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 VNET_DEFINE_STATIC(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 VNET_DEFINE_STATIC(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 #define TS_OFFSET_SECRET_LENGTH 32 237 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 238 #define V_ts_offset_secret VNET(ts_offset_secret) 239 240 static int tcp_default_fb_init(struct tcpcb *tp); 241 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 242 static int tcp_default_handoff_ok(struct tcpcb *tp); 243 static struct inpcb *tcp_notify(struct inpcb *, int); 244 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 245 static void tcp_mtudisc(struct inpcb *, int); 246 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 247 void *ip4hdr, const void *ip6hdr); 248 249 250 static struct tcp_function_block tcp_def_funcblk = { 251 .tfb_tcp_block_name = "freebsd", 252 .tfb_tcp_output = tcp_output, 253 .tfb_tcp_do_segment = tcp_do_segment, 254 .tfb_tcp_ctloutput = tcp_default_ctloutput, 255 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 256 .tfb_tcp_fb_init = tcp_default_fb_init, 257 .tfb_tcp_fb_fini = tcp_default_fb_fini, 258 }; 259 260 static int tcp_fb_cnt = 0; 261 struct tcp_funchead t_functions; 262 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 263 264 static struct tcp_function_block * 265 find_tcp_functions_locked(struct tcp_function_set *fs) 266 { 267 struct tcp_function *f; 268 struct tcp_function_block *blk=NULL; 269 270 TAILQ_FOREACH(f, &t_functions, tf_next) { 271 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 272 blk = f->tf_fb; 273 break; 274 } 275 } 276 return(blk); 277 } 278 279 static struct tcp_function_block * 280 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 281 { 282 struct tcp_function_block *rblk=NULL; 283 struct tcp_function *f; 284 285 TAILQ_FOREACH(f, &t_functions, tf_next) { 286 if (f->tf_fb == blk) { 287 rblk = blk; 288 if (s) { 289 *s = f; 290 } 291 break; 292 } 293 } 294 return (rblk); 295 } 296 297 struct tcp_function_block * 298 find_and_ref_tcp_functions(struct tcp_function_set *fs) 299 { 300 struct tcp_function_block *blk; 301 302 rw_rlock(&tcp_function_lock); 303 blk = find_tcp_functions_locked(fs); 304 if (blk) 305 refcount_acquire(&blk->tfb_refcnt); 306 rw_runlock(&tcp_function_lock); 307 return(blk); 308 } 309 310 struct tcp_function_block * 311 find_and_ref_tcp_fb(struct tcp_function_block *blk) 312 { 313 struct tcp_function_block *rblk; 314 315 rw_rlock(&tcp_function_lock); 316 rblk = find_tcp_fb_locked(blk, NULL); 317 if (rblk) 318 refcount_acquire(&rblk->tfb_refcnt); 319 rw_runlock(&tcp_function_lock); 320 return(rblk); 321 } 322 323 static struct tcp_function_block * 324 find_and_ref_tcp_default_fb(void) 325 { 326 struct tcp_function_block *rblk; 327 328 rw_rlock(&tcp_function_lock); 329 rblk = tcp_func_set_ptr; 330 refcount_acquire(&rblk->tfb_refcnt); 331 rw_runlock(&tcp_function_lock); 332 return (rblk); 333 } 334 335 void 336 tcp_switch_back_to_default(struct tcpcb *tp) 337 { 338 struct tcp_function_block *tfb; 339 340 KASSERT(tp->t_fb != &tcp_def_funcblk, 341 ("%s: called by the built-in default stack", __func__)); 342 343 /* 344 * Release the old stack. This function will either find a new one 345 * or panic. 346 */ 347 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 348 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 349 refcount_release(&tp->t_fb->tfb_refcnt); 350 351 /* 352 * Now, we'll find a new function block to use. 353 * Start by trying the current user-selected 354 * default, unless this stack is the user-selected 355 * default. 356 */ 357 tfb = find_and_ref_tcp_default_fb(); 358 if (tfb == tp->t_fb) { 359 refcount_release(&tfb->tfb_refcnt); 360 tfb = NULL; 361 } 362 /* Does the stack accept this connection? */ 363 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 364 (*tfb->tfb_tcp_handoff_ok)(tp)) { 365 refcount_release(&tfb->tfb_refcnt); 366 tfb = NULL; 367 } 368 /* Try to use that stack. */ 369 if (tfb != NULL) { 370 /* Initialize the new stack. If it succeeds, we are done. */ 371 tp->t_fb = tfb; 372 if (tp->t_fb->tfb_tcp_fb_init == NULL || 373 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 374 return; 375 376 /* 377 * Initialization failed. Release the reference count on 378 * the stack. 379 */ 380 refcount_release(&tfb->tfb_refcnt); 381 } 382 383 /* 384 * If that wasn't feasible, use the built-in default 385 * stack which is not allowed to reject anyone. 386 */ 387 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 388 if (tfb == NULL) { 389 /* there always should be a default */ 390 panic("Can't refer to tcp_def_funcblk"); 391 } 392 if (tfb->tfb_tcp_handoff_ok != NULL) { 393 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 394 /* The default stack cannot say no */ 395 panic("Default stack rejects a new session?"); 396 } 397 } 398 tp->t_fb = tfb; 399 if (tp->t_fb->tfb_tcp_fb_init != NULL && 400 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 401 /* The default stack cannot fail */ 402 panic("Default stack initialization failed"); 403 } 404 } 405 406 static int 407 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 408 { 409 int error=ENOENT; 410 struct tcp_function_set fs; 411 struct tcp_function_block *blk; 412 413 memset(&fs, 0, sizeof(fs)); 414 rw_rlock(&tcp_function_lock); 415 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 416 if (blk) { 417 /* Found him */ 418 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 419 fs.pcbcnt = blk->tfb_refcnt; 420 } 421 rw_runlock(&tcp_function_lock); 422 error = sysctl_handle_string(oidp, fs.function_set_name, 423 sizeof(fs.function_set_name), req); 424 425 /* Check for error or no change */ 426 if (error != 0 || req->newptr == NULL) 427 return(error); 428 429 rw_wlock(&tcp_function_lock); 430 blk = find_tcp_functions_locked(&fs); 431 if ((blk == NULL) || 432 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 433 error = ENOENT; 434 goto done; 435 } 436 tcp_func_set_ptr = blk; 437 done: 438 rw_wunlock(&tcp_function_lock); 439 return (error); 440 } 441 442 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 443 CTLTYPE_STRING | CTLFLAG_RW, 444 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 445 "Set/get the default TCP functions"); 446 447 static int 448 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 449 { 450 int error, cnt, linesz; 451 struct tcp_function *f; 452 char *buffer, *cp; 453 size_t bufsz, outsz; 454 bool alias; 455 456 cnt = 0; 457 rw_rlock(&tcp_function_lock); 458 TAILQ_FOREACH(f, &t_functions, tf_next) { 459 cnt++; 460 } 461 rw_runlock(&tcp_function_lock); 462 463 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 464 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 465 466 error = 0; 467 cp = buffer; 468 469 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 470 "Alias", "PCB count"); 471 cp += linesz; 472 bufsz -= linesz; 473 outsz = linesz; 474 475 rw_rlock(&tcp_function_lock); 476 TAILQ_FOREACH(f, &t_functions, tf_next) { 477 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 478 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 479 f->tf_fb->tfb_tcp_block_name, 480 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 481 alias ? f->tf_name : "-", 482 f->tf_fb->tfb_refcnt); 483 if (linesz >= bufsz) { 484 error = EOVERFLOW; 485 break; 486 } 487 cp += linesz; 488 bufsz -= linesz; 489 outsz += linesz; 490 } 491 rw_runlock(&tcp_function_lock); 492 if (error == 0) 493 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 494 free(buffer, M_TEMP); 495 return (error); 496 } 497 498 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 499 CTLTYPE_STRING|CTLFLAG_RD, 500 NULL, 0, sysctl_net_inet_list_available, "A", 501 "list available TCP Function sets"); 502 503 /* 504 * Exports one (struct tcp_function_info) for each alias/name. 505 */ 506 static int 507 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 508 { 509 int cnt, error; 510 struct tcp_function *f; 511 struct tcp_function_info tfi; 512 513 /* 514 * We don't allow writes. 515 */ 516 if (req->newptr != NULL) 517 return (EINVAL); 518 519 /* 520 * Wire the old buffer so we can directly copy the functions to 521 * user space without dropping the lock. 522 */ 523 if (req->oldptr != NULL) { 524 error = sysctl_wire_old_buffer(req, 0); 525 if (error) 526 return (error); 527 } 528 529 /* 530 * Walk the list and copy out matching entries. If INVARIANTS 531 * is compiled in, also walk the list to verify the length of 532 * the list matches what we have recorded. 533 */ 534 rw_rlock(&tcp_function_lock); 535 536 cnt = 0; 537 #ifndef INVARIANTS 538 if (req->oldptr == NULL) { 539 cnt = tcp_fb_cnt; 540 goto skip_loop; 541 } 542 #endif 543 TAILQ_FOREACH(f, &t_functions, tf_next) { 544 #ifdef INVARIANTS 545 cnt++; 546 #endif 547 if (req->oldptr != NULL) { 548 bzero(&tfi, sizeof(tfi)); 549 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 550 tfi.tfi_id = f->tf_fb->tfb_id; 551 (void)strlcpy(tfi.tfi_alias, f->tf_name, 552 sizeof(tfi.tfi_alias)); 553 (void)strlcpy(tfi.tfi_name, 554 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 555 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 556 /* 557 * Don't stop on error, as that is the 558 * mechanism we use to accumulate length 559 * information if the buffer was too short. 560 */ 561 } 562 } 563 KASSERT(cnt == tcp_fb_cnt, 564 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 565 #ifndef INVARIANTS 566 skip_loop: 567 #endif 568 rw_runlock(&tcp_function_lock); 569 if (req->oldptr == NULL) 570 error = SYSCTL_OUT(req, NULL, 571 (cnt + 1) * sizeof(struct tcp_function_info)); 572 573 return (error); 574 } 575 576 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 577 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 578 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 579 "List TCP function block name-to-ID mappings"); 580 581 /* 582 * tfb_tcp_handoff_ok() function for the default stack. 583 * Note that we'll basically try to take all comers. 584 */ 585 static int 586 tcp_default_handoff_ok(struct tcpcb *tp) 587 { 588 589 return (0); 590 } 591 592 /* 593 * tfb_tcp_fb_init() function for the default stack. 594 * 595 * This handles making sure we have appropriate timers set if you are 596 * transitioning a socket that has some amount of setup done. 597 * 598 * The init() fuction from the default can *never* return non-zero i.e. 599 * it is required to always succeed since it is the stack of last resort! 600 */ 601 static int 602 tcp_default_fb_init(struct tcpcb *tp) 603 { 604 605 struct socket *so; 606 607 INP_WLOCK_ASSERT(tp->t_inpcb); 608 609 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 610 ("%s: connection %p in unexpected state %d", __func__, tp, 611 tp->t_state)); 612 613 /* 614 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 615 * know what to do for unexpected states (which includes TIME_WAIT). 