1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/callout.h> 45 #include <sys/eventhandler.h> 46 #ifdef TCP_HHOOK 47 #include <sys/hhook.h> 48 #endif 49 #include <sys/kernel.h> 50 #ifdef TCP_HHOOK 51 #include <sys/khelp.h> 52 #endif 53 #include <sys/sysctl.h> 54 #include <sys/jail.h> 55 #include <sys/malloc.h> 56 #include <sys/refcount.h> 57 #include <sys/mbuf.h> 58 #ifdef INET6 59 #include <sys/domain.h> 60 #endif 61 #include <sys/priv.h> 62 #include <sys/proc.h> 63 #include <sys/sdt.h> 64 #include <sys/socket.h> 65 #include <sys/socketvar.h> 66 #include <sys/protosw.h> 67 #include <sys/random.h> 68 69 #include <vm/uma.h> 70 71 #include <net/route.h> 72 #include <net/if.h> 73 #include <net/if_var.h> 74 #include <net/vnet.h> 75 76 #include <netinet/in.h> 77 #include <netinet/in_fib.h> 78 #include <netinet/in_kdtrace.h> 79 #include <netinet/in_pcb.h> 80 #include <netinet/in_systm.h> 81 #include <netinet/in_var.h> 82 #include <netinet/ip.h> 83 #include <netinet/ip_icmp.h> 84 #include <netinet/ip_var.h> 85 #ifdef INET6 86 #include <netinet/icmp6.h> 87 #include <netinet/ip6.h> 88 #include <netinet6/in6_fib.h> 89 #include <netinet6/in6_pcb.h> 90 #include <netinet6/ip6_var.h> 91 #include <netinet6/scope6_var.h> 92 #include <netinet6/nd6.h> 93 #endif 94 95 #include <netinet/tcp.h> 96 #include <netinet/tcp_fsm.h> 97 #include <netinet/tcp_seq.h> 98 #include <netinet/tcp_timer.h> 99 #include <netinet/tcp_var.h> 100 #include <netinet/tcp_log_buf.h> 101 #include <netinet/tcp_syncache.h> 102 #include <netinet/tcp_hpts.h> 103 #include <netinet/cc/cc.h> 104 #ifdef INET6 105 #include <netinet6/tcp6_var.h> 106 #endif 107 #include <netinet/tcpip.h> 108 #include <netinet/tcp_fastopen.h> 109 #ifdef TCPPCAP 110 #include <netinet/tcp_pcap.h> 111 #endif 112 #ifdef TCPDEBUG 113 #include <netinet/tcp_debug.h> 114 #endif 115 #ifdef INET6 116 #include <netinet6/ip6protosw.h> 117 #endif 118 #ifdef TCP_OFFLOAD 119 #include <netinet/tcp_offload.h> 120 #endif 121 122 #include <netipsec/ipsec_support.h> 123 124 #include <machine/in_cksum.h> 125 #include <sys/md5.h> 126 127 #include <security/mac/mac_framework.h> 128 129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 130 #ifdef INET6 131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 132 #endif 133 134 struct rwlock tcp_function_lock; 135 136 static int 137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 138 { 139 int error, new; 140 141 new = V_tcp_mssdflt; 142 error = sysctl_handle_int(oidp, &new, 0, req); 143 if (error == 0 && req->newptr) { 144 if (new < TCP_MINMSS) 145 error = EINVAL; 146 else 147 V_tcp_mssdflt = new; 148 } 149 return (error); 150 } 151 152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 153 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 154 &sysctl_net_inet_tcp_mss_check, "I", 155 "Default TCP Maximum Segment Size"); 156 157 #ifdef INET6 158 static int 159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 160 { 161 int error, new; 162 163 new = V_tcp_v6mssdflt; 164 error = sysctl_handle_int(oidp, &new, 0, req); 165 if (error == 0 && req->newptr) { 166 if (new < TCP_MINMSS) 167 error = EINVAL; 168 else 169 V_tcp_v6mssdflt = new; 170 } 171 return (error); 172 } 173 174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 175 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 176 &sysctl_net_inet_tcp_mss_v6_check, "I", 177 "Default TCP Maximum Segment Size for IPv6"); 178 #endif /* INET6 */ 179 180 /* 181 * Minimum MSS we accept and use. This prevents DoS attacks where 182 * we are forced to a ridiculous low MSS like 20 and send hundreds 183 * of packets instead of one. The effect scales with the available 184 * bandwidth and quickly saturates the CPU and network interface 185 * with packet generation and sending. Set to zero to disable MINMSS 186 * checking. This setting prevents us from sending too small packets. 187 */ 188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 190 &VNET_NAME(tcp_minmss), 0, 191 "Minimum TCP Maximum Segment Size"); 192 193 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 195 &VNET_NAME(tcp_do_rfc1323), 0, 196 "Enable rfc1323 (high performance TCP) extensions"); 197 198 static int tcp_log_debug = 0; 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 200 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 201 202 static int tcp_tcbhashsize; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 204 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 205 206 static int do_tcpdrain = 1; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 208 "Enable tcp_drain routine for extra help when low on mbufs"); 209 210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 211 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 212 213 static VNET_DEFINE(int, icmp_may_rst) = 1; 214 #define V_icmp_may_rst VNET(icmp_may_rst) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(icmp_may_rst), 0, 217 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 218 219 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 220 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_isn_reseed_interval), 0, 223 "Seconds between reseeding of ISN secret"); 224 225 static int tcp_soreceive_stream; 226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 227 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 228 229 VNET_DEFINE(uma_zone_t, sack_hole_zone); 230 #define V_sack_hole_zone VNET(sack_hole_zone) 231 232 #ifdef TCP_HHOOK 233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 234 #endif 235 236 static int tcp_default_fb_init(struct tcpcb *tp); 237 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 238 static int tcp_default_handoff_ok(struct tcpcb *tp); 239 static struct inpcb *tcp_notify(struct inpcb *, int); 240 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 241 static void tcp_mtudisc(struct inpcb *, int); 242 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 243 void *ip4hdr, const void *ip6hdr); 244 245 246 static struct tcp_function_block tcp_def_funcblk = { 247 .tfb_tcp_block_name = "freebsd", 248 .tfb_tcp_output = tcp_output, 249 .tfb_tcp_do_segment = tcp_do_segment, 250 .tfb_tcp_ctloutput = tcp_default_ctloutput, 251 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 252 .tfb_tcp_fb_init = tcp_default_fb_init, 253 .tfb_tcp_fb_fini = tcp_default_fb_fini, 254 }; 255 256 int t_functions_inited = 0; 257 static int tcp_fb_cnt = 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 static struct tcp_function_block * 331 find_and_ref_tcp_default_fb(void) 332 { 333 struct tcp_function_block *rblk; 334 335 rw_rlock(&tcp_function_lock); 336 rblk = tcp_func_set_ptr; 337 refcount_acquire(&rblk->tfb_refcnt); 338 rw_runlock(&tcp_function_lock); 339 return (rblk); 340 } 341 342 void 343 tcp_switch_back_to_default(struct tcpcb *tp) 344 { 345 struct tcp_function_block *tfb; 346 347 KASSERT(tp->t_fb != &tcp_def_funcblk, 348 ("%s: called by the built-in default stack", __func__)); 349 350 /* 351 * Release the old stack. This function will either find a new one 352 * or panic. 353 */ 354 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 355 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 356 refcount_release(&tp->t_fb->tfb_refcnt); 357 358 /* 359 * Now, we'll find a new function block to use. 360 * Start by trying the current user-selected 361 * default, unless this stack is the user-selected 362 * default. 363 */ 364 tfb = find_and_ref_tcp_default_fb(); 365 if (tfb == tp->t_fb) { 366 refcount_release(&tfb->tfb_refcnt); 367 tfb = NULL; 368 } 369 /* Does the stack accept this connection? */ 370 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 371 (*tfb->tfb_tcp_handoff_ok)(tp)) { 372 refcount_release(&tfb->tfb_refcnt); 373 tfb = NULL; 374 } 375 /* Try to use that stack. */ 376 if (tfb != NULL) { 377 /* Initialize the new stack. If it succeeds, we are done. */ 378 tp->t_fb = tfb; 379 if (tp->t_fb->tfb_tcp_fb_init == NULL || 380 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 381 return; 382 383 /* 384 * Initialization failed. Release the reference count on 385 * the stack. 386 */ 387 refcount_release(&tfb->tfb_refcnt); 388 } 389 390 /* 391 * If that wasn't feasible, use the built-in default 392 * stack which is not allowed to reject anyone. 393 */ 394 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 395 if (tfb == NULL) { 396 /* there always should be a default */ 397 panic("Can't refer to tcp_def_funcblk"); 398 } 399 if (tfb->tfb_tcp_handoff_ok != NULL) { 400 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 401 /* The default stack cannot say no */ 402 panic("Default stack rejects a new session?"); 403 } 404 } 405 tp->t_fb = tfb; 406 if (tp->t_fb->tfb_tcp_fb_init != NULL && 407 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 408 /* The default stack cannot fail */ 409 panic("Default stack initialization failed"); 410 } 411 } 412 413 static int 414 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 415 { 416 int error=ENOENT; 417 struct tcp_function_set fs; 418 struct tcp_function_block *blk; 419 420 memset(&fs, 0, sizeof(fs)); 421 rw_rlock(&tcp_function_lock); 422 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 423 if (blk) { 424 /* Found him */ 425 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 426 fs.pcbcnt = blk->tfb_refcnt; 427 } 428 rw_runlock(&tcp_function_lock); 429 error = sysctl_handle_string(oidp, fs.function_set_name, 430 sizeof(fs.function_set_name), req); 431 432 /* Check for error or no change */ 433 if (error != 0 || req->newptr == NULL) 434 return(error); 435 436 rw_wlock(&tcp_function_lock); 437 blk = find_tcp_functions_locked(&fs); 438 if ((blk == NULL) || 439 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 440 error = ENOENT; 441 goto done; 442 } 443 tcp_func_set_ptr = blk; 444 done: 445 rw_wunlock(&tcp_function_lock); 446 return (error); 447 } 448 449 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 450 CTLTYPE_STRING | CTLFLAG_RW, 451 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 452 "Set/get the default TCP functions"); 453 454 static int 455 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 456 { 457 int error, cnt, linesz; 458 struct tcp_function *f; 459 char *buffer, *cp; 460 size_t bufsz, outsz; 461 bool alias; 462 463 cnt = 0; 464 rw_rlock(&tcp_function_lock); 465 TAILQ_FOREACH(f, &t_functions, tf_next) { 466 cnt++; 467 } 468 rw_runlock(&tcp_function_lock); 469 470 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 471 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 472 473 error = 0; 474 cp = buffer; 475 476 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 477 "Alias", "PCB count"); 478 cp += linesz; 479 bufsz -= linesz; 480 outsz = linesz; 481 482 rw_rlock(&tcp_function_lock); 483 TAILQ_FOREACH(f, &t_functions, tf_next) { 484 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 485 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 486 f->tf_fb->tfb_tcp_block_name, 487 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 488 alias ? f->tf_name : "-", 489 f->tf_fb->tfb_refcnt); 490 if (linesz >= bufsz) { 491 error = EOVERFLOW; 492 break; 493 } 494 cp += linesz; 495 bufsz -= linesz; 496 outsz += linesz; 497 } 498 rw_runlock(&tcp_function_lock); 499 if (error == 0) 500 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 501 free(buffer, M_TEMP); 502 return (error); 503 } 504 505 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 506 CTLTYPE_STRING|CTLFLAG_RD, 507 NULL, 0, sysctl_net_inet_list_available, "A", 508 "list available TCP Function sets"); 509 510 /* 511 * Exports one (struct tcp_function_info) for each alias/name. 512 */ 513 static int 514 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 515 { 516 int cnt, error; 517 struct tcp_function *f; 518 struct tcp_function_info tfi; 519 520 /* 521 * We don't allow writes. 