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_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/eventhandler.h> 47 #ifdef TCP_HHOOK 48 #include <sys/hhook.h> 49 #endif 50 #include <sys/kernel.h> 51 #ifdef TCP_HHOOK 52 #include <sys/khelp.h> 53 #endif 54 #ifdef KERN_TLS 55 #include <sys/ktls.h> 56 #endif 57 #include <sys/sysctl.h> 58 #include <sys/jail.h> 59 #include <sys/malloc.h> 60 #include <sys/refcount.h> 61 #include <sys/mbuf.h> 62 #ifdef INET6 63 #include <sys/domain.h> 64 #endif 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/sdt.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/protosw.h> 71 #include <sys/random.h> 72 73 #include <vm/uma.h> 74 75 #include <net/route.h> 76 #include <net/if.h> 77 #include <net/if_var.h> 78 #include <net/vnet.h> 79 80 #include <netinet/in.h> 81 #include <netinet/in_fib.h> 82 #include <netinet/in_kdtrace.h> 83 #include <netinet/in_pcb.h> 84 #include <netinet/in_systm.h> 85 #include <netinet/in_var.h> 86 #include <netinet/ip.h> 87 #include <netinet/ip_icmp.h> 88 #include <netinet/ip_var.h> 89 #ifdef INET6 90 #include <netinet/icmp6.h> 91 #include <netinet/ip6.h> 92 #include <netinet6/in6_fib.h> 93 #include <netinet6/in6_pcb.h> 94 #include <netinet6/ip6_var.h> 95 #include <netinet6/scope6_var.h> 96 #include <netinet6/nd6.h> 97 #endif 98 99 #include <netinet/tcp.h> 100 #include <netinet/tcp_fsm.h> 101 #include <netinet/tcp_seq.h> 102 #include <netinet/tcp_timer.h> 103 #include <netinet/tcp_var.h> 104 #include <netinet/tcp_log_buf.h> 105 #include <netinet/tcp_syncache.h> 106 #include <netinet/tcp_hpts.h> 107 #include <netinet/cc/cc.h> 108 #ifdef INET6 109 #include <netinet6/tcp6_var.h> 110 #endif 111 #include <netinet/tcpip.h> 112 #include <netinet/tcp_fastopen.h> 113 #ifdef TCPPCAP 114 #include <netinet/tcp_pcap.h> 115 #endif 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif 119 #ifdef INET6 120 #include <netinet6/ip6protosw.h> 121 #endif 122 #ifdef TCP_OFFLOAD 123 #include <netinet/tcp_offload.h> 124 #endif 125 126 #include <netipsec/ipsec_support.h> 127 128 #include <machine/in_cksum.h> 129 #include <sys/md5.h> 130 131 #include <security/mac/mac_framework.h> 132 133 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 134 #ifdef INET6 135 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 136 #endif 137 138 struct rwlock tcp_function_lock; 139 140 static int 141 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 142 { 143 int error, new; 144 145 new = V_tcp_mssdflt; 146 error = sysctl_handle_int(oidp, &new, 0, req); 147 if (error == 0 && req->newptr) { 148 if (new < TCP_MINMSS) 149 error = EINVAL; 150 else 151 V_tcp_mssdflt = new; 152 } 153 return (error); 154 } 155 156 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 157 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 158 &sysctl_net_inet_tcp_mss_check, "I", 159 "Default TCP Maximum Segment Size"); 160 161 #ifdef INET6 162 static int 163 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 164 { 165 int error, new; 166 167 new = V_tcp_v6mssdflt; 168 error = sysctl_handle_int(oidp, &new, 0, req); 169 if (error == 0 && req->newptr) { 170 if (new < TCP_MINMSS) 171 error = EINVAL; 172 else 173 V_tcp_v6mssdflt = new; 174 } 175 return (error); 176 } 177 178 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 179 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 180 &sysctl_net_inet_tcp_mss_v6_check, "I", 181 "Default TCP Maximum Segment Size for IPv6"); 182 #endif /* INET6 */ 183 184 /* 185 * Minimum MSS we accept and use. This prevents DoS attacks where 186 * we are forced to a ridiculous low MSS like 20 and send hundreds 187 * of packets instead of one. The effect scales with the available 188 * bandwidth and quickly saturates the CPU and network interface 189 * with packet generation and sending. Set to zero to disable MINMSS 190 * checking. This setting prevents us from sending too small packets. 191 */ 192 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_minmss), 0, 195 "Minimum TCP Maximum Segment Size"); 196 197 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 198 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_do_rfc1323), 0, 200 "Enable rfc1323 (high performance TCP) extensions"); 201 202 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_ts_offset_per_conn), 0, 205 "Initialize TCP timestamps per connection instead of per host pair"); 206 207 static int tcp_log_debug = 0; 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 209 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 210 211 static int tcp_tcbhashsize; 212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 213 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 214 215 static int do_tcpdrain = 1; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 217 "Enable tcp_drain routine for extra help when low on mbufs"); 218 219 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 220 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 221 222 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 223 #define V_icmp_may_rst VNET(icmp_may_rst) 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 225 &VNET_NAME(icmp_may_rst), 0, 226 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 227 228 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 229 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 231 &VNET_NAME(tcp_isn_reseed_interval), 0, 232 "Seconds between reseeding of ISN secret"); 233 234 static int tcp_soreceive_stream; 235 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 236 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 237 238 VNET_DEFINE(uma_zone_t, sack_hole_zone); 239 #define V_sack_hole_zone VNET(sack_hole_zone) 240 241 #ifdef TCP_HHOOK 242 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 243 #endif 244 245 #define TS_OFFSET_SECRET_LENGTH 32 246 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 247 #define V_ts_offset_secret VNET(ts_offset_secret) 248 249 static int tcp_default_fb_init(struct tcpcb *tp); 250 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 251 static int tcp_default_handoff_ok(struct tcpcb *tp); 252 static struct inpcb *tcp_notify(struct inpcb *, int); 253 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 254 static void tcp_mtudisc(struct inpcb *, int); 255 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 256 void *ip4hdr, const void *ip6hdr); 257 258 259 static struct tcp_function_block tcp_def_funcblk = { 260 .tfb_tcp_block_name = "freebsd", 261 .tfb_tcp_output = tcp_output, 262 .tfb_tcp_do_segment = tcp_do_segment, 263 .tfb_tcp_ctloutput = tcp_default_ctloutput, 264 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 265 .tfb_tcp_fb_init = tcp_default_fb_init, 266 .tfb_tcp_fb_fini = tcp_default_fb_fini, 267 }; 268 269 static int tcp_fb_cnt = 0; 270 struct tcp_funchead t_functions; 271 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 272 273 static struct tcp_function_block * 274 find_tcp_functions_locked(struct tcp_function_set *fs) 275 { 276 struct tcp_function *f; 277 struct tcp_function_block *blk=NULL; 278 279 TAILQ_FOREACH(f, &t_functions, tf_next) { 280 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 281 blk = f->tf_fb; 282 break; 283 } 284 } 285 return(blk); 286 } 287 288 static struct tcp_function_block * 289 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 290 { 291 struct tcp_function_block *rblk=NULL; 292 struct tcp_function *f; 293 294 TAILQ_FOREACH(f, &t_functions, tf_next) { 295 if (f->tf_fb == blk) { 296 rblk = blk; 297 if (s) { 298 *s = f; 299 } 300 break; 301 } 302 } 303 return (rblk); 304 } 305 306 struct tcp_function_block * 307 find_and_ref_tcp_functions(struct tcp_function_set *fs) 308 { 309 struct tcp_function_block *blk; 310 311 rw_rlock(&tcp_function_lock); 312 blk = find_tcp_functions_locked(fs); 313 if (blk) 314 refcount_acquire(&blk->tfb_refcnt); 315 rw_runlock(&tcp_function_lock); 316 return(blk); 317 } 318 319 struct tcp_function_block * 320 find_and_ref_tcp_fb(struct tcp_function_block *blk) 321 { 322 struct tcp_function_block *rblk; 323 324 rw_rlock(&tcp_function_lock); 325 rblk = find_tcp_fb_locked(blk, NULL); 326 if (rblk) 327 refcount_acquire(&rblk->tfb_refcnt); 328 rw_runlock(&tcp_function_lock); 329 return(rblk); 330 } 331 332 static struct tcp_function_block * 333 find_and_ref_tcp_default_fb(void) 334 { 335 struct tcp_function_block *rblk; 336 337 rw_rlock(&tcp_function_lock); 338 rblk = tcp_func_set_ptr; 339 refcount_acquire(&rblk->tfb_refcnt); 340 rw_runlock(&tcp_function_lock); 341 return (rblk); 342 } 343 344 void 345 tcp_switch_back_to_default(struct tcpcb *tp) 346 { 347 struct tcp_function_block *tfb; 348 349 KASSERT(tp->t_fb != &tcp_def_funcblk, 350 ("%s: called by the built-in default stack", __func__)); 351 352 /* 353 * Release the old stack. This function will either find a new one 354 * or panic. 355 */ 356 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 357 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 358 refcount_release(&tp->t_fb->tfb_refcnt); 359 360 /* 361 * Now, we'll find a new function block to use. 362 * Start by trying the current user-selected 363 * default, unless this stack is the user-selected 364 * default. 365 */ 366 tfb = find_and_ref_tcp_default_fb(); 367 if (tfb == tp->t_fb) { 368 refcount_release(&tfb->tfb_refcnt); 369 tfb = NULL; 370 } 371 /* Does the stack accept this connection? */ 372 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 373 (*tfb->tfb_tcp_handoff_ok)(tp)) { 374 refcount_release(&tfb->tfb_refcnt); 375 tfb = NULL; 376 } 377 /* Try to use that stack. */ 378 if (tfb != NULL) { 379 /* Initialize the new stack. If it succeeds, we are done. */ 380 tp->t_fb = tfb; 381 if (tp->t_fb->tfb_tcp_fb_init == NULL || 382 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 383 return; 384 385 /* 386 * Initialization failed. Release the reference count on 387 * the stack. 388 */ 389 refcount_release(&tfb->tfb_refcnt); 390 } 391 392 /* 393 * If that wasn't feasible, use the built-in default 394 * stack which is not allowed to reject anyone. 395 */ 396 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 397 if (tfb == NULL) { 398 /* there always should be a default */ 399 panic("Can't refer to tcp_def_funcblk"); 400 } 401 if (tfb->tfb_tcp_handoff_ok != NULL) { 402 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 403 /* The default stack cannot say no */ 404 panic("Default stack rejects a new session?"); 405 } 406 } 407 tp->t_fb = tfb; 408 if (tp->t_fb->tfb_tcp_fb_init != NULL && 409 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 410 /* The default stack cannot fail */ 411 panic("Default stack initialization failed"); 412 } 413 } 414 415 static int 416 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 417 { 418 int error=ENOENT; 419 struct tcp_function_set fs; 420 struct tcp_function_block *blk; 421 422 memset(&fs, 0, sizeof(fs)); 423 rw_rlock(&tcp_function_lock); 424 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 425 if (blk) { 426 /* Found him */ 427 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 428 fs.pcbcnt = blk->tfb_refcnt; 429 } 430 rw_runlock(&tcp_function_lock); 431 error = sysctl_handle_string(oidp, fs.function_set_name, 432 sizeof(fs.function_set_name), req); 433 434 /* Check for error or no change */ 435 if (error != 0 || req->newptr == NULL) 436 return(error); 437 438 rw_wlock(&tcp_function_lock); 439 blk = find_tcp_functions_locked(&fs); 440 if ((blk == NULL) || 441 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 442 error = ENOENT; 443 goto done; 444 } 445 tcp_func_set_ptr = blk; 446 done: 447 rw_wunlock(&tcp_function_lock); 448 return (error); 449 } 450 451 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 452 CTLTYPE_STRING | CTLFLAG_RW, 453 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 454 "Set/get the default TCP functions"); 455 456 static int 457 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 458 { 459 int error, cnt, linesz; 460 struct tcp_function *f; 461 char *buffer, *cp; 462 size_t bufsz, outsz; 463 bool alias; 464 465 cnt = 0; 466 rw_rlock(&tcp_function_lock); 467 TAILQ_FOREACH(f, &t_functions, tf_next) { 468 cnt++; 469 } 470 rw_runlock(&tcp_function_lock); 471 472 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 473 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 474 475 error = 0; 476 cp = buffer; 477 478 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 479 "Alias", "PCB count"); 480 cp += linesz; 481 bufsz -= linesz; 482 outsz = linesz; 483 484 rw_rlock(&tcp_function_lock); 485 TAILQ_FOREACH(f, &t_functions, tf_next) { 486 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 487 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 488 f->tf_fb->tfb_tcp_block_name, 489 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 490 alias ? f->tf_name : "-", 491 f->tf_fb->tfb_refcnt); 492 if (linesz >= bufsz) { 493 error = EOVERFLOW; 494 break; 495 } 496 cp += linesz; 497 bufsz -= linesz; 498 outsz += linesz; 499 } 500 rw_runlock(&tcp_function_lock); 501 if (error == 0) 502 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 503 free(buffer, M_TEMP); 504 return (error); 505 } 506 507 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 508 CTLTYPE_STRING|CTLFLAG_RD, 509 NULL, 0, sysctl_net_inet_list_available, "A", 510 "list available TCP Function sets"); 511 512 /* 513 * Exports one (struct tcp_function_info) for each alias/name. 514 */ 515 static int 516 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 517 { 518 int cnt, error; 519 struct tcp_function *f; 520 struct tcp_function_info tfi; 521 522 /* 523 * We don't allow writes. 