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 #ifdef TCP_HHOOK
  46 #include <sys/hhook.h>
  47 #endif
  48 #include <sys/kernel.h>
  49 #ifdef TCP_HHOOK
  50 #include <sys/khelp.h>
  51 #endif
  52 #include <sys/sysctl.h>
  53 #include <sys/jail.h>
  54 #include <sys/malloc.h>
  55 #include <sys/refcount.h>
  56 #include <sys/mbuf.h>
  57 #ifdef INET6
  58 #include <sys/domain.h>
  59 #endif
  60 #include <sys/priv.h>
  61 #include <sys/proc.h>
  62 #include <sys/sdt.h>
  63 #include <sys/socket.h>
  64 #include <sys/socketvar.h>
  65 #include <sys/protosw.h>
  66 #include <sys/random.h>
  67
  68 #include <vm/uma.h>
  69
  70 #include <net/route.h>
  71 #include <net/if.h>
  72 #include <net/if_var.h>
  73 #include <net/vnet.h>
  74
  75 #include <netinet/in.h>
  76 #include <netinet/in_fib.h>
  77 #include <netinet/in_kdtrace.h>
  78 #include <netinet/in_pcb.h>
  79 #include <netinet/in_systm.h>
  80 #include <netinet/in_var.h>
  81 #include <netinet/ip.h>
  82 #include <netinet/ip_icmp.h>
  83 #include <netinet/ip_var.h>
  84 #ifdef INET6
  85 #include <netinet/icmp6.h>
  86 #include <netinet/ip6.h>
  87 #include <netinet6/in6_fib.h>
  88 #include <netinet6/in6_pcb.h>
  89 #include <netinet6/ip6_var.h>
  90 #include <netinet6/scope6_var.h>
  91 #include <netinet6/nd6.h>
  92 #endif
  93
  94 #ifdef TCP_RFC7413
  95 #include <netinet/tcp_fastopen.h>
  96 #endif
  97 #include <netinet/tcp.h>
  98 #include <netinet/tcp_fsm.h>
  99 #include <netinet/tcp_seq.h>
 100 #include <netinet/tcp_timer.h>
 101 #include <netinet/tcp_var.h>
 102 #include <netinet/tcp_syncache.h>
 103 #include <netinet/cc/cc.h>
 104 #ifdef INET6
 105 #include <netinet6/tcp6_var.h>
 106 #endif
 107 #include <netinet/tcpip.h>
 108 #ifdef TCPPCAP
 109 #include <netinet/tcp_pcap.h>
 110 #endif
 111 #ifdef TCPDEBUG
 112 #include <netinet/tcp_debug.h>
 113 #endif
 114 #ifdef INET6
 115 #include <netinet6/ip6protosw.h>
 116 #endif
 117 #ifdef TCP_OFFLOAD
 118 #include <netinet/tcp_offload.h>
 119 #endif
 120
 121 #ifdef IPSEC
 122 #include <netipsec/ipsec.h>
 123 #include <netipsec/xform.h>
 124 #ifdef INET6
 125 #include <netipsec/ipsec6.h>
 126 #endif
 127 #include <netipsec/key.h>
 128 #include <sys/syslog.h>
 129 #endif /*IPSEC*/
 130
 131 #include <machine/in_cksum.h>
 132 #include <sys/md5.h>
 133
 134 #include <security/mac/mac_framework.h>
 135
 136 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
 137 #ifdef INET6
 138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
 139 #endif
 140
 141 struct rwlock tcp_function_lock;
 142
 143 static int
 144 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
 145 {
 146         int error, new;
 147
 148         new = V_tcp_mssdflt;
 149         error = sysctl_handle_int(oidp, &new, 0, req);
 150         if (error == 0 && req->newptr) {
 151                 if (new < TCP_MINMSS)
 152                         error = EINVAL;
 153                 else
 154                         V_tcp_mssdflt = new;
 155         }
 156         return (error);
 157 }
 158
 159 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
 160     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
 161     &sysctl_net_inet_tcp_mss_check, "I",
 162     "Default TCP Maximum Segment Size");
 163
 164 #ifdef INET6
 165 static int
 166 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
 167 {
 168         int error, new;
 169
 170         new = V_tcp_v6mssdflt;
 171         error = sysctl_handle_int(oidp, &new, 0, req);
 172         if (error == 0 && req->newptr) {
 173                 if (new < TCP_MINMSS)
 174                         error = EINVAL;
 175                 else
 176                         V_tcp_v6mssdflt = new;
 177         }
 178         return (error);
 179 }
 180
 181 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
 182     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
 183     &sysctl_net_inet_tcp_mss_v6_check, "I",
 184    "Default TCP Maximum Segment Size for IPv6");
 185 #endif /* INET6 */
 186
 187 /*
 188  * Minimum MSS we accept and use. This prevents DoS attacks where
 189  * we are forced to a ridiculous low MSS like 20 and send hundreds
 190  * of packets instead of one. The effect scales with the available
 191  * bandwidth and quickly saturates the CPU and network interface
 192  * with packet generation and sending. Set to zero to disable MINMSS
 193  * checking. This setting prevents us from sending too small packets.
 194  */
 195 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
 197      &VNET_NAME(tcp_minmss), 0,
 198     "Minimum TCP Maximum Segment Size");
 199
 200 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
 201 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
 202     &VNET_NAME(tcp_do_rfc1323), 0,
 203     "Enable rfc1323 (high performance TCP) extensions");
 204
 205 static int      tcp_log_debug = 0;
 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
 207     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
 208
 209 static int      tcp_tcbhashsize;
 210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
 211     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
 212
 213 static int      do_tcpdrain = 1;
 214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
 215     "Enable tcp_drain routine for extra help when low on mbufs");
 216
 217 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
 218     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
 219
 220 static VNET_DEFINE(int, icmp_may_rst) = 1;
 221 #define V_icmp_may_rst                  VNET(icmp_may_rst)
 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
 223     &VNET_NAME(icmp_may_rst), 0,
 224     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
 225
 226 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
 227 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
 229     &VNET_NAME(tcp_isn_reseed_interval), 0,
 230     "Seconds between reseeding of ISN secret");
 231
 232 static int      tcp_soreceive_stream;
 233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
 234     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
 235
 236 #ifdef TCP_SIGNATURE
 237 static int      tcp_sig_checksigs = 1;
 238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
 239     &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
 240 #endif
 241
 242 VNET_DEFINE(uma_zone_t, sack_hole_zone);
 243 #define V_sack_hole_zone                VNET(sack_hole_zone)
 244
 245 #ifdef TCP_HHOOK
 246 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
 247 #endif
 248
 249 static struct inpcb *tcp_notify(struct inpcb *, int);
 250 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
 251 static void tcp_mtudisc(struct inpcb *, int);
 252 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
 253                     void *ip4hdr, const void *ip6hdr);
 254
 255
 256 static struct tcp_function_block tcp_def_funcblk = {
 257         "default",
 258         tcp_output,
 259         tcp_do_segment,
 260         tcp_default_ctloutput,
 261         NULL,
 262         NULL,
 263         NULL,
 264         NULL,
 265         NULL,
 266         NULL,
 267         0,
 268         0
 269 };
 270
 271 int t_functions_inited = 0;
 272 struct tcp_funchead t_functions;
 273 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
 274
 275 static void
 276 init_tcp_functions(void)
 277 {
 278         if (t_functions_inited == 0) {
 279                 TAILQ_INIT(&t_functions);
 280                 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
 281                 t_functions_inited = 1;
 282         }
 283 }
 284
 285 static struct tcp_function_block *
 286 find_tcp_functions_locked(struct tcp_function_set *fs)
 287 {
 288         struct tcp_function *f;
 289         struct tcp_function_block *blk=NULL;
 290
 291         TAILQ_FOREACH(f, &t_functions, tf_next) {
 292                 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
 293                         blk = f->tf_fb;
 294                         break;
 295                 }
 296         }
 297         return(blk);
 298 }
 299
 300 static struct tcp_function_block *
 301 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
 302 {
 303         struct tcp_function_block *rblk=NULL;
 304         struct tcp_function *f;
 305
 306         TAILQ_FOREACH(f, &t_functions, tf_next) {
 307                 if (f->tf_fb == blk) {
 308                         rblk = blk;
 309                         if (s) {
 310                                 *s = f;
 311                         }
 312                         break;
 313                 }
 314         }
 315         return (rblk);
 316 }
 317
 318 struct tcp_function_block *
 319 find_and_ref_tcp_functions(struct tcp_function_set *fs)
 320 {
 321         struct tcp_function_block *blk;
 322
 323         rw_rlock(&tcp_function_lock);
 324         blk = find_tcp_functions_locked(fs);
 325         if (blk)
 326                 refcount_acquire(&blk->tfb_refcnt);
 327         rw_runlock(&tcp_function_lock);
 328         return(blk);
 329 }
 330
 331 struct tcp_function_block *
 332 find_and_ref_tcp_fb(struct tcp_function_block *blk)
 333 {
 334         struct tcp_function_block *rblk;
 335
 336         rw_rlock(&tcp_function_lock);
 337         rblk = find_tcp_fb_locked(blk, NULL);
 338         if (rblk)
 339                 refcount_acquire(&rblk->tfb_refcnt);
 340         rw_runlock(&tcp_function_lock);
 341         return(rblk);
 342 }
 343
 344
 345 static int
 346 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
 347 {
 348         int error=ENOENT;
 349         struct tcp_function_set fs;
 350         struct tcp_function_block *blk;
 351
 352         memset(&fs, 0, sizeof(fs));
 353         rw_rlock(&tcp_function_lock);
 354         blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
 355         if (blk) {
 356                 /* Found him */
 357                 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 358                 fs.pcbcnt = blk->tfb_refcnt;
 359         }
 360         rw_runlock(&tcp_function_lock);
 361         error = sysctl_handle_string(oidp, fs.function_set_name,
 362                                      sizeof(fs.function_set_name), req);
 363
 364         /* Check for error or no change */
 365         if (error != 0 || req->newptr == NULL)
 366                 return(error);
 367
 368         rw_wlock(&tcp_function_lock);
 369         blk = find_tcp_functions_locked(&fs);
 370         if ((blk == NULL) ||
 371             (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
 372                 error = ENOENT;
 373                 goto done;
 374         }
 375         tcp_func_set_ptr = blk;
 376 done:
 377         rw_wunlock(&tcp_function_lock);
 378         return (error);
 379 }
 380
 381 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
 382             CTLTYPE_STRING | CTLFLAG_RW,
 383             NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
 384             "Set/get the default TCP functions");
 385
 386 static int
 387 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
 388 {
 389         int error, cnt, linesz;
 390         struct tcp_function *f;
 391         char *buffer, *cp;
 392         size_t bufsz, outsz;
 393
 394         cnt = 0;
 395         rw_rlock(&tcp_function_lock);
 396         TAILQ_FOREACH(f, &t_functions, tf_next) {
 397                 cnt++;
 398         }
 399         rw_runlock(&tcp_function_lock);
 400
 401         bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
 402         buffer = malloc(bufsz, M_TEMP, M_WAITOK);
 403
 404         error = 0;
 405         cp = buffer;
 406
 407         linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
 408         cp += linesz;
 409         bufsz -= linesz;
 410         outsz = linesz;
 411
 412         rw_rlock(&tcp_function_lock);
 413         TAILQ_FOREACH(f, &t_functions, tf_next) {
 414                 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
 415                     f->tf_fb->tfb_tcp_block_name,
 416                     (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
 417                     f->tf_fb->tfb_refcnt);
 418                 if (linesz >= bufsz) {
 419                         error = EOVERFLOW;
 420                         break;
 421                 }
 422                 cp += linesz;
 423                 bufsz -= linesz;
 424                 outsz += linesz;
 425         }
 426         rw_runlock(&tcp_function_lock);
 427         if (error == 0)
 428                 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
 429         free(buffer, M_TEMP);
 430         return (error);
 431 }
 432
 433 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
 434             CTLTYPE_STRING|CTLFLAG_RD,
 435             NULL, 0, sysctl_net_inet_list_available, "A",
 436             "list available TCP Function sets");
 437
 438 /*
 439  * Target size of TCP PCB hash tables. Must be a power of two.
 440  *
 441  * Note that this can be overridden by the kernel environment
 442  * variable net.inet.tcp.tcbhashsize
 443  */
 444 #ifndef TCBHASHSIZE
 445 #define TCBHASHSIZE     0
 446 #endif
 447
 448 /*
 449  * XXX
 450  * Callouts should be moved into struct tcp directly.  They are currently
 451  * separate because the tcpcb structure is exported to userland for sysctl
 452  * parsing purposes, which do not know about callouts.