522 */ 523 if (req->newptr != NULL) 524 return (EINVAL); 525 526 /* 527 * Wire the old buffer so we can directly copy the functions to 528 * user space without dropping the lock. 529 */ 530 if (req->oldptr != NULL) { 531 error = sysctl_wire_old_buffer(req, 0); 532 if (error) 533 return (error); 534 } 535 536 /* 537 * Walk the list and copy out matching entries. If INVARIANTS 538 * is compiled in, also walk the list to verify the length of 539 * the list matches what we have recorded. 540 */ 541 rw_rlock(&tcp_function_lock); 542 543 cnt = 0; 544 #ifndef INVARIANTS 545 if (req->oldptr == NULL) { 546 cnt = tcp_fb_cnt; 547 goto skip_loop; 548 } 549 #endif 550 TAILQ_FOREACH(f, &t_functions, tf_next) { 551 #ifdef INVARIANTS 552 cnt++; 553 #endif 554 if (req->oldptr != NULL) { 555 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 556 tfi.tfi_id = f->tf_fb->tfb_id; 557 (void)strncpy(tfi.tfi_alias, f->tf_name, 558 TCP_FUNCTION_NAME_LEN_MAX); 559 tfi.tfi_alias[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 560 (void)strncpy(tfi.tfi_name, 561 f->tf_fb->tfb_tcp_block_name, 562 TCP_FUNCTION_NAME_LEN_MAX); 563 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 564 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 565 /* 566 * Don't stop on error, as that is the 567 * mechanism we use to accumulate length 568 * information if the buffer was too short. 569 */ 570 } 571 } 572 KASSERT(cnt == tcp_fb_cnt, 573 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 574 #ifndef INVARIANTS 575 skip_loop: 576 #endif 577 rw_runlock(&tcp_function_lock); 578 if (req->oldptr == NULL) 579 error = SYSCTL_OUT(req, NULL, 580 (cnt + 1) * sizeof(struct tcp_function_info)); 581 582 return (error); 583 } 584 585 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 586 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 587 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 588 "List TCP function block name-to-ID mappings"); 589 590 /* 591 * tfb_tcp_handoff_ok() function for the default stack. 592 * Note that we'll basically try to take all comers. 593 */ 594 static int 595 tcp_default_handoff_ok(struct tcpcb *tp) 596 { 597 598 return (0); 599 } 600 601 /* 602 * tfb_tcp_fb_init() function for the default stack. 603 * 604 * This handles making sure we have appropriate timers set if you are 605 * transitioning a socket that has some amount of setup done. 606 * 607 * The init() fuction from the default can *never* return non-zero i.e. 608 * it is required to always succeed since it is the stack of last resort! 609 */ 610 static int 611 tcp_default_fb_init(struct tcpcb *tp) 612 { 613 614 struct socket *so; 615 616 INP_WLOCK_ASSERT(tp->t_inpcb); 617 618 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 619 ("%s: connection %p in unexpected state %d", __func__, tp, 620 tp->t_state)); 621 622 /* 623 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 624 * know what to do for unexpected states (which includes TIME_WAIT). 625 */ 626 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 627 return (0); 628 629 /* 630 * Make sure some kind of transmission timer is set if there is 631 * outstanding data. 632 */ 633 so = tp->t_inpcb->inp_socket; 634 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 635 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 636 tcp_timer_active(tp, TT_PERSIST))) { 637 /* 638 * If the session has established and it looks like it should 639 * be in the persist state, set the persist timer. Otherwise, 640 * set the retransmit timer. 641 */ 642 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 643 (int32_t)(tp->snd_nxt - tp->snd_una) < 644 (int32_t)sbavail(&so->so_snd)) 645 tcp_setpersist(tp); 646 else 647 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 648 } 649 650 /* All non-embryonic sessions get a keepalive timer. */ 651 if (!tcp_timer_active(tp, TT_KEEP)) 652 tcp_timer_activate(tp, TT_KEEP, 653 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 654 TP_KEEPINIT(tp)); 655 656 return (0); 657 } 658 659 /* 660 * tfb_tcp_fb_fini() function for the default stack. 661 * 662 * This changes state as necessary (or prudent) to prepare for another stack 663 * to assume responsibility for the connection. 664 */ 665 static void 666 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 667 { 668 669 INP_WLOCK_ASSERT(tp->t_inpcb); 670 return; 671 } 672 673 /* 674 * Target size of TCP PCB hash tables. Must be a power of two. 675 * 676 * Note that this can be overridden by the kernel environment 677 * variable net.inet.tcp.tcbhashsize 678 */ 679 #ifndef TCBHASHSIZE 680 #define TCBHASHSIZE 0 681 #endif 682 683 /* 684 * XXX 685 * Callouts should be moved into struct tcp directly. They are currently 686 * separate because the tcpcb structure is exported to userland for sysctl 687 * parsing purposes, which do not know about callouts. 688 */ 689 struct tcpcb_mem { 690 struct tcpcb tcb; 691 struct tcp_timer tt; 692 struct cc_var ccv; 693 #ifdef TCP_HHOOK 694 struct osd osd; 695 #endif 696 }; 697 698 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 699 #define V_tcpcb_zone VNET(tcpcb_zone) 700 701 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 702 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 703 704 static struct mtx isn_mtx; 705 706 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 707 #define ISN_LOCK() mtx_lock(&isn_mtx) 708 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 709 710 /* 711 * TCP initialization. 712 */ 713 static void 714 tcp_zone_change(void *tag) 715 { 716 717 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 718 uma_zone_set_max(V_tcpcb_zone, maxsockets); 719 tcp_tw_zone_change(); 720 } 721 722 static int 723 tcp_inpcb_init(void *mem, int size, int flags) 724 { 725 struct inpcb *inp = mem; 726 727 INP_LOCK_INIT(inp, "inp", "tcpinp"); 728 return (0); 729 } 730 731 /* 732 * Take a value and get the next power of 2 that doesn't overflow. 733 * Used to size the tcp_inpcb hash buckets. 734 */ 735 static int 736 maketcp_hashsize(int size) 737 { 738 int hashsize; 739 740 /* 741 * auto tune. 742 * get the next power of 2 higher than maxsockets. 743 */ 744 hashsize = 1 << fls(size); 745 /* catch overflow, and just go one power of 2 smaller */ 746 if (hashsize < size) { 747 hashsize = 1 << (fls(size) - 1); 748 } 749 return (hashsize); 750 } 751 752 static volatile int next_tcp_stack_id = 1; 753 754 /* 755 * Register a TCP function block with the name provided in the names 756 * array. (Note that this function does NOT automatically register 757 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 758 * explicitly include blk->tfb_tcp_block_name in the list of names if 759 * you wish to register the stack with that name.) 760 * 761 * Either all name registrations will succeed or all will fail. If 762 * a name registration fails, the function will update the num_names 763 * argument to point to the array index of the name that encountered 764 * the failure. 765 * 766 * Returns 0 on success, or an error code on failure. 767 */ 768 int 769 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 770 const char *names[], int *num_names) 771 { 772 struct tcp_function *n; 773 struct tcp_function_set fs; 774 int error, i; 775 776 KASSERT(names != NULL && *num_names > 0, 777 ("%s: Called with 0-length name list", __func__)); 778 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 779 780 if (t_functions_inited == 0) { 781 init_tcp_functions(); 782 } 783 if ((blk->tfb_tcp_output == NULL) || 784 (blk->tfb_tcp_do_segment == NULL) || 785 (blk->tfb_tcp_ctloutput == NULL) || 786 (strlen(blk->tfb_tcp_block_name) == 0)) { 787 /* 788 * These functions are required and you 789 * need a name. 790 */ 791 *num_names = 0; 792 return (EINVAL); 793 } 794 if (blk->tfb_tcp_timer_stop_all || 795 blk->tfb_tcp_timer_activate || 796 blk->tfb_tcp_timer_active || 797 blk->tfb_tcp_timer_stop) { 798 /* 799 * If you define one timer function you 800 * must have them all. 801 */ 802 if ((blk->tfb_tcp_timer_stop_all == NULL) || 803 (blk->tfb_tcp_timer_activate == NULL) || 804 (blk->tfb_tcp_timer_active == NULL) || 805 (blk->tfb_tcp_timer_stop == NULL)) { 806 *num_names = 0; 807 return (EINVAL); 808 } 809 } 810 811 refcount_init(&blk->tfb_refcnt, 0); 812 blk->tfb_flags = 0; 813 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 814 for (i = 0; i < *num_names; i++) { 815 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 816 if (n == NULL) { 817 error = ENOMEM; 818 goto cleanup; 819 } 820 n->tf_fb = blk; 821 822 (void)strncpy(fs.function_set_name, names[i], 823 TCP_FUNCTION_NAME_LEN_MAX); 824 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 825 rw_wlock(&tcp_function_lock); 826 if (find_tcp_functions_locked(&fs) != NULL) { 827 /* Duplicate name space not allowed */ 828 rw_wunlock(&tcp_function_lock); 829 free(n, M_TCPFUNCTIONS); 830 error = EALREADY; 831 goto cleanup; 832 } 833 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX); 834 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 835 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 836 tcp_fb_cnt++; 837 rw_wunlock(&tcp_function_lock); 838 } 839 return(0); 840 841 cleanup: 842 /* 843 * Deregister the names we just added. Because registration failed 844 * for names[i], we don't need to deregister that name. 845 */ 846 *num_names = i; 847 rw_wlock(&tcp_function_lock); 848 while (--i >= 0) { 849 TAILQ_FOREACH(n, &t_functions, tf_next) { 850 if (!strncmp(n->tf_name, names[i], 851 TCP_FUNCTION_NAME_LEN_MAX)) { 852 TAILQ_REMOVE(&t_functions, n, tf_next); 853 tcp_fb_cnt--; 854 n->tf_fb = NULL; 855 free(n, M_TCPFUNCTIONS); 856 break; 857 } 858 } 859 } 860 rw_wunlock(&tcp_function_lock); 861 return (error); 862 } 863 864 /* 865 * Register a TCP function block using the name provided in the name 866 * argument. 867 * 868 * Returns 0 on success, or an error code on failure. 869 */ 870 int 871 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 872 int wait) 873 { 874 const char *name_list[1]; 875 int num_names, rv; 876 877 num_names = 1; 878 if (name != NULL) 879 name_list[0] = name; 880 else 881 name_list[0] = blk->tfb_tcp_block_name; 882 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 883 return (rv); 884 } 885 886 /* 887 * Register a TCP function block using the name defined in 888 * blk->tfb_tcp_block_name. 889 * 890 * Returns 0 on success, or an error code on failure. 891 */ 892 int 893 register_tcp_functions(struct tcp_function_block *blk, int wait) 894 { 895 896 return (register_tcp_functions_as_name(blk, NULL, wait)); 897 } 898 899 /* 900 * Deregister all names associated with a function block. This 901 * functionally removes the function block from use within the system. 902 * 903 * When called with a true quiesce argument, mark the function block 904 * as being removed so no more stacks will use it and determine 905 * whether the removal would succeed. 906 * 907 * When called with a false quiesce argument, actually attempt the 908 * removal. 909 * 910 * When called with a force argument, attempt to switch all TCBs to 911 * use the default stack instead of returning EBUSY. 