616 */ 617 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 618 return (0); 619 620 /* 621 * Make sure some kind of transmission timer is set if there is 622 * outstanding data. 623 */ 624 so = tp->t_inpcb->inp_socket; 625 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 626 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 627 tcp_timer_active(tp, TT_PERSIST))) { 628 /* 629 * If the session has established and it looks like it should 630 * be in the persist state, set the persist timer. Otherwise, 631 * set the retransmit timer. 632 */ 633 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 634 (int32_t)(tp->snd_nxt - tp->snd_una) < 635 (int32_t)sbavail(&so->so_snd)) 636 tcp_setpersist(tp); 637 else 638 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 639 } 640 641 /* All non-embryonic sessions get a keepalive timer. */ 642 if (!tcp_timer_active(tp, TT_KEEP)) 643 tcp_timer_activate(tp, TT_KEEP, 644 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 645 TP_KEEPINIT(tp)); 646 647 return (0); 648 } 649 650 /* 651 * tfb_tcp_fb_fini() function for the default stack. 652 * 653 * This changes state as necessary (or prudent) to prepare for another stack 654 * to assume responsibility for the connection. 655 */ 656 static void 657 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 658 { 659 660 INP_WLOCK_ASSERT(tp->t_inpcb); 661 return; 662 } 663 664 /* 665 * Target size of TCP PCB hash tables. Must be a power of two. 666 * 667 * Note that this can be overridden by the kernel environment 668 * variable net.inet.tcp.tcbhashsize 669 */ 670 #ifndef TCBHASHSIZE 671 #define TCBHASHSIZE 0 672 #endif 673 674 /* 675 * XXX 676 * Callouts should be moved into struct tcp directly. They are currently 677 * separate because the tcpcb structure is exported to userland for sysctl 678 * parsing purposes, which do not know about callouts. 679 */ 680 struct tcpcb_mem { 681 struct tcpcb tcb; 682 struct tcp_timer tt; 683 struct cc_var ccv; 684 #ifdef TCP_HHOOK 685 struct osd osd; 686 #endif 687 }; 688 689 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 690 #define V_tcpcb_zone VNET(tcpcb_zone) 691 692 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 693 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 694 695 static struct mtx isn_mtx; 696 697 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 698 #define ISN_LOCK() mtx_lock(&isn_mtx) 699 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 700 701 /* 702 * TCP initialization. 703 */ 704 static void 705 tcp_zone_change(void *tag) 706 { 707 708 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 709 uma_zone_set_max(V_tcpcb_zone, maxsockets); 710 tcp_tw_zone_change(); 711 } 712 713 static int 714 tcp_inpcb_init(void *mem, int size, int flags) 715 { 716 struct inpcb *inp = mem; 717 718 INP_LOCK_INIT(inp, "inp", "tcpinp"); 719 return (0); 720 } 721 722 /* 723 * Take a value and get the next power of 2 that doesn't overflow. 724 * Used to size the tcp_inpcb hash buckets. 725 */ 726 static int 727 maketcp_hashsize(int size) 728 { 729 int hashsize; 730 731 /* 732 * auto tune. 733 * get the next power of 2 higher than maxsockets. 734 */ 735 hashsize = 1 << fls(size); 736 /* catch overflow, and just go one power of 2 smaller */ 737 if (hashsize < size) { 738 hashsize = 1 << (fls(size) - 1); 739 } 740 return (hashsize); 741 } 742 743 static volatile int next_tcp_stack_id = 1; 744 745 /* 746 * Register a TCP function block with the name provided in the names 747 * array. (Note that this function does NOT automatically register 748 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 749 * explicitly include blk->tfb_tcp_block_name in the list of names if 750 * you wish to register the stack with that name.) 751 * 752 * Either all name registrations will succeed or all will fail. If 753 * a name registration fails, the function will update the num_names 754 * argument to point to the array index of the name that encountered 755 * the failure. 756 * 757 * Returns 0 on success, or an error code on failure. 758 */ 759 int 760 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 761 const char *names[], int *num_names) 762 { 763 struct tcp_function *n; 764 struct tcp_function_set fs; 765 int error, i; 766 767 KASSERT(names != NULL && *num_names > 0, 768 ("%s: Called with 0-length name list", __func__)); 769 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 770 KASSERT(rw_initialized(&tcp_function_lock), 771 ("%s: called too early", __func__)); 772 773 if ((blk->tfb_tcp_output == NULL) || 774 (blk->tfb_tcp_do_segment == NULL) || 775 (blk->tfb_tcp_ctloutput == NULL) || 776 (strlen(blk->tfb_tcp_block_name) == 0)) { 777 /* 778 * These functions are required and you 779 * need a name. 780 */ 781 *num_names = 0; 782 return (EINVAL); 783 } 784 if (blk->tfb_tcp_timer_stop_all || 785 blk->tfb_tcp_timer_activate || 786 blk->tfb_tcp_timer_active || 787 blk->tfb_tcp_timer_stop) { 788 /* 789 * If you define one timer function you 790 * must have them all. 791 */ 792 if ((blk->tfb_tcp_timer_stop_all == NULL) || 793 (blk->tfb_tcp_timer_activate == NULL) || 794 (blk->tfb_tcp_timer_active == NULL) || 795 (blk->tfb_tcp_timer_stop == NULL)) { 796 *num_names = 0; 797 return (EINVAL); 798 } 799 } 800 801 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 802 *num_names = 0; 803 return (EINVAL); 804 } 805 806 refcount_init(&blk->tfb_refcnt, 0); 807 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 808 for (i = 0; i < *num_names; i++) { 809 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 810 if (n == NULL) { 811 error = ENOMEM; 812 goto cleanup; 813 } 814 n->tf_fb = blk; 815 816 (void)strlcpy(fs.function_set_name, names[i], 817 sizeof(fs.function_set_name)); 818 rw_wlock(&tcp_function_lock); 819 if (find_tcp_functions_locked(&fs) != NULL) { 820 /* Duplicate name space not allowed */ 821 rw_wunlock(&tcp_function_lock); 822 free(n, M_TCPFUNCTIONS); 823 error = EALREADY; 824 goto cleanup; 825 } 826 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 827 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 828 tcp_fb_cnt++; 829 rw_wunlock(&tcp_function_lock); 830 } 831 return(0); 832 833 cleanup: 834 /* 835 * Deregister the names we just added. Because registration failed 836 * for names[i], we don't need to deregister that name. 837 */ 838 *num_names = i; 839 rw_wlock(&tcp_function_lock); 840 while (--i >= 0) { 841 TAILQ_FOREACH(n, &t_functions, tf_next) { 842 if (!strncmp(n->tf_name, names[i], 843 TCP_FUNCTION_NAME_LEN_MAX)) { 844 TAILQ_REMOVE(&t_functions, n, tf_next); 845 tcp_fb_cnt--; 846 n->tf_fb = NULL; 847 free(n, M_TCPFUNCTIONS); 848 break; 849 } 850 } 851 } 852 rw_wunlock(&tcp_function_lock); 853 return (error); 854 } 855 856 /* 857 * Register a TCP function block using the name provided in the name 858 * argument. 859 * 860 * Returns 0 on success, or an error code on failure. 861 */ 862 int 863 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 864 int wait) 865 { 866 const char *name_list[1]; 867 int num_names, rv; 868 869 num_names = 1; 870 if (name != NULL) 871 name_list[0] = name; 872 else 873 name_list[0] = blk->tfb_tcp_block_name; 874 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 875 return (rv); 876 } 877 878 /* 879 * Register a TCP function block using the name defined in 880 * blk->tfb_tcp_block_name. 881 * 882 * Returns 0 on success, or an error code on failure. 883 */ 884 int 885 register_tcp_functions(struct tcp_function_block *blk, int wait) 886 { 887 888 return (register_tcp_functions_as_name(blk, NULL, wait)); 889 } 890 891 /* 892 * Deregister all names associated with a function block. This 893 * functionally removes the function block from use within the system. 894 * 895 * When called with a true quiesce argument, mark the function block 896 * as being removed so no more stacks will use it and determine 897 * whether the removal would succeed. 898 * 899 * When called with a false quiesce argument, actually attempt the 900 * removal. 901 * 902 * When called with a force argument, attempt to switch all TCBs to 903 * use the default stack instead of returning EBUSY. 904 * 905 * Returns 0 on success (or if the removal would succeed, or an error 906 * code on failure. 907 */ 908 int 909 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 910 bool force) 911 { 912 struct tcp_function *f; 913 914 if (blk == &tcp_def_funcblk) { 915 /* You can't un-register the default */ 916 return (EPERM); 917 } 918 rw_wlock(&tcp_function_lock); 919 if (blk == tcp_func_set_ptr) { 920 /* You can't free the current default */ 921 rw_wunlock(&tcp_function_lock); 922 return (EBUSY); 923 } 924 /* Mark the block so no more stacks can use it. */ 925 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 926 /* 927 * If TCBs are still attached to the stack, attempt to switch them 928 * to the default stack. 929 */ 930 if (force && blk->tfb_refcnt) { 931 struct inpcb *inp; 932 struct tcpcb *tp; 933 VNET_ITERATOR_DECL(vnet_iter); 934 935 rw_wunlock(&tcp_function_lock); 936 937 VNET_LIST_RLOCK(); 938 VNET_FOREACH(vnet_iter) { 939 CURVNET_SET(vnet_iter); 940 INP_INFO_WLOCK(&V_tcbinfo); 941 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 942 INP_WLOCK(inp); 943 if (inp->inp_flags & INP_TIMEWAIT) { 944 INP_WUNLOCK(inp); 945 continue; 946 } 947 tp = intotcpcb(inp); 948 if (tp == NULL || tp->t_fb != blk) { 949 INP_WUNLOCK(inp); 950 continue; 951 } 952 tcp_switch_back_to_default(tp); 953 INP_WUNLOCK(inp); 954 } 955 INP_INFO_WUNLOCK(&V_tcbinfo); 956 CURVNET_RESTORE(); 957 } 958 VNET_LIST_RUNLOCK(); 959 960 rw_wlock(&tcp_function_lock); 961 } 962 if (blk->tfb_refcnt) { 963 /* TCBs still attached. */ 964 rw_wunlock(&tcp_function_lock); 965 return (EBUSY); 966 } 967 if (quiesce) { 968 /* Skip removal. */ 969 rw_wunlock(&tcp_function_lock); 970 return (0); 971 } 972 /* Remove any function names that map to this function block. */ 973 while (find_tcp_fb_locked(blk, &f) != NULL) { 974 TAILQ_REMOVE(&t_functions, f, tf_next); 975 tcp_fb_cnt--; 976 f->tf_fb = NULL; 977 free(f, M_TCPFUNCTIONS); 978 } 979 rw_wunlock(&tcp_function_lock); 980 return (0); 981 } 982 983 void 984 tcp_init(void) 985 { 986 const char *tcbhash_tuneable; 987 int hashsize; 988 989 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 990 991 #ifdef TCP_HHOOK 992 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 993 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 994 printf("%s: WARNING: unable to register helper hook\n", __func__); 995 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 996 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 997 printf("%s: WARNING: unable to register helper hook\n", __func__); 998 #endif 999 hashsize = TCBHASHSIZE; 1000 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1001 if (hashsize == 0) { 1002 /* 1003 * Auto tune the hash size based on maxsockets. 1004 * A perfect hash would have a 1:1 mapping 1005 * (hashsize = maxsockets) however it's been 1006 * suggested that O(2) average is better. 1007 */ 1008 hashsize = maketcp_hashsize(maxsockets / 4); 1009 /* 1010 * Our historical default is 512, 1011 * do not autotune lower than this. 1012 */ 1013 if (hashsize < 512) 1014 hashsize = 512; 1015 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1016 printf("%s: %s auto tuned to %d\n", __func__, 1017 tcbhash_tuneable, hashsize); 1018 } 1019 /* 1020 * We require a hashsize to be a power of two. 1021 * Previously if it was not a power of two we would just reset it 1022 * back to 512, which could be a nasty surprise if you did not notice 1023 * the error message. 