sys/netinet/tcp_subr.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-3-Clause
   3  *
   4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   5  *      The Regents of the University of California.  All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  * 3. Neither the name of the University nor the names of its contributors
  16  *    may be used to endorse or promote products derived from this software
  17  *    without specific prior written permission.
  18  *
  19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  29  * SUCH DAMAGE.
  30  *
  31  *      @(#)tcp_subr.c  8.2 (Berkeley) 5/24/95
  32  */
  33
  34 #include <sys/cdefs.h>
  35 __FBSDID("$FreeBSD$");
  36
  37 #include "opt_inet.h"
  38 #include "opt_inet6.h"
  39 #include "opt_ipsec.h"
  40 #include "opt_kern_tls.h"
  41 #include "opt_tcpdebug.h"
  42
  43 #include <sys/param.h>
  44 #include <sys/systm.h>
  45 #include <sys/arb.h>
  46 #include <sys/callout.h>
  47 #include <sys/eventhandler.h>
  48 #ifdef TCP_HHOOK
  49 #include <sys/hhook.h>
  50 #endif
  51 #include <sys/kernel.h>
  52 #ifdef TCP_HHOOK
  53 #include <sys/khelp.h>
  54 #endif
  55 #ifdef KERN_TLS
  56 #include <sys/ktls.h>
  57 #endif
  58 #include <sys/qmath.h>
  59 #include <sys/stats.h>
  60 #include <sys/sysctl.h>
  61 #include <sys/jail.h>
  62 #include <sys/malloc.h>
  63 #include <sys/refcount.h>
  64 #include <sys/mbuf.h>
  65 #ifdef INET6
  66 #include <sys/domain.h>
  67 #endif
  68 #include <sys/priv.h>
  69 #include <sys/proc.h>
  70 #include <sys/sdt.h>
  71 #include <sys/socket.h>
  72 #include <sys/socketvar.h>
  73 #include <sys/protosw.h>
  74 #include <sys/random.h>
  75
  76 #include <vm/uma.h>
  77
  78 #include <net/route.h>
  79 #include <net/route/nhop.h>
  80 #include <net/if.h>
  81 #include <net/if_var.h>
  82 #include <net/vnet.h>
  83
  84 #include <netinet/in.h>
  85 #include <netinet/in_fib.h>
  86 #include <netinet/in_kdtrace.h>
  87 #include <netinet/in_pcb.h>
  88 #include <netinet/in_systm.h>
  89 #include <netinet/in_var.h>
  90 #include <netinet/ip.h>
  91 #include <netinet/ip_icmp.h>
  92 #include <netinet/ip_var.h>
  93 #ifdef INET6
  94 #include <netinet/icmp6.h>
  95 #include <netinet/ip6.h>
  96 #include <netinet6/in6_fib.h>
  97 #include <netinet6/in6_pcb.h>
  98 #include <netinet6/ip6_var.h>
  99 #include <netinet6/scope6_var.h>
 100 #include <netinet6/nd6.h>
 101 #endif
 102
 103 #include <netinet/tcp.h>
 104 #include <netinet/tcp_fsm.h>
 105 #include <netinet/tcp_seq.h>
 106 #include <netinet/tcp_timer.h>
 107 #include <netinet/tcp_var.h>
 108 #include <netinet/tcp_log_buf.h>
 109 #include <netinet/tcp_syncache.h>
 110 #include <netinet/tcp_hpts.h>
 111 #include <netinet/cc/cc.h>
 112 #ifdef INET6
 113 #include <netinet6/tcp6_var.h>
 114 #endif
 115 #include <netinet/tcpip.h>
 116 #include <netinet/tcp_fastopen.h>
 117 #ifdef TCPPCAP
 118 #include <netinet/tcp_pcap.h>
 119 #endif
 120 #ifdef TCPDEBUG
 121 #include <netinet/tcp_debug.h>
 122 #endif
 123 #ifdef INET6
 124 #include <netinet6/ip6protosw.h>
 125 #endif
 126 #ifdef TCP_OFFLOAD
 127 #include <netinet/tcp_offload.h>
 128 #endif
 129 #include <netinet/udp.h>
 130 #include <netinet/udp_var.h>
 131
 132 #include <netipsec/ipsec_support.h>
 133
 134 #include <machine/in_cksum.h>
 135 #include <crypto/siphash/siphash.h>
 136
 137 #include <security/mac/mac_framework.h>
 138
 139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
 140 #ifdef INET6
 141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
 142 #endif
 143
 144 #ifdef NETFLIX_EXP_DETECTION
 145 /*  Sack attack detection thresholds and such */
 146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack,
 147     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
 148     "Sack Attack detection thresholds");
 149 int32_t tcp_force_detection = 0;
 150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection,
 151     CTLFLAG_RW,
 152     &tcp_force_detection, 0,
 153     "Do we force detection even if the INP has it off?");
 154 int32_t tcp_sack_to_ack_thresh = 700;   /* 70 % */
 155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh,
 156     CTLFLAG_RW,
 157     &tcp_sack_to_ack_thresh, 700,
 158     "Percentage of sacks to acks we must see above (10.1 percent is 101)?");
 159 int32_t tcp_sack_to_move_thresh = 600;  /* 60 % */
 160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh,
 161     CTLFLAG_RW,
 162     &tcp_sack_to_move_thresh, 600,
 163     "Percentage of sack moves we must see above (10.1 percent is 101)");
 164 int32_t tcp_restoral_thresh = 650;      /* 65 % (sack:2:ack -5%) */
 165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh,
 166     CTLFLAG_RW,
 167     &tcp_restoral_thresh, 550,
 168     "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)");
 169 int32_t tcp_sad_decay_val = 800;
 170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per,
 171     CTLFLAG_RW,
 172     &tcp_sad_decay_val, 800,
 173     "The decay percentage (10.1 percent equals 101 )");
 174 int32_t tcp_map_minimum = 500;
 175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps,
 176     CTLFLAG_RW,
 177     &tcp_map_minimum, 500,
 178     "Number of Map enteries before we start detection");
 179 int32_t tcp_attack_on_turns_on_logging = 0;
 180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged,
 181     CTLFLAG_RW,
 182     &tcp_attack_on_turns_on_logging, 0,
 183    "When we have a positive hit on attack, do we turn on logging?");
 184 int32_t tcp_sad_pacing_interval = 2000;
 185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int,
 186     CTLFLAG_RW,
 187     &tcp_sad_pacing_interval, 2000,
 188     "What is the minimum pacing interval for a classified attacker?");
 189
 190 int32_t tcp_sad_low_pps = 100;
 191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps,
 192     CTLFLAG_RW,
 193     &tcp_sad_low_pps, 100,
 194     "What is the input pps that below which we do not decay?");
 195 #endif
 196 uint32_t tcp_ack_war_time_window = 1000;
 197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow,
 198     CTLFLAG_RW,
 199     &tcp_ack_war_time_window, 1000,
 200    "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?");
 201 uint32_t tcp_ack_war_cnt = 5;
 202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt,
 203     CTLFLAG_RW,
 204     &tcp_ack_war_cnt, 5,
 205    "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?");
 206
 207 struct rwlock tcp_function_lock;
 208
 209 static int
 210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
 211 {
 212         int error, new;
 213
 214         new = V_tcp_mssdflt;
 215         error = sysctl_handle_int(oidp, &new, 0, req);
 216         if (error == 0 && req->newptr) {
 217                 if (new < TCP_MINMSS)
 218                         error = EINVAL;
 219                 else
 220                         V_tcp_mssdflt = new;
 221         }
 222         return (error);
 223 }
 224
 225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
 226     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 227     &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I",
 228     "Default TCP Maximum Segment Size");
 229
 230 #ifdef INET6
 231 static int
 232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
 233 {
 234         int error, new;
 235
 236         new = V_tcp_v6mssdflt;
 237         error = sysctl_handle_int(oidp, &new, 0, req);
 238         if (error == 0 && req->newptr) {
 239                 if (new < TCP_MINMSS)
 240                         error = EINVAL;
 241                 else
 242                         V_tcp_v6mssdflt = new;
 243         }
 244         return (error);
 245 }
 246
 247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
 248     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 249     &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
 250    "Default TCP Maximum Segment Size for IPv6");
 251 #endif /* INET6 */
 252
 253 /*
 254  * Minimum MSS we accept and use. This prevents DoS attacks where
 255  * we are forced to a ridiculous low MSS like 20 and send hundreds
 256  * of packets instead of one. The effect scales with the available
 257  * bandwidth and quickly saturates the CPU and network interface
 258  * with packet generation and sending. Set to zero to disable MINMSS
 259  * checking. This setting prevents us from sending too small packets.
 260  */
 261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
 262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
 263      &VNET_NAME(tcp_minmss), 0,
 264     "Minimum TCP Maximum Segment Size");
 265
 266 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
 267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
 268     &VNET_NAME(tcp_do_rfc1323), 0,
 269     "Enable rfc1323 (high performance TCP) extensions");
 270
 271 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 0;
 272 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW,
 273     &VNET_NAME(tcp_tolerate_missing_ts), 0,
 274     "Tolerate missing TCP timestamps");
 275
 276 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1;
 277 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW,
 278     &VNET_NAME(tcp_ts_offset_per_conn), 0,
 279     "Initialize TCP timestamps per connection instead of per host pair");
 280
 281 /* How many connections are pacing */
 282 static volatile uint32_t number_of_tcp_connections_pacing = 0;
 283 static uint32_t shadow_num_connections = 0;
 284
 285 static int tcp_pacing_limit = 10000;
 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW,
 287     &tcp_pacing_limit, 1000,
 288     "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)");
 289
 290 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD,
 291     &shadow_num_connections, 0, "Number of TCP connections being paced");
 292
 293 static int      tcp_log_debug = 0;
 294 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
 295     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
 296
 297 static int      tcp_tcbhashsize;
 298 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
 299     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
 300
 301 static int      do_tcpdrain = 1;
 302 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
 303     "Enable tcp_drain routine for extra help when low on mbufs");
 304
 305 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
 306     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
 307
 308 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1;
 309 #define V_icmp_may_rst                  VNET(icmp_may_rst)
 310 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
 311     &VNET_NAME(icmp_may_rst), 0,
 312     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
 313
 314 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0;
 315 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
 316 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
 317     &VNET_NAME(tcp_isn_reseed_interval), 0,
 318     "Seconds between reseeding of ISN secret");
 319
 320 static int      tcp_soreceive_stream;
 321 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
 322     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
 323
 324 VNET_DEFINE(uma_zone_t, sack_hole_zone);
 325 #define V_sack_hole_zone                VNET(sack_hole_zone)
 326 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0;       /* unlimited */
 327 static int
 328 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS)
 329 {
 330         int error;
 331         uint32_t new;
 332
 333         new = V_tcp_map_entries_limit;
 334         error = sysctl_handle_int(oidp, &new, 0, req);
 335         if (error == 0 && req->newptr) {
 336                 /* only allow "0" and value > minimum */
 337                 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT)
 338                         error = EINVAL;
 339                 else
 340                         V_tcp_map_entries_limit = new;
 341         }
 342         return (error);
 343 }
 344 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit,
 345     CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 346     &VNET_NAME(tcp_map_entries_limit), 0,
 347     &sysctl_net_inet_tcp_map_limit_check, "IU",
 348     "Total sendmap entries limit");
 349
 350 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */
 351 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW,
 352      &VNET_NAME(tcp_map_split_limit), 0,
 353     "Total sendmap split entries limit");
 354
 355 #ifdef TCP_HHOOK
 356 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
 357 #endif
 358
 359 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH
 360 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]);
 361 #define V_ts_offset_secret      VNET(ts_offset_secret)
 362
 363 static int      tcp_default_fb_init(struct tcpcb *tp);
 364 static void     tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged);
 365 static int      tcp_default_handoff_ok(struct tcpcb *tp);
 366 static struct inpcb *tcp_notify(struct inpcb *, int);
 367 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
 368 static void tcp_mtudisc(struct inpcb *, int);
 369 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
 370                     void *ip4hdr, const void *ip6hdr);
 371
 372 static struct tcp_function_block tcp_def_funcblk = {
 373         .tfb_tcp_block_name = "freebsd",
 374         .tfb_tcp_output = tcp_output,
 375         .tfb_tcp_do_segment = tcp_do_segment,
 376         .tfb_tcp_ctloutput = tcp_default_ctloutput,
 377         .tfb_tcp_handoff_ok = tcp_default_handoff_ok,
 378         .tfb_tcp_fb_init = tcp_default_fb_init,
 379         .tfb_tcp_fb_fini = tcp_default_fb_fini,
 380 };
 381
 382 static int tcp_fb_cnt = 0;
 383 struct tcp_funchead t_functions;
 384 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
 385
 386 static struct tcp_function_block *
 387 find_tcp_functions_locked(struct tcp_function_set *fs)
 388 {
 389         struct tcp_function *f;
 390         struct tcp_function_block *blk=NULL;
 391
 392         TAILQ_FOREACH(f, &t_functions, tf_next) {
 393                 if (strcmp(f->tf_name, fs->function_set_name) == 0) {
 394                         blk = f->tf_fb;
 395                         break;
 396                 }
 397         }
 398         return(blk);
 399 }
 400
 401 static struct tcp_function_block *
 402 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
 403 {
 404         struct tcp_function_block *rblk=NULL;
 405         struct tcp_function *f;
 406
 407         TAILQ_FOREACH(f, &t_functions, tf_next) {
 408                 if (f->tf_fb == blk) {
 409                         rblk = blk;
 410                         if (s) {
 411                                 *s = f;
 412                         }
 413                         break;
 414                 }
 415         }
 416         return (rblk);
 417 }
 418
 419 struct tcp_function_block *
 420 find_and_ref_tcp_functions(struct tcp_function_set *fs)
 421 {
 422         struct tcp_function_block *blk;
 423
 424         rw_rlock(&tcp_function_lock);
 425         blk = find_tcp_functions_locked(fs);
 426         if (blk)
 427                 refcount_acquire(&blk->tfb_refcnt);
 428         rw_runlock(&tcp_function_lock);
 429         return(blk);
 430 }
 431
 432 struct tcp_function_block *
 433 find_and_ref_tcp_fb(struct tcp_function_block *blk)
 434 {
 435         struct tcp_function_block *rblk;
 436
 437         rw_rlock(&tcp_function_lock);
 438         rblk = find_tcp_fb_locked(blk, NULL);
 439         if (rblk)
 440                 refcount_acquire(&rblk->tfb_refcnt);
 441         rw_runlock(&tcp_function_lock);
 442         return(rblk);
 443 }
 444
 445 static struct tcp_function_block *
 446 find_and_ref_tcp_default_fb(void)
 447 {
 448         struct tcp_function_block *rblk;
 449
 450         rw_rlock(&tcp_function_lock);
 451         rblk = tcp_func_set_ptr;
 452         refcount_acquire(&rblk->tfb_refcnt);
 453         rw_runlock(&tcp_function_lock);
 454         return (rblk);
 455 }
 456
 457 void
 458 tcp_switch_back_to_default(struct tcpcb *tp)
 459 {
 460         struct tcp_function_block *tfb;
 461
 462         KASSERT(tp->t_fb != &tcp_def_funcblk,
 463             ("%s: called by the built-in default stack", __func__));
 464
 465         /*
 466          * Release the old stack. This function will either find a new one
 467          * or panic.
 468          */
 469         if (tp->t_fb->tfb_tcp_fb_fini != NULL)
 470                 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
 471         refcount_release(&tp->t_fb->tfb_refcnt);
 472
 473         /*
 474          * Now, we'll find a new function block to use.
 475          * Start by trying the current user-selected
 476          * default, unless this stack is the user-selected
 477          * default.
 478          */
 479         tfb = find_and_ref_tcp_default_fb();
 480         if (tfb == tp->t_fb) {
 481                 refcount_release(&tfb->tfb_refcnt);
 482                 tfb = NULL;
 483         }
 484         /* Does the stack accept this connection? */
 485         if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL &&
 486             (*tfb->tfb_tcp_handoff_ok)(tp)) {
 487                 refcount_release(&tfb->tfb_refcnt);
 488                 tfb = NULL;
 489         }
 490         /* Try to use that stack. */
 491         if (tfb != NULL) {
 492                 /* Initialize the new stack. If it succeeds, we are done. */
 493                 tp->t_fb = tfb;
 494                 if (tp->t_fb->tfb_tcp_fb_init == NULL ||
 495                     (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0)
 496                         return;
 497
 498                 /*
 499                  * Initialization failed. Release the reference count on
 500                  * the stack.
 501                  */
 502                 refcount_release(&tfb->tfb_refcnt);
 503         }
 504
 505         /*
 506          * If that wasn't feasible, use the built-in default
 507          * stack which is not allowed to reject anyone.