912 * 913 * Returns 0 on success (or if the removal would succeed, or an error 914 * code on failure. 915 */ 916 int 917 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 918 bool force) 919 { 920 struct tcp_function *f; 921 922 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 923 /* You can't un-register the default */ 924 return (EPERM); 925 } 926 rw_wlock(&tcp_function_lock); 927 if (blk == tcp_func_set_ptr) { 928 /* You can't free the current default */ 929 rw_wunlock(&tcp_function_lock); 930 return (EBUSY); 931 } 932 /* Mark the block so no more stacks can use it. */ 933 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 934 /* 935 * If TCBs are still attached to the stack, attempt to switch them 936 * to the default stack. 937 */ 938 if (force && blk->tfb_refcnt) { 939 struct inpcb *inp; 940 struct tcpcb *tp; 941 VNET_ITERATOR_DECL(vnet_iter); 942 943 rw_wunlock(&tcp_function_lock); 944 945 VNET_LIST_RLOCK(); 946 /* XXX handle */ 947 VNET_FOREACH(vnet_iter) { 948 CURVNET_SET(vnet_iter); 949 INP_INFO_WLOCK(&V_tcbinfo); 950 LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 951 INP_WLOCK(inp); 952 if (inp->inp_flags & INP_TIMEWAIT) { 953 INP_WUNLOCK(inp); 954 continue; 955 } 956 tp = intotcpcb(inp); 957 if (tp == NULL || tp->t_fb != blk) { 958 INP_WUNLOCK(inp); 959 continue; 960 } 961 tcp_switch_back_to_default(tp); 962 INP_WUNLOCK(inp); 963 } 964 INP_INFO_WUNLOCK(&V_tcbinfo); 965 CURVNET_RESTORE(); 966 } 967 VNET_LIST_RUNLOCK(); 968 969 rw_wlock(&tcp_function_lock); 970 } 971 if (blk->tfb_refcnt) { 972 /* TCBs still attached. */ 973 rw_wunlock(&tcp_function_lock); 974 return (EBUSY); 975 } 976 if (quiesce) { 977 /* Skip removal. */ 978 rw_wunlock(&tcp_function_lock); 979 return (0); 980 } 981 /* Remove any function names that map to this function block. */ 982 while (find_tcp_fb_locked(blk, &f) != NULL) { 983 TAILQ_REMOVE(&t_functions, f, tf_next); 984 tcp_fb_cnt--; 985 f->tf_fb = NULL; 986 free(f, M_TCPFUNCTIONS); 987 } 988 rw_wunlock(&tcp_function_lock); 989 return (0); 990 } 991 992 void 993 tcp_init(void) 994 { 995 const char *tcbhash_tuneable; 996 int hashsize; 997 998 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 999 1000 #ifdef TCP_HHOOK 1001 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1002 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1003 printf("%s: WARNING: unable to register helper hook\n", __func__); 1004 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1005 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1006 printf("%s: WARNING: unable to register helper hook\n", __func__); 1007 #endif 1008 hashsize = TCBHASHSIZE; 1009 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1010 if (hashsize == 0) { 1011 /* 1012 * Auto tune the hash size based on maxsockets. 1013 * A perfect hash would have a 1:1 mapping 1014 * (hashsize = maxsockets) however it's been 1015 * suggested that O(2) average is better. 1016 */ 1017 hashsize = maketcp_hashsize(maxsockets / 4); 1018 /* 1019 * Our historical default is 512, 1020 * do not autotune lower than this. 1021 */ 1022 if (hashsize < 512) 1023 hashsize = 512; 1024 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1025 printf("%s: %s auto tuned to %d\n", __func__, 1026 tcbhash_tuneable, hashsize); 1027 } 1028 /* 1029 * We require a hashsize to be a power of two. 1030 * Previously if it was not a power of two we would just reset it 1031 * back to 512, which could be a nasty surprise if you did not notice 1032 * the error message. 1033 * Instead what we do is clip it to the closest power of two lower 1034 * than the specified hash value. 1035 */ 1036 if (!powerof2(hashsize)) { 1037 int oldhashsize = hashsize; 1038 1039 hashsize = maketcp_hashsize(hashsize); 1040 /* prevent absurdly low value */ 1041 if (hashsize < 16) 1042 hashsize = 16; 1043 printf("%s: WARNING: TCB hash size not a power of 2, " 1044 "clipped from %d to %d.\n", __func__, oldhashsize, 1045 hashsize); 1046 } 1047 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1048 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1049 1050 /* 1051 * These have to be type stable for the benefit of the timers. 1052 */ 1053 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1054 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1055 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1056 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1057 1058 tcp_tw_init(); 1059 syncache_init(); 1060 tcp_hc_init(); 1061 1062 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1063 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1064 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1065 1066 tcp_fastopen_init(); 1067 1068 /* Skip initialization of globals for non-default instances. */ 1069 if (!IS_DEFAULT_VNET(curvnet)) 1070 return; 1071 1072 tcp_reass_global_init(); 1073 1074 /* XXX virtualize those bellow? */ 1075 tcp_delacktime = TCPTV_DELACK; 1076 tcp_keepinit = TCPTV_KEEP_INIT; 1077 tcp_keepidle = TCPTV_KEEP_IDLE; 1078 tcp_keepintvl = TCPTV_KEEPINTVL; 1079 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1080 tcp_msl = TCPTV_MSL; 1081 tcp_rexmit_min = TCPTV_MIN; 1082 if (tcp_rexmit_min < 1) 1083 tcp_rexmit_min = 1; 1084 tcp_persmin = TCPTV_PERSMIN; 1085 tcp_persmax = TCPTV_PERSMAX; 1086 tcp_rexmit_slop = TCPTV_CPU_VAR; 1087 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1088 tcp_tcbhashsize = hashsize; 1089 /* Setup the tcp function block list */ 1090 init_tcp_functions(); 1091 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1092 #ifdef TCP_BLACKBOX 1093 /* Initialize the TCP logging data. */ 1094 tcp_log_init(); 1095 #endif 1096 1097 if (tcp_soreceive_stream) { 1098 #ifdef INET 1099 tcp_usrreqs.pru_soreceive = soreceive_stream; 1100 #endif 1101 #ifdef INET6 1102 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1103 #endif /* INET6 */ 1104 } 1105 1106 #ifdef INET6 1107 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1108 #else /* INET6 */ 1109 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1110 #endif /* INET6 */ 1111 if (max_protohdr < TCP_MINPROTOHDR) 1112 max_protohdr = TCP_MINPROTOHDR; 1113 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1114 panic("tcp_init"); 1115 #undef TCP_MINPROTOHDR 1116 1117 ISN_LOCK_INIT(); 1118 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1119 SHUTDOWN_PRI_DEFAULT); 1120 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1121 EVENTHANDLER_PRI_ANY); 1122 #ifdef TCPPCAP 1123 tcp_pcap_init(); 1124 #endif 1125 } 1126 1127 #ifdef VIMAGE 1128 static void 1129 tcp_destroy(void *unused __unused) 1130 { 1131 int n; 1132 #ifdef TCP_HHOOK 1133 int error; 1134 #endif 1135 1136 /* 1137 * All our processes are gone, all our sockets should be cleaned 1138 * up, which means, we should be past the tcp_discardcb() calls. 1139 * Sleep to let all tcpcb timers really disappear and cleanup. 1140 */ 1141 for (;;) { 1142 INP_LIST_RLOCK(&V_tcbinfo); 1143 n = V_tcbinfo.ipi_count; 1144 INP_LIST_RUNLOCK(&V_tcbinfo); 1145 if (n == 0) 1146 break; 1147 pause("tcpdes", hz / 10); 1148 } 1149 tcp_hc_destroy(); 1150 syncache_destroy(); 1151 tcp_tw_destroy(); 1152 in_pcbinfo_destroy(&V_tcbinfo); 1153 /* tcp_discardcb() clears the sack_holes up. */ 1154 uma_zdestroy(V_sack_hole_zone); 1155 uma_zdestroy(V_tcpcb_zone); 1156 1157 /* 1158 * Cannot free the zone until all tcpcbs are released as we attach 1159 * the allocations to them. 1160 */ 1161 tcp_fastopen_destroy(); 1162 1163 #ifdef TCP_HHOOK 1164 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1165 if (error != 0) { 1166 printf("%s: WARNING: unable to deregister helper hook " 1167 "type=%d, id=%d: error %d returned\n", __func__, 1168 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1169 } 1170 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1171 if (error != 0) { 1172 printf("%s: WARNING: unable to deregister helper hook " 1173 "type=%d, id=%d: error %d returned\n", __func__, 1174 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1175 } 1176 #endif 1177 } 1178 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1179 #endif 1180 1181 void 1182 tcp_fini(void *xtp) 1183 { 1184 1185 } 1186 1187 /* 1188 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1189 * tcp_template used to store this data in mbufs, but we now recopy it out 1190 * of the tcpcb each time to conserve mbufs. 1191 */ 1192 void 1193 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1194 { 1195 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1196 1197 INP_WLOCK_ASSERT(inp); 1198 1199 #ifdef INET6 1200 if ((inp->inp_vflag & INP_IPV6) != 0) { 1201 struct ip6_hdr *ip6; 1202 1203 ip6 = (struct ip6_hdr *)ip_ptr; 1204 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1205 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1206 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1207 (IPV6_VERSION & IPV6_VERSION_MASK); 1208 ip6->ip6_nxt = IPPROTO_TCP; 1209 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1210 ip6->ip6_src = inp->in6p_laddr; 1211 ip6->ip6_dst = inp->in6p_faddr; 1212 } 1213 #endif /* INET6 */ 1214 #if defined(INET6) && defined(INET) 1215 else 1216 #endif 1217 #ifdef INET 1218 { 1219 struct ip *ip; 1220 1221 ip = (struct ip *)ip_ptr; 1222 ip->ip_v = IPVERSION; 1223 ip->ip_hl = 5; 1224 ip->ip_tos = inp->inp_ip_tos; 1225 ip->ip_len = 0; 1226 ip->ip_id = 0; 1227 ip->ip_off = 0; 1228 ip->ip_ttl = inp->inp_ip_ttl; 1229 ip->ip_sum = 0; 1230 ip->ip_p = IPPROTO_TCP; 1231 ip->ip_src = inp->inp_laddr; 1232 ip->ip_dst = inp->inp_faddr; 1233 } 1234 #endif /* INET */ 1235 th->th_sport = inp->inp_lport; 1236 th->th_dport = inp->inp_fport; 1237 th->th_seq = 0; 1238 th->th_ack = 0; 1239 th->th_x2 = 0; 1240 th->th_off = 5; 1241 th->th_flags = 0; 1242 th->th_win = 0; 1243 th->th_urp = 0; 1244 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1245 } 1246 1247 /* 1248 * Create template to be used to send tcp packets on a connection. 1249 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1250 * use for this function is in keepalives, which use tcp_respond. 1251 */ 1252 struct tcptemp * 1253 tcpip_maketemplate(struct inpcb *inp) 1254 { 1255 struct tcptemp *t; 1256 1257 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1258 if (t == NULL) 1259 return (NULL); 1260 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1261 return (t); 1262 } 1263 1264 /* 1265 * Send a single message to the TCP at address specified by 1266 * the given TCP/IP header. If m == NULL, then we make a copy 1267 * of the tcpiphdr at th and send directly to the addressed host. 1268 * This is used to force keep alive messages out using the TCP 1269 * template for a connection. If flags are given then we send 1270 * a message back to the TCP which originated the segment th, 1271 * and discard the mbuf containing it and any other attached mbufs. 1272 * 1273 * In any case the ack and sequence number of the transmitted 1274 * segment are as specified by the parameters. 1275 * 1276 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1277 */ 1278 void 1279 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1280 tcp_seq ack, tcp_seq seq, int flags) 1281 { 1282 struct tcpopt to; 1283 struct inpcb *inp; 1284 struct ip *ip; 1285 struct mbuf *optm; 1286 struct tcphdr *nth; 1287 u_char *optp; 1288 #ifdef INET6 1289 struct ip6_hdr *ip6; 1290 int isipv6; 1291 #endif /* INET6 */ 1292 int optlen, tlen, win; 1293 bool incl_opts; 1294 1295 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1296 1297 #ifdef INET6 1298 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1299 ip6 = ipgen; 1300 #endif /* INET6 */ 1301 ip = ipgen; 1302 1303 if (tp != NULL) { 1304 inp = tp->t_inpcb; 1305 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1306 INP_WLOCK_ASSERT(inp); 1307 } else 1308 inp = NULL; 1309 1310 incl_opts = false; 1311 win = 0; 1312 if (tp != NULL) { 1313 if (!