sys/netinet/tcp_subr.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)tcp_subr.c  8.2 (Berkeley) 5/24/95
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include "opt_compat.h"
  36 #include "opt_inet.h"
  37 #include "opt_inet6.h"
  38 #include "opt_ipsec.h"
  39 #include "opt_tcpdebug.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/callout.h>
  44 #include <sys/eventhandler.h>
  45 #ifdef TCP_HHOOK
  46 #include <sys/hhook.h>
  47 #endif
  48 #include <sys/kernel.h>
  49 #ifdef TCP_HHOOK
  50 #include <sys/khelp.h>
  51 #endif
  52 #include <sys/sysctl.h>
  53 #include <sys/jail.h>
  54 #include <sys/malloc.h>
  55 #include <sys/refcount.h>
  56 #include <sys/mbuf.h>
  57 #ifdef INET6
  58 #include <sys/domain.h>
  59 #endif
  60 #include <sys/priv.h>
  61 #include <sys/proc.h>
  62 #include <sys/sdt.h>
  63 #include <sys/socket.h>
  64 #include <sys/socketvar.h>
  65 #include <sys/protosw.h>
  66 #include <sys/random.h>
  67
  68 #include <vm/uma.h>
  69
  70 #include <net/route.h>
  71 #include <net/if.h>
  72 #include <net/if_var.h>
  73 #include <net/vnet.h>
  74
  75 #include <netinet/in.h>
  76 #include <netinet/in_fib.h>
  77 #include <netinet/in_kdtrace.h>
  78 #include <netinet/in_pcb.h>
  79 #include <netinet/in_systm.h>
  80 #include <netinet/in_var.h>
  81 #include <netinet/ip.h>
  82 #include <netinet/ip_icmp.h>
  83 #include <netinet/ip_var.h>
  84 #ifdef INET6
  85 #include <netinet/icmp6.h>
  86 #include <netinet/ip6.h>
  87 #include <netinet6/in6_fib.h>
  88 #include <netinet6/in6_pcb.h>
  89 #include <netinet6/ip6_var.h>
  90 #include <netinet6/scope6_var.h>
  91 #include <netinet6/nd6.h>
  92 #endif
  93
  94 #ifdef TCP_RFC7413
  95 #include <netinet/tcp_fastopen.h>
  96 #endif
  97 #include <netinet/tcp.h>
  98 #include <netinet/tcp_fsm.h>
  99 #include <netinet/tcp_seq.h>
 100 #include <netinet/tcp_timer.h>
 101 #include <netinet/tcp_var.h>
 102 #include <netinet/tcp_syncache.h>
 103 #include <netinet/cc/cc.h>
 104 #ifdef INET6
 105 #include <netinet6/tcp6_var.h>
 106 #endif
 107 #include <netinet/tcpip.h>
 108 #ifdef TCPPCAP
 109 #include <netinet/tcp_pcap.h>
 110 #endif
 111 #ifdef TCPDEBUG
 112 #include <netinet/tcp_debug.h>
 113 #endif
 114 #ifdef INET6
 115 #include <netinet6/ip6protosw.h>
 116 #endif
 117 #ifdef TCP_OFFLOAD
 118 #include <netinet/tcp_offload.h>
 119 #endif
 120
 121 #include <netipsec/ipsec_support.h>
 122
 123 #include <machine/in_cksum.h>
 124 #include <sys/md5.h>
 125
 126 #include <security/mac/mac_framework.h>
 127
 128 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
 129 #ifdef INET6
 130 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
 131 #endif
 132
 133 struct rwlock tcp_function_lock;
 134
 135 static int
 136 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
 137 {
 138         int error, new;
 139
 140         new = V_tcp_mssdflt;
 141         error = sysctl_handle_int(oidp, &new, 0, req);
 142         if (error == 0 && req->newptr) {
 143                 if (new < TCP_MINMSS)
 144                         error = EINVAL;
 145                 else
 146                         V_tcp_mssdflt = new;
 147         }
 148         return (error);
 149 }
 150
 151 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
 152     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
 153     &sysctl_net_inet_tcp_mss_check, "I",
 154     "Default TCP Maximum Segment Size");
 155
 156 #ifdef INET6
 157 static int
 158 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
 159 {
 160         int error, new;
 161
 162         new = V_tcp_v6mssdflt;
 163         error = sysctl_handle_int(oidp, &new, 0, req);
 164         if (error == 0 && req->newptr) {
 165                 if (new < TCP_MINMSS)
 166                         error = EINVAL;
 167                 else
 168                         V_tcp_v6mssdflt = new;
 169         }
 170         return (error);
 171 }
 172
 173 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
 174     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
 175     &sysctl_net_inet_tcp_mss_v6_check, "I",
 176    "Default TCP Maximum Segment Size for IPv6");
 177 #endif /* INET6 */
 178
 179 /*
 180  * Minimum MSS we accept and use. This prevents DoS attacks where
 181  * we are forced to a ridiculous low MSS like 20 and send hundreds
 182  * of packets instead of one. The effect scales with the available
 183  * bandwidth and quickly saturates the CPU and network interface
 184  * with packet generation and sending. Set to zero to disable MINMSS
 185  * checking. This setting prevents us from sending too small packets.
 186  */
 187 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
 189      &VNET_NAME(tcp_minmss), 0,
 190     "Minimum TCP Maximum Segment Size");
 191
 192 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
 193 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
 194     &VNET_NAME(tcp_do_rfc1323), 0,
 195     "Enable rfc1323 (high performance TCP) extensions");
 196
 197 static int      tcp_log_debug = 0;
 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
 199     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
 200
 201 static int      tcp_tcbhashsize;
 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
 203     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
 204
 205 static int      do_tcpdrain = 1;
 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
 207     "Enable tcp_drain routine for extra help when low on mbufs");
 208
 209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
 210     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
 211
 212 static VNET_DEFINE(int, icmp_may_rst) = 1;
 213 #define V_icmp_may_rst                  VNET(icmp_may_rst)
 214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
 215     &VNET_NAME(icmp_may_rst), 0,
 216     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
 217
 218 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
 219 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
 221     &VNET_NAME(tcp_isn_reseed_interval), 0,
 222     "Seconds between reseeding of ISN secret");
 223
 224 static int      tcp_soreceive_stream;
 225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
 226     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
 227
 228 VNET_DEFINE(uma_zone_t, sack_hole_zone);
 229 #define V_sack_hole_zone                VNET(sack_hole_zone)
 230
 231 #ifdef TCP_HHOOK
 232 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
 233 #endif
 234
 235 static struct inpcb *tcp_notify(struct inpcb *, int);
 236 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
 237 static void tcp_mtudisc(struct inpcb *, int);
 238 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
 239                     void *ip4hdr, const void *ip6hdr);
 240
 241
 242 static struct tcp_function_block tcp_def_funcblk = {
 243         "default",
 244         tcp_output,
 245         tcp_do_segment,
 246         tcp_default_ctloutput,
 247         NULL,
 248         NULL,
 249         NULL,
 250         NULL,
 251         NULL,
 252         NULL,
 253         0,
 254         0
 255 };
 256
 257 int t_functions_inited = 0;
 258 struct tcp_funchead t_functions;
 259 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
 260
 261 static void
 262 init_tcp_functions(void)
 263 {
 264         if (t_functions_inited == 0) {
 265                 TAILQ_INIT(&t_functions);
 266                 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
 267                 t_functions_inited = 1;
 268         }
 269 }
 270
 271 static struct tcp_function_block *
 272 find_tcp_functions_locked(struct tcp_function_set *fs)
 273 {
 274         struct tcp_function *f;
 275         struct tcp_function_block *blk=NULL;
 276
 277         TAILQ_FOREACH(f, &t_functions, tf_next) {
 278                 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
 279                         blk = f->tf_fb;
 280                         break;
 281                 }
 282         }
 283         return(blk);
 284 }
 285
 286 static struct tcp_function_block *
 287 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
 288 {
 289         struct tcp_function_block *rblk=NULL;
 290         struct tcp_function *f;
 291
 292         TAILQ_FOREACH(f, &t_functions, tf_next) {
 293                 if (f->tf_fb == blk) {
 294                         rblk = blk;
 295                         if (s) {
 296                                 *s = f;
 297                         }
 298                         break;
 299                 }
 300         }
 301         return (rblk);
 302 }
 303
 304 struct tcp_function_block *
 305 find_and_ref_tcp_functions(struct tcp_function_set *fs)
 306 {
 307         struct tcp_function_block *blk;
 308
 309         rw_rlock(&tcp_function_lock);
 310         blk = find_tcp_functions_locked(fs);
 311         if (blk)
 312                 refcount_acquire(&blk->tfb_refcnt);
 313         rw_runlock(&tcp_function_lock);
 314         return(blk);
 315 }
 316
 317 struct tcp_function_block *
 318 find_and_ref_tcp_fb(struct tcp_function_block *blk)
 319 {
 320         struct tcp_function_block *rblk;
 321
 322         rw_rlock(&tcp_function_lock);
 323         rblk = find_tcp_fb_locked(blk, NULL);
 324         if (rblk)
 325                 refcount_acquire(&rblk->tfb_refcnt);
 326         rw_runlock(&tcp_function_lock);
 327         return(rblk);
 328 }
 329
 330
 331 static int
 332 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
 333 {
 334         int error=ENOENT;
 335         struct tcp_function_set fs;
 336         struct tcp_function_block *blk;
 337
 338         memset(&fs, 0, sizeof(fs));
 339         rw_rlock(&tcp_function_lock);
 340         blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
 341         if (blk) {
 342                 /* Found him */
 343                 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 344                 fs.pcbcnt = blk->tfb_refcnt;
 345         }
 346         rw_runlock(&tcp_function_lock);
 347         error = sysctl_handle_string(oidp, fs.function_set_name,
 348                                      sizeof(fs.function_set_name), req);
 349
 350         /* Check for error or no change */
 351         if (error != 0 || req->newptr == NULL)
 352                 return(error);
 353
 354         rw_wlock(&tcp_function_lock);
 355         blk = find_tcp_functions_locked(&fs);
 356         if ((blk == NULL) ||
 357             (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
 358                 error = ENOENT;
 359                 goto done;
 360         }
 361         tcp_func_set_ptr = blk;
 362 done:
 363         rw_wunlock(&tcp_function_lock);
 364         return (error);
 365 }
 366
 367 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
 368             CTLTYPE_STRING | CTLFLAG_RW,
 369             NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
 370             "Set/get the default TCP functions");
 371
 372 static int
 373 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
 374 {
 375         int error, cnt, linesz;
 376         struct tcp_function *f;
 377         char *buffer, *cp;
 378         size_t bufsz, outsz;
 379         bool alias;
 380
 381         cnt = 0;
 382         rw_rlock(&tcp_function_lock);
 383         TAILQ_FOREACH(f, &t_functions, tf_next) {
 384                 cnt++;
 385         }
 386         rw_runlock(&tcp_function_lock);
 387
 388         bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
 389         buffer = malloc(bufsz, M_TEMP, M_WAITOK);
 390
 391         error = 0;
 392         cp = buffer;
 393
 394         linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
 395             "Alias", "PCB count");
 396         cp += linesz;
 397         bufsz -= linesz;
 398         outsz = linesz;
 399
 400         rw_rlock(&tcp_function_lock);
 401         TAILQ_FOREACH(f, &t_functions, tf_next) {
 402                 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
 403                 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
 404                     f->tf_fb->tfb_tcp_block_name,
 405                     (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
 406                     alias ? f->tf_name : "-",
 407                     f->tf_fb->tfb_refcnt);
 408                 if (linesz >= bufsz) {
 409                         error = EOVERFLOW;
 410                         break;
 411                 }
 412                 cp += linesz;
 413                 bufsz -= linesz;
 414                 outsz += linesz;
 415         }
 416         rw_runlock(&tcp_function_lock);
 417         if (error == 0)
 418                 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
 419         free(buffer, M_TEMP);
 420         return (error);
 421 }
 422
 423 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
 424             CTLTYPE_STRING|CTLFLAG_RD,
 425             NULL, 0, sysctl_net_inet_list_available, "A",
 426             "list available TCP Function sets");
 427
 428 /*
 429  * Target size of TCP PCB hash tables. Must be a power of two.