1024 * Instead what we do is clip it to the closest power of two lower 1025 * than the specified hash value. 1026 */ 1027 if (!powerof2(hashsize)) { 1028 int oldhashsize = hashsize; 1029 1030 hashsize = maketcp_hashsize(hashsize); 1031 /* prevent absurdly low value */ 1032 if (hashsize < 16) 1033 hashsize = 16; 1034 printf("%s: WARNING: TCB hash size not a power of 2, " 1035 "clipped from %d to %d.\n", __func__, oldhashsize, 1036 hashsize); 1037 } 1038 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1039 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1040 1041 /* 1042 * These have to be type stable for the benefit of the timers. 1043 */ 1044 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1045 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1046 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1047 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1048 1049 tcp_tw_init(); 1050 syncache_init(); 1051 tcp_hc_init(); 1052 1053 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1054 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1055 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1056 1057 tcp_fastopen_init(); 1058 1059 /* Skip initialization of globals for non-default instances. */ 1060 if (!IS_DEFAULT_VNET(curvnet)) 1061 return; 1062 1063 tcp_reass_global_init(); 1064 1065 /* XXX virtualize those bellow? */ 1066 tcp_delacktime = TCPTV_DELACK; 1067 tcp_keepinit = TCPTV_KEEP_INIT; 1068 tcp_keepidle = TCPTV_KEEP_IDLE; 1069 tcp_keepintvl = TCPTV_KEEPINTVL; 1070 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1071 tcp_msl = TCPTV_MSL; 1072 tcp_rexmit_initial = TCPTV_RTOBASE; 1073 if (tcp_rexmit_initial < 1) 1074 tcp_rexmit_initial = 1; 1075 tcp_rexmit_min = TCPTV_MIN; 1076 if (tcp_rexmit_min < 1) 1077 tcp_rexmit_min = 1; 1078 tcp_persmin = TCPTV_PERSMIN; 1079 tcp_persmax = TCPTV_PERSMAX; 1080 tcp_rexmit_slop = TCPTV_CPU_VAR; 1081 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1082 tcp_tcbhashsize = hashsize; 1083 1084 /* Setup the tcp function block list */ 1085 TAILQ_INIT(&t_functions); 1086 rw_init(&tcp_function_lock, "tcp_func_lock"); 1087 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1088 #ifdef TCP_BLACKBOX 1089 /* Initialize the TCP logging data. */ 1090 tcp_log_init(); 1091 #endif 1092 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1093 1094 if (tcp_soreceive_stream) { 1095 #ifdef INET 1096 tcp_usrreqs.pru_soreceive = soreceive_stream; 1097 #endif 1098 #ifdef INET6 1099 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1100 #endif /* INET6 */ 1101 } 1102 1103 #ifdef INET6 1104 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1105 #else /* INET6 */ 1106 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1107 #endif /* INET6 */ 1108 if (max_protohdr < TCP_MINPROTOHDR) 1109 max_protohdr = TCP_MINPROTOHDR; 1110 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1111 panic("tcp_init"); 1112 #undef TCP_MINPROTOHDR 1113 1114 ISN_LOCK_INIT(); 1115 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1116 SHUTDOWN_PRI_DEFAULT); 1117 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1118 EVENTHANDLER_PRI_ANY); 1119 #ifdef TCPPCAP 1120 tcp_pcap_init(); 1121 #endif 1122 } 1123 1124 #ifdef VIMAGE 1125 static void 1126 tcp_destroy(void *unused __unused) 1127 { 1128 int n; 1129 #ifdef TCP_HHOOK 1130 int error; 1131 #endif 1132 1133 /* 1134 * All our processes are gone, all our sockets should be cleaned 1135 * up, which means, we should be past the tcp_discardcb() calls. 1136 * Sleep to let all tcpcb timers really disappear and cleanup. 1137 */ 1138 for (;;) { 1139 INP_LIST_RLOCK(&V_tcbinfo); 1140 n = V_tcbinfo.ipi_count; 1141 INP_LIST_RUNLOCK(&V_tcbinfo); 1142 if (n == 0) 1143 break; 1144 pause("tcpdes", hz / 10); 1145 } 1146 tcp_hc_destroy(); 1147 syncache_destroy(); 1148 tcp_tw_destroy(); 1149 in_pcbinfo_destroy(&V_tcbinfo); 1150 /* tcp_discardcb() clears the sack_holes up. */ 1151 uma_zdestroy(V_sack_hole_zone); 1152 uma_zdestroy(V_tcpcb_zone); 1153 1154 /* 1155 * Cannot free the zone until all tcpcbs are released as we attach 1156 * the allocations to them. 1157 */ 1158 tcp_fastopen_destroy(); 1159 1160 #ifdef TCP_HHOOK 1161 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1162 if (error != 0) { 1163 printf("%s: WARNING: unable to deregister helper hook " 1164 "type=%d, id=%d: error %d returned\n", __func__, 1165 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1166 } 1167 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1168 if (error != 0) { 1169 printf("%s: WARNING: unable to deregister helper hook " 1170 "type=%d, id=%d: error %d returned\n", __func__, 1171 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1172 } 1173 #endif 1174 } 1175 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1176 #endif 1177 1178 void 1179 tcp_fini(void *xtp) 1180 { 1181 1182 } 1183 1184 /* 1185 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1186 * tcp_template used to store this data in mbufs, but we now recopy it out 1187 * of the tcpcb each time to conserve mbufs. 1188 */ 1189 void 1190 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1191 { 1192 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1193 1194 INP_WLOCK_ASSERT(inp); 1195 1196 #ifdef INET6 1197 if ((inp->inp_vflag & INP_IPV6) != 0) { 1198 struct ip6_hdr *ip6; 1199 1200 ip6 = (struct ip6_hdr *)ip_ptr; 1201 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1202 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1203 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1204 (IPV6_VERSION & IPV6_VERSION_MASK); 1205 ip6->ip6_nxt = IPPROTO_TCP; 1206 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1207 ip6->ip6_src = inp->in6p_laddr; 1208 ip6->ip6_dst = inp->in6p_faddr; 1209 } 1210 #endif /* INET6 */ 1211 #if defined(INET6) && defined(INET) 1212 else 1213 #endif 1214 #ifdef INET 1215 { 1216 struct ip *ip; 1217 1218 ip = (struct ip *)ip_ptr; 1219 ip->ip_v = IPVERSION; 1220 ip->ip_hl = 5; 1221 ip->ip_tos = inp->inp_ip_tos; 1222 ip->ip_len = 0; 1223 ip->ip_id = 0; 1224 ip->ip_off = 0; 1225 ip->ip_ttl = inp->inp_ip_ttl; 1226 ip->ip_sum = 0; 1227 ip->ip_p = IPPROTO_TCP; 1228 ip->ip_src = inp->inp_laddr; 1229 ip->ip_dst = inp->inp_faddr; 1230 } 1231 #endif /* INET */ 1232 th->th_sport = inp->inp_lport; 1233 th->th_dport = inp->inp_fport; 1234 th->th_seq = 0; 1235 th->th_ack = 0; 1236 th->th_x2 = 0; 1237 th->th_off = 5; 1238 th->th_flags = 0; 1239 th->th_win = 0; 1240 th->th_urp = 0; 1241 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1242 } 1243 1244 /* 1245 * Create template to be used to send tcp packets on a connection. 1246 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1247 * use for this function is in keepalives, which use tcp_respond. 1248 */ 1249 struct tcptemp * 1250 tcpip_maketemplate(struct inpcb *inp) 1251 { 1252 struct tcptemp *t; 1253 1254 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1255 if (t == NULL) 1256 return (NULL); 1257 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1258 return (t); 1259 } 1260 1261 /* 1262 * Send a single message to the TCP at address specified by 1263 * the given TCP/IP header. If m == NULL, then we make a copy 1264 * of the tcpiphdr at th and send directly to the addressed host. 1265 * This is used to force keep alive messages out using the TCP 1266 * template for a connection. If flags are given then we send 1267 * a message back to the TCP which originated the segment th, 1268 * and discard the mbuf containing it and any other attached mbufs. 1269 * 1270 * In any case the ack and sequence number of the transmitted 1271 * segment are as specified by the parameters. 1272 * 1273 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1274 */ 1275 void 1276 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1277 tcp_seq ack, tcp_seq seq, int flags) 1278 { 1279 struct tcpopt to; 1280 struct inpcb *inp; 1281 struct ip *ip; 1282 struct mbuf *optm; 1283 struct tcphdr *nth; 1284 u_char *optp; 1285 #ifdef INET6 1286 struct ip6_hdr *ip6; 1287 int isipv6; 1288 #endif /* INET6 */ 1289 int optlen, tlen, win; 1290 bool incl_opts; 1291 1292 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1293 1294 #ifdef INET6 1295 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1296 ip6 = ipgen; 1297 #endif /* INET6 */ 1298 ip = ipgen; 1299 1300 if (tp != NULL) { 1301 inp = tp->t_inpcb; 1302 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1303 INP_WLOCK_ASSERT(inp); 1304 } else 1305 inp = NULL; 1306 1307 incl_opts = false; 1308 win = 0; 1309 if (tp != NULL) { 1310 if (!(flags & TH_RST)) { 1311 win = sbspace(&inp->inp_socket->so_rcv); 1312 if (win > TCP_MAXWIN << tp->rcv_scale) 1313 win = TCP_MAXWIN << tp->rcv_scale; 1314 } 1315 if ((tp->t_flags & TF_NOOPT) == 0) 1316 incl_opts = true; 1317 } 1318 if (m == NULL) { 1319 m = m_gethdr(M_NOWAIT, MT_DATA); 1320 if (m == NULL) 1321 return; 1322 m->m_data += max_linkhdr; 1323 #ifdef INET6 1324 if (isipv6) { 1325 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1326 sizeof(struct ip6_hdr)); 1327 ip6 = mtod(m, struct ip6_hdr *); 1328 nth = (struct tcphdr *)(ip6 + 1); 1329 } else 1330 #endif /* INET6 */ 1331 { 1332 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1333 ip = mtod(m, struct ip *); 1334 nth = (struct tcphdr *)(ip + 1); 1335 } 1336 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1337 flags = TH_ACK; 1338 } else if (!M_WRITABLE(m)) { 1339 struct mbuf *n; 1340 1341 /* Can't reuse 'm', allocate a new mbuf. */ 1342 n = m_gethdr(M_NOWAIT, MT_DATA); 1343 if (n == NULL) { 1344 m_freem(m); 1345 return; 1346 } 1347 1348 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1349 m_freem(m); 1350 m_freem(n); 1351 return; 1352 } 1353 1354 n->m_data += max_linkhdr; 1355 /* m_len is set later */ 1356 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1357 #ifdef INET6 1358 if (isipv6) { 1359 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1360 sizeof(struct ip6_hdr)); 1361 ip6 = mtod(n, struct ip6_hdr *); 1362 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1363 nth = (struct tcphdr *)(ip6 + 1); 1364 } else 1365 #endif /* INET6 */ 1366 { 1367 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1368 ip = mtod(n, struct ip *); 1369 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1370 nth = (struct tcphdr *)(ip + 1); 1371 } 1372 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1373 xchg(nth->th_dport, nth->th_sport, uint16_t); 1374 th = nth; 1375 m_freem(m); 1376 m = n; 1377 } else { 1378 /* 1379 * reuse the mbuf. 1380 * XXX MRT We inherit the FIB, which is lucky. 1381 */ 1382 m_freem(m->m_next); 1383 m->m_next = NULL; 1384 m->m_data = (caddr_t)ipgen; 1385 /* m_len is set later */ 1386 #ifdef INET6 1387 if (isipv6) { 1388 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1389 nth = (struct tcphdr *)(ip6 + 1); 1390 } else 1391 #endif /* INET6 */ 1392 { 1393 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1394 nth = (struct tcphdr *)(ip + 1); 1395 } 1396 if (th != nth) { 1397 /* 1398 * this is usually a case when an extension header 1399 * exists between the IPv6 header and the 1400 * TCP header. 