(flags & TH_RST)) { 1314 win = sbspace(&inp->inp_socket->so_rcv); 1315 if (win > TCP_MAXWIN << tp->rcv_scale) 1316 win = TCP_MAXWIN << tp->rcv_scale; 1317 } 1318 if ((tp->t_flags & TF_NOOPT) == 0) 1319 incl_opts = true; 1320 } 1321 if (m == NULL) { 1322 m = m_gethdr(M_NOWAIT, MT_DATA); 1323 if (m == NULL) 1324 return; 1325 m->m_data += max_linkhdr; 1326 #ifdef INET6 1327 if (isipv6) { 1328 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1329 sizeof(struct ip6_hdr)); 1330 ip6 = mtod(m, struct ip6_hdr *); 1331 nth = (struct tcphdr *)(ip6 + 1); 1332 } else 1333 #endif /* INET6 */ 1334 { 1335 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1336 ip = mtod(m, struct ip *); 1337 nth = (struct tcphdr *)(ip + 1); 1338 } 1339 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1340 flags = TH_ACK; 1341 } else if (!M_WRITABLE(m)) { 1342 struct mbuf *n; 1343 1344 /* Can't reuse 'm', allocate a new mbuf. */ 1345 n = m_gethdr(M_NOWAIT, MT_DATA); 1346 if (n == NULL) { 1347 m_freem(m); 1348 return; 1349 } 1350 1351 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1352 m_freem(m); 1353 m_freem(n); 1354 return; 1355 } 1356 1357 n->m_data += max_linkhdr; 1358 /* m_len is set later */ 1359 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1360 #ifdef INET6 1361 if (isipv6) { 1362 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1363 sizeof(struct ip6_hdr)); 1364 ip6 = mtod(n, struct ip6_hdr *); 1365 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1366 nth = (struct tcphdr *)(ip6 + 1); 1367 } else 1368 #endif /* INET6 */ 1369 { 1370 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1371 ip = mtod(n, struct ip *); 1372 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1373 nth = (struct tcphdr *)(ip + 1); 1374 } 1375 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1376 xchg(nth->th_dport, nth->th_sport, uint16_t); 1377 th = nth; 1378 m_freem(m); 1379 m = n; 1380 } else { 1381 /* 1382 * reuse the mbuf. 1383 * XXX MRT We inherit the FIB, which is lucky. 1384 */ 1385 m_freem(m->m_next); 1386 m->m_next = NULL; 1387 m->m_data = (caddr_t)ipgen; 1388 /* m_len is set later */ 1389 #ifdef INET6 1390 if (isipv6) { 1391 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1392 nth = (struct tcphdr *)(ip6 + 1); 1393 } else 1394 #endif /* INET6 */ 1395 { 1396 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1397 nth = (struct tcphdr *)(ip + 1); 1398 } 1399 if (th != nth) { 1400 /* 1401 * this is usually a case when an extension header 1402 * exists between the IPv6 header and the 1403 * TCP header. 1404 */ 1405 nth->th_sport = th->th_sport; 1406 nth->th_dport = th->th_dport; 1407 } 1408 xchg(nth->th_dport, nth->th_sport, uint16_t); 1409 #undef xchg 1410 } 1411 tlen = 0; 1412 #ifdef INET6 1413 if (isipv6) 1414 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1415 #endif 1416 #if defined(INET) && defined(INET6) 1417 else 1418 #endif 1419 #ifdef INET 1420 tlen = sizeof (struct tcpiphdr); 1421 #endif 1422 #ifdef INVARIANTS 1423 m->m_len = 0; 1424 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1425 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1426 m, tlen, (long)M_TRAILINGSPACE(m))); 1427 #endif 1428 m->m_len = tlen; 1429 to.to_flags = 0; 1430 if (incl_opts) { 1431 /* Make sure we have room. */ 1432 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1433 m->m_next = m_get(M_NOWAIT, MT_DATA); 1434 if (m->m_next) { 1435 optp = mtod(m->m_next, u_char *); 1436 optm = m->m_next; 1437 } else 1438 incl_opts = false; 1439 } else { 1440 optp = (u_char *) (nth + 1); 1441 optm = m; 1442 } 1443 } 1444 if (incl_opts) { 1445 /* Timestamps. */ 1446 if (tp->t_flags & TF_RCVD_TSTMP) { 1447 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1448 to.to_tsecr = tp->ts_recent; 1449 to.to_flags |= TOF_TS; 1450 } 1451 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1452 /* TCP-MD5 (RFC2385). */ 1453 if (tp->t_flags & TF_SIGNATURE) 1454 to.to_flags |= TOF_SIGNATURE; 1455 #endif 1456 /* Add the options. */ 1457 tlen += optlen = tcp_addoptions(&to, optp); 1458 1459 /* Update m_len in the correct mbuf. */ 1460 optm->m_len += optlen; 1461 } else 1462 optlen = 0; 1463 #ifdef INET6 1464 if (isipv6) { 1465 ip6->ip6_flow = 0; 1466 ip6->ip6_vfc = IPV6_VERSION; 1467 ip6->ip6_nxt = IPPROTO_TCP; 1468 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1469 } 1470 #endif 1471 #if defined(INET) && defined(INET6) 1472 else 1473 #endif 1474 #ifdef INET 1475 { 1476 ip->ip_len = htons(tlen); 1477 ip->ip_ttl = V_ip_defttl; 1478 if (V_path_mtu_discovery) 1479 ip->ip_off |= htons(IP_DF); 1480 } 1481 #endif 1482 m->m_pkthdr.len = tlen; 1483 m->m_pkthdr.rcvif = NULL; 1484 #ifdef MAC 1485 if (inp != NULL) { 1486 /* 1487 * Packet is associated with a socket, so allow the 1488 * label of the response to reflect the socket label. 1489 */ 1490 INP_WLOCK_ASSERT(inp); 1491 mac_inpcb_create_mbuf(inp, m); 1492 } else { 1493 /* 1494 * Packet is not associated with a socket, so possibly 1495 * update the label in place. 1496 */ 1497 mac_netinet_tcp_reply(m); 1498 } 1499 #endif 1500 nth->th_seq = htonl(seq); 1501 nth->th_ack = htonl(ack); 1502 nth->th_x2 = 0; 1503 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1504 nth->th_flags = flags; 1505 if (tp != NULL) 1506 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1507 else 1508 nth->th_win = htons((u_short)win); 1509 nth->th_urp = 0; 1510 1511 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1512 if (to.to_flags & TOF_SIGNATURE) { 1513 if (!TCPMD5_ENABLED() || 1514 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1515 m_freem(m); 1516 return; 1517 } 1518 } 1519 #endif 1520 1521 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1522 #ifdef INET6 1523 if (isipv6) { 1524 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1525 nth->th_sum = in6_cksum_pseudo(ip6, 1526 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1527 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1528 NULL, NULL); 1529 } 1530 #endif /* INET6 */ 1531 #if defined(INET6) && defined(INET) 1532 else 1533 #endif 1534 #ifdef INET 1535 { 1536 m->m_pkthdr.csum_flags = CSUM_TCP; 1537 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1538 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1539 } 1540 #endif /* INET */ 1541 #ifdef TCPDEBUG 1542 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1543 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1544 #endif 1545 TCP_PROBE3(debug__output, tp, th, m); 1546 if (flags & TH_RST) 1547 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1548 1549 #ifdef INET6 1550 if (isipv6) { 1551 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1552 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1553 } 1554 #endif /* INET6 */ 1555 #if defined(INET) && defined(INET6) 1556 else 1557 #endif 1558 #ifdef INET 1559 { 1560 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1561 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1562 } 1563 #endif 1564 } 1565 1566 /* 1567 * Create a new TCP control block, making an 1568 * empty reassembly queue and hooking it to the argument 1569 * protocol control block. The `inp' parameter must have 1570 * come from the zone allocator set up in tcp_init(). 1571 */ 1572 struct tcpcb * 1573 tcp_newtcpcb(struct inpcb *inp) 1574 { 1575 struct tcpcb_mem *tm; 1576 struct tcpcb *tp; 1577 #ifdef INET6 1578 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1579 #endif /* INET6 */ 1580 1581 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1582 if (tm == NULL) 1583 return (NULL); 1584 tp = &tm->tcb; 1585 1586 /* Initialise cc_var struct for this tcpcb. */ 1587 tp->ccv = &tm->ccv; 1588 tp->ccv->type = IPPROTO_TCP; 1589 tp->ccv->ccvc.tcp = tp; 1590 rw_rlock(&tcp_function_lock); 1591 tp->t_fb = tcp_func_set_ptr; 1592 refcount_acquire(&tp->t_fb->tfb_refcnt); 1593 rw_runlock(&tcp_function_lock); 1594 /* 1595 * Use the current system default CC algorithm. 1596 */ 1597 CC_LIST_RLOCK(); 1598 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1599 CC_ALGO(tp) = CC_DEFAULT(); 1600 CC_LIST_RUNLOCK(); 1601 1602 if (CC_ALGO(tp)->cb_init != NULL) 1603 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1604 if (tp->t_fb->tfb_tcp_fb_fini) 1605 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1606 refcount_release(&tp->t_fb->tfb_refcnt); 1607 uma_zfree(V_tcpcb_zone, tm); 1608 return (NULL); 1609 } 1610 1611 #ifdef TCP_HHOOK 1612 tp->osd = &tm->osd; 1613 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1614 if (tp->t_fb->tfb_tcp_fb_fini) 1615 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1616 refcount_release(&tp->t_fb->tfb_refcnt); 1617 uma_zfree(V_tcpcb_zone, tm); 1618 return (NULL); 1619 } 1620 #endif 1621 1622 #ifdef VIMAGE 1623 tp->t_vnet = inp->inp_vnet; 1624 #endif 1625 tp->t_timers = &tm->tt; 1626 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1627 tp->t_maxseg = 1628 #ifdef INET6 1629 isipv6 ? V_tcp_v6mssdflt : 1630 #endif /* INET6 */ 1631 V_tcp_mssdflt; 1632 1633 /* Set up our timeouts. */ 1634 callout_init(&tp->t_timers->tt_rexmt, 1); 1635 callout_init(&tp->t_timers->tt_persist, 1); 1636 callout_init(&tp->t_timers->tt_keep, 1); 1637 callout_init(&tp->t_timers->tt_2msl, 1); 1638 callout_init(&tp->t_timers->tt_delack, 1); 1639 1640 if (V_tcp_do_rfc1323) 1641 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1642 if (V_tcp_do_sack) 1643 tp->t_flags |= TF_SACK_PERMIT; 1644 TAILQ_INIT(&tp->snd_holes); 1645 /* 1646 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1647 * is called. 1648 */ 1649 in_pcbref(inp); /* Reference for tcpcb */ 1650 tp->t_inpcb = inp; 1651 1652 /* 1653 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1654 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1655 * reasonable initial retransmit time. 1656 */ 1657 tp->t_srtt = TCPTV_SRTTBASE; 1658 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1659 tp->t_rttmin = tcp_rexmit_min; 1660 tp->t_rxtcur = TCPTV_RTOBASE; 1661 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1662 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1663 tp->t_rcvtime = ticks; 1664 /* 1665 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1666 * because the socket may be bound to an IPv6 wildcard address, 1667 * which may match an IPv4-mapped IPv6 address. 1668 */ 1669 inp->inp_ip_ttl = V_ip_defttl; 1670 inp->inp_ppcb = tp; 1671 #ifdef TCPPCAP 1672 /* 1673 * Init the TCP PCAP queues. 1674 */ 1675 tcp_pcap_tcpcb_init(tp); 1676 #endif 1677 #ifdef TCP_BLACKBOX 1678 /* Initialize the per-TCPCB log data. */ 1679 tcp_log_tcpcbinit(tp); 1680 #endif 1681 if (tp->t_fb->tfb_tcp_fb_init) { 1682 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1683 } 1684 return (tp); /* XXX */ 1685 } 1686 1687 /* 1688 * Switch the congestion control algorithm back to NewReno for any active 1689 * control blocks using an algorithm which is about to go away. 1690 * This ensures the CC framework can allow the unload to proceed without leaving 1691 * any dangling pointers which would trigger a panic. 