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/protosw.h>
  62 #include <sys/smp.h>
  63 #include <sys/socket.h>
  64 #include <sys/socketvar.h>
  65 #include <sys/sockio.h>
  66 #include <sys/priv.h>
  67 #include <sys/proc.h>
  68 #include <sys/refcount.h>
  69 #include <sys/jail.h>
  70 #include <sys/kernel.h>
  71 #include <sys/sysctl.h>
  72
  73 #ifdef DDB
  74 #include <ddb/ddb.h>
  75 #endif
  76
  77 #include <vm/uma.h>
  78 #include <vm/vm.h>
  79
  80 #include <net/if.h>
  81 #include <net/if_var.h>
  82 #include <net/if_private.h>
  83 #include <net/if_types.h>
  84 #include <net/if_llatbl.h>
  85 #include <net/route.h>
  86 #include <net/rss_config.h>
  87 #include <net/vnet.h>
  88
  89 #if defined(INET) || defined(INET6)
  90 #include <netinet/in.h>
  91 #include <netinet/in_pcb.h>
  92 #include <netinet/in_pcb_var.h>
  93 #include <netinet/tcp.h>
  94 #ifdef INET
  95 #include <netinet/in_var.h>
  96 #include <netinet/in_fib.h>
  97 #endif
  98 #include <netinet/ip_var.h>
  99 #ifdef INET6
 100 #include <netinet/ip6.h>
 101 #include <netinet6/in6_pcb.h>
 102 #include <netinet6/in6_var.h>
 103 #include <netinet6/ip6_var.h>
 104 #endif /* INET6 */
 105 #include <net/route/nhop.h>
 106 #endif
 107
 108 #include <netipsec/ipsec_support.h>
 109
 110 #include <security/mac/mac_framework.h>
 111
 112 #define INPCBLBGROUP_SIZMIN     8
 113 #define INPCBLBGROUP_SIZMAX     256
 114 #define INP_FREED       0x00000200      /* See in_pcb.h. */
 115
 116 /*
 117  * These configure the range of local port addresses assigned to
 118  * "unspecified" outgoing connections/packets/whatever.
 119  */
 120 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;    /* 1023 */
 121 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;    /* 600 */
 122 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;     /* 10000 */
 123 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;       /* 65535 */
 124 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;      /* 49152 */
 125 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;        /* 65535 */
 126
 127 /*
 128  * Reserved ports accessible only to root. There are significant
 129  * security considerations that must be accounted for when changing these,
 130  * but the security benefits can be great. Please be careful.
 131  */
 132 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;    /* 1023 */
 133 VNET_DEFINE(int, ipport_reservedlow);
 134
 135 /* Enable random ephemeral port allocation by default. */
 136 VNET_DEFINE(int, ipport_randomized) = 1;
 137
 138 #ifdef INET
 139 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 140                             struct in_addr faddr, u_int fport_arg,
 141                             struct in_addr laddr, u_int lport_arg,
 142                             int lookupflags, struct ifnet *ifp,
 143                             uint8_t numa_domain);
 144
 145 #define RANGECHK(var, min, max) \
 146         if ((var) < (min)) { (var) = (min); } \
 147         else if ((var) > (max)) { (var) = (max); }
 148
 149 static int
 150 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 151 {
 152         int error;
 153
 154         error = sysctl_handle_int(oidp, arg1, arg2, req);
 155         if (error == 0) {
 156                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 157                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 158                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 159                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 160                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 161                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 162         }
 163         return (error);
 164 }
 165
 166 #undef RANGECHK
 167
 168 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange,
 169     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 170     "IP Ports");
 171
 172 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 173     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 174     &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I",
 175     "");
 176 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 177     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 178     &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I",
 179     "");
 180 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 181     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 182     &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I",
 183     "");
 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 185     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 186     &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I",
 187     "");
 188 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 189     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 190     &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I",
 191     "");
 192 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 193     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 194     &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I",
 195     "");
 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 197         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 198         &VNET_NAME(ipport_reservedhigh), 0, "");
 199 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 200         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 202         CTLFLAG_VNET | CTLFLAG_RW,
 203         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 204
 205 #ifdef RATELIMIT
 206 counter_u64_t rate_limit_new;
 207 counter_u64_t rate_limit_chg;
 208 counter_u64_t rate_limit_active;
 209 counter_u64_t rate_limit_alloc_fail;
 210 counter_u64_t rate_limit_set_ok;
 211
 212 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, rl, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
 213     "IP Rate Limiting");
 214 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, active, CTLFLAG_RD,
 215     &rate_limit_active, "Active rate limited connections");
 216 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, alloc_fail, CTLFLAG_RD,
 217    &rate_limit_alloc_fail, "Rate limited connection failures");
 218 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD,
 219    &rate_limit_set_ok, "Rate limited setting succeeded");
 220 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD,
 221    &rate_limit_new, "Total Rate limit new attempts");
 222 SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD,
 223    &rate_limit_chg, "Total Rate limited change attempts");
 224
 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_hashbase = hashinit(hash_nelements, M_PCB,
 515             &pcbinfo->ipi_hashmask);
 516         porthash_nelements = imin(porthash_nelements, IPPORT_MAX + 1);
 517         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 518             &pcbinfo->ipi_porthashmask);
 519         pcbinfo->ipi_lbgrouphashbase = hashinit(porthash_nelements, M_PCB,
 520             &pcbinfo->ipi_lbgrouphashmask);
 521         pcbinfo->ipi_zone = pcbstor->ips_zone;
 522         pcbinfo->ipi_portzone = pcbstor->ips_portzone;
 523         pcbinfo->ipi_smr = uma_zone_get_smr(pcbinfo->ipi_zone);
 524 }
 525
 526 /*
 527  * Destroy an inpcbinfo.
 528  */
 529 void
 530 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 531 {
 532
 533         KASSERT(pcbinfo->ipi_count == 0,
 534             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 535
 536         hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
 537         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 538             pcbinfo->ipi_porthashmask);
 539         hashdestroy(pcbinfo->ipi_lbgrouphashbase, M_PCB,
 540             pcbinfo->ipi_lbgrouphashmask);
 541         mtx_destroy(&pcbinfo->ipi_hash_lock);
 542         mtx_destroy(&pcbinfo->ipi_lock);
 543 }
 544
 545 /*
 546  * Initialize a pcbstorage - per protocol zones to allocate inpcbs.
 547  */
 548 static void inpcb_dtor(void *, int, void *);
 549 static void inpcb_fini(void *, int);
 550 void
 551 in_pcbstorage_init(void *arg)
 552 {
 553         struct inpcbstorage *pcbstor = arg;
 554
 555         pcbstor->ips_zone = uma_zcreate(pcbstor->ips_zone_name,
 556             pcbstor->ips_size, NULL, inpcb_dtor, pcbstor->ips_pcbinit,
 557             inpcb_fini, UMA_ALIGN_CACHE, UMA_ZONE_SMR);
 558         pcbstor->ips_portzone = uma_zcreate(pcbstor->ips_portzone_name,
 559             sizeof(struct inpcbport), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 560         uma_zone_set_smr(pcbstor->ips_portzone,
 561             uma_zone_get_smr(pcbstor->ips_zone));
 562 }
 563
 564 /*
 565  * Destroy a pcbstorage - used by unloadable protocols.
 566  */
 567 void
 568 in_pcbstorage_destroy(void *arg)
 569 {
 570         struct inpcbstorage *pcbstor = arg;
 571
 572         uma_zdestroy(pcbstor->ips_zone);
 573         uma_zdestroy(pcbstor->ips_portzone);
 574 }
 575
 576 /*
 577  * Allocate a PCB and associate it with the socket.
 578  * On success return with the PCB locked.
 579  */
 580 int
 581 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 582 {
 583         struct inpcb *inp;
 584 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 585         int error;
 586 #endif
 587
 588         inp = uma_zalloc_smr(pcbinfo->ipi_zone, M_NOWAIT);
 589         if (inp == NULL)
 590                 return (ENOBUFS);
 591         bzero(&inp->inp_start_zero, inp_zero_size);
 592 #ifdef NUMA
 593         inp->inp_numa_domain = M_NODOM;
 594 #endif
 595         inp->inp_pcbinfo = pcbinfo;
 596         inp->inp_socket = so;
 597         inp->inp_cred = crhold(so->so_cred);
 598         inp->inp_inc.inc_fibnum = so->so_fibnum;
 599 #ifdef MAC
 600         error = mac_inpcb_init(inp, M_NOWAIT);
 601         if (error != 0)
 602                 goto out;
 603         mac_inpcb_create(so, inp);
 604 #endif
 605 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 606         error = ipsec_init_pcbpolicy(inp);
 607         if (error != 0) {
 608 #ifdef MAC
 609                 mac_inpcb_destroy(inp);
 610 #endif
 611                 goto out;
 612         }
 613 #endif /*IPSEC*/
 614 #ifdef INET6
 615         if (INP_SOCKAF(so) == AF_INET6) {
 616                 inp->inp_vflag |= INP_IPV6PROTO | INP_IPV6;
 617                 if (V_ip6_v6only)
 618                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 619 #ifdef INET
 620                 else
 621                         inp->inp_vflag |= INP_IPV4;
 622 #endif
 623                 if (V_ip6_auto_flowlabel)
 624                         inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 625                 inp->in6p_hops = -1;    /* use kernel default */
 626         }
 627 #endif
 628 #if defined(INET) && defined(INET6)
 629         else
 630 #endif
 631 #ifdef INET
 632                 inp->inp_vflag |= INP_IPV4;
 633 #endif
 634         /*
 635          * Routes in inpcb's can cache L2 as well; they are guaranteed
 636          * to be cleaned up.
 637          */
 638         inp->inp_route.ro_flags = RT_LLE_CACHE;
 639         refcount_init(&inp->inp_refcount, 1);   /* Reference from socket. */
 640         INP_WLOCK(inp);
 641         INP_INFO_WLOCK(pcbinfo);
 642         pcbinfo->ipi_count++;
 643         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 644         CK_LIST_INSERT_HEAD(&pcbinfo->ipi_listhead, inp, inp_list);
 645         INP_INFO_WUNLOCK(pcbinfo);
 646         so->so_pcb = inp;
 647
 648         return (0);
 649
 650 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 651 out:
 652         uma_zfree_smr(pcbinfo->ipi_zone, inp);
 653         return (error);
 654 #endif
 655 }
 656
 657 #ifdef INET
 658 int
 659 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 660 {
 661         int anonport, error;
 662
 663         KASSERT(nam == NULL || nam->sa_family == AF_INET,
 664             ("%s: invalid address family for %p", __func__, nam));
 665         KASSERT(nam == NULL || nam->sa_len == sizeof(struct sockaddr_in),
 666             ("%s: invalid address length for %p", __func__, nam));
 667         INP_WLOCK_ASSERT(inp);
 668         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 669
 670         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 671                 return (EINVAL);
 672         anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
 673         error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
 674             &inp->inp_lport, cred);
 675         if (error)
 676                 return (error);
 677         if (in_pcbinshash(inp) != 0) {
 678                 inp->inp_laddr.s_addr = INADDR_ANY;
 679                 inp->inp_lport = 0;
 680                 return (EAGAIN);
 681         }
 682         if (anonport)
 683                 inp->inp_flags |= INP_ANONPORT;
 684         return (0);
 685 }
 686 #endif
 687
 688 #if defined(INET) || defined(INET6)
 689 /*
 690  * Assign a local port like in_pcb_lport(), but also used with connect()
 691  * and a foreign address and port.  If fsa is non-NULL, choose a local port
 692  * that is unused with those, otherwise one that is completely unused.
 693  * lsa can be NULL for IPv6.
 694  */
 695 int
 696 in_pcb_lport_dest(struct inpcb *inp, struct sockaddr *lsa, u_short *lportp,
 697     struct sockaddr *fsa, u_short fport, struct ucred *cred, int lookupflags)
 698 {
 699         struct inpcbinfo *pcbinfo;
 700         struct inpcb *tmpinp;
 701         unsigned short *lastport;
 702         int count, error;
 703         u_short aux, first, last, lport;
 704 #ifdef INET
 705         struct in_addr laddr, faddr;
 706 #endif
 707 #ifdef INET6
 708         struct in6_addr *laddr6, *faddr6;
 709 #endif
 710
 711         pcbinfo = inp->inp_pcbinfo;
 712
 713         /*
 714          * Because no actual state changes occur here, a global write lock on
 715          * the pcbinfo isn't required.