 453  */
 454 struct tcpcb_mem {
 455         struct  tcpcb           tcb;
 456         struct  tcp_timer       tt;
 457         struct  cc_var          ccv;
 458 #ifdef TCP_HHOOK
 459         struct  osd             osd;
 460 #endif
 461 };
 462
 463 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
 464 #define V_tcpcb_zone                    VNET(tcpcb_zone)
 465
 466 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
 467 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
 468
 469 static struct mtx isn_mtx;
 470
 471 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
 472 #define ISN_LOCK()      mtx_lock(&isn_mtx)
 473 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
 474
 475 /*
 476  * TCP initialization.
 477  */
 478 static void
 479 tcp_zone_change(void *tag)
 480 {
 481
 482         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
 483         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 484         tcp_tw_zone_change();
 485 }
 486
 487 static int
 488 tcp_inpcb_init(void *mem, int size, int flags)
 489 {
 490         struct inpcb *inp = mem;
 491
 492         INP_LOCK_INIT(inp, "inp", "tcpinp");
 493         return (0);
 494 }
 495
 496 /*
 497  * Take a value and get the next power of 2 that doesn't overflow.
 498  * Used to size the tcp_inpcb hash buckets.
 499  */
 500 static int
 501 maketcp_hashsize(int size)
 502 {
 503         int hashsize;
 504
 505         /*
 506          * auto tune.
 507          * get the next power of 2 higher than maxsockets.
 508          */
 509         hashsize = 1 << fls(size);
 510         /* catch overflow, and just go one power of 2 smaller */
 511         if (hashsize < size) {
 512                 hashsize = 1 << (fls(size) - 1);
 513         }
 514         return (hashsize);
 515 }
 516
 517 int
 518 register_tcp_functions(struct tcp_function_block *blk, int wait)
 519 {
 520         struct tcp_function_block *lblk;
 521         struct tcp_function *n;
 522         struct tcp_function_set fs;
 523
 524         if (t_functions_inited == 0) {
 525                 init_tcp_functions();
 526         }
 527         if ((blk->tfb_tcp_output == NULL) ||
 528             (blk->tfb_tcp_do_segment == NULL) ||
 529             (blk->tfb_tcp_ctloutput == NULL) ||
 530             (strlen(blk->tfb_tcp_block_name) == 0)) {
 531                 /*
 532                  * These functions are required and you
 533                  * need a name.
 534                  */
 535                 return (EINVAL);
 536         }
 537         if (blk->tfb_tcp_timer_stop_all ||
 538             blk->tfb_tcp_timer_activate ||
 539             blk->tfb_tcp_timer_active ||
 540             blk->tfb_tcp_timer_stop) {
 541                 /*
 542                  * If you define one timer function you
 543                  * must have them all.
 544                  */
 545                 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
 546                     (blk->tfb_tcp_timer_activate == NULL) ||
 547                     (blk->tfb_tcp_timer_active == NULL) ||
 548                     (blk->tfb_tcp_timer_stop == NULL)) {
 549                         return (EINVAL);
 550                 }
 551         }
 552         n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
 553         if (n == NULL) {
 554                 return (ENOMEM);
 555         }
 556         n->tf_fb = blk;
 557         strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 558         rw_wlock(&tcp_function_lock);
 559         lblk = find_tcp_functions_locked(&fs);
 560         if (lblk) {
 561                 /* Duplicate name space not allowed */
 562                 rw_wunlock(&tcp_function_lock);
 563                 free(n, M_TCPFUNCTIONS);
 564                 return (EALREADY);
 565         }
 566         refcount_init(&blk->tfb_refcnt, 0);
 567         blk->tfb_flags = 0;
 568         TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
 569         rw_wunlock(&tcp_function_lock);
 570         return(0);
 571 }
 572
 573 int
 574 deregister_tcp_functions(struct tcp_function_block *blk)
 575 {
 576         struct tcp_function_block *lblk;
 577         struct tcp_function *f;
 578         int error=ENOENT;
 579
 580         if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
 581                 /* You can't un-register the default */
 582                 return (EPERM);
 583         }
 584         rw_wlock(&tcp_function_lock);
 585         if (blk == tcp_func_set_ptr) {
 586                 /* You can't free the current default */
 587                 rw_wunlock(&tcp_function_lock);
 588                 return (EBUSY);
 589         }
 590         if (blk->tfb_refcnt) {
 591                 /* Still tcb attached, mark it. */
 592                 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
 593                 rw_wunlock(&tcp_function_lock);
 594                 return (EBUSY);
 595         }
 596         lblk = find_tcp_fb_locked(blk, &f);
 597         if (lblk) {
 598                 /* Found */
 599                 TAILQ_REMOVE(&t_functions, f, tf_next);
 600                 f->tf_fb = NULL;
 601                 free(f, M_TCPFUNCTIONS);
 602                 error = 0;
 603         }
 604         rw_wunlock(&tcp_function_lock);
 605         return (error);
 606 }
 607
 608 void
 609 tcp_init(void)
 610 {
 611         const char *tcbhash_tuneable;
 612         int hashsize;
 613
 614         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
 615
 616 #ifdef TCP_HHOOK
 617         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
 618             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 619                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 620         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
 621             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 622                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 623 #endif
 624         hashsize = TCBHASHSIZE;
 625         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
 626         if (hashsize == 0) {
 627                 /*
 628                  * Auto tune the hash size based on maxsockets.
 629                  * A perfect hash would have a 1:1 mapping
 630                  * (hashsize = maxsockets) however it's been
 631                  * suggested that O(2) average is better.
 632                  */
 633                 hashsize = maketcp_hashsize(maxsockets / 4);
 634                 /*
 635                  * Our historical default is 512,
 636                  * do not autotune lower than this.
 637                  */
 638                 if (hashsize < 512)
 639                         hashsize = 512;
 640                 if (bootverbose && IS_DEFAULT_VNET(curvnet))
 641                         printf("%s: %s auto tuned to %d\n", __func__,
 642                             tcbhash_tuneable, hashsize);
 643         }
 644         /*
 645          * We require a hashsize to be a power of two.
 646          * Previously if it was not a power of two we would just reset it
 647          * back to 512, which could be a nasty surprise if you did not notice
 648          * the error message.
 649          * Instead what we do is clip it to the closest power of two lower
 650          * than the specified hash value.
 651          */
 652         if (!powerof2(hashsize)) {
 653                 int oldhashsize = hashsize;
 654
 655                 hashsize = maketcp_hashsize(hashsize);
 656                 /* prevent absurdly low value */
 657                 if (hashsize < 16)
 658                         hashsize = 16;
 659                 printf("%s: WARNING: TCB hash size not a power of 2, "
 660                     "clipped from %d to %d.\n", __func__, oldhashsize,
 661                     hashsize);
 662         }
 663         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
 664             "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
 665
 666         /*
 667          * These have to be type stable for the benefit of the timers.
 668          */
 669         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
 670             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 671         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 672         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
 673
 674         tcp_tw_init();
 675         syncache_init();
 676         tcp_hc_init();
 677
 678         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
 679         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
 680             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 681
 682 #ifdef TCP_RFC7413
 683         tcp_fastopen_init();
 684 #endif
 685
 686         /* Skip initialization of globals for non-default instances. */
 687         if (!IS_DEFAULT_VNET(curvnet))
 688                 return;
 689
 690         tcp_reass_global_init();
 691
 692         /* XXX virtualize those bellow? */
 693         tcp_delacktime = TCPTV_DELACK;
 694         tcp_keepinit = TCPTV_KEEP_INIT;
 695         tcp_keepidle = TCPTV_KEEP_IDLE;
 696         tcp_keepintvl = TCPTV_KEEPINTVL;
 697         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
 698         tcp_msl = TCPTV_MSL;
 699         tcp_rexmit_min = TCPTV_MIN;
 700         if (tcp_rexmit_min < 1)
 701                 tcp_rexmit_min = 1;
 702         tcp_persmin = TCPTV_PERSMIN;
 703         tcp_persmax = TCPTV_PERSMAX;
 704         tcp_rexmit_slop = TCPTV_CPU_VAR;
 705         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
 706         tcp_tcbhashsize = hashsize;
 707         /* Setup the tcp function block list */
 708         init_tcp_functions();
 709         register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
 710
 711         if (tcp_soreceive_stream) {
 712 #ifdef INET
 713                 tcp_usrreqs.pru_soreceive = soreceive_stream;
 714 #endif
 715 #ifdef INET6
 716                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
 717 #endif /* INET6 */
 718         }
 719
 720 #ifdef INET6
 721 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
 722 #else /* INET6 */
 723 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
 724 #endif /* INET6 */
 725         if (max_protohdr < TCP_MINPROTOHDR)
 726                 max_protohdr = TCP_MINPROTOHDR;
 727         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
 728                 panic("tcp_init");
 729 #undef TCP_MINPROTOHDR
 730
 731         ISN_LOCK_INIT();
 732         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
 733                 SHUTDOWN_PRI_DEFAULT);
 734         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
 735                 EVENTHANDLER_PRI_ANY);
 736 #ifdef TCPPCAP
 737         tcp_pcap_init();
 738 #endif
 739 }
 740
 741 #ifdef VIMAGE
 742 static void
 743 tcp_destroy(void *unused __unused)
 744 {
 745         int n;
 746 #ifdef TCP_HHOOK
 747         int error;
 748 #endif
 749
 750         /*
 751          * All our processes are gone, all our sockets should be cleaned
 752          * up, which means, we should be past the tcp_discardcb() calls.
 753          * Sleep to let all tcpcb timers really disappear and cleanup.
 754          */
 755         for (;;) {
 756                 INP_LIST_RLOCK(&V_tcbinfo);
 757                 n = V_tcbinfo.ipi_count;
 758                 INP_LIST_RUNLOCK(&V_tcbinfo);
 759                 if (n == 0)
 760                         break;
 761                 pause("tcpdes", hz / 10);
 762         }
 763         tcp_hc_destroy();
 764         syncache_destroy();
 765         tcp_tw_destroy();
 766         in_pcbinfo_destroy(&V_tcbinfo);
 767         /* tcp_discardcb() clears the sack_holes up. */
 768         uma_zdestroy(V_sack_hole_zone);
 769         uma_zdestroy(V_tcpcb_zone);
 770
 771 #ifdef TCP_RFC7413
 772         /*
 773          * Cannot free the zone until all tcpcbs are released as we attach
 774          * the allocations to them.
 775          */
 776         tcp_fastopen_destroy();
 777 #endif
 778
 779 #ifdef TCP_HHOOK
 780         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
 781         if (error != 0) {
 782                 printf("%s: WARNING: unable to deregister helper hook "
 783                     "type=%d, id=%d: error %d returned\n", __func__,
 784                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
 785         }
 786         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
 787         if (error != 0) {
 788                 printf("%s: WARNING: unable to deregister helper hook "
 789                     "type=%d, id=%d: error %d returned\n", __func__,
 790                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
 791         }
 792 #endif
 793 }
 794 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
 795 #endif
 796
 797 void
 798 tcp_fini(void *xtp)
 799 {
 800
 801 }
 802
 803 /*
 804  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
 805  * tcp_template used to store this data in mbufs, but we now recopy it out
 806  * of the tcpcb each time to conserve mbufs.
 807  */
 808 void
 809 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
 810 {
 811         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
 812
 813         INP_WLOCK_ASSERT(inp);
 814
 815 #ifdef INET6
 816         if ((inp->inp_vflag & INP_IPV6) != 0) {
 817                 struct ip6_hdr *ip6;
 818
 819                 ip6 = (struct ip6_hdr *)ip_ptr;
 820                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
 821                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
 822                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
 823                         (IPV6_VERSION & IPV6_VERSION_MASK);
 824                 ip6->ip6_nxt = IPPROTO_TCP;
 825                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
 826                 ip6->ip6_src = inp->in6p_laddr;
 827                 ip6->ip6_dst = inp->in6p_faddr;
 828         }
 829 #endif /* INET6 */
 830 #if defined(INET6) && defined(INET)
 831         else
 832 #endif
 833 #ifdef INET
 834         {
 835                 struct ip *ip;
 836
 837                 ip = (struct ip *)ip_ptr;
 838                 ip->ip_v = IPVERSION;
 839                 ip->ip_hl = 5;
 840                 ip->ip_tos = inp->inp_ip_tos;
 841                 ip->ip_len = 0;
 842                 ip->ip_id = 0;
 843                 ip->ip_off = 0;
 844                 ip->ip_ttl = inp->inp_ip_ttl;
 845                 ip->ip_sum = 0;
 846                 ip->ip_p = IPPROTO_TCP;
 847                 ip->ip_src = inp->inp_laddr;
 848                 ip->ip_dst = inp->inp_faddr;
 849         }
 850 #endif /* INET */
 851         th->th_sport = inp->inp_lport;
 852         th->th_dport = inp->inp_fport;
 853         th->th_seq = 0;
 854         th->th_ack = 0;
 855         th->th_x2 = 0;
 856         th->th_off = 5;
 857         th->th_flags = 0;
 858         th->th_win = 0;
 859         th->th_urp = 0;
 860         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
 861 }
 862
 863 /*
 864  * Create template to be used to send tcp packets on a connection.