1692 * Returning non-zero would inform the CC framework that something went wrong 1693 * and it would be unsafe to allow the unload to proceed. However, there is no 1694 * way for this to occur with this implementation so we always return zero. 1695 */ 1696 int 1697 tcp_ccalgounload(struct cc_algo *unload_algo) 1698 { 1699 struct cc_algo *tmpalgo; 1700 struct inpcb *inp; 1701 struct tcpcb *tp; 1702 VNET_ITERATOR_DECL(vnet_iter); 1703 1704 /* 1705 * Check all active control blocks across all network stacks and change 1706 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1707 * requires cleanup code to be run, call it. 1708 */ 1709 VNET_LIST_RLOCK(); 1710 VNET_FOREACH(vnet_iter) { 1711 CURVNET_SET(vnet_iter); 1712 INP_INFO_WLOCK(&V_tcbinfo); 1713 /* 1714 * New connections already part way through being initialised 1715 * with the CC algo we're removing will not race with this code 1716 * because the INP_INFO_WLOCK is held during initialisation. We 1717 * therefore don't enter the loop below until the connection 1718 * list has stabilised. 1719 */ 1720 LIST_FOREACH(inp, &V_tcb, inp_list) { 1721 INP_WLOCK(inp); 1722 /* Important to skip tcptw structs. */ 1723 if (!(inp->inp_flags & INP_TIMEWAIT) && 1724 (tp = intotcpcb(inp)) != NULL) { 1725 /* 1726 * By holding INP_WLOCK here, we are assured 1727 * that the connection is not currently 1728 * executing inside the CC module's functions 1729 * i.e. it is safe to make the switch back to 1730 * NewReno. 1731 */ 1732 if (CC_ALGO(tp) == unload_algo) { 1733 tmpalgo = CC_ALGO(tp); 1734 /* NewReno does not require any init. */ 1735 CC_ALGO(tp) = &newreno_cc_algo; 1736 /* XXX defer to epoch_call */ 1737 if (tmpalgo->cb_destroy != NULL) 1738 tmpalgo->cb_destroy(tp->ccv); 1739 } 1740 } 1741 INP_WUNLOCK(inp); 1742 } 1743 INP_INFO_WUNLOCK(&V_tcbinfo); 1744 CURVNET_RESTORE(); 1745 } 1746 VNET_LIST_RUNLOCK(); 1747 1748 return (0); 1749 } 1750 1751 /* 1752 * Drop a TCP connection, reporting 1753 * the specified error. If connection is synchronized, 1754 * then send a RST to peer. 1755 */ 1756 struct tcpcb * 1757 tcp_drop(struct tcpcb *tp, int errno) 1758 { 1759 struct socket *so = tp->t_inpcb->inp_socket; 1760 1761 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1762 INP_WLOCK_ASSERT(tp->t_inpcb); 1763 1764 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1765 tcp_state_change(tp, TCPS_CLOSED); 1766 (void) tp->t_fb->tfb_tcp_output(tp); 1767 TCPSTAT_INC(tcps_drops); 1768 } else 1769 TCPSTAT_INC(tcps_conndrops); 1770 if (errno == ETIMEDOUT && tp->t_softerror) 1771 errno = tp->t_softerror; 1772 so->so_error = errno; 1773 return (tcp_close(tp)); 1774 } 1775 1776 void 1777 tcp_discardcb(struct tcpcb *tp) 1778 { 1779 struct inpcb *inp = tp->t_inpcb; 1780 struct socket *so = inp->inp_socket; 1781 #ifdef INET6 1782 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1783 #endif /* INET6 */ 1784 int released __unused; 1785 1786 INP_WLOCK_ASSERT(inp); 1787 1788 /* 1789 * Make sure that all of our timers are stopped before we delete the 1790 * PCB. 1791 * 1792 * If stopping a timer fails, we schedule a discard function in same 1793 * callout, and the last discard function called will take care of 1794 * deleting the tcpcb. 1795 */ 1796 tp->t_timers->tt_draincnt = 0; 1797 tcp_timer_stop(tp, TT_REXMT); 1798 tcp_timer_stop(tp, TT_PERSIST); 1799 tcp_timer_stop(tp, TT_KEEP); 1800 tcp_timer_stop(tp, TT_2MSL); 1801 tcp_timer_stop(tp, TT_DELACK); 1802 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1803 /* 1804 * Call the stop-all function of the methods, 1805 * this function should call the tcp_timer_stop() 1806 * method with each of the function specific timeouts. 1807 * That stop will be called via the tfb_tcp_timer_stop() 1808 * which should use the async drain function of the 1809 * callout system (see tcp_var.h). 1810 */ 1811 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1812 } 1813 1814 /* 1815 * If we got enough samples through the srtt filter, 1816 * save the rtt and rttvar in the routing entry. 1817 * 'Enough' is arbitrarily defined as 4 rtt samples. 1818 * 4 samples is enough for the srtt filter to converge 1819 * to within enough % of the correct value; fewer samples 1820 * and we could save a bogus rtt. The danger is not high 1821 * as tcp quickly recovers from everything. 1822 * XXX: Works very well but needs some more statistics! 1823 */ 1824 if (tp->t_rttupdated >= 4) { 1825 struct hc_metrics_lite metrics; 1826 uint32_t ssthresh; 1827 1828 bzero(&metrics, sizeof(metrics)); 1829 /* 1830 * Update the ssthresh always when the conditions below 1831 * are satisfied. This gives us better new start value 1832 * for the congestion avoidance for new connections. 1833 * ssthresh is only set if packet loss occurred on a session. 1834 * 1835 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1836 * being torn down. Ideally this code would not use 'so'. 1837 */ 1838 ssthresh = tp->snd_ssthresh; 1839 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1840 /* 1841 * convert the limit from user data bytes to 1842 * packets then to packet data bytes. 1843 */ 1844 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1845 if (ssthresh < 2) 1846 ssthresh = 2; 1847 ssthresh *= (tp->t_maxseg + 1848 #ifdef INET6 1849 (isipv6 ? sizeof (struct ip6_hdr) + 1850 sizeof (struct tcphdr) : 1851 #endif 1852 sizeof (struct tcpiphdr) 1853 #ifdef INET6 1854 ) 1855 #endif 1856 ); 1857 } else 1858 ssthresh = 0; 1859 metrics.rmx_ssthresh = ssthresh; 1860 1861 metrics.rmx_rtt = tp->t_srtt; 1862 metrics.rmx_rttvar = tp->t_rttvar; 1863 metrics.rmx_cwnd = tp->snd_cwnd; 1864 metrics.rmx_sendpipe = 0; 1865 metrics.rmx_recvpipe = 0; 1866 1867 tcp_hc_update(&inp->inp_inc, &metrics); 1868 } 1869 1870 /* free the reassembly queue, if any */ 1871 tcp_reass_flush(tp); 1872 1873 #ifdef TCP_OFFLOAD 1874 /* Disconnect offload device, if any. */ 1875 if (tp->t_flags & TF_TOE) 1876 tcp_offload_detach(tp); 1877 #endif 1878 1879 tcp_free_sackholes(tp); 1880 1881 #ifdef TCPPCAP 1882 /* Free the TCP PCAP queues. */ 1883 tcp_pcap_drain(&(tp->t_inpkts)); 1884 tcp_pcap_drain(&(tp->t_outpkts)); 1885 #endif 1886 1887 /* Allow the CC algorithm to clean up after itself. */ 1888 if (CC_ALGO(tp)->cb_destroy != NULL) 1889 CC_ALGO(tp)->cb_destroy(tp->ccv); 1890 1891 #ifdef TCP_HHOOK 1892 khelp_destroy_osd(tp->osd); 1893 #endif 1894 1895 CC_ALGO(tp) = NULL; 1896 inp->inp_ppcb = NULL; 1897 if (tp->t_timers->tt_draincnt == 0) { 1898 /* We own the last reference on tcpcb, let's free it. */ 1899 #ifdef TCP_BLACKBOX 1900 tcp_log_tcpcbfini(tp); 1901 #endif 1902 TCPSTATES_DEC(tp->t_state); 1903 if (tp->t_fb->tfb_tcp_fb_fini) 1904 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1905 refcount_release(&tp->t_fb->tfb_refcnt); 1906 tp->t_inpcb = NULL; 1907 uma_zfree(V_tcpcb_zone, tp); 1908 released = in_pcbrele_wlocked(inp); 1909 KASSERT(!released, ("%s: inp %p should not have been released " 1910 "here", __func__, inp)); 1911 } 1912 } 1913 1914 void 1915 tcp_timer_discard(void *ptp) 1916 { 1917 struct inpcb *inp; 1918 struct tcpcb *tp; 1919 1920 tp = (struct tcpcb *)ptp; 1921 CURVNET_SET(tp->t_vnet); 1922 INP_INFO_RLOCK(&V_tcbinfo); 1923 inp = tp->t_inpcb; 1924 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1925 __func__, tp)); 1926 INP_WLOCK(inp); 1927 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1928 ("%s: tcpcb has to be stopped here", __func__)); 1929 tp->t_timers->tt_draincnt--; 1930 if (tp->t_timers->tt_draincnt == 0) { 1931 /* We own the last reference on this tcpcb, let's free it. */ 1932 #ifdef TCP_BLACKBOX 1933 tcp_log_tcpcbfini(tp); 1934 #endif 1935 TCPSTATES_DEC(tp->t_state); 1936 if (tp->t_fb->tfb_tcp_fb_fini) 1937 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1938 refcount_release(&tp->t_fb->tfb_refcnt); 1939 tp->t_inpcb = NULL; 1940 uma_zfree(V_tcpcb_zone, tp); 1941 if (in_pcbrele_wlocked(inp)) { 1942 INP_INFO_RUNLOCK(&V_tcbinfo); 1943 CURVNET_RESTORE(); 1944 return; 1945 } 1946 } 1947 INP_WUNLOCK(inp); 1948 INP_INFO_RUNLOCK(&V_tcbinfo); 1949 CURVNET_RESTORE(); 1950 } 1951 1952 /* 1953 * Attempt to close a TCP control block, marking it as dropped, and freeing 1954 * the socket if we hold the only reference. 1955 */ 1956 struct tcpcb * 1957 tcp_close(struct tcpcb *tp) 1958 { 1959 struct inpcb *inp = tp->t_inpcb; 1960 struct socket *so; 1961 1962 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1963 INP_WLOCK_ASSERT(inp); 1964 1965 #ifdef TCP_OFFLOAD 1966 if (tp->t_state == TCPS_LISTEN) 1967 tcp_offload_listen_stop(tp); 1968 #endif 1969 /* 1970 * This releases the TFO pending counter resource for TFO listen 1971 * sockets as well as passively-created TFO sockets that transition 1972 * from SYN_RECEIVED to CLOSED. 1973 */ 1974 if (tp->t_tfo_pending) { 1975 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1976 tp->t_tfo_pending = NULL; 1977 } 1978 in_pcbdrop(inp); 1979 TCPSTAT_INC(tcps_closed); 1980 if (tp->t_state != TCPS_CLOSED) 1981 tcp_state_change(tp, TCPS_CLOSED); 1982 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1983 so = inp->inp_socket; 1984 soisdisconnected(so); 1985 if (inp->inp_flags & INP_SOCKREF) { 1986 KASSERT(so->so_state & SS_PROTOREF, 1987 ("tcp_close: !SS_PROTOREF")); 1988 inp->inp_flags &= ~INP_SOCKREF; 1989 INP_WUNLOCK(inp); 1990 SOCK_LOCK(so); 1991 so->so_state &= ~SS_PROTOREF; 1992 sofree(so); 1993 return (NULL); 1994 } 1995 return (tp); 1996 } 1997 1998 void 1999 tcp_drain(void) 2000 { 2001 VNET_ITERATOR_DECL(vnet_iter); 2002 2003 if (!do_tcpdrain) 2004 return; 2005 2006 VNET_LIST_RLOCK_NOSLEEP(); 2007 VNET_FOREACH(vnet_iter) { 2008 CURVNET_SET(vnet_iter); 2009 struct inpcb *inpb; 2010 struct tcpcb *tcpb; 2011 2012 /* 2013 * Walk the tcpbs, if existing, and flush the reassembly queue, 2014 * if there is one... 2015 * XXX: The "Net/3" implementation doesn't imply that the TCP 2016 * reassembly queue should be flushed, but in a situation 2017 * where we're really low on mbufs, this is potentially 2018 * useful. 2019 */ 2020 INP_INFO_WLOCK(&V_tcbinfo); 2021 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2022 if (inpb->inp_flags & INP_TIMEWAIT) 2023 continue; 2024 INP_WLOCK(inpb); 2025 if ((tcpb = intotcpcb(inpb)) != NULL) { 2026 tcp_reass_flush(tcpb); 2027 tcp_clean_sackreport(tcpb); 2028 #ifdef TCP_BLACKBOX 2029 tcp_log_drain(tcpb); 2030 #endif 2031 #ifdef TCPPCAP 2032 if (tcp_pcap_aggressive_free) { 2033 /* Free the TCP PCAP queues. */ 2034 tcp_pcap_drain(&(tcpb->t_inpkts)); 2035 tcp_pcap_drain(&(tcpb->t_outpkts)); 2036 } 2037 #endif 2038 } 2039 INP_WUNLOCK(inpb); 2040 } 2041 INP_INFO_WUNLOCK(&V_tcbinfo); 2042 CURVNET_RESTORE(); 2043 } 2044 VNET_LIST_RUNLOCK_NOSLEEP(); 2045 } 2046 2047 /* 2048 * Notify a tcp user of an asynchronous error; 2049 * store error as soft error, but wake up user 2050 * (for now, won't do anything until can select for soft error). 2051 * 2052 * Do not wake up user since there currently is no mechanism for 2053 * reporting soft errors (yet - a kqueue filter may be added). 2054 */ 2055 static struct inpcb * 2056 tcp_notify(struct inpcb *inp, int error) 2057 { 2058 struct tcpcb *tp; 2059 2060 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2061 INP_WLOCK_ASSERT(inp); 2062 2063 if ((inp->inp_flags & INP_TIMEWAIT) || 2064 (inp->inp_flags & INP_DROPPED)) 2065 return (inp); 2066 2067 tp = intotcpcb(inp); 2068 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2069 2070 /* 2071 * Ignore some errors if we are hooked up. 2072 * If connection hasn't completed, has retransmitted several times, 2073 * and receives a second error, give up now. This is better 2074 * than waiting a long time to establish a connection that 2075 * can never complete. 