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  * 4. 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 #include <sys/hhook.h>
  46 #include <sys/kernel.h>
  47 #include <sys/khelp.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/jail.h>
  50 #include <sys/malloc.h>
  51 #include <sys/refcount.h>
  52 #include <sys/mbuf.h>
  53 #ifdef INET6
  54 #include <sys/domain.h>
  55 #endif
  56 #include <sys/priv.h>
  57 #include <sys/proc.h>
  58 #include <sys/sdt.h>
  59 #include <sys/socket.h>
  60 #include <sys/socketvar.h>
  61 #include <sys/protosw.h>
  62 #include <sys/random.h>
  63
  64 #include <vm/uma.h>
  65
  66 #include <net/route.h>
  67 #include <net/if.h>
  68 #include <net/if_var.h>
  69 #include <net/vnet.h>
  70
  71 #include <netinet/in.h>
  72 #include <netinet/in_fib.h>
  73 #include <netinet/in_kdtrace.h>
  74 #include <netinet/in_pcb.h>
  75 #include <netinet/in_systm.h>
  76 #include <netinet/in_var.h>
  77 #include <netinet/ip.h>
  78 #include <netinet/ip_icmp.h>
  79 #include <netinet/ip_var.h>
  80 #ifdef INET6
  81 #include <netinet/icmp6.h>
  82 #include <netinet/ip6.h>
  83 #include <netinet6/in6_fib.h>
  84 #include <netinet6/in6_pcb.h>
  85 #include <netinet6/ip6_var.h>
  86 #include <netinet6/scope6_var.h>
  87 #include <netinet6/nd6.h>
  88 #endif
  89
  90 #ifdef TCP_RFC7413
  91 #include <netinet/tcp_fastopen.h>
  92 #endif
  93 #include <netinet/tcp.h>
  94 #include <netinet/tcp_fsm.h>
  95 #include <netinet/tcp_seq.h>
  96 #include <netinet/tcp_timer.h>
  97 #include <netinet/tcp_var.h>
  98 #include <netinet/tcp_syncache.h>
  99 #include <netinet/cc/cc.h>
 100 #ifdef INET6
 101 #include <netinet6/tcp6_var.h>
 102 #endif
 103 #include <netinet/tcpip.h>
 104 #ifdef TCPPCAP
 105 #include <netinet/tcp_pcap.h>
 106 #endif
 107 #ifdef TCPDEBUG
 108 #include <netinet/tcp_debug.h>
 109 #endif
 110 #ifdef INET6
 111 #include <netinet6/ip6protosw.h>
 112 #endif
 113 #ifdef TCP_OFFLOAD
 114 #include <netinet/tcp_offload.h>
 115 #endif
 116
 117 #ifdef IPSEC
 118 #include <netipsec/ipsec.h>
 119 #include <netipsec/xform.h>
 120 #ifdef INET6
 121 #include <netipsec/ipsec6.h>
 122 #endif
 123 #include <netipsec/key.h>
 124 #include <sys/syslog.h>
 125 #endif /*IPSEC*/
 126
 127 #include <machine/in_cksum.h>
 128 #include <sys/md5.h>
 129
 130 #include <security/mac/mac_framework.h>
 131
 132 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
 133 #ifdef INET6
 134 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
 135 #endif
 136
 137 struct rwlock tcp_function_lock;
 138
 139 static int
 140 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
 141 {
 142         int error, new;
 143
 144         new = V_tcp_mssdflt;
 145         error = sysctl_handle_int(oidp, &new, 0, req);
 146         if (error == 0 && req->newptr) {
 147                 if (new < TCP_MINMSS)
 148                         error = EINVAL;
 149                 else
 150                         V_tcp_mssdflt = new;
 151         }
 152         return (error);
 153 }
 154
 155 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
 156     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
 157     &sysctl_net_inet_tcp_mss_check, "I",
 158     "Default TCP Maximum Segment Size");
 159
 160 #ifdef INET6
 161 static int
 162 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
 163 {
 164         int error, new;
 165
 166         new = V_tcp_v6mssdflt;
 167         error = sysctl_handle_int(oidp, &new, 0, req);
 168         if (error == 0 && req->newptr) {
 169                 if (new < TCP_MINMSS)
 170                         error = EINVAL;
 171                 else
 172                         V_tcp_v6mssdflt = new;
 173         }
 174         return (error);
 175 }
 176
 177 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
 178     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
 179     &sysctl_net_inet_tcp_mss_v6_check, "I",
 180    "Default TCP Maximum Segment Size for IPv6");
 181 #endif /* INET6 */
 182
 183 /*
 184  * Minimum MSS we accept and use. This prevents DoS attacks where
 185  * we are forced to a ridiculous low MSS like 20 and send hundreds
 186  * of packets instead of one. The effect scales with the available
 187  * bandwidth and quickly saturates the CPU and network interface
 188  * with packet generation and sending. Set to zero to disable MINMSS
 189  * checking. This setting prevents us from sending too small packets.
 190  */
 191 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
 193      &VNET_NAME(tcp_minmss), 0,
 194     "Minimum TCP Maximum Segment Size");
 195
 196 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
 197 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
 198     &VNET_NAME(tcp_do_rfc1323), 0,
 199     "Enable rfc1323 (high performance TCP) extensions");
 200
 201 static int      tcp_log_debug = 0;
 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
 203     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
 204
 205 static int      tcp_tcbhashsize;
 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
 207     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
 208
 209 static int      do_tcpdrain = 1;
 210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
 211     "Enable tcp_drain routine for extra help when low on mbufs");
 212
 213 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
 214     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
 215
 216 static VNET_DEFINE(int, icmp_may_rst) = 1;
 217 #define V_icmp_may_rst                  VNET(icmp_may_rst)
 218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
 219     &VNET_NAME(icmp_may_rst), 0,
 220     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
 221
 222 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
 223 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
 225     &VNET_NAME(tcp_isn_reseed_interval), 0,
 226     "Seconds between reseeding of ISN secret");
 227
 228 static int      tcp_soreceive_stream;
 229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
 230     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
 231
 232 #ifdef TCP_SIGNATURE
 233 static int      tcp_sig_checksigs = 1;
 234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
 235     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
 236 #endif
 237
 238 VNET_DEFINE(uma_zone_t, sack_hole_zone);
 239 #define V_sack_hole_zone                VNET(sack_hole_zone)
 240
 241 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
 242
 243 static struct inpcb *tcp_notify(struct inpcb *, int);
 244 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
 245 static void tcp_mtudisc(struct inpcb *, int);
 246 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
 247                     void *ip4hdr, const void *ip6hdr);
 248
 249
 250 static struct tcp_function_block tcp_def_funcblk = {
 251         "default",
 252         tcp_output,
 253         tcp_do_segment,
 254         tcp_default_ctloutput,
 255         NULL,
 256         NULL,
 257         NULL,
 258         NULL,
 259         NULL,
 260         NULL,
 261         0,
 262         0
 263 };
 264
 265 int t_functions_inited = 0;
 266 struct tcp_funchead t_functions;
 267 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
 268
 269 static void
 270 init_tcp_functions(void)
 271 {
 272         if (t_functions_inited == 0) {
 273                 TAILQ_INIT(&t_functions);
 274                 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
 275                 t_functions_inited = 1;
 276         }
 277 }
 278
 279 static struct tcp_function_block *
 280 find_tcp_functions_locked(struct tcp_function_set *fs)
 281 {
 282         struct tcp_function *f;
 283         struct tcp_function_block *blk=NULL;
 284
 285         TAILQ_FOREACH(f, &t_functions, tf_next) {
 286                 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
 287                         blk = f->tf_fb;
 288                         break;
 289                 }
 290         }
 291         return(blk);
 292 }
 293
 294 static struct tcp_function_block *
 295 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
 296 {
 297         struct tcp_function_block *rblk=NULL;
 298         struct tcp_function *f;
 299
 300         TAILQ_FOREACH(f, &t_functions, tf_next) {
 301                 if (f->tf_fb == blk) {
 302                         rblk = blk;
 303                         if (s) {
 304                                 *s = f;
 305                         }
 306                         break;
 307                 }
 308         }
 309         return (rblk);
 310 }
 311
 312 struct tcp_function_block *
 313 find_and_ref_tcp_functions(struct tcp_function_set *fs)
 314 {
 315         struct tcp_function_block *blk;
 316
 317         rw_rlock(&tcp_function_lock);
 318         blk = find_tcp_functions_locked(fs);
 319         if (blk)
 320                 refcount_acquire(&blk->tfb_refcnt);
 321         rw_runlock(&tcp_function_lock);
 322         return(blk);
 323 }
 324
 325 struct tcp_function_block *
 326 find_and_ref_tcp_fb(struct tcp_function_block *blk)
 327 {
 328         struct tcp_function_block *rblk;
 329
 330         rw_rlock(&tcp_function_lock);
 331         rblk = find_tcp_fb_locked(blk, NULL);
 332         if (rblk)
 333                 refcount_acquire(&rblk->tfb_refcnt);
 334         rw_runlock(&tcp_function_lock);
 335         return(rblk);
 336 }
 337
 338
 339 static int
 340 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
 341 {
 342         int error=ENOENT;
 343         struct tcp_function_set fs;
 344         struct tcp_function_block *blk;
 345
 346         memset(&fs, 0, sizeof(fs));
 347         rw_rlock(&tcp_function_lock);
 348         blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
 349         if (blk) {
 350                 /* Found him */
 351                 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 352                 fs.pcbcnt = blk->tfb_refcnt;
 353         }
 354         rw_runlock(&tcp_function_lock);
 355         error = sysctl_handle_string(oidp, fs.function_set_name,
 356                                      sizeof(fs.function_set_name), req);
 357
 358         /* Check for error or no change */
 359         if (error != 0 || req->newptr == NULL)
 360                 return(error);
 361
 362         rw_wlock(&tcp_function_lock);
 363         blk = find_tcp_functions_locked(&fs);
 364         if ((blk == NULL) ||
 365             (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
 366                 error = ENOENT;
 367                 goto done;
 368         }
 369         tcp_func_set_ptr = blk;
 370 done:
 371         rw_wunlock(&tcp_function_lock);
 372         return (error);
 373 }
 374
 375 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
 376             CTLTYPE_STRING | CTLFLAG_RW,
 377             NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
 378             "Set/get the default TCP functions");
 379
 380 static int
 381 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
 382 {
 383         int error, cnt, linesz;
 384         struct tcp_function *f;
 385         char *buffer, *cp;
 386         size_t bufsz, outsz;
 387
 388         cnt = 0;
 389         rw_rlock(&tcp_function_lock);
 390         TAILQ_FOREACH(f, &t_functions, tf_next) {
 391                 cnt++;
 392         }
 393         rw_runlock(&tcp_function_lock);
 394
 395         bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
 396         buffer = malloc(bufsz, M_TEMP, M_WAITOK);
 397
 398         error = 0;
 399         cp = buffer;
 400
 401         linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
 402         cp += linesz;
 403         bufsz -= linesz;
 404         outsz = linesz;
 405
 406         rw_rlock(&tcp_function_lock);
 407         TAILQ_FOREACH(f, &t_functions, tf_next) {
 408                 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
 409                     f->tf_fb->tfb_tcp_block_name,
 410                     (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
 411                     f->tf_fb->tfb_refcnt);
 412                 if (linesz >= bufsz) {
 413                         error = EOVERFLOW;
 414                         break;
 415                 }
 416                 cp += linesz;
 417                 bufsz -= linesz;
 418                 outsz += linesz;
 419         }
 420         rw_runlock(&tcp_function_lock);
 421         if (error == 0)
 422                 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
 423         free(buffer, M_TEMP);
 424         return (error);
 425 }
 426
 427 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
 428             CTLTYPE_STRING|CTLFLAG_RD,
 429             NULL, 0, sysctl_net_inet_list_available, "A",
 430             "list available TCP Function sets");
 431
 432 /*
 433  * Target size of TCP PCB hash tables. Must be a power of two.
 434  *
 435  * Note that this can be overridden by the kernel environment
 436  * variable net.inet.tcp.tcbhashsize
 437  */
 438 #ifndef TCBHASHSIZE
 439 #define TCBHASHSIZE     0
 440 #endif
 441
 442 /*
 443  * XXX
 444  * Callouts should be moved into struct tcp directly.  They are currently
 445  * separate because the tcpcb structure is exported to userland for sysctl
 446  * parsing purposes, which do not know about callouts.