 430  *
 431  * Note that this can be overridden by the kernel environment
 432  * variable net.inet.tcp.tcbhashsize
 433  */
 434 #ifndef TCBHASHSIZE
 435 #define TCBHASHSIZE     0
 436 #endif
 437
 438 /*
 439  * XXX
 440  * Callouts should be moved into struct tcp directly.  They are currently
 441  * separate because the tcpcb structure is exported to userland for sysctl
 442  * parsing purposes, which do not know about callouts.
 443  */
 444 struct tcpcb_mem {
 445         struct  tcpcb           tcb;
 446         struct  tcp_timer       tt;
 447         struct  cc_var          ccv;
 448 #ifdef TCP_HHOOK
 449         struct  osd             osd;
 450 #endif
 451 };
 452
 453 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
 454 #define V_tcpcb_zone                    VNET(tcpcb_zone)
 455
 456 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
 457 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
 458
 459 static struct mtx isn_mtx;
 460
 461 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
 462 #define ISN_LOCK()      mtx_lock(&isn_mtx)
 463 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
 464
 465 /*
 466  * TCP initialization.
 467  */
 468 static void
 469 tcp_zone_change(void *tag)
 470 {
 471
 472         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
 473         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 474         tcp_tw_zone_change();
 475 }
 476
 477 static int
 478 tcp_inpcb_init(void *mem, int size, int flags)
 479 {
 480         struct inpcb *inp = mem;
 481
 482         INP_LOCK_INIT(inp, "inp", "tcpinp");
 483         return (0);
 484 }
 485
 486 /*
 487  * Take a value and get the next power of 2 that doesn't overflow.
 488  * Used to size the tcp_inpcb hash buckets.
 489  */
 490 static int
 491 maketcp_hashsize(int size)
 492 {
 493         int hashsize;
 494
 495         /*
 496          * auto tune.
 497          * get the next power of 2 higher than maxsockets.
 498          */
 499         hashsize = 1 << fls(size);
 500         /* catch overflow, and just go one power of 2 smaller */
 501         if (hashsize < size) {
 502                 hashsize = 1 << (fls(size) - 1);
 503         }
 504         return (hashsize);
 505 }
 506
 507 /*
 508  * Register a TCP function block with the name provided in the names
 509  * array.  (Note that this function does NOT automatically register
 510  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
 511  * explicitly include blk->tfb_tcp_block_name in the list of names if
 512  * you wish to register the stack with that name.)
 513  *
 514  * Either all name registrations will succeed or all will fail.  If
 515  * a name registration fails, the function will update the num_names
 516  * argument to point to the array index of the name that encountered
 517  * the failure.
 518  *
 519  * Returns 0 on success, or an error code on failure.
 520  */
 521 int
 522 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
 523     const char *names[], int *num_names)
 524 {
 525         struct tcp_function *n;
 526         struct tcp_function_set fs;
 527         int error, i;
 528
 529         KASSERT(names != NULL && *num_names > 0,
 530             ("%s: Called with 0-length name list", __func__));
 531         KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
 532
 533         if (t_functions_inited == 0) {
 534                 init_tcp_functions();
 535         }
 536         if ((blk->tfb_tcp_output == NULL) ||
 537             (blk->tfb_tcp_do_segment == NULL) ||
 538             (blk->tfb_tcp_ctloutput == NULL) ||
 539             (strlen(blk->tfb_tcp_block_name) == 0)) {
 540                 /*
 541                  * These functions are required and you
 542                  * need a name.
 543                  */
 544                 *num_names = 0;
 545                 return (EINVAL);
 546         }
 547         if (blk->tfb_tcp_timer_stop_all ||
 548             blk->tfb_tcp_timer_activate ||
 549             blk->tfb_tcp_timer_active ||
 550             blk->tfb_tcp_timer_stop) {
 551                 /*
 552                  * If you define one timer function you
 553                  * must have them all.
 554                  */
 555                 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
 556                     (blk->tfb_tcp_timer_activate == NULL) ||
 557                     (blk->tfb_tcp_timer_active == NULL) ||
 558                     (blk->tfb_tcp_timer_stop == NULL)) {
 559                         *num_names = 0;
 560                         return (EINVAL);
 561                 }
 562         }
 563
 564         refcount_init(&blk->tfb_refcnt, 0);
 565         blk->tfb_flags = 0;
 566         for (i = 0; i < *num_names; i++) {
 567                 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
 568                 if (n == NULL) {
 569                         error = ENOMEM;
 570                         goto cleanup;
 571                 }
 572                 n->tf_fb = blk;
 573
 574                 (void)strncpy(fs.function_set_name, names[i],
 575                     TCP_FUNCTION_NAME_LEN_MAX);
 576                 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
 577                 rw_wlock(&tcp_function_lock);
 578                 if (find_tcp_functions_locked(&fs) != NULL) {
 579                         /* Duplicate name space not allowed */
 580                         rw_wunlock(&tcp_function_lock);
 581                         free(n, M_TCPFUNCTIONS);
 582                         error = EALREADY;
 583                         goto cleanup;
 584                 }
 585                 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX);
 586                 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
 587                 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
 588                 rw_wunlock(&tcp_function_lock);
 589         }
 590         return(0);
 591
 592 cleanup:
 593         /*
 594          * Deregister the names we just added. Because registration failed
 595          * for names[i], we don't need to deregister that name.
 596          */
 597         *num_names = i;
 598         rw_wlock(&tcp_function_lock);
 599         while (--i >= 0) {
 600                 TAILQ_FOREACH(n, &t_functions, tf_next) {
 601                         if (!strncmp(n->tf_name, names[i],
 602                             TCP_FUNCTION_NAME_LEN_MAX)) {
 603                                 TAILQ_REMOVE(&t_functions, n, tf_next);
 604                                 n->tf_fb = NULL;
 605                                 free(n, M_TCPFUNCTIONS);
 606                                 break;
 607                         }
 608                 }
 609         }
 610         rw_wunlock(&tcp_function_lock);
 611         return (error);
 612 }
 613
 614 /*
 615  * Register a TCP function block using the name provided in the name
 616  * argument.
 617  *
 618  * Returns 0 on success, or an error code on failure.
 619  */
 620 int
 621 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
 622     int wait)
 623 {
 624         const char *name_list[1];
 625         int num_names, rv;
 626
 627         num_names = 1;
 628         if (name != NULL)
 629                 name_list[0] = name;
 630         else
 631                 name_list[0] = blk->tfb_tcp_block_name;
 632         rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
 633         return (rv);
 634 }
 635
 636 /*
 637  * Register a TCP function block using the name defined in
 638  * blk->tfb_tcp_block_name.
 639  *
 640  * Returns 0 on success, or an error code on failure.
 641  */
 642 int
 643 register_tcp_functions(struct tcp_function_block *blk, int wait)
 644 {
 645
 646         return (register_tcp_functions_as_name(blk, NULL, wait));
 647 }
 648
 649 int
 650 deregister_tcp_functions(struct tcp_function_block *blk)
 651 {
 652         struct tcp_function *f;
 653         int error=ENOENT;
 654
 655         if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
 656                 /* You can't un-register the default */
 657                 return (EPERM);
 658         }
 659         rw_wlock(&tcp_function_lock);
 660         if (blk == tcp_func_set_ptr) {
 661                 /* You can't free the current default */
 662                 rw_wunlock(&tcp_function_lock);
 663                 return (EBUSY);
 664         }
 665         if (blk->tfb_refcnt) {
 666                 /* Still tcb attached, mark it. */
 667                 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
 668                 rw_wunlock(&tcp_function_lock);
 669                 return (EBUSY);
 670         }
 671         while (find_tcp_fb_locked(blk, &f) != NULL) {
 672                 /* Found */
 673                 TAILQ_REMOVE(&t_functions, f, tf_next);
 674                 f->tf_fb = NULL;
 675                 free(f, M_TCPFUNCTIONS);
 676                 error = 0;
 677         }
 678         rw_wunlock(&tcp_function_lock);
 679         return (error);
 680 }
 681
 682 void
 683 tcp_init(void)
 684 {
 685         const char *tcbhash_tuneable;
 686         int hashsize;
 687
 688         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
 689
 690 #ifdef TCP_HHOOK
 691         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
 692             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 693                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 694         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
 695             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 696                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 697 #endif
 698         hashsize = TCBHASHSIZE;
 699         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
 700         if (hashsize == 0) {
 701                 /*
 702                  * Auto tune the hash size based on maxsockets.
 703                  * A perfect hash would have a 1:1 mapping
 704                  * (hashsize = maxsockets) however it's been
 705                  * suggested that O(2) average is better.
 706                  */
 707                 hashsize = maketcp_hashsize(maxsockets / 4);
 708                 /*
 709                  * Our historical default is 512,
 710                  * do not autotune lower than this.
 711                  */
 712                 if (hashsize < 512)
 713                         hashsize = 512;
 714                 if (bootverbose && IS_DEFAULT_VNET(curvnet))
 715                         printf("%s: %s auto tuned to %d\n", __func__,
 716                             tcbhash_tuneable, hashsize);
 717         }
 718         /*
 719          * We require a hashsize to be a power of two.
 720          * Previously if it was not a power of two we would just reset it
 721          * back to 512, which could be a nasty surprise if you did not notice
 722          * the error message.
 723          * Instead what we do is clip it to the closest power of two lower
 724          * than the specified hash value.
 725          */
 726         if (!powerof2(hashsize)) {
 727                 int oldhashsize = hashsize;
 728
 729                 hashsize = maketcp_hashsize(hashsize);
 730                 /* prevent absurdly low value */
 731                 if (hashsize < 16)
 732                         hashsize = 16;
 733                 printf("%s: WARNING: TCB hash size not a power of 2, "
 734                     "clipped from %d to %d.\n", __func__, oldhashsize,
 735                     hashsize);
 736         }
 737         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
 738             "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
 739
 740         /*
 741          * These have to be type stable for the benefit of the timers.
 742          */
 743         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
 744             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 745         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 746         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
 747
 748         tcp_tw_init();
 749         syncache_init();
 750         tcp_hc_init();
 751
 752         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
 753         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
 754             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 755
 756 #ifdef TCP_RFC7413
 757         tcp_fastopen_init();
 758 #endif
 759
 760         /* Skip initialization of globals for non-default instances. */
 761         if (!IS_DEFAULT_VNET(curvnet))
 762                 return;
 763
 764         tcp_reass_global_init();
 765
 766         /* XXX virtualize those bellow? */
 767         tcp_delacktime = TCPTV_DELACK;
 768         tcp_keepinit = TCPTV_KEEP_INIT;
 769         tcp_keepidle = TCPTV_KEEP_IDLE;
 770         tcp_keepintvl = TCPTV_KEEPINTVL;
 771         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
 772         tcp_msl = TCPTV_MSL;
 773         tcp_rexmit_min = TCPTV_MIN;
 774         if (tcp_rexmit_min < 1)
 775                 tcp_rexmit_min = 1;
 776         tcp_persmin = TCPTV_PERSMIN;
 777         tcp_persmax = TCPTV_PERSMAX;
 778         tcp_rexmit_slop = TCPTV_CPU_VAR;
 779         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
 780         tcp_tcbhashsize = hashsize;
 781         /* Setup the tcp function block list */
 782         init_tcp_functions();
 783         register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
 784
 785         if (tcp_soreceive_stream) {
 786 #ifdef INET
 787                 tcp_usrreqs.pru_soreceive = soreceive_stream;
 788 #endif
 789 #ifdef INET6
 790                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
 791 #endif /* INET6 */
 792         }
 793
 794 #ifdef INET6
 795 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
 796 #else /* INET6 */
 797 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
 798 #endif /* INET6 */
 799         if (max_protohdr < TCP_MINPROTOHDR)
 800                 max_protohdr = TCP_MINPROTOHDR;
 801         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
 802                 panic("tcp_init");
 803 #undef TCP_MINPROTOHDR
 804
 805         ISN_LOCK_INIT();
 806         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
 807                 SHUTDOWN_PRI_DEFAULT);
 808         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
 809                 EVENTHANDLER_PRI_ANY);
 810 #ifdef TCPPCAP
 811         tcp_pcap_init();
 812 #endif
 813 }
 814
 815 #ifdef VIMAGE
 816 static void
 817 tcp_destroy(void *unused __unused)
 818 {
 819         int n;
 820 #ifdef TCP_HHOOK
 821         int error;
 822 #endif
 823
 824         /*
 825          * All our processes are gone, all our sockets should be cleaned
 826          * up, which means, we should be past the tcp_discardcb() calls.
