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/ip6.h> 82 #include <netinet6/in6_fib.h> 83 #include <netinet6/in6_pcb.h> 84 #include <netinet6/ip6_var.h> 85 #include <netinet6/scope6_var.h> 86 #include <netinet6/nd6.h> 87 #endif 88 89 #ifdef TCP_RFC7413 90 #include <netinet/tcp_fastopen.h> 91 #endif 92 #include <netinet/tcp.h> 93 #include <netinet/tcp_fsm.h> 94 #include <netinet/tcp_seq.h> 95 #include <netinet/tcp_timer.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/tcp_syncache.h> 98 #include <netinet/cc/cc.h> 99 #ifdef INET6 100 #include <netinet6/tcp6_var.h> 101 #endif 102 #include <netinet/tcpip.h> 103 #ifdef TCPPCAP 104 #include <netinet/tcp_pcap.h> 105 #endif 106 #ifdef TCPDEBUG 107 #include <netinet/tcp_debug.h> 108 #endif 109 #ifdef INET6 110 #include <netinet6/ip6protosw.h> 111 #endif 112 #ifdef TCP_OFFLOAD 113 #include <netinet/tcp_offload.h> 114 #endif 115 116 #ifdef IPSEC 117 #include <netipsec/ipsec.h> 118 #include <netipsec/xform.h> 119 #ifdef INET6 120 #include <netipsec/ipsec6.h> 121 #endif 122 #include <netipsec/key.h> 123 #include <sys/syslog.h> 124 #endif /*IPSEC*/ 125 126 #include <machine/in_cksum.h> 127 #include <sys/md5.h> 128 129 #include <security/mac/mac_framework.h> 130 131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 132 #ifdef INET6 133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 134 #endif 135 136 struct rwlock tcp_function_lock; 137 138 static int 139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 140 { 141 int error, new; 142 143 new = V_tcp_mssdflt; 144 error = sysctl_handle_int(oidp, &new, 0, req); 145 if (error == 0 && req->newptr) { 146 if (new < TCP_MINMSS) 147 error = EINVAL; 148 else 149 V_tcp_mssdflt = new; 150 } 151 return (error); 152 } 153 154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 155 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 156 &sysctl_net_inet_tcp_mss_check, "I", 157 "Default TCP Maximum Segment Size"); 158 159 #ifdef INET6 160 static int 161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 162 { 163 int error, new; 164 165 new = V_tcp_v6mssdflt; 166 error = sysctl_handle_int(oidp, &new, 0, req); 167 if (error == 0 && req->newptr) { 168 if (new < TCP_MINMSS) 169 error = EINVAL; 170 else 171 V_tcp_v6mssdflt = new; 172 } 173 return (error); 174 } 175 176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 178 &sysctl_net_inet_tcp_mss_v6_check, "I", 179 "Default TCP Maximum Segment Size for IPv6"); 180 #endif /* INET6 */ 181 182 /* 183 * Minimum MSS we accept and use. This prevents DoS attacks where 184 * we are forced to a ridiculous low MSS like 20 and send hundreds 185 * of packets instead of one. The effect scales with the available 186 * bandwidth and quickly saturates the CPU and network interface 187 * with packet generation and sending. Set to zero to disable MINMSS 188 * checking. This setting prevents us from sending too small packets. 189 */ 190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_minmss), 0, 193 "Minimum TCP Maximum Segment Size"); 194 195 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_do_rfc1323), 0, 198 "Enable rfc1323 (high performance TCP) extensions"); 199 200 static int tcp_log_debug = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 202 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 203 204 static int tcp_tcbhashsize; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 206 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 207 208 static int do_tcpdrain = 1; 209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 210 "Enable tcp_drain routine for extra help when low on mbufs"); 211 212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 213 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 214 215 static VNET_DEFINE(int, icmp_may_rst) = 1; 216 #define V_icmp_may_rst VNET(icmp_may_rst) 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(icmp_may_rst), 0, 219 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 220 221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 222 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_isn_reseed_interval), 0, 225 "Seconds between reseeding of ISN secret"); 226 227 static int tcp_soreceive_stream; 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 229 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 230 231 #ifdef TCP_SIGNATURE 232 static int tcp_sig_checksigs = 1; 233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, 234 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); 235 #endif 236 237 VNET_DEFINE(uma_zone_t, sack_hole_zone); 238 #define V_sack_hole_zone VNET(sack_hole_zone) 239 240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 241 242 static struct inpcb *tcp_notify(struct inpcb *, int); 243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 244 static void tcp_mtudisc(struct inpcb *, int); 245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 246 void *ip4hdr, const void *ip6hdr); 247 248 249 static struct tcp_function_block tcp_def_funcblk = { 250 "default", 251 tcp_output, 252 tcp_do_segment, 253 tcp_default_ctloutput, 254 NULL, 255 NULL, 256 NULL, 257 NULL, 258 NULL, 259 NULL, 260 0, 261 0 262 }; 263 264 int t_functions_inited = 0; 265 struct tcp_funchead t_functions; 266 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 267 268 static void 269 init_tcp_functions(void) 270 { 271 if (t_functions_inited == 0) { 272 TAILQ_INIT(&t_functions); 273 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 274 t_functions_inited = 1; 275 } 276 } 277 278 static struct tcp_function_block * 279 find_tcp_functions_locked(struct tcp_function_set *fs) 280 { 281 struct tcp_function *f; 282 struct tcp_function_block *blk=NULL; 283 284 TAILQ_FOREACH(f, &t_functions, tf_next) { 285 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) { 286 blk = f->tf_fb; 287 break; 288 } 289 } 290 return(blk); 291 } 292 293 static struct tcp_function_block * 294 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 295 { 296 struct tcp_function_block *rblk=NULL; 297 struct tcp_function *f; 298 299 TAILQ_FOREACH(f, &t_functions, tf_next) { 300 if (f->tf_fb == blk) { 301 rblk = blk; 302 if (s) { 303 *s = f; 304 } 305 break; 306 } 307 } 308 return (rblk); 309 } 310 311 struct tcp_function_block * 312 find_and_ref_tcp_functions(struct tcp_function_set *fs) 313 { 314 struct tcp_function_block *blk; 315 316 rw_rlock(&tcp_function_lock); 317 blk = find_tcp_functions_locked(fs); 318 if (blk) 319 refcount_acquire(&blk->tfb_refcnt); 320 rw_runlock(&tcp_function_lock); 321 return(blk); 322 } 323 324 struct tcp_function_block * 325 find_and_ref_tcp_fb(struct tcp_function_block *blk) 326 { 327 struct tcp_function_block *rblk; 328 329 rw_rlock(&tcp_function_lock); 330 rblk = find_tcp_fb_locked(blk, NULL); 331 if (rblk) 332 refcount_acquire(&rblk->tfb_refcnt); 333 rw_runlock(&tcp_function_lock); 334 return(rblk); 335 } 336 337 338 static int 339 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 340 { 341 int error=ENOENT; 342 struct tcp_function_set fs; 343 struct tcp_function_block *blk; 344 345 memset(&fs, 0, sizeof(fs)); 346 rw_rlock(&tcp_function_lock); 347 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 348 if (blk) { 349 /* Found him */ 350 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 351 fs.pcbcnt = blk->tfb_refcnt; 352 } 353 rw_runlock(&tcp_function_lock); 354 error = sysctl_handle_string(oidp, fs.function_set_name, 355 sizeof(fs.function_set_name), req); 356 357 /* Check for error or no change */ 358 if (error != 0 || req->newptr == NULL) 359 return(error); 360 361 rw_wlock(&tcp_function_lock); 362 blk = find_tcp_functions_locked(&fs); 363 if ((blk == NULL) || 364 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 365 error = ENOENT; 366 goto done; 367 } 368 tcp_func_set_ptr = blk; 369 done: 370 rw_wunlock(&tcp_function_lock); 371 return (error); 372 } 373 374 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 375 CTLTYPE_STRING | CTLFLAG_RW, 376 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 377 "Set/get the default TCP functions"); 378 379 static int 380 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 381 { 382 int error, cnt, linesz; 383 struct tcp_function *f; 384 char *buffer, *cp; 385 size_t bufsz, outsz; 386 387 cnt = 0; 388 rw_rlock(&tcp_function_lock); 389 TAILQ_FOREACH(f, &t_functions, tf_next) { 390 cnt++; 391 } 392 rw_runlock(&tcp_function_lock); 393 394 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1; 395 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 396 397 error = 0; 398 cp = buffer; 399 400 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count"); 401 cp += linesz; 402 bufsz -= linesz; 403 outsz = linesz; 404 405 rw_rlock(&tcp_function_lock); 406 TAILQ_FOREACH(f, &t_functions, tf_next) { 407 linesz = snprintf(cp, bufsz, "%-32s%c %u\n", 408 f->tf_fb->tfb_tcp_block_name, 409 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 410 f->tf_fb->tfb_refcnt); 411 if (linesz >= bufsz) { 412 error = EOVERFLOW; 413 break; 414 } 415 cp += linesz; 416 bufsz -= linesz; 417 outsz += linesz; 418 } 419 rw_runlock(&tcp_function_lock); 420 if (error == 0) 421 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 422 free(buffer, M_TEMP); 423 return (error); 424 } 425 426 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 427 CTLTYPE_STRING|CTLFLAG_RD, 428 NULL, 0, sysctl_net_inet_list_available, "A", 429 "list available TCP Function sets"); 430 431 /* 432 * Target size of TCP PCB hash tables. Must be a power of two. 433 * 434 * Note that this can be overridden by the kernel environment 435 * variable net.inet.tcp.tcbhashsize 436 */ 437 #ifndef TCBHASHSIZE 438 #define TCBHASHSIZE 0 439 #endif 440 441 /* 442 * XXX 443 * Callouts should be moved into struct tcp directly. They are currently 444 * separate because the tcpcb structure is exported to userland for sysctl 445 * parsing purposes, which do not know about callouts. 