sys/netinet/tcp_subr.c

   1 /*-
   2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 4. Neither the name of the University nor the names of its contributors
  14  *    may be used to endorse or promote products derived from this software
  15  *    without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  *
  29  *      @(#)tcp_subr.c  8.2 (Berkeley) 5/24/95
  30  */
  31
  32 #include <sys/cdefs.h>
  33 __FBSDID("$FreeBSD$");
  34
  35 #include "opt_compat.h"
  36 #include "opt_inet.h"
  37 #include "opt_inet6.h"
  38 #include "opt_ipsec.h"
  39 #include "opt_tcpdebug.h"
  40
  41 #include <sys/param.h>
  42 #include <sys/systm.h>
  43 #include <sys/callout.h>
  44 #include <sys/eventhandler.h>
  45 #include <sys/hhook.h>
  46 #include <sys/kernel.h>
  47 #include <sys/khelp.h>
  48 #include <sys/sysctl.h>
  49 #include <sys/jail.h>
  50 #include <sys/malloc.h>
  51 #include <sys/refcount.h>
  52 #include <sys/mbuf.h>
  53 #ifdef INET6
  54 #include <sys/domain.h>
  55 #endif
  56 #include <sys/priv.h>
  57 #include <sys/proc.h>
  58 #include <sys/sdt.h>
  59 #include <sys/socket.h>
  60 #include <sys/socketvar.h>
  61 #include <sys/protosw.h>
  62 #include <sys/random.h>
  63
  64 #include <vm/uma.h>
  65
  66 #include <net/route.h>
  67 #include <net/if.h>
  68 #include <net/if_var.h>
  69 #include <net/vnet.h>
  70
  71 #include <netinet/in.h>
  72 #include <netinet/in_fib.h>
  73 #include <netinet/in_kdtrace.h>
  74 #include <netinet/in_pcb.h>
  75 #include <netinet/in_systm.h>
  76 #include <netinet/in_var.h>
  77 #include <netinet/ip.h>
  78 #include <netinet/ip_icmp.h>
  79 #include <netinet/ip_var.h>
  80 #ifdef INET6
  81 #include <netinet/icmp6.h>
  82 #include <netinet/ip6.h>
  83 #include <netinet6/in6_fib.h>
  84 #include <netinet6/in6_pcb.h>
  85 #include <netinet6/ip6_var.h>
  86 #include <netinet6/scope6_var.h>
  87 #include <netinet6/nd6.h>
  88 #endif
  89
  90 #ifdef TCP_RFC7413
  91 #include <netinet/tcp_fastopen.h>
  92 #endif
  93 #include <netinet/tcp.h>
  94 #include <netinet/tcp_fsm.h>
  95 #include <netinet/tcp_seq.h>
  96 #include <netinet/tcp_timer.h>
  97 #include <netinet/tcp_var.h>
  98 #include <netinet/tcp_syncache.h>
  99 #include <netinet/cc/cc.h>
 100 #ifdef INET6
 101 #include <netinet6/tcp6_var.h>
 102 #endif
 103 #include <netinet/tcpip.h>
 104 #ifdef TCPPCAP
 105 #include <netinet/tcp_pcap.h>
 106 #endif
 107 #ifdef TCPDEBUG
 108 #include <netinet/tcp_debug.h>
 109 #endif
 110 #ifdef INET6
 111 #include <netinet6/ip6protosw.h>
 112 #endif
 113 #ifdef TCP_OFFLOAD
 114 #include <netinet/tcp_offload.h>
 115 #endif
 116
 117 #include <netipsec/ipsec_support.h>
 118
 119 #include <machine/in_cksum.h>
 120 #include <sys/md5.h>
 121
 122 #include <security/mac/mac_framework.h>
 123
 124 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
 125 #ifdef INET6
 126 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
 127 #endif
 128
 129 struct rwlock tcp_function_lock;
 130
 131 static int
 132 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
 133 {
 134         int error, new;
 135
 136         new = V_tcp_mssdflt;
 137         error = sysctl_handle_int(oidp, &new, 0, req);
 138         if (error == 0 && req->newptr) {
 139                 if (new < TCP_MINMSS)
 140                         error = EINVAL;
 141                 else
 142                         V_tcp_mssdflt = new;
 143         }
 144         return (error);
 145 }
 146
 147 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
 148     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
 149     &sysctl_net_inet_tcp_mss_check, "I",
 150     "Default TCP Maximum Segment Size");
 151
 152 #ifdef INET6
 153 static int
 154 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
 155 {
 156         int error, new;
 157
 158         new = V_tcp_v6mssdflt;
 159         error = sysctl_handle_int(oidp, &new, 0, req);
 160         if (error == 0 && req->newptr) {
 161                 if (new < TCP_MINMSS)
 162                         error = EINVAL;
 163                 else
 164                         V_tcp_v6mssdflt = new;
 165         }
 166         return (error);
 167 }
 168
 169 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
 170     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
 171     &sysctl_net_inet_tcp_mss_v6_check, "I",
 172    "Default TCP Maximum Segment Size for IPv6");
 173 #endif /* INET6 */
 174
 175 /*
 176  * Minimum MSS we accept and use. This prevents DoS attacks where
 177  * we are forced to a ridiculous low MSS like 20 and send hundreds
 178  * of packets instead of one. The effect scales with the available
 179  * bandwidth and quickly saturates the CPU and network interface
 180  * with packet generation and sending. Set to zero to disable MINMSS
 181  * checking. This setting prevents us from sending too small packets.
 182  */
 183 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
 184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
 185      &VNET_NAME(tcp_minmss), 0,
 186     "Minimum TCP Maximum Segment Size");
 187
 188 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
 189 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
 190     &VNET_NAME(tcp_do_rfc1323), 0,
 191     "Enable rfc1323 (high performance TCP) extensions");
 192
 193 static int      tcp_log_debug = 0;
 194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
 195     &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
 196
 197 static int      tcp_tcbhashsize;
 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
 199     &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
 200
 201 static int      do_tcpdrain = 1;
 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
 203     "Enable tcp_drain routine for extra help when low on mbufs");
 204
 205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
 206     &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
 207
 208 static VNET_DEFINE(int, icmp_may_rst) = 1;
 209 #define V_icmp_may_rst                  VNET(icmp_may_rst)
 210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
 211     &VNET_NAME(icmp_may_rst), 0,
 212     "Certain ICMP unreachable messages may abort connections in SYN_SENT");
 213
 214 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
 215 #define V_tcp_isn_reseed_interval       VNET(tcp_isn_reseed_interval)
 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
 217     &VNET_NAME(tcp_isn_reseed_interval), 0,
 218     "Seconds between reseeding of ISN secret");
 219
 220 static int      tcp_soreceive_stream;
 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
 222     &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
 223
 224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
 225 #define V_sack_hole_zone                VNET(sack_hole_zone)
 226
 227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
 228
 229 VNET_DEFINE(u_char, ts_offset_secret[32]);
 230 #define V_ts_offset_secret      VNET(ts_offset_secret)
 231
 232 static struct inpcb *tcp_notify(struct inpcb *, int);
 233 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
 234 static void tcp_mtudisc(struct inpcb *, int);
 235 static char *   tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
 236                     void *ip4hdr, const void *ip6hdr);
 237
 238
 239 static struct tcp_function_block tcp_def_funcblk = {
 240         "default",
 241         tcp_output,
 242         tcp_do_segment,
 243         tcp_default_ctloutput,
 244         NULL,
 245         NULL,
 246         NULL,
 247         NULL,
 248         NULL,
 249         NULL,
 250         0,
 251         0
 252 };
 253
 254 int t_functions_inited = 0;
 255 struct tcp_funchead t_functions;
 256 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk;
 257
 258 static void
 259 init_tcp_functions(void)
 260 {
 261         if (t_functions_inited == 0) {
 262                 TAILQ_INIT(&t_functions);
 263                 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0);
 264                 t_functions_inited = 1;
 265         }
 266 }
 267
 268 static struct tcp_function_block *
 269 find_tcp_functions_locked(struct tcp_function_set *fs)
 270 {
 271         struct tcp_function *f;
 272         struct tcp_function_block *blk=NULL;
 273
 274         TAILQ_FOREACH(f, &t_functions, tf_next) {
 275                 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) {
 276                         blk = f->tf_fb;
 277                         break;
 278                 }
 279         }
 280         return(blk);
 281 }
 282
 283 static struct tcp_function_block *
 284 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s)
 285 {
 286         struct tcp_function_block *rblk=NULL;
 287         struct tcp_function *f;
 288
 289         TAILQ_FOREACH(f, &t_functions, tf_next) {
 290                 if (f->tf_fb == blk) {
 291                         rblk = blk;
 292                         if (s) {
 293                                 *s = f;
 294                         }
 295                         break;
 296                 }
 297         }
 298         return (rblk);
 299 }
 300
 301 struct tcp_function_block *
 302 find_and_ref_tcp_functions(struct tcp_function_set *fs)
 303 {
 304         struct tcp_function_block *blk;
 305
 306         rw_rlock(&tcp_function_lock);
 307         blk = find_tcp_functions_locked(fs);
 308         if (blk)
 309                 refcount_acquire(&blk->tfb_refcnt);
 310         rw_runlock(&tcp_function_lock);
 311         return(blk);
 312 }
 313
 314 struct tcp_function_block *
 315 find_and_ref_tcp_fb(struct tcp_function_block *blk)
 316 {
 317         struct tcp_function_block *rblk;
 318
 319         rw_rlock(&tcp_function_lock);
 320         rblk = find_tcp_fb_locked(blk, NULL);
 321         if (rblk)
 322                 refcount_acquire(&rblk->tfb_refcnt);
 323         rw_runlock(&tcp_function_lock);
 324         return(rblk);
 325 }
 326
 327
 328 static int
 329 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS)
 330 {
 331         int error=ENOENT;
 332         struct tcp_function_set fs;
 333         struct tcp_function_block *blk;
 334
 335         memset(&fs, 0, sizeof(fs));
 336         rw_rlock(&tcp_function_lock);
 337         blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL);
 338         if (blk) {
 339                 /* Found him */
 340                 strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 341                 fs.pcbcnt = blk->tfb_refcnt;
 342         }
 343         rw_runlock(&tcp_function_lock);
 344         error = sysctl_handle_string(oidp, fs.function_set_name,
 345                                      sizeof(fs.function_set_name), req);
 346
 347         /* Check for error or no change */
 348         if (error != 0 || req->newptr == NULL)
 349                 return(error);
 350
 351         rw_wlock(&tcp_function_lock);
 352         blk = find_tcp_functions_locked(&fs);
 353         if ((blk == NULL) ||
 354             (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) {
 355                 error = ENOENT;
 356                 goto done;
 357         }
 358         tcp_func_set_ptr = blk;
 359 done:
 360         rw_wunlock(&tcp_function_lock);
 361         return (error);
 362 }
 363
 364 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default,
 365             CTLTYPE_STRING | CTLFLAG_RW,
 366             NULL, 0, sysctl_net_inet_default_tcp_functions, "A",
 367             "Set/get the default TCP functions");
 368
 369 static int
 370 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS)
 371 {
 372         int error, cnt, linesz;
 373         struct tcp_function *f;
 374         char *buffer, *cp;
 375         size_t bufsz, outsz;
 376
 377         cnt = 0;
 378         rw_rlock(&tcp_function_lock);
 379         TAILQ_FOREACH(f, &t_functions, tf_next) {
 380                 cnt++;
 381         }
 382         rw_runlock(&tcp_function_lock);
 383
 384         bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1;
 385         buffer = malloc(bufsz, M_TEMP, M_WAITOK);
 386
 387         error = 0;
 388         cp = buffer;
 389
 390         linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count");
 391         cp += linesz;
 392         bufsz -= linesz;
 393         outsz = linesz;
 394
 395         rw_rlock(&tcp_function_lock);
 396         TAILQ_FOREACH(f, &t_functions, tf_next) {
 397                 linesz = snprintf(cp, bufsz, "%-32s%c %u\n",
 398                     f->tf_fb->tfb_tcp_block_name,
 399                     (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ',
 400                     f->tf_fb->tfb_refcnt);
 401                 if (linesz >= bufsz) {
 402                         error = EOVERFLOW;
 403                         break;
 404                 }
 405                 cp += linesz;
 406                 bufsz -= linesz;
 407                 outsz += linesz;
 408         }
 409         rw_runlock(&tcp_function_lock);
 410         if (error == 0)
 411                 error = sysctl_handle_string(oidp, buffer, outsz + 1, req);
 412         free(buffer, M_TEMP);
 413         return (error);
 414 }
 415
 416 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available,
 417             CTLTYPE_STRING|CTLFLAG_RD,
 418             NULL, 0, sysctl_net_inet_list_available, "A",
 419             "list available TCP Function sets");
 420
 421 /*
 422  * Target size of TCP PCB hash tables. Must be a power of two.