 716          */
 717         INP_LOCK_ASSERT(inp);
 718         INP_HASH_LOCK_ASSERT(pcbinfo);
 719
 720         if (inp->inp_flags & INP_HIGHPORT) {
 721                 first = V_ipport_hifirstauto;   /* sysctl */
 722                 last  = V_ipport_hilastauto;
 723                 lastport = &pcbinfo->ipi_lasthi;
 724         } else if (inp->inp_flags & INP_LOWPORT) {
 725                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT);
 726                 if (error)
 727                         return (error);
 728                 first = V_ipport_lowfirstauto;  /* 1023 */
 729                 last  = V_ipport_lowlastauto;   /* 600 */
 730                 lastport = &pcbinfo->ipi_lastlow;
 731         } else {
 732                 first = V_ipport_firstauto;     /* sysctl */
 733                 last  = V_ipport_lastauto;
 734                 lastport = &pcbinfo->ipi_lastport;
 735         }
 736
 737         /*
 738          * Instead of having two loops further down counting up or down
 739          * make sure that first is always <= last and go with only one
 740          * code path implementing all logic.
 741          */
 742         if (first > last) {
 743                 aux = first;
 744                 first = last;
 745                 last = aux;
 746         }
 747
 748 #ifdef INET
 749         laddr.s_addr = INADDR_ANY;      /* used by INET6+INET below too */
 750         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 751                 if (lsa != NULL)
 752                         laddr = ((struct sockaddr_in *)lsa)->sin_addr;
 753                 if (fsa != NULL)
 754                         faddr = ((struct sockaddr_in *)fsa)->sin_addr;
 755         }
 756 #endif
 757 #ifdef INET6
 758         laddr6 = NULL;
 759         if ((inp->inp_vflag & INP_IPV6) != 0) {
 760                 if (lsa != NULL)
 761                         laddr6 = &((struct sockaddr_in6 *)lsa)->sin6_addr;
 762                 if (fsa != NULL)
 763                         faddr6 = &((struct sockaddr_in6 *)fsa)->sin6_addr;
 764         }
 765 #endif
 766
 767         tmpinp = NULL;
 768         lport = *lportp;
 769
 770         if (V_ipport_randomized)
 771                 *lastport = first + (arc4random() % (last - first));
 772
 773         count = last - first;
 774
 775         do {
 776                 if (count-- < 0)        /* completely used? */
 777                         return (EADDRNOTAVAIL);
 778                 ++*lastport;
 779                 if (*lastport < first || *lastport > last)
 780                         *lastport = first;
 781                 lport = htons(*lastport);
 782
 783                 if (fsa != NULL) {
 784 #ifdef INET
 785                         if (lsa->sa_family == AF_INET) {
 786                                 tmpinp = in_pcblookup_hash_locked(pcbinfo,
 787                                     faddr, fport, laddr, lport, lookupflags,
 788                                     NULL, M_NODOM);
 789                         }
 790 #endif
 791 #ifdef INET6
 792                         if (lsa->sa_family == AF_INET6) {
 793                                 tmpinp = in6_pcblookup_hash_locked(pcbinfo,
 794                                     faddr6, fport, laddr6, lport, lookupflags,
 795                                     NULL, M_NODOM);
 796                         }
 797 #endif
 798                 } else {
 799 #ifdef INET6
 800                         if ((inp->inp_vflag & INP_IPV6) != 0) {
 801                                 tmpinp = in6_pcblookup_local(pcbinfo,
 802                                     &inp->in6p_laddr, lport, lookupflags, cred);
 803 #ifdef INET
 804                                 if (tmpinp == NULL &&
 805                                     (inp->inp_vflag & INP_IPV4))
 806                                         tmpinp = in_pcblookup_local(pcbinfo,
 807                                             laddr, lport, lookupflags, cred);
 808 #endif
 809                         }
 810 #endif
 811 #if defined(INET) && defined(INET6)
 812                         else
 813 #endif
 814 #ifdef INET
 815                                 tmpinp = in_pcblookup_local(pcbinfo, laddr,
 816                                     lport, lookupflags, cred);
 817 #endif
 818                 }
 819         } while (tmpinp != NULL);
 820
 821         *lportp = lport;
 822
 823         return (0);
 824 }
 825
 826 /*
 827  * Select a local port (number) to use.
 828  */
 829 int
 830 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 831     struct ucred *cred, int lookupflags)
 832 {
 833         struct sockaddr_in laddr;
 834
 835         if (laddrp) {
 836                 bzero(&laddr, sizeof(laddr));
 837                 laddr.sin_family = AF_INET;
 838                 laddr.sin_addr = *laddrp;
 839         }
 840         return (in_pcb_lport_dest(inp, laddrp ? (struct sockaddr *) &laddr :
 841             NULL, lportp, NULL, 0, cred, lookupflags));
 842 }
 843
 844 /*
 845  * Return cached socket options.
 846  */
 847 int
 848 inp_so_options(const struct inpcb *inp)
 849 {
 850         int so_options;
 851
 852         so_options = 0;
 853
 854         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
 855                 so_options |= SO_REUSEPORT_LB;
 856         if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 857                 so_options |= SO_REUSEPORT;
 858         if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 859                 so_options |= SO_REUSEADDR;
 860         return (so_options);
 861 }
 862 #endif /* INET || INET6 */
 863
 864 /*
 865  * Check if a new BINDMULTI socket is allowed to be created.
 866  *
 867  * ni points to the new inp.
 868  * oi points to the existing inp.
 869  *
 870  * This checks whether the existing inp also has BINDMULTI and
 871  * whether the credentials match.
 872  */
 873 int
 874 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
 875 {
 876         /* Check permissions match */
 877         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 878             (ni->inp_cred->cr_uid !=
 879             oi->inp_cred->cr_uid))
 880                 return (0);
 881
 882         /* Check the existing inp has BINDMULTI set */
 883         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 884             ((oi->inp_flags2 & INP_BINDMULTI) == 0))
 885                 return (0);
 886
 887         /*
 888          * We're okay - either INP_BINDMULTI isn't set on ni, or
 889          * it is and it matches the checks.
 890          */
 891         return (1);
 892 }
 893
 894 #ifdef INET
 895 /*
 896  * Set up a bind operation on a PCB, performing port allocation
 897  * as required, but do not actually modify the PCB. Callers can
 898  * either complete the bind by setting inp_laddr/inp_lport and
 899  * calling in_pcbinshash(), or they can just use the resulting
 900  * port and address to authorise the sending of a once-off packet.
 901  *
 902  * On error, the values of *laddrp and *lportp are not changed.
 903  */
 904 int
 905 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
 906     u_short *lportp, struct ucred *cred)
 907 {
 908         struct socket *so = inp->inp_socket;
 909         struct sockaddr_in *sin;
 910         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 911         struct in_addr laddr;
 912         u_short lport = 0;
 913         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 914         int error;
 915
 916         /*
 917          * XXX: Maybe we could let SO_REUSEPORT_LB set SO_REUSEPORT bit here
 918          * so that we don't have to add to the (already messy) code below.
 919          */
 920         int reuseport_lb = (so->so_options & SO_REUSEPORT_LB);
 921
 922         /*
 923          * No state changes, so read locks are sufficient here.
 924          */
 925         INP_LOCK_ASSERT(inp);
 926         INP_HASH_LOCK_ASSERT(pcbinfo);
 927
 928         laddr.s_addr = *laddrp;
 929         if (nam != NULL && laddr.s_addr != INADDR_ANY)
 930                 return (EINVAL);
 931         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT|SO_REUSEPORT_LB)) == 0)
 932                 lookupflags = INPLOOKUP_WILDCARD;
 933         if (nam == NULL) {
 934                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 935                         return (error);
 936         } else {
 937                 sin = (struct sockaddr_in *)nam;
 938                 KASSERT(sin->sin_family == AF_INET,
 939                     ("%s: invalid family for address %p", __func__, sin));
 940                 KASSERT(sin->sin_len == sizeof(*sin),
 941                     ("%s: invalid length for address %p", __func__, sin));
 942
 943                 error = prison_local_ip4(cred, &sin->sin_addr);
 944                 if (error)
 945                         return (error);
 946                 if (sin->sin_port != *lportp) {
 947                         /* Don't allow the port to change. */
 948                         if (*lportp != 0)
 949                                 return (EINVAL);
 950                         lport = sin->sin_port;
 951                 }
 952                 /* NB: lport is left as 0 if the port isn't being changed. */
 953                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 954                         /*
 955                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 956                          * allow complete duplication of binding if
 957                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 958                          * and a multicast address is bound on both
 959                          * new and duplicated sockets.
 960                          */
 961                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 962                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 963                         /*
 964                          * XXX: How to deal with SO_REUSEPORT_LB here?
 965                          * Treat same as SO_REUSEPORT for now.
 966                          */
 967                         if ((so->so_options &
 968                             (SO_REUSEADDR|SO_REUSEPORT_LB)) != 0)
 969                                 reuseport_lb = SO_REUSEADDR|SO_REUSEPORT_LB;
 970                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 971                         sin->sin_port = 0;              /* yech... */
 972                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 973                         /*
 974                          * Is the address a local IP address?
 975                          * If INP_BINDANY is set, then the socket may be bound
 976                          * to any endpoint address, local or not.
 977                          */
 978                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 979                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 980                                 return (EADDRNOTAVAIL);
 981                 }
 982                 laddr = sin->sin_addr;
 983                 if (lport) {
 984                         struct inpcb *t;
 985
 986                         /* GROSS */
 987                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 988                             ntohs(lport) >= V_ipport_reservedlow &&
 989                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT))
 990                                 return (EACCES);
 991                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 992                             priv_check_cred(inp->inp_cred, PRIV_NETINET_REUSEPORT) != 0) {
 993                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 994                                     lport, INPLOOKUP_WILDCARD, cred);
 995         /*
 996          * XXX
 997          * This entire block sorely needs a rewrite.
 998          */
 999                                 if (t &&
1000                                     ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
1001                                     (so->so_type != SOCK_STREAM ||
1002                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
1003                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
1004                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
1005                                      (t->inp_flags2 & INP_REUSEPORT) ||
1006                                      (t->inp_flags2 & INP_REUSEPORT_LB) == 0) &&
1007                                     (inp->inp_cred->cr_uid !=
1008                                      t->inp_cred->cr_uid))
1009                                         return (EADDRINUSE);
1010
1011                                 /*
1012                                  * If the socket is a BINDMULTI socket, then
1013                                  * the credentials need to match and the
1014                                  * original socket also has to have been bound
1015                                  * with BINDMULTI.
1016                                  */
1017                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1018                                         return (EADDRINUSE);
1019                         }
1020                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
1021                             lport, lookupflags, cred);
1022                         if (t && ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
1023                             (reuseport & inp_so_options(t)) == 0 &&
1024                             (reuseport_lb & inp_so_options(t)) == 0) {
1025 #ifdef INET6
1026                                 if (ntohl(sin->sin_addr.s_addr) !=
1027                                     INADDR_ANY ||
1028                                     ntohl(t->inp_laddr.s_addr) !=
1029                                     INADDR_ANY ||
1030                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
1031                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
1032 #endif
1033                                                 return (EADDRINUSE);
1034                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
1035                                         return (EADDRINUSE);
1036                         }
1037                 }
1038         }
1039         if (*lportp != 0)
1040                 lport = *lportp;
1041         if (lport == 0) {
1042                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
1043                 if (error != 0)
1044                         return (error);
1045         }
1046         *laddrp = laddr.s_addr;
1047         *lportp = lport;
1048         return (0);
1049 }
1050
1051 /*
1052  * Connect from a socket to a specified address.
1053  * Both address and port must be specified in argument sin.
1054  * If don't have a local address for this socket yet,
1055  * then pick one.