 827          * Sleep to let all tcpcb timers really disappear and cleanup.
 828          */
 829         for (;;) {
 830                 INP_LIST_RLOCK(&V_tcbinfo);
 831                 n = V_tcbinfo.ipi_count;
 832                 INP_LIST_RUNLOCK(&V_tcbinfo);
 833                 if (n == 0)
 834                         break;
 835                 pause("tcpdes", hz / 10);
 836         }
 837         tcp_hc_destroy();
 838         syncache_destroy();
 839         tcp_tw_destroy();
 840         in_pcbinfo_destroy(&V_tcbinfo);
 841         /* tcp_discardcb() clears the sack_holes up. */
 842         uma_zdestroy(V_sack_hole_zone);
 843         uma_zdestroy(V_tcpcb_zone);
 844
 845 #ifdef TCP_RFC7413
 846         /*
 847          * Cannot free the zone until all tcpcbs are released as we attach
 848          * the allocations to them.
 849          */
 850         tcp_fastopen_destroy();
 851 #endif
 852
 853 #ifdef TCP_HHOOK
 854         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
 855         if (error != 0) {
 856                 printf("%s: WARNING: unable to deregister helper hook "
 857                     "type=%d, id=%d: error %d returned\n", __func__,
 858                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
 859         }
 860         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
 861         if (error != 0) {
 862                 printf("%s: WARNING: unable to deregister helper hook "
 863                     "type=%d, id=%d: error %d returned\n", __func__,
 864                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
 865         }
 866 #endif
 867 }
 868 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
 869 #endif
 870
 871 void
 872 tcp_fini(void *xtp)
 873 {
 874
 875 }
 876
 877 /*
 878  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
 879  * tcp_template used to store this data in mbufs, but we now recopy it out
 880  * of the tcpcb each time to conserve mbufs.
 881  */
 882 void
 883 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
 884 {
 885         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
 886
 887         INP_WLOCK_ASSERT(inp);
 888
 889 #ifdef INET6
 890         if ((inp->inp_vflag & INP_IPV6) != 0) {
 891                 struct ip6_hdr *ip6;
 892
 893                 ip6 = (struct ip6_hdr *)ip_ptr;
 894                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
 895                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
 896                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
 897                         (IPV6_VERSION & IPV6_VERSION_MASK);
 898                 ip6->ip6_nxt = IPPROTO_TCP;
 899                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
 900                 ip6->ip6_src = inp->in6p_laddr;
 901                 ip6->ip6_dst = inp->in6p_faddr;
 902         }
 903 #endif /* INET6 */
 904 #if defined(INET6) && defined(INET)
 905         else
 906 #endif
 907 #ifdef INET
 908         {
 909                 struct ip *ip;
 910
 911                 ip = (struct ip *)ip_ptr;
 912                 ip->ip_v = IPVERSION;
 913                 ip->ip_hl = 5;
 914                 ip->ip_tos = inp->inp_ip_tos;
 915                 ip->ip_len = 0;
 916                 ip->ip_id = 0;
 917                 ip->ip_off = 0;
 918                 ip->ip_ttl = inp->inp_ip_ttl;
 919                 ip->ip_sum = 0;
 920                 ip->ip_p = IPPROTO_TCP;
 921                 ip->ip_src = inp->inp_laddr;
 922                 ip->ip_dst = inp->inp_faddr;
 923         }
 924 #endif /* INET */
 925         th->th_sport = inp->inp_lport;
 926         th->th_dport = inp->inp_fport;
 927         th->th_seq = 0;
 928         th->th_ack = 0;
 929         th->th_x2 = 0;
 930         th->th_off = 5;
 931         th->th_flags = 0;
 932         th->th_win = 0;
 933         th->th_urp = 0;
 934         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
 935 }
 936
 937 /*
 938  * Create template to be used to send tcp packets on a connection.
 939  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
 940  * use for this function is in keepalives, which use tcp_respond.
 941  */
 942 struct tcptemp *
 943 tcpip_maketemplate(struct inpcb *inp)
 944 {
 945         struct tcptemp *t;
 946
 947         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
 948         if (t == NULL)
 949                 return (NULL);
 950         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
 951         return (t);
 952 }
 953
 954 /*
 955  * Send a single message to the TCP at address specified by
 956  * the given TCP/IP header.  If m == NULL, then we make a copy
 957  * of the tcpiphdr at th and send directly to the addressed host.
 958  * This is used to force keep alive messages out using the TCP
 959  * template for a connection.  If flags are given then we send
 960  * a message back to the TCP which originated the segment th,
 961  * and discard the mbuf containing it and any other attached mbufs.
 962  *
 963  * In any case the ack and sequence number of the transmitted
 964  * segment are as specified by the parameters.
 965  *
 966  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
 967  */
 968 void
 969 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
 970     tcp_seq ack, tcp_seq seq, int flags)
 971 {
 972         struct tcpopt to;
 973         struct inpcb *inp;
 974         struct ip *ip;
 975         struct mbuf *optm;
 976         struct tcphdr *nth;
 977         u_char *optp;
 978 #ifdef INET6
 979         struct ip6_hdr *ip6;
 980         int isipv6;
 981 #endif /* INET6 */
 982         int optlen, tlen, win;
 983         bool incl_opts;
 984
 985         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
 986
 987 #ifdef INET6
 988         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
 989         ip6 = ipgen;
 990 #endif /* INET6 */
 991         ip = ipgen;
 992
 993         if (tp != NULL) {
 994                 inp = tp->t_inpcb;
 995                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
 996                 INP_WLOCK_ASSERT(inp);
 997         } else
 998                 inp = NULL;
 999
1000         incl_opts = false;
1001         win = 0;
1002         if (tp != NULL) {
1003                 if (!(flags & TH_RST)) {
1004                         win = sbspace(&inp->inp_socket->so_rcv);
1005                         if (win > TCP_MAXWIN << tp->rcv_scale)
1006                                 win = TCP_MAXWIN << tp->rcv_scale;
1007                 }
1008                 if ((tp->t_flags & TF_NOOPT) == 0)
1009                         incl_opts = true;
1010         }
1011         if (m == NULL) {
1012                 m = m_gethdr(M_NOWAIT, MT_DATA);
1013                 if (m == NULL)
1014                         return;
1015                 m->m_data += max_linkhdr;
1016 #ifdef INET6
1017                 if (isipv6) {
1018                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
1019                               sizeof(struct ip6_hdr));
1020                         ip6 = mtod(m, struct ip6_hdr *);
1021                         nth = (struct tcphdr *)(ip6 + 1);
1022                 } else
1023 #endif /* INET6 */
1024                 {
1025                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1026                         ip = mtod(m, struct ip *);
1027                         nth = (struct tcphdr *)(ip + 1);
1028                 }
1029                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1030                 flags = TH_ACK;
1031         } else if (!M_WRITABLE(m)) {
1032                 struct mbuf *n;
1033
1034                 /* Can't reuse 'm', allocate a new mbuf. */
1035                 n = m_gethdr(M_NOWAIT, MT_DATA);
1036                 if (n == NULL) {
1037                         m_freem(m);
1038                         return;
1039                 }
1040
1041                 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1042                         m_freem(m);
1043                         m_freem(n);
1044                         return;
1045                 }
1046
1047                 n->m_data += max_linkhdr;
1048                 /* m_len is set later */
1049 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1050 #ifdef INET6
1051                 if (isipv6) {
1052                         bcopy((caddr_t)ip6, mtod(n, caddr_t),
1053                               sizeof(struct ip6_hdr));
1054                         ip6 = mtod(n, struct ip6_hdr *);
1055                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1056                         nth = (struct tcphdr *)(ip6 + 1);
1057                 } else
1058 #endif /* INET6 */
1059                 {
1060                         bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1061                         ip = mtod(n, struct ip *);
1062                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1063                         nth = (struct tcphdr *)(ip + 1);
1064                 }
1065                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1066                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1067                 th = nth;
1068                 m_freem(m);
1069                 m = n;
1070         } else {
1071                 /*
1072                  *  reuse the mbuf.
1073                  * XXX MRT We inherit the FIB, which is lucky.
1074                  */
1075                 m_freem(m->m_next);
1076                 m->m_next = NULL;
1077                 m->m_data = (caddr_t)ipgen;
1078                 /* m_len is set later */
1079 #ifdef INET6
1080                 if (isipv6) {
1081                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1082                         nth = (struct tcphdr *)(ip6 + 1);
1083                 } else
1084 #endif /* INET6 */
1085                 {
1086                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1087                         nth = (struct tcphdr *)(ip + 1);
1088                 }
1089                 if (th != nth) {
1090                         /*
1091                          * this is usually a case when an extension header
1092                          * exists between the IPv6 header and the
1093                          * TCP header.
1094                          */
1095                         nth->th_sport = th->th_sport;
1096                         nth->th_dport = th->th_dport;
1097                 }
1098                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1099 #undef xchg
1100         }
1101         tlen = 0;
1102 #ifdef INET6
1103         if (isipv6)
1104                 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1105 #endif
1106 #if defined(INET) && defined(INET6)
1107         else
1108 #endif
1109 #ifdef INET
1110                 tlen = sizeof (struct tcpiphdr);
1111 #endif
1112 #ifdef INVARIANTS
1113         m->m_len = 0;
1114         KASSERT(M_TRAILINGSPACE(m) >= tlen,
1115             ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1116             m, tlen, (long)M_TRAILINGSPACE(m)));
1117 #endif
1118         m->m_len = tlen;
1119         to.to_flags = 0;
1120         if (incl_opts) {
1121                 /* Make sure we have room. */
1122                 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1123                         m->m_next = m_get(M_NOWAIT, MT_DATA);
1124                         if (m->m_next) {
1125                                 optp = mtod(m->m_next, u_char *);
1126                                 optm = m->m_next;
1127                         } else
1128                                 incl_opts = false;
1129                 } else {
1130                         optp = (u_char *) (nth + 1);
1131                         optm = m;
1132                 }
1133         }
1134         if (incl_opts) {
1135                 /* Timestamps. */
1136                 if (tp->t_flags & TF_RCVD_TSTMP) {
1137                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1138                         to.to_tsecr = tp->ts_recent;
1139                         to.to_flags |= TOF_TS;
1140                 }
1141 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1142                 /* TCP-MD5 (RFC2385). */
1143                 if (tp->t_flags & TF_SIGNATURE)
1144                         to.to_flags |= TOF_SIGNATURE;
1145 #endif
1146                 /* Add the options. */
1147                 tlen += optlen = tcp_addoptions(&to, optp);
1148
1149                 /* Update m_len in the correct mbuf. */
1150                 optm->m_len += optlen;
1151         } else
1152                 optlen = 0;
1153 #ifdef INET6
1154         if (isipv6) {
1155                 ip6->ip6_flow = 0;
1156                 ip6->ip6_vfc = IPV6_VERSION;
1157                 ip6->ip6_nxt = IPPROTO_TCP;
1158                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1159         }
1160 #endif
1161 #if defined(INET) && defined(INET6)
1162         else
1163 #endif
1164 #ifdef INET
1165         {
1166                 ip->ip_len = htons(tlen);
1167                 ip->ip_ttl = V_ip_defttl;
1168                 if (V_path_mtu_discovery)
1169                         ip->ip_off |= htons(IP_DF);
1170         }
1171 #endif
1172         m->m_pkthdr.len = tlen;
1173         m->m_pkthdr.rcvif = NULL;
1174 #ifdef MAC
1175         if (inp != NULL) {
1176                 /*
1177                  * Packet is associated with a socket, so allow the
1178                  * label of the response to reflect the socket label.