 447  */
 448 struct tcpcb_mem {
 449         struct  tcpcb           tcb;
 450         struct  tcp_timer       tt;
 451         struct  cc_var          ccv;
 452         struct  osd             osd;
 453 };
 454
 455 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
 456 #define V_tcpcb_zone                    VNET(tcpcb_zone)
 457
 458 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
 459 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
 460
 461 static struct mtx isn_mtx;
 462
 463 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
 464 #define ISN_LOCK()      mtx_lock(&isn_mtx)
 465 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
 466
 467 /*
 468  * TCP initialization.
 469  */
 470 static void
 471 tcp_zone_change(void *tag)
 472 {
 473
 474         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
 475         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 476         tcp_tw_zone_change();
 477 }
 478
 479 static int
 480 tcp_inpcb_init(void *mem, int size, int flags)
 481 {
 482         struct inpcb *inp = mem;
 483
 484         INP_LOCK_INIT(inp, "inp", "tcpinp");
 485         return (0);
 486 }
 487
 488 /*
 489  * Take a value and get the next power of 2 that doesn't overflow.
 490  * Used to size the tcp_inpcb hash buckets.
 491  */
 492 static int
 493 maketcp_hashsize(int size)
 494 {
 495         int hashsize;
 496
 497         /*
 498          * auto tune.
 499          * get the next power of 2 higher than maxsockets.
 500          */
 501         hashsize = 1 << fls(size);
 502         /* catch overflow, and just go one power of 2 smaller */
 503         if (hashsize < size) {
 504                 hashsize = 1 << (fls(size) - 1);
 505         }
 506         return (hashsize);
 507 }
 508
 509 int
 510 register_tcp_functions(struct tcp_function_block *blk, int wait)
 511 {
 512         struct tcp_function_block *lblk;
 513         struct tcp_function *n;
 514         struct tcp_function_set fs;
 515
 516         if (t_functions_inited == 0) {
 517                 init_tcp_functions();
 518         }
 519         if ((blk->tfb_tcp_output == NULL) ||
 520             (blk->tfb_tcp_do_segment == NULL) ||
 521             (blk->tfb_tcp_ctloutput == NULL) ||
 522             (strlen(blk->tfb_tcp_block_name) == 0)) {
 523                 /*
 524                  * These functions are required and you
 525                  * need a name.
 526                  */
 527                 return (EINVAL);
 528         }
 529         if (blk->tfb_tcp_timer_stop_all ||
 530             blk->tfb_tcp_timer_activate ||
 531             blk->tfb_tcp_timer_active ||
 532             blk->tfb_tcp_timer_stop) {
 533                 /*
 534                  * If you define one timer function you
 535                  * must have them all.
 536                  */
 537                 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
 538                     (blk->tfb_tcp_timer_activate == NULL) ||
 539                     (blk->tfb_tcp_timer_active == NULL) ||
 540                     (blk->tfb_tcp_timer_stop == NULL)) {
 541                         return (EINVAL);
 542                 }
 543         }
 544         n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
 545         if (n == NULL) {
 546                 return (ENOMEM);
 547         }
 548         n->tf_fb = blk;
 549         strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 550         rw_wlock(&tcp_function_lock);
 551         lblk = find_tcp_functions_locked(&fs);
 552         if (lblk) {
 553                 /* Duplicate name space not allowed */
 554                 rw_wunlock(&tcp_function_lock);
 555                 free(n, M_TCPFUNCTIONS);
 556                 return (EALREADY);
 557         }
 558         refcount_init(&blk->tfb_refcnt, 0);
 559         blk->tfb_flags = 0;
 560         TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
 561         rw_wunlock(&tcp_function_lock);
 562         return(0);
 563 }
 564
 565 int
 566 deregister_tcp_functions(struct tcp_function_block *blk)
 567 {
 568         struct tcp_function_block *lblk;
 569         struct tcp_function *f;
 570         int error=ENOENT;
 571
 572         if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
 573                 /* You can't un-register the default */
 574                 return (EPERM);
 575         }
 576         rw_wlock(&tcp_function_lock);
 577         if (blk == tcp_func_set_ptr) {
 578                 /* You can't free the current default */
 579                 rw_wunlock(&tcp_function_lock);
 580                 return (EBUSY);
 581         }
 582         if (blk->tfb_refcnt) {
 583                 /* Still tcb attached, mark it. */
 584                 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
 585                 rw_wunlock(&tcp_function_lock);
 586                 return (EBUSY);
 587         }
 588         lblk = find_tcp_fb_locked(blk, &f);
 589         if (lblk) {
 590                 /* Found */
 591                 TAILQ_REMOVE(&t_functions, f, tf_next);
 592                 f->tf_fb = NULL;
 593                 free(f, M_TCPFUNCTIONS);
 594                 error = 0;
 595         }
 596         rw_wunlock(&tcp_function_lock);
 597         return (error);
 598 }
 599
 600 void
 601 tcp_init(void)
 602 {
 603         const char *tcbhash_tuneable;
 604         int hashsize;
 605
 606         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
 607
 608         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
 609             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 610                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 611         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
 612             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 613                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 614         hashsize = TCBHASHSIZE;
 615         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
 616         if (hashsize == 0) {
 617                 /*
 618                  * Auto tune the hash size based on maxsockets.
 619                  * A perfect hash would have a 1:1 mapping
 620                  * (hashsize = maxsockets) however it's been
 621                  * suggested that O(2) average is better.
 622                  */
 623                 hashsize = maketcp_hashsize(maxsockets / 4);
 624                 /*
 625                  * Our historical default is 512,
 626                  * do not autotune lower than this.
 627                  */
 628                 if (hashsize < 512)
 629                         hashsize = 512;
 630                 if (bootverbose && IS_DEFAULT_VNET(curvnet))
 631                         printf("%s: %s auto tuned to %d\n", __func__,
 632                             tcbhash_tuneable, hashsize);
 633         }
 634         /*
 635          * We require a hashsize to be a power of two.
 636          * Previously if it was not a power of two we would just reset it
 637          * back to 512, which could be a nasty surprise if you did not notice
 638          * the error message.
 639          * Instead what we do is clip it to the closest power of two lower
 640          * than the specified hash value.
 641          */
 642         if (!powerof2(hashsize)) {
 643                 int oldhashsize = hashsize;
 644
 645                 hashsize = maketcp_hashsize(hashsize);
 646                 /* prevent absurdly low value */
 647                 if (hashsize < 16)
 648                         hashsize = 16;
 649                 printf("%s: WARNING: TCB hash size not a power of 2, "
 650                     "clipped from %d to %d.\n", __func__, oldhashsize,
 651                     hashsize);
 652         }
 653         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
 654             "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
 655
 656         /*
 657          * These have to be type stable for the benefit of the timers.
 658          */
 659         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
 660             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 661         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 662         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
 663
 664         tcp_tw_init();
 665         syncache_init();
 666         tcp_hc_init();
 667
 668         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
 669         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
 670             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 671
 672         /* Skip initialization of globals for non-default instances. */
 673         if (!IS_DEFAULT_VNET(curvnet))
 674                 return;
 675
 676         tcp_reass_global_init();
 677
 678         /* XXX virtualize those bellow? */
 679         tcp_delacktime = TCPTV_DELACK;
 680         tcp_keepinit = TCPTV_KEEP_INIT;
 681         tcp_keepidle = TCPTV_KEEP_IDLE;
 682         tcp_keepintvl = TCPTV_KEEPINTVL;
 683         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
 684         tcp_msl = TCPTV_MSL;
 685         tcp_rexmit_min = TCPTV_MIN;
 686         if (tcp_rexmit_min < 1)
 687                 tcp_rexmit_min = 1;
 688         tcp_persmin = TCPTV_PERSMIN;
 689         tcp_persmax = TCPTV_PERSMAX;
 690         tcp_rexmit_slop = TCPTV_CPU_VAR;
 691         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
 692         tcp_tcbhashsize = hashsize;
 693         /* Setup the tcp function block list */
 694         init_tcp_functions();
 695         register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
 696
 697         if (tcp_soreceive_stream) {
 698 #ifdef INET
 699                 tcp_usrreqs.pru_soreceive = soreceive_stream;
 700 #endif
 701 #ifdef INET6
 702                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
 703 #endif /* INET6 */
 704         }
 705
 706 #ifdef INET6
 707 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
 708 #else /* INET6 */
 709 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
 710 #endif /* INET6 */
 711         if (max_protohdr < TCP_MINPROTOHDR)
 712                 max_protohdr = TCP_MINPROTOHDR;
 713         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
 714                 panic("tcp_init");
 715 #undef TCP_MINPROTOHDR
 716
 717         ISN_LOCK_INIT();
 718         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
 719                 SHUTDOWN_PRI_DEFAULT);
 720         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
 721                 EVENTHANDLER_PRI_ANY);
 722 #ifdef TCPPCAP
 723         tcp_pcap_init();
 724 #endif
 725
 726 #ifdef TCP_RFC7413
 727         tcp_fastopen_init();
 728 #endif
 729 }
 730
 731 #ifdef VIMAGE
 732 static void
 733 tcp_destroy(void *unused __unused)
 734 {
 735         int error, n;
 736
 737         /*
 738          * All our processes are gone, all our sockets should be cleaned
 739          * up, which means, we should be past the tcp_discardcb() calls.
 740          * Sleep to let all tcpcb timers really disappear and cleanup.
 741          */
 742         for (;;) {
 743                 INP_LIST_RLOCK(&V_tcbinfo);
 744                 n = V_tcbinfo.ipi_count;
 745                 INP_LIST_RUNLOCK(&V_tcbinfo);
 746                 if (n == 0)
 747                         break;
 748                 pause("tcpdes", hz / 10);
 749         }
 750         tcp_hc_destroy();
 751         syncache_destroy();
 752         tcp_tw_destroy();
 753         in_pcbinfo_destroy(&V_tcbinfo);
 754         /* tcp_discardcb() clears the sack_holes up. */
 755         uma_zdestroy(V_sack_hole_zone);
 756         uma_zdestroy(V_tcpcb_zone);
 757
 758 #ifdef TCP_RFC7413
 759         /*
 760          * Cannot free the zone until all tcpcbs are released as we attach
 761          * the allocations to them.
 762          */
 763         tcp_fastopen_destroy();
 764 #endif
 765
 766         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
 767         if (error != 0) {
 768                 printf("%s: WARNING: unable to deregister helper hook "
 769                     "type=%d, id=%d: error %d returned\n", __func__,
 770                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
 771         }
 772         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
 773         if (error != 0) {
 774                 printf("%s: WARNING: unable to deregister helper hook "
 775                     "type=%d, id=%d: error %d returned\n", __func__,
 776                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
 777         }
 778 }
 779 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
 780 #endif
 781
 782 void
 783 tcp_fini(void *xtp)
 784 {
 785
 786 }
 787
 788 /*
 789  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
 790  * tcp_template used to store this data in mbufs, but we now recopy it out
 791  * of the tcpcb each time to conserve mbufs.
 792  */
 793 void
 794 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
 795 {
 796         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
 797
 798         INP_WLOCK_ASSERT(inp);
 799
 800 #ifdef INET6
 801         if ((inp->inp_vflag & INP_IPV6) != 0) {
 802                 struct ip6_hdr *ip6;
 803
 804                 ip6 = (struct ip6_hdr *)ip_ptr;
 805                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
 806                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
 807                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
 808                         (IPV6_VERSION & IPV6_VERSION_MASK);
 809                 ip6->ip6_nxt = IPPROTO_TCP;
 810                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
 811                 ip6->ip6_src = inp->in6p_laddr;
 812                 ip6->ip6_dst = inp->in6p_faddr;
 813         }
 814 #endif /* INET6 */
 815 #if defined(INET6) && defined(INET)
 816         else
 817 #endif
 818 #ifdef INET
 819         {
 820                 struct ip *ip;
 821
 822                 ip = (struct ip *)ip_ptr;
 823                 ip->ip_v = IPVERSION;
 824                 ip->ip_hl = 5;
 825                 ip->ip_tos = inp->inp_ip_tos;
 826                 ip->ip_len = 0;
 827                 ip->ip_id = 0;
 828                 ip->ip_off = 0;
 829                 ip->ip_ttl = inp->inp_ip_ttl;
 830                 ip->ip_sum = 0;
 831                 ip->ip_p = IPPROTO_TCP;
 832                 ip->ip_src = inp->inp_laddr;
 833                 ip->ip_dst = inp->inp_faddr;
 834         }
 835 #endif /* INET */
 836         th->th_sport = inp->inp_lport;
 837         th->th_dport = inp->inp_fport;
 838         th->th_seq = 0;
 839         th->th_ack = 0;
 840         th->th_x2 = 0;
 841         th->th_off = 5;
 842         th->th_flags = 0;
 843         th->th_win = 0;
 844         th->th_urp = 0;
 845         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
 846 }
 847
 848 /*
 849  * Create template to be used to send tcp packets on a connection.