1056  */
1057 int
1058 in_pcbconnect(struct inpcb *inp, struct sockaddr_in *sin, struct ucred *cred,
1059     bool rehash)
1060 {
1061         u_short lport, fport;
1062         in_addr_t laddr, faddr;
1063         int anonport, error;
1064
1065         INP_WLOCK_ASSERT(inp);
1066         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1067
1068         lport = inp->inp_lport;
1069         laddr = inp->inp_laddr.s_addr;
1070         anonport = (lport == 0);
1071         error = in_pcbconnect_setup(inp, sin, &laddr, &lport, &faddr, &fport,
1072             cred);
1073         if (error)
1074                 return (error);
1075
1076         /* Do the initial binding of the local address if required. */
1077         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
1078                 KASSERT(rehash == true,
1079                     ("Rehashing required for unbound inps"));
1080                 inp->inp_lport = lport;
1081                 inp->inp_laddr.s_addr = laddr;
1082                 if (in_pcbinshash(inp) != 0) {
1083                         inp->inp_laddr.s_addr = INADDR_ANY;
1084                         inp->inp_lport = 0;
1085                         return (EAGAIN);
1086                 }
1087         }
1088
1089         /* Commit the remaining changes. */
1090         inp->inp_lport = lport;
1091         inp->inp_laddr.s_addr = laddr;
1092         inp->inp_faddr.s_addr = faddr;
1093         inp->inp_fport = fport;
1094         if (rehash) {
1095                 in_pcbrehash(inp);
1096         } else {
1097                 in_pcbinshash(inp);
1098         }
1099
1100         if (anonport)
1101                 inp->inp_flags |= INP_ANONPORT;
1102         return (0);
1103 }
1104
1105 /*
1106  * Do proper source address selection on an unbound socket in case
1107  * of connect. Take jails into account as well.
1108  */
1109 int
1110 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
1111     struct ucred *cred)
1112 {
1113         struct ifaddr *ifa;
1114         struct sockaddr *sa;
1115         struct sockaddr_in *sin, dst;
1116         struct nhop_object *nh;
1117         int error;
1118
1119         NET_EPOCH_ASSERT();
1120         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
1121
1122         /*
1123          * Bypass source address selection and use the primary jail IP
1124          * if requested.
1125          */
1126         if (!prison_saddrsel_ip4(cred, laddr))
1127                 return (0);
1128
1129         error = 0;
1130
1131         nh = NULL;
1132         bzero(&dst, sizeof(dst));
1133         sin = &dst;
1134         sin->sin_family = AF_INET;
1135         sin->sin_len = sizeof(struct sockaddr_in);
1136         sin->sin_addr.s_addr = faddr->s_addr;
1137
1138         /*
1139          * If route is known our src addr is taken from the i/f,
1140          * else punt.
1141          *
1142          * Find out route to destination.
1143          */
1144         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
1145                 nh = fib4_lookup(inp->inp_inc.inc_fibnum, *faddr,
1146                     0, NHR_NONE, 0);
1147
1148         /*
1149          * If we found a route, use the address corresponding to
1150          * the outgoing interface.
1151          *
1152          * Otherwise assume faddr is reachable on a directly connected
1153          * network and try to find a corresponding interface to take
1154          * the source address from.
1155          */
1156         if (nh == NULL || nh->nh_ifp == NULL) {
1157                 struct in_ifaddr *ia;
1158                 struct ifnet *ifp;
1159
1160                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
1161                                         inp->inp_socket->so_fibnum));
1162                 if (ia == NULL) {
1163                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
1164                                                 inp->inp_socket->so_fibnum));
1165                 }
1166                 if (ia == NULL) {
1167                         error = ENETUNREACH;
1168                         goto done;
1169                 }
1170
1171                 if (!prison_flag(cred, PR_IP4)) {
1172                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1173                         goto done;
1174                 }
1175
1176                 ifp = ia->ia_ifp;
1177                 ia = NULL;
1178                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1179                         sa = ifa->ifa_addr;
1180                         if (sa->sa_family != AF_INET)
1181                                 continue;
1182                         sin = (struct sockaddr_in *)sa;
1183                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1184                                 ia = (struct in_ifaddr *)ifa;
1185                                 break;
1186                         }
1187                 }
1188                 if (ia != NULL) {
1189                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1190                         goto done;
1191                 }
1192
1193                 /* 3. As a last resort return the 'default' jail address. */
1194                 error = prison_get_ip4(cred, laddr);
1195                 goto done;
1196         }
1197
1198         /*
1199          * If the outgoing interface on the route found is not
1200          * a loopback interface, use the address from that interface.
1201          * In case of jails do those three steps:
1202          * 1. check if the interface address belongs to the jail. If so use it.
1203          * 2. check if we have any address on the outgoing interface
1204          *    belonging to this jail. If so use it.
1205          * 3. as a last resort return the 'default' jail address.
1206          */
1207         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) == 0) {
1208                 struct in_ifaddr *ia;
1209                 struct ifnet *ifp;
1210
1211                 /* If not jailed, use the default returned. */
1212                 if (!prison_flag(cred, PR_IP4)) {
1213                         ia = (struct in_ifaddr *)nh->nh_ifa;
1214                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1215                         goto done;
1216                 }
1217
1218                 /* Jailed. */
1219                 /* 1. Check if the iface address belongs to the jail. */
1220                 sin = (struct sockaddr_in *)nh->nh_ifa->ifa_addr;
1221                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1222                         ia = (struct in_ifaddr *)nh->nh_ifa;
1223                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1224                         goto done;
1225                 }
1226
1227                 /*
1228                  * 2. Check if we have any address on the outgoing interface
1229                  *    belonging to this jail.
1230                  */
1231                 ia = NULL;
1232                 ifp = nh->nh_ifp;
1233                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1234                         sa = ifa->ifa_addr;
1235                         if (sa->sa_family != AF_INET)
1236                                 continue;
1237                         sin = (struct sockaddr_in *)sa;
1238                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
1239                                 ia = (struct in_ifaddr *)ifa;
1240                                 break;
1241                         }
1242                 }
1243                 if (ia != NULL) {
1244                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1245                         goto done;
1246                 }
1247
1248                 /* 3. As a last resort return the 'default' jail address. */
1249                 error = prison_get_ip4(cred, laddr);
1250                 goto done;
1251         }
1252
1253         /*
1254          * The outgoing interface is marked with 'loopback net', so a route
1255          * to ourselves is here.
1256          * Try to find the interface of the destination address and then
1257          * take the address from there. That interface is not necessarily
1258          * a loopback interface.
1259          * In case of jails, check that it is an address of the jail
1260          * and if we cannot find, fall back to the 'default' jail address.
1261          */
1262         if ((nh->nh_ifp->if_flags & IFF_LOOPBACK) != 0) {
1263                 struct in_ifaddr *ia;
1264
1265                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&dst),
1266                                         inp->inp_socket->so_fibnum));
1267                 if (ia == NULL)
1268                         ia = ifatoia(ifa_ifwithnet(sintosa(&dst), 0,
1269                                                 inp->inp_socket->so_fibnum));
1270                 if (ia == NULL)
1271                         ia = ifatoia(ifa_ifwithaddr(sintosa(&dst)));
1272
1273                 if (!prison_flag(cred, PR_IP4)) {
1274                         if (ia == NULL) {
1275                                 error = ENETUNREACH;
1276                                 goto done;
1277                         }
1278                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1279                         goto done;
1280                 }
1281
1282                 /* Jailed. */
1283                 if (ia != NULL) {
1284                         struct ifnet *ifp;
1285
1286                         ifp = ia->ia_ifp;
1287                         ia = NULL;
1288                         CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1289                                 sa = ifa->ifa_addr;
1290                                 if (sa->sa_family != AF_INET)
1291                                         continue;
1292                                 sin = (struct sockaddr_in *)sa;
1293                                 if (prison_check_ip4(cred,
1294                                     &sin->sin_addr) == 0) {
1295                                         ia = (struct in_ifaddr *)ifa;
1296                                         break;
1297                                 }
1298                         }
1299                         if (ia != NULL) {
1300                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
1301                                 goto done;
1302                         }
1303                 }
1304
1305                 /* 3. As a last resort return the 'default' jail address. */
1306                 error = prison_get_ip4(cred, laddr);
1307                 goto done;
1308         }
1309
1310 done:
1311         return (error);
1312 }
1313
1314 /*
1315  * Set up for a connect from a socket to the specified address.
1316  * On entry, *laddrp and *lportp should contain the current local
1317  * address and port for the PCB; these are updated to the values
1318  * that should be placed in inp_laddr and inp_lport to complete
1319  * the connect.
1320  *
1321  * On success, *faddrp and *fportp will be set to the remote address
1322  * and port. These are not updated in the error case.
1323  */
1324 int
1325 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr_in *sin,
1326     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1327     struct ucred *cred)
1328 {
1329         struct in_ifaddr *ia;
1330         struct in_addr laddr, faddr;
1331         u_short lport, fport;
1332         int error;
1333
1334         KASSERT(sin->sin_family == AF_INET,
1335             ("%s: invalid address family for %p", __func__, sin));
1336         KASSERT(sin->sin_len == sizeof(*sin),
1337             ("%s: invalid address length for %p", __func__, sin));
1338
1339         /*
1340          * Because a global state change doesn't actually occur here, a read
1341          * lock is sufficient.
1342          */
1343         NET_EPOCH_ASSERT();
1344         INP_LOCK_ASSERT(inp);
1345         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1346
1347         if (sin->sin_port == 0)
1348                 return (EADDRNOTAVAIL);
1349         laddr.s_addr = *laddrp;
1350         lport = *lportp;
1351         faddr = sin->sin_addr;
1352         fport = sin->sin_port;
1353 #ifdef ROUTE_MPATH
1354         if (CALC_FLOWID_OUTBOUND) {
1355                 uint32_t hash_val, hash_type;
1356
1357                 hash_val = fib4_calc_software_hash(laddr, faddr, 0, fport,
1358                     inp->inp_socket->so_proto->pr_protocol, &hash_type);
1359
1360                 inp->inp_flowid = hash_val;
1361                 inp->inp_flowtype = hash_type;
1362         }
1363 #endif
1364         if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1365                 /*
1366                  * If the destination address is INADDR_ANY,
1367                  * use the primary local address.
1368                  * If the supplied address is INADDR_BROADCAST,
1369                  * and the primary interface supports broadcast,
1370                  * choose the broadcast address for that interface.
1371                  */
1372                 if (faddr.s_addr == INADDR_ANY) {
1373                         faddr =
1374                             IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1375                         if ((error = prison_get_ip4(cred, &faddr)) != 0)
1376                                 return (error);
1377                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1378                         if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1379                             IFF_BROADCAST)
1380                                 faddr = satosin(&CK_STAILQ_FIRST(
1381                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1382                 }
1383         }
1384         if (laddr.s_addr == INADDR_ANY) {
1385                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1386                 /*
1387                  * If the destination address is multicast and an outgoing
1388                  * interface has been set as a multicast option, prefer the
1389                  * address of that interface as our source address.
1390                  */
1391                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1392                     inp->inp_moptions != NULL) {
1393                         struct ip_moptions *imo;
1394                         struct ifnet *ifp;
1395
1396                         imo = inp->inp_moptions;
1397                         if (imo->imo_multicast_ifp != NULL) {
1398                                 ifp = imo->imo_multicast_ifp;
1399                                 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1400                                         if (ia->ia_ifp == ifp &&
1401                                             prison_check_ip4(cred,
1402                                             &ia->ia_addr.sin_addr) == 0)
1403                                                 break;
1404                                 }
1405                                 if (ia == NULL)
1406                                         error = EADDRNOTAVAIL;
1407                                 else {
1408                                         laddr = ia->ia_addr.sin_addr;
1409                                         error = 0;
1410                                 }
1411                         }
1412                 }
1413                 if (error)
1414                         return (error);
1415         }
1416
1417         if (lport != 0) {
1418                 if (in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr,
1419                     fport, laddr, lport, 0, NULL, M_NODOM) != NULL)
1420                         return (EADDRINUSE);
1421         } else {
1422                 struct sockaddr_in lsin, fsin;
1423
1424                 bzero(&lsin, sizeof(lsin));
1425                 bzero(&fsin, sizeof(fsin));
1426                 lsin.sin_family = AF_INET;
1427                 lsin.sin_addr = laddr;
1428                 fsin.sin_family = AF_INET;
1429                 fsin.sin_addr = faddr;
1430                 error = in_pcb_lport_dest(inp, (struct sockaddr *) &lsin,
1431                     &lport, (struct sockaddr *)& fsin, fport, cred,
1432                     INPLOOKUP_WILDCARD);
1433                 if (error)
1434                         return (error);
1435         }
1436         *laddrp = laddr.s_addr;
1437         *lportp = lport;
1438         *faddrp = faddr.s_addr;
1439         *fportp = fport;
1440         return (0);
1441 }
1442
1443 void
1444 in_pcbdisconnect(struct inpcb *inp)
1445 {
1446
1447         INP_WLOCK_ASSERT(inp);
1448         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1449
1450         inp->inp_laddr.s_addr = INADDR_ANY;
1451         inp->inp_faddr.s_addr = INADDR_ANY;
1452         inp->inp_fport = 0;
1453         in_pcbrehash(inp);
1454 }
1455 #endif /* INET */
1456
1457 /*
1458  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1459  * For most protocols, this will be invoked immediately prior to calling
1460  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1461  * socket, in which case in_pcbfree() is deferred.
