1751 #ifdef PCBGROUP
1752                 in_pcbgroup_remove(inp);
1753 #endif
1754         }
1755 }
1756
1757 #ifdef INET
1758 /*
1759  * Common routines to return the socket addresses associated with inpcbs.
1760  */
1761 struct sockaddr *
1762 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1763 {
1764         struct sockaddr_in *sin;
1765
1766         sin = malloc(sizeof *sin, M_SONAME,
1767                 M_WAITOK | M_ZERO);
1768         sin->sin_family = AF_INET;
1769         sin->sin_len = sizeof(*sin);
1770         sin->sin_addr = *addr_p;
1771         sin->sin_port = port;
1772
1773         return (struct sockaddr *)sin;
1774 }
1775
1776 int
1777 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1778 {
1779         struct inpcb *inp;
1780         struct in_addr addr;
1781         in_port_t port;
1782
1783         inp = sotoinpcb(so);
1784         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1785
1786         INP_RLOCK(inp);
1787         port = inp->inp_lport;
1788         addr = inp->inp_laddr;
1789         INP_RUNLOCK(inp);
1790
1791         *nam = in_sockaddr(port, &addr);
1792         return 0;
1793 }
1794
1795 int
1796 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1797 {
1798         struct inpcb *inp;
1799         struct in_addr addr;
1800         in_port_t port;
1801
1802         inp = sotoinpcb(so);
1803         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1804
1805         INP_RLOCK(inp);
1806         port = inp->inp_fport;
1807         addr = inp->inp_faddr;
1808         INP_RUNLOCK(inp);
1809
1810         *nam = in_sockaddr(port, &addr);
1811         return 0;
1812 }
1813
1814 void
1815 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1816     struct inpcb *(*notify)(struct inpcb *, int))
1817 {
1818         struct inpcb *inp, *inp_temp;
1819
1820         INP_INFO_WLOCK(pcbinfo);
1821         CK_LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1822                 INP_WLOCK(inp);
1823 #ifdef INET6
1824                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1825                         INP_WUNLOCK(inp);
1826                         continue;
1827                 }
1828 #endif
1829                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1830                     inp->inp_socket == NULL) {
1831                         INP_WUNLOCK(inp);
1832                         continue;
1833                 }
1834                 if ((*notify)(inp, errno))
1835                         INP_WUNLOCK(inp);
1836         }
1837         INP_INFO_WUNLOCK(pcbinfo);
1838 }
1839
1840 void
1841 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1842 {
1843         struct inpcb *inp;
1844         struct in_multi *inm;
1845         struct in_mfilter *imf;
1846         struct ip_moptions *imo;
1847
1848         INP_INFO_WLOCK(pcbinfo);
1849         CK_LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1850                 INP_WLOCK(inp);
1851                 imo = inp->inp_moptions;
1852                 if ((inp->inp_vflag & INP_IPV4) &&
1853                     imo != NULL) {
1854                         /*
1855                          * Unselect the outgoing interface if it is being
1856                          * detached.
1857                          */
1858                         if (imo->imo_multicast_ifp == ifp)
1859                                 imo->imo_multicast_ifp = NULL;
1860
1861                         /*
1862                          * Drop multicast group membership if we joined
1863                          * through the interface being detached.
1864                          *
1865                          * XXX This can all be deferred to an epoch_call
1866                          */
1867 restart:
1868                         IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1869                                 if ((inm = imf->imf_inm) == NULL)
1870                                         continue;
1871                                 if (inm->inm_ifp != ifp)
1872                                         continue;
1873                                 ip_mfilter_remove(&imo->imo_head, imf);
1874                                 IN_MULTI_LOCK_ASSERT();
1875                                 in_leavegroup_locked(inm, NULL);
1876                                 ip_mfilter_free(imf);
1877                                 goto restart;
1878                         }
1879                 }
1880                 INP_WUNLOCK(inp);
1881         }
1882         INP_INFO_WUNLOCK(pcbinfo);
1883 }
1884
1885 /*
1886  * Lookup a PCB based on the local address and port.  Caller must hold the
1887  * hash lock.  No inpcb locks or references are acquired.
1888  */
1889 #define INP_LOOKUP_MAPPED_PCB_COST      3
1890 struct inpcb *
1891 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1892     u_short lport, int lookupflags, struct ucred *cred)
1893 {
1894         struct inpcb *inp;
1895 #ifdef INET6
1896         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1897 #else
1898         int matchwild = 3;
1899 #endif
1900         int wildcard;
1901
1902         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1903             ("%s: invalid lookup flags %d", __func__, lookupflags));
1904
1905         INP_HASH_LOCK_ASSERT(pcbinfo);
1906
1907         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1908                 struct inpcbhead *head;
1909                 /*
1910                  * Look for an unconnected (wildcard foreign addr) PCB that
1911                  * matches the local address and port we're looking for.
1912                  */
1913                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1914                     0, pcbinfo->ipi_hashmask)];
1915                 CK_LIST_FOREACH(inp, head, inp_hash) {
1916 #ifdef INET6
1917                         /* XXX inp locking */
1918                         if ((inp->inp_vflag & INP_IPV4) == 0)
1919                                 continue;
1920 #endif
1921                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
1922                             inp->inp_laddr.s_addr == laddr.s_addr &&
1923                             inp->inp_lport == lport) {
1924                                 /*
1925                                  * Found?
1926                                  */
1927                                 if (cred == NULL ||
1928                                     prison_equal_ip4(cred->cr_prison,
1929                                         inp->inp_cred->cr_prison))
1930                                         return (inp);
1931                         }
1932                 }
1933                 /*
1934                  * Not found.
1935                  */
1936                 return (NULL);
1937         } else {
1938                 struct inpcbporthead *porthash;
1939                 struct inpcbport *phd;
1940                 struct inpcb *match = NULL;
1941                 /*
1942                  * Best fit PCB lookup.
1943                  *
1944                  * First see if this local port is in use by looking on the
1945                  * port hash list.
1946                  */
1947                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1948                     pcbinfo->ipi_porthashmask)];
1949                 CK_LIST_FOREACH(phd, porthash, phd_hash) {
1950                         if (phd->phd_port == lport)
1951                                 break;
1952                 }
1953                 if (phd != NULL) {
1954                         /*
1955                          * Port is in use by one or more PCBs. Look for best
1956                          * fit.
1957                          */
1958                         CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1959                                 wildcard = 0;
1960                                 if (cred != NULL &&
1961                                     !prison_equal_ip4(inp->inp_cred->cr_prison,
1962                                         cred->cr_prison))
1963                                         continue;
1964 #ifdef INET6
1965                                 /* XXX inp locking */
1966                                 if ((inp->inp_vflag & INP_IPV4) == 0)
1967                                         continue;
1968                                 /*
1969                                  * We never select the PCB that has
1970                                  * INP_IPV6 flag and is bound to :: if
1971                                  * we have another PCB which is bound
1972                                  * to 0.0.0.0.  If a PCB has the
1973                                  * INP_IPV6 flag, then we set its cost
1974                                  * higher than IPv4 only PCBs.