 865  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
 866  * use for this function is in keepalives, which use tcp_respond.
 867  */
 868 struct tcptemp *
 869 tcpip_maketemplate(struct inpcb *inp)
 870 {
 871         struct tcptemp *t;
 872
 873         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
 874         if (t == NULL)
 875                 return (NULL);
 876         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
 877         return (t);
 878 }
 879
 880 /*
 881  * Send a single message to the TCP at address specified by
 882  * the given TCP/IP header.  If m == NULL, then we make a copy
 883  * of the tcpiphdr at th and send directly to the addressed host.
 884  * This is used to force keep alive messages out using the TCP
 885  * template for a connection.  If flags are given then we send
 886  * a message back to the TCP which originated the segment th,
 887  * and discard the mbuf containing it and any other attached mbufs.
 888  *
 889  * In any case the ack and sequence number of the transmitted
 890  * segment are as specified by the parameters.
 891  *
 892  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
 893  */
 894 void
 895 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
 896     tcp_seq ack, tcp_seq seq, int flags)
 897 {
 898         struct tcpopt to;
 899         struct inpcb *inp;
 900         struct ip *ip;
 901         struct mbuf *optm;
 902         struct tcphdr *nth;
 903         u_char *optp;
 904 #ifdef INET6
 905         struct ip6_hdr *ip6;
 906         int isipv6;
 907 #endif /* INET6 */
 908         int optlen, tlen, win;
 909         bool incl_opts;
 910
 911         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
 912
 913 #ifdef INET6
 914         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
 915         ip6 = ipgen;
 916 #endif /* INET6 */
 917         ip = ipgen;
 918
 919         if (tp != NULL) {
 920                 inp = tp->t_inpcb;
 921                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
 922                 INP_WLOCK_ASSERT(inp);
 923         } else
 924                 inp = NULL;
 925
 926         incl_opts = false;
 927         win = 0;
 928         if (tp != NULL) {
 929                 if (!(flags & TH_RST)) {
 930                         win = sbspace(&inp->inp_socket->so_rcv);
 931                         if (win > TCP_MAXWIN << tp->rcv_scale)
 932                                 win = TCP_MAXWIN << tp->rcv_scale;
 933                 }
 934                 if ((tp->t_flags & TF_NOOPT) == 0)
 935                         incl_opts = true;
 936         }
 937         if (m == NULL) {
 938                 m = m_gethdr(M_NOWAIT, MT_DATA);
 939                 if (m == NULL)
 940                         return;
 941                 m->m_data += max_linkhdr;
 942 #ifdef INET6
 943                 if (isipv6) {
 944                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
 945                               sizeof(struct ip6_hdr));
 946                         ip6 = mtod(m, struct ip6_hdr *);
 947                         nth = (struct tcphdr *)(ip6 + 1);
 948                 } else
 949 #endif /* INET6 */
 950                 {
 951                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
 952                         ip = mtod(m, struct ip *);
 953                         nth = (struct tcphdr *)(ip + 1);
 954                 }
 955                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 956                 flags = TH_ACK;
 957         } else if (!M_WRITABLE(m)) {
 958                 struct mbuf *n;
 959
 960                 /* Can't reuse 'm', allocate a new mbuf. */
 961                 n = m_gethdr(M_NOWAIT, MT_DATA);
 962                 if (n == NULL) {
 963                         m_freem(m);
 964                         return;
 965                 }
 966
 967                 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
 968                         m_freem(m);
 969                         m_freem(n);
 970                         return;
 971                 }
 972
 973                 n->m_data += max_linkhdr;
 974                 /* m_len is set later */
 975 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
 976 #ifdef INET6
 977                 if (isipv6) {
 978                         bcopy((caddr_t)ip6, mtod(n, caddr_t),
 979                               sizeof(struct ip6_hdr));
 980                         ip6 = mtod(n, struct ip6_hdr *);
 981                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
 982                         nth = (struct tcphdr *)(ip6 + 1);
 983                 } else
 984 #endif /* INET6 */
 985                 {
 986                         bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
 987                         ip = mtod(n, struct ip *);
 988                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
 989                         nth = (struct tcphdr *)(ip + 1);
 990                 }
 991                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 992                 xchg(nth->th_dport, nth->th_sport, uint16_t);
 993                 th = nth;
 994                 m_freem(m);
 995                 m = n;
 996         } else {
 997                 /*
 998                  *  reuse the mbuf.
 999                  * XXX MRT We inherit the FIB, which is lucky.
1000                  */
1001                 m_freem(m->m_next);
1002                 m->m_next = NULL;
1003                 m->m_data = (caddr_t)ipgen;
1004                 /* m_len is set later */
1005 #ifdef INET6
1006                 if (isipv6) {
1007                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1008                         nth = (struct tcphdr *)(ip6 + 1);
1009                 } else
1010 #endif /* INET6 */
1011                 {
1012                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1013                         nth = (struct tcphdr *)(ip + 1);
1014                 }
1015                 if (th != nth) {
1016                         /*
1017                          * this is usually a case when an extension header
1018                          * exists between the IPv6 header and the
1019                          * TCP header.
1020                          */
1021                         nth->th_sport = th->th_sport;
1022                         nth->th_dport = th->th_dport;
1023                 }
1024                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1025 #undef xchg
1026         }
1027         tlen = 0;
1028 #ifdef INET6
1029         if (isipv6)
1030                 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1031 #endif
1032 #if defined(INET) && defined(INET6)
1033         else
1034 #endif
1035 #ifdef INET
1036                 tlen = sizeof (struct tcpiphdr);
1037 #endif
1038 #ifdef INVARIANTS
1039         m->m_len = 0;
1040         KASSERT(M_TRAILINGSPACE(m) >= tlen,
1041             ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1042             m, tlen, (long)M_TRAILINGSPACE(m)));
1043 #endif
1044         m->m_len = tlen;
1045         to.to_flags = 0;
1046         if (incl_opts) {
1047                 /* Make sure we have room. */
1048                 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1049                         m->m_next = m_get(M_NOWAIT, MT_DATA);
1050                         if (m->m_next) {
1051                                 optp = mtod(m->m_next, u_char *);
1052                                 optm = m->m_next;
1053                         } else
1054                                 incl_opts = false;
1055                 } else {
1056                         optp = (u_char *) (nth + 1);
1057                         optm = m;
1058                 }
1059         }
1060         if (incl_opts) {
1061                 /* Timestamps. */
1062                 if (tp->t_flags & TF_RCVD_TSTMP) {
1063                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1064                         to.to_tsecr = tp->ts_recent;
1065                         to.to_flags |= TOF_TS;
1066                 }
1067 #ifdef TCP_SIGNATURE
1068                 /* TCP-MD5 (RFC2385). */
1069                 if (tp->t_flags & TF_SIGNATURE)
1070                         to.to_flags |= TOF_SIGNATURE;
1071 #endif
1072
1073                 /* Add the options. */
1074                 tlen += optlen = tcp_addoptions(&to, optp);
1075
1076                 /* Update m_len in the correct mbuf. */
1077                 optm->m_len += optlen;
1078         } else
1079                 optlen = 0;
1080 #ifdef INET6
1081         if (isipv6) {
1082                 ip6->ip6_flow = 0;
1083                 ip6->ip6_vfc = IPV6_VERSION;
1084                 ip6->ip6_nxt = IPPROTO_TCP;
1085                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1086         }
1087 #endif
1088 #if defined(INET) && defined(INET6)
1089         else
1090 #endif
1091 #ifdef INET
1092         {
1093                 ip->ip_len = htons(tlen);
1094                 ip->ip_ttl = V_ip_defttl;
1095                 if (V_path_mtu_discovery)
1096                         ip->ip_off |= htons(IP_DF);
1097         }
1098 #endif
1099         m->m_pkthdr.len = tlen;
1100         m->m_pkthdr.rcvif = NULL;
1101 #ifdef MAC
1102         if (inp != NULL) {
1103                 /*
1104                  * Packet is associated with a socket, so allow the
1105                  * label of the response to reflect the socket label.
1106                  */
1107                 INP_WLOCK_ASSERT(inp);
1108                 mac_inpcb_create_mbuf(inp, m);
1109         } else {
1110                 /*
1111                  * Packet is not associated with a socket, so possibly
1112                  * update the label in place.
1113                  */
1114                 mac_netinet_tcp_reply(m);
1115         }
1116 #endif
1117         nth->th_seq = htonl(seq);
1118         nth->th_ack = htonl(ack);
1119         nth->th_x2 = 0;
1120         nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1121         nth->th_flags = flags;
1122         if (tp != NULL)
1123                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1124         else
1125                 nth->th_win = htons((u_short)win);
1126         nth->th_urp = 0;
1127
1128 #ifdef TCP_SIGNATURE
1129         if (to.to_flags & TOF_SIGNATURE) {
1130                 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
1131                     IPSEC_DIR_OUTBOUND);
1132         }
1133 #endif
1134
1135         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1136 #ifdef INET6
1137         if (isipv6) {
1138                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1139                 nth->th_sum = in6_cksum_pseudo(ip6,
1140                     tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1141                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1142                     NULL, NULL);
1143         }
1144 #endif /* INET6 */
1145 #if defined(INET6) && defined(INET)
1146         else
1147 #endif
1148 #ifdef INET
1149         {
1150                 m->m_pkthdr.csum_flags = CSUM_TCP;
1151                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1152                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1153         }
1154 #endif /* INET */
1155 #ifdef TCPDEBUG
1156         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1157                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1158 #endif
1159         TCP_PROBE3(debug__output, tp, th, m);
1160         if (flags & TH_RST)
1161                 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1162
1163         TCP_PROBE5(send, NULL, tp, m, tp, nth);
1164 #ifdef INET6
1165         if (isipv6)
1166                 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1167 #endif /* INET6 */
1168 #if defined(INET) && defined(INET6)
1169         else
1170 #endif
1171 #ifdef INET
1172                 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
1173 #endif
1174 }
1175
1176 /*
1177  * Create a new TCP control block, making an
1178  * empty reassembly queue and hooking it to the argument
1179  * protocol control block.  The `inp' parameter must have
1180  * come from the zone allocator set up in tcp_init().
1181  */
1182 struct tcpcb *
1183 tcp_newtcpcb(struct inpcb *inp)
1184 {
1185         struct tcpcb_mem *tm;
1186         struct tcpcb *tp;
1187 #ifdef INET6
1188         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1189 #endif /* INET6 */
1190
1191         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1192         if (tm == NULL)
1193                 return (NULL);
1194         tp = &tm->tcb;
1195
1196         /* Initialise cc_var struct for this tcpcb. */
1197         tp->ccv = &tm->ccv;
1198         tp->ccv->type = IPPROTO_TCP;
1199         tp->ccv->ccvc.tcp = tp;
1200         rw_rlock(&tcp_function_lock);
1201         tp->t_fb = tcp_func_set_ptr;
1202         refcount_acquire(&tp->t_fb->tfb_refcnt);
1203         rw_runlock(&tcp_function_lock);
1204         /*
1205          * Use the current system default CC algorithm.
1206          */
1207         CC_LIST_RLOCK();
1208         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1209         CC_ALGO(tp) = CC_DEFAULT();
1210         CC_LIST_RUNLOCK();
1211
1212         if (CC_ALGO(tp)->cb_init != NULL)
1213                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1214                         if (tp->t_fb->tfb_tcp_fb_fini)
1215                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1216                         refcount_release(&tp->t_fb->tfb_refcnt);
1217                         uma_zfree(V_tcpcb_zone, tm);
1218                         return (NULL);
1219                 }
1220
1221 #ifdef TCP_HHOOK
1222         tp->osd = &tm->osd;
1223         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1224                 if (tp->t_fb->tfb_tcp_fb_fini)
1225                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1226                 refcount_release(&tp->t_fb->tfb_refcnt);
1227                 uma_zfree(V_tcpcb_zone, tm);
1228                 return (NULL);
1229         }
1230 #endif
1231
1232 #ifdef VIMAGE
1233         tp->t_vnet = inp->inp_vnet;
1234 #endif
1235         tp->t_timers = &tm->tt;
1236         /*      LIST_INIT(&tp->t_segq); */      /* XXX covered by M_ZERO */
1237         tp->t_maxseg =
1238 #ifdef INET6
1239                 isipv6 ? V_tcp_v6mssdflt :
1240 #endif /* INET6 */
1241                 V_tcp_mssdflt;
1242
1243         /* Set up our timeouts. */
1244         callout_init(&tp->t_timers->tt_rexmt, 1);
1245         callout_init(&tp->t_timers->tt_persist, 1);
1246         callout_init(&tp->t_timers->tt_keep, 1);
1247         callout_init(&tp->t_timers->tt_2msl, 1);
1248         callout_init(&tp->t_timers->tt_delack, 1);
1249
1250         if (V_tcp_do_rfc1323)
1251                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1252         if (V_tcp_do_sack)
1253                 tp->t_flags |= TF_SACK_PERMIT;
1254         TAILQ_INIT(&tp->snd_holes);
1255         /*
1256          * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1257          * is called.