 508          */
 509         tfb = find_and_ref_tcp_fb(&tcp_def_funcblk);
 510         if (tfb == NULL) {
 511                 /* there always should be a default */
 512                 panic("Can't refer to tcp_def_funcblk");
 513         }
 514         if (tfb->tfb_tcp_handoff_ok != NULL) {
 515                 if ((*tfb->tfb_tcp_handoff_ok) (tp)) {
 516                         /* The default stack cannot say no */
 517                         panic("Default stack rejects a new session?");
 518                 }
 519         }
 520         tp->t_fb = tfb;
 521         if (tp->t_fb->tfb_tcp_fb_init != NULL &&
 522             (*tp->t_fb->tfb_tcp_fb_init)(tp)) {
 523                 /* The default stack cannot fail */
 524                 panic("Default stack initialization failed");
 525         }
 526 }
 527
 528 static void
 529 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp,
 530     const struct sockaddr *sa, void *ctx)
 531 {
 532         struct ip *iph;
 533 #ifdef INET6
 534         struct ip6_hdr *ip6;
 535 #endif
 536         struct udphdr *uh;
 537         struct tcphdr *th;
 538         int thlen;
 539         uint16_t port;
 540
 541         TCPSTAT_INC(tcps_tunneled_pkts);
 542         if ((m->m_flags & M_PKTHDR) == 0) {
 543                 /* Can't handle one that is not a pkt hdr */
 544                 TCPSTAT_INC(tcps_tunneled_errs);
 545                 goto out;
 546         }
 547         thlen = sizeof(struct tcphdr);
 548         if (m->m_len < off + sizeof(struct udphdr) + thlen &&
 549             (m =  m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) {
 550                 TCPSTAT_INC(tcps_tunneled_errs);
 551                 goto out;
 552         }
 553         iph = mtod(m, struct ip *);
 554         uh = (struct udphdr *)((caddr_t)iph + off);
 555         th = (struct tcphdr *)(uh + 1);
 556         thlen = th->th_off << 2;
 557         if (m->m_len < off + sizeof(struct udphdr) + thlen) {
 558                 m =  m_pullup(m, off + sizeof(struct udphdr) + thlen);
 559                 if (m == NULL) {
 560                         TCPSTAT_INC(tcps_tunneled_errs);
 561                         goto out;
 562                 } else {
 563                         iph = mtod(m, struct ip *);
 564                         uh = (struct udphdr *)((caddr_t)iph + off);
 565                         th = (struct tcphdr *)(uh + 1);
 566                 }
 567         }
 568         m->m_pkthdr.tcp_tun_port = port = uh->uh_sport;
 569         bcopy(th, uh, m->m_len - off);
 570         m->m_len -= sizeof(struct udphdr);
 571         m->m_pkthdr.len -= sizeof(struct udphdr);
 572         /*
 573          * We use the same algorithm for
 574          * both UDP and TCP for c-sum. So
 575          * the code in tcp_input will skip
 576          * the checksum. So we do nothing
 577          * with the flag (m->m_pkthdr.csum_flags).
 578          */
 579         switch (iph->ip_v) {
 580 #ifdef INET
 581         case IPVERSION:
 582                 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr));
 583                 tcp_input_with_port(&m, &off, IPPROTO_TCP, port);
 584                 break;
 585 #endif
 586 #ifdef INET6
 587         case IPV6_VERSION >> 4:
 588                 ip6 = mtod(m, struct ip6_hdr *);
 589                 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr));
 590                 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port);
 591                 break;
 592 #endif
 593         default:
 594                 goto out;
 595                 break;
 596         }
 597         return;
 598 out:
 599         m_freem(m);
 600 }
 601
 602 static int
 603 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
 604 {
 605         int error=ENOENT;
 606         struct tcp_function_set fs;
 607         struct tcp_function_block *blk;
 608
 609         memset(&fs, 0, sizeof(fs));
 610         rw_rlock(&tcp_function_lock);
 611         blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
 612         if (blk) {
 613                 /* Found him */
 614                 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 615                 fs.pcbcnt = blk->tfb_refcnt;
 616         }
 617         rw_runlock(&tcp_function_lock);
 618         error = sysctl_handle_string(oidp, fs.function_set_name,
 619                                      sizeof(fs.function_set_name), req);
 620
 621         /* Check for error or no change */
 622         if (error != 0 || req->newptr == NULL)
 623                 return(error);
 624
 625         rw_wlock(&tcp_function_lock);
 626         blk = find_tcp_functions_locked(&fs);
 627         if ((blk == NULL) ||
 628             (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
 629                 error = ENOENT;
 630                 goto done;
 631         }
 632         tcp_func_set_ptr = blk;
 633 done:
 634         rw_wunlock(&tcp_function_lock);
 635         return (error);
 636 }
 637
 638 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
 639     CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
 640     NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
 641     "Set/get the default TCP functions");
 642
 643 static int
 644 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
 645 {
 646         int error, cnt, linesz;
 647         struct tcp_function *f;
 648         char *buffer, *cp;
 649         size_t bufsz, outsz;
 650         bool alias;
 651
 652         cnt = 0;
 653         rw_rlock(&tcp_function_lock);
 654         TAILQ_FOREACH(f, &t_functions, tf_next) {
 655                 cnt++;
 656         }
 657         rw_runlock(&tcp_function_lock);
 658
 659         bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1;
 660         buffer = malloc(bufsz, M_TEMP, M_WAITOK);
 661
 662         error = 0;
 663         cp = buffer;
 664
 665         linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D',
 666             "Alias", "PCB count");
 667         cp += linesz;
 668         bufsz -= linesz;
 669         outsz = linesz;
 670
 671         rw_rlock(&tcp_function_lock);
 672         TAILQ_FOREACH(f, &t_functions, tf_next) {
 673                 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name);
 674                 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n",
 675                     f->tf_fb->tfb_tcp_block_name,
 676                     (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
 677                     alias ? f->tf_name : "-",
 678                     f->tf_fb->tfb_refcnt);
 679                 if (linesz >= bufsz) {
 680                         error = EOVERFLOW;
 681                         break;
 682                 }
 683                 cp += linesz;
 684                 bufsz -= linesz;
 685                 outsz += linesz;
 686         }
 687         rw_runlock(&tcp_function_lock);
 688         if (error == 0)
 689                 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
 690         free(buffer, M_TEMP);
 691         return (error);
 692 }
 693
 694 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
 695     CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
 696     NULL, 0, sysctl_net_inet_list_available, "A",
 697     "list available TCP Function sets");
 698
 699 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT;
 700
 701 #ifdef INET
 702 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL;
 703 #define V_udp4_tun_socket       VNET(udp4_tun_socket)
 704 #endif
 705 #ifdef INET6
 706 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL;
 707 #define V_udp6_tun_socket       VNET(udp6_tun_socket)
 708 #endif
 709
 710 static void
 711 tcp_over_udp_stop(void)
 712 {
 713         /*
 714          * This function assumes sysctl caller holds inp_rinfo_lock()
 715          * for writting!
 716          */
 717 #ifdef INET
 718         if (V_udp4_tun_socket != NULL) {
 719                 soclose(V_udp4_tun_socket);
 720                 V_udp4_tun_socket = NULL;
 721         }
 722 #endif
 723 #ifdef INET6
 724         if (V_udp6_tun_socket != NULL) {
 725                 soclose(V_udp6_tun_socket);
 726                 V_udp6_tun_socket = NULL;
 727         }
 728 #endif
 729 }
 730
 731 static int
 732 tcp_over_udp_start(void)
 733 {
 734         uint16_t port;
 735         int ret;
 736 #ifdef INET
 737         struct sockaddr_in sin;
 738 #endif
 739 #ifdef INET6
 740         struct sockaddr_in6 sin6;
 741 #endif
 742         /*
 743          * This function assumes sysctl caller holds inp_info_rlock()
 744          * for writting!
 745          */
 746         port = V_tcp_udp_tunneling_port;
 747         if (ntohs(port) == 0) {
 748                 /* Must have a port set */
 749                 return (EINVAL);
 750         }
 751 #ifdef INET
 752         if (V_udp4_tun_socket != NULL) {
 753                 /* Already running -- must stop first */
 754                 return (EALREADY);
 755         }
 756 #endif
 757 #ifdef INET6
 758         if (V_udp6_tun_socket != NULL) {
 759                 /* Already running -- must stop first */
 760                 return (EALREADY);
 761         }
 762 #endif
 763 #ifdef INET
 764         if ((ret = socreate(PF_INET, &V_udp4_tun_socket,
 765             SOCK_DGRAM, IPPROTO_UDP,
 766             curthread->td_ucred, curthread))) {
 767                 tcp_over_udp_stop();
 768                 return (ret);
 769         }
 770         /* Call the special UDP hook. */
 771         if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket,
 772             tcp_recv_udp_tunneled_packet,
 773             tcp_ctlinput_viaudp,
 774             NULL))) {
 775                 tcp_over_udp_stop();
 776                 return (ret);
 777         }
 778         /* Ok, we have a socket, bind it to the port. */
 779         memset(&sin, 0, sizeof(struct sockaddr_in));
 780         sin.sin_len = sizeof(struct sockaddr_in);
 781         sin.sin_family = AF_INET;
 782         sin.sin_port = htons(port);
 783         if ((ret = sobind(V_udp4_tun_socket,
 784             (struct sockaddr *)&sin, curthread))) {
 785                 tcp_over_udp_stop();
 786                 return (ret);
 787         }
 788 #endif
 789 #ifdef INET6
 790         if ((ret = socreate(PF_INET6, &V_udp6_tun_socket,
 791             SOCK_DGRAM, IPPROTO_UDP,
 792             curthread->td_ucred, curthread))) {
 793                 tcp_over_udp_stop();
 794                 return (ret);
 795         }
 796         /* Call the special UDP hook. */
 797         if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket,
 798             tcp_recv_udp_tunneled_packet,
 799             tcp6_ctlinput_viaudp,
 800             NULL))) {
 801                 tcp_over_udp_stop();
 802                 return (ret);
 803         }
 804         /* Ok, we have a socket, bind it to the port. */
 805         memset(&sin6, 0, sizeof(struct sockaddr_in6));
 806         sin6.sin6_len = sizeof(struct sockaddr_in6);
 807         sin6.sin6_family = AF_INET6;
 808         sin6.sin6_port = htons(port);
 809         if ((ret = sobind(V_udp6_tun_socket,
 810             (struct sockaddr *)&sin6, curthread))) {
 811                 tcp_over_udp_stop();
 812                 return (ret);
 813         }
 814 #endif
 815         return (0);
 816 }
 817
 818 static int
 819 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS)
 820 {
 821         int error;
 822         uint32_t old, new;
 823
 824         old = V_tcp_udp_tunneling_port;
 825         new = old;
 826         error = sysctl_handle_int(oidp, &new, 0, req);
 827         if ((error == 0) &&
 828             (req->newptr != NULL)) {
 829                 if ((new < TCP_TUNNELING_PORT_MIN) ||
 830                     (new > TCP_TUNNELING_PORT_MAX)) {
 831                         error = EINVAL;
 832                 } else {
 833                         V_tcp_udp_tunneling_port = new;
 834                         if (old != 0) {
 835                                 tcp_over_udp_stop();
 836                         }
 837                         if (new != 0) {
 838                                 error = tcp_over_udp_start();
 839                         }
 840                 }
 841         }
 842         return (error);
 843 }
 844
 845 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port,
 846     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
 847     &VNET_NAME(tcp_udp_tunneling_port),
 848     0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU",
 849     "Tunneling port for tcp over udp");
 850
 851 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT;
 852
 853 static int
 854 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS)
 855 {
 856         int error, new;
 857
 858         new = V_tcp_udp_tunneling_overhead;
 859         error = sysctl_handle_int(oidp, &new, 0, req);
 860         if (error == 0 && req->newptr) {
 861                 if ((new < TCP_TUNNELING_OVERHEAD_MIN) ||
 862                     (new > TCP_TUNNELING_OVERHEAD_MAX))
 863                         error = EINVAL;
 864                 else
 865                         V_tcp_udp_tunneling_overhead = new;
 866         }
 867         return (error);
 868 }
 869
 870 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead,
 871     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
 872     &VNET_NAME(tcp_udp_tunneling_overhead),
 873     0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU",
 874     "MSS reduction when using tcp over udp");
 875
 876 /*
 877  * Exports one (struct tcp_function_info) for each alias/name.
 878  */
 879 static int
 880 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS)
 881 {
 882         int cnt, error;
 883         struct tcp_function *f;
 884         struct tcp_function_info tfi;
 885
 886         /*
 887          * We don't allow writes.
 888          */
 889         if (req->newptr != NULL)
 890                 return (EINVAL);
 891
 892         /*
 893          * Wire the old buffer so we can directly copy the functions to
 894          * user space without dropping the lock.
 895          */
 896         if (req->oldptr != NULL) {
 897                 error = sysctl_wire_old_buffer(req, 0);
 898                 if (error)
 899                         return (error);
 900         }
 901
 902         /*
 903          * Walk the list and copy out matching entries. If INVARIANTS
 904          * is compiled in, also walk the list to verify the length of
 905          * the list matches what we have recorded.
 906          */
 907         rw_rlock(&tcp_function_lock);
 908
 909         cnt = 0;
 910 #ifndef INVARIANTS
 911         if (req->oldptr == NULL) {
 912                 cnt = tcp_fb_cnt;
 913                 goto skip_loop;
 914         }
 915 #endif
 916         TAILQ_FOREACH(f, &t_functions, tf_next) {
 917 #ifdef INVARIANTS
 918                 cnt++;
 919 #endif
 920                 if (req->oldptr != NULL) {
 921                         bzero(&tfi, sizeof(tfi));
 922                         tfi.tfi_refcnt = f->tf_fb->tfb_refcnt;
 923                         tfi.tfi_id = f->tf_fb->tfb_id;
 924                         (void)strlcpy(tfi.tfi_alias, f->tf_name,
 925                             sizeof(tfi.tfi_alias));
 926                         (void)strlcpy(tfi.tfi_name,
 927                             f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name));
 928                         error = SYSCTL_OUT(req, &tfi, sizeof(tfi));
 929                         /*
 930                          * Don't stop on error, as that is the
 931                          * mechanism we use to accumulate length
 932                          * information if the buffer was too short.
 933                          */
 934                 }
 935         }
 936         KASSERT(cnt == tcp_fb_cnt,
 937             ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt));
 938 #ifndef INVARIANTS
 939 skip_loop:
 940 #endif
 941         rw_runlock(&tcp_function_lock);
 942         if (req->oldptr == NULL)
 943                 error = SYSCTL_OUT(req, NULL,
 944                     (cnt + 1) * sizeof(struct tcp_function_info));
 945
 946         return (error);
 947 }
 948
 949 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info,
 950             CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE,
 951             NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info",
 952             "List TCP function block name-to-ID mappings");
 953
 954 /*
 955  * tfb_tcp_handoff_ok() function for the default stack.
 956  * Note that we'll basically try to take all comers.
 957  */
 958 static int
 959 tcp_default_handoff_ok(struct tcpcb *tp)
 960 {
 961
 962         return (0);
 963 }
 964
 965 /*
 966  * tfb_tcp_fb_init() function for the default stack.
 967  *
 968  * This handles making sure we have appropriate timers set if you are
 969  * transitioning a socket that has some amount of setup done.
 970  *
 971  * The init() fuction from the default can *never* return non-zero i.e.
 972  * it is required to always succeed since it is the stack of last resort!
 973  */
 974 static int
 975 tcp_default_fb_init(struct tcpcb *tp)
 976 {
 977
 978         struct socket *so;
 979
 980         INP_WLOCK_ASSERT(tp->t_inpcb);
 981
 982         KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT,
 983             ("%s: connection %p in unexpected state %d", __func__, tp,
 984             tp->t_state));
 985
 986         /*
 987          * Nothing to do for ESTABLISHED or LISTEN states. And, we don't
 988          * know what to do for unexpected states (which includes TIME_WAIT).
 989          */
 990         if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT)
 991                 return (0);
 992
 993         /*
 994          * Make sure some kind of transmission timer is set if there is
 995          * outstanding data.
 996          */
 997         so = tp->t_inpcb->inp_socket;
 998         if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) ||
 999             tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) ||
1000             tcp_timer_active(tp, TT_PERSIST))) {
1001                 /*
1002                  * If the session has established and it looks like it should
1003                  * be in the persist state, set the persist timer. Otherwise,
1004                  * set the retransmit timer.
1005                  */
1006                 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 &&
1007                     (int32_t)(tp->snd_nxt - tp->snd_una) <
1008                     (int32_t)sbavail(&so->so_snd))
1009                         tcp_setpersist(tp);
1010                 else
1011                         tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
1012         }
1013
1014         /* All non-embryonic sessions get a keepalive timer. */
1015         if (!tcp_timer_active(tp, TT_KEEP))
1016                 tcp_timer_activate(tp, TT_KEEP,
1017                     TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) :
1018                     TP_KEEPINIT(tp));
1019
1020         /*
1021          * Make sure critical variables are initialized
1022          * if transitioning while in Recovery.
1023          */
1024         if IN_FASTRECOVERY(tp->t_flags) {
1025                 if (tp->sackhint.recover_fs == 0)
1026                         tp->sackhint.recover_fs = max(1,
1027                             tp->snd_nxt - tp->snd_una);
1028         }
1029
1030         return (0);
1031 }
1032
1033 /*
1034  * tfb_tcp_fb_fini() function for the default stack.
1035  *
1036  * This changes state as necessary (or prudent) to prepare for another stack
1037  * to assume responsibility for the connection.
1038  */
1039 static void
1040 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged)
1041 {
1042
1043         INP_WLOCK_ASSERT(tp->t_inpcb);
1044         return;
1045 }
1046
1047 /*
1048  * Target size of TCP PCB hash tables. Must be a power of two.