446 */ 447 struct tcpcb_mem { 448 struct tcpcb tcb; 449 struct tcp_timer tt; 450 struct cc_var ccv; 451 struct osd osd; 452 }; 453 454 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 455 #define V_tcpcb_zone VNET(tcpcb_zone) 456 457 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 458 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 459 460 static struct mtx isn_mtx; 461 462 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 463 #define ISN_LOCK() mtx_lock(&isn_mtx) 464 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 465 466 /* 467 * TCP initialization. 468 */ 469 static void 470 tcp_zone_change(void *tag) 471 { 472 473 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 474 uma_zone_set_max(V_tcpcb_zone, maxsockets); 475 tcp_tw_zone_change(); 476 } 477 478 static int 479 tcp_inpcb_init(void *mem, int size, int flags) 480 { 481 struct inpcb *inp = mem; 482 483 INP_LOCK_INIT(inp, "inp", "tcpinp"); 484 return (0); 485 } 486 487 /* 488 * Take a value and get the next power of 2 that doesn't overflow. 489 * Used to size the tcp_inpcb hash buckets. 490 */ 491 static int 492 maketcp_hashsize(int size) 493 { 494 int hashsize; 495 496 /* 497 * auto tune. 498 * get the next power of 2 higher than maxsockets. 499 */ 500 hashsize = 1 << fls(size); 501 /* catch overflow, and just go one power of 2 smaller */ 502 if (hashsize < size) { 503 hashsize = 1 << (fls(size) - 1); 504 } 505 return (hashsize); 506 } 507 508 int 509 register_tcp_functions(struct tcp_function_block *blk, int wait) 510 { 511 struct tcp_function_block *lblk; 512 struct tcp_function *n; 513 struct tcp_function_set fs; 514 515 if (t_functions_inited == 0) { 516 init_tcp_functions(); 517 } 518 if ((blk->tfb_tcp_output == NULL) || 519 (blk->tfb_tcp_do_segment == NULL) || 520 (blk->tfb_tcp_ctloutput == NULL) || 521 (strlen(blk->tfb_tcp_block_name) == 0)) { 522 /* 523 * These functions are required and you 524 * need a name. 525 */ 526 return (EINVAL); 527 } 528 if (blk->tfb_tcp_timer_stop_all || 529 blk->tfb_tcp_timer_activate || 530 blk->tfb_tcp_timer_active || 531 blk->tfb_tcp_timer_stop) { 532 /* 533 * If you define one timer function you 534 * must have them all. 535 */ 536 if ((blk->tfb_tcp_timer_stop_all == NULL) || 537 (blk->tfb_tcp_timer_activate == NULL) || 538 (blk->tfb_tcp_timer_active == NULL) || 539 (blk->tfb_tcp_timer_stop == NULL)) { 540 return (EINVAL); 541 } 542 } 543 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 544 if (n == NULL) { 545 return (ENOMEM); 546 } 547 n->tf_fb = blk; 548 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 549 rw_wlock(&tcp_function_lock); 550 lblk = find_tcp_functions_locked(&fs); 551 if (lblk) { 552 /* Duplicate name space not allowed */ 553 rw_wunlock(&tcp_function_lock); 554 free(n, M_TCPFUNCTIONS); 555 return (EALREADY); 556 } 557 refcount_init(&blk->tfb_refcnt, 0); 558 blk->tfb_flags = 0; 559 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 560 rw_wunlock(&tcp_function_lock); 561 return(0); 562 } 563 564 int 565 deregister_tcp_functions(struct tcp_function_block *blk) 566 { 567 struct tcp_function_block *lblk; 568 struct tcp_function *f; 569 int error=ENOENT; 570 571 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 572 /* You can't un-register the default */ 573 return (EPERM); 574 } 575 rw_wlock(&tcp_function_lock); 576 if (blk == tcp_func_set_ptr) { 577 /* You can't free the current default */ 578 rw_wunlock(&tcp_function_lock); 579 return (EBUSY); 580 } 581 if (blk->tfb_refcnt) { 582 /* Still tcb attached, mark it. */ 583 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 584 rw_wunlock(&tcp_function_lock); 585 return (EBUSY); 586 } 587 lblk = find_tcp_fb_locked(blk, &f); 588 if (lblk) { 589 /* Found */ 590 TAILQ_REMOVE(&t_functions, f, tf_next); 591 f->tf_fb = NULL; 592 free(f, M_TCPFUNCTIONS); 593 error = 0; 594 } 595 rw_wunlock(&tcp_function_lock); 596 return (error); 597 } 598 599 void 600 tcp_init(void) 601 { 602 const char *tcbhash_tuneable; 603 int hashsize; 604 605 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 606 607 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 608 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 609 printf("%s: WARNING: unable to register helper hook\n", __func__); 610 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 611 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 612 printf("%s: WARNING: unable to register helper hook\n", __func__); 613 hashsize = TCBHASHSIZE; 614 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 615 if (hashsize == 0) { 616 /* 617 * Auto tune the hash size based on maxsockets. 618 * A perfect hash would have a 1:1 mapping 619 * (hashsize = maxsockets) however it's been 620 * suggested that O(2) average is better. 621 */ 622 hashsize = maketcp_hashsize(maxsockets / 4); 623 /* 624 * Our historical default is 512, 625 * do not autotune lower than this. 626 */ 627 if (hashsize < 512) 628 hashsize = 512; 629 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 630 printf("%s: %s auto tuned to %d\n", __func__, 631 tcbhash_tuneable, hashsize); 632 } 633 /* 634 * We require a hashsize to be a power of two. 635 * Previously if it was not a power of two we would just reset it 636 * back to 512, which could be a nasty surprise if you did not notice 637 * the error message. 638 * Instead what we do is clip it to the closest power of two lower 639 * than the specified hash value. 640 */ 641 if (!powerof2(hashsize)) { 642 int oldhashsize = hashsize; 643 644 hashsize = maketcp_hashsize(hashsize); 645 /* prevent absurdly low value */ 646 if (hashsize < 16) 647 hashsize = 16; 648 printf("%s: WARNING: TCB hash size not a power of 2, " 649 "clipped from %d to %d.\n", __func__, oldhashsize, 650 hashsize); 651 } 652 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 653 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE); 654 655 /* 656 * These have to be type stable for the benefit of the timers. 657 */ 658 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 659 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 660 uma_zone_set_max(V_tcpcb_zone, maxsockets); 661 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 662 663 tcp_tw_init(); 664 syncache_init(); 665 tcp_hc_init(); 666 667 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 668 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 669 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 670 671 /* Skip initialization of globals for non-default instances. */ 672 if (!IS_DEFAULT_VNET(curvnet)) 673 return; 674 675 tcp_reass_global_init(); 676 677 /* XXX virtualize those bellow? */ 678 tcp_delacktime = TCPTV_DELACK; 679 tcp_keepinit = TCPTV_KEEP_INIT; 680 tcp_keepidle = TCPTV_KEEP_IDLE; 681 tcp_keepintvl = TCPTV_KEEPINTVL; 682 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 683 tcp_msl = TCPTV_MSL; 684 tcp_rexmit_min = TCPTV_MIN; 685 if (tcp_rexmit_min < 1) 686 tcp_rexmit_min = 1; 687 tcp_persmin = TCPTV_PERSMIN; 688 tcp_persmax = TCPTV_PERSMAX; 689 tcp_rexmit_slop = TCPTV_CPU_VAR; 690 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 691 tcp_tcbhashsize = hashsize; 692 /* Setup the tcp function block list */ 693 init_tcp_functions(); 694 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 695 696 if (tcp_soreceive_stream) { 697 #ifdef INET 698 tcp_usrreqs.pru_soreceive = soreceive_stream; 699 #endif 700 #ifdef INET6 701 tcp6_usrreqs.pru_soreceive = soreceive_stream; 702 #endif /* INET6 */ 703 } 704 705 #ifdef INET6 706 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 707 #else /* INET6 */ 708 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 709 #endif /* INET6 */ 710 if (max_protohdr < TCP_MINPROTOHDR) 711 max_protohdr = TCP_MINPROTOHDR; 712 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 713 panic("tcp_init"); 714 #undef TCP_MINPROTOHDR 715 716 ISN_LOCK_INIT(); 717 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 718 SHUTDOWN_PRI_DEFAULT); 719 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 720 EVENTHANDLER_PRI_ANY); 721 #ifdef TCPPCAP 722 tcp_pcap_init(); 723 #endif 724 725 #ifdef TCP_RFC7413 726 tcp_fastopen_init(); 727 #endif 728 } 729 730 #ifdef VIMAGE 731 static void 732 tcp_destroy(void *unused __unused) 733 { 734 int error, n; 735 736 /* 737 * All our processes are gone, all our sockets should be cleaned 738 * up, which means, we should be past the tcp_discardcb() calls. 739 * Sleep to let all tcpcb timers really disappear and cleanup. 740 */ 741 for (;;) { 742 INP_LIST_RLOCK(&V_tcbinfo); 743 n = V_tcbinfo.ipi_count; 744 INP_LIST_RUNLOCK(&V_tcbinfo); 745 if (n == 0) 746 break; 747 pause("tcpdes", hz / 10); 748 } 749 tcp_hc_destroy(); 750 syncache_destroy(); 751 tcp_tw_destroy(); 752 in_pcbinfo_destroy(&V_tcbinfo); 753 /* tcp_discardcb() clears the sack_holes up. */ 754 uma_zdestroy(V_sack_hole_zone); 755 uma_zdestroy(V_tcpcb_zone); 756 757 #ifdef TCP_RFC7413 758 /* 759 * Cannot free the zone until all tcpcbs are released as we attach 760 * the allocations to them. 761 */ 762 tcp_fastopen_destroy(); 763 #endif 764 765 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 766 if (error != 0) { 767 printf("%s: WARNING: unable to deregister helper hook " 768 "type=%d, id=%d: error %d returned\n", __func__, 769 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 770 } 771 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 772 if (error != 0) { 773 printf("%s: WARNING: unable to deregister helper hook " 774 "type=%d, id=%d: error %d returned\n", __func__, 775 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 776 } 777 } 778 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 779 #endif 780 781 void 782 tcp_fini(void *xtp) 783 { 784 785 } 786 787 /* 788 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 789 * tcp_template used to store this data in mbufs, but we now recopy it out 790 * of the tcpcb each time to conserve mbufs. 