1462  */
1463 void
1464 in_pcbdetach(struct inpcb *inp)
1465 {
1466
1467         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1468
1469 #ifdef RATELIMIT
1470         if (inp->inp_snd_tag != NULL)
1471                 in_pcbdetach_txrtlmt(inp);
1472 #endif
1473         inp->inp_socket->so_pcb = NULL;
1474         inp->inp_socket = NULL;
1475 }
1476
1477 /*
1478  * inpcb hash lookups are protected by SMR section.
1479  *
1480  * Once desired pcb has been found, switching from SMR section to a pcb
1481  * lock is performed with inp_smr_lock(). We can not use INP_(W|R)LOCK
1482  * here because SMR is a critical section.
1483  * In 99%+ cases inp_smr_lock() would obtain the lock immediately.
1484  */
1485 static inline void
1486 inp_lock(struct inpcb *inp, const inp_lookup_t lock)
1487 {
1488
1489         lock == INPLOOKUP_RLOCKPCB ?
1490             rw_rlock(&inp->inp_lock) : rw_wlock(&inp->inp_lock);
1491 }
1492
1493 static inline void
1494 inp_unlock(struct inpcb *inp, const inp_lookup_t lock)
1495 {
1496
1497         lock == INPLOOKUP_RLOCKPCB ?
1498             rw_runlock(&inp->inp_lock) : rw_wunlock(&inp->inp_lock);
1499 }
1500
1501 static inline int
1502 inp_trylock(struct inpcb *inp, const inp_lookup_t lock)
1503 {
1504
1505         return (lock == INPLOOKUP_RLOCKPCB ?
1506             rw_try_rlock(&inp->inp_lock) : rw_try_wlock(&inp->inp_lock));
1507 }
1508
1509 static inline bool
1510 in_pcbrele(struct inpcb *inp, const inp_lookup_t lock)
1511 {
1512
1513         return (lock == INPLOOKUP_RLOCKPCB ?
1514             in_pcbrele_rlocked(inp) : in_pcbrele_wlocked(inp));
1515 }
1516
1517 static inline bool
1518 _inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock, const int ignflags)
1519 {
1520
1521         MPASS(lock == INPLOOKUP_RLOCKPCB || lock == INPLOOKUP_WLOCKPCB);
1522         SMR_ASSERT_ENTERED(inp->inp_pcbinfo->ipi_smr);
1523
1524         if (__predict_true(inp_trylock(inp, lock))) {
1525                 if (__predict_false(inp->inp_flags & ignflags)) {
1526                         smr_exit(inp->inp_pcbinfo->ipi_smr);
1527                         inp_unlock(inp, lock);
1528                         return (false);
1529                 }
1530                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1531                 return (true);
1532         }
1533
1534         if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1535                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1536                 inp_lock(inp, lock);
1537                 if (__predict_false(in_pcbrele(inp, lock)))
1538                         return (false);
1539                 /*
1540                  * inp acquired through refcount & lock for sure didn't went
1541                  * through uma_zfree().  However, it may have already went
1542                  * through in_pcbfree() and has another reference, that
1543                  * prevented its release by our in_pcbrele().
1544                  */
1545                 if (__predict_false(inp->inp_flags & ignflags)) {
1546                         inp_unlock(inp, lock);
1547                         return (false);
1548                 }
1549                 return (true);
1550         } else {
1551                 smr_exit(inp->inp_pcbinfo->ipi_smr);
1552                 return (false);
1553         }
1554 }
1555
1556 bool
1557 inp_smr_lock(struct inpcb *inp, const inp_lookup_t lock)
1558 {
1559
1560         /*
1561          * in_pcblookup() family of functions ignore not only freed entries,
1562          * that may be found due to lockless access to the hash, but dropped
1563          * entries, too.
1564          */
1565         return (_inp_smr_lock(inp, lock, INP_FREED | INP_DROPPED));
1566 }
1567
1568 /*
1569  * inp_next() - inpcb hash/list traversal iterator
1570  *
1571  * Requires initialized struct inpcb_iterator for context.
1572  * The structure can be initialized with INP_ITERATOR() or INP_ALL_ITERATOR().
1573  *
1574  * - Iterator can have either write-lock or read-lock semantics, that can not
1575  *   be changed later.
1576  * - Iterator can iterate either over all pcbs list (INP_ALL_LIST), or through
1577  *   a single hash slot.  Note: only rip_input() does the latter.
1578  * - Iterator may have optional bool matching function.  The matching function
1579  *   will be executed for each inpcb in the SMR context, so it can not acquire
1580  *   locks and can safely access only immutable fields of inpcb.
1581  *
1582  * A fresh initialized iterator has NULL inpcb in its context and that
1583  * means that inp_next() call would return the very first inpcb on the list
1584  * locked with desired semantic.  In all following calls the context pointer
1585  * shall hold the current inpcb pointer.  The KPI user is not supposed to
1586  * unlock the current inpcb!  Upon end of traversal inp_next() will return NULL
1587  * and write NULL to its context.  After end of traversal an iterator can be
1588  * reused.
1589  *
1590  * List traversals have the following features/constraints:
1591  * - New entries won't be seen, as they are always added to the head of a list.
1592  * - Removed entries won't stop traversal as long as they are not added to
1593  *   a different list. This is violated by in_pcbrehash().
1594  */
1595 #define II_LIST_FIRST(ipi, hash)                                        \
1596                 (((hash) == INP_ALL_LIST) ?                             \
1597                     CK_LIST_FIRST(&(ipi)->ipi_listhead) :               \
1598                     CK_LIST_FIRST(&(ipi)->ipi_hashbase[(hash)]))
1599 #define II_LIST_NEXT(inp, hash)                                         \
1600                 (((hash) == INP_ALL_LIST) ?                             \
1601                     CK_LIST_NEXT((inp), inp_list) :                     \
1602                     CK_LIST_NEXT((inp), inp_hash))
1603 #define II_LOCK_ASSERT(inp, lock)                                       \
1604                 rw_assert(&(inp)->inp_lock,                             \
1605                     (lock) == INPLOOKUP_RLOCKPCB ?  RA_RLOCKED : RA_WLOCKED )
1606 struct inpcb *
1607 inp_next(struct inpcb_iterator *ii)
1608 {
1609         const struct inpcbinfo *ipi = ii->ipi;
1610         inp_match_t *match = ii->match;
1611         void *ctx = ii->ctx;
1612         inp_lookup_t lock = ii->lock;
1613         int hash = ii->hash;
1614         struct inpcb *inp;
1615
1616         if (ii->inp == NULL) {          /* First call. */
1617                 smr_enter(ipi->ipi_smr);
1618                 /* This is unrolled CK_LIST_FOREACH(). */
1619                 for (inp = II_LIST_FIRST(ipi, hash);
1620                     inp != NULL;
1621                     inp = II_LIST_NEXT(inp, hash)) {
1622                         if (match != NULL && (match)(inp, ctx) == false)
1623                                 continue;
1624                         if (__predict_true(_inp_smr_lock(inp, lock, INP_FREED)))
1625                                 break;
1626                         else {
1627                                 smr_enter(ipi->ipi_smr);
1628                                 MPASS(inp != II_LIST_FIRST(ipi, hash));
1629                                 inp = II_LIST_FIRST(ipi, hash);
1630                                 if (inp == NULL)
1631                                         break;
1632                         }
1633                 }
1634
1635                 if (inp == NULL)
1636                         smr_exit(ipi->ipi_smr);
1637                 else
1638                         ii->inp = inp;
1639
1640                 return (inp);
1641         }
1642
1643         /* Not a first call. */
1644         smr_enter(ipi->ipi_smr);
1645 restart:
1646         inp = ii->inp;
1647         II_LOCK_ASSERT(inp, lock);
1648 next:
1649         inp = II_LIST_NEXT(inp, hash);
1650         if (inp == NULL) {
1651                 smr_exit(ipi->ipi_smr);
1652                 goto found;
1653         }
1654
1655         if (match != NULL && (match)(inp, ctx) == false)
1656                 goto next;
1657
1658         if (__predict_true(inp_trylock(inp, lock))) {
1659                 if (__predict_false(inp->inp_flags & INP_FREED)) {
1660                         /*
1661                          * Entries are never inserted in middle of a list, thus
1662                          * as long as we are in SMR, we can continue traversal.
1663                          * Jump to 'restart' should yield in the same result,
1664                          * but could produce unnecessary looping.  Could this
1665                          * looping be unbound?
1666                          */
1667                         inp_unlock(inp, lock);
1668                         goto next;
1669                 } else {
1670                         smr_exit(ipi->ipi_smr);
1671                         goto found;
1672                 }
1673         }
1674
1675         /*
1676          * Can't obtain lock immediately, thus going hard.  Once we exit the
1677          * SMR section we can no longer jump to 'next', and our only stable
1678          * anchoring point is ii->inp, which we keep locked for this case, so
1679          * we jump to 'restart'.
1680          */
1681         if (__predict_true(refcount_acquire_if_not_zero(&inp->inp_refcount))) {
1682                 smr_exit(ipi->ipi_smr);
1683                 inp_lock(inp, lock);
1684                 if (__predict_false(in_pcbrele(inp, lock))) {
1685                         smr_enter(ipi->ipi_smr);
1686                         goto restart;
1687                 }
1688                 /*
1689                  * See comment in inp_smr_lock().
1690                  */
1691                 if (__predict_false(inp->inp_flags & INP_FREED)) {
1692                         inp_unlock(inp, lock);
1693                         smr_enter(ipi->ipi_smr);
1694                         goto restart;
1695                 }
1696         } else
1697                 goto next;
1698
1699 found:
1700         inp_unlock(ii->inp, lock);
1701         ii->inp = inp;
1702
1703         return (ii->inp);
1704 }
1705
1706 /*
1707  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1708  * stability of an inpcb pointer despite the inpcb lock being released or
1709  * SMR section exited.
1710  *
1711  * To free a reference later in_pcbrele_(r|w)locked() must be performed.
1712  */
1713 void
1714 in_pcbref(struct inpcb *inp)
1715 {
1716         u_int old __diagused;
1717
1718         old = refcount_acquire(&inp->inp_refcount);
1719         KASSERT(old > 0, ("%s: refcount 0", __func__));
1720 }
1721
1722 /*
1723  * Drop a refcount on an inpcb elevated using in_pcbref(), potentially
1724  * freeing the pcb, if the reference was very last.
1725  */
1726 bool
1727 in_pcbrele_rlocked(struct inpcb *inp)
1728 {
1729
1730         INP_RLOCK_ASSERT(inp);
1731
1732         if (refcount_release(&inp->inp_refcount) == 0)
1733                 return (false);
1734
1735         MPASS(inp->inp_flags & INP_FREED);
1736         MPASS(inp->inp_socket == NULL);
1737         MPASS(inp->inp_in_hpts == 0);
1738         INP_RUNLOCK(inp);
1739         uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1740         return (true);
1741 }
1742
1743 bool
1744 in_pcbrele_wlocked(struct inpcb *inp)
1745 {
1746
1747         INP_WLOCK_ASSERT(inp);
1748
1749         if (refcount_release(&inp->inp_refcount) == 0)
1750                 return (false);
1751
1752         MPASS(inp->inp_flags & INP_FREED);
1753         MPASS(inp->inp_socket == NULL);
1754         MPASS(inp->inp_in_hpts == 0);
1755         INP_WUNLOCK(inp);
1756         uma_zfree_smr(inp->inp_pcbinfo->ipi_zone, inp);
1757         return (true);
1758 }
1759
1760 /*
1761  * Unconditionally schedule an inpcb to be freed by decrementing its
1762  * reference count, which should occur only after the inpcb has been detached
1763  * from its socket.  If another thread holds a temporary reference (acquired
1764  * using in_pcbref()) then the free is deferred until that reference is
1765  * released using in_pcbrele_(r|w)locked(), but the inpcb is still unlocked.
1766  *  Almost all work, including removal from global lists, is done in this
1767  * context, where the pcbinfo lock is held.
