sys/netinet/in_pcb.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1991, 1993, 1995
   5  *      The Regents of the University of California.
   6  * Copyright (c) 2007-2009 Robert N. M. Watson
   7  * Copyright (c) 2010-2011 Juniper Networks, Inc.
   8  * All rights reserved.
   9  *
  10  * Portions of this software were developed by Robert N. M. Watson under
  11  * contract to Juniper Networks, Inc.
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  * 1. Redistributions of source code must retain the above copyright
  17  *    notice, this list of conditions and the following disclaimer.
  18  * 2. Redistributions in binary form must reproduce the above copyright
  19  *    notice, this list of conditions and the following disclaimer in the
  20  *    documentation and/or other materials provided with the distribution.
  21  * 3. Neither the name of the University nor the names of its contributors
  22  *    may be used to endorse or promote products derived from this software
  23  *    without specific prior written permission.
  24  *
  25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  35  * SUCH DAMAGE.
  36  *
  37  *      @(#)in_pcb.c    8.4 (Berkeley) 5/24/95
  38  */
  39
  40 #include <sys/cdefs.h>
  41 __FBSDID("$FreeBSD$");
  42
  43 #include "opt_ddb.h"
  44 #include "opt_ipsec.h"
  45 #include "opt_inet.h"
  46 #include "opt_inet6.h"
  47 #include "opt_ratelimit.h"
  48 #include "opt_pcbgroup.h"
  49 #include "opt_rss.h"
  50
  51 #include <sys/param.h>
  52 #include <sys/systm.h>
  53 #include <sys/lock.h>
  54 #include <sys/malloc.h>
  55 #include <sys/mbuf.h>
  56 #include <sys/callout.h>
  57 #include <sys/eventhandler.h>
  58 #include <sys/domain.h>
  59 #include <sys/protosw.h>
  60 #include <sys/rmlock.h>
  61 #include <sys/smp.h>
  62 #include <sys/socket.h>
  63 #include <sys/socketvar.h>
  64 #include <sys/sockio.h>
  65 #include <sys/priv.h>
  66 #include <sys/proc.h>
  67 #include <sys/refcount.h>
  68 #include <sys/jail.h>
  69 #include <sys/kernel.h>
  70 #include <sys/sysctl.h>
  71
  72 #ifdef DDB
  73 #include <ddb/ddb.h>
  74 #endif
  75
  76 #include <vm/uma.h>
  77
  78 #include <net/if.h>
  79 #include <net/if_var.h>
  80 #include <net/if_types.h>
  81 #include <net/if_llatbl.h>
  82 #include <net/route.h>
  83 #include <net/rss_config.h>
  84 #include <net/vnet.h>
  85
  86 #if defined(INET) || defined(INET6)
  87 #include <netinet/in.h>
  88 #include <netinet/in_pcb.h>
  89 #include <netinet/ip_var.h>
  90 #include <netinet/tcp_var.h>
  91 #ifdef TCPHPTS
  92 #include <netinet/tcp_hpts.h>
  93 #endif
  94 #include <netinet/udp.h>
  95 #include <netinet/udp_var.h>
  96 #endif
  97 #ifdef INET
  98 #include <netinet/in_var.h>
  99 #endif
 100 #ifdef INET6
 101 #include <netinet/ip6.h>
 102 #include <netinet6/in6_pcb.h>
 103 #include <netinet6/in6_var.h>
 104 #include <netinet6/ip6_var.h>
 105 #endif /* INET6 */
 106
 107 #include <netipsec/ipsec_support.h>
 108
 109 #include <security/mac/mac_framework.h>
 110
 111 static struct callout   ipport_tick_callout;
 112
 113 /*
 114  * These configure the range of local port addresses assigned to
 115  * "unspecified" outgoing connections/packets/whatever.
 116  */
 117 VNET_DEFINE(int, ipport_lowfirstauto) = IPPORT_RESERVED - 1;    /* 1023 */
 118 VNET_DEFINE(int, ipport_lowlastauto) = IPPORT_RESERVEDSTART;    /* 600 */
 119 VNET_DEFINE(int, ipport_firstauto) = IPPORT_EPHEMERALFIRST;     /* 10000 */
 120 VNET_DEFINE(int, ipport_lastauto) = IPPORT_EPHEMERALLAST;       /* 65535 */
 121 VNET_DEFINE(int, ipport_hifirstauto) = IPPORT_HIFIRSTAUTO;      /* 49152 */
 122 VNET_DEFINE(int, ipport_hilastauto) = IPPORT_HILASTAUTO;        /* 65535 */
 123
 124 /*
 125  * Reserved ports accessible only to root. There are significant
 126  * security considerations that must be accounted for when changing these,
 127  * but the security benefits can be great. Please be careful.
 128  */
 129 VNET_DEFINE(int, ipport_reservedhigh) = IPPORT_RESERVED - 1;    /* 1023 */
 130 VNET_DEFINE(int, ipport_reservedlow);
 131
 132 /* Variables dealing with random ephemeral port allocation. */
 133 VNET_DEFINE(int, ipport_randomized) = 1;        /* user controlled via sysctl */
 134 VNET_DEFINE(int, ipport_randomcps) = 10;        /* user controlled via sysctl */
 135 VNET_DEFINE(int, ipport_randomtime) = 45;       /* user controlled via sysctl */
 136 VNET_DEFINE(int, ipport_stoprandom);            /* toggled by ipport_tick */
 137 VNET_DEFINE(int, ipport_tcpallocs);
 138 static VNET_DEFINE(int, ipport_tcplastcount);
 139
 140 #define V_ipport_tcplastcount           VNET(ipport_tcplastcount)
 141
 142 static void     in_pcbremlists(struct inpcb *inp);
 143 #ifdef INET
 144 static struct inpcb     *in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo,
 145                             struct in_addr faddr, u_int fport_arg,
 146                             struct in_addr laddr, u_int lport_arg,
 147                             int lookupflags, struct ifnet *ifp);
 148
 149 #define RANGECHK(var, min, max) \
 150         if ((var) < (min)) { (var) = (min); } \
 151         else if ((var) > (max)) { (var) = (max); }
 152
 153 static int
 154 sysctl_net_ipport_check(SYSCTL_HANDLER_ARGS)
 155 {
 156         int error;
 157
 158         error = sysctl_handle_int(oidp, arg1, arg2, req);
 159         if (error == 0) {
 160                 RANGECHK(V_ipport_lowfirstauto, 1, IPPORT_RESERVED - 1);
 161                 RANGECHK(V_ipport_lowlastauto, 1, IPPORT_RESERVED - 1);
 162                 RANGECHK(V_ipport_firstauto, IPPORT_RESERVED, IPPORT_MAX);
 163                 RANGECHK(V_ipport_lastauto, IPPORT_RESERVED, IPPORT_MAX);
 164                 RANGECHK(V_ipport_hifirstauto, IPPORT_RESERVED, IPPORT_MAX);
 165                 RANGECHK(V_ipport_hilastauto, IPPORT_RESERVED, IPPORT_MAX);
 166         }
 167         return (error);
 168 }
 169
 170 #undef RANGECHK
 171
 172 static SYSCTL_NODE(_net_inet_ip, IPPROTO_IP, portrange, CTLFLAG_RW, 0,
 173     "IP Ports");
 174
 175 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowfirst,
 176         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 177         &VNET_NAME(ipport_lowfirstauto), 0, &sysctl_net_ipport_check, "I", "");
 178 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, lowlast,
 179         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 180         &VNET_NAME(ipport_lowlastauto), 0, &sysctl_net_ipport_check, "I", "");
 181 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, first,
 182         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 183         &VNET_NAME(ipport_firstauto), 0, &sysctl_net_ipport_check, "I", "");
 184 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, last,
 185         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 186         &VNET_NAME(ipport_lastauto), 0, &sysctl_net_ipport_check, "I", "");
 187 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hifirst,
 188         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 189         &VNET_NAME(ipport_hifirstauto), 0, &sysctl_net_ipport_check, "I", "");
 190 SYSCTL_PROC(_net_inet_ip_portrange, OID_AUTO, hilast,
 191         CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
 192         &VNET_NAME(ipport_hilastauto), 0, &sysctl_net_ipport_check, "I", "");
 193 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedhigh,
 194         CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_SECURE,
 195         &VNET_NAME(ipport_reservedhigh), 0, "");
 196 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, reservedlow,
 197         CTLFLAG_RW|CTLFLAG_SECURE, &VNET_NAME(ipport_reservedlow), 0, "");
 198 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomized,
 199         CTLFLAG_VNET | CTLFLAG_RW,
 200         &VNET_NAME(ipport_randomized), 0, "Enable random port allocation");
 201 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomcps,
 202         CTLFLAG_VNET | CTLFLAG_RW,
 203         &VNET_NAME(ipport_randomcps), 0, "Maximum number of random port "
 204         "allocations before switching to a sequental one");
 205 SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, randomtime,
 206         CTLFLAG_VNET | CTLFLAG_RW,
 207         &VNET_NAME(ipport_randomtime), 0,
 208         "Minimum time to keep sequental port "
 209         "allocation before switching to a random one");
 210 #endif /* INET */
 211
 212 /*
 213  * in_pcb.c: manage the Protocol Control Blocks.
 214  *
 215  * NOTE: It is assumed that most of these functions will be called with
 216  * the pcbinfo lock held, and often, the inpcb lock held, as these utility
 217  * functions often modify hash chains or addresses in pcbs.
 218  */
 219
 220 /*
 221  * Different protocols initialize their inpcbs differently - giving
 222  * different name to the lock.  But they all are disposed the same.
 223  */
 224 static void
 225 inpcb_fini(void *mem, int size)
 226 {
 227         struct inpcb *inp = mem;
 228
 229         INP_LOCK_DESTROY(inp);
 230 }
 231
 232 /*
 233  * Initialize an inpcbinfo -- we should be able to reduce the number of
 234  * arguments in time.
 235  */
 236 void
 237 in_pcbinfo_init(struct inpcbinfo *pcbinfo, const char *name,
 238     struct inpcbhead *listhead, int hash_nelements, int porthash_nelements,
 239     char *inpcbzone_name, uma_init inpcbzone_init, u_int hashfields)
 240 {
 241
 242         INP_INFO_LOCK_INIT(pcbinfo, name);
 243         INP_HASH_LOCK_INIT(pcbinfo, "pcbinfohash");     /* XXXRW: argument? */
 244         INP_LIST_LOCK_INIT(pcbinfo, "pcbinfolist");
 245 #ifdef VIMAGE
 246         pcbinfo->ipi_vnet = curvnet;
 247 #endif
 248         pcbinfo->ipi_listhead = listhead;
 249         LIST_INIT(pcbinfo->ipi_listhead);
 250         pcbinfo->ipi_count = 0;
 251         pcbinfo->ipi_hashbase = hashinit(hash_nelements, M_PCB,
 252             &pcbinfo->ipi_hashmask);
 253         pcbinfo->ipi_porthashbase = hashinit(porthash_nelements, M_PCB,
 254             &pcbinfo->ipi_porthashmask);
 255 #ifdef PCBGROUP
 256         in_pcbgroup_init(pcbinfo, hashfields, hash_nelements);
 257 #endif
 258         pcbinfo->ipi_zone = uma_zcreate(inpcbzone_name, sizeof(struct inpcb),
 259             NULL, NULL, inpcbzone_init, inpcb_fini, UMA_ALIGN_PTR, 0);
 260         uma_zone_set_max(pcbinfo->ipi_zone, maxsockets);
 261         uma_zone_set_warning(pcbinfo->ipi_zone,
 262             "kern.ipc.maxsockets limit reached");
 263 }
 264
 265 /*
 266  * Destroy an inpcbinfo.
 267  */
 268 void
 269 in_pcbinfo_destroy(struct inpcbinfo *pcbinfo)
 270 {
 271
 272         KASSERT(pcbinfo->ipi_count == 0,
 273             ("%s: ipi_count = %u", __func__, pcbinfo->ipi_count));
 274
 275         hashdestroy(pcbinfo->ipi_hashbase, M_PCB, pcbinfo->ipi_hashmask);
 276         hashdestroy(pcbinfo->ipi_porthashbase, M_PCB,
 277             pcbinfo->ipi_porthashmask);
 278 #ifdef PCBGROUP
 279         in_pcbgroup_destroy(pcbinfo);
 280 #endif
 281         uma_zdestroy(pcbinfo->ipi_zone);
 282         INP_LIST_LOCK_DESTROY(pcbinfo);
 283         INP_HASH_LOCK_DESTROY(pcbinfo);
 284         INP_INFO_LOCK_DESTROY(pcbinfo);
 285 }
 286
 287 /*
 288  * Allocate a PCB and associate it with the socket.
 289  * On success return with the PCB locked.
 290  */
 291 int
 292 in_pcballoc(struct socket *so, struct inpcbinfo *pcbinfo)
 293 {
 294         struct inpcb *inp;
 295         int error;
 296
 297 #ifdef INVARIANTS
 298         if (pcbinfo == &V_tcbinfo) {
 299                 INP_INFO_RLOCK_ASSERT(pcbinfo);
 300         } else {
 301                 INP_INFO_WLOCK_ASSERT(pcbinfo);
 302         }
 303 #endif
 304
 305         error = 0;
 306         inp = uma_zalloc(pcbinfo->ipi_zone, M_NOWAIT);
 307         if (inp == NULL)
 308                 return (ENOBUFS);
 309         bzero(&inp->inp_start_zero, inp_zero_size);
 310         inp->inp_pcbinfo = pcbinfo;
 311         inp->inp_socket = so;
 312         inp->inp_cred = crhold(so->so_cred);
 313         inp->inp_inc.inc_fibnum = so->so_fibnum;
 314 #ifdef MAC
 315         error = mac_inpcb_init(inp, M_NOWAIT);
 316         if (error != 0)
 317                 goto out;
 318         mac_inpcb_create(so, inp);
 319 #endif
 320 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
 321         error = ipsec_init_pcbpolicy(inp);
 322         if (error != 0) {
 323 #ifdef MAC
 324                 mac_inpcb_destroy(inp);
 325 #endif
 326                 goto out;
 327         }
 328 #endif /*IPSEC*/
 329 #ifdef INET6
 330         if (INP_SOCKAF(so) == AF_INET6) {
 331                 inp->inp_vflag |= INP_IPV6PROTO;
 332                 if (V_ip6_v6only)
 333                         inp->inp_flags |= IN6P_IPV6_V6ONLY;
 334         }
 335 #endif
 336         INP_WLOCK(inp);
 337         INP_LIST_WLOCK(pcbinfo);
 338         LIST_INSERT_HEAD(pcbinfo->ipi_listhead, inp, inp_list);
 339         pcbinfo->ipi_count++;
 340         so->so_pcb = (caddr_t)inp;
 341 #ifdef INET6
 342         if (V_ip6_auto_flowlabel)
 343                 inp->inp_flags |= IN6P_AUTOFLOWLABEL;
 344 #endif
 345         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
 346         refcount_init(&inp->inp_refcount, 1);   /* Reference from inpcbinfo */
 347
 348         /*
 349          * Routes in inpcb's can cache L2 as well; they are guaranteed
 350          * to be cleaned up.