1049  *
1050  * Note that this can be overridden by the kernel environment
1051  * variable net.inet.tcp.tcbhashsize
1052  */
1053 #ifndef TCBHASHSIZE
1054 #define TCBHASHSIZE     0
1055 #endif
1056
1057 /*
1058  * XXX
1059  * Callouts should be moved into struct tcp directly.  They are currently
1060  * separate because the tcpcb structure is exported to userland for sysctl
1061  * parsing purposes, which do not know about callouts.
1062  */
1063 struct tcpcb_mem {
1064         struct  tcpcb           tcb;
1065         struct  tcp_timer       tt;
1066         struct  cc_var          ccv;
1067 #ifdef TCP_HHOOK
1068         struct  osd             osd;
1069 #endif
1070 };
1071
1072 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone);
1073 #define V_tcpcb_zone                    VNET(tcpcb_zone)
1074
1075 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
1076 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
1077
1078 static struct mtx isn_mtx;
1079
1080 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
1081 #define ISN_LOCK()      mtx_lock(&isn_mtx)
1082 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
1083
1084 /*
1085  * TCP initialization.
1086  */
1087 static void
1088 tcp_zone_change(void *tag)
1089 {
1090
1091         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
1092         uma_zone_set_max(V_tcpcb_zone, maxsockets);
1093         tcp_tw_zone_change();
1094 }
1095
1096 static int
1097 tcp_inpcb_init(void *mem, int size, int flags)
1098 {
1099         struct inpcb *inp = mem;
1100
1101         INP_LOCK_INIT(inp, "inp", "tcpinp");
1102         return (0);
1103 }
1104
1105 /*
1106  * Take a value and get the next power of 2 that doesn't overflow.
1107  * Used to size the tcp_inpcb hash buckets.
1108  */
1109 static int
1110 maketcp_hashsize(int size)
1111 {
1112         int hashsize;
1113
1114         /*
1115          * auto tune.
1116          * get the next power of 2 higher than maxsockets.
1117          */
1118         hashsize = 1 << fls(size);
1119         /* catch overflow, and just go one power of 2 smaller */
1120         if (hashsize < size) {
1121                 hashsize = 1 << (fls(size) - 1);
1122         }
1123         return (hashsize);
1124 }
1125
1126 static volatile int next_tcp_stack_id = 1;
1127
1128 /*
1129  * Register a TCP function block with the name provided in the names
1130  * array.  (Note that this function does NOT automatically register
1131  * blk->tfb_tcp_block_name as a stack name.  Therefore, you should
1132  * explicitly include blk->tfb_tcp_block_name in the list of names if
1133  * you wish to register the stack with that name.)
1134  *
1135  * Either all name registrations will succeed or all will fail.  If
1136  * a name registration fails, the function will update the num_names
1137  * argument to point to the array index of the name that encountered
1138  * the failure.
1139  *
1140  * Returns 0 on success, or an error code on failure.
1141  */
1142 int
1143 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait,
1144     const char *names[], int *num_names)
1145 {
1146         struct tcp_function *n;
1147         struct tcp_function_set fs;
1148         int error, i;
1149
1150         KASSERT(names != NULL && *num_names > 0,
1151             ("%s: Called with 0-length name list", __func__));
1152         KASSERT(names != NULL, ("%s: Called with NULL name list", __func__));
1153         KASSERT(rw_initialized(&tcp_function_lock),
1154             ("%s: called too early", __func__));
1155
1156         if ((blk->tfb_tcp_output == NULL) ||
1157             (blk->tfb_tcp_do_segment == NULL) ||
1158             (blk->tfb_tcp_ctloutput == NULL) ||
1159             (strlen(blk->tfb_tcp_block_name) == 0)) {
1160                 /*
1161                  * These functions are required and you
1162                  * need a name.
1163                  */
1164                 *num_names = 0;
1165                 return (EINVAL);
1166         }
1167         if (blk->tfb_tcp_timer_stop_all ||
1168             blk->tfb_tcp_timer_activate ||
1169             blk->tfb_tcp_timer_active ||
1170             blk->tfb_tcp_timer_stop) {
1171                 /*
1172                  * If you define one timer function you
1173                  * must have them all.
1174                  */
1175                 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
1176                     (blk->tfb_tcp_timer_activate == NULL) ||
1177                     (blk->tfb_tcp_timer_active == NULL) ||
1178                     (blk->tfb_tcp_timer_stop == NULL)) {
1179                         *num_names = 0;
1180                         return (EINVAL);
1181                 }
1182         }
1183
1184         if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1185                 *num_names = 0;
1186                 return (EINVAL);
1187         }
1188
1189         refcount_init(&blk->tfb_refcnt, 0);
1190         blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1);
1191         for (i = 0; i < *num_names; i++) {
1192                 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
1193                 if (n == NULL) {
1194                         error = ENOMEM;
1195                         goto cleanup;
1196                 }
1197                 n->tf_fb = blk;
1198
1199                 (void)strlcpy(fs.function_set_name, names[i],
1200                     sizeof(fs.function_set_name));
1201                 rw_wlock(&tcp_function_lock);
1202                 if (find_tcp_functions_locked(&fs) != NULL) {
1203                         /* Duplicate name space not allowed */
1204                         rw_wunlock(&tcp_function_lock);
1205                         free(n, M_TCPFUNCTIONS);
1206                         error = EALREADY;
1207                         goto cleanup;
1208                 }
1209                 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name));
1210                 TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
1211                 tcp_fb_cnt++;
1212                 rw_wunlock(&tcp_function_lock);
1213         }
1214         return(0);
1215
1216 cleanup:
1217         /*
1218          * Deregister the names we just added. Because registration failed
1219          * for names[i], we don't need to deregister that name.
1220          */
1221         *num_names = i;
1222         rw_wlock(&tcp_function_lock);
1223         while (--i >= 0) {
1224                 TAILQ_FOREACH(n, &t_functions, tf_next) {
1225                         if (!strncmp(n->tf_name, names[i],
1226                             TCP_FUNCTION_NAME_LEN_MAX)) {
1227                                 TAILQ_REMOVE(&t_functions, n, tf_next);
1228                                 tcp_fb_cnt--;
1229                                 n->tf_fb = NULL;
1230                                 free(n, M_TCPFUNCTIONS);
1231                                 break;
1232                         }
1233                 }
1234         }
1235         rw_wunlock(&tcp_function_lock);
1236         return (error);
1237 }
1238
1239 /*
1240  * Register a TCP function block using the name provided in the name
1241  * argument.
1242  *
1243  * Returns 0 on success, or an error code on failure.
1244  */
1245 int
1246 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name,
1247     int wait)
1248 {
1249         const char *name_list[1];
1250         int num_names, rv;
1251
1252         num_names = 1;
1253         if (name != NULL)
1254                 name_list[0] = name;
1255         else
1256                 name_list[0] = blk->tfb_tcp_block_name;
1257         rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names);
1258         return (rv);
1259 }
1260
1261 /*
1262  * Register a TCP function block using the name defined in
1263  * blk->tfb_tcp_block_name.
1264  *
1265  * Returns 0 on success, or an error code on failure.
1266  */
1267 int
1268 register_tcp_functions(struct tcp_function_block *blk, int wait)
1269 {
1270
1271         return (register_tcp_functions_as_name(blk, NULL, wait));
1272 }
1273
1274 /*
1275  * Deregister all names associated with a function block. This
1276  * functionally removes the function block from use within the system.
1277  *
1278  * When called with a true quiesce argument, mark the function block
1279  * as being removed so no more stacks will use it and determine
1280  * whether the removal would succeed.
1281  *
1282  * When called with a false quiesce argument, actually attempt the
1283  * removal.
1284  *
1285  * When called with a force argument, attempt to switch all TCBs to
1286  * use the default stack instead of returning EBUSY.
1287  *
1288  * Returns 0 on success (or if the removal would succeed, or an error
1289  * code on failure.
1290  */
1291 int
1292 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce,
1293     bool force)
1294 {
1295         struct tcp_function *f;
1296
1297         if (blk == &tcp_def_funcblk) {
1298                 /* You can't un-register the default */
1299                 return (EPERM);
1300         }
1301         rw_wlock(&tcp_function_lock);
1302         if (blk == tcp_func_set_ptr) {
1303                 /* You can't free the current default */
1304                 rw_wunlock(&tcp_function_lock);
1305                 return (EBUSY);
1306         }
1307         /* Mark the block so no more stacks can use it. */
1308         blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
1309         /*
1310          * If TCBs are still attached to the stack, attempt to switch them
1311          * to the default stack.
1312          */
1313         if (force && blk->tfb_refcnt) {
1314                 struct inpcb *inp;
1315                 struct tcpcb *tp;
1316                 VNET_ITERATOR_DECL(vnet_iter);
1317
1318                 rw_wunlock(&tcp_function_lock);
1319
1320                 VNET_LIST_RLOCK();
1321                 VNET_FOREACH(vnet_iter) {
1322                         CURVNET_SET(vnet_iter);
1323                         INP_INFO_WLOCK(&V_tcbinfo);
1324                         CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) {
1325                                 INP_WLOCK(inp);
1326                                 if (inp->inp_flags & INP_TIMEWAIT) {
1327                                         INP_WUNLOCK(inp);
1328                                         continue;
1329                                 }
1330                                 tp = intotcpcb(inp);
1331                                 if (tp == NULL || tp->t_fb != blk) {
1332                                         INP_WUNLOCK(inp);
1333                                         continue;
1334                                 }
1335                                 tcp_switch_back_to_default(tp);
1336                                 INP_WUNLOCK(inp);
1337                         }
1338                         INP_INFO_WUNLOCK(&V_tcbinfo);
1339                         CURVNET_RESTORE();
1340                 }
1341                 VNET_LIST_RUNLOCK();
1342
1343                 rw_wlock(&tcp_function_lock);
1344         }
1345         if (blk->tfb_refcnt) {
1346                 /* TCBs still attached. */
1347                 rw_wunlock(&tcp_function_lock);
1348                 return (EBUSY);
1349         }
1350         if (quiesce) {
1351                 /* Skip removal. */
1352                 rw_wunlock(&tcp_function_lock);
1353                 return (0);
1354         }
1355         /* Remove any function names that map to this function block. */
1356         while (find_tcp_fb_locked(blk, &f) != NULL) {
1357                 TAILQ_REMOVE(&t_functions, f, tf_next);
1358                 tcp_fb_cnt--;
1359                 f->tf_fb = NULL;
1360                 free(f, M_TCPFUNCTIONS);
1361         }
1362         rw_wunlock(&tcp_function_lock);
1363         return (0);
1364 }
1365
1366 void
1367 tcp_init(void)
1368 {
1369         const char *tcbhash_tuneable;
1370         int hashsize;
1371
1372         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
1373
1374 #ifdef TCP_HHOOK
1375         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
1376             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1377                 printf("%s: WARNING: unable to register helper hook\n", __func__);
1378         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
1379             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
1380                 printf("%s: WARNING: unable to register helper hook\n", __func__);
1381 #endif
1382 #ifdef STATS
1383         if (tcp_stats_init())
1384                 printf("%s: WARNING: unable to initialise TCP stats\n",
1385                     __func__);
1386 #endif
1387         hashsize = TCBHASHSIZE;
1388         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
1389         if (hashsize == 0) {
1390                 /*
1391                  * Auto tune the hash size based on maxsockets.
1392                  * A perfect hash would have a 1:1 mapping
1393                  * (hashsize = maxsockets) however it's been
1394                  * suggested that O(2) average is better.
1395                  */
1396                 hashsize = maketcp_hashsize(maxsockets / 4);
1397                 /*
1398                  * Our historical default is 512,
1399                  * do not autotune lower than this.
1400                  */
1401                 if (hashsize < 512)
1402                         hashsize = 512;
1403                 if (bootverbose && IS_DEFAULT_VNET(curvnet))
1404                         printf("%s: %s auto tuned to %d\n", __func__,
1405                             tcbhash_tuneable, hashsize);
1406         }
1407         /*
1408          * We require a hashsize to be a power of two.
1409          * Previously if it was not a power of two we would just reset it
1410          * back to 512, which could be a nasty surprise if you did not notice
1411          * the error message.
1412          * Instead what we do is clip it to the closest power of two lower
1413          * than the specified hash value.
1414          */
1415         if (!powerof2(hashsize)) {
1416                 int oldhashsize = hashsize;
1417
1418                 hashsize = maketcp_hashsize(hashsize);
1419                 /* prevent absurdly low value */
1420                 if (hashsize < 16)
1421                         hashsize = 16;
1422                 printf("%s: WARNING: TCB hash size not a power of 2, "
1423                     "clipped from %d to %d.\n", __func__, oldhashsize,
1424                     hashsize);
1425         }
1426         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
1427             "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE);
1428
1429         /*
1430          * These have to be type stable for the benefit of the timers.
1431          */
1432         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
1433             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1434         uma_zone_set_max(V_tcpcb_zone, maxsockets);
1435         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
1436
1437         tcp_tw_init();
1438         syncache_init();
1439         tcp_hc_init();
1440
1441         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
1442         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
1443             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1444
1445         tcp_fastopen_init();
1446
1447         /* Skip initialization of globals for non-default instances. */
1448         if (!IS_DEFAULT_VNET(curvnet))
1449                 return;
1450
1451         tcp_reass_global_init();
1452
1453         /* XXX virtualize those bellow? */
1454         tcp_delacktime = TCPTV_DELACK;
1455         tcp_keepinit = TCPTV_KEEP_INIT;
1456         tcp_keepidle = TCPTV_KEEP_IDLE;
1457         tcp_keepintvl = TCPTV_KEEPINTVL;
1458         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
1459         tcp_msl = TCPTV_MSL;
1460         tcp_rexmit_initial = TCPTV_RTOBASE;
1461         if (tcp_rexmit_initial < 1)
1462                 tcp_rexmit_initial = 1;
1463         tcp_rexmit_min = TCPTV_MIN;
1464         if (tcp_rexmit_min < 1)
1465                 tcp_rexmit_min = 1;
1466         tcp_persmin = TCPTV_PERSMIN;
1467         tcp_persmax = TCPTV_PERSMAX;
1468         tcp_rexmit_slop = TCPTV_CPU_VAR;
1469         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
1470         tcp_tcbhashsize = hashsize;
1471
1472         /* Setup the tcp function block list */
1473         TAILQ_INIT(&t_functions);
1474         rw_init(&tcp_function_lock, "tcp_func_lock");
1475         register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
1476 #ifdef TCP_BLACKBOX
1477         /* Initialize the TCP logging data. */
1478         tcp_log_init();
1479 #endif
1480         arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0);
1481
1482         if (tcp_soreceive_stream) {
1483 #ifdef INET
1484                 tcp_usrreqs.pru_soreceive = soreceive_stream;
1485 #endif
1486 #ifdef INET6
1487                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
1488 #endif /* INET6 */
1489         }
1490
1491 #ifdef INET6
1492 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
1493 #else /* INET6 */
1494 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
1495 #endif /* INET6 */
1496         if (max_protohdr < TCP_MINPROTOHDR)
1497                 max_protohdr = TCP_MINPROTOHDR;
1498         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
1499                 panic("tcp_init");
1500 #undef TCP_MINPROTOHDR
1501
1502         ISN_LOCK_INIT();
1503         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
1504                 SHUTDOWN_PRI_DEFAULT);
1505         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
1506                 EVENTHANDLER_PRI_ANY);
1507
1508         tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK);
1509         tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK);
1510         tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK);
1511         tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK);
1512         tcp_extra_mbuf = counter_u64_alloc(M_WAITOK);
1513         tcp_would_have_but = counter_u64_alloc(M_WAITOK);
1514         tcp_comp_total = counter_u64_alloc(M_WAITOK);
1515         tcp_uncomp_total = counter_u64_alloc(M_WAITOK);
1516 #ifdef TCPPCAP
1517         tcp_pcap_init();
1518 #endif
1519 }
1520
1521 #ifdef VIMAGE
1522 static void
1523 tcp_destroy(void *unused __unused)
1524 {
1525         int n;
1526 #ifdef TCP_HHOOK
1527         int error;
1528 #endif
1529
1530         /*
1531          * All our processes are gone, all our sockets should be cleaned
1532          * up, which means, we should be past the tcp_discardcb() calls.
1533          * Sleep to let all tcpcb timers really disappear and cleanup.
1534          */
1535         for (;;) {
1536                 INP_LIST_RLOCK(&V_tcbinfo);
1537                 n = V_tcbinfo.ipi_count;
1538                 INP_LIST_RUNLOCK(&V_tcbinfo);
1539                 if (n == 0)
1540                         break;
1541                 pause("tcpdes", hz / 10);
1542         }
1543         tcp_hc_destroy();
1544         syncache_destroy();
1545         tcp_tw_destroy();
1546         in_pcbinfo_destroy(&V_tcbinfo);
1547         /* tcp_discardcb() clears the sack_holes up. */
1548         uma_zdestroy(V_sack_hole_zone);
1549         uma_zdestroy(V_tcpcb_zone);
1550
1551         /*
1552          * Cannot free the zone until all tcpcbs are released as we attach
1553          * the allocations to them.
1554          */
1555         tcp_fastopen_destroy();
1556
1557 #ifdef TCP_HHOOK
1558         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
1559         if (error != 0) {
1560                 printf("%s: WARNING: unable to deregister helper hook "
1561                     "type=%d, id=%d: error %d returned\n", __func__,
1562                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
1563         }
1564         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
1565         if (error != 0) {
1566                 printf("%s: WARNING: unable to deregister helper hook "
1567                     "type=%d, id=%d: error %d returned\n", __func__,
1568                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
1569         }
1570 #endif
1571 }
1572 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
1573 #endif
1574
1575 void
1576 tcp_fini(void *xtp)
1577 {
1578
1579 }
1580
1581 /*
1582  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
1583  * tcp_template used to store this data in mbufs, but we now recopy it out
1584  * of the tcpcb each time to conserve mbufs.