1258          */
1259         in_pcbref(inp); /* Reference for tcpcb */
1260         tp->t_inpcb = inp;
1261
1262         /*
1263          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1264          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1265          * reasonable initial retransmit time.
1266          */
1267         tp->t_srtt = TCPTV_SRTTBASE;
1268         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1269         tp->t_rttmin = tcp_rexmit_min;
1270         tp->t_rxtcur = TCPTV_RTOBASE;
1271         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1272         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1273         tp->t_rcvtime = ticks;
1274         /*
1275          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1276          * because the socket may be bound to an IPv6 wildcard address,
1277          * which may match an IPv4-mapped IPv6 address.
1278          */
1279         inp->inp_ip_ttl = V_ip_defttl;
1280         inp->inp_ppcb = tp;
1281 #ifdef TCPPCAP
1282         /*
1283          * Init the TCP PCAP queues.
1284          */
1285         tcp_pcap_tcpcb_init(tp);
1286 #endif
1287         if (tp->t_fb->tfb_tcp_fb_init) {
1288                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1289         }
1290         return (tp);            /* XXX */
1291 }
1292
1293 /*
1294  * Switch the congestion control algorithm back to NewReno for any active
1295  * control blocks using an algorithm which is about to go away.
1296  * This ensures the CC framework can allow the unload to proceed without leaving
1297  * any dangling pointers which would trigger a panic.
1298  * Returning non-zero would inform the CC framework that something went wrong
1299  * and it would be unsafe to allow the unload to proceed. However, there is no
1300  * way for this to occur with this implementation so we always return zero.
1301  */
1302 int
1303 tcp_ccalgounload(struct cc_algo *unload_algo)
1304 {
1305         struct cc_algo *tmpalgo;
1306         struct inpcb *inp;
1307         struct tcpcb *tp;
1308         VNET_ITERATOR_DECL(vnet_iter);
1309
1310         /*
1311          * Check all active control blocks across all network stacks and change
1312          * any that are using "unload_algo" back to NewReno. If "unload_algo"
1313          * requires cleanup code to be run, call it.
1314          */
1315         VNET_LIST_RLOCK();
1316         VNET_FOREACH(vnet_iter) {
1317                 CURVNET_SET(vnet_iter);
1318                 INP_INFO_WLOCK(&V_tcbinfo);
1319                 /*
1320                  * New connections already part way through being initialised
1321                  * with the CC algo we're removing will not race with this code
1322                  * because the INP_INFO_WLOCK is held during initialisation. We
1323                  * therefore don't enter the loop below until the connection
1324                  * list has stabilised.
1325                  */
1326                 LIST_FOREACH(inp, &V_tcb, inp_list) {
1327                         INP_WLOCK(inp);
1328                         /* Important to skip tcptw structs. */
1329                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1330                             (tp = intotcpcb(inp)) != NULL) {
1331                                 /*
1332                                  * By holding INP_WLOCK here, we are assured
1333                                  * that the connection is not currently
1334                                  * executing inside the CC module's functions
1335                                  * i.e. it is safe to make the switch back to
1336                                  * NewReno.
1337                                  */
1338                                 if (CC_ALGO(tp) == unload_algo) {
1339                                         tmpalgo = CC_ALGO(tp);
1340                                         /* NewReno does not require any init. */
1341                                         CC_ALGO(tp) = &newreno_cc_algo;
1342                                         if (tmpalgo->cb_destroy != NULL)
1343                                                 tmpalgo->cb_destroy(tp->ccv);
1344                                 }
1345                         }
1346                         INP_WUNLOCK(inp);
1347                 }
1348                 INP_INFO_WUNLOCK(&V_tcbinfo);
1349                 CURVNET_RESTORE();
1350         }
1351         VNET_LIST_RUNLOCK();
1352
1353         return (0);
1354 }
1355
1356 /*
1357  * Drop a TCP connection, reporting
1358  * the specified error.  If connection is synchronized,
1359  * then send a RST to peer.
1360  */
1361 struct tcpcb *
1362 tcp_drop(struct tcpcb *tp, int errno)
1363 {
1364         struct socket *so = tp->t_inpcb->inp_socket;
1365
1366         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1367         INP_WLOCK_ASSERT(tp->t_inpcb);
1368
1369         if (TCPS_HAVERCVDSYN(tp->t_state)) {
1370                 tcp_state_change(tp, TCPS_CLOSED);
1371                 (void) tp->t_fb->tfb_tcp_output(tp);
1372                 TCPSTAT_INC(tcps_drops);
1373         } else
1374                 TCPSTAT_INC(tcps_conndrops);
1375         if (errno == ETIMEDOUT && tp->t_softerror)
1376                 errno = tp->t_softerror;
1377         so->so_error = errno;
1378         return (tcp_close(tp));
1379 }
1380
1381 void
1382 tcp_discardcb(struct tcpcb *tp)
1383 {
1384         struct inpcb *inp = tp->t_inpcb;
1385         struct socket *so = inp->inp_socket;
1386 #ifdef INET6
1387         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1388 #endif /* INET6 */
1389         int released;
1390
1391         INP_WLOCK_ASSERT(inp);
1392
1393         /*
1394          * Make sure that all of our timers are stopped before we delete the
1395          * PCB.
1396          *
1397          * If stopping a timer fails, we schedule a discard function in same
1398          * callout, and the last discard function called will take care of
1399          * deleting the tcpcb.
1400          */
1401         tp->t_timers->tt_draincnt = 0;
1402         tcp_timer_stop(tp, TT_REXMT);
1403         tcp_timer_stop(tp, TT_PERSIST);
1404         tcp_timer_stop(tp, TT_KEEP);
1405         tcp_timer_stop(tp, TT_2MSL);
1406         tcp_timer_stop(tp, TT_DELACK);
1407         if (tp->t_fb->tfb_tcp_timer_stop_all) {
1408                 /*
1409                  * Call the stop-all function of the methods,
1410                  * this function should call the tcp_timer_stop()
1411                  * method with each of the function specific timeouts.
1412                  * That stop will be called via the tfb_tcp_timer_stop()
1413                  * which should use the async drain function of the
1414                  * callout system (see tcp_var.h).
1415                  */
1416                 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1417         }
1418
1419         /*
1420          * If we got enough samples through the srtt filter,
1421          * save the rtt and rttvar in the routing entry.
1422          * 'Enough' is arbitrarily defined as 4 rtt samples.
1423          * 4 samples is enough for the srtt filter to converge
1424          * to within enough % of the correct value; fewer samples
1425          * and we could save a bogus rtt. The danger is not high
1426          * as tcp quickly recovers from everything.
1427          * XXX: Works very well but needs some more statistics!
1428          */
1429         if (tp->t_rttupdated >= 4) {
1430                 struct hc_metrics_lite metrics;
1431                 uint32_t ssthresh;
1432
1433                 bzero(&metrics, sizeof(metrics));
1434                 /*
1435                  * Update the ssthresh always when the conditions below
1436                  * are satisfied. This gives us better new start value
1437                  * for the congestion avoidance for new connections.
1438                  * ssthresh is only set if packet loss occurred on a session.
1439                  *
1440                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1441                  * being torn down.  Ideally this code would not use 'so'.
1442                  */
1443                 ssthresh = tp->snd_ssthresh;
1444                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1445                         /*
1446                          * convert the limit from user data bytes to
1447                          * packets then to packet data bytes.
1448                          */
1449                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1450                         if (ssthresh < 2)
1451                                 ssthresh = 2;
1452                         ssthresh *= (tp->t_maxseg +
1453 #ifdef INET6
1454                             (isipv6 ? sizeof (struct ip6_hdr) +
1455                                 sizeof (struct tcphdr) :
1456 #endif
1457                                 sizeof (struct tcpiphdr)
1458 #ifdef INET6
1459                             )
1460 #endif
1461                             );
1462                 } else
1463                         ssthresh = 0;
1464                 metrics.rmx_ssthresh = ssthresh;
1465
1466                 metrics.rmx_rtt = tp->t_srtt;
1467                 metrics.rmx_rttvar = tp->t_rttvar;
1468                 metrics.rmx_cwnd = tp->snd_cwnd;
1469                 metrics.rmx_sendpipe = 0;
1470                 metrics.rmx_recvpipe = 0;
1471
1472                 tcp_hc_update(&inp->inp_inc, &metrics);
1473         }
1474
1475         /* free the reassembly queue, if any */
1476         tcp_reass_flush(tp);
1477
1478 #ifdef TCP_OFFLOAD
1479         /* Disconnect offload device, if any. */
1480         if (tp->t_flags & TF_TOE)
1481                 tcp_offload_detach(tp);
1482 #endif
1483
1484         tcp_free_sackholes(tp);
1485
1486 #ifdef TCPPCAP
1487         /* Free the TCP PCAP queues. */
1488         tcp_pcap_drain(&(tp->t_inpkts));
1489         tcp_pcap_drain(&(tp->t_outpkts));
1490 #endif
1491
1492         /* Allow the CC algorithm to clean up after itself. */
1493         if (CC_ALGO(tp)->cb_destroy != NULL)
1494                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1495
1496 #ifdef TCP_HHOOK
1497         khelp_destroy_osd(tp->osd);
1498 #endif
1499
1500         CC_ALGO(tp) = NULL;
1501         inp->inp_ppcb = NULL;
1502         if (tp->t_timers->tt_draincnt == 0) {
1503                 /* We own the last reference on tcpcb, let's free it. */
1504                 TCPSTATES_DEC(tp->t_state);
1505                 if (tp->t_fb->tfb_tcp_fb_fini)
1506                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1507                 refcount_release(&tp->t_fb->tfb_refcnt);
1508                 tp->t_inpcb = NULL;
1509                 uma_zfree(V_tcpcb_zone, tp);
1510                 released = in_pcbrele_wlocked(inp);
1511                 KASSERT(!released, ("%s: inp %p should not have been released "
1512                         "here", __func__, inp));
1513         }
1514 }
1515
1516 void
1517 tcp_timer_discard(void *ptp)
1518 {
1519         struct inpcb *inp;
1520         struct tcpcb *tp;
1521
1522         tp = (struct tcpcb *)ptp;
1523         CURVNET_SET(tp->t_vnet);
1524         INP_INFO_RLOCK(&V_tcbinfo);
1525         inp = tp->t_inpcb;
1526         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1527                 __func__, tp));
1528         INP_WLOCK(inp);
1529         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1530                 ("%s: tcpcb has to be stopped here", __func__));
1531         tp->t_timers->tt_draincnt--;
1532         if (tp->t_timers->tt_draincnt == 0) {
1533                 /* We own the last reference on this tcpcb, let's free it. */
1534                 TCPSTATES_DEC(tp->t_state);
1535                 if (tp->t_fb->tfb_tcp_fb_fini)
1536                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
1537                 refcount_release(&tp->t_fb->tfb_refcnt);
1538                 tp->t_inpcb = NULL;
1539                 uma_zfree(V_tcpcb_zone, tp);
1540                 if (in_pcbrele_wlocked(inp)) {
1541                         INP_INFO_RUNLOCK(&V_tcbinfo);
1542                         CURVNET_RESTORE();
1543                         return;
1544                 }
1545         }
1546         INP_WUNLOCK(inp);
1547         INP_INFO_RUNLOCK(&V_tcbinfo);
1548         CURVNET_RESTORE();
1549 }
1550
1551 /*
1552  * Attempt to close a TCP control block, marking it as dropped, and freeing
1553  * the socket if we hold the only reference.
1554  */
1555 struct tcpcb *
1556 tcp_close(struct tcpcb *tp)
1557 {
1558         struct inpcb *inp = tp->t_inpcb;
1559         struct socket *so;
1560
1561         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1562         INP_WLOCK_ASSERT(inp);
1563
1564 #ifdef TCP_OFFLOAD
1565         if (tp->t_state == TCPS_LISTEN)
1566                 tcp_offload_listen_stop(tp);
1567 #endif
1568 #ifdef TCP_RFC7413
1569         /*
1570          * This releases the TFO pending counter resource for TFO listen
1571          * sockets as well as passively-created TFO sockets that transition
1572          * from SYN_RECEIVED to CLOSED.