 351          */
 352         inp->inp_route.ro_flags = RT_LLE_CACHE;
 353         INP_LIST_WUNLOCK(pcbinfo);
 354 #if defined(IPSEC) || defined(IPSEC_SUPPORT) || defined(MAC)
 355 out:
 356         if (error != 0) {
 357                 crfree(inp->inp_cred);
 358                 uma_zfree(pcbinfo->ipi_zone, inp);
 359         }
 360 #endif
 361         return (error);
 362 }
 363
 364 #ifdef INET
 365 int
 366 in_pcbbind(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 367 {
 368         int anonport, error;
 369
 370         INP_WLOCK_ASSERT(inp);
 371         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 372
 373         if (inp->inp_lport != 0 || inp->inp_laddr.s_addr != INADDR_ANY)
 374                 return (EINVAL);
 375         anonport = nam == NULL || ((struct sockaddr_in *)nam)->sin_port == 0;
 376         error = in_pcbbind_setup(inp, nam, &inp->inp_laddr.s_addr,
 377             &inp->inp_lport, cred);
 378         if (error)
 379                 return (error);
 380         if (in_pcbinshash(inp) != 0) {
 381                 inp->inp_laddr.s_addr = INADDR_ANY;
 382                 inp->inp_lport = 0;
 383                 return (EAGAIN);
 384         }
 385         if (anonport)
 386                 inp->inp_flags |= INP_ANONPORT;
 387         return (0);
 388 }
 389 #endif
 390
 391 /*
 392  * Select a local port (number) to use.
 393  */
 394 #if defined(INET) || defined(INET6)
 395 int
 396 in_pcb_lport(struct inpcb *inp, struct in_addr *laddrp, u_short *lportp,
 397     struct ucred *cred, int lookupflags)
 398 {
 399         struct inpcbinfo *pcbinfo;
 400         struct inpcb *tmpinp;
 401         unsigned short *lastport;
 402         int count, dorandom, error;
 403         u_short aux, first, last, lport;
 404 #ifdef INET
 405         struct in_addr laddr;
 406 #endif
 407
 408         pcbinfo = inp->inp_pcbinfo;
 409
 410         /*
 411          * Because no actual state changes occur here, a global write lock on
 412          * the pcbinfo isn't required.
 413          */
 414         INP_LOCK_ASSERT(inp);
 415         INP_HASH_LOCK_ASSERT(pcbinfo);
 416
 417         if (inp->inp_flags & INP_HIGHPORT) {
 418                 first = V_ipport_hifirstauto;   /* sysctl */
 419                 last  = V_ipport_hilastauto;
 420                 lastport = &pcbinfo->ipi_lasthi;
 421         } else if (inp->inp_flags & INP_LOWPORT) {
 422                 error = priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT, 0);
 423                 if (error)
 424                         return (error);
 425                 first = V_ipport_lowfirstauto;  /* 1023 */
 426                 last  = V_ipport_lowlastauto;   /* 600 */
 427                 lastport = &pcbinfo->ipi_lastlow;
 428         } else {
 429                 first = V_ipport_firstauto;     /* sysctl */
 430                 last  = V_ipport_lastauto;
 431                 lastport = &pcbinfo->ipi_lastport;
 432         }
 433         /*
 434          * For UDP(-Lite), use random port allocation as long as the user
 435          * allows it.  For TCP (and as of yet unknown) connections,
 436          * use random port allocation only if the user allows it AND
 437          * ipport_tick() allows it.
 438          */
 439         if (V_ipport_randomized &&
 440                 (!V_ipport_stoprandom || pcbinfo == &V_udbinfo ||
 441                 pcbinfo == &V_ulitecbinfo))
 442                 dorandom = 1;
 443         else
 444                 dorandom = 0;
 445         /*
 446          * It makes no sense to do random port allocation if
 447          * we have the only port available.
 448          */
 449         if (first == last)
 450                 dorandom = 0;
 451         /* Make sure to not include UDP(-Lite) packets in the count. */
 452         if (pcbinfo != &V_udbinfo || pcbinfo != &V_ulitecbinfo)
 453                 V_ipport_tcpallocs++;
 454         /*
 455          * Instead of having two loops further down counting up or down
 456          * make sure that first is always <= last and go with only one
 457          * code path implementing all logic.
 458          */
 459         if (first > last) {
 460                 aux = first;
 461                 first = last;
 462                 last = aux;
 463         }
 464
 465 #ifdef INET
 466         /* Make the compiler happy. */
 467         laddr.s_addr = 0;
 468         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4) {
 469                 KASSERT(laddrp != NULL, ("%s: laddrp NULL for v4 inp %p",
 470                     __func__, inp));
 471                 laddr = *laddrp;
 472         }
 473 #endif
 474         tmpinp = NULL;  /* Make compiler happy. */
 475         lport = *lportp;
 476
 477         if (dorandom)
 478                 *lastport = first + (arc4random() % (last - first));
 479
 480         count = last - first;
 481
 482         do {
 483                 if (count-- < 0)        /* completely used? */
 484                         return (EADDRNOTAVAIL);
 485                 ++*lastport;
 486                 if (*lastport < first || *lastport > last)
 487                         *lastport = first;
 488                 lport = htons(*lastport);
 489
 490 #ifdef INET6
 491                 if ((inp->inp_vflag & INP_IPV6) != 0)
 492                         tmpinp = in6_pcblookup_local(pcbinfo,
 493                             &inp->in6p_laddr, lport, lookupflags, cred);
 494 #endif
 495 #if defined(INET) && defined(INET6)
 496                 else
 497 #endif
 498 #ifdef INET
 499                         tmpinp = in_pcblookup_local(pcbinfo, laddr,
 500                             lport, lookupflags, cred);
 501 #endif
 502         } while (tmpinp != NULL);
 503
 504 #ifdef INET
 505         if ((inp->inp_vflag & (INP_IPV4|INP_IPV6)) == INP_IPV4)
 506                 laddrp->s_addr = laddr.s_addr;
 507 #endif
 508         *lportp = lport;
 509
 510         return (0);
 511 }
 512
 513 /*
 514  * Return cached socket options.
 515  */
 516 short
 517 inp_so_options(const struct inpcb *inp)
 518 {
 519    short so_options;
 520
 521    so_options = 0;
 522
 523    if ((inp->inp_flags2 & INP_REUSEPORT) != 0)
 524            so_options |= SO_REUSEPORT;
 525    if ((inp->inp_flags2 & INP_REUSEADDR) != 0)
 526            so_options |= SO_REUSEADDR;
 527    return (so_options);
 528 }
 529 #endif /* INET || INET6 */
 530
 531 /*
 532  * Check if a new BINDMULTI socket is allowed to be created.
 533  *
 534  * ni points to the new inp.
 535  * oi points to the exisitng inp.
 536  *
 537  * This checks whether the existing inp also has BINDMULTI and
 538  * whether the credentials match.
 539  */
 540 int
 541 in_pcbbind_check_bindmulti(const struct inpcb *ni, const struct inpcb *oi)
 542 {
 543         /* Check permissions match */
 544         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 545             (ni->inp_cred->cr_uid !=
 546             oi->inp_cred->cr_uid))
 547                 return (0);
 548
 549         /* Check the existing inp has BINDMULTI set */
 550         if ((ni->inp_flags2 & INP_BINDMULTI) &&
 551             ((oi->inp_flags2 & INP_BINDMULTI) == 0))
 552                 return (0);
 553
 554         /*
 555          * We're okay - either INP_BINDMULTI isn't set on ni, or
 556          * it is and it matches the checks.
 557          */
 558         return (1);
 559 }
 560
 561 #ifdef INET
 562 /*
 563  * Set up a bind operation on a PCB, performing port allocation
 564  * as required, but do not actually modify the PCB. Callers can
 565  * either complete the bind by setting inp_laddr/inp_lport and
 566  * calling in_pcbinshash(), or they can just use the resulting
 567  * port and address to authorise the sending of a once-off packet.
 568  *
 569  * On error, the values of *laddrp and *lportp are not changed.
 570  */
 571 int
 572 in_pcbbind_setup(struct inpcb *inp, struct sockaddr *nam, in_addr_t *laddrp,
 573     u_short *lportp, struct ucred *cred)
 574 {
 575         struct socket *so = inp->inp_socket;
 576         struct sockaddr_in *sin;
 577         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
 578         struct in_addr laddr;
 579         u_short lport = 0;
 580         int lookupflags = 0, reuseport = (so->so_options & SO_REUSEPORT);
 581         int error;
 582
 583         /*
 584          * No state changes, so read locks are sufficient here.
 585          */
 586         INP_LOCK_ASSERT(inp);
 587         INP_HASH_LOCK_ASSERT(pcbinfo);
 588
 589         if (CK_STAILQ_EMPTY(&V_in_ifaddrhead)) /* XXX broken! */
 590                 return (EADDRNOTAVAIL);
 591         laddr.s_addr = *laddrp;
 592         if (nam != NULL && laddr.s_addr != INADDR_ANY)
 593                 return (EINVAL);
 594         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) == 0)
 595                 lookupflags = INPLOOKUP_WILDCARD;
 596         if (nam == NULL) {
 597                 if ((error = prison_local_ip4(cred, &laddr)) != 0)
 598                         return (error);
 599         } else {
 600                 sin = (struct sockaddr_in *)nam;
 601                 if (nam->sa_len != sizeof (*sin))
 602                         return (EINVAL);
 603 #ifdef notdef
 604                 /*
 605                  * We should check the family, but old programs
 606                  * incorrectly fail to initialize it.
 607                  */
 608                 if (sin->sin_family != AF_INET)
 609                         return (EAFNOSUPPORT);
 610 #endif
 611                 error = prison_local_ip4(cred, &sin->sin_addr);
 612                 if (error)
 613                         return (error);
 614                 if (sin->sin_port != *lportp) {
 615                         /* Don't allow the port to change. */
 616                         if (*lportp != 0)
 617                                 return (EINVAL);
 618                         lport = sin->sin_port;
 619                 }
 620                 /* NB: lport is left as 0 if the port isn't being changed. */
 621                 if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr))) {
 622                         /*
 623                          * Treat SO_REUSEADDR as SO_REUSEPORT for multicast;
 624                          * allow complete duplication of binding if
 625                          * SO_REUSEPORT is set, or if SO_REUSEADDR is set
 626                          * and a multicast address is bound on both
 627                          * new and duplicated sockets.
 628                          */
 629                         if ((so->so_options & (SO_REUSEADDR|SO_REUSEPORT)) != 0)
 630                                 reuseport = SO_REUSEADDR|SO_REUSEPORT;
 631                 } else if (sin->sin_addr.s_addr != INADDR_ANY) {
 632                         sin->sin_port = 0;              /* yech... */
 633                         bzero(&sin->sin_zero, sizeof(sin->sin_zero));
 634                         /*
 635                          * Is the address a local IP address?
 636                          * If INP_BINDANY is set, then the socket may be bound
 637                          * to any endpoint address, local or not.
 638                          */
 639                         if ((inp->inp_flags & INP_BINDANY) == 0 &&
 640                             ifa_ifwithaddr_check((struct sockaddr *)sin) == 0)
 641                                 return (EADDRNOTAVAIL);
 642                 }
 643                 laddr = sin->sin_addr;
 644                 if (lport) {
 645                         struct inpcb *t;
 646                         struct tcptw *tw;
 647
 648                         /* GROSS */
 649                         if (ntohs(lport) <= V_ipport_reservedhigh &&
 650                             ntohs(lport) >= V_ipport_reservedlow &&
 651                             priv_check_cred(cred, PRIV_NETINET_RESERVEDPORT,
 652                             0))
 653                                 return (EACCES);
 654                         if (!IN_MULTICAST(ntohl(sin->sin_addr.s_addr)) &&
 655                             priv_check_cred(inp->inp_cred,
 656                             PRIV_NETINET_REUSEPORT, 0) != 0) {
 657                                 t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 658                                     lport, INPLOOKUP_WILDCARD, cred);
 659         /*
 660          * XXX
 661          * This entire block sorely needs a rewrite.
 662          */
 663                                 if (t &&
 664                                     ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 665                                     ((t->inp_flags & INP_TIMEWAIT) == 0) &&
 666                                     (so->so_type != SOCK_STREAM ||
 667                                      ntohl(t->inp_faddr.s_addr) == INADDR_ANY) &&
 668                                     (ntohl(sin->sin_addr.s_addr) != INADDR_ANY ||
 669                                      ntohl(t->inp_laddr.s_addr) != INADDR_ANY ||
 670                                      (t->inp_flags2 & INP_REUSEPORT) == 0) &&
 671                                     (inp->inp_cred->cr_uid !=
 672                                      t->inp_cred->cr_uid))
 673                                         return (EADDRINUSE);
 674
 675                                 /*
 676                                  * If the socket is a BINDMULTI socket, then
 677                                  * the credentials need to match and the
 678                                  * original socket also has to have been bound
 679                                  * with BINDMULTI.
 680                                  */
 681                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 682                                         return (EADDRINUSE);
 683                         }
 684                         t = in_pcblookup_local(pcbinfo, sin->sin_addr,
 685                             lport, lookupflags, cred);
 686                         if (t && (t->inp_flags & INP_TIMEWAIT)) {
 687                                 /*
 688                                  * XXXRW: If an incpb has had its timewait
 689                                  * state recycled, we treat the address as
 690                                  * being in use (for now).  This is better
 691                                  * than a panic, but not desirable.
 692                                  */
 693                                 tw = intotw(t);
 694                                 if (tw == NULL ||
 695                                     (reuseport & tw->tw_so_options) == 0)
 696                                         return (EADDRINUSE);
 697                         } else if (t &&
 698                             ((inp->inp_flags2 & INP_BINDMULTI) == 0) &&
 699                             (reuseport & inp_so_options(t)) == 0) {
 700 #ifdef INET6
 701                                 if (ntohl(sin->sin_addr.s_addr) !=
 702                                     INADDR_ANY ||
 703                                     ntohl(t->inp_laddr.s_addr) !=
 704                                     INADDR_ANY ||
 705                                     (inp->inp_vflag & INP_IPV6PROTO) == 0 ||
 706                                     (t->inp_vflag & INP_IPV6PROTO) == 0)
 707 #endif
 708                                 return (EADDRINUSE);
 709                                 if (t && (! in_pcbbind_check_bindmulti(inp, t)))
 710                                         return (EADDRINUSE);
 711                         }
 712                 }
 713         }
 714         if (*lportp != 0)
 715                 lport = *lportp;
 716         if (lport == 0) {
 717                 error = in_pcb_lport(inp, &laddr, &lport, cred, lookupflags);
 718                 if (error != 0)
 719                         return (error);
 720
 721         }
 722         *laddrp = laddr.s_addr;
 723         *lportp = lport;
 724         return (0);
 725 }
 726
 727 /*
 728  * Connect from a socket to a specified address.
 729  * Both address and port must be specified in argument sin.
