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