524 */ 525 if (req->newptr != NULL) 526 return (EINVAL); 527 528 /* 529 * Wire the old buffer so we can directly copy the functions to 530 * user space without dropping the lock. 531 */ 532 if (req->oldptr != NULL) { 533 error = sysctl_wire_old_buffer(req, 0); 534 if (error) 535 return (error); 536 } 537 538 /* 539 * Walk the list and copy out matching entries. If INVARIANTS 540 * is compiled in, also walk the list to verify the length of 541 * the list matches what we have recorded. 542 */ 543 rw_rlock(&tcp_function_lock); 544 545 cnt = 0; 546 #ifndef INVARIANTS 547 if (req->oldptr == NULL) { 548 cnt = tcp_fb_cnt; 549 goto skip_loop; 550 } 551 #endif 552 TAILQ_FOREACH(f, &t_functions, tf_next) { 553 #ifdef INVARIANTS 554 cnt++; 555 #endif 556 if (req->oldptr != NULL) { 557 bzero(&tfi, sizeof(tfi)); 558 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 559 tfi.tfi_id = f->tf_fb->tfb_id; 560 (void)strlcpy(tfi.tfi_alias, f->tf_name, 561 sizeof(tfi.tfi_alias)); 562 (void)strlcpy(tfi.tfi_name, 563 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 564 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 565 /* 566 * Don't stop on error, as that is the 567 * mechanism we use to accumulate length 568 * information if the buffer was too short. 569 */ 570 } 571 } 572 KASSERT(cnt == tcp_fb_cnt, 573 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 574 #ifndef INVARIANTS 575 skip_loop: 576 #endif 577 rw_runlock(&tcp_function_lock); 578 if (req->oldptr == NULL) 579 error = SYSCTL_OUT(req, NULL, 580 (cnt + 1) * sizeof(struct tcp_function_info)); 581 582 return (error); 583 } 584 585 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 586 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 587 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 588 "List TCP function block name-to-ID mappings"); 589 590 /* 591 * tfb_tcp_handoff_ok() function for the default stack. 592 * Note that we'll basically try to take all comers. 593 */ 594 static int 595 tcp_default_handoff_ok(struct tcpcb *tp) 596 { 597 598 return (0); 599 } 600 601 /* 602 * tfb_tcp_fb_init() function for the default stack. 603 * 604 * This handles making sure we have appropriate timers set if you are 605 * transitioning a socket that has some amount of setup done. 606 * 607 * The init() fuction from the default can *never* return non-zero i.e. 608 * it is required to always succeed since it is the stack of last resort! 609 */ 610 static int 611 tcp_default_fb_init(struct tcpcb *tp) 612 { 613 614 struct socket *so; 615 616 INP_WLOCK_ASSERT(tp->t_inpcb); 617 618 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 619 ("%s: connection %p in unexpected state %d", __func__, tp, 620 tp->t_state)); 621 622 /* 623 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 624 * know what to do for unexpected states (which includes TIME_WAIT). 625 */ 626 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 627 return (0); 628 629 /* 630 * Make sure some kind of transmission timer is set if there is 631 * outstanding data. 632 */ 633 so = tp->t_inpcb->inp_socket; 634 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 635 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 636 tcp_timer_active(tp, TT_PERSIST))) { 637 /* 638 * If the session has established and it looks like it should 639 * be in the persist state, set the persist timer. Otherwise, 640 * set the retransmit timer. 641 */ 642 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 643 (int32_t)(tp->snd_nxt - tp->snd_una) < 644 (int32_t)sbavail(&so->so_snd)) 645 tcp_setpersist(tp); 646 else 647 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 648 } 649 650 /* All non-embryonic sessions get a keepalive timer. */ 651 if (!tcp_timer_active(tp, TT_KEEP)) 652 tcp_timer_activate(tp, TT_KEEP, 653 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 654 TP_KEEPINIT(tp)); 655 656 return (0); 657 } 658 659 /* 660 * tfb_tcp_fb_fini() function for the default stack. 661 * 662 * This changes state as necessary (or prudent) to prepare for another stack 663 * to assume responsibility for the connection. 664 */ 665 static void 666 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 667 { 668 669 INP_WLOCK_ASSERT(tp->t_inpcb); 670 return; 671 } 672 673 /* 674 * Target size of TCP PCB hash tables. Must be a power of two. 675 * 676 * Note that this can be overridden by the kernel environment 677 * variable net.inet.tcp.tcbhashsize 678 */ 679 #ifndef TCBHASHSIZE 680 #define TCBHASHSIZE 0 681 #endif 682 683 /* 684 * XXX 685 * Callouts should be moved into struct tcp directly. They are currently 686 * separate because the tcpcb structure is exported to userland for sysctl 687 * parsing purposes, which do not know about callouts. 688 */ 689 struct tcpcb_mem { 690 struct tcpcb tcb; 691 struct tcp_timer tt; 692 struct cc_var ccv; 693 #ifdef TCP_HHOOK 694 struct osd osd; 695 #endif 696 }; 697 698 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 699 #define V_tcpcb_zone VNET(tcpcb_zone) 700 701 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 702 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 703 704 static struct mtx isn_mtx; 705 706 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 707 #define ISN_LOCK() mtx_lock(&isn_mtx) 708 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 709 710 /* 711 * TCP initialization. 712 */ 713 static void 714 tcp_zone_change(void *tag) 715 { 716 717 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 718 uma_zone_set_max(V_tcpcb_zone, maxsockets); 719 tcp_tw_zone_change(); 720 } 721 722 static int 723 tcp_inpcb_init(void *mem, int size, int flags) 724 { 725 struct inpcb *inp = mem; 726 727 INP_LOCK_INIT(inp, "inp", "tcpinp"); 728 return (0); 729 } 730 731 /* 732 * Take a value and get the next power of 2 that doesn't overflow. 733 * Used to size the tcp_inpcb hash buckets. 734 */ 735 static int 736 maketcp_hashsize(int size) 737 { 738 int hashsize; 739 740 /* 741 * auto tune. 742 * get the next power of 2 higher than maxsockets. 743 */ 744 hashsize = 1 << fls(size); 745 /* catch overflow, and just go one power of 2 smaller */ 746 if (hashsize < size) { 747 hashsize = 1 << (fls(size) - 1); 748 } 749 return (hashsize); 750 } 751 752 static volatile int next_tcp_stack_id = 1; 753 754 /* 755 * Register a TCP function block with the name provided in the names 756 * array. (Note that this function does NOT automatically register 757 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 758 * explicitly include blk->tfb_tcp_block_name in the list of names if 759 * you wish to register the stack with that name.) 760 * 761 * Either all name registrations will succeed or all will fail. If 762 * a name registration fails, the function will update the num_names 763 * argument to point to the array index of the name that encountered 764 * the failure. 765 * 766 * Returns 0 on success, or an error code on failure. 767 */ 768 int 769 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 770 const char *names[], int *num_names) 771 { 772 struct tcp_function *n; 773 struct tcp_function_set fs; 774 int error, i; 775 776 KASSERT(names != NULL && *num_names > 0, 777 ("%s: Called with 0-length name list", __func__)); 778 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 779 KASSERT(rw_initialized(&tcp_function_lock), 780 ("%s: called too early", __func__)); 781 782 if ((blk->tfb_tcp_output == NULL) || 783 (blk->tfb_tcp_do_segment == NULL) || 784 (blk->tfb_tcp_ctloutput == NULL) || 785 (strlen(blk->tfb_tcp_block_name) == 0)) { 786 /* 787 * These functions are required and you 788 * need a name. 789 */ 790 *num_names = 0; 791 return (EINVAL); 792 } 793 if (blk->tfb_tcp_timer_stop_all || 794 blk->tfb_tcp_timer_activate || 795 blk->tfb_tcp_timer_active || 796 blk->tfb_tcp_timer_stop) { 797 /* 798 * If you define one timer function you 799 * must have them all. 800 */ 801 if ((blk->tfb_tcp_timer_stop_all == NULL) || 802 (blk->tfb_tcp_timer_activate == NULL) || 803 (blk->tfb_tcp_timer_active == NULL) || 804 (blk->tfb_tcp_timer_stop == NULL)) { 805 *num_names = 0; 806 return (EINVAL); 807 } 808 } 809 810 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 811 *num_names = 0; 812 return (EINVAL); 813 } 814 815 refcount_init(&blk->tfb_refcnt, 0); 816 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 817 for (i = 0; i < *num_names; i++) { 818 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 819 if (n == NULL) { 820 error = ENOMEM; 821 goto cleanup; 822 } 823 n->tf_fb = blk; 824 825 (void)strlcpy(fs.function_set_name, names[i], 826 sizeof(fs.function_set_name)); 827 rw_wlock(&tcp_function_lock); 828 if (find_tcp_functions_locked(&fs) != NULL) { 829 /* Duplicate name space not allowed */ 830 rw_wunlock(&tcp_function_lock); 831 free(n, M_TCPFUNCTIONS); 832 error = EALREADY; 833 goto cleanup; 834 } 835 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 836 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 837 tcp_fb_cnt++; 838 rw_wunlock(&tcp_function_lock); 839 } 840 return(0); 841 842 cleanup: 843 /* 844 * Deregister the names we just added. Because registration failed 845 * for names[i], we don't need to deregister that name. 846 */ 847 *num_names = i; 848 rw_wlock(&tcp_function_lock); 849 while (--i >= 0) { 850 TAILQ_FOREACH(n, &t_functions, tf_next) { 851 if (!strncmp(n->tf_name, names[i], 852 TCP_FUNCTION_NAME_LEN_MAX)) { 853 TAILQ_REMOVE(&t_functions, n, tf_next); 854 tcp_fb_cnt--; 855 n->tf_fb = NULL; 856 free(n, M_TCPFUNCTIONS); 857 break; 858 } 859 } 860 } 861 rw_wunlock(&tcp_function_lock); 862 return (error); 863 } 864 865 /* 866 * Register a TCP function block using the name provided in the name 867 * argument. 868 * 869 * Returns 0 on success, or an error code on failure. 870 */ 871 int 872 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 873 int wait) 874 { 875 const char *name_list[1]; 876 int num_names, rv; 877 878 num_names = 1; 879 if (name != NULL) 880 name_list[0] = name; 881 else 882 name_list[0] = blk->tfb_tcp_block_name; 883 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 884 return (rv); 885 } 886 887 /* 888 * Register a TCP function block using the name defined in 889 * blk->tfb_tcp_block_name. 890 * 891 * Returns 0 on success, or an error code on failure. 892 */ 893 int 894 register_tcp_functions(struct tcp_function_block *blk, int wait) 895 { 896 897 return (register_tcp_functions_as_name(blk, NULL, wait)); 898 } 899 900 /* 901 * Deregister all names associated with a function block. This 902 * functionally removes the function block from use within the system. 903 * 904 * When called with a true quiesce argument, mark the function block 905 * as being removed so no more stacks will use it and determine 906 * whether the removal would succeed. 907 * 908 * When called with a false quiesce argument, actually attempt the 909 * removal. 910 * 911 * When called with a force argument, attempt to switch all TCBs to 912 * use the default stack instead of returning EBUSY. 913 * 914 * Returns 0 on success (or if the removal would succeed, or an error 915 * code on failure. 