1401 */ 1402 nth->th_sport = th->th_sport; 1403 nth->th_dport = th->th_dport; 1404 } 1405 xchg(nth->th_dport, nth->th_sport, uint16_t); 1406 #undef xchg 1407 } 1408 tlen = 0; 1409 #ifdef INET6 1410 if (isipv6) 1411 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1412 #endif 1413 #if defined(INET) && defined(INET6) 1414 else 1415 #endif 1416 #ifdef INET 1417 tlen = sizeof (struct tcpiphdr); 1418 #endif 1419 #ifdef INVARIANTS 1420 m->m_len = 0; 1421 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1422 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1423 m, tlen, (long)M_TRAILINGSPACE(m))); 1424 #endif 1425 m->m_len = tlen; 1426 to.to_flags = 0; 1427 if (incl_opts) { 1428 /* Make sure we have room. */ 1429 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1430 m->m_next = m_get(M_NOWAIT, MT_DATA); 1431 if (m->m_next) { 1432 optp = mtod(m->m_next, u_char *); 1433 optm = m->m_next; 1434 } else 1435 incl_opts = false; 1436 } else { 1437 optp = (u_char *) (nth + 1); 1438 optm = m; 1439 } 1440 } 1441 if (incl_opts) { 1442 /* Timestamps. */ 1443 if (tp->t_flags & TF_RCVD_TSTMP) { 1444 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1445 to.to_tsecr = tp->ts_recent; 1446 to.to_flags |= TOF_TS; 1447 } 1448 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1449 /* TCP-MD5 (RFC2385). */ 1450 if (tp->t_flags & TF_SIGNATURE) 1451 to.to_flags |= TOF_SIGNATURE; 1452 #endif 1453 /* Add the options. */ 1454 tlen += optlen = tcp_addoptions(&to, optp); 1455 1456 /* Update m_len in the correct mbuf. */ 1457 optm->m_len += optlen; 1458 } else 1459 optlen = 0; 1460 #ifdef INET6 1461 if (isipv6) { 1462 ip6->ip6_flow = 0; 1463 ip6->ip6_vfc = IPV6_VERSION; 1464 ip6->ip6_nxt = IPPROTO_TCP; 1465 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1466 } 1467 #endif 1468 #if defined(INET) && defined(INET6) 1469 else 1470 #endif 1471 #ifdef INET 1472 { 1473 ip->ip_len = htons(tlen); 1474 ip->ip_ttl = V_ip_defttl; 1475 if (V_path_mtu_discovery) 1476 ip->ip_off |= htons(IP_DF); 1477 } 1478 #endif 1479 m->m_pkthdr.len = tlen; 1480 m->m_pkthdr.rcvif = NULL; 1481 #ifdef MAC 1482 if (inp != NULL) { 1483 /* 1484 * Packet is associated with a socket, so allow the 1485 * label of the response to reflect the socket label. 1486 */ 1487 INP_WLOCK_ASSERT(inp); 1488 mac_inpcb_create_mbuf(inp, m); 1489 } else { 1490 /* 1491 * Packet is not associated with a socket, so possibly 1492 * update the label in place. 1493 */ 1494 mac_netinet_tcp_reply(m); 1495 } 1496 #endif 1497 nth->th_seq = htonl(seq); 1498 nth->th_ack = htonl(ack); 1499 nth->th_x2 = 0; 1500 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1501 nth->th_flags = flags; 1502 if (tp != NULL) 1503 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1504 else 1505 nth->th_win = htons((u_short)win); 1506 nth->th_urp = 0; 1507 1508 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1509 if (to.to_flags & TOF_SIGNATURE) { 1510 if (!TCPMD5_ENABLED() || 1511 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1512 m_freem(m); 1513 return; 1514 } 1515 } 1516 #endif 1517 1518 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1519 #ifdef INET6 1520 if (isipv6) { 1521 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1522 nth->th_sum = in6_cksum_pseudo(ip6, 1523 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1524 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1525 NULL, NULL); 1526 } 1527 #endif /* INET6 */ 1528 #if defined(INET6) && defined(INET) 1529 else 1530 #endif 1531 #ifdef INET 1532 { 1533 m->m_pkthdr.csum_flags = CSUM_TCP; 1534 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1535 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1536 } 1537 #endif /* INET */ 1538 #ifdef TCPDEBUG 1539 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1540 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1541 #endif 1542 TCP_PROBE3(debug__output, tp, th, m); 1543 if (flags & TH_RST) 1544 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1545 1546 #ifdef INET6 1547 if (isipv6) { 1548 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1549 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1550 } 1551 #endif /* INET6 */ 1552 #if defined(INET) && defined(INET6) 1553 else 1554 #endif 1555 #ifdef INET 1556 { 1557 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1558 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1559 } 1560 #endif 1561 } 1562 1563 /* 1564 * Create a new TCP control block, making an 1565 * empty reassembly queue and hooking it to the argument 1566 * protocol control block. The `inp' parameter must have 1567 * come from the zone allocator set up in tcp_init(). 1568 */ 1569 struct tcpcb * 1570 tcp_newtcpcb(struct inpcb *inp) 1571 { 1572 struct tcpcb_mem *tm; 1573 struct tcpcb *tp; 1574 #ifdef INET6 1575 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1576 #endif /* INET6 */ 1577 1578 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1579 if (tm == NULL) 1580 return (NULL); 1581 tp = &tm->tcb; 1582 1583 /* Initialise cc_var struct for this tcpcb. */ 1584 tp->ccv = &tm->ccv; 1585 tp->ccv->type = IPPROTO_TCP; 1586 tp->ccv->ccvc.tcp = tp; 1587 rw_rlock(&tcp_function_lock); 1588 tp->t_fb = tcp_func_set_ptr; 1589 refcount_acquire(&tp->t_fb->tfb_refcnt); 1590 rw_runlock(&tcp_function_lock); 1591 /* 1592 * Use the current system default CC algorithm. 1593 */ 1594 CC_LIST_RLOCK(); 1595 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1596 CC_ALGO(tp) = CC_DEFAULT(); 1597 CC_LIST_RUNLOCK(); 1598 1599 if (CC_ALGO(tp)->cb_init != NULL) 1600 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1601 if (tp->t_fb->tfb_tcp_fb_fini) 1602 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1603 refcount_release(&tp->t_fb->tfb_refcnt); 1604 uma_zfree(V_tcpcb_zone, tm); 1605 return (NULL); 1606 } 1607 1608 #ifdef TCP_HHOOK 1609 tp->osd = &tm->osd; 1610 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1611 if (tp->t_fb->tfb_tcp_fb_fini) 1612 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1613 refcount_release(&tp->t_fb->tfb_refcnt); 1614 uma_zfree(V_tcpcb_zone, tm); 1615 return (NULL); 1616 } 1617 #endif 1618 1619 #ifdef VIMAGE 1620 tp->t_vnet = inp->inp_vnet; 1621 #endif 1622 tp->t_timers = &tm->tt; 1623 TAILQ_INIT(&tp->t_segq); 1624 tp->t_maxseg = 1625 #ifdef INET6 1626 isipv6 ? V_tcp_v6mssdflt : 1627 #endif /* INET6 */ 1628 V_tcp_mssdflt; 1629 1630 /* Set up our timeouts. */ 1631 callout_init(&tp->t_timers->tt_rexmt, 1); 1632 callout_init(&tp->t_timers->tt_persist, 1); 1633 callout_init(&tp->t_timers->tt_keep, 1); 1634 callout_init(&tp->t_timers->tt_2msl, 1); 1635 callout_init(&tp->t_timers->tt_delack, 1); 1636 1637 if (V_tcp_do_rfc1323) 1638 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1639 if (V_tcp_do_sack) 1640 tp->t_flags |= TF_SACK_PERMIT; 1641 TAILQ_INIT(&tp->snd_holes); 1642 /* 1643 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1644 * is called. 1645 */ 1646 in_pcbref(inp); /* Reference for tcpcb */ 1647 tp->t_inpcb = inp; 1648 1649 /* 1650 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1651 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1652 * reasonable initial retransmit time. 1653 */ 1654 tp->t_srtt = TCPTV_SRTTBASE; 1655 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1656 tp->t_rttmin = tcp_rexmit_min; 1657 tp->t_rxtcur = tcp_rexmit_initial; 1658 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1659 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1660 tp->t_rcvtime = ticks; 1661 /* 1662 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1663 * because the socket may be bound to an IPv6 wildcard address, 1664 * which may match an IPv4-mapped IPv6 address. 1665 */ 1666 inp->inp_ip_ttl = V_ip_defttl; 1667 inp->inp_ppcb = tp; 1668 #ifdef TCPPCAP 1669 /* 1670 * Init the TCP PCAP queues. 1671 */ 1672 tcp_pcap_tcpcb_init(tp); 1673 #endif 1674 #ifdef TCP_BLACKBOX 1675 /* Initialize the per-TCPCB log data. */ 1676 tcp_log_tcpcbinit(tp); 1677 #endif 1678 if (tp->t_fb->tfb_tcp_fb_init) { 1679 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1680 } 1681 return (tp); /* XXX */ 1682 } 1683 1684 /* 1685 * Switch the congestion control algorithm back to NewReno for any active 1686 * control blocks using an algorithm which is about to go away. 1687 * This ensures the CC framework can allow the unload to proceed without leaving 1688 * any dangling pointers which would trigger a panic. 1689 * Returning non-zero would inform the CC framework that something went wrong 1690 * and it would be unsafe to allow the unload to proceed. However, there is no 1691 * way for this to occur with this implementation so we always return zero. 1692 */ 1693 int 1694 tcp_ccalgounload(struct cc_algo *unload_algo) 1695 { 1696 struct cc_algo *tmpalgo; 1697 struct inpcb *inp; 1698 struct tcpcb *tp; 1699 VNET_ITERATOR_DECL(vnet_iter); 1700 1701 /* 1702 * Check all active control blocks across all network stacks and change 1703 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1704 * requires cleanup code to be run, call it. 1705 */ 1706 VNET_LIST_RLOCK(); 1707 VNET_FOREACH(vnet_iter) { 1708 CURVNET_SET(vnet_iter); 1709 INP_INFO_WLOCK(&V_tcbinfo); 1710 /* 1711 * New connections already part way through being initialised 1712 * with the CC algo we're removing will not race with this code 1713 * because the INP_INFO_WLOCK is held during initialisation. We 1714 * therefore don't enter the loop below until the connection 1715 * list has stabilised. 1716 */ 1717 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 1718 INP_WLOCK(inp); 1719 /* Important to skip tcptw structs. */ 1720 if (!(inp->inp_flags & INP_TIMEWAIT) && 1721 (tp = intotcpcb(inp)) != NULL) { 1722 /* 1723 * By holding INP_WLOCK here, we are assured 1724 * that the connection is not currently 1725 * executing inside the CC module's functions 1726 * i.e. it is safe to make the switch back to 1727 * NewReno. 1728 */ 1729 if (CC_ALGO(tp) == unload_algo) { 1730 tmpalgo = CC_ALGO(tp); 1731 if (tmpalgo->cb_destroy != NULL) 1732 tmpalgo->cb_destroy(tp->ccv); 1733 CC_DATA(tp) = NULL; 1734 /* 1735 * NewReno may allocate memory on 1736 * demand for certain stateful 1737 * configuration as needed, but is 1738 * coded to never fail on memory 1739 * allocation failure so it is a safe 1740 * fallback. 