1975                                  *
1976                                  * Note that the case only happens
1977                                  * when a socket is bound to ::, under
1978                                  * the condition that the use of the
1979                                  * mapped address is allowed.
1980                                  */
1981                                 if ((inp->inp_vflag & INP_IPV6) != 0)
1982                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1983 #endif
1984                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
1985                                         wildcard++;
1986                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1987                                         if (laddr.s_addr == INADDR_ANY)
1988                                                 wildcard++;
1989                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
1990                                                 continue;
1991                                 } else {
1992                                         if (laddr.s_addr != INADDR_ANY)
1993                                                 wildcard++;
1994                                 }
1995                                 if (wildcard < matchwild) {
1996                                         match = inp;
1997                                         matchwild = wildcard;
1998                                         if (matchwild == 0)
1999                                                 break;
2000                                 }
2001                         }
2002                 }
2003                 return (match);
2004         }
2005 }
2006 #undef INP_LOOKUP_MAPPED_PCB_COST
2007
2008 static struct inpcb *
2009 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2010     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2011     uint16_t fport, int lookupflags)
2012 {
2013         struct inpcb *local_wild;
2014         const struct inpcblbgrouphead *hdr;
2015         struct inpcblbgroup *grp;
2016         uint32_t idx;
2017
2018         INP_HASH_LOCK_ASSERT(pcbinfo);
2019
2020         hdr = &pcbinfo->ipi_lbgrouphashbase[
2021             INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2022
2023         /*
2024          * Order of socket selection:
2025          * 1. non-wild.
2026          * 2. wild (if lookupflags contains INPLOOKUP_WILDCARD).
2027          *
2028          * NOTE:
2029          * - Load balanced group does not contain jailed sockets
2030          * - Load balanced group does not contain IPv4 mapped INET6 wild sockets
2031          */
2032         local_wild = NULL;
2033         CK_LIST_FOREACH(grp, hdr, il_list) {
2034 #ifdef INET6
2035                 if (!(grp->il_vflag & INP_IPV4))
2036                         continue;
2037 #endif
2038                 if (grp->il_lport != lport)
2039                         continue;
2040
2041                 idx = INP_PCBLBGROUP_PKTHASH(faddr->s_addr, lport, fport) %
2042                     grp->il_inpcnt;
2043                 if (grp->il_laddr.s_addr == laddr->s_addr)
2044                         return (grp->il_inp[idx]);
2045                 if (grp->il_laddr.s_addr == INADDR_ANY &&
2046                     (lookupflags & INPLOOKUP_WILDCARD) != 0)
2047                         local_wild = grp->il_inp[idx];
2048         }
2049         return (local_wild);
2050 }
2051
2052 #ifdef PCBGROUP
2053 /*
2054  * Lookup PCB in hash list, using pcbgroup tables.
2055  */
2056 static struct inpcb *
2057 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
2058     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
2059     u_int lport_arg, int lookupflags, struct ifnet *ifp)
2060 {
2061         struct inpcbhead *head;
2062         struct inpcb *inp, *tmpinp;
2063         u_short fport = fport_arg, lport = lport_arg;
2064         bool locked;
2065
2066         /*
2067          * First look for an exact match.
2068          */
2069         tmpinp = NULL;
2070         INP_GROUP_LOCK(pcbgroup);
2071         head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2072             pcbgroup->ipg_hashmask)];
2073         CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2074 #ifdef INET6
2075                 /* XXX inp locking */
2076                 if ((inp->inp_vflag & INP_IPV4) == 0)
2077                         continue;
2078 #endif
2079                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2080                     inp->inp_laddr.s_addr == laddr.s_addr &&
2081                     inp->inp_fport == fport &&
2082                     inp->inp_lport == lport) {
2083                         /*
2084                          * XXX We should be able to directly return
2085                          * the inp here, without any checks.
2086                          * Well unless both bound with SO_REUSEPORT?
2087                          */
2088                         if (prison_flag(inp->inp_cred, PR_IP4))
2089                                 goto found;
2090                         if (tmpinp == NULL)
2091                                 tmpinp = inp;
2092                 }
2093         }
2094         if (tmpinp != NULL) {
2095                 inp = tmpinp;
2096                 goto found;
2097         }
2098
2099 #ifdef  RSS
2100         /*
2101          * For incoming connections, we may wish to do a wildcard
2102          * match for an RSS-local socket.
2103          */
2104         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2105                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2106 #ifdef INET6
2107                 struct inpcb *local_wild_mapped = NULL;
2108 #endif
2109                 struct inpcb *jail_wild = NULL;
2110                 struct inpcbhead *head;
2111                 int injail;
2112
2113                 /*
2114                  * Order of socket selection - we always prefer jails.
2115                  *      1. jailed, non-wild.
2116                  *      2. jailed, wild.
2117                  *      3. non-jailed, non-wild.
2118                  *      4. non-jailed, wild.
2119                  */
2120
2121                 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
2122                     lport, 0, pcbgroup->ipg_hashmask)];
2123                 CK_LIST_FOREACH(inp, head, inp_pcbgrouphash) {
2124 #ifdef INET6
2125                         /* XXX inp locking */
2126                         if ((inp->inp_vflag & INP_IPV4) == 0)
2127                                 continue;
2128 #endif
2129                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2130                             inp->inp_lport != lport)
2131                                 continue;
2132
2133                         injail = prison_flag(inp->inp_cred, PR_IP4);
2134                         if (injail) {
2135                                 if (prison_check_ip4(inp->inp_cred,
2136                                     &laddr) != 0)
2137                                         continue;
2138                         } else {
2139                                 if (local_exact != NULL)
2140                                         continue;
2141                         }
2142
2143                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2144                                 if (injail)
2145                                         goto found;
2146                                 else
2147                                         local_exact = inp;
2148                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2149 #ifdef INET6
2150                                 /* XXX inp locking, NULL check */
2151                                 if (inp->inp_vflag & INP_IPV6PROTO)
2152                                         local_wild_mapped = inp;
2153                                 else
2154 #endif
2155                                         if (injail)
2156                                                 jail_wild = inp;
2157                                         else
2158                                                 local_wild = inp;
2159                         }
2160                 } /* LIST_FOREACH */
2161
2162                 inp = jail_wild;
2163                 if (inp == NULL)
2164                         inp = local_exact;
2165                 if (inp == NULL)
2166                         inp = local_wild;
2167 #ifdef INET6
2168                 if (inp == NULL)
2169                         inp = local_wild_mapped;
2170 #endif
2171                 if (inp != NULL)
2172                         goto found;
2173         }
2174 #endif
2175
2176         /*
2177          * Then look for a wildcard match, if requested.
