1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_compat.h" 38 #include "opt_inet.h" 39 #include "opt_inet6.h" 40 #include "opt_ipsec.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/eventhandler.h> 47 #ifdef TCP_HHOOK 48 #include <sys/hhook.h> 49 #endif 50 #include <sys/kernel.h> 51 #ifdef TCP_HHOOK 52 #include <sys/khelp.h> 53 #endif 54 #include <sys/sysctl.h> 55 #include <sys/jail.h> 56 #include <sys/malloc.h> 57 #include <sys/refcount.h> 58 #include <sys/mbuf.h> 59 #ifdef INET6 60 #include <sys/domain.h> 61 #endif 62 #include <sys/priv.h> 63 #include <sys/proc.h> 64 #include <sys/sdt.h> 65 #include <sys/socket.h> 66 #include <sys/socketvar.h> 67 #include <sys/protosw.h> 68 #include <sys/random.h> 69 70 #include <vm/uma.h> 71 72 #include <net/route.h> 73 #include <net/if.h> 74 #include <net/if_var.h> 75 #include <net/vnet.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_fib.h> 79 #include <netinet/in_kdtrace.h> 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_systm.h> 82 #include <netinet/in_var.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_icmp.h> 85 #include <netinet/ip_var.h> 86 #ifdef INET6 87 #include <netinet/icmp6.h> 88 #include <netinet/ip6.h> 89 #include <netinet6/in6_fib.h> 90 #include <netinet6/in6_pcb.h> 91 #include <netinet6/ip6_var.h> 92 #include <netinet6/scope6_var.h> 93 #include <netinet6/nd6.h> 94 #endif 95 96 #ifdef TCP_RFC7413 97 #include <netinet/tcp_fastopen.h> 98 #endif 99 #include <netinet/tcp.h> 100 #include <netinet/tcp_fsm.h> 101 #include <netinet/tcp_seq.h> 102 #include <netinet/tcp_timer.h> 103 #include <netinet/tcp_var.h> 104 #include <netinet/tcp_syncache.h> 105 #include <netinet/cc/cc.h> 106 #ifdef INET6 107 #include <netinet6/tcp6_var.h> 108 #endif 109 #include <netinet/tcpip.h> 110 #ifdef TCPPCAP 111 #include <netinet/tcp_pcap.h> 112 #endif 113 #ifdef TCPDEBUG 114 #include <netinet/tcp_debug.h> 115 #endif 116 #ifdef INET6 117 #include <netinet6/ip6protosw.h> 118 #endif 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #include <netipsec/ipsec_support.h> 124 125 #include <machine/in_cksum.h> 126 #include <sys/md5.h> 127 128 #include <security/mac/mac_framework.h> 129 130 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 131 #ifdef INET6 132 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 133 #endif 134 135 struct rwlock tcp_function_lock; 136 137 static int 138 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 139 { 140 int error, new; 141 142 new = V_tcp_mssdflt; 143 error = sysctl_handle_int(oidp, &new, 0, req); 144 if (error == 0 && req->newptr) { 145 if (new < TCP_MINMSS) 146 error = EINVAL; 147 else 148 V_tcp_mssdflt = new; 149 } 150 return (error); 151 } 152 153 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 154 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 155 &sysctl_net_inet_tcp_mss_check, "I", 156 "Default TCP Maximum Segment Size"); 157 158 #ifdef INET6 159 static int 160 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 161 { 162 int error, new; 163 164 new = V_tcp_v6mssdflt; 165 error = sysctl_handle_int(oidp, &new, 0, req); 166 if (error == 0 && req->newptr) { 167 if (new < TCP_MINMSS) 168 error = EINVAL; 169 else 170 V_tcp_v6mssdflt = new; 171 } 172 return (error); 173 } 174 175 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 176 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 177 &sysctl_net_inet_tcp_mss_v6_check, "I", 178 "Default TCP Maximum Segment Size for IPv6"); 179 #endif /* INET6 */ 180 181 /* 182 * Minimum MSS we accept and use. This prevents DoS attacks where 183 * we are forced to a ridiculous low MSS like 20 and send hundreds 184 * of packets instead of one. The effect scales with the available 185 * bandwidth and quickly saturates the CPU and network interface 186 * with packet generation and sending. Set to zero to disable MINMSS 187 * checking. This setting prevents us from sending too small packets. 188 */ 189 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 191 &VNET_NAME(tcp_minmss), 0, 192 "Minimum TCP Maximum Segment Size"); 193 194 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 195 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 196 &VNET_NAME(tcp_do_rfc1323), 0, 197 "Enable rfc1323 (high performance TCP) extensions"); 198 199 static int tcp_log_debug = 0; 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 201 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 202 203 static int tcp_tcbhashsize; 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 205 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 206 207 static int do_tcpdrain = 1; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 209 "Enable tcp_drain routine for extra help when low on mbufs"); 210 211 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 212 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 213 214 static VNET_DEFINE(int, icmp_may_rst) = 1; 215 #define V_icmp_may_rst VNET(icmp_may_rst) 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(icmp_may_rst), 0, 218 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 219 220 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 221 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_isn_reseed_interval), 0, 224 "Seconds between reseeding of ISN secret"); 225 226 static int tcp_soreceive_stream; 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 228 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 229 230 VNET_DEFINE(uma_zone_t, sack_hole_zone); 231 #define V_sack_hole_zone VNET(sack_hole_zone) 232 233 #ifdef TCP_HHOOK 234 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 235 #endif 236 237 static struct inpcb *tcp_notify(struct inpcb *, int); 238 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 239 static void tcp_mtudisc(struct inpcb *, int); 240 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 241 void *ip4hdr, const void *ip6hdr); 242 243 244 static struct tcp_function_block tcp_def_funcblk = { 245 "default", 246 tcp_output, 247 tcp_do_segment, 248 tcp_default_ctloutput, 249 NULL, 250 NULL, 251 NULL, 252 NULL, 253 NULL, 254 NULL, 255 0, 256 0 257 }; 258 259 int t_functions_inited = 0; 260 struct tcp_funchead t_functions; 261 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 262 263 static void 264 init_tcp_functions(void) 265 { 266 if (t_functions_inited == 0) { 267 TAILQ_INIT(&t_functions); 268 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 269 t_functions_inited = 1; 270 } 271 } 272 273 static struct tcp_function_block * 274 find_tcp_functions_locked(struct tcp_function_set *fs) 275 { 276 struct tcp_function *f; 277 struct tcp_function_block *blk=NULL; 278 279 TAILQ_FOREACH(f, &t_functions, tf_next) { 280 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 281 blk = f->tf_fb; 282 break; 283 } 284 } 285 return(blk); 286 } 287 288 static struct tcp_function_block * 289 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 290 { 291 struct tcp_function_block *rblk=NULL; 292 struct tcp_function *f; 293 294 TAILQ_FOREACH(f, &t_functions, tf_next) { 295 if (f->tf_fb == blk) { 296 rblk = blk; 297 if (s) { 298 *s = f; 299 } 300 break; 301 } 302 } 303 return (rblk); 304 } 305 306 struct tcp_function_block * 307 find_and_ref_tcp_functions(struct tcp_function_set *fs) 308 { 309 struct tcp_function_block *blk; 310 311 rw_rlock(&tcp_function_lock); 312 blk = find_tcp_functions_locked(fs); 313 if (blk) 314 refcount_acquire(&blk->tfb_refcnt); 315 rw_runlock(&tcp_function_lock); 316 return(blk); 317 } 318 319 struct tcp_function_block * 320 find_and_ref_tcp_fb(struct tcp_function_block *blk) 321 { 322 struct tcp_function_block *rblk; 323 324 rw_rlock(&tcp_function_lock); 325 rblk = find_tcp_fb_locked(blk, NULL); 326 if (rblk) 327 refcount_acquire(&rblk->tfb_refcnt); 328 rw_runlock(&tcp_function_lock); 329 return(rblk); 330 } 331 332 333 static int 334 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 335 { 336 int error=ENOENT; 337 struct tcp_function_set fs; 338 struct tcp_function_block *blk; 339 340 memset(&fs, 0, sizeof(fs)); 341 rw_rlock(&tcp_function_lock); 342 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 343 if (blk) { 344 /* Found him */ 345 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 346 fs.pcbcnt = blk->tfb_refcnt; 347 } 348 rw_runlock(&tcp_function_lock); 349 error = sysctl_handle_string(oidp, fs.function_set_name, 350 sizeof(fs.function_set_name), req); 351 352 /* Check for error or no change */ 353 if (error != 0 || req->newptr == NULL) 354 return(error); 355 356 rw_wlock(&tcp_function_lock); 357 blk = find_tcp_functions_locked(&fs); 358 if ((blk == NULL) || 359 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 360 error = ENOENT; 361 goto done; 362 } 363 tcp_func_set_ptr = blk; 364 done: 365 rw_wunlock(&tcp_function_lock); 366 return (error); 367 } 368 369 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 370 CTLTYPE_STRING | CTLFLAG_RW, 371 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 372 "Set/get the default TCP functions"); 373 374 static int 375 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 376 { 377 int error, cnt, linesz; 378 struct tcp_function *f; 379 char *buffer, *cp; 380 size_t bufsz, outsz; 381 bool alias; 382 383 cnt = 0; 384 rw_rlock(&tcp_function_lock); 385 TAILQ_FOREACH(f, &t_functions, tf_next) { 386 cnt++; 387 } 388 rw_runlock(&tcp_function_lock); 389 390 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 391 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 392 393 error = 0; 394 cp = buffer; 395 396 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 397 "Alias", "PCB count"); 398 cp += linesz; 399 bufsz -= linesz; 400 outsz = linesz; 401 402 rw_rlock(&tcp_function_lock); 403 TAILQ_FOREACH(f, &t_functions, tf_next) { 404 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 405 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 406 f->tf_fb->tfb_tcp_block_name, 407 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 408 alias ? f->tf_name : "-", 409 f->tf_fb->tfb_refcnt); 410 if (linesz >= bufsz) { 411 error = EOVERFLOW; 412 break; 413 } 414 cp += linesz; 415 bufsz -= linesz; 416 outsz += linesz; 417 } 418 rw_runlock(&tcp_function_lock); 419 if (error == 0) 420 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 421 free(buffer, M_TEMP); 422 return (error); 423 } 424 425 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 426 CTLTYPE_STRING|CTLFLAG_RD, 427 NULL, 0, sysctl_net_inet_list_available, "A", 428 "list available TCP Function sets"); 429 430 /* 431 * Target size of TCP PCB hash tables. Must be a power of two. 432 * 433 * Note that this can be overridden by the kernel environment 434 * variable net.inet.tcp.tcbhashsize 435 */ 436 #ifndef TCBHASHSIZE 437 #define TCBHASHSIZE 0 438 #endif 439 440 /* 441 * XXX 442 * Callouts should be moved into struct tcp directly. They are currently 443 * separate because the tcpcb structure is exported to userland for sysctl 444 * parsing purposes, which do not know about callouts. 