 423  *
 424  * Note that this can be overridden by the kernel environment
 425  * variable net.inet.tcp.tcbhashsize
 426  */
 427 #ifndef TCBHASHSIZE
 428 #define TCBHASHSIZE     0
 429 #endif
 430
 431 /*
 432  * XXX
 433  * Callouts should be moved into struct tcp directly.  They are currently
 434  * separate because the tcpcb structure is exported to userland for sysctl
 435  * parsing purposes, which do not know about callouts.
 436  */
 437 struct tcpcb_mem {
 438         struct  tcpcb           tcb;
 439         struct  tcp_timer       tt;
 440         struct  cc_var          ccv;
 441         struct  osd             osd;
 442 };
 443
 444 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
 445 #define V_tcpcb_zone                    VNET(tcpcb_zone)
 446
 447 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
 448 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory");
 449
 450 static struct mtx isn_mtx;
 451
 452 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
 453 #define ISN_LOCK()      mtx_lock(&isn_mtx)
 454 #define ISN_UNLOCK()    mtx_unlock(&isn_mtx)
 455
 456 /*
 457  * TCP initialization.
 458  */
 459 static void
 460 tcp_zone_change(void *tag)
 461 {
 462
 463         uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
 464         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 465         tcp_tw_zone_change();
 466 }
 467
 468 static int
 469 tcp_inpcb_init(void *mem, int size, int flags)
 470 {
 471         struct inpcb *inp = mem;
 472
 473         INP_LOCK_INIT(inp, "inp", "tcpinp");
 474         return (0);
 475 }
 476
 477 /*
 478  * Take a value and get the next power of 2 that doesn't overflow.
 479  * Used to size the tcp_inpcb hash buckets.
 480  */
 481 static int
 482 maketcp_hashsize(int size)
 483 {
 484         int hashsize;
 485
 486         /*
 487          * auto tune.
 488          * get the next power of 2 higher than maxsockets.
 489          */
 490         hashsize = 1 << fls(size);
 491         /* catch overflow, and just go one power of 2 smaller */
 492         if (hashsize < size) {
 493                 hashsize = 1 << (fls(size) - 1);
 494         }
 495         return (hashsize);
 496 }
 497
 498 int
 499 register_tcp_functions(struct tcp_function_block *blk, int wait)
 500 {
 501         struct tcp_function_block *lblk;
 502         struct tcp_function *n;
 503         struct tcp_function_set fs;
 504
 505         if (t_functions_inited == 0) {
 506                 init_tcp_functions();
 507         }
 508         if ((blk->tfb_tcp_output == NULL) ||
 509             (blk->tfb_tcp_do_segment == NULL) ||
 510             (blk->tfb_tcp_ctloutput == NULL) ||
 511             (strlen(blk->tfb_tcp_block_name) == 0)) {
 512                 /*
 513                  * These functions are required and you
 514                  * need a name.
 515                  */
 516                 return (EINVAL);
 517         }
 518         if (blk->tfb_tcp_timer_stop_all ||
 519             blk->tfb_tcp_timer_activate ||
 520             blk->tfb_tcp_timer_active ||
 521             blk->tfb_tcp_timer_stop) {
 522                 /*
 523                  * If you define one timer function you
 524                  * must have them all.
 525                  */
 526                 if ((blk->tfb_tcp_timer_stop_all == NULL) ||
 527                     (blk->tfb_tcp_timer_activate == NULL) ||
 528                     (blk->tfb_tcp_timer_active == NULL) ||
 529                     (blk->tfb_tcp_timer_stop == NULL)) {
 530                         return (EINVAL);
 531                 }
 532         }
 533         n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait);
 534         if (n == NULL) {
 535                 return (ENOMEM);
 536         }
 537         n->tf_fb = blk;
 538         strcpy(fs.function_set_name, blk->tfb_tcp_block_name);
 539         rw_wlock(&tcp_function_lock);
 540         lblk = find_tcp_functions_locked(&fs);
 541         if (lblk) {
 542                 /* Duplicate name space not allowed */
 543                 rw_wunlock(&tcp_function_lock);
 544                 free(n, M_TCPFUNCTIONS);
 545                 return (EALREADY);
 546         }
 547         refcount_init(&blk->tfb_refcnt, 0);
 548         blk->tfb_flags = 0;
 549         TAILQ_INSERT_TAIL(&t_functions, n, tf_next);
 550         rw_wunlock(&tcp_function_lock);
 551         return(0);
 552 }
 553
 554 int
 555 deregister_tcp_functions(struct tcp_function_block *blk)
 556 {
 557         struct tcp_function_block *lblk;
 558         struct tcp_function *f;
 559         int error=ENOENT;
 560
 561         if (strcmp(blk->tfb_tcp_block_name, "default") == 0) {
 562                 /* You can't un-register the default */
 563                 return (EPERM);
 564         }
 565         rw_wlock(&tcp_function_lock);
 566         if (blk == tcp_func_set_ptr) {
 567                 /* You can't free the current default */
 568                 rw_wunlock(&tcp_function_lock);
 569                 return (EBUSY);
 570         }
 571         if (blk->tfb_refcnt) {
 572                 /* Still tcb attached, mark it. */
 573                 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED;
 574                 rw_wunlock(&tcp_function_lock);
 575                 return (EBUSY);
 576         }
 577         lblk = find_tcp_fb_locked(blk, &f);
 578         if (lblk) {
 579                 /* Found */
 580                 TAILQ_REMOVE(&t_functions, f, tf_next);
 581                 f->tf_fb = NULL;
 582                 free(f, M_TCPFUNCTIONS);
 583                 error = 0;
 584         }
 585         rw_wunlock(&tcp_function_lock);
 586         return (error);
 587 }
 588
 589 void
 590 tcp_init(void)
 591 {
 592         const char *tcbhash_tuneable;
 593         int hashsize;
 594
 595         tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
 596
 597         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
 598             &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 599                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 600         if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
 601             &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
 602                 printf("%s: WARNING: unable to register helper hook\n", __func__);
 603         hashsize = TCBHASHSIZE;
 604         TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
 605         if (hashsize == 0) {
 606                 /*
 607                  * Auto tune the hash size based on maxsockets.
 608                  * A perfect hash would have a 1:1 mapping
 609                  * (hashsize = maxsockets) however it's been
 610                  * suggested that O(2) average is better.
 611                  */
 612                 hashsize = maketcp_hashsize(maxsockets / 4);
 613                 /*
 614                  * Our historical default is 512,
 615                  * do not autotune lower than this.
 616                  */
 617                 if (hashsize < 512)
 618                         hashsize = 512;
 619                 if (bootverbose && IS_DEFAULT_VNET(curvnet))
 620                         printf("%s: %s auto tuned to %d\n", __func__,
 621                             tcbhash_tuneable, hashsize);
 622         }
 623         /*
 624          * We require a hashsize to be a power of two.
 625          * Previously if it was not a power of two we would just reset it
 626          * back to 512, which could be a nasty surprise if you did not notice
 627          * the error message.
 628          * Instead what we do is clip it to the closest power of two lower
 629          * than the specified hash value.
 630          */
 631         if (!powerof2(hashsize)) {
 632                 int oldhashsize = hashsize;
 633
 634                 hashsize = maketcp_hashsize(hashsize);
 635                 /* prevent absurdly low value */
 636                 if (hashsize < 16)
 637                         hashsize = 16;
 638                 printf("%s: WARNING: TCB hash size not a power of 2, "
 639                     "clipped from %d to %d.\n", __func__, oldhashsize,
 640                     hashsize);
 641         }
 642         in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
 643             "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE);
 644
 645         /*
 646          * These have to be type stable for the benefit of the timers.
 647          */
 648         V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
 649             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 650         uma_zone_set_max(V_tcpcb_zone, maxsockets);
 651         uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
 652
 653         tcp_tw_init();
 654         syncache_init();
 655         tcp_hc_init();
 656
 657         TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
 658         V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
 659             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
 660
 661 #ifdef TCP_RFC7413
 662         tcp_fastopen_init();
 663 #endif
 664
 665         /* Skip initialization of globals for non-default instances. */
 666         if (!IS_DEFAULT_VNET(curvnet))
 667                 return;
 668
 669         tcp_reass_global_init();
 670
 671         /* XXX virtualize those bellow? */
 672         tcp_delacktime = TCPTV_DELACK;
 673         tcp_keepinit = TCPTV_KEEP_INIT;
 674         tcp_keepidle = TCPTV_KEEP_IDLE;
 675         tcp_keepintvl = TCPTV_KEEPINTVL;
 676         tcp_maxpersistidle = TCPTV_KEEP_IDLE;
 677         tcp_msl = TCPTV_MSL;
 678         tcp_rexmit_min = TCPTV_MIN;
 679         if (tcp_rexmit_min < 1)
 680                 tcp_rexmit_min = 1;
 681         tcp_persmin = TCPTV_PERSMIN;
 682         tcp_persmax = TCPTV_PERSMAX;
 683         tcp_rexmit_slop = TCPTV_CPU_VAR;
 684         tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
 685         tcp_tcbhashsize = hashsize;
 686         /* Setup the tcp function block list */
 687         init_tcp_functions();
 688         register_tcp_functions(&tcp_def_funcblk, M_WAITOK);
 689         read_random(&V_ts_offset_secret, sizeof(V_ts_offset_secret));
 690
 691         if (tcp_soreceive_stream) {
 692 #ifdef INET
 693                 tcp_usrreqs.pru_soreceive = soreceive_stream;
 694 #endif
 695 #ifdef INET6
 696                 tcp6_usrreqs.pru_soreceive = soreceive_stream;
 697 #endif /* INET6 */
 698         }
 699
 700 #ifdef INET6
 701 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
 702 #else /* INET6 */
 703 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
 704 #endif /* INET6 */
 705         if (max_protohdr < TCP_MINPROTOHDR)
 706                 max_protohdr = TCP_MINPROTOHDR;
 707         if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
 708                 panic("tcp_init");
 709 #undef TCP_MINPROTOHDR
 710
 711         ISN_LOCK_INIT();
 712         EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
 713                 SHUTDOWN_PRI_DEFAULT);
 714         EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
 715                 EVENTHANDLER_PRI_ANY);
 716 #ifdef TCPPCAP
 717         tcp_pcap_init();
 718 #endif
 719 }
 720
 721 #ifdef VIMAGE
 722 static void
 723 tcp_destroy(void *unused __unused)
 724 {
 725         int error, n;
 726
 727         /*
 728          * All our processes are gone, all our sockets should be cleaned
 729          * up, which means, we should be past the tcp_discardcb() calls.
 730          * Sleep to let all tcpcb timers really disappear and cleanup.