2178          */
2179         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2180                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2181 #ifdef INET6
2182                 struct inpcb *local_wild_mapped = NULL;
2183 #endif
2184                 struct inpcb *jail_wild = NULL;
2185                 struct inpcbhead *head;
2186                 int injail;
2187
2188                 /*
2189                  * Order of socket selection - we always prefer jails.
2190                  *      1. jailed, non-wild.
2191                  *      2. jailed, wild.
2192                  *      3. non-jailed, non-wild.
2193                  *      4. non-jailed, wild.
2194                  */
2195                 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
2196                     0, pcbinfo->ipi_wildmask)];
2197                 CK_LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
2198 #ifdef INET6
2199                         /* XXX inp locking */
2200                         if ((inp->inp_vflag & INP_IPV4) == 0)
2201                                 continue;
2202 #endif
2203                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2204                             inp->inp_lport != lport)
2205                                 continue;
2206
2207                         injail = prison_flag(inp->inp_cred, PR_IP4);
2208                         if (injail) {
2209                                 if (prison_check_ip4(inp->inp_cred,
2210                                     &laddr) != 0)
2211                                         continue;
2212                         } else {
2213                                 if (local_exact != NULL)
2214                                         continue;
2215                         }
2216
2217                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2218                                 if (injail)
2219                                         goto found;
2220                                 else
2221                                         local_exact = inp;
2222                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2223 #ifdef INET6
2224                                 /* XXX inp locking, NULL check */
2225                                 if (inp->inp_vflag & INP_IPV6PROTO)
2226                                         local_wild_mapped = inp;
2227                                 else
2228 #endif
2229                                         if (injail)
2230                                                 jail_wild = inp;
2231                                         else
2232                                                 local_wild = inp;
2233                         }
2234                 } /* LIST_FOREACH */
2235                 inp = jail_wild;
2236                 if (inp == NULL)
2237                         inp = local_exact;
2238                 if (inp == NULL)
2239                         inp = local_wild;
2240 #ifdef INET6
2241                 if (inp == NULL)
2242                         inp = local_wild_mapped;
2243 #endif
2244                 if (inp != NULL)
2245                         goto found;
2246         } /* if (lookupflags & INPLOOKUP_WILDCARD) */
2247         INP_GROUP_UNLOCK(pcbgroup);
2248         return (NULL);
2249
2250 found:
2251         if (lookupflags & INPLOOKUP_WLOCKPCB)
2252                 locked = INP_TRY_WLOCK(inp);
2253         else if (lookupflags & INPLOOKUP_RLOCKPCB)
2254                 locked = INP_TRY_RLOCK(inp);
2255         else
2256                 panic("%s: locking bug", __func__);
2257         if (__predict_false(locked && (inp->inp_flags2 & INP_FREED))) {
2258                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2259                         INP_WUNLOCK(inp);
2260                 else
2261                         INP_RUNLOCK(inp);
2262                 return (NULL);
2263         } else if (!locked)
2264                 in_pcbref(inp);
2265         INP_GROUP_UNLOCK(pcbgroup);
2266         if (!locked) {
2267                 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2268                         INP_WLOCK(inp);
2269                         if (in_pcbrele_wlocked(inp))
2270                                 return (NULL);
2271                 } else {
2272                         INP_RLOCK(inp);
2273                         if (in_pcbrele_rlocked(inp))
2274                                 return (NULL);
2275                 }
2276         }
2277 #ifdef INVARIANTS
2278         if (lookupflags & INPLOOKUP_WLOCKPCB)
2279                 INP_WLOCK_ASSERT(inp);
2280         else
2281                 INP_RLOCK_ASSERT(inp);
2282 #endif
2283         return (inp);
2284 }
2285 #endif /* PCBGROUP */
2286
2287 /*
2288  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2289  * that the caller has locked the hash list, and will not perform any further
2290  * locking or reference operations on either the hash list or the connection.
2291  */
2292 static struct inpcb *
2293 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2294     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2295     struct ifnet *ifp)
2296 {
2297         struct inpcbhead *head;
2298         struct inpcb *inp, *tmpinp;
2299         u_short fport = fport_arg, lport = lport_arg;
2300
2301         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2302             ("%s: invalid lookup flags %d", __func__, lookupflags));
2303         INP_HASH_LOCK_ASSERT(pcbinfo);
2304
2305         /*
2306          * First look for an exact match.
2307          */
2308         tmpinp = NULL;
2309         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
2310             pcbinfo->ipi_hashmask)];
2311         CK_LIST_FOREACH(inp, head, inp_hash) {
2312 #ifdef INET6
2313                 /* XXX inp locking */
2314                 if ((inp->inp_vflag & INP_IPV4) == 0)
2315                         continue;
2316 #endif
2317                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2318                     inp->inp_laddr.s_addr == laddr.s_addr &&
2319                     inp->inp_fport == fport &&
2320                     inp->inp_lport == lport) {
2321                         /*
2322                          * XXX We should be able to directly return
2323                          * the inp here, without any checks.
2324                          * Well unless both bound with SO_REUSEPORT?
2325                          */
2326                         if (prison_flag(inp->inp_cred, PR_IP4))
2327                                 return (inp);
2328                         if (tmpinp == NULL)
2329                                 tmpinp = inp;
2330                 }
2331         }
2332         if (tmpinp != NULL)
2333                 return (tmpinp);
2334
2335         /*
2336          * Then look in lb group (for wildcard match).
2337          */
2338         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2339                 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2340                     fport, lookupflags);
2341                 if (inp != NULL)
2342                         return (inp);
2343         }
2344
2345         /*
2346          * Then look for a wildcard match, if requested.
2347          */
2348         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2349                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2350 #ifdef INET6
2351                 struct inpcb *local_wild_mapped = NULL;
2352 #endif
2353                 struct inpcb *jail_wild = NULL;
2354                 int injail;
2355
2356                 /*
2357                  * Order of socket selection - we always prefer jails.
2358                  *      1. jailed, non-wild.
2359                  *      2. jailed, wild.
2360                  *      3. non-jailed, non-wild.
2361                  *      4. non-jailed, wild.