 850  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
 851  * use for this function is in keepalives, which use tcp_respond.
 852  */
 853 struct tcptemp *
 854 tcpip_maketemplate(struct inpcb *inp)
 855 {
 856         struct tcptemp *t;
 857
 858         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
 859         if (t == NULL)
 860                 return (NULL);
 861         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
 862         return (t);
 863 }
 864
 865 /*
 866  * Send a single message to the TCP at address specified by
 867  * the given TCP/IP header.  If m == NULL, then we make a copy
 868  * of the tcpiphdr at th and send directly to the addressed host.
 869  * This is used to force keep alive messages out using the TCP
 870  * template for a connection.  If flags are given then we send
 871  * a message back to the TCP which originated the segment th,
 872  * and discard the mbuf containing it and any other attached mbufs.
 873  *
 874  * In any case the ack and sequence number of the transmitted
 875  * segment are as specified by the parameters.
 876  *
 877  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
 878  */
 879 void
 880 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
 881     tcp_seq ack, tcp_seq seq, int flags)
 882 {
 883         struct tcpopt to;
 884         struct inpcb *inp;
 885         struct ip *ip;
 886         struct mbuf *optm;
 887         struct tcphdr *nth;
 888         u_char *optp;
 889 #ifdef INET6
 890         struct ip6_hdr *ip6;
 891         int isipv6;
 892 #endif /* INET6 */
 893         int optlen, tlen, win;
 894         bool incl_opts;
 895
 896         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
 897
 898 #ifdef INET6
 899         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
 900         ip6 = ipgen;
 901 #endif /* INET6 */
 902         ip = ipgen;
 903
 904         if (tp != NULL) {
 905                 inp = tp->t_inpcb;
 906                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
 907                 INP_WLOCK_ASSERT(inp);
 908         } else
 909                 inp = NULL;
 910
 911         incl_opts = false;
 912         win = 0;
 913         if (tp != NULL) {
 914                 if (!(flags & TH_RST)) {
 915                         win = sbspace(&inp->inp_socket->so_rcv);
 916                         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
 917                                 win = (long)TCP_MAXWIN << tp->rcv_scale;
 918                 }
 919                 if ((tp->t_flags & TF_NOOPT) == 0)
 920                         incl_opts = true;
 921         }
 922         if (m == NULL) {
 923                 m = m_gethdr(M_NOWAIT, MT_DATA);
 924                 if (m == NULL)
 925                         return;
 926                 m->m_data += max_linkhdr;
 927 #ifdef INET6
 928                 if (isipv6) {
 929                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
 930                               sizeof(struct ip6_hdr));
 931                         ip6 = mtod(m, struct ip6_hdr *);
 932                         nth = (struct tcphdr *)(ip6 + 1);
 933                 } else
 934 #endif /* INET6 */
 935                 {
 936                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
 937                         ip = mtod(m, struct ip *);
 938                         nth = (struct tcphdr *)(ip + 1);
 939                 }
 940                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 941                 flags = TH_ACK;
 942         } else if (!M_WRITABLE(m)) {
 943                 struct mbuf *n;
 944
 945                 /* Can't reuse 'm', allocate a new mbuf. */
 946                 n = m_gethdr(M_NOWAIT, MT_DATA);
 947                 if (n == NULL) {
 948                         m_freem(m);
 949                         return;
 950                 }
 951
 952                 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
 953                         m_freem(m);
 954                         m_freem(n);
 955                         return;
 956                 }
 957
 958                 n->m_data += max_linkhdr;
 959                 /* m_len is set later */
 960 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
 961 #ifdef INET6
 962                 if (isipv6) {
 963                         bcopy((caddr_t)ip6, mtod(n, caddr_t),
 964                               sizeof(struct ip6_hdr));
 965                         ip6 = mtod(n, struct ip6_hdr *);
 966                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
 967                         nth = (struct tcphdr *)(ip6 + 1);
 968                 } else
 969 #endif /* INET6 */
 970                 {
 971                         bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
 972                         ip = mtod(n, struct ip *);
 973                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
 974                         nth = (struct tcphdr *)(ip + 1);
 975                 }
 976                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 977                 xchg(nth->th_dport, nth->th_sport, uint16_t);
 978                 th = nth;
 979                 m_freem(m);
 980                 m = n;
 981         } else {
 982                 /*
 983                  *  reuse the mbuf.
 984                  * XXX MRT We inherit the FIB, which is lucky.
 985                  */
 986                 m_freem(m->m_next);
 987                 m->m_next = NULL;
 988                 m->m_data = (caddr_t)ipgen;
 989                 /* m_len is set later */
 990 #ifdef INET6
 991                 if (isipv6) {
 992                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
 993                         nth = (struct tcphdr *)(ip6 + 1);
 994                 } else
 995 #endif /* INET6 */
 996                 {
 997                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
 998                         nth = (struct tcphdr *)(ip + 1);
 999                 }
1000                 if (th != nth) {
1001                         /*
1002                          * this is usually a case when an extension header
1003                          * exists between the IPv6 header and the
1004                          * TCP header.
1005                          */
1006                         nth->th_sport = th->th_sport;
1007                         nth->th_dport = th->th_dport;
1008                 }
1009                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1010 #undef xchg
1011         }
1012         tlen = 0;
1013 #ifdef INET6
1014         if (isipv6)
1015                 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1016 #endif
1017 #if defined(INET) && defined(INET6)
1018         else
1019 #endif
1020 #ifdef INET
1021                 tlen = sizeof (struct tcpiphdr);
1022 #endif
1023 #ifdef INVARIANTS
1024         m->m_len = 0;
1025         KASSERT(M_TRAILINGSPACE(m) >= tlen,
1026             ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1027             m, tlen, (long)M_TRAILINGSPACE(m)));
1028 #endif
1029         m->m_len = tlen;
1030         to.to_flags = 0;
1031         if (incl_opts) {
1032                 /* Make sure we have room. */
1033                 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1034                         m->m_next = m_get(M_NOWAIT, MT_DATA);
1035                         if (m->m_next) {
1036                                 optp = mtod(m->m_next, u_char *);
1037                                 optm = m->m_next;
1038                         } else
1039                                 incl_opts = false;
1040                 } else {
1041                         optp = (u_char *) (nth + 1);
1042                         optm = m;
1043                 }
1044         }
1045         if (incl_opts) {
1046                 /* Timestamps. */
1047                 if (tp->t_flags & TF_RCVD_TSTMP) {
1048                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1049                         to.to_tsecr = tp->ts_recent;
1050                         to.to_flags |= TOF_TS;
1051                 }
1052 #ifdef TCP_SIGNATURE
1053                 /* TCP-MD5 (RFC2385). */
1054                 if (tp->t_flags & TF_SIGNATURE)
1055                         to.to_flags |= TOF_SIGNATURE;
1056 #endif
1057
1058                 /* Add the options. */
1059                 tlen += optlen = tcp_addoptions(&to, optp);
1060
1061                 /* Update m_len in the correct mbuf. */
1062                 optm->m_len += optlen;
1063         } else
1064                 optlen = 0;
1065 #ifdef INET6
1066         if (isipv6) {
1067                 ip6->ip6_flow = 0;
1068                 ip6->ip6_vfc = IPV6_VERSION;
1069                 ip6->ip6_nxt = IPPROTO_TCP;
1070                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1071         }
1072 #endif
1073 #if defined(INET) && defined(INET6)
1074         else
1075 #endif
1076 #ifdef INET
1077         {
1078                 ip->ip_len = htons(tlen);
1079                 ip->ip_ttl = V_ip_defttl;
1080                 if (V_path_mtu_discovery)
1081                         ip->ip_off |= htons(IP_DF);
1082         }
1083 #endif
1084         m->m_pkthdr.len = tlen;
1085         m->m_pkthdr.rcvif = NULL;
1086 #ifdef MAC
1087         if (inp != NULL) {
1088                 /*
1089                  * Packet is associated with a socket, so allow the
1090                  * label of the response to reflect the socket label.
1091                  */
1092                 INP_WLOCK_ASSERT(inp);
1093                 mac_inpcb_create_mbuf(inp, m);
1094         } else {
1095                 /*
1096                  * Packet is not associated with a socket, so possibly
1097                  * update the label in place.
1098                  */
1099                 mac_netinet_tcp_reply(m);
1100         }
1101 #endif
1102         nth->th_seq = htonl(seq);
1103         nth->th_ack = htonl(ack);
1104         nth->th_x2 = 0;
1105         nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1106         nth->th_flags = flags;
1107         if (tp != NULL)
1108                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1109         else
1110                 nth->th_win = htons((u_short)win);
1111         nth->th_urp = 0;
1112
1113 #ifdef TCP_SIGNATURE
1114         if (to.to_flags & TOF_SIGNATURE) {
1115                 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1116                     IPSEC_DIR_OUTBOUND);
1117         }
1118 #endif
1119
1120         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1121 #ifdef INET6
1122         if (isipv6) {
1123                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1124                 nth->th_sum = in6_cksum_pseudo(ip6,
1125                     tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1126                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1127                     NULL, NULL);
1128         }
1129 #endif /* INET6 */
1130 #if defined(INET6) && defined(INET)
1131         else
1132 #endif
1133 #ifdef INET
1134         {
1135                 m->m_pkthdr.csum_flags = CSUM_TCP;
1136                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1137                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1138         }
1139 #endif /* INET */
1140 #ifdef TCPDEBUG
1141         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1142                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1143 #endif
1144         TCP_PROBE3(debug__output, tp, th, mtod(m, const char *));
1145         if (flags & TH_RST)
1146                 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
1147                     tp, nth);
1148
1149         TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
1150 #ifdef INET6
1151         if (isipv6)
1152                 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1153 #endif /* INET6 */
1154 #if defined(INET) && defined(INET6)
1155         else
1156 #endif
1157 #ifdef INET
1158                 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1159 #endif
1160 }
1161
1162 /*
1163  * Create a new TCP control block, making an
1164  * empty reassembly queue and hooking it to the argument
1165  * protocol control block.  The `inp' parameter must have
1166  * come from the zone allocator set up in tcp_init().
1167  */
1168 struct tcpcb *
1169 tcp_newtcpcb(struct inpcb *inp)
1170 {
1171         struct tcpcb_mem *tm;
1172         struct tcpcb *tp;
1173 #ifdef INET6
1174         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1175 #endif /* INET6 */
1176
1177         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1178         if (tm == NULL)
1179                 return (NULL);
1180         tp = &tm->tcb;
1181
1182         /* Initialise cc_var struct for this tcpcb. */
1183         tp->ccv = &tm->ccv;
1184         tp->ccv->type = IPPROTO_TCP;
1185         tp->ccv->ccvc.tcp = tp;
1186         rw_rlock(&tcp_function_lock);
1187         tp->t_fb = tcp_func_set_ptr;
1188         refcount_acquire(&tp->t_fb->tfb_refcnt);
1189         rw_runlock(&tcp_function_lock);
1190         if (tp->t_fb->tfb_tcp_fb_init) {
1191                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1192         }
1193         /*
1194          * Use the current system default CC algorithm.
1195          */
1196         CC_LIST_RLOCK();
1197         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1198         CC_ALGO(tp) = CC_DEFAULT();
1199         CC_LIST_RUNLOCK();
1200
1201         if (CC_ALGO(tp)->cb_init != NULL)
1202                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1203                         if (tp->t_fb->tfb_tcp_fb_fini)
1204                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1205                         refcount_release(&tp->t_fb->tfb_refcnt);
1206                         uma_zfree(V_tcpcb_zone, tm);
1207                         return (NULL);
1208                 }
1209
1210         tp->osd = &tm->osd;
1211         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1212                 if (tp->t_fb->tfb_tcp_fb_fini)
1213                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1214                 refcount_release(&tp->t_fb->tfb_refcnt);
1215                 uma_zfree(V_tcpcb_zone, tm);
1216                 return (NULL);
1217         }
1218
1219 #ifdef VIMAGE
1220         tp->t_vnet = inp->inp_vnet;
1221 #endif
1222         tp->t_timers = &tm->tt;
1223         /*      LIST_INIT(&tp->t_segq); */      /* XXX covered by M_ZERO */
1224         tp->t_maxseg =
1225 #ifdef INET6
1226                 isipv6 ? V_tcp_v6mssdflt :
1227 #endif /* INET6 */
1228                 V_tcp_mssdflt;
1229
1230         /* Set up our timeouts. */
1231         callout_init(&tp->t_timers->tt_rexmt, 1);
1232         callout_init(&tp->t_timers->tt_persist, 1);
1233         callout_init(&tp->t_timers->tt_keep, 1);
1234         callout_init(&tp->t_timers->tt_2msl, 1);
1235         callout_init(&tp->t_timers->tt_delack, 1);
1236
1237         if (V_tcp_do_rfc1323)
1238                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1239         if (V_tcp_do_sack)
1240                 tp->t_flags |= TF_SACK_PERMIT;
1241         TAILQ_INIT(&tp->snd_holes);
1242         /*
1243          * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1244          * is called.