1768  */
1769 void
1770 in_pcbfree(struct inpcb *inp)
1771 {
1772         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1773 #ifdef INET
1774         struct ip_moptions *imo;
1775 #endif
1776 #ifdef INET6
1777         struct ip6_moptions *im6o;
1778 #endif
1779
1780         INP_WLOCK_ASSERT(inp);
1781         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1782         KASSERT((inp->inp_flags & INP_FREED) == 0,
1783             ("%s: called twice for pcb %p", __func__, inp));
1784
1785         inp->inp_flags |= INP_FREED;
1786         INP_INFO_WLOCK(pcbinfo);
1787         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1788         pcbinfo->ipi_count--;
1789         CK_LIST_REMOVE(inp, inp_list);
1790         INP_INFO_WUNLOCK(pcbinfo);
1791
1792         if (inp->inp_flags & INP_INHASHLIST)
1793                 in_pcbremhash(inp);
1794
1795         RO_INVALIDATE_CACHE(&inp->inp_route);
1796 #ifdef MAC
1797         mac_inpcb_destroy(inp);
1798 #endif
1799 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1800         if (inp->inp_sp != NULL)
1801                 ipsec_delete_pcbpolicy(inp);
1802 #endif
1803 #ifdef INET
1804         if (inp->inp_options)
1805                 (void)m_free(inp->inp_options);
1806         imo = inp->inp_moptions;
1807 #endif
1808 #ifdef INET6
1809         if (inp->inp_vflag & INP_IPV6PROTO) {
1810                 ip6_freepcbopts(inp->in6p_outputopts);
1811                 im6o = inp->in6p_moptions;
1812         } else
1813                 im6o = NULL;
1814 #endif
1815
1816         if (__predict_false(in_pcbrele_wlocked(inp) == false)) {
1817                 INP_WUNLOCK(inp);
1818         }
1819 #ifdef INET6
1820         ip6_freemoptions(im6o);
1821 #endif
1822 #ifdef INET
1823         inp_freemoptions(imo);
1824 #endif
1825         /* Destruction is finalized in inpcb_dtor(). */
1826 }
1827
1828 static void
1829 inpcb_dtor(void *mem, int size, void *arg)
1830 {
1831         struct inpcb *inp = mem;
1832
1833         crfree(inp->inp_cred);
1834 #ifdef INVARIANTS
1835         inp->inp_cred = NULL;
1836 #endif
1837 }
1838
1839 /*
1840  * Different protocols initialize their inpcbs differently - giving
1841  * different name to the lock.  But they all are disposed the same.
1842  */
1843 static void
1844 inpcb_fini(void *mem, int size)
1845 {
1846         struct inpcb *inp = mem;
1847
1848         INP_LOCK_DESTROY(inp);
1849 }
1850
1851 /*
1852  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1853  * port reservation, and preventing it from being returned by inpcb lookups.
1854  *
1855  * It is used by TCP to mark an inpcb as unused and avoid future packet
1856  * delivery or event notification when a socket remains open but TCP has
1857  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1858  * or a RST on the wire, and allows the port binding to be reused while still
1859  * maintaining the invariant that so_pcb always points to a valid inpcb until
1860  * in_pcbdetach().
1861  *
1862  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1863  * in_pcbnotifyall() and in_pcbpurgeif0()?
1864  */
1865 void
1866 in_pcbdrop(struct inpcb *inp)
1867 {
1868
1869         INP_WLOCK_ASSERT(inp);
1870 #ifdef INVARIANTS
1871         if (inp->inp_socket != NULL && inp->inp_ppcb != NULL)
1872                 MPASS(inp->inp_refcount > 1);
1873 #endif
1874
1875         inp->inp_flags |= INP_DROPPED;
1876         if (inp->inp_flags & INP_INHASHLIST)
1877                 in_pcbremhash(inp);
1878 }
1879
1880 #ifdef INET
1881 /*
1882  * Common routines to return the socket addresses associated with inpcbs.
1883  */
1884 struct sockaddr *
1885 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1886 {
1887         struct sockaddr_in *sin;
1888
1889         sin = malloc(sizeof *sin, M_SONAME,
1890                 M_WAITOK | M_ZERO);
1891         sin->sin_family = AF_INET;
1892         sin->sin_len = sizeof(*sin);
1893         sin->sin_addr = *addr_p;
1894         sin->sin_port = port;
1895
1896         return (struct sockaddr *)sin;
1897 }
1898
1899 int
1900 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1901 {
1902         struct inpcb *inp;
1903         struct in_addr addr;
1904         in_port_t port;
1905
1906         inp = sotoinpcb(so);
1907         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1908
1909         INP_RLOCK(inp);
1910         port = inp->inp_lport;
1911         addr = inp->inp_laddr;
1912         INP_RUNLOCK(inp);
1913
1914         *nam = in_sockaddr(port, &addr);
1915         return 0;
1916 }
1917
1918 int
1919 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1920 {
1921         struct inpcb *inp;
1922         struct in_addr addr;
1923         in_port_t port;
1924
1925         inp = sotoinpcb(so);
1926         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1927
1928         INP_RLOCK(inp);
1929         port = inp->inp_fport;
1930         addr = inp->inp_faddr;
1931         INP_RUNLOCK(inp);
1932
1933         *nam = in_sockaddr(port, &addr);
1934         return 0;
1935 }
1936
1937 void
1938 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1939     struct inpcb *(*notify)(struct inpcb *, int))
1940 {
1941         struct inpcb *inp, *inp_temp;
1942
1943         INP_INFO_WLOCK(pcbinfo);
1944         CK_LIST_FOREACH_SAFE(inp, &pcbinfo->ipi_listhead, inp_list, inp_temp) {
1945                 INP_WLOCK(inp);
1946 #ifdef INET6
1947                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1948                         INP_WUNLOCK(inp);
1949                         continue;
1950                 }
1951 #endif
1952                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1953                     inp->inp_socket == NULL) {
1954                         INP_WUNLOCK(inp);
1955                         continue;
1956                 }
1957                 if ((*notify)(inp, errno))
1958                         INP_WUNLOCK(inp);
1959         }
1960         INP_INFO_WUNLOCK(pcbinfo);
1961 }
1962
1963 static bool
1964 inp_v4_multi_match(const struct inpcb *inp, void *v __unused)
1965 {
1966
1967         if ((inp->inp_vflag & INP_IPV4) && inp->inp_moptions != NULL)
1968                 return (true);
1969         else
1970                 return (false);
1971 }
1972
1973 void
1974 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1975 {
1976         struct inpcb_iterator inpi = INP_ITERATOR(pcbinfo, INPLOOKUP_WLOCKPCB,
1977             inp_v4_multi_match, NULL);
1978         struct inpcb *inp;
1979         struct in_multi *inm;
1980         struct in_mfilter *imf;
1981         struct ip_moptions *imo;
1982
1983         IN_MULTI_LOCK_ASSERT();
1984
1985         while ((inp = inp_next(&inpi)) != NULL) {
1986                 INP_WLOCK_ASSERT(inp);
1987
1988                 imo = inp->inp_moptions;
1989                 /*
1990                  * Unselect the outgoing interface if it is being
1991                  * detached.
1992                  */
1993                 if (imo->imo_multicast_ifp == ifp)
1994                         imo->imo_multicast_ifp = NULL;
1995
1996                 /*
1997                  * Drop multicast group membership if we joined
1998                  * through the interface being detached.
1999                  *
2000                  * XXX This can all be deferred to an epoch_call
2001                  */
2002 restart:
2003                 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
2004                         if ((inm = imf->imf_inm) == NULL)
2005                                 continue;
2006                         if (inm->inm_ifp != ifp)
2007                                 continue;
2008                         ip_mfilter_remove(&imo->imo_head, imf);
2009                         in_leavegroup_locked(inm, NULL);
2010                         ip_mfilter_free(imf);
2011                         goto restart;
2012                 }
2013         }
2014 }
2015
2016 /*
2017  * Lookup a PCB based on the local address and port.  Caller must hold the
2018  * hash lock.  No inpcb locks or references are acquired.
2019  */
2020 #define INP_LOOKUP_MAPPED_PCB_COST      3
2021 struct inpcb *
2022 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
2023     u_short lport, int lookupflags, struct ucred *cred)
2024 {
2025         struct inpcb *inp;
2026 #ifdef INET6
2027         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
2028 #else
2029         int matchwild = 3;
2030 #endif
2031         int wildcard;
2032
2033         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2034             ("%s: invalid lookup flags %d", __func__, lookupflags));
2035         INP_HASH_LOCK_ASSERT(pcbinfo);
2036
2037         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
2038                 struct inpcbhead *head;
2039                 /*
2040                  * Look for an unconnected (wildcard foreign addr) PCB that
2041                  * matches the local address and port we're looking for.
2042                  */
2043                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2044                     pcbinfo->ipi_hashmask)];
2045                 CK_LIST_FOREACH(inp, head, inp_hash) {
2046 #ifdef INET6
2047                         /* XXX inp locking */
2048                         if ((inp->inp_vflag & INP_IPV4) == 0)
2049                                 continue;
2050 #endif
2051                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
2052                             inp->inp_laddr.s_addr == laddr.s_addr &&
2053                             inp->inp_lport == lport) {
2054                                 /*
2055                                  * Found?
2056                                  */
2057                                 if (prison_equal_ip4(cred->cr_prison,
2058                                     inp->inp_cred->cr_prison))
2059                                         return (inp);
2060                         }
2061                 }
2062                 /*
2063                  * Not found.
2064                  */
2065                 return (NULL);
2066         } else {
2067                 struct inpcbporthead *porthash;
2068                 struct inpcbport *phd;
2069                 struct inpcb *match = NULL;
2070                 /*
2071                  * Best fit PCB lookup.
2072                  *
2073                  * First see if this local port is in use by looking on the
2074                  * port hash list.
2075                  */
2076                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
2077                     pcbinfo->ipi_porthashmask)];
2078                 CK_LIST_FOREACH(phd, porthash, phd_hash) {
2079                         if (phd->phd_port == lport)
2080                                 break;
2081                 }
2082                 if (phd != NULL) {
2083                         /*
2084                          * Port is in use by one or more PCBs. Look for best
2085                          * fit.
2086                          */
2087                         CK_LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
2088                                 wildcard = 0;
2089                                 if (!prison_equal_ip4(inp->inp_cred->cr_prison,
2090                                     cred->cr_prison))
2091                                         continue;
2092 #ifdef INET6
2093                                 /* XXX inp locking */
2094                                 if ((inp->inp_vflag & INP_IPV4) == 0)
2095                                         continue;
2096                                 /*
2097                                  * We never select the PCB that has
2098                                  * INP_IPV6 flag and is bound to :: if
2099                                  * we have another PCB which is bound
2100                                  * to 0.0.0.0.  If a PCB has the
2101                                  * INP_IPV6 flag, then we set its cost
2102                                  * higher than IPv4 only PCBs.
2103                                  *
2104                                  * Note that the case only happens
2105                                  * when a socket is bound to ::, under
2106                                  * the condition that the use of the
2107                                  * mapped address is allowed.