445 */ 446 struct tcpcb_mem { 447 struct tcpcb tcb; 448 struct tcp_timer tt; 449 struct cc_var ccv; 450 #ifdef TCP_HHOOK 451 struct osd osd; 452 #endif 453 }; 454 455 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 456 #define V_tcpcb_zone VNET(tcpcb_zone) 457 458 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 459 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 460 461 static struct mtx isn_mtx; 462 463 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 464 #define ISN_LOCK() mtx_lock(&isn_mtx) 465 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 466 467 /* 468 * TCP initialization. 469 */ 470 static void 471 tcp_zone_change(void *tag) 472 { 473 474 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 475 uma_zone_set_max(V_tcpcb_zone, maxsockets); 476 tcp_tw_zone_change(); 477 } 478 479 static int 480 tcp_inpcb_init(void *mem, int size, int flags) 481 { 482 struct inpcb *inp = mem; 483 484 INP_LOCK_INIT(inp, "inp", "tcpinp"); 485 return (0); 486 } 487 488 /* 489 * Take a value and get the next power of 2 that doesn't overflow. 490 * Used to size the tcp_inpcb hash buckets. 491 */ 492 static int 493 maketcp_hashsize(int size) 494 { 495 int hashsize; 496 497 /* 498 * auto tune. 499 * get the next power of 2 higher than maxsockets. 500 */ 501 hashsize = 1 << fls(size); 502 /* catch overflow, and just go one power of 2 smaller */ 503 if (hashsize < size) { 504 hashsize = 1 << (fls(size) - 1); 505 } 506 return (hashsize); 507 } 508 509 /* 510 * Register a TCP function block with the name provided in the names 511 * array. (Note that this function does NOT automatically register 512 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 513 * explicitly include blk->tfb_tcp_block_name in the list of names if 514 * you wish to register the stack with that name.) 515 * 516 * Either all name registrations will succeed or all will fail. If 517 * a name registration fails, the function will update the num_names 518 * argument to point to the array index of the name that encountered 519 * the failure. 520 * 521 * Returns 0 on success, or an error code on failure. 522 */ 523 int 524 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 525 const char *names[], int *num_names) 526 { 527 struct tcp_function *n; 528 struct tcp_function_set fs; 529 int error, i; 530 531 KASSERT(names != NULL && *num_names > 0, 532 ("%s: Called with 0-length name list", __func__)); 533 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 534 535 if (t_functions_inited == 0) { 536 init_tcp_functions(); 537 } 538 if ((blk->tfb_tcp_output == NULL) || 539 (blk->tfb_tcp_do_segment == NULL) || 540 (blk->tfb_tcp_ctloutput == NULL) || 541 (strlen(blk->tfb_tcp_block_name) == 0)) { 542 /* 543 * These functions are required and you 544 * need a name. 545 */ 546 *num_names = 0; 547 return (EINVAL); 548 } 549 if (blk->tfb_tcp_timer_stop_all || 550 blk->tfb_tcp_timer_activate || 551 blk->tfb_tcp_timer_active || 552 blk->tfb_tcp_timer_stop) { 553 /* 554 * If you define one timer function you 555 * must have them all. 556 */ 557 if ((blk->tfb_tcp_timer_stop_all == NULL) || 558 (blk->tfb_tcp_timer_activate == NULL) || 559 (blk->tfb_tcp_timer_active == NULL) || 560 (blk->tfb_tcp_timer_stop == NULL)) { 561 *num_names = 0; 562 return (EINVAL); 563 } 564 } 565 566 refcount_init(&blk->tfb_refcnt, 0); 567 blk->tfb_flags = 0; 568 for (i = 0; i < *num_names; i++) { 569 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 570 if (n == NULL) { 571 error = ENOMEM; 572 goto cleanup; 573 } 574 n->tf_fb = blk; 575 576 (void)strncpy(fs.function_set_name, names[i], 577 TCP_FUNCTION_NAME_LEN_MAX); 578 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 579 rw_wlock(&tcp_function_lock); 580 if (find_tcp_functions_locked(&fs) != NULL) { 581 /* Duplicate name space not allowed */ 582 rw_wunlock(&tcp_function_lock); 583 free(n, M_TCPFUNCTIONS); 584 error = EALREADY; 585 goto cleanup; 586 } 587 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX); 588 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 589 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 590 rw_wunlock(&tcp_function_lock); 591 } 592 return(0); 593 594 cleanup: 595 /* 596 * Deregister the names we just added. Because registration failed 597 * for names[i], we don't need to deregister that name. 598 */ 599 *num_names = i; 600 rw_wlock(&tcp_function_lock); 601 while (--i >= 0) { 602 TAILQ_FOREACH(n, &t_functions, tf_next) { 603 if (!strncmp(n->tf_name, names[i], 604 TCP_FUNCTION_NAME_LEN_MAX)) { 605 TAILQ_REMOVE(&t_functions, n, tf_next); 606 n->tf_fb = NULL; 607 free(n, M_TCPFUNCTIONS); 608 break; 609 } 610 } 611 } 612 rw_wunlock(&tcp_function_lock); 613 return (error); 614 } 615 616 /* 617 * Register a TCP function block using the name provided in the name 618 * argument. 619 * 620 * Returns 0 on success, or an error code on failure. 621 */ 622 int 623 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 624 int wait) 625 { 626 const char *name_list[1]; 627 int num_names, rv; 628 629 num_names = 1; 630 if (name != NULL) 631 name_list[0] = name; 632 else 633 name_list[0] = blk->tfb_tcp_block_name; 634 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 635 return (rv); 636 } 637 638 /* 639 * Register a TCP function block using the name defined in 640 * blk->tfb_tcp_block_name. 641 * 642 * Returns 0 on success, or an error code on failure. 643 */ 644 int 645 register_tcp_functions(struct tcp_function_block *blk, int wait) 646 { 647 648 return (register_tcp_functions_as_name(blk, NULL, wait)); 649 } 650 651 int 652 deregister_tcp_functions(struct tcp_function_block *blk) 653 { 654 struct tcp_function *f; 655 int error=ENOENT; 656 657 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 658 /* You can't un-register the default */ 659 return (EPERM); 660 } 661 rw_wlock(&tcp_function_lock); 662 if (blk == tcp_func_set_ptr) { 663 /* You can't free the current default */ 664 rw_wunlock(&tcp_function_lock); 665 return (EBUSY); 666 } 667 if (blk->tfb_refcnt) { 668 /* Still tcb attached, mark it. */ 669 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 670 rw_wunlock(&tcp_function_lock); 671 return (EBUSY); 672 } 673 while (find_tcp_fb_locked(blk, &f) != NULL) { 674 /* Found */ 675 TAILQ_REMOVE(&t_functions, f, tf_next); 676 f->tf_fb = NULL; 677 free(f, M_TCPFUNCTIONS); 678 error = 0; 679 } 680 rw_wunlock(&tcp_function_lock); 681 return (error); 682 } 683 684 void 685 tcp_init(void) 686 { 687 const char *tcbhash_tuneable; 688 int hashsize; 689 690 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 691 692 #ifdef TCP_HHOOK 693 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 694 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 695 printf("%s: WARNING: unable to register helper hook\n", __func__); 696 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 697 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 698 printf("%s: WARNING: unable to register helper hook\n", __func__); 699 #endif 700 hashsize = TCBHASHSIZE; 701 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 702 if (hashsize == 0) { 703 /* 704 * Auto tune the hash size based on maxsockets. 705 * A perfect hash would have a 1:1 mapping 706 * (hashsize = maxsockets) however it's been 707 * suggested that O(2) average is better. 708 */ 709 hashsize = maketcp_hashsize(maxsockets / 4); 710 /* 711 * Our historical default is 512, 712 * do not autotune lower than this. 713 */ 714 if (hashsize < 512) 715 hashsize = 512; 716 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 717 printf("%s: %s auto tuned to %d\n", __func__, 718 tcbhash_tuneable, hashsize); 719 } 720 /* 721 * We require a hashsize to be a power of two. 722 * Previously if it was not a power of two we would just reset it 723 * back to 512, which could be a nasty surprise if you did not notice 724 * the error message. 725 * Instead what we do is clip it to the closest power of two lower 726 * than the specified hash value. 727 */ 728 if (!powerof2(hashsize)) { 729 int oldhashsize = hashsize; 730 731 hashsize = maketcp_hashsize(hashsize); 732 /* prevent absurdly low value */ 733 if (hashsize < 16) 734 hashsize = 16; 735 printf("%s: WARNING: TCB hash size not a power of 2, " 736 "clipped from %d to %d.\n", __func__, oldhashsize, 737 hashsize); 738 } 739 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 740 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 741 742 /* 743 * These have to be type stable for the benefit of the timers. 744 */ 745 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 746 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 747 uma_zone_set_max(V_tcpcb_zone, maxsockets); 748 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 749 750 tcp_tw_init(); 751 syncache_init(); 752 tcp_hc_init(); 753 754 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 755 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 756 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 757 758 #ifdef TCP_RFC7413 759 tcp_fastopen_init(); 760 #endif 761 762 /* Skip initialization of globals for non-default instances. */ 763 if (!IS_DEFAULT_VNET(curvnet)) 764 return; 765 766 tcp_reass_global_init(); 767 768 /* XXX virtualize those bellow? */ 769 tcp_delacktime = TCPTV_DELACK; 770 tcp_keepinit = TCPTV_KEEP_INIT; 771 tcp_keepidle = TCPTV_KEEP_IDLE; 772 tcp_keepintvl = TCPTV_KEEPINTVL; 773 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 774 tcp_msl = TCPTV_MSL; 775 tcp_rexmit_min = TCPTV_MIN; 776 if (tcp_rexmit_min < 1) 777 tcp_rexmit_min = 1; 778 tcp_persmin = TCPTV_PERSMIN; 779 tcp_persmax = TCPTV_PERSMAX; 780 tcp_rexmit_slop = TCPTV_CPU_VAR; 781 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 782 tcp_tcbhashsize = hashsize; 783 /* Setup the tcp function block list */ 784 init_tcp_functions(); 785 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 786 787 if (tcp_soreceive_stream) { 788 #ifdef INET 789 tcp_usrreqs.pru_soreceive = soreceive_stream; 790 #endif 791 #ifdef INET6 792 tcp6_usrreqs.pru_soreceive = soreceive_stream; 793 #endif /* INET6 */ 794 } 795 796 #ifdef INET6 797 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 798 #else /* INET6 */ 799 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 800 #endif /* INET6 */ 801 if (max_protohdr < TCP_MINPROTOHDR) 802 max_protohdr = TCP_MINPROTOHDR; 803 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 804 panic("tcp_init"); 805 #undef TCP_MINPROTOHDR 806 807 ISN_LOCK_INIT(); 808 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 809 SHUTDOWN_PRI_DEFAULT); 810 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 811 EVENTHANDLER_PRI_ANY); 812 #ifdef TCPPCAP 813 tcp_pcap_init(); 814 #endif 815 } 816 817 #ifdef VIMAGE 818 static void 819 tcp_destroy(void *unused __unused) 820 { 821 int n; 822 #ifdef TCP_HHOOK 823 int error; 824 #endif 825 826 /* 827 * All our processes are gone, all our sockets should be cleaned 828 * up, which means, we should be past the tcp_discardcb() calls. 