1245          */
1246         in_pcbref(inp); /* Reference for tcpcb */
1247         tp->t_inpcb = inp;
1248
1249         /*
1250          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1251          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1252          * reasonable initial retransmit time.
1253          */
1254         tp->t_srtt = TCPTV_SRTTBASE;
1255         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1256         tp->t_rttmin = tcp_rexmit_min;
1257         tp->t_rxtcur = TCPTV_RTOBASE;
1258         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1259         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1260         tp->t_rcvtime = ticks;
1261         /*
1262          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1263          * because the socket may be bound to an IPv6 wildcard address,
1264          * which may match an IPv4-mapped IPv6 address.
1265          */
1266         inp->inp_ip_ttl = V_ip_defttl;
1267         inp->inp_ppcb = tp;
1268 #ifdef TCPPCAP
1269         /*
1270          * Init the TCP PCAP queues.
1271          */
1272         tcp_pcap_tcpcb_init(tp);
1273 #endif
1274         return (tp);            /* XXX */
1275 }
1276
1277 /*
1278  * Switch the congestion control algorithm back to NewReno for any active
1279  * control blocks using an algorithm which is about to go away.
1280  * This ensures the CC framework can allow the unload to proceed without leaving
1281  * any dangling pointers which would trigger a panic.
1282  * Returning non-zero would inform the CC framework that something went wrong
1283  * and it would be unsafe to allow the unload to proceed. However, there is no
1284  * way for this to occur with this implementation so we always return zero.
1285  */
1286 int
1287 tcp_ccalgounload(struct cc_algo *unload_algo)
1288 {
1289         struct cc_algo *tmpalgo;
1290         struct inpcb *inp;
1291         struct tcpcb *tp;
1292         VNET_ITERATOR_DECL(vnet_iter);
1293
1294         /*
1295          * Check all active control blocks across all network stacks and change
1296          * any that are using "unload_algo" back to NewReno. If "unload_algo"
1297          * requires cleanup code to be run, call it.
1298          */
1299         VNET_LIST_RLOCK();
1300         VNET_FOREACH(vnet_iter) {
1301                 CURVNET_SET(vnet_iter);
1302                 INP_INFO_WLOCK(&V_tcbinfo);
1303                 /*
1304                  * New connections already part way through being initialised
1305                  * with the CC algo we're removing will not race with this code
1306                  * because the INP_INFO_WLOCK is held during initialisation. We
1307                  * therefore don't enter the loop below until the connection
1308                  * list has stabilised.
1309                  */
1310                 LIST_FOREACH(inp, &V_tcb, inp_list) {
1311                         INP_WLOCK(inp);
1312                         /* Important to skip tcptw structs. */
1313                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1314                             (tp = intotcpcb(inp)) != NULL) {
1315                                 /*
1316                                  * By holding INP_WLOCK here, we are assured
1317                                  * that the connection is not currently
1318                                  * executing inside the CC module's functions
1319                                  * i.e. it is safe to make the switch back to
1320                                  * NewReno.
1321                                  */
1322                                 if (CC_ALGO(tp) == unload_algo) {
1323                                         tmpalgo = CC_ALGO(tp);
1324                                         /* NewReno does not require any init. */
1325                                         CC_ALGO(tp) = &newreno_cc_algo;
1326                                         if (tmpalgo->cb_destroy != NULL)
1327                                                 tmpalgo->cb_destroy(tp->ccv);
1328                                 }
1329                         }
1330                         INP_WUNLOCK(inp);
1331                 }
1332                 INP_INFO_WUNLOCK(&V_tcbinfo);
1333                 CURVNET_RESTORE();
1334         }
1335         VNET_LIST_RUNLOCK();
1336
1337         return (0);
1338 }
1339
1340 /*
1341  * Drop a TCP connection, reporting
1342  * the specified error.  If connection is synchronized,
1343  * then send a RST to peer.
1344  */
1345 struct tcpcb *
1346 tcp_drop(struct tcpcb *tp, int errno)
1347 {
1348         struct socket *so = tp->t_inpcb->inp_socket;
1349
1350         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1351         INP_WLOCK_ASSERT(tp->t_inpcb);
1352
1353         if (TCPS_HAVERCVDSYN(tp->t_state)) {
1354                 tcp_state_change(tp, TCPS_CLOSED);
1355                 (void) tp->t_fb->tfb_tcp_output(tp);
1356                 TCPSTAT_INC(tcps_drops);
1357         } else
1358                 TCPSTAT_INC(tcps_conndrops);
1359         if (errno == ETIMEDOUT && tp->t_softerror)
1360                 errno = tp->t_softerror;
1361         so->so_error = errno;
1362         return (tcp_close(tp));
1363 }
1364
1365 void
1366 tcp_discardcb(struct tcpcb *tp)
1367 {
1368         struct inpcb *inp = tp->t_inpcb;
1369         struct socket *so = inp->inp_socket;
1370 #ifdef INET6
1371         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1372 #endif /* INET6 */
1373         int released;
1374
1375         INP_WLOCK_ASSERT(inp);
1376
1377         /*
1378          * Make sure that all of our timers are stopped before we delete the
1379          * PCB.
1380          *
1381          * If stopping a timer fails, we schedule a discard function in same
1382          * callout, and the last discard function called will take care of
1383          * deleting the tcpcb.
1384          */
1385         tp->t_timers->tt_draincnt = 0;
1386         tcp_timer_stop(tp, TT_REXMT);
1387         tcp_timer_stop(tp, TT_PERSIST);
1388         tcp_timer_stop(tp, TT_KEEP);
1389         tcp_timer_stop(tp, TT_2MSL);
1390         tcp_timer_stop(tp, TT_DELACK);
1391         if (tp->t_fb->tfb_tcp_timer_stop_all) {
1392                 /*
1393                  * Call the stop-all function of the methods,
1394                  * this function should call the tcp_timer_stop()
1395                  * method with each of the function specific timeouts.
1396                  * That stop will be called via the tfb_tcp_timer_stop()
1397                  * which should use the async drain function of the
1398                  * callout system (see tcp_var.h).
1399                  */
1400                 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1401         }
1402
1403         /*
1404          * If we got enough samples through the srtt filter,
1405          * save the rtt and rttvar in the routing entry.
1406          * 'Enough' is arbitrarily defined as 4 rtt samples.
1407          * 4 samples is enough for the srtt filter to converge
1408          * to within enough % of the correct value; fewer samples
1409          * and we could save a bogus rtt. The danger is not high
1410          * as tcp quickly recovers from everything.
1411          * XXX: Works very well but needs some more statistics!
1412          */
1413         if (tp->t_rttupdated >= 4) {
1414                 struct hc_metrics_lite metrics;
1415                 u_long ssthresh;
1416
1417                 bzero(&metrics, sizeof(metrics));
1418                 /*
1419                  * Update the ssthresh always when the conditions below
1420                  * are satisfied. This gives us better new start value
1421                  * for the congestion avoidance for new connections.
1422                  * ssthresh is only set if packet loss occurred on a session.
1423                  *
1424                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1425                  * being torn down.  Ideally this code would not use 'so'.
1426                  */
1427                 ssthresh = tp->snd_ssthresh;
1428                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1429                         /*
1430                          * convert the limit from user data bytes to
1431                          * packets then to packet data bytes.
1432                          */
1433                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1434                         if (ssthresh < 2)
1435                                 ssthresh = 2;
1436                         ssthresh *= (u_long)(tp->t_maxseg +
1437 #ifdef INET6
1438                             (isipv6 ? sizeof (struct ip6_hdr) +
1439                                 sizeof (struct tcphdr) :
1440 #endif
1441                                 sizeof (struct tcpiphdr)
1442 #ifdef INET6
1443                             )
1444 #endif
1445                             );
1446                 } else
1447                         ssthresh = 0;
1448                 metrics.rmx_ssthresh = ssthresh;
1449
1450                 metrics.rmx_rtt = tp->t_srtt;
1451                 metrics.rmx_rttvar = tp->t_rttvar;
1452                 metrics.rmx_cwnd = tp->snd_cwnd;
1453                 metrics.rmx_sendpipe = 0;
1454                 metrics.rmx_recvpipe = 0;
1455
1456                 tcp_hc_update(&inp->inp_inc, &metrics);
1457         }
1458
1459         /* free the reassembly queue, if any */
1460         tcp_reass_flush(tp);
1461
1462 #ifdef TCP_OFFLOAD
1463         /* Disconnect offload device, if any. */
1464         if (tp->t_flags & TF_TOE)
1465                 tcp_offload_detach(tp);
1466 #endif
1467
1468         tcp_free_sackholes(tp);
1469
1470 #ifdef TCPPCAP
1471         /* Free the TCP PCAP queues. */
1472         tcp_pcap_drain(&(tp->t_inpkts));
1473         tcp_pcap_drain(&(tp->t_outpkts));
1474 #endif
1475
1476         /* Allow the CC algorithm to clean up after itself. */
1477         if (CC_ALGO(tp)->cb_destroy != NULL)
1478                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1479
1480         khelp_destroy_osd(tp->osd);
1481
1482         CC_ALGO(tp) = NULL;
1483         inp->inp_ppcb = NULL;
1484         if (tp->t_timers->tt_draincnt == 0) {
1485                 /* We own the last reference on tcpcb, let's free it. */
1486                 if (tp->t_fb->tfb_tcp_fb_fini)
1487                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1488                 refcount_release(&tp->t_fb->tfb_refcnt);
1489                 tp->t_inpcb = NULL;
1490                 uma_zfree(V_tcpcb_zone, tp);
1491                 released = in_pcbrele_wlocked(inp);
1492                 KASSERT(!released, ("%s: inp %p should not have been released "
1493                         "here", __func__, inp));
1494         }
1495 }
1496
1497 void
1498 tcp_timer_discard(void *ptp)
1499 {
1500         struct inpcb *inp;
1501         struct tcpcb *tp;
1502
1503         tp = (struct tcpcb *)ptp;
1504         CURVNET_SET(tp->t_vnet);
1505         INP_INFO_RLOCK(&V_tcbinfo);
1506         inp = tp->t_inpcb;
1507         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1508                 __func__, tp));
1509         INP_WLOCK(inp);
1510         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1511                 ("%s: tcpcb has to be stopped here", __func__));
1512         tp->t_timers->tt_draincnt--;
1513         if (tp->t_timers->tt_draincnt == 0) {
1514                 /* We own the last reference on this tcpcb, let's free it. */
1515                 if (tp->t_fb->tfb_tcp_fb_fini)
1516                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1517                 refcount_release(&tp->t_fb->tfb_refcnt);
1518                 tp->t_inpcb = NULL;
1519                 uma_zfree(V_tcpcb_zone, tp);
1520                 if (in_pcbrele_wlocked(inp)) {
1521                         INP_INFO_RUNLOCK(&V_tcbinfo);
1522                         CURVNET_RESTORE();
1523                         return;
1524                 }
1525         }
1526         INP_WUNLOCK(inp);
1527         INP_INFO_RUNLOCK(&V_tcbinfo);
1528         CURVNET_RESTORE();
1529 }
1530
1531 /*
1532  * Attempt to close a TCP control block, marking it as dropped, and freeing
1533  * the socket if we hold the only reference.
1534  */
1535 struct tcpcb *
1536 tcp_close(struct tcpcb *tp)
1537 {
1538         struct inpcb *inp = tp->t_inpcb;
1539         struct socket *so;
1540
1541         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1542         INP_WLOCK_ASSERT(inp);
1543
1544 #ifdef TCP_OFFLOAD
1545         if (tp->t_state == TCPS_LISTEN)
1546                 tcp_offload_listen_stop(tp);
1547 #endif
1548 #ifdef TCP_RFC7413
1549         /*
1550          * This releases the TFO pending counter resource for TFO listen
1551          * sockets as well as passively-created TFO sockets that transition
1552          * from SYN_RECEIVED to CLOSED.