916 */ 917 int 918 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 919 bool force) 920 { 921 struct tcp_function *f; 922 923 if (blk == &tcp_def_funcblk) { 924 /* You can't un-register the default */ 925 return (EPERM); 926 } 927 rw_wlock(&tcp_function_lock); 928 if (blk == tcp_func_set_ptr) { 929 /* You can't free the current default */ 930 rw_wunlock(&tcp_function_lock); 931 return (EBUSY); 932 } 933 /* Mark the block so no more stacks can use it. */ 934 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 935 /* 936 * If TCBs are still attached to the stack, attempt to switch them 937 * to the default stack. 938 */ 939 if (force && blk->tfb_refcnt) { 940 struct inpcb *inp; 941 struct tcpcb *tp; 942 VNET_ITERATOR_DECL(vnet_iter); 943 944 rw_wunlock(&tcp_function_lock); 945 946 VNET_LIST_RLOCK(); 947 VNET_FOREACH(vnet_iter) { 948 CURVNET_SET(vnet_iter); 949 INP_INFO_WLOCK(&V_tcbinfo); 950 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 951 INP_WLOCK(inp); 952 if (inp->inp_flags & INP_TIMEWAIT) { 953 INP_WUNLOCK(inp); 954 continue; 955 } 956 tp = intotcpcb(inp); 957 if (tp == NULL || tp->t_fb != blk) { 958 INP_WUNLOCK(inp); 959 continue; 960 } 961 tcp_switch_back_to_default(tp); 962 INP_WUNLOCK(inp); 963 } 964 INP_INFO_WUNLOCK(&V_tcbinfo); 965 CURVNET_RESTORE(); 966 } 967 VNET_LIST_RUNLOCK(); 968 969 rw_wlock(&tcp_function_lock); 970 } 971 if (blk->tfb_refcnt) { 972 /* TCBs still attached. */ 973 rw_wunlock(&tcp_function_lock); 974 return (EBUSY); 975 } 976 if (quiesce) { 977 /* Skip removal. */ 978 rw_wunlock(&tcp_function_lock); 979 return (0); 980 } 981 /* Remove any function names that map to this function block. */ 982 while (find_tcp_fb_locked(blk, &f) != NULL) { 983 TAILQ_REMOVE(&t_functions, f, tf_next); 984 tcp_fb_cnt--; 985 f->tf_fb = NULL; 986 free(f, M_TCPFUNCTIONS); 987 } 988 rw_wunlock(&tcp_function_lock); 989 return (0); 990 } 991 992 void 993 tcp_init(void) 994 { 995 const char *tcbhash_tuneable; 996 int hashsize; 997 998 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 999 1000 #ifdef TCP_HHOOK 1001 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1002 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1003 printf("%s: WARNING: unable to register helper hook\n", __func__); 1004 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1005 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1006 printf("%s: WARNING: unable to register helper hook\n", __func__); 1007 #endif 1008 hashsize = TCBHASHSIZE; 1009 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1010 if (hashsize == 0) { 1011 /* 1012 * Auto tune the hash size based on maxsockets. 1013 * A perfect hash would have a 1:1 mapping 1014 * (hashsize = maxsockets) however it's been 1015 * suggested that O(2) average is better. 1016 */ 1017 hashsize = maketcp_hashsize(maxsockets / 4); 1018 /* 1019 * Our historical default is 512, 1020 * do not autotune lower than this. 1021 */ 1022 if (hashsize < 512) 1023 hashsize = 512; 1024 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1025 printf("%s: %s auto tuned to %d\n", __func__, 1026 tcbhash_tuneable, hashsize); 1027 } 1028 /* 1029 * We require a hashsize to be a power of two. 1030 * Previously if it was not a power of two we would just reset it 1031 * back to 512, which could be a nasty surprise if you did not notice 1032 * the error message. 1033 * Instead what we do is clip it to the closest power of two lower 1034 * than the specified hash value. 1035 */ 1036 if (!powerof2(hashsize)) { 1037 int oldhashsize = hashsize; 1038 1039 hashsize = maketcp_hashsize(hashsize); 1040 /* prevent absurdly low value */ 1041 if (hashsize < 16) 1042 hashsize = 16; 1043 printf("%s: WARNING: TCB hash size not a power of 2, " 1044 "clipped from %d to %d.\n", __func__, oldhashsize, 1045 hashsize); 1046 } 1047 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1048 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1049 1050 /* 1051 * These have to be type stable for the benefit of the timers. 1052 */ 1053 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1054 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1055 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1056 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1057 1058 tcp_tw_init(); 1059 syncache_init(); 1060 tcp_hc_init(); 1061 1062 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1063 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1064 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1065 1066 tcp_fastopen_init(); 1067 1068 /* Skip initialization of globals for non-default instances. */ 1069 if (!IS_DEFAULT_VNET(curvnet)) 1070 return; 1071 1072 tcp_reass_global_init(); 1073 1074 /* XXX virtualize those bellow? */ 1075 tcp_delacktime = TCPTV_DELACK; 1076 tcp_keepinit = TCPTV_KEEP_INIT; 1077 tcp_keepidle = TCPTV_KEEP_IDLE; 1078 tcp_keepintvl = TCPTV_KEEPINTVL; 1079 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1080 tcp_msl = TCPTV_MSL; 1081 tcp_rexmit_initial = TCPTV_RTOBASE; 1082 if (tcp_rexmit_initial < 1) 1083 tcp_rexmit_initial = 1; 1084 tcp_rexmit_min = TCPTV_MIN; 1085 if (tcp_rexmit_min < 1) 1086 tcp_rexmit_min = 1; 1087 tcp_persmin = TCPTV_PERSMIN; 1088 tcp_persmax = TCPTV_PERSMAX; 1089 tcp_rexmit_slop = TCPTV_CPU_VAR; 1090 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1091 tcp_tcbhashsize = hashsize; 1092 1093 /* Setup the tcp function block list */ 1094 TAILQ_INIT(&t_functions); 1095 rw_init(&tcp_function_lock, "tcp_func_lock"); 1096 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1097 #ifdef TCP_BLACKBOX 1098 /* Initialize the TCP logging data. */ 1099 tcp_log_init(); 1100 #endif 1101 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1102 1103 if (tcp_soreceive_stream) { 1104 #ifdef INET 1105 tcp_usrreqs.pru_soreceive = soreceive_stream; 1106 #endif 1107 #ifdef INET6 1108 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1109 #endif /* INET6 */ 1110 } 1111 1112 #ifdef INET6 1113 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1114 #else /* INET6 */ 1115 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1116 #endif /* INET6 */ 1117 if (max_protohdr < TCP_MINPROTOHDR) 1118 max_protohdr = TCP_MINPROTOHDR; 1119 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1120 panic("tcp_init"); 1121 #undef TCP_MINPROTOHDR 1122 1123 ISN_LOCK_INIT(); 1124 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1125 SHUTDOWN_PRI_DEFAULT); 1126 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1127 EVENTHANDLER_PRI_ANY); 1128 1129 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1130 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1131 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1132 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); 1133 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1134 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); 1135 #ifdef TCPPCAP 1136 tcp_pcap_init(); 1137 #endif 1138 } 1139 1140 #ifdef VIMAGE 1141 static void 1142 tcp_destroy(void *unused __unused) 1143 { 1144 int n; 1145 #ifdef TCP_HHOOK 1146 int error; 1147 #endif 1148 1149 /* 1150 * All our processes are gone, all our sockets should be cleaned 1151 * up, which means, we should be past the tcp_discardcb() calls. 1152 * Sleep to let all tcpcb timers really disappear and cleanup. 1153 */ 1154 for (;;) { 1155 INP_LIST_RLOCK(&V_tcbinfo); 1156 n = V_tcbinfo.ipi_count; 1157 INP_LIST_RUNLOCK(&V_tcbinfo); 1158 if (n == 0) 1159 break; 1160 pause("tcpdes", hz / 10); 1161 } 1162 tcp_hc_destroy(); 1163 syncache_destroy(); 1164 tcp_tw_destroy(); 1165 in_pcbinfo_destroy(&V_tcbinfo); 1166 /* tcp_discardcb() clears the sack_holes up. */ 1167 uma_zdestroy(V_sack_hole_zone); 1168 uma_zdestroy(V_tcpcb_zone); 1169 1170 /* 1171 * Cannot free the zone until all tcpcbs are released as we attach 1172 * the allocations to them. 1173 */ 1174 tcp_fastopen_destroy(); 1175 1176 #ifdef TCP_HHOOK 1177 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1178 if (error != 0) { 1179 printf("%s: WARNING: unable to deregister helper hook " 1180 "type=%d, id=%d: error %d returned\n", __func__, 1181 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1182 } 1183 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1184 if (error != 0) { 1185 printf("%s: WARNING: unable to deregister helper hook " 1186 "type=%d, id=%d: error %d returned\n", __func__, 1187 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1188 } 1189 #endif 1190 } 1191 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1192 #endif 1193 1194 void 1195 tcp_fini(void *xtp) 1196 { 1197 1198 } 1199 1200 /* 1201 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1202 * tcp_template used to store this data in mbufs, but we now recopy it out 1203 * of the tcpcb each time to conserve mbufs. 1204 */ 1205 void 1206 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1207 { 1208 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1209 1210 INP_WLOCK_ASSERT(inp); 1211 1212 #ifdef INET6 1213 if ((inp->inp_vflag & INP_IPV6) != 0) { 1214 struct ip6_hdr *ip6; 1215 1216 ip6 = (struct ip6_hdr *)ip_ptr; 1217 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1218 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1219 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1220 (IPV6_VERSION & IPV6_VERSION_MASK); 1221 ip6->ip6_nxt = IPPROTO_TCP; 1222 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1223 ip6->ip6_src = inp->in6p_laddr; 1224 ip6->ip6_dst = inp->in6p_faddr; 1225 } 1226 #endif /* INET6 */ 1227 #if defined(INET6) && defined(INET) 1228 else 1229 #endif 1230 #ifdef INET 1231 { 1232 struct ip *ip; 1233 1234 ip = (struct ip *)ip_ptr; 1235 ip->ip_v = IPVERSION; 1236 ip->ip_hl = 5; 1237 ip->ip_tos = inp->inp_ip_tos; 1238 ip->ip_len = 0; 1239 ip->ip_id = 0; 1240 ip->ip_off = 0; 1241 ip->ip_ttl = inp->inp_ip_ttl; 1242 ip->ip_sum = 0; 1243 ip->ip_p = IPPROTO_TCP; 1244 ip->ip_src = inp->inp_laddr; 1245 ip->ip_dst = inp->inp_faddr; 1246 } 1247 #endif /* INET */ 1248 th->th_sport = inp->inp_lport; 1249 th->th_dport = inp->inp_fport; 1250 th->th_seq = 0; 1251 th->th_ack = 0; 1252 th->th_x2 = 0; 1253 th->th_off = 5; 1254 th->th_flags = 0; 1255 th->th_win = 0; 1256 th->th_urp = 0; 1257 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1258 } 1259 1260 /* 1261 * Create template to be used to send tcp packets on a connection. 1262 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1263 * use for this function is in keepalives, which use tcp_respond. 1264 */ 1265 struct tcptemp * 1266 tcpip_maketemplate(struct inpcb *inp) 1267 { 1268 struct tcptemp *t; 1269 1270 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1271 if (t == NULL) 1272 return (NULL); 1273 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1274 return (t); 1275 } 1276 1277 /* 1278 * Send a single message to the TCP at address specified by 1279 * the given TCP/IP header. If m == NULL, then we make a copy 1280 * of the tcpiphdr at th and send directly to the addressed host. 1281 * This is used to force keep alive messages out using the TCP 1282 * template for a connection. If flags are given then we send 1283 * a message back to the TCP which originated the segment th, 1284 * and discard the mbuf containing it and any other attached mbufs. 1285 * 1286 * In any case the ack and sequence number of the transmitted 1287 * segment are as specified by the parameters. 1288 * 1289 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1290 */ 1291 void 1292 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1293 tcp_seq ack, tcp_seq seq, int flags) 1294 { 1295 struct tcpopt to; 1296 struct inpcb *inp; 1297 struct ip *ip; 1298 struct mbuf *optm; 1299 struct tcphdr *nth; 1300 u_char *optp; 1301 #ifdef INET6 1302 struct ip6_hdr *ip6; 1303 int isipv6; 1304 #endif /* INET6 */ 1305 int optlen, tlen, win; 1306 bool incl_opts; 1307 1308 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1309 1310 #ifdef INET6 1311 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1312 ip6 = ipgen; 1313 #endif /* INET6 */ 1314 ip = ipgen; 1315 1316 if (tp != NULL) { 1317 inp = tp->t_inpcb; 1318 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1319 INP_WLOCK_ASSERT(inp); 1320 } else 1321 inp = NULL; 1322 1323 incl_opts = false; 1324 win = 0; 1325 if (tp != NULL) { 1326 if (!