2076 */ 2077 if (tp->t_state == TCPS_ESTABLISHED && 2078 (error == EHOSTUNREACH || error == ENETUNREACH || 2079 error == EHOSTDOWN)) { 2080 if (inp->inp_route.ro_rt) { 2081 RTFREE(inp->inp_route.ro_rt); 2082 inp->inp_route.ro_rt = (struct rtentry *)NULL; 2083 } 2084 return (inp); 2085 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2086 tp->t_softerror) { 2087 tp = tcp_drop(tp, error); 2088 if (tp != NULL) 2089 return (inp); 2090 else 2091 return (NULL); 2092 } else { 2093 tp->t_softerror = error; 2094 return (inp); 2095 } 2096 #if 0 2097 wakeup( &so->so_timeo); 2098 sorwakeup(so); 2099 sowwakeup(so); 2100 #endif 2101 } 2102 2103 static int 2104 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2105 { 2106 int error, i, m, n, pcb_count; 2107 struct in_pcblist *il; 2108 struct inpcb *inp, **inp_list; 2109 inp_gen_t gencnt; 2110 struct xinpgen xig; 2111 2112 /* 2113 * The process of preparing the TCB list is too time-consuming and 2114 * resource-intensive to repeat twice on every request. 2115 */ 2116 if (req->oldptr == NULL) { 2117 n = V_tcbinfo.ipi_count + 2118 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2119 n += imax(n / 8, 10); 2120 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2121 return (0); 2122 } 2123 2124 if (req->newptr != NULL) 2125 return (EPERM); 2126 2127 /* 2128 * OK, now we're committed to doing something. 2129 */ 2130 INP_LIST_RLOCK(&V_tcbinfo); 2131 gencnt = V_tcbinfo.ipi_gencnt; 2132 n = V_tcbinfo.ipi_count; 2133 INP_LIST_RUNLOCK(&V_tcbinfo); 2134 2135 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2136 2137 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 2138 + (n + m) * sizeof(struct xtcpcb)); 2139 if (error != 0) 2140 return (error); 2141 2142 xig.xig_len = sizeof xig; 2143 xig.xig_count = n + m; 2144 xig.xig_gen = gencnt; 2145 xig.xig_sogen = so_gencnt; 2146 error = SYSCTL_OUT(req, &xig, sizeof xig); 2147 if (error) 2148 return (error); 2149 2150 error = syncache_pcblist(req, m, &pcb_count); 2151 if (error) 2152 return (error); 2153 2154 il = malloc(sizeof(struct in_pcblist) + n * sizeof(struct inpcb *), M_TEMP, M_WAITOK|M_ZERO_INVARIANTS); 2155 inp_list = il->il_inp_list; 2156 2157 INP_INFO_WLOCK(&V_tcbinfo); 2158 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 2159 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 2160 INP_WLOCK(inp); 2161 if (inp->inp_gencnt <= gencnt) { 2162 /* 2163 * XXX: This use of cr_cansee(), introduced with 2164 * TCP state changes, is not quite right, but for 2165 * now, better than nothing. 2166 */ 2167 if (inp->inp_flags & INP_TIMEWAIT) { 2168 if (intotw(inp) != NULL) 2169 error = cr_cansee(req->td->td_ucred, 2170 intotw(inp)->tw_cred); 2171 else 2172 error = EINVAL; /* Skip this inp. */ 2173 } else 2174 error = cr_canseeinpcb(req->td->td_ucred, inp); 2175 if (error == 0) { 2176 in_pcbref(inp); 2177 inp_list[i++] = inp; 2178 } 2179 } 2180 INP_WUNLOCK(inp); 2181 } 2182 INP_INFO_WUNLOCK(&V_tcbinfo); 2183 n = i; 2184 2185 error = 0; 2186 for (i = 0; i < n; i++) { 2187 inp = inp_list[i]; 2188 INP_RLOCK(inp); 2189 if (inp->inp_gencnt <= gencnt) { 2190 struct xtcpcb xt; 2191 2192 tcp_inptoxtp(inp, &xt); 2193 INP_RUNLOCK(inp); 2194 error = SYSCTL_OUT(req, &xt, sizeof xt); 2195 } else 2196 INP_RUNLOCK(inp); 2197 } 2198 2199 il->il_count = n; 2200 il->il_pcbinfo = &V_tcbinfo; 2201 epoch_call(net_epoch_preempt, &il->il_epoch_ctx, in_pcblist_rele_rlocked); 2202 2203 if (!error) { 2204 /* 2205 * Give the user an updated idea of our state. 2206 * If the generation differs from what we told 2207 * her before, she knows that something happened 2208 * while we were processing this request, and it 2209 * might be necessary to retry. 2210 */ 2211 INP_LIST_RLOCK(&V_tcbinfo); 2212 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2213 xig.xig_sogen = so_gencnt; 2214 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 2215 INP_LIST_RUNLOCK(&V_tcbinfo); 2216 error = SYSCTL_OUT(req, &xig, sizeof xig); 2217 } 2218 return (error); 2219 } 2220 2221 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2222 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 2223 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 2224 2225 #ifdef INET 2226 static int 2227 tcp_getcred(SYSCTL_HANDLER_ARGS) 2228 { 2229 struct xucred xuc; 2230 struct sockaddr_in addrs[2]; 2231 struct inpcb *inp; 2232 int error; 2233 2234 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2235 if (error) 2236 return (error); 2237 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2238 if (error) 2239 return (error); 2240 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2241 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2242 if (inp != NULL) { 2243 if (inp->inp_socket == NULL) 2244 error = ENOENT; 2245 if (error == 0) 2246 error = cr_canseeinpcb(req->td->td_ucred, inp); 2247 if (error == 0) 2248 cru2x(inp->inp_cred, &xuc); 2249 INP_RUNLOCK(inp); 2250 } else 2251 error = ENOENT; 2252 if (error == 0) 2253 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2254 return (error); 2255 } 2256 2257 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2258 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2259 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2260 #endif /* INET */ 2261 2262 #ifdef INET6 2263 static int 2264 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2265 { 2266 struct xucred xuc; 2267 struct sockaddr_in6 addrs[2]; 2268 struct inpcb *inp; 2269 int error; 2270 #ifdef INET 2271 int mapped = 0; 2272 #endif 2273 2274 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2275 if (error) 2276 return (error); 2277 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2278 if (error) 2279 return (error); 2280 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2281 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2282 return (error); 2283 } 2284 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2285 #ifdef INET 2286 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2287 mapped = 1; 2288 else 2289 #endif 2290 return (EINVAL); 2291 } 2292 2293 #ifdef INET 2294 if (mapped == 1) 2295 inp = in_pcblookup(&V_tcbinfo, 2296 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2297 addrs[1].sin6_port, 2298 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2299 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2300 else 2301 #endif 2302 inp = in6_pcblookup(&V_tcbinfo, 2303 &addrs[1].sin6_addr, addrs[1].sin6_port, 2304 &addrs[0].sin6_addr, addrs[0].sin6_port, 2305 INPLOOKUP_RLOCKPCB, NULL); 2306 if (inp != NULL) { 2307 if (inp->inp_socket == NULL) 2308 error = ENOENT; 2309 if (error == 0) 2310 error = cr_canseeinpcb(req->td->td_ucred, inp); 2311 if (error == 0) 2312 cru2x(inp->inp_cred, &xuc); 2313 INP_RUNLOCK(inp); 2314 } else 2315 error = ENOENT; 2316 if (error == 0) 2317 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2318 return (error); 2319 } 2320 2321 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2322 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2323 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2324 #endif /* INET6 */ 2325 2326 2327 #ifdef INET 2328 void 2329 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2330 { 2331 struct ip *ip = vip; 2332 struct tcphdr *th; 2333 struct in_addr faddr; 2334 struct inpcb *inp; 2335 struct tcpcb *tp; 2336 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2337 struct icmp *icp; 2338 struct in_conninfo inc; 2339 tcp_seq icmp_tcp_seq; 2340 int mtu; 2341 2342 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2343 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2344 return; 2345 2346 if (cmd == PRC_MSGSIZE) 2347 notify = tcp_mtudisc_notify; 2348 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2349 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2350 cmd == PRC_TIMXCEED_INTRANS) && ip) 2351 notify = tcp_drop_syn_sent; 2352 2353 /* 2354 * Hostdead is ugly because it goes linearly through all PCBs. 2355 * XXX: We never get this from ICMP, otherwise it makes an 2356 * excellent DoS attack on machines with many connections. 2357 */ 2358 else if (cmd == PRC_HOSTDEAD) 2359 ip = NULL; 2360 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2361 return; 2362 2363 if (ip == NULL) { 2364 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2365 return; 2366 } 2367 2368 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2369 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2370 INP_INFO_RLOCK(&V_tcbinfo); 2371 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2372 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2373 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2374 /* signal EHOSTDOWN, as it flushes the cached route */ 2375 inp = (*notify)(inp, EHOSTDOWN); 2376 goto out; 2377 } 2378 icmp_tcp_seq = th->th_seq; 2379 if (inp != NULL) { 2380 if (!(inp->inp_flags & INP_TIMEWAIT) && 2381 !(inp->inp_flags & INP_DROPPED) && 2382 !(inp->inp_socket == NULL)) { 2383 tp = intotcpcb(inp); 2384 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2385 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2386 if (cmd == PRC_MSGSIZE) { 2387 /* 2388 * MTU discovery: 2389 * If we got a needfrag set the MTU 2390 * in the route to the suggested new 2391 * value (if given) and then notify. 2392 */ 2393 mtu = ntohs(icp->icmp_nextmtu); 2394 /* 2395 * If no alternative MTU was 2396 * proposed, try the next smaller 2397 * one. 2398 */ 2399 if (!mtu) 2400 mtu = ip_next_mtu( 2401 ntohs(ip->ip_len), 1); 2402 if (mtu < V_tcp_minmss + 2403 sizeof(struct tcpiphdr)) 2404 mtu = V_tcp_minmss + 2405 sizeof(struct tcpiphdr); 2406 /* 2407 * Only process the offered MTU if it 2408 * is smaller than the current one. 2409 */ 2410 if (mtu < tp->t_maxseg + 2411 sizeof(struct tcpiphdr)) { 2412 bzero(&inc, sizeof(inc)); 2413 inc.inc_faddr = faddr; 2414 inc.inc_fibnum = 2415 inp->inp_inc.inc_fibnum; 2416 tcp_hc_updatemtu(&inc, mtu); 2417 tcp_mtudisc(inp, mtu); 2418 } 2419 } else 2420 inp = (*notify)(inp, 2421 inetctlerrmap[cmd]); 2422 } 2423 } 2424 } else { 2425 bzero(&inc, sizeof(inc)); 2426 inc.inc_fport = th->th_dport; 2427 inc.inc_lport = th->th_sport; 2428 inc.inc_faddr = faddr; 2429 inc.inc_laddr = ip->ip_src; 2430 syncache_unreach(&inc, icmp_tcp_seq); 2431 } 2432 out: 2433 if (inp != NULL) 2434 INP_WUNLOCK(inp); 2435 INP_INFO_RUNLOCK(&V_tcbinfo); 2436 } 2437 #endif /* INET */ 2438 2439 #ifdef INET6 2440 void 2441 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2442 { 2443 struct in6_addr *dst; 2444 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2445 struct ip6_hdr *ip6; 2446 struct mbuf *m; 2447 struct inpcb *inp; 2448 struct tcpcb *tp; 2449 struct icmp6_hdr *icmp6; 2450 struct ip6ctlparam *ip6cp = NULL; 2451 const struct sockaddr_in6 *sa6_src = NULL; 2452 struct in_conninfo inc; 2453 struct tcp_ports { 2454 uint16_t th_sport; 2455 uint16_t th_dport; 2456 } t_ports; 2457 tcp_seq icmp_tcp_seq; 2458 unsigned int mtu; 2459 unsigned int off; 2460 2461 if (sa->sa_family != AF_INET6 || 2462 sa->sa_len != sizeof(struct sockaddr_in6)) 2463 return; 2464 2465 /* if the parameter is from icmp6, decode it. */ 2466 if (d != NULL) { 2467 ip6cp = (struct ip6ctlparam *)d; 2468 icmp6 = ip6cp->ip6c_icmp6; 2469 m = ip6cp->ip6c_m; 2470 ip6 = ip6cp->ip6c_ip6; 2471 off = ip6cp->ip6c_off; 2472 sa6_src = ip6cp->ip6c_src; 2473 dst = ip6cp->ip6c_finaldst; 2474 } else { 2475 m = NULL; 2476 ip6 = NULL; 2477 off = 0; /* fool gcc */ 2478 sa6_src = &sa6_any; 2479 dst = NULL; 2480 } 2481 2482 if (cmd == PRC_MSGSIZE) 2483 notify = tcp_mtudisc_notify; 2484 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2485 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2486 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2487 notify = tcp_drop_syn_sent; 2488 2489 /* 2490 * Hostdead is ugly because it goes linearly through all PCBs. 2491 * XXX: We never get this from ICMP, otherwise it makes an 2492 * excellent DoS attack on machines with many connections. 