1553          */
1554         if (tp->t_tfo_pending) {
1555                 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1556                 tp->t_tfo_pending = NULL;
1557         }
1558 #endif
1559         in_pcbdrop(inp);
1560         TCPSTAT_INC(tcps_closed);
1561         TCPSTATES_DEC(tp->t_state);
1562         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1563         so = inp->inp_socket;
1564         soisdisconnected(so);
1565         if (inp->inp_flags & INP_SOCKREF) {
1566                 KASSERT(so->so_state & SS_PROTOREF,
1567                     ("tcp_close: !SS_PROTOREF"));
1568                 inp->inp_flags &= ~INP_SOCKREF;
1569                 INP_WUNLOCK(inp);
1570                 ACCEPT_LOCK();
1571                 SOCK_LOCK(so);
1572                 so->so_state &= ~SS_PROTOREF;
1573                 sofree(so);
1574                 return (NULL);
1575         }
1576         return (tp);
1577 }
1578
1579 void
1580 tcp_drain(void)
1581 {
1582         VNET_ITERATOR_DECL(vnet_iter);
1583
1584         if (!do_tcpdrain)
1585                 return;
1586
1587         VNET_LIST_RLOCK_NOSLEEP();
1588         VNET_FOREACH(vnet_iter) {
1589                 CURVNET_SET(vnet_iter);
1590                 struct inpcb *inpb;
1591                 struct tcpcb *tcpb;
1592
1593         /*
1594          * Walk the tcpbs, if existing, and flush the reassembly queue,
1595          * if there is one...
1596          * XXX: The "Net/3" implementation doesn't imply that the TCP
1597          *      reassembly queue should be flushed, but in a situation
1598          *      where we're really low on mbufs, this is potentially
1599          *      useful.
1600          */
1601                 INP_INFO_WLOCK(&V_tcbinfo);
1602                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1603                         if (inpb->inp_flags & INP_TIMEWAIT)
1604                                 continue;
1605                         INP_WLOCK(inpb);
1606                         if ((tcpb = intotcpcb(inpb)) != NULL) {
1607                                 tcp_reass_flush(tcpb);
1608                                 tcp_clean_sackreport(tcpb);
1609 #ifdef TCPPCAP
1610                                 if (tcp_pcap_aggressive_free) {
1611                                         /* Free the TCP PCAP queues. */
1612                                         tcp_pcap_drain(&(tcpb->t_inpkts));
1613                                         tcp_pcap_drain(&(tcpb->t_outpkts));
1614                                 }
1615 #endif
1616                         }
1617                         INP_WUNLOCK(inpb);
1618                 }
1619                 INP_INFO_WUNLOCK(&V_tcbinfo);
1620                 CURVNET_RESTORE();
1621         }
1622         VNET_LIST_RUNLOCK_NOSLEEP();
1623 }
1624
1625 /*
1626  * Notify a tcp user of an asynchronous error;
1627  * store error as soft error, but wake up user
1628  * (for now, won't do anything until can select for soft error).
1629  *
1630  * Do not wake up user since there currently is no mechanism for
1631  * reporting soft errors (yet - a kqueue filter may be added).
1632  */
1633 static struct inpcb *
1634 tcp_notify(struct inpcb *inp, int error)
1635 {
1636         struct tcpcb *tp;
1637
1638         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1639         INP_WLOCK_ASSERT(inp);
1640
1641         if ((inp->inp_flags & INP_TIMEWAIT) ||
1642             (inp->inp_flags & INP_DROPPED))
1643                 return (inp);
1644
1645         tp = intotcpcb(inp);
1646         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1647
1648         /*
1649          * Ignore some errors if we are hooked up.
1650          * If connection hasn't completed, has retransmitted several times,
1651          * and receives a second error, give up now.  This is better
1652          * than waiting a long time to establish a connection that
1653          * can never complete.
1654          */
1655         if (tp->t_state == TCPS_ESTABLISHED &&
1656             (error == EHOSTUNREACH || error == ENETUNREACH ||
1657              error == EHOSTDOWN)) {
1658                 if (inp->inp_route.ro_rt) {
1659                         RTFREE(inp->inp_route.ro_rt);
1660                         inp->inp_route.ro_rt = (struct rtentry *)NULL;
1661                 }
1662                 return (inp);
1663         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1664             tp->t_softerror) {
1665                 tp = tcp_drop(tp, error);
1666                 if (tp != NULL)
1667                         return (inp);
1668                 else
1669                         return (NULL);
1670         } else {
1671                 tp->t_softerror = error;
1672                 return (inp);
1673         }
1674 #if 0
1675         wakeup( &so->so_timeo);
1676         sorwakeup(so);
1677         sowwakeup(so);
1678 #endif
1679 }
1680
1681 static int
1682 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1683 {
1684         int error, i, m, n, pcb_count;
1685         struct inpcb *inp, **inp_list;
1686         inp_gen_t gencnt;
1687         struct xinpgen xig;
1688
1689         /*
1690          * The process of preparing the TCB list is too time-consuming and
1691          * resource-intensive to repeat twice on every request.
1692          */
1693         if (req->oldptr == NULL) {
1694                 n = V_tcbinfo.ipi_count +
1695                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1696                 n += imax(n / 8, 10);
1697                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1698                 return (0);
1699         }
1700
1701         if (req->newptr != NULL)
1702                 return (EPERM);
1703
1704         /*
1705          * OK, now we're committed to doing something.
1706          */
1707         INP_LIST_RLOCK(&V_tcbinfo);
1708         gencnt = V_tcbinfo.ipi_gencnt;
1709         n = V_tcbinfo.ipi_count;
1710         INP_LIST_RUNLOCK(&V_tcbinfo);
1711
1712         m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1713
1714         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1715                 + (n + m) * sizeof(struct xtcpcb));
1716         if (error != 0)
1717                 return (error);
1718
1719         xig.xig_len = sizeof xig;
1720         xig.xig_count = n + m;
1721         xig.xig_gen = gencnt;
1722         xig.xig_sogen = so_gencnt;
1723         error = SYSCTL_OUT(req, &xig, sizeof xig);
1724         if (error)
1725                 return (error);
1726
1727         error = syncache_pcblist(req, m, &pcb_count);
1728         if (error)
1729                 return (error);
1730
1731         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1732
1733         INP_INFO_WLOCK(&V_tcbinfo);
1734         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1735             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1736                 INP_WLOCK(inp);
1737                 if (inp->inp_gencnt <= gencnt) {
1738                         /*
1739                          * XXX: This use of cr_cansee(), introduced with
1740                          * TCP state changes, is not quite right, but for
1741                          * now, better than nothing.
1742                          */
1743                         if (inp->inp_flags & INP_TIMEWAIT) {
1744                                 if (intotw(inp) != NULL)
1745                                         error = cr_cansee(req->td->td_ucred,
1746                                             intotw(inp)->tw_cred);
1747                                 else
1748                                         error = EINVAL; /* Skip this inp. */
1749                         } else
1750                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
1751                         if (error == 0) {
1752                                 in_pcbref(inp);
1753                                 inp_list[i++] = inp;
1754                         }
1755                 }
1756                 INP_WUNLOCK(inp);
1757         }
1758         INP_INFO_WUNLOCK(&V_tcbinfo);
1759         n = i;
1760
1761         error = 0;
1762         for (i = 0; i < n; i++) {
1763                 inp = inp_list[i];
1764                 INP_RLOCK(inp);
1765                 if (inp->inp_gencnt <= gencnt) {
1766                         struct xtcpcb xt;
1767                         void *inp_ppcb;
1768
1769                         bzero(&xt, sizeof(xt));
1770                         xt.xt_len = sizeof xt;
1771                         /* XXX should avoid extra copy */
1772                         bcopy(inp, &xt.xt_inp, sizeof *inp);
1773                         inp_ppcb = inp->inp_ppcb;
1774                         if (inp_ppcb == NULL)
1775                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1776                         else if (inp->inp_flags & INP_TIMEWAIT) {
1777                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1778                                 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1779                         } else {
1780                                 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1781                                 if (xt.xt_tp.t_timers)
1782                                         tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1783                         }
1784                         if (inp->inp_socket != NULL)
1785                                 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1786                         else {
1787                                 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1788                                 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1789                         }
1790                         xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1791                         INP_RUNLOCK(inp);
1792                         error = SYSCTL_OUT(req, &xt, sizeof xt);
1793                 } else
1794                         INP_RUNLOCK(inp);
1795         }
1796         INP_INFO_RLOCK(&V_tcbinfo);
1797         for (i = 0; i < n; i++) {
1798                 inp = inp_list[i];
1799                 INP_RLOCK(inp);
1800                 if (!in_pcbrele_rlocked(inp))
1801                         INP_RUNLOCK(inp);
1802         }
1803         INP_INFO_RUNLOCK(&V_tcbinfo);
1804
1805         if (!error) {
1806                 /*
1807                  * Give the user an updated idea of our state.
1808                  * If the generation differs from what we told
1809                  * her before, she knows that something happened
1810                  * while we were processing this request, and it
1811                  * might be necessary to retry.
1812                  */
1813                 INP_LIST_RLOCK(&V_tcbinfo);
1814                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1815                 xig.xig_sogen = so_gencnt;
1816                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1817                 INP_LIST_RUNLOCK(&V_tcbinfo);
1818                 error = SYSCTL_OUT(req, &xig, sizeof xig);
1819         }
1820         free(inp_list, M_TEMP);
1821         return (error);
1822 }
1823
1824 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1825     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1826     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1827
1828 #ifdef INET
1829 static int
1830 tcp_getcred(SYSCTL_HANDLER_ARGS)
1831 {
1832         struct xucred xuc;
1833         struct sockaddr_in addrs[2];
1834         struct inpcb *inp;
1835         int error;
1836
1837         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1838         if (error)
1839                 return (error);
1840         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1841         if (error)
1842                 return (error);
1843         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1844             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1845         if (inp != NULL) {
1846                 if (inp->inp_socket == NULL)
1847                         error = ENOENT;
1848                 if (error == 0)
1849                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1850                 if (error == 0)
1851                         cru2x(inp->inp_cred, &xuc);
1852                 INP_RUNLOCK(inp);
1853         } else
1854                 error = ENOENT;
1855         if (error == 0)
1856                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1857         return (error);
1858 }
1859
1860 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1861     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1862     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1863 #endif /* INET */
1864
1865 #ifdef INET6
1866 static int
1867 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1868 {
1869         struct xucred xuc;
1870         struct sockaddr_in6 addrs[2];
1871         struct inpcb *inp;
1872         int error;
1873 #ifdef INET
1874         int mapped = 0;
1875 #endif
1876
1877         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1878         if (error)
1879                 return (error);
1880         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1881         if (error)
1882                 return (error);
1883         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1884             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1885                 return (error);
1886         }
1887         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1888 #ifdef INET
1889                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1890                         mapped = 1;
1891                 else
1892 #endif
1893                         return (EINVAL);
1894         }
1895
1896 #ifdef INET
1897         if (mapped == 1)
1898                 inp = in_pcblookup(&V_tcbinfo,
1899                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1900                         addrs[1].sin6_port,
1901                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1902                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1903         else
1904 #endif
1905                 inp = in6_pcblookup(&V_tcbinfo,
1906                         &addrs[1].sin6_addr, addrs[1].sin6_port,
1907                         &addrs[0].sin6_addr, addrs[0].sin6_port,
1908                         INPLOOKUP_RLOCKPCB, NULL);
1909         if (inp != NULL) {
1910                 if (inp->inp_socket == NULL)
1911                         error = ENOENT;
1912                 if (error == 0)
1913                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1914                 if (error == 0)
1915                         cru2x(inp->inp_cred, &xuc);
1916                 INP_RUNLOCK(inp);
1917         } else
1918                 error = ENOENT;
1919         if (error == 0)
1920                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1921         return (error);
1922 }
1923
1924 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1925     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1926     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1927 #endif /* INET6 */
1928
1929
1930 #ifdef INET
1931 void
1932 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1933 {
1934         struct ip *ip = vip;
1935         struct tcphdr *th;
1936         struct in_addr faddr;
1937         struct inpcb *inp;
1938         struct tcpcb *tp;
1939         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1940         struct icmp *icp;
1941         struct in_conninfo inc;
1942         tcp_seq icmp_tcp_seq;
1943         int mtu;
1944
1945         faddr = ((struct sockaddr_in *)sa)->sin_addr;
1946         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1947                 return;
1948
1949         if (cmd == PRC_MSGSIZE)
1950                 notify = tcp_mtudisc_notify;
1951         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1952                 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1953                 notify = tcp_drop_syn_sent;
1954
1955         /*
1956          * Hostdead is ugly because it goes linearly through all PCBs.