 731          */
 732         for (;;) {
 733                 INP_LIST_RLOCK(&V_tcbinfo);
 734                 n = V_tcbinfo.ipi_count;
 735                 INP_LIST_RUNLOCK(&V_tcbinfo);
 736                 if (n == 0)
 737                         break;
 738                 pause("tcpdes", hz / 10);
 739         }
 740         tcp_hc_destroy();
 741         syncache_destroy();
 742         tcp_tw_destroy();
 743         in_pcbinfo_destroy(&V_tcbinfo);
 744         /* tcp_discardcb() clears the sack_holes up. */
 745         uma_zdestroy(V_sack_hole_zone);
 746         uma_zdestroy(V_tcpcb_zone);
 747
 748 #ifdef TCP_RFC7413
 749         /*
 750          * Cannot free the zone until all tcpcbs are released as we attach
 751          * the allocations to them.
 752          */
 753         tcp_fastopen_destroy();
 754 #endif
 755
 756         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
 757         if (error != 0) {
 758                 printf("%s: WARNING: unable to deregister helper hook "
 759                     "type=%d, id=%d: error %d returned\n", __func__,
 760                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
 761         }
 762         error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
 763         if (error != 0) {
 764                 printf("%s: WARNING: unable to deregister helper hook "
 765                     "type=%d, id=%d: error %d returned\n", __func__,
 766                     HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
 767         }
 768 }
 769 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL);
 770 #endif
 771
 772 void
 773 tcp_fini(void *xtp)
 774 {
 775
 776 }
 777
 778 /*
 779  * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
 780  * tcp_template used to store this data in mbufs, but we now recopy it out
 781  * of the tcpcb each time to conserve mbufs.
 782  */
 783 void
 784 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
 785 {
 786         struct tcphdr *th = (struct tcphdr *)tcp_ptr;
 787
 788         INP_WLOCK_ASSERT(inp);
 789
 790 #ifdef INET6
 791         if ((inp->inp_vflag & INP_IPV6) != 0) {
 792                 struct ip6_hdr *ip6;
 793
 794                 ip6 = (struct ip6_hdr *)ip_ptr;
 795                 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
 796                         (inp->inp_flow & IPV6_FLOWINFO_MASK);
 797                 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
 798                         (IPV6_VERSION & IPV6_VERSION_MASK);
 799                 ip6->ip6_nxt = IPPROTO_TCP;
 800                 ip6->ip6_plen = htons(sizeof(struct tcphdr));
 801                 ip6->ip6_src = inp->in6p_laddr;
 802                 ip6->ip6_dst = inp->in6p_faddr;
 803         }
 804 #endif /* INET6 */
 805 #if defined(INET6) && defined(INET)
 806         else
 807 #endif
 808 #ifdef INET
 809         {
 810                 struct ip *ip;
 811
 812                 ip = (struct ip *)ip_ptr;
 813                 ip->ip_v = IPVERSION;
 814                 ip->ip_hl = 5;
 815                 ip->ip_tos = inp->inp_ip_tos;
 816                 ip->ip_len = 0;
 817                 ip->ip_id = 0;
 818                 ip->ip_off = 0;
 819                 ip->ip_ttl = inp->inp_ip_ttl;
 820                 ip->ip_sum = 0;
 821                 ip->ip_p = IPPROTO_TCP;
 822                 ip->ip_src = inp->inp_laddr;
 823                 ip->ip_dst = inp->inp_faddr;
 824         }
 825 #endif /* INET */
 826         th->th_sport = inp->inp_lport;
 827         th->th_dport = inp->inp_fport;
 828         th->th_seq = 0;
 829         th->th_ack = 0;
 830         th->th_x2 = 0;
 831         th->th_off = 5;
 832         th->th_flags = 0;
 833         th->th_win = 0;
 834         th->th_urp = 0;
 835         th->th_sum = 0;         /* in_pseudo() is called later for ipv4 */
 836 }
 837
 838 /*
 839  * Create template to be used to send tcp packets on a connection.
 840  * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
 841  * use for this function is in keepalives, which use tcp_respond.
 842  */
 843 struct tcptemp *
 844 tcpip_maketemplate(struct inpcb *inp)
 845 {
 846         struct tcptemp *t;
 847
 848         t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
 849         if (t == NULL)
 850                 return (NULL);
 851         tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
 852         return (t);
 853 }
 854
 855 /*
 856  * Send a single message to the TCP at address specified by
 857  * the given TCP/IP header.  If m == NULL, then we make a copy
 858  * of the tcpiphdr at th and send directly to the addressed host.
 859  * This is used to force keep alive messages out using the TCP
 860  * template for a connection.  If flags are given then we send
 861  * a message back to the TCP which originated the segment th,
 862  * and discard the mbuf containing it and any other attached mbufs.
 863  *
 864  * In any case the ack and sequence number of the transmitted
 865  * segment are as specified by the parameters.
 866  *
 867  * NOTE: If m != NULL, then th must point to *inside* the mbuf.
 868  */
 869 void
 870 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
 871     tcp_seq ack, tcp_seq seq, int flags)
 872 {
 873         struct tcpopt to;
 874         struct inpcb *inp;
 875         struct ip *ip;
 876         struct mbuf *optm;
 877         struct tcphdr *nth;
 878         u_char *optp;
 879 #ifdef INET6
 880         struct ip6_hdr *ip6;
 881         int isipv6;
 882 #endif /* INET6 */
 883         int optlen, tlen, win;
 884         bool incl_opts;
 885
 886         KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
 887
 888 #ifdef INET6
 889         isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
 890         ip6 = ipgen;
 891 #endif /* INET6 */
 892         ip = ipgen;
 893
 894         if (tp != NULL) {
 895                 inp = tp->t_inpcb;
 896                 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
 897                 INP_WLOCK_ASSERT(inp);
 898         } else
 899                 inp = NULL;
 900
 901         incl_opts = false;
 902         win = 0;
 903         if (tp != NULL) {
 904                 if (!(flags & TH_RST)) {
 905                         win = sbspace(&inp->inp_socket->so_rcv);
 906                         if (win > (long)TCP_MAXWIN << tp->rcv_scale)
 907                                 win = (long)TCP_MAXWIN << tp->rcv_scale;
 908                 }
 909                 if ((tp->t_flags & TF_NOOPT) == 0)
 910                         incl_opts = true;
 911         }
 912         if (m == NULL) {
 913                 m = m_gethdr(M_NOWAIT, MT_DATA);
 914                 if (m == NULL)
 915                         return;
 916                 m->m_data += max_linkhdr;
 917 #ifdef INET6
 918                 if (isipv6) {
 919                         bcopy((caddr_t)ip6, mtod(m, caddr_t),
 920                               sizeof(struct ip6_hdr));
 921                         ip6 = mtod(m, struct ip6_hdr *);
 922                         nth = (struct tcphdr *)(ip6 + 1);
 923                 } else
 924 #endif /* INET6 */
 925                 {
 926                         bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
 927                         ip = mtod(m, struct ip *);
 928                         nth = (struct tcphdr *)(ip + 1);
 929                 }
 930                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 931                 flags = TH_ACK;
 932         } else if (!M_WRITABLE(m)) {
 933                 struct mbuf *n;
 934
 935                 /* Can't reuse 'm', allocate a new mbuf. */
 936                 n = m_gethdr(M_NOWAIT, MT_DATA);
 937                 if (n == NULL) {
 938                         m_freem(m);
 939                         return;
 940                 }
 941
 942                 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
 943                         m_freem(m);
 944                         m_freem(n);
 945                         return;
 946                 }
 947
 948                 n->m_data += max_linkhdr;
 949                 /* m_len is set later */
 950 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
 951 #ifdef INET6
 952                 if (isipv6) {
 953                         bcopy((caddr_t)ip6, mtod(n, caddr_t),
 954                               sizeof(struct ip6_hdr));
 955                         ip6 = mtod(n, struct ip6_hdr *);
 956                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
 957                         nth = (struct tcphdr *)(ip6 + 1);
 958                 } else
 959 #endif /* INET6 */
 960                 {
 961                         bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
 962                         ip = mtod(n, struct ip *);
 963                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
 964                         nth = (struct tcphdr *)(ip + 1);
 965                 }
 966                 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
 967                 xchg(nth->th_dport, nth->th_sport, uint16_t);
 968                 th = nth;
 969                 m_freem(m);
 970                 m = n;
 971         } else {
 972                 /*
 973                  *  reuse the mbuf.
 974                  * XXX MRT We inherit the FIB, which is lucky.
 975                  */
 976                 m_freem(m->m_next);
 977                 m->m_next = NULL;
 978                 m->m_data = (caddr_t)ipgen;
 979                 /* m_len is set later */
 980 #ifdef INET6
 981                 if (isipv6) {
 982                         xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
 983                         nth = (struct tcphdr *)(ip6 + 1);
 984                 } else
 985 #endif /* INET6 */
 986                 {
 987                         xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
 988                         nth = (struct tcphdr *)(ip + 1);
 989                 }
 990                 if (th != nth) {
 991                         /*
 992                          * this is usually a case when an extension header
 993                          * exists between the IPv6 header and the
 994                          * TCP header.
 995                          */
 996                         nth->th_sport = th->th_sport;
 997                         nth->th_dport = th->th_dport;
 998                 }
 999                 xchg(nth->th_dport, nth->th_sport, uint16_t);
1000 #undef xchg
1001         }
1002         tlen = 0;
1003 #ifdef INET6
1004         if (isipv6)
1005                 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
1006 #endif
1007 #if defined(INET) && defined(INET6)
1008         else
1009 #endif
1010 #ifdef INET
1011                 tlen = sizeof (struct tcpiphdr);
1012 #endif
1013 #ifdef INVARIANTS
1014         m->m_len = 0;
1015         KASSERT(M_TRAILINGSPACE(m) >= tlen,
1016             ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
1017             m, tlen, (long)M_TRAILINGSPACE(m)));
1018 #endif
1019         m->m_len = tlen;
1020         to.to_flags = 0;
1021         if (incl_opts) {
1022                 /* Make sure we have room. */
1023                 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
1024                         m->m_next = m_get(M_NOWAIT, MT_DATA);
1025                         if (m->m_next) {
1026                                 optp = mtod(m->m_next, u_char *);
1027                                 optm = m->m_next;
1028                         } else
1029                                 incl_opts = false;
1030                 } else {
1031                         optp = (u_char *) (nth + 1);
1032                         optm = m;
1033                 }
1034         }
1035         if (incl_opts) {
1036                 /* Timestamps. */
1037                 if (tp->t_flags & TF_RCVD_TSTMP) {
1038                         to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
1039                         to.to_tsecr = tp->ts_recent;
1040                         to.to_flags |= TOF_TS;
1041                 }
1042 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1043                 /* TCP-MD5 (RFC2385). */
1044                 if (tp->t_flags & TF_SIGNATURE)
1045                         to.to_flags |= TOF_SIGNATURE;
1046 #endif
1047                 /* Add the options. */
1048                 tlen += optlen = tcp_addoptions(&to, optp);
1049
1050                 /* Update m_len in the correct mbuf. */
1051                 optm->m_len += optlen;
1052         } else
1053                 optlen = 0;
1054 #ifdef INET6
1055         if (isipv6) {
1056                 ip6->ip6_flow = 0;
1057                 ip6->ip6_vfc = IPV6_VERSION;
1058                 ip6->ip6_nxt = IPPROTO_TCP;
1059                 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
1060         }
1061 #endif
1062 #if defined(INET) && defined(INET6)
1063         else
1064 #endif
1065 #ifdef INET
1066         {
1067                 ip->ip_len = htons(tlen);
1068                 ip->ip_ttl = V_ip_defttl;
1069                 if (V_path_mtu_discovery)
1070                         ip->ip_off |= htons(IP_DF);
1071         }
1072 #endif
1073         m->m_pkthdr.len = tlen;
1074         m->m_pkthdr.rcvif = NULL;
1075 #ifdef MAC
1076         if (inp != NULL) {
1077                 /*
1078                  * Packet is associated with a socket, so allow the
1079                  * label of the response to reflect the socket label.
1080                  */
1081                 INP_WLOCK_ASSERT(inp);
1082                 mac_inpcb_create_mbuf(inp, m);
1083         } else {
1084                 /*
1085                  * Packet is not associated with a socket, so possibly
1086                  * update the label in place.