 730  * If don't have a local address for this socket yet,
 731  * then pick one.
 732  */
 733 int
 734 in_pcbconnect_mbuf(struct inpcb *inp, struct sockaddr *nam,
 735     struct ucred *cred, struct mbuf *m)
 736 {
 737         u_short lport, fport;
 738         in_addr_t laddr, faddr;
 739         int anonport, error;
 740
 741         INP_WLOCK_ASSERT(inp);
 742         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
 743
 744         lport = inp->inp_lport;
 745         laddr = inp->inp_laddr.s_addr;
 746         anonport = (lport == 0);
 747         error = in_pcbconnect_setup(inp, nam, &laddr, &lport, &faddr, &fport,
 748             NULL, cred);
 749         if (error)
 750                 return (error);
 751
 752         /* Do the initial binding of the local address if required. */
 753         if (inp->inp_laddr.s_addr == INADDR_ANY && inp->inp_lport == 0) {
 754                 inp->inp_lport = lport;
 755                 inp->inp_laddr.s_addr = laddr;
 756                 if (in_pcbinshash(inp) != 0) {
 757                         inp->inp_laddr.s_addr = INADDR_ANY;
 758                         inp->inp_lport = 0;
 759                         return (EAGAIN);
 760                 }
 761         }
 762
 763         /* Commit the remaining changes. */
 764         inp->inp_lport = lport;
 765         inp->inp_laddr.s_addr = laddr;
 766         inp->inp_faddr.s_addr = faddr;
 767         inp->inp_fport = fport;
 768         in_pcbrehash_mbuf(inp, m);
 769
 770         if (anonport)
 771                 inp->inp_flags |= INP_ANONPORT;
 772         return (0);
 773 }
 774
 775 int
 776 in_pcbconnect(struct inpcb *inp, struct sockaddr *nam, struct ucred *cred)
 777 {
 778
 779         return (in_pcbconnect_mbuf(inp, nam, cred, NULL));
 780 }
 781
 782 /*
 783  * Do proper source address selection on an unbound socket in case
 784  * of connect. Take jails into account as well.
 785  */
 786 int
 787 in_pcbladdr(struct inpcb *inp, struct in_addr *faddr, struct in_addr *laddr,
 788     struct ucred *cred)
 789 {
 790         struct ifaddr *ifa;
 791         struct sockaddr *sa;
 792         struct sockaddr_in *sin;
 793         struct route sro;
 794         int error;
 795
 796         KASSERT(laddr != NULL, ("%s: laddr NULL", __func__));
 797         /*
 798          * Bypass source address selection and use the primary jail IP
 799          * if requested.
 800          */
 801         if (cred != NULL && !prison_saddrsel_ip4(cred, laddr))
 802                 return (0);
 803
 804         error = 0;
 805         bzero(&sro, sizeof(sro));
 806
 807         sin = (struct sockaddr_in *)&sro.ro_dst;
 808         sin->sin_family = AF_INET;
 809         sin->sin_len = sizeof(struct sockaddr_in);
 810         sin->sin_addr.s_addr = faddr->s_addr;
 811
 812         /*
 813          * If route is known our src addr is taken from the i/f,
 814          * else punt.
 815          *
 816          * Find out route to destination.
 817          */
 818         if ((inp->inp_socket->so_options & SO_DONTROUTE) == 0)
 819                 in_rtalloc_ign(&sro, 0, inp->inp_inc.inc_fibnum);
 820
 821         /*
 822          * If we found a route, use the address corresponding to
 823          * the outgoing interface.
 824          *
 825          * Otherwise assume faddr is reachable on a directly connected
 826          * network and try to find a corresponding interface to take
 827          * the source address from.
 828          */
 829         NET_EPOCH_ENTER();
 830         if (sro.ro_rt == NULL || sro.ro_rt->rt_ifp == NULL) {
 831                 struct in_ifaddr *ia;
 832                 struct ifnet *ifp;
 833
 834                 ia = ifatoia(ifa_ifwithdstaddr((struct sockaddr *)sin,
 835                                         inp->inp_socket->so_fibnum));
 836                 if (ia == NULL) {
 837                         ia = ifatoia(ifa_ifwithnet((struct sockaddr *)sin, 0,
 838                                                 inp->inp_socket->so_fibnum));
 839
 840                 }
 841                 if (ia == NULL) {
 842                         error = ENETUNREACH;
 843                         goto done;
 844                 }
 845
 846                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 847                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 848                         goto done;
 849                 }
 850
 851                 ifp = ia->ia_ifp;
 852                 ia = NULL;
 853                 IF_ADDR_RLOCK(ifp);
 854                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 855
 856                         sa = ifa->ifa_addr;
 857                         if (sa->sa_family != AF_INET)
 858                                 continue;
 859                         sin = (struct sockaddr_in *)sa;
 860                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 861                                 ia = (struct in_ifaddr *)ifa;
 862                                 break;
 863                         }
 864                 }
 865                 if (ia != NULL) {
 866                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 867                         IF_ADDR_RUNLOCK(ifp);
 868                         goto done;
 869                 }
 870                 IF_ADDR_RUNLOCK(ifp);
 871
 872                 /* 3. As a last resort return the 'default' jail address. */
 873                 error = prison_get_ip4(cred, laddr);
 874                 goto done;
 875         }
 876
 877         /*
 878          * If the outgoing interface on the route found is not
 879          * a loopback interface, use the address from that interface.
 880          * In case of jails do those three steps:
 881          * 1. check if the interface address belongs to the jail. If so use it.
 882          * 2. check if we have any address on the outgoing interface
 883          *    belonging to this jail. If so use it.
 884          * 3. as a last resort return the 'default' jail address.
 885          */
 886         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) == 0) {
 887                 struct in_ifaddr *ia;
 888                 struct ifnet *ifp;
 889
 890                 /* If not jailed, use the default returned. */
 891                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 892                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
 893                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 894                         goto done;
 895                 }
 896
 897                 /* Jailed. */
 898                 /* 1. Check if the iface address belongs to the jail. */
 899                 sin = (struct sockaddr_in *)sro.ro_rt->rt_ifa->ifa_addr;
 900                 if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 901                         ia = (struct in_ifaddr *)sro.ro_rt->rt_ifa;
 902                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 903                         goto done;
 904                 }
 905
 906                 /*
 907                  * 2. Check if we have any address on the outgoing interface
 908                  *    belonging to this jail.
 909                  */
 910                 ia = NULL;
 911                 ifp = sro.ro_rt->rt_ifp;
 912                 IF_ADDR_RLOCK(ifp);
 913                 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 914                         sa = ifa->ifa_addr;
 915                         if (sa->sa_family != AF_INET)
 916                                 continue;
 917                         sin = (struct sockaddr_in *)sa;
 918                         if (prison_check_ip4(cred, &sin->sin_addr) == 0) {
 919                                 ia = (struct in_ifaddr *)ifa;
 920                                 break;
 921                         }
 922                 }
 923                 if (ia != NULL) {
 924                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 925                         IF_ADDR_RUNLOCK(ifp);
 926                         goto done;
 927                 }
 928                 IF_ADDR_RUNLOCK(ifp);
 929
 930                 /* 3. As a last resort return the 'default' jail address. */
 931                 error = prison_get_ip4(cred, laddr);
 932                 goto done;
 933         }
 934
 935         /*
 936          * The outgoing interface is marked with 'loopback net', so a route
 937          * to ourselves is here.
 938          * Try to find the interface of the destination address and then
 939          * take the address from there. That interface is not necessarily
 940          * a loopback interface.
 941          * In case of jails, check that it is an address of the jail
 942          * and if we cannot find, fall back to the 'default' jail address.
 943          */
 944         if ((sro.ro_rt->rt_ifp->if_flags & IFF_LOOPBACK) != 0) {
 945                 struct sockaddr_in sain;
 946                 struct in_ifaddr *ia;
 947
 948                 bzero(&sain, sizeof(struct sockaddr_in));
 949                 sain.sin_family = AF_INET;
 950                 sain.sin_len = sizeof(struct sockaddr_in);
 951                 sain.sin_addr.s_addr = faddr->s_addr;
 952
 953                 ia = ifatoia(ifa_ifwithdstaddr(sintosa(&sain),
 954                                         inp->inp_socket->so_fibnum));
 955                 if (ia == NULL)
 956                         ia = ifatoia(ifa_ifwithnet(sintosa(&sain), 0,
 957                                                 inp->inp_socket->so_fibnum));
 958                 if (ia == NULL)
 959                         ia = ifatoia(ifa_ifwithaddr(sintosa(&sain)));
 960
 961                 if (cred == NULL || !prison_flag(cred, PR_IP4)) {
 962                         if (ia == NULL) {
 963                                 error = ENETUNREACH;
 964                                 goto done;
 965                         }
 966                         laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 967                         goto done;
 968                 }
 969
 970                 /* Jailed. */
 971                 if (ia != NULL) {
 972                         struct ifnet *ifp;
 973
 974                         ifp = ia->ia_ifp;
 975                         ia = NULL;
 976                         IF_ADDR_RLOCK(ifp);
 977                         CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
 978
 979                                 sa = ifa->ifa_addr;
 980                                 if (sa->sa_family != AF_INET)
 981                                         continue;
 982                                 sin = (struct sockaddr_in *)sa;
 983                                 if (prison_check_ip4(cred,
 984                                     &sin->sin_addr) == 0) {
 985                                         ia = (struct in_ifaddr *)ifa;
 986                                         break;
 987                                 }
 988                         }
 989                         if (ia != NULL) {
 990                                 laddr->s_addr = ia->ia_addr.sin_addr.s_addr;
 991                                 IF_ADDR_RUNLOCK(ifp);
 992                                 goto done;
 993                         }
 994                         IF_ADDR_RUNLOCK(ifp);
 995                 }
 996
 997                 /* 3. As a last resort return the 'default' jail address. */
 998                 error = prison_get_ip4(cred, laddr);
 999                 goto done;
1000         }
1001
1002 done:
1003         NET_EPOCH_EXIT();
1004         if (sro.ro_rt != NULL)
1005                 RTFREE(sro.ro_rt);
1006         return (error);
1007 }
1008
1009 /*
1010  * Set up for a connect from a socket to the specified address.
1011  * On entry, *laddrp and *lportp should contain the current local
1012  * address and port for the PCB; these are updated to the values
1013  * that should be placed in inp_laddr and inp_lport to complete
1014  * the connect.
1015  *
1016  * On success, *faddrp and *fportp will be set to the remote address
1017  * and port. These are not updated in the error case.
1018  *
1019  * If the operation fails because the connection already exists,
1020  * *oinpp will be set to the PCB of that connection so that the
1021  * caller can decide to override it. In all other cases, *oinpp
1022  * is set to NULL.
1023  */
1024 int
1025 in_pcbconnect_setup(struct inpcb *inp, struct sockaddr *nam,
1026     in_addr_t *laddrp, u_short *lportp, in_addr_t *faddrp, u_short *fportp,
1027     struct inpcb **oinpp, struct ucred *cred)
1028 {
1029         struct rm_priotracker in_ifa_tracker;
1030         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1031         struct in_ifaddr *ia;
1032         struct inpcb *oinp;
1033         struct in_addr laddr, faddr;
1034         u_short lport, fport;
1035         int error;
1036
1037         /*
1038          * Because a global state change doesn't actually occur here, a read
1039          * lock is sufficient.
1040          */
1041         INP_LOCK_ASSERT(inp);
1042         INP_HASH_LOCK_ASSERT(inp->inp_pcbinfo);
1043
1044         if (oinpp != NULL)
1045                 *oinpp = NULL;
1046         if (nam->sa_len != sizeof (*sin))
1047                 return (EINVAL);
1048         if (sin->sin_family != AF_INET)
1049                 return (EAFNOSUPPORT);
1050         if (sin->sin_port == 0)
1051                 return (EADDRNOTAVAIL);
1052         laddr.s_addr = *laddrp;
1053         lport = *lportp;
1054         faddr = sin->sin_addr;
1055         fport = sin->sin_port;
1056
1057         if (!CK_STAILQ_EMPTY(&V_in_ifaddrhead)) {
1058                 /*
1059                  * If the destination address is INADDR_ANY,
1060                  * use the primary local address.
1061                  * If the supplied address is INADDR_BROADCAST,
1062                  * and the primary interface supports broadcast,
1063                  * choose the broadcast address for that interface.
1064                  */
1065                 if (faddr.s_addr == INADDR_ANY) {
1066                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1067                         faddr =
1068                             IA_SIN(CK_STAILQ_FIRST(&V_in_ifaddrhead))->sin_addr;
1069                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1070                         if (cred != NULL &&
1071                             (error = prison_get_ip4(cred, &faddr)) != 0)
1072                                 return (error);
1073                 } else if (faddr.s_addr == (u_long)INADDR_BROADCAST) {
1074                         IN_IFADDR_RLOCK(&in_ifa_tracker);
1075                         if (CK_STAILQ_FIRST(&V_in_ifaddrhead)->ia_ifp->if_flags &
1076                             IFF_BROADCAST)
1077                                 faddr = satosin(&CK_STAILQ_FIRST(
1078                                     &V_in_ifaddrhead)->ia_broadaddr)->sin_addr;
1079                         IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1080                 }
1081         }
1082         if (laddr.s_addr == INADDR_ANY) {
1083                 error = in_pcbladdr(inp, &faddr, &laddr, cred);
1084                 /*
1085                  * If the destination address is multicast and an outgoing
1086                  * interface has been set as a multicast option, prefer the
1087                  * address of that interface as our source address.
1088                  */
1089                 if (IN_MULTICAST(ntohl(faddr.s_addr)) &&
1090                     inp->inp_moptions != NULL) {
1091                         struct ip_moptions *imo;
1092                         struct ifnet *ifp;
1093
1094                         imo = inp->inp_moptions;
1095                         if (imo->imo_multicast_ifp != NULL) {
1096                                 ifp = imo->imo_multicast_ifp;
1097                                 IN_IFADDR_RLOCK(&in_ifa_tracker);
1098                                 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1099                                         if ((ia->ia_ifp == ifp) &&
1100                                             (cred == NULL ||
1101                                             prison_check_ip4(cred,
1102                                             &ia->ia_addr.sin_addr) == 0))
1103                                                 break;
1104                                 }
1105                                 if (ia == NULL)
1106                                         error = EADDRNOTAVAIL;
1107                                 else {
1108                                         laddr = ia->ia_addr.sin_addr;
1109                                         error = 0;
1110                                 }
1111                                 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1112                         }
1113                 }
1114                 if (error)
1115                         return (error);
1116         }
1117         oinp = in_pcblookup_hash_locked(inp->inp_pcbinfo, faddr, fport,
1118             laddr, lport, 0, NULL);
1119         if (oinp != NULL) {
1120                 if (oinpp != NULL)
1121                         *oinpp = oinp;
1122                 return (EADDRINUSE);
1123         }
1124         if (lport == 0) {
1125                 error = in_pcbbind_setup(inp, NULL, &laddr.s_addr, &lport,
1126                     cred);
1127                 if (error)
1128                         return (error);
1129         }
1130         *laddrp = laddr.s_addr;
1131         *lportp = lport;
1132         *faddrp = faddr.s_addr;
1133         *fportp = fport;
1134         return (0);
1135 }
1136
1137 void
1138 in_pcbdisconnect(struct inpcb *inp)
1139 {
1140
1141         INP_WLOCK_ASSERT(inp);
1142         INP_HASH_WLOCK_ASSERT(inp->inp_pcbinfo);
1143
1144         inp->inp_faddr.s_addr = INADDR_ANY;
1145         inp->inp_fport = 0;
1146         in_pcbrehash(inp);
1147 }
1148 #endif /* INET */
1149
1150 /*
1151  * in_pcbdetach() is responsibe for disassociating a socket from an inpcb.