1585  */
1586 void
1587 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr)
1588 {
1589         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
1590
1591         INP_WLOCK_ASSERT(inp);
1592
1593 #ifdef INET6
1594         if ((inp->inp_vflag & INP_IPV6) != 0) {
1595                 struct ip6_hdr *ip6;
1596
1597                 ip6 = (struct ip6_hdr *)ip_ptr;
1598                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
1599                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
1600                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
1601                         (IPV6_VERSION & IPV6_VERSION_MASK);
1602                 if (port == 0)
1603                         ip6->ip6_nxt = IPPROTO_TCP;
1604                 else
1605                         ip6->ip6_nxt = IPPROTO_UDP;
1606                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
1607                 ip6->ip6_src = inp->in6p_laddr;
1608                 ip6->ip6_dst = inp->in6p_faddr;
1609         }
1610 #endif /* INET6 */
1611 #if defined(INET6) && defined(INET)
1612         else
1613 #endif
1614 #ifdef INET
1615         {
1616                 struct ip *ip;
1617
1618                 ip = (struct ip *)ip_ptr;
1619                 ip->ip_v = IPVERSION;
1620                 ip->ip_hl = 5;
1621                 ip->ip_tos = inp->inp_ip_tos;
1622                 ip->ip_len = 0;
1623                 ip->ip_id = 0;
1624                 ip->ip_off = 0;
1625                 ip->ip_ttl = inp->inp_ip_ttl;
1626                 ip->ip_sum = 0;
1627                 if (port == 0)
1628                         ip->ip_p = IPPROTO_TCP;
1629                 else
1630                         ip->ip_p = IPPROTO_UDP;
1631                 ip->ip_src = inp->inp_laddr;
1632                 ip->ip_dst = inp->inp_faddr;
1633         }
1634 #endif /* INET */
1635         th->th_sport = inp->inp_lport;
1636         th->th_dport = inp->inp_fport;
1637         th->th_seq = 0;
1638         th->th_ack = 0;
1639         th->th_x2 = 0;
1640         th->th_off = 5;
1641         th->th_flags = 0;
1642         th->th_win = 0;
1643         th->th_urp = 0;
1644         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
1645 }
1646
1647 /*
1648  * Create template to be used to send tcp packets on a connection.
1649  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
1650  * use for this function is in keepalives, which use tcp_respond.
1651  */
1652 struct tcptemp *
1653 tcpip_maketemplate(struct inpcb *inp)
1654 {
1655         struct tcptemp *t;
1656
1657         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
1658         if (t == NULL)
1659                 return (NULL);
1660         tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t);
1661         return (t);
1662 }
1663
1664 /*
1665  * Send a single message to the TCP at address specified by
1666  * the given TCP/IP header.  If m == NULL, then we make a copy
1667  * of the tcpiphdr at th and send directly to the addressed host.
1668  * This is used to force keep alive messages out using the TCP
1669  * template for a connection.  If flags are given then we send
1670  * a message back to the TCP which originated the segment th,
1671  * and discard the mbuf containing it and any other attached mbufs.
1672  *
1673  * In any case the ack and sequence number of the transmitted
1674  * segment are as specified by the parameters.
1675  *
1676  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
1677  */
1678 void
1679 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
1680     tcp_seq ack, tcp_seq seq, int flags)
1681 {
1682         struct tcpopt to;
1683         struct inpcb *inp;
1684         struct ip *ip;
1685         struct mbuf *optm;
1686         struct udphdr *uh = NULL;
1687         struct tcphdr *nth;
1688         u_char *optp;
1689 #ifdef INET6
1690         struct ip6_hdr *ip6;
1691         int isipv6;
1692 #endif /* INET6 */
1693         int optlen, tlen, win, ulen;
1694         bool incl_opts;
1695         uint16_t port;
1696
1697         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
1698         NET_EPOCH_ASSERT();
1699
1700 #ifdef INET6
1701         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
1702         ip6 = ipgen;
1703 #endif /* INET6 */
1704         ip = ipgen;
1705
1706         if (tp != NULL) {
1707                 inp = tp->t_inpcb;
1708                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
1709                 INP_LOCK_ASSERT(inp);
1710         } else
1711                 inp = NULL;
1712
1713         if (m != NULL) {
1714 #ifdef INET6
1715                 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP))
1716                         port = m->m_pkthdr.tcp_tun_port;
1717                 else
1718 #endif
1719                 if (ip && (ip->ip_p == IPPROTO_UDP))
1720                         port = m->m_pkthdr.tcp_tun_port;
1721                 else
1722                         port = 0;
1723         } else
1724                 port = tp->t_port;
1725
1726         incl_opts = false;
1727         win = 0;
1728         if (tp != NULL) {
1729                 if (!(flags & TH_RST)) {
1730                         win = sbspace(&inp->inp_socket->so_rcv);
1731                         if (win > TCP_MAXWIN << tp->rcv_scale)
1732                                 win = TCP_MAXWIN << tp->rcv_scale;
1733                 }
1734                 if ((tp->t_flags & TF_NOOPT) == 0)
1735                         incl_opts = true;
1736         }
1737         if (m == NULL) {
1738                 m = m_gethdr(M_NOWAIT, MT_DATA);
1739                 if (m == NULL)
1740                         return;
1741                 m->m_data += max_linkhdr;
1742 #ifdef INET6
1743                 if (isipv6) {
1744                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
1745                               sizeof(struct ip6_hdr));
1746                         ip6 = mtod(m, struct ip6_hdr *);
1747                         nth = (struct tcphdr *)(ip6 + 1);
1748                         if (port) {
1749                                 /* Insert a UDP header */
1750                                 uh = (struct udphdr *)nth;
1751                                 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1752                                 uh->uh_dport = port;
1753                                 nth = (struct tcphdr *)(uh + 1);
1754                         }
1755                 } else
1756 #endif /* INET6 */
1757                 {
1758                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
1759                         ip = mtod(m, struct ip *);
1760                         nth = (struct tcphdr *)(ip + 1);
1761                         if (port) {
1762                                 /* Insert a UDP header */
1763                                 uh = (struct udphdr *)nth;
1764                                 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1765                                 uh->uh_dport = port;
1766                                 nth = (struct tcphdr *)(uh + 1);
1767                         }
1768                 }
1769                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1770                 flags = TH_ACK;
1771         } else if ((!M_WRITABLE(m)) || (port != 0)) {
1772                 struct mbuf *n;
1773
1774                 /* Can't reuse 'm', allocate a new mbuf. */
1775                 n = m_gethdr(M_NOWAIT, MT_DATA);
1776                 if (n == NULL) {
1777                         m_freem(m);
1778                         return;
1779                 }
1780
1781                 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
1782                         m_freem(m);
1783                         m_freem(n);
1784                         return;
1785                 }
1786
1787                 n->m_data += max_linkhdr;
1788                 /* m_len is set later */
1789 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
1790 #ifdef INET6
1791                 if (isipv6) {
1792                         bcopy((caddr_t)ip6, mtod(n, caddr_t),
1793                               sizeof(struct ip6_hdr));
1794                         ip6 = mtod(n, struct ip6_hdr *);
1795                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1796                         nth = (struct tcphdr *)(ip6 + 1);
1797                         if (port) {
1798                                 /* Insert a UDP header */
1799                                 uh = (struct udphdr *)nth;
1800                                 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1801                                 uh->uh_dport = port;
1802                                 nth = (struct tcphdr *)(uh + 1);
1803                         }
1804                 } else
1805 #endif /* INET6 */
1806                 {
1807                         bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
1808                         ip = mtod(n, struct ip *);
1809                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1810                         nth = (struct tcphdr *)(ip + 1);
1811                         if (port) {
1812                                 /* Insert a UDP header */
1813                                 uh = (struct udphdr *)nth;
1814                                 uh->uh_sport = htons(V_tcp_udp_tunneling_port);
1815                                 uh->uh_dport = port;
1816                                 nth = (struct tcphdr *)(uh + 1);
1817                         }
1818                 }
1819                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
1820                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1821                 th = nth;
1822                 m_freem(m);
1823                 m = n;
1824         } else {
1825                 /*
1826                  *  reuse the mbuf.
1827                  * XXX MRT We inherit the FIB, which is lucky.
1828                  */
1829                 m_freem(m->m_next);
1830                 m->m_next = NULL;
1831                 m->m_data = (caddr_t)ipgen;
1832                 /* m_len is set later */
1833 #ifdef INET6
1834                 if (isipv6) {
1835                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
1836                         nth = (struct tcphdr *)(ip6 + 1);
1837                 } else
1838 #endif /* INET6 */
1839                 {
1840                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
1841                         nth = (struct tcphdr *)(ip + 1);
1842                 }
1843                 if (th != nth) {
1844                         /*
1845                          * this is usually a case when an extension header
1846                          * exists between the IPv6 header and the
1847                          * TCP header.
1848                          */
1849                         nth->th_sport = th->th_sport;
1850                         nth->th_dport = th->th_dport;
1851                 }
1852                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1853 #undef xchg
1854         }
1855         tlen = 0;
1856 #ifdef INET6
1857         if (isipv6)
1858                 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1859 #endif
1860 #if defined(INET) && defined(INET6)
1861         else
1862 #endif
1863 #ifdef INET
1864                 tlen = sizeof (struct tcpiphdr);
1865 #endif
1866         if (port)
1867                 tlen += sizeof (struct udphdr);
1868 #ifdef INVARIANTS
1869         m->m_len = 0;
1870         KASSERT(M_TRAILINGSPACE(m) >= tlen,
1871             ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1872             m, tlen, (long)M_TRAILINGSPACE(m)));
1873 #endif
1874         m->m_len = tlen;
1875         to.to_flags = 0;
1876         if (incl_opts) {
1877                 /* Make sure we have room. */
1878                 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1879                         m->m_next = m_get(M_NOWAIT, MT_DATA);
1880                         if (m->m_next) {
1881                                 optp = mtod(m->m_next, u_char *);
1882                                 optm = m->m_next;
1883                         } else
1884                                 incl_opts = false;
1885                 } else {
1886                         optp = (u_char *) (nth + 1);
1887                         optm = m;
1888                 }
1889         }
1890         if (incl_opts) {
1891                 /* Timestamps. */
1892                 if (tp->t_flags & TF_RCVD_TSTMP) {
1893                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1894                         to.to_tsecr = tp->ts_recent;
1895                         to.to_flags |= TOF_TS;
1896                 }
1897 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1898                 /* TCP-MD5 (RFC2385). */
1899                 if (tp->t_flags & TF_SIGNATURE)
1900                         to.to_flags |= TOF_SIGNATURE;
1901 #endif
1902                 /* Add the options. */
1903                 tlen += optlen = tcp_addoptions(&to, optp);
1904
1905                 /* Update m_len in the correct mbuf. */
1906                 optm->m_len += optlen;
1907         } else
1908                 optlen = 0;
1909 #ifdef INET6
1910         if (isipv6) {
1911                 if (uh) {
1912                         ulen = tlen - sizeof(struct ip6_hdr);
1913                         uh->uh_ulen = htons(ulen);
1914                 }
1915                 ip6->ip6_flow = 0;
1916                 ip6->ip6_vfc = IPV6_VERSION;
1917                 if (port)
1918                         ip6->ip6_nxt = IPPROTO_UDP;
1919                 else
1920                         ip6->ip6_nxt = IPPROTO_TCP;
1921                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1922         }
1923 #endif
1924 #if defined(INET) && defined(INET6)
1925         else
1926 #endif
1927 #ifdef INET
1928         {
1929                 if (uh) {
1930                         ulen = tlen - sizeof(struct ip);
1931                         uh->uh_ulen = htons(ulen);
1932                 }
1933                 ip->ip_len = htons(tlen);
1934                 ip->ip_ttl = V_ip_defttl;
1935                 if (port) {
1936                         ip->ip_p = IPPROTO_UDP;
1937                 } else {
1938                         ip->ip_p = IPPROTO_TCP;
1939                 }
1940                 if (V_path_mtu_discovery)
1941                         ip->ip_off |= htons(IP_DF);
1942         }
1943 #endif
1944         m->m_pkthdr.len = tlen;
1945         m->m_pkthdr.rcvif = NULL;
1946 #ifdef MAC
1947         if (inp != NULL) {
1948                 /*
1949                  * Packet is associated with a socket, so allow the
1950                  * label of the response to reflect the socket label.
1951                  */
1952                 INP_LOCK_ASSERT(inp);
1953                 mac_inpcb_create_mbuf(inp, m);
1954         } else {
1955                 /*
1956                  * Packet is not associated with a socket, so possibly
1957                  * update the label in place.
1958                  */
1959                 mac_netinet_tcp_reply(m);
1960         }
1961 #endif
1962         nth->th_seq = htonl(seq);
1963         nth->th_ack = htonl(ack);
1964         nth->th_x2 = 0;
1965         nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1966         nth->th_flags = flags;
1967         if (tp != NULL)
1968                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1969         else
1970                 nth->th_win = htons((u_short)win);
1971         nth->th_urp = 0;
1972
1973 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1974         if (to.to_flags & TOF_SIGNATURE) {
1975                 if (!TCPMD5_ENABLED() ||
1976                     TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1977                         m_freem(m);
1978                         return;
1979                 }
1980         }
1981 #endif
1982
1983 #ifdef INET6
1984         if (isipv6) {
1985                 if (port) {
1986                         m->m_pkthdr.csum_flags = CSUM_UDP_IPV6;
1987                         m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
1988                         uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0);
1989                         nth->th_sum = 0;
1990                 } else {
1991                         m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1992                         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1993                         nth->th_sum = in6_cksum_pseudo(ip6,
1994                             tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1995                 }
1996                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1997                     NULL, NULL);
1998         }
1999 #endif /* INET6 */
2000 #if defined(INET6) && defined(INET)
2001         else
2002 #endif
2003 #ifdef INET
2004         {
2005                 if (port) {
2006                         uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2007                             htons(ulen + IPPROTO_UDP));
2008                         m->m_pkthdr.csum_flags = CSUM_UDP;
2009                         m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum);
2010                         nth->th_sum = 0;
2011                 } else {
2012                         m->m_pkthdr.csum_flags = CSUM_TCP;
2013                         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
2014                         nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
2015                             htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
2016                 }
2017         }
2018 #endif /* INET */
2019 #ifdef TCPDEBUG
2020         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
2021                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
2022 #endif
2023         TCP_PROBE3(debug__output, tp, th, m);
2024         if (flags & TH_RST)
2025                 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
2026
2027 #ifdef INET6
2028         if (isipv6) {
2029                 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
2030                 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
2031         }
2032 #endif /* INET6 */
2033 #if defined(INET) && defined(INET6)
2034         else
2035 #endif
2036 #ifdef INET
2037         {
2038                 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
2039                 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
2040         }
2041 #endif
2042 }
2043
2044 /*
2045  * Create a new TCP control block, making an
2046  * empty reassembly queue and hooking it to the argument
2047  * protocol control block.  The `inp' parameter must have
2048  * come from the zone allocator set up in tcp_init().
2049  */
2050 struct tcpcb *
2051 tcp_newtcpcb(struct inpcb *inp)
2052 {
2053         struct tcpcb_mem *tm;
2054         struct tcpcb *tp;
2055 #ifdef INET6
2056         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2057 #endif /* INET6 */
2058
2059         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
2060         if (tm == NULL)
2061                 return (NULL);
2062         tp = &tm->tcb;
2063
2064         /* Initialise cc_var struct for this tcpcb. */
2065         tp->ccv = &tm->ccv;
2066         tp->ccv->type = IPPROTO_TCP;
2067         tp->ccv->ccvc.tcp = tp;
2068         rw_rlock(&tcp_function_lock);
2069         tp->t_fb = tcp_func_set_ptr;
2070         refcount_acquire(&tp->t_fb->tfb_refcnt);
2071         rw_runlock(&tcp_function_lock);
2072         /*
2073          * Use the current system default CC algorithm.
2074          */
2075         CC_LIST_RLOCK();
2076         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
2077         CC_ALGO(tp) = CC_DEFAULT();
2078         CC_LIST_RUNLOCK();
2079         /*
2080          * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
2081          * is called.