791 */ 792 void 793 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 794 { 795 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 796 797 INP_WLOCK_ASSERT(inp); 798 799 #ifdef INET6 800 if ((inp->inp_vflag & INP_IPV6) != 0) { 801 struct ip6_hdr *ip6; 802 803 ip6 = (struct ip6_hdr *)ip_ptr; 804 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 805 (inp->inp_flow & IPV6_FLOWINFO_MASK); 806 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 807 (IPV6_VERSION & IPV6_VERSION_MASK); 808 ip6->ip6_nxt = IPPROTO_TCP; 809 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 810 ip6->ip6_src = inp->in6p_laddr; 811 ip6->ip6_dst = inp->in6p_faddr; 812 } 813 #endif /* INET6 */ 814 #if defined(INET6) && defined(INET) 815 else 816 #endif 817 #ifdef INET 818 { 819 struct ip *ip; 820 821 ip = (struct ip *)ip_ptr; 822 ip->ip_v = IPVERSION; 823 ip->ip_hl = 5; 824 ip->ip_tos = inp->inp_ip_tos; 825 ip->ip_len = 0; 826 ip->ip_id = 0; 827 ip->ip_off = 0; 828 ip->ip_ttl = inp->inp_ip_ttl; 829 ip->ip_sum = 0; 830 ip->ip_p = IPPROTO_TCP; 831 ip->ip_src = inp->inp_laddr; 832 ip->ip_dst = inp->inp_faddr; 833 } 834 #endif /* INET */ 835 th->th_sport = inp->inp_lport; 836 th->th_dport = inp->inp_fport; 837 th->th_seq = 0; 838 th->th_ack = 0; 839 th->th_x2 = 0; 840 th->th_off = 5; 841 th->th_flags = 0; 842 th->th_win = 0; 843 th->th_urp = 0; 844 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 845 } 846 847 /* 848 * Create template to be used to send tcp packets on a connection. 849 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 850 * use for this function is in keepalives, which use tcp_respond. 851 */ 852 struct tcptemp * 853 tcpip_maketemplate(struct inpcb *inp) 854 { 855 struct tcptemp *t; 856 857 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 858 if (t == NULL) 859 return (NULL); 860 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 861 return (t); 862 } 863 864 /* 865 * Send a single message to the TCP at address specified by 866 * the given TCP/IP header. If m == NULL, then we make a copy 867 * of the tcpiphdr at th and send directly to the addressed host. 868 * This is used to force keep alive messages out using the TCP 869 * template for a connection. If flags are given then we send 870 * a message back to the TCP which originated the segment th, 871 * and discard the mbuf containing it and any other attached mbufs. 872 * 873 * In any case the ack and sequence number of the transmitted 874 * segment are as specified by the parameters. 875 * 876 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 877 */ 878 void 879 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 880 tcp_seq ack, tcp_seq seq, int flags) 881 { 882 struct tcpopt to; 883 struct inpcb *inp; 884 struct ip *ip; 885 struct mbuf *optm; 886 struct tcphdr *nth; 887 u_char *optp; 888 #ifdef INET6 889 struct ip6_hdr *ip6; 890 int isipv6; 891 #endif /* INET6 */ 892 int optlen, tlen, win; 893 bool incl_opts; 894 895 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 896 897 #ifdef INET6 898 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 899 ip6 = ipgen; 900 #endif /* INET6 */ 901 ip = ipgen; 902 903 if (tp != NULL) { 904 inp = tp->t_inpcb; 905 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 906 INP_WLOCK_ASSERT(inp); 907 } else 908 inp = NULL; 909 910 incl_opts = false; 911 win = 0; 912 if (tp != NULL) { 913 if (!(flags & TH_RST)) { 914 win = sbspace(&inp->inp_socket->so_rcv); 915 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 916 win = (long)TCP_MAXWIN << tp->rcv_scale; 917 } 918 if ((tp->t_flags & TF_NOOPT) == 0) 919 incl_opts = true; 920 } 921 if (m == NULL) { 922 m = m_gethdr(M_NOWAIT, MT_DATA); 923 if (m == NULL) 924 return; 925 m->m_data += max_linkhdr; 926 #ifdef INET6 927 if (isipv6) { 928 bcopy((caddr_t)ip6, mtod(m, caddr_t), 929 sizeof(struct ip6_hdr)); 930 ip6 = mtod(m, struct ip6_hdr *); 931 nth = (struct tcphdr *)(ip6 + 1); 932 } else 933 #endif /* INET6 */ 934 { 935 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 936 ip = mtod(m, struct ip *); 937 nth = (struct tcphdr *)(ip + 1); 938 } 939 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 940 flags = TH_ACK; 941 } else if (!M_WRITABLE(m)) { 942 struct mbuf *n; 943 944 /* Can't reuse 'm', allocate a new mbuf. */ 945 n = m_gethdr(M_NOWAIT, MT_DATA); 946 if (n == NULL) { 947 m_freem(m); 948 return; 949 } 950 951 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 952 m_freem(m); 953 m_freem(n); 954 return; 955 } 956 957 n->m_data += max_linkhdr; 958 /* m_len is set later */ 959 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 960 #ifdef INET6 961 if (isipv6) { 962 bcopy((caddr_t)ip6, mtod(n, caddr_t), 963 sizeof(struct ip6_hdr)); 964 ip6 = mtod(n, struct ip6_hdr *); 965 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 966 nth = (struct tcphdr *)(ip6 + 1); 967 } else 968 #endif /* INET6 */ 969 { 970 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 971 ip = mtod(n, struct ip *); 972 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 973 nth = (struct tcphdr *)(ip + 1); 974 } 975 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 976 xchg(nth->th_dport, nth->th_sport, uint16_t); 977 th = nth; 978 m_freem(m); 979 m = n; 980 } else { 981 /* 982 * reuse the mbuf. 983 * XXX MRT We inherit the FIB, which is lucky. 984 */ 985 m_freem(m->m_next); 986 m->m_next = NULL; 987 m->m_data = (caddr_t)ipgen; 988 /* m_len is set later */ 989 #ifdef INET6 990 if (isipv6) { 991 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 992 nth = (struct tcphdr *)(ip6 + 1); 993 } else 994 #endif /* INET6 */ 995 { 996 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 997 nth = (struct tcphdr *)(ip + 1); 998 } 999 if (th != nth) { 1000 /* 1001 * this is usually a case when an extension header 1002 * exists between the IPv6 header and the 1003 * TCP header. 1004 */ 1005 nth->th_sport = th->th_sport; 1006 nth->th_dport = th->th_dport; 1007 } 1008 xchg(nth->th_dport, nth->th_sport, uint16_t); 1009 #undef xchg 1010 } 1011 tlen = 0; 1012 #ifdef INET6 1013 if (isipv6) 1014 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1015 #endif 1016 #if defined(INET) && defined(INET6) 1017 else 1018 #endif 1019 #ifdef INET 1020 tlen = sizeof (struct tcpiphdr); 1021 #endif 1022 #ifdef INVARIANTS 1023 m->m_len = 0; 1024 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1025 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1026 m, tlen, (long)M_TRAILINGSPACE(m))); 1027 #endif 1028 m->m_len = tlen; 1029 to.to_flags = 0; 1030 if (incl_opts) { 1031 /* Make sure we have room. */ 1032 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1033 m->m_next = m_get(M_NOWAIT, MT_DATA); 1034 if (m->m_next) { 1035 optp = mtod(m->m_next, u_char *); 1036 optm = m->m_next; 1037 } else 1038 incl_opts = false; 1039 } else { 1040 optp = (u_char *) (nth + 1); 1041 optm = m; 1042 } 1043 } 1044 if (incl_opts) { 1045 /* Timestamps. */ 1046 if (tp->t_flags & TF_RCVD_TSTMP) { 1047 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1048 to.to_tsecr = tp->ts_recent; 1049 to.to_flags |= TOF_TS; 1050 } 1051 #ifdef TCP_SIGNATURE 1052 /* TCP-MD5 (RFC2385). */ 1053 if (tp->t_flags & TF_SIGNATURE) 1054 to.to_flags |= TOF_SIGNATURE; 1055 #endif 1056 1057 /* Add the options. */ 1058 tlen += optlen = tcp_addoptions(&to, optp); 1059 1060 /* Update m_len in the correct mbuf. */ 1061 optm->m_len += optlen; 1062 } else 1063 optlen = 0; 1064 #ifdef INET6 1065 if (isipv6) { 1066 ip6->ip6_flow = 0; 1067 ip6->ip6_vfc = IPV6_VERSION; 1068 ip6->ip6_nxt = IPPROTO_TCP; 1069 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1070 } 1071 #endif 1072 #if defined(INET) && defined(INET6) 1073 else 1074 #endif 1075 #ifdef INET 1076 { 1077 ip->ip_len = htons(tlen); 1078 ip->ip_ttl = V_ip_defttl; 1079 if (V_path_mtu_discovery) 1080 ip->ip_off |= htons(IP_DF); 1081 } 1082 #endif 1083 m->m_pkthdr.len = tlen; 1084 m->m_pkthdr.rcvif = NULL; 1085 #ifdef MAC 1086 if (inp != NULL) { 1087 /* 1088 * Packet is associated with a socket, so allow the 1089 * label of the response to reflect the socket label. 1090 */ 1091 INP_WLOCK_ASSERT(inp); 1092 mac_inpcb_create_mbuf(inp, m); 1093 } else { 1094 /* 1095 * Packet is not associated with a socket, so possibly 1096 * update the label in place. 1097 */ 1098 mac_netinet_tcp_reply(m); 1099 } 1100 #endif 1101 nth->th_seq = htonl(seq); 1102 nth->th_ack = htonl(ack); 1103 nth->th_x2 = 0; 1104 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1105 nth->th_flags = flags; 1106 if (tp != NULL) 1107 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1108 else 1109 nth->th_win = htons((u_short)win); 1110 nth->th_urp = 0; 1111 1112 #ifdef TCP_SIGNATURE 1113 if (to.to_flags & TOF_SIGNATURE) { 1114 tcp_signature_compute(m, 0, 0, optlen, to.to_signature, 1115 IPSEC_DIR_OUTBOUND); 1116 } 1117 #endif 1118 1119 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1120 #ifdef INET6 1121 if (isipv6) { 1122 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1123 nth->th_sum = in6_cksum_pseudo(ip6, 1124 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1125 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1126 NULL, NULL); 1127 } 1128 #endif /* INET6 */ 1129 #if defined(INET6) && defined(INET) 1130 else 1131 #endif 1132 #ifdef INET 1133 { 1134 m->m_pkthdr.csum_flags = CSUM_TCP; 1135 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1136 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1137 } 1138 #endif /* INET */ 1139 #ifdef TCPDEBUG 1140 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1141 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1142 #endif 1143 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *)); 1144 if (flags & TH_RST) 1145 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), 1146 tp, nth); 1147 1148 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); 1149 #ifdef INET6 1150 if (isipv6) 1151 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1152 #endif /* INET6 */ 1153 #if defined(INET) && defined(INET6) 1154 else 1155 #endif 1156 #ifdef INET 1157 (void) ip_output(m, NULL, NULL, 0, NULL, inp); 1158 #endif 1159 } 1160 1161 /* 1162 * Create a new TCP control block, making an 1163 * empty reassembly queue and hooking it to the argument 1164 * protocol control block. The `inp' parameter must have 1165 * come from the zone allocator set up in tcp_init(). 