2362                  */
2363
2364                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2365                     0, pcbinfo->ipi_hashmask)];
2366                 CK_LIST_FOREACH(inp, head, inp_hash) {
2367 #ifdef INET6
2368                         /* XXX inp locking */
2369                         if ((inp->inp_vflag & INP_IPV4) == 0)
2370                                 continue;
2371 #endif
2372                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2373                             inp->inp_lport != lport)
2374                                 continue;
2375
2376                         injail = prison_flag(inp->inp_cred, PR_IP4);
2377                         if (injail) {
2378                                 if (prison_check_ip4(inp->inp_cred,
2379                                     &laddr) != 0)
2380                                         continue;
2381                         } else {
2382                                 if (local_exact != NULL)
2383                                         continue;
2384                         }
2385
2386                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2387                                 if (injail)
2388                                         return (inp);
2389                                 else
2390                                         local_exact = inp;
2391                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2392 #ifdef INET6
2393                                 /* XXX inp locking, NULL check */
2394                                 if (inp->inp_vflag & INP_IPV6PROTO)
2395                                         local_wild_mapped = inp;
2396                                 else
2397 #endif
2398                                         if (injail)
2399                                                 jail_wild = inp;
2400                                         else
2401                                                 local_wild = inp;
2402                         }
2403                 } /* LIST_FOREACH */
2404                 if (jail_wild != NULL)
2405                         return (jail_wild);
2406                 if (local_exact != NULL)
2407                         return (local_exact);
2408                 if (local_wild != NULL)
2409                         return (local_wild);
2410 #ifdef INET6
2411                 if (local_wild_mapped != NULL)
2412                         return (local_wild_mapped);
2413 #endif
2414         } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2415
2416         return (NULL);
2417 }
2418
2419 /*
2420  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2421  * hash list lock, and will return the inpcb locked (i.e., requires
2422  * INPLOOKUP_LOCKPCB).
2423  */
2424 static struct inpcb *
2425 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2426     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2427     struct ifnet *ifp)
2428 {
2429         struct inpcb *inp;
2430
2431         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2432             (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2433         if (inp != NULL) {
2434                 if (lookupflags & INPLOOKUP_WLOCKPCB) {
2435                         INP_WLOCK(inp);
2436                         if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2437                                 INP_WUNLOCK(inp);
2438                                 inp = NULL;
2439                         }
2440                 } else if (lookupflags & INPLOOKUP_RLOCKPCB) {
2441                         INP_RLOCK(inp);
2442                         if (__predict_false(inp->inp_flags2 & INP_FREED)) {
2443                                 INP_RUNLOCK(inp);
2444                                 inp = NULL;
2445                         }
2446                 } else
2447                         panic("%s: locking bug", __func__);
2448 #ifdef INVARIANTS
2449                 if (inp != NULL) {
2450                         if (lookupflags & INPLOOKUP_WLOCKPCB)
2451                                 INP_WLOCK_ASSERT(inp);
2452                         else
2453                                 INP_RLOCK_ASSERT(inp);
2454                 }
2455 #endif
2456         }
2457
2458         return (inp);
2459 }
2460
2461 /*
2462  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2463  * from which a pre-calculated hash value may be extracted.
2464  *
2465  * Possibly more of this logic should be in in_pcbgroup.c.
2466  */
2467 struct inpcb *
2468 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2469     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2470 {
2471 #if defined(PCBGROUP) && !defined(RSS)
2472         struct inpcbgroup *pcbgroup;
2473 #endif
2474
2475         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2476             ("%s: invalid lookup flags %d", __func__, lookupflags));
2477         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2478             ("%s: LOCKPCB not set", __func__));
2479
2480         /*
2481          * When not using RSS, use connection groups in preference to the
2482          * reservation table when looking up 4-tuples.  When using RSS, just
2483          * use the reservation table, due to the cost of the Toeplitz hash
2484          * in software.
2485          *
2486          * XXXRW: This policy belongs in the pcbgroup code, as in principle
2487          * we could be doing RSS with a non-Toeplitz hash that is affordable
2488          * in software.
2489          */
2490 #if defined(PCBGROUP) && !defined(RSS)
2491         if (in_pcbgroup_enabled(pcbinfo)) {
2492                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2493                     fport);
2494                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2495                     laddr, lport, lookupflags, ifp));
2496         }
2497 #endif
2498         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2499             lookupflags, ifp));
2500 }
2501
2502 struct inpcb *
2503 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2504     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2505     struct ifnet *ifp, struct mbuf *m)
2506 {
2507 #ifdef PCBGROUP
2508         struct inpcbgroup *pcbgroup;
2509 #endif
2510
2511         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2512             ("%s: invalid lookup flags %d", __func__, lookupflags));
2513         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2514             ("%s: LOCKPCB not set", __func__));
2515
2516 #ifdef PCBGROUP
2517         /*
2518          * If we can use a hardware-generated hash to look up the connection
2519          * group, use that connection group to find the inpcb.  Otherwise
2520          * fall back on a software hash -- or the reservation table if we're
2521          * using RSS.
2522          *
2523          * XXXRW: As above, that policy belongs in the pcbgroup code.
2524          */
2525         if (in_pcbgroup_enabled(pcbinfo) &&
2526             !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2527                 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2528                     m->m_pkthdr.flowid);
2529                 if (pcbgroup != NULL)
2530                         return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2531                             fport, laddr, lport, lookupflags, ifp));
2532 #ifndef RSS
2533                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2534                     fport);
2535                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2536                     laddr, lport, lookupflags, ifp));
2537 #endif
2538         }
2539 #endif
2540         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2541             lookupflags, ifp));
2542 }
2543 #endif /* INET */
2544
2545 /*
2546  * Insert PCB onto various hash lists.
2547  */
2548 static int
2549 in_pcbinshash_internal(struct inpcb *inp, struct mbuf *m)
2550 {
2551         struct inpcbhead *pcbhash;
2552         struct inpcbporthead *pcbporthash;
2553         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2554         struct inpcbport *phd;
2555         u_int32_t hashkey_faddr;
2556         int so_options;
2557
2558         INP_WLOCK_ASSERT(inp);
2559         INP_HASH_WLOCK_ASSERT(pcbinfo);
2560
2561         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2562             ("in_pcbinshash: INP_INHASHLIST"));
2563
2564 #ifdef INET6
2565         if (inp->inp_vflag & INP_IPV6)
2566                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2567         else
2568 #endif
2569         hashkey_faddr = inp->inp_faddr.s_addr;
2570
2571         pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2572                  inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2573
2574         pcbporthash = &pcbinfo->ipi_porthashbase[
2575             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2576
2577         /*
2578          * Add entry to load balance group.
2579          * Only do this if SO_REUSEPORT_LB is set.
2580          */
2581         so_options = inp_so_options(inp);
2582         if (so_options & SO_REUSEPORT_LB) {
2583                 int ret = in_pcbinslbgrouphash(inp);
2584                 if (ret) {
2585                         /* pcb lb group malloc fail (ret=ENOBUFS). */
2586                         return (ret);
2587                 }
2588         }
2589
2590         /*
2591          * Go through port list and look for a head for this lport.
2592          */
2593         CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2594                 if (phd->phd_port == inp->inp_lport)
2595                         break;
2596         }
2597         /*
2598          * If none exists, malloc one and tack it on.