2082          */
2083         in_pcbref(inp); /* Reference for tcpcb */
2084         tp->t_inpcb = inp;
2085
2086         if (CC_ALGO(tp)->cb_init != NULL)
2087                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
2088                         if (tp->t_fb->tfb_tcp_fb_fini)
2089                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2090                         in_pcbrele_wlocked(inp);
2091                         refcount_release(&tp->t_fb->tfb_refcnt);
2092                         uma_zfree(V_tcpcb_zone, tm);
2093                         return (NULL);
2094                 }
2095
2096 #ifdef TCP_HHOOK
2097         tp->osd = &tm->osd;
2098         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
2099                 if (tp->t_fb->tfb_tcp_fb_fini)
2100                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2101                 in_pcbrele_wlocked(inp);
2102                 refcount_release(&tp->t_fb->tfb_refcnt);
2103                 uma_zfree(V_tcpcb_zone, tm);
2104                 return (NULL);
2105         }
2106 #endif
2107
2108 #ifdef VIMAGE
2109         tp->t_vnet = inp->inp_vnet;
2110 #endif
2111         tp->t_timers = &tm->tt;
2112         TAILQ_INIT(&tp->t_segq);
2113         tp->t_maxseg =
2114 #ifdef INET6
2115                 isipv6 ? V_tcp_v6mssdflt :
2116 #endif /* INET6 */
2117                 V_tcp_mssdflt;
2118
2119         /* Set up our timeouts. */
2120         callout_init(&tp->t_timers->tt_rexmt, 1);
2121         callout_init(&tp->t_timers->tt_persist, 1);
2122         callout_init(&tp->t_timers->tt_keep, 1);
2123         callout_init(&tp->t_timers->tt_2msl, 1);
2124         callout_init(&tp->t_timers->tt_delack, 1);
2125
2126         if (V_tcp_do_rfc1323)
2127                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
2128         if (V_tcp_do_sack)
2129                 tp->t_flags |= TF_SACK_PERMIT;
2130         TAILQ_INIT(&tp->snd_holes);
2131
2132         /*
2133          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
2134          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
2135          * reasonable initial retransmit time.
2136          */
2137         tp->t_srtt = TCPTV_SRTTBASE;
2138         tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
2139         tp->t_rttmin = tcp_rexmit_min;
2140         tp->t_rxtcur = tcp_rexmit_initial;
2141         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2142         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
2143         tp->t_rcvtime = ticks;
2144         /*
2145          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
2146          * because the socket may be bound to an IPv6 wildcard address,
2147          * which may match an IPv4-mapped IPv6 address.
2148          */
2149         inp->inp_ip_ttl = V_ip_defttl;
2150         inp->inp_ppcb = tp;
2151 #ifdef TCPPCAP
2152         /*
2153          * Init the TCP PCAP queues.
2154          */
2155         tcp_pcap_tcpcb_init(tp);
2156 #endif
2157 #ifdef TCP_BLACKBOX
2158         /* Initialize the per-TCPCB log data. */
2159         tcp_log_tcpcbinit(tp);
2160 #endif
2161         tp->t_pacing_rate = -1;
2162         if (tp->t_fb->tfb_tcp_fb_init) {
2163                 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) {
2164                         refcount_release(&tp->t_fb->tfb_refcnt);
2165                         in_pcbrele_wlocked(inp);
2166                         uma_zfree(V_tcpcb_zone, tm);
2167                         return (NULL);
2168                 }
2169         }
2170 #ifdef STATS
2171         if (V_tcp_perconn_stats_enable == 1)
2172                 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0);
2173 #endif
2174         if (V_tcp_do_lrd)
2175                 tp->t_flags |= TF_LRD;
2176         return (tp);            /* XXX */
2177 }
2178
2179 /*
2180  * Switch the congestion control algorithm back to NewReno for any active
2181  * control blocks using an algorithm which is about to go away.
2182  * This ensures the CC framework can allow the unload to proceed without leaving
2183  * any dangling pointers which would trigger a panic.
2184  * Returning non-zero would inform the CC framework that something went wrong
2185  * and it would be unsafe to allow the unload to proceed. However, there is no
2186  * way for this to occur with this implementation so we always return zero.
2187  */
2188 int
2189 tcp_ccalgounload(struct cc_algo *unload_algo)
2190 {
2191         struct cc_algo *tmpalgo;
2192         struct inpcb *inp;
2193         struct tcpcb *tp;
2194         VNET_ITERATOR_DECL(vnet_iter);
2195
2196         /*
2197          * Check all active control blocks across all network stacks and change
2198          * any that are using "unload_algo" back to NewReno. If "unload_algo"
2199          * requires cleanup code to be run, call it.
2200          */
2201         VNET_LIST_RLOCK();
2202         VNET_FOREACH(vnet_iter) {
2203                 CURVNET_SET(vnet_iter);
2204                 INP_INFO_WLOCK(&V_tcbinfo);
2205                 /*
2206                  * New connections already part way through being initialised
2207                  * with the CC algo we're removing will not race with this code
2208                  * because the INP_INFO_WLOCK is held during initialisation. We
2209                  * therefore don't enter the loop below until the connection
2210                  * list has stabilised.
2211                  */
2212                 CK_LIST_FOREACH(inp, &V_tcb, inp_list) {
2213                         INP_WLOCK(inp);
2214                         /* Important to skip tcptw structs. */
2215                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
2216                             (tp = intotcpcb(inp)) != NULL) {
2217                                 /*
2218                                  * By holding INP_WLOCK here, we are assured
2219                                  * that the connection is not currently
2220                                  * executing inside the CC module's functions
2221                                  * i.e. it is safe to make the switch back to
2222                                  * NewReno.
2223                                  */
2224                                 if (CC_ALGO(tp) == unload_algo) {
2225                                         tmpalgo = CC_ALGO(tp);
2226                                         if (tmpalgo->cb_destroy != NULL)
2227                                                 tmpalgo->cb_destroy(tp->ccv);
2228                                         CC_DATA(tp) = NULL;
2229                                         /*
2230                                          * NewReno may allocate memory on
2231                                          * demand for certain stateful
2232                                          * configuration as needed, but is
2233                                          * coded to never fail on memory
2234                                          * allocation failure so it is a safe
2235                                          * fallback.
2236                                          */
2237                                         CC_ALGO(tp) = &newreno_cc_algo;
2238                                 }
2239                         }
2240                         INP_WUNLOCK(inp);
2241                 }
2242                 INP_INFO_WUNLOCK(&V_tcbinfo);
2243                 CURVNET_RESTORE();
2244         }
2245         VNET_LIST_RUNLOCK();
2246
2247         return (0);
2248 }
2249
2250 /*
2251  * Drop a TCP connection, reporting
2252  * the specified error.  If connection is synchronized,
2253  * then send a RST to peer.
2254  */
2255 struct tcpcb *
2256 tcp_drop(struct tcpcb *tp, int errno)
2257 {
2258         struct socket *so = tp->t_inpcb->inp_socket;
2259
2260         NET_EPOCH_ASSERT();
2261         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2262         INP_WLOCK_ASSERT(tp->t_inpcb);
2263
2264         if (TCPS_HAVERCVDSYN(tp->t_state)) {
2265                 tcp_state_change(tp, TCPS_CLOSED);
2266                 (void) tp->t_fb->tfb_tcp_output(tp);
2267                 TCPSTAT_INC(tcps_drops);
2268         } else
2269                 TCPSTAT_INC(tcps_conndrops);
2270         if (errno == ETIMEDOUT && tp->t_softerror)
2271                 errno = tp->t_softerror;
2272         so->so_error = errno;
2273         return (tcp_close(tp));
2274 }
2275
2276 void
2277 tcp_discardcb(struct tcpcb *tp)
2278 {
2279         struct inpcb *inp = tp->t_inpcb;
2280         struct socket *so = inp->inp_socket;
2281 #ifdef INET6
2282         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2283 #endif /* INET6 */
2284         int released __unused;
2285
2286         INP_WLOCK_ASSERT(inp);
2287
2288         /*
2289          * Make sure that all of our timers are stopped before we delete the
2290          * PCB.
2291          *
2292          * If stopping a timer fails, we schedule a discard function in same
2293          * callout, and the last discard function called will take care of
2294          * deleting the tcpcb.
2295          */
2296         tp->t_timers->tt_draincnt = 0;
2297         tcp_timer_stop(tp, TT_REXMT);
2298         tcp_timer_stop(tp, TT_PERSIST);
2299         tcp_timer_stop(tp, TT_KEEP);
2300         tcp_timer_stop(tp, TT_2MSL);
2301         tcp_timer_stop(tp, TT_DELACK);
2302         if (tp->t_fb->tfb_tcp_timer_stop_all) {
2303                 /*
2304                  * Call the stop-all function of the methods,
2305                  * this function should call the tcp_timer_stop()
2306                  * method with each of the function specific timeouts.
2307                  * That stop will be called via the tfb_tcp_timer_stop()
2308                  * which should use the async drain function of the
2309                  * callout system (see tcp_var.h).
2310                  */
2311                 tp->t_fb->tfb_tcp_timer_stop_all(tp);
2312         }
2313
2314         /* free the reassembly queue, if any */
2315         tcp_reass_flush(tp);
2316
2317 #ifdef TCP_OFFLOAD
2318         /* Disconnect offload device, if any. */
2319         if (tp->t_flags & TF_TOE)
2320                 tcp_offload_detach(tp);
2321 #endif
2322
2323         tcp_free_sackholes(tp);
2324
2325 #ifdef TCPPCAP
2326         /* Free the TCP PCAP queues. */
2327         tcp_pcap_drain(&(tp->t_inpkts));
2328         tcp_pcap_drain(&(tp->t_outpkts));
2329 #endif
2330
2331         /* Allow the CC algorithm to clean up after itself. */
2332         if (CC_ALGO(tp)->cb_destroy != NULL)
2333                 CC_ALGO(tp)->cb_destroy(tp->ccv);
2334         CC_DATA(tp) = NULL;
2335
2336 #ifdef TCP_HHOOK
2337         khelp_destroy_osd(tp->osd);
2338 #endif
2339 #ifdef STATS
2340         stats_blob_destroy(tp->t_stats);
2341 #endif
2342
2343         CC_ALGO(tp) = NULL;
2344         inp->inp_ppcb = NULL;
2345         if (tp->t_timers->tt_draincnt == 0) {
2346                 /* We own the last reference on tcpcb, let's free it. */
2347 #ifdef TCP_BLACKBOX
2348                 tcp_log_tcpcbfini(tp);
2349 #endif
2350                 TCPSTATES_DEC(tp->t_state);
2351                 if (tp->t_fb->tfb_tcp_fb_fini)
2352                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2353
2354                 /*
2355                  * If we got enough samples through the srtt filter,
2356                  * save the rtt and rttvar in the routing entry.
2357                  * 'Enough' is arbitrarily defined as 4 rtt samples.
2358                  * 4 samples is enough for the srtt filter to converge
2359                  * to within enough % of the correct value; fewer samples
2360                  * and we could save a bogus rtt. The danger is not high
2361                  * as tcp quickly recovers from everything.
2362                  * XXX: Works very well but needs some more statistics!
2363                  *
2364                  * XXXRRS: Updating must be after the stack fini() since
2365                  * that may be converting some internal representation of
2366                  * say srtt etc into the general one used by other stacks.
2367                  * Lets also at least protect against the so being NULL
2368                  * as RW stated below.
2369                  */
2370                 if ((tp->t_rttupdated >= 4) && (so != NULL)) {
2371                         struct hc_metrics_lite metrics;
2372                         uint32_t ssthresh;
2373
2374                         bzero(&metrics, sizeof(metrics));
2375                         /*
2376                          * Update the ssthresh always when the conditions below
2377                          * are satisfied. This gives us better new start value
2378                          * for the congestion avoidance for new connections.
2379                          * ssthresh is only set if packet loss occurred on a session.
2380                          *
2381                          * XXXRW: 'so' may be NULL here, and/or socket buffer may be
2382                          * being torn down.  Ideally this code would not use 'so'.
2383                          */
2384                         ssthresh = tp->snd_ssthresh;
2385                         if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
2386                                 /*
2387                                  * convert the limit from user data bytes to
2388                                  * packets then to packet data bytes.
2389                                  */
2390                                 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
2391                                 if (ssthresh < 2)
2392                                         ssthresh = 2;
2393                                 ssthresh *= (tp->t_maxseg +
2394 #ifdef INET6
2395                                              (isipv6 ? sizeof (struct ip6_hdr) +
2396                                               sizeof (struct tcphdr) :
2397 #endif
2398                                               sizeof (struct tcpiphdr)
2399 #ifdef INET6
2400                                                      )
2401 #endif
2402                                         );
2403                         } else
2404                                 ssthresh = 0;
2405                         metrics.rmx_ssthresh = ssthresh;
2406
2407                         metrics.rmx_rtt = tp->t_srtt;
2408                         metrics.rmx_rttvar = tp->t_rttvar;
2409                         metrics.rmx_cwnd = tp->snd_cwnd;
2410                         metrics.rmx_sendpipe = 0;
2411                         metrics.rmx_recvpipe = 0;
2412
2413                         tcp_hc_update(&inp->inp_inc, &metrics);
2414                 }
2415                 refcount_release(&tp->t_fb->tfb_refcnt);
2416                 tp->t_inpcb = NULL;
2417                 uma_zfree(V_tcpcb_zone, tp);
2418                 released = in_pcbrele_wlocked(inp);
2419                 KASSERT(!released, ("%s: inp %p should not have been released "
2420                         "here", __func__, inp));
2421         }
2422 }
2423
2424 void
2425 tcp_timer_discard(void *ptp)
2426 {
2427         struct inpcb *inp;
2428         struct tcpcb *tp;
2429         struct epoch_tracker et;
2430
2431         tp = (struct tcpcb *)ptp;
2432         CURVNET_SET(tp->t_vnet);
2433         NET_EPOCH_ENTER(et);
2434         inp = tp->t_inpcb;
2435         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
2436                 __func__, tp));
2437         INP_WLOCK(inp);
2438         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
2439                 ("%s: tcpcb has to be stopped here", __func__));
2440         tp->t_timers->tt_draincnt--;
2441         if (tp->t_timers->tt_draincnt == 0) {
2442                 /* We own the last reference on this tcpcb, let's free it. */
2443 #ifdef TCP_BLACKBOX
2444                 tcp_log_tcpcbfini(tp);
2445 #endif
2446                 TCPSTATES_DEC(tp->t_state);
2447                 if (tp->t_fb->tfb_tcp_fb_fini)
2448                         (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1);
2449                 refcount_release(&tp->t_fb->tfb_refcnt);
2450                 tp->t_inpcb = NULL;
2451                 uma_zfree(V_tcpcb_zone, tp);
2452                 if (in_pcbrele_wlocked(inp)) {
2453                         NET_EPOCH_EXIT(et);
2454                         CURVNET_RESTORE();
2455                         return;
2456                 }
2457         }
2458         INP_WUNLOCK(inp);
2459         NET_EPOCH_EXIT(et);
2460         CURVNET_RESTORE();
2461 }
2462
2463 /*
2464  * Attempt to close a TCP control block, marking it as dropped, and freeing
2465  * the socket if we hold the only reference.
2466  */
2467 struct tcpcb *
2468 tcp_close(struct tcpcb *tp)
2469 {
2470         struct inpcb *inp = tp->t_inpcb;
2471         struct socket *so;
2472
2473         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2474         INP_WLOCK_ASSERT(inp);
2475
2476 #ifdef TCP_OFFLOAD
2477         if (tp->t_state == TCPS_LISTEN)
2478                 tcp_offload_listen_stop(tp);
2479 #endif
2480         /*
2481          * This releases the TFO pending counter resource for TFO listen
2482          * sockets as well as passively-created TFO sockets that transition
2483          * from SYN_RECEIVED to CLOSED.
2484          */
2485         if (tp->t_tfo_pending) {
2486                 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2487                 tp->t_tfo_pending = NULL;
2488         }
2489         in_pcbdrop(inp);
2490         TCPSTAT_INC(tcps_closed);
2491         if (tp->t_state != TCPS_CLOSED)
2492                 tcp_state_change(tp, TCPS_CLOSED);
2493         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
2494         so = inp->inp_socket;
2495         soisdisconnected(so);
2496         if (inp->inp_flags & INP_SOCKREF) {
2497                 KASSERT(so->so_state & SS_PROTOREF,
2498                     ("tcp_close: !SS_PROTOREF"));
2499                 inp->inp_flags &= ~INP_SOCKREF;
2500                 INP_WUNLOCK(inp);
2501                 SOCK_LOCK(so);
2502                 so->so_state &= ~SS_PROTOREF;
2503                 sofree(so);
2504                 return (NULL);
2505         }
2506         return (tp);
2507 }
2508
2509 void
2510 tcp_drain(void)
2511 {
2512         VNET_ITERATOR_DECL(vnet_iter);
2513
2514         if (!do_tcpdrain)
2515                 return;
2516
2517         VNET_LIST_RLOCK_NOSLEEP();
2518         VNET_FOREACH(vnet_iter) {
2519                 CURVNET_SET(vnet_iter);
2520                 struct inpcb *inpb;
2521                 struct tcpcb *tcpb;
2522
2523         /*
2524          * Walk the tcpbs, if existing, and flush the reassembly queue,
2525          * if there is one...
2526          * XXX: The "Net/3" implementation doesn't imply that the TCP
2527          *      reassembly queue should be flushed, but in a situation
2528          *      where we're really low on mbufs, this is potentially
2529          *      useful.
2530          */
2531                 INP_INFO_WLOCK(&V_tcbinfo);
2532                 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
2533                         INP_WLOCK(inpb);
2534                         if (inpb->inp_flags & INP_TIMEWAIT) {
2535                                 INP_WUNLOCK(inpb);
2536                                 continue;
2537                         }
2538                         if ((tcpb = intotcpcb(inpb)) != NULL) {
2539                                 tcp_reass_flush(tcpb);
2540                                 tcp_clean_sackreport(tcpb);
2541 #ifdef TCP_BLACKBOX
2542                                 tcp_log_drain(tcpb);
2543 #endif
2544 #ifdef TCPPCAP
2545                                 if (tcp_pcap_aggressive_free) {
2546                                         /* Free the TCP PCAP queues. */
2547                                         tcp_pcap_drain(&(tcpb->t_inpkts));
2548                                         tcp_pcap_drain(&(tcpb->t_outpkts));
2549                                 }
2550 #endif
2551                         }
2552                         INP_WUNLOCK(inpb);
2553                 }
2554                 INP_INFO_WUNLOCK(&V_tcbinfo);
2555                 CURVNET_RESTORE();
2556         }
2557         VNET_LIST_RUNLOCK_NOSLEEP();
2558 }
2559
2560 /*
2561  * Notify a tcp user of an asynchronous error;
2562  * store error as soft error, but wake up user
2563  * (for now, won't do anything until can select for soft error).