1166 */ 1167 struct tcpcb * 1168 tcp_newtcpcb(struct inpcb *inp) 1169 { 1170 struct tcpcb_mem *tm; 1171 struct tcpcb *tp; 1172 #ifdef INET6 1173 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1174 #endif /* INET6 */ 1175 1176 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1177 if (tm == NULL) 1178 return (NULL); 1179 tp = &tm->tcb; 1180 1181 /* Initialise cc_var struct for this tcpcb. */ 1182 tp->ccv = &tm->ccv; 1183 tp->ccv->type = IPPROTO_TCP; 1184 tp->ccv->ccvc.tcp = tp; 1185 rw_rlock(&tcp_function_lock); 1186 tp->t_fb = tcp_func_set_ptr; 1187 refcount_acquire(&tp->t_fb->tfb_refcnt); 1188 rw_runlock(&tcp_function_lock); 1189 if (tp->t_fb->tfb_tcp_fb_init) { 1190 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1191 } 1192 /* 1193 * Use the current system default CC algorithm. 1194 */ 1195 CC_LIST_RLOCK(); 1196 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1197 CC_ALGO(tp) = CC_DEFAULT(); 1198 CC_LIST_RUNLOCK(); 1199 1200 if (CC_ALGO(tp)->cb_init != NULL) 1201 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1202 if (tp->t_fb->tfb_tcp_fb_fini) 1203 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1204 refcount_release(&tp->t_fb->tfb_refcnt); 1205 uma_zfree(V_tcpcb_zone, tm); 1206 return (NULL); 1207 } 1208 1209 tp->osd = &tm->osd; 1210 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1211 if (tp->t_fb->tfb_tcp_fb_fini) 1212 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1213 refcount_release(&tp->t_fb->tfb_refcnt); 1214 uma_zfree(V_tcpcb_zone, tm); 1215 return (NULL); 1216 } 1217 1218 #ifdef VIMAGE 1219 tp->t_vnet = inp->inp_vnet; 1220 #endif 1221 tp->t_timers = &tm->tt; 1222 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1223 tp->t_maxseg = 1224 #ifdef INET6 1225 isipv6 ? V_tcp_v6mssdflt : 1226 #endif /* INET6 */ 1227 V_tcp_mssdflt; 1228 1229 /* Set up our timeouts. */ 1230 callout_init(&tp->t_timers->tt_rexmt, 1); 1231 callout_init(&tp->t_timers->tt_persist, 1); 1232 callout_init(&tp->t_timers->tt_keep, 1); 1233 callout_init(&tp->t_timers->tt_2msl, 1); 1234 callout_init(&tp->t_timers->tt_delack, 1); 1235 1236 if (V_tcp_do_rfc1323) 1237 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1238 if (V_tcp_do_sack) 1239 tp->t_flags |= TF_SACK_PERMIT; 1240 TAILQ_INIT(&tp->snd_holes); 1241 /* 1242 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1243 * is called. 1244 */ 1245 in_pcbref(inp); /* Reference for tcpcb */ 1246 tp->t_inpcb = inp; 1247 1248 /* 1249 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1250 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1251 * reasonable initial retransmit time. 1252 */ 1253 tp->t_srtt = TCPTV_SRTTBASE; 1254 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1255 tp->t_rttmin = tcp_rexmit_min; 1256 tp->t_rxtcur = TCPTV_RTOBASE; 1257 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1258 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1259 tp->t_rcvtime = ticks; 1260 /* 1261 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1262 * because the socket may be bound to an IPv6 wildcard address, 1263 * which may match an IPv4-mapped IPv6 address. 1264 */ 1265 inp->inp_ip_ttl = V_ip_defttl; 1266 inp->inp_ppcb = tp; 1267 #ifdef TCPPCAP 1268 /* 1269 * Init the TCP PCAP queues. 1270 */ 1271 tcp_pcap_tcpcb_init(tp); 1272 #endif 1273 return (tp); /* XXX */ 1274 } 1275 1276 /* 1277 * Switch the congestion control algorithm back to NewReno for any active 1278 * control blocks using an algorithm which is about to go away. 1279 * This ensures the CC framework can allow the unload to proceed without leaving 1280 * any dangling pointers which would trigger a panic. 1281 * Returning non-zero would inform the CC framework that something went wrong 1282 * and it would be unsafe to allow the unload to proceed. However, there is no 1283 * way for this to occur with this implementation so we always return zero. 1284 */ 1285 int 1286 tcp_ccalgounload(struct cc_algo *unload_algo) 1287 { 1288 struct cc_algo *tmpalgo; 1289 struct inpcb *inp; 1290 struct tcpcb *tp; 1291 VNET_ITERATOR_DECL(vnet_iter); 1292 1293 /* 1294 * Check all active control blocks across all network stacks and change 1295 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1296 * requires cleanup code to be run, call it. 1297 */ 1298 VNET_LIST_RLOCK(); 1299 VNET_FOREACH(vnet_iter) { 1300 CURVNET_SET(vnet_iter); 1301 INP_INFO_WLOCK(&V_tcbinfo); 1302 /* 1303 * New connections already part way through being initialised 1304 * with the CC algo we're removing will not race with this code 1305 * because the INP_INFO_WLOCK is held during initialisation. We 1306 * therefore don't enter the loop below until the connection 1307 * list has stabilised. 1308 */ 1309 LIST_FOREACH(inp, &V_tcb, inp_list) { 1310 INP_WLOCK(inp); 1311 /* Important to skip tcptw structs. */ 1312 if (!(inp->inp_flags & INP_TIMEWAIT) && 1313 (tp = intotcpcb(inp)) != NULL) { 1314 /* 1315 * By holding INP_WLOCK here, we are assured 1316 * that the connection is not currently 1317 * executing inside the CC module's functions 1318 * i.e. it is safe to make the switch back to 1319 * NewReno. 1320 */ 1321 if (CC_ALGO(tp) == unload_algo) { 1322 tmpalgo = CC_ALGO(tp); 1323 /* NewReno does not require any init. */ 1324 CC_ALGO(tp) = &newreno_cc_algo; 1325 if (tmpalgo->cb_destroy != NULL) 1326 tmpalgo->cb_destroy(tp->ccv); 1327 } 1328 } 1329 INP_WUNLOCK(inp); 1330 } 1331 INP_INFO_WUNLOCK(&V_tcbinfo); 1332 CURVNET_RESTORE(); 1333 } 1334 VNET_LIST_RUNLOCK(); 1335 1336 return (0); 1337 } 1338 1339 /* 1340 * Drop a TCP connection, reporting 1341 * the specified error. If connection is synchronized, 1342 * then send a RST to peer. 1343 */ 1344 struct tcpcb * 1345 tcp_drop(struct tcpcb *tp, int errno) 1346 { 1347 struct socket *so = tp->t_inpcb->inp_socket; 1348 1349 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1350 INP_WLOCK_ASSERT(tp->t_inpcb); 1351 1352 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1353 tcp_state_change(tp, TCPS_CLOSED); 1354 (void) tp->t_fb->tfb_tcp_output(tp); 1355 TCPSTAT_INC(tcps_drops); 1356 } else 1357 TCPSTAT_INC(tcps_conndrops); 1358 if (errno == ETIMEDOUT && tp->t_softerror) 1359 errno = tp->t_softerror; 1360 so->so_error = errno; 1361 return (tcp_close(tp)); 1362 } 1363 1364 void 1365 tcp_discardcb(struct tcpcb *tp) 1366 { 1367 struct inpcb *inp = tp->t_inpcb; 1368 struct socket *so = inp->inp_socket; 1369 #ifdef INET6 1370 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1371 #endif /* INET6 */ 1372 int released; 1373 1374 INP_WLOCK_ASSERT(inp); 1375 1376 /* 1377 * Make sure that all of our timers are stopped before we delete the 1378 * PCB. 1379 * 1380 * If stopping a timer fails, we schedule a discard function in same 1381 * callout, and the last discard function called will take care of 1382 * deleting the tcpcb. 1383 */ 1384 tp->t_timers->tt_draincnt = 0; 1385 tcp_timer_stop(tp, TT_REXMT); 1386 tcp_timer_stop(tp, TT_PERSIST); 1387 tcp_timer_stop(tp, TT_KEEP); 1388 tcp_timer_stop(tp, TT_2MSL); 1389 tcp_timer_stop(tp, TT_DELACK); 1390 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1391 /* 1392 * Call the stop-all function of the methods, 1393 * this function should call the tcp_timer_stop() 1394 * method with each of the function specific timeouts. 1395 * That stop will be called via the tfb_tcp_timer_stop() 1396 * which should use the async drain function of the 1397 * callout system (see tcp_var.h). 1398 */ 1399 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1400 } 1401 1402 /* 1403 * If we got enough samples through the srtt filter, 1404 * save the rtt and rttvar in the routing entry. 1405 * 'Enough' is arbitrarily defined as 4 rtt samples. 1406 * 4 samples is enough for the srtt filter to converge 1407 * to within enough % of the correct value; fewer samples 1408 * and we could save a bogus rtt. The danger is not high 1409 * as tcp quickly recovers from everything. 1410 * XXX: Works very well but needs some more statistics! 1411 */ 1412 if (tp->t_rttupdated >= 4) { 1413 struct hc_metrics_lite metrics; 1414 u_long ssthresh; 1415 1416 bzero(&metrics, sizeof(metrics)); 1417 /* 1418 * Update the ssthresh always when the conditions below 1419 * are satisfied. This gives us better new start value 1420 * for the congestion avoidance for new connections. 1421 * ssthresh is only set if packet loss occurred on a session. 1422 * 1423 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1424 * being torn down. Ideally this code would not use 'so'. 1425 */ 1426 ssthresh = tp->snd_ssthresh; 1427 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1428 /* 1429 * convert the limit from user data bytes to 1430 * packets then to packet data bytes. 