1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_compat.h" 38 #include "opt_inet.h" 39 #include "opt_inet6.h" 40 #include "opt_ipsec.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/callout.h> 46 #include <sys/eventhandler.h> 47 #ifdef TCP_HHOOK 48 #include <sys/hhook.h> 49 #endif 50 #include <sys/kernel.h> 51 #ifdef TCP_HHOOK 52 #include <sys/khelp.h> 53 #endif 54 #include <sys/sysctl.h> 55 #include <sys/jail.h> 56 #include <sys/malloc.h> 57 #include <sys/refcount.h> 58 #include <sys/mbuf.h> 59 #ifdef INET6 60 #include <sys/domain.h> 61 #endif 62 #include <sys/priv.h> 63 #include <sys/proc.h> 64 #include <sys/sdt.h> 65 #include <sys/socket.h> 66 #include <sys/socketvar.h> 67 #include <sys/protosw.h> 68 #include <sys/random.h> 69 70 #include <vm/uma.h> 71 72 #include <net/route.h> 73 #include <net/if.h> 74 #include <net/if_var.h> 75 #include <net/vnet.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_fib.h> 79 #include <netinet/in_kdtrace.h> 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_systm.h> 82 #include <netinet/in_var.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_icmp.h> 85 #include <netinet/ip_var.h> 86 #ifdef INET6 87 #include <netinet/icmp6.h> 88 #include <netinet/ip6.h> 89 #include <netinet6/in6_fib.h> 90 #include <netinet6/in6_pcb.h> 91 #include <netinet6/ip6_var.h> 92 #include <netinet6/scope6_var.h> 93 #include <netinet6/nd6.h> 94 #endif 95 96 #include <netinet/tcp.h> 97 #include <netinet/tcp_fsm.h> 98 #include <netinet/tcp_seq.h> 99 #include <netinet/tcp_timer.h> 100 #include <netinet/tcp_var.h> 101 #include <netinet/tcp_log_buf.h> 102 #include <netinet/tcp_syncache.h> 103 #include <netinet/cc/cc.h> 104 #ifdef INET6 105 #include <netinet6/tcp6_var.h> 106 #endif 107 #include <netinet/tcpip.h> 108 #include <netinet/tcp_fastopen.h> 109 #ifdef TCPPCAP 110 #include <netinet/tcp_pcap.h> 111 #endif 112 #ifdef TCPDEBUG 113 #include <netinet/tcp_debug.h> 114 #endif 115 #ifdef INET6 116 #include <netinet6/ip6protosw.h> 117 #endif 118 #ifdef TCP_OFFLOAD 119 #include <netinet/tcp_offload.h> 120 #endif 121 122 #include <netipsec/ipsec_support.h> 123 124 #include <machine/in_cksum.h> 125 #include <sys/md5.h> 126 127 #include <security/mac/mac_framework.h> 128 129 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 130 #ifdef INET6 131 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 132 #endif 133 134 struct rwlock tcp_function_lock; 135 136 static int 137 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 138 { 139 int error, new; 140 141 new = V_tcp_mssdflt; 142 error = sysctl_handle_int(oidp, &new, 0, req); 143 if (error == 0 && req->newptr) { 144 if (new < TCP_MINMSS) 145 error = EINVAL; 146 else 147 V_tcp_mssdflt = new; 148 } 149 return (error); 150 } 151 152 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 153 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 154 &sysctl_net_inet_tcp_mss_check, "I", 155 "Default TCP Maximum Segment Size"); 156 157 #ifdef INET6 158 static int 159 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 160 { 161 int error, new; 162 163 new = V_tcp_v6mssdflt; 164 error = sysctl_handle_int(oidp, &new, 0, req); 165 if (error == 0 && req->newptr) { 166 if (new < TCP_MINMSS) 167 error = EINVAL; 168 else 169 V_tcp_v6mssdflt = new; 170 } 171 return (error); 172 } 173 174 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 175 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 176 &sysctl_net_inet_tcp_mss_v6_check, "I", 177 "Default TCP Maximum Segment Size for IPv6"); 178 #endif /* INET6 */ 179 180 /* 181 * Minimum MSS we accept and use. This prevents DoS attacks where 182 * we are forced to a ridiculous low MSS like 20 and send hundreds 183 * of packets instead of one. The effect scales with the available 184 * bandwidth and quickly saturates the CPU and network interface 185 * with packet generation and sending. Set to zero to disable MINMSS 186 * checking. This setting prevents us from sending too small packets. 