2564  *
2565  * Do not wake up user since there currently is no mechanism for
2566  * reporting soft errors (yet - a kqueue filter may be added).
2567  */
2568 static struct inpcb *
2569 tcp_notify(struct inpcb *inp, int error)
2570 {
2571         struct tcpcb *tp;
2572
2573         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2574         INP_WLOCK_ASSERT(inp);
2575
2576         if ((inp->inp_flags & INP_TIMEWAIT) ||
2577             (inp->inp_flags & INP_DROPPED))
2578                 return (inp);
2579
2580         tp = intotcpcb(inp);
2581         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
2582
2583         /*
2584          * Ignore some errors if we are hooked up.
2585          * If connection hasn't completed, has retransmitted several times,
2586          * and receives a second error, give up now.  This is better
2587          * than waiting a long time to establish a connection that
2588          * can never complete.
2589          */
2590         if (tp->t_state == TCPS_ESTABLISHED &&
2591             (error == EHOSTUNREACH || error == ENETUNREACH ||
2592              error == EHOSTDOWN)) {
2593                 if (inp->inp_route.ro_nh) {
2594                         NH_FREE(inp->inp_route.ro_nh);
2595                         inp->inp_route.ro_nh = (struct nhop_object *)NULL;
2596                 }
2597                 return (inp);
2598         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
2599             tp->t_softerror) {
2600                 tp = tcp_drop(tp, error);
2601                 if (tp != NULL)
2602                         return (inp);
2603                 else
2604                         return (NULL);
2605         } else {
2606                 tp->t_softerror = error;
2607                 return (inp);
2608         }
2609 #if 0
2610         wakeup( &so->so_timeo);
2611         sorwakeup(so);
2612         sowwakeup(so);
2613 #endif
2614 }
2615
2616 static int
2617 tcp_pcblist(SYSCTL_HANDLER_ARGS)
2618 {
2619         struct epoch_tracker et;
2620         struct inpcb *inp;
2621         struct xinpgen xig;
2622         int error;
2623
2624         if (req->newptr != NULL)
2625                 return (EPERM);
2626
2627         if (req->oldptr == NULL) {
2628                 int n;
2629
2630                 n = V_tcbinfo.ipi_count +
2631                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2632                 n += imax(n / 8, 10);
2633                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
2634                 return (0);
2635         }
2636
2637         if ((error = sysctl_wire_old_buffer(req, 0)) != 0)
2638                 return (error);
2639
2640         bzero(&xig, sizeof(xig));
2641         xig.xig_len = sizeof xig;
2642         xig.xig_count = V_tcbinfo.ipi_count +
2643             counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2644         xig.xig_gen = V_tcbinfo.ipi_gencnt;
2645         xig.xig_sogen = so_gencnt;
2646         error = SYSCTL_OUT(req, &xig, sizeof xig);
2647         if (error)
2648                 return (error);
2649
2650         error = syncache_pcblist(req);
2651         if (error)
2652                 return (error);
2653
2654         NET_EPOCH_ENTER(et);
2655         for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead);
2656             inp != NULL;
2657             inp = CK_LIST_NEXT(inp, inp_list)) {
2658                 INP_RLOCK(inp);
2659                 if (inp->inp_gencnt <= xig.xig_gen) {
2660                         int crerr;
2661
2662                         /*
2663                          * XXX: This use of cr_cansee(), introduced with
2664                          * TCP state changes, is not quite right, but for
2665                          * now, better than nothing.
2666                          */
2667                         if (inp->inp_flags & INP_TIMEWAIT) {
2668                                 if (intotw(inp) != NULL)
2669                                         crerr = cr_cansee(req->td->td_ucred,
2670                                             intotw(inp)->tw_cred);
2671                                 else
2672                                         crerr = EINVAL; /* Skip this inp. */
2673                         } else
2674                                 crerr = cr_canseeinpcb(req->td->td_ucred, inp);
2675                         if (crerr == 0) {
2676                                 struct xtcpcb xt;
2677
2678                                 tcp_inptoxtp(inp, &xt);
2679                                 INP_RUNLOCK(inp);
2680                                 error = SYSCTL_OUT(req, &xt, sizeof xt);
2681                                 if (error)
2682                                         break;
2683                                 else
2684                                         continue;
2685                         }
2686                 }
2687                 INP_RUNLOCK(inp);
2688         }
2689         NET_EPOCH_EXIT(et);
2690
2691         if (!error) {
2692                 /*
2693                  * Give the user an updated idea of our state.
2694                  * If the generation differs from what we told
2695                  * her before, she knows that something happened
2696                  * while we were processing this request, and it
2697                  * might be necessary to retry.
2698                  */
2699                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
2700                 xig.xig_sogen = so_gencnt;
2701                 xig.xig_count = V_tcbinfo.ipi_count +
2702                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
2703                 error = SYSCTL_OUT(req, &xig, sizeof xig);
2704         }
2705
2706         return (error);
2707 }
2708
2709 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2710     CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT,
2711     NULL, 0, tcp_pcblist, "S,xtcpcb",
2712     "List of active TCP connections");
2713
2714 #ifdef INET
2715 static int
2716 tcp_getcred(SYSCTL_HANDLER_ARGS)
2717 {
2718         struct xucred xuc;
2719         struct sockaddr_in addrs[2];
2720         struct epoch_tracker et;
2721         struct inpcb *inp;
2722         int error;
2723
2724         error = priv_check(req->td, PRIV_NETINET_GETCRED);
2725         if (error)
2726                 return (error);
2727         error = SYSCTL_IN(req, addrs, sizeof(addrs));
2728         if (error)
2729                 return (error);
2730         NET_EPOCH_ENTER(et);
2731         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
2732             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
2733         NET_EPOCH_EXIT(et);
2734         if (inp != NULL) {
2735                 if (inp->inp_socket == NULL)
2736                         error = ENOENT;
2737                 if (error == 0)
2738                         error = cr_canseeinpcb(req->td->td_ucred, inp);
2739                 if (error == 0)
2740                         cru2x(inp->inp_cred, &xuc);
2741                 INP_RUNLOCK(inp);
2742         } else
2743                 error = ENOENT;
2744         if (error == 0)
2745                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2746         return (error);
2747 }
2748
2749 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
2750     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2751     0, 0, tcp_getcred, "S,xucred",
2752     "Get the xucred of a TCP connection");
2753 #endif /* INET */
2754
2755 #ifdef INET6
2756 static int
2757 tcp6_getcred(SYSCTL_HANDLER_ARGS)
2758 {
2759         struct epoch_tracker et;
2760         struct xucred xuc;
2761         struct sockaddr_in6 addrs[2];
2762         struct inpcb *inp;
2763         int error;
2764 #ifdef INET
2765         int mapped = 0;
2766 #endif
2767
2768         error = priv_check(req->td, PRIV_NETINET_GETCRED);
2769         if (error)
2770                 return (error);
2771         error = SYSCTL_IN(req, addrs, sizeof(addrs));
2772         if (error)
2773                 return (error);
2774         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
2775             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
2776                 return (error);
2777         }
2778         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
2779 #ifdef INET
2780                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
2781                         mapped = 1;
2782                 else
2783 #endif
2784                         return (EINVAL);
2785         }
2786
2787         NET_EPOCH_ENTER(et);
2788 #ifdef INET
2789         if (mapped == 1)
2790                 inp = in_pcblookup(&V_tcbinfo,
2791                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
2792                         addrs[1].sin6_port,
2793                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
2794                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
2795         else
2796 #endif
2797                 inp = in6_pcblookup(&V_tcbinfo,
2798                         &addrs[1].sin6_addr, addrs[1].sin6_port,
2799                         &addrs[0].sin6_addr, addrs[0].sin6_port,
2800                         INPLOOKUP_RLOCKPCB, NULL);
2801         NET_EPOCH_EXIT(et);
2802         if (inp != NULL) {
2803                 if (inp->inp_socket == NULL)
2804                         error = ENOENT;
2805                 if (error == 0)
2806                         error = cr_canseeinpcb(req->td->td_ucred, inp);
2807                 if (error == 0)
2808                         cru2x(inp->inp_cred, &xuc);
2809                 INP_RUNLOCK(inp);
2810         } else
2811                 error = ENOENT;
2812         if (error == 0)
2813                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
2814         return (error);
2815 }
2816
2817 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
2818     CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT,
2819     0, 0, tcp6_getcred, "S,xucred",
2820     "Get the xucred of a TCP6 connection");
2821 #endif /* INET6 */
2822
2823 #ifdef INET
2824 /* Path MTU to try next when a fragmentation-needed message is received. */
2825 static inline int
2826 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip)
2827 {
2828         int mtu = ntohs(icp->icmp_nextmtu);
2829
2830         /* If no alternative MTU was proposed, try the next smaller one. */
2831         if (!mtu)
2832                 mtu = ip_next_mtu(ntohs(ip->ip_len), 1);
2833         if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr))
2834                 mtu = V_tcp_minmss + sizeof(struct tcpiphdr);
2835
2836         return (mtu);
2837 }
2838
2839 static void
2840 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port)
2841 {
2842         struct ip *ip = vip;
2843         struct tcphdr *th;
2844         struct in_addr faddr;
2845         struct inpcb *inp;
2846         struct tcpcb *tp;
2847         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2848         struct icmp *icp;
2849         struct in_conninfo inc;
2850         tcp_seq icmp_tcp_seq;
2851         int mtu;
2852
2853         faddr = ((struct sockaddr_in *)sa)->sin_addr;
2854         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2855                 return;
2856
2857         if (cmd == PRC_MSGSIZE)
2858                 notify = tcp_mtudisc_notify;
2859         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2860                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2861                 cmd == PRC_TIMXCEED_INTRANS) && ip)
2862                 notify = tcp_drop_syn_sent;
2863
2864         /*
2865          * Hostdead is ugly because it goes linearly through all PCBs.
2866          * XXX: We never get this from ICMP, otherwise it makes an
2867          * excellent DoS attack on machines with many connections.
2868          */
2869         else if (cmd == PRC_HOSTDEAD)
2870                 ip = NULL;
2871         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
2872                 return;
2873
2874         if (ip == NULL) {
2875                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
2876                 return;
2877         }
2878
2879         icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2880         th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
2881         inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
2882             th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
2883         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2884                 /* signal EHOSTDOWN, as it flushes the cached route */
2885                 inp = (*notify)(inp, EHOSTDOWN);
2886                 goto out;
2887         }
2888         icmp_tcp_seq = th->th_seq;
2889         if (inp != NULL)  {
2890                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2891                     !(inp->inp_flags & INP_DROPPED) &&
2892                     !(inp->inp_socket == NULL)) {
2893                         tp = intotcpcb(inp);
2894 #ifdef TCP_OFFLOAD
2895                         if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
2896                                 /*
2897                                  * MTU discovery for offloaded connections.  Let
2898                                  * the TOE driver verify seq# and process it.
2899                                  */
2900                                 mtu = tcp_next_pmtu(icp, ip);
2901                                 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
2902                                 goto out;
2903                         }
2904 #endif
2905                         if (tp->t_port != port) {
2906                                 goto out;
2907                         }
2908                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2909                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2910                                 if (cmd == PRC_MSGSIZE) {
2911                                         /*
2912                                          * MTU discovery: we got a needfrag and
2913                                          * will potentially try a lower MTU.
2914                                          */
2915                                         mtu = tcp_next_pmtu(icp, ip);
2916
2917                                         /*
2918                                          * Only process the offered MTU if it
2919                                          * is smaller than the current one.
2920                                          */
2921                                         if (mtu < tp->t_maxseg +
2922                                             sizeof(struct tcpiphdr)) {
2923                                                 bzero(&inc, sizeof(inc));
2924                                                 inc.inc_faddr = faddr;
2925                                                 inc.inc_fibnum =
2926                                                     inp->inp_inc.inc_fibnum;
2927                                                 tcp_hc_updatemtu(&inc, mtu);
2928                                                 tcp_mtudisc(inp, mtu);
2929                                         }
2930                                 } else
2931                                         inp = (*notify)(inp,
2932                                             inetctlerrmap[cmd]);
2933                         }
2934                 }
2935         } else {
2936                 bzero(&inc, sizeof(inc));
2937                 inc.inc_fport = th->th_dport;
2938                 inc.inc_lport = th->th_sport;
2939                 inc.inc_faddr = faddr;
2940                 inc.inc_laddr = ip->ip_src;
2941                 syncache_unreach(&inc, icmp_tcp_seq, port);
2942         }
2943 out:
2944         if (inp != NULL)
2945                 INP_WUNLOCK(inp);
2946 }
2947
2948 void
2949 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
2950 {
2951         tcp_ctlinput_with_port(cmd, sa, vip, htons(0));
2952 }
2953
2954 void
2955 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused)
2956 {
2957         /* Its a tunneled TCP over UDP icmp */
2958         struct ip *outer_ip, *inner_ip;
2959         struct icmp *icmp;
2960         struct udphdr *udp;
2961         struct tcphdr *th, ttemp;
2962         int i_hlen, o_len;
2963         uint16_t port;
2964
2965         inner_ip = (struct ip *)vip;
2966         icmp = (struct icmp *)((caddr_t)inner_ip -
2967             (sizeof(struct icmp) - sizeof(struct ip)));
2968         outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip));
2969         i_hlen = inner_ip->ip_hl << 2;
2970         o_len = ntohs(outer_ip->ip_len);
2971         if (o_len <
2972             (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) {
2973                 /* Not enough data present */
2974                 return;
2975         }
2976         /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */
2977         udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen);
2978         if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
2979                 return;
2980         }
2981         port = udp->uh_dport;
2982         th = (struct tcphdr *)(udp + 1);
2983         memcpy(&ttemp, th, sizeof(struct tcphdr));
2984         memcpy(udp, &ttemp, sizeof(struct tcphdr));
2985         /* Now adjust down the size of the outer IP header */
2986         o_len -= sizeof(struct udphdr);
2987         outer_ip->ip_len = htons(o_len);
2988         /* Now call in to the normal handling code */
2989         tcp_ctlinput_with_port(cmd, sa, vip, port);
2990 }
2991 #endif /* INET */
2992
2993 #ifdef INET6
2994 static inline int
2995 tcp6_next_pmtu(const struct icmp6_hdr *icmp6)
2996 {
2997         int mtu = ntohl(icmp6->icmp6_mtu);
2998
2999         /*
3000          * If no alternative MTU was proposed, or the proposed MTU was too
3001          * small, set to the min.
3002          */
3003         if (mtu < IPV6_MMTU)
3004                 mtu = IPV6_MMTU - 8;    /* XXXNP: what is the adjustment for? */
3005         return (mtu);
3006 }
3007
3008 static void
3009 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port)
3010 {
3011         struct in6_addr *dst;
3012         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
3013         struct ip6_hdr *ip6;
3014         struct mbuf *m;
3015         struct inpcb *inp;
3016         struct tcpcb *tp;
3017         struct icmp6_hdr *icmp6;
3018         struct ip6ctlparam *ip6cp = NULL;
3019         const struct sockaddr_in6 *sa6_src = NULL;
3020         struct in_conninfo inc;
3021         struct tcp_ports {
3022                 uint16_t th_sport;
3023                 uint16_t th_dport;
3024         } t_ports;
3025         tcp_seq icmp_tcp_seq;
3026         unsigned int mtu;
3027         unsigned int off;
3028
3029         if (sa->sa_family != AF_INET6 ||
3030             sa->sa_len != sizeof(struct sockaddr_in6))
3031                 return;
3032
3033         /* if the parameter is from icmp6, decode it. */
3034         if (d != NULL) {
3035                 ip6cp = (struct ip6ctlparam *)d;
3036                 icmp6 = ip6cp->ip6c_icmp6;
3037                 m = ip6cp->ip6c_m;
3038                 ip6 = ip6cp->ip6c_ip6;
3039                 off = ip6cp->ip6c_off;
3040                 sa6_src = ip6cp->ip6c_src;
3041                 dst = ip6cp->ip6c_finaldst;
3042         } else {
3043                 m = NULL;
3044                 ip6 = NULL;
3045                 off = 0;        /* fool gcc */
3046                 sa6_src = &sa6_any;
3047                 dst = NULL;
3048         }
3049
3050         if (cmd == PRC_MSGSIZE)
3051                 notify = tcp_mtudisc_notify;
3052         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
3053                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
3054                 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
3055                 notify = tcp_drop_syn_sent;
3056
3057         /*
3058          * Hostdead is ugly because it goes linearly through all PCBs.
3059          * XXX: We never get this from ICMP, otherwise it makes an
3060          * excellent DoS attack on machines with many connections.