1179                  */
1180                 INP_WLOCK_ASSERT(inp);
1181                 mac_inpcb_create_mbuf(inp, m);
1182         } else {
1183                 /*
1184                  * Packet is not associated with a socket, so possibly
1185                  * update the label in place.
1186                  */
1187                 mac_netinet_tcp_reply(m);
1188         }
1189 #endif
1190         nth->th_seq = htonl(seq);
1191         nth->th_ack = htonl(ack);
1192         nth->th_x2 = 0;
1193         nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1194         nth->th_flags = flags;
1195         if (tp != NULL)
1196                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1197         else
1198                 nth->th_win = htons((u_short)win);
1199         nth->th_urp = 0;
1200
1201 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1202         if (to.to_flags & TOF_SIGNATURE) {
1203                 if (!TCPMD5_ENABLED() ||
1204                     TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1205                         m_freem(m);
1206                         return;
1207                 }
1208         }
1209 #endif
1210
1211         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1212 #ifdef INET6
1213         if (isipv6) {
1214                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1215                 nth->th_sum = in6_cksum_pseudo(ip6,
1216                     tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1217                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1218                     NULL, NULL);
1219         }
1220 #endif /* INET6 */
1221 #if defined(INET6) && defined(INET)
1222         else
1223 #endif
1224 #ifdef INET
1225         {
1226                 m->m_pkthdr.csum_flags = CSUM_TCP;
1227                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1228                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1229         }
1230 #endif /* INET */
1231 #ifdef TCPDEBUG
1232         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1233                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1234 #endif
1235         TCP_PROBE3(debug__output, tp, th, m);
1236         if (flags & TH_RST)
1237                 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1238
1239         TCP_PROBE5(send, NULL, tp, m, tp, nth);
1240 #ifdef INET6
1241         if (isipv6)
1242                 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1243 #endif /* INET6 */
1244 #if defined(INET) && defined(INET6)
1245         else
1246 #endif
1247 #ifdef INET
1248                 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1249 #endif
1250 }
1251
1252 /*
1253  * Create a new TCP control block, making an
1254  * empty reassembly queue and hooking it to the argument
1255  * protocol control block.  The `inp' parameter must have
1256  * come from the zone allocator set up in tcp_init().
1257  */
1258 struct tcpcb *
1259 tcp_newtcpcb(struct inpcb *inp)
1260 {
1261         struct tcpcb_mem *tm;
1262         struct tcpcb *tp;
1263 #ifdef INET6
1264         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1265 #endif /* INET6 */
1266
1267         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1268         if (tm == NULL)
1269                 return (NULL);
1270         tp = &tm->tcb;
1271
1272         /* Initialise cc_var struct for this tcpcb. */
1273         tp->ccv = &tm->ccv;
1274         tp->ccv->type = IPPROTO_TCP;
1275         tp->ccv->ccvc.tcp = tp;
1276         rw_rlock(&tcp_function_lock);
1277         tp->t_fb = tcp_func_set_ptr;
1278         refcount_acquire(&tp->t_fb->tfb_refcnt);
1279         rw_runlock(&tcp_function_lock);
1280         /*
1281          * Use the current system default CC algorithm.
1282          */
1283         CC_LIST_RLOCK();
1284         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1285         CC_ALGO(tp) = CC_DEFAULT();
1286         CC_LIST_RUNLOCK();
1287
1288         if (CC_ALGO(tp)->cb_init != NULL)
1289                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1290                         if (tp->t_fb->tfb_tcp_fb_fini)
1291                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1292                         refcount_release(&tp->t_fb->tfb_refcnt);
1293                         uma_zfree(V_tcpcb_zone, tm);
1294                         return (NULL);
1295                 }
1296
1297 #ifdef TCP_HHOOK
1298         tp->osd = &tm->osd;
1299         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1300                 if (tp->t_fb->tfb_tcp_fb_fini)
1301                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1302                 refcount_release(&tp->t_fb->tfb_refcnt);
1303                 uma_zfree(V_tcpcb_zone, tm);
1304                 return (NULL);
1305         }
1306 #endif
1307
1308 #ifdef VIMAGE
1309         tp->t_vnet = inp->inp_vnet;
1310 #endif
1311         tp->t_timers = &tm->tt;
1312         /*      LIST_INIT(&tp->t_segq); */      /* XXX covered by M_ZERO */
1313         tp->t_maxseg =
1314 #ifdef INET6
1315                 isipv6 ? V_tcp_v6mssdflt :
1316 #endif /* INET6 */
1317                 V_tcp_mssdflt;
1318
1319         /* Set up our timeouts. */
1320         callout_init(&tp->t_timers->tt_rexmt, 1);
1321         callout_init(&tp->t_timers->tt_persist, 1);
1322         callout_init(&tp->t_timers->tt_keep, 1);
1323         callout_init(&tp->t_timers->tt_2msl, 1);
1324         callout_init(&tp->t_timers->tt_delack, 1);
1325
1326         if (V_tcp_do_rfc1323)
1327                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1328         if (V_tcp_do_sack)
1329                 tp->t_flags |= TF_SACK_PERMIT;
1330         TAILQ_INIT(&tp->snd_holes);
1331         /*
1332          * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1333          * is called.
1334          */
1335         in_pcbref(inp); /* Reference for tcpcb */
1336         tp->t_inpcb = inp;
1337
1338         /*
1339          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1340          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1341          * reasonable initial retransmit time.
1342          */
1343         tp->t_srtt = TCPTV_SRTTBASE;
1344         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1345         tp->t_rttmin = tcp_rexmit_min;
1346         tp->t_rxtcur = TCPTV_RTOBASE;
1347         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1348         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1349         tp->t_rcvtime = ticks;
1350         /*
1351          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1352          * because the socket may be bound to an IPv6 wildcard address,
1353          * which may match an IPv4-mapped IPv6 address.
1354          */
1355         inp->inp_ip_ttl = V_ip_defttl;
1356         inp->inp_ppcb = tp;
1357 #ifdef TCPPCAP
1358         /*
1359          * Init the TCP PCAP queues.
1360          */
1361         tcp_pcap_tcpcb_init(tp);
1362 #endif
1363         if (tp->t_fb->tfb_tcp_fb_init) {
1364                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1365         }
1366         return (tp);            /* XXX */
1367 }
1368
1369 /*
1370  * Switch the congestion control algorithm back to NewReno for any active
1371  * control blocks using an algorithm which is about to go away.
1372  * This ensures the CC framework can allow the unload to proceed without leaving
1373  * any dangling pointers which would trigger a panic.
1374  * Returning non-zero would inform the CC framework that something went wrong
1375  * and it would be unsafe to allow the unload to proceed. However, there is no
1376  * way for this to occur with this implementation so we always return zero.
1377  */
1378 int
1379 tcp_ccalgounload(struct cc_algo *unload_algo)
1380 {
1381         struct cc_algo *tmpalgo;
1382         struct inpcb *inp;
1383         struct tcpcb *tp;
1384         VNET_ITERATOR_DECL(vnet_iter);
1385
1386         /*
1387          * Check all active control blocks across all network stacks and change
1388          * any that are using "unload_algo" back to NewReno. If "unload_algo"
1389          * requires cleanup code to be run, call it.
1390          */
1391         VNET_LIST_RLOCK();
1392         VNET_FOREACH(vnet_iter) {
1393                 CURVNET_SET(vnet_iter);
1394                 INP_INFO_WLOCK(&V_tcbinfo);
1395                 /*
1396                  * New connections already part way through being initialised
1397                  * with the CC algo we're removing will not race with this code
1398                  * because the INP_INFO_WLOCK is held during initialisation. We
1399                  * therefore don't enter the loop below until the connection
1400                  * list has stabilised.
1401                  */
1402                 LIST_FOREACH(inp, &V_tcb, inp_list) {
1403                         INP_WLOCK(inp);
1404                         /* Important to skip tcptw structs. */
1405                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1406                             (tp = intotcpcb(inp)) != NULL) {
1407                                 /*
1408                                  * By holding INP_WLOCK here, we are assured
1409                                  * that the connection is not currently
1410                                  * executing inside the CC module's functions
1411                                  * i.e. it is safe to make the switch back to
1412                                  * NewReno.
1413                                  */
1414                                 if (CC_ALGO(tp) == unload_algo) {
1415                                         tmpalgo = CC_ALGO(tp);
1416                                         /* NewReno does not require any init. */
1417                                         CC_ALGO(tp) = &newreno_cc_algo;
1418                                         if (tmpalgo->cb_destroy != NULL)
1419                                                 tmpalgo->cb_destroy(tp->ccv);
1420                                 }
1421                         }
1422                         INP_WUNLOCK(inp);
1423                 }
1424                 INP_INFO_WUNLOCK(&V_tcbinfo);
1425                 CURVNET_RESTORE();
1426         }
1427         VNET_LIST_RUNLOCK();
1428
1429         return (0);
1430 }
1431
1432 /*
1433  * Drop a TCP connection, reporting
1434  * the specified error.  If connection is synchronized,
1435  * then send a RST to peer.
1436  */
1437 struct tcpcb *
1438 tcp_drop(struct tcpcb *tp, int errno)
1439 {
1440         struct socket *so = tp->t_inpcb->inp_socket;
1441
1442         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1443         INP_WLOCK_ASSERT(tp->t_inpcb);
1444
1445         if (TCPS_HAVERCVDSYN(tp->t_state)) {
1446                 tcp_state_change(tp, TCPS_CLOSED);
1447                 (void) tp->t_fb->tfb_tcp_output(tp);
1448                 TCPSTAT_INC(tcps_drops);
1449         } else
1450                 TCPSTAT_INC(tcps_conndrops);
1451         if (errno == ETIMEDOUT && tp->t_softerror)
1452                 errno = tp->t_softerror;
1453         so->so_error = errno;
1454         return (tcp_close(tp));
1455 }
1456
1457 void
1458 tcp_discardcb(struct tcpcb *tp)
1459 {
1460         struct inpcb *inp = tp->t_inpcb;
1461         struct socket *so = inp->inp_socket;
1462 #ifdef INET6
1463         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1464 #endif /* INET6 */
1465         int released;
1466
1467         INP_WLOCK_ASSERT(inp);
1468
1469         /*
1470          * Make sure that all of our timers are stopped before we delete the
1471          * PCB.
1472          *
1473          * If stopping a timer fails, we schedule a discard function in same
1474          * callout, and the last discard function called will take care of
1475          * deleting the tcpcb.
1476          */
1477         tp->t_timers->tt_draincnt = 0;
1478         tcp_timer_stop(tp, TT_REXMT);
1479         tcp_timer_stop(tp, TT_PERSIST);
1480         tcp_timer_stop(tp, TT_KEEP);
1481         tcp_timer_stop(tp, TT_2MSL);
1482         tcp_timer_stop(tp, TT_DELACK);
1483         if (tp->t_fb->tfb_tcp_timer_stop_all) {
1484                 /*
1485                  * Call the stop-all function of the methods,
1486                  * this function should call the tcp_timer_stop()
1487                  * method with each of the function specific timeouts.
1488                  * That stop will be called via the tfb_tcp_timer_stop()
1489                  * which should use the async drain function of the
1490                  * callout system (see tcp_var.h).
1491                  */
1492                 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1493         }
1494
1495         /*
1496          * If we got enough samples through the srtt filter,
1497          * save the rtt and rttvar in the routing entry.