1431 */ 1432 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1433 if (ssthresh < 2) 1434 ssthresh = 2; 1435 ssthresh *= (u_long)(tp->t_maxseg + 1436 #ifdef INET6 1437 (isipv6 ? sizeof (struct ip6_hdr) + 1438 sizeof (struct tcphdr) : 1439 #endif 1440 sizeof (struct tcpiphdr) 1441 #ifdef INET6 1442 ) 1443 #endif 1444 ); 1445 } else 1446 ssthresh = 0; 1447 metrics.rmx_ssthresh = ssthresh; 1448 1449 metrics.rmx_rtt = tp->t_srtt; 1450 metrics.rmx_rttvar = tp->t_rttvar; 1451 metrics.rmx_cwnd = tp->snd_cwnd; 1452 metrics.rmx_sendpipe = 0; 1453 metrics.rmx_recvpipe = 0; 1454 1455 tcp_hc_update(&inp->inp_inc, &metrics); 1456 } 1457 1458 /* free the reassembly queue, if any */ 1459 tcp_reass_flush(tp); 1460 1461 #ifdef TCP_OFFLOAD 1462 /* Disconnect offload device, if any. */ 1463 if (tp->t_flags & TF_TOE) 1464 tcp_offload_detach(tp); 1465 #endif 1466 1467 tcp_free_sackholes(tp); 1468 1469 #ifdef TCPPCAP 1470 /* Free the TCP PCAP queues. */ 1471 tcp_pcap_drain(&(tp->t_inpkts)); 1472 tcp_pcap_drain(&(tp->t_outpkts)); 1473 #endif 1474 1475 /* Allow the CC algorithm to clean up after itself. */ 1476 if (CC_ALGO(tp)->cb_destroy != NULL) 1477 CC_ALGO(tp)->cb_destroy(tp->ccv); 1478 1479 khelp_destroy_osd(tp->osd); 1480 1481 CC_ALGO(tp) = NULL; 1482 inp->inp_ppcb = NULL; 1483 if (tp->t_timers->tt_draincnt == 0) { 1484 /* We own the last reference on tcpcb, let's free it. */ 1485 if (tp->t_fb->tfb_tcp_fb_fini) 1486 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1487 refcount_release(&tp->t_fb->tfb_refcnt); 1488 tp->t_inpcb = NULL; 1489 uma_zfree(V_tcpcb_zone, tp); 1490 released = in_pcbrele_wlocked(inp); 1491 KASSERT(!released, ("%s: inp %p should not have been released " 1492 "here", __func__, inp)); 1493 } 1494 } 1495 1496 void 1497 tcp_timer_discard(void *ptp) 1498 { 1499 struct inpcb *inp; 1500 struct tcpcb *tp; 1501 1502 tp = (struct tcpcb *)ptp; 1503 CURVNET_SET(tp->t_vnet); 1504 INP_INFO_RLOCK(&V_tcbinfo); 1505 inp = tp->t_inpcb; 1506 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1507 __func__, tp)); 1508 INP_WLOCK(inp); 1509 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1510 ("%s: tcpcb has to be stopped here", __func__)); 1511 tp->t_timers->tt_draincnt--; 1512 if (tp->t_timers->tt_draincnt == 0) { 1513 /* We own the last reference on this tcpcb, let's free it. */ 1514 if (tp->t_fb->tfb_tcp_fb_fini) 1515 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1516 refcount_release(&tp->t_fb->tfb_refcnt); 1517 tp->t_inpcb = NULL; 1518 uma_zfree(V_tcpcb_zone, tp); 1519 if (in_pcbrele_wlocked(inp)) { 1520 INP_INFO_RUNLOCK(&V_tcbinfo); 1521 CURVNET_RESTORE(); 1522 return; 1523 } 1524 } 1525 INP_WUNLOCK(inp); 1526 INP_INFO_RUNLOCK(&V_tcbinfo); 1527 CURVNET_RESTORE(); 1528 } 1529 1530 /* 1531 * Attempt to close a TCP control block, marking it as dropped, and freeing 1532 * the socket if we hold the only reference. 1533 */ 1534 struct tcpcb * 1535 tcp_close(struct tcpcb *tp) 1536 { 1537 struct inpcb *inp = tp->t_inpcb; 1538 struct socket *so; 1539 1540 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1541 INP_WLOCK_ASSERT(inp); 1542 1543 #ifdef TCP_OFFLOAD 1544 if (tp->t_state == TCPS_LISTEN) 1545 tcp_offload_listen_stop(tp); 1546 #endif 1547 #ifdef TCP_RFC7413 1548 /* 1549 * This releases the TFO pending counter resource for TFO listen 1550 * sockets as well as passively-created TFO sockets that transition 1551 * from SYN_RECEIVED to CLOSED. 1552 */ 1553 if (tp->t_tfo_pending) { 1554 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1555 tp->t_tfo_pending = NULL; 1556 } 1557 #endif 1558 in_pcbdrop(inp); 1559 TCPSTAT_INC(tcps_closed); 1560 TCPSTATES_DEC(tp->t_state); 1561 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1562 so = inp->inp_socket; 1563 soisdisconnected(so); 1564 if (inp->inp_flags & INP_SOCKREF) { 1565 KASSERT(so->so_state & SS_PROTOREF, 1566 ("tcp_close: !SS_PROTOREF")); 1567 inp->inp_flags &= ~INP_SOCKREF; 1568 INP_WUNLOCK(inp); 1569 ACCEPT_LOCK(); 1570 SOCK_LOCK(so); 1571 so->so_state &= ~SS_PROTOREF; 1572 sofree(so); 1573 return (NULL); 1574 } 1575 return (tp); 1576 } 1577 1578 void 1579 tcp_drain(void) 1580 { 1581 VNET_ITERATOR_DECL(vnet_iter); 1582 1583 if (!do_tcpdrain) 1584 return; 1585 1586 VNET_LIST_RLOCK_NOSLEEP(); 1587 VNET_FOREACH(vnet_iter) { 1588 CURVNET_SET(vnet_iter); 1589 struct inpcb *inpb; 1590 struct tcpcb *tcpb; 1591 1592 /* 1593 * Walk the tcpbs, if existing, and flush the reassembly queue, 1594 * if there is one... 1595 * XXX: The "Net/3" implementation doesn't imply that the TCP 1596 * reassembly queue should be flushed, but in a situation 1597 * where we're really low on mbufs, this is potentially 1598 * useful. 1599 */ 1600 INP_INFO_WLOCK(&V_tcbinfo); 1601 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1602 if (inpb->inp_flags & INP_TIMEWAIT) 1603 continue; 1604 INP_WLOCK(inpb); 1605 if ((tcpb = intotcpcb(inpb)) != NULL) { 1606 tcp_reass_flush(tcpb); 1607 tcp_clean_sackreport(tcpb); 1608 #ifdef TCPPCAP 1609 if (tcp_pcap_aggressive_free) { 1610 /* Free the TCP PCAP queues. */ 1611 tcp_pcap_drain(&(tcpb->t_inpkts)); 1612 tcp_pcap_drain(&(tcpb->t_outpkts)); 1613 } 1614 #endif 1615 } 1616 INP_WUNLOCK(inpb); 1617 } 1618 INP_INFO_WUNLOCK(&V_tcbinfo); 1619 CURVNET_RESTORE(); 1620 } 1621 VNET_LIST_RUNLOCK_NOSLEEP(); 1622 } 1623 1624 /* 1625 * Notify a tcp user of an asynchronous error; 1626 * store error as soft error, but wake up user 1627 * (for now, won't do anything until can select for soft error). 1628 * 1629 * Do not wake up user since there currently is no mechanism for 1630 * reporting soft errors (yet - a kqueue filter may be added). 1631 */ 1632 static struct inpcb * 1633 tcp_notify(struct inpcb *inp, int error) 1634 { 1635 struct tcpcb *tp; 1636 1637 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1638 INP_WLOCK_ASSERT(inp); 1639 1640 if ((inp->inp_flags & INP_TIMEWAIT) || 1641 (inp->inp_flags & INP_DROPPED)) 1642 return (inp); 1643 1644 tp = intotcpcb(inp); 1645 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1646 1647 /* 1648 * Ignore some errors if we are hooked up. 1649 * If connection hasn't completed, has retransmitted several times, 1650 * and receives a second error, give up now. This is better 1651 * than waiting a long time to establish a connection that 1652 * can never complete. 1653 */ 1654 if (tp->t_state == TCPS_ESTABLISHED && 1655 (error == EHOSTUNREACH || error == ENETUNREACH || 1656 error == EHOSTDOWN)) { 1657 if (inp->inp_route.ro_rt) { 1658 RTFREE(inp->inp_route.ro_rt); 1659 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1660 } 1661 return (inp); 1662 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1663 tp->t_softerror) { 1664 tp = tcp_drop(tp, error); 1665 if (tp != NULL) 1666 return (inp); 1667 else 1668 return (NULL); 1669 } else { 1670 tp->t_softerror = error; 1671 return (inp); 1672 } 1673 #if 0 1674 wakeup( &so->so_timeo); 1675 sorwakeup(so); 1676 sowwakeup(so); 1677 #endif 1678 } 1679 1680 static int 1681 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1682 { 1683 int error, i, m, n, pcb_count; 1684 struct inpcb *inp, **inp_list; 1685 inp_gen_t gencnt; 1686 struct xinpgen xig; 1687 1688 /* 1689 * The process of preparing the TCB list is too time-consuming and 1690 * resource-intensive to repeat twice on every request. 1691 */ 1692 if (req->oldptr == NULL) { 1693 n = V_tcbinfo.ipi_count + 1694 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1695 n += imax(n / 8, 10); 1696 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1697 return (0); 1698 } 1699 1700 if (req->newptr != NULL) 1701 return (EPERM); 1702 1703 /* 1704 * OK, now we're committed to doing something. 1705 */ 1706 INP_LIST_RLOCK(&V_tcbinfo); 1707 gencnt = V_tcbinfo.ipi_gencnt; 1708 n = V_tcbinfo.ipi_count; 1709 INP_LIST_RUNLOCK(&V_tcbinfo); 1710 1711 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1712 1713 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1714 + (n + m) * sizeof(struct xtcpcb)); 1715 if (error != 0) 1716 return (error); 1717 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_TIMXCEED_INTRANS) && ip) 1952 notify = tcp_drop_syn_sent; 1953 else if (PRC_IS_REDIRECT(cmd)) { 1954 /* signal EHOSTDOWN, as it flushes the cached route */ 1955 in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify); 1956 return; 1957 } 1958 /* 1959 * Hostdead is ugly because it goes linearly through all PCBs. 1960 * XXX: We never get this from ICMP, otherwise it makes an 1961 * excellent DoS attack on machines with many connections. 1962 */ 1963 else if (cmd == PRC_HOSTDEAD) 1964 ip = NULL; 1965 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1966 return; 1967 1968 if (ip == NULL) { 1969 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1970 return; 1971 } 1972 1973 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1974 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1975 INP_INFO_RLOCK(&V_tcbinfo); 1976 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 1977 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1978 if (inp != NULL) { 1979 if (!(inp->inp_flags & INP_TIMEWAIT) && 1980 !(inp->inp_flags & INP_DROPPED) && 1981 !(inp->inp_socket == NULL)) { 1982 icmp_tcp_seq = ntohl(th->th_seq); 1983 tp = intotcpcb(inp); 1984 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1985 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1986 if (cmd == PRC_MSGSIZE) { 1987 /* 1988 * MTU discovery: 1989 * If we got a needfrag set the MTU 1990 * in the route to the suggested new 1991 * value (if given) and then notify. 