1741 */ 1742 CC_ALGO(tp) = &newreno_cc_algo; 1743 } 1744 } 1745 INP_WUNLOCK(inp); 1746 } 1747 INP_INFO_WUNLOCK(&V_tcbinfo); 1748 CURVNET_RESTORE(); 1749 } 1750 VNET_LIST_RUNLOCK(); 1751 1752 return (0); 1753 } 1754 1755 /* 1756 * Drop a TCP connection, reporting 1757 * the specified error. If connection is synchronized, 1758 * then send a RST to peer. 1759 */ 1760 struct tcpcb * 1761 tcp_drop(struct tcpcb *tp, int errno) 1762 { 1763 struct socket *so = tp->t_inpcb->inp_socket; 1764 1765 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1766 INP_WLOCK_ASSERT(tp->t_inpcb); 1767 1768 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1769 tcp_state_change(tp, TCPS_CLOSED); 1770 (void) tp->t_fb->tfb_tcp_output(tp); 1771 TCPSTAT_INC(tcps_drops); 1772 } else 1773 TCPSTAT_INC(tcps_conndrops); 1774 if (errno == ETIMEDOUT && tp->t_softerror) 1775 errno = tp->t_softerror; 1776 so->so_error = errno; 1777 return (tcp_close(tp)); 1778 } 1779 1780 void 1781 tcp_discardcb(struct tcpcb *tp) 1782 { 1783 struct inpcb *inp = tp->t_inpcb; 1784 struct socket *so = inp->inp_socket; 1785 #ifdef INET6 1786 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1787 #endif /* INET6 */ 1788 int released __unused; 1789 1790 INP_WLOCK_ASSERT(inp); 1791 1792 /* 1793 * Make sure that all of our timers are stopped before we delete the 1794 * PCB. 1795 * 1796 * If stopping a timer fails, we schedule a discard function in same 1797 * callout, and the last discard function called will take care of 1798 * deleting the tcpcb. 1799 */ 1800 tp->t_timers->tt_draincnt = 0; 1801 tcp_timer_stop(tp, TT_REXMT); 1802 tcp_timer_stop(tp, TT_PERSIST); 1803 tcp_timer_stop(tp, TT_KEEP); 1804 tcp_timer_stop(tp, TT_2MSL); 1805 tcp_timer_stop(tp, TT_DELACK); 1806 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1807 /* 1808 * Call the stop-all function of the methods, 1809 * this function should call the tcp_timer_stop() 1810 * method with each of the function specific timeouts. 1811 * That stop will be called via the tfb_tcp_timer_stop() 1812 * which should use the async drain function of the 1813 * callout system (see tcp_var.h). 1814 */ 1815 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1816 } 1817 1818 /* 1819 * If we got enough samples through the srtt filter, 1820 * save the rtt and rttvar in the routing entry. 1821 * 'Enough' is arbitrarily defined as 4 rtt samples. 1822 * 4 samples is enough for the srtt filter to converge 1823 * to within enough % of the correct value; fewer samples 1824 * and we could save a bogus rtt. The danger is not high 1825 * as tcp quickly recovers from everything. 1826 * XXX: Works very well but needs some more statistics! 1827 */ 1828 if (tp->t_rttupdated >= 4) { 1829 struct hc_metrics_lite metrics; 1830 uint32_t ssthresh; 1831 1832 bzero(&metrics, sizeof(metrics)); 1833 /* 1834 * Update the ssthresh always when the conditions below 1835 * are satisfied. This gives us better new start value 1836 * for the congestion avoidance for new connections. 1837 * ssthresh is only set if packet loss occurred on a session. 1838 * 1839 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1840 * being torn down. Ideally this code would not use 'so'. 1841 */ 1842 ssthresh = tp->snd_ssthresh; 1843 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1844 /* 1845 * convert the limit from user data bytes to 1846 * packets then to packet data bytes. 1847 */ 1848 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1849 if (ssthresh < 2) 1850 ssthresh = 2; 1851 ssthresh *= (tp->t_maxseg + 1852 #ifdef INET6 1853 (isipv6 ? sizeof (struct ip6_hdr) + 1854 sizeof (struct tcphdr) : 1855 #endif 1856 sizeof (struct tcpiphdr) 1857 #ifdef INET6 1858 ) 1859 #endif 1860 ); 1861 } else 1862 ssthresh = 0; 1863 metrics.rmx_ssthresh = ssthresh; 1864 1865 metrics.rmx_rtt = tp->t_srtt; 1866 metrics.rmx_rttvar = tp->t_rttvar; 1867 metrics.rmx_cwnd = tp->snd_cwnd; 1868 metrics.rmx_sendpipe = 0; 1869 metrics.rmx_recvpipe = 0; 1870 1871 tcp_hc_update(&inp->inp_inc, &metrics); 1872 } 1873 1874 /* free the reassembly queue, if any */ 1875 tcp_reass_flush(tp); 1876 1877 #ifdef TCP_OFFLOAD 1878 /* Disconnect offload device, if any. */ 1879 if (tp->t_flags & TF_TOE) 1880 tcp_offload_detach(tp); 1881 #endif 1882 1883 tcp_free_sackholes(tp); 1884 1885 #ifdef TCPPCAP 1886 /* Free the TCP PCAP queues. */ 1887 tcp_pcap_drain(&(tp->t_inpkts)); 1888 tcp_pcap_drain(&(tp->t_outpkts)); 1889 #endif 1890 1891 /* Allow the CC algorithm to clean up after itself. */ 1892 if (CC_ALGO(tp)->cb_destroy != NULL) 1893 CC_ALGO(tp)->cb_destroy(tp->ccv); 1894 CC_DATA(tp) = NULL; 1895 1896 #ifdef TCP_HHOOK 1897 khelp_destroy_osd(tp->osd); 1898 #endif 1899 1900 CC_ALGO(tp) = NULL; 1901 inp->inp_ppcb = NULL; 1902 if (tp->t_timers->tt_draincnt == 0) { 1903 /* We own the last reference on tcpcb, let's free it. */ 1904 #ifdef TCP_BLACKBOX 1905 tcp_log_tcpcbfini(tp); 1906 #endif 1907 TCPSTATES_DEC(tp->t_state); 1908 if (tp->t_fb->tfb_tcp_fb_fini) 1909 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1910 refcount_release(&tp->t_fb->tfb_refcnt); 1911 tp->t_inpcb = NULL; 1912 uma_zfree(V_tcpcb_zone, tp); 1913 released = in_pcbrele_wlocked(inp); 1914 KASSERT(!released, ("%s: inp %p should not have been released " 1915 "here", __func__, inp)); 1916 } 1917 } 1918 1919 void 1920 tcp_timer_discard(void *ptp) 1921 { 1922 struct inpcb *inp; 1923 struct tcpcb *tp; 1924 struct epoch_tracker et; 1925 1926 tp = (struct tcpcb *)ptp; 1927 CURVNET_SET(tp->t_vnet); 1928 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 1929 inp = tp->t_inpcb; 1930 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1931 __func__, tp)); 1932 INP_WLOCK(inp); 1933 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1934 ("%s: tcpcb has to be stopped here", __func__)); 1935 tp->t_timers->tt_draincnt--; 1936 if (tp->t_timers->tt_draincnt == 0) { 1937 /* We own the last reference on this tcpcb, let's free it. */ 1938 #ifdef TCP_BLACKBOX 1939 tcp_log_tcpcbfini(tp); 1940 #endif 1941 TCPSTATES_DEC(tp->t_state); 1942 if (tp->t_fb->tfb_tcp_fb_fini) 1943 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1944 refcount_release(&tp->t_fb->tfb_refcnt); 1945 tp->t_inpcb = NULL; 1946 uma_zfree(V_tcpcb_zone, tp); 1947 if (in_pcbrele_wlocked(inp)) { 1948 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1949 CURVNET_RESTORE(); 1950 return; 1951 } 1952 } 1953 INP_WUNLOCK(inp); 1954 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1955 CURVNET_RESTORE(); 1956 } 1957 1958 /* 1959 * Attempt to close a TCP control block, marking it as dropped, and freeing 1960 * the socket if we hold the only reference. 1961 */ 1962 struct tcpcb * 1963 tcp_close(struct tcpcb *tp) 1964 { 1965 struct inpcb *inp = tp->t_inpcb; 1966 struct socket *so; 1967 1968 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1969 INP_WLOCK_ASSERT(inp); 1970 1971 #ifdef TCP_OFFLOAD 1972 if (tp->t_state == TCPS_LISTEN) 1973 tcp_offload_listen_stop(tp); 1974 #endif 1975 /* 1976 * This releases the TFO pending counter resource for TFO listen 1977 * sockets as well as passively-created TFO sockets that transition 1978 * from SYN_RECEIVED to CLOSED. 1979 */ 1980 if (tp->t_tfo_pending) { 1981 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1982 tp->t_tfo_pending = NULL; 1983 } 1984 in_pcbdrop(inp); 1985 TCPSTAT_INC(tcps_closed); 1986 if (tp->t_state != TCPS_CLOSED) 1987 tcp_state_change(tp, TCPS_CLOSED); 1988 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1989 so = inp->inp_socket; 1990 soisdisconnected(so); 1991 if (inp->inp_flags & INP_SOCKREF) { 1992 KASSERT(so->so_state & SS_PROTOREF, 1993 ("tcp_close: !SS_PROTOREF")); 1994 inp->inp_flags &= ~INP_SOCKREF; 1995 INP_WUNLOCK(inp); 1996 SOCK_LOCK(so); 1997 so->so_state &= ~SS_PROTOREF; 1998 sofree(so); 1999 return (NULL); 2000 } 2001 return (tp); 2002 } 2003 2004 void 2005 tcp_drain(void) 2006 { 2007 VNET_ITERATOR_DECL(vnet_iter); 2008 2009 if (!do_tcpdrain) 2010 return; 2011 2012 VNET_LIST_RLOCK_NOSLEEP(); 2013 VNET_FOREACH(vnet_iter) { 2014 CURVNET_SET(vnet_iter); 2015 struct inpcb *inpb; 2016 struct tcpcb *tcpb; 2017 2018 /* 2019 * Walk the tcpbs, if existing, and flush the reassembly queue, 2020 * if there is one... 2021 * XXX: The "Net/3" implementation doesn't imply that the TCP 2022 * reassembly queue should be flushed, but in a situation 2023 * where we're really low on mbufs, this is potentially 2024 * useful. 2025 */ 2026 INP_INFO_WLOCK(&V_tcbinfo); 2027 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2028 INP_WLOCK(inpb); 2029 if (inpb->inp_flags & INP_TIMEWAIT) { 2030 INP_WUNLOCK(inpb); 2031 continue; 2032 } 2033 if ((tcpb = intotcpcb(inpb)) != NULL) { 2034 tcp_reass_flush(tcpb); 2035 tcp_clean_sackreport(tcpb); 2036 #ifdef TCP_BLACKBOX 2037 tcp_log_drain(tcpb); 2038 #endif 2039 #ifdef TCPPCAP 2040 if (tcp_pcap_aggressive_free) { 2041 /* Free the TCP PCAP queues. */ 2042 tcp_pcap_drain(&(tcpb->t_inpkts)); 2043 tcp_pcap_drain(&(tcpb->t_outpkts)); 2044 } 2045 #endif 2046 } 2047 INP_WUNLOCK(inpb); 2048 } 2049 INP_INFO_WUNLOCK(&V_tcbinfo); 2050 CURVNET_RESTORE(); 2051 } 2052 VNET_LIST_RUNLOCK_NOSLEEP(); 2053 } 2054 2055 /* 2056 * Notify a tcp user of an asynchronous error; 2057 * store error as soft error, but wake up user 2058 * (for now, won't do anything until can select for soft error). 2059 * 2060 * Do not wake up user since there currently is no mechanism for 2061 * reporting soft errors (yet - a kqueue filter may be added). 2062 */ 2063 static struct inpcb * 2064 tcp_notify(struct inpcb *inp, int error) 2065 { 2066 struct tcpcb *tp; 2067 2068 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2069 INP_WLOCK_ASSERT(inp); 2070 2071 if ((inp->inp_flags & INP_TIMEWAIT) || 2072 (inp->inp_flags & INP_DROPPED)) 2073 return (inp); 2074 2075 tp = intotcpcb(inp); 2076 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2077 2078 /* 2079 * Ignore some errors if we are hooked up. 2080 * If connection hasn't completed, has retransmitted several times, 2081 * and receives a second error, give up now. This is better 2082 * than waiting a long time to establish a connection that 2083 * can never complete. 2084 */ 2085 if (tp->t_state == TCPS_ESTABLISHED && 2086 (error == EHOSTUNREACH || error == ENETUNREACH || 2087 error == EHOSTDOWN)) { 2088 if (inp->inp_route.ro_rt) { 2089 RTFREE(inp->inp_route.ro_rt); 2090 inp->inp_route.ro_rt = (struct rtentry *)NULL; 2091 } 2092 return (inp); 2093 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2094 tp->t_softerror) { 2095 tp = tcp_drop(tp, error); 2096 if (tp != NULL) 2097 return (inp); 2098 else 2099 return (NULL); 2100 } else { 2101 tp->t_softerror = error; 2102 return (inp); 2103 } 2104 #if 0 2105 wakeup( &so->so_timeo); 2106 sorwakeup(so); 2107 sowwakeup(so); 2108 #endif 2109 } 2110 2111 static int 2112 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2113 { 2114 int error, i, m, n, pcb_count; 2115 struct inpcb *inp, **inp_list; 2116 inp_gen_t gencnt; 2117 struct xinpgen xig; 2118 struct epoch_tracker et; 2119 2120 /* 2121 * The process of preparing the TCB list is too time-consuming and 2122 * resource-intensive to repeat twice on every request. 2123 */ 2124 if (req->oldptr == NULL) { 2125 n = V_tcbinfo.ipi_count + 2126 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2127 n += imax(n / 8, 10); 2128 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2129 return (0); 2130 } 2131 2132 if (req->newptr != NULL) 2133 return (EPERM); 2134 2135 /* 2136 * OK, now we're committed to doing something. 