1087                  */
1088                 mac_netinet_tcp_reply(m);
1089         }
1090 #endif
1091         nth->th_seq = htonl(seq);
1092         nth->th_ack = htonl(ack);
1093         nth->th_x2 = 0;
1094         nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
1095         nth->th_flags = flags;
1096         if (tp != NULL)
1097                 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
1098         else
1099                 nth->th_win = htons((u_short)win);
1100         nth->th_urp = 0;
1101
1102 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1103         if (to.to_flags & TOF_SIGNATURE) {
1104                 if (!TCPMD5_ENABLED() ||
1105                     TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) {
1106                         m_freem(m);
1107                         return;
1108                 }
1109         }
1110 #endif
1111
1112         m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1113 #ifdef INET6
1114         if (isipv6) {
1115                 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
1116                 nth->th_sum = in6_cksum_pseudo(ip6,
1117                     tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
1118                 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
1119                     NULL, NULL);
1120         }
1121 #endif /* INET6 */
1122 #if defined(INET6) && defined(INET)
1123         else
1124 #endif
1125 #ifdef INET
1126         {
1127                 m->m_pkthdr.csum_flags = CSUM_TCP;
1128                 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1129                     htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
1130         }
1131 #endif /* INET */
1132 #ifdef TCPDEBUG
1133         if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
1134                 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
1135 #endif
1136         TCP_PROBE3(debug__output, tp, th, m);
1137         if (flags & TH_RST)
1138                 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth);
1139
1140 #ifdef INET6
1141         if (isipv6) {
1142                 TCP_PROBE5(send, NULL, tp, ip6, tp, nth);
1143                 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
1144         }
1145 #endif /* INET6 */
1146 #if defined(INET) && defined(INET6)
1147         else
1148 #endif
1149 #ifdef INET
1150         {
1151                 TCP_PROBE5(send, NULL, tp, ip, tp, nth);
1152                 (void)ip_output(m, NULL, NULL, 0, NULL, inp);
1153         }
1154 #endif
1155 }
1156
1157 /*
1158  * Create a new TCP control block, making an
1159  * empty reassembly queue and hooking it to the argument
1160  * protocol control block.  The `inp' parameter must have
1161  * come from the zone allocator set up in tcp_init().
1162  */
1163 struct tcpcb *
1164 tcp_newtcpcb(struct inpcb *inp)
1165 {
1166         struct tcpcb_mem *tm;
1167         struct tcpcb *tp;
1168 #ifdef INET6
1169         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1170 #endif /* INET6 */
1171
1172         tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
1173         if (tm == NULL)
1174                 return (NULL);
1175         tp = &tm->tcb;
1176
1177         /* Initialise cc_var struct for this tcpcb. */
1178         tp->ccv = &tm->ccv;
1179         tp->ccv->type = IPPROTO_TCP;
1180         tp->ccv->ccvc.tcp = tp;
1181         rw_rlock(&tcp_function_lock);
1182         tp->t_fb = tcp_func_set_ptr;
1183         refcount_acquire(&tp->t_fb->tfb_refcnt);
1184         rw_runlock(&tcp_function_lock);
1185         if (tp->t_fb->tfb_tcp_fb_init) {
1186                 (*tp->t_fb->tfb_tcp_fb_init)(tp);
1187         }
1188         /*
1189          * Use the current system default CC algorithm.
1190          */
1191         CC_LIST_RLOCK();
1192         KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
1193         CC_ALGO(tp) = CC_DEFAULT();
1194         CC_LIST_RUNLOCK();
1195
1196         if (CC_ALGO(tp)->cb_init != NULL)
1197                 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
1198                         if (tp->t_fb->tfb_tcp_fb_fini)
1199                                 (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1200                         refcount_release(&tp->t_fb->tfb_refcnt);
1201                         uma_zfree(V_tcpcb_zone, tm);
1202                         return (NULL);
1203                 }
1204
1205         tp->osd = &tm->osd;
1206         if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
1207                 if (tp->t_fb->tfb_tcp_fb_fini)
1208                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1209                 refcount_release(&tp->t_fb->tfb_refcnt);
1210                 uma_zfree(V_tcpcb_zone, tm);
1211                 return (NULL);
1212         }
1213
1214 #ifdef VIMAGE
1215         tp->t_vnet = inp->inp_vnet;
1216 #endif
1217         tp->t_timers = &tm->tt;
1218         TAILQ_INIT(&tp->t_segq);
1219         tp->t_maxseg =
1220 #ifdef INET6
1221                 isipv6 ? V_tcp_v6mssdflt :
1222 #endif /* INET6 */
1223                 V_tcp_mssdflt;
1224
1225         /* Set up our timeouts. */
1226         callout_init(&tp->t_timers->tt_rexmt, 1);
1227         callout_init(&tp->t_timers->tt_persist, 1);
1228         callout_init(&tp->t_timers->tt_keep, 1);
1229         callout_init(&tp->t_timers->tt_2msl, 1);
1230         callout_init(&tp->t_timers->tt_delack, 1);
1231
1232         if (V_tcp_do_rfc1323)
1233                 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1234         if (V_tcp_do_sack)
1235                 tp->t_flags |= TF_SACK_PERMIT;
1236         TAILQ_INIT(&tp->snd_holes);
1237         /*
1238          * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
1239          * is called.
1240          */
1241         in_pcbref(inp); /* Reference for tcpcb */
1242         tp->t_inpcb = inp;
1243
1244         /*
1245          * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1246          * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
1247          * reasonable initial retransmit time.
1248          */
1249         tp->t_srtt = TCPTV_SRTTBASE;
1250         tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1251         tp->t_rttmin = tcp_rexmit_min;
1252         tp->t_rxtcur = TCPTV_RTOBASE;
1253         tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1254         tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1255         tp->t_rcvtime = ticks;
1256         /*
1257          * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1258          * because the socket may be bound to an IPv6 wildcard address,
1259          * which may match an IPv4-mapped IPv6 address.
1260          */
1261         inp->inp_ip_ttl = V_ip_defttl;
1262         inp->inp_ppcb = tp;
1263 #ifdef TCPPCAP
1264         /*
1265          * Init the TCP PCAP queues.
1266          */
1267         tcp_pcap_tcpcb_init(tp);
1268 #endif
1269         return (tp);            /* XXX */
1270 }
1271
1272 /*
1273  * Switch the congestion control algorithm back to NewReno for any active
1274  * control blocks using an algorithm which is about to go away.
1275  * This ensures the CC framework can allow the unload to proceed without leaving
1276  * any dangling pointers which would trigger a panic.
1277  * Returning non-zero would inform the CC framework that something went wrong
1278  * and it would be unsafe to allow the unload to proceed. However, there is no
1279  * way for this to occur with this implementation so we always return zero.
1280  */
1281 int
1282 tcp_ccalgounload(struct cc_algo *unload_algo)
1283 {
1284         struct cc_algo *tmpalgo;
1285         struct inpcb *inp;
1286         struct tcpcb *tp;
1287         VNET_ITERATOR_DECL(vnet_iter);
1288
1289         /*
1290          * Check all active control blocks across all network stacks and change
1291          * any that are using "unload_algo" back to NewReno. If "unload_algo"
1292          * requires cleanup code to be run, call it.
1293          */
1294         VNET_LIST_RLOCK();
1295         VNET_FOREACH(vnet_iter) {
1296                 CURVNET_SET(vnet_iter);
1297                 INP_INFO_WLOCK(&V_tcbinfo);
1298                 /*
1299                  * New connections already part way through being initialised
1300                  * with the CC algo we're removing will not race with this code
1301                  * because the INP_INFO_WLOCK is held during initialisation. We
1302                  * therefore don't enter the loop below until the connection
1303                  * list has stabilised.
1304                  */
1305                 LIST_FOREACH(inp, &V_tcb, inp_list) {
1306                         INP_WLOCK(inp);
1307                         /* Important to skip tcptw structs. */
1308                         if (!(inp->inp_flags & INP_TIMEWAIT) &&
1309                             (tp = intotcpcb(inp)) != NULL) {
1310                                 /*
1311                                  * By holding INP_WLOCK here, we are assured
1312                                  * that the connection is not currently
1313                                  * executing inside the CC module's functions
1314                                  * i.e. it is safe to make the switch back to
1315                                  * NewReno.
1316                                  */
1317                                 if (CC_ALGO(tp) == unload_algo) {
1318                                         tmpalgo = CC_ALGO(tp);
1319                                         /* NewReno does not require any init. */
1320                                         CC_ALGO(tp) = &newreno_cc_algo;
1321                                         if (tmpalgo->cb_destroy != NULL)
1322                                                 tmpalgo->cb_destroy(tp->ccv);
1323                                 }
1324                         }
1325                         INP_WUNLOCK(inp);
1326                 }
1327                 INP_INFO_WUNLOCK(&V_tcbinfo);
1328                 CURVNET_RESTORE();
1329         }
1330         VNET_LIST_RUNLOCK();
1331
1332         return (0);
1333 }
1334
1335 /*
1336  * Drop a TCP connection, reporting
1337  * the specified error.  If connection is synchronized,
1338  * then send a RST to peer.
1339  */
1340 struct tcpcb *
1341 tcp_drop(struct tcpcb *tp, int errno)
1342 {
1343         struct socket *so = tp->t_inpcb->inp_socket;
1344
1345         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1346         INP_WLOCK_ASSERT(tp->t_inpcb);
1347
1348         if (TCPS_HAVERCVDSYN(tp->t_state)) {
1349                 tcp_state_change(tp, TCPS_CLOSED);
1350                 (void) tp->t_fb->tfb_tcp_output(tp);
1351                 TCPSTAT_INC(tcps_drops);
1352         } else
1353                 TCPSTAT_INC(tcps_conndrops);
1354         if (errno == ETIMEDOUT && tp->t_softerror)
1355                 errno = tp->t_softerror;
1356         so->so_error = errno;
1357         return (tcp_close(tp));
1358 }
1359
1360 void
1361 tcp_discardcb(struct tcpcb *tp)
1362 {
1363         struct inpcb *inp = tp->t_inpcb;
1364         struct socket *so = inp->inp_socket;
1365 #ifdef INET6
1366         int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1367 #endif /* INET6 */
1368         int released;
1369
1370         INP_WLOCK_ASSERT(inp);
1371
1372         /*
1373          * Make sure that all of our timers are stopped before we delete the
1374          * PCB.
1375          *
1376          * If stopping a timer fails, we schedule a discard function in same
1377          * callout, and the last discard function called will take care of
1378          * deleting the tcpcb.
1379          */
1380         tp->t_timers->tt_draincnt = 0;
1381         tcp_timer_stop(tp, TT_REXMT);
1382         tcp_timer_stop(tp, TT_PERSIST);
1383         tcp_timer_stop(tp, TT_KEEP);
1384         tcp_timer_stop(tp, TT_2MSL);
1385         tcp_timer_stop(tp, TT_DELACK);
1386         if (tp->t_fb->tfb_tcp_timer_stop_all) {
1387                 /*
1388                  * Call the stop-all function of the methods,
1389                  * this function should call the tcp_timer_stop()
1390                  * method with each of the function specific timeouts.
1391                  * That stop will be called via the tfb_tcp_timer_stop()
1392                  * which should use the async drain function of the
1393                  * callout system (see tcp_var.h).
1394                  */
1395                 tp->t_fb->tfb_tcp_timer_stop_all(tp);
1396         }
1397
1398         /*
1399          * If we got enough samples through the srtt filter,
1400          * save the rtt and rttvar in the routing entry.
1401          * 'Enough' is arbitrarily defined as 4 rtt samples.
1402          * 4 samples is enough for the srtt filter to converge
1403          * to within enough % of the correct value; fewer samples
1404          * and we could save a bogus rtt. The danger is not high
1405          * as tcp quickly recovers from everything.
1406          * XXX: Works very well but needs some more statistics!