829 * Sleep to let all tcpcb timers really disappear and cleanup. 830 */ 831 for (;;) { 832 INP_LIST_RLOCK(&V_tcbinfo); 833 n = V_tcbinfo.ipi_count; 834 INP_LIST_RUNLOCK(&V_tcbinfo); 835 if (n == 0) 836 break; 837 pause("tcpdes", hz / 10); 838 } 839 tcp_hc_destroy(); 840 syncache_destroy(); 841 tcp_tw_destroy(); 842 in_pcbinfo_destroy(&V_tcbinfo); 843 /* tcp_discardcb() clears the sack_holes up. */ 844 uma_zdestroy(V_sack_hole_zone); 845 uma_zdestroy(V_tcpcb_zone); 846 847 #ifdef TCP_RFC7413 848 /* 849 * Cannot free the zone until all tcpcbs are released as we attach 850 * the allocations to them. 851 */ 852 tcp_fastopen_destroy(); 853 #endif 854 855 #ifdef TCP_HHOOK 856 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 857 if (error != 0) { 858 printf("%s: WARNING: unable to deregister helper hook " 859 "type=%d, id=%d: error %d returned\n", __func__, 860 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 861 } 862 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 863 if (error != 0) { 864 printf("%s: WARNING: unable to deregister helper hook " 865 "type=%d, id=%d: error %d returned\n", __func__, 866 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 867 } 868 #endif 869 } 870 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 871 #endif 872 873 void 874 tcp_fini(void *xtp) 875 { 876 877 } 878 879 /* 880 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 881 * tcp_template used to store this data in mbufs, but we now recopy it out 882 * of the tcpcb each time to conserve mbufs. 883 */ 884 void 885 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 886 { 887 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 888 889 INP_WLOCK_ASSERT(inp); 890 891 #ifdef INET6 892 if ((inp->inp_vflag & INP_IPV6) != 0) { 893 struct ip6_hdr *ip6; 894 895 ip6 = (struct ip6_hdr *)ip_ptr; 896 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 897 (inp->inp_flow & IPV6_FLOWINFO_MASK); 898 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 899 (IPV6_VERSION & IPV6_VERSION_MASK); 900 ip6->ip6_nxt = IPPROTO_TCP; 901 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 902 ip6->ip6_src = inp->in6p_laddr; 903 ip6->ip6_dst = inp->in6p_faddr; 904 } 905 #endif /* INET6 */ 906 #if defined(INET6) && defined(INET) 907 else 908 #endif 909 #ifdef INET 910 { 911 struct ip *ip; 912 913 ip = (struct ip *)ip_ptr; 914 ip->ip_v = IPVERSION; 915 ip->ip_hl = 5; 916 ip->ip_tos = inp->inp_ip_tos; 917 ip->ip_len = 0; 918 ip->ip_id = 0; 919 ip->ip_off = 0; 920 ip->ip_ttl = inp->inp_ip_ttl; 921 ip->ip_sum = 0; 922 ip->ip_p = IPPROTO_TCP; 923 ip->ip_src = inp->inp_laddr; 924 ip->ip_dst = inp->inp_faddr; 925 } 926 #endif /* INET */ 927 th->th_sport = inp->inp_lport; 928 th->th_dport = inp->inp_fport; 929 th->th_seq = 0; 930 th->th_ack = 0; 931 th->th_x2 = 0; 932 th->th_off = 5; 933 th->th_flags = 0; 934 th->th_win = 0; 935 th->th_urp = 0; 936 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 937 } 938 939 /* 940 * Create template to be used to send tcp packets on a connection. 941 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 942 * use for this function is in keepalives, which use tcp_respond. 943 */ 944 struct tcptemp * 945 tcpip_maketemplate(struct inpcb *inp) 946 { 947 struct tcptemp *t; 948 949 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 950 if (t == NULL) 951 return (NULL); 952 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 953 return (t); 954 } 955 956 /* 957 * Send a single message to the TCP at address specified by 958 * the given TCP/IP header. If m == NULL, then we make a copy 959 * of the tcpiphdr at th and send directly to the addressed host. 960 * This is used to force keep alive messages out using the TCP 961 * template for a connection. If flags are given then we send 962 * a message back to the TCP which originated the segment th, 963 * and discard the mbuf containing it and any other attached mbufs. 964 * 965 * In any case the ack and sequence number of the transmitted 966 * segment are as specified by the parameters. 967 * 968 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 969 */ 970 void 971 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 972 tcp_seq ack, tcp_seq seq, int flags) 973 { 974 struct tcpopt to; 975 struct inpcb *inp; 976 struct ip *ip; 977 struct mbuf *optm; 978 struct tcphdr *nth; 979 u_char *optp; 980 #ifdef INET6 981 struct ip6_hdr *ip6; 982 int isipv6; 983 #endif /* INET6 */ 984 int optlen, tlen, win; 985 bool incl_opts; 986 987 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 988 989 #ifdef INET6 990 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 991 ip6 = ipgen; 992 #endif /* INET6 */ 993 ip = ipgen; 994 995 if (tp != NULL) { 996 inp = tp->t_inpcb; 997 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 998 INP_WLOCK_ASSERT(inp); 999 } else 1000 inp = NULL; 1001 1002 incl_opts = false; 1003 win = 0; 1004 if (tp != NULL) { 1005 if (!(flags & TH_RST)) { 1006 win = sbspace(&inp->inp_socket->so_rcv); 1007 if (win > TCP_MAXWIN << tp->rcv_scale) 1008 win = TCP_MAXWIN << tp->rcv_scale; 1009 } 1010 if ((tp->t_flags & TF_NOOPT) == 0) 1011 incl_opts = true; 1012 } 1013 if (m == NULL) { 1014 m = m_gethdr(M_NOWAIT, MT_DATA); 1015 if (m == NULL) 1016 return; 1017 m->m_data += max_linkhdr; 1018 #ifdef INET6 1019 if (isipv6) { 1020 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1021 sizeof(struct ip6_hdr)); 1022 ip6 = mtod(m, struct ip6_hdr *); 1023 nth = (struct tcphdr *)(ip6 + 1); 1024 } else 1025 #endif /* INET6 */ 1026 { 1027 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1028 ip = mtod(m, struct ip *); 1029 nth = (struct tcphdr *)(ip + 1); 1030 } 1031 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1032 flags = TH_ACK; 1033 } else if (!M_WRITABLE(m)) { 1034 struct mbuf *n; 1035 1036 /* Can't reuse 'm', allocate a new mbuf. */ 1037 n = m_gethdr(M_NOWAIT, MT_DATA); 1038 if (n == NULL) { 1039 m_freem(m); 1040 return; 1041 } 1042 1043 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1044 m_freem(m); 1045 m_freem(n); 1046 return; 1047 } 1048 1049 n->m_data += max_linkhdr; 1050 /* m_len is set later */ 1051 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1052 #ifdef INET6 1053 if (isipv6) { 1054 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1055 sizeof(struct ip6_hdr)); 1056 ip6 = mtod(n, struct ip6_hdr *); 1057 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1058 nth = (struct tcphdr *)(ip6 + 1); 1059 } else 1060 #endif /* INET6 */ 1061 { 1062 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1063 ip = mtod(n, struct ip *); 1064 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1065 nth = (struct tcphdr *)(ip + 1); 1066 } 1067 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1068 xchg(nth->th_dport, nth->th_sport, uint16_t); 1069 th = nth; 1070 m_freem(m); 1071 m = n; 1072 } else { 1073 /* 1074 * reuse the mbuf. 1075 * XXX MRT We inherit the FIB, which is lucky. 1076 */ 1077 m_freem(m->m_next); 1078 m->m_next = NULL; 1079 m->m_data = (caddr_t)ipgen; 1080 /* m_len is set later */ 1081 #ifdef INET6 1082 if (isipv6) { 1083 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1084 nth = (struct tcphdr *)(ip6 + 1); 1085 } else 1086 #endif /* INET6 */ 1087 { 1088 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1089 nth = (struct tcphdr *)(ip + 1); 1090 } 1091 if (th != nth) { 1092 /* 1093 * this is usually a case when an extension header 1094 * exists between the IPv6 header and the 1095 * TCP header. 1096 */ 1097 nth->th_sport = th->th_sport; 1098 nth->th_dport = th->th_dport; 1099 } 1100 xchg(nth->th_dport, nth->th_sport, uint16_t); 1101 #undef xchg 1102 } 1103 tlen = 0; 1104 #ifdef INET6 1105 if (isipv6) 1106 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1107 #endif 1108 #if defined(INET) && defined(INET6) 1109 else 1110 #endif 1111 #ifdef INET 1112 tlen = sizeof (struct tcpiphdr); 1113 #endif 1114 #ifdef INVARIANTS 1115 m->m_len = 0; 1116 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1117 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1118 m, tlen, (long)M_TRAILINGSPACE(m))); 1119 #endif 1120 m->m_len = tlen; 1121 to.to_flags = 0; 1122 if (incl_opts) { 1123 /* Make sure we have room. */ 1124 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1125 m->m_next = m_get(M_NOWAIT, MT_DATA); 1126 if (m->m_next) { 1127 optp = mtod(m->m_next, u_char *); 1128 optm = m->m_next; 1129 } else 1130 incl_opts = false; 1131 } else { 1132 optp = (u_char *) (nth + 1); 1133 optm = m; 1134 } 1135 } 1136 if (incl_opts) { 1137 /* Timestamps. */ 1138 if (tp->t_flags & TF_RCVD_TSTMP) { 1139 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1140 to.to_tsecr = tp->ts_recent; 1141 to.to_flags |= TOF_TS; 1142 } 1143 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1144 /* TCP-MD5 (RFC2385). */ 1145 if (tp->t_flags & TF_SIGNATURE) 1146 to.to_flags |= TOF_SIGNATURE; 1147 #endif 1148 /* Add the options. */ 1149 tlen += optlen = tcp_addoptions(&to, optp); 1150 1151 /* Update m_len in the correct mbuf. */ 1152 optm->m_len += optlen; 1153 } else 1154 optlen = 0; 1155 #ifdef INET6 1156 if (isipv6) { 1157 ip6->ip6_flow = 0; 1158 ip6->ip6_vfc = IPV6_VERSION; 1159 ip6->ip6_nxt = IPPROTO_TCP; 1160 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1161 } 1162 #endif 1163 #if defined(INET) && defined(INET6) 1164 else 1165 #endif 1166 #ifdef INET 1167 { 1168 ip->ip_len = htons(tlen); 1169 ip->ip_ttl = V_ip_defttl; 1170 if (V_path_mtu_discovery) 1171 ip->ip_off |= htons(IP_DF); 1172 } 1173 #endif 1174 m->m_pkthdr.len = tlen; 1175 m->m_pkthdr.rcvif = NULL; 1176 #ifdef MAC 1177 if (inp != NULL) { 1178 /* 1179 * Packet is associated with a socket, so allow the 1180 * label of the response to reflect the socket label. 