2108                                  */
2109                                 if ((inp->inp_vflag & INP_IPV6) != 0)
2110                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
2111 #endif
2112                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
2113                                         wildcard++;
2114                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
2115                                         if (laddr.s_addr == INADDR_ANY)
2116                                                 wildcard++;
2117                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
2118                                                 continue;
2119                                 } else {
2120                                         if (laddr.s_addr != INADDR_ANY)
2121                                                 wildcard++;
2122                                 }
2123                                 if (wildcard < matchwild) {
2124                                         match = inp;
2125                                         matchwild = wildcard;
2126                                         if (matchwild == 0)
2127                                                 break;
2128                                 }
2129                         }
2130                 }
2131                 return (match);
2132         }
2133 }
2134 #undef INP_LOOKUP_MAPPED_PCB_COST
2135
2136 static bool
2137 in_pcblookup_lb_numa_match(const struct inpcblbgroup *grp, int domain)
2138 {
2139         return (domain == M_NODOM || domain == grp->il_numa_domain);
2140 }
2141
2142 static struct inpcb *
2143 in_pcblookup_lbgroup(const struct inpcbinfo *pcbinfo,
2144     const struct in_addr *laddr, uint16_t lport, const struct in_addr *faddr,
2145     uint16_t fport, int lookupflags, int domain)
2146 {
2147         const struct inpcblbgrouphead *hdr;
2148         struct inpcblbgroup *grp;
2149         struct inpcblbgroup *jail_exact, *jail_wild, *local_exact, *local_wild;
2150
2151         INP_HASH_LOCK_ASSERT(pcbinfo);
2152
2153         hdr = &pcbinfo->ipi_lbgrouphashbase[
2154             INP_PCBPORTHASH(lport, pcbinfo->ipi_lbgrouphashmask)];
2155
2156         /*
2157          * Search for an LB group match based on the following criteria:
2158          * - prefer jailed groups to non-jailed groups
2159          * - prefer exact source address matches to wildcard matches
2160          * - prefer groups bound to the specified NUMA domain
2161          */
2162         jail_exact = jail_wild = local_exact = local_wild = NULL;
2163         CK_LIST_FOREACH(grp, hdr, il_list) {
2164                 bool injail;
2165
2166 #ifdef INET6
2167                 if (!(grp->il_vflag & INP_IPV4))
2168                         continue;
2169 #endif
2170                 if (grp->il_lport != lport)
2171                         continue;
2172
2173                 injail = prison_flag(grp->il_cred, PR_IP4) != 0;
2174                 if (injail && prison_check_ip4_locked(grp->il_cred->cr_prison,
2175                     laddr) != 0)
2176                         continue;
2177
2178                 if (grp->il_laddr.s_addr == laddr->s_addr) {
2179                         if (injail) {
2180                                 jail_exact = grp;
2181                                 if (in_pcblookup_lb_numa_match(grp, domain))
2182                                         /* This is a perfect match. */
2183                                         goto out;
2184                         } else if (local_exact == NULL ||
2185                             in_pcblookup_lb_numa_match(grp, domain)) {
2186                                 local_exact = grp;
2187                         }
2188                 } else if (grp->il_laddr.s_addr == INADDR_ANY &&
2189                     (lookupflags & INPLOOKUP_WILDCARD) != 0) {
2190                         if (injail) {
2191                                 if (jail_wild == NULL ||
2192                                     in_pcblookup_lb_numa_match(grp, domain))
2193                                         jail_wild = grp;
2194                         } else if (local_wild == NULL ||
2195                             in_pcblookup_lb_numa_match(grp, domain)) {
2196                                 local_wild = grp;
2197                         }
2198                 }
2199         }
2200
2201         if (jail_exact != NULL)
2202                 grp = jail_exact;
2203         else if (jail_wild != NULL)
2204                 grp = jail_wild;
2205         else if (local_exact != NULL)
2206                 grp = local_exact;
2207         else
2208                 grp = local_wild;
2209         if (grp == NULL)
2210                 return (NULL);
2211 out:
2212         return (grp->il_inp[INP_PCBLBGROUP_PKTHASH(faddr, lport, fport) %
2213             grp->il_inpcnt]);
2214 }
2215
2216 /*
2217  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
2218  * that the caller has either locked the hash list, which usually happens
2219  * for bind(2) operations, or is in SMR section, which happens when sorting
2220  * out incoming packets.
2221  */
2222 static struct inpcb *
2223 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2224     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
2225     struct ifnet *ifp, uint8_t numa_domain)
2226 {
2227         struct inpcbhead *head;
2228         struct inpcb *inp;
2229         u_short fport = fport_arg, lport = lport_arg;
2230
2231         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
2232             ("%s: invalid lookup flags %d", __func__, lookupflags));
2233         KASSERT(faddr.s_addr != INADDR_ANY,
2234             ("%s: invalid foreign address", __func__));
2235         KASSERT(laddr.s_addr != INADDR_ANY,
2236             ("%s: invalid local address", __func__));
2237         INP_HASH_LOCK_ASSERT(pcbinfo);
2238
2239         /*
2240          * First look for an exact match.
2241          */
2242         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&faddr, lport, fport,
2243             pcbinfo->ipi_hashmask)];
2244         CK_LIST_FOREACH(inp, head, inp_hash) {
2245 #ifdef INET6
2246                 /* XXX inp locking */
2247                 if ((inp->inp_vflag & INP_IPV4) == 0)
2248                         continue;
2249 #endif
2250                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
2251                     inp->inp_laddr.s_addr == laddr.s_addr &&
2252                     inp->inp_fport == fport &&
2253                     inp->inp_lport == lport)
2254                         return (inp);
2255         }
2256
2257         /*
2258          * Then look for a wildcard match, if requested.
2259          */
2260         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
2261                 struct inpcb *local_wild = NULL, *local_exact = NULL;
2262 #ifdef INET6
2263                 struct inpcb *local_wild_mapped = NULL;
2264 #endif
2265                 struct inpcb *jail_wild = NULL;
2266                 int injail;
2267
2268                 /*
2269                  * First see if an LB group matches the request before scanning
2270                  * all sockets on this port.
2271                  */
2272                 inp = in_pcblookup_lbgroup(pcbinfo, &laddr, lport, &faddr,
2273                     fport, lookupflags, numa_domain);
2274                 if (inp != NULL)
2275                         return (inp);
2276
2277                 /*
2278                  * Order of socket selection - we always prefer jails.
2279                  *      1. jailed, non-wild.
2280                  *      2. jailed, wild.
2281                  *      3. non-jailed, non-wild.
2282                  *      4. non-jailed, wild.
2283                  */
2284
2285                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH_WILD(lport,
2286                     pcbinfo->ipi_hashmask)];
2287                 CK_LIST_FOREACH(inp, head, inp_hash) {
2288 #ifdef INET6
2289                         /* XXX inp locking */
2290                         if ((inp->inp_vflag & INP_IPV4) == 0)
2291                                 continue;
2292 #endif
2293                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2294                             inp->inp_lport != lport)
2295                                 continue;
2296
2297                         injail = prison_flag(inp->inp_cred, PR_IP4);
2298                         if (injail) {
2299                                 if (prison_check_ip4_locked(
2300                                     inp->inp_cred->cr_prison, &laddr) != 0)
2301                                         continue;
2302                         } else {
2303                                 if (local_exact != NULL)
2304                                         continue;
2305                         }
2306
2307                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2308                                 if (injail)
2309                                         return (inp);
2310                                 else
2311                                         local_exact = inp;
2312                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2313 #ifdef INET6
2314                                 /* XXX inp locking, NULL check */
2315                                 if (inp->inp_vflag & INP_IPV6PROTO)
2316                                         local_wild_mapped = inp;
2317                                 else
2318 #endif
2319                                         if (injail)
2320                                                 jail_wild = inp;
2321                                         else
2322                                                 local_wild = inp;
2323                         }
2324                 } /* LIST_FOREACH */
2325                 if (jail_wild != NULL)
2326                         return (jail_wild);
2327                 if (local_exact != NULL)
2328                         return (local_exact);
2329                 if (local_wild != NULL)
2330                         return (local_wild);
2331 #ifdef INET6
2332                 if (local_wild_mapped != NULL)
2333                         return (local_wild_mapped);
2334 #endif
2335         } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2336
2337         return (NULL);
2338 }
2339
2340 /*
2341  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2342  * hash list lock, and will return the inpcb locked (i.e., requires
2343  * INPLOOKUP_LOCKPCB).
2344  */
2345 static struct inpcb *
2346 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2347     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2348     struct ifnet *ifp, uint8_t numa_domain)
2349 {
2350         struct inpcb *inp;
2351
2352         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2353             ("%s: invalid lookup flags %d", __func__, lookupflags));
2354         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2355             ("%s: LOCKPCB not set", __func__));
2356
2357         smr_enter(pcbinfo->ipi_smr);
2358         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2359             lookupflags & INPLOOKUP_WILDCARD, ifp, numa_domain);
2360         if (inp != NULL) {
2361                 if (__predict_false(inp_smr_lock(inp,
2362                     (lookupflags & INPLOOKUP_LOCKMASK)) == false))
2363                         inp = NULL;
2364         } else
2365                 smr_exit(pcbinfo->ipi_smr);
2366
2367         return (inp);
2368 }
2369
2370 /*
2371  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2372  * from which a pre-calculated hash value may be extracted.
2373  */
2374 struct inpcb *
2375 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2376     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2377 {
2378         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2379             lookupflags, ifp, M_NODOM));
2380 }
2381
2382 struct inpcb *
2383 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2384     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2385     struct ifnet *ifp, struct mbuf *m)
2386 {
2387         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2388             lookupflags, ifp, m->m_pkthdr.numa_domain));
2389 }
2390 #endif /* INET */
2391
2392 /*
2393  * Insert PCB onto various hash lists.
2394  */
2395 int
2396 in_pcbinshash(struct inpcb *inp)
2397 {
2398         struct inpcbhead *pcbhash;
2399         struct inpcbporthead *pcbporthash;
2400         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2401         struct inpcbport *phd;
2402
2403         INP_WLOCK_ASSERT(inp);
2404         INP_HASH_WLOCK_ASSERT(pcbinfo);
2405
2406         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2407             ("in_pcbinshash: INP_INHASHLIST"));
2408
2409 #ifdef INET6
2410         if (inp->inp_vflag & INP_IPV6)
2411                 pcbhash = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2412                     inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2413         else
2414 #endif
2415                 pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2416                     inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2417
2418         pcbporthash = &pcbinfo->ipi_porthashbase[
2419             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2420
2421         /*
2422          * Add entry to load balance group.
2423          * Only do this if SO_REUSEPORT_LB is set.
2424          */
2425         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0) {
2426                 int error = in_pcbinslbgrouphash(inp, M_NODOM);
2427                 if (error != 0)
2428                         return (error);
2429         }
2430
2431         /*
2432          * Go through port list and look for a head for this lport.
2433          */
2434         CK_LIST_FOREACH(phd, pcbporthash, phd_hash) {
2435                 if (phd->phd_port == inp->inp_lport)
2436                         break;
2437         }
2438
2439         /*
2440          * If none exists, malloc one and tack it on.
2441          */
2442         if (phd == NULL) {
2443                 phd = uma_zalloc_smr(pcbinfo->ipi_portzone, M_NOWAIT);
2444                 if (phd == NULL) {
2445                         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2446                                 in_pcbremlbgrouphash(inp);
2447                         return (ENOMEM);
2448                 }
2449                 phd->phd_port = inp->inp_lport;
2450                 CK_LIST_INIT(&phd->phd_pcblist);
2451                 CK_LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2452         }
2453         inp->inp_phd = phd;
2454         CK_LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2455         CK_LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2456         inp->inp_flags |= INP_INHASHLIST;
2457
2458         return (0);
2459 }
2460
2461 static void
2462 in_pcbremhash(struct inpcb *inp)
2463 {
2464         struct inpcbport *phd = inp->inp_phd;
2465
2466         INP_WLOCK_ASSERT(inp);
2467         MPASS(inp->inp_flags & INP_INHASHLIST);
2468
2469         INP_HASH_WLOCK(inp->inp_pcbinfo);
2470         if ((inp->inp_flags2 & INP_REUSEPORT_LB) != 0)
2471                 in_pcbremlbgrouphash(inp);
2472         CK_LIST_REMOVE(inp, inp_hash);
2473         CK_LIST_REMOVE(inp, inp_portlist);
2474         if (CK_LIST_FIRST(&phd->phd_pcblist) == NULL) {
2475                 CK_LIST_REMOVE(phd, phd_hash);
2476                 uma_zfree_smr(inp->inp_pcbinfo->ipi_portzone, phd);
2477         }
2478         INP_HASH_WUNLOCK(inp->inp_pcbinfo);
2479         inp->inp_flags &= ~INP_INHASHLIST;
2480 }
2481
2482 /*
2483  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2484  * changed. NOTE: This does not handle the case of the lport changing (the
2485  * hashed port list would have to be updated as well), so the lport must
2486  * not change after in_pcbinshash() has been called.
2487  *
2488  * XXXGL: a race between this function and SMR-protected hash iterator
2489  * will lead to iterator traversing a possibly wrong hash list. However,
2490  * this race should have been here since change from rwlock to epoch.
2491  */
2492 void
2493 in_pcbrehash(struct inpcb *inp)
2494 {
2495         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2496         struct inpcbhead *head;
2497
2498         INP_WLOCK_ASSERT(inp);
2499         INP_HASH_WLOCK_ASSERT(pcbinfo);
2500
2501         KASSERT(inp->inp_flags & INP_INHASHLIST,
2502             ("in_pcbrehash: !INP_INHASHLIST"));
2503
2504 #ifdef INET6
2505         if (inp->inp_vflag & INP_IPV6)
2506                 head = &pcbinfo->ipi_hashbase[INP6_PCBHASH(&inp->in6p_faddr,
2507                     inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2508         else
2509 #endif
2510                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(&inp->inp_faddr,
2511                     inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2512
2513         CK_LIST_REMOVE(inp, inp_hash);
2514         CK_LIST_INSERT_HEAD(head, inp, inp_hash);
2515 }
2516
2517 /*
2518  * Check for alternatives when higher level complains
2519  * about service problems.  For now, invalidate cached
2520  * routing information.  If the route was created dynamically
2521  * (by a redirect), time to try a default gateway again.