1407          */
1408         if (tp->t_rttupdated >= 4) {
1409                 struct hc_metrics_lite metrics;
1410                 u_long ssthresh;
1411
1412                 bzero(&metrics, sizeof(metrics));
1413                 /*
1414                  * Update the ssthresh always when the conditions below
1415                  * are satisfied. This gives us better new start value
1416                  * for the congestion avoidance for new connections.
1417                  * ssthresh is only set if packet loss occurred on a session.
1418                  *
1419                  * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1420                  * being torn down.  Ideally this code would not use 'so'.
1421                  */
1422                 ssthresh = tp->snd_ssthresh;
1423                 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1424                         /*
1425                          * convert the limit from user data bytes to
1426                          * packets then to packet data bytes.
1427                          */
1428                         ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1429                         if (ssthresh < 2)
1430                                 ssthresh = 2;
1431                         ssthresh *= (u_long)(tp->t_maxseg +
1432 #ifdef INET6
1433                             (isipv6 ? sizeof (struct ip6_hdr) +
1434                                 sizeof (struct tcphdr) :
1435 #endif
1436                                 sizeof (struct tcpiphdr)
1437 #ifdef INET6
1438                             )
1439 #endif
1440                             );
1441                 } else
1442                         ssthresh = 0;
1443                 metrics.rmx_ssthresh = ssthresh;
1444
1445                 metrics.rmx_rtt = tp->t_srtt;
1446                 metrics.rmx_rttvar = tp->t_rttvar;
1447                 metrics.rmx_cwnd = tp->snd_cwnd;
1448                 metrics.rmx_sendpipe = 0;
1449                 metrics.rmx_recvpipe = 0;
1450
1451                 tcp_hc_update(&inp->inp_inc, &metrics);
1452         }
1453
1454         /* free the reassembly queue, if any */
1455         tcp_reass_flush(tp);
1456
1457 #ifdef TCP_OFFLOAD
1458         /* Disconnect offload device, if any. */
1459         if (tp->t_flags & TF_TOE)
1460                 tcp_offload_detach(tp);
1461 #endif
1462
1463         tcp_free_sackholes(tp);
1464
1465 #ifdef TCPPCAP
1466         /* Free the TCP PCAP queues. */
1467         tcp_pcap_drain(&(tp->t_inpkts));
1468         tcp_pcap_drain(&(tp->t_outpkts));
1469 #endif
1470
1471         /* Allow the CC algorithm to clean up after itself. */
1472         if (CC_ALGO(tp)->cb_destroy != NULL)
1473                 CC_ALGO(tp)->cb_destroy(tp->ccv);
1474
1475         khelp_destroy_osd(tp->osd);
1476
1477         CC_ALGO(tp) = NULL;
1478         inp->inp_ppcb = NULL;
1479         if (tp->t_timers->tt_draincnt == 0) {
1480                 /* We own the last reference on tcpcb, let's free it. */
1481                 TCPSTATES_DEC(tp->t_state);
1482                 if (tp->t_fb->tfb_tcp_fb_fini)
1483                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1484                 refcount_release(&tp->t_fb->tfb_refcnt);
1485                 tp->t_inpcb = NULL;
1486                 uma_zfree(V_tcpcb_zone, tp);
1487                 released = in_pcbrele_wlocked(inp);
1488                 KASSERT(!released, ("%s: inp %p should not have been released "
1489                         "here", __func__, inp));
1490         }
1491 }
1492
1493 void
1494 tcp_timer_discard(void *ptp)
1495 {
1496         struct inpcb *inp;
1497         struct tcpcb *tp;
1498
1499         tp = (struct tcpcb *)ptp;
1500         CURVNET_SET(tp->t_vnet);
1501         INP_INFO_RLOCK(&V_tcbinfo);
1502         inp = tp->t_inpcb;
1503         KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1504                 __func__, tp));
1505         INP_WLOCK(inp);
1506         KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1507                 ("%s: tcpcb has to be stopped here", __func__));
1508         tp->t_timers->tt_draincnt--;
1509         if (tp->t_timers->tt_draincnt == 0) {
1510                 /* We own the last reference on this tcpcb, let's free it. */
1511                 TCPSTATES_DEC(tp->t_state);
1512                 if (tp->t_fb->tfb_tcp_fb_fini)
1513                         (*tp->t_fb->tfb_tcp_fb_fini)(tp);
1514                 refcount_release(&tp->t_fb->tfb_refcnt);
1515                 tp->t_inpcb = NULL;
1516                 uma_zfree(V_tcpcb_zone, tp);
1517                 if (in_pcbrele_wlocked(inp)) {
1518                         INP_INFO_RUNLOCK(&V_tcbinfo);
1519                         CURVNET_RESTORE();
1520                         return;
1521                 }
1522         }
1523         INP_WUNLOCK(inp);
1524         INP_INFO_RUNLOCK(&V_tcbinfo);
1525         CURVNET_RESTORE();
1526 }
1527
1528 /*
1529  * Attempt to close a TCP control block, marking it as dropped, and freeing
1530  * the socket if we hold the only reference.
1531  */
1532 struct tcpcb *
1533 tcp_close(struct tcpcb *tp)
1534 {
1535         struct inpcb *inp = tp->t_inpcb;
1536         struct socket *so;
1537
1538         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1539         INP_WLOCK_ASSERT(inp);
1540
1541 #ifdef TCP_OFFLOAD
1542         if (tp->t_state == TCPS_LISTEN)
1543                 tcp_offload_listen_stop(tp);
1544 #endif
1545 #ifdef TCP_RFC7413
1546         /*
1547          * This releases the TFO pending counter resource for TFO listen
1548          * sockets as well as passively-created TFO sockets that transition
1549          * from SYN_RECEIVED to CLOSED.
1550          */
1551         if (tp->t_tfo_pending) {
1552                 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1553                 tp->t_tfo_pending = NULL;
1554         }
1555 #endif
1556         in_pcbdrop(inp);
1557         TCPSTAT_INC(tcps_closed);
1558         if (tp->t_state != TCPS_CLOSED)
1559                 tcp_state_change(tp, TCPS_CLOSED);
1560         KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1561         so = inp->inp_socket;
1562         soisdisconnected(so);
1563         if (inp->inp_flags & INP_SOCKREF) {
1564                 KASSERT(so->so_state & SS_PROTOREF,
1565                     ("tcp_close: !SS_PROTOREF"));
1566                 inp->inp_flags &= ~INP_SOCKREF;
1567                 INP_WUNLOCK(inp);
1568                 ACCEPT_LOCK();
1569                 SOCK_LOCK(so);
1570                 so->so_state &= ~SS_PROTOREF;
1571                 sofree(so);
1572                 return (NULL);
1573         }
1574         return (tp);
1575 }
1576
1577 void
1578 tcp_drain(void)
1579 {
1580         VNET_ITERATOR_DECL(vnet_iter);
1581
1582         if (!do_tcpdrain)
1583                 return;
1584
1585         VNET_LIST_RLOCK_NOSLEEP();
1586         VNET_FOREACH(vnet_iter) {
1587                 CURVNET_SET(vnet_iter);
1588                 struct inpcb *inpb;
1589                 struct tcpcb *tcpb;
1590
1591         /*
1592          * Walk the tcpbs, if existing, and flush the reassembly queue,
1593          * if there is one...
1594          * XXX: The "Net/3" implementation doesn't imply that the TCP
1595          *      reassembly queue should be flushed, but in a situation
1596          *      where we're really low on mbufs, this is potentially
1597          *      useful.
1598          */
1599                 INP_INFO_WLOCK(&V_tcbinfo);
1600                 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1601                         if (inpb->inp_flags & INP_TIMEWAIT)
1602                                 continue;
1603                         INP_WLOCK(inpb);
1604                         if ((tcpb = intotcpcb(inpb)) != NULL) {
1605                                 tcp_reass_flush(tcpb);
1606                                 tcp_clean_sackreport(tcpb);
1607 #ifdef TCPPCAP
1608                                 if (tcp_pcap_aggressive_free) {
1609                                         /* Free the TCP PCAP queues. */
1610                                         tcp_pcap_drain(&(tcpb->t_inpkts));
1611                                         tcp_pcap_drain(&(tcpb->t_outpkts));
1612                                 }
1613 #endif
1614                         }
1615                         INP_WUNLOCK(inpb);
1616                 }
1617                 INP_INFO_WUNLOCK(&V_tcbinfo);
1618                 CURVNET_RESTORE();
1619         }
1620         VNET_LIST_RUNLOCK_NOSLEEP();
1621 }
1622
1623 /*
1624  * Notify a tcp user of an asynchronous error;
1625  * store error as soft error, but wake up user
1626  * (for now, won't do anything until can select for soft error).
1627  *
1628  * Do not wake up user since there currently is no mechanism for
1629  * reporting soft errors (yet - a kqueue filter may be added).
1630  */
1631 static struct inpcb *
1632 tcp_notify(struct inpcb *inp, int error)
1633 {
1634         struct tcpcb *tp;
1635
1636         INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1637         INP_WLOCK_ASSERT(inp);
1638
1639         if ((inp->inp_flags & INP_TIMEWAIT) ||
1640             (inp->inp_flags & INP_DROPPED))
1641                 return (inp);
1642
1643         tp = intotcpcb(inp);
1644         KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1645
1646         /*
1647          * Ignore some errors if we are hooked up.
1648          * If connection hasn't completed, has retransmitted several times,
1649          * and receives a second error, give up now.  This is better
1650          * than waiting a long time to establish a connection that
1651          * can never complete.
1652          */
1653         if (tp->t_state == TCPS_ESTABLISHED &&
1654             (error == EHOSTUNREACH || error == ENETUNREACH ||
1655              error == EHOSTDOWN)) {
1656                 if (inp->inp_route.ro_rt) {
1657                         RTFREE(inp->inp_route.ro_rt);
1658                         inp->inp_route.ro_rt = (struct rtentry *)NULL;
1659                 }
1660                 return (inp);
1661         } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1662             tp->t_softerror) {
1663                 tp = tcp_drop(tp, error);
1664                 if (tp != NULL)
1665                         return (inp);
1666                 else
1667                         return (NULL);
1668         } else {
1669                 tp->t_softerror = error;
1670                 return (inp);
1671         }
1672 #if 0
1673         wakeup( &so->so_timeo);
1674         sorwakeup(so);
1675         sowwakeup(so);
1676 #endif
1677 }
1678
1679 static int
1680 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1681 {
1682         int error, i, m, n, pcb_count;
1683         struct inpcb *inp, **inp_list;
1684         inp_gen_t gencnt;
1685         struct xinpgen xig;
1686
1687         /*
1688          * The process of preparing the TCB list is too time-consuming and
1689          * resource-intensive to repeat twice on every request.
1690          */
1691         if (req->oldptr == NULL) {
1692                 n = V_tcbinfo.ipi_count +
1693                     counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1694                 n += imax(n / 8, 10);
1695                 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1696                 return (0);
1697         }
1698
1699         if (req->newptr != NULL)
1700                 return (EPERM);
1701
1702         /*
1703          * OK, now we're committed to doing something.
1704          */
1705         INP_LIST_RLOCK(&V_tcbinfo);
1706         gencnt = V_tcbinfo.ipi_gencnt;
1707         n = V_tcbinfo.ipi_count;
1708         INP_LIST_RUNLOCK(&V_tcbinfo);
1709
1710         m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]);
1711
1712         error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1713                 + (n + m) * sizeof(struct xtcpcb));
1714         if (error != 0)
1715                 return (error);
1716
1717         bzero(&xig, sizeof(xig));
1718         xig.xig_len = sizeof xig;
1719         xig.xig_count = n + m;
1720         xig.xig_gen = gencnt;
1721         xig.xig_sogen = so_gencnt;
1722         error = SYSCTL_OUT(req, &xig, sizeof xig);
1723         if (error)
1724                 return (error);
1725
1726         error = syncache_pcblist(req, m, &pcb_count);
1727         if (error)
1728                 return (error);
1729
1730         inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1731
1732         INP_INFO_WLOCK(&V_tcbinfo);
1733         for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1734             inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1735                 INP_WLOCK(inp);
1736                 if (inp->inp_gencnt <= gencnt) {
1737                         /*
1738                          * XXX: This use of cr_cansee(), introduced with
1739                          * TCP state changes, is not quite right, but for
1740                          * now, better than nothing.