1181 */ 1182 INP_WLOCK_ASSERT(inp); 1183 mac_inpcb_create_mbuf(inp, m); 1184 } else { 1185 /* 1186 * Packet is not associated with a socket, so possibly 1187 * update the label in place. 1188 */ 1189 mac_netinet_tcp_reply(m); 1190 } 1191 #endif 1192 nth->th_seq = htonl(seq); 1193 nth->th_ack = htonl(ack); 1194 nth->th_x2 = 0; 1195 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1196 nth->th_flags = flags; 1197 if (tp != NULL) 1198 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1199 else 1200 nth->th_win = htons((u_short)win); 1201 nth->th_urp = 0; 1202 1203 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1204 if (to.to_flags & TOF_SIGNATURE) { 1205 if (!TCPMD5_ENABLED() || 1206 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1207 m_freem(m); 1208 return; 1209 } 1210 } 1211 #endif 1212 1213 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1214 #ifdef INET6 1215 if (isipv6) { 1216 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1217 nth->th_sum = in6_cksum_pseudo(ip6, 1218 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1219 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1220 NULL, NULL); 1221 } 1222 #endif /* INET6 */ 1223 #if defined(INET6) && defined(INET) 1224 else 1225 #endif 1226 #ifdef INET 1227 { 1228 m->m_pkthdr.csum_flags = CSUM_TCP; 1229 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1230 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1231 } 1232 #endif /* INET */ 1233 #ifdef TCPDEBUG 1234 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1235 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1236 #endif 1237 TCP_PROBE3(debug__output, tp, th, m); 1238 if (flags & TH_RST) 1239 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1240 1241 TCP_PROBE5(send, NULL, tp, m, tp, nth); 1242 #ifdef INET6 1243 if (isipv6) 1244 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1245 #endif /* INET6 */ 1246 #if defined(INET) && defined(INET6) 1247 else 1248 #endif 1249 #ifdef INET 1250 (void) ip_output(m, NULL, NULL, 0, NULL, inp); 1251 #endif 1252 } 1253 1254 /* 1255 * Create a new TCP control block, making an 1256 * empty reassembly queue and hooking it to the argument 1257 * protocol control block. The `inp' parameter must have 1258 * come from the zone allocator set up in tcp_init(). 1259 */ 1260 struct tcpcb * 1261 tcp_newtcpcb(struct inpcb *inp) 1262 { 1263 struct tcpcb_mem *tm; 1264 struct tcpcb *tp; 1265 #ifdef INET6 1266 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1267 #endif /* INET6 */ 1268 1269 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1270 if (tm == NULL) 1271 return (NULL); 1272 tp = &tm->tcb; 1273 1274 /* Initialise cc_var struct for this tcpcb. */ 1275 tp->ccv = &tm->ccv; 1276 tp->ccv->type = IPPROTO_TCP; 1277 tp->ccv->ccvc.tcp = tp; 1278 rw_rlock(&tcp_function_lock); 1279 tp->t_fb = tcp_func_set_ptr; 1280 refcount_acquire(&tp->t_fb->tfb_refcnt); 1281 rw_runlock(&tcp_function_lock); 1282 /* 1283 * Use the current system default CC algorithm. 1284 */ 1285 CC_LIST_RLOCK(); 1286 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1287 CC_ALGO(tp) = CC_DEFAULT(); 1288 CC_LIST_RUNLOCK(); 1289 1290 if (CC_ALGO(tp)->cb_init != NULL) 1291 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1292 if (tp->t_fb->tfb_tcp_fb_fini) 1293 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1294 refcount_release(&tp->t_fb->tfb_refcnt); 1295 uma_zfree(V_tcpcb_zone, tm); 1296 return (NULL); 1297 } 1298 1299 #ifdef TCP_HHOOK 1300 tp->osd = &tm->osd; 1301 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1302 if (tp->t_fb->tfb_tcp_fb_fini) 1303 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1304 refcount_release(&tp->t_fb->tfb_refcnt); 1305 uma_zfree(V_tcpcb_zone, tm); 1306 return (NULL); 1307 } 1308 #endif 1309 1310 #ifdef VIMAGE 1311 tp->t_vnet = inp->inp_vnet; 1312 #endif 1313 tp->t_timers = &tm->tt; 1314 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1315 tp->t_maxseg = 1316 #ifdef INET6 1317 isipv6 ? V_tcp_v6mssdflt : 1318 #endif /* INET6 */ 1319 V_tcp_mssdflt; 1320 1321 /* Set up our timeouts. */ 1322 callout_init(&tp->t_timers->tt_rexmt, 1); 1323 callout_init(&tp->t_timers->tt_persist, 1); 1324 callout_init(&tp->t_timers->tt_keep, 1); 1325 callout_init(&tp->t_timers->tt_2msl, 1); 1326 callout_init(&tp->t_timers->tt_delack, 1); 1327 1328 if (V_tcp_do_rfc1323) 1329 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1330 if (V_tcp_do_sack) 1331 tp->t_flags |= TF_SACK_PERMIT; 1332 TAILQ_INIT(&tp->snd_holes); 1333 /* 1334 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1335 * is called. 1336 */ 1337 in_pcbref(inp); /* Reference for tcpcb */ 1338 tp->t_inpcb = inp; 1339 1340 /* 1341 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1342 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1343 * reasonable initial retransmit time. 1344 */ 1345 tp->t_srtt = TCPTV_SRTTBASE; 1346 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1347 tp->t_rttmin = tcp_rexmit_min; 1348 tp->t_rxtcur = TCPTV_RTOBASE; 1349 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1350 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1351 tp->t_rcvtime = ticks; 1352 /* 1353 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1354 * because the socket may be bound to an IPv6 wildcard address, 1355 * which may match an IPv4-mapped IPv6 address. 1356 */ 1357 inp->inp_ip_ttl = V_ip_defttl; 1358 inp->inp_ppcb = tp; 1359 #ifdef TCPPCAP 1360 /* 1361 * Init the TCP PCAP queues. 1362 */ 1363 tcp_pcap_tcpcb_init(tp); 1364 #endif 1365 if (tp->t_fb->tfb_tcp_fb_init) { 1366 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1367 } 1368 return (tp); /* XXX */ 1369 } 1370 1371 /* 1372 * Switch the congestion control algorithm back to NewReno for any active 1373 * control blocks using an algorithm which is about to go away. 1374 * This ensures the CC framework can allow the unload to proceed without leaving 1375 * any dangling pointers which would trigger a panic. 1376 * Returning non-zero would inform the CC framework that something went wrong 1377 * and it would be unsafe to allow the unload to proceed. However, there is no 1378 * way for this to occur with this implementation so we always return zero. 1379 */ 1380 int 1381 tcp_ccalgounload(struct cc_algo *unload_algo) 1382 { 1383 struct cc_algo *tmpalgo; 1384 struct inpcb *inp; 1385 struct tcpcb *tp; 1386 VNET_ITERATOR_DECL(vnet_iter); 1387 1388 /* 1389 * Check all active control blocks across all network stacks and change 1390 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1391 * requires cleanup code to be run, call it. 1392 */ 1393 VNET_LIST_RLOCK(); 1394 VNET_FOREACH(vnet_iter) { 1395 CURVNET_SET(vnet_iter); 1396 INP_INFO_WLOCK(&V_tcbinfo); 1397 /* 1398 * New connections already part way through being initialised 1399 * with the CC algo we're removing will not race with this code 1400 * because the INP_INFO_WLOCK is held during initialisation. We 1401 * therefore don't enter the loop below until the connection 1402 * list has stabilised. 1403 */ 1404 LIST_FOREACH(inp, &V_tcb, inp_list) { 1405 INP_WLOCK(inp); 1406 /* Important to skip tcptw structs. */ 1407 if (!(inp->inp_flags & INP_TIMEWAIT) && 1408 (tp = intotcpcb(inp)) != NULL) { 1409 /* 1410 * By holding INP_WLOCK here, we are assured 1411 * that the connection is not currently 1412 * executing inside the CC module's functions 1413 * i.e. it is safe to make the switch back to 1414 * NewReno. 1415 */ 1416 if (CC_ALGO(tp) == unload_algo) { 1417 tmpalgo = CC_ALGO(tp); 1418 /* NewReno does not require any init. */ 1419 CC_ALGO(tp) = &newreno_cc_algo; 1420 if (tmpalgo->cb_destroy != NULL) 1421 tmpalgo->cb_destroy(tp->ccv); 1422 } 1423 } 1424 INP_WUNLOCK(inp); 1425 } 1426 INP_INFO_WUNLOCK(&V_tcbinfo); 1427 CURVNET_RESTORE(); 1428 } 1429 VNET_LIST_RUNLOCK(); 1430 1431 return (0); 1432 } 1433 1434 /* 1435 * Drop a TCP connection, reporting 1436 * the specified error. If connection is synchronized, 1437 * then send a RST to peer. 1438 */ 1439 struct tcpcb * 1440 tcp_drop(struct tcpcb *tp, int errno) 1441 { 1442 struct socket *so = tp->t_inpcb->inp_socket; 1443 1444 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1445 INP_WLOCK_ASSERT(tp->t_inpcb); 1446 1447 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1448 tcp_state_change(tp, TCPS_CLOSED); 1449 (void) tp->t_fb->tfb_tcp_output(tp); 1450 TCPSTAT_INC(tcps_drops); 1451 } else 1452 TCPSTAT_INC(tcps_conndrops); 1453 if (errno == ETIMEDOUT && tp->t_softerror) 1454 errno = tp->t_softerror; 1455 so->so_error = errno; 1456 return (tcp_close(tp)); 1457 } 1458 1459 void 1460 tcp_discardcb(struct tcpcb *tp) 1461 { 1462 struct inpcb *inp = tp->t_inpcb; 1463 struct socket *so = inp->inp_socket; 1464 #ifdef INET6 1465 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1466 #endif /* INET6 */ 1467 int released; 1468 1469 INP_WLOCK_ASSERT(inp); 1470 1471 /* 1472 * Make sure that all of our timers are stopped before we delete the 1473 * PCB. 1474 * 1475 * If stopping a timer fails, we schedule a discard function in same 1476 * callout, and the last discard function called will take care of 1477 * deleting the tcpcb. 1478 */ 1479 tp->t_timers->tt_draincnt = 0; 1480 tcp_timer_stop(tp, TT_REXMT); 1481 tcp_timer_stop(tp, TT_PERSIST); 1482 tcp_timer_stop(tp, TT_KEEP); 1483 tcp_timer_stop(tp, TT_2MSL); 1484 tcp_timer_stop(tp, TT_DELACK); 1485 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1486 /* 1487 * Call the stop-all function of the methods, 1488 * this function should call the tcp_timer_stop() 1489 * method with each of the function specific timeouts. 1490 * That stop will be called via the tfb_tcp_timer_stop() 1491 * which should use the async drain function of the 1492 * callout system (see tcp_var.h). 1493 */ 1494 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1495 } 1496 1497 /* 1498 * If we got enough samples through the srtt filter, 1499 * save the rtt and rttvar in the routing entry. 1500 * 'Enough' is arbitrarily defined as 4 rtt samples. 