2522  */
2523 void
2524 in_losing(struct inpcb *inp)
2525 {
2526
2527         RO_INVALIDATE_CACHE(&inp->inp_route);
2528         return;
2529 }
2530
2531 /*
2532  * A set label operation has occurred at the socket layer, propagate the
2533  * label change into the in_pcb for the socket.
2534  */
2535 void
2536 in_pcbsosetlabel(struct socket *so)
2537 {
2538 #ifdef MAC
2539         struct inpcb *inp;
2540
2541         inp = sotoinpcb(so);
2542         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2543
2544         INP_WLOCK(inp);
2545         SOCK_LOCK(so);
2546         mac_inpcb_sosetlabel(so, inp);
2547         SOCK_UNLOCK(so);
2548         INP_WUNLOCK(inp);
2549 #endif
2550 }
2551
2552 void
2553 inp_wlock(struct inpcb *inp)
2554 {
2555
2556         INP_WLOCK(inp);
2557 }
2558
2559 void
2560 inp_wunlock(struct inpcb *inp)
2561 {
2562
2563         INP_WUNLOCK(inp);
2564 }
2565
2566 void
2567 inp_rlock(struct inpcb *inp)
2568 {
2569
2570         INP_RLOCK(inp);
2571 }
2572
2573 void
2574 inp_runlock(struct inpcb *inp)
2575 {
2576
2577         INP_RUNLOCK(inp);
2578 }
2579
2580 #ifdef INVARIANT_SUPPORT
2581 void
2582 inp_lock_assert(struct inpcb *inp)
2583 {
2584
2585         INP_WLOCK_ASSERT(inp);
2586 }
2587
2588 void
2589 inp_unlock_assert(struct inpcb *inp)
2590 {
2591
2592         INP_UNLOCK_ASSERT(inp);
2593 }
2594 #endif
2595
2596 void
2597 inp_apply_all(struct inpcbinfo *pcbinfo,
2598     void (*func)(struct inpcb *, void *), void *arg)
2599 {
2600         struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2601             INPLOOKUP_WLOCKPCB);
2602         struct inpcb *inp;
2603
2604         while ((inp = inp_next(&inpi)) != NULL)
2605                 func(inp, arg);
2606 }
2607
2608 struct socket *
2609 inp_inpcbtosocket(struct inpcb *inp)
2610 {
2611
2612         INP_WLOCK_ASSERT(inp);
2613         return (inp->inp_socket);
2614 }
2615
2616 struct tcpcb *
2617 inp_inpcbtotcpcb(struct inpcb *inp)
2618 {
2619
2620         INP_WLOCK_ASSERT(inp);
2621         return ((struct tcpcb *)inp->inp_ppcb);
2622 }
2623
2624 int
2625 inp_ip_tos_get(const struct inpcb *inp)
2626 {
2627
2628         return (inp->inp_ip_tos);
2629 }
2630
2631 void
2632 inp_ip_tos_set(struct inpcb *inp, int val)
2633 {
2634
2635         inp->inp_ip_tos = val;
2636 }
2637
2638 void
2639 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2640     uint32_t *faddr, uint16_t *fp)
2641 {
2642
2643         INP_LOCK_ASSERT(inp);
2644         *laddr = inp->inp_laddr.s_addr;
2645         *faddr = inp->inp_faddr.s_addr;
2646         *lp = inp->inp_lport;
2647         *fp = inp->inp_fport;
2648 }
2649
2650 struct inpcb *
2651 so_sotoinpcb(struct socket *so)
2652 {
2653
2654         return (sotoinpcb(so));
2655 }
2656
2657 /*
2658  * Create an external-format (``xinpcb'') structure using the information in
2659  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2660  * reduce the spew of irrelevant information over this interface, to isolate
2661  * user code from changes in the kernel structure, and potentially to provide
2662  * information-hiding if we decide that some of this information should be
2663  * hidden from users.
2664  */
2665 void
2666 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2667 {
2668
2669         bzero(xi, sizeof(*xi));
2670         xi->xi_len = sizeof(struct xinpcb);
2671         if (inp->inp_socket)
2672                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2673         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2674         xi->inp_gencnt = inp->inp_gencnt;
2675         xi->inp_ppcb = (uintptr_t)inp->inp_ppcb;
2676         xi->inp_flow = inp->inp_flow;
2677         xi->inp_flowid = inp->inp_flowid;
2678         xi->inp_flowtype = inp->inp_flowtype;
2679         xi->inp_flags = inp->inp_flags;
2680         xi->inp_flags2 = inp->inp_flags2;
2681         xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2682         xi->in6p_cksum = inp->in6p_cksum;
2683         xi->in6p_hops = inp->in6p_hops;
2684         xi->inp_ip_tos = inp->inp_ip_tos;
2685         xi->inp_vflag = inp->inp_vflag;
2686         xi->inp_ip_ttl = inp->inp_ip_ttl;
2687         xi->inp_ip_p = inp->inp_ip_p;
2688         xi->inp_ip_minttl = inp->inp_ip_minttl;
2689 }
2690
2691 int
2692 sysctl_setsockopt(SYSCTL_HANDLER_ARGS, struct inpcbinfo *pcbinfo,
2693     int (*ctloutput_set)(struct inpcb *, struct sockopt *))
2694 {
2695         struct sockopt sopt;
2696         struct inpcb_iterator inpi = INP_ALL_ITERATOR(pcbinfo,
2697             INPLOOKUP_WLOCKPCB);
2698         struct inpcb *inp;
2699         struct sockopt_parameters *params;
2700         struct socket *so;
2701         int error;
2702         char buf[1024];
2703
2704         if (req->oldptr != NULL || req->oldlen != 0)
2705                 return (EINVAL);
2706         if (req->newptr == NULL)
2707                 return (EPERM);
2708         if (req->newlen > sizeof(buf))
2709                 return (ENOMEM);
2710         error = SYSCTL_IN(req, buf, req->newlen);
2711         if (error != 0)
2712                 return (error);
2713         if (req->newlen < sizeof(struct sockopt_parameters))
2714                 return (EINVAL);
2715         params = (struct sockopt_parameters *)buf;
2716         sopt.sopt_level = params->sop_level;
2717         sopt.sopt_name = params->sop_optname;
2718         sopt.sopt_dir = SOPT_SET;
2719         sopt.sopt_val = params->sop_optval;
2720         sopt.sopt_valsize = req->newlen - sizeof(struct sockopt_parameters);
2721         sopt.sopt_td = NULL;
2722 #ifdef INET6
2723         if (params->sop_inc.inc_flags & INC_ISIPV6) {
2724                 if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_laddr))
2725                         params->sop_inc.inc6_laddr.s6_addr16[1] =
2726                             htons(params->sop_inc.inc6_zoneid & 0xffff);
2727                 if (IN6_IS_SCOPE_LINKLOCAL(&params->sop_inc.inc6_faddr))
2728                         params->sop_inc.inc6_faddr.s6_addr16[1] =
2729                             htons(params->sop_inc.inc6_zoneid & 0xffff);
2730         }
2731 #endif
2732         if (params->sop_inc.inc_lport != htons(0)) {
2733                 if (params->sop_inc.inc_fport == htons(0))
2734                         inpi.hash = INP_PCBHASH_WILD(params->sop_inc.inc_lport,
2735                             pcbinfo->ipi_hashmask);
2736                 else
2737 #ifdef INET6
2738                         if (params->sop_inc.inc_flags & INC_ISIPV6)
2739                                 inpi.hash = INP6_PCBHASH(
2740                                     &params->sop_inc.inc6_faddr,
2741                                     params->sop_inc.inc_lport,
2742                                     params->sop_inc.inc_fport,
2743                                     pcbinfo->ipi_hashmask);
2744                         else
2745 #endif
2746                                 inpi.hash = INP_PCBHASH(
2747                                     &params->sop_inc.inc_faddr,
2748                                     params->sop_inc.inc_lport,
2749                                     params->sop_inc.inc_fport,
2750                                     pcbinfo->ipi_hashmask);
2751         }
2752         while ((inp = inp_next(&inpi)) != NULL)
2753                 if (inp->inp_gencnt == params->sop_id) {
2754                         if (inp->inp_flags & INP_DROPPED) {
2755                                 INP_WUNLOCK(inp);
2756                                 return (ECONNRESET);
2757                         }
2758                         so = inp->inp_socket;
2759                         KASSERT(so != NULL, ("inp_socket == NULL"));
2760                         soref(so);
2761                         error = (*ctloutput_set)(inp, &sopt);
2762                         sorele(so);
2763                         break;
2764                 }
2765         if (inp == NULL)
2766                 error = ESRCH;
2767         return (error);
2768 }
2769
2770 #ifdef DDB
2771 static void
2772 db_print_indent(int indent)
2773 {
2774         int i;
2775
2776         for (i = 0; i < indent; i++)
2777                 db_printf(" ");
2778 }
2779
2780 static void
2781 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2782 {
2783         char faddr_str[48], laddr_str[48];
2784
2785         db_print_indent(indent);
2786         db_printf("%s at %p\n", name, inc);
2787
2788         indent += 2;
2789
2790 #ifdef INET6
2791         if (inc->inc_flags & INC_ISIPV6) {
2792                 /* IPv6. */
2793                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2794                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2795         } else
2796 #endif
2797         {
2798                 /* IPv4. */
2799                 inet_ntoa_r(inc->inc_laddr, laddr_str);
2800                 inet_ntoa_r(inc->inc_faddr, faddr_str);
2801         }
2802         db_print_indent(indent);
2803         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2804             ntohs(inc->inc_lport));
2805         db_print_indent(indent);
2806         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2807             ntohs(inc->inc_fport));
2808 }
2809
2810 static void
2811 db_print_inpflags(int inp_flags)
2812 {
2813         int comma;
2814
2815         comma = 0;
2816         if (inp_flags & INP_RECVOPTS) {
2817                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2818                 comma = 1;
2819         }
2820         if (inp_flags & INP_RECVRETOPTS) {
2821                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2822                 comma = 1;
2823         }
2824         if (inp_flags & INP_RECVDSTADDR) {
2825                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2826                 comma = 1;
2827         }
2828         if (inp_flags & INP_ORIGDSTADDR) {
2829                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2830                 comma = 1;
2831         }
2832         if (inp_flags & INP_HDRINCL) {
2833                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2834                 comma = 1;
2835         }
2836         if (inp_flags & INP_HIGHPORT) {
2837                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2838                 comma = 1;
2839         }
2840         if (inp_flags & INP_LOWPORT) {
2841                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2842                 comma = 1;
2843         }
2844         if (inp_flags & INP_ANONPORT) {
2845                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2846                 comma = 1;
2847         }
2848         if (inp_flags & INP_RECVIF) {
2849                 db_printf("%sINP_RECVIF", comma ? ", " : "");
2850                 comma = 1;
2851         }
2852         if (inp_flags & INP_MTUDISC) {
2853                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2854                 comma = 1;
2855         }
2856         if (inp_flags & INP_RECVTTL) {
2857                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2858                 comma = 1;
2859         }
2860         if (inp_flags & INP_DONTFRAG) {
2861                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2862                 comma = 1;
2863         }
2864         if (inp_flags & INP_RECVTOS) {
2865                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2866                 comma = 1;
2867         }
2868         if (inp_flags & IN6P_IPV6_V6ONLY) {
2869                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2870                 comma = 1;
2871         }
2872         if (inp_flags & IN6P_PKTINFO) {
2873                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2874                 comma = 1;
2875         }
2876         if (inp_flags & IN6P_HOPLIMIT) {
2877                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2878                 comma = 1;
2879         }
2880         if (inp_flags & IN6P_HOPOPTS) {
2881                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2882                 comma = 1;
2883         }
2884         if (inp_flags & IN6P_DSTOPTS) {
2885                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2886                 comma = 1;
2887         }
2888         if (inp_flags & IN6P_RTHDR) {
2889                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2890                 comma = 1;
2891         }
2892         if (inp_flags & IN6P_RTHDRDSTOPTS) {
2893                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2894                 comma = 1;
2895         }
2896         if (inp_flags & IN6P_TCLASS) {
2897                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2898                 comma = 1;
2899         }
2900         if (inp_flags & IN6P_AUTOFLOWLABEL) {
2901                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2902                 comma = 1;
2903         }
2904         if (inp_flags & INP_ONESBCAST) {
2905                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2906                 comma  = 1;
2907         }
2908         if (inp_flags & INP_DROPPED) {
2909                 db_printf("%sINP_DROPPED", comma ? ", " : "");
2910                 comma  = 1;
2911         }
2912         if (inp_flags & INP_SOCKREF) {
2913                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2914                 comma  = 1;
2915         }
2916         if (inp_flags & IN6P_RFC2292) {
2917                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2918                 comma = 1;
2919         }
2920         if (inp_flags & IN6P_MTU) {
2921                 db_printf("IN6P_MTU%s", comma ? ", " : "");
2922                 comma = 1;
2923         }
2924 }
2925
2926 static void
2927 db_print_inpvflag(u_char inp_vflag)
2928 {
2929         int comma;
2930
2931         comma = 0;
2932         if (inp_vflag & INP_IPV4) {
2933                 db_printf("%sINP_IPV4", comma ? ", " : "");
2934                 comma  = 1;
2935         }
2936         if (inp_vflag & INP_IPV6) {
2937                 db_printf("%sINP_IPV6", comma ? ", " : "");
2938                 comma  = 1;
2939         }
2940         if (inp_vflag & INP_IPV6PROTO) {
2941                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2942                 comma  = 1;
2943         }
2944 }
2945
2946 static void
2947 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2948 {
2949
2950         db_print_indent(indent);
2951         db_printf("%s at %p\n", name, inp);
2952
2953         indent += 2;
2954
2955         db_print_indent(indent);
2956         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2957
2958         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2959
2960         db_print_indent(indent);
2961         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2962             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2963
2964         db_print_indent(indent);
2965         db_printf("inp_label: %p   inp_flags: 0x%x (",
2966            inp->inp_label, inp->inp_flags);
2967         db_print_inpflags(inp->inp_flags);
2968         db_printf(")\n");
2969
2970         db_print_indent(indent);
2971         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2972             inp->inp_vflag);
2973         db_print_inpvflag(inp->inp_vflag);
2974         db_printf(")\n");
2975
2976         db_print_indent(indent);
2977         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2978             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2979
2980         db_print_indent(indent);
2981 #ifdef INET6
2982         if (inp->inp_vflag & INP_IPV6) {
2983                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
2984                     "in6p_moptions: %p\n", inp->in6p_options,
2985                     inp->in6p_outputopts, inp->in6p_moptions);
2986                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2987                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2988                     inp->in6p_hops);
2989         } else
2990 #endif
2991         {
2992                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2993                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2994                     inp->inp_options, inp->inp_moptions);
2995         }
2996
2997         db_print_indent(indent);
2998         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2999             (uintmax_t)inp->inp_gencnt);
3000 }
3001
3002 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
3003 {
3004         struct inpcb *inp;
3005
3006         if (!have_addr) {
3007                 db_printf("usage: show inpcb <addr>\n");
3008                 return;
3009         }
3010         inp = (struct inpcb *)addr;
3011
3012         db_print_inpcb(inp, "inpcb", 0);
3013 }
3014 #endif /* DDB */
3015
3016 #ifdef RATELIMIT
3017 /*
3018  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
3019  * if any.