(flags & TH_RST)) { 1327 win = sbspace(&inp->inp_socket->so_rcv); 1328 if (win > TCP_MAXWIN << tp->rcv_scale) 1329 win = TCP_MAXWIN << tp->rcv_scale; 1330 } 1331 if ((tp->t_flags & TF_NOOPT) == 0) 1332 incl_opts = true; 1333 } 1334 if (m == NULL) { 1335 m = m_gethdr(M_NOWAIT, MT_DATA); 1336 if (m == NULL) 1337 return; 1338 m->m_data += max_linkhdr; 1339 #ifdef INET6 1340 if (isipv6) { 1341 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1342 sizeof(struct ip6_hdr)); 1343 ip6 = mtod(m, struct ip6_hdr *); 1344 nth = (struct tcphdr *)(ip6 + 1); 1345 } else 1346 #endif /* INET6 */ 1347 { 1348 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1349 ip = mtod(m, struct ip *); 1350 nth = (struct tcphdr *)(ip + 1); 1351 } 1352 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1353 flags = TH_ACK; 1354 } else if (!M_WRITABLE(m)) { 1355 struct mbuf *n; 1356 1357 /* Can't reuse 'm', allocate a new mbuf. */ 1358 n = m_gethdr(M_NOWAIT, MT_DATA); 1359 if (n == NULL) { 1360 m_freem(m); 1361 return; 1362 } 1363 1364 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1365 m_freem(m); 1366 m_freem(n); 1367 return; 1368 } 1369 1370 n->m_data += max_linkhdr; 1371 /* m_len is set later */ 1372 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1373 #ifdef INET6 1374 if (isipv6) { 1375 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1376 sizeof(struct ip6_hdr)); 1377 ip6 = mtod(n, struct ip6_hdr *); 1378 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1379 nth = (struct tcphdr *)(ip6 + 1); 1380 } else 1381 #endif /* INET6 */ 1382 { 1383 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1384 ip = mtod(n, struct ip *); 1385 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1386 nth = (struct tcphdr *)(ip + 1); 1387 } 1388 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1389 xchg(nth->th_dport, nth->th_sport, uint16_t); 1390 th = nth; 1391 m_freem(m); 1392 m = n; 1393 } else { 1394 /* 1395 * reuse the mbuf. 1396 * XXX MRT We inherit the FIB, which is lucky. 1397 */ 1398 m_freem(m->m_next); 1399 m->m_next = NULL; 1400 m->m_data = (caddr_t)ipgen; 1401 /* m_len is set later */ 1402 #ifdef INET6 1403 if (isipv6) { 1404 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1405 nth = (struct tcphdr *)(ip6 + 1); 1406 } else 1407 #endif /* INET6 */ 1408 { 1409 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1410 nth = (struct tcphdr *)(ip + 1); 1411 } 1412 if (th != nth) { 1413 /* 1414 * this is usually a case when an extension header 1415 * exists between the IPv6 header and the 1416 * TCP header. 1417 */ 1418 nth->th_sport = th->th_sport; 1419 nth->th_dport = th->th_dport; 1420 } 1421 xchg(nth->th_dport, nth->th_sport, uint16_t); 1422 #undef xchg 1423 } 1424 tlen = 0; 1425 #ifdef INET6 1426 if (isipv6) 1427 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1428 #endif 1429 #if defined(INET) && defined(INET6) 1430 else 1431 #endif 1432 #ifdef INET 1433 tlen = sizeof (struct tcpiphdr); 1434 #endif 1435 #ifdef INVARIANTS 1436 m->m_len = 0; 1437 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1438 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1439 m, tlen, (long)M_TRAILINGSPACE(m))); 1440 #endif 1441 m->m_len = tlen; 1442 to.to_flags = 0; 1443 if (incl_opts) { 1444 /* Make sure we have room. */ 1445 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1446 m->m_next = m_get(M_NOWAIT, MT_DATA); 1447 if (m->m_next) { 1448 optp = mtod(m->m_next, u_char *); 1449 optm = m->m_next; 1450 } else 1451 incl_opts = false; 1452 } else { 1453 optp = (u_char *) (nth + 1); 1454 optm = m; 1455 } 1456 } 1457 if (incl_opts) { 1458 /* Timestamps. */ 1459 if (tp->t_flags & TF_RCVD_TSTMP) { 1460 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1461 to.to_tsecr = tp->ts_recent; 1462 to.to_flags |= TOF_TS; 1463 } 1464 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1465 /* TCP-MD5 (RFC2385). */ 1466 if (tp->t_flags & TF_SIGNATURE) 1467 to.to_flags |= TOF_SIGNATURE; 1468 #endif 1469 /* Add the options. */ 1470 tlen += optlen = tcp_addoptions(&to, optp); 1471 1472 /* Update m_len in the correct mbuf. */ 1473 optm->m_len += optlen; 1474 } else 1475 optlen = 0; 1476 #ifdef INET6 1477 if (isipv6) { 1478 ip6->ip6_flow = 0; 1479 ip6->ip6_vfc = IPV6_VERSION; 1480 ip6->ip6_nxt = IPPROTO_TCP; 1481 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1482 } 1483 #endif 1484 #if defined(INET) && defined(INET6) 1485 else 1486 #endif 1487 #ifdef INET 1488 { 1489 ip->ip_len = htons(tlen); 1490 ip->ip_ttl = V_ip_defttl; 1491 if (V_path_mtu_discovery) 1492 ip->ip_off |= htons(IP_DF); 1493 } 1494 #endif 1495 m->m_pkthdr.len = tlen; 1496 m->m_pkthdr.rcvif = NULL; 1497 #ifdef MAC 1498 if (inp != NULL) { 1499 /* 1500 * Packet is associated with a socket, so allow the 1501 * label of the response to reflect the socket label. 1502 */ 1503 INP_WLOCK_ASSERT(inp); 1504 mac_inpcb_create_mbuf(inp, m); 1505 } else { 1506 /* 1507 * Packet is not associated with a socket, so possibly 1508 * update the label in place. 1509 */ 1510 mac_netinet_tcp_reply(m); 1511 } 1512 #endif 1513 nth->th_seq = htonl(seq); 1514 nth->th_ack = htonl(ack); 1515 nth->th_x2 = 0; 1516 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1517 nth->th_flags = flags; 1518 if (tp != NULL) 1519 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1520 else 1521 nth->th_win = htons((u_short)win); 1522 nth->th_urp = 0; 1523 1524 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1525 if (to.to_flags & TOF_SIGNATURE) { 1526 if (!TCPMD5_ENABLED() || 1527 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1528 m_freem(m); 1529 return; 1530 } 1531 } 1532 #endif 1533 1534 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1535 #ifdef INET6 1536 if (isipv6) { 1537 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1538 nth->th_sum = in6_cksum_pseudo(ip6, 1539 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1540 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1541 NULL, NULL); 1542 } 1543 #endif /* INET6 */ 1544 #if defined(INET6) && defined(INET) 1545 else 1546 #endif 1547 #ifdef INET 1548 { 1549 m->m_pkthdr.csum_flags = CSUM_TCP; 1550 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1551 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1552 } 1553 #endif /* INET */ 1554 #ifdef TCPDEBUG 1555 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1556 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1557 #endif 1558 TCP_PROBE3(debug__output, tp, th, m); 1559 if (flags & TH_RST) 1560 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1561 1562 #ifdef INET6 1563 if (isipv6) { 1564 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1565 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1566 } 1567 #endif /* INET6 */ 1568 #if defined(INET) && defined(INET6) 1569 else 1570 #endif 1571 #ifdef INET 1572 { 1573 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1574 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1575 } 1576 #endif 1577 } 1578 1579 /* 1580 * Create a new TCP control block, making an 1581 * empty reassembly queue and hooking it to the argument 1582 * protocol control block. The `inp' parameter must have 1583 * come from the zone allocator set up in tcp_init(). 1584 */ 1585 struct tcpcb * 1586 tcp_newtcpcb(struct inpcb *inp) 1587 { 1588 struct tcpcb_mem *tm; 1589 struct tcpcb *tp; 1590 #ifdef INET6 1591 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1592 #endif /* INET6 */ 1593 1594 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1595 if (tm == NULL) 1596 return (NULL); 1597 tp = &tm->tcb; 1598 1599 /* Initialise cc_var struct for this tcpcb. */ 1600 tp->ccv = &tm->ccv; 1601 tp->ccv->type = IPPROTO_TCP; 1602 tp->ccv->ccvc.tcp = tp; 1603 rw_rlock(&tcp_function_lock); 1604 tp->t_fb = tcp_func_set_ptr; 1605 refcount_acquire(&tp->t_fb->tfb_refcnt); 1606 rw_runlock(&tcp_function_lock); 1607 /* 1608 * Use the current system default CC algorithm. 1609 */ 1610 CC_LIST_RLOCK(); 1611 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1612 CC_ALGO(tp) = CC_DEFAULT(); 1613 CC_LIST_RUNLOCK(); 1614 1615 if (CC_ALGO(tp)->cb_init != NULL) 1616 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1617 if (tp->t_fb->tfb_tcp_fb_fini) 1618 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1619 refcount_release(&tp->t_fb->tfb_refcnt); 1620 uma_zfree(V_tcpcb_zone, tm); 1621 return (NULL); 1622 } 1623 1624 #ifdef TCP_HHOOK 1625 tp->osd = &tm->osd; 1626 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1627 if (tp->t_fb->tfb_tcp_fb_fini) 1628 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1629 refcount_release(&tp->t_fb->tfb_refcnt); 1630 uma_zfree(V_tcpcb_zone, tm); 1631 return (NULL); 1632 } 1633 #endif 1634 1635 #ifdef VIMAGE 1636 tp->t_vnet = inp->inp_vnet; 1637 #endif 1638 tp->t_timers = &tm->tt; 1639 TAILQ_INIT(&tp->t_segq); 1640 tp->t_maxseg = 1641 #ifdef INET6 1642 isipv6 ? V_tcp_v6mssdflt : 1643 #endif /* INET6 */ 1644 V_tcp_mssdflt; 1645 1646 /* Set up our timeouts. */ 1647 callout_init(&tp->t_timers->tt_rexmt, 1); 1648 callout_init(&tp->t_timers->tt_persist, 1); 1649 callout_init(&tp->t_timers->tt_keep, 1); 1650 callout_init(&tp->t_timers->tt_2msl, 1); 1651 callout_init(&tp->t_timers->tt_delack, 1); 1652 1653 if (V_tcp_do_rfc1323) 1654 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1655 if (V_tcp_do_sack) 1656 tp->t_flags |= TF_SACK_PERMIT; 1657 TAILQ_INIT(&tp->snd_holes); 1658 /* 1659 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1660 * is called. 1661 */ 1662 in_pcbref(inp); /* Reference for tcpcb */ 1663 tp->t_inpcb = inp; 1664 1665 /* 1666 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1667 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1668 * reasonable initial retransmit time. 1669 */ 1670 tp->t_srtt = TCPTV_SRTTBASE; 1671 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1672 tp->t_rttmin = tcp_rexmit_min; 1673 tp->t_rxtcur = tcp_rexmit_initial; 1674 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1675 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1676 tp->t_rcvtime = ticks; 1677 /* 1678 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1679 * because the socket may be bound to an IPv6 wildcard address, 1680 * which may match an IPv4-mapped IPv6 address. 1681 */ 1682 inp->inp_ip_ttl = V_ip_defttl; 1683 inp->inp_ppcb = tp; 1684 #ifdef TCPPCAP 1685 /* 1686 * Init the TCP PCAP queues. 1687 */ 1688 tcp_pcap_tcpcb_init(tp); 1689 #endif 1690 #ifdef TCP_BLACKBOX 1691 /* Initialize the per-TCPCB log data. */ 1692 tcp_log_tcpcbinit(tp); 1693 #endif 1694 if (tp->t_fb->tfb_tcp_fb_init) { 1695 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1696 } 1697 return (tp); /* XXX */ 1698 } 1699 1700 /* 1701 * Switch the congestion control algorithm back to NewReno for any active 1702 * control blocks using an algorithm which is about to go away. 1703 * This ensures the CC framework can allow the unload to proceed without leaving 1704 * any dangling pointers which would trigger a panic. 1705 * Returning non-zero would inform the CC framework that something went wrong 1706 * and it would be unsafe to allow the unload to proceed. However, there is no 1707 * way for this to occur with this implementation so we always return zero. 1708 */ 1709 int 1710 tcp_ccalgounload(struct cc_algo *unload_algo) 1711 { 1712 struct cc_algo *tmpalgo; 1713 struct inpcb *inp; 1714 struct tcpcb *tp; 1715 VNET_ITERATOR_DECL(vnet_iter); 1716 1717 /* 1718 * Check all active control blocks across all network stacks and change 1719 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1720 * requires cleanup code to be run, call it. 1721 */ 1722 VNET_LIST_RLOCK(); 1723 VNET_FOREACH(vnet_iter) { 1724 CURVNET_SET(vnet_iter); 1725 INP_INFO_WLOCK(&V_tcbinfo); 1726 /* 1727 * New connections already part way through being initialised 1728 * with the CC algo we're removing will not race with this code 1729 * because the INP_INFO_WLOCK is held during initialisation. We 1730 * therefore don't enter the loop below until the connection 1731 * list has stabilised. 1732 */ 1733 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 1734 INP_WLOCK(inp); 1735 /* Important to skip tcptw structs. */ 1736 if (!(inp->inp_flags & INP_TIMEWAIT) && 1737 (tp = intotcpcb(inp)) != NULL) { 1738 /* 1739 * By holding INP_WLOCK here, we are assured 1740 * that the connection is not currently 1741 * executing inside the CC module's functions 1742 * i.e. it is safe to make the switch back to 1743 * NewReno. 