1992 */ 1993 mtu = ntohs(icp->icmp_nextmtu); 1994 /* 1995 * If no alternative MTU was 1996 * proposed, try the next smaller 1997 * one. 1998 */ 1999 if (!mtu) 2000 mtu = ip_next_mtu( 2001 ntohs(ip->ip_len), 1); 2002 if (mtu < V_tcp_minmss + 2003 sizeof(struct tcpiphdr)) 2004 mtu = V_tcp_minmss + 2005 sizeof(struct tcpiphdr); 2006 /* 2007 * Only process the offered MTU if it 2008 * is smaller than the current one. 2009 */ 2010 if (mtu < tp->t_maxseg + 2011 sizeof(struct tcpiphdr)) { 2012 bzero(&inc, sizeof(inc)); 2013 inc.inc_faddr = faddr; 2014 inc.inc_fibnum = 2015 inp->inp_inc.inc_fibnum; 2016 tcp_hc_updatemtu(&inc, mtu); 2017 tcp_mtudisc(inp, mtu); 2018 } 2019 } else 2020 inp = (*notify)(inp, 2021 inetctlerrmap[cmd]); 2022 } 2023 } 2024 if (inp != NULL) 2025 INP_WUNLOCK(inp); 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, th); 2033 } 2034 INP_INFO_RUNLOCK(&V_tcbinfo); 2035 } 2036 #endif /* INET */ 2037 2038 #ifdef INET6 2039 void 2040 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2041 { 2042 struct tcphdr th; 2043 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2044 struct ip6_hdr *ip6; 2045 struct mbuf *m; 2046 struct ip6ctlparam *ip6cp = NULL; 2047 const struct sockaddr_in6 *sa6_src = NULL; 2048 int off; 2049 struct tcp_portonly { 2050 u_int16_t th_sport; 2051 u_int16_t th_dport; 2052 } *thp; 2053 2054 if (sa->sa_family != AF_INET6 || 2055 sa->sa_len != sizeof(struct sockaddr_in6)) 2056 return; 2057 2058 if (cmd == PRC_MSGSIZE) 2059 notify = tcp_mtudisc_notify; 2060 else if (!PRC_IS_REDIRECT(cmd) && 2061 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 2062 return; 2063 2064 /* if the parameter is from icmp6, decode it. */ 2065 if (d != NULL) { 2066 ip6cp = (struct ip6ctlparam *)d; 2067 m = ip6cp->ip6c_m; 2068 ip6 = ip6cp->ip6c_ip6; 2069 off = ip6cp->ip6c_off; 2070 sa6_src = ip6cp->ip6c_src; 2071 } else { 2072 m = NULL; 2073 ip6 = NULL; 2074 off = 0; /* fool gcc */ 2075 sa6_src = &sa6_any; 2076 } 2077 2078 if (ip6 != NULL) { 2079 struct in_conninfo inc; 2080 /* 2081 * XXX: We assume that when IPV6 is non NULL, 2082 * M and OFF are valid. 2083 */ 2084 2085 /* check if we can safely examine src and dst ports */ 2086 if (m->m_pkthdr.len < off + sizeof(*thp)) 2087 return; 2088 2089 bzero(&th, sizeof(th)); 2090 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 2091 2092 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 2093 (struct sockaddr *)ip6cp->ip6c_src, 2094 th.th_sport, cmd, NULL, notify); 2095 2096 bzero(&inc, sizeof(inc)); 2097 inc.inc_fport = th.th_dport; 2098 inc.inc_lport = th.th_sport; 2099 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 2100 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 2101 inc.inc_flags |= INC_ISIPV6; 2102 INP_INFO_RLOCK(&V_tcbinfo); 2103 syncache_unreach(&inc, &th); 2104 INP_INFO_RUNLOCK(&V_tcbinfo); 2105 } else 2106 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 2107 0, cmd, NULL, notify); 2108 } 2109 #endif /* INET6 */ 2110 2111 2112 /* 2113 * Following is where TCP initial sequence number generation occurs. 2114 * 2115 * There are two places where we must use initial sequence numbers: 2116 * 1. In SYN-ACK packets. 2117 * 2. In SYN packets. 2118 * 2119 * All ISNs for SYN-ACK packets are generated by the syncache. See 2120 * tcp_syncache.c for details. 2121 * 2122 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2123 * depends on this property. In addition, these ISNs should be 2124 * unguessable so as to prevent connection hijacking. To satisfy 2125 * the requirements of this situation, the algorithm outlined in 2126 * RFC 1948 is used, with only small modifications. 2127 * 2128 * Implementation details: 2129 * 2130 * Time is based off the system timer, and is corrected so that it 2131 * increases by one megabyte per second. This allows for proper 2132 * recycling on high speed LANs while still leaving over an hour 2133 * before rollover. 2134 * 2135 * As reading the *exact* system time is too expensive to be done 2136 * whenever setting up a TCP connection, we increment the time 2137 * offset in two ways. First, a small random positive increment 2138 * is added to isn_offset for each connection that is set up. 2139 * Second, the function tcp_isn_tick fires once per clock tick 2140 * and increments isn_offset as necessary so that sequence numbers 2141 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2142 * random positive increments serve only to ensure that the same 2143 * exact sequence number is never sent out twice (as could otherwise 2144 * happen when a port is recycled in less than the system tick 2145 * interval.) 2146 * 2147 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2148 * between seeding of isn_secret. This is normally set to zero, 2149 * as reseeding should not be necessary. 2150 * 2151 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2152 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2153 * general, this means holding an exclusive (write) lock. 2154 */ 2155 2156 #define ISN_BYTES_PER_SECOND 1048576 2157 #define ISN_STATIC_INCREMENT 4096 2158 #define ISN_RANDOM_INCREMENT (4096 - 1) 2159 2160 static VNET_DEFINE(u_char, isn_secret[32]); 2161 static VNET_DEFINE(int, isn_last); 2162 static VNET_DEFINE(int, isn_last_reseed); 2163 static VNET_DEFINE(u_int32_t, isn_offset); 2164 static VNET_DEFINE(u_int32_t, isn_offset_old); 2165 2166 #define V_isn_secret VNET(isn_secret) 2167 #define V_isn_last VNET(isn_last) 2168 #define V_isn_last_reseed VNET(isn_last_reseed) 2169 #define V_isn_offset VNET(isn_offset) 2170 #define V_isn_offset_old VNET(isn_offset_old) 2171 2172 tcp_seq 2173 tcp_new_isn(struct tcpcb *tp) 2174 { 2175 MD5_CTX isn_ctx; 2176 u_int32_t md5_buffer[4]; 2177 tcp_seq new_isn; 2178 u_int32_t projected_offset; 2179 2180 INP_WLOCK_ASSERT(tp->t_inpcb); 2181 2182 ISN_LOCK(); 2183 /* Seed if this is the first use, reseed if requested. */ 2184 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2185 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2186 < (u_int)ticks))) { 2187 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2188 V_isn_last_reseed = ticks; 2189 } 2190 2191 /* Compute the md5 hash and return the ISN. */ 2192 MD5Init(&isn_ctx); 2193 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2194 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2195 #ifdef INET6 2196 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2197 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2198 sizeof(struct in6_addr)); 2199 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2200 sizeof(struct in6_addr)); 2201 } else 2202 #endif 2203 { 2204 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2205 sizeof(struct in_addr)); 2206 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2207 sizeof(struct in_addr)); 2208 } 2209 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2210 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2211 new_isn = (tcp_seq) md5_buffer[0]; 2212 V_isn_offset += ISN_STATIC_INCREMENT + 2213 (arc4random() & ISN_RANDOM_INCREMENT); 2214 if (ticks != V_isn_last) { 2215 projected_offset = V_isn_offset_old + 2216 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2217 if (SEQ_GT(projected_offset, V_isn_offset)) 2218 V_isn_offset = projected_offset; 2219 V_isn_offset_old = V_isn_offset; 2220 V_isn_last = ticks; 2221 } 2222 new_isn += V_isn_offset; 2223 ISN_UNLOCK(); 2224 return (new_isn); 2225 } 2226 2227 /* 2228 * When a specific ICMP unreachable message is received and the 2229 * connection state is SYN-SENT, drop the connection. This behavior 2230 * is controlled by the icmp_may_rst sysctl. 2231 */ 2232 struct inpcb * 2233 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2234 { 2235 struct tcpcb *tp; 2236 2237 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2238 INP_WLOCK_ASSERT(inp); 2239 2240 if ((inp->inp_flags & INP_TIMEWAIT) || 2241 (inp->inp_flags & INP_DROPPED)) 2242 return (inp); 2243 2244 tp = intotcpcb(inp); 2245 if (tp->t_state != TCPS_SYN_SENT) 2246 return (inp); 2247 2248 tp = tcp_drop(tp, errno); 2249 if (tp != NULL) 2250 return (inp); 2251 else 2252 return (NULL); 2253 } 2254 2255 /* 2256 * When `need fragmentation' ICMP is received, update our idea of the MSS 2257 * based on the new value. Also nudge TCP to send something, since we 2258 * know the packet we just sent was dropped. 2259 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2260 */ 2261 static struct inpcb * 2262 tcp_mtudisc_notify(struct inpcb *inp, int error) 2263 { 2264 2265 tcp_mtudisc(inp, -1); 2266 return (inp); 2267 } 2268 2269 static void 2270 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2271 { 2272 struct tcpcb *tp; 2273 struct socket *so; 2274 2275 INP_WLOCK_ASSERT(inp); 2276 if ((inp->inp_flags & INP_TIMEWAIT) || 2277 (inp->inp_flags & INP_DROPPED)) 2278 return; 2279 2280 tp = intotcpcb(inp); 2281 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2282 2283 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2284 2285 so = inp->inp_socket; 2286 SOCKBUF_LOCK(&so->so_snd); 2287 /* If the mss is larger than the socket buffer, decrease the mss. */ 2288 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2289 tp->t_maxseg = so->so_snd.sb_hiwat; 2290 SOCKBUF_UNLOCK(&so->so_snd); 2291 2292 TCPSTAT_INC(tcps_mturesent); 2293 tp->t_rtttime = 0; 2294 tp->snd_nxt = tp->snd_una; 2295 tcp_free_sackholes(tp); 2296 tp->snd_recover = tp->snd_max; 2297 if (tp->t_flags & TF_SACK_PERMIT) 2298 EXIT_FASTRECOVERY(tp->t_flags); 2299 tp->t_fb->tfb_tcp_output(tp); 2300 } 2301 2302 #ifdef INET 2303 /* 2304 * Look-up the routing entry to the peer of this inpcb. If no route 2305 * is found and it cannot be allocated, then return 0. This routine 2306 * is called by TCP routines that access the rmx structure and by 2307 * tcp_mss_update to get the peer/interface MTU. 2308 */ 2309 u_long 2310 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2311 { 2312 struct nhop4_extended nh4; 2313 struct ifnet *ifp; 2314 u_long maxmtu = 0; 2315 2316 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2317 2318 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2319 2320 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2321 NHR_REF, 0, &nh4) != 0) 2322 return (0); 2323 2324 ifp = nh4.nh_ifp; 2325 maxmtu = nh4.nh_mtu; 2326 2327 /* Report additional interface capabilities. */ 2328 if (cap != NULL) { 2329 if (ifp->if_capenable & IFCAP_TSO4 && 2330 ifp->if_hwassist & CSUM_TSO) { 2331 cap->ifcap |= CSUM_TSO; 2332 cap->tsomax = ifp->if_hw_tsomax; 2333 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2334 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2335 } 2336 } 2337 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2338 } 2339 return (maxmtu); 2340 } 2341 #endif /* INET */ 2342 2343 #ifdef INET6 2344 u_long 2345 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2346 { 2347 struct nhop6_extended nh6; 2348 struct in6_addr dst6; 2349 uint32_t scopeid; 2350 struct ifnet *ifp; 2351 u_long maxmtu = 0; 2352 2353 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2354 2355 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2356 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2357 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2358 0, &nh6) != 0) 2359 return (0); 2360 2361 ifp = nh6.nh_ifp; 2362 maxmtu = nh6.nh_mtu; 2363 2364 /* Report additional interface capabilities. */ 2365 if (cap != NULL) { 2366 if (ifp->if_capenable & IFCAP_TSO6 && 2367 ifp->if_hwassist & CSUM_TSO) { 2368 cap->ifcap |= CSUM_TSO; 2369 cap->tsomax = ifp->if_hw_tsomax; 2370 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2371 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2372 } 2373 } 2374 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2375 } 2376 2377 return (maxmtu); 2378 } 2379 #endif /* INET6 */ 2380 2381 /* 2382 * Calculate effective SMSS per RFC5681 definition for a given TCP 2383 * connection at its current state, taking into account SACK and etc. 2384 */ 2385 u_int 2386 tcp_maxseg(const struct tcpcb *tp) 2387 { 2388 u_int optlen; 2389 2390 if (tp->t_flags & TF_NOOPT) 2391 return (tp->t_maxseg); 2392 2393 /* 2394 * Here we have a simplified code from tcp_addoptions(), 2395 * without a proper loop, and having most of paddings hardcoded. 