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  * Copyright (c) 2021-2022 Gleb Smirnoff <glebius@FreeBSD.org>
   9  * All rights reserved.
  10  *
  11  * Portions of this software were developed by Robert N. M. Watson under
  12  * contract to Juniper Networks, Inc.
  13  *
  14  * Redistribution and use in source and binary forms, with or without
  15  * modification, are permitted provided that the following conditions
  16  * are met:
  17  * 1. Redistributions of source code must retain the above copyright
  18  *    notice, this list of conditions and the following disclaimer.
  19  * 2. Redistributions in binary form must reproduce the above copyright
  20  *    notice, this list of conditions and the following disclaimer in the
  21  *    documentation and/or other materials provided with the distribution.
  22  * 3. Neither the name of the University nor the names of its contributors
  23  *    may be used to endorse or promote products derived from this software
  24  *    without specific prior written permission.
  25  *
  26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  36  * SUCH DAMAGE.
  37  *
  38  *      @(#)in_pcb.c    8.4 (Berkeley) 5/24/95
  39  */
  40
  41 #include <sys/cdefs.h>
  42 __FBSDID("$FreeBSD$");
  43
  44 #include "opt_ddb.h"
  45 #include "opt_ipsec.h"
  46 #include "opt_inet.h"
  47 #include "opt_inet6.h"
  48 #include "opt_ratelimit.h"
  49 #include "opt_route.h"
  50 #include "opt_rss.h"
  51
  52 #include <sys/param.h>
  53 #include <sys/hash.h>
  54 #include <sys/systm.h>
  55 #include <sys/libkern.h>
  56 #include <sys/lock.h>
  57 #include <sys/malloc.h>
  58 #include <sys/mbuf.h>
  59 #include <sys/eventhandler.h>
  60 #include <sys/domain.h>
  61 #include <sys/proc.h>
  62 #include <sys/protosw.h>
  63 #include <sys/smp.h>
  64 #include <sys/smr.h>
  65 #include <sys/socket.h>
  66 #include <sys/socketvar.h>
  67 #include <sys/sockio.h>
  68 #include <sys/priv.h>
  69 #include <sys/proc.h>
  70 #include <sys/refcount.h>
  71 #include <sys/jail.h>
  72 #include <sys/kernel.h>
  73 #include <sys/sysctl.h>
  74
  75 #ifdef DDB
  76 #include <ddb/ddb.h>
  77 #endif
  78
  79 #include <vm/uma.h>
  80 #include <vm/vm.h>
  81
  82 #include <net/if.h>
  83 #include <net/if_var.h>
  84 #include <net/if_private.h>
  85 #include <net/if_types.h>
  86 #include <net/if_llatbl.h>
  87 #include <net/route.h>
  88 #include <net/rss_config.h>
  89 #include <net/vnet.h>
  90
  91 #if defined(INET) || defined(INET6)
  92 #include <netinet/in.h>
  93 #include <netinet/in_pcb.h>
  94 #include <netinet/in_pcb_var.h>
  95 #include <netinet/tcp.h>
  96 #ifdef INET
  97 #include <netinet/in_var.h>
  98 #include <netinet/in_fib.h>
  99 #endif
 100 #include <netinet/ip_var.h>
 101 #ifdef INET6
 102 #include <netinet/ip6.h>
 103 #include <netinet6/in6_pcb.h>
 104 #include <netinet6/in6_var.h>
 105 #include <netinet6/ip6_var.h>
 106 #endif /* INET6 */
 107 #include <net/route/nhop.h>
 108 #endif
 109
 110 #include <netipsec/ipsec_support.h>
 111
 112 #include <security/mac/mac_framework.h>
 113
 114 #define INPCBLBGROUP_SIZMIN     8
 115 #define INPCBLBGROUP_SIZMAX     256
 116 #define INP_FREED       0x00000200      /* See in_pcb.h. */
 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 /* Enable random ephemeral port allocation by default. */
 138 VNET_DEFINE(int, ipport_randomized) = 1;
 139
 140 #ifdef INET
 141 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 142                             struct in_addr faddr, u_int fport_arg,
 143                             struct in_addr laddr, u_int lport_arg,
 144                             int lookupflags, uint8_t numa_domain);
 145
 146 #define RANGECHK(var, min, max) \
 147         if ((var) < (min)) { (var) = (min); } \
 148         else if ((var) > (max)) { (var) = (max); }
 149
 150 static int
 151 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 152 {
 153         int error;
 154
 155         error = sysctl_handle_int(oidp, arg1, arg2, req);
 156         if (error == 0) {
 157                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 158                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 159                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 160                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 161                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 162                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 163         }
 164         return (error);
 165 }
 166
 167 #undef RANGECHK
 168
 169 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
 170     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 171     "IP Ports");
 172
 173 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 174     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 175     &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
 176     "");
 177 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 178     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 179     &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
 180     "");
 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 182     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 183     &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
 184     "");
 185 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 186     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 187     &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
 188     "");
 189 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 190     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 191     &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
 192     "");
 193 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 194     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 195     &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
 196     "");
 197 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 198         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 199         &VNET_NAME(ipport_reservedhigh), 0, "");
 200 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 201         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 202 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 203         CTLFLAG_VNET | CTLFLAG_RW,
 204         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 205
 206 #ifdef RATELIMIT
 207 counter_u64_t rate_limit_new;
 208 counter_u64_t rate_limit_chg;
 209 counter_u64_t rate_limit_active;
 210 counter_u64_t rate_limit_alloc_fail;
 211 counter_u64_t rate_limit_set_ok;
 212
 213 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 214     "IP Rate Limiting");
 215 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
 216     &rate_limit_active, "Active rate limited connections");
 217 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
 218    &rate_limit_alloc_fail, "Rate limited connection failures");
 219 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
 220    &rate_limit_set_ok, "Rate limited setting succeeded");
 221 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD,
 222    &rate_limit_new, "Total Rate limit new attempts");
 223 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD,
 224    &rate_limit_chg, "Total Rate limited change attempts");
 225 #endif /* RATELIMIT */
 226
 227 #endif /* INET */
 228
 229 VNET_DEFINE(uint32_t, in_pcbhashseed);
 230 static void
 231 in_pcbhashseed_init(void)
 232 {
 233
 234         V_in_pcbhashseed = arc4random();
 235 }
 236 VNET_SYSINIT(in_pcbhashseed_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
 237     in_pcbhashseed_init, 0);
 238
 239 static void in_pcbremhash(struct inpcb *);
 240
 241 /*
 242  * in_pcb.c: manage the Protocol Control Blocks.
 243  *
 244  * NOTE: It is assumed that most of these functions will be called with
 245  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
 246  * functions often modify hash chains or addresses in pcbs.
 247  */
 248
 249 static struct inpcblbgroup *
 250 in_pcblbgroup_alloc(struct inpcblbgrouphead *hdr, struct ucred *cred,
 251     u_char vflag, uint16_t port, const union in_dependaddr *addr, int size,
 252     uint8_t numa_domain)
 253 {
 254         struct inpcblbgroup *grp;
 255         size_t bytes;
 256
 257         bytes = __offsetof(struct inpcblbgroup, il_inp[size]);
 258         grp = malloc(bytes, M_PCB, M_ZERO | M_NOWAIT);
 259         if (grp == NULL)
 260                 return (NULL);
 261         grp->il_cred = crhold(cred);
 262         grp->il_vflag = vflag;
 263         grp->il_lport = port;
 264         grp->il_numa_domain = numa_domain;
 265         grp->il_dependladdr = *addr;
 266         grp->il_inpsiz = size;
 267         CK_LIST_INSERT_HEAD(hdr, grp, il_list);
 268         return (grp);
 269 }
 270
 271 static void
 272 in_pcblbgroup_free_deferred(epoch_context_t ctx)
 273 {
 274         struct inpcblbgroup *grp;
 275
 276         grp = __containerof(ctx, struct inpcblbgroup, il_epoch_ctx);
 277         crfree(grp->il_cred);
 278         free(grp, M_PCB);
 279 }
 280
 281 static void
 282 in_pcblbgroup_free(struct inpcblbgroup *grp)
 283 {
 284
 285         CK_LIST_REMOVE(grp, il_list);
 286         NET_EPOCH_CALL(in_pcblbgroup_free_deferred, &grp->il_epoch_ctx);
 287 }
 288
 289 static struct inpcblbgroup *
 290 in_pcblbgroup_resize(struct inpcblbgrouphead *hdr,
 291     struct inpcblbgroup *old_grp, int size)
 292 {
 293         struct inpcblbgroup *grp;
 294         int i;
 295
 296         grp = in_pcblbgroup_alloc(hdr, old_grp->il_cred, old_grp->il_vflag,
 297             old_grp->il_lport, &old_grp->il_dependladdr, size,
 298             old_grp->il_numa_domain);
 299         if (grp == NULL)
 300                 return (NULL);
 301
 302         KASSERT(old_grp->il_inpcnt < grp->il_inpsiz,
 303             ("invalid new local group size %d and old local group count %d",
 304              grp->il_inpsiz, old_grp->il_inpcnt));
 305
 306         for (i = 0; i < old_grp->il_inpcnt; ++i)
 307                 grp->il_inp[i] = old_grp->il_inp[i];
 308         grp->il_inpcnt = old_grp->il_inpcnt;
 309         in_pcblbgroup_free(old_grp);
 310         return (grp);
 311 }
 312
 313 /*
 314  * PCB at index 'i' is removed from the group. Pull up the ones below il_inp[i]
 315  * and shrink group if possible.
 316  */
 317 static void
 318 in_pcblbgroup_reorder(struct inpcblbgrouphead *hdr, struct inpcblbgroup **grpp,
 319     int i)
 320 {
 321         struct inpcblbgroup *grp, *new_grp;
 322
 323         grp = *grpp;
 324         for (; i + 1 < grp->il_inpcnt; ++i)
 325                 grp->il_inp[i] = grp->il_inp[i + 1];
 326         grp->il_inpcnt--;
 327
 328         if (grp->il_inpsiz > INPCBLBGROUP_SIZMIN &&
 329             grp->il_inpcnt <= grp->il_inpsiz / 4) {
 330                 /* Shrink this group. */
 331                 new_grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz / 2);
 332                 if (new_grp != NULL)
 333                         *grpp = new_grp;
 334         }
 335 }
 336
 337 /*
 338  * Add PCB to load balance group for SO_REUSEPORT_LB option.
 339  */
 340 static int
 341 in_pcbinslbgrouphash(struct inpcb *inp, uint8_t numa_domain)
 342 {
 343         const static struct timeval interval = { 60, 0 };
 344         static struct timeval lastprint;
 345         struct inpcbinfo *pcbinfo;
 346         struct inpcblbgrouphead *hdr;
 347         struct inpcblbgroup *grp;
 348         uint32_t idx;
 349
 350         pcbinfo = inp->inp_pcbinfo;
 351
 352         INP_WLOCK_ASSERT(inp);
 353         INP_HASH_WLOCK_ASSERT(pcbinfo);
 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_cred->cr_prison == inp->inp_cred->cr_prison &&
 370                     grp->il_vflag == inp->inp_vflag &&
 371                     grp->il_lport == inp->inp_lport &&
 372                     grp->il_numa_domain == numa_domain &&
 373                     memcmp(&grp->il_dependladdr,
 374                     &inp->inp_inc.inc_ie.ie_dependladdr,
 375                     sizeof(grp->il_dependladdr)) == 0) {
 376                         break;
 377                 }
 378         }
 379         if (grp == NULL) {
 380                 /* Create new load balance group. */
 381                 grp = in_pcblbgroup_alloc(hdr, inp->inp_cred, inp->inp_vflag,
 382                     inp->inp_lport, &inp->inp_inc.inc_ie.ie_dependladdr,
 383                     INPCBLBGROUP_SIZMIN, numa_domain);
 384                 if (grp == NULL)
 385                         return (ENOBUFS);
 386         } else if (grp->il_inpcnt == grp->il_inpsiz) {
 387                 if (grp->il_inpsiz >= INPCBLBGROUP_SIZMAX) {
 388                         if (ratecheck(&lastprint, &interval))
 389                                 printf("lb group port %d, limit reached\n",
 390                                     ntohs(grp->il_lport));
 391                         return (0);
 392                 }
 393
 394                 /* Expand this local group. */
 395                 grp = in_pcblbgroup_resize(hdr, grp, grp->il_inpsiz * 2);
 396                 if (grp == NULL)
 397                         return (ENOBUFS);
 398         }
 399
 400         KASSERT(grp->il_inpcnt < grp->il_inpsiz,
 401             ("invalid local group size %d and count %d", grp->il_inpsiz,
 402             grp->il_inpcnt));
 403
 404         grp->il_inp[grp->il_inpcnt] = inp;
 405         grp->il_inpcnt++;
 406         return (0);
 407 }
 408
 409 /*
 410  * Remove PCB from load balance group.