1957          * XXX: We never get this from ICMP, otherwise it makes an
1958          * excellent DoS attack on machines with many connections.
1959          */
1960         else if (cmd == PRC_HOSTDEAD)
1961                 ip = NULL;
1962         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1963                 return;
1964
1965         if (ip == NULL) {
1966                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1967                 return;
1968         }
1969
1970         icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1971         th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1972         INP_INFO_RLOCK(&V_tcbinfo);
1973         inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1974             th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1975         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1976                 /* signal EHOSTDOWN, as it flushes the cached route */
1977                 inp = (*notify)(inp, EHOSTDOWN);
1978                 if (inp != NULL)
1979                         INP_WUNLOCK(inp);
1980         } else if (inp != NULL)  {
1981                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1982                     !(inp->inp_flags & INP_DROPPED) &&
1983                     !(inp->inp_socket == NULL)) {
1984                         icmp_tcp_seq = ntohl(th->th_seq);
1985                         tp = intotcpcb(inp);
1986                         if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1987                             SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1988                                 if (cmd == PRC_MSGSIZE) {
1989                                         /*
1990                                          * MTU discovery:
1991                                          * If we got a needfrag set the MTU
1992                                          * in the route to the suggested new
1993                                          * value (if given) and then notify.
1994                                          */
1995                                         mtu = ntohs(icp->icmp_nextmtu);
1996                                         /*
1997                                          * If no alternative MTU was
1998                                          * proposed, try the next smaller
1999                                          * one.
2000                                          */
2001                                         if (!mtu)
2002                                                 mtu = ip_next_mtu(
2003                                                     ntohs(ip->ip_len), 1);
2004                                         if (mtu < V_tcp_minmss +
2005                                             sizeof(struct tcpiphdr))
2006                                                 mtu = V_tcp_minmss +
2007                                                     sizeof(struct tcpiphdr);
2008                                         /*
2009                                          * Only process the offered MTU if it
2010                                          * is smaller than the current one.
2011                                          */
2012                                         if (mtu < tp->t_maxseg +
2013                                             sizeof(struct tcpiphdr)) {
2014                                                 bzero(&inc, sizeof(inc));
2015                                                 inc.inc_faddr = faddr;
2016                                                 inc.inc_fibnum =
2017                                                     inp->inp_inc.inc_fibnum;
2018                                                 tcp_hc_updatemtu(&inc, mtu);
2019                                                 tcp_mtudisc(inp, mtu);
2020                                         }
2021                                 } else
2022                                         inp = (*notify)(inp,
2023                                             inetctlerrmap[cmd]);
2024                         }
2025                 }
2026                 if (inp != NULL)
2027                         INP_WUNLOCK(inp);
2028         } else {
2029                 bzero(&inc, sizeof(inc));
2030                 inc.inc_fport = th->th_dport;
2031                 inc.inc_lport = th->th_sport;
2032                 inc.inc_faddr = faddr;
2033                 inc.inc_laddr = ip->ip_src;
2034                 syncache_unreach(&inc, th);
2035         }
2036         INP_INFO_RUNLOCK(&V_tcbinfo);
2037 }
2038 #endif /* INET */
2039
2040 #ifdef INET6
2041 void
2042 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2043 {
2044         struct in6_addr *dst;
2045         struct tcphdr *th;
2046         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2047         struct ip6_hdr *ip6;
2048         struct mbuf *m;
2049         struct inpcb *inp;
2050         struct tcpcb *tp;
2051         struct icmp6_hdr *icmp6;
2052         struct ip6ctlparam *ip6cp = NULL;
2053         const struct sockaddr_in6 *sa6_src = NULL;
2054         struct in_conninfo inc;
2055         tcp_seq icmp_tcp_seq;
2056         unsigned int mtu;
2057         unsigned int off;
2058
2059
2060         if (sa->sa_family != AF_INET6 ||
2061             sa->sa_len != sizeof(struct sockaddr_in6))
2062                 return;
2063
2064         /* if the parameter is from icmp6, decode it. */
2065         if (d != NULL) {
2066                 ip6cp = (struct ip6ctlparam *)d;
2067                 icmp6 = ip6cp->ip6c_icmp6;
2068                 m = ip6cp->ip6c_m;
2069                 ip6 = ip6cp->ip6c_ip6;
2070                 off = ip6cp->ip6c_off;
2071                 sa6_src = ip6cp->ip6c_src;
2072                 dst = ip6cp->ip6c_finaldst;
2073         } else {
2074                 m = NULL;
2075                 ip6 = NULL;
2076                 off = 0;        /* fool gcc */
2077                 sa6_src = &sa6_any;
2078                 dst = NULL;
2079         }
2080
2081         if (cmd == PRC_MSGSIZE)
2082                 notify = tcp_mtudisc_notify;
2083         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2084                 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) &&
2085                 ip6 != NULL)
2086                 notify = tcp_drop_syn_sent;
2087
2088         /*
2089          * Hostdead is ugly because it goes linearly through all PCBs.
2090          * XXX: We never get this from ICMP, otherwise it makes an
2091          * excellent DoS attack on machines with many connections.
2092          */
2093         else if (cmd == PRC_HOSTDEAD)
2094                 ip6 = NULL;
2095         else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2096                 return;
2097
2098         if (ip6 == NULL) {
2099                 in6_pcbnotify(&V_tcbinfo, sa, 0,
2100                               (const struct sockaddr *)sa6_src,
2101                               0, cmd, NULL, notify);
2102                 return;
2103         }
2104
2105         /* Check if we can safely get the ports from the tcp hdr */
2106         if (m == NULL ||
2107             (m->m_pkthdr.len <
2108                 (int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2109                 return;
2110         }
2111
2112         th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2113         INP_INFO_RLOCK(&V_tcbinfo);
2114         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2115             &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2116         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2117                 /* signal EHOSTDOWN, as it flushes the cached route */
2118                 inp = (*notify)(inp, EHOSTDOWN);
2119                 if (inp != NULL)
2120                         INP_WUNLOCK(inp);
2121         } else if (inp != NULL)  {
2122                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2123                     !(inp->inp_flags & INP_DROPPED) &&
2124                     !(inp->inp_socket == NULL)) {
2125                         icmp_tcp_seq = ntohl(th->th_seq);
2126                         tp = intotcpcb(inp);
2127                         if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2128                             SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2129                                 if (cmd == PRC_MSGSIZE) {
2130                                         /*
2131                                          * MTU discovery:
2132                                          * If we got a needfrag set the MTU
2133                                          * in the route to the suggested new
2134                                          * value (if given) and then notify.
2135                                          */
2136                                         mtu = ntohl(icmp6->icmp6_mtu);
2137                                         /*
2138                                          * If no alternative MTU was
2139                                          * proposed, or the proposed
2140                                          * MTU was too small, set to
2141                                          * the min.
2142                                          */
2143                                         if (mtu < IPV6_MMTU)
2144                                                 mtu = IPV6_MMTU - 8;
2145
2146
2147                                         bzero(&inc, sizeof(inc));
2148                                         inc.inc_fibnum = M_GETFIB(m);
2149                                         inc.inc_flags |= INC_ISIPV6;
2150                                         inc.inc6_faddr = *dst;
2151                                         if (in6_setscope(&inc.inc6_faddr,
2152                                                 m->m_pkthdr.rcvif, NULL))
2153                                                 goto unlock_inp;
2154
2155                                         /*
2156                                          * Only process the offered MTU if it
2157                                          * is smaller than the current one.
2158                                          */
2159                                         if (mtu < tp->t_maxseg +
2160                                             (sizeof (*th) + sizeof (*ip6))) {
2161                                                 tcp_hc_updatemtu(&inc, mtu);
2162                                                 tcp_mtudisc(inp, mtu);
2163                                                 ICMP6STAT_INC(icp6s_pmtuchg);
2164                                         }
2165                                 } else
2166                                         inp = (*notify)(inp,
2167                                             inet6ctlerrmap[cmd]);
2168                         }
2169                 }
2170 unlock_inp:
2171                 if (inp != NULL)
2172                         INP_WUNLOCK(inp);
2173         } else {
2174                 bzero(&inc, sizeof(inc));
2175                 inc.inc_fibnum = M_GETFIB(m);
2176                 inc.inc_flags |= INC_ISIPV6;
2177                 inc.inc_fport = th->th_dport;
2178                 inc.inc_lport = th->th_sport;
2179                 inc.inc6_faddr = *dst;
2180                 inc.inc6_laddr = ip6->ip6_src;
2181                 syncache_unreach(&inc, th);
2182         }
2183         INP_INFO_RUNLOCK(&V_tcbinfo);
2184 }
2185 #endif /* INET6 */
2186
2187
2188 /*
2189  * Following is where TCP initial sequence number generation occurs.
2190  *
2191  * There are two places where we must use initial sequence numbers:
2192  * 1.  In SYN-ACK packets.
2193  * 2.  In SYN packets.
2194  *
2195  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2196  * tcp_syncache.c for details.
2197  *
2198  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2199  * depends on this property.  In addition, these ISNs should be
2200  * unguessable so as to prevent connection hijacking.  To satisfy
2201  * the requirements of this situation, the algorithm outlined in
2202  * RFC 1948 is used, with only small modifications.
2203  *
2204  * Implementation details:
2205  *
2206  * Time is based off the system timer, and is corrected so that it
2207  * increases by one megabyte per second.  This allows for proper
2208  * recycling on high speed LANs while still leaving over an hour
2209  * before rollover.
2210  *
2211  * As reading the *exact* system time is too expensive to be done
2212  * whenever setting up a TCP connection, we increment the time
2213  * offset in two ways.  First, a small random positive increment
2214  * is added to isn_offset for each connection that is set up.
2215  * Second, the function tcp_isn_tick fires once per clock tick
2216  * and increments isn_offset as necessary so that sequence numbers
2217  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2218  * random positive increments serve only to ensure that the same
2219  * exact sequence number is never sent out twice (as could otherwise
2220  * happen when a port is recycled in less than the system tick
2221  * interval.)
2222  *
2223  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2224  * between seeding of isn_secret.  This is normally set to zero,
2225  * as reseeding should not be necessary.
2226  *
2227  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2228  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2229  * general, this means holding an exclusive (write) lock.
2230  */
2231
2232 #define ISN_BYTES_PER_SECOND 1048576
2233 #define ISN_STATIC_INCREMENT 4096
2234 #define ISN_RANDOM_INCREMENT (4096 - 1)
2235
2236 static VNET_DEFINE(u_char, isn_secret[32]);
2237 static VNET_DEFINE(int, isn_last);
2238 static VNET_DEFINE(int, isn_last_reseed);
2239 static VNET_DEFINE(u_int32_t, isn_offset);
2240 static VNET_DEFINE(u_int32_t, isn_offset_old);
2241
2242 #define V_isn_secret                    VNET(isn_secret)
2243 #define V_isn_last                      VNET(isn_last)
2244 #define V_isn_last_reseed               VNET(isn_last_reseed)
2245 #define V_isn_offset                    VNET(isn_offset)
2246 #define V_isn_offset_old                VNET(isn_offset_old)
2247
2248 tcp_seq
2249 tcp_new_isn(struct tcpcb *tp)
2250 {
2251         MD5_CTX isn_ctx;
2252         u_int32_t md5_buffer[4];
2253         tcp_seq new_isn;
2254         u_int32_t projected_offset;
2255
2256         INP_WLOCK_ASSERT(tp->t_inpcb);
2257
2258         ISN_LOCK();
2259         /* Seed if this is the first use, reseed if requested. */
2260         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2261              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2262                 < (u_int)ticks))) {
2263                 read_random(&V_isn_secret, sizeof(V_isn_secret));
2264                 V_isn_last_reseed = ticks;
2265         }
2266
2267         /* Compute the md5 hash and return the ISN. */
2268         MD5Init(&isn_ctx);
2269         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2270         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2271 #ifdef INET6
2272         if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2273                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2274                           sizeof(struct in6_addr));
2275                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2276                           sizeof(struct in6_addr));
2277         } else
2278 #endif
2279         {
2280                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2281                           sizeof(struct in_addr));
2282                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2283                           sizeof(struct in_addr));
2284         }
2285         MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2286         MD5Final((u_char *) &md5_buffer, &isn_ctx);
2287         new_isn = (tcp_seq) md5_buffer[0];
2288         V_isn_offset += ISN_STATIC_INCREMENT +
2289                 (arc4random() & ISN_RANDOM_INCREMENT);
2290         if (ticks != V_isn_last) {
2291                 projected_offset = V_isn_offset_old +
2292                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2293                 if (SEQ_GT(projected_offset, V_isn_offset))
2294                         V_isn_offset = projected_offset;
2295                 V_isn_offset_old = V_isn_offset;
2296                 V_isn_last = ticks;
2297         }
2298         new_isn += V_isn_offset;
2299         ISN_UNLOCK();
2300         return (new_isn);
2301 }
2302
2303 /*
2304  * When a specific ICMP unreachable message is received and the
2305  * connection state is SYN-SENT, drop the connection.  This behavior
2306  * is controlled by the icmp_may_rst sysctl.