187 */ 188 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 190 &VNET_NAME(tcp_minmss), 0, 191 "Minimum TCP Maximum Segment Size"); 192 193 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 194 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 195 &VNET_NAME(tcp_do_rfc1323), 0, 196 "Enable rfc1323 (high performance TCP) extensions"); 197 198 static int tcp_log_debug = 0; 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 200 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 201 202 static int tcp_tcbhashsize; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 204 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 205 206 static int do_tcpdrain = 1; 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 208 "Enable tcp_drain routine for extra help when low on mbufs"); 209 210 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 211 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 212 213 static VNET_DEFINE(int, icmp_may_rst) = 1; 214 #define V_icmp_may_rst VNET(icmp_may_rst) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(icmp_may_rst), 0, 217 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 218 219 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 220 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_isn_reseed_interval), 0, 223 "Seconds between reseeding of ISN secret"); 224 225 static int tcp_soreceive_stream; 226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 227 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 228 229 VNET_DEFINE(uma_zone_t, sack_hole_zone); 230 #define V_sack_hole_zone VNET(sack_hole_zone) 231 232 #ifdef TCP_HHOOK 233 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 234 #endif 235 236 static struct inpcb *tcp_notify(struct inpcb *, int); 237 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 238 static void tcp_mtudisc(struct inpcb *, int); 239 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 240 void *ip4hdr, const void *ip6hdr); 241 242 243 static struct tcp_function_block tcp_def_funcblk = { 244 "default", 245 tcp_output, 246 tcp_do_segment, 247 tcp_default_ctloutput, 248 NULL, 249 NULL, 250 NULL, 251 NULL, 252 NULL, 253 NULL, 254 0, 255 0 256 }; 257 258 int t_functions_inited = 0; 259 struct tcp_funchead t_functions; 260 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 261 262 static void 263 init_tcp_functions(void) 264 { 265 if (t_functions_inited == 0) { 266 TAILQ_INIT(&t_functions); 267 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 268 t_functions_inited = 1; 269 } 270 } 271 272 static struct tcp_function_block * 273 find_tcp_functions_locked(struct tcp_function_set *fs) 274 { 275 struct tcp_function *f; 276 struct tcp_function_block *blk=NULL; 277 278 TAILQ_FOREACH(f, &t_functions, tf_next) { 279 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 280 blk = f->tf_fb; 281 break; 282 } 283 } 284 return(blk); 285 } 286 287 static struct tcp_function_block * 288 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 289 { 290 struct tcp_function_block *rblk=NULL; 291 struct tcp_function *f; 292 293 TAILQ_FOREACH(f, &t_functions, tf_next) { 294 if (f->tf_fb == blk) { 295 rblk = blk; 296 if (s) { 297 *s = f; 298 } 299 break; 300 } 301 } 302 return (rblk); 303 } 304 305 struct tcp_function_block * 306 find_and_ref_tcp_functions(struct tcp_function_set *fs) 307 { 308 struct tcp_function_block *blk; 309 310 rw_rlock(&tcp_function_lock); 311 blk = find_tcp_functions_locked(fs); 312 if (blk) 313 refcount_acquire(&blk->tfb_refcnt); 314 rw_runlock(&tcp_function_lock); 315 return(blk); 316 } 317 318 struct tcp_function_block * 319 find_and_ref_tcp_fb(struct tcp_function_block *blk) 320 { 321 struct tcp_function_block *rblk; 322 323 rw_rlock(&tcp_function_lock); 324 rblk = find_tcp_fb_locked(blk, NULL); 325 if (rblk) 326 refcount_acquire(&rblk->tfb_refcnt); 327 rw_runlock(&tcp_function_lock); 328 return(rblk); 329 } 330 331 332 static int 333 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 334 { 335 int error=ENOENT; 336 struct tcp_function_set fs; 337 struct tcp_function_block *blk; 338 339 memset(&fs, 0, sizeof(fs)); 340 rw_rlock(&tcp_function_lock); 341 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 342 if (blk) { 343 /* Found him */ 344 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 345 fs.pcbcnt = blk->tfb_refcnt; 346 } 347 rw_runlock(&tcp_function_lock); 348 error = sysctl_handle_string(oidp, fs.function_set_name, 349 sizeof(fs.function_set_name), req); 350 351 /* Check for error or no change */ 352 if (error != 0 || req->newptr == NULL) 353 return(error); 354 355 rw_wlock(&tcp_function_lock); 356 blk = find_tcp_functions_locked(&fs); 357 if ((blk == NULL) || 358 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 359 error = ENOENT; 360 goto done; 361 } 362 tcp_func_set_ptr = blk; 363 done: 364 rw_wunlock(&tcp_function_lock); 365 return (error); 366 } 367 368 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 369 CTLTYPE_STRING | CTLFLAG_RW, 370 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 371 "Set/get the default TCP functions"); 372 373 static int 374 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 375 { 376 int error, cnt, linesz; 377 struct tcp_function *f; 378 char *buffer, *cp; 379 size_t bufsz, outsz; 380 bool alias; 381 382 cnt = 0; 383 rw_rlock(&tcp_function_lock); 384 TAILQ_FOREACH(f, &t_functions, tf_next) { 385 cnt++; 386 } 387 rw_runlock(&tcp_function_lock); 388 389 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 390 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 391 392 error = 0; 393 cp = buffer; 394 395 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 396 "Alias", "PCB count"); 397 cp += linesz; 398 bufsz -= linesz; 399 outsz = linesz; 400 401 rw_rlock(&tcp_function_lock); 402 TAILQ_FOREACH(f, &t_functions, tf_next) { 403 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 404 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 405 f->tf_fb->tfb_tcp_block_name, 406 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 407 alias ? f->tf_name : "-", 408 f->tf_fb->tfb_refcnt); 409 if (linesz >= bufsz) { 410 error = EOVERFLOW; 411 break; 412 } 413 cp += linesz; 414 bufsz -= linesz; 415 outsz += linesz; 416 } 417 rw_runlock(&tcp_function_lock); 418 if (error == 0) 419 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 420 free(buffer, M_TEMP); 421 return (error); 422 } 423 424 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 425 CTLTYPE_STRING|CTLFLAG_RD, 426 NULL, 0, sysctl_net_inet_list_available, "A", 427 "list available TCP Function sets"); 428 429 /* 430 * Exports one (struct tcp_function_id) for each non-alias. 431 */ 432 static int 433 sysctl_net_inet_list_func_ids(SYSCTL_HANDLER_ARGS) 434 { 435 int error, cnt; 436 struct tcp_function *f; 437 struct tcp_function_id tfi; 438 439 /* 440 * We don't allow writes. 441 */ 442 if (req->newptr != NULL) 443 return (EINVAL); 444 445 /* 446 * Wire the old buffer so we can directly copy the functions to 447 * user space without dropping the lock. 448 */ 449 if (req->oldptr != NULL) { 450 error = sysctl_wire_old_buffer(req, 0); 451 if (error) 452 return (error); 453 } 454 455 /* 456 * Walk the list, comparing the name of the function entry and 457 * function block to determine which is an alias. 458 * If exporting the list, copy out matching entries. Otherwise, 459 * just record the total length. 460 */ 461 cnt = 0; 462 rw_rlock(&tcp_function_lock); 463 TAILQ_FOREACH(f, &t_functions, tf_next) { 464 if (strncmp(f->tf_name, f->tf_fb->tfb_tcp_block_name, 465 TCP_FUNCTION_NAME_LEN_MAX)) 466 continue; 467 if (req->oldptr != NULL) { 468 tfi.tfi_id = f->tf_fb->tfb_id; 469 (void)strncpy(tfi.tfi_name, f->tf_name, 470 TCP_FUNCTION_NAME_LEN_MAX); 471 tfi.tfi_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 472 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 473 /* 474 * Don't stop on error, as that is the 475 * mechanism we use to accumulate length 476 * information if the buffer was too short. 