1744 */ 1745 if (CC_ALGO(tp) == unload_algo) { 1746 tmpalgo = CC_ALGO(tp); 1747 if (tmpalgo->cb_destroy != NULL) 1748 tmpalgo->cb_destroy(tp->ccv); 1749 CC_DATA(tp) = NULL; 1750 /* 1751 * NewReno may allocate memory on 1752 * demand for certain stateful 1753 * configuration as needed, but is 1754 * coded to never fail on memory 1755 * allocation failure so it is a safe 1756 * fallback. 1757 */ 1758 CC_ALGO(tp) = &newreno_cc_algo; 1759 } 1760 } 1761 INP_WUNLOCK(inp); 1762 } 1763 INP_INFO_WUNLOCK(&V_tcbinfo); 1764 CURVNET_RESTORE(); 1765 } 1766 VNET_LIST_RUNLOCK(); 1767 1768 return (0); 1769 } 1770 1771 /* 1772 * Drop a TCP connection, reporting 1773 * the specified error. If connection is synchronized, 1774 * then send a RST to peer. 1775 */ 1776 struct tcpcb * 1777 tcp_drop(struct tcpcb *tp, int errno) 1778 { 1779 struct socket *so = tp->t_inpcb->inp_socket; 1780 1781 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1782 INP_WLOCK_ASSERT(tp->t_inpcb); 1783 1784 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1785 tcp_state_change(tp, TCPS_CLOSED); 1786 (void) tp->t_fb->tfb_tcp_output(tp); 1787 TCPSTAT_INC(tcps_drops); 1788 } else 1789 TCPSTAT_INC(tcps_conndrops); 1790 if (errno == ETIMEDOUT && tp->t_softerror) 1791 errno = tp->t_softerror; 1792 so->so_error = errno; 1793 return (tcp_close(tp)); 1794 } 1795 1796 void 1797 tcp_discardcb(struct tcpcb *tp) 1798 { 1799 struct inpcb *inp = tp->t_inpcb; 1800 struct socket *so = inp->inp_socket; 1801 #ifdef INET6 1802 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1803 #endif /* INET6 */ 1804 int released __unused; 1805 1806 INP_WLOCK_ASSERT(inp); 1807 1808 /* 1809 * Make sure that all of our timers are stopped before we delete the 1810 * PCB. 1811 * 1812 * If stopping a timer fails, we schedule a discard function in same 1813 * callout, and the last discard function called will take care of 1814 * deleting the tcpcb. 1815 */ 1816 tp->t_timers->tt_draincnt = 0; 1817 tcp_timer_stop(tp, TT_REXMT); 1818 tcp_timer_stop(tp, TT_PERSIST); 1819 tcp_timer_stop(tp, TT_KEEP); 1820 tcp_timer_stop(tp, TT_2MSL); 1821 tcp_timer_stop(tp, TT_DELACK); 1822 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1823 /* 1824 * Call the stop-all function of the methods, 1825 * this function should call the tcp_timer_stop() 1826 * method with each of the function specific timeouts. 1827 * That stop will be called via the tfb_tcp_timer_stop() 1828 * which should use the async drain function of the 1829 * callout system (see tcp_var.h). 1830 */ 1831 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1832 } 1833 1834 /* 1835 * If we got enough samples through the srtt filter, 1836 * save the rtt and rttvar in the routing entry. 1837 * 'Enough' is arbitrarily defined as 4 rtt samples. 1838 * 4 samples is enough for the srtt filter to converge 1839 * to within enough % of the correct value; fewer samples 1840 * and we could save a bogus rtt. The danger is not high 1841 * as tcp quickly recovers from everything. 1842 * XXX: Works very well but needs some more statistics! 1843 */ 1844 if (tp->t_rttupdated >= 4) { 1845 struct hc_metrics_lite metrics; 1846 uint32_t ssthresh; 1847 1848 bzero(&metrics, sizeof(metrics)); 1849 /* 1850 * Update the ssthresh always when the conditions below 1851 * are satisfied. This gives us better new start value 1852 * for the congestion avoidance for new connections. 1853 * ssthresh is only set if packet loss occurred on a session. 1854 * 1855 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1856 * being torn down. Ideally this code would not use 'so'. 1857 */ 1858 ssthresh = tp->snd_ssthresh; 1859 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1860 /* 1861 * convert the limit from user data bytes to 1862 * packets then to packet data bytes. 1863 */ 1864 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1865 if (ssthresh < 2) 1866 ssthresh = 2; 1867 ssthresh *= (tp->t_maxseg + 1868 #ifdef INET6 1869 (isipv6 ? sizeof (struct ip6_hdr) + 1870 sizeof (struct tcphdr) : 1871 #endif 1872 sizeof (struct tcpiphdr) 1873 #ifdef INET6 1874 ) 1875 #endif 1876 ); 1877 } else 1878 ssthresh = 0; 1879 metrics.rmx_ssthresh = ssthresh; 1880 1881 metrics.rmx_rtt = tp->t_srtt; 1882 metrics.rmx_rttvar = tp->t_rttvar; 1883 metrics.rmx_cwnd = tp->snd_cwnd; 1884 metrics.rmx_sendpipe = 0; 1885 metrics.rmx_recvpipe = 0; 1886 1887 tcp_hc_update(&inp->inp_inc, &metrics); 1888 } 1889 1890 /* free the reassembly queue, if any */ 1891 tcp_reass_flush(tp); 1892 1893 #ifdef TCP_OFFLOAD 1894 /* Disconnect offload device, if any. */ 1895 if (tp->t_flags & TF_TOE) 1896 tcp_offload_detach(tp); 1897 #endif 1898 1899 tcp_free_sackholes(tp); 1900 1901 #ifdef TCPPCAP 1902 /* Free the TCP PCAP queues. */ 1903 tcp_pcap_drain(&(tp->t_inpkts)); 1904 tcp_pcap_drain(&(tp->t_outpkts)); 1905 #endif 1906 1907 /* Allow the CC algorithm to clean up after itself. */ 1908 if (CC_ALGO(tp)->cb_destroy != NULL) 1909 CC_ALGO(tp)->cb_destroy(tp->ccv); 1910 CC_DATA(tp) = NULL; 1911 1912 #ifdef TCP_HHOOK 1913 khelp_destroy_osd(tp->osd); 1914 #endif 1915 1916 CC_ALGO(tp) = NULL; 1917 inp->inp_ppcb = NULL; 1918 if (tp->t_timers->tt_draincnt == 0) { 1919 /* We own the last reference on tcpcb, let's free it. */ 1920 #ifdef TCP_BLACKBOX 1921 tcp_log_tcpcbfini(tp); 1922 #endif 1923 TCPSTATES_DEC(tp->t_state); 1924 if (tp->t_fb->tfb_tcp_fb_fini) 1925 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1926 refcount_release(&tp->t_fb->tfb_refcnt); 1927 tp->t_inpcb = NULL; 1928 uma_zfree(V_tcpcb_zone, tp); 1929 released = in_pcbrele_wlocked(inp); 1930 KASSERT(!released, ("%s: inp %p should not have been released " 1931 "here", __func__, inp)); 1932 } 1933 } 1934 1935 void 1936 tcp_timer_discard(void *ptp) 1937 { 1938 struct inpcb *inp; 1939 struct tcpcb *tp; 1940 struct epoch_tracker et; 1941 1942 tp = (struct tcpcb *)ptp; 1943 CURVNET_SET(tp->t_vnet); 1944 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 1945 inp = tp->t_inpcb; 1946 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1947 __func__, tp)); 1948 INP_WLOCK(inp); 1949 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1950 ("%s: tcpcb has to be stopped here", __func__)); 1951 tp->t_timers->tt_draincnt--; 1952 if (tp->t_timers->tt_draincnt == 0) { 1953 /* We own the last reference on this tcpcb, let's free it. */ 1954 #ifdef TCP_BLACKBOX 1955 tcp_log_tcpcbfini(tp); 1956 #endif 1957 TCPSTATES_DEC(tp->t_state); 1958 if (tp->t_fb->tfb_tcp_fb_fini) 1959 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1960 refcount_release(&tp->t_fb->tfb_refcnt); 1961 tp->t_inpcb = NULL; 1962 uma_zfree(V_tcpcb_zone, tp); 1963 if (in_pcbrele_wlocked(inp)) { 1964 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1965 CURVNET_RESTORE(); 1966 return; 1967 } 1968 } 1969 INP_WUNLOCK(inp); 1970 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1971 CURVNET_RESTORE(); 1972 } 1973 1974 /* 1975 * Attempt to close a TCP control block, marking it as dropped, and freeing 1976 * the socket if we hold the only reference. 1977 */ 1978 struct tcpcb * 1979 tcp_close(struct tcpcb *tp) 1980 { 1981 struct inpcb *inp = tp->t_inpcb; 1982 struct socket *so; 1983 1984 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1985 INP_WLOCK_ASSERT(inp); 1986 1987 #ifdef TCP_OFFLOAD 1988 if (tp->t_state == TCPS_LISTEN) 1989 tcp_offload_listen_stop(tp); 1990 #endif 1991 /* 1992 * This releases the TFO pending counter resource for TFO listen 1993 * sockets as well as passively-created TFO sockets that transition 1994 * from SYN_RECEIVED to CLOSED. 1995 */ 1996 if (tp->t_tfo_pending) { 1997 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1998 tp->t_tfo_pending = NULL; 1999 } 2000 in_pcbdrop(inp); 2001 TCPSTAT_INC(tcps_closed); 2002 if (tp->t_state != TCPS_CLOSED) 2003 tcp_state_change(tp, TCPS_CLOSED); 2004 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2005 so = inp->inp_socket; 2006 soisdisconnected(so); 2007 if (inp->inp_flags & INP_SOCKREF) { 2008 KASSERT(so->so_state & SS_PROTOREF, 2009 ("tcp_close: !SS_PROTOREF")); 2010 inp->inp_flags &= ~INP_SOCKREF; 2011 INP_WUNLOCK(inp); 2012 SOCK_LOCK(so); 2013 so->so_state &= ~SS_PROTOREF; 2014 sofree(so); 2015 return (NULL); 2016 } 2017 return (tp); 2018 } 2019 2020 void 2021 tcp_drain(void) 2022 { 2023 VNET_ITERATOR_DECL(vnet_iter); 2024 2025 if (!do_tcpdrain) 2026 return; 2027 2028 VNET_LIST_RLOCK_NOSLEEP(); 2029 VNET_FOREACH(vnet_iter) { 2030 CURVNET_SET(vnet_iter); 2031 struct inpcb *inpb; 2032 struct tcpcb *tcpb; 2033 2034 /* 2035 * Walk the tcpbs, if existing, and flush the reassembly queue, 2036 * if there is one... 2037 * XXX: The "Net/3" implementation doesn't imply that the TCP 2038 * reassembly queue should be flushed, but in a situation 2039 * where we're really low on mbufs, this is potentially 2040 * useful. 2041 */ 2042 INP_INFO_WLOCK(&V_tcbinfo); 2043 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2044 INP_WLOCK(inpb); 2045 if (inpb->inp_flags & INP_TIMEWAIT) { 2046 INP_WUNLOCK(inpb); 2047 continue; 2048 } 2049 if ((tcpb = intotcpcb(inpb)) != NULL) { 2050 tcp_reass_flush(tcpb); 2051 tcp_clean_sackreport(tcpb); 2052 #ifdef TCP_BLACKBOX 2053 tcp_log_drain(tcpb); 2054 #endif 2055 #ifdef TCPPCAP 2056 if (tcp_pcap_aggressive_free) { 2057 /* Free the TCP PCAP queues. */ 2058 tcp_pcap_drain(&(tcpb->t_inpkts)); 2059 tcp_pcap_drain(&(tcpb->t_outpkts)); 2060 } 2061 #endif 2062 } 2063 INP_WUNLOCK(inpb); 2064 } 2065 INP_INFO_WUNLOCK(&V_tcbinfo); 2066 CURVNET_RESTORE(); 2067 } 2068 VNET_LIST_RUNLOCK_NOSLEEP(); 2069 } 2070 2071 /* 2072 * Notify a tcp user of an asynchronous error; 2073 * store error as soft error, but wake up user 2074 * (for now, won't do anything until can select for soft error). 2075 * 2076 * Do not wake up user since there currently is no mechanism for 2077 * reporting soft errors (yet - a kqueue filter may be added). 2078 */ 2079 static struct inpcb * 2080 tcp_notify(struct inpcb *inp, int error) 2081 { 2082 struct tcpcb *tp; 2083 2084 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2085 INP_WLOCK_ASSERT(inp); 2086 2087 if ((inp->inp_flags & INP_TIMEWAIT) || 2088 (inp->inp_flags & INP_DROPPED)) 2089 return (inp); 2090 2091 tp = intotcpcb(inp); 2092 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2093 2094 /* 2095 * Ignore some errors if we are hooked up. 2096 * If connection hasn't completed, has retransmitted several times, 2097 * and receives a second error, give up now. This is better 2098 * than waiting a long time to establish a connection that 2099 * can never complete. 2100 */ 2101 if (tp->t_state == TCPS_ESTABLISHED && 2102 (error == EHOSTUNREACH || error == ENETUNREACH || 2103 error == EHOSTDOWN)) { 2104 if (inp->inp_route.ro_rt) { 2105 RTFREE(inp->inp_route.ro_rt); 2106 inp->inp_route.ro_rt = (struct rtentry *)NULL; 2107 } 2108 return (inp); 2109 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2110 tp->t_softerror) { 2111 tp = tcp_drop(tp, error); 2112 if (tp != NULL) 2113 return (inp); 2114 else 2115 return (NULL); 2116 } else { 2117 tp->t_softerror = error; 2118 return (inp); 2119 } 2120 #if 0 2121 wakeup( &so->so_timeo); 2122 sorwakeup(so); 2123 sowwakeup(so); 2124 #endif 2125 } 2126 2127 static int 2128 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2129 { 2130 int error, i, m, n, pcb_count; 2131 struct inpcb *inp, **inp_list; 2132 inp_gen_t gencnt; 2133 struct xinpgen xig; 2134 struct epoch_tracker et; 2135 2136 /* 2137 * The process of preparing the TCB list is too time-consuming and 2138 * resource-intensive to repeat twice on every request. 2139 */ 2140 if (req->oldptr == NULL) { 2141 n = V_tcbinfo.ipi_count + 2142 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2143 n += imax(n / 8, 10); 2144 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2145 return (0); 2146 } 2147 2148 if (req->newptr != NULL) 2149 return (EPERM); 2150 2151 /* 2152 * OK, now we're committed to doing something. 