1498          * 'Enough' is arbitrarily defined as 4 rtt samples.
1499          * 4 samples is enough for the srtt filter to converge
1500          * to within enough % of the correct value; fewer samples
1501          * and we could save a bogus rtt. The danger is not high
1502          * as tcp quickly recovers from everything.
1503          * XXX: Works very well but needs some more statistics!
1504          */
1505         if (tp->t_rttupdated >= 4) {
1506                 struct hc_metrics_lite metrics;
1507                 uint32_t ssthresh;
1508
1509                 bzero(&metrics, sizeof(metrics));
1510                 /*
1511                  * Update the ssthresh always when the conditions below
1512                  * are satisfied. This gives us better new start value
1513                  * for the congestion avoidance for new connections.
1514                  * ssthresh is only set if packet loss occurred on a session.
1515                  *
1516                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1517                  * being torn down.  Ideally this code would not use 'so'.
1518                  */
1519                 ssthresh = tp->snd_ssthresh;
1520                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1521                         /*
1522                          * convert the limit from user data bytes to
1523                          * packets then to packet data bytes.
1524                          */
1525                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1526                         if (ssthresh < 2)
1527                                 ssthresh = 2;
1528                         ssthresh *= (tp->t_maxseg +
1529 #ifdef INET6
1530                             (isipv6 ? sizeof (struct ip6_hdr) +
1531                                 sizeof (struct tcphdr) :
1532 #endif
1533                                 sizeof (struct tcpiphdr)
1534 #ifdef INET6
1535                             )
1536 #endif
1537                             );
1538                 } else
1539                         ssthresh = 0;
1540                 metrics.rmx_ssthresh = ssthresh;
1541
1542                 metrics.rmx_rtt = tp->t_srtt;
1543                 metrics.rmx_rttvar = tp->t_rttvar;
1544                 metrics.rmx_cwnd = tp->snd_cwnd;
1545                 metrics.rmx_sendpipe = 0;
1546                 metrics.rmx_recvpipe = 0;
1547
1548                 tcp_hc_update(&inp->inp_inc, &metrics);
1549         }
1550
1551         /* free the reassembly queue, if any */
1552         tcp_reass_flush(tp);
1553
1554 #ifdef TCP_OFFLOAD
1555         /* Disconnect offload device, if any. */
1556         if (tp->t_flags & TF_TOE)
1557                 tcp_offload_detach(tp);
1558 #endif
1559
1560         tcp_free_sackholes(tp);
1561
1562 #ifdef TCPPCAP
1563         /* Free the TCP PCAP queues. */
1564         tcp_pcap_drain(&(tp->t_inpkts));
1565         tcp_pcap_drain(&(tp->t_outpkts));
1566 #endif
1567
1568         /* Allow the CC algorithm to clean up after itself. */
1569         if (CC_ALGO(tp)->cb_destroy != NULL)
1570                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1571
1572 #ifdef TCP_HHOOK
1573         khelp_destroy_osd(tp->osd);
1574 #endif
1575
1576         CC_ALGO(tp) = NULL;
1577         inp->inp_ppcb = NULL;
1578         if (tp->t_timers->tt_draincnt == 0) {
1579                 /* We own the last reference on tcpcb, let's free it. */
1580                 TCPSTATES_DEC(tp->t_state);
1581                 if (tp->t_fb->tfb_tcp_fb_fini)
1582                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1583                 refcount_release(&tp->t_fb->tfb_refcnt);
1584                 tp->t_inpcb = NULL;
1585                 uma_zfree(V_tcpcb_zone, tp);
1586                 released = in_pcbrele_wlocked(inp);
1587                 KASSERT(!released, ("%s: inp %p should not have been released "
1588                         "here", __func__, inp));
1589         }
1590 }
1591
1592 void
1593 tcp_timer_discard(void *ptp)
1594 {
1595         struct inpcb *inp;
1596         struct tcpcb *tp;
1597
1598         tp = (struct tcpcb *)ptp;
1599         CURVNET_SET(tp->t_vnet);
1600         INP_INFO_RLOCK(&V_tcbinfo);
1601         inp = tp->t_inpcb;
1602         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1603                 __func__, tp));
1604         INP_WLOCK(inp);
1605         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1606                 ("%s: tcpcb has to be stopped here", __func__));
1607         tp->t_timers->tt_draincnt--;
1608         if (tp->t_timers->tt_draincnt == 0) {
1609                 /* We own the last reference on this tcpcb, let's free it. */
1610                 TCPSTATES_DEC(tp->t_state);
1611                 if (tp->t_fb->tfb_tcp_fb_fini)
1612                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1613                 refcount_release(&tp->t_fb->tfb_refcnt);
1614                 tp->t_inpcb = NULL;
1615                 uma_zfree(V_tcpcb_zone, tp);
1616                 if (in_pcbrele_wlocked(inp)) {
1617                         INP_INFO_RUNLOCK(&V_tcbinfo);
1618                         CURVNET_RESTORE();
1619                         return;
1620                 }
1621         }
1622         INP_WUNLOCK(inp);
1623         INP_INFO_RUNLOCK(&V_tcbinfo);
1624         CURVNET_RESTORE();
1625 }
1626
1627 /*
1628  * Attempt to close a TCP control block, marking it as dropped, and freeing
1629  * the socket if we hold the only reference.
1630  */
1631 struct tcpcb *
1632 tcp_close(struct tcpcb *tp)
1633 {
1634         struct inpcb *inp = tp->t_inpcb;
1635         struct socket *so;
1636
1637         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1638         INP_WLOCK_ASSERT(inp);
1639
1640 #ifdef TCP_OFFLOAD
1641         if (tp->t_state == TCPS_LISTEN)
1642                 tcp_offload_listen_stop(tp);
1643 #endif
1644 #ifdef TCP_RFC7413
1645         /*
1646          * This releases the TFO pending counter resource for TFO listen
1647          * sockets as well as passively-created TFO sockets that transition
1648          * from SYN_RECEIVED to CLOSED.
1649          */
1650         if (tp->t_tfo_pending) {
1651                 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1652                 tp->t_tfo_pending = NULL;
1653         }
1654 #endif
1655         in_pcbdrop(inp);
1656         TCPSTAT_INC(tcps_closed);
1657         if (tp->t_state != TCPS_CLOSED)
1658                 tcp_state_change(tp, TCPS_CLOSED);
1659         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1660         so = inp->inp_socket;
1661         soisdisconnected(so);
1662         if (inp->inp_flags & INP_SOCKREF) {
1663                 KASSERT(so->so_state & SS_PROTOREF,
1664                     ("tcp_close: !SS_PROTOREF"));
1665                 inp->inp_flags &= ~INP_SOCKREF;
1666                 INP_WUNLOCK(inp);
1667                 SOCK_LOCK(so);
1668                 so->so_state &= ~SS_PROTOREF;
1669                 sofree(so);
1670                 return (NULL);
1671         }
1672         return (tp);
1673 }
1674
1675 void
1676 tcp_drain(void)
1677 {
1678         VNET_ITERATOR_DECL(vnet_iter);
1679
1680         if (!do_tcpdrain)
1681                 return;
1682
1683         VNET_LIST_RLOCK_NOSLEEP();
1684         VNET_FOREACH(vnet_iter) {
1685                 CURVNET_SET(vnet_iter);
1686                 struct inpcb *inpb;
1687                 struct tcpcb *tcpb;
1688
1689         /*
1690          * Walk the tcpbs, if existing, and flush the reassembly queue,
1691          * if there is one...
1692          * XXX: The "Net/3" implementation doesn't imply that the TCP
1693          *      reassembly queue should be flushed, but in a situation
1694          *      where we're really low on mbufs, this is potentially
1695          *      useful.
1696          */
1697                 INP_INFO_WLOCK(&V_tcbinfo);
1698                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1699                         if (inpb->inp_flags & INP_TIMEWAIT)
1700                                 continue;
1701                         INP_WLOCK(inpb);
1702                         if ((tcpb = intotcpcb(inpb)) != NULL) {
1703                                 tcp_reass_flush(tcpb);
1704                                 tcp_clean_sackreport(tcpb);
1705 #ifdef TCPPCAP
1706                                 if (tcp_pcap_aggressive_free) {
1707                                         /* Free the TCP PCAP queues. */
1708                                         tcp_pcap_drain(&(tcpb->t_inpkts));
1709                                         tcp_pcap_drain(&(tcpb->t_outpkts));
1710                                 }
1711 #endif
1712                         }
1713                         INP_WUNLOCK(inpb);
1714                 }
1715                 INP_INFO_WUNLOCK(&V_tcbinfo);
1716                 CURVNET_RESTORE();
1717         }
1718         VNET_LIST_RUNLOCK_NOSLEEP();
1719 }
1720
1721 /*
1722  * Notify a tcp user of an asynchronous error;
1723  * store error as soft error, but wake up user
1724  * (for now, won't do anything until can select for soft error).
1725  *
1726  * Do not wake up user since there currently is no mechanism for
1727  * reporting soft errors (yet - a kqueue filter may be added).
1728  */
1729 static struct inpcb *
1730 tcp_notify(struct inpcb *inp, int error)
1731 {
1732         struct tcpcb *tp;
1733
1734         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1735         INP_WLOCK_ASSERT(inp);
1736
1737         if ((inp->inp_flags & INP_TIMEWAIT) ||
1738             (inp->inp_flags & INP_DROPPED))
1739                 return (inp);
1740
1741         tp = intotcpcb(inp);
1742         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1743
1744         /*
1745          * Ignore some errors if we are hooked up.
1746          * If connection hasn't completed, has retransmitted several times,
1747          * and receives a second error, give up now.  This is better
1748          * than waiting a long time to establish a connection that
1749          * can never complete.
1750          */
1751         if (tp->t_state == TCPS_ESTABLISHED &&
1752             (error == EHOSTUNREACH || error == ENETUNREACH ||
1753              error == EHOSTDOWN)) {
1754                 if (inp->inp_route.ro_rt) {
1755                         RTFREE(inp->inp_route.ro_rt);
1756                         inp->inp_route.ro_rt = (struct rtentry *)NULL;
1757                 }
1758                 return (inp);
1759         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1760             tp->t_softerror) {
1761                 tp = tcp_drop(tp, error);
1762                 if (tp != NULL)
1763                         return (inp);
1764                 else
1765                         return (NULL);
1766         } else {
1767                 tp->t_softerror = error;
1768                 return (inp);
1769         }
1770 #if 0
1771         wakeup( &so->so_timeo);
1772         sorwakeup(so);
1773         sowwakeup(so);
1774 #endif
1775 }
1776
1777 static int
1778 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1779 {
1780         int error, i, m, n, pcb_count;
1781         struct inpcb *inp, **inp_list;
1782         inp_gen_t gencnt;
1783         struct xinpgen xig;
1784
1785         /*
1786          * The process of preparing the TCB list is too time-consuming and
1787          * resource-intensive to repeat twice on every request.
1788          */
1789         if (req->oldptr == NULL) {
1790                 n = V_tcbinfo.ipi_count +
1791                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1792                 n += imax(n / 8, 10);
1793                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1794                 return (0);
1795         }
1796
1797         if (req->newptr != NULL)
1798                 return (EPERM);
1799
1800         /*
1801          * OK, now we're committed to doing something.