1152  * For most protocols, this will be invoked immediately prior to calling
1153  * in_pcbfree().  However, with TCP the inpcb may significantly outlive the
1154  * socket, in which case in_pcbfree() is deferred.
1155  */
1156 void
1157 in_pcbdetach(struct inpcb *inp)
1158 {
1159
1160         KASSERT(inp->inp_socket != NULL, ("%s: inp_socket == NULL", __func__));
1161
1162 #ifdef RATELIMIT
1163         if (inp->inp_snd_tag != NULL)
1164                 in_pcbdetach_txrtlmt(inp);
1165 #endif
1166         inp->inp_socket->so_pcb = NULL;
1167         inp->inp_socket = NULL;
1168 }
1169
1170 /*
1171  * in_pcbref() bumps the reference count on an inpcb in order to maintain
1172  * stability of an inpcb pointer despite the inpcb lock being released.  This
1173  * is used in TCP when the inpcbinfo lock needs to be acquired or upgraded,
1174  * but where the inpcb lock may already held, or when acquiring a reference
1175  * via a pcbgroup.
1176  *
1177  * in_pcbref() should be used only to provide brief memory stability, and
1178  * must always be followed by a call to INP_WLOCK() and in_pcbrele() to
1179  * garbage collect the inpcb if it has been in_pcbfree()'d from another
1180  * context.  Until in_pcbrele() has returned that the inpcb is still valid,
1181  * lock and rele are the *only* safe operations that may be performed on the
1182  * inpcb.
1183  *
1184  * While the inpcb will not be freed, releasing the inpcb lock means that the
1185  * connection's state may change, so the caller should be careful to
1186  * revalidate any cached state on reacquiring the lock.  Drop the reference
1187  * using in_pcbrele().
1188  */
1189 void
1190 in_pcbref(struct inpcb *inp)
1191 {
1192
1193         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1194
1195         refcount_acquire(&inp->inp_refcount);
1196 }
1197
1198 /*
1199  * Drop a refcount on an inpcb elevated using in_pcbref(); because a call to
1200  * in_pcbfree() may have been made between in_pcbref() and in_pcbrele(), we
1201  * return a flag indicating whether or not the inpcb remains valid.  If it is
1202  * valid, we return with the inpcb lock held.
1203  *
1204  * Notice that, unlike in_pcbref(), the inpcb lock must be held to drop a
1205  * reference on an inpcb.  Historically more work was done here (actually, in
1206  * in_pcbfree_internal()) but has been moved to in_pcbfree() to avoid the
1207  * need for the pcbinfo lock in in_pcbrele().  Deferring the free is entirely
1208  * about memory stability (and continued use of the write lock).
1209  */
1210 int
1211 in_pcbrele_rlocked(struct inpcb *inp)
1212 {
1213         struct inpcbinfo *pcbinfo;
1214
1215         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1216
1217         INP_RLOCK_ASSERT(inp);
1218
1219         if (refcount_release(&inp->inp_refcount) == 0) {
1220                 /*
1221                  * If the inpcb has been freed, let the caller know, even if
1222                  * this isn't the last reference.
1223                  */
1224                 if (inp->inp_flags2 & INP_FREED) {
1225                         INP_RUNLOCK(inp);
1226                         return (1);
1227                 }
1228                 return (0);
1229         }
1230
1231         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1232 #ifdef TCPHPTS
1233         if (inp->inp_in_hpts || inp->inp_in_input) {
1234                 struct tcp_hpts_entry *hpts;
1235                 /*
1236                  * We should not be on the hpts at
1237                  * this point in any form. we must
1238                  * get the lock to be sure.
1239                  */
1240                 hpts = tcp_hpts_lock(inp);
1241                 if (inp->inp_in_hpts)
1242                         panic("Hpts:%p inp:%p at free still on hpts",
1243                               hpts, inp);
1244                 mtx_unlock(&hpts->p_mtx);
1245                 hpts = tcp_input_lock(inp);
1246                 if (inp->inp_in_input)
1247                         panic("Hpts:%p inp:%p at free still on input hpts",
1248                               hpts, inp);
1249                 mtx_unlock(&hpts->p_mtx);
1250         }
1251 #endif
1252         INP_RUNLOCK(inp);
1253         pcbinfo = inp->inp_pcbinfo;
1254         uma_zfree(pcbinfo->ipi_zone, inp);
1255         return (1);
1256 }
1257
1258 int
1259 in_pcbrele_wlocked(struct inpcb *inp)
1260 {
1261         struct inpcbinfo *pcbinfo;
1262
1263         KASSERT(inp->inp_refcount > 0, ("%s: refcount 0", __func__));
1264
1265         INP_WLOCK_ASSERT(inp);
1266
1267         if (refcount_release(&inp->inp_refcount) == 0) {
1268                 /*
1269                  * If the inpcb has been freed, let the caller know, even if
1270                  * this isn't the last reference.
1271                  */
1272                 if (inp->inp_flags2 & INP_FREED) {
1273                         INP_WUNLOCK(inp);
1274                         return (1);
1275                 }
1276                 return (0);
1277         }
1278
1279         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1280 #ifdef TCPHPTS
1281         if (inp->inp_in_hpts || inp->inp_in_input) {
1282                 struct tcp_hpts_entry *hpts;
1283                 /*
1284                  * We should not be on the hpts at
1285                  * this point in any form. we must
1286                  * get the lock to be sure.
1287                  */
1288                 hpts = tcp_hpts_lock(inp);
1289                 if (inp->inp_in_hpts)
1290                         panic("Hpts:%p inp:%p at free still on hpts",
1291                               hpts, inp);
1292                 mtx_unlock(&hpts->p_mtx);
1293                 hpts = tcp_input_lock(inp);
1294                 if (inp->inp_in_input)
1295                         panic("Hpts:%p inp:%p at free still on input hpts",
1296                               hpts, inp);
1297                 mtx_unlock(&hpts->p_mtx);
1298         }
1299 #endif
1300         INP_WUNLOCK(inp);
1301         pcbinfo = inp->inp_pcbinfo;
1302         uma_zfree(pcbinfo->ipi_zone, inp);
1303         return (1);
1304 }
1305
1306 /*
1307  * Temporary wrapper.
1308  */
1309 int
1310 in_pcbrele(struct inpcb *inp)
1311 {
1312
1313         return (in_pcbrele_wlocked(inp));
1314 }
1315
1316 void
1317 in_pcblist_rele_rlocked(epoch_context_t ctx)
1318 {
1319         struct in_pcblist *il;
1320         struct inpcb *inp;
1321         struct inpcbinfo *pcbinfo;
1322         int i, n;
1323
1324         il = __containerof(ctx, struct in_pcblist, il_epoch_ctx);
1325         pcbinfo = il->il_pcbinfo;
1326         n = il->il_count;
1327         INP_INFO_WLOCK(pcbinfo);
1328         for (i = 0; i < n; i++) {
1329                 inp = il->il_inp_list[i];
1330                 INP_RLOCK(inp);
1331                 if (!in_pcbrele_rlocked(inp))
1332                         INP_RUNLOCK(inp);
1333         }
1334         INP_INFO_WUNLOCK(pcbinfo);
1335         free(il, M_TEMP);
1336 }
1337
1338 /*
1339  * Unconditionally schedule an inpcb to be freed by decrementing its
1340  * reference count, which should occur only after the inpcb has been detached
1341  * from its socket.  If another thread holds a temporary reference (acquired
1342  * using in_pcbref()) then the free is deferred until that reference is
1343  * released using in_pcbrele(), but the inpcb is still unlocked.  Almost all
1344  * work, including removal from global lists, is done in this context, where
1345  * the pcbinfo lock is held.
1346  */
1347 void
1348 in_pcbfree(struct inpcb *inp)
1349 {
1350         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
1351
1352 #ifdef INET6
1353         struct ip6_moptions *im6o = NULL;
1354 #endif
1355 #ifdef INET
1356         struct ip_moptions *imo = NULL;
1357 #endif
1358         KASSERT(inp->inp_socket == NULL, ("%s: inp_socket != NULL", __func__));
1359
1360         KASSERT((inp->inp_flags2 & INP_FREED) == 0,
1361             ("%s: called twice for pcb %p", __func__, inp));
1362         if (inp->inp_flags2 & INP_FREED) {
1363                 INP_WUNLOCK(inp);
1364                 return;
1365         }
1366
1367 #ifdef INVARIANTS
1368         if (pcbinfo == &V_tcbinfo) {
1369                 INP_INFO_LOCK_ASSERT(pcbinfo);
1370         } else {
1371                 INP_INFO_WLOCK_ASSERT(pcbinfo);
1372         }
1373 #endif
1374         INP_WLOCK_ASSERT(inp);
1375
1376 #ifdef INET
1377         imo = inp->inp_moptions;
1378         inp->inp_moptions = NULL;
1379 #endif
1380         /* XXXRW: Do as much as possible here. */
1381 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
1382         if (inp->inp_sp != NULL)
1383                 ipsec_delete_pcbpolicy(inp);
1384 #endif
1385         INP_LIST_WLOCK(pcbinfo);
1386         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
1387         in_pcbremlists(inp);
1388         INP_LIST_WUNLOCK(pcbinfo);
1389 #ifdef INET6
1390         if (inp->inp_vflag & INP_IPV6PROTO) {
1391                 ip6_freepcbopts(inp->in6p_outputopts);
1392                 im6o = inp->in6p_moptions;
1393                 inp->in6p_moptions = NULL;
1394         }
1395 #endif
1396         if (inp->inp_options)
1397                 (void)m_free(inp->inp_options);
1398         RO_INVALIDATE_CACHE(&inp->inp_route);
1399
1400         inp->inp_vflag = 0;
1401         inp->inp_flags2 |= INP_FREED;
1402         crfree(inp->inp_cred);
1403 #ifdef MAC
1404         mac_inpcb_destroy(inp);
1405 #endif
1406 #ifdef INET6
1407         ip6_freemoptions(im6o);
1408 #endif
1409 #ifdef INET
1410         inp_freemoptions(imo);
1411 #endif
1412         if (!in_pcbrele_wlocked(inp))
1413                 INP_WUNLOCK(inp);
1414 }
1415
1416 /*
1417  * in_pcbdrop() removes an inpcb from hashed lists, releasing its address and
1418  * port reservation, and preventing it from being returned by inpcb lookups.
1419  *
1420  * It is used by TCP to mark an inpcb as unused and avoid future packet
1421  * delivery or event notification when a socket remains open but TCP has
1422  * closed.  This might occur as a result of a shutdown()-initiated TCP close
1423  * or a RST on the wire, and allows the port binding to be reused while still
1424  * maintaining the invariant that so_pcb always points to a valid inpcb until
1425  * in_pcbdetach().
1426  *
1427  * XXXRW: Possibly in_pcbdrop() should also prevent future notifications by
1428  * in_pcbnotifyall() and in_pcbpurgeif0()?
1429  */
1430 void
1431 in_pcbdrop(struct inpcb *inp)
1432 {
1433
1434         INP_WLOCK_ASSERT(inp);
1435
1436         /*
1437          * XXXRW: Possibly we should protect the setting of INP_DROPPED with
1438          * the hash lock...?
1439          */
1440         inp->inp_flags |= INP_DROPPED;
1441         if (inp->inp_flags & INP_INHASHLIST) {
1442                 struct inpcbport *phd = inp->inp_phd;
1443
1444                 INP_HASH_WLOCK(inp->inp_pcbinfo);
1445                 LIST_REMOVE(inp, inp_hash);
1446                 LIST_REMOVE(inp, inp_portlist);
1447                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
1448                         LIST_REMOVE(phd, phd_hash);
1449                         free(phd, M_PCB);
1450                 }
1451                 INP_HASH_WUNLOCK(inp->inp_pcbinfo);
1452                 inp->inp_flags &= ~INP_INHASHLIST;
1453 #ifdef PCBGROUP
1454                 in_pcbgroup_remove(inp);
1455 #endif
1456         }
1457 }
1458
1459 #ifdef INET
1460 /*
1461  * Common routines to return the socket addresses associated with inpcbs.
1462  */
1463 struct sockaddr *
1464 in_sockaddr(in_port_t port, struct in_addr *addr_p)
1465 {
1466         struct sockaddr_in *sin;
1467
1468         sin = malloc(sizeof *sin, M_SONAME,
1469                 M_WAITOK | M_ZERO);
1470         sin->sin_family = AF_INET;
1471         sin->sin_len = sizeof(*sin);
1472         sin->sin_addr = *addr_p;
1473         sin->sin_port = port;
1474
1475         return (struct sockaddr *)sin;
1476 }
1477
1478 int
1479 in_getsockaddr(struct socket *so, struct sockaddr **nam)
1480 {
1481         struct inpcb *inp;
1482         struct in_addr addr;
1483         in_port_t port;
1484
1485         inp = sotoinpcb(so);
1486         KASSERT(inp != NULL, ("in_getsockaddr: inp == NULL"));
1487
1488         INP_RLOCK(inp);
1489         port = inp->inp_lport;
1490         addr = inp->inp_laddr;
1491         INP_RUNLOCK(inp);
1492
1493         *nam = in_sockaddr(port, &addr);
1494         return 0;
1495 }
1496
1497 int
1498 in_getpeeraddr(struct socket *so, struct sockaddr **nam)
1499 {
1500         struct inpcb *inp;
1501         struct in_addr addr;
1502         in_port_t port;
1503
1504         inp = sotoinpcb(so);
1505         KASSERT(inp != NULL, ("in_getpeeraddr: inp == NULL"));
1506
1507         INP_RLOCK(inp);
1508         port = inp->inp_fport;
1509         addr = inp->inp_faddr;
1510         INP_RUNLOCK(inp);
1511
1512         *nam = in_sockaddr(port, &addr);
1513         return 0;
1514 }
1515
1516 void
1517 in_pcbnotifyall(struct inpcbinfo *pcbinfo, struct in_addr faddr, int errno,
1518     struct inpcb *(*notify)(struct inpcb *, int))
1519 {
1520         struct inpcb *inp, *inp_temp;
1521
1522         INP_INFO_WLOCK(pcbinfo);
1523         LIST_FOREACH_SAFE(inp, pcbinfo->ipi_listhead, inp_list, inp_temp) {
1524                 INP_WLOCK(inp);
1525 #ifdef INET6
1526                 if ((inp->inp_vflag & INP_IPV4) == 0) {
1527                         INP_WUNLOCK(inp);
1528                         continue;
1529                 }
1530 #endif
1531                 if (inp->inp_faddr.s_addr != faddr.s_addr ||
1532                     inp->inp_socket == NULL) {
1533                         INP_WUNLOCK(inp);
1534                         continue;
1535                 }
1536                 if ((*notify)(inp, errno))
1537                         INP_WUNLOCK(inp);
1538         }
1539         INP_INFO_WUNLOCK(pcbinfo);
1540 }
1541
1542 void
1543 in_pcbpurgeif0(struct inpcbinfo *pcbinfo, struct ifnet *ifp)
1544 {
1545         struct inpcb *inp;
1546         struct ip_moptions *imo;
1547         int i, gap;
1548
1549         INP_INFO_WLOCK(pcbinfo);
1550         LIST_FOREACH(inp, pcbinfo->ipi_listhead, inp_list) {
1551                 INP_WLOCK(inp);
1552                 imo = inp->inp_moptions;
1553                 if ((inp->inp_vflag & INP_IPV4) &&
1554                     imo != NULL) {
1555                         /*
1556                          * Unselect the outgoing interface if it is being
1557                          * detached.