2153 */ 2154 INP_LIST_RLOCK(&V_tcbinfo); 2155 gencnt = V_tcbinfo.ipi_gencnt; 2156 n = V_tcbinfo.ipi_count; 2157 INP_LIST_RUNLOCK(&V_tcbinfo); 2158 2159 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2160 2161 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 2162 + (n + m) * sizeof(struct xtcpcb)); 2163 if (error != 0) 2164 return (error); 2165 2166 bzero(&xig, sizeof(xig)); 2167 xig.xig_len = sizeof xig; 2168 xig.xig_count = n + m; 2169 xig.xig_gen = gencnt; 2170 xig.xig_sogen = so_gencnt; 2171 error = SYSCTL_OUT(req, &xig, sizeof xig); 2172 if (error) 2173 return (error); 2174 2175 error = syncache_pcblist(req, m, &pcb_count); 2176 if (error) 2177 return (error); 2178 2179 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 2180 2181 INP_INFO_WLOCK(&V_tcbinfo); 2182 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 2183 inp != NULL && i < n; inp = CK_LIST_NEXT(inp, inp_list)) { 2184 INP_WLOCK(inp); 2185 if (inp->inp_gencnt <= gencnt) { 2186 /* 2187 * XXX: This use of cr_cansee(), introduced with 2188 * TCP state changes, is not quite right, but for 2189 * now, better than nothing. 2190 */ 2191 if (inp->inp_flags & INP_TIMEWAIT) { 2192 if (intotw(inp) != NULL) 2193 error = cr_cansee(req->td->td_ucred, 2194 intotw(inp)->tw_cred); 2195 else 2196 error = EINVAL; /* Skip this inp. */ 2197 } else 2198 error = cr_canseeinpcb(req->td->td_ucred, inp); 2199 if (error == 0) { 2200 in_pcbref(inp); 2201 inp_list[i++] = inp; 2202 } 2203 } 2204 INP_WUNLOCK(inp); 2205 } 2206 INP_INFO_WUNLOCK(&V_tcbinfo); 2207 n = i; 2208 2209 error = 0; 2210 for (i = 0; i < n; i++) { 2211 inp = inp_list[i]; 2212 INP_RLOCK(inp); 2213 if (inp->inp_gencnt <= gencnt) { 2214 struct xtcpcb xt; 2215 2216 tcp_inptoxtp(inp, &xt); 2217 INP_RUNLOCK(inp); 2218 error = SYSCTL_OUT(req, &xt, sizeof xt); 2219 } else 2220 INP_RUNLOCK(inp); 2221 } 2222 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2223 for (i = 0; i < n; i++) { 2224 inp = inp_list[i]; 2225 INP_RLOCK(inp); 2226 if (!in_pcbrele_rlocked(inp)) 2227 INP_RUNLOCK(inp); 2228 } 2229 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2230 2231 if (!error) { 2232 /* 2233 * Give the user an updated idea of our state. 2234 * If the generation differs from what we told 2235 * her before, she knows that something happened 2236 * while we were processing this request, and it 2237 * might be necessary to retry. 2238 */ 2239 INP_LIST_RLOCK(&V_tcbinfo); 2240 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2241 xig.xig_sogen = so_gencnt; 2242 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 2243 INP_LIST_RUNLOCK(&V_tcbinfo); 2244 error = SYSCTL_OUT(req, &xig, sizeof xig); 2245 } 2246 free(inp_list, M_TEMP); 2247 return (error); 2248 } 2249 2250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2251 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 2252 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 2253 2254 #ifdef INET 2255 static int 2256 tcp_getcred(SYSCTL_HANDLER_ARGS) 2257 { 2258 struct xucred xuc; 2259 struct sockaddr_in addrs[2]; 2260 struct inpcb *inp; 2261 int error; 2262 2263 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2264 if (error) 2265 return (error); 2266 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2267 if (error) 2268 return (error); 2269 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2270 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2271 if (inp != NULL) { 2272 if (inp->inp_socket == NULL) 2273 error = ENOENT; 2274 if (error == 0) 2275 error = cr_canseeinpcb(req->td->td_ucred, inp); 2276 if (error == 0) 2277 cru2x(inp->inp_cred, &xuc); 2278 INP_RUNLOCK(inp); 2279 } else 2280 error = ENOENT; 2281 if (error == 0) 2282 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2283 return (error); 2284 } 2285 2286 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2287 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2288 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2289 #endif /* INET */ 2290 2291 #ifdef INET6 2292 static int 2293 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2294 { 2295 struct xucred xuc; 2296 struct sockaddr_in6 addrs[2]; 2297 struct inpcb *inp; 2298 int error; 2299 #ifdef INET 2300 int mapped = 0; 2301 #endif 2302 2303 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2304 if (error) 2305 return (error); 2306 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2307 if (error) 2308 return (error); 2309 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2310 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2311 return (error); 2312 } 2313 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2314 #ifdef INET 2315 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2316 mapped = 1; 2317 else 2318 #endif 2319 return (EINVAL); 2320 } 2321 2322 #ifdef INET 2323 if (mapped == 1) 2324 inp = in_pcblookup(&V_tcbinfo, 2325 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2326 addrs[1].sin6_port, 2327 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2328 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2329 else 2330 #endif 2331 inp = in6_pcblookup(&V_tcbinfo, 2332 &addrs[1].sin6_addr, addrs[1].sin6_port, 2333 &addrs[0].sin6_addr, addrs[0].sin6_port, 2334 INPLOOKUP_RLOCKPCB, NULL); 2335 if (inp != NULL) { 2336 if (inp->inp_socket == NULL) 2337 error = ENOENT; 2338 if (error == 0) 2339 error = cr_canseeinpcb(req->td->td_ucred, inp); 2340 if (error == 0) 2341 cru2x(inp->inp_cred, &xuc); 2342 INP_RUNLOCK(inp); 2343 } else 2344 error = ENOENT; 2345 if (error == 0) 2346 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2347 return (error); 2348 } 2349 2350 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2351 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2352 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2353 #endif /* INET6 */ 2354 2355 2356 #ifdef INET 2357 void 2358 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2359 { 2360 struct ip *ip = vip; 2361 struct tcphdr *th; 2362 struct in_addr faddr; 2363 struct inpcb *inp; 2364 struct tcpcb *tp; 2365 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2366 struct icmp *icp; 2367 struct in_conninfo inc; 2368 struct epoch_tracker et; 2369 tcp_seq icmp_tcp_seq; 2370 int mtu; 2371 2372 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2373 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2374 return; 2375 2376 if (cmd == PRC_MSGSIZE) 2377 notify = tcp_mtudisc_notify; 2378 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2379 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2380 cmd == PRC_TIMXCEED_INTRANS) && ip) 2381 notify = tcp_drop_syn_sent; 2382 2383 /* 2384 * Hostdead is ugly because it goes linearly through all PCBs. 2385 * XXX: We never get this from ICMP, otherwise it makes an 2386 * excellent DoS attack on machines with many connections. 2387 */ 2388 else if (cmd == PRC_HOSTDEAD) 2389 ip = NULL; 2390 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2391 return; 2392 2393 if (ip == NULL) { 2394 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2395 return; 2396 } 2397 2398 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2399 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2400 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2401 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2402 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2403 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2404 /* signal EHOSTDOWN, as it flushes the cached route */ 2405 inp = (*notify)(inp, EHOSTDOWN); 2406 goto out; 2407 } 2408 icmp_tcp_seq = th->th_seq; 2409 if (inp != NULL) { 2410 if (!(inp->inp_flags & INP_TIMEWAIT) && 2411 !(inp->inp_flags & INP_DROPPED) && 2412 !(inp->inp_socket == NULL)) { 2413 tp = intotcpcb(inp); 2414 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2415 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2416 if (cmd == PRC_MSGSIZE) { 2417 /* 2418 * MTU discovery: 2419 * If we got a needfrag set the MTU 2420 * in the route to the suggested new 2421 * value (if given) and then notify. 2422 */ 2423 mtu = ntohs(icp->icmp_nextmtu); 2424 /* 2425 * If no alternative MTU was 2426 * proposed, try the next smaller 2427 * one. 2428 */ 2429 if (!mtu) 2430 mtu = ip_next_mtu( 2431 ntohs(ip->ip_len), 1); 2432 if (mtu < V_tcp_minmss + 2433 sizeof(struct tcpiphdr)) 2434 mtu = V_tcp_minmss + 2435 sizeof(struct tcpiphdr); 2436 /* 2437 * Only process the offered MTU if it 2438 * is smaller than the current one. 2439 */ 2440 if (mtu < tp->t_maxseg + 2441 sizeof(struct tcpiphdr)) { 2442 bzero(&inc, sizeof(inc)); 2443 inc.inc_faddr = faddr; 2444 inc.inc_fibnum = 2445 inp->inp_inc.inc_fibnum; 2446 tcp_hc_updatemtu(&inc, mtu); 2447 tcp_mtudisc(inp, mtu); 2448 } 2449 } else 2450 inp = (*notify)(inp, 2451 inetctlerrmap[cmd]); 2452 } 2453 } 2454 } else { 2455 bzero(&inc, sizeof(inc)); 2456 inc.inc_fport = th->th_dport; 2457 inc.inc_lport = th->th_sport; 2458 inc.inc_faddr = faddr; 2459 inc.inc_laddr = ip->ip_src; 2460 syncache_unreach(&inc, icmp_tcp_seq); 2461 } 2462 out: 2463 if (inp != NULL) 2464 INP_WUNLOCK(inp); 2465 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2466 } 2467 #endif /* INET */ 2468 2469 #ifdef INET6 2470 void 2471 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2472 { 2473 struct in6_addr *dst; 2474 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2475 struct ip6_hdr *ip6; 2476 struct mbuf *m; 2477 struct inpcb *inp; 2478 struct tcpcb *tp; 2479 struct icmp6_hdr *icmp6; 2480 struct ip6ctlparam *ip6cp = NULL; 2481 const struct sockaddr_in6 *sa6_src = NULL; 2482 struct in_conninfo inc; 2483 struct epoch_tracker et; 2484 struct tcp_ports { 2485 uint16_t th_sport; 2486 uint16_t th_dport; 2487 } t_ports; 2488 tcp_seq icmp_tcp_seq; 2489 unsigned int mtu; 2490 unsigned int off; 2491 2492 if (sa->sa_family != AF_INET6 || 2493 sa->sa_len != sizeof(struct sockaddr_in6)) 2494 return; 2495 2496 /* if the parameter is from icmp6, decode it. */ 2497 if (d != NULL) { 2498 ip6cp = (struct ip6ctlparam *)d; 2499 icmp6 = ip6cp->ip6c_icmp6; 2500 m = ip6cp->ip6c_m; 2501 ip6 = ip6cp->ip6c_ip6; 2502 off = ip6cp->ip6c_off; 2503 sa6_src = ip6cp->ip6c_src; 2504 dst = ip6cp->ip6c_finaldst; 2505 } else { 2506 m = NULL; 2507 ip6 = NULL; 2508 off = 0; /* fool gcc */ 2509 sa6_src = &sa6_any; 2510 dst = NULL; 2511 } 2512 2513 if (cmd == PRC_MSGSIZE) 2514 notify = tcp_mtudisc_notify; 2515 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2516 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2517 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2518 notify = tcp_drop_syn_sent; 2519 2520 /* 2521 * Hostdead is ugly because it goes linearly through all PCBs. 2522 * XXX: We never get this from ICMP, otherwise it makes an 2523 * excellent DoS attack on machines with many connections. 2524 */ 2525 else if (cmd == PRC_HOSTDEAD) 2526 ip6 = NULL; 2527 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2528 return; 2529 2530 if (ip6 == NULL) { 2531 in6_pcbnotify(&V_tcbinfo, sa, 0, 2532 (const struct sockaddr *)sa6_src, 2533 0, cmd, NULL, notify); 2534 return; 2535 } 2536 2537 /* Check if we can safely get the ports from the tcp hdr */ 2538 if (m == NULL || 2539 (m->m_pkthdr.len < 2540 (int32_t) (off + sizeof(struct tcp_ports)))) { 2541 return; 2542 } 2543 bzero(&t_ports, sizeof(struct tcp_ports)); 2544 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2545 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 2546 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2547 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2548 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2549 /* signal EHOSTDOWN, as it flushes the cached route */ 2550 inp = (*notify)(inp, EHOSTDOWN); 2551 goto out; 2552 } 2553 off += sizeof(struct tcp_ports); 2554 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2555 goto out; 2556 } 2557 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2558 if (inp != NULL) { 2559 if (!(inp->inp_flags & INP_TIMEWAIT) && 2560 !(inp->inp_flags & INP_DROPPED) && 2561 !(inp->inp_socket == NULL)) { 2562 tp = intotcpcb(inp); 2563 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2564 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2565 if (cmd == PRC_MSGSIZE) { 2566 /* 2567 * MTU discovery: 2568 * If we got a needfrag set the MTU 2569 * in the route to the suggested new 2570 * value (if given) and then notify. 