2137 */ 2138 INP_LIST_RLOCK(&V_tcbinfo); 2139 gencnt = V_tcbinfo.ipi_gencnt; 2140 n = V_tcbinfo.ipi_count; 2141 INP_LIST_RUNLOCK(&V_tcbinfo); 2142 2143 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2144 2145 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 2146 + (n + m) * sizeof(struct xtcpcb)); 2147 if (error != 0) 2148 return (error); 2149 2150 bzero(&xig, sizeof(xig)); 2151 xig.xig_len = sizeof xig; 2152 xig.xig_count = n + m; 2153 xig.xig_gen = gencnt; 2154 xig.xig_sogen = so_gencnt; 2155 error = SYSCTL_OUT(req, &xig, sizeof xig); 2156 if (error) 2157 return (error); 2158 2159 error = syncache_pcblist(req, m, &pcb_count); 2160 if (error) 2161 return (error); 2162 2163 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 2164 2165 INP_INFO_WLOCK(&V_tcbinfo); 2166 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 2167 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) { 2168 INP_WLOCK(inp); 2169 if (inp->inp_gencnt <= gencnt) { 2170 /* 2171 * XXX: This use of cr_cansee(), introduced with 2172 * TCP state changes, is not quite right, but for 2173 * now, better than nothing. 2174 */ 2175 if (inp->inp_flags & INP_TIMEWAIT) { 2176 if (intotw(inp) != NULL) 2177 error = cr_cansee(req->td->td_ucred, 2178 intotw(inp)->tw_cred); 2179 else 2180 error = EINVAL; /* Skip this inp. */ 2181 } else 2182 error = cr_canseeinpcb(req->td->td_ucred, inp); 2183 if (error == 0) { 2184 in_pcbref(inp); 2185 inp_list[i++] = inp; 2186 } 2187 } 2188 INP_WUNLOCK(inp); 2189 } 2190 INP_INFO_WUNLOCK(&V_tcbinfo); 2191 n = i; 2192 2193 error = 0; 2194 for (i = 0; i < n; i++) { 2195 inp = inp_list[i]; 2196 INP_RLOCK(inp); 2197 if (inp->inp_gencnt <= gencnt) { 2198 struct xtcpcb xt; 2199 2200 tcp_inptoxtp(inp, &xt); 2201 INP_RUNLOCK(inp); 2202 error = SYSCTL_OUT(req, &xt, sizeof xt); 2203 } else 2204 INP_RUNLOCK(inp); 2205 } 2206 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2207 for (i = 0; i < n; i++) { 2208 inp = inp_list[i]; 2209 INP_RLOCK(inp); 2210 if (!in_pcbrele_rlocked(inp)) 2211 INP_RUNLOCK(inp); 2212 } 2213 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2214 2215 if (!error) { 2216 /* 2217 * Give the user an updated idea of our state. 2218 * If the generation differs from what we told 2219 * her before, she knows that something happened 2220 * while we were processing this request, and it 2221 * might be necessary to retry. 2222 */ 2223 INP_LIST_RLOCK(&V_tcbinfo); 2224 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2225 xig.xig_sogen = so_gencnt; 2226 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 2227 INP_LIST_RUNLOCK(&V_tcbinfo); 2228 error = SYSCTL_OUT(req, &xig, sizeof xig); 2229 } 2230 free(inp_list, M_TEMP); 2231 return (error); 2232 } 2233 2234 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2235 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 2236 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 2237 2238 #ifdef INET 2239 static int 2240 tcp_getcred(SYSCTL_HANDLER_ARGS) 2241 { 2242 struct xucred xuc; 2243 struct sockaddr_in addrs[2]; 2244 struct inpcb *inp; 2245 int error; 2246 2247 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2248 if (error) 2249 return (error); 2250 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2251 if (error) 2252 return (error); 2253 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2254 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2255 if (inp != NULL) { 2256 if (inp->inp_socket == NULL) 2257 error = ENOENT; 2258 if (error == 0) 2259 error = cr_canseeinpcb(req->td->td_ucred, inp); 2260 if (error == 0) 2261 cru2x(inp->inp_cred, &xuc); 2262 INP_RUNLOCK(inp); 2263 } else 2264 error = ENOENT; 2265 if (error == 0) 2266 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2267 return (error); 2268 } 2269 2270 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2271 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2272 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2273 #endif /* INET */ 2274 2275 #ifdef INET6 2276 static int 2277 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2278 { 2279 struct xucred xuc; 2280 struct sockaddr_in6 addrs[2]; 2281 struct inpcb *inp; 2282 int error; 2283 #ifdef INET 2284 int mapped = 0; 2285 #endif 2286 2287 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2288 if (error) 2289 return (error); 2290 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2291 if (error) 2292 return (error); 2293 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2294 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2295 return (error); 2296 } 2297 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2298 #ifdef INET 2299 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2300 mapped = 1; 2301 else 2302 #endif 2303 return (EINVAL); 2304 } 2305 2306 #ifdef INET 2307 if (mapped == 1) 2308 inp = in_pcblookup(&V_tcbinfo, 2309 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2310 addrs[1].sin6_port, 2311 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2312 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2313 else 2314 #endif 2315 inp = in6_pcblookup(&V_tcbinfo, 2316 &addrs[1].sin6_addr, addrs[1].sin6_port, 2317 &addrs[0].sin6_addr, addrs[0].sin6_port, 2318 INPLOOKUP_RLOCKPCB, NULL); 2319 if (inp != NULL) { 2320 if (inp->inp_socket == NULL) 2321 error = ENOENT; 2322 if (error == 0) 2323 error = cr_canseeinpcb(req->td->td_ucred, inp); 2324 if (error == 0) 2325 cru2x(inp->inp_cred, &xuc); 2326 INP_RUNLOCK(inp); 2327 } else 2328 error = ENOENT; 2329 if (error == 0) 2330 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2331 return (error); 2332 } 2333 2334 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2335 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2336 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2337 #endif /* INET6 */ 2338 2339 2340 #ifdef INET 2341 void 2342 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2343 { 2344 struct ip *ip = vip; 2345 struct tcphdr *th; 2346 struct in_addr faddr; 2347 struct inpcb *inp; 2348 struct tcpcb *tp; 2349 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2350 struct icmp *icp; 2351 struct in_conninfo inc; 2352 struct epoch_tracker et; 2353 tcp_seq icmp_tcp_seq; 2354 int mtu; 2355 2356 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2357 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2358 return; 2359 2360 if (cmd == PRC_MSGSIZE) 2361 notify = tcp_mtudisc_notify; 2362 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2363 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2364 cmd == PRC_TIMXCEED_INTRANS) && ip) 2365 notify = tcp_drop_syn_sent; 2366 2367 /* 2368 * Hostdead is ugly because it goes linearly through all PCBs. 2369 * XXX: We never get this from ICMP, otherwise it makes an 2370 * excellent DoS attack on machines with many connections. 2371 */ 2372 else if (cmd == PRC_HOSTDEAD) 2373 ip = NULL; 2374 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2375 return; 2376 2377 if (ip == NULL) { 2378 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2379 return; 2380 } 2381 2382 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2383 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2384 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2385 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2386 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2387 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2388 /* signal EHOSTDOWN, as it flushes the cached route */ 2389 inp = (*notify)(inp, EHOSTDOWN); 2390 goto out; 2391 } 2392 icmp_tcp_seq = th->th_seq; 2393 if (inp != NULL) { 2394 if (!(inp->inp_flags & INP_TIMEWAIT) && 2395 !(inp->inp_flags & INP_DROPPED) && 2396 !(inp->inp_socket == NULL)) { 2397 tp = intotcpcb(inp); 2398 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2399 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2400 if (cmd == PRC_MSGSIZE) { 2401 /* 2402 * MTU discovery: 2403 * If we got a needfrag set the MTU 2404 * in the route to the suggested new 2405 * value (if given) and then notify. 2406 */ 2407 mtu = ntohs(icp->icmp_nextmtu); 2408 /* 2409 * If no alternative MTU was 2410 * proposed, try the next smaller 2411 * one. 2412 */ 2413 if (!mtu) 2414 mtu = ip_next_mtu( 2415 ntohs(ip->ip_len), 1); 2416 if (mtu < V_tcp_minmss + 2417 sizeof(struct tcpiphdr)) 2418 mtu = V_tcp_minmss + 2419 sizeof(struct tcpiphdr); 2420 /* 2421 * Only process the offered MTU if it 2422 * is smaller than the current one. 2423 */ 2424 if (mtu < tp->t_maxseg + 2425 sizeof(struct tcpiphdr)) { 2426 bzero(&inc, sizeof(inc)); 2427 inc.inc_faddr = faddr; 2428 inc.inc_fibnum = 2429 inp->inp_inc.inc_fibnum; 2430 tcp_hc_updatemtu(&inc, mtu); 2431 tcp_mtudisc(inp, mtu); 2432 } 2433 } else 2434 inp = (*notify)(inp, 2435 inetctlerrmap[cmd]); 2436 } 2437 } 2438 } else { 2439 bzero(&inc, sizeof(inc)); 2440 inc.inc_fport = th->th_dport; 2441 inc.inc_lport = th->th_sport; 2442 inc.inc_faddr = faddr; 2443 inc.inc_laddr = ip->ip_src; 2444 syncache_unreach(&inc, icmp_tcp_seq); 2445 } 2446 out: 2447 if (inp != NULL) 2448 INP_WUNLOCK(inp); 2449 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2450 } 2451 #endif /* INET */ 2452 2453 #ifdef INET6 2454 void 2455 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2456 { 2457 struct in6_addr *dst; 2458 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2459 struct ip6_hdr *ip6; 2460 struct mbuf *m; 2461 struct inpcb *inp; 2462 struct tcpcb *tp; 2463 struct icmp6_hdr *icmp6; 2464 struct ip6ctlparam *ip6cp = NULL; 2465 const struct sockaddr_in6 *sa6_src = NULL; 2466 struct in_conninfo inc; 2467 struct epoch_tracker et; 2468 struct tcp_ports { 2469 uint16_t th_sport; 2470 uint16_t th_dport; 2471 } t_ports; 2472 tcp_seq icmp_tcp_seq; 2473 unsigned int mtu; 2474 unsigned int off; 2475 2476 if (sa->sa_family != AF_INET6 || 2477 sa->sa_len != sizeof(struct sockaddr_in6)) 2478 return; 2479 2480 /* if the parameter is from icmp6, decode it. */ 2481 if (d != NULL) { 2482 ip6cp = (struct ip6ctlparam *)d; 2483 icmp6 = ip6cp->ip6c_icmp6; 2484 m = ip6cp->ip6c_m; 2485 ip6 = ip6cp->ip6c_ip6; 2486 off = ip6cp->ip6c_off; 2487 sa6_src = ip6cp->ip6c_src; 2488 dst = ip6cp->ip6c_finaldst; 2489 } else { 2490 m = NULL; 2491 ip6 = NULL; 2492 off = 0; /* fool gcc */ 2493 sa6_src = &sa6_any; 2494 dst = NULL; 2495 } 2496 2497 if (cmd == PRC_MSGSIZE) 2498 notify = tcp_mtudisc_notify; 2499 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2500 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2501 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2502 notify = tcp_drop_syn_sent; 2503 2504 /* 2505 * Hostdead is ugly because it goes linearly through all PCBs. 2506 * XXX: We never get this from ICMP, otherwise it makes an 2507 * excellent DoS attack on machines with many connections. 2508 */ 2509 else if (cmd == PRC_HOSTDEAD) 2510 ip6 = NULL; 2511 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2512 return; 2513 2514 if (ip6 == NULL) { 2515 in6_pcbnotify(&V_tcbinfo, sa, 0, 2516 (const struct sockaddr *)sa6_src, 2517 0, cmd, NULL, notify); 2518 return; 2519 } 2520 2521 /* Check if we can safely get the ports from the tcp hdr */ 2522 if (m == NULL || 2523 (m->m_pkthdr.len < 2524 (int32_t) (off + sizeof(struct tcp_ports)))) { 2525 return; 2526 } 2527 bzero(&t_ports, sizeof(struct tcp_ports)); 2528 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2529 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2530 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2531 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2532 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2533 /* signal EHOSTDOWN, as it flushes the cached route */ 2534 inp = (*notify)(inp, EHOSTDOWN); 2535 goto out; 2536 } 2537 off += sizeof(struct tcp_ports); 2538 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2539 goto out; 2540 } 2541 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2542 if (inp != NULL) { 2543 if (!