1501 * 4 samples is enough for the srtt filter to converge 1502 * to within enough % of the correct value; fewer samples 1503 * and we could save a bogus rtt. The danger is not high 1504 * as tcp quickly recovers from everything. 1505 * XXX: Works very well but needs some more statistics! 1506 */ 1507 if (tp->t_rttupdated >= 4) { 1508 struct hc_metrics_lite metrics; 1509 uint32_t ssthresh; 1510 1511 bzero(&metrics, sizeof(metrics)); 1512 /* 1513 * Update the ssthresh always when the conditions below 1514 * are satisfied. This gives us better new start value 1515 * for the congestion avoidance for new connections. 1516 * ssthresh is only set if packet loss occurred on a session. 1517 * 1518 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1519 * being torn down. Ideally this code would not use 'so'. 1520 */ 1521 ssthresh = tp->snd_ssthresh; 1522 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1523 /* 1524 * convert the limit from user data bytes to 1525 * packets then to packet data bytes. 1526 */ 1527 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1528 if (ssthresh < 2) 1529 ssthresh = 2; 1530 ssthresh *= (tp->t_maxseg + 1531 #ifdef INET6 1532 (isipv6 ? sizeof (struct ip6_hdr) + 1533 sizeof (struct tcphdr) : 1534 #endif 1535 sizeof (struct tcpiphdr) 1536 #ifdef INET6 1537 ) 1538 #endif 1539 ); 1540 } else 1541 ssthresh = 0; 1542 metrics.rmx_ssthresh = ssthresh; 1543 1544 metrics.rmx_rtt = tp->t_srtt; 1545 metrics.rmx_rttvar = tp->t_rttvar; 1546 metrics.rmx_cwnd = tp->snd_cwnd; 1547 metrics.rmx_sendpipe = 0; 1548 metrics.rmx_recvpipe = 0; 1549 1550 tcp_hc_update(&inp->inp_inc, &metrics); 1551 } 1552 1553 /* free the reassembly queue, if any */ 1554 tcp_reass_flush(tp); 1555 1556 #ifdef TCP_OFFLOAD 1557 /* Disconnect offload device, if any. */ 1558 if (tp->t_flags & TF_TOE) 1559 tcp_offload_detach(tp); 1560 #endif 1561 1562 tcp_free_sackholes(tp); 1563 1564 #ifdef TCPPCAP 1565 /* Free the TCP PCAP queues. */ 1566 tcp_pcap_drain(&(tp->t_inpkts)); 1567 tcp_pcap_drain(&(tp->t_outpkts)); 1568 #endif 1569 1570 /* Allow the CC algorithm to clean up after itself. */ 1571 if (CC_ALGO(tp)->cb_destroy != NULL) 1572 CC_ALGO(tp)->cb_destroy(tp->ccv); 1573 1574 #ifdef TCP_HHOOK 1575 khelp_destroy_osd(tp->osd); 1576 #endif 1577 1578 CC_ALGO(tp) = NULL; 1579 inp->inp_ppcb = NULL; 1580 if (tp->t_timers->tt_draincnt == 0) { 1581 /* We own the last reference on tcpcb, let's free it. */ 1582 TCPSTATES_DEC(tp->t_state); 1583 if (tp->t_fb->tfb_tcp_fb_fini) 1584 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1585 refcount_release(&tp->t_fb->tfb_refcnt); 1586 tp->t_inpcb = NULL; 1587 uma_zfree(V_tcpcb_zone, tp); 1588 released = in_pcbrele_wlocked(inp); 1589 KASSERT(!released, ("%s: inp %p should not have been released " 1590 "here", __func__, inp)); 1591 } 1592 } 1593 1594 void 1595 tcp_timer_discard(void *ptp) 1596 { 1597 struct inpcb *inp; 1598 struct tcpcb *tp; 1599 1600 tp = (struct tcpcb *)ptp; 1601 CURVNET_SET(tp->t_vnet); 1602 INP_INFO_RLOCK(&V_tcbinfo); 1603 inp = tp->t_inpcb; 1604 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1605 __func__, tp)); 1606 INP_WLOCK(inp); 1607 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1608 ("%s: tcpcb has to be stopped here", __func__)); 1609 tp->t_timers->tt_draincnt--; 1610 if (tp->t_timers->tt_draincnt == 0) { 1611 /* We own the last reference on this tcpcb, let's free it. */ 1612 TCPSTATES_DEC(tp->t_state); 1613 if (tp->t_fb->tfb_tcp_fb_fini) 1614 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1615 refcount_release(&tp->t_fb->tfb_refcnt); 1616 tp->t_inpcb = NULL; 1617 uma_zfree(V_tcpcb_zone, tp); 1618 if (in_pcbrele_wlocked(inp)) { 1619 INP_INFO_RUNLOCK(&V_tcbinfo); 1620 CURVNET_RESTORE(); 1621 return; 1622 } 1623 } 1624 INP_WUNLOCK(inp); 1625 INP_INFO_RUNLOCK(&V_tcbinfo); 1626 CURVNET_RESTORE(); 1627 } 1628 1629 /* 1630 * Attempt to close a TCP control block, marking it as dropped, and freeing 1631 * the socket if we hold the only reference. 1632 */ 1633 struct tcpcb * 1634 tcp_close(struct tcpcb *tp) 1635 { 1636 struct inpcb *inp = tp->t_inpcb; 1637 struct socket *so; 1638 1639 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1640 INP_WLOCK_ASSERT(inp); 1641 1642 #ifdef TCP_OFFLOAD 1643 if (tp->t_state == TCPS_LISTEN) 1644 tcp_offload_listen_stop(tp); 1645 #endif 1646 #ifdef TCP_RFC7413 1647 /* 1648 * This releases the TFO pending counter resource for TFO listen 1649 * sockets as well as passively-created TFO sockets that transition 1650 * from SYN_RECEIVED to CLOSED. 1651 */ 1652 if (tp->t_tfo_pending) { 1653 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1654 tp->t_tfo_pending = NULL; 1655 } 1656 #endif 1657 in_pcbdrop(inp); 1658 TCPSTAT_INC(tcps_closed); 1659 if (tp->t_state != TCPS_CLOSED) 1660 tcp_state_change(tp, TCPS_CLOSED); 1661 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1662 so = inp->inp_socket; 1663 soisdisconnected(so); 1664 if (inp->inp_flags & INP_SOCKREF) { 1665 KASSERT(so->so_state & SS_PROTOREF, 1666 ("tcp_close: !SS_PROTOREF")); 1667 inp->inp_flags &= ~INP_SOCKREF; 1668 INP_WUNLOCK(inp); 1669 SOCK_LOCK(so); 1670 so->so_state &= ~SS_PROTOREF; 1671 sofree(so); 1672 return (NULL); 1673 } 1674 return (tp); 1675 } 1676 1677 void 1678 tcp_drain(void) 1679 { 1680 VNET_ITERATOR_DECL(vnet_iter); 1681 1682 if (!do_tcpdrain) 1683 return; 1684 1685 VNET_LIST_RLOCK_NOSLEEP(); 1686 VNET_FOREACH(vnet_iter) { 1687 CURVNET_SET(vnet_iter); 1688 struct inpcb *inpb; 1689 struct tcpcb *tcpb; 1690 1691 /* 1692 * Walk the tcpbs, if existing, and flush the reassembly queue, 1693 * if there is one... 1694 * XXX: The "Net/3" implementation doesn't imply that the TCP 1695 * reassembly queue should be flushed, but in a situation 1696 * where we're really low on mbufs, this is potentially 1697 * useful. 1698 */ 1699 INP_INFO_WLOCK(&V_tcbinfo); 1700 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1701 if (inpb->inp_flags & INP_TIMEWAIT) 1702 continue; 1703 INP_WLOCK(inpb); 1704 if ((tcpb = intotcpcb(inpb)) != NULL) { 1705 tcp_reass_flush(tcpb); 1706 tcp_clean_sackreport(tcpb); 1707 #ifdef TCPPCAP 1708 if (tcp_pcap_aggressive_free) { 1709 /* Free the TCP PCAP queues. */ 1710 tcp_pcap_drain(&(tcpb->t_inpkts)); 1711 tcp_pcap_drain(&(tcpb->t_outpkts)); 1712 } 1713 #endif 1714 } 1715 INP_WUNLOCK(inpb); 1716 } 1717 INP_INFO_WUNLOCK(&V_tcbinfo); 1718 CURVNET_RESTORE(); 1719 } 1720 VNET_LIST_RUNLOCK_NOSLEEP(); 1721 } 1722 1723 /* 1724 * Notify a tcp user of an asynchronous error; 1725 * store error as soft error, but wake up user 1726 * (for now, won't do anything until can select for soft error). 1727 * 1728 * Do not wake up user since there currently is no mechanism for 1729 * reporting soft errors (yet - a kqueue filter may be added). 1730 */ 1731 static struct inpcb * 1732 tcp_notify(struct inpcb *inp, int error) 1733 { 1734 struct tcpcb *tp; 1735 1736 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1737 INP_WLOCK_ASSERT(inp); 1738 1739 if ((inp->inp_flags & INP_TIMEWAIT) || 1740 (inp->inp_flags & INP_DROPPED)) 1741 return (inp); 1742 1743 tp = intotcpcb(inp); 1744 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1745 1746 /* 1747 * Ignore some errors if we are hooked up. 1748 * If connection hasn't completed, has retransmitted several times, 1749 * and receives a second error, give up now. This is better 1750 * than waiting a long time to establish a connection that 1751 * can never complete. 1752 */ 1753 if (tp->t_state == TCPS_ESTABLISHED && 1754 (error == EHOSTUNREACH || error == ENETUNREACH || 1755 error == EHOSTDOWN)) { 1756 if (inp->inp_route.ro_rt) { 1757 RTFREE(inp->inp_route.ro_rt); 1758 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1759 } 1760 return (inp); 1761 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1762 tp->t_softerror) { 1763 tp = tcp_drop(tp, error); 1764 if (tp != NULL) 1765 return (inp); 1766 else 1767 return (NULL); 1768 } else { 1769 tp->t_softerror = error; 1770 return (inp); 1771 } 1772 #if 0 1773 wakeup( &so->so_timeo); 1774 sorwakeup(so); 1775 sowwakeup(so); 1776 #endif 1777 } 1778 1779 static int 1780 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1781 { 1782 int error, i, m, n, pcb_count; 1783 struct inpcb *inp, **inp_list; 1784 inp_gen_t gencnt; 1785 struct xinpgen xig; 1786 1787 /* 1788 * The process of preparing the TCB list is too time-consuming and 1789 * resource-intensive to repeat twice on every request. 1790 */ 1791 if (req->oldptr == NULL) { 1792 n = V_tcbinfo.ipi_count + 1793 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1794 n += imax(n / 8, 10); 1795 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1796 return (0); 1797 } 1798 1799 if (req->newptr != NULL) 1800 return (EPERM); 1801 1802 /* 1803 * OK, now we're committed to doing something. 1804 */ 1805 INP_LIST_RLOCK(&V_tcbinfo); 1806 gencnt = V_tcbinfo.ipi_gencnt; 1807 n = V_tcbinfo.ipi_count; 1808 INP_LIST_RUNLOCK(&V_tcbinfo); 1809 1810 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1811 1812 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1813 + (n + m) * sizeof(struct xtcpcb)); 1814 if (error != 0) 1815 return (error); 1816 1817 xig.xig_len = sizeof xig; 1818 xig.xig_count = n + m; 1819 xig.xig_gen = gencnt; 1820 xig.xig_sogen = so_gencnt; 1821 error = SYSCTL_OUT(req, &xig, sizeof xig); 1822 if (error) 1823 return (error); 1824 1825 error = syncache_pcblist(req, m, &pcb_count); 1826 if (error) 1827 return (error); 1828 1829 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1830 1831 INP_INFO_WLOCK(&V_tcbinfo); 1832 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1833 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1834 INP_WLOCK(inp); 1835 if (inp->inp_gencnt <= gencnt) { 1836 /* 1837 * XXX: This use of cr_cansee(), introduced with 1838 * TCP state changes, is not quite right, but for 1839 * now, better than nothing. 1840 */ 1841 if (inp->inp_flags & INP_TIMEWAIT) { 1842 if (intotw(inp) != NULL) 1843 error = cr_cansee(req->td->td_ucred, 1844 intotw(inp)->tw_cred); 1845 else 1846 error = EINVAL; /* Skip this inp. */ 1847 } else 1848 error = cr_canseeinpcb(req->td->td_ucred, inp); 1849 if (error == 0) { 1850 in_pcbref(inp); 1851 inp_list[i++] = inp; 1852 } 1853 } 1854 INP_WUNLOCK(inp); 1855 } 1856 INP_INFO_WUNLOCK(&V_tcbinfo); 1857 n = i; 1858 1859 error = 0; 1860 for (i = 0; i < n; i++) { 1861 inp = inp_list[i]; 1862 INP_RLOCK(inp); 1863 if (inp->inp_gencnt <= gencnt) { 1864 struct xtcpcb xt; 1865 1866 tcp_inptoxtp(inp, &xt); 1867 INP_RUNLOCK(inp); 1868 error = SYSCTL_OUT(req, &xt, sizeof xt); 1869 } else 1870 INP_RUNLOCK(inp); 1871 } 1872 INP_INFO_RLOCK(&V_tcbinfo); 1873 for (i = 0; i < n; i++) { 1874 inp = inp_list[i]; 1875 INP_RLOCK(inp); 1876 if (!in_pcbrele_rlocked(inp)) 1877 INP_RUNLOCK(inp); 1878 } 1879 INP_INFO_RUNLOCK(&V_tcbinfo); 1880 1881 if (!error) { 1882 /* 1883 * Give the user an updated idea of our state. 