2599          */
2600         if (phd == NULL) {
2601                 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2602                 if (phd == NULL) {
2603                         return (ENOBUFS); /* XXX */
2604                 }
2605                 bzero(&phd->phd_epoch_ctx, sizeof(struct epoch_context));
2606                 phd->phd_port = inp->inp_lport;
2607                 CK_LIST_INIT(&phd->phd_pcblist);
2608                 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2609         }
2610         inp->inp_phd = phd;
2611         CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2612         CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2613         inp->inp_flags |= INP_INHASHLIST;
2614 #ifdef PCBGROUP
2615         if (m != NULL) {
2616                 in_pcbgroup_update_mbuf(inp, m);
2617         } else {
2618                 in_pcbgroup_update(inp);
2619         }
2620 #endif
2621         return (0);
2622 }
2623
2624 int
2625 in_pcbinshash(struct inpcb *inp)
2626 {
2627
2628         return (in_pcbinshash_internal(inp, NULL));
2629 }
2630
2631 int
2632 in_pcbinshash_mbuf(struct inpcb *inp, struct mbuf *m)
2633 {
2634
2635         return (in_pcbinshash_internal(inp, m));
2636 }
2637
2638 /*
2639  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2640  * changed. NOTE: This does not handle the case of the lport changing (the
2641  * hashed port list would have to be updated as well), so the lport must
2642  * not change after in_pcbinshash() has been called.
2643  */
2644 void
2645 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2646 {
2647         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2648         struct inpcbhead *head;
2649         u_int32_t hashkey_faddr;
2650
2651         INP_WLOCK_ASSERT(inp);
2652         INP_HASH_WLOCK_ASSERT(pcbinfo);
2653
2654         KASSERT(inp->inp_flags & INP_INHASHLIST,
2655             ("in_pcbrehash: !INP_INHASHLIST"));
2656
2657 #ifdef INET6
2658         if (inp->inp_vflag & INP_IPV6)
2659                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2660         else
2661 #endif
2662         hashkey_faddr = inp->inp_faddr.s_addr;
2663
2664         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2665                 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2666
2667         CK_LIST_REMOVE(inp, inp_hash);
2668         CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2669
2670 #ifdef PCBGROUP
2671         if (m != NULL)
2672                 in_pcbgroup_update_mbuf(inp, m);
2673         else
2674                 in_pcbgroup_update(inp);
2675 #endif
2676 }
2677
2678 void
2679 in_pcbrehash(struct inpcb *inp)
2680 {
2681
2682         in_pcbrehash_mbuf(inp, NULL);
2683 }
2684
2685 /*
2686  * Remove PCB from various lists.
2687  */
2688 static void
2689 in_pcbremlists(struct inpcb *inp)
2690 {
2691         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2692
2693         INP_WLOCK_ASSERT(inp);
2694         INP_LIST_WLOCK_ASSERT(pcbinfo);
2695
2696         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2697         if (inp->inp_flags & INP_INHASHLIST) {
2698                 struct inpcbport *phd = inp->inp_phd;
2699
2700                 INP_HASH_WLOCK(pcbinfo);
2701
2702                 /* XXX: Only do if SO_REUSEPORT_LB set? */
2703                 in_pcbremlbgrouphash(inp);
2704
2705                 CK_LIST_REMOVE(inp, inp_hash);
2706                 CK_LIST_REMOVE(inp, inp_portlist);
2707                 if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2708                         CK_LIST_REMOVE(phd, phd_hash);
2709                         NET_EPOCH_CALL(inpcbport_free, &phd->phd_epoch_ctx);
2710                 }
2711                 INP_HASH_WUNLOCK(pcbinfo);
2712                 inp->inp_flags &= ~INP_INHASHLIST;
2713         }
2714         CK_LIST_REMOVE(inp, inp_list);
2715         pcbinfo->ipi_count--;
2716 #ifdef PCBGROUP
2717         in_pcbgroup_remove(inp);
2718 #endif
2719 }
2720
2721 /*
2722  * Check for alternatives when higher level complains
2723  * about service problems.  For now, invalidate cached
2724  * routing information.  If the route was created dynamically
2725  * (by a redirect), time to try a default gateway again.
2726  */
2727 void
2728 in_losing(struct inpcb *inp)
2729 {
2730
2731         RO_INVALIDATE_CACHE(&inp->inp_route);
2732         return;
2733 }
2734
2735 /*
2736  * A set label operation has occurred at the socket layer, propagate the
2737  * label change into the in_pcb for the socket.
2738  */
2739 void
2740 in_pcbsosetlabel(struct socket *so)
2741 {
2742 #ifdef MAC
2743         struct inpcb *inp;
2744
2745         inp = sotoinpcb(so);
2746         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2747
2748         INP_WLOCK(inp);
2749         SOCK_LOCK(so);
2750         mac_inpcb_sosetlabel(so, inp);
2751         SOCK_UNLOCK(so);
2752         INP_WUNLOCK(inp);
2753 #endif
2754 }
2755
2756 /*
2757  * ipport_tick runs once per second, determining if random port allocation
2758  * should be continued.  If more than ipport_randomcps ports have been
2759  * allocated in the last second, then we return to sequential port
2760  * allocation. We return to random allocation only once we drop below
2761  * ipport_randomcps for at least ipport_randomtime seconds.
2762  */
2763 static void
2764 ipport_tick(void *xtp)
2765 {
2766         VNET_ITERATOR_DECL(vnet_iter);
2767
2768         VNET_LIST_RLOCK_NOSLEEP();
2769         VNET_FOREACH(vnet_iter) {
2770                 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2771                 if (V_ipport_tcpallocs <=
2772                     V_ipport_tcplastcount + V_ipport_randomcps) {
2773                         if (V_ipport_stoprandom > 0)
2774                                 V_ipport_stoprandom--;
2775                 } else
2776                         V_ipport_stoprandom = V_ipport_randomtime;
2777                 V_ipport_tcplastcount = V_ipport_tcpallocs;
2778                 CURVNET_RESTORE();
2779         }
2780         VNET_LIST_RUNLOCK_NOSLEEP();
2781         callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2782 }
2783
2784 static void
2785 ip_fini(void *xtp)
2786 {
2787
2788         callout_stop(&ipport_tick_callout);
2789 }
2790
2791 /*
2792  * The ipport_callout should start running at about the time we attach the
2793  * inet or inet6 domains.