1573          */
1574         if (tp->t_tfo_pending) {
1575                 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1576                 tp->t_tfo_pending = NULL;
1577         }
1578 #endif
1579         in_pcbdrop(inp);
1580         TCPSTAT_INC(tcps_closed);
1581         if (tp->t_state != TCPS_CLOSED)
1582                 tcp_state_change(tp, TCPS_CLOSED);
1583         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1584         so = inp->inp_socket;
1585         soisdisconnected(so);
1586         if (inp->inp_flags & INP_SOCKREF) {
1587                 KASSERT(so->so_state & SS_PROTOREF,
1588                     ("tcp_close: !SS_PROTOREF"));
1589                 inp->inp_flags &= ~INP_SOCKREF;
1590                 INP_WUNLOCK(inp);
1591                 ACCEPT_LOCK();
1592                 SOCK_LOCK(so);
1593                 so->so_state &= ~SS_PROTOREF;
1594                 sofree(so);
1595                 return (NULL);
1596         }
1597         return (tp);
1598 }
1599
1600 void
1601 tcp_drain(void)
1602 {
1603         VNET_ITERATOR_DECL(vnet_iter);
1604
1605         if (!do_tcpdrain)
1606                 return;
1607
1608         VNET_LIST_RLOCK_NOSLEEP();
1609         VNET_FOREACH(vnet_iter) {
1610                 CURVNET_SET(vnet_iter);
1611                 struct inpcb *inpb;
1612                 struct tcpcb *tcpb;
1613
1614         /*
1615          * Walk the tcpbs, if existing, and flush the reassembly queue,
1616          * if there is one...
1617          * XXX: The "Net/3" implementation doesn't imply that the TCP
1618          *      reassembly queue should be flushed, but in a situation
1619          *      where we're really low on mbufs, this is potentially
1620          *      useful.
1621          */
1622                 INP_INFO_WLOCK(&V_tcbinfo);
1623                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1624                         if (inpb->inp_flags & INP_TIMEWAIT)
1625                                 continue;
1626                         INP_WLOCK(inpb);
1627                         if ((tcpb = intotcpcb(inpb)) != NULL) {
1628                                 tcp_reass_flush(tcpb);
1629                                 tcp_clean_sackreport(tcpb);
1630 #ifdef TCPPCAP
1631                                 if (tcp_pcap_aggressive_free) {
1632                                         /* Free the TCP PCAP queues. */
1633                                         tcp_pcap_drain(&(tcpb->t_inpkts));
1634                                         tcp_pcap_drain(&(tcpb->t_outpkts));
1635                                 }
1636 #endif
1637                         }
1638                         INP_WUNLOCK(inpb);
1639                 }
1640                 INP_INFO_WUNLOCK(&V_tcbinfo);
1641                 CURVNET_RESTORE();
1642         }
1643         VNET_LIST_RUNLOCK_NOSLEEP();
1644 }
1645
1646 /*
1647  * Notify a tcp user of an asynchronous error;
1648  * store error as soft error, but wake up user
1649  * (for now, won't do anything until can select for soft error).
1650  *
1651  * Do not wake up user since there currently is no mechanism for
1652  * reporting soft errors (yet - a kqueue filter may be added).
1653  */
1654 static struct inpcb *
1655 tcp_notify(struct inpcb *inp, int error)
1656 {
1657         struct tcpcb *tp;
1658
1659         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1660         INP_WLOCK_ASSERT(inp);
1661
1662         if ((inp->inp_flags & INP_TIMEWAIT) ||
1663             (inp->inp_flags & INP_DROPPED))
1664                 return (inp);
1665
1666         tp = intotcpcb(inp);
1667         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1668
1669         /*
1670          * Ignore some errors if we are hooked up.
1671          * If connection hasn't completed, has retransmitted several times,
1672          * and receives a second error, give up now.  This is better
1673          * than waiting a long time to establish a connection that
1674          * can never complete.
1675          */
1676         if (tp->t_state == TCPS_ESTABLISHED &&
1677             (error == EHOSTUNREACH || error == ENETUNREACH ||
1678              error == EHOSTDOWN)) {
1679                 if (inp->inp_route.ro_rt) {
1680                         RTFREE(inp->inp_route.ro_rt);
1681                         inp->inp_route.ro_rt = (struct rtentry *)NULL;
1682                 }
1683                 return (inp);
1684         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1685             tp->t_softerror) {
1686                 tp = tcp_drop(tp, error);
1687                 if (tp != NULL)
1688                         return (inp);
1689                 else
1690                         return (NULL);
1691         } else {
1692                 tp->t_softerror = error;
1693                 return (inp);
1694         }
1695 #if 0
1696         wakeup( &so->so_timeo);
1697         sorwakeup(so);
1698         sowwakeup(so);
1699 #endif
1700 }
1701
1702 static int
1703 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1704 {
1705         int error, i, m, n, pcb_count;
1706         struct inpcb *inp, **inp_list;
1707         inp_gen_t gencnt;
1708         struct xinpgen xig;
1709
1710         /*
1711          * The process of preparing the TCB list is too time-consuming and
1712          * resource-intensive to repeat twice on every request.
1713          */
1714         if (req->oldptr == NULL) {
1715                 n = V_tcbinfo.ipi_count +
1716                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1717                 n += imax(n / 8, 10);
1718                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1719                 return (0);
1720         }
1721
1722         if (req->newptr != NULL)
1723                 return (EPERM);
1724
1725         /*
1726          * OK, now we're committed to doing something.
1727          */
1728         INP_LIST_RLOCK(&V_tcbinfo);
1729         gencnt = V_tcbinfo.ipi_gencnt;
1730         n = V_tcbinfo.ipi_count;
1731         INP_LIST_RUNLOCK(&V_tcbinfo);
1732
1733         m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1734
1735         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1736                 + (n + m) * sizeof(struct xtcpcb));
1737         if (error != 0)
1738                 return (error);
1739
1740         xig.xig_len = sizeof xig;
1741         xig.xig_count = n + m;
1742         xig.xig_gen = gencnt;
1743         xig.xig_sogen = so_gencnt;
1744         error = SYSCTL_OUT(req, &xig, sizeof xig);
1745         if (error)
1746                 return (error);
1747
1748         error = syncache_pcblist(req, m, &pcb_count);
1749         if (error)
1750                 return (error);
1751
1752         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1753
1754         INP_INFO_WLOCK(&V_tcbinfo);
1755         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1756             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1757                 INP_WLOCK(inp);
1758                 if (inp->inp_gencnt <= gencnt) {
1759                         /*
1760                          * XXX: This use of cr_cansee(), introduced with
1761                          * TCP state changes, is not quite right, but for
1762                          * now, better than nothing.
1763                          */
1764                         if (inp->inp_flags & INP_TIMEWAIT) {
1765                                 if (intotw(inp) != NULL)
1766                                         error = cr_cansee(req->td->td_ucred,
1767                                             intotw(inp)->tw_cred);
1768                                 else
1769                                         error = EINVAL; /* Skip this inp. */
1770                         } else
1771                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
1772                         if (error == 0) {
1773                                 in_pcbref(inp);
1774                                 inp_list[i++] = inp;
1775                         }
1776                 }
1777                 INP_WUNLOCK(inp);
1778         }
1779         INP_INFO_WUNLOCK(&V_tcbinfo);
1780         n = i;
1781
1782         error = 0;
1783         for (i = 0; i < n; i++) {
1784                 inp = inp_list[i];
1785                 INP_RLOCK(inp);
1786                 if (inp->inp_gencnt <= gencnt) {
1787                         struct xtcpcb xt;
1788                         void *inp_ppcb;
1789
1790                         bzero(&xt, sizeof(xt));
1791                         xt.xt_len = sizeof xt;
1792                         /* XXX should avoid extra copy */
1793                         bcopy(inp, &xt.xt_inp, sizeof *inp);
1794                         inp_ppcb = inp->inp_ppcb;
1795                         if (inp_ppcb == NULL)
1796                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1797                         else if (inp->inp_flags & INP_TIMEWAIT) {
1798                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1799                                 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1800                         } else {
1801                                 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1802                                 if (xt.xt_tp.t_timers)
1803                                         tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1804                         }
1805                         if (inp->inp_socket != NULL)
1806                                 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1807                         else {
1808                                 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1809                                 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1810                         }
1811                         xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1812                         INP_RUNLOCK(inp);
1813                         error = SYSCTL_OUT(req, &xt, sizeof xt);
1814                 } else
1815                         INP_RUNLOCK(inp);
1816         }
1817         INP_INFO_RLOCK(&V_tcbinfo);
1818         for (i = 0; i < n; i++) {
1819                 inp = inp_list[i];
1820                 INP_RLOCK(inp);
1821                 if (!in_pcbrele_rlocked(inp))
1822                         INP_RUNLOCK(inp);
1823         }
1824         INP_INFO_RUNLOCK(&V_tcbinfo);
1825
1826         if (!error) {
1827                 /*
1828                  * Give the user an updated idea of our state.
1829                  * If the generation differs from what we told
1830                  * her before, she knows that something happened
1831                  * while we were processing this request, and it
1832                  * might be necessary to retry.
1833                  */
1834                 INP_LIST_RLOCK(&V_tcbinfo);
1835                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1836                 xig.xig_sogen = so_gencnt;
1837                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1838                 INP_LIST_RUNLOCK(&V_tcbinfo);
1839                 error = SYSCTL_OUT(req, &xig, sizeof xig);
1840         }
1841         free(inp_list, M_TEMP);
1842         return (error);
1843 }
1844
1845 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1846     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1847     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1848
1849 #ifdef INET
1850 static int
1851 tcp_getcred(SYSCTL_HANDLER_ARGS)
1852 {
1853         struct xucred xuc;
1854         struct sockaddr_in addrs[2];
1855         struct inpcb *inp;
1856         int error;
1857
1858         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1859         if (error)
1860                 return (error);
1861         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1862         if (error)
1863                 return (error);
1864         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1865             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1866         if (inp != NULL) {
1867                 if (inp->inp_socket == NULL)
1868                         error = ENOENT;
1869                 if (error == 0)
1870                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1871                 if (error == 0)
1872                         cru2x(inp->inp_cred, &xuc);
1873                 INP_RUNLOCK(inp);
1874         } else
1875                 error = ENOENT;
1876         if (error == 0)
1877                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1878         return (error);
1879 }
1880
1881 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1882     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1883     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1884 #endif /* INET */
1885
1886 #ifdef INET6
1887 static int
1888 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1889 {
1890         struct xucred xuc;
1891         struct sockaddr_in6 addrs[2];
1892         struct inpcb *inp;
1893         int error;
1894 #ifdef INET
1895         int mapped = 0;
1896 #endif
1897
1898         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1899         if (error)
1900                 return (error);
1901         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1902         if (error)
1903                 return (error);
1904         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1905             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1906                 return (error);
1907         }
1908         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1909 #ifdef INET
1910                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1911                         mapped = 1;
1912                 else
1913 #endif
1914                         return (EINVAL);
1915         }
1916
1917 #ifdef INET
1918         if (mapped == 1)
1919                 inp = in_pcblookup(&V_tcbinfo,
1920                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1921                         addrs[1].sin6_port,
1922                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1923                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1924         else
1925 #endif
1926                 inp = in6_pcblookup(&V_tcbinfo,
1927                         &addrs[1].sin6_addr, addrs[1].sin6_port,
1928                         &addrs[0].sin6_addr, addrs[0].sin6_port,
1929                         INPLOOKUP_RLOCKPCB, NULL);
1930         if (inp != NULL) {
1931                 if (inp->inp_socket == NULL)
1932                         error = ENOENT;
1933                 if (error == 0)
1934                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1935                 if (error == 0)
1936                         cru2x(inp->inp_cred, &xuc);
1937                 INP_RUNLOCK(inp);
1938         } else
1939                 error = ENOENT;
1940         if (error == 0)
1941                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1942         return (error);
1943 }
1944
1945 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1946     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1947     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1948 #endif /* INET6 */
1949
1950
1951 #ifdef INET
1952 void
1953 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1954 {
1955         struct ip *ip = vip;
1956         struct tcphdr *th;
1957         struct in_addr faddr;
1958         struct inpcb *inp;
1959         struct tcpcb *tp;
1960         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1961         struct icmp *icp;
1962         struct in_conninfo inc;
1963         tcp_seq icmp_tcp_seq;
1964         int mtu;
1965
1966         faddr = ((struct sockaddr_in *)sa)->sin_addr;
1967         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1968                 return;
1969
1970         if (cmd == PRC_MSGSIZE)
1971                 notify = tcp_mtudisc_notify;
1972         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1973                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1974                 cmd == PRC_TIMXCEED_INTRANS) && ip)
1975                 notify = tcp_drop_syn_sent;
1976
1977         /*
1978          * Hostdead is ugly because it goes linearly through all PCBs.