1741                          */
1742                         if (inp->inp_flags & INP_TIMEWAIT) {
1743                                 if (intotw(inp) != NULL)
1744                                         error = cr_cansee(req->td->td_ucred,
1745                                             intotw(inp)->tw_cred);
1746                                 else
1747                                         error = EINVAL; /* Skip this inp. */
1748                         } else
1749                                 error = cr_canseeinpcb(req->td->td_ucred, inp);
1750                         if (error == 0) {
1751                                 in_pcbref(inp);
1752                                 inp_list[i++] = inp;
1753                         }
1754                 }
1755                 INP_WUNLOCK(inp);
1756         }
1757         INP_INFO_WUNLOCK(&V_tcbinfo);
1758         n = i;
1759
1760         error = 0;
1761         for (i = 0; i < n; i++) {
1762                 inp = inp_list[i];
1763                 INP_RLOCK(inp);
1764                 if (inp->inp_gencnt <= gencnt) {
1765                         struct xtcpcb xt;
1766                         void *inp_ppcb;
1767
1768                         bzero(&xt, sizeof(xt));
1769                         xt.xt_len = sizeof xt;
1770                         /* XXX should avoid extra copy */
1771                         bcopy(inp, &xt.xt_inp, sizeof *inp);
1772                         inp_ppcb = inp->inp_ppcb;
1773                         if (inp_ppcb == NULL)
1774                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1775                         else if (inp->inp_flags & INP_TIMEWAIT) {
1776                                 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1777                                 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1778                         } else {
1779                                 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1780                                 if (xt.xt_tp.t_timers)
1781                                         tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1782                         }
1783                         if (inp->inp_socket != NULL)
1784                                 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1785                         else {
1786                                 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1787                                 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1788                         }
1789                         xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1790                         INP_RUNLOCK(inp);
1791                         error = SYSCTL_OUT(req, &xt, sizeof xt);
1792                 } else
1793                         INP_RUNLOCK(inp);
1794         }
1795         INP_INFO_RLOCK(&V_tcbinfo);
1796         for (i = 0; i < n; i++) {
1797                 inp = inp_list[i];
1798                 INP_RLOCK(inp);
1799                 if (!in_pcbrele_rlocked(inp))
1800                         INP_RUNLOCK(inp);
1801         }
1802         INP_INFO_RUNLOCK(&V_tcbinfo);
1803
1804         if (!error) {
1805                 /*
1806                  * Give the user an updated idea of our state.
1807                  * If the generation differs from what we told
1808                  * her before, she knows that something happened
1809                  * while we were processing this request, and it
1810                  * might be necessary to retry.
1811                  */
1812                 INP_LIST_RLOCK(&V_tcbinfo);
1813                 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1814                 xig.xig_sogen = so_gencnt;
1815                 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1816                 INP_LIST_RUNLOCK(&V_tcbinfo);
1817                 error = SYSCTL_OUT(req, &xig, sizeof xig);
1818         }
1819         free(inp_list, M_TEMP);
1820         return (error);
1821 }
1822
1823 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1824     CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1825     tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1826
1827 #ifdef INET
1828 static int
1829 tcp_getcred(SYSCTL_HANDLER_ARGS)
1830 {
1831         struct xucred xuc;
1832         struct sockaddr_in addrs[2];
1833         struct inpcb *inp;
1834         int error;
1835
1836         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1837         if (error)
1838                 return (error);
1839         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1840         if (error)
1841                 return (error);
1842         inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1843             addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1844         if (inp != NULL) {
1845                 if (inp->inp_socket == NULL)
1846                         error = ENOENT;
1847                 if (error == 0)
1848                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1849                 if (error == 0)
1850                         cru2x(inp->inp_cred, &xuc);
1851                 INP_RUNLOCK(inp);
1852         } else
1853                 error = ENOENT;
1854         if (error == 0)
1855                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1856         return (error);
1857 }
1858
1859 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1860     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1861     tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1862 #endif /* INET */
1863
1864 #ifdef INET6
1865 static int
1866 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1867 {
1868         struct xucred xuc;
1869         struct sockaddr_in6 addrs[2];
1870         struct inpcb *inp;
1871         int error;
1872 #ifdef INET
1873         int mapped = 0;
1874 #endif
1875
1876         error = priv_check(req->td, PRIV_NETINET_GETCRED);
1877         if (error)
1878                 return (error);
1879         error = SYSCTL_IN(req, addrs, sizeof(addrs));
1880         if (error)
1881                 return (error);
1882         if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1883             (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1884                 return (error);
1885         }
1886         if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1887 #ifdef INET
1888                 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1889                         mapped = 1;
1890                 else
1891 #endif
1892                         return (EINVAL);
1893         }
1894
1895 #ifdef INET
1896         if (mapped == 1)
1897                 inp = in_pcblookup(&V_tcbinfo,
1898                         *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1899                         addrs[1].sin6_port,
1900                         *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1901                         addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1902         else
1903 #endif
1904                 inp = in6_pcblookup(&V_tcbinfo,
1905                         &addrs[1].sin6_addr, addrs[1].sin6_port,
1906                         &addrs[0].sin6_addr, addrs[0].sin6_port,
1907                         INPLOOKUP_RLOCKPCB, NULL);
1908         if (inp != NULL) {
1909                 if (inp->inp_socket == NULL)
1910                         error = ENOENT;
1911                 if (error == 0)
1912                         error = cr_canseeinpcb(req->td->td_ucred, inp);
1913                 if (error == 0)
1914                         cru2x(inp->inp_cred, &xuc);
1915                 INP_RUNLOCK(inp);
1916         } else
1917                 error = ENOENT;
1918         if (error == 0)
1919                 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1920         return (error);
1921 }
1922
1923 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1924     CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1925     tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1926 #endif /* INET6 */
1927
1928
1929 #ifdef INET
1930 void
1931 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1932 {
1933         struct ip *ip = vip;
1934         struct tcphdr *th;
1935         struct in_addr faddr;
1936         struct inpcb *inp;
1937         struct tcpcb *tp;
1938         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1939         struct icmp *icp;
1940         struct in_conninfo inc;
1941         tcp_seq icmp_tcp_seq;
1942         int mtu;
1943
1944         faddr = ((struct sockaddr_in *)sa)->sin_addr;
1945         if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1946                 return;
1947
1948         if (cmd == PRC_MSGSIZE)
1949                 notify = tcp_mtudisc_notify;
1950         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1951                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
1952                 cmd == PRC_TIMXCEED_INTRANS) && ip)
1953                 notify = tcp_drop_syn_sent;
1954
1955         /*
1956          * Hostdead is ugly because it goes linearly through all PCBs.
1957          * XXX: We never get this from ICMP, otherwise it makes an
1958          * excellent DoS attack on machines with many connections.
1959          */
1960         else if (cmd == PRC_HOSTDEAD)
1961                 ip = NULL;
1962         else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1963                 return;
1964
1965         if (ip == NULL) {
1966                 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1967                 return;
1968         }
1969
1970         icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1971         th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1972         INP_INFO_RLOCK(&V_tcbinfo);
1973         inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1974             th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1975         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
1976                 /* signal EHOSTDOWN, as it flushes the cached route */
1977                 inp = (*notify)(inp, EHOSTDOWN);
1978                 goto out;
1979         }
1980         icmp_tcp_seq = th->th_seq;
1981         if (inp != NULL)  {
1982                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1983                     !(inp->inp_flags & INP_DROPPED) &&
1984                     !(inp->inp_socket == NULL)) {
1985                         tp = intotcpcb(inp);
1986                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
1987                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
1988                                 if (cmd == PRC_MSGSIZE) {
1989                                         /*
1990                                          * MTU discovery:
1991                                          * If we got a needfrag set the MTU
1992                                          * in the route to the suggested new
1993                                          * value (if given) and then notify.
1994                                          */
1995                                         mtu = ntohs(icp->icmp_nextmtu);
1996                                         /*
1997                                          * If no alternative MTU was
1998                                          * proposed, try the next smaller
1999                                          * one.
2000                                          */
2001                                         if (!mtu)
2002                                                 mtu = ip_next_mtu(
2003                                                     ntohs(ip->ip_len), 1);
2004                                         if (mtu < V_tcp_minmss +
2005                                             sizeof(struct tcpiphdr))
2006                                                 mtu = V_tcp_minmss +
2007                                                     sizeof(struct tcpiphdr);
2008                                         /*
2009                                          * Only process the offered MTU if it
2010                                          * is smaller than the current one.
2011                                          */
2012                                         if (mtu < tp->t_maxseg +
2013                                             sizeof(struct tcpiphdr)) {
2014                                                 bzero(&inc, sizeof(inc));
2015                                                 inc.inc_faddr = faddr;
2016                                                 inc.inc_fibnum =
2017                                                     inp->inp_inc.inc_fibnum;
2018                                                 tcp_hc_updatemtu(&inc, mtu);
2019                                                 tcp_mtudisc(inp, mtu);
2020                                         }
2021                                 } else
2022                                         inp = (*notify)(inp,
2023                                             inetctlerrmap[cmd]);
2024                         }
2025                 }
2026         } else {
2027                 bzero(&inc, sizeof(inc));
2028                 inc.inc_fport = th->th_dport;
2029                 inc.inc_lport = th->th_sport;
2030                 inc.inc_faddr = faddr;
2031                 inc.inc_laddr = ip->ip_src;
2032                 syncache_unreach(&inc, icmp_tcp_seq);
2033         }
2034 out:
2035         if (inp != NULL)
2036                 INP_WUNLOCK(inp);
2037         INP_INFO_RUNLOCK(&V_tcbinfo);
2038 }
2039 #endif /* INET */
2040
2041 #ifdef INET6
2042 void
2043 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
2044 {
2045         struct in6_addr *dst;
2046         struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
2047         struct ip6_hdr *ip6;
2048         struct mbuf *m;
2049         struct inpcb *inp;
2050         struct tcpcb *tp;
2051         struct icmp6_hdr *icmp6;
2052         struct ip6ctlparam *ip6cp = NULL;
2053         const struct sockaddr_in6 *sa6_src = NULL;
2054         struct in_conninfo inc;
2055         struct tcp_ports {
2056                 uint16_t th_sport;
2057                 uint16_t th_dport;
2058         } t_ports;
2059         tcp_seq icmp_tcp_seq;
2060         unsigned int mtu;
2061         unsigned int off;
2062
2063         if (sa->sa_family != AF_INET6 ||
2064             sa->sa_len != sizeof(struct sockaddr_in6))
2065                 return;
2066
2067         /* if the parameter is from icmp6, decode it. */
2068         if (d != NULL) {
2069                 ip6cp = (struct ip6ctlparam *)d;
2070                 icmp6 = ip6cp->ip6c_icmp6;
2071                 m = ip6cp->ip6c_m;
2072                 ip6 = ip6cp->ip6c_ip6;
2073                 off = ip6cp->ip6c_off;
2074                 sa6_src = ip6cp->ip6c_src;
2075                 dst = ip6cp->ip6c_finaldst;
2076         } else {
2077                 m = NULL;
2078                 ip6 = NULL;
2079                 off = 0;        /* fool gcc */
2080                 sa6_src = &sa6_any;
2081                 dst = NULL;
2082         }
2083
2084         if (cmd == PRC_MSGSIZE)
2085                 notify = tcp_mtudisc_notify;
2086         else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2087                 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2088                 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL)
2089                 notify = tcp_drop_syn_sent;
2090
2091         /*
2092          * Hostdead is ugly because it goes linearly through all PCBs.
2093          * XXX: We never get this from ICMP, otherwise it makes an
2094          * excellent DoS attack on machines with many connections.