2493 */ 2494 else if (cmd == PRC_HOSTDEAD) 2495 ip6 = NULL; 2496 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2497 return; 2498 2499 if (ip6 == NULL) { 2500 in6_pcbnotify(&V_tcbinfo, sa, 0, 2501 (const struct sockaddr *)sa6_src, 2502 0, cmd, NULL, notify); 2503 return; 2504 } 2505 2506 /* Check if we can safely get the ports from the tcp hdr */ 2507 if (m == NULL || 2508 (m->m_pkthdr.len < 2509 (int32_t) (off + sizeof(struct tcp_ports)))) { 2510 return; 2511 } 2512 bzero(&t_ports, sizeof(struct tcp_ports)); 2513 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2514 INP_INFO_RLOCK(&V_tcbinfo); 2515 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2516 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2517 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2518 /* signal EHOSTDOWN, as it flushes the cached route */ 2519 inp = (*notify)(inp, EHOSTDOWN); 2520 goto out; 2521 } 2522 off += sizeof(struct tcp_ports); 2523 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2524 goto out; 2525 } 2526 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2527 if (inp != NULL) { 2528 if (!(inp->inp_flags & INP_TIMEWAIT) && 2529 !(inp->inp_flags & INP_DROPPED) && 2530 !(inp->inp_socket == NULL)) { 2531 tp = intotcpcb(inp); 2532 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2533 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2534 if (cmd == PRC_MSGSIZE) { 2535 /* 2536 * MTU discovery: 2537 * If we got a needfrag set the MTU 2538 * in the route to the suggested new 2539 * value (if given) and then notify. 2540 */ 2541 mtu = ntohl(icmp6->icmp6_mtu); 2542 /* 2543 * If no alternative MTU was 2544 * proposed, or the proposed 2545 * MTU was too small, set to 2546 * the min. 2547 */ 2548 if (mtu < IPV6_MMTU) 2549 mtu = IPV6_MMTU - 8; 2550 bzero(&inc, sizeof(inc)); 2551 inc.inc_fibnum = M_GETFIB(m); 2552 inc.inc_flags |= INC_ISIPV6; 2553 inc.inc6_faddr = *dst; 2554 if (in6_setscope(&inc.inc6_faddr, 2555 m->m_pkthdr.rcvif, NULL)) 2556 goto out; 2557 /* 2558 * Only process the offered MTU if it 2559 * is smaller than the current one. 2560 */ 2561 if (mtu < tp->t_maxseg + 2562 sizeof (struct tcphdr) + 2563 sizeof (struct ip6_hdr)) { 2564 tcp_hc_updatemtu(&inc, mtu); 2565 tcp_mtudisc(inp, mtu); 2566 ICMP6STAT_INC(icp6s_pmtuchg); 2567 } 2568 } else 2569 inp = (*notify)(inp, 2570 inet6ctlerrmap[cmd]); 2571 } 2572 } 2573 } else { 2574 bzero(&inc, sizeof(inc)); 2575 inc.inc_fibnum = M_GETFIB(m); 2576 inc.inc_flags |= INC_ISIPV6; 2577 inc.inc_fport = t_ports.th_dport; 2578 inc.inc_lport = t_ports.th_sport; 2579 inc.inc6_faddr = *dst; 2580 inc.inc6_laddr = ip6->ip6_src; 2581 syncache_unreach(&inc, icmp_tcp_seq); 2582 } 2583 out: 2584 if (inp != NULL) 2585 INP_WUNLOCK(inp); 2586 INP_INFO_RUNLOCK(&V_tcbinfo); 2587 } 2588 #endif /* INET6 */ 2589 2590 2591 /* 2592 * Following is where TCP initial sequence number generation occurs. 2593 * 2594 * There are two places where we must use initial sequence numbers: 2595 * 1. In SYN-ACK packets. 2596 * 2. In SYN packets. 2597 * 2598 * All ISNs for SYN-ACK packets are generated by the syncache. See 2599 * tcp_syncache.c for details. 2600 * 2601 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2602 * depends on this property. In addition, these ISNs should be 2603 * unguessable so as to prevent connection hijacking. To satisfy 2604 * the requirements of this situation, the algorithm outlined in 2605 * RFC 1948 is used, with only small modifications. 2606 * 2607 * Implementation details: 2608 * 2609 * Time is based off the system timer, and is corrected so that it 2610 * increases by one megabyte per second. This allows for proper 2611 * recycling on high speed LANs while still leaving over an hour 2612 * before rollover. 2613 * 2614 * As reading the *exact* system time is too expensive to be done 2615 * whenever setting up a TCP connection, we increment the time 2616 * offset in two ways. First, a small random positive increment 2617 * is added to isn_offset for each connection that is set up. 2618 * Second, the function tcp_isn_tick fires once per clock tick 2619 * and increments isn_offset as necessary so that sequence numbers 2620 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2621 * random positive increments serve only to ensure that the same 2622 * exact sequence number is never sent out twice (as could otherwise 2623 * happen when a port is recycled in less than the system tick 2624 * interval.) 2625 * 2626 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2627 * between seeding of isn_secret. This is normally set to zero, 2628 * as reseeding should not be necessary. 2629 * 2630 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2631 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2632 * general, this means holding an exclusive (write) lock. 2633 */ 2634 2635 #define ISN_BYTES_PER_SECOND 1048576 2636 #define ISN_STATIC_INCREMENT 4096 2637 #define ISN_RANDOM_INCREMENT (4096 - 1) 2638 2639 static VNET_DEFINE(u_char, isn_secret[32]); 2640 static VNET_DEFINE(int, isn_last); 2641 static VNET_DEFINE(int, isn_last_reseed); 2642 static VNET_DEFINE(u_int32_t, isn_offset); 2643 static VNET_DEFINE(u_int32_t, isn_offset_old); 2644 2645 #define V_isn_secret VNET(isn_secret) 2646 #define V_isn_last VNET(isn_last) 2647 #define V_isn_last_reseed VNET(isn_last_reseed) 2648 #define V_isn_offset VNET(isn_offset) 2649 #define V_isn_offset_old VNET(isn_offset_old) 2650 2651 tcp_seq 2652 tcp_new_isn(struct tcpcb *tp) 2653 { 2654 MD5_CTX isn_ctx; 2655 u_int32_t md5_buffer[4]; 2656 tcp_seq new_isn; 2657 u_int32_t projected_offset; 2658 2659 INP_WLOCK_ASSERT(tp->t_inpcb); 2660 2661 ISN_LOCK(); 2662 /* Seed if this is the first use, reseed if requested. */ 2663 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2664 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2665 < (u_int)ticks))) { 2666 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2667 V_isn_last_reseed = ticks; 2668 } 2669 2670 /* Compute the md5 hash and return the ISN. */ 2671 MD5Init(&isn_ctx); 2672 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2673 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2674 #ifdef INET6 2675 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2676 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2677 sizeof(struct in6_addr)); 2678 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2679 sizeof(struct in6_addr)); 2680 } else 2681 #endif 2682 { 2683 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2684 sizeof(struct in_addr)); 2685 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2686 sizeof(struct in_addr)); 2687 } 2688 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2689 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2690 new_isn = (tcp_seq) md5_buffer[0]; 2691 V_isn_offset += ISN_STATIC_INCREMENT + 2692 (arc4random() & ISN_RANDOM_INCREMENT); 2693 if (ticks != V_isn_last) { 2694 projected_offset = V_isn_offset_old + 2695 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2696 if (SEQ_GT(projected_offset, V_isn_offset)) 2697 V_isn_offset = projected_offset; 2698 V_isn_offset_old = V_isn_offset; 2699 V_isn_last = ticks; 2700 } 2701 new_isn += V_isn_offset; 2702 ISN_UNLOCK(); 2703 return (new_isn); 2704 } 2705 2706 /* 2707 * When a specific ICMP unreachable message is received and the 2708 * connection state is SYN-SENT, drop the connection. This behavior 2709 * is controlled by the icmp_may_rst sysctl. 2710 */ 2711 struct inpcb * 2712 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2713 { 2714 struct tcpcb *tp; 2715 2716 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2717 INP_WLOCK_ASSERT(inp); 2718 2719 if ((inp->inp_flags & INP_TIMEWAIT) || 2720 (inp->inp_flags & INP_DROPPED)) 2721 return (inp); 2722 2723 tp = intotcpcb(inp); 2724 if (tp->t_state != TCPS_SYN_SENT) 2725 return (inp); 2726 2727 if (IS_FASTOPEN(tp->t_flags)) 2728 tcp_fastopen_disable_path(tp); 2729 2730 tp = tcp_drop(tp, errno); 2731 if (tp != NULL) 2732 return (inp); 2733 else 2734 return (NULL); 2735 } 2736 2737 /* 2738 * When `need fragmentation' ICMP is received, update our idea of the MSS 2739 * based on the new value. Also nudge TCP to send something, since we 2740 * know the packet we just sent was dropped. 2741 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2742 */ 2743 static struct inpcb * 2744 tcp_mtudisc_notify(struct inpcb *inp, int error) 2745 { 2746 2747 tcp_mtudisc(inp, -1); 2748 return (inp); 2749 } 2750 2751 static void 2752 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2753 { 2754 struct tcpcb *tp; 2755 struct socket *so; 2756 2757 INP_WLOCK_ASSERT(inp); 2758 if ((inp->inp_flags & INP_TIMEWAIT) || 2759 (inp->inp_flags & INP_DROPPED)) 2760 return; 2761 2762 tp = intotcpcb(inp); 2763 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2764 2765 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2766 2767 so = inp->inp_socket; 2768 SOCKBUF_LOCK(&so->so_snd); 2769 /* If the mss is larger than the socket buffer, decrease the mss. */ 2770 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2771 tp->t_maxseg = so->so_snd.sb_hiwat; 2772 SOCKBUF_UNLOCK(&so->so_snd); 2773 2774 TCPSTAT_INC(tcps_mturesent); 2775 tp->t_rtttime = 0; 2776 tp->snd_nxt = tp->snd_una; 2777 tcp_free_sackholes(tp); 2778 tp->snd_recover = tp->snd_max; 2779 if (tp->t_flags & TF_SACK_PERMIT) 2780 EXIT_FASTRECOVERY(tp->t_flags); 2781 tp->t_fb->tfb_tcp_output(tp); 2782 } 2783 2784 #ifdef INET 2785 /* 2786 * Look-up the routing entry to the peer of this inpcb. If no route 2787 * is found and it cannot be allocated, then return 0. This routine 2788 * is called by TCP routines that access the rmx structure and by 2789 * tcp_mss_update to get the peer/interface MTU. 2790 */ 2791 uint32_t 2792 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2793 { 2794 struct nhop4_extended nh4; 2795 struct ifnet *ifp; 2796 uint32_t maxmtu = 0; 2797 2798 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2799 2800 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2801 2802 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2803 NHR_REF, 0, &nh4) != 0) 2804 return (0); 2805 2806 ifp = nh4.nh_ifp; 2807 maxmtu = nh4.nh_mtu; 2808 2809 /* Report additional interface capabilities. */ 2810 if (cap != NULL) { 2811 if (ifp->if_capenable & IFCAP_TSO4 && 2812 ifp->if_hwassist & CSUM_TSO) { 2813 cap->ifcap |= CSUM_TSO; 2814 cap->tsomax = ifp->if_hw_tsomax; 2815 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2816 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2817 } 2818 } 2819 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2820 } 2821 return (maxmtu); 2822 } 2823 #endif /* INET */ 2824 2825 #ifdef INET6 2826 uint32_t 2827 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2828 { 2829 struct nhop6_extended nh6; 2830 struct in6_addr dst6; 2831 uint32_t scopeid; 2832 struct ifnet *ifp; 2833 uint32_t maxmtu = 0; 2834 2835 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2836 2837 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2838 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2839 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2840 0, &nh6) != 0) 2841 return (0); 2842 2843 ifp = nh6.nh_ifp; 2844 maxmtu = nh6.nh_mtu; 2845 2846 /* Report additional interface capabilities. */ 2847 if (cap != NULL) { 2848 if (ifp->if_capenable & IFCAP_TSO6 && 2849 ifp->if_hwassist & CSUM_TSO) { 2850 cap->ifcap |= CSUM_TSO; 2851 cap->tsomax = ifp->if_hw_tsomax; 2852 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2853 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2854 } 2855 } 2856 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2857 } 2858 2859 return (maxmtu); 2860 } 2861 #endif /* INET6 */ 2862 2863 /* 2864 * Calculate effective SMSS per RFC5681 definition for a given TCP 2865 * connection at its current state, taking into account SACK and etc. 2866 */ 2867 u_int 2868 tcp_maxseg(const struct tcpcb *tp) 2869 { 2870 u_int optlen; 2871 2872 if (tp->t_flags & TF_NOOPT) 2873 return (tp->t_maxseg); 2874 2875 /* 2876 * Here we have a simplified code from tcp_addoptions(), 2877 * without a proper loop, and having most of paddings hardcoded. 