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