2095          */
2096         else if (cmd == PRC_HOSTDEAD)
2097                 ip6 = NULL;
2098         else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)
2099                 return;
2100
2101         if (ip6 == NULL) {
2102                 in6_pcbnotify(&V_tcbinfo, sa, 0,
2103                               (const struct sockaddr *)sa6_src,
2104                               0, cmd, NULL, notify);
2105                 return;
2106         }
2107
2108         /* Check if we can safely get the ports from the tcp hdr */
2109         if (m == NULL ||
2110             (m->m_pkthdr.len <
2111                 (int32_t) (off + sizeof(struct tcp_ports)))) {
2112                 return;
2113         }
2114         bzero(&t_ports, sizeof(struct tcp_ports));
2115         m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2116         INP_INFO_RLOCK(&V_tcbinfo);
2117         inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2118             &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL);
2119         if (inp != NULL && PRC_IS_REDIRECT(cmd)) {
2120                 /* signal EHOSTDOWN, as it flushes the cached route */
2121                 inp = (*notify)(inp, EHOSTDOWN);
2122                 goto out;
2123         }
2124         off += sizeof(struct tcp_ports);
2125         if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2126                 goto out;
2127         }
2128         m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2129         if (inp != NULL)  {
2130                 if (!(inp->inp_flags & INP_TIMEWAIT) &&
2131                     !(inp->inp_flags & INP_DROPPED) &&
2132                     !(inp->inp_socket == NULL)) {
2133                         tp = intotcpcb(inp);
2134                         if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) &&
2135                             SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) {
2136                                 if (cmd == PRC_MSGSIZE) {
2137                                         /*
2138                                          * MTU discovery:
2139                                          * If we got a needfrag set the MTU
2140                                          * in the route to the suggested new
2141                                          * value (if given) and then notify.
2142                                          */
2143                                         mtu = ntohl(icmp6->icmp6_mtu);
2144                                         /*
2145                                          * If no alternative MTU was
2146                                          * proposed, or the proposed
2147                                          * MTU was too small, set to
2148                                          * the min.
2149                                          */
2150                                         if (mtu < IPV6_MMTU)
2151                                                 mtu = IPV6_MMTU - 8;
2152                                         bzero(&inc, sizeof(inc));
2153                                         inc.inc_fibnum = M_GETFIB(m);
2154                                         inc.inc_flags |= INC_ISIPV6;
2155                                         inc.inc6_faddr = *dst;
2156                                         if (in6_setscope(&inc.inc6_faddr,
2157                                                 m->m_pkthdr.rcvif, NULL))
2158                                                 goto out;
2159                                         /*
2160                                          * Only process the offered MTU if it
2161                                          * is smaller than the current one.
2162                                          */
2163                                         if (mtu < tp->t_maxseg +
2164                                             sizeof (struct tcphdr) +
2165                                             sizeof (struct ip6_hdr)) {
2166                                                 tcp_hc_updatemtu(&inc, mtu);
2167                                                 tcp_mtudisc(inp, mtu);
2168                                                 ICMP6STAT_INC(icp6s_pmtuchg);
2169                                         }
2170                                 } else
2171                                         inp = (*notify)(inp,
2172                                             inet6ctlerrmap[cmd]);
2173                         }
2174                 }
2175         } else {
2176                 bzero(&inc, sizeof(inc));
2177                 inc.inc_fibnum = M_GETFIB(m);
2178                 inc.inc_flags |= INC_ISIPV6;
2179                 inc.inc_fport = t_ports.th_dport;
2180                 inc.inc_lport = t_ports.th_sport;
2181                 inc.inc6_faddr = *dst;
2182                 inc.inc6_laddr = ip6->ip6_src;
2183                 syncache_unreach(&inc, icmp_tcp_seq);
2184         }
2185 out:
2186         if (inp != NULL)
2187                 INP_WUNLOCK(inp);
2188         INP_INFO_RUNLOCK(&V_tcbinfo);
2189 }
2190 #endif /* INET6 */
2191
2192 static uint32_t
2193 tcp_keyed_hash(struct in_conninfo *inc, u_char *key)
2194 {
2195         MD5_CTX ctx;
2196         uint32_t hash[4];
2197
2198         MD5Init(&ctx);
2199         MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t));
2200         MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t));
2201         switch (inc->inc_flags & INC_ISIPV6) {
2202 #ifdef INET
2203         case 0:
2204                 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr));
2205                 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr));
2206                 break;
2207 #endif
2208 #ifdef INET6
2209         case INC_ISIPV6:
2210                 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr));
2211                 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr));
2212                 break;
2213 #endif
2214         }
2215         MD5Update(&ctx, key, 32);
2216         MD5Final((unsigned char *)hash, &ctx);
2217
2218         return (hash[0]);
2219 }
2220
2221 uint32_t
2222 tcp_new_ts_offset(struct in_conninfo *inc)
2223 {
2224         return (tcp_keyed_hash(inc, V_ts_offset_secret));
2225 }
2226
2227 /*
2228  * Following is where TCP initial sequence number generation occurs.
2229  *
2230  * There are two places where we must use initial sequence numbers:
2231  * 1.  In SYN-ACK packets.
2232  * 2.  In SYN packets.
2233  *
2234  * All ISNs for SYN-ACK packets are generated by the syncache.  See
2235  * tcp_syncache.c for details.
2236  *
2237  * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2238  * depends on this property.  In addition, these ISNs should be
2239  * unguessable so as to prevent connection hijacking.  To satisfy
2240  * the requirements of this situation, the algorithm outlined in
2241  * RFC 1948 is used, with only small modifications.
2242  *
2243  * Implementation details:
2244  *
2245  * Time is based off the system timer, and is corrected so that it
2246  * increases by one megabyte per second.  This allows for proper
2247  * recycling on high speed LANs while still leaving over an hour
2248  * before rollover.
2249  *
2250  * As reading the *exact* system time is too expensive to be done
2251  * whenever setting up a TCP connection, we increment the time
2252  * offset in two ways.  First, a small random positive increment
2253  * is added to isn_offset for each connection that is set up.
2254  * Second, the function tcp_isn_tick fires once per clock tick
2255  * and increments isn_offset as necessary so that sequence numbers
2256  * are incremented at approximately ISN_BYTES_PER_SECOND.  The
2257  * random positive increments serve only to ensure that the same
2258  * exact sequence number is never sent out twice (as could otherwise
2259  * happen when a port is recycled in less than the system tick
2260  * interval.)
2261  *
2262  * net.inet.tcp.isn_reseed_interval controls the number of seconds
2263  * between seeding of isn_secret.  This is normally set to zero,
2264  * as reseeding should not be necessary.
2265  *
2266  * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
2267  * isn_offset_old, and isn_ctx is performed using the ISN lock.  In
2268  * general, this means holding an exclusive (write) lock.
2269  */
2270
2271 #define ISN_BYTES_PER_SECOND 1048576
2272 #define ISN_STATIC_INCREMENT 4096
2273 #define ISN_RANDOM_INCREMENT (4096 - 1)
2274
2275 static VNET_DEFINE(u_char, isn_secret[32]);
2276 static VNET_DEFINE(int, isn_last);
2277 static VNET_DEFINE(int, isn_last_reseed);
2278 static VNET_DEFINE(u_int32_t, isn_offset);
2279 static VNET_DEFINE(u_int32_t, isn_offset_old);
2280
2281 #define V_isn_secret                    VNET(isn_secret)
2282 #define V_isn_last                      VNET(isn_last)
2283 #define V_isn_last_reseed               VNET(isn_last_reseed)
2284 #define V_isn_offset                    VNET(isn_offset)
2285 #define V_isn_offset_old                VNET(isn_offset_old)
2286
2287 tcp_seq
2288 tcp_new_isn(struct in_conninfo *inc)
2289 {
2290         tcp_seq new_isn;
2291         u_int32_t projected_offset;
2292
2293         ISN_LOCK();
2294         /* Seed if this is the first use, reseed if requested. */
2295         if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
2296              (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
2297                 < (u_int)ticks))) {
2298                 read_random(&V_isn_secret, sizeof(V_isn_secret));
2299                 V_isn_last_reseed = ticks;
2300         }
2301
2302         /* Compute the md5 hash and return the ISN. */
2303         new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret);
2304         V_isn_offset += ISN_STATIC_INCREMENT +
2305                 (arc4random() & ISN_RANDOM_INCREMENT);
2306         if (ticks != V_isn_last) {
2307                 projected_offset = V_isn_offset_old +
2308                     ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
2309                 if (SEQ_GT(projected_offset, V_isn_offset))
2310                         V_isn_offset = projected_offset;
2311                 V_isn_offset_old = V_isn_offset;
2312                 V_isn_last = ticks;
2313         }
2314         new_isn += V_isn_offset;
2315         ISN_UNLOCK();
2316         return (new_isn);
2317 }
2318
2319 /*
2320  * When a specific ICMP unreachable message is received and the
2321  * connection state is SYN-SENT, drop the connection.  This behavior
2322  * is controlled by the icmp_may_rst sysctl.
2323  */
2324 struct inpcb *
2325 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2326 {
2327         struct tcpcb *tp;
2328
2329         INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2330         INP_WLOCK_ASSERT(inp);
2331
2332         if ((inp->inp_flags & INP_TIMEWAIT) ||
2333             (inp->inp_flags & INP_DROPPED))
2334                 return (inp);
2335
2336         tp = intotcpcb(inp);
2337         if (tp->t_state != TCPS_SYN_SENT)
2338                 return (inp);
2339
2340         tp = tcp_drop(tp, errno);
2341         if (tp != NULL)
2342                 return (inp);
2343         else
2344                 return (NULL);
2345 }
2346
2347 /*
2348  * When `need fragmentation' ICMP is received, update our idea of the MSS
2349  * based on the new value. Also nudge TCP to send something, since we
2350  * know the packet we just sent was dropped.
2351  * This duplicates some code in the tcp_mss() function in tcp_input.c.
2352  */
2353 static struct inpcb *
2354 tcp_mtudisc_notify(struct inpcb *inp, int error)
2355 {
2356
2357         tcp_mtudisc(inp, -1);
2358         return (inp);
2359 }
2360
2361 static void
2362 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
2363 {
2364         struct tcpcb *tp;
2365         struct socket *so;
2366
2367         INP_WLOCK_ASSERT(inp);
2368         if ((inp->inp_flags & INP_TIMEWAIT) ||
2369             (inp->inp_flags & INP_DROPPED))
2370                 return;
2371
2372         tp = intotcpcb(inp);
2373         KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
2374
2375         tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
2376
2377         so = inp->inp_socket;
2378         SOCKBUF_LOCK(&so->so_snd);
2379         /* If the mss is larger than the socket buffer, decrease the mss. */
2380         if (so->so_snd.sb_hiwat < tp->t_maxseg)
2381                 tp->t_maxseg = so->so_snd.sb_hiwat;
2382         SOCKBUF_UNLOCK(&so->so_snd);
2383
2384         TCPSTAT_INC(tcps_mturesent);
2385         tp->t_rtttime = 0;
2386         tp->snd_nxt = tp->snd_una;
2387         tcp_free_sackholes(tp);
2388         tp->snd_recover = tp->snd_max;
2389         if (tp->t_flags & TF_SACK_PERMIT)
2390                 EXIT_FASTRECOVERY(tp->t_flags);
2391         tp->t_fb->tfb_tcp_output(tp);
2392 }
2393
2394 #ifdef INET
2395 /*
2396  * Look-up the routing entry to the peer of this inpcb.  If no route
2397  * is found and it cannot be allocated, then return 0.  This routine
2398  * is called by TCP routines that access the rmx structure and by
2399  * tcp_mss_update to get the peer/interface MTU.