1558                          */
1559                         if (imo->imo_multicast_ifp == ifp)
1560                                 imo->imo_multicast_ifp = NULL;
1561
1562                         /*
1563                          * Drop multicast group membership if we joined
1564                          * through the interface being detached.
1565                          *
1566                          * XXX This can all be deferred to an epoch_call
1567                          */
1568                         for (i = 0, gap = 0; i < imo->imo_num_memberships;
1569                             i++) {
1570                                 if (imo->imo_membership[i]->inm_ifp == ifp) {
1571                                         IN_MULTI_LOCK_ASSERT();
1572                                         in_leavegroup_locked(imo->imo_membership[i], NULL);
1573                                         gap++;
1574                                 } else if (gap != 0)
1575                                         imo->imo_membership[i - gap] =
1576                                             imo->imo_membership[i];
1577                         }
1578                         imo->imo_num_memberships -= gap;
1579                 }
1580                 INP_WUNLOCK(inp);
1581         }
1582         INP_INFO_WUNLOCK(pcbinfo);
1583 }
1584
1585 /*
1586  * Lookup a PCB based on the local address and port.  Caller must hold the
1587  * hash lock.  No inpcb locks or references are acquired.
1588  */
1589 #define INP_LOOKUP_MAPPED_PCB_COST      3
1590 struct inpcb *
1591 in_pcblookup_local(struct inpcbinfo *pcbinfo, struct in_addr laddr,
1592     u_short lport, int lookupflags, struct ucred *cred)
1593 {
1594         struct inpcb *inp;
1595 #ifdef INET6
1596         int matchwild = 3 + INP_LOOKUP_MAPPED_PCB_COST;
1597 #else
1598         int matchwild = 3;
1599 #endif
1600         int wildcard;
1601
1602         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1603             ("%s: invalid lookup flags %d", __func__, lookupflags));
1604
1605         INP_HASH_LOCK_ASSERT(pcbinfo);
1606
1607         if ((lookupflags & INPLOOKUP_WILDCARD) == 0) {
1608                 struct inpcbhead *head;
1609                 /*
1610                  * Look for an unconnected (wildcard foreign addr) PCB that
1611                  * matches the local address and port we're looking for.
1612                  */
1613                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
1614                     0, pcbinfo->ipi_hashmask)];
1615                 LIST_FOREACH(inp, head, inp_hash) {
1616 #ifdef INET6
1617                         /* XXX inp locking */
1618                         if ((inp->inp_vflag & INP_IPV4) == 0)
1619                                 continue;
1620 #endif
1621                         if (inp->inp_faddr.s_addr == INADDR_ANY &&
1622                             inp->inp_laddr.s_addr == laddr.s_addr &&
1623                             inp->inp_lport == lport) {
1624                                 /*
1625                                  * Found?
1626                                  */
1627                                 if (cred == NULL ||
1628                                     prison_equal_ip4(cred->cr_prison,
1629                                         inp->inp_cred->cr_prison))
1630                                         return (inp);
1631                         }
1632                 }
1633                 /*
1634                  * Not found.
1635                  */
1636                 return (NULL);
1637         } else {
1638                 struct inpcbporthead *porthash;
1639                 struct inpcbport *phd;
1640                 struct inpcb *match = NULL;
1641                 /*
1642                  * Best fit PCB lookup.
1643                  *
1644                  * First see if this local port is in use by looking on the
1645                  * port hash list.
1646                  */
1647                 porthash = &pcbinfo->ipi_porthashbase[INP_PCBPORTHASH(lport,
1648                     pcbinfo->ipi_porthashmask)];
1649                 LIST_FOREACH(phd, porthash, phd_hash) {
1650                         if (phd->phd_port == lport)
1651                                 break;
1652                 }
1653                 if (phd != NULL) {
1654                         /*
1655                          * Port is in use by one or more PCBs. Look for best
1656                          * fit.
1657                          */
1658                         LIST_FOREACH(inp, &phd->phd_pcblist, inp_portlist) {
1659                                 wildcard = 0;
1660                                 if (cred != NULL &&
1661                                     !prison_equal_ip4(inp->inp_cred->cr_prison,
1662                                         cred->cr_prison))
1663                                         continue;
1664 #ifdef INET6
1665                                 /* XXX inp locking */
1666                                 if ((inp->inp_vflag & INP_IPV4) == 0)
1667                                         continue;
1668                                 /*
1669                                  * We never select the PCB that has
1670                                  * INP_IPV6 flag and is bound to :: if
1671                                  * we have another PCB which is bound
1672                                  * to 0.0.0.0.  If a PCB has the
1673                                  * INP_IPV6 flag, then we set its cost
1674                                  * higher than IPv4 only PCBs.
1675                                  *
1676                                  * Note that the case only happens
1677                                  * when a socket is bound to ::, under
1678                                  * the condition that the use of the
1679                                  * mapped address is allowed.
1680                                  */
1681                                 if ((inp->inp_vflag & INP_IPV6) != 0)
1682                                         wildcard += INP_LOOKUP_MAPPED_PCB_COST;
1683 #endif
1684                                 if (inp->inp_faddr.s_addr != INADDR_ANY)
1685                                         wildcard++;
1686                                 if (inp->inp_laddr.s_addr != INADDR_ANY) {
1687                                         if (laddr.s_addr == INADDR_ANY)
1688                                                 wildcard++;
1689                                         else if (inp->inp_laddr.s_addr != laddr.s_addr)
1690                                                 continue;
1691                                 } else {
1692                                         if (laddr.s_addr != INADDR_ANY)
1693                                                 wildcard++;
1694                                 }
1695                                 if (wildcard < matchwild) {
1696                                         match = inp;
1697                                         matchwild = wildcard;
1698                                         if (matchwild == 0)
1699                                                 break;
1700                                 }
1701                         }
1702                 }
1703                 return (match);
1704         }
1705 }
1706 #undef INP_LOOKUP_MAPPED_PCB_COST
1707
1708 #ifdef PCBGROUP
1709 /*
1710  * Lookup PCB in hash list, using pcbgroup tables.
1711  */
1712 static struct inpcb *
1713 in_pcblookup_group(struct inpcbinfo *pcbinfo, struct inpcbgroup *pcbgroup,
1714     struct in_addr faddr, u_int fport_arg, struct in_addr laddr,
1715     u_int lport_arg, int lookupflags, struct ifnet *ifp)
1716 {
1717         struct inpcbhead *head;
1718         struct inpcb *inp, *tmpinp;
1719         u_short fport = fport_arg, lport = lport_arg;
1720         bool locked;
1721
1722         /*
1723          * First look for an exact match.
1724          */
1725         tmpinp = NULL;
1726         INP_GROUP_LOCK(pcbgroup);
1727         head = &pcbgroup->ipg_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1728             pcbgroup->ipg_hashmask)];
1729         LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1730 #ifdef INET6
1731                 /* XXX inp locking */
1732                 if ((inp->inp_vflag & INP_IPV4) == 0)
1733                         continue;
1734 #endif
1735                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1736                     inp->inp_laddr.s_addr == laddr.s_addr &&
1737                     inp->inp_fport == fport &&
1738                     inp->inp_lport == lport) {
1739                         /*
1740                          * XXX We should be able to directly return
1741                          * the inp here, without any checks.
1742                          * Well unless both bound with SO_REUSEPORT?
1743                          */
1744                         if (prison_flag(inp->inp_cred, PR_IP4))
1745                                 goto found;
1746                         if (tmpinp == NULL)
1747                                 tmpinp = inp;
1748                 }
1749         }
1750         if (tmpinp != NULL) {
1751                 inp = tmpinp;
1752                 goto found;
1753         }
1754
1755 #ifdef  RSS
1756         /*
1757          * For incoming connections, we may wish to do a wildcard
1758          * match for an RSS-local socket.
1759          */
1760         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1761                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1762 #ifdef INET6
1763                 struct inpcb *local_wild_mapped = NULL;
1764 #endif
1765                 struct inpcb *jail_wild = NULL;
1766                 struct inpcbhead *head;
1767                 int injail;
1768
1769                 /*
1770                  * Order of socket selection - we always prefer jails.
1771                  *      1. jailed, non-wild.
1772                  *      2. jailed, wild.
1773                  *      3. non-jailed, non-wild.
1774                  *      4. non-jailed, wild.
1775                  */
1776
1777                 head = &pcbgroup->ipg_hashbase[INP_PCBHASH(INADDR_ANY,
1778                     lport, 0, pcbgroup->ipg_hashmask)];
1779                 LIST_FOREACH(inp, head, inp_pcbgrouphash) {
1780 #ifdef INET6
1781                         /* XXX inp locking */
1782                         if ((inp->inp_vflag & INP_IPV4) == 0)
1783                                 continue;
1784 #endif
1785                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
1786                             inp->inp_lport != lport)
1787                                 continue;
1788
1789                         injail = prison_flag(inp->inp_cred, PR_IP4);
1790                         if (injail) {
1791                                 if (prison_check_ip4(inp->inp_cred,
1792                                     &laddr) != 0)
1793                                         continue;
1794                         } else {
1795                                 if (local_exact != NULL)
1796                                         continue;
1797                         }
1798
1799                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
1800                                 if (injail)
1801                                         goto found;
1802                                 else
1803                                         local_exact = inp;
1804                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1805 #ifdef INET6
1806                                 /* XXX inp locking, NULL check */
1807                                 if (inp->inp_vflag & INP_IPV6PROTO)
1808                                         local_wild_mapped = inp;
1809                                 else
1810 #endif
1811                                         if (injail)
1812                                                 jail_wild = inp;
1813                                         else
1814                                                 local_wild = inp;
1815                         }
1816                 } /* LIST_FOREACH */
1817
1818                 inp = jail_wild;
1819                 if (inp == NULL)
1820                         inp = local_exact;
1821                 if (inp == NULL)
1822                         inp = local_wild;
1823 #ifdef INET6
1824                 if (inp == NULL)
1825                         inp = local_wild_mapped;
1826 #endif
1827                 if (inp != NULL)
1828                         goto found;
1829         }
1830 #endif
1831
1832         /*
1833          * Then look for a wildcard match, if requested.
1834          */
1835         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1836                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1837 #ifdef INET6
1838                 struct inpcb *local_wild_mapped = NULL;
1839 #endif
1840                 struct inpcb *jail_wild = NULL;
1841                 struct inpcbhead *head;
1842                 int injail;
1843
1844                 /*
1845                  * Order of socket selection - we always prefer jails.
1846                  *      1. jailed, non-wild.
1847                  *      2. jailed, wild.
1848                  *      3. non-jailed, non-wild.
1849                  *      4. non-jailed, wild.
1850                  */
1851                 head = &pcbinfo->ipi_wildbase[INP_PCBHASH(INADDR_ANY, lport,
1852                     0, pcbinfo->ipi_wildmask)];
1853                 LIST_FOREACH(inp, head, inp_pcbgroup_wild) {
1854 #ifdef INET6
1855                         /* XXX inp locking */
1856                         if ((inp->inp_vflag & INP_IPV4) == 0)
1857                                 continue;
1858 #endif
1859                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
1860                             inp->inp_lport != lport)
1861                                 continue;
1862
1863                         injail = prison_flag(inp->inp_cred, PR_IP4);
1864                         if (injail) {
1865                                 if (prison_check_ip4(inp->inp_cred,
1866                                     &laddr) != 0)
1867                                         continue;
1868                         } else {
1869                                 if (local_exact != NULL)
1870                                         continue;
1871                         }
1872
1873                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
1874                                 if (injail)
1875                                         goto found;
1876                                 else
1877                                         local_exact = inp;
1878                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
1879 #ifdef INET6
1880                                 /* XXX inp locking, NULL check */
1881                                 if (inp->inp_vflag & INP_IPV6PROTO)
1882                                         local_wild_mapped = inp;
1883                                 else
1884 #endif
1885                                         if (injail)
1886                                                 jail_wild = inp;
1887                                         else
1888                                                 local_wild = inp;
1889                         }
1890                 } /* LIST_FOREACH */
1891                 inp = jail_wild;
1892                 if (inp == NULL)
1893                         inp = local_exact;
1894                 if (inp == NULL)
1895                         inp = local_wild;
1896 #ifdef INET6
1897                 if (inp == NULL)
1898                         inp = local_wild_mapped;
1899 #endif
1900                 if (inp != NULL)
1901                         goto found;
1902         } /* if (lookupflags & INPLOOKUP_WILDCARD) */
1903         INP_GROUP_UNLOCK(pcbgroup);
1904         return (NULL);
1905
1906 found:
1907         if (lookupflags & INPLOOKUP_WLOCKPCB)
1908                 locked = INP_TRY_WLOCK(inp);
1909         else if (lookupflags & INPLOOKUP_RLOCKPCB)
1910                 locked = INP_TRY_RLOCK(inp);
1911         else
1912                 panic("%s: locking bug", __func__);
1913         if (!locked)
1914                 in_pcbref(inp);
1915         INP_GROUP_UNLOCK(pcbgroup);
1916         if (!locked) {
1917                 if (lookupflags & INPLOOKUP_WLOCKPCB) {
1918                         INP_WLOCK(inp);
1919                         if (in_pcbrele_wlocked(inp))
1920                                 return (NULL);
1921                 } else {
1922                         INP_RLOCK(inp);
1923                         if (in_pcbrele_rlocked(inp))
1924                                 return (NULL);
1925                 }
1926         }
1927 #ifdef INVARIANTS
1928         if (lookupflags & INPLOOKUP_WLOCKPCB)
1929                 INP_WLOCK_ASSERT(inp);
1930         else
1931                 INP_RLOCK_ASSERT(inp);
1932 #endif
1933         return (inp);
1934 }
1935 #endif /* PCBGROUP */
1936
1937 /*
1938  * Lookup PCB in hash list, using pcbinfo tables.  This variation assumes
1939  * that the caller has locked the hash list, and will not perform any further
1940  * locking or reference operations on either the hash list or the connection.