 411  */
 412 static void
 413 in_pcbremlbgrouphash(struct inpcb *inp)
 414 {
 415         struct inpcbinfo *pcbinfo;
 416         struct inpcblbgrouphead *hdr;
 417         struct inpcblbgroup *grp;
 418         int i;
 419
 420         pcbinfo = inp->inp_pcbinfo;
 421
 422         INP_WLOCK_ASSERT(inp);
 423         INP_HASH_WLOCK_ASSERT(pcbinfo);
 424
 425         hdr = &pcbinfo->ipi_lbgrouphashbase[
 426             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
 427         CK_LIST_FOREACH(grp, hdr, il_list) {
 428                 for (i = 0; i < grp->il_inpcnt; ++i) {
 429                         if (grp->il_inp[i] != inp)
 430                                 continue;
 431
 432                         if (grp->il_inpcnt == 1) {
 433                                 /* We are the last, free this local group. */
 434                                 in_pcblbgroup_free(grp);
 435                         } else {
 436                                 /* Pull up inpcbs, shrink group if possible. */
 437                                 in_pcblbgroup_reorder(hdr, &grp, i);
 438                         }
 439                         return;
 440                 }
 441         }
 442 }
 443
 444 int
 445 in_pcblbgroup_numa(struct inpcb *inp, int arg)
 446 {
 447         struct inpcbinfo *pcbinfo;
 448         struct inpcblbgrouphead *hdr;
 449         struct inpcblbgroup *grp;
 450         int err, i;
 451         uint8_t numa_domain;
 452
 453         switch (arg) {
 454         case TCP_REUSPORT_LB_NUMA_NODOM:
 455                 numa_domain = M_NODOM;
 456                 break;
 457         case TCP_REUSPORT_LB_NUMA_CURDOM:
 458                 numa_domain = PCPU_GET(domain);
 459                 break;
 460         default:
 461                 if (arg < 0 || arg >= vm_ndomains)
 462                         return (EINVAL);
 463                 numa_domain = arg;
 464         }
 465
 466         err = 0;
 467         pcbinfo = inp->inp_pcbinfo;
 468         INP_WLOCK_ASSERT(inp);
 469         INP_HASH_WLOCK(pcbinfo);
 470         hdr = &pcbinfo->ipi_lbgrouphashbase[
 471             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_lbgrouphashmask)];
 472         CK_LIST_FOREACH(grp, hdr, il_list) {
 473                 for (i = 0; i < grp->il_inpcnt; ++i) {
 474                         if (grp->il_inp[i] != inp)
 475                                 continue;
 476
 477                         if (grp->il_numa_domain == numa_domain) {
 478                                 goto abort_with_hash_wlock;
 479                         }
 480
 481                         /* Remove it from the old group. */
 482                         in_pcbremlbgrouphash(inp);
 483
 484                         /* Add it to the new group based on numa domain. */
 485                         in_pcbinslbgrouphash(inp, numa_domain);
 486                         goto abort_with_hash_wlock;
 487                 }
 488         }
 489         err = ENOENT;
 490 abort_with_hash_wlock:
 491         INP_HASH_WUNLOCK(pcbinfo);
 492         return (err);
 493 }
 494
 495 /* Make sure it is safe to use hashinit(9) on CK_LIST. */
 496 CTASSERT(sizeof(struct inpcbhead) == sizeof(LIST_HEAD(, inpcb)));
 497
 498 /*
 499  * Initialize an inpcbinfo - a per-VNET instance of connections db.
 500  */
 501 void
 502 in_pcbinfo_init(struct inpcbinfo *pcbinfo, struct inpcbstorage *pcbstor,
 503     u_int hash_nelements, u_int porthash_nelements)
 504 {
 505
 506         mtx_init(&pcbinfo->ipi_lock, pcbstor->ips_infolock_name, NULL, MTX_DEF);
 507         mtx_init(&pcbinfo->ipi_hash_lock, pcbstor->ips_hashlock_name,
 508             NULL, MTX_DEF);
 509 #ifdef VIMAGE
 510         pcbinfo->ipi_vnet = curvnet;
 511 #endif
 512         CK_LIST_INIT(&pcbinfo->ipi_listhead);
 513         pcbinfo->ipi_count = 0;
 514         pcbinfo->ipi_hash_exact = hashinit(hash_nelements, M_PCB,
 515             &pcbinfo->ipi_hashmask);
 516         pcbinfo->ipi_hash_wild = hashinit(hash_nelements, M_PCB,
 517             &pcbinfo->ipi_hashmask);
 518         porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
 519         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 520             &pcbinfo->ipi_porthashmask);
 521         pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
 522             &pcbinfo->ipi_lbgrouphashmask);
 523         pcbinfo->ipi_zone = pcbstor->ips_zone;
 524         pcbinfo->ipi_portzone = pcbstor->ips_portzone;
 525         pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
 526 }
 527
 528 /*
 529  * Destroy an inpcbinfo.
 530  */
 531 void
 532 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 533 {
 534
 535         KASSERT(pcbinfo->ipi_count == 0,
 536             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 537
 538         hashdestroy(pcbinfo->ipi_hash_exact, M_PCB, pcbinfo->ipi_hashmask);
 539         hashdestroy(pcbinfo->ipi_hash_wild, M_PCB, pcbinfo->ipi_hashmask);
 540         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 541             pcbinfo->ipi_porthashmask);
 542         hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
 543             pcbinfo->ipi_lbgrouphashmask);
 544         mtx_destroy(&pcbinfo->ipi_hash_lock);
 545         mtx_destroy(&pcbinfo->ipi_lock);
 546 }
 547
 548 /*
 549  * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
 550  */
 551 static void inpcb_fini(void *, int);
 552 void
 553 in_pcbstorage_init(void *arg)
 554 {
 555         struct inpcbstorage *pcbstor = arg;
 556
 557         pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
 558             pcbstor->ips_size, NULL, NULL, pcbstor->ips_pcbinit,
 559             inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR);
 560         pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
 561             sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 562         uma_zone_set_smr(pcbstor->ips_portzone,
 563             uma_zone_get_smr(pcbstor->ips_zone));
 564 }
 565
 566 /*
 567  * Destroy a pcbstorage - used by unloadable protocols.
 568  */
 569 void
 570 in_pcbstorage_destroy(void *arg)
 571 {
 572         struct inpcbstorage *pcbstor = arg;
 573
 574         uma_zdestroy(pcbstor->ips_zone);
 575         uma_zdestroy(pcbstor->ips_portzone);
 576 }
 577
 578 /*
 579  * Allocate a PCB and associate it with the socket.
 580  * On success return with the PCB locked.
 581  */
 582 int
 583 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 584 {
 585         struct inpcb *inp;
 586 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 587         int error;
 588 #endif
 589
 590         inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
 591         if (inp == NULL)
 592                 return (ENOBUFS);
 593         bzero(&inp->inp_start_zero, inp_zero_size);
 594 #ifdef NUMA
 595         inp->inp_numa_domain = M_NODOM;
 596 #endif
 597         inp->inp_pcbinfo = pcbinfo;
 598         inp->inp_socket = so;
 599         inp->inp_cred = crhold(so->so_cred);
 600         inp->inp_inc.inc_fibnum = so->so_fibnum;
 601 #ifdef MAC
 602         error = mac_inpcb_init(inp, M_NOWAIT);
 603         if (error != 0)
 604                 goto out;
 605         mac_inpcb_create(so, inp);
 606 #endif
 607 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 608         error = ipsec_init_pcbpolicy(inp);
 609         if (error != 0) {
 610 #ifdef MAC
 611                 mac_inpcb_destroy(inp);
 612 #endif
 613                 goto out;
 614         }
 615 #endif /*IPSEC*/
 616 #ifdef INET6
 617         if (INP_SOCKAF(so) == AF_INET6) {
 618                 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6;
 619                 if (V_ip6_v6only)
 620                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 621 #ifdef INET
 622                 else
 623                         inp->inp_vflag |= INP_IPV4;
 624 #endif
 625                 if (V_ip6_auto_flowlabel)
 626                         inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 627                 inp->in6p_hops = -1;    /* use kernel default */
 628         }
 629 #endif
 630 #if defined(INET) && defined(INET6)
 631         else
 632 #endif
 633 #ifdef INET
 634                 inp->inp_vflag |= INP_IPV4;
 635 #endif
 636         inp->inp_smr = SMR_SEQ_INVALID;
 637
 638         /*
 639          * Routes in inpcb's can cache L2 as well; they are guaranteed
 640          * to be cleaned up.
 641          */
 642         inp->inp_route.ro_flags = RT_LLE_CACHE;
 643         refcount_init(&inp->inp_refcount, 1);   /* Reference from socket. */
 644         INP_WLOCK(inp);
 645         INP_INFO_WLOCK(pcbinfo);
 646         pcbinfo->ipi_count++;
 647         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 648         CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
 649         INP_INFO_WUNLOCK(pcbinfo);
 650         so->so_pcb = inp;
 651
 652         return (0);
 653
 654 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 655 out:
 656         uma_zfree_smr(pcbinfo->ipi_zone, inp);
 657         return (error);
 658 #endif
 659 }
 660
 661 #ifdef INET
 662 int
 663 in_pcbbind(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred)
 664 {
 665         int anonport, error;
 666
 667         KASSERT(sin == NULL || sin->sin_family == AF_INET,
 668             ("%s: invalid address family for %p", __func__, sin));
 669         KASSERT(sin == NULL || sin->sin_len == sizeof(struct sockaddr_in),
 670             ("%s: invalid address length for %p", __func__, sin));
 671         INP_WLOCK_ASSERT(inp);
 672         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 673
 674         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 675                 return (EINVAL);
 676         anonport = sin == NULL || sin->sin_port == 0;
 677         error = in_pcbbind_setup(inp, sin, &inp->inp_laddr.s_addr,
 678             &inp->inp_lport, cred);
 679         if (error)
 680                 return (error);
 681         if (in_pcbinshash(inp) != 0) {
 682                 inp->inp_laddr.s_addr = INADDR_ANY;
 683                 inp->inp_lport = 0;
 684                 return (EAGAIN);
 685         }
 686         if (anonport)
 687                 inp->inp_flags |= INP_ANONPORT;
 688         return (0);
 689 }
 690 #endif
 691
 692 #if defined(INET) || defined(INET6)
 693 /*
 694  * Assign a local port like in_pcb_lport(), but also used with connect()
 695  * and a foreign address and port.  If fsa is non-NULL, choose a local port
 696  * that is unused with those, otherwise one that is completely unused.
 697  * lsa can be NULL for IPv6.
 698  */
 699 int
 700 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
 701     struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
 702 {
 703         struct inpcbinfo *pcbinfo;
 704         struct inpcb *tmpinp;
 705         unsigned short *lastport;
 706         int count, error;
 707         u_short aux, first, last, lport;
 708 #ifdef INET
 709         struct in_addr laddr, faddr;
 710 #endif
 711 #ifdef INET6
 712         struct in6_addr *laddr6, *faddr6;
 713 #endif
 714
 715         pcbinfo = inp->inp_pcbinfo;
 716
 717         /*
 718          * Because no actual state changes occur here, a global write lock on
 719          * the pcbinfo isn't required.
 720          */
 721         INP_LOCK_ASSERT(inp);
 722         INP_HASH_LOCK_ASSERT(pcbinfo);
 723
 724         if (inp->inp_flags & INP_HIGHPORT) {
 725                 first = V_ipport_hifirstauto;   /* sysctl */
 726                 last  = V_ipport_hilastauto;
 727                 lastport = &pcbinfo->ipi_lasthi;
 728         } else if (inp->inp_flags & INP_LOWPORT) {
 729                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
 730                 if (error)
 731                         return (error);
 732                 first = V_ipport_lowfirstauto;  /* 1023 */
 733                 last  = V_ipport_lowlastauto;   /* 600 */
 734                 lastport = &pcbinfo->ipi_lastlow;
 735         } else {
 736                 first = V_ipport_firstauto;     /* sysctl */
 737                 last  = V_ipport_lastauto;
 738                 lastport = &pcbinfo->ipi_lastport;
 739         }
 740
 741         /*
 742          * Instead of having two loops further down counting up or down
 743          * make sure that first is always <= last and go with only one
 744          * code path implementing all logic.
 745          */
 746         if (first > last) {
 747                 aux = first;
 748                 first = last;
 749                 last = aux;
 750         }
 751
 752 #ifdef INET
 753         laddr.s_addr = INADDR_ANY;      /* used by INET6+INET below too */
 754         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 755                 if (lsa != NULL)
 756                         laddr = ((struct sockaddr_in *)lsa)->sin_addr;
 757                 if (fsa != NULL)
 758                         faddr = ((struct sockaddr_in *)fsa)->sin_addr;
 759         }
 760 #endif
 761 #ifdef INET6
 762         laddr6 = NULL;
 763         if ((inp->inp_vflag & INP_IPV6) != 0) {
 764                 if (lsa != NULL)
 765                         laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
 766                 if (fsa != NULL)
 767                         faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
 768         }
 769 #endif
 770
 771         tmpinp = NULL;
 772         lport = *lportp;
 773
 774         if (V_ipport_randomized)
 775                 *lastport = first + (arc4random() % (last - first));
 776
 777         count = last - first;
 778
 779         do {
 780                 if (count-- < 0)        /* completely used? */
 781                         return (EADDRNOTAVAIL);
 782                 ++*lastport;
 783                 if (*lastport < first || *lastport > last)
 784                         *lastport = first;
 785                 lport = htons(*lastport);
 786
 787                 if (fsa != NULL) {
 788 #ifdef INET
 789                         if (lsa->sa_family == AF_INET) {
 790                                 tmpinp = in_pcblookup_hash_locked(pcbinfo,
 791                                     faddr, fport, laddr, lport, lookupflags,
 792                                     M_NODOM);
 793                         }
 794 #endif
 795 #ifdef INET6
 796                         if (lsa->sa_family == AF_INET6) {
 797                                 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
 798                                     faddr6, fport, laddr6, lport, lookupflags,
 799                                     M_NODOM);
 800                         }
 801 #endif
 802                 } else {
 803 #ifdef INET6
 804                         if ((inp->inp_vflag & INP_IPV6) != 0) {
 805                                 tmpinp = in6_pcblookup_local(pcbinfo,
 806                                     &inp->in6p_laddr, lport, lookupflags, cred);
 807 #ifdef INET
 808                                 if (tmpinp == NULL &&
 809                                     (inp->inp_vflag & INP_IPV4))
 810                                         tmpinp = in_pcblookup_local(pcbinfo,
 811                                             laddr, lport, lookupflags, cred);
 812 #endif
 813                         }
 814 #endif
 815 #if defined(INET) && defined(INET6)
 816                         else
 817 #endif
 818 #ifdef INET
 819                                 tmpinp = in_pcblookup_local(pcbinfo, laddr,
 820                                     lport, lookupflags, cred);
 821 #endif
 822                 }
 823         } while (tmpinp != NULL);
 824
 825         *lportp = lport;
 826
 827         return (0);
 828 }
 829
 830 /*
 831  * Select a local port (number) to use.
 832  */
 833 int
 834 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 835     struct ucred *cred, int lookupflags)
 836 {
 837         struct sockaddr_in laddr;
 838
 839         if (laddrp) {
 840                 bzero(&laddr, sizeof(laddr));
 841                 laddr.sin_family = AF_INET;
 842                 laddr.sin_addr = *laddrp;
 843         }
 844         return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
 845             NULL, lportp, NULL, 0, cred, lookupflags));
 846 }
 847
 848 /*
 849  * Return cached socket options.
 850  */
 851 int
 852 inp_so_options(const struct inpcb *inp)
 853 {
 854         int so_options;
 855
 856         so_options = 0;
 857
 858         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
 859                 so_options |= SO_REUSEPORT_LB;
 860         if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 861                 so_options |= SO_REUSEPORT;
 862         if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 863                 so_options |= SO_REUSEADDR;
 864         return (so_options);
 865 }
 866 #endif /* INET || INET6 */
 867
 868 #ifdef INET
 869 /*
 870  * Set up a bind operation on a PCB, performing port allocation
 871  * as required, but do not actually modify the PCB. Callers can
 872  * either complete the bind by setting inp_laddr/inp_lport and
 873  * calling in_pcbinshash(), or they can just use the resulting
 874  * port and address to authorise the sending of a once-off packet.