1979          * XXX: We never get this from ICMP, otherwise it makes an
1980          * excellent DoS attack on machines with many connections.
1981          */
1982         else if (cmd == PRC_HOSTDEAD)
1983                 ip = NULL;
1984         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1985                 return;
1986
1987         if (ip == NULL) {
1988                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1989                 return;
1990         }
1991
1992         icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1993         th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1994         INP_INFO_RLOCK(&V_tcbinfo);
1995         inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1996             th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1997         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1998                 /* signal EHOSTDOWN, as it flushes the cached route */
1999                 inp = (*notify)(inp, EHOSTDOWN);
2000                 if (inp != NULL)
2001                         INP_WUNLOCK(inp);
2002         } else if (inp != NULL)  {
2003                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2004                     !(inp->inp_flags & INP_DROPPED) &&
2005                     !(inp->inp_socket == NULL)) {
2006                         icmp_tcp_seq = ntohl(th->th_seq);
2007                         tp = intotcpcb(inp);
2008                         if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2009                             SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2010                                 if (cmd == PRC_MSGSIZE) {
2011                                         /*
2012                                          * MTU discovery:
2013                                          * If we got a needfrag set the MTU
2014                                          * in the route to the suggested new
2015                                          * value (if given) and then notify.
2016                                          */
2017                                         mtu = ntohs(icp->icmp_nextmtu);
2018                                         /*
2019                                          * If no alternative MTU was
2020                                          * proposed, try the next smaller
2021                                          * one.
2022                                          */
2023                                         if (!mtu)
2024                                                 mtu = ip_next_mtu(
2025                                                     ntohs(ip->ip_len), 1);
2026                                         if (mtu < V_tcp_minmss +
2027                                             sizeof(struct tcpiphdr))
2028                                                 mtu = V_tcp_minmss +
2029                                                     sizeof(struct tcpiphdr);
2030                                         /*
2031                                          * Only process the offered MTU if it
2032                                          * is smaller than the current one.
2033                                          */
2034                                         if (mtu < tp->t_maxseg +
2035                                             sizeof(struct tcpiphdr)) {
2036                                                 bzero(&inc, sizeof(inc));
2037                                                 inc.inc_faddr = faddr;
2038                                                 inc.inc_fibnum =
2039                                                     inp->inp_inc.inc_fibnum;
2040                                                 tcp_hc_updatemtu(&inc, mtu);
2041                                                 tcp_mtudisc(inp, mtu);
2042                                         }
2043                                 } else
2044                                         inp = (*notify)(inp,
2045                                             inetctlerrmap[cmd]);
2046                         }
2047                 }
2048                 if (inp != NULL)
2049                         INP_WUNLOCK(inp);
2050         } else {
2051                 bzero(&inc, sizeof(inc));
2052                 inc.inc_fport = th->th_dport;
2053                 inc.inc_lport = th->th_sport;
2054                 inc.inc_faddr = faddr;
2055                 inc.inc_laddr = ip->ip_src;
2056                 syncache_unreach(&inc, th);
2057         }
2058         INP_INFO_RUNLOCK(&V_tcbinfo);
2059 }
2060 #endif /* INET */
2061
2062 #ifdef INET6
2063 void
2064 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2065 {
2066         struct in6_addr *dst;
2067         struct tcphdr *th;
2068         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2069         struct ip6_hdr *ip6;
2070         struct mbuf *m;
2071         struct inpcb *inp;
2072         struct tcpcb *tp;
2073         struct icmp6_hdr *icmp6;
2074         struct ip6ctlparam *ip6cp = NULL;
2075         const struct sockaddr_in6 *sa6_src = NULL;
2076         struct in_conninfo inc;
2077         tcp_seq icmp_tcp_seq;
2078         unsigned int mtu;
2079         unsigned int off;
2080
2081
2082         if (sa->sa_family != AF_INET6 ||
2083             sa->sa_len != sizeof(struct sockaddr_in6))
2084                 return;
2085
2086         /* if the parameter is from icmp6, decode it. */
2087         if (d != NULL) {
2088                 ip6cp = (struct ip6ctlparam *)d;
2089                 icmp6 = ip6cp->ip6c_icmp6;
2090                 m = ip6cp->ip6c_m;
2091                 ip6 = ip6cp->ip6c_ip6;
2092                 off = ip6cp->ip6c_off;
2093                 sa6_src = ip6cp->ip6c_src;
2094                 dst = ip6cp->ip6c_finaldst;
2095         } else {
2096                 m = NULL;
2097                 ip6 = NULL;
2098                 off = 0;        /* fool gcc */
2099                 sa6_src = &sa6_any;
2100                 dst = NULL;
2101         }
2102
2103         if (cmd == PRC_MSGSIZE)
2104                 notify = tcp_mtudisc_notify;
2105         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2106                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2107                 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2108                 notify = tcp_drop_syn_sent;
2109
2110         /*
2111          * Hostdead is ugly because it goes linearly through all PCBs.
2112          * XXX: We never get this from ICMP, otherwise it makes an
2113          * excellent DoS attack on machines with many connections.
2114          */
2115         else if (cmd == PRC_HOSTDEAD)
2116                 ip6 = NULL;
2117         else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2118                 return;
2119
2120         if (ip6 == NULL) {
2121                 in6_pcbnotify(&V_tcbinfo, sa, 0,
2122                               (const struct sockaddr *)sa6_src,
2123                               0, cmd, NULL, notify);
2124                 return;
2125         }
2126
2127         /* Check if we can safely get the ports from the tcp hdr */
2128         if (m == NULL ||
2129             (m->m_pkthdr.len <
2130                 (int32_t) (off + offsetof(struct tcphdr, th_seq)))) {
2131                 return;
2132         }
2133
2134         th = (struct tcphdr *) mtodo(ip6cp->ip6c_m, ip6cp->ip6c_off);
2135         INP_INFO_RLOCK(&V_tcbinfo);
2136         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, th->th_dport,
2137             &ip6->ip6_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2138         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2139                 /* signal EHOSTDOWN, as it flushes the cached route */
2140                 inp = (*notify)(inp, EHOSTDOWN);
2141                 if (inp != NULL)
2142                         INP_WUNLOCK(inp);
2143         } else if (inp != NULL)  {
2144                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2145                     !(inp->inp_flags & INP_DROPPED) &&
2146                     !(inp->inp_socket == NULL)) {
2147                         icmp_tcp_seq = ntohl(th->th_seq);
2148                         tp = intotcpcb(inp);
2149                         if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2150                             SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2151                                 if (cmd == PRC_MSGSIZE) {
2152                                         /*
2153                                          * MTU discovery:
2154                                          * If we got a needfrag set the MTU
2155                                          * in the route to the suggested new
2156                                          * value (if given) and then notify.
2157                                          */
2158                                         mtu = ntohl(icmp6->icmp6_mtu);
2159                                         /*
2160                                          * If no alternative MTU was
2161                                          * proposed, or the proposed
2162                                          * MTU was too small, set to
2163                                          * the min.
2164                                          */
2165                                         if (mtu < IPV6_MMTU)
2166                                                 mtu = IPV6_MMTU - 8;
2167
2168
2169                                         bzero(&inc, sizeof(inc));
2170                                         inc.inc_fibnum = M_GETFIB(m);
2171                                         inc.inc_flags |= INC_ISIPV6;
2172                                         inc.inc6_faddr = *dst;
2173                                         if (in6_setscope(&inc.inc6_faddr,
2174                                                 m->m_pkthdr.rcvif, NULL))
2175                                                 goto unlock_inp;
2176
2177                                         /*
2178                                          * Only process the offered MTU if it
2179                                          * is smaller than the current one.
2180                                          */
2181                                         if (mtu < tp->t_maxseg +
2182                                             (sizeof (*th) + sizeof (*ip6))) {
2183                                                 tcp_hc_updatemtu(&inc, mtu);
2184                                                 tcp_mtudisc(inp, mtu);
2185                                                 ICMP6STAT_INC(icp6s_pmtuchg);
2186                                         }
2187                                 } else
2188                                         inp = (*notify)(inp,
2189                                             inet6ctlerrmap[cmd]);
2190                         }
2191                 }
2192 unlock_inp:
2193                 if (inp != NULL)
2194                         INP_WUNLOCK(inp);
2195         } else {
2196                 bzero(&inc, sizeof(inc));
2197                 inc.inc_fibnum = M_GETFIB(m);
2198                 inc.inc_flags |= INC_ISIPV6;
2199                 inc.inc_fport = th->th_dport;
2200                 inc.inc_lport = th->th_sport;
2201                 inc.inc6_faddr = *dst;
2202                 inc.inc6_laddr = ip6->ip6_src;
2203                 syncache_unreach(&inc, th);
2204         }
2205         INP_INFO_RUNLOCK(&V_tcbinfo);
2206 }
2207 #endif /* INET6 */
2208
2209
2210 /*
2211  * Following is where TCP initial sequence number generation occurs.
2212  *
2213  * There are two places where we must use initial sequence numbers:
2214  * 1.  In SYN-ACK packets.
2215  * 2.  In SYN packets.
2216  *
2217  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2218  * tcp_syncache.c for details.
2219  *
2220  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2221  * depends on this property.  In addition, these ISNs should be
2222  * unguessable so as to prevent connection hijacking.  To satisfy
2223  * the requirements of this situation, the algorithm outlined in
2224  * RFC 1948 is used, with only small modifications.
2225  *
2226  * Implementation details:
2227  *
2228  * Time is based off the system timer, and is corrected so that it
2229  * increases by one megabyte per second.  This allows for proper
2230  * recycling on high speed LANs while still leaving over an hour
2231  * before rollover.
2232  *
2233  * As reading the *exact* system time is too expensive to be done
2234  * whenever setting up a TCP connection, we increment the time
2235  * offset in two ways.  First, a small random positive increment
2236  * is added to isn_offset for each connection that is set up.
2237  * Second, the function tcp_isn_tick fires once per clock tick
2238  * and increments isn_offset as necessary so that sequence numbers
2239  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2240  * random positive increments serve only to ensure that the same
2241  * exact sequence number is never sent out twice (as could otherwise
2242  * happen when a port is recycled in less than the system tick
2243  * interval.)
2244  *
2245  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2246  * between seeding of isn_secret.  This is normally set to zero,
2247  * as reseeding should not be necessary.
2248  *
2249  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2250  * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock.  In
2251  * general, this means holding an exclusive (write) lock.
2252  */
2253
2254 #define ISN_BYTES_PER_SECOND 1048576
2255 #define ISN_STATIC_INCREMENT 4096
2256 #define ISN_RANDOM_INCREMENT (4096 - 1)
2257
2258 static VNET_DEFINE(u_char, isn_secret[32]);
2259 static VNET_DEFINE(int, isn_last);
2260 static VNET_DEFINE(int, isn_last_reseed);
2261 static VNET_DEFINE(u_int32_t, isn_offset);
2262 static VNET_DEFINE(u_int32_t, isn_offset_old);
2263
2264 #define V_isn_secret                    VNET(isn_secret)
2265 #define V_isn_last                      VNET(isn_last)
2266 #define V_isn_last_reseed               VNET(isn_last_reseed)
2267 #define V_isn_offset                    VNET(isn_offset)
2268 #define V_isn_offset_old                VNET(isn_offset_old)
2269
2270 tcp_seq
2271 tcp_new_isn(struct tcpcb *tp)
2272 {
2273         MD5_CTX isn_ctx;
2274         u_int32_t md5_buffer[4];
2275         tcp_seq new_isn;
2276         u_int32_t projected_offset;
2277
2278         INP_WLOCK_ASSERT(tp->t_inpcb);
2279
2280         ISN_LOCK();
2281         /* Seed if this is the first use, reseed if requested. */
2282         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2283              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2284                 < (u_int)ticks))) {
2285                 read_random(&V_isn_secret, sizeof(V_isn_secret));
2286                 V_isn_last_reseed = ticks;
2287         }
2288
2289         /* Compute the md5 hash and return the ISN. */
2290         MD5Init(&isn_ctx);
2291         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2292         MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2293 #ifdef INET6
2294         if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2295                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2296                           sizeof(struct in6_addr));
2297                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2298                           sizeof(struct in6_addr));
2299         } else
2300 #endif
2301         {
2302                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2303                           sizeof(struct in_addr));
2304                 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2305                           sizeof(struct in_addr));
2306         }
2307         MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
2308         MD5Final((u_char *) &md5_buffer, &isn_ctx);
2309         new_isn = (tcp_seq) md5_buffer[0];
2310         V_isn_offset += ISN_STATIC_INCREMENT +
2311                 (arc4random() & ISN_RANDOM_INCREMENT);
2312         if (ticks != V_isn_last) {
2313                 projected_offset = V_isn_offset_old +
2314                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2315                 if (SEQ_GT(projected_offset, V_isn_offset))
2316                         V_isn_offset = projected_offset;
2317                 V_isn_offset_old = V_isn_offset;
2318                 V_isn_last = ticks;
2319         }
2320         new_isn += V_isn_offset;
2321         ISN_UNLOCK();
2322         return (new_isn);
2323 }
2324
2325 /*
2326  * When a specific ICMP unreachable message is received and the
2327  * connection state is SYN-SENT, drop the connection.  This behavior
2328  * is controlled by the icmp_may_rst sysctl.