1941  */
1942 static struct inpcb *
1943 in_pcblookup_hash_locked(struct inpcbinfo *pcbinfo, struct in_addr faddr,
1944     u_int fport_arg, struct in_addr laddr, u_int lport_arg, int lookupflags,
1945     struct ifnet *ifp)
1946 {
1947         struct inpcbhead *head;
1948         struct inpcb *inp, *tmpinp;
1949         u_short fport = fport_arg, lport = lport_arg;
1950
1951         KASSERT((lookupflags & ~(INPLOOKUP_WILDCARD)) == 0,
1952             ("%s: invalid lookup flags %d", __func__, lookupflags));
1953
1954         INP_HASH_LOCK_ASSERT(pcbinfo);
1955
1956         /*
1957          * First look for an exact match.
1958          */
1959         tmpinp = NULL;
1960         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(faddr.s_addr, lport, fport,
1961             pcbinfo->ipi_hashmask)];
1962         LIST_FOREACH(inp, head, inp_hash) {
1963 #ifdef INET6
1964                 /* XXX inp locking */
1965                 if ((inp->inp_vflag & INP_IPV4) == 0)
1966                         continue;
1967 #endif
1968                 if (inp->inp_faddr.s_addr == faddr.s_addr &&
1969                     inp->inp_laddr.s_addr == laddr.s_addr &&
1970                     inp->inp_fport == fport &&
1971                     inp->inp_lport == lport) {
1972                         /*
1973                          * XXX We should be able to directly return
1974                          * the inp here, without any checks.
1975                          * Well unless both bound with SO_REUSEPORT?
1976                          */
1977                         if (prison_flag(inp->inp_cred, PR_IP4))
1978                                 return (inp);
1979                         if (tmpinp == NULL)
1980                                 tmpinp = inp;
1981                 }
1982         }
1983         if (tmpinp != NULL)
1984                 return (tmpinp);
1985
1986         /*
1987          * Then look for a wildcard match, if requested.
1988          */
1989         if ((lookupflags & INPLOOKUP_WILDCARD) != 0) {
1990                 struct inpcb *local_wild = NULL, *local_exact = NULL;
1991 #ifdef INET6
1992                 struct inpcb *local_wild_mapped = NULL;
1993 #endif
1994                 struct inpcb *jail_wild = NULL;
1995                 int injail;
1996
1997                 /*
1998                  * Order of socket selection - we always prefer jails.
1999                  *      1. jailed, non-wild.
2000                  *      2. jailed, wild.
2001                  *      3. non-jailed, non-wild.
2002                  *      4. non-jailed, wild.
2003                  */
2004
2005                 head = &pcbinfo->ipi_hashbase[INP_PCBHASH(INADDR_ANY, lport,
2006                     0, pcbinfo->ipi_hashmask)];
2007                 LIST_FOREACH(inp, head, inp_hash) {
2008 #ifdef INET6
2009                         /* XXX inp locking */
2010                         if ((inp->inp_vflag & INP_IPV4) == 0)
2011                                 continue;
2012 #endif
2013                         if (inp->inp_faddr.s_addr != INADDR_ANY ||
2014                             inp->inp_lport != lport)
2015                                 continue;
2016
2017                         injail = prison_flag(inp->inp_cred, PR_IP4);
2018                         if (injail) {
2019                                 if (prison_check_ip4(inp->inp_cred,
2020                                     &laddr) != 0)
2021                                         continue;
2022                         } else {
2023                                 if (local_exact != NULL)
2024                                         continue;
2025                         }
2026
2027                         if (inp->inp_laddr.s_addr == laddr.s_addr) {
2028                                 if (injail)
2029                                         return (inp);
2030                                 else
2031                                         local_exact = inp;
2032                         } else if (inp->inp_laddr.s_addr == INADDR_ANY) {
2033 #ifdef INET6
2034                                 /* XXX inp locking, NULL check */
2035                                 if (inp->inp_vflag & INP_IPV6PROTO)
2036                                         local_wild_mapped = inp;
2037                                 else
2038 #endif
2039                                         if (injail)
2040                                                 jail_wild = inp;
2041                                         else
2042                                                 local_wild = inp;
2043                         }
2044                 } /* LIST_FOREACH */
2045                 if (jail_wild != NULL)
2046                         return (jail_wild);
2047                 if (local_exact != NULL)
2048                         return (local_exact);
2049                 if (local_wild != NULL)
2050                         return (local_wild);
2051 #ifdef INET6
2052                 if (local_wild_mapped != NULL)
2053                         return (local_wild_mapped);
2054 #endif
2055         } /* if ((lookupflags & INPLOOKUP_WILDCARD) != 0) */
2056
2057         return (NULL);
2058 }
2059
2060 /*
2061  * Lookup PCB in hash list, using pcbinfo tables.  This variation locks the
2062  * hash list lock, and will return the inpcb locked (i.e., requires
2063  * INPLOOKUP_LOCKPCB).
2064  */
2065 static struct inpcb *
2066 in_pcblookup_hash(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2067     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2068     struct ifnet *ifp)
2069 {
2070         struct inpcb *inp;
2071         bool locked;
2072
2073         INP_HASH_RLOCK(pcbinfo);
2074         inp = in_pcblookup_hash_locked(pcbinfo, faddr, fport, laddr, lport,
2075             (lookupflags & ~(INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)), ifp);
2076         if (inp != NULL) {
2077                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2078                         locked = INP_TRY_WLOCK(inp);
2079                 else if (lookupflags & INPLOOKUP_RLOCKPCB)
2080                         locked = INP_TRY_RLOCK(inp);
2081                 else
2082                         panic("%s: locking bug", __func__);
2083                 if (!locked)
2084                         in_pcbref(inp);
2085                 INP_HASH_RUNLOCK(pcbinfo);
2086                 if (!locked) {
2087                         if (lookupflags & INPLOOKUP_WLOCKPCB) {
2088                                 INP_WLOCK(inp);
2089                                 if (in_pcbrele_wlocked(inp))
2090                                         return (NULL);
2091                         } else {
2092                                 INP_RLOCK(inp);
2093                                 if (in_pcbrele_rlocked(inp))
2094                                         return (NULL);
2095                         }
2096                 }
2097 #ifdef INVARIANTS
2098                 if (lookupflags & INPLOOKUP_WLOCKPCB)
2099                         INP_WLOCK_ASSERT(inp);
2100                 else
2101                         INP_RLOCK_ASSERT(inp);
2102 #endif
2103         } else
2104                 INP_HASH_RUNLOCK(pcbinfo);
2105         return (inp);
2106 }
2107
2108 /*
2109  * Public inpcb lookup routines, accepting a 4-tuple, and optionally, an mbuf
2110  * from which a pre-calculated hash value may be extracted.
2111  *
2112  * Possibly more of this logic should be in in_pcbgroup.c.
2113  */
2114 struct inpcb *
2115 in_pcblookup(struct inpcbinfo *pcbinfo, struct in_addr faddr, u_int fport,
2116     struct in_addr laddr, u_int lport, int lookupflags, struct ifnet *ifp)
2117 {
2118 #if defined(PCBGROUP) && !defined(RSS)
2119         struct inpcbgroup *pcbgroup;
2120 #endif
2121
2122         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2123             ("%s: invalid lookup flags %d", __func__, lookupflags));
2124         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2125             ("%s: LOCKPCB not set", __func__));
2126
2127         /*
2128          * When not using RSS, use connection groups in preference to the
2129          * reservation table when looking up 4-tuples.  When using RSS, just
2130          * use the reservation table, due to the cost of the Toeplitz hash
2131          * in software.
2132          *
2133          * XXXRW: This policy belongs in the pcbgroup code, as in principle
2134          * we could be doing RSS with a non-Toeplitz hash that is affordable
2135          * in software.
2136          */
2137 #if defined(PCBGROUP) && !defined(RSS)
2138         if (in_pcbgroup_enabled(pcbinfo)) {
2139                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2140                     fport);
2141                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2142                     laddr, lport, lookupflags, ifp));
2143         }
2144 #endif
2145         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2146             lookupflags, ifp));
2147 }
2148
2149 struct inpcb *
2150 in_pcblookup_mbuf(struct inpcbinfo *pcbinfo, struct in_addr faddr,
2151     u_int fport, struct in_addr laddr, u_int lport, int lookupflags,
2152     struct ifnet *ifp, struct mbuf *m)
2153 {
2154 #ifdef PCBGROUP
2155         struct inpcbgroup *pcbgroup;
2156 #endif
2157
2158         KASSERT((lookupflags & ~INPLOOKUP_MASK) == 0,
2159             ("%s: invalid lookup flags %d", __func__, lookupflags));
2160         KASSERT((lookupflags & (INPLOOKUP_RLOCKPCB | INPLOOKUP_WLOCKPCB)) != 0,
2161             ("%s: LOCKPCB not set", __func__));
2162
2163 #ifdef PCBGROUP
2164         /*
2165          * If we can use a hardware-generated hash to look up the connection
2166          * group, use that connection group to find the inpcb.  Otherwise
2167          * fall back on a software hash -- or the reservation table if we're
2168          * using RSS.
2169          *
2170          * XXXRW: As above, that policy belongs in the pcbgroup code.
2171          */
2172         if (in_pcbgroup_enabled(pcbinfo) &&
2173             !(M_HASHTYPE_TEST(m, M_HASHTYPE_NONE))) {
2174                 pcbgroup = in_pcbgroup_byhash(pcbinfo, M_HASHTYPE_GET(m),
2175                     m->m_pkthdr.flowid);
2176                 if (pcbgroup != NULL)
2177                         return (in_pcblookup_group(pcbinfo, pcbgroup, faddr,
2178                             fport, laddr, lport, lookupflags, ifp));
2179 #ifndef RSS
2180                 pcbgroup = in_pcbgroup_bytuple(pcbinfo, laddr, lport, faddr,
2181                     fport);
2182                 return (in_pcblookup_group(pcbinfo, pcbgroup, faddr, fport,
2183                     laddr, lport, lookupflags, ifp));
2184 #endif
2185         }
2186 #endif
2187         return (in_pcblookup_hash(pcbinfo, faddr, fport, laddr, lport,
2188             lookupflags, ifp));
2189 }
2190 #endif /* INET */
2191
2192 /*
2193  * Insert PCB onto various hash lists.
2194  */
2195 static int
2196 in_pcbinshash_internal(struct inpcb *inp, int do_pcbgroup_update)
2197 {
2198         struct inpcbhead *pcbhash;
2199         struct inpcbporthead *pcbporthash;
2200         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2201         struct inpcbport *phd;
2202         u_int32_t hashkey_faddr;
2203
2204         INP_WLOCK_ASSERT(inp);
2205         INP_HASH_WLOCK_ASSERT(pcbinfo);
2206
2207         KASSERT((inp->inp_flags & INP_INHASHLIST) == 0,
2208             ("in_pcbinshash: INP_INHASHLIST"));
2209
2210 #ifdef INET6
2211         if (inp->inp_vflag & INP_IPV6)
2212                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2213         else
2214 #endif
2215         hashkey_faddr = inp->inp_faddr.s_addr;
2216
2217         pcbhash = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2218                  inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2219
2220         pcbporthash = &pcbinfo->ipi_porthashbase[
2221             INP_PCBPORTHASH(inp->inp_lport, pcbinfo->ipi_porthashmask)];
2222
2223         /*
2224          * Go through port list and look for a head for this lport.
2225          */
2226         LIST_FOREACH(phd, pcbporthash, phd_hash) {
2227                 if (phd->phd_port == inp->inp_lport)
2228                         break;
2229         }
2230         /*
2231          * If none exists, malloc one and tack it on.
2232          */
2233         if (phd == NULL) {
2234                 phd = malloc(sizeof(struct inpcbport), M_PCB, M_NOWAIT);
2235                 if (phd == NULL) {
2236                         return (ENOBUFS); /* XXX */
2237                 }
2238                 phd->phd_port = inp->inp_lport;
2239                 LIST_INIT(&phd->phd_pcblist);
2240                 LIST_INSERT_HEAD(pcbporthash, phd, phd_hash);
2241         }
2242         inp->inp_phd = phd;
2243         LIST_INSERT_HEAD(&phd->phd_pcblist, inp, inp_portlist);
2244         LIST_INSERT_HEAD(pcbhash, inp, inp_hash);
2245         inp->inp_flags |= INP_INHASHLIST;
2246 #ifdef PCBGROUP
2247         if (do_pcbgroup_update)
2248                 in_pcbgroup_update(inp);
2249 #endif
2250         return (0);
2251 }
2252
2253 /*
2254  * For now, there are two public interfaces to insert an inpcb into the hash
2255  * lists -- one that does update pcbgroups, and one that doesn't.  The latter
2256  * is used only in the TCP syncache, where in_pcbinshash is called before the
2257  * full 4-tuple is set for the inpcb, and we don't want to install in the
2258  * pcbgroup until later.
2259  *
2260  * XXXRW: This seems like a misfeature.  in_pcbinshash should always update
2261  * connection groups, and partially initialised inpcbs should not be exposed
2262  * to either reservation hash tables or pcbgroups.
2263  */
2264 int
2265 in_pcbinshash(struct inpcb *inp)
2266 {
2267
2268         return (in_pcbinshash_internal(inp, 1));
2269 }
2270
2271 int
2272 in_pcbinshash_nopcbgroup(struct inpcb *inp)
2273 {
2274
2275         return (in_pcbinshash_internal(inp, 0));
2276 }
2277
2278 /*
2279  * Move PCB to the proper hash bucket when { faddr, fport } have  been
2280  * changed. NOTE: This does not handle the case of the lport changing (the
2281  * hashed port list would have to be updated as well), so the lport must
2282  * not change after in_pcbinshash() has been called.
2283  */
2284 void
2285 in_pcbrehash_mbuf(struct inpcb *inp, struct mbuf *m)
2286 {
2287         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2288         struct inpcbhead *head;
2289         u_int32_t hashkey_faddr;
2290
2291         INP_WLOCK_ASSERT(inp);
2292         INP_HASH_WLOCK_ASSERT(pcbinfo);
2293
2294         KASSERT(inp->inp_flags & INP_INHASHLIST,
2295             ("in_pcbrehash: !INP_INHASHLIST"));
2296
2297 #ifdef INET6
2298         if (inp->inp_vflag & INP_IPV6)
2299                 hashkey_faddr = INP6_PCBHASHKEY(&inp->in6p_faddr);
2300         else
2301 #endif
2302         hashkey_faddr = inp->inp_faddr.s_addr;
2303
2304         head = &pcbinfo->ipi_hashbase[INP_PCBHASH(hashkey_faddr,
2305                 inp->inp_lport, inp->inp_fport, pcbinfo->ipi_hashmask)];
2306
2307         LIST_REMOVE(inp, inp_hash);
2308         LIST_INSERT_HEAD(head, inp, inp_hash);
2309
2310 #ifdef PCBGROUP
2311         if (m != NULL)
2312                 in_pcbgroup_update_mbuf(inp, m);
2313         else
2314                 in_pcbgroup_update(inp);
2315 #endif
2316 }
2317
2318 void
2319 in_pcbrehash(struct inpcb *inp)
2320 {
2321
2322         in_pcbrehash_mbuf(inp, NULL);
2323 }
2324
2325 /*
2326  * Remove PCB from various lists.