2400  */
2401 u_long
2402 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
2403 {
2404         struct nhop4_extended nh4;
2405         struct ifnet *ifp;
2406         u_long maxmtu = 0;
2407
2408         KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
2409
2410         if (inc->inc_faddr.s_addr != INADDR_ANY) {
2411
2412                 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr,
2413                     NHR_REF, 0, &nh4) != 0)
2414                         return (0);
2415
2416                 ifp = nh4.nh_ifp;
2417                 maxmtu = nh4.nh_mtu;
2418
2419                 /* Report additional interface capabilities. */
2420                 if (cap != NULL) {
2421                         if (ifp->if_capenable & IFCAP_TSO4 &&
2422                             ifp->if_hwassist & CSUM_TSO) {
2423                                 cap->ifcap |= CSUM_TSO;
2424                                 cap->tsomax = ifp->if_hw_tsomax;
2425                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2426                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2427                         }
2428                 }
2429                 fib4_free_nh_ext(inc->inc_fibnum, &nh4);
2430         }
2431         return (maxmtu);
2432 }
2433 #endif /* INET */
2434
2435 #ifdef INET6
2436 u_long
2437 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2438 {
2439         struct nhop6_extended nh6;
2440         struct in6_addr dst6;
2441         uint32_t scopeid;
2442         struct ifnet *ifp;
2443         u_long maxmtu = 0;
2444
2445         KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2446
2447         if (inc->inc_flags & INC_IPV6MINMTU)
2448                 return (IPV6_MMTU);
2449
2450         if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2451                 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid);
2452                 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0,
2453                     0, &nh6) != 0)
2454                         return (0);
2455
2456                 ifp = nh6.nh_ifp;
2457                 maxmtu = nh6.nh_mtu;
2458
2459                 /* Report additional interface capabilities. */
2460                 if (cap != NULL) {
2461                         if (ifp->if_capenable & IFCAP_TSO6 &&
2462                             ifp->if_hwassist & CSUM_TSO) {
2463                                 cap->ifcap |= CSUM_TSO;
2464                                 cap->tsomax = ifp->if_hw_tsomax;
2465                                 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2466                                 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2467                         }
2468                 }
2469                 fib6_free_nh_ext(inc->inc_fibnum, &nh6);
2470         }
2471
2472         return (maxmtu);
2473 }
2474 #endif /* INET6 */
2475
2476 /*
2477  * Calculate effective SMSS per RFC5681 definition for a given TCP
2478  * connection at its current state, taking into account SACK and etc.
2479  */
2480 u_int
2481 tcp_maxseg(const struct tcpcb *tp)
2482 {
2483         u_int optlen;
2484
2485         if (tp->t_flags & TF_NOOPT)
2486                 return (tp->t_maxseg);
2487
2488         /*
2489          * Here we have a simplified code from tcp_addoptions(),
2490          * without a proper loop, and having most of paddings hardcoded.
2491          * We might make mistakes with padding here in some edge cases,
2492          * but this is harmless, since result of tcp_maxseg() is used
2493          * only in cwnd and ssthresh estimations.
2494          */
2495 #define PAD(len)        ((((len) / 4) + !!((len) % 4)) * 4)
2496         if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2497                 if (tp->t_flags & TF_RCVD_TSTMP)
2498                         optlen = TCPOLEN_TSTAMP_APPA;
2499                 else
2500                         optlen = 0;
2501 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2502                 if (tp->t_flags & TF_SIGNATURE)
2503                         optlen += PAD(TCPOLEN_SIGNATURE);
2504 #endif
2505                 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) {
2506                         optlen += TCPOLEN_SACKHDR;
2507                         optlen += tp->rcv_numsacks * TCPOLEN_SACK;
2508                         optlen = PAD(optlen);
2509                 }
2510         } else {
2511                 if (tp->t_flags & TF_REQ_TSTMP)
2512                         optlen = TCPOLEN_TSTAMP_APPA;
2513                 else
2514                         optlen = PAD(TCPOLEN_MAXSEG);
2515                 if (tp->t_flags & TF_REQ_SCALE)
2516                         optlen += PAD(TCPOLEN_WINDOW);
2517 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2518                 if (tp->t_flags & TF_SIGNATURE)
2519                         optlen += PAD(TCPOLEN_SIGNATURE);
2520 #endif
2521                 if (tp->t_flags & TF_SACK_PERMIT)
2522                         optlen += PAD(TCPOLEN_SACK_PERMITTED);
2523         }
2524 #undef PAD
2525         optlen = min(optlen, TCP_MAXOLEN);
2526         return (tp->t_maxseg - optlen);
2527 }
2528
2529 static int
2530 sysctl_drop(SYSCTL_HANDLER_ARGS)
2531 {
2532         /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2533         struct sockaddr_storage addrs[2];
2534         struct inpcb *inp;
2535         struct tcpcb *tp;
2536         struct tcptw *tw;
2537         struct sockaddr_in *fin, *lin;
2538 #ifdef INET6
2539         struct sockaddr_in6 *fin6, *lin6;
2540 #endif
2541         int error;
2542
2543         inp = NULL;
2544         fin = lin = NULL;
2545 #ifdef INET6
2546         fin6 = lin6 = NULL;
2547 #endif
2548         error = 0;
2549
2550         if (req->oldptr != NULL || req->oldlen != 0)
2551                 return (EINVAL);
2552         if (req->newptr == NULL)
2553                 return (EPERM);
2554         if (req->newlen < sizeof(addrs))
2555                 return (ENOMEM);
2556         error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2557         if (error)
2558                 return (error);
2559
2560         switch (addrs[0].ss_family) {
2561 #ifdef INET6
2562         case AF_INET6:
2563                 fin6 = (struct sockaddr_in6 *)&addrs[0];
2564                 lin6 = (struct sockaddr_in6 *)&addrs[1];
2565                 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2566                     lin6->sin6_len != sizeof(struct sockaddr_in6))
2567                         return (EINVAL);
2568                 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2569                         if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2570                                 return (EINVAL);
2571                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2572                         in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2573                         fin = (struct sockaddr_in *)&addrs[0];
2574                         lin = (struct sockaddr_in *)&addrs[1];
2575                         break;
2576                 }
2577                 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2578                 if (error)
2579                         return (error);
2580                 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2581                 if (error)
2582                         return (error);
2583                 break;
2584 #endif
2585 #ifdef INET
2586         case AF_INET:
2587                 fin = (struct sockaddr_in *)&addrs[0];
2588                 lin = (struct sockaddr_in *)&addrs[1];
2589                 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2590                     lin->sin_len != sizeof(struct sockaddr_in))
2591                         return (EINVAL);
2592                 break;
2593 #endif
2594         default:
2595                 return (EINVAL);
2596         }
2597         INP_INFO_RLOCK(&V_tcbinfo);
2598         switch (addrs[0].ss_family) {
2599 #ifdef INET6
2600         case AF_INET6:
2601                 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2602                     fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2603                     INPLOOKUP_WLOCKPCB, NULL);
2604                 break;
2605 #endif
2606 #ifdef INET
2607         case AF_INET:
2608                 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2609                     lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2610                 break;
2611 #endif
2612         }
2613         if (inp != NULL) {
2614                 if (inp->inp_flags & INP_TIMEWAIT) {
2615                         /*
2616                          * XXXRW: There currently exists a state where an
2617                          * inpcb is present, but its timewait state has been
2618                          * discarded.  For now, don't allow dropping of this
2619                          * type of inpcb.
2620                          */
2621                         tw = intotw(inp);
2622                         if (tw != NULL)
2623                                 tcp_twclose(tw, 0);
2624                         else
2625                                 INP_WUNLOCK(inp);
2626                 } else if (!(inp->inp_flags & INP_DROPPED) &&
2627                            !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2628                         tp = intotcpcb(inp);
2629                         tp = tcp_drop(tp, ECONNABORTED);
2630                         if (tp != NULL)
2631                                 INP_WUNLOCK(inp);
2632                 } else
2633                         INP_WUNLOCK(inp);
2634         } else
2635                 error = ESRCH;
2636         INP_INFO_RUNLOCK(&V_tcbinfo);
2637         return (error);
2638 }
2639
2640 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2641     CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2642     0, sysctl_drop, "", "Drop TCP connection");
2643
2644 /*
2645  * Generate a standardized TCP log line for use throughout the
2646  * tcp subsystem.  Memory allocation is done with M_NOWAIT to
2647  * allow use in the interrupt context.
2648  *
2649  * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2650  * NB: The function may return NULL if memory allocation failed.
2651  *
2652  * Due to header inclusion and ordering limitations the struct ip
2653  * and ip6_hdr pointers have to be passed as void pointers.
2654  */
2655 char *
2656 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2657     const void *ip6hdr)
2658 {
2659
2660         /* Is logging enabled? */
2661         if (tcp_log_in_vain == 0)
2662                 return (NULL);
2663
2664         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2665 }
2666
2667 char *
2668 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2669     const void *ip6hdr)
2670 {
2671
2672         /* Is logging enabled? */
2673         if (tcp_log_debug == 0)
2674                 return (NULL);
2675
2676         return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2677 }
2678
2679 static char *
2680 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2681     const void *ip6hdr)
2682 {
2683         char *s, *sp;
2684         size_t size;
2685         struct ip *ip;
2686 #ifdef INET6
2687         const struct ip6_hdr *ip6;
2688
2689         ip6 = (const struct ip6_hdr *)ip6hdr;
2690 #endif /* INET6 */
2691         ip = (struct ip *)ip4hdr;
2692
2693         /*
2694          * The log line looks like this:
2695          * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2696          */
2697         size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2698             sizeof(PRINT_TH_FLAGS) + 1 +
2699 #ifdef INET6
2700             2 * INET6_ADDRSTRLEN;
2701 #else
2702             2 * INET_ADDRSTRLEN;
2703 #endif /* INET6 */
2704
2705         s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2706         if (s == NULL)
2707                 return (NULL);
2708
2709         strcat(s, "TCP: [");
2710         sp = s + strlen(s);
2711
2712         if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2713                 inet_ntoa_r(inc->inc_faddr, sp);
2714                 sp = s + strlen(s);
2715                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2716                 sp = s + strlen(s);
2717                 inet_ntoa_r(inc->inc_laddr, sp);
2718                 sp = s + strlen(s);
2719                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2720 #ifdef INET6
2721         } else if (inc) {
2722                 ip6_sprintf(sp, &inc->inc6_faddr);
2723                 sp = s + strlen(s);
2724                 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2725                 sp = s + strlen(s);
2726                 ip6_sprintf(sp, &inc->inc6_laddr);
2727                 sp = s + strlen(s);
2728                 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2729         } else if (ip6 && th) {
2730                 ip6_sprintf(sp, &ip6->ip6_src);
2731                 sp = s + strlen(s);
2732                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2733                 sp = s + strlen(s);
2734                 ip6_sprintf(sp, &ip6->ip6_dst);
2735                 sp = s + strlen(s);
2736                 sprintf(sp, "]:%i", ntohs(th->th_dport));
2737 #endif /* INET6 */
2738 #ifdef INET
2739         } else if (ip && th) {
2740                 inet_ntoa_r(ip->ip_src, sp);
2741                 sp = s + strlen(s);
2742                 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2743                 sp = s + strlen(s);
2744                 inet_ntoa_r(ip->ip_dst, sp);
2745                 sp = s + strlen(s);
2746                 sprintf(sp, "]:%i", ntohs(th->th_dport));
2747 #endif /* INET */
2748         } else {
2749                 free(s, M_TCPLOG);
2750                 return (NULL);
2751         }
2752         sp = s + strlen(s);
2753         if (th)
2754                 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2755         if (*(s + size - 1) != '\0')
2756                 panic("%s: string too long", __func__);
2757         return (s);
2758 }
2759
2760 /*
2761  * A subroutine which makes it easy to track TCP state changes with DTrace.
2762  * This function shouldn't be called for t_state initializations that don't
2763  * correspond to actual TCP state transitions.
2764  */
2765 void
2766 tcp_state_change(struct tcpcb *tp, int newstate)
2767 {
2768 #if defined(KDTRACE_HOOKS)
2769         int pstate = tp->t_state;
2770 #endif
2771
2772         TCPSTATES_DEC(tp->t_state);
2773         TCPSTATES_INC(newstate);
2774         tp->t_state = newstate;
2775         TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2776 }