2794  */
2795 static void
2796 ipport_tick_init(const void *unused __unused)
2797 {
2798
2799         /* Start ipport_tick. */
2800         callout_init(&ipport_tick_callout, 1);
2801         callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2802         EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2803                 SHUTDOWN_PRI_DEFAULT);
2804 }
2805 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2806     ipport_tick_init, NULL);
2807
2808 void
2809 inp_wlock(struct inpcb *inp)
2810 {
2811
2812         INP_WLOCK(inp);
2813 }
2814
2815 void
2816 inp_wunlock(struct inpcb *inp)
2817 {
2818
2819         INP_WUNLOCK(inp);
2820 }
2821
2822 void
2823 inp_rlock(struct inpcb *inp)
2824 {
2825
2826         INP_RLOCK(inp);
2827 }
2828
2829 void
2830 inp_runlock(struct inpcb *inp)
2831 {
2832
2833         INP_RUNLOCK(inp);
2834 }
2835
2836 #ifdef INVARIANT_SUPPORT
2837 void
2838 inp_lock_assert(struct inpcb *inp)
2839 {
2840
2841         INP_WLOCK_ASSERT(inp);
2842 }
2843
2844 void
2845 inp_unlock_assert(struct inpcb *inp)
2846 {
2847
2848         INP_UNLOCK_ASSERT(inp);
2849 }
2850 #endif
2851
2852 void
2853 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2854 {
2855         struct inpcb *inp;
2856
2857         INP_INFO_WLOCK(&V_tcbinfo);
2858         CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2859                 INP_WLOCK(inp);
2860                 func(inp, arg);
2861                 INP_WUNLOCK(inp);
2862         }
2863         INP_INFO_WUNLOCK(&V_tcbinfo);
2864 }
2865
2866 struct socket *
2867 inp_inpcbtosocket(struct inpcb *inp)
2868 {
2869
2870         INP_WLOCK_ASSERT(inp);
2871         return (inp->inp_socket);
2872 }
2873
2874 struct tcpcb *
2875 inp_inpcbtotcpcb(struct inpcb *inp)
2876 {
2877
2878         INP_WLOCK_ASSERT(inp);
2879         return ((struct tcpcb *)inp->inp_ppcb);
2880 }
2881
2882 int
2883 inp_ip_tos_get(const struct inpcb *inp)
2884 {
2885
2886         return (inp->inp_ip_tos);
2887 }
2888
2889 void
2890 inp_ip_tos_set(struct inpcb *inp, int val)
2891 {
2892
2893         inp->inp_ip_tos = val;
2894 }
2895
2896 void
2897 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2898     uint32_t *faddr, uint16_t *fp)
2899 {
2900
2901         INP_LOCK_ASSERT(inp);
2902         *laddr = inp->inp_laddr.s_addr;
2903         *faddr = inp->inp_faddr.s_addr;
2904         *lp = inp->inp_lport;
2905         *fp = inp->inp_fport;
2906 }
2907
2908 struct inpcb *
2909 so_sotoinpcb(struct socket *so)
2910 {
2911
2912         return (sotoinpcb(so));
2913 }
2914
2915 struct tcpcb *
2916 so_sototcpcb(struct socket *so)
2917 {
2918
2919         return (sototcpcb(so));
2920 }
2921
2922 /*
2923  * Create an external-format (``xinpcb'') structure using the information in
2924  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2925  * reduce the spew of irrelevant information over this interface, to isolate
2926  * user code from changes in the kernel structure, and potentially to provide
2927  * information-hiding if we decide that some of this information should be
2928  * hidden from users.
2929  */
2930 void
2931 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2932 {
2933
2934         bzero(xi, sizeof(*xi));
2935         xi->xi_len = sizeof(struct xinpcb);
2936         if (inp->inp_socket)
2937                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2938         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2939         xi->inp_gencnt = inp->inp_gencnt;
2940         xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2941         xi->inp_flow = inp->inp_flow;
2942         xi->inp_flowid = inp->inp_flowid;
2943         xi->inp_flowtype = inp->inp_flowtype;
2944         xi->inp_flags = inp->inp_flags;
2945         xi->inp_flags2 = inp->inp_flags2;
2946         xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2947         xi->in6p_cksum = inp->in6p_cksum;
2948         xi->in6p_hops = inp->in6p_hops;
2949         xi->inp_ip_tos = inp->inp_ip_tos;
2950         xi->inp_vflag = inp->inp_vflag;
2951         xi->inp_ip_ttl = inp->inp_ip_ttl;
2952         xi->inp_ip_p = inp->inp_ip_p;
2953         xi->inp_ip_minttl = inp->inp_ip_minttl;
2954 }
2955
2956 #ifdef DDB
2957 static void
2958 db_print_indent(int indent)
2959 {
2960         int i;
2961
2962         for (i = 0; i < indent; i++)
2963                 db_printf(" ");
2964 }
2965
2966 static void
2967 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2968 {
2969         char faddr_str[48], laddr_str[48];
2970
2971         db_print_indent(indent);
2972         db_printf("%s at %p\n", name, inc);
2973
2974         indent += 2;
2975
2976 #ifdef INET6
2977         if (inc->inc_flags & INC_ISIPV6) {
2978                 /* IPv6. */
2979                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2980                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2981         } else
2982 #endif
2983         {
2984                 /* IPv4. */
2985                 inet_ntoa_r(inc->inc_laddr, laddr_str);
2986                 inet_ntoa_r(inc->inc_faddr, faddr_str);
2987         }
2988         db_print_indent(indent);
2989         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2990             ntohs(inc->inc_lport));
2991         db_print_indent(indent);
2992         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2993             ntohs(inc->inc_fport));
2994 }
2995
2996 static void
2997 db_print_inpflags(int inp_flags)
2998 {
2999         int comma;
3000
3001         comma = 0;
3002         if (inp_flags & INP_RECVOPTS) {
3003                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3004                 comma = 1;
3005         }
3006         if (inp_flags & INP_RECVRETOPTS) {
3007                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3008                 comma = 1;
3009         }
3010         if (inp_flags & INP_RECVDSTADDR) {
3011                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3012                 comma = 1;
3013         }
3014         if (inp_flags & INP_ORIGDSTADDR) {
3015                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3016                 comma = 1;
3017         }
3018         if (inp_flags & INP_HDRINCL) {
3019                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3020                 comma = 1;
3021         }
3022         if (inp_flags & INP_HIGHPORT) {
3023                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3024                 comma = 1;
3025         }
3026         if (inp_flags & INP_LOWPORT) {
3027                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3028                 comma = 1;
3029         }
3030         if (inp_flags & INP_ANONPORT) {
3031                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3032                 comma = 1;
3033         }
3034         if (inp_flags & INP_RECVIF) {
3035                 db_printf("%sINP_RECVIF", comma ? ", " : "");
3036                 comma = 1;
3037         }
3038         if (inp_flags & INP_MTUDISC) {
3039                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3040                 comma = 1;
3041         }
3042         if (inp_flags & INP_RECVTTL) {
3043                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3044                 comma = 1;
3045         }
3046         if (inp_flags & INP_DONTFRAG) {
3047                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3048                 comma = 1;
3049         }
3050         if (inp_flags & INP_RECVTOS) {
3051                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3052                 comma = 1;