2307  */
2308 struct inpcb *
2309 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2310 {
2311         struct tcpcb *tp;
2312
2313         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2314         INP_WLOCK_ASSERT(inp);
2315
2316         if ((inp->inp_flags & INP_TIMEWAIT) ||
2317             (inp->inp_flags & INP_DROPPED))
2318                 return (inp);
2319
2320         tp = intotcpcb(inp);
2321         if (tp->t_state != TCPS_SYN_SENT)
2322                 return (inp);
2323
2324         tp = tcp_drop(tp, errno);
2325         if (tp != NULL)
2326                 return (inp);
2327         else
2328                 return (NULL);
2329 }
2330
2331 /*
2332  * When `need fragmentation' ICMP is received, update our idea of the MSS
2333  * based on the new value. Also nudge TCP to send something, since we
2334  * know the packet we just sent was dropped.
2335  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2336  */
2337 static struct inpcb *
2338 tcp_mtudisc_notify(struct inpcb *inp, int error)
2339 {
2340
2341         tcp_mtudisc(inp, -1);
2342         return (inp);
2343 }
2344
2345 static void
2346 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2347 {
2348         struct tcpcb *tp;
2349         struct socket *so;
2350
2351         INP_WLOCK_ASSERT(inp);
2352         if ((inp->inp_flags & INP_TIMEWAIT) ||
2353             (inp->inp_flags & INP_DROPPED))
2354                 return;
2355
2356         tp = intotcpcb(inp);
2357         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2358
2359         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2360
2361         so = inp->inp_socket;
2362         SOCKBUF_LOCK(&so->so_snd);
2363         /* If the mss is larger than the socket buffer, decrease the mss. */
2364         if (so->so_snd.sb_hiwat < tp->t_maxseg)
2365                 tp->t_maxseg = so->so_snd.sb_hiwat;
2366         SOCKBUF_UNLOCK(&so->so_snd);
2367
2368         TCPSTAT_INC(tcps_mturesent);
2369         tp->t_rtttime = 0;
2370         tp->snd_nxt = tp->snd_una;
2371         tcp_free_sackholes(tp);
2372         tp->snd_recover = tp->snd_max;
2373         if (tp->t_flags & TF_SACK_PERMIT)
2374                 EXIT_FASTRECOVERY(tp->t_flags);
2375         tp->t_fb->tfb_tcp_output(tp);
2376 }
2377
2378 #ifdef INET
2379 /*
2380  * Look-up the routing entry to the peer of this inpcb.  If no route
2381  * is found and it cannot be allocated, then return 0.  This routine
2382  * is called by TCP routines that access the rmx structure and by
2383  * tcp_mss_update to get the peer/interface MTU.
2384  */
2385 u_long
2386 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2387 {
2388         struct nhop4_extended nh4;
2389         struct ifnet *ifp;
2390         u_long maxmtu = 0;
2391
2392         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2393
2394         if (inc->inc_faddr.s_addr != INADDR_ANY) {
2395
2396                 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2397                     NHR_REF, 0, &nh4) != 0)
2398                         return (0);
2399
2400                 ifp = nh4.nh_ifp;
2401                 maxmtu = nh4.nh_mtu;
2402
2403                 /* Report additional interface capabilities. */
2404                 if (cap != NULL) {
2405                         if (ifp->if_capenable & IFCAP_TSO4 &&
2406                             ifp->if_hwassist & CSUM_TSO) {
2407                                 cap->ifcap |= CSUM_TSO;
2408                                 cap->tsomax = ifp->if_hw_tsomax;
2409                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2410                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2411                         }
2412                 }
2413                 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2414         }
2415         return (maxmtu);
2416 }
2417 #endif /* INET */
2418
2419 #ifdef INET6
2420 u_long
2421 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2422 {
2423         struct nhop6_extended nh6;
2424         struct in6_addr dst6;
2425         uint32_t scopeid;
2426         struct ifnet *ifp;
2427         u_long maxmtu = 0;
2428
2429         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2430
2431         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2432                 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2433                 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2434                     0, &nh6) != 0)
2435                         return (0);
2436
2437                 ifp = nh6.nh_ifp;
2438                 maxmtu = nh6.nh_mtu;
2439
2440                 /* Report additional interface capabilities. */
2441                 if (cap != NULL) {
2442                         if (ifp->if_capenable & IFCAP_TSO6 &&
2443                             ifp->if_hwassist & CSUM_TSO) {
2444                                 cap->ifcap |= CSUM_TSO;
2445                                 cap->tsomax = ifp->if_hw_tsomax;
2446                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2447                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2448                         }
2449                 }
2450                 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2451         }
2452
2453         return (maxmtu);
2454 }
2455 #endif /* INET6 */
2456
2457 /*
2458  * Calculate effective SMSS per RFC5681 definition for a given TCP
2459  * connection at its current state, taking into account SACK and etc.
2460  */
2461 u_int
2462 tcp_maxseg(const struct tcpcb *tp)
2463 {
2464         u_int optlen;
2465
2466         if (tp->t_flags & TF_NOOPT)
2467                 return (tp->t_maxseg);
2468
2469         /*
2470          * Here we have a simplified code from tcp_addoptions(),
2471          * without a proper loop, and having most of paddings hardcoded.
2472          * We might make mistakes with padding here in some edge cases,
2473          * but this is harmless, since result of tcp_maxseg() is used
2474          * only in cwnd and ssthresh estimations.
2475          */
2476 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
2477         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2478                 if (tp->t_flags & TF_RCVD_TSTMP)
2479                         optlen = TCPOLEN_TSTAMP_APPA;
2480                 else
2481                         optlen = 0;
2482 #ifdef TCP_SIGNATURE
2483                 if (tp->t_flags & TF_SIGNATURE)
2484                         optlen += PAD(TCPOLEN_SIGNATURE);
2485 #endif
2486                 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2487                         optlen += TCPOLEN_SACKHDR;
2488                         optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2489                         optlen = PAD(optlen);
2490                 }
2491         } else {
2492                 if (tp->t_flags & TF_REQ_TSTMP)
2493                         optlen = TCPOLEN_TSTAMP_APPA;
2494                 else
2495                         optlen = PAD(TCPOLEN_MAXSEG);
2496                 if (tp->t_flags & TF_REQ_SCALE)
2497                         optlen += PAD(TCPOLEN_WINDOW);
2498 #ifdef TCP_SIGNATURE
2499                 if (tp->t_flags & TF_SIGNATURE)
2500                         optlen += PAD(TCPOLEN_SIGNATURE);
2501 #endif
2502                 if (tp->t_flags & TF_SACK_PERMIT)
2503                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
2504         }
2505 #undef PAD
2506         optlen = min(optlen, TCP_MAXOLEN);
2507         return (tp->t_maxseg - optlen);
2508 }
2509
2510 #ifdef IPSEC
2511 /* compute ESP/AH header size for TCP, including outer IP header. */
2512 size_t
2513 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2514 {
2515         struct inpcb *inp;
2516         struct mbuf *m;
2517         size_t hdrsiz;
2518         struct ip *ip;
2519 #ifdef INET6
2520         struct ip6_hdr *ip6;
2521 #endif
2522         struct tcphdr *th;
2523
2524         if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2525                 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2526                 return (0);
2527         m = m_gethdr(M_NOWAIT, MT_DATA);
2528         if (!m)
2529                 return (0);
2530
2531 #ifdef INET6
2532         if ((inp->inp_vflag & INP_IPV6) != 0) {
2533                 ip6 = mtod(m, struct ip6_hdr *);
2534                 th = (struct tcphdr *)(ip6 + 1);
2535                 m->m_pkthdr.len = m->m_len =
2536                         sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2537                 tcpip_fillheaders(inp, ip6, th);
2538                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2539         } else
2540 #endif /* INET6 */
2541         {
2542                 ip = mtod(m, struct ip *);
2543                 th = (struct tcphdr *)(ip + 1);
2544                 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2545                 tcpip_fillheaders(inp, ip, th);
2546                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2547         }
2548
2549         m_free(m);
2550         return (hdrsiz);
2551 }
2552 #endif /* IPSEC */
2553
2554 #ifdef TCP_SIGNATURE
2555 /*
2556  * Callback function invoked by m_apply() to digest TCP segment data
2557  * contained within an mbuf chain.
2558  */
2559 static int
2560 tcp_signature_apply(void *fstate, void *data, u_int len)
2561 {
2562
2563         MD5Update(fstate, (u_char *)data, len);
2564         return (0);
2565 }
2566
2567 /*
2568  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2569  * search with the destination IP address, and a 'magic SPI' to be
2570  * determined by the application. This is hardcoded elsewhere to 1179
2571 */
2572 struct secasvar *
2573 tcp_get_sav(struct mbuf *m, u_int direction)
2574 {
2575         union sockaddr_union dst;
2576         struct secasvar *sav;
2577         struct ip *ip;
2578 #ifdef INET6
2579         struct ip6_hdr *ip6;
2580         char ip6buf[INET6_ADDRSTRLEN];
2581 #endif
2582
2583         /* Extract the destination from the IP header in the mbuf. */
2584         bzero(&dst, sizeof(union sockaddr_union));
2585         ip = mtod(m, struct ip *);
2586 #ifdef INET6
2587         ip6 = NULL;     /* Make the compiler happy. */
2588 #endif
2589         switch (ip->ip_v) {
2590 #ifdef INET
2591         case IPVERSION:
2592                 dst.sa.sa_len = sizeof(struct sockaddr_in);
2593                 dst.sa.sa_family = AF_INET;
2594                 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2595                     ip->ip_src : ip->ip_dst;
2596                 break;
2597 #endif
2598 #ifdef INET6
2599         case (IPV6_VERSION >> 4):
2600                 ip6 = mtod(m, struct ip6_hdr *);
2601                 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2602                 dst.sa.sa_family = AF_INET6;
2603                 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2604                     ip6->ip6_src : ip6->ip6_dst;
2605                 break;
2606 #endif
2607         default:
2608                 return (NULL);
2609                 /* NOTREACHED */
2610                 break;
2611         }
2612
2613         /* Look up an SADB entry which matches the address of the peer. */
2614         sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2615         if (sav == NULL) {
2616                 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2617                     (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2618 #ifdef INET6
2619                         (ip->ip_v == (IPV6_VERSION >> 4)) ?
2620                             ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2621 #endif
2622                         "(unsupported)"));
2623         }
2624
2625         return (sav);
2626 }
2627
2628 /*
2629  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2630  *
2631  * Parameters:
2632  * m            pointer to head of mbuf chain
2633  * len          length of TCP segment data, excluding options
2634  * optlen       length of TCP segment options
2635  * buf          pointer to storage for computed MD5 digest
2636  * sav          pointer to security assosiation
2637  *
2638  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2639  * When called from tcp_input(), we can be sure that th_sum has been
2640  * zeroed out and verified already.
2641  *
2642  * Releases reference to SADB key before return.
2643  *
2644  * Return 0 if successful, otherwise return -1.
2645  *
2646  */
2647 int
2648 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2649     u_char *buf, struct secasvar *sav)
2650 {
2651 #ifdef INET
2652         struct ippseudo ippseudo;
2653 #endif
2654         MD5_CTX ctx;
2655         int doff;
2656         struct ip *ip;
2657 #ifdef INET
2658         struct ipovly *ipovly;
2659 #endif
2660         struct tcphdr *th;
2661 #ifdef INET6
2662         struct ip6_hdr *ip6;
2663         struct in6_addr in6;
2664         uint32_t plen;
2665         uint16_t nhdr;
2666 #endif
2667         u_short savecsum;
2668
2669         KASSERT(m != NULL, ("NULL mbuf chain"));
2670         KASSERT(buf != NULL, ("NULL signature pointer"));
2671
2672         /* Extract the destination from the IP header in the mbuf. */
2673         ip = mtod(m, struct ip *);
2674 #ifdef INET6
2675         ip6 = NULL;     /* Make the compiler happy. */
2676 #endif
2677
2678         MD5Init(&ctx);
2679         /*
2680          * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2681          *
2682          * XXX The ippseudo header MUST be digested in network byte order,
2683          * or else we'll fail the regression test. Assume all fields we've
2684          * been doing arithmetic on have been in host byte order.