 875  *
 876  * On error, the values of *laddrp and *lportp are not changed.
 877  */
 878 int
 879 in_pcbbind_setup(struct inpcb *inp, struct sockaddr_in *sin, in_addr_t *laddrp,
 880     u_short *lportp, struct ucred *cred)
 881 {
 882         struct socket *so = inp->inp_socket;
 883         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 884         struct in_addr laddr;
 885         u_short lport = 0;
 886         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 887         int error;
 888
 889         /*
 890          * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
 891          * so that we don't have to add to the (already messy) code below.
 892          */
 893         int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
 894
 895         /*
 896          * No state changes, so read locks are sufficient here.
 897          */
 898         INP_LOCK_ASSERT(inp);
 899         INP_HASH_LOCK_ASSERT(pcbinfo);
 900
 901         laddr.s_addr = *laddrp;
 902         if (sin != NULL && laddr.s_addr != INADDR_ANY)
 903                 return (EINVAL);
 904         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
 905                 lookupflags = INPLOOKUP_WILDCARD;
 906         if (sin == NULL) {
 907                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 908                         return (error);
 909         } else {
 910                 KASSERT(sin->sin_family == AF_INET,
 911                     ("%s: invalid family for address %p", __func__, sin));
 912                 KASSERT(sin->sin_len == sizeof(*sin),
 913                     ("%s: invalid length for address %p", __func__, sin));
 914
 915                 error = prison_local_ip4(cred, &sin->sin_addr);
 916                 if (error)
 917                         return (error);
 918                 if (sin->sin_port != *lportp) {
 919                         /* Don't allow the port to change. */
 920                         if (*lportp != 0)
 921                                 return (EINVAL);
 922                         lport = sin->sin_port;
 923                 }
 924                 /* NB: lport is left as 0 if the port isn't being changed. */
 925                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 926                         /*
 927                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 928                          * allow complete duplication of binding if
 929                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 930                          * and a multicast address is bound on both
 931                          * new and duplicated sockets.
 932                          */
 933                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 934                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 935                         /*
 936                          * XXX: How to deal with SO_REUSEPORT_LB here?
 937                          * Treat same as SO_REUSEPORT for now.
 938                          */
 939                         if ((so->so_options &
 940                             (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
 941                                 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
 942                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 943                         sin->sin_port = 0;              /* yech... */
 944                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 945                         /*
 946                          * Is the address a local IP address?
 947                          * If INP_BINDANY is set, then the socket may be bound
 948                          * to any endpoint address, local or not.
 949                          */
 950                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 951                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 952                                 return (EADDRNOTAVAIL);
 953                 }
 954                 laddr = sin->sin_addr;
 955                 if (lport) {
 956                         struct inpcb *t;
 957
 958                         /* GROSS */
 959                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 960                             ntohs(lport) >= V_ipport_reservedlow &&
 961                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
 962                                 return (EACCES);
 963                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 964                             priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
 965                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 966                                     lport, INPLOOKUP_WILDCARD, cred);
 967         /*
 968          * XXX
 969          * This entire block sorely needs a rewrite.
 970          */
 971                                 if (t != NULL &&
 972                                     (so->so_type != SOCK_STREAM ||
 973                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
 974                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
 975                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
 976                                      (t->inp_flags2 & INP_REUSEPORT) ||
 977                                      (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
 978                                     (inp->inp_cred->cr_uid !=
 979                                      t->inp_cred->cr_uid))
 980                                         return (EADDRINUSE);
 981                         }
 982                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 983                             lport, lookupflags, cred);
 984                         if (t != NULL && (reuseport & inp_so_options(t)) == 0 &&
 985                             (reuseport_lb & inp_so_options(t)) == 0) {
 986 #ifdef INET6
 987                                 if (ntohl(sin->sin_addr.s_addr) !=
 988                                     INADDR_ANY ||
 989                                     ntohl(t->inp_laddr.s_addr) !=
 990                                     INADDR_ANY ||
 991                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
 992                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
 993 #endif
 994                                                 return (EADDRINUSE);
 995                         }
 996                 }
 997         }
 998         if (*lportp != 0)
 999                 lport = *lportp;
1000         if (lport == 0) {
1001                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1002                 if (error != 0)
1003                         return (error);
1004         }
1005         *laddrp = laddr.s_addr;
1006         *lportp = lport;
1007         return (0);
1008 }
1009
1010 /*
1011  * Connect from a socket to a specified address.
1012  * Both address and port must be specified in argument sin.
1013  * If don't have a local address for this socket yet,
1014  * then pick one.
1015  */
1016 int
1017 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1018     bool rehash __unused)
1019 {
1020         u_short lport, fport;
1021         in_addr_t laddr, faddr;
1022         int anonport, error;
1023
1024         INP_WLOCK_ASSERT(inp);
1025         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1026         KASSERT(in_nullhost(inp->inp_faddr),
1027             ("%s: inp is already connected", __func__));
1028
1029         lport = inp->inp_lport;
1030         laddr = inp->inp_laddr.s_addr;
1031         anonport = (lport == 0);
1032         error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1033             cred);
1034         if (error)
1035                 return (error);
1036
1037         inp->inp_faddr.s_addr = faddr;
1038         inp->inp_fport = fport;
1039
1040         /* Do the initial binding of the local address if required. */
1041         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1042                 inp->inp_lport = lport;
1043                 inp->inp_laddr.s_addr = laddr;
1044                 if (in_pcbinshash(inp) != 0) {
1045                         inp->inp_laddr.s_addr = inp->inp_faddr.s_addr =
1046                             INADDR_ANY;
1047                         inp->inp_lport = inp->inp_fport = 0;
1048                         return (EAGAIN);
1049                 }
1050         } else {
1051                 inp->inp_lport = lport;
1052                 inp->inp_laddr.s_addr = laddr;
1053                 if ((inp->inp_flags & INP_INHASHLIST) != 0)
1054                         in_pcbrehash(inp);
1055                 else
1056                         in_pcbinshash(inp);
1057         }
1058
1059         if (anonport)
1060                 inp->inp_flags |= INP_ANONPORT;
1061         return (0);
1062 }
1063
1064 /*
1065  * Do proper source address selection on an unbound socket in case
1066  * of connect. Take jails into account as well.
1067  */
1068 int
1069 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1070     struct ucred *cred)
1071 {
1072         struct ifaddr *ifa;
1073         struct sockaddr *sa;
1074         struct sockaddr_in *sin, dst;
1075         struct nhop_object *nh;
1076         int error;
1077
1078         NET_EPOCH_ASSERT();
1079         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1080
1081         /*
1082          * Bypass source address selection and use the primary jail IP
1083          * if requested.
1084          */
1085         if (!prison_saddrsel_ip4(cred, laddr))
1086                 return (0);
1087
1088         error = 0;
1089
1090         nh = NULL;
1091         bzero(&dst, sizeof(dst));
1092         sin = &dst;
1093         sin->sin_family = AF_INET;
1094         sin->sin_len = sizeof(struct sockaddr_in);
1095         sin->sin_addr.s_addr = faddr->s_addr;
1096
1097         /*
1098          * If route is known our src addr is taken from the i/f,
1099          * else punt.
1100          *
1101          * Find out route to destination.
1102          */
1103         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1104                 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1105                     0, NHR_NONE, 0);
1106
1107         /*
1108          * If we found a route, use the address corresponding to
1109          * the outgoing interface.
1110          *
1111          * Otherwise assume faddr is reachable on a directly connected
1112          * network and try to find a corresponding interface to take
1113          * the source address from.
1114          */
1115         if (nh == NULL || nh->nh_ifp == NULL) {
1116                 struct in_ifaddr *ia;
1117                 struct ifnet *ifp;
1118
1119                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1120                                         inp->inp_socket->so_fibnum));
1121                 if (ia == NULL) {
1122                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1123                                                 inp->inp_socket->so_fibnum));
1124                 }
1125                 if (ia == NULL) {
1126                         error = ENETUNREACH;
1127                         goto done;
1128                 }
1129
1130                 if (!prison_flag(cred, PR_IP4)) {
1131                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1132                         goto done;
1133                 }
1134
1135                 ifp = ia->ia_ifp;
1136                 ia = NULL;
1137                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1138                         sa = ifa->ifa_addr;
1139                         if (sa->sa_family != AF_INET)
1140                                 continue;
1141                         sin = (struct sockaddr_in *)sa;
1142                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1143                                 ia = (struct in_ifaddr *)ifa;
1144                                 break;
1145                         }
1146                 }
1147                 if (ia != NULL) {
1148                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1149                         goto done;
1150                 }
1151
1152                 /* 3. As a last resort return the 'default' jail address. */
1153                 error = prison_get_ip4(cred, laddr);
1154                 goto done;
1155         }
1156
1157         /*
1158          * If the outgoing interface on the route found is not
1159          * a loopback interface, use the address from that interface.
1160          * In case of jails do those three steps:
1161          * 1. check if the interface address belongs to the jail. If so use it.
1162          * 2. check if we have any address on the outgoing interface
1163          *    belonging to this jail. If so use it.
1164          * 3. as a last resort return the 'default' jail address.
1165          */
1166         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1167                 struct in_ifaddr *ia;
1168                 struct ifnet *ifp;
1169
1170                 /* If not jailed, use the default returned. */
1171                 if (!prison_flag(cred, PR_IP4)) {
1172                         ia = (struct in_ifaddr *)nh->nh_ifa;
1173                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1174                         goto done;
1175                 }
1176
1177                 /* Jailed. */
1178                 /* 1. Check if the iface address belongs to the jail. */
1179                 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1180                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
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                 /*
1187                  * 2. Check if we have any address on the outgoing interface
1188                  *    belonging to this jail.
1189                  */
1190                 ia = NULL;
1191                 ifp = nh->nh_ifp;
1192                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1193                         sa = ifa->ifa_addr;
1194                         if (sa->sa_family != AF_INET)
1195                                 continue;
1196                         sin = (struct sockaddr_in *)sa;
1197                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1198                                 ia = (struct in_ifaddr *)ifa;
1199                                 break;
1200                         }
1201                 }
1202                 if (ia != NULL) {
1203                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1204                         goto done;
1205                 }
1206
1207                 /* 3. As a last resort return the 'default' jail address. */
1208                 error = prison_get_ip4(cred, laddr);
1209                 goto done;
1210         }
1211
1212         /*
1213          * The outgoing interface is marked with 'loopback net', so a route
1214          * to ourselves is here.
1215          * Try to find the interface of the destination address and then
1216          * take the address from there. That interface is not necessarily
1217          * a loopback interface.
1218          * In case of jails, check that it is an address of the jail
1219          * and if we cannot find, fall back to the 'default' jail address.
1220          */
1221         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1222                 struct in_ifaddr *ia;
1223
1224                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1225                                         inp->inp_socket->so_fibnum));
1226                 if (ia == NULL)
1227                         ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1228                                                 inp->inp_socket->so_fibnum));
1229                 if (ia == NULL)
1230                         ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1231
1232                 if (!prison_flag(cred, PR_IP4)) {
1233                         if (ia == NULL) {
1234                                 error = ENETUNREACH;
1235                                 goto done;
1236                         }
1237                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1238                         goto done;
1239                 }
1240
1241                 /* Jailed. */
1242                 if (ia != NULL) {
1243                         struct ifnet *ifp;
1244
1245                         ifp = ia->ia_ifp;
1246                         ia = NULL;
1247                         CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1248                                 sa = ifa->ifa_addr;
1249                                 if (sa->sa_family != AF_INET)
1250                                         continue;
1251                                 sin = (struct sockaddr_in *)sa;
1252                                 if (prison_check_ip4(cred,
1253                                     &sin->sin_addr) == 0) {
1254                                         ia = (struct in_ifaddr *)ifa;
1255                                         break;
1256                                 }
1257                         }
1258                         if (ia != NULL) {
1259                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1260                                 goto done;
1261                         }
1262                 }
1263
1264                 /* 3. As a last resort return the 'default' jail address. */
1265                 error = prison_get_ip4(cred, laddr);
1266                 goto done;
1267         }
1268
1269 done:
1270         if (error == 0 && laddr->s_addr == INADDR_ANY)
1271                 return (EHOSTUNREACH);
1272         return (error);
1273 }
1274
1275 /*
1276  * Set up for a connect from a socket to the specified address.
1277  * On entry, *laddrp and *lportp should contain the current local
1278  * address and port for the PCB; these are updated to the values
1279  * that should be placed in inp_laddr and inp_lport to complete
1280  * the connect.
1281  *
1282  * On success, *faddrp and *fportp will be set to the remote address
1283  * and port. These are not updated in the error case.
1284  */
1285 int
1286 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1287     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1288     struct ucred *cred)
1289 {
1290         struct in_ifaddr *ia;
1291         struct in_addr laddr, faddr;
1292         u_short lport, fport;
1293         int error;
1294
1295         KASSERT(sin->sin_family == AF_INET,
1296             ("%s: invalid address family for %p", __func__, sin));
1297         KASSERT(sin->sin_len == sizeof(*sin),
1298             ("%s: invalid address length for %p", __func__, sin));
1299
1300         /*
1301          * Because a global state change doesn't actually occur here, a read
1302          * lock is sufficient.
1303          */
1304         NET_EPOCH_ASSERT();
1305         INP_LOCK_ASSERT(inp);
1306         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1307
1308         if (sin->sin_port == 0)
1309                 return (EADDRNOTAVAIL);
1310         laddr.s_addr = *laddrp;
1311         lport = *lportp;
1312         faddr = sin->sin_addr;
1313         fport = sin->sin_port;
1314 #ifdef ROUTE_MPATH
1315         if (CALC_FLOWID_OUTBOUND) {
1316                 uint32_t hash_val, hash_type;
1317
1318                 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1319                     inp->inp_socket->so_proto->pr_protocol, &hash_type);
1320
1321                 inp->inp_flowid = hash_val;
1322                 inp->inp_flowtype = hash_type;
1323         }
1324 #endif
1325         if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1326                 /*
1327                  * If the destination address is INADDR_ANY,
1328                  * use the primary local address.
1329                  * If the supplied address is INADDR_BROADCAST,
1330                  * and the primary interface supports broadcast,
1331                  * choose the broadcast address for that interface.