2329  */
2330 struct inpcb *
2331 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2332 {
2333         struct tcpcb *tp;
2334
2335         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2336         INP_WLOCK_ASSERT(inp);
2337
2338         if ((inp->inp_flags & INP_TIMEWAIT) ||
2339             (inp->inp_flags & INP_DROPPED))
2340                 return (inp);
2341
2342         tp = intotcpcb(inp);
2343         if (tp->t_state != TCPS_SYN_SENT)
2344                 return (inp);
2345
2346         tp = tcp_drop(tp, errno);
2347         if (tp != NULL)
2348                 return (inp);
2349         else
2350                 return (NULL);
2351 }
2352
2353 /*
2354  * When `need fragmentation' ICMP is received, update our idea of the MSS
2355  * based on the new value. Also nudge TCP to send something, since we
2356  * know the packet we just sent was dropped.
2357  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2358  */
2359 static struct inpcb *
2360 tcp_mtudisc_notify(struct inpcb *inp, int error)
2361 {
2362
2363         tcp_mtudisc(inp, -1);
2364         return (inp);
2365 }
2366
2367 static void
2368 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2369 {
2370         struct tcpcb *tp;
2371         struct socket *so;
2372
2373         INP_WLOCK_ASSERT(inp);
2374         if ((inp->inp_flags & INP_TIMEWAIT) ||
2375             (inp->inp_flags & INP_DROPPED))
2376                 return;
2377
2378         tp = intotcpcb(inp);
2379         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2380
2381         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2382
2383         so = inp->inp_socket;
2384         SOCKBUF_LOCK(&so->so_snd);
2385         /* If the mss is larger than the socket buffer, decrease the mss. */
2386         if (so->so_snd.sb_hiwat < tp->t_maxseg)
2387                 tp->t_maxseg = so->so_snd.sb_hiwat;
2388         SOCKBUF_UNLOCK(&so->so_snd);
2389
2390         TCPSTAT_INC(tcps_mturesent);
2391         tp->t_rtttime = 0;
2392         tp->snd_nxt = tp->snd_una;
2393         tcp_free_sackholes(tp);
2394         tp->snd_recover = tp->snd_max;
2395         if (tp->t_flags & TF_SACK_PERMIT)
2396                 EXIT_FASTRECOVERY(tp->t_flags);
2397         tp->t_fb->tfb_tcp_output(tp);
2398 }
2399
2400 #ifdef INET
2401 /*
2402  * Look-up the routing entry to the peer of this inpcb.  If no route
2403  * is found and it cannot be allocated, then return 0.  This routine
2404  * is called by TCP routines that access the rmx structure and by
2405  * tcp_mss_update to get the peer/interface MTU.
2406  */
2407 uint32_t
2408 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2409 {
2410         struct nhop4_extended nh4;
2411         struct ifnet *ifp;
2412         uint32_t maxmtu = 0;
2413
2414         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2415
2416         if (inc->inc_faddr.s_addr != INADDR_ANY) {
2417
2418                 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2419                     NHR_REF, 0, &nh4) != 0)
2420                         return (0);
2421
2422                 ifp = nh4.nh_ifp;
2423                 maxmtu = nh4.nh_mtu;
2424
2425                 /* Report additional interface capabilities. */
2426                 if (cap != NULL) {
2427                         if (ifp->if_capenable & IFCAP_TSO4 &&
2428                             ifp->if_hwassist & CSUM_TSO) {
2429                                 cap->ifcap |= CSUM_TSO;
2430                                 cap->tsomax = ifp->if_hw_tsomax;
2431                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2432                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2433                         }
2434                 }
2435                 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2436         }
2437         return (maxmtu);
2438 }
2439 #endif /* INET */
2440
2441 #ifdef INET6
2442 uint32_t
2443 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2444 {
2445         struct nhop6_extended nh6;
2446         struct in6_addr dst6;
2447         uint32_t scopeid;
2448         struct ifnet *ifp;
2449         uint32_t maxmtu = 0;
2450
2451         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2452
2453         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2454                 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2455                 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2456                     0, &nh6) != 0)
2457                         return (0);
2458
2459                 ifp = nh6.nh_ifp;
2460                 maxmtu = nh6.nh_mtu;
2461
2462                 /* Report additional interface capabilities. */
2463                 if (cap != NULL) {
2464                         if (ifp->if_capenable & IFCAP_TSO6 &&
2465                             ifp->if_hwassist & CSUM_TSO) {
2466                                 cap->ifcap |= CSUM_TSO;
2467                                 cap->tsomax = ifp->if_hw_tsomax;
2468                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2469                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2470                         }
2471                 }
2472                 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2473         }
2474
2475         return (maxmtu);
2476 }
2477 #endif /* INET6 */
2478
2479 /*
2480  * Calculate effective SMSS per RFC5681 definition for a given TCP
2481  * connection at its current state, taking into account SACK and etc.
2482  */
2483 u_int
2484 tcp_maxseg(const struct tcpcb *tp)
2485 {
2486         u_int optlen;
2487
2488         if (tp->t_flags & TF_NOOPT)
2489                 return (tp->t_maxseg);
2490
2491         /*
2492          * Here we have a simplified code from tcp_addoptions(),
2493          * without a proper loop, and having most of paddings hardcoded.
2494          * We might make mistakes with padding here in some edge cases,
2495          * but this is harmless, since result of tcp_maxseg() is used
2496          * only in cwnd and ssthresh estimations.
2497          */
2498 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
2499         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2500                 if (tp->t_flags & TF_RCVD_TSTMP)
2501                         optlen = TCPOLEN_TSTAMP_APPA;
2502                 else
2503                         optlen = 0;
2504 #ifdef TCP_SIGNATURE
2505                 if (tp->t_flags & TF_SIGNATURE)
2506                         optlen += PAD(TCPOLEN_SIGNATURE);
2507 #endif
2508                 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2509                         optlen += TCPOLEN_SACKHDR;
2510                         optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2511                         optlen = PAD(optlen);
2512                 }
2513         } else {
2514                 if (tp->t_flags & TF_REQ_TSTMP)
2515                         optlen = TCPOLEN_TSTAMP_APPA;
2516                 else
2517                         optlen = PAD(TCPOLEN_MAXSEG);
2518                 if (tp->t_flags & TF_REQ_SCALE)
2519                         optlen += PAD(TCPOLEN_WINDOW);
2520 #ifdef TCP_SIGNATURE
2521                 if (tp->t_flags & TF_SIGNATURE)
2522                         optlen += PAD(TCPOLEN_SIGNATURE);
2523 #endif
2524                 if (tp->t_flags & TF_SACK_PERMIT)
2525                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
2526         }
2527 #undef PAD
2528         optlen = min(optlen, TCP_MAXOLEN);
2529         return (tp->t_maxseg - optlen);
2530 }
2531
2532 #ifdef IPSEC
2533 /* compute ESP/AH header size for TCP, including outer IP header. */
2534 size_t
2535 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2536 {
2537         struct inpcb *inp;
2538         struct mbuf *m;
2539         size_t hdrsiz;
2540         struct ip *ip;
2541 #ifdef INET6
2542         struct ip6_hdr *ip6;
2543 #endif
2544         struct tcphdr *th;
2545
2546         if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2547                 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2548                 return (0);
2549         m = m_gethdr(M_NOWAIT, MT_DATA);
2550         if (!m)
2551                 return (0);
2552
2553 #ifdef INET6
2554         if ((inp->inp_vflag & INP_IPV6) != 0) {
2555                 ip6 = mtod(m, struct ip6_hdr *);
2556                 th = (struct tcphdr *)(ip6 + 1);
2557                 m->m_pkthdr.len = m->m_len =
2558                         sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2559                 tcpip_fillheaders(inp, ip6, th);
2560                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2561         } else
2562 #endif /* INET6 */
2563         {
2564                 ip = mtod(m, struct ip *);
2565                 th = (struct tcphdr *)(ip + 1);
2566                 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2567                 tcpip_fillheaders(inp, ip, th);
2568                 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2569         }
2570
2571         m_free(m);
2572         return (hdrsiz);
2573 }
2574 #endif /* IPSEC */
2575
2576 #ifdef TCP_SIGNATURE
2577 /*
2578  * Callback function invoked by m_apply() to digest TCP segment data
2579  * contained within an mbuf chain.
2580  */
2581 static int
2582 tcp_signature_apply(void *fstate, void *data, u_int len)
2583 {
2584
2585         MD5Update(fstate, (u_char *)data, len);
2586         return (0);
2587 }
2588
2589 /*
2590  * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2591  * search with the destination IP address, and a 'magic SPI' to be
2592  * determined by the application. This is hardcoded elsewhere to 1179
2593 */
2594 struct secasvar *
2595 tcp_get_sav(struct mbuf *m, u_int direction)
2596 {
2597         union sockaddr_union dst;
2598         struct secasvar *sav;
2599         struct ip *ip;
2600 #ifdef INET6
2601         struct ip6_hdr *ip6;
2602         char ip6buf[INET6_ADDRSTRLEN];
2603 #endif
2604
2605         /* Extract the destination from the IP header in the mbuf. */
2606         bzero(&dst, sizeof(union sockaddr_union));
2607         ip = mtod(m, struct ip *);
2608 #ifdef INET6
2609         ip6 = NULL;     /* Make the compiler happy. */
2610 #endif
2611         switch (ip->ip_v) {
2612 #ifdef INET
2613         case IPVERSION:
2614                 dst.sa.sa_len = sizeof(struct sockaddr_in);
2615                 dst.sa.sa_family = AF_INET;
2616                 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2617                     ip->ip_src : ip->ip_dst;
2618                 break;
2619 #endif
2620 #ifdef INET6
2621         case (IPV6_VERSION >> 4):
2622                 ip6 = mtod(m, struct ip6_hdr *);
2623                 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2624                 dst.sa.sa_family = AF_INET6;
2625                 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2626                     ip6->ip6_src : ip6->ip6_dst;
2627                 break;
2628 #endif
2629         default:
2630                 return (NULL);
2631                 /* NOTREACHED */
2632                 break;
2633         }
2634
2635         /* Look up an SADB entry which matches the address of the peer. */
2636         sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2637         if (sav == NULL) {
2638                 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2639                     (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2640 #ifdef INET6
2641                         (ip->ip_v == (IPV6_VERSION >> 4)) ?
2642                             ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2643 #endif
2644                         "(unsupported)"));
2645         }
2646
2647         return (sav);
2648 }
2649
2650 /*
2651  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2652  *
2653  * Parameters:
2654  * m            pointer to head of mbuf chain
2655  * len          length of TCP segment data, excluding options
2656  * optlen       length of TCP segment options
2657  * buf          pointer to storage for computed MD5 digest
2658  * sav          pointer to security assosiation
2659  *
2660  * We do this over ip, tcphdr, segment data, and the key in the SADB.
2661  * When called from tcp_input(), we can be sure that th_sum has been
2662  * zeroed out and verified already.
2663  *
2664  * Releases reference to SADB key before return.
2665  *
2666  * Return 0 if successful, otherwise return -1.
2667  *
2668  */
2669 int
2670 tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2671     u_char *buf, struct secasvar *sav)
2672 {
2673 #ifdef INET
2674         struct ippseudo ippseudo;
2675 #endif
2676         MD5_CTX ctx;
2677         int doff;
2678         struct ip *ip;
2679 #ifdef INET
2680         struct ipovly *ipovly;
2681 #endif
2682         struct tcphdr *th;
2683 #ifdef INET6
2684         struct ip6_hdr *ip6;
2685         struct in6_addr in6;
2686         uint32_t plen;
2687         uint16_t nhdr;
2688 #endif
2689         u_short savecsum;
2690
2691         KASSERT(m != NULL, ("NULL mbuf chain"));
2692         KASSERT(buf != NULL, ("NULL signature pointer"));
2693
2694         /* Extract the destination from the IP header in the mbuf. */
2695         ip = mtod(m, struct ip *);
2696 #ifdef INET6
2697         ip6 = NULL;     /* Make the compiler happy. */
2698 #endif
2699
2700         MD5Init(&ctx);
2701         /*
2702          * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2703          *
2704          * XXX The ippseudo header MUST be digested in network byte order,
2705          * or else we'll fail the regression test. Assume all fields we've
2706          * been doing arithmetic on have been in host byte order.