2878 * We might make mistakes with padding here in some edge cases, 2879 * but this is harmless, since result of tcp_maxseg() is used 2880 * only in cwnd and ssthresh estimations. 2881 */ 2882 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2883 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2884 if (tp->t_flags & TF_RCVD_TSTMP) 2885 optlen = TCPOLEN_TSTAMP_APPA; 2886 else 2887 optlen = 0; 2888 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2889 if (tp->t_flags & TF_SIGNATURE) 2890 optlen += PAD(TCPOLEN_SIGNATURE); 2891 #endif 2892 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2893 optlen += TCPOLEN_SACKHDR; 2894 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2895 optlen = PAD(optlen); 2896 } 2897 } else { 2898 if (tp->t_flags & TF_REQ_TSTMP) 2899 optlen = TCPOLEN_TSTAMP_APPA; 2900 else 2901 optlen = PAD(TCPOLEN_MAXSEG); 2902 if (tp->t_flags & TF_REQ_SCALE) 2903 optlen += PAD(TCPOLEN_WINDOW); 2904 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2905 if (tp->t_flags & TF_SIGNATURE) 2906 optlen += PAD(TCPOLEN_SIGNATURE); 2907 #endif 2908 if (tp->t_flags & TF_SACK_PERMIT) 2909 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2910 } 2911 #undef PAD 2912 optlen = min(optlen, TCP_MAXOLEN); 2913 return (tp->t_maxseg - optlen); 2914 } 2915 2916 static int 2917 sysctl_drop(SYSCTL_HANDLER_ARGS) 2918 { 2919 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2920 struct sockaddr_storage addrs[2]; 2921 struct inpcb *inp; 2922 struct tcpcb *tp; 2923 struct tcptw *tw; 2924 struct sockaddr_in *fin, *lin; 2925 #ifdef INET6 2926 struct sockaddr_in6 *fin6, *lin6; 2927 #endif 2928 int error; 2929 2930 inp = NULL; 2931 fin = lin = NULL; 2932 #ifdef INET6 2933 fin6 = lin6 = NULL; 2934 #endif 2935 error = 0; 2936 2937 if (req->oldptr != NULL || req->oldlen != 0) 2938 return (EINVAL); 2939 if (req->newptr == NULL) 2940 return (EPERM); 2941 if (req->newlen < sizeof(addrs)) 2942 return (ENOMEM); 2943 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2944 if (error) 2945 return (error); 2946 2947 switch (addrs[0].ss_family) { 2948 #ifdef INET6 2949 case AF_INET6: 2950 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2951 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2952 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2953 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2954 return (EINVAL); 2955 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2956 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2957 return (EINVAL); 2958 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2959 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2960 fin = (struct sockaddr_in *)&addrs[0]; 2961 lin = (struct sockaddr_in *)&addrs[1]; 2962 break; 2963 } 2964 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2965 if (error) 2966 return (error); 2967 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2968 if (error) 2969 return (error); 2970 break; 2971 #endif 2972 #ifdef INET 2973 case AF_INET: 2974 fin = (struct sockaddr_in *)&addrs[0]; 2975 lin = (struct sockaddr_in *)&addrs[1]; 2976 if (fin->sin_len != sizeof(struct sockaddr_in) || 2977 lin->sin_len != sizeof(struct sockaddr_in)) 2978 return (EINVAL); 2979 break; 2980 #endif 2981 default: 2982 return (EINVAL); 2983 } 2984 INP_INFO_RLOCK(&V_tcbinfo); 2985 switch (addrs[0].ss_family) { 2986 #ifdef INET6 2987 case AF_INET6: 2988 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2989 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2990 INPLOOKUP_WLOCKPCB, NULL); 2991 break; 2992 #endif 2993 #ifdef INET 2994 case AF_INET: 2995 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2996 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2997 break; 2998 #endif 2999 } 3000 if (inp != NULL) { 3001 if (inp->inp_flags & INP_TIMEWAIT) { 3002 /* 3003 * XXXRW: There currently exists a state where an 3004 * inpcb is present, but its timewait state has been 3005 * discarded. For now, don't allow dropping of this 3006 * type of inpcb. 3007 */ 3008 tw = intotw(inp); 3009 if (tw != NULL) 3010 tcp_twclose(tw, 0); 3011 else 3012 INP_WUNLOCK(inp); 3013 } else if (!(inp->inp_flags & INP_DROPPED) && 3014 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3015 tp = intotcpcb(inp); 3016 tp = tcp_drop(tp, ECONNABORTED); 3017 if (tp != NULL) 3018 INP_WUNLOCK(inp); 3019 } else 3020 INP_WUNLOCK(inp); 3021 } else 3022 error = ESRCH; 3023 INP_INFO_RUNLOCK(&V_tcbinfo); 3024 return (error); 3025 } 3026 3027 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3028 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3029 0, sysctl_drop, "", "Drop TCP connection"); 3030 3031 /* 3032 * Generate a standardized TCP log line for use throughout the 3033 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3034 * allow use in the interrupt context. 3035 * 3036 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3037 * NB: The function may return NULL if memory allocation failed. 3038 * 3039 * Due to header inclusion and ordering limitations the struct ip 3040 * and ip6_hdr pointers have to be passed as void pointers. 3041 */ 3042 char * 3043 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3044 const void *ip6hdr) 3045 { 3046 3047 /* Is logging enabled? */ 3048 if (tcp_log_in_vain == 0) 3049 return (NULL); 3050 3051 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3052 } 3053 3054 char * 3055 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3056 const void *ip6hdr) 3057 { 3058 3059 /* Is logging enabled? */ 3060 if (tcp_log_debug == 0) 3061 return (NULL); 3062 3063 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3064 } 3065 3066 static char * 3067 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3068 const void *ip6hdr) 3069 { 3070 char *s, *sp; 3071 size_t size; 3072 struct ip *ip; 3073 #ifdef INET6 3074 const struct ip6_hdr *ip6; 3075 3076 ip6 = (const struct ip6_hdr *)ip6hdr; 3077 #endif /* INET6 */ 3078 ip = (struct ip *)ip4hdr; 3079 3080 /* 3081 * The log line looks like this: 3082 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3083 */ 3084 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3085 sizeof(PRINT_TH_FLAGS) + 1 + 3086 #ifdef INET6 3087 2 * INET6_ADDRSTRLEN; 3088 #else 3089 2 * INET_ADDRSTRLEN; 3090 #endif /* INET6 */ 3091 3092 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3093 if (s == NULL) 3094 return (NULL); 3095 3096 strcat(s, "TCP: ["); 3097 sp = s + strlen(s); 3098 3099 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3100 inet_ntoa_r(inc->inc_faddr, sp); 3101 sp = s + strlen(s); 3102 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3103 sp = s + strlen(s); 3104 inet_ntoa_r(inc->inc_laddr, sp); 3105 sp = s + strlen(s); 3106 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3107 #ifdef INET6 3108 } else if (inc) { 3109 ip6_sprintf(sp, &inc->inc6_faddr); 3110 sp = s + strlen(s); 3111 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3112 sp = s + strlen(s); 3113 ip6_sprintf(sp, &inc->inc6_laddr); 3114 sp = s + strlen(s); 3115 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3116 } else if (ip6 && th) { 3117 ip6_sprintf(sp, &ip6->ip6_src); 3118 sp = s + strlen(s); 3119 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3120 sp = s + strlen(s); 3121 ip6_sprintf(sp, &ip6->ip6_dst); 3122 sp = s + strlen(s); 3123 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3124 #endif /* INET6 */ 3125 #ifdef INET 3126 } else if (ip && th) { 3127 inet_ntoa_r(ip->ip_src, sp); 3128 sp = s + strlen(s); 3129 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3130 sp = s + strlen(s); 3131 inet_ntoa_r(ip->ip_dst, sp); 3132 sp = s + strlen(s); 3133 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3134 #endif /* INET */ 3135 } else { 3136 free(s, M_TCPLOG); 3137 return (NULL); 3138 } 3139 sp = s + strlen(s); 3140 if (th) 3141 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3142 if (*(s + size - 1) != '\0') 3143 panic("%s: string too long", __func__); 3144 return (s); 3145 } 3146 3147 /* 3148 * A subroutine which makes it easy to track TCP state changes with DTrace. 3149 * This function shouldn't be called for t_state initializations that don't 3150 * correspond to actual TCP state transitions. 3151 */ 3152 void 3153 tcp_state_change(struct tcpcb *tp, int newstate) 3154 { 3155 #if defined(KDTRACE_HOOKS) 3156 int pstate = tp->t_state; 3157 #endif 3158 3159 TCPSTATES_DEC(tp->t_state); 3160 TCPSTATES_INC(newstate); 3161 tp->t_state = newstate; 3162 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3163 } 3164 3165 /* 3166 * Create an external-format (``xtcpcb'') structure using the information in 3167 * the kernel-format tcpcb structure pointed to by tp. This is done to 3168 * reduce the spew of irrelevant information over this interface, to isolate 3169 * user code from changes in the kernel structure, and potentially to provide 3170 * information-hiding if we decide that some of this information should be 3171 * hidden from users. 3172 */ 3173 void 3174 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3175 { 3176 struct tcpcb *tp = intotcpcb(inp); 3177 sbintime_t now; 3178 3179 if (inp->inp_flags & INP_TIMEWAIT) { 3180 bzero(xt, sizeof(struct xtcpcb)); 3181 xt->t_state = TCPS_TIME_WAIT; 3182 } else { 3183 xt->t_state = tp->t_state; 3184 xt->t_logstate = tp->t_logstate; 3185 xt->t_flags = tp->t_flags; 3186 xt->t_sndzerowin = tp->t_sndzerowin; 3187 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3188 xt->t_rcvoopack = tp->t_rcvoopack; 3189 3190 now = getsbinuptime(); 3191 #define COPYTIMER(ttt) do { \ 3192 if (callout_active(&tp->t_timers->ttt)) \ 3193 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3194 SBT_1MS; \ 3195 else \ 3196 xt->ttt = 0; \ 3197 } while (0) 3198 COPYTIMER(tt_delack); 3199 COPYTIMER(tt_rexmt); 3200 COPYTIMER(tt_persist); 3201 COPYTIMER(tt_keep); 3202 COPYTIMER(tt_2msl); 3203 #undef COPYTIMER 3204 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3205 3206 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3207 TCP_FUNCTION_NAME_LEN_MAX); 3208 bzero(xt->xt_logid, TCP_LOG_ID_LEN); 3209 #ifdef TCP_BLACKBOX 3210 (void)tcp_log_get_id(tp, xt->xt_logid); 3211 #endif 3212 } 3213 3214 xt->xt_len = sizeof(struct xtcpcb); 3215 in_pcbtoxinpcb(inp, &xt->xt_inp); 3216 if (inp->inp_socket == NULL) 3217 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3218 } 3219