3061          */
3062         else if (cmd == PRC_HOSTDEAD)
3063                 ip6 = NULL;
3064         else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
3065                 return;
3066
3067         if (ip6 == NULL) {
3068                 in6_pcbnotify(&V_tcbinfo, sa, 0,
3069                               (const struct sockaddr *)sa6_src,
3070                               0, cmd, NULL, notify);
3071                 return;
3072         }
3073
3074         /* Check if we can safely get the ports from the tcp hdr */
3075         if (m == NULL ||
3076             (m->m_pkthdr.len <
3077                 (int32_t) (off + sizeof(struct tcp_ports)))) {
3078                 return;
3079         }
3080         bzero(&t_ports, sizeof(struct tcp_ports));
3081         m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
3082         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
3083             &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
3084         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
3085                 /* signal EHOSTDOWN, as it flushes the cached route */
3086                 inp = (*notify)(inp, EHOSTDOWN);
3087                 goto out;
3088         }
3089         off += sizeof(struct tcp_ports);
3090         if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
3091                 goto out;
3092         }
3093         m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
3094         if (inp != NULL)  {
3095                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
3096                     !(inp->inp_flags & INP_DROPPED) &&
3097                     !(inp->inp_socket == NULL)) {
3098                         tp = intotcpcb(inp);
3099 #ifdef TCP_OFFLOAD
3100                         if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) {
3101                                 /* MTU discovery for offloaded connections. */
3102                                 mtu = tcp6_next_pmtu(icmp6);
3103                                 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu);
3104                                 goto out;
3105                         }
3106 #endif
3107                         if (tp->t_port != port) {
3108                                 goto out;
3109                         }
3110                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
3111                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
3112                                 if (cmd == PRC_MSGSIZE) {
3113                                         /*
3114                                          * MTU discovery:
3115                                          * If we got a needfrag set the MTU
3116                                          * in the route to the suggested new
3117                                          * value (if given) and then notify.
3118                                          */
3119                                         mtu = tcp6_next_pmtu(icmp6);
3120
3121                                         bzero(&inc, sizeof(inc));
3122                                         inc.inc_fibnum = M_GETFIB(m);
3123                                         inc.inc_flags |= INC_ISIPV6;
3124                                         inc.inc6_faddr = *dst;
3125                                         if (in6_setscope(&inc.inc6_faddr,
3126                                                 m->m_pkthdr.rcvif, NULL))
3127                                                 goto out;
3128                                         /*
3129                                          * Only process the offered MTU if it
3130                                          * is smaller than the current one.
3131                                          */
3132                                         if (mtu < tp->t_maxseg +
3133                                             sizeof (struct tcphdr) +
3134                                             sizeof (struct ip6_hdr)) {
3135                                                 tcp_hc_updatemtu(&inc, mtu);
3136                                                 tcp_mtudisc(inp, mtu);
3137                                                 ICMP6STAT_INC(icp6s_pmtuchg);
3138                                         }
3139                                 } else
3140                                         inp = (*notify)(inp,
3141                                             inet6ctlerrmap[cmd]);
3142                         }
3143                 }
3144         } else {
3145                 bzero(&inc, sizeof(inc));
3146                 inc.inc_fibnum = M_GETFIB(m);
3147                 inc.inc_flags |= INC_ISIPV6;
3148                 inc.inc_fport = t_ports.th_dport;
3149                 inc.inc_lport = t_ports.th_sport;
3150                 inc.inc6_faddr = *dst;
3151                 inc.inc6_laddr = ip6->ip6_src;
3152                 syncache_unreach(&inc, icmp_tcp_seq, port);
3153         }
3154 out:
3155         if (inp != NULL)
3156                 INP_WUNLOCK(inp);
3157 }
3158
3159 void
3160 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
3161 {
3162         tcp6_ctlinput_with_port(cmd, sa, d, htons(0));
3163 }
3164
3165 void
3166 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused)
3167 {
3168         struct ip6ctlparam *ip6cp;
3169         struct mbuf *m;
3170         struct udphdr *udp;
3171         uint16_t port;
3172
3173         ip6cp = (struct ip6ctlparam *)d;
3174         m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL);
3175         if (m == NULL) {
3176                 return;
3177         }
3178         udp = mtod(m, struct udphdr *);
3179         if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) {
3180                 return;
3181         }
3182         port = udp->uh_dport;
3183         m_adj(m, sizeof(struct udphdr));
3184         if ((m->m_flags & M_PKTHDR) == 0) {
3185                 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr);
3186         }
3187         /* Now call in to the normal handling code */
3188         tcp6_ctlinput_with_port(cmd, sa, d, port);
3189 }
3190
3191 #endif /* INET6 */
3192
3193 static uint32_t
3194 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len)
3195 {
3196         SIPHASH_CTX ctx;
3197         uint32_t hash[2];
3198
3199         KASSERT(len >= SIPHASH_KEY_LENGTH,
3200             ("%s: keylen %u too short ", __func__, len));
3201         SipHash24_Init(&ctx);
3202         SipHash_SetKey(&ctx, (uint8_t *)key);
3203         SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
3204         SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
3205         switch (inc->inc_flags & INC_ISIPV6) {
3206 #ifdef INET
3207         case 0:
3208                 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
3209                 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
3210                 break;
3211 #endif
3212 #ifdef INET6
3213         case INC_ISIPV6:
3214                 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
3215                 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
3216                 break;
3217 #endif
3218         }
3219         SipHash_Final((uint8_t *)hash, &ctx);
3220
3221         return (hash[0] ^ hash[1]);
3222 }
3223
3224 uint32_t
3225 tcp_new_ts_offset(struct in_conninfo *inc)
3226 {
3227         struct in_conninfo inc_store, *local_inc;
3228
3229         if (!V_tcp_ts_offset_per_conn) {
3230                 memcpy(&inc_store, inc, sizeof(struct in_conninfo));
3231                 inc_store.inc_lport = 0;
3232                 inc_store.inc_fport = 0;
3233                 local_inc = &inc_store;
3234         } else {
3235                 local_inc = inc;
3236         }
3237         return (tcp_keyed_hash(local_inc, V_ts_offset_secret,
3238             sizeof(V_ts_offset_secret)));
3239 }
3240
3241 /*
3242  * Following is where TCP initial sequence number generation occurs.
3243  *
3244  * There are two places where we must use initial sequence numbers:
3245  * 1.  In SYN-ACK packets.
3246  * 2.  In SYN packets.
3247  *
3248  * All ISNs for SYN-ACK packets are generated by the syncache.  See
3249  * tcp_syncache.c for details.
3250  *
3251  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
3252  * depends on this property.  In addition, these ISNs should be
3253  * unguessable so as to prevent connection hijacking.  To satisfy
3254  * the requirements of this situation, the algorithm outlined in
3255  * RFC 1948 is used, with only small modifications.
3256  *
3257  * Implementation details:
3258  *
3259  * Time is based off the system timer, and is corrected so that it
3260  * increases by one megabyte per second.  This allows for proper
3261  * recycling on high speed LANs while still leaving over an hour
3262  * before rollover.
3263  *
3264  * As reading the *exact* system time is too expensive to be done
3265  * whenever setting up a TCP connection, we increment the time
3266  * offset in two ways.  First, a small random positive increment
3267  * is added to isn_offset for each connection that is set up.
3268  * Second, the function tcp_isn_tick fires once per clock tick
3269  * and increments isn_offset as necessary so that sequence numbers
3270  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
3271  * random positive increments serve only to ensure that the same
3272  * exact sequence number is never sent out twice (as could otherwise
3273  * happen when a port is recycled in less than the system tick
3274  * interval.)
3275  *
3276  * net.inet.tcp.isn_reseed_interval controls the number of seconds
3277  * between seeding of isn_secret.  This is normally set to zero,
3278  * as reseeding should not be necessary.
3279  *
3280  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
3281  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
3282  * general, this means holding an exclusive (write) lock.
3283  */
3284
3285 #define ISN_BYTES_PER_SECOND 1048576
3286 #define ISN_STATIC_INCREMENT 4096
3287 #define ISN_RANDOM_INCREMENT (4096 - 1)
3288 #define ISN_SECRET_LENGTH    SIPHASH_KEY_LENGTH
3289
3290 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]);
3291 VNET_DEFINE_STATIC(int, isn_last);
3292 VNET_DEFINE_STATIC(int, isn_last_reseed);
3293 VNET_DEFINE_STATIC(u_int32_t, isn_offset);
3294 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old);
3295
3296 #define V_isn_secret                    VNET(isn_secret)
3297 #define V_isn_last                      VNET(isn_last)
3298 #define V_isn_last_reseed               VNET(isn_last_reseed)
3299 #define V_isn_offset                    VNET(isn_offset)
3300 #define V_isn_offset_old                VNET(isn_offset_old)
3301
3302 tcp_seq
3303 tcp_new_isn(struct in_conninfo *inc)
3304 {
3305         tcp_seq new_isn;
3306         u_int32_t projected_offset;
3307
3308         ISN_LOCK();
3309         /* Seed if this is the first use, reseed if requested. */
3310         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
3311              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
3312                 < (u_int)ticks))) {
3313                 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0);
3314                 V_isn_last_reseed = ticks;
3315         }
3316
3317         /* Compute the hash and return the ISN. */
3318         new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret,
3319             sizeof(V_isn_secret));
3320         V_isn_offset += ISN_STATIC_INCREMENT +
3321                 (arc4random() & ISN_RANDOM_INCREMENT);
3322         if (ticks != V_isn_last) {
3323                 projected_offset = V_isn_offset_old +
3324                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
3325                 if (SEQ_GT(projected_offset, V_isn_offset))
3326                         V_isn_offset = projected_offset;
3327                 V_isn_offset_old = V_isn_offset;
3328                 V_isn_last = ticks;
3329         }
3330         new_isn += V_isn_offset;
3331         ISN_UNLOCK();
3332         return (new_isn);
3333 }
3334
3335 /*
3336  * When a specific ICMP unreachable message is received and the
3337  * connection state is SYN-SENT, drop the connection.  This behavior
3338  * is controlled by the icmp_may_rst sysctl.
3339  */
3340 struct inpcb *
3341 tcp_drop_syn_sent(struct inpcb *inp, int errno)
3342 {
3343         struct tcpcb *tp;
3344
3345         NET_EPOCH_ASSERT();
3346         INP_WLOCK_ASSERT(inp);
3347
3348         if ((inp->inp_flags & INP_TIMEWAIT) ||
3349             (inp->inp_flags & INP_DROPPED))
3350                 return (inp);
3351
3352         tp = intotcpcb(inp);
3353         if (tp->t_state != TCPS_SYN_SENT)
3354                 return (inp);
3355
3356         if (IS_FASTOPEN(tp->t_flags))
3357                 tcp_fastopen_disable_path(tp);
3358
3359         tp = tcp_drop(tp, errno);
3360         if (tp != NULL)
3361                 return (inp);
3362         else
3363                 return (NULL);
3364 }
3365
3366 /*
3367  * When `need fragmentation' ICMP is received, update our idea of the MSS
3368  * based on the new value. Also nudge TCP to send something, since we
3369  * know the packet we just sent was dropped.
3370  * This duplicates some code in the tcp_mss() function in tcp_input.c.
3371  */
3372 static struct inpcb *
3373 tcp_mtudisc_notify(struct inpcb *inp, int error)
3374 {
3375
3376         tcp_mtudisc(inp, -1);
3377         return (inp);
3378 }
3379
3380 static void
3381 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
3382 {
3383         struct tcpcb *tp;
3384         struct socket *so;
3385
3386         INP_WLOCK_ASSERT(inp);
3387         if ((inp->inp_flags & INP_TIMEWAIT) ||
3388             (inp->inp_flags & INP_DROPPED))
3389                 return;
3390
3391         tp = intotcpcb(inp);
3392         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
3393
3394         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
3395
3396         so = inp->inp_socket;
3397         SOCKBUF_LOCK(&so->so_snd);
3398         /* If the mss is larger than the socket buffer, decrease the mss. */
3399         if (so->so_snd.sb_hiwat < tp->t_maxseg)
3400                 tp->t_maxseg = so->so_snd.sb_hiwat;
3401         SOCKBUF_UNLOCK(&so->so_snd);
3402
3403         TCPSTAT_INC(tcps_mturesent);
3404         tp->t_rtttime = 0;
3405         tp->snd_nxt = tp->snd_una;
3406         tcp_free_sackholes(tp);
3407         tp->snd_recover = tp->snd_max;
3408         if (tp->t_flags & TF_SACK_PERMIT)
3409                 EXIT_FASTRECOVERY(tp->t_flags);
3410         if (tp->t_fb->tfb_tcp_mtu_chg != NULL) {
3411                 /*
3412                  * Conceptually the snd_nxt setting
3413                  * and freeing sack holes should
3414                  * be done by the default stacks
3415                  * own tfb_tcp_mtu_chg().
3416                  */
3417                 tp->t_fb->tfb_tcp_mtu_chg(tp);
3418         }
3419         tp->t_fb->tfb_tcp_output(tp);
3420 }
3421
3422 #ifdef INET
3423 /*
3424  * Look-up the routing entry to the peer of this inpcb.  If no route
3425  * is found and it cannot be allocated, then return 0.  This routine
3426  * is called by TCP routines that access the rmx structure and by
3427  * tcp_mss_update to get the peer/interface MTU.
3428  */
3429 uint32_t
3430 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
3431 {
3432         struct nhop_object *nh;
3433         struct ifnet *ifp;
3434         uint32_t maxmtu = 0;
3435
3436         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
3437
3438         if (inc->inc_faddr.s_addr != INADDR_ANY) {
3439                 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0);
3440                 if (nh == NULL)
3441                         return (0);
3442
3443                 ifp = nh->nh_ifp;
3444                 maxmtu = nh->nh_mtu;
3445
3446                 /* Report additional interface capabilities. */
3447                 if (cap != NULL) {
3448                         if (ifp->if_capenable & IFCAP_TSO4 &&
3449                             ifp->if_hwassist & CSUM_TSO) {
3450                                 cap->ifcap |= CSUM_TSO;
3451                                 cap->tsomax = ifp->if_hw_tsomax;
3452                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3453                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3454                         }
3455                 }
3456         }
3457         return (maxmtu);
3458 }
3459 #endif /* INET */
3460
3461 #ifdef INET6
3462 uint32_t
3463 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
3464 {
3465         struct nhop_object *nh;
3466         struct in6_addr dst6;
3467         uint32_t scopeid;
3468         struct ifnet *ifp;
3469         uint32_t maxmtu = 0;
3470
3471         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
3472
3473         if (inc->inc_flags & INC_IPV6MINMTU)
3474                 return (IPV6_MMTU);
3475
3476         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
3477                 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
3478                 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0);
3479                 if (nh == NULL)
3480                         return (0);
3481
3482                 ifp = nh->nh_ifp;
3483                 maxmtu = nh->nh_mtu;
3484
3485                 /* Report additional interface capabilities. */
3486                 if (cap != NULL) {
3487                         if (ifp->if_capenable & IFCAP_TSO6 &&
3488                             ifp->if_hwassist & CSUM_TSO) {
3489                                 cap->ifcap |= CSUM_TSO;
3490                                 cap->tsomax = ifp->if_hw_tsomax;
3491                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
3492                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
3493                         }
3494                 }
3495         }
3496
3497         return (maxmtu);
3498 }
3499 #endif /* INET6 */
3500
3501 /*
3502  * Calculate effective SMSS per RFC5681 definition for a given TCP
3503  * connection at its current state, taking into account SACK and etc.
3504  */
3505 u_int
3506 tcp_maxseg(const struct tcpcb *tp)
3507 {
3508         u_int optlen;
3509
3510         if (tp->t_flags & TF_NOOPT)
3511                 return (tp->t_maxseg);
3512
3513         /*
3514          * Here we have a simplified code from tcp_addoptions(),
3515          * without a proper loop, and having most of paddings hardcoded.
3516          * We might make mistakes with padding here in some edge cases,
3517          * but this is harmless, since result of tcp_maxseg() is used
3518          * only in cwnd and ssthresh estimations.
3519          */
3520         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3521                 if (tp->t_flags & TF_RCVD_TSTMP)
3522                         optlen = TCPOLEN_TSTAMP_APPA;
3523                 else
3524                         optlen = 0;
3525 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3526                 if (tp->t_flags & TF_SIGNATURE)
3527                         optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3528 #endif
3529                 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
3530                         optlen += TCPOLEN_SACKHDR;
3531                         optlen += tp->rcv_numsacks * TCPOLEN_SACK;
3532                         optlen = PADTCPOLEN(optlen);
3533                 }
3534         } else {
3535                 if (tp->t_flags & TF_REQ_TSTMP)
3536                         optlen = TCPOLEN_TSTAMP_APPA;
3537                 else
3538                         optlen = PADTCPOLEN(TCPOLEN_MAXSEG);
3539                 if (tp->t_flags & TF_REQ_SCALE)
3540                         optlen += PADTCPOLEN(TCPOLEN_WINDOW);
3541 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3542                 if (tp->t_flags & TF_SIGNATURE)
3543                         optlen += PADTCPOLEN(TCPOLEN_SIGNATURE);
3544 #endif
3545                 if (tp->t_flags & TF_SACK_PERMIT)
3546                         optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED);
3547         }
3548 #undef PAD
3549         optlen = min(optlen, TCP_MAXOLEN);
3550         return (tp->t_maxseg - optlen);
3551 }
3552
3553
3554 u_int
3555 tcp_fixed_maxseg(const struct tcpcb *tp)
3556 {
3557         int optlen;
3558
3559         if (tp->t_flags & TF_NOOPT)
3560                 return (tp->t_maxseg);
3561
3562         /*
3563          * Here we have a simplified code from tcp_addoptions(),
3564          * without a proper loop, and having most of paddings hardcoded.