1884 * If the generation differs from what we told 1885 * her before, she knows that something happened 1886 * while we were processing this request, and it 1887 * might be necessary to retry. 1888 */ 1889 INP_LIST_RLOCK(&V_tcbinfo); 1890 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1891 xig.xig_sogen = so_gencnt; 1892 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1893 INP_LIST_RUNLOCK(&V_tcbinfo); 1894 error = SYSCTL_OUT(req, &xig, sizeof xig); 1895 } 1896 free(inp_list, M_TEMP); 1897 return (error); 1898 } 1899 1900 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1901 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1902 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1903 1904 #ifdef INET 1905 static int 1906 tcp_getcred(SYSCTL_HANDLER_ARGS) 1907 { 1908 struct xucred xuc; 1909 struct sockaddr_in addrs[2]; 1910 struct inpcb *inp; 1911 int error; 1912 1913 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1914 if (error) 1915 return (error); 1916 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1917 if (error) 1918 return (error); 1919 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1920 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1921 if (inp != NULL) { 1922 if (inp->inp_socket == NULL) 1923 error = ENOENT; 1924 if (error == 0) 1925 error = cr_canseeinpcb(req->td->td_ucred, inp); 1926 if (error == 0) 1927 cru2x(inp->inp_cred, &xuc); 1928 INP_RUNLOCK(inp); 1929 } else 1930 error = ENOENT; 1931 if (error == 0) 1932 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1933 return (error); 1934 } 1935 1936 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1937 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1938 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1939 #endif /* INET */ 1940 1941 #ifdef INET6 1942 static int 1943 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1944 { 1945 struct xucred xuc; 1946 struct sockaddr_in6 addrs[2]; 1947 struct inpcb *inp; 1948 int error; 1949 #ifdef INET 1950 int mapped = 0; 1951 #endif 1952 1953 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1954 if (error) 1955 return (error); 1956 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1957 if (error) 1958 return (error); 1959 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1960 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1961 return (error); 1962 } 1963 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1964 #ifdef INET 1965 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1966 mapped = 1; 1967 else 1968 #endif 1969 return (EINVAL); 1970 } 1971 1972 #ifdef INET 1973 if (mapped == 1) 1974 inp = in_pcblookup(&V_tcbinfo, 1975 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1976 addrs[1].sin6_port, 1977 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1978 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1979 else 1980 #endif 1981 inp = in6_pcblookup(&V_tcbinfo, 1982 &addrs[1].sin6_addr, addrs[1].sin6_port, 1983 &addrs[0].sin6_addr, addrs[0].sin6_port, 1984 INPLOOKUP_RLOCKPCB, NULL); 1985 if (inp != NULL) { 1986 if (inp->inp_socket == NULL) 1987 error = ENOENT; 1988 if (error == 0) 1989 error = cr_canseeinpcb(req->td->td_ucred, inp); 1990 if (error == 0) 1991 cru2x(inp->inp_cred, &xuc); 1992 INP_RUNLOCK(inp); 1993 } else 1994 error = ENOENT; 1995 if (error == 0) 1996 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1997 return (error); 1998 } 1999 2000 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2001 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2002 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2003 #endif /* INET6 */ 2004 2005 2006 #ifdef INET 2007 void 2008 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2009 { 2010 struct ip *ip = vip; 2011 struct tcphdr *th; 2012 struct in_addr faddr; 2013 struct inpcb *inp; 2014 struct tcpcb *tp; 2015 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2016 struct icmp *icp; 2017 struct in_conninfo inc; 2018 tcp_seq icmp_tcp_seq; 2019 int mtu; 2020 2021 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2022 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2023 return; 2024 2025 if (cmd == PRC_MSGSIZE) 2026 notify = tcp_mtudisc_notify; 2027 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2028 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2029 cmd == PRC_TIMXCEED_INTRANS) && ip) 2030 notify = tcp_drop_syn_sent; 2031 2032 /* 2033 * Hostdead is ugly because it goes linearly through all PCBs. 2034 * XXX: We never get this from ICMP, otherwise it makes an 2035 * excellent DoS attack on machines with many connections. 2036 */ 2037 else if (cmd == PRC_HOSTDEAD) 2038 ip = NULL; 2039 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2040 return; 2041 2042 if (ip == NULL) { 2043 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2044 return; 2045 } 2046 2047 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2048 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2049 INP_INFO_RLOCK(&V_tcbinfo); 2050 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2051 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2052 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2053 /* signal EHOSTDOWN, as it flushes the cached route */ 2054 inp = (*notify)(inp, EHOSTDOWN); 2055 goto out; 2056 } 2057 icmp_tcp_seq = th->th_seq; 2058 if (inp != NULL) { 2059 if (!(inp->inp_flags & INP_TIMEWAIT) && 2060 !(inp->inp_flags & INP_DROPPED) && 2061 !(inp->inp_socket == NULL)) { 2062 tp = intotcpcb(inp); 2063 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2064 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2065 if (cmd == PRC_MSGSIZE) { 2066 /* 2067 * MTU discovery: 2068 * If we got a needfrag set the MTU 2069 * in the route to the suggested new 2070 * value (if given) and then notify. 2071 */ 2072 mtu = ntohs(icp->icmp_nextmtu); 2073 /* 2074 * If no alternative MTU was 2075 * proposed, try the next smaller 2076 * one. 2077 */ 2078 if (!mtu) 2079 mtu = ip_next_mtu( 2080 ntohs(ip->ip_len), 1); 2081 if (mtu < V_tcp_minmss + 2082 sizeof(struct tcpiphdr)) 2083 mtu = V_tcp_minmss + 2084 sizeof(struct tcpiphdr); 2085 /* 2086 * Only process the offered MTU if it 2087 * is smaller than the current one. 2088 */ 2089 if (mtu < tp->t_maxseg + 2090 sizeof(struct tcpiphdr)) { 2091 bzero(&inc, sizeof(inc)); 2092 inc.inc_faddr = faddr; 2093 inc.inc_fibnum = 2094 inp->inp_inc.inc_fibnum; 2095 tcp_hc_updatemtu(&inc, mtu); 2096 tcp_mtudisc(inp, mtu); 2097 } 2098 } else 2099 inp = (*notify)(inp, 2100 inetctlerrmap[cmd]); 2101 } 2102 } 2103 } else { 2104 bzero(&inc, sizeof(inc)); 2105 inc.inc_fport = th->th_dport; 2106 inc.inc_lport = th->th_sport; 2107 inc.inc_faddr = faddr; 2108 inc.inc_laddr = ip->ip_src; 2109 syncache_unreach(&inc, icmp_tcp_seq); 2110 } 2111 out: 2112 if (inp != NULL) 2113 INP_WUNLOCK(inp); 2114 INP_INFO_RUNLOCK(&V_tcbinfo); 2115 } 2116 #endif /* INET */ 2117 2118 #ifdef INET6 2119 void 2120 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2121 { 2122 struct in6_addr *dst; 2123 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2124 struct ip6_hdr *ip6; 2125 struct mbuf *m; 2126 struct inpcb *inp; 2127 struct tcpcb *tp; 2128 struct icmp6_hdr *icmp6; 2129 struct ip6ctlparam *ip6cp = NULL; 2130 const struct sockaddr_in6 *sa6_src = NULL; 2131 struct in_conninfo inc; 2132 struct tcp_ports { 2133 uint16_t th_sport; 2134 uint16_t th_dport; 2135 } t_ports; 2136 tcp_seq icmp_tcp_seq; 2137 unsigned int mtu; 2138 unsigned int off; 2139 2140 if (sa->sa_family != AF_INET6 || 2141 sa->sa_len != sizeof(struct sockaddr_in6)) 2142 return; 2143 2144 /* if the parameter is from icmp6, decode it. */ 2145 if (d != NULL) { 2146 ip6cp = (struct ip6ctlparam *)d; 2147 icmp6 = ip6cp->ip6c_icmp6; 2148 m = ip6cp->ip6c_m; 2149 ip6 = ip6cp->ip6c_ip6; 2150 off = ip6cp->ip6c_off; 2151 sa6_src = ip6cp->ip6c_src; 2152 dst = ip6cp->ip6c_finaldst; 2153 } else { 2154 m = NULL; 2155 ip6 = NULL; 2156 off = 0; /* fool gcc */ 2157 sa6_src = &sa6_any; 2158 dst = NULL; 2159 } 2160 2161 if (cmd == PRC_MSGSIZE) 2162 notify = tcp_mtudisc_notify; 2163 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2164 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2165 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2166 notify = tcp_drop_syn_sent; 2167 2168 /* 2169 * Hostdead is ugly because it goes linearly through all PCBs. 2170 * XXX: We never get this from ICMP, otherwise it makes an 2171 * excellent DoS attack on machines with many connections. 2172 */ 2173 else if (cmd == PRC_HOSTDEAD) 2174 ip6 = NULL; 2175 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2176 return; 2177 2178 if (ip6 == NULL) { 2179 in6_pcbnotify(&V_tcbinfo, sa, 0, 2180 (const struct sockaddr *)sa6_src, 2181 0, cmd, NULL, notify); 2182 return; 2183 } 2184 2185 /* Check if we can safely get the ports from the tcp hdr */ 2186 if (m == NULL || 2187 (m->m_pkthdr.len < 2188 (int32_t) (off + sizeof(struct tcp_ports)))) { 2189 return; 2190 } 2191 bzero(&t_ports, sizeof(struct tcp_ports)); 2192 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2193 INP_INFO_RLOCK(&V_tcbinfo); 2194 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2195 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2196 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2197 /* signal EHOSTDOWN, as it flushes the cached route */ 2198 inp = (*notify)(inp, EHOSTDOWN); 2199 goto out; 2200 } 2201 off += sizeof(struct tcp_ports); 2202 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2203 goto out; 2204 } 2205 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2206 if (inp != NULL) { 2207 if (!(inp->inp_flags & INP_TIMEWAIT) && 2208 !(inp->inp_flags & INP_DROPPED) && 2209 !(inp->inp_socket == NULL)) { 2210 tp = intotcpcb(inp); 2211 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2212 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2213 if (cmd == PRC_MSGSIZE) { 2214 /* 2215 * MTU discovery: 2216 * If we got a needfrag set the MTU 2217 * in the route to the suggested new 2218 * value (if given) and then notify. 