2396 * We might make mistakes with padding here in some edge cases, 2397 * but this is harmless, since result of tcp_maxseg() is used 2398 * only in cwnd and ssthresh estimations. 2399 */ 2400 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2401 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2402 if (tp->t_flags & TF_RCVD_TSTMP) 2403 optlen = TCPOLEN_TSTAMP_APPA; 2404 else 2405 optlen = 0; 2406 #ifdef TCP_SIGNATURE 2407 if (tp->t_flags & TF_SIGNATURE) 2408 optlen += PAD(TCPOLEN_SIGNATURE); 2409 #endif 2410 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2411 optlen += TCPOLEN_SACKHDR; 2412 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2413 optlen = PAD(optlen); 2414 } 2415 } else { 2416 if (tp->t_flags & TF_REQ_TSTMP) 2417 optlen = TCPOLEN_TSTAMP_APPA; 2418 else 2419 optlen = PAD(TCPOLEN_MAXSEG); 2420 if (tp->t_flags & TF_REQ_SCALE) 2421 optlen += PAD(TCPOLEN_WINDOW); 2422 #ifdef TCP_SIGNATURE 2423 if (tp->t_flags & TF_SIGNATURE) 2424 optlen += PAD(TCPOLEN_SIGNATURE); 2425 #endif 2426 if (tp->t_flags & TF_SACK_PERMIT) 2427 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2428 } 2429 #undef PAD 2430 optlen = min(optlen, TCP_MAXOLEN); 2431 return (tp->t_maxseg - optlen); 2432 } 2433 2434 #ifdef IPSEC 2435 /* compute ESP/AH header size for TCP, including outer IP header. */ 2436 size_t 2437 ipsec_hdrsiz_tcp(struct tcpcb *tp) 2438 { 2439 struct inpcb *inp; 2440 struct mbuf *m; 2441 size_t hdrsiz; 2442 struct ip *ip; 2443 #ifdef INET6 2444 struct ip6_hdr *ip6; 2445 #endif 2446 struct tcphdr *th; 2447 2448 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || 2449 (!key_havesp(IPSEC_DIR_OUTBOUND))) 2450 return (0); 2451 m = m_gethdr(M_NOWAIT, MT_DATA); 2452 if (!m) 2453 return (0); 2454 2455 #ifdef INET6 2456 if ((inp->inp_vflag & INP_IPV6) != 0) { 2457 ip6 = mtod(m, struct ip6_hdr *); 2458 th = (struct tcphdr *)(ip6 + 1); 2459 m->m_pkthdr.len = m->m_len = 2460 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2461 tcpip_fillheaders(inp, ip6, th); 2462 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2463 } else 2464 #endif /* INET6 */ 2465 { 2466 ip = mtod(m, struct ip *); 2467 th = (struct tcphdr *)(ip + 1); 2468 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 2469 tcpip_fillheaders(inp, ip, th); 2470 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2471 } 2472 2473 m_free(m); 2474 return (hdrsiz); 2475 } 2476 #endif /* IPSEC */ 2477 2478 #ifdef TCP_SIGNATURE 2479 /* 2480 * Callback function invoked by m_apply() to digest TCP segment data 2481 * contained within an mbuf chain. 2482 */ 2483 static int 2484 tcp_signature_apply(void *fstate, void *data, u_int len) 2485 { 2486 2487 MD5Update(fstate, (u_char *)data, len); 2488 return (0); 2489 } 2490 2491 /* 2492 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2493 * search with the destination IP address, and a 'magic SPI' to be 2494 * determined by the application. This is hardcoded elsewhere to 1179 2495 */ 2496 struct secasvar * 2497 tcp_get_sav(struct mbuf *m, u_int direction) 2498 { 2499 union sockaddr_union dst; 2500 struct secasvar *sav; 2501 struct ip *ip; 2502 #ifdef INET6 2503 struct ip6_hdr *ip6; 2504 char ip6buf[INET6_ADDRSTRLEN]; 2505 #endif 2506 2507 /* Extract the destination from the IP header in the mbuf. */ 2508 bzero(&dst, sizeof(union sockaddr_union)); 2509 ip = mtod(m, struct ip *); 2510 #ifdef INET6 2511 ip6 = NULL; /* Make the compiler happy. */ 2512 #endif 2513 switch (ip->ip_v) { 2514 #ifdef INET 2515 case IPVERSION: 2516 dst.sa.sa_len = sizeof(struct sockaddr_in); 2517 dst.sa.sa_family = AF_INET; 2518 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2519 ip->ip_src : ip->ip_dst; 2520 break; 2521 #endif 2522 #ifdef INET6 2523 case (IPV6_VERSION >> 4): 2524 ip6 = mtod(m, struct ip6_hdr *); 2525 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2526 dst.sa.sa_family = AF_INET6; 2527 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 2528 ip6->ip6_src : ip6->ip6_dst; 2529 break; 2530 #endif 2531 default: 2532 return (NULL); 2533 /* NOTREACHED */ 2534 break; 2535 } 2536 2537 /* Look up an SADB entry which matches the address of the peer. */ 2538 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2539 if (sav == NULL) { 2540 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 2541 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 2542 #ifdef INET6 2543 (ip->ip_v == (IPV6_VERSION >> 4)) ? 2544 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 2545 #endif 2546 "(unsupported)")); 2547 } 2548 2549 return (sav); 2550 } 2551 2552 /* 2553 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2554 * 2555 * Parameters: 2556 * m pointer to head of mbuf chain 2557 * len length of TCP segment data, excluding options 2558 * optlen length of TCP segment options 2559 * buf pointer to storage for computed MD5 digest 2560 * sav pointer to security assosiation 2561 * 2562 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2563 * When called from tcp_input(), we can be sure that th_sum has been 2564 * zeroed out and verified already. 2565 * 2566 * Releases reference to SADB key before return. 2567 * 2568 * Return 0 if successful, otherwise return -1. 2569 * 2570 */ 2571 int 2572 tcp_signature_do_compute(struct mbuf *m, int len, int optlen, 2573 u_char *buf, struct secasvar *sav) 2574 { 2575 #ifdef INET 2576 struct ippseudo ippseudo; 2577 #endif 2578 MD5_CTX ctx; 2579 int doff; 2580 struct ip *ip; 2581 #ifdef INET 2582 struct ipovly *ipovly; 2583 #endif 2584 struct tcphdr *th; 2585 #ifdef INET6 2586 struct ip6_hdr *ip6; 2587 struct in6_addr in6; 2588 uint32_t plen; 2589 uint16_t nhdr; 2590 #endif 2591 u_short savecsum; 2592 2593 KASSERT(m != NULL, ("NULL mbuf chain")); 2594 KASSERT(buf != NULL, ("NULL signature pointer")); 2595 2596 /* Extract the destination from the IP header in the mbuf. */ 2597 ip = mtod(m, struct ip *); 2598 #ifdef INET6 2599 ip6 = NULL; /* Make the compiler happy. */ 2600 #endif 2601 2602 MD5Init(&ctx); 2603 /* 2604 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 2605 * 2606 * XXX The ippseudo header MUST be digested in network byte order, 2607 * or else we'll fail the regression test. Assume all fields we've 2608 * been doing arithmetic on have been in host byte order. 2609 * XXX One cannot depend on ipovly->ih_len here. When called from 2610 * tcp_output(), the underlying ip_len member has not yet been set. 2611 */ 2612 switch (ip->ip_v) { 2613 #ifdef INET 2614 case IPVERSION: 2615 ipovly = (struct ipovly *)ip; 2616 ippseudo.ippseudo_src = ipovly->ih_src; 2617 ippseudo.ippseudo_dst = ipovly->ih_dst; 2618 ippseudo.ippseudo_pad = 0; 2619 ippseudo.ippseudo_p = IPPROTO_TCP; 2620 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 2621 optlen); 2622 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2623 2624 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2625 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2626 break; 2627 #endif 2628 #ifdef INET6 2629 /* 2630 * RFC 2385, 2.0 Proposal 2631 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2632 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2633 * extended next header value (to form 32 bits), and 32-bit segment 2634 * length. 2635 * Note: Upper-Layer Packet Length comes before Next Header. 2636 */ 2637 case (IPV6_VERSION >> 4): 2638 in6 = ip6->ip6_src; 2639 in6_clearscope(&in6); 2640 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2641 in6 = ip6->ip6_dst; 2642 in6_clearscope(&in6); 2643 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2644 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2645 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2646 nhdr = 0; 2647 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2648 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2649 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2650 nhdr = IPPROTO_TCP; 2651 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2652 2653 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2654 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2655 break; 2656 #endif 2657 default: 2658 KEY_FREESAV(&sav); 2659 return (-1); 2660 /* NOTREACHED */ 2661 break; 2662 } 2663 2664 2665 /* 2666 * Step 2: Update MD5 hash with TCP header, excluding options. 2667 * The TCP checksum must be set to zero. 2668 */ 2669 savecsum = th->th_sum; 2670 th->th_sum = 0; 2671 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2672 th->th_sum = savecsum; 2673 2674 /* 2675 * Step 3: Update MD5 hash with TCP segment data. 2676 * Use m_apply() to avoid an early m_pullup(). 