3565          * We only consider fixed options that we would send every
3566          * time I.e. SACK is not considered. This is important
3567          * for cc modules to figure out what the modulo of the
3568          * cwnd should be.
3569          */
3570 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
3571         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3572                 if (tp->t_flags & TF_RCVD_TSTMP)
3573                         optlen = TCPOLEN_TSTAMP_APPA;
3574                 else
3575                         optlen = 0;
3576 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3577                 if (tp->t_flags & TF_SIGNATURE)
3578                         optlen += PAD(TCPOLEN_SIGNATURE);
3579 #endif
3580         } else {
3581                 if (tp->t_flags & TF_REQ_TSTMP)
3582                         optlen = TCPOLEN_TSTAMP_APPA;
3583                 else
3584                         optlen = PAD(TCPOLEN_MAXSEG);
3585                 if (tp->t_flags & TF_REQ_SCALE)
3586                         optlen += PAD(TCPOLEN_WINDOW);
3587 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
3588                 if (tp->t_flags & TF_SIGNATURE)
3589                         optlen += PAD(TCPOLEN_SIGNATURE);
3590 #endif
3591                 if (tp->t_flags & TF_SACK_PERMIT)
3592                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
3593         }
3594 #undef PAD
3595         optlen = min(optlen, TCP_MAXOLEN);
3596         return (tp->t_maxseg - optlen);
3597 }
3598
3599
3600
3601 static int
3602 sysctl_drop(SYSCTL_HANDLER_ARGS)
3603 {
3604         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3605         struct sockaddr_storage addrs[2];
3606         struct inpcb *inp;
3607         struct tcpcb *tp;
3608         struct tcptw *tw;
3609         struct sockaddr_in *fin, *lin;
3610         struct epoch_tracker et;
3611 #ifdef INET6
3612         struct sockaddr_in6 *fin6, *lin6;
3613 #endif
3614         int error;
3615
3616         inp = NULL;
3617         fin = lin = NULL;
3618 #ifdef INET6
3619         fin6 = lin6 = NULL;
3620 #endif
3621         error = 0;
3622
3623         if (req->oldptr != NULL || req->oldlen != 0)
3624                 return (EINVAL);
3625         if (req->newptr == NULL)
3626                 return (EPERM);
3627         if (req->newlen < sizeof(addrs))
3628                 return (ENOMEM);
3629         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3630         if (error)
3631                 return (error);
3632
3633         switch (addrs[0].ss_family) {
3634 #ifdef INET6
3635         case AF_INET6:
3636                 fin6 = (struct sockaddr_in6 *)&addrs[0];
3637                 lin6 = (struct sockaddr_in6 *)&addrs[1];
3638                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3639                     lin6->sin6_len != sizeof(struct sockaddr_in6))
3640                         return (EINVAL);
3641                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3642                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3643                                 return (EINVAL);
3644                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3645                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3646                         fin = (struct sockaddr_in *)&addrs[0];
3647                         lin = (struct sockaddr_in *)&addrs[1];
3648                         break;
3649                 }
3650                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3651                 if (error)
3652                         return (error);
3653                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3654                 if (error)
3655                         return (error);
3656                 break;
3657 #endif
3658 #ifdef INET
3659         case AF_INET:
3660                 fin = (struct sockaddr_in *)&addrs[0];
3661                 lin = (struct sockaddr_in *)&addrs[1];
3662                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3663                     lin->sin_len != sizeof(struct sockaddr_in))
3664                         return (EINVAL);
3665                 break;
3666 #endif
3667         default:
3668                 return (EINVAL);
3669         }
3670         NET_EPOCH_ENTER(et);
3671         switch (addrs[0].ss_family) {
3672 #ifdef INET6
3673         case AF_INET6:
3674                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3675                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3676                     INPLOOKUP_WLOCKPCB, NULL);
3677                 break;
3678 #endif
3679 #ifdef INET
3680         case AF_INET:
3681                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3682                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3683                 break;
3684 #endif
3685         }
3686         if (inp != NULL) {
3687                 if (inp->inp_flags & INP_TIMEWAIT) {
3688                         /*
3689                          * XXXRW: There currently exists a state where an
3690                          * inpcb is present, but its timewait state has been
3691                          * discarded.  For now, don't allow dropping of this
3692                          * type of inpcb.
3693                          */
3694                         tw = intotw(inp);
3695                         if (tw != NULL)
3696                                 tcp_twclose(tw, 0);
3697                         else
3698                                 INP_WUNLOCK(inp);
3699                 } else if (!(inp->inp_flags & INP_DROPPED) &&
3700                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
3701                         tp = intotcpcb(inp);
3702                         tp = tcp_drop(tp, ECONNABORTED);
3703                         if (tp != NULL)
3704                                 INP_WUNLOCK(inp);
3705                 } else
3706                         INP_WUNLOCK(inp);
3707         } else
3708                 error = ESRCH;
3709         NET_EPOCH_EXIT(et);
3710         return (error);
3711 }
3712
3713 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
3714     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3715     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "",
3716     "Drop TCP connection");
3717
3718 #ifdef KERN_TLS
3719 static int
3720 sysctl_switch_tls(SYSCTL_HANDLER_ARGS)
3721 {
3722         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
3723         struct sockaddr_storage addrs[2];
3724         struct inpcb *inp;
3725         struct sockaddr_in *fin, *lin;
3726         struct epoch_tracker et;
3727 #ifdef INET6
3728         struct sockaddr_in6 *fin6, *lin6;
3729 #endif
3730         int error;
3731
3732         inp = NULL;
3733         fin = lin = NULL;
3734 #ifdef INET6
3735         fin6 = lin6 = NULL;
3736 #endif
3737         error = 0;
3738
3739         if (req->oldptr != NULL || req->oldlen != 0)
3740                 return (EINVAL);
3741         if (req->newptr == NULL)
3742                 return (EPERM);
3743         if (req->newlen < sizeof(addrs))
3744                 return (ENOMEM);
3745         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
3746         if (error)
3747                 return (error);
3748
3749         switch (addrs[0].ss_family) {
3750 #ifdef INET6
3751         case AF_INET6:
3752                 fin6 = (struct sockaddr_in6 *)&addrs[0];
3753                 lin6 = (struct sockaddr_in6 *)&addrs[1];
3754                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
3755                     lin6->sin6_len != sizeof(struct sockaddr_in6))
3756                         return (EINVAL);
3757                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
3758                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
3759                                 return (EINVAL);
3760                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
3761                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
3762                         fin = (struct sockaddr_in *)&addrs[0];
3763                         lin = (struct sockaddr_in *)&addrs[1];
3764                         break;
3765                 }
3766                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
3767                 if (error)
3768                         return (error);
3769                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
3770                 if (error)
3771                         return (error);
3772                 break;
3773 #endif
3774 #ifdef INET
3775         case AF_INET:
3776                 fin = (struct sockaddr_in *)&addrs[0];
3777                 lin = (struct sockaddr_in *)&addrs[1];
3778                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
3779                     lin->sin_len != sizeof(struct sockaddr_in))
3780                         return (EINVAL);
3781                 break;
3782 #endif
3783         default:
3784                 return (EINVAL);
3785         }
3786         NET_EPOCH_ENTER(et);
3787         switch (addrs[0].ss_family) {
3788 #ifdef INET6
3789         case AF_INET6:
3790                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
3791                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
3792                     INPLOOKUP_WLOCKPCB, NULL);
3793                 break;
3794 #endif
3795 #ifdef INET
3796         case AF_INET:
3797                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
3798                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
3799                 break;
3800 #endif
3801         }
3802         NET_EPOCH_EXIT(et);
3803         if (inp != NULL) {
3804                 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 ||
3805                     inp->inp_socket == NULL) {
3806                         error = ECONNRESET;
3807                         INP_WUNLOCK(inp);
3808                 } else {
3809                         struct socket *so;
3810
3811                         so = inp->inp_socket;
3812                         soref(so);
3813                         error = ktls_set_tx_mode(so,
3814                             arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET);
3815                         INP_WUNLOCK(inp);
3816                         SOCK_LOCK(so);
3817                         sorele(so);
3818                 }
3819         } else
3820                 error = ESRCH;
3821         return (error);
3822 }
3823
3824 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls,
3825     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3826     CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "",
3827     "Switch TCP connection to SW TLS");
3828 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls,
3829     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP |
3830     CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "",
3831     "Switch TCP connection to ifnet TLS");
3832 #endif
3833
3834 /*
3835  * Generate a standardized TCP log line for use throughout the
3836  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
3837  * allow use in the interrupt context.
3838  *
3839  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
3840  * NB: The function may return NULL if memory allocation failed.
3841  *
3842  * Due to header inclusion and ordering limitations the struct ip
3843  * and ip6_hdr pointers have to be passed as void pointers.
3844  */
3845 char *
3846 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3847     const void *ip6hdr)
3848 {
3849
3850         /* Is logging enabled? */
3851         if (V_tcp_log_in_vain == 0)
3852                 return (NULL);
3853
3854         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3855 }
3856
3857 char *
3858 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3859     const void *ip6hdr)
3860 {
3861
3862         /* Is logging enabled? */
3863         if (tcp_log_debug == 0)
3864                 return (NULL);
3865
3866         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
3867 }
3868
3869 static char *
3870 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
3871     const void *ip6hdr)
3872 {
3873         char *s, *sp;
3874         size_t size;
3875         struct ip *ip;
3876 #ifdef INET6
3877         const struct ip6_hdr *ip6;
3878
3879         ip6 = (const struct ip6_hdr *)ip6hdr;
3880 #endif /* INET6 */
3881         ip = (struct ip *)ip4hdr;
3882
3883         /*
3884          * The log line looks like this:
3885          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
3886          */
3887         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
3888             sizeof(PRINT_TH_FLAGS) + 1 +
3889 #ifdef INET6
3890             2 * INET6_ADDRSTRLEN;
3891 #else
3892             2 * INET_ADDRSTRLEN;
3893 #endif /* INET6 */
3894
3895         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
3896         if (s == NULL)
3897                 return (NULL);
3898
3899         strcat(s, "TCP: [");
3900         sp = s + strlen(s);
3901
3902         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
3903                 inet_ntoa_r(inc->inc_faddr, sp);
3904                 sp = s + strlen(s);
3905                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3906                 sp = s + strlen(s);
3907                 inet_ntoa_r(inc->inc_laddr, sp);
3908                 sp = s + strlen(s);
3909                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3910 #ifdef INET6
3911         } else if (inc) {
3912                 ip6_sprintf(sp, &inc->inc6_faddr);
3913                 sp = s + strlen(s);
3914                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
3915                 sp = s + strlen(s);
3916                 ip6_sprintf(sp, &inc->inc6_laddr);
3917                 sp = s + strlen(s);
3918                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
3919         } else if (ip6 && th) {
3920                 ip6_sprintf(sp, &ip6->ip6_src);
3921                 sp = s + strlen(s);
3922                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3923                 sp = s + strlen(s);
3924                 ip6_sprintf(sp, &ip6->ip6_dst);
3925                 sp = s + strlen(s);
3926                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3927 #endif /* INET6 */
3928 #ifdef INET
3929         } else if (ip && th) {
3930                 inet_ntoa_r(ip->ip_src, sp);
3931                 sp = s + strlen(s);
3932                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
3933                 sp = s + strlen(s);
3934                 inet_ntoa_r(ip->ip_dst, sp);
3935                 sp = s + strlen(s);
3936                 sprintf(sp, "]:%i", ntohs(th->th_dport));
3937 #endif /* INET */
3938         } else {
3939                 free(s, M_TCPLOG);
3940                 return (NULL);
3941         }
3942         sp = s + strlen(s);
3943         if (th)
3944                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
3945         if (*(s + size - 1) != '\0')
3946                 panic("%s: string too long", __func__);
3947         return (s);
3948 }
3949
3950 /*
3951  * A subroutine which makes it easy to track TCP state changes with DTrace.
3952  * This function shouldn't be called for t_state initializations that don't
3953  * correspond to actual TCP state transitions.
3954  */
3955 void
3956 tcp_state_change(struct tcpcb *tp, int newstate)
3957 {
3958 #if defined(KDTRACE_HOOKS)
3959         int pstate = tp->t_state;
3960 #endif
3961
3962         TCPSTATES_DEC(tp->t_state);
3963         TCPSTATES_INC(newstate);
3964         tp->t_state = newstate;
3965         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
3966 }
3967
3968 /*
3969  * Create an external-format (``xtcpcb'') structure using the information in
3970  * the kernel-format tcpcb structure pointed to by tp.  This is done to
3971  * reduce the spew of irrelevant information over this interface, to isolate
3972  * user code from changes in the kernel structure, and potentially to provide
3973  * information-hiding if we decide that some of this information should be
3974  * hidden from users.
3975  */
3976 void
3977 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt)
3978 {
3979         struct tcpcb *tp = intotcpcb(inp);
3980         struct tcptw *tw = intotw(inp);
3981         sbintime_t now;
3982
3983         bzero(xt, sizeof(*xt));
3984         if (inp->inp_flags & INP_TIMEWAIT) {
3985                 xt->t_state = TCPS_TIME_WAIT;
3986                 xt->xt_encaps_port = tw->t_port;
3987         } else {
3988                 xt->t_state = tp->t_state;
3989                 xt->t_logstate = tp->t_logstate;
3990                 xt->t_flags = tp->t_flags;
3991                 xt->t_sndzerowin = tp->t_sndzerowin;
3992                 xt->t_sndrexmitpack = tp->t_sndrexmitpack;
3993                 xt->t_rcvoopack = tp->t_rcvoopack;
3994                 xt->t_rcv_wnd = tp->rcv_wnd;
3995                 xt->t_snd_wnd = tp->snd_wnd;
3996                 xt->t_snd_cwnd = tp->snd_cwnd;
3997                 xt->t_snd_ssthresh = tp->snd_ssthresh;
3998                 xt->t_maxseg = tp->t_maxseg;
3999                 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 +
4000                              (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0;
4001
4002                 now = getsbinuptime();
4003 #define COPYTIMER(ttt)  do {                                            \
4004                 if (callout_active(&tp->t_timers->ttt))                 \
4005                         xt->ttt = (tp->t_timers->ttt.c_time - now) /    \
4006                             SBT_1MS;                                    \
4007                 else                                                    \
4008                         xt->ttt = 0;                                    \
4009 } while (0)
4010                 COPYTIMER(tt_delack);
4011                 COPYTIMER(tt_rexmt);
4012                 COPYTIMER(tt_persist);
4013                 COPYTIMER(tt_keep);
4014                 COPYTIMER(tt_2msl);
4015 #undef COPYTIMER
4016                 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz;
4017
4018                 xt->xt_encaps_port = tp->t_port;
4019                 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack,
4020                     TCP_FUNCTION_NAME_LEN_MAX);
4021                 bcopy(CC_ALGO(tp)->name, xt->xt_cc,
4022                     TCP_CA_NAME_MAX);
4023 #ifdef TCP_BLACKBOX
4024                 (void)tcp_log_get_id(tp, xt->xt_logid);
4025 #endif
4026         }
4027
4028         xt->xt_len = sizeof(struct xtcpcb);
4029         in_pcbtoxinpcb(inp, &xt->xt_inp);
4030         if (inp->inp_socket == NULL)
4031                 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
4032 }
4033
4034 void
4035 tcp_log_end_status(struct tcpcb *tp, uint8_t status)
4036 {
4037         uint32_t bit, i;
4038
4039         if ((tp == NULL) ||
4040             (status > TCP_EI_STATUS_MAX_VALUE) ||
4041             (status == 0)) {
4042                 /* Invalid */
4043                 return;
4044         }
4045         if (status > (sizeof(uint32_t) * 8)) {
4046                 /* Should this be a KASSERT? */
4047                 return;
4048         }
4049         bit = 1U << (status - 1);
4050         if (bit & tp->t_end_info_status) {
4051                 /* already logged */
4052                 return;
4053         }
4054         for (i = 0; i < TCP_END_BYTE_INFO; i++) {
4055                 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) {
4056                         tp->t_end_info_bytes[i] = status;
4057                         tp->t_end_info_status |= bit;
4058                         break;
4059                 }
4060         }
4061 }
4062
4063 int
4064 tcp_can_enable_pacing(void)
4065 {
4066
4067         if ((tcp_pacing_limit == -1) ||
4068             (tcp_pacing_limit > number_of_tcp_connections_pacing)) {
4069                 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1);
4070                 shadow_num_connections = number_of_tcp_connections_pacing;
4071                 return (1);
4072         } else {
4073                 return (0);
4074         }
4075 }
4076
4077 static uint8_t tcp_pacing_warning = 0;
4078
4079 void
4080 tcp_decrement_paced_conn(void)
4081 {
4082         uint32_t ret;
4083
4084         ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1);
4085         shadow_num_connections = number_of_tcp_connections_pacing;
4086         KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?"));
4087         if (ret == 0) {
4088                 if (tcp_pacing_limit != -1) {
4089                         printf("Warning all pacing is now disabled, count decrements invalidly!\n");
4090                         tcp_pacing_limit = 0;
4091                 } else if (tcp_pacing_warning == 0) {
4092                         printf("Warning pacing count is invalid, invalid decrement\n");
4093                         tcp_pacing_warning = 1;
4094                 }
4095         }
4096 }