2219 */ 2220 mtu = ntohl(icmp6->icmp6_mtu); 2221 /* 2222 * If no alternative MTU was 2223 * proposed, or the proposed 2224 * MTU was too small, set to 2225 * the min. 2226 */ 2227 if (mtu < IPV6_MMTU) 2228 mtu = IPV6_MMTU - 8; 2229 bzero(&inc, sizeof(inc)); 2230 inc.inc_fibnum = M_GETFIB(m); 2231 inc.inc_flags |= INC_ISIPV6; 2232 inc.inc6_faddr = *dst; 2233 if (in6_setscope(&inc.inc6_faddr, 2234 m->m_pkthdr.rcvif, NULL)) 2235 goto out; 2236 /* 2237 * Only process the offered MTU if it 2238 * is smaller than the current one. 2239 */ 2240 if (mtu < tp->t_maxseg + 2241 sizeof (struct tcphdr) + 2242 sizeof (struct ip6_hdr)) { 2243 tcp_hc_updatemtu(&inc, mtu); 2244 tcp_mtudisc(inp, mtu); 2245 ICMP6STAT_INC(icp6s_pmtuchg); 2246 } 2247 } else 2248 inp = (*notify)(inp, 2249 inet6ctlerrmap[cmd]); 2250 } 2251 } 2252 } else { 2253 bzero(&inc, sizeof(inc)); 2254 inc.inc_fibnum = M_GETFIB(m); 2255 inc.inc_flags |= INC_ISIPV6; 2256 inc.inc_fport = t_ports.th_dport; 2257 inc.inc_lport = t_ports.th_sport; 2258 inc.inc6_faddr = *dst; 2259 inc.inc6_laddr = ip6->ip6_src; 2260 syncache_unreach(&inc, icmp_tcp_seq); 2261 } 2262 out: 2263 if (inp != NULL) 2264 INP_WUNLOCK(inp); 2265 INP_INFO_RUNLOCK(&V_tcbinfo); 2266 } 2267 #endif /* INET6 */ 2268 2269 2270 /* 2271 * Following is where TCP initial sequence number generation occurs. 2272 * 2273 * There are two places where we must use initial sequence numbers: 2274 * 1. In SYN-ACK packets. 2275 * 2. In SYN packets. 2276 * 2277 * All ISNs for SYN-ACK packets are generated by the syncache. See 2278 * tcp_syncache.c for details. 2279 * 2280 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2281 * depends on this property. In addition, these ISNs should be 2282 * unguessable so as to prevent connection hijacking. To satisfy 2283 * the requirements of this situation, the algorithm outlined in 2284 * RFC 1948 is used, with only small modifications. 2285 * 2286 * Implementation details: 2287 * 2288 * Time is based off the system timer, and is corrected so that it 2289 * increases by one megabyte per second. This allows for proper 2290 * recycling on high speed LANs while still leaving over an hour 2291 * before rollover. 2292 * 2293 * As reading the *exact* system time is too expensive to be done 2294 * whenever setting up a TCP connection, we increment the time 2295 * offset in two ways. First, a small random positive increment 2296 * is added to isn_offset for each connection that is set up. 2297 * Second, the function tcp_isn_tick fires once per clock tick 2298 * and increments isn_offset as necessary so that sequence numbers 2299 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2300 * random positive increments serve only to ensure that the same 2301 * exact sequence number is never sent out twice (as could otherwise 2302 * happen when a port is recycled in less than the system tick 2303 * interval.) 2304 * 2305 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2306 * between seeding of isn_secret. This is normally set to zero, 2307 * as reseeding should not be necessary. 2308 * 2309 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2310 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2311 * general, this means holding an exclusive (write) lock. 2312 */ 2313 2314 #define ISN_BYTES_PER_SECOND 1048576 2315 #define ISN_STATIC_INCREMENT 4096 2316 #define ISN_RANDOM_INCREMENT (4096 - 1) 2317 2318 static VNET_DEFINE(u_char, isn_secret[32]); 2319 static VNET_DEFINE(int, isn_last); 2320 static VNET_DEFINE(int, isn_last_reseed); 2321 static VNET_DEFINE(u_int32_t, isn_offset); 2322 static VNET_DEFINE(u_int32_t, isn_offset_old); 2323 2324 #define V_isn_secret VNET(isn_secret) 2325 #define V_isn_last VNET(isn_last) 2326 #define V_isn_last_reseed VNET(isn_last_reseed) 2327 #define V_isn_offset VNET(isn_offset) 2328 #define V_isn_offset_old VNET(isn_offset_old) 2329 2330 tcp_seq 2331 tcp_new_isn(struct tcpcb *tp) 2332 { 2333 MD5_CTX isn_ctx; 2334 u_int32_t md5_buffer[4]; 2335 tcp_seq new_isn; 2336 u_int32_t projected_offset; 2337 2338 INP_WLOCK_ASSERT(tp->t_inpcb); 2339 2340 ISN_LOCK(); 2341 /* Seed if this is the first use, reseed if requested. */ 2342 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2343 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2344 < (u_int)ticks))) { 2345 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2346 V_isn_last_reseed = ticks; 2347 } 2348 2349 /* Compute the md5 hash and return the ISN. */ 2350 MD5Init(&isn_ctx); 2351 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2352 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2353 #ifdef INET6 2354 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2355 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2356 sizeof(struct in6_addr)); 2357 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2358 sizeof(struct in6_addr)); 2359 } else 2360 #endif 2361 { 2362 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2363 sizeof(struct in_addr)); 2364 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2365 sizeof(struct in_addr)); 2366 } 2367 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2368 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2369 new_isn = (tcp_seq) md5_buffer[0]; 2370 V_isn_offset += ISN_STATIC_INCREMENT + 2371 (arc4random() & ISN_RANDOM_INCREMENT); 2372 if (ticks != V_isn_last) { 2373 projected_offset = V_isn_offset_old + 2374 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2375 if (SEQ_GT(projected_offset, V_isn_offset)) 2376 V_isn_offset = projected_offset; 2377 V_isn_offset_old = V_isn_offset; 2378 V_isn_last = ticks; 2379 } 2380 new_isn += V_isn_offset; 2381 ISN_UNLOCK(); 2382 return (new_isn); 2383 } 2384 2385 /* 2386 * When a specific ICMP unreachable message is received and the 2387 * connection state is SYN-SENT, drop the connection. This behavior 2388 * is controlled by the icmp_may_rst sysctl. 2389 */ 2390 struct inpcb * 2391 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2392 { 2393 struct tcpcb *tp; 2394 2395 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2396 INP_WLOCK_ASSERT(inp); 2397 2398 if ((inp->inp_flags & INP_TIMEWAIT) || 2399 (inp->inp_flags & INP_DROPPED)) 2400 return (inp); 2401 2402 tp = intotcpcb(inp); 2403 if (tp->t_state != TCPS_SYN_SENT) 2404 return (inp); 2405 2406 tp = tcp_drop(tp, errno); 2407 if (tp != NULL) 2408 return (inp); 2409 else 2410 return (NULL); 2411 } 2412 2413 /* 2414 * When `need fragmentation' ICMP is received, update our idea of the MSS 2415 * based on the new value. Also nudge TCP to send something, since we 2416 * know the packet we just sent was dropped. 2417 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2418 */ 2419 static struct inpcb * 2420 tcp_mtudisc_notify(struct inpcb *inp, int error) 2421 { 2422 2423 tcp_mtudisc(inp, -1); 2424 return (inp); 2425 } 2426 2427 static void 2428 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2429 { 2430 struct tcpcb *tp; 2431 struct socket *so; 2432 2433 INP_WLOCK_ASSERT(inp); 2434 if ((inp->inp_flags & INP_TIMEWAIT) || 2435 (inp->inp_flags & INP_DROPPED)) 2436 return; 2437 2438 tp = intotcpcb(inp); 2439 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2440 2441 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2442 2443 so = inp->inp_socket; 2444 SOCKBUF_LOCK(&so->so_snd); 2445 /* If the mss is larger than the socket buffer, decrease the mss. */ 2446 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2447 tp->t_maxseg = so->so_snd.sb_hiwat; 2448 SOCKBUF_UNLOCK(&so->so_snd); 2449 2450 TCPSTAT_INC(tcps_mturesent); 2451 tp->t_rtttime = 0; 2452 tp->snd_nxt = tp->snd_una; 2453 tcp_free_sackholes(tp); 2454 tp->snd_recover = tp->snd_max; 2455 if (tp->t_flags & TF_SACK_PERMIT) 2456 EXIT_FASTRECOVERY(tp->t_flags); 2457 tp->t_fb->tfb_tcp_output(tp); 2458 } 2459 2460 #ifdef INET 2461 /* 2462 * Look-up the routing entry to the peer of this inpcb. If no route 2463 * is found and it cannot be allocated, then return 0. This routine 2464 * is called by TCP routines that access the rmx structure and by 2465 * tcp_mss_update to get the peer/interface MTU. 2466 */ 2467 uint32_t 2468 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2469 { 2470 struct nhop4_extended nh4; 2471 struct ifnet *ifp; 2472 uint32_t maxmtu = 0; 2473 2474 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2475 2476 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2477 2478 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2479 NHR_REF, 0, &nh4) != 0) 2480 return (0); 2481 2482 ifp = nh4.nh_ifp; 2483 maxmtu = nh4.nh_mtu; 2484 2485 /* Report additional interface capabilities. */ 2486 if (cap != NULL) { 2487 if (ifp->if_capenable & IFCAP_TSO4 && 2488 ifp->if_hwassist & CSUM_TSO) { 2489 cap->ifcap |= CSUM_TSO; 2490 cap->tsomax = ifp->if_hw_tsomax; 2491 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2492 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2493 } 2494 } 2495 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2496 } 2497 return (maxmtu); 2498 } 2499 #endif /* INET */ 2500 2501 #ifdef INET6 2502 uint32_t 2503 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2504 { 2505 struct nhop6_extended nh6; 2506 struct in6_addr dst6; 2507 uint32_t scopeid; 2508 struct ifnet *ifp; 2509 uint32_t maxmtu = 0; 2510 2511 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2512 2513 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2514 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2515 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2516 0, &nh6) != 0) 2517 return (0); 2518 2519 ifp = nh6.nh_ifp; 2520 maxmtu = nh6.nh_mtu; 2521 2522 /* Report additional interface capabilities. */ 2523 if (cap != NULL) { 2524 if (ifp->if_capenable & IFCAP_TSO6 && 2525 ifp->if_hwassist & CSUM_TSO) { 2526 cap->ifcap |= CSUM_TSO; 2527 cap->tsomax = ifp->if_hw_tsomax; 2528 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2529 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2530 } 2531 } 2532 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2533 } 2534 2535 return (maxmtu); 2536 } 2537 #endif /* INET6 */ 2538 2539 /* 2540 * Calculate effective SMSS per RFC5681 definition for a given TCP 2541 * connection at its current state, taking into account SACK and etc. 2542 */ 2543 u_int 2544 tcp_maxseg(const struct tcpcb *tp) 2545 { 2546 u_int optlen; 2547 2548 if (tp->t_flags & TF_NOOPT) 2549 return (tp->t_maxseg); 2550 2551 /* 2552 * Here we have a simplified code from tcp_addoptions(), 2553 * without a proper loop, and having most of paddings hardcoded. 