3053         }
3054         if (inp_flags & IN6P_IPV6_V6ONLY) {
3055                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3056                 comma = 1;
3057         }
3058         if (inp_flags & IN6P_PKTINFO) {
3059                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3060                 comma = 1;
3061         }
3062         if (inp_flags & IN6P_HOPLIMIT) {
3063                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3064                 comma = 1;
3065         }
3066         if (inp_flags & IN6P_HOPOPTS) {
3067                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3068                 comma = 1;
3069         }
3070         if (inp_flags & IN6P_DSTOPTS) {
3071                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3072                 comma = 1;
3073         }
3074         if (inp_flags & IN6P_RTHDR) {
3075                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3076                 comma = 1;
3077         }
3078         if (inp_flags & IN6P_RTHDRDSTOPTS) {
3079                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3080                 comma = 1;
3081         }
3082         if (inp_flags & IN6P_TCLASS) {
3083                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3084                 comma = 1;
3085         }
3086         if (inp_flags & IN6P_AUTOFLOWLABEL) {
3087                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3088                 comma = 1;
3089         }
3090         if (inp_flags & INP_TIMEWAIT) {
3091                 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
3092                 comma  = 1;
3093         }
3094         if (inp_flags & INP_ONESBCAST) {
3095                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3096                 comma  = 1;
3097         }
3098         if (inp_flags & INP_DROPPED) {
3099                 db_printf("%sINP_DROPPED", comma ? ", " : "");
3100                 comma  = 1;
3101         }
3102         if (inp_flags & INP_SOCKREF) {
3103                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3104                 comma  = 1;
3105         }
3106         if (inp_flags & IN6P_RFC2292) {
3107                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3108                 comma = 1;
3109         }
3110         if (inp_flags & IN6P_MTU) {
3111                 db_printf("IN6P_MTU%s", comma ? ", " : "");
3112                 comma = 1;
3113         }
3114 }
3115
3116 static void
3117 db_print_inpvflag(u_char inp_vflag)
3118 {
3119         int comma;
3120
3121         comma = 0;
3122         if (inp_vflag & INP_IPV4) {
3123                 db_printf("%sINP_IPV4", comma ? ", " : "");
3124                 comma  = 1;
3125         }
3126         if (inp_vflag & INP_IPV6) {
3127                 db_printf("%sINP_IPV6", comma ? ", " : "");
3128                 comma  = 1;
3129         }
3130         if (inp_vflag & INP_IPV6PROTO) {
3131                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3132                 comma  = 1;
3133         }
3134 }
3135
3136 static void
3137 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3138 {
3139
3140         db_print_indent(indent);
3141         db_printf("%s at %p\n", name, inp);
3142
3143         indent += 2;
3144
3145         db_print_indent(indent);
3146         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3147
3148         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3149
3150         db_print_indent(indent);
3151         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3152             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3153
3154         db_print_indent(indent);
3155         db_printf("inp_label: %p   inp_flags: 0x%x (",
3156            inp->inp_label, inp->inp_flags);
3157         db_print_inpflags(inp->inp_flags);
3158         db_printf(")\n");
3159
3160         db_print_indent(indent);
3161         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3162             inp->inp_vflag);
3163         db_print_inpvflag(inp->inp_vflag);
3164         db_printf(")\n");
3165
3166         db_print_indent(indent);
3167         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3168             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3169
3170         db_print_indent(indent);
3171 #ifdef INET6
3172         if (inp->inp_vflag & INP_IPV6) {
3173                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
3174                     "in6p_moptions: %p\n", inp->in6p_options,
3175                     inp->in6p_outputopts, inp->in6p_moptions);
3176                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3177                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3178                     inp->in6p_hops);
3179         } else
3180 #endif
3181         {
3182                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3183                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3184                     inp->inp_options, inp->inp_moptions);
3185         }
3186
3187         db_print_indent(indent);
3188         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3189             (uintmax_t)inp->inp_gencnt);
3190 }
3191
3192 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3193 {
3194         struct inpcb *inp;
3195
3196         if (!have_addr) {
3197                 db_printf("usage: show inpcb <addr>\n");
3198                 return;
3199         }
3200         inp = (struct inpcb *)addr;
3201
3202         db_print_inpcb(inp, "inpcb", 0);
3203 }
3204 #endif /* DDB */
3205
3206 #ifdef RATELIMIT
3207 /*
3208  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3209  * if any.
3210  */
3211 int
3212 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3213 {
3214         union if_snd_tag_modify_params params = {
3215                 .rate_limit.max_rate = max_pacing_rate,
3216                 .rate_limit.flags = M_NOWAIT,
3217         };
3218         struct m_snd_tag *mst;
3219         struct ifnet *ifp;
3220         int error;
3221
3222         mst = inp->inp_snd_tag;
3223         if (mst == NULL)
3224                 return (EINVAL);
3225
3226         ifp = mst->ifp;
3227         if (ifp == NULL)
3228                 return (EINVAL);
3229
3230         if (ifp->if_snd_tag_modify == NULL) {
3231                 error = EOPNOTSUPP;
3232         } else {
3233                 error = ifp->if_snd_tag_modify(mst, &params);
3234         }
3235         return (error);
3236 }
3237
3238 /*
3239  * Query existing TX rate limit based on the existing
3240  * "inp->inp_snd_tag", if any.
3241  */
3242 int
3243 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3244 {
3245         union if_snd_tag_query_params params = { };
3246         struct m_snd_tag *mst;
3247         struct ifnet *ifp;
3248         int error;
3249
3250         mst = inp->inp_snd_tag;
3251         if (mst == NULL)
3252                 return (EINVAL);
3253
3254         ifp = mst->ifp;
3255         if (ifp == NULL)
3256                 return (EINVAL);
3257
3258         if (ifp->if_snd_tag_query == NULL) {
3259                 error = EOPNOTSUPP;
3260         } else {
3261                 error = ifp->if_snd_tag_query(mst, &params);
3262                 if (error == 0 &&  p_max_pacing_rate != NULL)
3263                         *p_max_pacing_rate = params.rate_limit.max_rate;
3264         }
3265         return (error);
3266 }
3267
3268 /*
3269  * Query existing TX queue level based on the existing
3270  * "inp->inp_snd_tag", if any.