1802          */
1803         INP_LIST_RLOCK(&V_tcbinfo);
1804         gencnt = V_tcbinfo.ipi_gencnt;
1805         n = V_tcbinfo.ipi_count;
1806         INP_LIST_RUNLOCK(&V_tcbinfo);
1807
1808         m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1809
1810         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1811                 + (n + m) * sizeof(struct xtcpcb));
1812         if (error != 0)
1813                 return (error);
1814
1815         xig.xig_len = sizeof xig;
1816         xig.xig_count = n + m;
1817         xig.xig_gen = gencnt;
1818         xig.xig_sogen = so_gencnt;
1819         error = SYSCTL_OUT(req, &xig, sizeof xig);
1820         if (error)
1821                 return (error);
1822
1823         error = syncache_pcblist(req, m, &pcb_count);
1824         if (error)
1825                 return (error);
1826
1827         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1828
1829         INP_INFO_WLOCK(&V_tcbinfo);
1830         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1831             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1832                 INP_WLOCK(inp);
1833                 if (inp->inp_gencnt <= gencnt) {
1834                         /*
1835                          * XXX: This use of cr_cansee(), introduced with
1836                          * TCP state changes, is not quite right, but for
1837                          * now, better than nothing.
1838                          */
1839                         if (inp->inp_flags & INP_TIMEWAIT) {
1840                                 if (intotw(inp) != NULL)
1841                                         error = cr_cansee(req->td->td_ucred,
1842                                             intotw(inp)->tw_cred);
1843                                 else
1844                                         error = EINVAL; /* Skip this inp. */
1845                         } else
1846                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
1847                         if (error == 0) {
1848                                 in_pcbref(inp);
1849                                 inp_list[i++] = inp;
1850                         }
1851                 }
1852                 INP_WUNLOCK(inp);
1853         }
1854         INP_INFO_WUNLOCK(&V_tcbinfo);
1855         n = i;
1856
1857         error = 0;
1858         for (i = 0; i < n; i++) {
1859                 inp = inp_list[i];
1860                 INP_RLOCK(inp);
1861                 if (inp->inp_gencnt <= gencnt) {
1862                         struct xtcpcb xt;
1863
1864                         tcp_inptoxtp(inp, &xt);
1865                         INP_RUNLOCK(inp);
1866                         error = SYSCTL_OUT(req, &xt, sizeof xt);
1867                 } else
1868                         INP_RUNLOCK(inp);
1869         }
1870         INP_INFO_RLOCK(&V_tcbinfo);
1871         for (i = 0; i < n; i++) {
1872                 inp = inp_list[i];
1873                 INP_RLOCK(inp);
1874                 if (!in_pcbrele_rlocked(inp))
1875                         INP_RUNLOCK(inp);
1876         }
1877         INP_INFO_RUNLOCK(&V_tcbinfo);
1878
1879         if (!error) {
1880                 /*
1881                  * Give the user an updated idea of our state.
1882                  * If the generation differs from what we told
1883                  * her before, she knows that something happened
1884                  * while we were processing this request, and it
1885                  * might be necessary to retry.
1886                  */
1887                 INP_LIST_RLOCK(&V_tcbinfo);
1888                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1889                 xig.xig_sogen = so_gencnt;
1890                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1891                 INP_LIST_RUNLOCK(&V_tcbinfo);
1892                 error = SYSCTL_OUT(req, &xig, sizeof xig);
1893         }
1894         free(inp_list, M_TEMP);
1895         return (error);
1896 }
1897
1898 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1899     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1900     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1901
1902 #ifdef INET
1903 static int
1904 tcp_getcred(SYSCTL_HANDLER_ARGS)
1905 {
1906         struct xucred xuc;
1907         struct sockaddr_in addrs[2];
1908         struct inpcb *inp;
1909         int error;
1910
1911         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1912         if (error)
1913                 return (error);
1914         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1915         if (error)
1916                 return (error);
1917         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1918             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1919         if (inp != NULL) {
1920                 if (inp->inp_socket == NULL)
1921                         error = ENOENT;
1922                 if (error == 0)
1923                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1924                 if (error == 0)
1925                         cru2x(inp->inp_cred, &xuc);
1926                 INP_RUNLOCK(inp);
1927         } else
1928                 error = ENOENT;
1929         if (error == 0)
1930                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1931         return (error);
1932 }
1933
1934 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1935     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1936     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1937 #endif /* INET */
1938
1939 #ifdef INET6
1940 static int
1941 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1942 {
1943         struct xucred xuc;
1944         struct sockaddr_in6 addrs[2];
1945         struct inpcb *inp;
1946         int error;
1947 #ifdef INET
1948         int mapped = 0;
1949 #endif
1950
1951         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1952         if (error)
1953                 return (error);
1954         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1955         if (error)
1956                 return (error);
1957         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1958             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1959                 return (error);
1960         }
1961         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1962 #ifdef INET
1963                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1964                         mapped = 1;
1965                 else
1966 #endif
1967                         return (EINVAL);
1968         }
1969
1970 #ifdef INET
1971         if (mapped == 1)
1972                 inp = in_pcblookup(&V_tcbinfo,
1973                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1974                         addrs[1].sin6_port,
1975                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1976                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1977         else
1978 #endif
1979                 inp = in6_pcblookup(&V_tcbinfo,
1980                         &addrs[1].sin6_addr, addrs[1].sin6_port,
1981                         &addrs[0].sin6_addr, addrs[0].sin6_port,
1982                         INPLOOKUP_RLOCKPCB, NULL);
1983         if (inp != NULL) {
1984                 if (inp->inp_socket == NULL)
1985                         error = ENOENT;
1986                 if (error == 0)
1987                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1988                 if (error == 0)
1989                         cru2x(inp->inp_cred, &xuc);
1990                 INP_RUNLOCK(inp);
1991         } else
1992                 error = ENOENT;
1993         if (error == 0)
1994                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1995         return (error);
1996 }
1997
1998 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1999     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
2000     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
2001 #endif /* INET6 */
2002
2003
2004 #ifdef INET
2005 void
2006 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2007 {
2008         struct ip *ip = vip;
2009         struct tcphdr *th;
2010         struct in_addr faddr;
2011         struct inpcb *inp;
2012         struct tcpcb *tp;
2013         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2014         struct icmp *icp;
2015         struct in_conninfo inc;
2016         tcp_seq icmp_tcp_seq;
2017         int mtu;
2018
2019         faddr = ((struct sockaddr_in *)sa)->sin_addr;
2020         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2021                 return;
2022
2023         if (cmd == PRC_MSGSIZE)
2024                 notify = tcp_mtudisc_notify;
2025         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2026                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2027                 cmd == PRC_TIMXCEED_INTRANS) && ip)
2028                 notify = tcp_drop_syn_sent;
2029
2030         /*
2031          * Hostdead is ugly because it goes linearly through all PCBs.
2032          * XXX: We never get this from ICMP, otherwise it makes an
2033          * excellent DoS attack on machines with many connections.
2034          */
2035         else if (cmd == PRC_HOSTDEAD)
2036                 ip = NULL;
2037         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2038                 return;
2039
2040         if (ip == NULL) {
2041                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2042                 return;
2043         }
2044
2045         icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2046         th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2047         INP_INFO_RLOCK(&V_tcbinfo);
2048         inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2049             th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2050         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2051                 /* signal EHOSTDOWN, as it flushes the cached route */
2052                 inp = (*notify)(inp, EHOSTDOWN);
2053                 goto out;
2054         }
2055         icmp_tcp_seq = th->th_seq;
2056         if (inp != NULL)  {
2057                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2058                     !(inp->inp_flags & INP_DROPPED) &&
2059                     !(inp->inp_socket == NULL)) {
2060                         tp = intotcpcb(inp);
2061                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2062                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2063                                 if (cmd == PRC_MSGSIZE) {
2064                                         /*
2065                                          * MTU discovery:
2066                                          * If we got a needfrag set the MTU
2067                                          * in the route to the suggested new
2068                                          * value (if given) and then notify.
2069                                          */
2070                                         mtu = ntohs(icp->icmp_nextmtu);
2071                                         /*
2072                                          * If no alternative MTU was
2073                                          * proposed, try the next smaller
2074                                          * one.
2075                                          */
2076                                         if (!mtu)
2077                                                 mtu = ip_next_mtu(
2078                                                     ntohs(ip->ip_len), 1);
2079                                         if (mtu < V_tcp_minmss +
2080                                             sizeof(struct tcpiphdr))
2081                                                 mtu = V_tcp_minmss +
2082                                                     sizeof(struct tcpiphdr);
2083                                         /*
2084                                          * Only process the offered MTU if it
2085                                          * is smaller than the current one.
2086                                          */
2087                                         if (mtu < tp->t_maxseg +
2088                                             sizeof(struct tcpiphdr)) {
2089                                                 bzero(&inc, sizeof(inc));
2090                                                 inc.inc_faddr = faddr;
2091                                                 inc.inc_fibnum =
2092                                                     inp->inp_inc.inc_fibnum;
2093                                                 tcp_hc_updatemtu(&inc, mtu);
2094                                                 tcp_mtudisc(inp, mtu);
2095                                         }
2096                                 } else
2097                                         inp = (*notify)(inp,
2098                                             inetctlerrmap[cmd]);
2099                         }
2100                 }
2101         } else {
2102                 bzero(&inc, sizeof(inc));
2103                 inc.inc_fport = th->th_dport;
2104                 inc.inc_lport = th->th_sport;
2105                 inc.inc_faddr = faddr;
2106                 inc.inc_laddr = ip->ip_src;
2107                 syncache_unreach(&inc, icmp_tcp_seq);
2108         }
2109 out:
2110         if (inp != NULL)
2111                 INP_WUNLOCK(inp);
2112         INP_INFO_RUNLOCK(&V_tcbinfo);
2113 }
2114 #endif /* INET */
2115
2116 #ifdef INET6
2117 void
2118 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2119 {
2120         struct in6_addr *dst;
2121         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2122         struct ip6_hdr *ip6;
2123         struct mbuf *m;
2124         struct inpcb *inp;
2125         struct tcpcb *tp;
2126         struct icmp6_hdr *icmp6;
2127         struct ip6ctlparam *ip6cp = NULL;
2128         const struct sockaddr_in6 *sa6_src = NULL;
2129         struct in_conninfo inc;
2130         struct tcp_ports {
2131                 uint16_t th_sport;
2132                 uint16_t th_dport;
2133         } t_ports;
2134         tcp_seq icmp_tcp_seq;
2135         unsigned int mtu;
2136         unsigned int off;
2137
2138         if (sa->sa_family != AF_INET6 ||
2139             sa->sa_len != sizeof(struct sockaddr_in6))
2140                 return;
2141
2142         /* if the parameter is from icmp6, decode it. */
2143         if (d != NULL) {
2144                 ip6cp = (struct ip6ctlparam *)d;
2145                 icmp6 = ip6cp->ip6c_icmp6;
2146                 m = ip6cp->ip6c_m;
2147                 ip6 = ip6cp->ip6c_ip6;
2148                 off = ip6cp->ip6c_off;
2149                 sa6_src = ip6cp->ip6c_src;
2150                 dst = ip6cp->ip6c_finaldst;
2151         } else {
2152                 m = NULL;
2153                 ip6 = NULL;
2154                 off = 0;        /* fool gcc */
2155                 sa6_src = &sa6_any;
2156                 dst = NULL;
2157         }
2158
2159         if (cmd == PRC_MSGSIZE)
2160                 notify = tcp_mtudisc_notify;
2161         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2162                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2163                 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2164                 notify = tcp_drop_syn_sent;
2165
2166         /*
2167          * Hostdead is ugly because it goes linearly through all PCBs.
2168          * XXX: We never get this from ICMP, otherwise it makes an
2169          * excellent DoS attack on machines with many connections.