1332                  */
1333                 if (faddr.s_addr == INADDR_ANY) {
1334                         faddr =
1335                             IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1336                         if ((error = prison_get_ip4(cred, &faddr)) != 0)
1337                                 return (error);
1338                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1339                         if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1340                             IFF_BROADCAST)
1341                                 faddr = satosin(&CK_STAILQ_FIRST(
1342                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1343                 }
1344         }
1345         if (laddr.s_addr == INADDR_ANY) {
1346                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1347                 /*
1348                  * If the destination address is multicast and an outgoing
1349                  * interface has been set as a multicast option, prefer the
1350                  * address of that interface as our source address.
1351                  */
1352                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1353                     inp->inp_moptions != NULL) {
1354                         struct ip_moptions *imo;
1355                         struct ifnet *ifp;
1356
1357                         imo = inp->inp_moptions;
1358                         if (imo->imo_multicast_ifp != NULL) {
1359                                 ifp = imo->imo_multicast_ifp;
1360                                 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1361                                         if (ia->ia_ifp == ifp &&
1362                                             prison_check_ip4(cred,
1363                                             &ia->ia_addr.sin_addr) == 0)
1364                                                 break;
1365                                 }
1366                                 if (ia == NULL)
1367                                         error = EADDRNOTAVAIL;
1368                                 else {
1369                                         laddr = ia->ia_addr.sin_addr;
1370                                         error = 0;
1371                                 }
1372                         }
1373                 }
1374                 if (error)
1375                         return (error);
1376         }
1377
1378         if (lport != 0) {
1379                 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1380                     fport, laddr, lport, 0, M_NODOM) != NULL)
1381                         return (EADDRINUSE);
1382         } else {
1383                 struct sockaddr_in lsin, fsin;
1384
1385                 bzero(&lsin, sizeof(lsin));
1386                 bzero(&fsin, sizeof(fsin));
1387                 lsin.sin_family = AF_INET;
1388                 lsin.sin_addr = laddr;
1389                 fsin.sin_family = AF_INET;
1390                 fsin.sin_addr = faddr;
1391                 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1392                     &lport, (struct sockaddr *)& fsin, fport, cred,
1393                     INPLOOKUP_WILDCARD);
1394                 if (error)
1395                         return (error);
1396         }
1397         *laddrp = laddr.s_addr;
1398         *lportp = lport;
1399         *faddrp = faddr.s_addr;
1400         *fportp = fport;
1401         return (0);
1402 }
1403
1404 void
1405 in_pcbdisconnect(struct inpcb *inp)
1406 {
1407
1408         INP_WLOCK_ASSERT(inp);
1409         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1410         KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
1411             ("%s: inp %p was already disconnected", __func__, inp));
1412
1413         in_pcbremhash_locked(inp);
1414
1415         /* See the comment in in_pcbinshash(). */
1416         inp->inp_smr = smr_advance(inp->inp_pcbinfo->ipi_smr);
1417         inp->inp_laddr.s_addr = INADDR_ANY;
1418         inp->inp_faddr.s_addr = INADDR_ANY;
1419         inp->inp_fport = 0;
1420 }
1421 #endif /* INET */
1422
1423 /*
1424  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1425  * For most protocols, this will be invoked immediately prior to calling
1426  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1427  * socket, in which case in_pcbfree() is deferred.
1428  */
1429 void
1430 in_pcbdetach(struct inpcb *inp)
1431 {
1432
1433         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1434
1435 #ifdef RATELIMIT
1436         if (inp->inp_snd_tag != NULL)
1437                 in_pcbdetach_txrtlmt(inp);
1438 #endif
1439         inp->inp_socket->so_pcb = NULL;
1440         inp->inp_socket = NULL;
1441 }
1442
1443 /*
1444  * inpcb hash lookups are protected by SMR section.
1445  *
1446  * Once desired pcb has been found, switching from SMR section to a pcb
1447  * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1448  * here because SMR is a critical section.
1449  * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1450  */
1451 void
1452 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1453 {
1454
1455         lock == INPLOOKUP_RLOCKPCB ?
1456             rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1457 }
1458
1459 void
1460 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1461 {
1462
1463         lock == INPLOOKUP_RLOCKPCB ?
1464             rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1465 }
1466
1467 int
1468 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1469 {
1470
1471         return (lock == INPLOOKUP_RLOCKPCB ?
1472             rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1473 }
1474
1475 static inline bool
1476 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1477 {
1478
1479         MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1480         SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1481
1482         if (__predict_true(inp_trylock(inp, lock))) {
1483                 if (__predict_false(inp->inp_flags & ignflags)) {
1484                         smr_exit(inp->inp_pcbinfo->ipi_smr);
1485                         inp_unlock(inp, lock);
1486                         return (false);
1487                 }
1488                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1489                 return (true);
1490         }
1491
1492         if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1493                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1494                 inp_lock(inp, lock);
1495                 if (__predict_false(in_pcbrele(inp, lock)))
1496                         return (false);
1497                 /*
1498                  * inp acquired through refcount & lock for sure didn't went
1499                  * through uma_zfree().  However, it may have already went
1500                  * through in_pcbfree() and has another reference, that
1501                  * prevented its release by our in_pcbrele().
1502                  */
1503                 if (__predict_false(inp->inp_flags & ignflags)) {
1504                         inp_unlock(inp, lock);
1505                         return (false);
1506                 }
1507                 return (true);
1508         } else {
1509                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1510                 return (false);
1511         }
1512 }
1513
1514 bool
1515 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1516 {
1517
1518         /*
1519          * in_pcblookup() family of functions ignore not only freed entries,
1520          * that may be found due to lockless access to the hash, but dropped
1521          * entries, too.
1522          */
1523         return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1524 }
1525
1526 /*
1527  * inp_next() - inpcb hash/list traversal iterator
1528  *
1529  * Requires initialized struct inpcb_iterator for context.
1530  * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1531  *
1532  * - Iterator can have either write-lock or read-lock semantics, that can not
1533  *   be changed later.
1534  * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1535  *   a single hash slot.  Note: only rip_input() does the latter.
1536  * - Iterator may have optional bool matching function.  The matching function
1537  *   will be executed for each inpcb in the SMR context, so it can not acquire
1538  *   locks and can safely access only immutable fields of inpcb.
1539  *
1540  * A fresh initialized iterator has NULL inpcb in its context and that
1541  * means that inp_next() call would return the very first inpcb on the list
1542  * locked with desired semantic.  In all following calls the context pointer
1543  * shall hold the current inpcb pointer.  The KPI user is not supposed to
1544  * unlock the current inpcb!  Upon end of traversal inp_next() will return NULL
1545  * and write NULL to its context.  After end of traversal an iterator can be
1546  * reused.
1547  *
1548  * List traversals have the following features/constraints:
1549  * - New entries won't be seen, as they are always added to the head of a list.
1550  * - Removed entries won't stop traversal as long as they are not added to
1551  *   a different list. This is violated by in_pcbrehash().
1552  */
1553 #define II_LIST_FIRST(ipi, hash)                                        \
1554                 (((hash) == INP_ALL_LIST) ?                             \
1555                     CK_LIST_FIRST(&(ipi)->ipi_listhead) :               \
1556                     CK_LIST_FIRST(&(ipi)->ipi_hash_exact[(hash)]))
1557 #define II_LIST_NEXT(inp, hash)                                         \
1558                 (((hash) == INP_ALL_LIST) ?                             \
1559                     CK_LIST_NEXT((inp), inp_list) :                     \
1560                     CK_LIST_NEXT((inp), inp_hash_exact))
1561 #define II_LOCK_ASSERT(inp, lock)                                       \
1562                 rw_assert(&(inp)->inp_lock,                             \
1563                     (lock) == INPLOOKUP_RLOCKPCB ?  RA_RLOCKED : RA_WLOCKED )
1564 struct inpcb *
1565 inp_next(struct inpcb_iterator *ii)
1566 {
1567         const struct inpcbinfo *ipi = ii->ipi;
1568         inp_match_t *match = ii->match;
1569         void *ctx = ii->ctx;
1570         inp_lookup_t lock = ii->lock;
1571         int hash = ii->hash;
1572         struct inpcb *inp;
1573
1574         if (ii->inp == NULL) {          /* First call. */
1575                 smr_enter(ipi->ipi_smr);
1576                 /* This is unrolled CK_LIST_FOREACH(). */
1577                 for (inp = II_LIST_FIRST(ipi, hash);
1578                     inp != NULL;
1579                     inp = II_LIST_NEXT(inp, hash)) {
1580                         if (match != NULL && (match)(inp, ctx) == false)
1581                                 continue;
1582                         if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1583                                 break;
1584                         else {
1585                                 smr_enter(ipi->ipi_smr);
1586                                 MPASS(inp != II_LIST_FIRST(ipi, hash));
1587                                 inp = II_LIST_FIRST(ipi, hash);
1588                                 if (inp == NULL)
1589                                         break;
1590                         }
1591                 }
1592
1593                 if (inp == NULL)
1594                         smr_exit(ipi->ipi_smr);
1595                 else
1596                         ii->inp = inp;
1597
1598                 return (inp);
1599         }
1600
1601         /* Not a first call. */
1602         smr_enter(ipi->ipi_smr);
1603 restart:
1604         inp = ii->inp;
1605         II_LOCK_ASSERT(inp, lock);
1606 next:
1607         inp = II_LIST_NEXT(inp, hash);
1608         if (inp == NULL) {
1609                 smr_exit(ipi->ipi_smr);
1610                 goto found;
1611         }
1612
1613         if (match != NULL && (match)(inp, ctx) == false)
1614                 goto next;
1615
1616         if (__predict_true(inp_trylock(inp, lock))) {
1617                 if (__predict_false(inp->inp_flags & INP_FREED)) {
1618                         /*
1619                          * Entries are never inserted in middle of a list, thus
1620                          * as long as we are in SMR, we can continue traversal.
1621                          * Jump to 'restart' should yield in the same result,
1622                          * but could produce unnecessary looping.  Could this
1623                          * looping be unbound?
1624                          */
1625                         inp_unlock(inp, lock);
1626                         goto next;
1627                 } else {
1628                         smr_exit(ipi->ipi_smr);
1629                         goto found;
1630                 }
1631         }
1632
1633         /*
1634          * Can't obtain lock immediately, thus going hard.  Once we exit the
1635          * SMR section we can no longer jump to 'next', and our only stable
1636          * anchoring point is ii->inp, which we keep locked for this case, so
1637          * we jump to 'restart'.
1638          */
1639         if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1640                 smr_exit(ipi->ipi_smr);
1641                 inp_lock(inp, lock);
1642                 if (__predict_false(in_pcbrele(inp, lock))) {
1643                         smr_enter(ipi->ipi_smr);
1644                         goto restart;
1645                 }
1646                 /*
1647                  * See comment in inp_smr_lock().
1648                  */
1649                 if (__predict_false(inp->inp_flags & INP_FREED)) {
1650                         inp_unlock(inp, lock);
1651                         smr_enter(ipi->ipi_smr);
1652                         goto restart;
1653                 }
1654         } else
1655                 goto next;
1656
1657 found:
1658         inp_unlock(ii->inp, lock);
1659         ii->inp = inp;
1660
1661         return (ii->inp);
1662 }
1663
1664 /*
1665  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1666  * stability of an inpcb pointer despite the inpcb lock being released or
1667  * SMR section exited.
1668  *
1669  * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1670  */
1671 void
1672 in_pcbref(struct inpcb *inp)
1673 {
1674         u_int old __diagused;
1675
1676         old = refcount_acquire(&inp->inp_refcount);
1677         KASSERT(old > 0, ("%s: refcount 0", __func__));
1678 }
1679
1680 /*
1681  * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1682  * freeing the pcb, if the reference was very last.
1683  */
1684 bool
1685 in_pcbrele_rlocked(struct inpcb *inp)
1686 {
1687
1688         INP_RLOCK_ASSERT(inp);
1689
1690         if (!refcount_release(&inp->inp_refcount))
1691                 return (false);
1692
1693         MPASS(inp->inp_flags & INP_FREED);
1694         MPASS(inp->inp_socket == NULL);
1695         crfree(inp->inp_cred);
1696 #ifdef INVARIANTS
1697         inp->inp_cred = NULL;
1698 #endif
1699         INP_RUNLOCK(inp);
1700         uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1701         return (true);
1702 }
1703
1704 bool
1705 in_pcbrele_wlocked(struct inpcb *inp)
1706 {
1707
1708         INP_WLOCK_ASSERT(inp);
1709
1710         if (!refcount_release(&inp->inp_refcount))
1711                 return (false);
1712
1713         MPASS(inp->inp_flags & INP_FREED);
1714         MPASS(inp->inp_socket == NULL);
1715         crfree(inp->inp_cred);
1716 #ifdef INVARIANTS
1717         inp->inp_cred = NULL;
1718 #endif
1719         INP_WUNLOCK(inp);
1720         uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1721         return (true);
1722 }
1723
1724 bool
1725 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1726 {
1727
1728         return (lock == INPLOOKUP_RLOCKPCB ?
1729             in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1730 }
1731
1732 /*
1733  * Unconditionally schedule an inpcb to be freed by decrementing its
1734  * reference count, which should occur only after the inpcb has been detached
1735  * from its socket.  If another thread holds a temporary reference (acquired
1736  * using in_pcbref()) then the free is deferred until that reference is
1737  * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1738  *  Almost all work, including removal from global lists, is done in this
1739  * context, where the pcbinfo lock is held.