3271  */
3272 int
3273 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3274 {
3275         union if_snd_tag_query_params params = { };
3276         struct m_snd_tag *mst;
3277         struct ifnet *ifp;
3278         int error;
3279
3280         mst = inp->inp_snd_tag;
3281         if (mst == NULL)
3282                 return (EINVAL);
3283
3284         ifp = mst->ifp;
3285         if (ifp == NULL)
3286                 return (EINVAL);
3287
3288         if (ifp->if_snd_tag_query == NULL)
3289                 return (EOPNOTSUPP);
3290
3291         error = ifp->if_snd_tag_query(mst, &params);
3292         if (error == 0 &&  p_txqueue_level != NULL)
3293                 *p_txqueue_level = params.rate_limit.queue_level;
3294         return (error);
3295 }
3296
3297 /*
3298  * Allocate a new TX rate limit send tag from the network interface
3299  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3300  */
3301 int
3302 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3303     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3304
3305 {
3306         union if_snd_tag_alloc_params params = {
3307                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3308                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3309                 .rate_limit.hdr.flowid = flowid,
3310                 .rate_limit.hdr.flowtype = flowtype,
3311                 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3312                 .rate_limit.max_rate = max_pacing_rate,
3313                 .rate_limit.flags = M_NOWAIT,
3314         };
3315         int error;
3316
3317         INP_WLOCK_ASSERT(inp);
3318
3319         if (*st != NULL)
3320                 return (EINVAL);
3321
3322         if (ifp->if_snd_tag_alloc == NULL) {
3323                 error = EOPNOTSUPP;
3324         } else {
3325                 error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
3326
3327 #ifdef INET
3328                 if (error == 0) {
3329                         counter_u64_add(rate_limit_set_ok, 1);
3330                         counter_u64_add(rate_limit_active, 1);
3331                 } else
3332                         counter_u64_add(rate_limit_alloc_fail, 1);
3333 #endif
3334         }
3335         return (error);
3336 }
3337
3338 void
3339 in_pcbdetach_tag(struct ifnet *ifp, struct m_snd_tag *mst)
3340 {
3341         if (ifp == NULL)
3342                 return;
3343
3344         /*
3345          * If the device was detached while we still had reference(s)
3346          * on the ifp, we assume if_snd_tag_free() was replaced with
3347          * stubs.
3348          */
3349         ifp->if_snd_tag_free(mst);
3350
3351         /* release reference count on network interface */
3352         if_rele(ifp);
3353 #ifdef INET
3354         counter_u64_add(rate_limit_active, -1);
3355 #endif
3356 }
3357
3358 /*
3359  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3360  * if any:
3361  */
3362 void
3363 in_pcbdetach_txrtlmt(struct inpcb *inp)
3364 {
3365         struct m_snd_tag *mst;
3366
3367         INP_WLOCK_ASSERT(inp);
3368
3369         mst = inp->inp_snd_tag;
3370         inp->inp_snd_tag = NULL;
3371
3372         if (mst == NULL)
3373                 return;
3374
3375         m_snd_tag_rele(mst);
3376 }
3377
3378 int
3379 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3380 {
3381         int error;
3382
3383         /*
3384          * If the existing send tag is for the wrong interface due to
3385          * a route change, first drop the existing tag.  Set the
3386          * CHANGED flag so that we will keep trying to allocate a new
3387          * tag if we fail to allocate one this time.
3388          */
3389         if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3390                 in_pcbdetach_txrtlmt(inp);
3391                 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3392         }
3393
3394         /*
3395          * NOTE: When attaching to a network interface a reference is
3396          * made to ensure the network interface doesn't go away until
3397          * all ratelimit connections are gone. The network interface
3398          * pointers compared below represent valid network interfaces,
3399          * except when comparing towards NULL.
3400          */
3401         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3402                 error = 0;
3403         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3404                 if (inp->inp_snd_tag != NULL)
3405                         in_pcbdetach_txrtlmt(inp);
3406                 error = 0;
3407         } else if (inp->inp_snd_tag == NULL) {
3408                 /*
3409                  * In order to utilize packet pacing with RSS, we need
3410                  * to wait until there is a valid RSS hash before we
3411                  * can proceed:
3412                  */
3413                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3414                         error = EAGAIN;
3415                 } else {
3416                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3417                             mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3418                 }
3419         } else {
3420                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3421         }
3422         if (error == 0 || error == EOPNOTSUPP)
3423                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3424
3425         return (error);
3426 }
3427
3428 /*
3429  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3430  * is set in the fast path and will attach/detach/modify the TX rate
3431  * limit send tag based on the socket's so_max_pacing_rate value.
3432  */
3433 void
3434 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3435 {
3436         struct socket *socket;
3437         uint32_t max_pacing_rate;
3438         bool did_upgrade;
3439         int error;
3440
3441         if (inp == NULL)
3442                 return;
3443
3444         socket = inp->inp_socket;
3445         if (socket == NULL)
3446                 return;
3447
3448         if (!INP_WLOCKED(inp)) {
3449                 /*
3450                  * NOTE: If the write locking fails, we need to bail
3451                  * out and use the non-ratelimited ring for the
3452                  * transmit until there is a new chance to get the
3453                  * write lock.
3454                  */
3455                 if (!INP_TRY_UPGRADE(inp))
3456                         return;
3457                 did_upgrade = 1;
3458         } else {
3459                 did_upgrade = 0;
3460         }
3461
3462         /*
3463          * NOTE: The so_max_pacing_rate value is read unlocked,
3464          * because atomic updates are not required since the variable
3465          * is checked at every mbuf we send. It is assumed that the
3466          * variable read itself will be atomic.
3467          */
3468         max_pacing_rate = socket->so_max_pacing_rate;
3469
3470         error = in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3471
3472         if (did_upgrade)
3473                 INP_DOWNGRADE(inp);
3474 }
3475
3476 /*
3477  * Track route changes for TX rate limiting.
3478  */
3479 void
3480 in_pcboutput_eagain(struct inpcb *inp)
3481 {
3482         bool did_upgrade;
3483
3484         if (inp == NULL)
3485                 return;
3486
3487         if (inp->inp_snd_tag == NULL)
3488                 return;
3489
3490         if (!INP_WLOCKED(inp)) {
3491                 /*
3492                  * NOTE: If the write locking fails, we need to bail
3493                  * out and use the non-ratelimited ring for the
3494                  * transmit until there is a new chance to get the
3495                  * write lock.
3496                  */
3497                 if (!INP_TRY_UPGRADE(inp))
3498                         return;
3499                 did_upgrade = 1;
3500         } else {
3501                 did_upgrade = 0;
3502         }
3503
3504         /* detach rate limiting */
3505         in_pcbdetach_txrtlmt(inp);
3506
3507         /* make sure new mbuf send tag allocation is made */
3508         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3509
3510         if (did_upgrade)
3511                 INP_DOWNGRADE(inp);
3512 }
3513
3514 #ifdef INET
3515 static void
3516 rl_init(void *st)
3517 {
3518         rate_limit_active = counter_u64_alloc(M_WAITOK);
3519         rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3520         rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3521 }
3522
3523 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3524 #endif
3525 #endif /* RATELIMIT */