2707          * XXX One cannot depend on ipovly->ih_len here. When called from
2708          * tcp_output(), the underlying ip_len member has not yet been set.
2709          */
2710         switch (ip->ip_v) {
2711 #ifdef INET
2712         case IPVERSION:
2713                 ipovly = (struct ipovly *)ip;
2714                 ippseudo.ippseudo_src = ipovly->ih_src;
2715                 ippseudo.ippseudo_dst = ipovly->ih_dst;
2716                 ippseudo.ippseudo_pad = 0;
2717                 ippseudo.ippseudo_p = IPPROTO_TCP;
2718                 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2719                     optlen);
2720                 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2721
2722                 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2723                 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2724                 break;
2725 #endif
2726 #ifdef INET6
2727         /*
2728          * RFC 2385, 2.0  Proposal
2729          * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2730          * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2731          * extended next header value (to form 32 bits), and 32-bit segment
2732          * length.
2733          * Note: Upper-Layer Packet Length comes before Next Header.
2734          */
2735         case (IPV6_VERSION >> 4):
2736                 ip6 = mtod(m, struct ip6_hdr *);
2737                 in6 = ip6->ip6_src;
2738                 in6_clearscope(&in6);
2739                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2740                 in6 = ip6->ip6_dst;
2741                 in6_clearscope(&in6);
2742                 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2743                 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2744                 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2745                 nhdr = 0;
2746                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2747                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2748                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2749                 nhdr = IPPROTO_TCP;
2750                 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2751
2752                 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2753                 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2754                 break;
2755 #endif
2756         default:
2757                 KEY_FREESAV(&sav);
2758                 return (-1);
2759                 /* NOTREACHED */
2760                 break;
2761         }
2762
2763
2764         /*
2765          * Step 2: Update MD5 hash with TCP header, excluding options.
2766          * The TCP checksum must be set to zero.
2767          */
2768         savecsum = th->th_sum;
2769         th->th_sum = 0;
2770         MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2771         th->th_sum = savecsum;
2772
2773         /*
2774          * Step 3: Update MD5 hash with TCP segment data.
2775          *         Use m_apply() to avoid an early m_pullup().
2776          */
2777         if (len > 0)
2778                 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2779
2780         /*
2781          * Step 4: Update MD5 hash with shared secret.
2782          */
2783         MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2784         MD5Final(buf, &ctx);
2785
2786         key_sa_recordxfer(sav, m);
2787         KEY_FREESAV(&sav);
2788         return (0);
2789 }
2790
2791 /*
2792  * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2793  *
2794  * Return 0 if successful, otherwise return -1.
2795  */
2796 int
2797 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2798     u_char *buf, u_int direction)
2799 {
2800         struct secasvar *sav;
2801
2802         if ((sav = tcp_get_sav(m, direction)) == NULL)
2803                 return (-1);
2804
2805         return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2806 }
2807
2808 /*
2809  * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2810  *
2811  * Parameters:
2812  * m            pointer to head of mbuf chain
2813  * len          length of TCP segment data, excluding options
2814  * optlen       length of TCP segment options
2815  * buf          pointer to storage for computed MD5 digest
2816  * direction    direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2817  *
2818  * Return 1 if successful, otherwise return 0.
2819  */
2820 int
2821 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2822     struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2823 {
2824         char tmpdigest[TCP_SIGLEN];
2825
2826         if (tcp_sig_checksigs == 0)
2827                 return (1);
2828         if ((tcpbflag & TF_SIGNATURE) == 0) {
2829                 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2830
2831                         /*
2832                          * If this socket is not expecting signature but
2833                          * the segment contains signature just fail.
2834                          */
2835                         TCPSTAT_INC(tcps_sig_err_sigopt);
2836                         TCPSTAT_INC(tcps_sig_rcvbadsig);
2837                         return (0);
2838                 }
2839
2840                 /* Signature is not expected, and not present in segment. */
2841                 return (1);
2842         }
2843
2844         /*
2845          * If this socket is expecting signature but the segment does not
2846          * contain any just fail.
2847          */
2848         if ((to->to_flags & TOF_SIGNATURE) == 0) {
2849                 TCPSTAT_INC(tcps_sig_err_nosigopt);
2850                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2851                 return (0);
2852         }
2853         if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2854             IPSEC_DIR_INBOUND) == -1) {
2855                 TCPSTAT_INC(tcps_sig_err_buildsig);
2856                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2857                 return (0);
2858         }
2859
2860         if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2861                 TCPSTAT_INC(tcps_sig_rcvbadsig);
2862                 return (0);
2863         }
2864         TCPSTAT_INC(tcps_sig_rcvgoodsig);
2865         return (1);
2866 }
2867 #endif /* TCP_SIGNATURE */
2868
2869 static int
2870 sysctl_drop(SYSCTL_HANDLER_ARGS)
2871 {
2872         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2873         struct sockaddr_storage addrs[2];
2874         struct inpcb *inp;
2875         struct tcpcb *tp;
2876         struct tcptw *tw;
2877         struct sockaddr_in *fin, *lin;
2878 #ifdef INET6
2879         struct sockaddr_in6 *fin6, *lin6;
2880 #endif
2881         int error;
2882
2883         inp = NULL;
2884         fin = lin = NULL;
2885 #ifdef INET6
2886         fin6 = lin6 = NULL;
2887 #endif
2888         error = 0;
2889
2890         if (req->oldptr != NULL || req->oldlen != 0)
2891                 return (EINVAL);
2892         if (req->newptr == NULL)
2893                 return (EPERM);
2894         if (req->newlen < sizeof(addrs))
2895                 return (ENOMEM);
2896         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2897         if (error)
2898                 return (error);
2899
2900         switch (addrs[0].ss_family) {
2901 #ifdef INET6
2902         case AF_INET6:
2903                 fin6 = (struct sockaddr_in6 *)&addrs[0];
2904                 lin6 = (struct sockaddr_in6 *)&addrs[1];
2905                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2906                     lin6->sin6_len != sizeof(struct sockaddr_in6))
2907                         return (EINVAL);
2908                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2909                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2910                                 return (EINVAL);
2911                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2912                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2913                         fin = (struct sockaddr_in *)&addrs[0];
2914                         lin = (struct sockaddr_in *)&addrs[1];
2915                         break;
2916                 }
2917                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2918                 if (error)
2919                         return (error);
2920                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2921                 if (error)
2922                         return (error);
2923                 break;
2924 #endif
2925 #ifdef INET
2926         case AF_INET:
2927                 fin = (struct sockaddr_in *)&addrs[0];
2928                 lin = (struct sockaddr_in *)&addrs[1];
2929                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2930                     lin->sin_len != sizeof(struct sockaddr_in))
2931                         return (EINVAL);
2932                 break;
2933 #endif
2934         default:
2935                 return (EINVAL);
2936         }
2937         INP_INFO_RLOCK(&V_tcbinfo);
2938         switch (addrs[0].ss_family) {
2939 #ifdef INET6
2940         case AF_INET6:
2941                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2942                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2943                     INPLOOKUP_WLOCKPCB, NULL);
2944                 break;
2945 #endif
2946 #ifdef INET
2947         case AF_INET:
2948                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2949                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2950                 break;
2951 #endif
2952         }
2953         if (inp != NULL) {
2954                 if (inp->inp_flags & INP_TIMEWAIT) {
2955                         /*
2956                          * XXXRW: There currently exists a state where an
2957                          * inpcb is present, but its timewait state has been
2958                          * discarded.  For now, don't allow dropping of this
2959                          * type of inpcb.
2960                          */
2961                         tw = intotw(inp);
2962                         if (tw != NULL)
2963                                 tcp_twclose(tw, 0);
2964                         else
2965                                 INP_WUNLOCK(inp);
2966                 } else if (!(inp->inp_flags & INP_DROPPED) &&
2967                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2968                         tp = intotcpcb(inp);
2969                         tp = tcp_drop(tp, ECONNABORTED);
2970                         if (tp != NULL)
2971                                 INP_WUNLOCK(inp);
2972                 } else
2973                         INP_WUNLOCK(inp);
2974         } else
2975                 error = ESRCH;
2976         INP_INFO_RUNLOCK(&V_tcbinfo);
2977         return (error);
2978 }
2979
2980 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2981     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2982     0, sysctl_drop, "", "Drop TCP connection");
2983
2984 /*
2985  * Generate a standardized TCP log line for use throughout the
2986  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2987  * allow use in the interrupt context.
2988  *
2989  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2990  * NB: The function may return NULL if memory allocation failed.
2991  *
2992  * Due to header inclusion and ordering limitations the struct ip
2993  * and ip6_hdr pointers have to be passed as void pointers.
2994  */
2995 char *
2996 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2997     const void *ip6hdr)
2998 {
2999
3000         /* Is logging enabled? */
3001         if (tcp_log_in_vain == 0)
3002                 return (NULL);
3003
3004         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3005 }
3006
3007 char *
3008 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3009     const void *ip6hdr)
3010 {
3011
3012         /* Is logging enabled? */
3013         if (tcp_log_debug == 0)
3014                 return (NULL);
3015
3016         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3017 }
3018
3019 static char *
3020 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3021     const void *ip6hdr)
3022 {
3023         char *s, *sp;
3024         size_t size;
3025         struct ip *ip;
3026 #ifdef INET6
3027         const struct ip6_hdr *ip6;
3028
3029         ip6 = (const struct ip6_hdr *)ip6hdr;
3030 #endif /* INET6 */
3031         ip = (struct ip *)ip4hdr;
3032
3033         /*
3034          * The log line looks like this:
3035          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3036          */
3037         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3038             sizeof(PRINT_TH_FLAGS) + 1 +
3039 #ifdef INET6
3040             2 * INET6_ADDRSTRLEN;
3041 #else
3042             2 * INET_ADDRSTRLEN;
3043 #endif /* INET6 */
3044
3045         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3046         if (s == NULL)
3047                 return (NULL);
3048
3049         strcat(s, "TCP: [");
3050         sp = s + strlen(s);
3051
3052         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3053                 inet_ntoa_r(inc->inc_faddr, sp);
3054                 sp = s + strlen(s);
3055                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3056                 sp = s + strlen(s);
3057                 inet_ntoa_r(inc->inc_laddr, sp);
3058                 sp = s + strlen(s);
3059                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3060 #ifdef INET6
3061         } else if (inc) {
3062                 ip6_sprintf(sp, &inc->inc6_faddr);
3063                 sp = s + strlen(s);
3064                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3065                 sp = s + strlen(s);
3066                 ip6_sprintf(sp, &inc->inc6_laddr);
3067                 sp = s + strlen(s);
3068                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3069         } else if (ip6 && th) {
3070                 ip6_sprintf(sp, &ip6->ip6_src);
3071                 sp = s + strlen(s);
3072                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3073                 sp = s + strlen(s);
3074                 ip6_sprintf(sp, &ip6->ip6_dst);
3075                 sp = s + strlen(s);
3076                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3077 #endif /* INET6 */
3078 #ifdef INET
3079         } else if (ip && th) {
3080                 inet_ntoa_r(ip->ip_src, sp);
3081                 sp = s + strlen(s);
3082                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3083                 sp = s + strlen(s);
3084                 inet_ntoa_r(ip->ip_dst, sp);
3085                 sp = s + strlen(s);
3086                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3087 #endif /* INET */
3088         } else {
3089                 free(s, M_TCPLOG);
3090                 return (NULL);
3091         }
3092         sp = s + strlen(s);
3093         if (th)
3094                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3095         if (*(s + size - 1) != '\0')
3096                 panic("%s: string too long", __func__);
3097         return (s);
3098 }
3099
3100 /*
3101  * A subroutine which makes it easy to track TCP state changes with DTrace.
3102  * This function shouldn't be called for t_state initializations that don't
3103  * correspond to actual TCP state transitions.
3104  */
3105 void
3106 tcp_state_change(struct tcpcb *tp, int newstate)
3107 {
3108 #if defined(KDTRACE_HOOKS)
3109         int pstate = tp->t_state;
3110 #endif
3111
3112         TCPSTATES_DEC(tp->t_state);
3113         TCPSTATES_INC(newstate);
3114         tp->t_state = newstate;
3115         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3116 }