3020  */
3021 int
3022 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
3023 {
3024         union if_snd_tag_modify_params params = {
3025                 .rate_limit.max_rate = max_pacing_rate,
3026                 .rate_limit.flags = M_NOWAIT,
3027         };
3028         struct m_snd_tag *mst;
3029         int error;
3030
3031         mst = inp->inp_snd_tag;
3032         if (mst == NULL)
3033                 return (EINVAL);
3034
3035         if (mst->sw->snd_tag_modify == NULL) {
3036                 error = EOPNOTSUPP;
3037         } else {
3038                 error = mst->sw->snd_tag_modify(mst, &params);
3039         }
3040         return (error);
3041 }
3042
3043 /*
3044  * Query existing TX rate limit based on the existing
3045  * "inp->inp_snd_tag", if any.
3046  */
3047 int
3048 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
3049 {
3050         union if_snd_tag_query_params params = { };
3051         struct m_snd_tag *mst;
3052         int error;
3053
3054         mst = inp->inp_snd_tag;
3055         if (mst == NULL)
3056                 return (EINVAL);
3057
3058         if (mst->sw->snd_tag_query == NULL) {
3059                 error = EOPNOTSUPP;
3060         } else {
3061                 error = mst->sw->snd_tag_query(mst, &params);
3062                 if (error == 0 && p_max_pacing_rate != NULL)
3063                         *p_max_pacing_rate = params.rate_limit.max_rate;
3064         }
3065         return (error);
3066 }
3067
3068 /*
3069  * Query existing TX queue level based on the existing
3070  * "inp->inp_snd_tag", if any.
3071  */
3072 int
3073 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
3074 {
3075         union if_snd_tag_query_params params = { };
3076         struct m_snd_tag *mst;
3077         int error;
3078
3079         mst = inp->inp_snd_tag;
3080         if (mst == NULL)
3081                 return (EINVAL);
3082
3083         if (mst->sw->snd_tag_query == NULL)
3084                 return (EOPNOTSUPP);
3085
3086         error = mst->sw->snd_tag_query(mst, &params);
3087         if (error == 0 && p_txqueue_level != NULL)
3088                 *p_txqueue_level = params.rate_limit.queue_level;
3089         return (error);
3090 }
3091
3092 /*
3093  * Allocate a new TX rate limit send tag from the network interface
3094  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
3095  */
3096 int
3097 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
3098     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate, struct m_snd_tag **st)
3099
3100 {
3101         union if_snd_tag_alloc_params params = {
3102                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
3103                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
3104                 .rate_limit.hdr.flowid = flowid,
3105                 .rate_limit.hdr.flowtype = flowtype,
3106                 .rate_limit.hdr.numa_domain = inp->inp_numa_domain,
3107                 .rate_limit.max_rate = max_pacing_rate,
3108                 .rate_limit.flags = M_NOWAIT,
3109         };
3110         int error;
3111
3112         INP_WLOCK_ASSERT(inp);
3113
3114         /*
3115          * If there is already a send tag, or the INP is being torn
3116          * down, allocating a new send tag is not allowed. Else send
3117          * tags may leak.
3118          */
3119         if (*st != NULL || (inp->inp_flags & INP_DROPPED) != 0)
3120                 return (EINVAL);
3121
3122         error = m_snd_tag_alloc(ifp, &params, st);
3123 #ifdef INET
3124         if (error == 0) {
3125                 counter_u64_add(rate_limit_set_ok, 1);
3126                 counter_u64_add(rate_limit_active, 1);
3127         } else if (error != EOPNOTSUPP)
3128                   counter_u64_add(rate_limit_alloc_fail, 1);
3129 #endif
3130         return (error);
3131 }
3132
3133 void
3134 in_pcbdetach_tag(struct m_snd_tag *mst)
3135 {
3136
3137         m_snd_tag_rele(mst);
3138 #ifdef INET
3139         counter_u64_add(rate_limit_active, -1);
3140 #endif
3141 }
3142
3143 /*
3144  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
3145  * if any:
3146  */
3147 void
3148 in_pcbdetach_txrtlmt(struct inpcb *inp)
3149 {
3150         struct m_snd_tag *mst;
3151
3152         INP_WLOCK_ASSERT(inp);
3153
3154         mst = inp->inp_snd_tag;
3155         inp->inp_snd_tag = NULL;
3156
3157         if (mst == NULL)
3158                 return;
3159
3160         m_snd_tag_rele(mst);
3161 #ifdef INET
3162         counter_u64_add(rate_limit_active, -1);
3163 #endif
3164 }
3165
3166 int
3167 in_pcboutput_txrtlmt_locked(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb, uint32_t max_pacing_rate)
3168 {
3169         int error;
3170
3171         /*
3172          * If the existing send tag is for the wrong interface due to
3173          * a route change, first drop the existing tag.  Set the
3174          * CHANGED flag so that we will keep trying to allocate a new
3175          * tag if we fail to allocate one this time.
3176          */
3177         if (inp->inp_snd_tag != NULL && inp->inp_snd_tag->ifp != ifp) {
3178                 in_pcbdetach_txrtlmt(inp);
3179                 inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3180         }
3181
3182         /*
3183          * NOTE: When attaching to a network interface a reference is
3184          * made to ensure the network interface doesn't go away until
3185          * all ratelimit connections are gone. The network interface
3186          * pointers compared below represent valid network interfaces,
3187          * except when comparing towards NULL.
3188          */
3189         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3190                 error = 0;
3191         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3192                 if (inp->inp_snd_tag != NULL)
3193                         in_pcbdetach_txrtlmt(inp);
3194                 error = 0;
3195         } else if (inp->inp_snd_tag == NULL) {
3196                 /*
3197                  * In order to utilize packet pacing with RSS, we need
3198                  * to wait until there is a valid RSS hash before we
3199                  * can proceed:
3200                  */
3201                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3202                         error = EAGAIN;
3203                 } else {
3204                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3205                             mb->m_pkthdr.flowid, max_pacing_rate, &inp->inp_snd_tag);
3206                 }
3207         } else {
3208                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3209         }
3210         if (error == 0 || error == EOPNOTSUPP)
3211                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3212
3213         return (error);
3214 }
3215
3216 /*
3217  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3218  * is set in the fast path and will attach/detach/modify the TX rate
3219  * limit send tag based on the socket's so_max_pacing_rate value.
3220  */
3221 void
3222 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3223 {
3224         struct socket *socket;
3225         uint32_t max_pacing_rate;
3226         bool did_upgrade;
3227
3228         if (inp == NULL)
3229                 return;
3230
3231         socket = inp->inp_socket;
3232         if (socket == NULL)
3233                 return;
3234
3235         if (!INP_WLOCKED(inp)) {
3236                 /*
3237                  * NOTE: If the write locking fails, we need to bail
3238                  * out and use the non-ratelimited ring for the
3239                  * transmit until there is a new chance to get the
3240                  * write lock.
3241                  */
3242                 if (!INP_TRY_UPGRADE(inp))
3243                         return;
3244                 did_upgrade = 1;
3245         } else {
3246                 did_upgrade = 0;
3247         }
3248
3249         /*
3250          * NOTE: The so_max_pacing_rate value is read unlocked,
3251          * because atomic updates are not required since the variable
3252          * is checked at every mbuf we send. It is assumed that the
3253          * variable read itself will be atomic.
3254          */
3255         max_pacing_rate = socket->so_max_pacing_rate;
3256
3257         in_pcboutput_txrtlmt_locked(inp, ifp, mb, max_pacing_rate);
3258
3259         if (did_upgrade)
3260                 INP_DOWNGRADE(inp);
3261 }
3262
3263 /*
3264  * Track route changes for TX rate limiting.
3265  */
3266 void
3267 in_pcboutput_eagain(struct inpcb *inp)
3268 {
3269         bool did_upgrade;
3270
3271         if (inp == NULL)
3272                 return;
3273
3274         if (inp->inp_snd_tag == NULL)
3275                 return;
3276
3277         if (!INP_WLOCKED(inp)) {
3278                 /*
3279                  * NOTE: If the write locking fails, we need to bail
3280                  * out and use the non-ratelimited ring for the
3281                  * transmit until there is a new chance to get the
3282                  * write lock.
3283                  */
3284                 if (!INP_TRY_UPGRADE(inp))
3285                         return;
3286                 did_upgrade = 1;
3287         } else {
3288                 did_upgrade = 0;
3289         }
3290
3291         /* detach rate limiting */
3292         in_pcbdetach_txrtlmt(inp);
3293
3294         /* make sure new mbuf send tag allocation is made */
3295         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3296
3297         if (did_upgrade)
3298                 INP_DOWNGRADE(inp);
3299 }
3300
3301 #ifdef INET
3302 static void
3303 rl_init(void *st)
3304 {
3305         rate_limit_new = counter_u64_alloc(M_WAITOK);
3306         rate_limit_chg = counter_u64_alloc(M_WAITOK);
3307         rate_limit_active = counter_u64_alloc(M_WAITOK);
3308         rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
3309         rate_limit_set_ok = counter_u64_alloc(M_WAITOK);
3310 }
3311
3312 SYSINIT(rl, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, rl_init, NULL);
3313 #endif
3314 #endif /* RATELIMIT */