2554 * We might make mistakes with padding here in some edge cases, 2555 * but this is harmless, since result of tcp_maxseg() is used 2556 * only in cwnd and ssthresh estimations. 2557 */ 2558 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2559 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2560 if (tp->t_flags & TF_RCVD_TSTMP) 2561 optlen = TCPOLEN_TSTAMP_APPA; 2562 else 2563 optlen = 0; 2564 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2565 if (tp->t_flags & TF_SIGNATURE) 2566 optlen += PAD(TCPOLEN_SIGNATURE); 2567 #endif 2568 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2569 optlen += TCPOLEN_SACKHDR; 2570 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2571 optlen = PAD(optlen); 2572 } 2573 } else { 2574 if (tp->t_flags & TF_REQ_TSTMP) 2575 optlen = TCPOLEN_TSTAMP_APPA; 2576 else 2577 optlen = PAD(TCPOLEN_MAXSEG); 2578 if (tp->t_flags & TF_REQ_SCALE) 2579 optlen += PAD(TCPOLEN_WINDOW); 2580 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2581 if (tp->t_flags & TF_SIGNATURE) 2582 optlen += PAD(TCPOLEN_SIGNATURE); 2583 #endif 2584 if (tp->t_flags & TF_SACK_PERMIT) 2585 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2586 } 2587 #undef PAD 2588 optlen = min(optlen, TCP_MAXOLEN); 2589 return (tp->t_maxseg - optlen); 2590 } 2591 2592 static int 2593 sysctl_drop(SYSCTL_HANDLER_ARGS) 2594 { 2595 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2596 struct sockaddr_storage addrs[2]; 2597 struct inpcb *inp; 2598 struct tcpcb *tp; 2599 struct tcptw *tw; 2600 struct sockaddr_in *fin, *lin; 2601 #ifdef INET6 2602 struct sockaddr_in6 *fin6, *lin6; 2603 #endif 2604 int error; 2605 2606 inp = NULL; 2607 fin = lin = NULL; 2608 #ifdef INET6 2609 fin6 = lin6 = NULL; 2610 #endif 2611 error = 0; 2612 2613 if (req->oldptr != NULL || req->oldlen != 0) 2614 return (EINVAL); 2615 if (req->newptr == NULL) 2616 return (EPERM); 2617 if (req->newlen < sizeof(addrs)) 2618 return (ENOMEM); 2619 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2620 if (error) 2621 return (error); 2622 2623 switch (addrs[0].ss_family) { 2624 #ifdef INET6 2625 case AF_INET6: 2626 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2627 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2628 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2629 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2630 return (EINVAL); 2631 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2632 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2633 return (EINVAL); 2634 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2635 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2636 fin = (struct sockaddr_in *)&addrs[0]; 2637 lin = (struct sockaddr_in *)&addrs[1]; 2638 break; 2639 } 2640 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2641 if (error) 2642 return (error); 2643 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2644 if (error) 2645 return (error); 2646 break; 2647 #endif 2648 #ifdef INET 2649 case AF_INET: 2650 fin = (struct sockaddr_in *)&addrs[0]; 2651 lin = (struct sockaddr_in *)&addrs[1]; 2652 if (fin->sin_len != sizeof(struct sockaddr_in) || 2653 lin->sin_len != sizeof(struct sockaddr_in)) 2654 return (EINVAL); 2655 break; 2656 #endif 2657 default: 2658 return (EINVAL); 2659 } 2660 INP_INFO_RLOCK(&V_tcbinfo); 2661 switch (addrs[0].ss_family) { 2662 #ifdef INET6 2663 case AF_INET6: 2664 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2665 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2666 INPLOOKUP_WLOCKPCB, NULL); 2667 break; 2668 #endif 2669 #ifdef INET 2670 case AF_INET: 2671 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2672 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2673 break; 2674 #endif 2675 } 2676 if (inp != NULL) { 2677 if (inp->inp_flags & INP_TIMEWAIT) { 2678 /* 2679 * XXXRW: There currently exists a state where an 2680 * inpcb is present, but its timewait state has been 2681 * discarded. For now, don't allow dropping of this 2682 * type of inpcb. 2683 */ 2684 tw = intotw(inp); 2685 if (tw != NULL) 2686 tcp_twclose(tw, 0); 2687 else 2688 INP_WUNLOCK(inp); 2689 } else if (!(inp->inp_flags & INP_DROPPED) && 2690 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2691 tp = intotcpcb(inp); 2692 tp = tcp_drop(tp, ECONNABORTED); 2693 if (tp != NULL) 2694 INP_WUNLOCK(inp); 2695 } else 2696 INP_WUNLOCK(inp); 2697 } else 2698 error = ESRCH; 2699 INP_INFO_RUNLOCK(&V_tcbinfo); 2700 return (error); 2701 } 2702 2703 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2704 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2705 0, sysctl_drop, "", "Drop TCP connection"); 2706 2707 /* 2708 * Generate a standardized TCP log line for use throughout the 2709 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2710 * allow use in the interrupt context. 2711 * 2712 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2713 * NB: The function may return NULL if memory allocation failed. 2714 * 2715 * Due to header inclusion and ordering limitations the struct ip 2716 * and ip6_hdr pointers have to be passed as void pointers. 2717 */ 2718 char * 2719 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2720 const void *ip6hdr) 2721 { 2722 2723 /* Is logging enabled? */ 2724 if (tcp_log_in_vain == 0) 2725 return (NULL); 2726 2727 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2728 } 2729 2730 char * 2731 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2732 const void *ip6hdr) 2733 { 2734 2735 /* Is logging enabled? */ 2736 if (tcp_log_debug == 0) 2737 return (NULL); 2738 2739 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2740 } 2741 2742 static char * 2743 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2744 const void *ip6hdr) 2745 { 2746 char *s, *sp; 2747 size_t size; 2748 struct ip *ip; 2749 #ifdef INET6 2750 const struct ip6_hdr *ip6; 2751 2752 ip6 = (const struct ip6_hdr *)ip6hdr; 2753 #endif /* INET6 */ 2754 ip = (struct ip *)ip4hdr; 2755 2756 /* 2757 * The log line looks like this: 2758 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2759 */ 2760 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2761 sizeof(PRINT_TH_FLAGS) + 1 + 2762 #ifdef INET6 2763 2 * INET6_ADDRSTRLEN; 2764 #else 2765 2 * INET_ADDRSTRLEN; 2766 #endif /* INET6 */ 2767 2768 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2769 if (s == NULL) 2770 return (NULL); 2771 2772 strcat(s, "TCP: ["); 2773 sp = s + strlen(s); 2774 2775 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2776 inet_ntoa_r(inc->inc_faddr, sp); 2777 sp = s + strlen(s); 2778 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2779 sp = s + strlen(s); 2780 inet_ntoa_r(inc->inc_laddr, sp); 2781 sp = s + strlen(s); 2782 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2783 #ifdef INET6 2784 } else if (inc) { 2785 ip6_sprintf(sp, &inc->inc6_faddr); 2786 sp = s + strlen(s); 2787 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2788 sp = s + strlen(s); 2789 ip6_sprintf(sp, &inc->inc6_laddr); 2790 sp = s + strlen(s); 2791 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2792 } else if (ip6 && th) { 2793 ip6_sprintf(sp, &ip6->ip6_src); 2794 sp = s + strlen(s); 2795 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2796 sp = s + strlen(s); 2797 ip6_sprintf(sp, &ip6->ip6_dst); 2798 sp = s + strlen(s); 2799 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2800 #endif /* INET6 */ 2801 #ifdef INET 2802 } else if (ip && th) { 2803 inet_ntoa_r(ip->ip_src, sp); 2804 sp = s + strlen(s); 2805 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2806 sp = s + strlen(s); 2807 inet_ntoa_r(ip->ip_dst, sp); 2808 sp = s + strlen(s); 2809 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2810 #endif /* INET */ 2811 } else { 2812 free(s, M_TCPLOG); 2813 return (NULL); 2814 } 2815 sp = s + strlen(s); 2816 if (th) 2817 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2818 if (*(s + size - 1) != '\0') 2819 panic("%s: string too long", __func__); 2820 return (s); 2821 } 2822 2823 /* 2824 * A subroutine which makes it easy to track TCP state changes with DTrace. 2825 * This function shouldn't be called for t_state initializations that don't 2826 * correspond to actual TCP state transitions. 2827 */ 2828 void 2829 tcp_state_change(struct tcpcb *tp, int newstate) 2830 { 2831 #if defined(KDTRACE_HOOKS) 2832 int pstate = tp->t_state; 2833 #endif 2834 2835 TCPSTATES_DEC(tp->t_state); 2836 TCPSTATES_INC(newstate); 2837 tp->t_state = newstate; 2838 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2839 } 2840 2841 /* 2842 * Create an external-format (``xtcpcb'') structure using the information in 2843 * the kernel-format tcpcb structure pointed to by tp. This is done to 2844 * reduce the spew of irrelevant information over this interface, to isolate 2845 * user code from changes in the kernel structure, and potentially to provide 2846 * information-hiding if we decide that some of this information should be 2847 * hidden from users. 2848 */ 2849 void 2850 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 2851 { 2852 struct tcpcb *tp = intotcpcb(inp); 2853 sbintime_t now; 2854 2855 if (inp->inp_flags & INP_TIMEWAIT) { 2856 bzero(xt, sizeof(struct xtcpcb)); 2857 xt->t_state = TCPS_TIME_WAIT; 2858 } else { 2859 xt->t_state = tp->t_state; 2860 xt->t_flags = tp->t_flags; 2861 xt->t_sndzerowin = tp->t_sndzerowin; 2862 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 2863 xt->t_rcvoopack = tp->t_rcvoopack; 2864 2865 now = getsbinuptime(); 2866 #define COPYTIMER(ttt) do { \ 2867 if (callout_active(&tp->t_timers->ttt)) \ 2868 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 2869 SBT_1MS; \ 2870 else \ 2871 xt->ttt = 0; \ 2872 } while (0) 2873 COPYTIMER(tt_delack); 2874 COPYTIMER(tt_rexmt); 2875 COPYTIMER(tt_persist); 2876 COPYTIMER(tt_keep); 2877 COPYTIMER(tt_2msl); 2878 #undef COPYTIMER 2879 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 2880 2881 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 2882 TCP_FUNCTION_NAME_LEN_MAX); 2883 } 2884 2885 xt->xt_len = sizeof(struct xtcpcb); 2886 in_pcbtoxinpcb(inp, &xt->xt_inp); 2887 if (inp->inp_socket == NULL) 2888 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 2889 } 2890