1740  */
1741 void
1742 in_pcbfree(struct inpcb *inp)
1743 {
1744         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1745 #ifdef INET
1746         struct ip_moptions *imo;
1747 #endif
1748 #ifdef INET6
1749         struct ip6_moptions *im6o;
1750 #endif
1751
1752         INP_WLOCK_ASSERT(inp);
1753         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1754         KASSERT((inp->inp_flags & INP_FREED) == 0,
1755             ("%s: called twice for pcb %p", __func__, inp));
1756
1757         inp->inp_flags |= INP_FREED;
1758         INP_INFO_WLOCK(pcbinfo);
1759         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1760         pcbinfo->ipi_count--;
1761         CK_LIST_REMOVE(inp, inp_list);
1762         INP_INFO_WUNLOCK(pcbinfo);
1763
1764         if (inp->inp_flags & INP_INHASHLIST)
1765                 in_pcbremhash(inp);
1766
1767         RO_INVALIDATE_CACHE(&inp->inp_route);
1768 #ifdef MAC
1769         mac_inpcb_destroy(inp);
1770 #endif
1771 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1772         if (inp->inp_sp != NULL)
1773                 ipsec_delete_pcbpolicy(inp);
1774 #endif
1775 #ifdef INET
1776         if (inp->inp_options)
1777                 (void)m_free(inp->inp_options);
1778         imo = inp->inp_moptions;
1779 #endif
1780 #ifdef INET6
1781         if (inp->inp_vflag & INP_IPV6PROTO) {
1782                 ip6_freepcbopts(inp->in6p_outputopts);
1783                 im6o = inp->in6p_moptions;
1784         } else
1785                 im6o = NULL;
1786 #endif
1787
1788         if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1789                 INP_WUNLOCK(inp);
1790         }
1791 #ifdef INET6
1792         ip6_freemoptions(im6o);
1793 #endif
1794 #ifdef INET
1795         inp_freemoptions(imo);
1796 #endif
1797 }
1798
1799 /*
1800  * Different protocols initialize their inpcbs differently - giving
1801  * different name to the lock.  But they all are disposed the same.
1802  */
1803 static void
1804 inpcb_fini(void *mem, int size)
1805 {
1806         struct inpcb *inp = mem;
1807
1808         INP_LOCK_DESTROY(inp);
1809 }
1810
1811 /*
1812  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1813  * port reservation, and preventing it from being returned by inpcb lookups.
1814  *
1815  * It is used by TCP to mark an inpcb as unused and avoid future packet
1816  * delivery or event notification when a socket remains open but TCP has
1817  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1818  * or a RST on the wire, and allows the port binding to be reused while still
1819  * maintaining the invariant that so_pcb always points to a valid inpcb until
1820  * in_pcbdetach().
1821  *
1822  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1823  * in_pcbnotifyall() and in_pcbpurgeif0()?
1824  */
1825 void
1826 in_pcbdrop(struct inpcb *inp)
1827 {
1828
1829         INP_WLOCK_ASSERT(inp);
1830 #ifdef INVARIANTS
1831         if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1832                 MPASS(inp->inp_refcount > 1);
1833 #endif
1834
1835         inp->inp_flags |= INP_DROPPED;
1836         if (inp->inp_flags & INP_INHASHLIST)
1837                 in_pcbremhash(inp);
1838 }
1839
1840 #ifdef INET
1841 /*
1842  * Common routines to return the socket addresses associated with inpcbs.
1843  */
1844 struct sockaddr *
1845 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1846 {
1847         struct sockaddr_in *sin;
1848
1849         sin = malloc(sizeof *sin, M_SONAME,
1850                 M_WAITOK | M_ZERO);
1851         sin->sin_family = AF_INET;
1852         sin->sin_len = sizeof(*sin);
1853         sin->sin_addr = *addr_p;
1854         sin->sin_port = port;
1855
1856         return (struct sockaddr *)sin;
1857 }
1858
1859 int
1860 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1861 {
1862         struct inpcb *inp;
1863         struct in_addr addr;
1864         in_port_t port;
1865
1866         inp = sotoinpcb(so);
1867         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1868
1869         INP_RLOCK(inp);
1870         port = inp->inp_lport;
1871         addr = inp->inp_laddr;
1872         INP_RUNLOCK(inp);
1873
1874         *nam = in_sockaddr(port, &addr);
1875         return 0;
1876 }
1877
1878 int
1879 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1880 {
1881         struct inpcb *inp;
1882         struct in_addr addr;
1883         in_port_t port;
1884
1885         inp = sotoinpcb(so);
1886         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1887
1888         INP_RLOCK(inp);
1889         port = inp->inp_fport;
1890         addr = inp->inp_faddr;
1891         INP_RUNLOCK(inp);
1892
1893         *nam = in_sockaddr(port, &addr);
1894         return 0;
1895 }
1896
1897 void
1898 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1899     struct inpcb *(*notify)(struct inpcb *, int))
1900 {
1901         struct inpcb *inp, *inp_temp;
1902
1903         INP_INFO_WLOCK(pcbinfo);
1904         CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1905                 INP_WLOCK(inp);
1906 #ifdef INET6
1907                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1908                         INP_WUNLOCK(inp);
1909                         continue;
1910                 }
1911 #endif
1912                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1913                     inp->inp_socket == NULL) {
1914                         INP_WUNLOCK(inp);
1915                         continue;
1916                 }
1917                 if ((*notify)(inp, errno))
1918                         INP_WUNLOCK(inp);
1919         }
1920         INP_INFO_WUNLOCK(pcbinfo);
1921 }
1922
1923 static bool
1924 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1925 {
1926
1927         if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1928                 return (true);
1929         else
1930                 return (false);
1931 }
1932
1933 void
1934 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1935 {
1936         struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1937             inp_v4_multi_match, NULL);
1938         struct inpcb *inp;
1939         struct in_multi *inm;
1940         struct in_mfilter *imf;
1941         struct ip_moptions *imo;
1942
1943         IN_MULTI_LOCK_ASSERT();
1944
1945         while ((inp = inp_next(&inpi)) != NULL) {
1946                 INP_WLOCK_ASSERT(inp);
1947
1948                 imo = inp->inp_moptions;
1949                 /*
1950                  * Unselect the outgoing interface if it is being
1951                  * detached.
1952                  */
1953                 if (imo->imo_multicast_ifp == ifp)
1954                         imo->imo_multicast_ifp = NULL;
1955
1956                 /*
1957                  * Drop multicast group membership if we joined
1958                  * through the interface being detached.
1959                  *
1960                  * XXX This can all be deferred to an epoch_call
1961                  */
1962 restart:
1963                 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
1964                         if ((inm = imf->imf_inm) == NULL)
1965                                 continue;
1966                         if (inm->inm_ifp != ifp)
1967                                 continue;
1968                         ip_mfilter_remove(&imo->imo_head, imf);
1969                         in_leavegroup_locked(inm, NULL);
1970                         ip_mfilter_free(imf);
1971                         goto restart;
1972                 }
1973         }
1974 }
1975
1976 /*
1977  * Lookup a PCB based on the local address and port.  Caller must hold the
1978  * hash lock.  No inpcb locks or references are acquired.
1979  */
1980 #define INP_LOOKUP_MAPPED_PCB_COST      3
1981 struct inpcb *
1982 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1983     u_short lport, int lookupflags, struct ucred *cred)
1984 {
1985         struct inpcb *inp;
1986 #ifdef INET6
1987         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1988 #else
1989         int matchwild = 3;
1990 #endif
1991         int wildcard;
1992
1993         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1994             ("%s: invalid lookup flags %d", __func__, lookupflags));
1995         INP_HASH_LOCK_ASSERT(pcbinfo);
1996
1997         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1998                 struct inpcbhead *head;
1999                 /*
2000                  * Look for an unconnected (wildcard foreign addr) PCB that
2001                  * matches the local address and port we're looking for.
2002                  */
2003                 head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2004                     pcbinfo->ipi_hashmask)];
2005                 CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2006 #ifdef INET6
2007                         /* XXX inp locking */
2008                         if ((inp->inp_vflag & INP_IPV4) == 0)
2009                                 continue;
2010 #endif
2011                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
2012                             inp->inp_laddr.s_addr == laddr.s_addr &&
2013                             inp->inp_lport == lport) {
2014                                 /*
2015                                  * Found?
2016                                  */
2017                                 if (prison_equal_ip4(cred->cr_prison,
2018                                     inp->inp_cred->cr_prison))
2019                                         return (inp);
2020                         }
2021                 }
2022                 /*
2023                  * Not found.
2024                  */
2025                 return (NULL);
2026         } else {
2027                 struct inpcbporthead *porthash;
2028                 struct inpcbport *phd;
2029                 struct inpcb *match = NULL;
2030                 /*
2031                  * Best fit PCB lookup.
2032                  *
2033                  * First see if this local port is in use by looking on the
2034                  * port hash list.
2035                  */
2036                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2037                     pcbinfo->ipi_porthashmask)];
2038                 CK_LIST_FOREACH(phd, porthash, phd_hash) {
2039                         if (phd->phd_port == lport)
2040                                 break;
2041                 }
2042                 if (phd != NULL) {
2043                         /*
2044                          * Port is in use by one or more PCBs. Look for best
2045                          * fit.
2046                          */
2047                         CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2048                                 wildcard = 0;
2049                                 if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2050                                     cred->cr_prison))
2051                                         continue;
2052 #ifdef INET6
2053                                 /* XXX inp locking */
2054                                 if ((inp->inp_vflag & INP_IPV4) == 0)
2055                                         continue;
2056                                 /*
2057                                  * We never select the PCB that has
2058                                  * INP_IPV6 flag and is bound to :: if
2059                                  * we have another PCB which is bound
2060                                  * to 0.0.0.0.  If a PCB has the
2061                                  * INP_IPV6 flag, then we set its cost
2062                                  * higher than IPv4 only PCBs.
2063                                  *
2064                                  * Note that the case only happens
2065                                  * when a socket is bound to ::, under
2066                                  * the condition that the use of the
2067                                  * mapped address is allowed.
2068                                  */
2069                                 if ((inp->inp_vflag & INP_IPV6) != 0)
2070                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2071 #endif
2072                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
2073                                         wildcard++;
2074                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
2075                                         if (laddr.s_addr == INADDR_ANY)
2076                                                 wildcard++;
2077                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
2078                                                 continue;
2079                                 } else {
2080                                         if (laddr.s_addr != INADDR_ANY)
2081                                                 wildcard++;
2082                                 }
2083                                 if (wildcard < matchwild) {
2084                                         match = inp;
2085                                         matchwild = wildcard;
2086                                         if (matchwild == 0)
2087                                                 break;
2088                                 }
2089                         }
2090                 }
2091                 return (match);
2092         }
2093 }
2094 #undef INP_LOOKUP_MAPPED_PCB_COST
2095
2096 static bool
2097 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2098 {
2099         return (domain == M_NODOM || domain == grp->il_numa_domain);
2100 }
2101
2102 static struct inpcb *
2103 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2104     const struct in_addr *faddr, uint16_t fport, const struct in_addr *laddr,
2105     uint16_t lport, int domain)
2106 {
2107         const struct inpcblbgrouphead *hdr;
2108         struct inpcblbgroup *grp;
2109         struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2110
2111         INP_HASH_LOCK_ASSERT(pcbinfo);
2112
2113         hdr = &pcbinfo->ipi_lbgrouphashbase[
2114             INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2115
2116         /*
2117          * Search for an LB group match based on the following criteria:
2118          * - prefer jailed groups to non-jailed groups
2119          * - prefer exact source address matches to wildcard matches
2120          * - prefer groups bound to the specified NUMA domain
2121          */
2122         jail_exact = jail_wild = local_exact = local_wild = NULL;
2123         CK_LIST_FOREACH(grp, hdr, il_list) {
2124                 bool injail;
2125
2126 #ifdef INET6
2127                 if (!(grp->il_vflag & INP_IPV4))
2128                         continue;
2129 #endif
2130                 if (grp->il_lport != lport)
2131                         continue;
2132
2133                 injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2134                 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2135                     laddr) != 0)
2136                         continue;
2137
2138                 if (grp->il_laddr.s_addr == laddr->s_addr) {
2139                         if (injail) {
2140                                 jail_exact = grp;
2141                                 if (in_pcblookup_lb_numa_match(grp, domain))
2142                                         /* This is a perfect match. */
2143                                         goto out;
2144                         } else if (local_exact == NULL ||
2145                             in_pcblookup_lb_numa_match(grp, domain)) {
2146                                 local_exact = grp;
2147                         }
2148                 } else if (grp->il_laddr.s_addr == INADDR_ANY) {
2149                         if (injail) {
2150                                 if (jail_wild == NULL ||
2151                                     in_pcblookup_lb_numa_match(grp, domain))
2152                                         jail_wild = grp;
2153                         } else if (local_wild == NULL ||
2154                             in_pcblookup_lb_numa_match(grp, domain)) {
2155                                 local_wild = grp;
2156                         }
2157                 }
2158         }
2159
2160         if (jail_exact != NULL)
2161                 grp = jail_exact;
2162         else if (jail_wild != NULL)
2163                 grp = jail_wild;
2164         else if (local_exact != NULL)
2165                 grp = local_exact;
2166         else
2167                 grp = local_wild;
2168         if (grp == NULL)
2169                 return (NULL);
2170 out:
2171         return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2172             grp->il_inpcnt]);
2173 }
2174
2175 static bool
2176 in_pcblookup_exact_match(const struct inpcb *inp, struct in_addr faddr,
2177     u_short fport, struct in_addr laddr, u_short lport)
2178 {
2179 #ifdef INET6
2180         /* XXX inp locking */
2181         if ((inp->inp_vflag & INP_IPV4) == 0)
2182                 return (false);
2183 #endif
2184         if (inp->inp_faddr.s_addr == faddr.s_addr &&
2185             inp->inp_laddr.s_addr == laddr.s_addr &&
2186             inp->inp_fport == fport &&
2187             inp->inp_lport == lport)
2188                 return (true);
2189         return (false);
2190 }
2191
2192 static struct inpcb *
2193 in_pcblookup_hash_exact(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2194     u_short fport, struct in_addr laddr, u_short lport)
2195 {
2196         struct inpcbhead *head;
2197         struct inpcb *inp;
2198
2199         INP_HASH_LOCK_ASSERT(pcbinfo);
2200
2201         head = &pcbinfo->ipi_hash_exact[INP_PCBHASH(&faddr, lport, fport,
2202             pcbinfo->ipi_hashmask)];
2203         CK_LIST_FOREACH(inp, head, inp_hash_exact) {
2204                 if (in_pcblookup_exact_match(inp, faddr, fport, laddr, lport))
2205                         return (inp);
2206         }
2207         return (NULL);
2208 }
2209
2210 typedef enum {
2211         INPLOOKUP_MATCH_NONE = 0,
2212         INPLOOKUP_MATCH_WILD = 1,
2213         INPLOOKUP_MATCH_LADDR = 2,
2214 } inp_lookup_match_t;
2215
2216 static inp_lookup_match_t
2217 in_pcblookup_wild_match(const struct inpcb *inp, struct in_addr laddr,
2218     u_short lport)
2219 {
2220 #ifdef INET6
2221         /* XXX inp locking */
2222         if ((inp->inp_vflag & INP_IPV4) == 0)
2223                 return (INPLOOKUP_MATCH_NONE);
2224 #endif
2225         if (inp->inp_faddr.s_addr != INADDR_ANY || inp->inp_lport != lport)
2226                 return (INPLOOKUP_MATCH_NONE);
2227         if (inp->inp_laddr.s_addr == INADDR_ANY)
2228                 return (INPLOOKUP_MATCH_WILD);
2229         if (inp->inp_laddr.s_addr == laddr.s_addr)
2230                 return (INPLOOKUP_MATCH_LADDR);
2231         return (INPLOOKUP_MATCH_NONE);
2232 }
2233
2234 #define INP_LOOKUP_AGAIN        ((struct inpcb *)(uintptr_t)-1)
2235
2236 static struct inpcb *
2237 in_pcblookup_hash_wild_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2238     u_short fport, struct in_addr laddr, u_short lport,
2239     const inp_lookup_t lockflags)
2240 {
2241         struct inpcbhead *head;
2242         struct inpcb *inp;
2243
2244         KASSERT(SMR_ENTERED(pcbinfo->ipi_smr),
2245             ("%s: not in SMR read section", __func__));
2246
2247         head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2248             pcbinfo->ipi_hashmask)];
2249         CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2250                 inp_lookup_match_t match;
2251
2252                 match = in_pcblookup_wild_match(inp, laddr, lport);
2253                 if (match == INPLOOKUP_MATCH_NONE)
2254                         continue;
2255
2256                 if (__predict_true(inp_smr_lock(inp, lockflags))) {
2257                         match = in_pcblookup_wild_match(inp, laddr, lport);
2258                         if (match != INPLOOKUP_MATCH_NONE &&
2259                             prison_check_ip4_locked(inp->inp_cred->cr_prison,
2260                             &laddr) == 0)
2261                                 return (inp);
2262                         inp_unlock(inp, lockflags);
2263                 }
2264
2265                 /*
2266                  * The matching socket disappeared out from under us.  Fall back
2267                  * to a serialized lookup.