(inp->inp_flags & INP_TIMEWAIT) && 2544 !(inp->inp_flags & INP_DROPPED) && 2545 !(inp->inp_socket == NULL)) { 2546 tp = intotcpcb(inp); 2547 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2548 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2549 if (cmd == PRC_MSGSIZE) { 2550 /* 2551 * MTU discovery: 2552 * If we got a needfrag set the MTU 2553 * in the route to the suggested new 2554 * value (if given) and then notify. 2555 */ 2556 mtu = ntohl(icmp6->icmp6_mtu); 2557 /* 2558 * If no alternative MTU was 2559 * proposed, or the proposed 2560 * MTU was too small, set to 2561 * the min. 2562 */ 2563 if (mtu < IPV6_MMTU) 2564 mtu = IPV6_MMTU - 8; 2565 bzero(&inc, sizeof(inc)); 2566 inc.inc_fibnum = M_GETFIB(m); 2567 inc.inc_flags |= INC_ISIPV6; 2568 inc.inc6_faddr = *dst; 2569 if (in6_setscope(&inc.inc6_faddr, 2570 m->m_pkthdr.rcvif, NULL)) 2571 goto out; 2572 /* 2573 * Only process the offered MTU if it 2574 * is smaller than the current one. 2575 */ 2576 if (mtu < tp->t_maxseg + 2577 sizeof (struct tcphdr) + 2578 sizeof (struct ip6_hdr)) { 2579 tcp_hc_updatemtu(&inc, mtu); 2580 tcp_mtudisc(inp, mtu); 2581 ICMP6STAT_INC(icp6s_pmtuchg); 2582 } 2583 } else 2584 inp = (*notify)(inp, 2585 inet6ctlerrmap[cmd]); 2586 } 2587 } 2588 } else { 2589 bzero(&inc, sizeof(inc)); 2590 inc.inc_fibnum = M_GETFIB(m); 2591 inc.inc_flags |= INC_ISIPV6; 2592 inc.inc_fport = t_ports.th_dport; 2593 inc.inc_lport = t_ports.th_sport; 2594 inc.inc6_faddr = *dst; 2595 inc.inc6_laddr = ip6->ip6_src; 2596 syncache_unreach(&inc, icmp_tcp_seq); 2597 } 2598 out: 2599 if (inp != NULL) 2600 INP_WUNLOCK(inp); 2601 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2602 } 2603 #endif /* INET6 */ 2604 2605 static uint32_t 2606 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 2607 { 2608 MD5_CTX ctx; 2609 uint32_t hash[4]; 2610 2611 MD5Init(&ctx); 2612 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 2613 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 2614 switch (inc->inc_flags & INC_ISIPV6) { 2615 #ifdef INET 2616 case 0: 2617 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 2618 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 2619 break; 2620 #endif 2621 #ifdef INET6 2622 case INC_ISIPV6: 2623 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 2624 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 2625 break; 2626 #endif 2627 } 2628 MD5Update(&ctx, key, len); 2629 MD5Final((unsigned char *)hash, &ctx); 2630 2631 return (hash[0]); 2632 } 2633 2634 uint32_t 2635 tcp_new_ts_offset(struct in_conninfo *inc) 2636 { 2637 return (tcp_keyed_hash(inc, V_ts_offset_secret, 2638 sizeof(V_ts_offset_secret))); 2639 } 2640 2641 /* 2642 * Following is where TCP initial sequence number generation occurs. 2643 * 2644 * There are two places where we must use initial sequence numbers: 2645 * 1. In SYN-ACK packets. 2646 * 2. In SYN packets. 2647 * 2648 * All ISNs for SYN-ACK packets are generated by the syncache. See 2649 * tcp_syncache.c for details. 2650 * 2651 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2652 * depends on this property. In addition, these ISNs should be 2653 * unguessable so as to prevent connection hijacking. To satisfy 2654 * the requirements of this situation, the algorithm outlined in 2655 * RFC 1948 is used, with only small modifications. 2656 * 2657 * Implementation details: 2658 * 2659 * Time is based off the system timer, and is corrected so that it 2660 * increases by one megabyte per second. This allows for proper 2661 * recycling on high speed LANs while still leaving over an hour 2662 * before rollover. 2663 * 2664 * As reading the *exact* system time is too expensive to be done 2665 * whenever setting up a TCP connection, we increment the time 2666 * offset in two ways. First, a small random positive increment 2667 * is added to isn_offset for each connection that is set up. 2668 * Second, the function tcp_isn_tick fires once per clock tick 2669 * and increments isn_offset as necessary so that sequence numbers 2670 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2671 * random positive increments serve only to ensure that the same 2672 * exact sequence number is never sent out twice (as could otherwise 2673 * happen when a port is recycled in less than the system tick 2674 * interval.) 2675 * 2676 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2677 * between seeding of isn_secret. This is normally set to zero, 2678 * as reseeding should not be necessary. 2679 * 2680 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2681 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 2682 * general, this means holding an exclusive (write) lock. 2683 */ 2684 2685 #define ISN_BYTES_PER_SECOND 1048576 2686 #define ISN_STATIC_INCREMENT 4096 2687 #define ISN_RANDOM_INCREMENT (4096 - 1) 2688 #define ISN_SECRET_LENGTH 32 2689 2690 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 2691 VNET_DEFINE_STATIC(int, isn_last); 2692 VNET_DEFINE_STATIC(int, isn_last_reseed); 2693 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 2694 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 2695 2696 #define V_isn_secret VNET(isn_secret) 2697 #define V_isn_last VNET(isn_last) 2698 #define V_isn_last_reseed VNET(isn_last_reseed) 2699 #define V_isn_offset VNET(isn_offset) 2700 #define V_isn_offset_old VNET(isn_offset_old) 2701 2702 tcp_seq 2703 tcp_new_isn(struct in_conninfo *inc) 2704 { 2705 tcp_seq new_isn; 2706 u_int32_t projected_offset; 2707 2708 ISN_LOCK(); 2709 /* Seed if this is the first use, reseed if requested. */ 2710 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2711 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2712 < (u_int)ticks))) { 2713 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 2714 V_isn_last_reseed = ticks; 2715 } 2716 2717 /* Compute the md5 hash and return the ISN. */ 2718 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 2719 sizeof(V_isn_secret)); 2720 V_isn_offset += ISN_STATIC_INCREMENT + 2721 (arc4random() & ISN_RANDOM_INCREMENT); 2722 if (ticks != V_isn_last) { 2723 projected_offset = V_isn_offset_old + 2724 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2725 if (SEQ_GT(projected_offset, V_isn_offset)) 2726 V_isn_offset = projected_offset; 2727 V_isn_offset_old = V_isn_offset; 2728 V_isn_last = ticks; 2729 } 2730 new_isn += V_isn_offset; 2731 ISN_UNLOCK(); 2732 return (new_isn); 2733 } 2734 2735 /* 2736 * When a specific ICMP unreachable message is received and the 2737 * connection state is SYN-SENT, drop the connection. This behavior 2738 * is controlled by the icmp_may_rst sysctl. 2739 */ 2740 struct inpcb * 2741 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2742 { 2743 struct tcpcb *tp; 2744 2745 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2746 INP_WLOCK_ASSERT(inp); 2747 2748 if ((inp->inp_flags & INP_TIMEWAIT) || 2749 (inp->inp_flags & INP_DROPPED)) 2750 return (inp); 2751 2752 tp = intotcpcb(inp); 2753 if (tp->t_state != TCPS_SYN_SENT) 2754 return (inp); 2755 2756 if (IS_FASTOPEN(tp->t_flags)) 2757 tcp_fastopen_disable_path(tp); 2758 2759 tp = tcp_drop(tp, errno); 2760 if (tp != NULL) 2761 return (inp); 2762 else 2763 return (NULL); 2764 } 2765 2766 /* 2767 * When `need fragmentation' ICMP is received, update our idea of the MSS 2768 * based on the new value. Also nudge TCP to send something, since we 2769 * know the packet we just sent was dropped. 2770 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2771 */ 2772 static struct inpcb * 2773 tcp_mtudisc_notify(struct inpcb *inp, int error) 2774 { 2775 2776 tcp_mtudisc(inp, -1); 2777 return (inp); 2778 } 2779 2780 static void 2781 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2782 { 2783 struct tcpcb *tp; 2784 struct socket *so; 2785 2786 INP_WLOCK_ASSERT(inp); 2787 if ((inp->inp_flags & INP_TIMEWAIT) || 2788 (inp->inp_flags & INP_DROPPED)) 2789 return; 2790 2791 tp = intotcpcb(inp); 2792 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2793 2794 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2795 2796 so = inp->inp_socket; 2797 SOCKBUF_LOCK(&so->so_snd); 2798 /* If the mss is larger than the socket buffer, decrease the mss. */ 2799 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2800 tp->t_maxseg = so->so_snd.sb_hiwat; 2801 SOCKBUF_UNLOCK(&so->so_snd); 2802 2803 TCPSTAT_INC(tcps_mturesent); 2804 tp->t_rtttime = 0; 2805 tp->snd_nxt = tp->snd_una; 2806 tcp_free_sackholes(tp); 2807 tp->snd_recover = tp->snd_max; 2808 if (tp->t_flags & TF_SACK_PERMIT) 2809 EXIT_FASTRECOVERY(tp->t_flags); 2810 tp->t_fb->tfb_tcp_output(tp); 2811 } 2812 2813 #ifdef INET 2814 /* 2815 * Look-up the routing entry to the peer of this inpcb. If no route 2816 * is found and it cannot be allocated, then return 0. This routine 2817 * is called by TCP routines that access the rmx structure and by 2818 * tcp_mss_update to get the peer/interface MTU. 2819 */ 2820 uint32_t 2821 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2822 { 2823 struct nhop4_extended nh4; 2824 struct ifnet *ifp; 2825 uint32_t maxmtu = 0; 2826 2827 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2828 2829 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2830 2831 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2832 NHR_REF, 0, &nh4) != 0) 2833 return (0); 2834 2835 ifp = nh4.nh_ifp; 2836 maxmtu = nh4.nh_mtu; 2837 2838 /* Report additional interface capabilities. */ 2839 if (cap != NULL) { 2840 if (ifp->if_capenable & IFCAP_TSO4 && 2841 ifp->if_hwassist & CSUM_TSO) { 2842 cap->ifcap |= CSUM_TSO; 2843 cap->tsomax = ifp->if_hw_tsomax; 2844 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2845 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2846 } 2847 } 2848 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2849 } 2850 return (maxmtu); 2851 } 2852 #endif /* INET */ 2853 2854 #ifdef INET6 2855 uint32_t 2856 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2857 { 2858 struct nhop6_extended nh6; 2859 struct in6_addr dst6; 2860 uint32_t scopeid; 2861 struct ifnet *ifp; 2862 uint32_t maxmtu = 0; 2863 2864 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2865 2866 if (inc->inc_flags & INC_IPV6MINMTU) 2867 return (IPV6_MMTU); 2868 2869 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2870 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2871 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2872 0, &nh6) != 0) 2873 return (0); 2874 2875 ifp = nh6.nh_ifp; 2876 maxmtu = nh6.nh_mtu; 2877 2878 /* Report additional interface capabilities. */ 2879 if (cap != NULL) { 2880 if (ifp->if_capenable & IFCAP_TSO6 && 2881 ifp->if_hwassist & CSUM_TSO) { 2882 cap->ifcap |= CSUM_TSO; 2883 cap->tsomax = ifp->if_hw_tsomax; 2884 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2885 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2886 } 2887 } 2888 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2889 } 2890 2891 return (maxmtu); 2892 } 2893 #endif /* INET6 */ 2894 2895 /* 2896 * Calculate effective SMSS per RFC5681 definition for a given TCP 2897 * connection at its current state, taking into account SACK and etc. 2898 */ 2899 u_int 2900 tcp_maxseg(const struct tcpcb *tp) 2901 { 2902 u_int optlen; 2903 2904 if (tp->t_flags & TF_NOOPT) 2905 return (tp->t_maxseg); 2906 2907 /* 2908 * Here we have a simplified code from tcp_addoptions(), 2909 * without a proper loop, and having most of paddings hardcoded. 