2677 */ 2678 if (len > 0) 2679 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2680 2681 /* 2682 * Step 4: Update MD5 hash with shared secret. 2683 */ 2684 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2685 MD5Final(buf, &ctx); 2686 2687 key_sa_recordxfer(sav, m); 2688 KEY_FREESAV(&sav); 2689 return (0); 2690 } 2691 2692 /* 2693 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2694 * 2695 * Return 0 if successful, otherwise return -1. 2696 */ 2697 int 2698 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 2699 u_char *buf, u_int direction) 2700 { 2701 struct secasvar *sav; 2702 2703 if ((sav = tcp_get_sav(m, direction)) == NULL) 2704 return (-1); 2705 2706 return (tcp_signature_do_compute(m, len, optlen, buf, sav)); 2707 } 2708 2709 /* 2710 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2711 * 2712 * Parameters: 2713 * m pointer to head of mbuf chain 2714 * len length of TCP segment data, excluding options 2715 * optlen length of TCP segment options 2716 * buf pointer to storage for computed MD5 digest 2717 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2718 * 2719 * Return 1 if successful, otherwise return 0. 2720 */ 2721 int 2722 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2723 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2724 { 2725 char tmpdigest[TCP_SIGLEN]; 2726 2727 if (tcp_sig_checksigs == 0) 2728 return (1); 2729 if ((tcpbflag & TF_SIGNATURE) == 0) { 2730 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2731 2732 /* 2733 * If this socket is not expecting signature but 2734 * the segment contains signature just fail. 2735 */ 2736 TCPSTAT_INC(tcps_sig_err_sigopt); 2737 TCPSTAT_INC(tcps_sig_rcvbadsig); 2738 return (0); 2739 } 2740 2741 /* Signature is not expected, and not present in segment. */ 2742 return (1); 2743 } 2744 2745 /* 2746 * If this socket is expecting signature but the segment does not 2747 * contain any just fail. 2748 */ 2749 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2750 TCPSTAT_INC(tcps_sig_err_nosigopt); 2751 TCPSTAT_INC(tcps_sig_rcvbadsig); 2752 return (0); 2753 } 2754 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2755 IPSEC_DIR_INBOUND) == -1) { 2756 TCPSTAT_INC(tcps_sig_err_buildsig); 2757 TCPSTAT_INC(tcps_sig_rcvbadsig); 2758 return (0); 2759 } 2760 2761 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2762 TCPSTAT_INC(tcps_sig_rcvbadsig); 2763 return (0); 2764 } 2765 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2766 return (1); 2767 } 2768 #endif /* TCP_SIGNATURE */ 2769 2770 static int 2771 sysctl_drop(SYSCTL_HANDLER_ARGS) 2772 { 2773 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2774 struct sockaddr_storage addrs[2]; 2775 struct inpcb *inp; 2776 struct tcpcb *tp; 2777 struct tcptw *tw; 2778 struct sockaddr_in *fin, *lin; 2779 #ifdef INET6 2780 struct sockaddr_in6 *fin6, *lin6; 2781 #endif 2782 int error; 2783 2784 inp = NULL; 2785 fin = lin = NULL; 2786 #ifdef INET6 2787 fin6 = lin6 = NULL; 2788 #endif 2789 error = 0; 2790 2791 if (req->oldptr != NULL || req->oldlen != 0) 2792 return (EINVAL); 2793 if (req->newptr == NULL) 2794 return (EPERM); 2795 if (req->newlen < sizeof(addrs)) 2796 return (ENOMEM); 2797 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2798 if (error) 2799 return (error); 2800 2801 switch (addrs[0].ss_family) { 2802 #ifdef INET6 2803 case AF_INET6: 2804 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2805 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2806 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2807 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2808 return (EINVAL); 2809 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2810 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2811 return (EINVAL); 2812 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2813 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2814 fin = (struct sockaddr_in *)&addrs[0]; 2815 lin = (struct sockaddr_in *)&addrs[1]; 2816 break; 2817 } 2818 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2819 if (error) 2820 return (error); 2821 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2822 if (error) 2823 return (error); 2824 break; 2825 #endif 2826 #ifdef INET 2827 case AF_INET: 2828 fin = (struct sockaddr_in *)&addrs[0]; 2829 lin = (struct sockaddr_in *)&addrs[1]; 2830 if (fin->sin_len != sizeof(struct sockaddr_in) || 2831 lin->sin_len != sizeof(struct sockaddr_in)) 2832 return (EINVAL); 2833 break; 2834 #endif 2835 default: 2836 return (EINVAL); 2837 } 2838 INP_INFO_RLOCK(&V_tcbinfo); 2839 switch (addrs[0].ss_family) { 2840 #ifdef INET6 2841 case AF_INET6: 2842 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2843 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2844 INPLOOKUP_WLOCKPCB, NULL); 2845 break; 2846 #endif 2847 #ifdef INET 2848 case AF_INET: 2849 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2850 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2851 break; 2852 #endif 2853 } 2854 if (inp != NULL) { 2855 if (inp->inp_flags & INP_TIMEWAIT) { 2856 /* 2857 * XXXRW: There currently exists a state where an 2858 * inpcb is present, but its timewait state has been 2859 * discarded. For now, don't allow dropping of this 2860 * type of inpcb. 2861 */ 2862 tw = intotw(inp); 2863 if (tw != NULL) 2864 tcp_twclose(tw, 0); 2865 else 2866 INP_WUNLOCK(inp); 2867 } else if (!(inp->inp_flags & INP_DROPPED) && 2868 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2869 tp = intotcpcb(inp); 2870 tp = tcp_drop(tp, ECONNABORTED); 2871 if (tp != NULL) 2872 INP_WUNLOCK(inp); 2873 } else 2874 INP_WUNLOCK(inp); 2875 } else 2876 error = ESRCH; 2877 INP_INFO_RUNLOCK(&V_tcbinfo); 2878 return (error); 2879 } 2880 2881 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2882 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2883 0, sysctl_drop, "", "Drop TCP connection"); 2884 2885 /* 2886 * Generate a standardized TCP log line for use throughout the 2887 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2888 * allow use in the interrupt context. 2889 * 2890 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2891 * NB: The function may return NULL if memory allocation failed. 2892 * 2893 * Due to header inclusion and ordering limitations the struct ip 2894 * and ip6_hdr pointers have to be passed as void pointers. 2895 */ 2896 char * 2897 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2898 const void *ip6hdr) 2899 { 2900 2901 /* Is logging enabled? */ 2902 if (tcp_log_in_vain == 0) 2903 return (NULL); 2904 2905 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2906 } 2907 2908 char * 2909 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2910 const void *ip6hdr) 2911 { 2912 2913 /* Is logging enabled? */ 2914 if (tcp_log_debug == 0) 2915 return (NULL); 2916 2917 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2918 } 2919 2920 static char * 2921 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2922 const void *ip6hdr) 2923 { 2924 char *s, *sp; 2925 size_t size; 2926 struct ip *ip; 2927 #ifdef INET6 2928 const struct ip6_hdr *ip6; 2929 2930 ip6 = (const struct ip6_hdr *)ip6hdr; 2931 #endif /* INET6 */ 2932 ip = (struct ip *)ip4hdr; 2933 2934 /* 2935 * The log line looks like this: 2936 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2937 */ 2938 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2939 sizeof(PRINT_TH_FLAGS) + 1 + 2940 #ifdef INET6 2941 2 * INET6_ADDRSTRLEN; 2942 #else 2943 2 * INET_ADDRSTRLEN; 2944 #endif /* INET6 */ 2945 2946 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2947 if (s == NULL) 2948 return (NULL); 2949 2950 strcat(s, "TCP: ["); 2951 sp = s + strlen(s); 2952 2953 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2954 inet_ntoa_r(inc->inc_faddr, sp); 2955 sp = s + strlen(s); 2956 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2957 sp = s + strlen(s); 2958 inet_ntoa_r(inc->inc_laddr, sp); 2959 sp = s + strlen(s); 2960 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2961 #ifdef INET6 2962 } else if (inc) { 2963 ip6_sprintf(sp, &inc->inc6_faddr); 2964 sp = s + strlen(s); 2965 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2966 sp = s + strlen(s); 2967 ip6_sprintf(sp, &inc->inc6_laddr); 2968 sp = s + strlen(s); 2969 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2970 } else if (ip6 && th) { 2971 ip6_sprintf(sp, &ip6->ip6_src); 2972 sp = s + strlen(s); 2973 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2974 sp = s + strlen(s); 2975 ip6_sprintf(sp, &ip6->ip6_dst); 2976 sp = s + strlen(s); 2977 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2978 #endif /* INET6 */ 2979 #ifdef INET 2980 } else if (ip && th) { 2981 inet_ntoa_r(ip->ip_src, sp); 2982 sp = s + strlen(s); 2983 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2984 sp = s + strlen(s); 2985 inet_ntoa_r(ip->ip_dst, sp); 2986 sp = s + strlen(s); 2987 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2988 #endif /* INET */ 2989 } else { 2990 free(s, M_TCPLOG); 2991 return (NULL); 2992 } 2993 sp = s + strlen(s); 2994 if (th) 2995 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2996 if (*(s + size - 1) != '\0') 2997 panic("%s: string too long", __func__); 2998 return (s); 2999 } 3000 3001 /* 3002 * A subroutine which makes it easy to track TCP state changes with DTrace. 3003 * This function shouldn't be called for t_state initializations that don't 3004 * correspond to actual TCP state transitions. 3005 */ 3006 void 3007 tcp_state_change(struct tcpcb *tp, int newstate) 3008 { 3009 #if defined(KDTRACE_HOOKS) 3010 int pstate = tp->t_state; 3011 #endif 3012 3013 TCPSTATES_DEC(tp->t_state); 3014 TCPSTATES_INC(newstate); 3015 tp->t_state = newstate; 3016 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3017 } 3018