2327  */
2328 static void
2329 in_pcbremlists(struct inpcb *inp)
2330 {
2331         struct inpcbinfo *pcbinfo = inp->inp_pcbinfo;
2332
2333 #ifdef INVARIANTS
2334         if (pcbinfo == &V_tcbinfo) {
2335                 INP_INFO_RLOCK_ASSERT(pcbinfo);
2336         } else {
2337                 INP_INFO_WLOCK_ASSERT(pcbinfo);
2338         }
2339 #endif
2340
2341         INP_WLOCK_ASSERT(inp);
2342         INP_LIST_WLOCK_ASSERT(pcbinfo);
2343
2344         inp->inp_gencnt = ++pcbinfo->ipi_gencnt;
2345         if (inp->inp_flags & INP_INHASHLIST) {
2346                 struct inpcbport *phd = inp->inp_phd;
2347
2348                 INP_HASH_WLOCK(pcbinfo);
2349                 LIST_REMOVE(inp, inp_hash);
2350                 LIST_REMOVE(inp, inp_portlist);
2351                 if (LIST_FIRST(&phd->phd_pcblist) == NULL) {
2352                         LIST_REMOVE(phd, phd_hash);
2353                         free(phd, M_PCB);
2354                 }
2355                 INP_HASH_WUNLOCK(pcbinfo);
2356                 inp->inp_flags &= ~INP_INHASHLIST;
2357         }
2358         LIST_REMOVE(inp, inp_list);
2359         pcbinfo->ipi_count--;
2360 #ifdef PCBGROUP
2361         in_pcbgroup_remove(inp);
2362 #endif
2363 }
2364
2365 /*
2366  * Check for alternatives when higher level complains
2367  * about service problems.  For now, invalidate cached
2368  * routing information.  If the route was created dynamically
2369  * (by a redirect), time to try a default gateway again.
2370  */
2371 void
2372 in_losing(struct inpcb *inp)
2373 {
2374
2375         RO_INVALIDATE_CACHE(&inp->inp_route);
2376         return;
2377 }
2378
2379 /*
2380  * A set label operation has occurred at the socket layer, propagate the
2381  * label change into the in_pcb for the socket.
2382  */
2383 void
2384 in_pcbsosetlabel(struct socket *so)
2385 {
2386 #ifdef MAC
2387         struct inpcb *inp;
2388
2389         inp = sotoinpcb(so);
2390         KASSERT(inp != NULL, ("in_pcbsosetlabel: so->so_pcb == NULL"));
2391
2392         INP_WLOCK(inp);
2393         SOCK_LOCK(so);
2394         mac_inpcb_sosetlabel(so, inp);
2395         SOCK_UNLOCK(so);
2396         INP_WUNLOCK(inp);
2397 #endif
2398 }
2399
2400 /*
2401  * ipport_tick runs once per second, determining if random port allocation
2402  * should be continued.  If more than ipport_randomcps ports have been
2403  * allocated in the last second, then we return to sequential port
2404  * allocation. We return to random allocation only once we drop below
2405  * ipport_randomcps for at least ipport_randomtime seconds.
2406  */
2407 static void
2408 ipport_tick(void *xtp)
2409 {
2410         VNET_ITERATOR_DECL(vnet_iter);
2411
2412         VNET_LIST_RLOCK_NOSLEEP();
2413         VNET_FOREACH(vnet_iter) {
2414                 CURVNET_SET(vnet_iter); /* XXX appease INVARIANTS here */
2415                 if (V_ipport_tcpallocs <=
2416                     V_ipport_tcplastcount + V_ipport_randomcps) {
2417                         if (V_ipport_stoprandom > 0)
2418                                 V_ipport_stoprandom--;
2419                 } else
2420                         V_ipport_stoprandom = V_ipport_randomtime;
2421                 V_ipport_tcplastcount = V_ipport_tcpallocs;
2422                 CURVNET_RESTORE();
2423         }
2424         VNET_LIST_RUNLOCK_NOSLEEP();
2425         callout_reset(&ipport_tick_callout, hz, ipport_tick, NULL);
2426 }
2427
2428 static void
2429 ip_fini(void *xtp)
2430 {
2431
2432         callout_stop(&ipport_tick_callout);
2433 }
2434
2435 /*
2436  * The ipport_callout should start running at about the time we attach the
2437  * inet or inet6 domains.
2438  */
2439 static void
2440 ipport_tick_init(const void *unused __unused)
2441 {
2442
2443         /* Start ipport_tick. */
2444         callout_init(&ipport_tick_callout, 1);
2445         callout_reset(&ipport_tick_callout, 1, ipport_tick, NULL);
2446         EVENTHANDLER_REGISTER(shutdown_pre_sync, ip_fini, NULL,
2447                 SHUTDOWN_PRI_DEFAULT);
2448 }
2449 SYSINIT(ipport_tick_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_MIDDLE,
2450     ipport_tick_init, NULL);
2451
2452 void
2453 inp_wlock(struct inpcb *inp)
2454 {
2455
2456         INP_WLOCK(inp);
2457 }
2458
2459 void
2460 inp_wunlock(struct inpcb *inp)
2461 {
2462
2463         INP_WUNLOCK(inp);
2464 }
2465
2466 void
2467 inp_rlock(struct inpcb *inp)
2468 {
2469
2470         INP_RLOCK(inp);
2471 }
2472
2473 void
2474 inp_runlock(struct inpcb *inp)
2475 {
2476
2477         INP_RUNLOCK(inp);
2478 }
2479
2480 #ifdef INVARIANT_SUPPORT
2481 void
2482 inp_lock_assert(struct inpcb *inp)
2483 {
2484
2485         INP_WLOCK_ASSERT(inp);
2486 }
2487
2488 void
2489 inp_unlock_assert(struct inpcb *inp)
2490 {
2491
2492         INP_UNLOCK_ASSERT(inp);
2493 }
2494 #endif
2495
2496 void
2497 inp_apply_all(void (*func)(struct inpcb *, void *), void *arg)
2498 {
2499         struct inpcb *inp;
2500
2501         INP_INFO_WLOCK(&V_tcbinfo);
2502         LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
2503                 INP_WLOCK(inp);
2504                 func(inp, arg);
2505                 INP_WUNLOCK(inp);
2506         }
2507         INP_INFO_WUNLOCK(&V_tcbinfo);
2508 }
2509
2510 struct socket *
2511 inp_inpcbtosocket(struct inpcb *inp)
2512 {
2513
2514         INP_WLOCK_ASSERT(inp);
2515         return (inp->inp_socket);
2516 }
2517
2518 struct tcpcb *
2519 inp_inpcbtotcpcb(struct inpcb *inp)
2520 {
2521
2522         INP_WLOCK_ASSERT(inp);
2523         return ((struct tcpcb *)inp->inp_ppcb);
2524 }
2525
2526 int
2527 inp_ip_tos_get(const struct inpcb *inp)
2528 {
2529
2530         return (inp->inp_ip_tos);
2531 }
2532
2533 void
2534 inp_ip_tos_set(struct inpcb *inp, int val)
2535 {
2536
2537         inp->inp_ip_tos = val;
2538 }
2539
2540 void
2541 inp_4tuple_get(struct inpcb *inp, uint32_t *laddr, uint16_t *lp,
2542     uint32_t *faddr, uint16_t *fp)
2543 {
2544
2545         INP_LOCK_ASSERT(inp);
2546         *laddr = inp->inp_laddr.s_addr;
2547         *faddr = inp->inp_faddr.s_addr;
2548         *lp = inp->inp_lport;
2549         *fp = inp->inp_fport;
2550 }
2551
2552 struct inpcb *
2553 so_sotoinpcb(struct socket *so)
2554 {
2555
2556         return (sotoinpcb(so));
2557 }
2558
2559 struct tcpcb *
2560 so_sototcpcb(struct socket *so)
2561 {
2562
2563         return (sototcpcb(so));
2564 }
2565
2566 /*
2567  * Create an external-format (``xinpcb'') structure using the information in
2568  * the kernel-format in_pcb structure pointed to by inp.  This is done to
2569  * reduce the spew of irrelevant information over this interface, to isolate
2570  * user code from changes in the kernel structure, and potentially to provide
2571  * information-hiding if we decide that some of this information should be
2572  * hidden from users.
2573  */
2574 void
2575 in_pcbtoxinpcb(const struct inpcb *inp, struct xinpcb *xi)
2576 {
2577
2578         xi->xi_len = sizeof(struct xinpcb);
2579         if (inp->inp_socket)
2580                 sotoxsocket(inp->inp_socket, &xi->xi_socket);
2581         else
2582                 bzero(&xi->xi_socket, sizeof(struct xsocket));
2583         bcopy(&inp->inp_inc, &xi->inp_inc, sizeof(struct in_conninfo));
2584         xi->inp_gencnt = inp->inp_gencnt;
2585         xi->inp_ppcb = inp->inp_ppcb;
2586         xi->inp_flow = inp->inp_flow;
2587         xi->inp_flowid = inp->inp_flowid;
2588         xi->inp_flowtype = inp->inp_flowtype;
2589         xi->inp_flags = inp->inp_flags;
2590         xi->inp_flags2 = inp->inp_flags2;
2591         xi->inp_rss_listen_bucket = inp->inp_rss_listen_bucket;
2592         xi->in6p_cksum = inp->in6p_cksum;
2593         xi->in6p_hops = inp->in6p_hops;
2594         xi->inp_ip_tos = inp->inp_ip_tos;
2595         xi->inp_vflag = inp->inp_vflag;
2596         xi->inp_ip_ttl = inp->inp_ip_ttl;
2597         xi->inp_ip_p = inp->inp_ip_p;
2598         xi->inp_ip_minttl = inp->inp_ip_minttl;
2599 }
2600
2601 #ifdef DDB
2602 static void
2603 db_print_indent(int indent)
2604 {
2605         int i;
2606
2607         for (i = 0; i < indent; i++)
2608                 db_printf(" ");
2609 }
2610
2611 static void
2612 db_print_inconninfo(struct in_conninfo *inc, const char *name, int indent)
2613 {
2614         char faddr_str[48], laddr_str[48];
2615
2616         db_print_indent(indent);
2617         db_printf("%s at %p\n", name, inc);
2618
2619         indent += 2;
2620
2621 #ifdef INET6
2622         if (inc->inc_flags & INC_ISIPV6) {
2623                 /* IPv6. */
2624                 ip6_sprintf(laddr_str, &inc->inc6_laddr);
2625                 ip6_sprintf(faddr_str, &inc->inc6_faddr);
2626         } else
2627 #endif
2628         {
2629                 /* IPv4. */
2630                 inet_ntoa_r(inc->inc_laddr, laddr_str);
2631                 inet_ntoa_r(inc->inc_faddr, faddr_str);
2632         }
2633         db_print_indent(indent);
2634         db_printf("inc_laddr %s   inc_lport %u\n", laddr_str,
2635             ntohs(inc->inc_lport));
2636         db_print_indent(indent);
2637         db_printf("inc_faddr %s   inc_fport %u\n", faddr_str,
2638             ntohs(inc->inc_fport));
2639 }
2640
2641 static void
2642 db_print_inpflags(int inp_flags)
2643 {
2644         int comma;
2645
2646         comma = 0;
2647         if (inp_flags & INP_RECVOPTS) {
2648                 db_printf("%sINP_RECVOPTS", comma ? ", " : "");
2649                 comma = 1;
2650         }
2651         if (inp_flags & INP_RECVRETOPTS) {
2652                 db_printf("%sINP_RECVRETOPTS", comma ? ", " : "");
2653                 comma = 1;
2654         }
2655         if (inp_flags & INP_RECVDSTADDR) {
2656                 db_printf("%sINP_RECVDSTADDR", comma ? ", " : "");
2657                 comma = 1;
2658         }
2659         if (inp_flags & INP_ORIGDSTADDR) {
2660                 db_printf("%sINP_ORIGDSTADDR", comma ? ", " : "");
2661                 comma = 1;
2662         }
2663         if (inp_flags & INP_HDRINCL) {
2664                 db_printf("%sINP_HDRINCL", comma ? ", " : "");
2665                 comma = 1;
2666         }
2667         if (inp_flags & INP_HIGHPORT) {
2668                 db_printf("%sINP_HIGHPORT", comma ? ", " : "");
2669                 comma = 1;
2670         }
2671         if (inp_flags & INP_LOWPORT) {
2672                 db_printf("%sINP_LOWPORT", comma ? ", " : "");
2673                 comma = 1;
2674         }
2675         if (inp_flags & INP_ANONPORT) {
2676                 db_printf("%sINP_ANONPORT", comma ? ", " : "");
2677                 comma = 1;
2678         }
2679         if (inp_flags & INP_RECVIF) {
2680                 db_printf("%sINP_RECVIF", comma ? ", " : "");
2681                 comma = 1;
2682         }
2683         if (inp_flags & INP_MTUDISC) {
2684                 db_printf("%sINP_MTUDISC", comma ? ", " : "");
2685                 comma = 1;
2686         }
2687         if (inp_flags & INP_RECVTTL) {
2688                 db_printf("%sINP_RECVTTL", comma ? ", " : "");
2689                 comma = 1;
2690         }
2691         if (inp_flags & INP_DONTFRAG) {
2692                 db_printf("%sINP_DONTFRAG", comma ? ", " : "");
2693                 comma = 1;
2694         }
2695         if (inp_flags & INP_RECVTOS) {
2696                 db_printf("%sINP_RECVTOS", comma ? ", " : "");
2697                 comma = 1;
2698         }
2699         if (inp_flags & IN6P_IPV6_V6ONLY) {
2700                 db_printf("%sIN6P_IPV6_V6ONLY", comma ? ", " : "");
2701                 comma = 1;
2702         }
2703         if (inp_flags & IN6P_PKTINFO) {
2704                 db_printf("%sIN6P_PKTINFO", comma ? ", " : "");
2705                 comma = 1;
2706         }
2707         if (inp_flags & IN6P_HOPLIMIT) {
2708                 db_printf("%sIN6P_HOPLIMIT", comma ? ", " : "");
2709                 comma = 1;
2710         }
2711         if (inp_flags & IN6P_HOPOPTS) {
2712                 db_printf("%sIN6P_HOPOPTS", comma ? ", " : "");
2713                 comma = 1;
2714         }
2715         if (inp_flags & IN6P_DSTOPTS) {
2716                 db_printf("%sIN6P_DSTOPTS", comma ? ", " : "");
2717                 comma = 1;
2718         }
2719         if (inp_flags & IN6P_RTHDR) {
2720                 db_printf("%sIN6P_RTHDR", comma ? ", " : "");
2721                 comma = 1;
2722         }
2723         if (inp_flags & IN6P_RTHDRDSTOPTS) {
2724                 db_printf("%sIN6P_RTHDRDSTOPTS", comma ? ", " : "");
2725                 comma = 1;
2726         }
2727         if (inp_flags & IN6P_TCLASS) {
2728                 db_printf("%sIN6P_TCLASS", comma ? ", " : "");
2729                 comma = 1;
2730         }
2731         if (inp_flags & IN6P_AUTOFLOWLABEL) {
2732                 db_printf("%sIN6P_AUTOFLOWLABEL", comma ? ", " : "");
2733                 comma = 1;
2734         }
2735         if (inp_flags & INP_TIMEWAIT) {
2736                 db_printf("%sINP_TIMEWAIT", comma ? ", " : "");
2737                 comma  = 1;
2738         }
2739         if (inp_flags & INP_ONESBCAST) {
2740                 db_printf("%sINP_ONESBCAST", comma ? ", " : "");
2741                 comma  = 1;
2742         }