2268                  */
2269                 return (INP_LOOKUP_AGAIN);
2270         }
2271         return (NULL);
2272 }
2273
2274 static struct inpcb *
2275 in_pcblookup_hash_wild_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2276     u_short fport, struct in_addr laddr, u_short lport)
2277 {
2278         struct inpcbhead *head;
2279         struct inpcb *inp, *local_wild, *local_exact, *jail_wild;
2280 #ifdef INET6
2281         struct inpcb *local_wild_mapped;
2282 #endif
2283
2284         INP_HASH_LOCK_ASSERT(pcbinfo);
2285
2286         /*
2287          * Order of socket selection - we always prefer jails.
2288          *      1. jailed, non-wild.
2289          *      2. jailed, wild.
2290          *      3. non-jailed, non-wild.
2291          *      4. non-jailed, wild.
2292          */
2293         head = &pcbinfo->ipi_hash_wild[INP_PCBHASH_WILD(lport,
2294             pcbinfo->ipi_hashmask)];
2295         local_wild = local_exact = jail_wild = NULL;
2296 #ifdef INET6
2297         local_wild_mapped = NULL;
2298 #endif
2299         CK_LIST_FOREACH(inp, head, inp_hash_wild) {
2300                 inp_lookup_match_t match;
2301                 bool injail;
2302
2303                 match = in_pcblookup_wild_match(inp, laddr, lport);
2304                 if (match == INPLOOKUP_MATCH_NONE)
2305                         continue;
2306
2307                 injail = prison_flag(inp->inp_cred, PR_IP4) != 0;
2308                 if (injail) {
2309                         if (prison_check_ip4_locked(inp->inp_cred->cr_prison,
2310                             &laddr) != 0)
2311                                 continue;
2312                 } else {
2313                         if (local_exact != NULL)
2314                                 continue;
2315                 }
2316
2317                 if (match == INPLOOKUP_MATCH_LADDR) {
2318                         if (injail)
2319                                 return (inp);
2320                         local_exact = inp;
2321                 } else {
2322 #ifdef INET6
2323                         /* XXX inp locking, NULL check */
2324                         if (inp->inp_vflag & INP_IPV6PROTO)
2325                                 local_wild_mapped = inp;
2326                         else
2327 #endif
2328                                 if (injail)
2329                                         jail_wild = inp;
2330                                 else
2331                                         local_wild = inp;
2332                 }
2333         }
2334         if (jail_wild != NULL)
2335                 return (jail_wild);
2336         if (local_exact != NULL)
2337                 return (local_exact);
2338         if (local_wild != NULL)
2339                 return (local_wild);
2340 #ifdef INET6
2341         if (local_wild_mapped != NULL)
2342                 return (local_wild_mapped);
2343 #endif
2344         return (NULL);
2345 }
2346
2347 /*
2348  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2349  * that the caller has either locked the hash list, which usually happens
2350  * for bind(2) operations, or is in SMR section, which happens when sorting
2351  * out incoming packets.
2352  */
2353 static struct inpcb *
2354 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2355     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2356     uint8_t numa_domain)
2357 {
2358         struct inpcb *inp;
2359         const u_short fport = fport_arg, lport = lport_arg;
2360
2361         KASSERT((lookupflags & ~INPLOOKUP_WILDCARD) == 0,
2362             ("%s: invalid lookup flags %d", __func__, lookupflags));
2363         KASSERT(faddr.s_addr != INADDR_ANY,
2364             ("%s: invalid foreign address", __func__));
2365         KASSERT(laddr.s_addr != INADDR_ANY,
2366             ("%s: invalid local address", __func__));
2367         INP_HASH_WLOCK_ASSERT(pcbinfo);
2368
2369         inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2370         if (inp != NULL)
2371                 return (inp);
2372
2373         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2374                 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport,
2375                     &laddr, lport, numa_domain);
2376                 if (inp == NULL) {
2377                         inp = in_pcblookup_hash_wild_locked(pcbinfo, faddr,
2378                             fport, laddr, lport);
2379                 }
2380         }
2381
2382         return (inp);
2383 }
2384
2385 static struct inpcb *
2386 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2387     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2388     uint8_t numa_domain)
2389 {
2390         struct inpcb *inp;
2391         const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK;
2392
2393         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2394             ("%s: LOCKPCB not set", __func__));
2395
2396         INP_HASH_WLOCK(pcbinfo);
2397         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2398             lookupflags & ~INPLOOKUP_LOCKMASK, numa_domain);
2399         if (inp != NULL && !inp_trylock(inp, lockflags)) {
2400                 in_pcbref(inp);
2401                 INP_HASH_WUNLOCK(pcbinfo);
2402                 inp_lock(inp, lockflags);
2403                 if (in_pcbrele(inp, lockflags))
2404                         /* XXX-MJ or retry until we get a negative match? */
2405                         inp = NULL;
2406         } else {
2407                 INP_HASH_WUNLOCK(pcbinfo);
2408         }
2409         return (inp);
2410 }
2411
2412 static struct inpcb *
2413 in_pcblookup_hash_smr(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2414     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2415     uint8_t numa_domain)
2416 {
2417         struct inpcb *inp;
2418         const inp_lookup_t lockflags = lookupflags & INPLOOKUP_LOCKMASK;
2419         const u_short fport = fport_arg, lport = lport_arg;
2420
2421         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2422             ("%s: invalid lookup flags %d", __func__, lookupflags));
2423         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2424             ("%s: LOCKPCB not set", __func__));
2425
2426         smr_enter(pcbinfo->ipi_smr);
2427         inp = in_pcblookup_hash_exact(pcbinfo, faddr, fport, laddr, lport);
2428         if (inp != NULL) {
2429                 if (__predict_true(inp_smr_lock(inp, lockflags))) {
2430                         /*
2431                          * Revalidate the 4-tuple, the socket could have been
2432                          * disconnected.
2433                          */
2434                         if (__predict_true(in_pcblookup_exact_match(inp,
2435                             faddr, fport, laddr, lport)))
2436                                 return (inp);
2437                         inp_unlock(inp, lockflags);
2438                 }
2439
2440                 /*
2441                  * We failed to lock the inpcb, or its connection state changed
2442                  * out from under us.  Fall back to a precise search.
2443                  */
2444                 return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2445                     lookupflags, numa_domain));
2446         }
2447
2448         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2449                 inp = in_pcblookup_lbgroup(pcbinfo, &faddr, fport,
2450                     &laddr, lport, numa_domain);
2451                 if (inp != NULL) {
2452                         if (__predict_true(inp_smr_lock(inp, lockflags))) {
2453                                 if (__predict_true(in_pcblookup_wild_match(inp,
2454                                     laddr, lport) != INPLOOKUP_MATCH_NONE))
2455                                         return (inp);
2456                                 inp_unlock(inp, lockflags);
2457                         }
2458                         inp = INP_LOOKUP_AGAIN;
2459                 } else {
2460                         inp = in_pcblookup_hash_wild_smr(pcbinfo, faddr, fport,
2461                             laddr, lport, lockflags);
2462                 }
2463                 if (inp == INP_LOOKUP_AGAIN) {
2464                         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr,
2465                             lport, lookupflags, numa_domain));
2466                 }
2467         }
2468
2469         if (inp == NULL)
2470                 smr_exit(pcbinfo->ipi_smr);
2471
2472         return (inp);
2473 }
2474
2475 /*
2476  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2477  * from which a pre-calculated hash value may be extracted.
2478  */
2479 struct inpcb *
2480 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2481     struct in_addr laddr, u_int lport, int lookupflags,
2482     struct ifnet *ifp __unused)
2483 {
2484         return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2485             lookupflags, M_NODOM));
2486 }
2487
2488 struct inpcb *
2489 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2490     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2491     struct ifnet *ifp __unused, struct mbuf *m)
2492 {
2493         return (in_pcblookup_hash_smr(pcbinfo, faddr, fport, laddr, lport,
2494             lookupflags, m->m_pkthdr.numa_domain));
2495 }
2496 #endif /* INET */
2497
2498 static bool
2499 in_pcbjailed(const struct inpcb *inp, unsigned int flag)
2500 {
2501         return (prison_flag(inp->inp_cred, flag) != 0);
2502 }
2503
2504 /*
2505  * Insert the PCB into a hash chain using ordering rules which ensure that
2506  * in_pcblookup_hash_wild_*() always encounter the highest-ranking PCB first.
2507  *
2508  * Specifically, keep jailed PCBs in front of non-jailed PCBs, and keep PCBs
2509  * with exact local addresses ahead of wildcard PCBs.  Unbound v4-mapped v6 PCBs
2510  * always appear last no matter whether they are jailed.
2511  */
2512 static void
2513 _in_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp)
2514 {
2515         struct inpcb *last;
2516         bool bound, injail;
2517
2518         INP_LOCK_ASSERT(inp);
2519         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2520
2521         last = NULL;
2522         bound = inp->inp_laddr.s_addr != INADDR_ANY;
2523         if (!bound && (inp->inp_vflag & INP_IPV6PROTO) != 0) {
2524                 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2525                         if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2526                                 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2527                                 return;
2528                         }
2529                 }
2530                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2531                 return;
2532         }
2533
2534         injail = in_pcbjailed(inp, PR_IP4);
2535         if (!injail) {
2536                 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2537                         if (!in_pcbjailed(last, PR_IP4))
2538                                 break;
2539                         if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2540                                 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2541                                 return;
2542                         }
2543                 }
2544         } else if (!CK_LIST_EMPTY(pcbhash) &&
2545             !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP4)) {
2546                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2547                 return;
2548         }
2549         if (!bound) {
2550                 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) {
2551                         if (last->inp_laddr.s_addr == INADDR_ANY)
2552                                 break;
2553                         if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2554                                 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2555                                 return;
2556                         }
2557                 }
2558         }
2559         if (last == NULL)
2560                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2561         else
2562                 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild);
2563 }
2564
2565 #ifdef INET6
2566 /*
2567  * See the comment above _in_pcbinshash_wild().
2568  */
2569 static void
2570 _in6_pcbinshash_wild(struct inpcbhead *pcbhash, struct inpcb *inp)
2571 {
2572         struct inpcb *last;
2573         bool bound, injail;
2574
2575         INP_LOCK_ASSERT(inp);
2576         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2577
2578         last = NULL;
2579         bound = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr);
2580         injail = in_pcbjailed(inp, PR_IP6);
2581         if (!injail) {
2582                 CK_LIST_FOREACH(last, pcbhash, inp_hash_wild) {
2583                         if (!in_pcbjailed(last, PR_IP6))
2584                                 break;
2585                         if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2586                                 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2587                                 return;
2588                         }
2589                 }
2590         } else if (!CK_LIST_EMPTY(pcbhash) &&
2591             !in_pcbjailed(CK_LIST_FIRST(pcbhash), PR_IP6)) {
2592                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2593                 return;
2594         }
2595         if (!bound) {
2596                 CK_LIST_FOREACH_FROM(last, pcbhash, inp_hash_wild) {
2597                         if (IN6_IS_ADDR_UNSPECIFIED(&last->in6p_laddr))
2598                                 break;
2599                         if (CK_LIST_NEXT(last, inp_hash_wild) == NULL) {
2600                                 CK_LIST_INSERT_AFTER(last, inp, inp_hash_wild);
2601                                 return;
2602                         }
2603                 }
2604         }
2605         if (last == NULL)
2606                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_wild);
2607         else
2608                 CK_LIST_INSERT_BEFORE(last, inp, inp_hash_wild);
2609 }
2610 #endif
2611
2612 /*
2613  * Insert PCB onto various hash lists.
2614  */
2615 int
2616 in_pcbinshash(struct inpcb *inp)
2617 {
2618         struct inpcbhead *pcbhash;
2619         struct inpcbporthead *pcbporthash;
2620         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2621         struct inpcbport *phd;
2622         uint32_t hash;
2623         bool connected;
2624
2625         INP_WLOCK_ASSERT(inp);
2626         INP_HASH_WLOCK_ASSERT(pcbinfo);
2627         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2628             ("in_pcbinshash: INP_INHASHLIST"));
2629
2630 #ifdef INET6
2631         if (inp->inp_vflag & INP_IPV6) {
2632                 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2633                     inp->inp_fport, pcbinfo->ipi_hashmask);
2634                 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2635         } else
2636 #endif
2637         {
2638                 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2639                     inp->inp_fport, pcbinfo->ipi_hashmask);
2640                 connected = !in_nullhost(inp->inp_faddr);
2641         }
2642
2643         if (connected)
2644                 pcbhash = &pcbinfo->ipi_hash_exact[hash];
2645         else
2646                 pcbhash = &pcbinfo->ipi_hash_wild[hash];
2647
2648         pcbporthash = &pcbinfo->ipi_porthashbase[
2649             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2650
2651         /*
2652          * Add entry to load balance group.