2571 */ 2572 mtu = ntohl(icmp6->icmp6_mtu); 2573 /* 2574 * If no alternative MTU was 2575 * proposed, or the proposed 2576 * MTU was too small, set to 2577 * the min. 2578 */ 2579 if (mtu < IPV6_MMTU) 2580 mtu = IPV6_MMTU - 8; 2581 bzero(&inc, sizeof(inc)); 2582 inc.inc_fibnum = M_GETFIB(m); 2583 inc.inc_flags |= INC_ISIPV6; 2584 inc.inc6_faddr = *dst; 2585 if (in6_setscope(&inc.inc6_faddr, 2586 m->m_pkthdr.rcvif, NULL)) 2587 goto out; 2588 /* 2589 * Only process the offered MTU if it 2590 * is smaller than the current one. 2591 */ 2592 if (mtu < tp->t_maxseg + 2593 sizeof (struct tcphdr) + 2594 sizeof (struct ip6_hdr)) { 2595 tcp_hc_updatemtu(&inc, mtu); 2596 tcp_mtudisc(inp, mtu); 2597 ICMP6STAT_INC(icp6s_pmtuchg); 2598 } 2599 } else 2600 inp = (*notify)(inp, 2601 inet6ctlerrmap[cmd]); 2602 } 2603 } 2604 } else { 2605 bzero(&inc, sizeof(inc)); 2606 inc.inc_fibnum = M_GETFIB(m); 2607 inc.inc_flags |= INC_ISIPV6; 2608 inc.inc_fport = t_ports.th_dport; 2609 inc.inc_lport = t_ports.th_sport; 2610 inc.inc6_faddr = *dst; 2611 inc.inc6_laddr = ip6->ip6_src; 2612 syncache_unreach(&inc, icmp_tcp_seq); 2613 } 2614 out: 2615 if (inp != NULL) 2616 INP_WUNLOCK(inp); 2617 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 2618 } 2619 #endif /* INET6 */ 2620 2621 static uint32_t 2622 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 2623 { 2624 MD5_CTX ctx; 2625 uint32_t hash[4]; 2626 2627 MD5Init(&ctx); 2628 MD5Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 2629 MD5Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 2630 switch (inc->inc_flags & INC_ISIPV6) { 2631 #ifdef INET 2632 case 0: 2633 MD5Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 2634 MD5Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 2635 break; 2636 #endif 2637 #ifdef INET6 2638 case INC_ISIPV6: 2639 MD5Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 2640 MD5Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 2641 break; 2642 #endif 2643 } 2644 MD5Update(&ctx, key, len); 2645 MD5Final((unsigned char *)hash, &ctx); 2646 2647 return (hash[0]); 2648 } 2649 2650 uint32_t 2651 tcp_new_ts_offset(struct in_conninfo *inc) 2652 { 2653 struct in_conninfo inc_store, *local_inc; 2654 2655 if (!V_tcp_ts_offset_per_conn) { 2656 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 2657 inc_store.inc_lport = 0; 2658 inc_store.inc_fport = 0; 2659 local_inc = &inc_store; 2660 } else { 2661 local_inc = inc; 2662 } 2663 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 2664 sizeof(V_ts_offset_secret))); 2665 } 2666 2667 /* 2668 * Following is where TCP initial sequence number generation occurs. 2669 * 2670 * There are two places where we must use initial sequence numbers: 2671 * 1. In SYN-ACK packets. 2672 * 2. In SYN packets. 2673 * 2674 * All ISNs for SYN-ACK packets are generated by the syncache. See 2675 * tcp_syncache.c for details. 2676 * 2677 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2678 * depends on this property. In addition, these ISNs should be 2679 * unguessable so as to prevent connection hijacking. To satisfy 2680 * the requirements of this situation, the algorithm outlined in 2681 * RFC 1948 is used, with only small modifications. 2682 * 2683 * Implementation details: 2684 * 2685 * Time is based off the system timer, and is corrected so that it 2686 * increases by one megabyte per second. This allows for proper 2687 * recycling on high speed LANs while still leaving over an hour 2688 * before rollover. 2689 * 2690 * As reading the *exact* system time is too expensive to be done 2691 * whenever setting up a TCP connection, we increment the time 2692 * offset in two ways. First, a small random positive increment 2693 * is added to isn_offset for each connection that is set up. 2694 * Second, the function tcp_isn_tick fires once per clock tick 2695 * and increments isn_offset as necessary so that sequence numbers 2696 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2697 * random positive increments serve only to ensure that the same 2698 * exact sequence number is never sent out twice (as could otherwise 2699 * happen when a port is recycled in less than the system tick 2700 * interval.) 2701 * 2702 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2703 * between seeding of isn_secret. This is normally set to zero, 2704 * as reseeding should not be necessary. 2705 * 2706 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2707 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 2708 * general, this means holding an exclusive (write) lock. 2709 */ 2710 2711 #define ISN_BYTES_PER_SECOND 1048576 2712 #define ISN_STATIC_INCREMENT 4096 2713 #define ISN_RANDOM_INCREMENT (4096 - 1) 2714 #define ISN_SECRET_LENGTH 32 2715 2716 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 2717 VNET_DEFINE_STATIC(int, isn_last); 2718 VNET_DEFINE_STATIC(int, isn_last_reseed); 2719 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 2720 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 2721 2722 #define V_isn_secret VNET(isn_secret) 2723 #define V_isn_last VNET(isn_last) 2724 #define V_isn_last_reseed VNET(isn_last_reseed) 2725 #define V_isn_offset VNET(isn_offset) 2726 #define V_isn_offset_old VNET(isn_offset_old) 2727 2728 tcp_seq 2729 tcp_new_isn(struct in_conninfo *inc) 2730 { 2731 tcp_seq new_isn; 2732 u_int32_t projected_offset; 2733 2734 ISN_LOCK(); 2735 /* Seed if this is the first use, reseed if requested. */ 2736 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2737 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2738 < (u_int)ticks))) { 2739 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 2740 V_isn_last_reseed = ticks; 2741 } 2742 2743 /* Compute the md5 hash and return the ISN. */ 2744 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 2745 sizeof(V_isn_secret)); 2746 V_isn_offset += ISN_STATIC_INCREMENT + 2747 (arc4random() & ISN_RANDOM_INCREMENT); 2748 if (ticks != V_isn_last) { 2749 projected_offset = V_isn_offset_old + 2750 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2751 if (SEQ_GT(projected_offset, V_isn_offset)) 2752 V_isn_offset = projected_offset; 2753 V_isn_offset_old = V_isn_offset; 2754 V_isn_last = ticks; 2755 } 2756 new_isn += V_isn_offset; 2757 ISN_UNLOCK(); 2758 return (new_isn); 2759 } 2760 2761 /* 2762 * When a specific ICMP unreachable message is received and the 2763 * connection state is SYN-SENT, drop the connection. This behavior 2764 * is controlled by the icmp_may_rst sysctl. 2765 */ 2766 struct inpcb * 2767 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2768 { 2769 struct tcpcb *tp; 2770 2771 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2772 INP_WLOCK_ASSERT(inp); 2773 2774 if ((inp->inp_flags & INP_TIMEWAIT) || 2775 (inp->inp_flags & INP_DROPPED)) 2776 return (inp); 2777 2778 tp = intotcpcb(inp); 2779 if (tp->t_state != TCPS_SYN_SENT) 2780 return (inp); 2781 2782 if (IS_FASTOPEN(tp->t_flags)) 2783 tcp_fastopen_disable_path(tp); 2784 2785 tp = tcp_drop(tp, errno); 2786 if (tp != NULL) 2787 return (inp); 2788 else 2789 return (NULL); 2790 } 2791 2792 /* 2793 * When `need fragmentation' ICMP is received, update our idea of the MSS 2794 * based on the new value. Also nudge TCP to send something, since we 2795 * know the packet we just sent was dropped. 2796 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2797 */ 2798 static struct inpcb * 2799 tcp_mtudisc_notify(struct inpcb *inp, int error) 2800 { 2801 2802 tcp_mtudisc(inp, -1); 2803 return (inp); 2804 } 2805 2806 static void 2807 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2808 { 2809 struct tcpcb *tp; 2810 struct socket *so; 2811 2812 INP_WLOCK_ASSERT(inp); 2813 if ((inp->inp_flags & INP_TIMEWAIT) || 2814 (inp->inp_flags & INP_DROPPED)) 2815 return; 2816 2817 tp = intotcpcb(inp); 2818 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2819 2820 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2821 2822 so = inp->inp_socket; 2823 SOCKBUF_LOCK(&so->so_snd); 2824 /* If the mss is larger than the socket buffer, decrease the mss. */ 2825 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2826 tp->t_maxseg = so->so_snd.sb_hiwat; 2827 SOCKBUF_UNLOCK(&so->so_snd); 2828 2829 TCPSTAT_INC(tcps_mturesent); 2830 tp->t_rtttime = 0; 2831 tp->snd_nxt = tp->snd_una; 2832 tcp_free_sackholes(tp); 2833 tp->snd_recover = tp->snd_max; 2834 if (tp->t_flags & TF_SACK_PERMIT) 2835 EXIT_FASTRECOVERY(tp->t_flags); 2836 tp->t_fb->tfb_tcp_output(tp); 2837 } 2838 2839 #ifdef INET 2840 /* 2841 * Look-up the routing entry to the peer of this inpcb. If no route 2842 * is found and it cannot be allocated, then return 0. This routine 2843 * is called by TCP routines that access the rmx structure and by 2844 * tcp_mss_update to get the peer/interface MTU. 2845 */ 2846 uint32_t 2847 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2848 { 2849 struct nhop4_extended nh4; 2850 struct ifnet *ifp; 2851 uint32_t maxmtu = 0; 2852 2853 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2854 2855 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2856 2857 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2858 NHR_REF, 0, &nh4) != 0) 2859 return (0); 2860 2861 ifp = nh4.nh_ifp; 2862 maxmtu = nh4.nh_mtu; 2863 2864 /* Report additional interface capabilities. */ 2865 if (cap != NULL) { 2866 if (ifp->if_capenable & IFCAP_TSO4 && 2867 ifp->if_hwassist & CSUM_TSO) { 2868 cap->ifcap |= CSUM_TSO; 2869 cap->tsomax = ifp->if_hw_tsomax; 2870 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2871 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2872 } 2873 } 2874 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2875 } 2876 return (maxmtu); 2877 } 2878 #endif /* INET */ 2879 2880 #ifdef INET6 2881 uint32_t 2882 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2883 { 2884 struct nhop6_extended nh6; 2885 struct in6_addr dst6; 2886 uint32_t scopeid; 2887 struct ifnet *ifp; 2888 uint32_t maxmtu = 0; 2889 2890 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2891 2892 if (inc->inc_flags & INC_IPV6MINMTU) 2893 return (IPV6_MMTU); 2894 2895 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2896 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2897 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2898 0, &nh6) != 0) 2899 return (0); 2900 2901 ifp = nh6.nh_ifp; 2902 maxmtu = nh6.nh_mtu; 2903 2904 /* Report additional interface capabilities. */ 2905 if (cap != NULL) { 2906 if (ifp->if_capenable & IFCAP_TSO6 && 2907 ifp->if_hwassist & CSUM_TSO) { 2908 cap->ifcap |= CSUM_TSO; 2909 cap->tsomax = ifp->if_hw_tsomax; 2910 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2911 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2912 } 2913 } 2914 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2915 } 2916 2917 return (maxmtu); 2918 } 2919 #endif /* INET6 */ 2920 2921 /* 2922 * Calculate effective SMSS per RFC5681 definition for a given TCP 2923 * connection at its current state, taking into account SACK and etc. 2924 */ 2925 u_int 2926 tcp_maxseg(const struct tcpcb *tp) 2927 { 2928 u_int optlen; 2929 2930 if (tp->t_flags & TF_NOOPT) 2931 return (tp->t_maxseg); 2932 2933 /* 2934 * Here we have a simplified code from tcp_addoptions(), 2935 * without a proper loop, and having most of paddings hardcoded. 2936 * We might make mistakes with padding here in some edge cases, 2937 * but this is harmless, since result of tcp_maxseg() is used 2938 * only in cwnd and ssthresh estimations. 