2743         if (inp_flags & INP_DROPPED) {
2744                 db_printf("%sINP_DROPPED", comma ? ", " : "");
2745                 comma  = 1;
2746         }
2747         if (inp_flags & INP_SOCKREF) {
2748                 db_printf("%sINP_SOCKREF", comma ? ", " : "");
2749                 comma  = 1;
2750         }
2751         if (inp_flags & IN6P_RFC2292) {
2752                 db_printf("%sIN6P_RFC2292", comma ? ", " : "");
2753                 comma = 1;
2754         }
2755         if (inp_flags & IN6P_MTU) {
2756                 db_printf("IN6P_MTU%s", comma ? ", " : "");
2757                 comma = 1;
2758         }
2759 }
2760
2761 static void
2762 db_print_inpvflag(u_char inp_vflag)
2763 {
2764         int comma;
2765
2766         comma = 0;
2767         if (inp_vflag & INP_IPV4) {
2768                 db_printf("%sINP_IPV4", comma ? ", " : "");
2769                 comma  = 1;
2770         }
2771         if (inp_vflag & INP_IPV6) {
2772                 db_printf("%sINP_IPV6", comma ? ", " : "");
2773                 comma  = 1;
2774         }
2775         if (inp_vflag & INP_IPV6PROTO) {
2776                 db_printf("%sINP_IPV6PROTO", comma ? ", " : "");
2777                 comma  = 1;
2778         }
2779 }
2780
2781 static void
2782 db_print_inpcb(struct inpcb *inp, const char *name, int indent)
2783 {
2784
2785         db_print_indent(indent);
2786         db_printf("%s at %p\n", name, inp);
2787
2788         indent += 2;
2789
2790         db_print_indent(indent);
2791         db_printf("inp_flow: 0x%x\n", inp->inp_flow);
2792
2793         db_print_inconninfo(&inp->inp_inc, "inp_conninfo", indent);
2794
2795         db_print_indent(indent);
2796         db_printf("inp_ppcb: %p   inp_pcbinfo: %p   inp_socket: %p\n",
2797             inp->inp_ppcb, inp->inp_pcbinfo, inp->inp_socket);
2798
2799         db_print_indent(indent);
2800         db_printf("inp_label: %p   inp_flags: 0x%x (",
2801            inp->inp_label, inp->inp_flags);
2802         db_print_inpflags(inp->inp_flags);
2803         db_printf(")\n");
2804
2805         db_print_indent(indent);
2806         db_printf("inp_sp: %p   inp_vflag: 0x%x (", inp->inp_sp,
2807             inp->inp_vflag);
2808         db_print_inpvflag(inp->inp_vflag);
2809         db_printf(")\n");
2810
2811         db_print_indent(indent);
2812         db_printf("inp_ip_ttl: %d   inp_ip_p: %d   inp_ip_minttl: %d\n",
2813             inp->inp_ip_ttl, inp->inp_ip_p, inp->inp_ip_minttl);
2814
2815         db_print_indent(indent);
2816 #ifdef INET6
2817         if (inp->inp_vflag & INP_IPV6) {
2818                 db_printf("in6p_options: %p   in6p_outputopts: %p   "
2819                     "in6p_moptions: %p\n", inp->in6p_options,
2820                     inp->in6p_outputopts, inp->in6p_moptions);
2821                 db_printf("in6p_icmp6filt: %p   in6p_cksum %d   "
2822                     "in6p_hops %u\n", inp->in6p_icmp6filt, inp->in6p_cksum,
2823                     inp->in6p_hops);
2824         } else
2825 #endif
2826         {
2827                 db_printf("inp_ip_tos: %d   inp_ip_options: %p   "
2828                     "inp_ip_moptions: %p\n", inp->inp_ip_tos,
2829                     inp->inp_options, inp->inp_moptions);
2830         }
2831
2832         db_print_indent(indent);
2833         db_printf("inp_phd: %p   inp_gencnt: %ju\n", inp->inp_phd,
2834             (uintmax_t)inp->inp_gencnt);
2835 }
2836
2837 DB_SHOW_COMMAND(inpcb, db_show_inpcb)
2838 {
2839         struct inpcb *inp;
2840
2841         if (!have_addr) {
2842                 db_printf("usage: show inpcb <addr>\n");
2843                 return;
2844         }
2845         inp = (struct inpcb *)addr;
2846
2847         db_print_inpcb(inp, "inpcb", 0);
2848 }
2849 #endif /* DDB */
2850
2851 #ifdef RATELIMIT
2852 /*
2853  * Modify TX rate limit based on the existing "inp->inp_snd_tag",
2854  * if any.
2855  */
2856 int
2857 in_pcbmodify_txrtlmt(struct inpcb *inp, uint32_t max_pacing_rate)
2858 {
2859         union if_snd_tag_modify_params params = {
2860                 .rate_limit.max_rate = max_pacing_rate,
2861         };
2862         struct m_snd_tag *mst;
2863         struct ifnet *ifp;
2864         int error;
2865
2866         mst = inp->inp_snd_tag;
2867         if (mst == NULL)
2868                 return (EINVAL);
2869
2870         ifp = mst->ifp;
2871         if (ifp == NULL)
2872                 return (EINVAL);
2873
2874         if (ifp->if_snd_tag_modify == NULL) {
2875                 error = EOPNOTSUPP;
2876         } else {
2877                 error = ifp->if_snd_tag_modify(mst, &params);
2878         }
2879         return (error);
2880 }
2881
2882 /*
2883  * Query existing TX rate limit based on the existing
2884  * "inp->inp_snd_tag", if any.
2885  */
2886 int
2887 in_pcbquery_txrtlmt(struct inpcb *inp, uint32_t *p_max_pacing_rate)
2888 {
2889         union if_snd_tag_query_params params = { };
2890         struct m_snd_tag *mst;
2891         struct ifnet *ifp;
2892         int error;
2893
2894         mst = inp->inp_snd_tag;
2895         if (mst == NULL)
2896                 return (EINVAL);
2897
2898         ifp = mst->ifp;
2899         if (ifp == NULL)
2900                 return (EINVAL);
2901
2902         if (ifp->if_snd_tag_query == NULL) {
2903                 error = EOPNOTSUPP;
2904         } else {
2905                 error = ifp->if_snd_tag_query(mst, &params);
2906                 if (error == 0 &&  p_max_pacing_rate != NULL)
2907                         *p_max_pacing_rate = params.rate_limit.max_rate;
2908         }
2909         return (error);
2910 }
2911
2912 /*
2913  * Query existing TX queue level based on the existing
2914  * "inp->inp_snd_tag", if any.
2915  */
2916 int
2917 in_pcbquery_txrlevel(struct inpcb *inp, uint32_t *p_txqueue_level)
2918 {
2919         union if_snd_tag_query_params params = { };
2920         struct m_snd_tag *mst;
2921         struct ifnet *ifp;
2922         int error;
2923
2924         mst = inp->inp_snd_tag;
2925         if (mst == NULL)
2926                 return (EINVAL);
2927
2928         ifp = mst->ifp;
2929         if (ifp == NULL)
2930                 return (EINVAL);
2931
2932         if (ifp->if_snd_tag_query == NULL)
2933                 return (EOPNOTSUPP);
2934
2935         error = ifp->if_snd_tag_query(mst, &params);
2936         if (error == 0 &&  p_txqueue_level != NULL)
2937                 *p_txqueue_level = params.rate_limit.queue_level;
2938         return (error);
2939 }
2940
2941 /*
2942  * Allocate a new TX rate limit send tag from the network interface
2943  * given by the "ifp" argument and save it in "inp->inp_snd_tag":
2944  */
2945 int
2946 in_pcbattach_txrtlmt(struct inpcb *inp, struct ifnet *ifp,
2947     uint32_t flowtype, uint32_t flowid, uint32_t max_pacing_rate)
2948 {
2949         union if_snd_tag_alloc_params params = {
2950                 .rate_limit.hdr.type = (max_pacing_rate == -1U) ?
2951                     IF_SND_TAG_TYPE_UNLIMITED : IF_SND_TAG_TYPE_RATE_LIMIT,
2952                 .rate_limit.hdr.flowid = flowid,
2953                 .rate_limit.hdr.flowtype = flowtype,
2954                 .rate_limit.max_rate = max_pacing_rate,
2955         };
2956         int error;
2957
2958         INP_WLOCK_ASSERT(inp);
2959
2960         if (inp->inp_snd_tag != NULL)
2961                 return (EINVAL);
2962
2963         if (ifp->if_snd_tag_alloc == NULL) {
2964                 error = EOPNOTSUPP;
2965         } else {
2966                 error = ifp->if_snd_tag_alloc(ifp, &params, &inp->inp_snd_tag);
2967
2968                 /*
2969                  * At success increment the refcount on
2970                  * the send tag's network interface:
2971                  */
2972                 if (error == 0)
2973                         if_ref(inp->inp_snd_tag->ifp);
2974         }
2975         return (error);
2976 }
2977
2978 /*
2979  * Free an existing TX rate limit tag based on the "inp->inp_snd_tag",
2980  * if any:
2981  */
2982 void
2983 in_pcbdetach_txrtlmt(struct inpcb *inp)
2984 {
2985         struct m_snd_tag *mst;
2986         struct ifnet *ifp;
2987
2988         INP_WLOCK_ASSERT(inp);
2989
2990         mst = inp->inp_snd_tag;
2991         inp->inp_snd_tag = NULL;
2992
2993         if (mst == NULL)
2994                 return;
2995
2996         ifp = mst->ifp;
2997         if (ifp == NULL)
2998                 return;
2999
3000         /*
3001          * If the device was detached while we still had reference(s)
3002          * on the ifp, we assume if_snd_tag_free() was replaced with
3003          * stubs.
3004          */
3005         ifp->if_snd_tag_free(mst);
3006
3007         /* release reference count on network interface */
3008         if_rele(ifp);
3009 }
3010
3011 /*
3012  * This function should be called when the INP_RATE_LIMIT_CHANGED flag
3013  * is set in the fast path and will attach/detach/modify the TX rate
3014  * limit send tag based on the socket's so_max_pacing_rate value.
3015  */
3016 void
3017 in_pcboutput_txrtlmt(struct inpcb *inp, struct ifnet *ifp, struct mbuf *mb)
3018 {
3019         struct socket *socket;
3020         uint32_t max_pacing_rate;
3021         bool did_upgrade;
3022         int error;
3023
3024         if (inp == NULL)
3025                 return;
3026
3027         socket = inp->inp_socket;
3028         if (socket == NULL)
3029                 return;
3030
3031         if (!INP_WLOCKED(inp)) {
3032                 /*
3033                  * NOTE: If the write locking fails, we need to bail
3034                  * out and use the non-ratelimited ring for the
3035                  * transmit until there is a new chance to get the
3036                  * write lock.
3037                  */
3038                 if (!INP_TRY_UPGRADE(inp))
3039                         return;
3040                 did_upgrade = 1;
3041         } else {
3042                 did_upgrade = 0;
3043         }
3044
3045         /*
3046          * NOTE: The so_max_pacing_rate value is read unlocked,
3047          * because atomic updates are not required since the variable
3048          * is checked at every mbuf we send. It is assumed that the
3049          * variable read itself will be atomic.
3050          */
3051         max_pacing_rate = socket->so_max_pacing_rate;
3052
3053         /*
3054          * NOTE: When attaching to a network interface a reference is
3055          * made to ensure the network interface doesn't go away until
3056          * all ratelimit connections are gone. The network interface
3057          * pointers compared below represent valid network interfaces,
3058          * except when comparing towards NULL.
3059          */
3060         if (max_pacing_rate == 0 && inp->inp_snd_tag == NULL) {
3061                 error = 0;
3062         } else if (!(ifp->if_capenable & IFCAP_TXRTLMT)) {
3063                 if (inp->inp_snd_tag != NULL)
3064                         in_pcbdetach_txrtlmt(inp);
3065                 error = 0;
3066         } else if (inp->inp_snd_tag == NULL) {
3067                 /*
3068                  * In order to utilize packet pacing with RSS, we need
3069                  * to wait until there is a valid RSS hash before we
3070                  * can proceed:
3071                  */
3072                 if (M_HASHTYPE_GET(mb) == M_HASHTYPE_NONE) {
3073                         error = EAGAIN;
3074                 } else {
3075                         error = in_pcbattach_txrtlmt(inp, ifp, M_HASHTYPE_GET(mb),
3076                             mb->m_pkthdr.flowid, max_pacing_rate);
3077                 }
3078         } else {
3079                 error = in_pcbmodify_txrtlmt(inp, max_pacing_rate);
3080         }
3081         if (error == 0 || error == EOPNOTSUPP)
3082                 inp->inp_flags2 &= ~INP_RATE_LIMIT_CHANGED;
3083         if (did_upgrade)
3084                 INP_DOWNGRADE(inp);
3085 }
3086
3087 /*
3088  * Track route changes for TX rate limiting.
3089  */
3090 void
3091 in_pcboutput_eagain(struct inpcb *inp)
3092 {
3093         struct socket *socket;
3094         bool did_upgrade;
3095
3096         if (inp == NULL)
3097                 return;
3098
3099         socket = inp->inp_socket;
3100         if (socket == NULL)
3101                 return;
3102
3103         if (inp->inp_snd_tag == NULL)
3104                 return;
3105
3106         if (!INP_WLOCKED(inp)) {
3107                 /*
3108                  * NOTE: If the write locking fails, we need to bail
3109                  * out and use the non-ratelimited ring for the
3110                  * transmit until there is a new chance to get the
3111                  * write lock.
3112                  */
3113                 if (!INP_TRY_UPGRADE(inp))
3114                         return;
3115                 did_upgrade = 1;
3116         } else {
3117                 did_upgrade = 0;
3118         }
3119
3120         /* detach rate limiting */
3121         in_pcbdetach_txrtlmt(inp);
3122
3123         /* make sure new mbuf send tag allocation is made */
3124         inp->inp_flags2 |= INP_RATE_LIMIT_CHANGED;
3125
3126         if (did_upgrade)
3127                 INP_DOWNGRADE(inp);
3128 }
3129 #endif /* RATELIMIT */