2910 * We might make mistakes with padding here in some edge cases, 2911 * but this is harmless, since result of tcp_maxseg() is used 2912 * only in cwnd and ssthresh estimations. 2913 */ 2914 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2915 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2916 if (tp->t_flags & TF_RCVD_TSTMP) 2917 optlen = TCPOLEN_TSTAMP_APPA; 2918 else 2919 optlen = 0; 2920 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2921 if (tp->t_flags & TF_SIGNATURE) 2922 optlen += PAD(TCPOLEN_SIGNATURE); 2923 #endif 2924 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2925 optlen += TCPOLEN_SACKHDR; 2926 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2927 optlen = PAD(optlen); 2928 } 2929 } else { 2930 if (tp->t_flags & TF_REQ_TSTMP) 2931 optlen = TCPOLEN_TSTAMP_APPA; 2932 else 2933 optlen = PAD(TCPOLEN_MAXSEG); 2934 if (tp->t_flags & TF_REQ_SCALE) 2935 optlen += PAD(TCPOLEN_WINDOW); 2936 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2937 if (tp->t_flags & TF_SIGNATURE) 2938 optlen += PAD(TCPOLEN_SIGNATURE); 2939 #endif 2940 if (tp->t_flags & TF_SACK_PERMIT) 2941 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2942 } 2943 #undef PAD 2944 optlen = min(optlen, TCP_MAXOLEN); 2945 return (tp->t_maxseg - optlen); 2946 } 2947 2948 static int 2949 sysctl_drop(SYSCTL_HANDLER_ARGS) 2950 { 2951 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2952 struct sockaddr_storage addrs[2]; 2953 struct inpcb *inp; 2954 struct tcpcb *tp; 2955 struct tcptw *tw; 2956 struct sockaddr_in *fin, *lin; 2957 struct epoch_tracker et; 2958 #ifdef INET6 2959 struct sockaddr_in6 *fin6, *lin6; 2960 #endif 2961 int error; 2962 2963 inp = NULL; 2964 fin = lin = NULL; 2965 #ifdef INET6 2966 fin6 = lin6 = NULL; 2967 #endif 2968 error = 0; 2969 2970 if (req->oldptr != NULL || req->oldlen != 0) 2971 return (EINVAL); 2972 if (req->newptr == NULL) 2973 return (EPERM); 2974 if (req->newlen < sizeof(addrs)) 2975 return (ENOMEM); 2976 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2977 if (error) 2978 return (error); 2979 2980 switch (addrs[0].ss_family) { 2981 #ifdef INET6 2982 case AF_INET6: 2983 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2984 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2985 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2986 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2987 return (EINVAL); 2988 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2989 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2990 return (EINVAL); 2991 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2992 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2993 fin = (struct sockaddr_in *)&addrs[0]; 2994 lin = (struct sockaddr_in *)&addrs[1]; 2995 break; 2996 } 2997 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2998 if (error) 2999 return (error); 3000 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3001 if (error) 3002 return (error); 3003 break; 3004 #endif 3005 #ifdef INET 3006 case AF_INET: 3007 fin = (struct sockaddr_in *)&addrs[0]; 3008 lin = (struct sockaddr_in *)&addrs[1]; 3009 if (fin->sin_len != sizeof(struct sockaddr_in) || 3010 lin->sin_len != sizeof(struct sockaddr_in)) 3011 return (EINVAL); 3012 break; 3013 #endif 3014 default: 3015 return (EINVAL); 3016 } 3017 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 3018 switch (addrs[0].ss_family) { 3019 #ifdef INET6 3020 case AF_INET6: 3021 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3022 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3023 INPLOOKUP_WLOCKPCB, NULL); 3024 break; 3025 #endif 3026 #ifdef INET 3027 case AF_INET: 3028 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3029 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3030 break; 3031 #endif 3032 } 3033 if (inp != NULL) { 3034 if (inp->inp_flags & INP_TIMEWAIT) { 3035 /* 3036 * XXXRW: There currently exists a state where an 3037 * inpcb is present, but its timewait state has been 3038 * discarded. For now, don't allow dropping of this 3039 * type of inpcb. 3040 */ 3041 tw = intotw(inp); 3042 if (tw != NULL) 3043 tcp_twclose(tw, 0); 3044 else 3045 INP_WUNLOCK(inp); 3046 } else if (!(inp->inp_flags & INP_DROPPED) && 3047 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3048 tp = intotcpcb(inp); 3049 tp = tcp_drop(tp, ECONNABORTED); 3050 if (tp != NULL) 3051 INP_WUNLOCK(inp); 3052 } else 3053 INP_WUNLOCK(inp); 3054 } else 3055 error = ESRCH; 3056 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 3057 return (error); 3058 } 3059 3060 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3061 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3062 0, sysctl_drop, "", "Drop TCP connection"); 3063 3064 /* 3065 * Generate a standardized TCP log line for use throughout the 3066 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3067 * allow use in the interrupt context. 3068 * 3069 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3070 * NB: The function may return NULL if memory allocation failed. 3071 * 3072 * Due to header inclusion and ordering limitations the struct ip 3073 * and ip6_hdr pointers have to be passed as void pointers. 3074 */ 3075 char * 3076 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3077 const void *ip6hdr) 3078 { 3079 3080 /* Is logging enabled? */ 3081 if (tcp_log_in_vain == 0) 3082 return (NULL); 3083 3084 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3085 } 3086 3087 char * 3088 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3089 const void *ip6hdr) 3090 { 3091 3092 /* Is logging enabled? */ 3093 if (tcp_log_debug == 0) 3094 return (NULL); 3095 3096 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3097 } 3098 3099 static char * 3100 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3101 const void *ip6hdr) 3102 { 3103 char *s, *sp; 3104 size_t size; 3105 struct ip *ip; 3106 #ifdef INET6 3107 const struct ip6_hdr *ip6; 3108 3109 ip6 = (const struct ip6_hdr *)ip6hdr; 3110 #endif /* INET6 */ 3111 ip = (struct ip *)ip4hdr; 3112 3113 /* 3114 * The log line looks like this: 3115 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3116 */ 3117 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3118 sizeof(PRINT_TH_FLAGS) + 1 + 3119 #ifdef INET6 3120 2 * INET6_ADDRSTRLEN; 3121 #else 3122 2 * INET_ADDRSTRLEN; 3123 #endif /* INET6 */ 3124 3125 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3126 if (s == NULL) 3127 return (NULL); 3128 3129 strcat(s, "TCP: ["); 3130 sp = s + strlen(s); 3131 3132 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3133 inet_ntoa_r(inc->inc_faddr, sp); 3134 sp = s + strlen(s); 3135 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3136 sp = s + strlen(s); 3137 inet_ntoa_r(inc->inc_laddr, sp); 3138 sp = s + strlen(s); 3139 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3140 #ifdef INET6 3141 } else if (inc) { 3142 ip6_sprintf(sp, &inc->inc6_faddr); 3143 sp = s + strlen(s); 3144 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3145 sp = s + strlen(s); 3146 ip6_sprintf(sp, &inc->inc6_laddr); 3147 sp = s + strlen(s); 3148 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3149 } else if (ip6 && th) { 3150 ip6_sprintf(sp, &ip6->ip6_src); 3151 sp = s + strlen(s); 3152 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3153 sp = s + strlen(s); 3154 ip6_sprintf(sp, &ip6->ip6_dst); 3155 sp = s + strlen(s); 3156 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3157 #endif /* INET6 */ 3158 #ifdef INET 3159 } else if (ip && th) { 3160 inet_ntoa_r(ip->ip_src, sp); 3161 sp = s + strlen(s); 3162 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3163 sp = s + strlen(s); 3164 inet_ntoa_r(ip->ip_dst, sp); 3165 sp = s + strlen(s); 3166 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3167 #endif /* INET */ 3168 } else { 3169 free(s, M_TCPLOG); 3170 return (NULL); 3171 } 3172 sp = s + strlen(s); 3173 if (th) 3174 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3175 if (*(s + size - 1) != '\0') 3176 panic("%s: string too long", __func__); 3177 return (s); 3178 } 3179 3180 /* 3181 * A subroutine which makes it easy to track TCP state changes with DTrace. 3182 * This function shouldn't be called for t_state initializations that don't 3183 * correspond to actual TCP state transitions. 3184 */ 3185 void 3186 tcp_state_change(struct tcpcb *tp, int newstate) 3187 { 3188 #if defined(KDTRACE_HOOKS) 3189 int pstate = tp->t_state; 3190 #endif 3191 3192 TCPSTATES_DEC(tp->t_state); 3193 TCPSTATES_INC(newstate); 3194 tp->t_state = newstate; 3195 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3196 } 3197 3198 /* 3199 * Create an external-format (``xtcpcb'') structure using the information in 3200 * the kernel-format tcpcb structure pointed to by tp. This is done to 3201 * reduce the spew of irrelevant information over this interface, to isolate 3202 * user code from changes in the kernel structure, and potentially to provide 3203 * information-hiding if we decide that some of this information should be 3204 * hidden from users. 3205 */ 3206 void 3207 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3208 { 3209 struct tcpcb *tp = intotcpcb(inp); 3210 sbintime_t now; 3211 3212 bzero(xt, sizeof(*xt)); 3213 if (inp->inp_flags & INP_TIMEWAIT) { 3214 xt->t_state = TCPS_TIME_WAIT; 3215 } else { 3216 xt->t_state = tp->t_state; 3217 xt->t_logstate = tp->t_logstate; 3218 xt->t_flags = tp->t_flags; 3219 xt->t_sndzerowin = tp->t_sndzerowin; 3220 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3221 xt->t_rcvoopack = tp->t_rcvoopack; 3222 3223 now = getsbinuptime(); 3224 #define COPYTIMER(ttt) do { \ 3225 if (callout_active(&tp->t_timers->ttt)) \ 3226 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3227 SBT_1MS; \ 3228 else \ 3229 xt->ttt = 0; \ 3230 } while (0) 3231 COPYTIMER(tt_delack); 3232 COPYTIMER(tt_rexmt); 3233 COPYTIMER(tt_persist); 3234 COPYTIMER(tt_keep); 3235 COPYTIMER(tt_2msl); 3236 #undef COPYTIMER 3237 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3238 3239 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3240 TCP_FUNCTION_NAME_LEN_MAX); 3241 #ifdef TCP_BLACKBOX 3242 (void)tcp_log_get_id(tp, xt->xt_logid); 3243 #endif 3244 } 3245 3246 xt->xt_len = sizeof(struct xtcpcb); 3247 in_pcbtoxinpcb(inp, &xt->xt_inp); 3248 if (inp->inp_socket == NULL) 3249 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3250 } 3251