2939 */ 2940 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2941 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2942 if (tp->t_flags & TF_RCVD_TSTMP) 2943 optlen = TCPOLEN_TSTAMP_APPA; 2944 else 2945 optlen = 0; 2946 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2947 if (tp->t_flags & TF_SIGNATURE) 2948 optlen += PAD(TCPOLEN_SIGNATURE); 2949 #endif 2950 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2951 optlen += TCPOLEN_SACKHDR; 2952 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2953 optlen = PAD(optlen); 2954 } 2955 } else { 2956 if (tp->t_flags & TF_REQ_TSTMP) 2957 optlen = TCPOLEN_TSTAMP_APPA; 2958 else 2959 optlen = PAD(TCPOLEN_MAXSEG); 2960 if (tp->t_flags & TF_REQ_SCALE) 2961 optlen += PAD(TCPOLEN_WINDOW); 2962 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2963 if (tp->t_flags & TF_SIGNATURE) 2964 optlen += PAD(TCPOLEN_SIGNATURE); 2965 #endif 2966 if (tp->t_flags & TF_SACK_PERMIT) 2967 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2968 } 2969 #undef PAD 2970 optlen = min(optlen, TCP_MAXOLEN); 2971 return (tp->t_maxseg - optlen); 2972 } 2973 2974 static int 2975 sysctl_drop(SYSCTL_HANDLER_ARGS) 2976 { 2977 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2978 struct sockaddr_storage addrs[2]; 2979 struct inpcb *inp; 2980 struct tcpcb *tp; 2981 struct tcptw *tw; 2982 struct sockaddr_in *fin, *lin; 2983 struct epoch_tracker et; 2984 #ifdef INET6 2985 struct sockaddr_in6 *fin6, *lin6; 2986 #endif 2987 int error; 2988 2989 inp = NULL; 2990 fin = lin = NULL; 2991 #ifdef INET6 2992 fin6 = lin6 = NULL; 2993 #endif 2994 error = 0; 2995 2996 if (req->oldptr != NULL || req->oldlen != 0) 2997 return (EINVAL); 2998 if (req->newptr == NULL) 2999 return (EPERM); 3000 if (req->newlen < sizeof(addrs)) 3001 return (ENOMEM); 3002 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3003 if (error) 3004 return (error); 3005 3006 switch (addrs[0].ss_family) { 3007 #ifdef INET6 3008 case AF_INET6: 3009 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3010 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3011 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3012 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3013 return (EINVAL); 3014 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3015 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3016 return (EINVAL); 3017 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3018 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3019 fin = (struct sockaddr_in *)&addrs[0]; 3020 lin = (struct sockaddr_in *)&addrs[1]; 3021 break; 3022 } 3023 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3024 if (error) 3025 return (error); 3026 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3027 if (error) 3028 return (error); 3029 break; 3030 #endif 3031 #ifdef INET 3032 case AF_INET: 3033 fin = (struct sockaddr_in *)&addrs[0]; 3034 lin = (struct sockaddr_in *)&addrs[1]; 3035 if (fin->sin_len != sizeof(struct sockaddr_in) || 3036 lin->sin_len != sizeof(struct sockaddr_in)) 3037 return (EINVAL); 3038 break; 3039 #endif 3040 default: 3041 return (EINVAL); 3042 } 3043 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 3044 switch (addrs[0].ss_family) { 3045 #ifdef INET6 3046 case AF_INET6: 3047 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3048 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3049 INPLOOKUP_WLOCKPCB, NULL); 3050 break; 3051 #endif 3052 #ifdef INET 3053 case AF_INET: 3054 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3055 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3056 break; 3057 #endif 3058 } 3059 if (inp != NULL) { 3060 if (inp->inp_flags & INP_TIMEWAIT) { 3061 /* 3062 * XXXRW: There currently exists a state where an 3063 * inpcb is present, but its timewait state has been 3064 * discarded. For now, don't allow dropping of this 3065 * type of inpcb. 3066 */ 3067 tw = intotw(inp); 3068 if (tw != NULL) 3069 tcp_twclose(tw, 0); 3070 else 3071 INP_WUNLOCK(inp); 3072 } else if (!(inp->inp_flags & INP_DROPPED) && 3073 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3074 tp = intotcpcb(inp); 3075 tp = tcp_drop(tp, ECONNABORTED); 3076 if (tp != NULL) 3077 INP_WUNLOCK(inp); 3078 } else 3079 INP_WUNLOCK(inp); 3080 } else 3081 error = ESRCH; 3082 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 3083 return (error); 3084 } 3085 3086 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3087 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3088 0, sysctl_drop, "", "Drop TCP connection"); 3089 3090 #ifdef KERN_TLS 3091 static int 3092 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3093 { 3094 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3095 struct sockaddr_storage addrs[2]; 3096 struct inpcb *inp; 3097 struct sockaddr_in *fin, *lin; 3098 struct epoch_tracker et; 3099 #ifdef INET6 3100 struct sockaddr_in6 *fin6, *lin6; 3101 #endif 3102 int error; 3103 3104 inp = NULL; 3105 fin = lin = NULL; 3106 #ifdef INET6 3107 fin6 = lin6 = NULL; 3108 #endif 3109 error = 0; 3110 3111 if (req->oldptr != NULL || req->oldlen != 0) 3112 return (EINVAL); 3113 if (req->newptr == NULL) 3114 return (EPERM); 3115 if (req->newlen < sizeof(addrs)) 3116 return (ENOMEM); 3117 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3118 if (error) 3119 return (error); 3120 3121 switch (addrs[0].ss_family) { 3122 #ifdef INET6 3123 case AF_INET6: 3124 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3125 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3126 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3127 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3128 return (EINVAL); 3129 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3130 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3131 return (EINVAL); 3132 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3133 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3134 fin = (struct sockaddr_in *)&addrs[0]; 3135 lin = (struct sockaddr_in *)&addrs[1]; 3136 break; 3137 } 3138 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3139 if (error) 3140 return (error); 3141 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3142 if (error) 3143 return (error); 3144 break; 3145 #endif 3146 #ifdef INET 3147 case AF_INET: 3148 fin = (struct sockaddr_in *)&addrs[0]; 3149 lin = (struct sockaddr_in *)&addrs[1]; 3150 if (fin->sin_len != sizeof(struct sockaddr_in) || 3151 lin->sin_len != sizeof(struct sockaddr_in)) 3152 return (EINVAL); 3153 break; 3154 #endif 3155 default: 3156 return (EINVAL); 3157 } 3158 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 3159 switch (addrs[0].ss_family) { 3160 #ifdef INET6 3161 case AF_INET6: 3162 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3163 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3164 INPLOOKUP_WLOCKPCB, NULL); 3165 break; 3166 #endif 3167 #ifdef INET 3168 case AF_INET: 3169 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3170 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3171 break; 3172 #endif 3173 } 3174 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 3175 if (inp != NULL) { 3176 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3177 inp->inp_socket == NULL) { 3178 error = ECONNRESET; 3179 INP_WUNLOCK(inp); 3180 } else { 3181 struct socket *so; 3182 3183 so = inp->inp_socket; 3184 soref(so); 3185 error = ktls_set_tx_mode(so, 3186 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3187 INP_WUNLOCK(inp); 3188 SOCK_LOCK(so); 3189 sorele(so); 3190 } 3191 } else 3192 error = ESRCH; 3193 return (error); 3194 } 3195 3196 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3197 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3198 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS"); 3199 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3200 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3201 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS"); 3202 #endif 3203 3204 /* 3205 * Generate a standardized TCP log line for use throughout the 3206 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3207 * allow use in the interrupt context. 3208 * 3209 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3210 * NB: The function may return NULL if memory allocation failed. 3211 * 3212 * Due to header inclusion and ordering limitations the struct ip 3213 * and ip6_hdr pointers have to be passed as void pointers. 3214 */ 3215 char * 3216 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3217 const void *ip6hdr) 3218 { 3219 3220 /* Is logging enabled? */ 3221 if (tcp_log_in_vain == 0) 3222 return (NULL); 3223 3224 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3225 } 3226 3227 char * 3228 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3229 const void *ip6hdr) 3230 { 3231 3232 /* Is logging enabled? */ 3233 if (tcp_log_debug == 0) 3234 return (NULL); 3235 3236 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3237 } 3238 3239 static char * 3240 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3241 const void *ip6hdr) 3242 { 3243 char *s, *sp; 3244 size_t size; 3245 struct ip *ip; 3246 #ifdef INET6 3247 const struct ip6_hdr *ip6; 3248 3249 ip6 = (const struct ip6_hdr *)ip6hdr; 3250 #endif /* INET6 */ 3251 ip = (struct ip *)ip4hdr; 3252 3253 /* 3254 * The log line looks like this: 3255 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3256 */ 3257 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3258 sizeof(PRINT_TH_FLAGS) + 1 + 3259 #ifdef INET6 3260 2 * INET6_ADDRSTRLEN; 3261 #else 3262 2 * INET_ADDRSTRLEN; 3263 #endif /* INET6 */ 3264 3265 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3266 if (s == NULL) 3267 return (NULL); 3268 3269 strcat(s, "TCP: ["); 3270 sp = s + strlen(s); 3271 3272 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3273 inet_ntoa_r(inc->inc_faddr, sp); 3274 sp = s + strlen(s); 3275 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3276 sp = s + strlen(s); 3277 inet_ntoa_r(inc->inc_laddr, sp); 3278 sp = s + strlen(s); 3279 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3280 #ifdef INET6 3281 } else if (inc) { 3282 ip6_sprintf(sp, &inc->inc6_faddr); 3283 sp = s + strlen(s); 3284 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3285 sp = s + strlen(s); 3286 ip6_sprintf(sp, &inc->inc6_laddr); 3287 sp = s + strlen(s); 3288 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3289 } else if (ip6 && th) { 3290 ip6_sprintf(sp, &ip6->ip6_src); 3291 sp = s + strlen(s); 3292 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3293 sp = s + strlen(s); 3294 ip6_sprintf(sp, &ip6->ip6_dst); 3295 sp = s + strlen(s); 3296 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3297 #endif /* INET6 */ 3298 #ifdef INET 3299 } else if (ip && th) { 3300 inet_ntoa_r(ip->ip_src, sp); 3301 sp = s + strlen(s); 3302 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3303 sp = s + strlen(s); 3304 inet_ntoa_r(ip->ip_dst, sp); 3305 sp = s + strlen(s); 3306 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3307 #endif /* INET */ 3308 } else { 3309 free(s, M_TCPLOG); 3310 return (NULL); 3311 } 3312 sp = s + strlen(s); 3313 if (th) 3314 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3315 if (*(s + size - 1) != '\0') 3316 panic("%s: string too long", __func__); 3317 return (s); 3318 } 3319 3320 /* 3321 * A subroutine which makes it easy to track TCP state changes with DTrace. 3322 * This function shouldn't be called for t_state initializations that don't 3323 * correspond to actual TCP state transitions. 3324 */ 3325 void 3326 tcp_state_change(struct tcpcb *tp, int newstate) 3327 { 3328 #if defined(KDTRACE_HOOKS) 3329 int pstate = tp->t_state; 3330 #endif 3331 3332 TCPSTATES_DEC(tp->t_state); 3333 TCPSTATES_INC(newstate); 3334 tp->t_state = newstate; 3335 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3336 } 3337 3338 /* 3339 * Create an external-format (``xtcpcb'') structure using the information in 3340 * the kernel-format tcpcb structure pointed to by tp. This is done to 3341 * reduce the spew of irrelevant information over this interface, to isolate 3342 * user code from changes in the kernel structure, and potentially to provide 3343 * information-hiding if we decide that some of this information should be 3344 * hidden from users. 3345 */ 3346 void 3347 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3348 { 3349 struct tcpcb *tp = intotcpcb(inp); 3350 sbintime_t now; 3351 3352 bzero(xt, sizeof(*xt)); 3353 if (inp->inp_flags & INP_TIMEWAIT) { 3354 xt->t_state = TCPS_TIME_WAIT; 3355 } else { 3356 xt->t_state = tp->t_state; 3357 xt->t_logstate = tp->t_logstate; 3358 xt->t_flags = tp->t_flags; 3359 xt->t_sndzerowin = tp->t_sndzerowin; 3360 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3361 xt->t_rcvoopack = tp->t_rcvoopack; 3362 3363 now = getsbinuptime(); 3364 #define COPYTIMER(ttt) do { \ 3365 if (callout_active(&tp->t_timers->ttt)) \ 3366 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3367 SBT_1MS; \ 3368 else \ 3369 xt->ttt = 0; \ 3370 } while (0) 3371 COPYTIMER(tt_delack); 3372 COPYTIMER(tt_rexmt); 3373 COPYTIMER(tt_persist); 3374 COPYTIMER(tt_keep); 3375 COPYTIMER(tt_2msl); 3376 #undef COPYTIMER 3377 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3378 3379 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3380 TCP_FUNCTION_NAME_LEN_MAX); 3381 #ifdef TCP_BLACKBOX 3382 (void)tcp_log_get_id(tp, xt->xt_logid); 3383 #endif 3384 } 3385 3386 xt->xt_len = sizeof(struct xtcpcb); 3387 in_pcbtoxinpcb(inp, &xt->xt_inp); 3388 if (inp->inp_socket == NULL) 3389 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3390 } 3391