2170          */
2171         else if (cmd == PRC_HOSTDEAD)
2172                 ip6 = NULL;
2173         else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2174                 return;
2175
2176         if (ip6 == NULL) {
2177                 in6_pcbnotify(&V_tcbinfo, sa, 0,
2178                               (const struct sockaddr *)sa6_src,
2179                               0, cmd, NULL, notify);
2180                 return;
2181         }
2182
2183         /* Check if we can safely get the ports from the tcp hdr */
2184         if (m == NULL ||
2185             (m->m_pkthdr.len <
2186                 (int32_t) (off + sizeof(struct tcp_ports)))) {
2187                 return;
2188         }
2189         bzero(&t_ports, sizeof(struct tcp_ports));
2190         m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2191         INP_INFO_RLOCK(&V_tcbinfo);
2192         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2193             &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2194         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2195                 /* signal EHOSTDOWN, as it flushes the cached route */
2196                 inp = (*notify)(inp, EHOSTDOWN);
2197                 goto out;
2198         }
2199         off += sizeof(struct tcp_ports);
2200         if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2201                 goto out;
2202         }
2203         m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2204         if (inp != NULL)  {
2205                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2206                     !(inp->inp_flags & INP_DROPPED) &&
2207                     !(inp->inp_socket == NULL)) {
2208                         tp = intotcpcb(inp);
2209                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2210                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2211                                 if (cmd == PRC_MSGSIZE) {
2212                                         /*
2213                                          * MTU discovery:
2214                                          * If we got a needfrag set the MTU
2215                                          * in the route to the suggested new
2216                                          * value (if given) and then notify.
2217                                          */
2218                                         mtu = ntohl(icmp6->icmp6_mtu);
2219                                         /*
2220                                          * If no alternative MTU was
2221                                          * proposed, or the proposed
2222                                          * MTU was too small, set to
2223                                          * the min.
2224                                          */
2225                                         if (mtu < IPV6_MMTU)
2226                                                 mtu = IPV6_MMTU - 8;
2227                                         bzero(&inc, sizeof(inc));
2228                                         inc.inc_fibnum = M_GETFIB(m);
2229                                         inc.inc_flags |= INC_ISIPV6;
2230                                         inc.inc6_faddr = *dst;
2231                                         if (in6_setscope(&inc.inc6_faddr,
2232                                                 m->m_pkthdr.rcvif, NULL))
2233                                                 goto out;
2234                                         /*
2235                                          * Only process the offered MTU if it
2236                                          * is smaller than the current one.
2237                                          */
2238                                         if (mtu < tp->t_maxseg +
2239                                             sizeof (struct tcphdr) +
2240                                             sizeof (struct ip6_hdr)) {
2241                                                 tcp_hc_updatemtu(&inc, mtu);
2242                                                 tcp_mtudisc(inp, mtu);
2243                                                 ICMP6STAT_INC(icp6s_pmtuchg);
2244                                         }
2245                                 } else
2246                                         inp = (*notify)(inp,
2247                                             inet6ctlerrmap[cmd]);
2248                         }
2249                 }
2250         } else {
2251                 bzero(&inc, sizeof(inc));
2252                 inc.inc_fibnum = M_GETFIB(m);
2253                 inc.inc_flags |= INC_ISIPV6;
2254                 inc.inc_fport = t_ports.th_dport;
2255                 inc.inc_lport = t_ports.th_sport;
2256                 inc.inc6_faddr = *dst;
2257                 inc.inc6_laddr = ip6->ip6_src;
2258                 syncache_unreach(&inc, icmp_tcp_seq);
2259         }
2260 out:
2261         if (inp != NULL)
2262                 INP_WUNLOCK(inp);
2263         INP_INFO_RUNLOCK(&V_tcbinfo);
2264 }
2265 #endif /* INET6 */
2266
2267
2268 /*
2269  * Following is where TCP initial sequence number generation occurs.
2270  *
2271  * There are two places where we must use initial sequence numbers:
2272  * 1.  In SYN-ACK packets.
2273  * 2.  In SYN packets.
2274  *
2275  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2276  * tcp_syncache.c for details.
2277  *
2278  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2279  * depends on this property.  In addition, these ISNs should be
2280  * unguessable so as to prevent connection hijacking.  To satisfy
2281  * the requirements of this situation, the algorithm outlined in
2282  * RFC 1948 is used, with only small modifications.
2283  *
2284  * Implementation details:
2285  *
2286  * Time is based off the system timer, and is corrected so that it
2287  * increases by one megabyte per second.  This allows for proper
2288  * recycling on high speed LANs while still leaving over an hour
2289  * before rollover.
2290  *
2291  * As reading the *exact* system time is too expensive to be done
2292  * whenever setting up a TCP connection, we increment the time
2293  * offset in two ways.  First, a small random positive increment
2294  * is added to isn_offset for each connection that is set up.
2295  * Second, the function tcp_isn_tick fires once per clock tick
2296  * and increments isn_offset as necessary so that sequence numbers
2297  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2298  * random positive increments serve only to ensure that the same
2299  * exact sequence number is never sent out twice (as could otherwise
2300  * happen when a port is recycled in less than the system tick
2301  * interval.)
2302  *
2303  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2304  * between seeding of isn_secret.  This is normally set to zero,
2305  * as reseeding should not be necessary.
2306  *
2307  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2308  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2309  * general, this means holding an exclusive (write) lock.
2310  */
2311
2312 #define ISN_BYTES_PER_SECOND 1048576
2313 #define ISN_STATIC_INCREMENT 4096
2314 #define ISN_RANDOM_INCREMENT (4096 - 1)
2315
2316 static VNET_DEFINE(u_char, isn_secret[32]);
2317 static VNET_DEFINE(int, isn_last);
2318 static VNET_DEFINE(int, isn_last_reseed);
2319 static VNET_DEFINE(u_int32_t, isn_offset);
2320 static VNET_DEFINE(u_int32_t, isn_offset_old);
2321
2322 #define V_isn_secret                    VNET(isn_secret)
2323 #define V_isn_last                      VNET(isn_last)
2324 #define V_isn_last_reseed               VNET(isn_last_reseed)
2325 #define V_isn_offset                    VNET(isn_offset)
2326 #define V_isn_offset_old                VNET(isn_offset_old)
2327
2328 tcp_seq
2329 tcp_new_isn(struct tcpcb *tp)
2330 {
2331         MD5_CTX isn_ctx;
2332         u_int32_t md5_buffer[4];
2333         tcp_seq new_isn;
2334         u_int32_t projected_offset;
2335
2336         INP_WLOCK_ASSERT(tp->t_inpcb);
2337
2338         ISN_LOCK();
2339         /* Seed if this is the first use, reseed if requested. */
2340         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2341              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2342                 < (u_int)ticks))) {
2343                 read_random(&V_isn_secret, sizeof(V_isn_secret));
2344                 V_isn_last_reseed = ticks;
2345         }
2346
2347         /* Compute the md5 hash and return the ISN. */
2348         MD5Init(&isn_ctx);
2349         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2350         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2351 #ifdef INET6
2352         if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2353                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2354                           sizeof(struct in6_addr));
2355                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2356                           sizeof(struct in6_addr));
2357         } else
2358 #endif
2359         {
2360                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2361                           sizeof(struct in_addr));
2362                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2363                           sizeof(struct in_addr));
2364         }
2365         MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2366         MD5Final((u_char *) &md5_buffer, &isn_ctx);
2367         new_isn = (tcp_seq) md5_buffer[0];
2368         V_isn_offset += ISN_STATIC_INCREMENT +
2369                 (arc4random() & ISN_RANDOM_INCREMENT);
2370         if (ticks != V_isn_last) {
2371                 projected_offset = V_isn_offset_old +
2372                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2373                 if (SEQ_GT(projected_offset, V_isn_offset))
2374                         V_isn_offset = projected_offset;
2375                 V_isn_offset_old = V_isn_offset;
2376                 V_isn_last = ticks;
2377         }
2378         new_isn += V_isn_offset;
2379         ISN_UNLOCK();
2380         return (new_isn);
2381 }
2382
2383 /*
2384  * When a specific ICMP unreachable message is received and the
2385  * connection state is SYN-SENT, drop the connection.  This behavior
2386  * is controlled by the icmp_may_rst sysctl.
2387  */
2388 struct inpcb *
2389 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2390 {
2391         struct tcpcb *tp;
2392
2393         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2394         INP_WLOCK_ASSERT(inp);
2395
2396         if ((inp->inp_flags & INP_TIMEWAIT) ||
2397             (inp->inp_flags & INP_DROPPED))
2398                 return (inp);
2399
2400         tp = intotcpcb(inp);
2401         if (tp->t_state != TCPS_SYN_SENT)
2402                 return (inp);
2403
2404         tp = tcp_drop(tp, errno);
2405         if (tp != NULL)
2406                 return (inp);
2407         else
2408                 return (NULL);
2409 }
2410
2411 /*
2412  * When `need fragmentation' ICMP is received, update our idea of the MSS
2413  * based on the new value. Also nudge TCP to send something, since we
2414  * know the packet we just sent was dropped.
2415  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2416  */
2417 static struct inpcb *
2418 tcp_mtudisc_notify(struct inpcb *inp, int error)
2419 {
2420
2421         tcp_mtudisc(inp, -1);
2422         return (inp);
2423 }
2424
2425 static void
2426 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2427 {
2428         struct tcpcb *tp;
2429         struct socket *so;
2430
2431         INP_WLOCK_ASSERT(inp);
2432         if ((inp->inp_flags & INP_TIMEWAIT) ||
2433             (inp->inp_flags & INP_DROPPED))
2434                 return;
2435
2436         tp = intotcpcb(inp);
2437         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2438
2439         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2440
2441         so = inp->inp_socket;
2442         SOCKBUF_LOCK(&so->so_snd);
2443         /* If the mss is larger than the socket buffer, decrease the mss. */
2444         if (so->so_snd.sb_hiwat < tp->t_maxseg)
2445                 tp->t_maxseg = so->so_snd.sb_hiwat;
2446         SOCKBUF_UNLOCK(&so->so_snd);
2447
2448         TCPSTAT_INC(tcps_mturesent);
2449         tp->t_rtttime = 0;
2450         tp->snd_nxt = tp->snd_una;
2451         tcp_free_sackholes(tp);
2452         tp->snd_recover = tp->snd_max;
2453         if (tp->t_flags & TF_SACK_PERMIT)
2454                 EXIT_FASTRECOVERY(tp->t_flags);
2455         tp->t_fb->tfb_tcp_output(tp);
2456 }
2457
2458 #ifdef INET
2459 /*
2460  * Look-up the routing entry to the peer of this inpcb.  If no route
2461  * is found and it cannot be allocated, then return 0.  This routine
2462  * is called by TCP routines that access the rmx structure and by
2463  * tcp_mss_update to get the peer/interface MTU.
2464  */
2465 uint32_t
2466 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2467 {
2468         struct nhop4_extended nh4;
2469         struct ifnet *ifp;
2470         uint32_t maxmtu = 0;
2471
2472         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2473
2474         if (inc->inc_faddr.s_addr != INADDR_ANY) {
2475
2476                 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2477                     NHR_REF, 0, &nh4) != 0)
2478                         return (0);
2479
2480                 ifp = nh4.nh_ifp;
2481                 maxmtu = nh4.nh_mtu;
2482
2483                 /* Report additional interface capabilities. */
2484                 if (cap != NULL) {
2485                         if (ifp->if_capenable & IFCAP_TSO4 &&
2486                             ifp->if_hwassist & CSUM_TSO) {
2487                                 cap->ifcap |= CSUM_TSO;
2488                                 cap->tsomax = ifp->if_hw_tsomax;
2489                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2490                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2491                         }
2492                 }
2493                 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2494         }
2495         return (maxmtu);
2496 }
2497 #endif /* INET */
2498
2499 #ifdef INET6
2500 uint32_t
2501 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2502 {
2503         struct nhop6_extended nh6;
2504         struct in6_addr dst6;
2505         uint32_t scopeid;
2506         struct ifnet *ifp;
2507         uint32_t maxmtu = 0;
2508
2509         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2510
2511         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2512                 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2513                 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2514                     0, &nh6) != 0)
2515                         return (0);
2516
2517                 ifp = nh6.nh_ifp;
2518                 maxmtu = nh6.nh_mtu;
2519
2520                 /* Report additional interface capabilities. */
2521                 if (cap != NULL) {
2522                         if (ifp->if_capenable & IFCAP_TSO6 &&
2523                             ifp->if_hwassist & CSUM_TSO) {
2524                                 cap->ifcap |= CSUM_TSO;
2525                                 cap->tsomax = ifp->if_hw_tsomax;
2526                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2527                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2528                         }
2529                 }
2530                 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2531         }
2532
2533         return (maxmtu);
2534 }
2535 #endif /* INET6 */
2536
2537 /*
2538  * Calculate effective SMSS per RFC5681 definition for a given TCP
2539  * connection at its current state, taking into account SACK and etc.