2653          * Only do this if SO_REUSEPORT_LB is set.
2654          */
2655         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2656                 int error = in_pcbinslbgrouphash(inp, M_NODOM);
2657                 if (error != 0)
2658                         return (error);
2659         }
2660
2661         /*
2662          * Go through port list and look for a head for this lport.
2663          */
2664         CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2665                 if (phd->phd_port == inp->inp_lport)
2666                         break;
2667         }
2668
2669         /*
2670          * If none exists, malloc one and tack it on.
2671          */
2672         if (phd == NULL) {
2673                 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2674                 if (phd == NULL) {
2675                         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2676                                 in_pcbremlbgrouphash(inp);
2677                         return (ENOMEM);
2678                 }
2679                 phd->phd_port = inp->inp_lport;
2680                 CK_LIST_INIT(&phd->phd_pcblist);
2681                 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2682         }
2683         inp->inp_phd = phd;
2684         CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2685
2686         /*
2687          * The PCB may have been disconnected in the past.  Before we can safely
2688          * make it visible in the hash table, we must wait for all readers which
2689          * may be traversing this PCB to finish.
2690          */
2691         if (inp->inp_smr != SMR_SEQ_INVALID) {
2692                 smr_wait(pcbinfo->ipi_smr, inp->inp_smr);
2693                 inp->inp_smr = SMR_SEQ_INVALID;
2694         }
2695
2696         if (connected)
2697                 CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash_exact);
2698         else {
2699 #ifdef INET6
2700                 if ((inp->inp_vflag & INP_IPV6) != 0)
2701                         _in6_pcbinshash_wild(pcbhash, inp);
2702                 else
2703 #endif
2704                         _in_pcbinshash_wild(pcbhash, inp);
2705         }
2706         inp->inp_flags |= INP_INHASHLIST;
2707
2708         return (0);
2709 }
2710
2711 void
2712 in_pcbremhash_locked(struct inpcb *inp)
2713 {
2714         struct inpcbport *phd = inp->inp_phd;
2715
2716         INP_WLOCK_ASSERT(inp);
2717         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
2718         MPASS(inp->inp_flags & INP_INHASHLIST);
2719
2720         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2721                 in_pcbremlbgrouphash(inp);
2722 #ifdef INET6
2723         if (inp->inp_vflag & INP_IPV6) {
2724                 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
2725                         CK_LIST_REMOVE(inp, inp_hash_wild);
2726                 else
2727                         CK_LIST_REMOVE(inp, inp_hash_exact);
2728         } else
2729 #endif
2730         {
2731                 if (in_nullhost(inp->inp_faddr))
2732                         CK_LIST_REMOVE(inp, inp_hash_wild);
2733                 else
2734                         CK_LIST_REMOVE(inp, inp_hash_exact);
2735         }
2736         CK_LIST_REMOVE(inp, inp_portlist);
2737         if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2738                 CK_LIST_REMOVE(phd, phd_hash);
2739                 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2740         }
2741         inp->inp_flags &= ~INP_INHASHLIST;
2742 }
2743
2744 static void
2745 in_pcbremhash(struct inpcb *inp)
2746 {
2747         INP_HASH_WLOCK(inp->inp_pcbinfo);
2748         in_pcbremhash_locked(inp);
2749         INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2750 }
2751
2752 /*
2753  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2754  * changed. NOTE: This does not handle the case of the lport changing (the
2755  * hashed port list would have to be updated as well), so the lport must
2756  * not change after in_pcbinshash() has been called.
2757  */
2758 void
2759 in_pcbrehash(struct inpcb *inp)
2760 {
2761         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2762         struct inpcbhead *head;
2763         uint32_t hash;
2764         bool connected;
2765
2766         INP_WLOCK_ASSERT(inp);
2767         INP_HASH_WLOCK_ASSERT(pcbinfo);
2768         KASSERT(inp->inp_flags & INP_INHASHLIST,
2769             ("%s: !INP_INHASHLIST", __func__));
2770         KASSERT(inp->inp_smr == SMR_SEQ_INVALID,
2771             ("%s: inp was disconnected", __func__));
2772
2773 #ifdef INET6
2774         if (inp->inp_vflag & INP_IPV6) {
2775                 hash = INP6_PCBHASH(&inp->in6p_faddr, inp->inp_lport,
2776                     inp->inp_fport, pcbinfo->ipi_hashmask);
2777                 connected = !IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr);
2778         } else
2779 #endif
2780         {
2781                 hash = INP_PCBHASH(&inp->inp_faddr, inp->inp_lport,
2782                     inp->inp_fport, pcbinfo->ipi_hashmask);
2783                 connected = !in_nullhost(inp->inp_faddr);
2784         }
2785
2786         /*
2787          * When rehashing, the caller must ensure that either the new or the old
2788          * foreign address was unspecified.
2789          */
2790         if (connected)
2791                 CK_LIST_REMOVE(inp, inp_hash_wild);
2792         else
2793                 CK_LIST_REMOVE(inp, inp_hash_exact);
2794
2795         if (connected) {
2796                 head = &pcbinfo->ipi_hash_exact[hash];
2797                 CK_LIST_INSERT_HEAD(head, inp, inp_hash_exact);
2798         } else {
2799                 head = &pcbinfo->ipi_hash_wild[hash];
2800                 CK_LIST_INSERT_HEAD(head, inp, inp_hash_wild);
2801         }
2802 }
2803
2804 /*
2805  * Check for alternatives when higher level complains
2806  * about service problems.  For now, invalidate cached
2807  * routing information.  If the route was created dynamically
2808  * (by a redirect), time to try a default gateway again.
2809  */
2810 void
2811 in_losing(struct inpcb *inp)
2812 {
2813
2814         RO_INVALIDATE_CACHE(&inp->inp_route);
2815         return;
2816 }
2817
2818 /*
2819  * A set label operation has occurred at the socket layer, propagate the
2820  * label change into the in_pcb for the socket.
2821  */
2822 void
2823 in_pcbsosetlabel(struct socket *so)
2824 {
2825 #ifdef MAC
2826         struct inpcb *inp;
2827
2828         inp = sotoinpcb(so);
2829         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2830
2831         INP_WLOCK(inp);
2832         SOCK_LOCK(so);
2833         mac_inpcb_sosetlabel(so, inp);
2834         SOCK_UNLOCK(so);
2835         INP_WUNLOCK(inp);
2836 #endif
2837 }
2838
2839 void
2840 inp_wlock(struct inpcb *inp)
2841 {
2842
2843         INP_WLOCK(inp);
2844 }
2845
2846 void
2847 inp_wunlock(struct inpcb *inp)
2848 {
2849
2850         INP_WUNLOCK(inp);
2851 }
2852
2853 void
2854 inp_rlock(struct inpcb *inp)
2855 {
2856
2857         INP_RLOCK(inp);
2858 }
2859
2860 void
2861 inp_runlock(struct inpcb *inp)
2862 {
2863
2864         INP_RUNLOCK(inp);
2865 }
2866
2867 #ifdef INVARIANT_SUPPORT
2868 void
2869 inp_lock_assert(struct inpcb *inp)
2870 {
2871
2872         INP_WLOCK_ASSERT(inp);
2873 }
2874
2875 void
2876 inp_unlock_assert(struct inpcb *inp)
2877 {
2878
2879         INP_UNLOCK_ASSERT(inp);
2880 }
2881 #endif
2882
2883 void
2884 inp_apply_all(struct inpcbinfo *pcbinfo,
2885     void (*func)(struct inpcb *, void *), void *arg)
2886 {
2887         struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2888             INPLOOKUP_WLOCKPCB);
2889         struct inpcb *inp;
2890
2891         while ((inp = inp_next(&inpi)) != NULL)
2892                 func(inp, arg);
2893 }
2894
2895 struct socket *
2896 inp_inpcbtosocket(struct inpcb *inp)
2897 {
2898
2899         INP_WLOCK_ASSERT(inp);
2900         return (inp->inp_socket);
2901 }
2902
2903 struct tcpcb *
2904 inp_inpcbtotcpcb(struct inpcb *inp)
2905 {
2906
2907         INP_WLOCK_ASSERT(inp);
2908         return ((struct tcpcb *)inp->inp_ppcb);
2909 }
2910
2911 int
2912 inp_ip_tos_get(const struct inpcb *inp)
2913 {
2914
2915         return (inp->inp_ip_tos);
2916 }
2917
2918 void
2919 inp_ip_tos_set(struct inpcb *inp, int val)
2920 {
2921
2922         inp->inp_ip_tos = val;
2923 }
2924
2925 void
2926 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2927     uint32_t *faddr, uint16_t *fp)
2928 {
2929
2930         INP_LOCK_ASSERT(inp);
2931         *laddr = inp->inp_laddr.s_addr;
2932         *faddr = inp->inp_faddr.s_addr;
2933         *lp = inp->inp_lport;
2934         *fp = inp->inp_fport;
2935 }
2936
2937 struct inpcb *
2938 so_sotoinpcb(struct socket *so)
2939 {
2940
2941         return (sotoinpcb(so));
2942 }
2943
2944 /*
2945  * Create an external-format (``xinpcb'') structure using the information in
2946  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2947  * reduce the spew of irrelevant information over this interface, to isolate
2948  * user code from changes in the kernel structure, and potentially to provide
2949  * information-hiding if we decide that some of this information should be
2950  * hidden from users.
2951  */
2952 void
2953 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2954 {
2955
2956         bzero(xi, sizeof(*xi));
2957         xi->xi_len = sizeof(struct xinpcb);
2958         if (inp->inp_socket)
2959                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2960         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2961         xi->inp_gencnt = inp->inp_gencnt;
2962         xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2963         xi->inp_flow = inp->inp_flow;
2964         xi->inp_flowid = inp->inp_flowid;
2965         xi->inp_flowtype = inp->inp_flowtype;
2966         xi->inp_flags = inp->inp_flags;
2967         xi->inp_flags2 = inp->inp_flags2;
2968         xi->in6p_cksum = inp->in6p_cksum;
2969         xi->in6p_hops = inp->in6p_hops;
2970         xi->inp_ip_tos = inp->inp_ip_tos;
2971         xi->inp_vflag = inp->inp_vflag;
2972         xi->inp_ip_ttl = inp->inp_ip_ttl;
2973         xi->inp_ip_p = inp->inp_ip_p;
2974         xi->inp_ip_minttl = inp->inp_ip_minttl;
2975 }
2976
2977 int
2978 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2979     int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2980 {
2981         struct sockopt sopt;
2982         struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2983             INPLOOKUP_WLOCKPCB);
2984         struct inpcb *inp;
2985         struct sockopt_parameters *params;
2986         struct socket *so;
2987         int error;
2988         char buf[1024];
2989
2990         if (req->oldptr != NULL || req->oldlen != 0)
2991                 return (EINVAL);
2992         if (req->newptr == NULL)
2993                 return (EPERM);
2994         if (req->newlen > sizeof(buf))
2995                 return (ENOMEM);
2996         error = SYSCTL_IN(req, buf, req->newlen);
2997         if (error != 0)
2998                 return (error);
2999         if (req->newlen < sizeof(struct sockopt_parameters))
3000                 return (EINVAL);
3001         params = (struct sockopt_parameters *)buf;
3002         sopt.sopt_level = params->sop_level;
3003         sopt.sopt_name = params->sop_optname;
3004         sopt.sopt_dir = SOPT_SET;
3005         sopt.sopt_val = params->sop_optval;
3006         sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
3007         sopt.sopt_td = NULL;
3008 #ifdef INET6
3009         if (params->sop_inc.inc_flags & INC_ISIPV6) {
3010                 if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_laddr))
3011                         params->sop_inc.inc6_laddr.s6_addr16[1] =
3012                             htons(params->sop_inc.inc6_zoneid & 0xffff);
3013                 if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_faddr))
3014                         params->sop_inc.inc6_faddr.s6_addr16[1] =
3015                             htons(params->sop_inc.inc6_zoneid & 0xffff);
3016         }
3017 #endif
3018         if (params->sop_inc.inc_lport != htons(0)) {
3019                 if (params->sop_inc.inc_fport == htons(0))
3020                         inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
3021                             pcbinfo->ipi_hashmask);
3022                 else
3023 #ifdef INET6
3024                         if (params->sop_inc.inc_flags & INC_ISIPV6)
3025                                 inpi.hash = INP6_PCBHASH(
3026                                     &params->sop_inc.inc6_faddr,
3027                                     params->sop_inc.inc_lport,
3028                                     params->sop_inc.inc_fport,
3029                                     pcbinfo->ipi_hashmask);
3030                         else
3031 #endif
3032                                 inpi.hash = INP_PCBHASH(
3033                                     &params->sop_inc.inc_faddr,
3034                                     params->sop_inc.inc_lport,
3035                                     params->sop_inc.inc_fport,
3036                                     pcbinfo->ipi_hashmask);
3037         }
3038         while ((inp = inp_next(&inpi)) != NULL)
3039                 if (inp->inp_gencnt == params->sop_id) {
3040                         if (inp->inp_flags & INP_DROPPED) {
3041                                 INP_WUNLOCK(inp);
3042                                 return (ECONNRESET);
3043                         }
3044                         so = inp->inp_socket;
3045                         KASSERT(so != NULL, ("inp_socket == NULL"));
3046                         soref(so);
3047                         error = (*ctloutput_set)(inp, &sopt);
3048                         sorele(so);
3049                         break;
3050                 }
3051         if (inp == NULL)
3052                 error = ESRCH;
3053         return (error);
3054 }
3055
3056 #ifdef DDB
3057 static void
3058 db_print_indent(int indent)
3059 {
3060         int i;
3061
3062         for (i = 0; i < indent; i++)
3063                 db_printf(" ");
3064 }
3065
3066 static void
3067 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
3068 {
3069         char faddr_str[48], laddr_str[48];
3070
3071         db_print_indent(indent);
3072         db_printf("%s at %p\n", name, inc);
3073
3074         indent += 2;
3075
3076 #ifdef INET6
3077         if (inc->inc_flags & INC_ISIPV6) {
3078                 /* IPv6. */
3079                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
3080                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
3081         } else
3082 #endif
3083         {
3084                 /* IPv4. */
3085                 inet_ntoa_r(inc->inc_laddr, laddr_str);
3086                 inet_ntoa_r(inc->inc_faddr, faddr_str);
3087         }
3088         db_print_indent(indent);
3089         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
3090             ntohs(inc->inc_lport));
3091         db_print_indent(indent);
3092         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
3093             ntohs(inc->inc_fport));
3094 }
3095
3096 static void
3097 db_print_inpflags(int inp_flags)
3098 {
3099         int comma;
3100
3101         comma = 0;
3102         if (inp_flags & INP_RECVOPTS) {
3103                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
3104                 comma = 1;
3105         }
3106         if (inp_flags & INP_RECVRETOPTS) {
3107                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
3108                 comma = 1;
3109         }
3110         if (inp_flags & INP_RECVDSTADDR) {
3111                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
3112                 comma = 1;
3113         }
3114         if (inp_flags & INP_ORIGDSTADDR) {
3115                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
3116                 comma = 1;
3117         }
3118         if (inp_flags & INP_HDRINCL) {
3119                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
3120                 comma = 1;
3121         }
3122         if (inp_flags & INP_HIGHPORT) {
3123                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
3124                 comma = 1;
3125         }
3126         if (inp_flags & INP_LOWPORT) {
3127                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
3128                 comma = 1;
3129         }