2685          * XXX One cannot depend on ipovly->ih_len here. When called from
2686          * tcp_output(), the underlying ip_len member has not yet been set.
2687          */
2688         switch (ip->ip_v) {
2689 #ifdef INET
2690         case IPVERSION:
2691                 ipovly = (struct ipovly *)ip;
2692                 ippseudo.ippseudo_src = ipovly->ih_src;
2693                 ippseudo.ippseudo_dst = ipovly->ih_dst;
2694                 ippseudo.ippseudo_pad = 0;
2695                 ippseudo.ippseudo_p = IPPROTO_TCP;
2696                 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2697                     optlen);
2698                 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2699
2700                 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2701                 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2702                 break;
2703 #endif
2704 #ifdef INET6
2705         /*
2706          * RFC 2385, 2.0  Proposal
2707          * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2708          * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2709          * extended next header value (to form 32 bits), and 32-bit segment
2710          * length.
2711          * Note: Upper-Layer Packet Length comes before Next Header.
2712          */
2713         case (IPV6_VERSION >> 4):
2714                 in6 = ip6->ip6_src;
2715                 in6_clearscope(&in6);
2716                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2717                 in6 = ip6->ip6_dst;
2718                 in6_clearscope(&in6);
2719                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2720                 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2721                 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2722                 nhdr = 0;
2723                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2724                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2725                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2726                 nhdr = IPPROTO_TCP;
2727                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2728
2729                 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2730                 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2731                 break;
2732 #endif
2733         default:
2734                 KEY_FREESAV(&sav);
2735                 return (-1);
2736                 /* NOTREACHED */
2737                 break;
2738         }
2739
2740
2741         /*
2742          * Step 2: Update MD5 hash with TCP header, excluding options.
2743          * The TCP checksum must be set to zero.
2744          */
2745         savecsum = th->th_sum;
2746         th->th_sum = 0;
2747         MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2748         th->th_sum = savecsum;
2749
2750         /*
2751          * Step 3: Update MD5 hash with TCP segment data.
2752          *         Use m_apply() to avoid an early m_pullup().
2753          */
2754         if (len > 0)
2755                 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2756
2757         /*
2758          * Step 4: Update MD5 hash with shared secret.
2759          */
2760         MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2761         MD5Final(buf, &ctx);
2762
2763         key_sa_recordxfer(sav, m);
2764         KEY_FREESAV(&sav);
2765         return (0);
2766 }
2767
2768 /*
2769  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2770  *
2771  * Return 0 if successful, otherwise return -1.
2772  */
2773 int
2774 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2775     u_char *buf, u_int direction)
2776 {
2777         struct secasvar *sav;
2778
2779         if ((sav = tcp_get_sav(m, direction)) == NULL)
2780                 return (-1);
2781
2782         return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2783 }
2784
2785 /*
2786  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2787  *
2788  * Parameters:
2789  * m            pointer to head of mbuf chain
2790  * len          length of TCP segment data, excluding options
2791  * optlen       length of TCP segment options
2792  * buf          pointer to storage for computed MD5 digest
2793  * direction    direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2794  *
2795  * Return 1 if successful, otherwise return 0.
2796  */
2797 int
2798 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2799     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2800 {
2801         char tmpdigest[TCP_SIGLEN];
2802
2803         if (tcp_sig_checksigs == 0)
2804                 return (1);
2805         if ((tcpbflag & TF_SIGNATURE) == 0) {
2806                 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2807
2808                         /*
2809                          * If this socket is not expecting signature but
2810                          * the segment contains signature just fail.
2811                          */
2812                         TCPSTAT_INC(tcps_sig_err_sigopt);
2813                         TCPSTAT_INC(tcps_sig_rcvbadsig);
2814                         return (0);
2815                 }
2816
2817                 /* Signature is not expected, and not present in segment. */
2818                 return (1);
2819         }
2820
2821         /*
2822          * If this socket is expecting signature but the segment does not
2823          * contain any just fail.
2824          */
2825         if ((to->to_flags & TOF_SIGNATURE) == 0) {
2826                 TCPSTAT_INC(tcps_sig_err_nosigopt);
2827                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2828                 return (0);
2829         }
2830         if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2831             IPSEC_DIR_INBOUND) == -1) {
2832                 TCPSTAT_INC(tcps_sig_err_buildsig);
2833                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2834                 return (0);
2835         }
2836
2837         if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2838                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2839                 return (0);
2840         }
2841         TCPSTAT_INC(tcps_sig_rcvgoodsig);
2842         return (1);
2843 }
2844 #endif /* TCP_SIGNATURE */
2845
2846 static int
2847 sysctl_drop(SYSCTL_HANDLER_ARGS)
2848 {
2849         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2850         struct sockaddr_storage addrs[2];
2851         struct inpcb *inp;
2852         struct tcpcb *tp;
2853         struct tcptw *tw;
2854         struct sockaddr_in *fin, *lin;
2855 #ifdef INET6
2856         struct sockaddr_in6 *fin6, *lin6;
2857 #endif
2858         int error;
2859
2860         inp = NULL;
2861         fin = lin = NULL;
2862 #ifdef INET6
2863         fin6 = lin6 = NULL;
2864 #endif
2865         error = 0;
2866
2867         if (req->oldptr != NULL || req->oldlen != 0)
2868                 return (EINVAL);
2869         if (req->newptr == NULL)
2870                 return (EPERM);
2871         if (req->newlen < sizeof(addrs))
2872                 return (ENOMEM);
2873         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2874         if (error)
2875                 return (error);
2876
2877         switch (addrs[0].ss_family) {
2878 #ifdef INET6
2879         case AF_INET6:
2880                 fin6 = (struct sockaddr_in6 *)&addrs[0];
2881                 lin6 = (struct sockaddr_in6 *)&addrs[1];
2882                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2883                     lin6->sin6_len != sizeof(struct sockaddr_in6))
2884                         return (EINVAL);
2885                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2886                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2887                                 return (EINVAL);
2888                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2889                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2890                         fin = (struct sockaddr_in *)&addrs[0];
2891                         lin = (struct sockaddr_in *)&addrs[1];
2892                         break;
2893                 }
2894                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2895                 if (error)
2896                         return (error);
2897                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2898                 if (error)
2899                         return (error);
2900                 break;
2901 #endif
2902 #ifdef INET
2903         case AF_INET:
2904                 fin = (struct sockaddr_in *)&addrs[0];
2905                 lin = (struct sockaddr_in *)&addrs[1];
2906                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2907                     lin->sin_len != sizeof(struct sockaddr_in))
2908                         return (EINVAL);
2909                 break;
2910 #endif
2911         default:
2912                 return (EINVAL);
2913         }
2914         INP_INFO_RLOCK(&V_tcbinfo);
2915         switch (addrs[0].ss_family) {
2916 #ifdef INET6
2917         case AF_INET6:
2918                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2919                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2920                     INPLOOKUP_WLOCKPCB, NULL);
2921                 break;
2922 #endif
2923 #ifdef INET
2924         case AF_INET:
2925                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2926                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2927                 break;
2928 #endif
2929         }
2930         if (inp != NULL) {
2931                 if (inp->inp_flags & INP_TIMEWAIT) {
2932                         /*
2933                          * XXXRW: There currently exists a state where an
2934                          * inpcb is present, but its timewait state has been
2935                          * discarded.  For now, don't allow dropping of this
2936                          * type of inpcb.
2937                          */
2938                         tw = intotw(inp);
2939                         if (tw != NULL)
2940                                 tcp_twclose(tw, 0);
2941                         else
2942                                 INP_WUNLOCK(inp);
2943                 } else if (!(inp->inp_flags & INP_DROPPED) &&
2944                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2945                         tp = intotcpcb(inp);
2946                         tp = tcp_drop(tp, ECONNABORTED);
2947                         if (tp != NULL)
2948                                 INP_WUNLOCK(inp);
2949                 } else
2950                         INP_WUNLOCK(inp);
2951         } else
2952                 error = ESRCH;
2953         INP_INFO_RUNLOCK(&V_tcbinfo);
2954         return (error);
2955 }
2956
2957 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2958     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2959     0, sysctl_drop, "", "Drop TCP connection");
2960
2961 /*
2962  * Generate a standardized TCP log line for use throughout the
2963  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2964  * allow use in the interrupt context.
2965  *
2966  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2967  * NB: The function may return NULL if memory allocation failed.
2968  *
2969  * Due to header inclusion and ordering limitations the struct ip
2970  * and ip6_hdr pointers have to be passed as void pointers.
2971  */
2972 char *
2973 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2974     const void *ip6hdr)
2975 {
2976
2977         /* Is logging enabled? */
2978         if (tcp_log_in_vain == 0)
2979                 return (NULL);
2980
2981         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2982 }
2983
2984 char *
2985 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2986     const void *ip6hdr)
2987 {
2988
2989         /* Is logging enabled? */
2990         if (tcp_log_debug == 0)
2991                 return (NULL);
2992
2993         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2994 }
2995
2996 static char *
2997 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2998     const void *ip6hdr)
2999 {
3000         char *s, *sp;
3001         size_t size;
3002         struct ip *ip;
3003 #ifdef INET6
3004         const struct ip6_hdr *ip6;
3005
3006         ip6 = (const struct ip6_hdr *)ip6hdr;
3007 #endif /* INET6 */
3008         ip = (struct ip *)ip4hdr;
3009
3010         /*
3011          * The log line looks like this:
3012          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3013          */
3014         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3015             sizeof(PRINT_TH_FLAGS) + 1 +
3016 #ifdef INET6
3017             2 * INET6_ADDRSTRLEN;
3018 #else
3019             2 * INET_ADDRSTRLEN;
3020 #endif /* INET6 */
3021
3022         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3023         if (s == NULL)
3024                 return (NULL);
3025
3026         strcat(s, "TCP: [");
3027         sp = s + strlen(s);
3028
3029         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3030                 inet_ntoa_r(inc->inc_faddr, sp);
3031                 sp = s + strlen(s);
3032                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3033                 sp = s + strlen(s);
3034                 inet_ntoa_r(inc->inc_laddr, sp);
3035                 sp = s + strlen(s);
3036                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3037 #ifdef INET6
3038         } else if (inc) {
3039                 ip6_sprintf(sp, &inc->inc6_faddr);
3040                 sp = s + strlen(s);
3041                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3042                 sp = s + strlen(s);
3043                 ip6_sprintf(sp, &inc->inc6_laddr);
3044                 sp = s + strlen(s);
3045                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3046         } else if (ip6 && th) {
3047                 ip6_sprintf(sp, &ip6->ip6_src);
3048                 sp = s + strlen(s);
3049                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3050                 sp = s + strlen(s);
3051                 ip6_sprintf(sp, &ip6->ip6_dst);
3052                 sp = s + strlen(s);
3053                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3054 #endif /* INET6 */
3055 #ifdef INET
3056         } else if (ip && th) {
3057                 inet_ntoa_r(ip->ip_src, sp);
3058                 sp = s + strlen(s);
3059                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3060                 sp = s + strlen(s);
3061                 inet_ntoa_r(ip->ip_dst, sp);
3062                 sp = s + strlen(s);
3063                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3064 #endif /* INET */
3065         } else {
3066                 free(s, M_TCPLOG);
3067                 return (NULL);
3068         }
3069         sp = s + strlen(s);
3070         if (th)
3071                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3072         if (*(s + size - 1) != '\0')
3073                 panic("%s: string too long", __func__);
3074         return (s);
3075 }
3076
3077 /*
3078  * A subroutine which makes it easy to track TCP state changes with DTrace.
3079  * This function shouldn't be called for t_state initializations that don't
3080  * correspond to actual TCP state transitions.
3081  */
3082 void
3083 tcp_state_change(struct tcpcb *tp, int newstate)
3084 {
3085 #if defined(KDTRACE_HOOKS)
3086         int pstate = tp->t_state;
3087 #endif
3088
3089         TCPSTATES_DEC(tp->t_state);
3090         TCPSTATES_INC(newstate);
3091         tp->t_state = newstate;
3092         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3093 }