2540  */
2541 u_int
2542 tcp_maxseg(const struct tcpcb *tp)
2543 {
2544         u_int optlen;
2545
2546         if (tp->t_flags & TF_NOOPT)
2547                 return (tp->t_maxseg);
2548
2549         /*
2550          * Here we have a simplified code from tcp_addoptions(),
2551          * without a proper loop, and having most of paddings hardcoded.
2552          * We might make mistakes with padding here in some edge cases,
2553          * but this is harmless, since result of tcp_maxseg() is used
2554          * only in cwnd and ssthresh estimations.
2555          */
2556 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
2557         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2558                 if (tp->t_flags & TF_RCVD_TSTMP)
2559                         optlen = TCPOLEN_TSTAMP_APPA;
2560                 else
2561                         optlen = 0;
2562 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2563                 if (tp->t_flags & TF_SIGNATURE)
2564                         optlen += PAD(TCPOLEN_SIGNATURE);
2565 #endif
2566                 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2567                         optlen += TCPOLEN_SACKHDR;
2568                         optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2569                         optlen = PAD(optlen);
2570                 }
2571         } else {
2572                 if (tp->t_flags & TF_REQ_TSTMP)
2573                         optlen = TCPOLEN_TSTAMP_APPA;
2574                 else
2575                         optlen = PAD(TCPOLEN_MAXSEG);
2576                 if (tp->t_flags & TF_REQ_SCALE)
2577                         optlen += PAD(TCPOLEN_WINDOW);
2578 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2579                 if (tp->t_flags & TF_SIGNATURE)
2580                         optlen += PAD(TCPOLEN_SIGNATURE);
2581 #endif
2582                 if (tp->t_flags & TF_SACK_PERMIT)
2583                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
2584         }
2585 #undef PAD
2586         optlen = min(optlen, TCP_MAXOLEN);
2587         return (tp->t_maxseg - optlen);
2588 }
2589
2590 static int
2591 sysctl_drop(SYSCTL_HANDLER_ARGS)
2592 {
2593         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2594         struct sockaddr_storage addrs[2];
2595         struct inpcb *inp;
2596         struct tcpcb *tp;
2597         struct tcptw *tw;
2598         struct sockaddr_in *fin, *lin;
2599 #ifdef INET6
2600         struct sockaddr_in6 *fin6, *lin6;
2601 #endif
2602         int error;
2603
2604         inp = NULL;
2605         fin = lin = NULL;
2606 #ifdef INET6
2607         fin6 = lin6 = NULL;
2608 #endif
2609         error = 0;
2610
2611         if (req->oldptr != NULL || req->oldlen != 0)
2612                 return (EINVAL);
2613         if (req->newptr == NULL)
2614                 return (EPERM);
2615         if (req->newlen < sizeof(addrs))
2616                 return (ENOMEM);
2617         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2618         if (error)
2619                 return (error);
2620
2621         switch (addrs[0].ss_family) {
2622 #ifdef INET6
2623         case AF_INET6:
2624                 fin6 = (struct sockaddr_in6 *)&addrs[0];
2625                 lin6 = (struct sockaddr_in6 *)&addrs[1];
2626                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2627                     lin6->sin6_len != sizeof(struct sockaddr_in6))
2628                         return (EINVAL);
2629                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2630                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2631                                 return (EINVAL);
2632                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2633                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2634                         fin = (struct sockaddr_in *)&addrs[0];
2635                         lin = (struct sockaddr_in *)&addrs[1];
2636                         break;
2637                 }
2638                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2639                 if (error)
2640                         return (error);
2641                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2642                 if (error)
2643                         return (error);
2644                 break;
2645 #endif
2646 #ifdef INET
2647         case AF_INET:
2648                 fin = (struct sockaddr_in *)&addrs[0];
2649                 lin = (struct sockaddr_in *)&addrs[1];
2650                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2651                     lin->sin_len != sizeof(struct sockaddr_in))
2652                         return (EINVAL);
2653                 break;
2654 #endif
2655         default:
2656                 return (EINVAL);
2657         }
2658         INP_INFO_RLOCK(&V_tcbinfo);
2659         switch (addrs[0].ss_family) {
2660 #ifdef INET6
2661         case AF_INET6:
2662                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2663                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2664                     INPLOOKUP_WLOCKPCB, NULL);
2665                 break;
2666 #endif
2667 #ifdef INET
2668         case AF_INET:
2669                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2670                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2671                 break;
2672 #endif
2673         }
2674         if (inp != NULL) {
2675                 if (inp->inp_flags & INP_TIMEWAIT) {
2676                         /*
2677                          * XXXRW: There currently exists a state where an
2678                          * inpcb is present, but its timewait state has been
2679                          * discarded.  For now, don't allow dropping of this
2680                          * type of inpcb.
2681                          */
2682                         tw = intotw(inp);
2683                         if (tw != NULL)
2684                                 tcp_twclose(tw, 0);
2685                         else
2686                                 INP_WUNLOCK(inp);
2687                 } else if (!(inp->inp_flags & INP_DROPPED) &&
2688                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2689                         tp = intotcpcb(inp);
2690                         tp = tcp_drop(tp, ECONNABORTED);
2691                         if (tp != NULL)
2692                                 INP_WUNLOCK(inp);
2693                 } else
2694                         INP_WUNLOCK(inp);
2695         } else
2696                 error = ESRCH;
2697         INP_INFO_RUNLOCK(&V_tcbinfo);
2698         return (error);
2699 }
2700
2701 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2702     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2703     0, sysctl_drop, "", "Drop TCP connection");
2704
2705 /*
2706  * Generate a standardized TCP log line for use throughout the
2707  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2708  * allow use in the interrupt context.
2709  *
2710  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2711  * NB: The function may return NULL if memory allocation failed.
2712  *
2713  * Due to header inclusion and ordering limitations the struct ip
2714  * and ip6_hdr pointers have to be passed as void pointers.
2715  */
2716 char *
2717 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2718     const void *ip6hdr)
2719 {
2720
2721         /* Is logging enabled? */
2722         if (tcp_log_in_vain == 0)
2723                 return (NULL);
2724
2725         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2726 }
2727
2728 char *
2729 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2730     const void *ip6hdr)
2731 {
2732
2733         /* Is logging enabled? */
2734         if (tcp_log_debug == 0)
2735                 return (NULL);
2736
2737         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2738 }
2739
2740 static char *
2741 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2742     const void *ip6hdr)
2743 {
2744         char *s, *sp;
2745         size_t size;
2746         struct ip *ip;
2747 #ifdef INET6
2748         const struct ip6_hdr *ip6;
2749
2750         ip6 = (const struct ip6_hdr *)ip6hdr;
2751 #endif /* INET6 */
2752         ip = (struct ip *)ip4hdr;
2753
2754         /*
2755          * The log line looks like this:
2756          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2757          */
2758         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2759             sizeof(PRINT_TH_FLAGS) + 1 +
2760 #ifdef INET6
2761             2 * INET6_ADDRSTRLEN;
2762 #else
2763             2 * INET_ADDRSTRLEN;
2764 #endif /* INET6 */
2765
2766         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2767         if (s == NULL)
2768                 return (NULL);
2769
2770         strcat(s, "TCP: [");
2771         sp = s + strlen(s);
2772
2773         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2774                 inet_ntoa_r(inc->inc_faddr, sp);
2775                 sp = s + strlen(s);
2776                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2777                 sp = s + strlen(s);
2778                 inet_ntoa_r(inc->inc_laddr, sp);
2779                 sp = s + strlen(s);
2780                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2781 #ifdef INET6
2782         } else if (inc) {
2783                 ip6_sprintf(sp, &inc->inc6_faddr);
2784                 sp = s + strlen(s);
2785                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2786                 sp = s + strlen(s);
2787                 ip6_sprintf(sp, &inc->inc6_laddr);
2788                 sp = s + strlen(s);
2789                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2790         } else if (ip6 && th) {
2791                 ip6_sprintf(sp, &ip6->ip6_src);
2792                 sp = s + strlen(s);
2793                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2794                 sp = s + strlen(s);
2795                 ip6_sprintf(sp, &ip6->ip6_dst);
2796                 sp = s + strlen(s);
2797                 sprintf(sp, "]:%i", ntohs(th->th_dport));
2798 #endif /* INET6 */
2799 #ifdef INET
2800         } else if (ip && th) {
2801                 inet_ntoa_r(ip->ip_src, sp);
2802                 sp = s + strlen(s);
2803                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2804                 sp = s + strlen(s);
2805                 inet_ntoa_r(ip->ip_dst, sp);
2806                 sp = s + strlen(s);
2807                 sprintf(sp, "]:%i", ntohs(th->th_dport));
2808 #endif /* INET */
2809         } else {
2810                 free(s, M_TCPLOG);
2811                 return (NULL);
2812         }
2813         sp = s + strlen(s);
2814         if (th)
2815                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2816         if (*(s + size - 1) != '\0')
2817                 panic("%s: string too long", __func__);
2818         return (s);
2819 }
2820
2821 /*
2822  * A subroutine which makes it easy to track TCP state changes with DTrace.
2823  * This function shouldn't be called for t_state initializations that don't
2824  * correspond to actual TCP state transitions.
2825  */
2826 void
2827 tcp_state_change(struct tcpcb *tp, int newstate)
2828 {
2829 #if defined(KDTRACE_HOOKS)
2830         int pstate = tp->t_state;
2831 #endif
2832
2833         TCPSTATES_DEC(tp->t_state);
2834         TCPSTATES_INC(newstate);
2835         tp->t_state = newstate;
2836         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2837 }
2838
2839 /*
2840  * Create an external-format (``xtcpcb'') structure using the information in
2841  * the kernel-format tcpcb structure pointed to by tp.  This is done to
2842  * reduce the spew of irrelevant information over this interface, to isolate
2843  * user code from changes in the kernel structure, and potentially to provide
2844  * information-hiding if we decide that some of this information should be
2845  * hidden from users.
2846  */
2847 void
2848 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
2849 {
2850         struct tcpcb *tp = intotcpcb(inp);
2851         sbintime_t now;
2852
2853         if (inp->inp_flags & INP_TIMEWAIT) {
2854                 bzero(xt, sizeof(struct xtcpcb));
2855                 xt->t_state = TCPS_TIME_WAIT;
2856         } else {
2857                 xt->t_state = tp->t_state;
2858                 xt->t_flags = tp->t_flags;
2859                 xt->t_sndzerowin = tp->t_sndzerowin;
2860                 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
2861                 xt->t_rcvoopack = tp->t_rcvoopack;
2862
2863                 now = getsbinuptime();
2864 #define COPYTIMER(ttt)  do {                                            \
2865                 if (callout_active(&tp->t_timers->ttt))                 \
2866                         xt->ttt = (tp->t_timers->ttt.c_time - now) /    \
2867                             SBT_1MS;                                    \
2868                 else                                                    \
2869                         xt->ttt = 0;                                    \
2870 } while (0)
2871                 COPYTIMER(tt_delack);
2872                 COPYTIMER(tt_rexmt);
2873                 COPYTIMER(tt_persist);
2874                 COPYTIMER(tt_keep);
2875                 COPYTIMER(tt_2msl);
2876 #undef COPYTIMER
2877                 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
2878
2879                 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
2880                     TCP_FUNCTION_NAME_LEN_MAX);
2881         }
2882
2883         xt->xt_len = sizeof(struct xtcpcb);
2884         in_pcbtoxinpcb(inp, &xt->xt_inp);
2885         if (inp->inp_socket == NULL)
2886                 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
2887 }