3130         if (inp_flags & INP_ANONPORT) {
3131                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
3132                 comma = 1;
3133         }
3134         if (inp_flags & INP_RECVIF) {
3135                 db_printf("%sINP_RECVIF", comma ? ", " : "");
3136                 comma = 1;
3137         }
3138         if (inp_flags & INP_MTUDISC) {
3139                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
3140                 comma = 1;
3141         }
3142         if (inp_flags & INP_RECVTTL) {
3143                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
3144                 comma = 1;
3145         }
3146         if (inp_flags & INP_DONTFRAG) {
3147                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
3148                 comma = 1;
3149         }
3150         if (inp_flags & INP_RECVTOS) {
3151                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
3152                 comma = 1;
3153         }
3154         if (inp_flags & IN6P_IPV6_V6ONLY) {
3155                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
3156                 comma = 1;
3157         }
3158         if (inp_flags & IN6P_PKTINFO) {
3159                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
3160                 comma = 1;
3161         }
3162         if (inp_flags & IN6P_HOPLIMIT) {
3163                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
3164                 comma = 1;
3165         }
3166         if (inp_flags & IN6P_HOPOPTS) {
3167                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
3168                 comma = 1;
3169         }
3170         if (inp_flags & IN6P_DSTOPTS) {
3171                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
3172                 comma = 1;
3173         }
3174         if (inp_flags & IN6P_RTHDR) {
3175                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
3176                 comma = 1;
3177         }
3178         if (inp_flags & IN6P_RTHDRDSTOPTS) {
3179                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
3180                 comma = 1;
3181         }
3182         if (inp_flags & IN6P_TCLASS) {
3183                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
3184                 comma = 1;
3185         }
3186         if (inp_flags & IN6P_AUTOFLOWLABEL) {
3187                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
3188                 comma = 1;
3189         }
3190         if (inp_flags & INP_ONESBCAST) {
3191                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
3192                 comma  = 1;
3193         }
3194         if (inp_flags & INP_DROPPED) {
3195                 db_printf("%sINP_DROPPED", comma ? ", " : "");
3196                 comma  = 1;
3197         }
3198         if (inp_flags & INP_SOCKREF) {
3199                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
3200                 comma  = 1;
3201         }
3202         if (inp_flags & IN6P_RFC2292) {
3203                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
3204                 comma = 1;
3205         }
3206         if (inp_flags & IN6P_MTU) {
3207                 db_printf("IN6P_MTU%s", comma ? ", " : "");
3208                 comma = 1;
3209         }
3210 }
3211
3212 static void
3213 db_print_inpvflag(u_char inp_vflag)
3214 {
3215         int comma;
3216
3217         comma = 0;
3218         if (inp_vflag & INP_IPV4) {
3219                 db_printf("%sINP_IPV4", comma ? ", " : "");
3220                 comma  = 1;
3221         }
3222         if (inp_vflag & INP_IPV6) {
3223                 db_printf("%sINP_IPV6", comma ? ", " : "");
3224                 comma  = 1;
3225         }
3226         if (inp_vflag & INP_IPV6PROTO) {
3227                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
3228                 comma  = 1;
3229         }
3230 }
3231
3232 static void
3233 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
3234 {
3235
3236         db_print_indent(indent);
3237         db_printf("%s at %p\n", name, inp);
3238
3239         indent += 2;
3240
3241         db_print_indent(indent);
3242         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
3243
3244         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
3245
3246         db_print_indent(indent);
3247         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
3248             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
3249
3250         db_print_indent(indent);
3251         db_printf("inp_label: %p   inp_flags: 0x%x (",
3252            inp->inp_label, inp->inp_flags);
3253         db_print_inpflags(inp->inp_flags);
3254         db_printf(")\n");
3255
3256         db_print_indent(indent);
3257         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
3258             inp->inp_vflag);
3259         db_print_inpvflag(inp->inp_vflag);
3260         db_printf(")\n");
3261
3262         db_print_indent(indent);
3263         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
3264             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
3265
3266         db_print_indent(indent);
3267 #ifdef INET6
3268         if (inp->inp_vflag & INP_IPV6) {
3269                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
3270                     "in6p_moptions: %p\n", inp->in6p_options,
3271                     inp->in6p_outputopts, inp->in6p_moptions);
3272                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
3273                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
3274                     inp->in6p_hops);
3275         } else
3276 #endif
3277         {
3278                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
3279                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
3280                     inp->inp_options, inp->inp_moptions);
3281         }
3282
3283         db_print_indent(indent);
3284         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
3285             (uintmax_t)inp->inp_gencnt);
3286 }
3287
3288 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3289 {
3290         struct inpcb *inp;
3291
3292         if (!have_addr) {
3293                 db_printf("usage: show inpcb <addr>\n");
3294                 return;
3295         }
3296         inp = (struct inpcb *)addr;
3297
3298         db_print_inpcb(inp, "inpcb", 0);
3299 }
3300 #endif /* DDB */
3301
3302 #ifdef RATELIMIT
3303 /*
3304  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3305  * if any.
3306  */
3307 int
3308 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3309 {
3310         union if_snd_tag_modify_params params = {
3311                 .rate_limit.max_rate = max_pacing_rate,
3312                 .rate_limit.flags = M_NOWAIT,
3313         };
3314         struct m_snd_tag *mst;
3315         int error;
3316
3317         mst = inp->inp_snd_tag;
3318         if (mst == NULL)
3319                 return (EINVAL);
3320
3321         if (mst->sw->snd_tag_modify == NULL) {
3322                 error = EOPNOTSUPP;
3323         } else {
3324                 error = mst->sw->snd_tag_modify(mst, &params);
3325         }
3326         return (error);
3327 }
3328
3329 /*
3330  * Query existing TX rate limit based on the existing
3331  * "inp->inp_snd_tag", if any.
3332  */
3333 int
3334 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3335 {
3336         union if_snd_tag_query_params params = { };
3337         struct m_snd_tag *mst;
3338         int error;
3339
3340         mst = inp->inp_snd_tag;
3341         if (mst == NULL)
3342                 return (EINVAL);
3343
3344         if (mst->sw->snd_tag_query == NULL) {
3345                 error = EOPNOTSUPP;
3346         } else {
3347                 error = mst->sw->snd_tag_query(mst, &params);
3348                 if (error == 0 && p_max_pacing_rate != NULL)
3349                         *p_max_pacing_rate = params.rate_limit.max_rate;
3350         }
3351         return (error);
3352 }
3353
3354 /*
3355  * Query existing TX queue level based on the existing
3356  * "inp->inp_snd_tag", if any.
3357  */
3358 int
3359 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3360 {
3361         union if_snd_tag_query_params params = { };
3362         struct m_snd_tag *mst;
3363         int error;
3364
3365         mst = inp->inp_snd_tag;
3366         if (mst == NULL)
3367                 return (EINVAL);
3368
3369         if (mst->sw->snd_tag_query == NULL)
3370                 return (EOPNOTSUPP);
3371
3372         error = mst->sw->snd_tag_query(mst, &params);
3373         if (error == 0 && p_txqueue_level != NULL)
3374                 *p_txqueue_level = params.rate_limit.queue_level;
3375         return (error);
3376 }
3377
3378 /*
3379  * Allocate a new TX rate limit send tag from the network interface
3380  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3381  */
3382 int
3383 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3384     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3385
3386 {
3387         union if_snd_tag_alloc_params params = {
3388                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3389                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3390                 .rate_limit.hdr.flowid = flowid,
3391                 .rate_limit.hdr.flowtype = flowtype,
3392                 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3393                 .rate_limit.max_rate = max_pacing_rate,
3394                 .rate_limit.flags = M_NOWAIT,
3395         };
3396         int error;
3397
3398         INP_WLOCK_ASSERT(inp);
3399
3400         /*
3401          * If there is already a send tag, or the INP is being torn
3402          * down, allocating a new send tag is not allowed. Else send
3403          * tags may leak.
3404          */
3405         if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3406                 return (EINVAL);
3407
3408         error = m_snd_tag_alloc(ifp, &params, st);
3409 #ifdef INET
3410         if (error == 0) {
3411                 counter_u64_add(rate_limit_set_ok, 1);
3412                 counter_u64_add(rate_limit_active, 1);
3413         } else if (error != EOPNOTSUPP)
3414                   counter_u64_add(rate_limit_alloc_fail, 1);
3415 #endif
3416         return (error);
3417 }
3418
3419 void
3420 in_pcbdetach_tag(struct m_snd_tag *mst)
3421 {
3422
3423         m_snd_tag_rele(mst);
3424 #ifdef INET
3425         counter_u64_add(rate_limit_active, -1);
3426 #endif
3427 }
3428
3429 /*
3430  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3431  * if any:
3432  */
3433 void
3434 in_pcbdetach_txrtlmt(struct inpcb *inp)
3435 {
3436         struct m_snd_tag *mst;
3437
3438         INP_WLOCK_ASSERT(inp);
3439
3440         mst = inp->inp_snd_tag;
3441         inp->inp_snd_tag = NULL;
3442
3443         if (mst == NULL)
3444                 return;
3445
3446         m_snd_tag_rele(mst);
3447 #ifdef INET
3448         counter_u64_add(rate_limit_active, -1);
3449 #endif
3450 }
3451
3452 int
3453 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3454 {
3455         int error;
3456
3457         /*
3458          * If the existing send tag is for the wrong interface due to
3459          * a route change, first drop the existing tag.  Set the
3460          * CHANGED flag so that we will keep trying to allocate a new
3461          * tag if we fail to allocate one this time.
3462          */
3463         if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3464                 in_pcbdetach_txrtlmt(inp);
3465                 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3466         }
3467
3468         /*
3469          * NOTE: When attaching to a network interface a reference is
3470          * made to ensure the network interface doesn't go away until
3471          * all ratelimit connections are gone. The network interface
3472          * pointers compared below represent valid network interfaces,
3473          * except when comparing towards NULL.
3474          */
3475         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3476                 error = 0;
3477         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3478                 if (inp->inp_snd_tag != NULL)
3479                         in_pcbdetach_txrtlmt(inp);
3480                 error = 0;
3481         } else if (inp->inp_snd_tag == NULL) {
3482                 /*
3483                  * In order to utilize packet pacing with RSS, we need
3484                  * to wait until there is a valid RSS hash before we
3485                  * can proceed:
3486                  */
3487                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3488                         error = EAGAIN;
3489                 } else {
3490                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3491                             mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3492                 }
3493         } else {
3494                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3495         }
3496         if (error == 0 || error == EOPNOTSUPP)
3497                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3498
3499         return (error);
3500 }
3501
3502 /*
3503  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3504  * is set in the fast path and will attach/detach/modify the TX rate
3505  * limit send tag based on the socket's so_max_pacing_rate value.
3506  */
3507 void
3508 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3509 {
3510         struct socket *socket;
3511         uint32_t max_pacing_rate;
3512         bool did_upgrade;
3513
3514         if (inp == NULL)
3515                 return;
3516
3517         socket = inp->inp_socket;
3518         if (socket == NULL)
3519                 return;
3520
3521         if (!INP_WLOCKED(inp)) {
3522                 /*
3523                  * NOTE: If the write locking fails, we need to bail
3524                  * out and use the non-ratelimited ring for the
3525                  * transmit until there is a new chance to get the
3526                  * write lock.
3527                  */
3528                 if (!INP_TRY_UPGRADE(inp))
3529                         return;
3530                 did_upgrade = 1;
3531         } else {
3532                 did_upgrade = 0;
3533         }
3534
3535         /*
3536          * NOTE: The so_max_pacing_rate value is read unlocked,
3537          * because atomic updates are not required since the variable
3538          * is checked at every mbuf we send. It is assumed that the
3539          * variable read itself will be atomic.
3540          */
3541         max_pacing_rate = socket->so_max_pacing_rate;
3542
3543         in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3544
3545         if (did_upgrade)
3546                 INP_DOWNGRADE(inp);
3547 }
3548
3549 /*
3550  * Track route changes for TX rate limiting.
3551  */
3552 void
3553 in_pcboutput_eagain(struct inpcb *inp)
3554 {
3555         bool did_upgrade;
3556
3557         if (inp == NULL)
3558                 return;
3559
3560         if (inp->inp_snd_tag == NULL)
3561                 return;
3562
3563         if (!INP_WLOCKED(inp)) {
3564                 /*
3565                  * NOTE: If the write locking fails, we need to bail
3566                  * out and use the non-ratelimited ring for the
3567                  * transmit until there is a new chance to get the
3568                  * write lock.
3569                  */
3570                 if (!INP_TRY_UPGRADE(inp))
3571                         return;
3572                 did_upgrade = 1;
3573         } else {
3574                 did_upgrade = 0;
3575         }
3576
3577         /* detach rate limiting */
3578         in_pcbdetach_txrtlmt(inp);
3579
3580         /* make sure new mbuf send tag allocation is made */
3581         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3582
3583         if (did_upgrade)
3584                 INP_DOWNGRADE(inp);
3585 }
3586
3587 #ifdef INET
3588 static void
3589 rl_init(void *st)
3590 {
3591         rate_limit_new = counter_u64_alloc(M_WAITOK);
3592         rate_limit_chg = counter_u64_alloc(M_WAITOK);
3593         rate_limit_active = counter_u64_alloc(M_WAITOK);
3594         rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3595         rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3596 }
3597
3598 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3599 #endif
3600 #endif /* RATELIMIT */