477 */ 478 } else 479 cnt++; 480 } 481 rw_runlock(&tcp_function_lock); 482 if (req->oldptr == NULL) 483 error = SYSCTL_OUT(req, NULL, 484 (cnt + 1) * sizeof(struct tcp_function_id)); 485 486 return (error); 487 } 488 489 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_ids, 490 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 491 NULL, 0, sysctl_net_inet_list_func_ids, "S,tcp_function_id", 492 "List TCP function block name-to-ID mappings"); 493 494 /* 495 * Target size of TCP PCB hash tables. Must be a power of two. 496 * 497 * Note that this can be overridden by the kernel environment 498 * variable net.inet.tcp.tcbhashsize 499 */ 500 #ifndef TCBHASHSIZE 501 #define TCBHASHSIZE 0 502 #endif 503 504 /* 505 * XXX 506 * Callouts should be moved into struct tcp directly. They are currently 507 * separate because the tcpcb structure is exported to userland for sysctl 508 * parsing purposes, which do not know about callouts. 509 */ 510 struct tcpcb_mem { 511 struct tcpcb tcb; 512 struct tcp_timer tt; 513 struct cc_var ccv; 514 #ifdef TCP_HHOOK 515 struct osd osd; 516 #endif 517 }; 518 519 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 520 #define V_tcpcb_zone VNET(tcpcb_zone) 521 522 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 523 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 524 525 static struct mtx isn_mtx; 526 527 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 528 #define ISN_LOCK() mtx_lock(&isn_mtx) 529 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 530 531 /* 532 * TCP initialization. 533 */ 534 static void 535 tcp_zone_change(void *tag) 536 { 537 538 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 539 uma_zone_set_max(V_tcpcb_zone, maxsockets); 540 tcp_tw_zone_change(); 541 } 542 543 static int 544 tcp_inpcb_init(void *mem, int size, int flags) 545 { 546 struct inpcb *inp = mem; 547 548 INP_LOCK_INIT(inp, "inp", "tcpinp"); 549 return (0); 550 } 551 552 /* 553 * Take a value and get the next power of 2 that doesn't overflow. 554 * Used to size the tcp_inpcb hash buckets. 555 */ 556 static int 557 maketcp_hashsize(int size) 558 { 559 int hashsize; 560 561 /* 562 * auto tune. 563 * get the next power of 2 higher than maxsockets. 564 */ 565 hashsize = 1 << fls(size); 566 /* catch overflow, and just go one power of 2 smaller */ 567 if (hashsize < size) { 568 hashsize = 1 << (fls(size) - 1); 569 } 570 return (hashsize); 571 } 572 573 static volatile int next_tcp_stack_id = 1; 574 575 /* 576 * Register a TCP function block with the name provided in the names 577 * array. (Note that this function does NOT automatically register 578 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 579 * explicitly include blk->tfb_tcp_block_name in the list of names if 580 * you wish to register the stack with that name.) 581 * 582 * Either all name registrations will succeed or all will fail. If 583 * a name registration fails, the function will update the num_names 584 * argument to point to the array index of the name that encountered 585 * the failure. 586 * 587 * Returns 0 on success, or an error code on failure. 588 */ 589 int 590 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 591 const char *names[], int *num_names) 592 { 593 struct tcp_function *n; 594 struct tcp_function_set fs; 595 int error, i; 596 597 KASSERT(names != NULL && *num_names > 0, 598 ("%s: Called with 0-length name list", __func__)); 599 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 600 601 if (t_functions_inited == 0) { 602 init_tcp_functions(); 603 } 604 if ((blk->tfb_tcp_output == NULL) || 605 (blk->tfb_tcp_do_segment == NULL) || 606 (blk->tfb_tcp_ctloutput == NULL) || 607 (strlen(blk->tfb_tcp_block_name) == 0)) { 608 /* 609 * These functions are required and you 610 * need a name. 611 */ 612 *num_names = 0; 613 return (EINVAL); 614 } 615 if (blk->tfb_tcp_timer_stop_all || 616 blk->tfb_tcp_timer_activate || 617 blk->tfb_tcp_timer_active || 618 blk->tfb_tcp_timer_stop) { 619 /* 620 * If you define one timer function you 621 * must have them all. 622 */ 623 if ((blk->tfb_tcp_timer_stop_all == NULL) || 624 (blk->tfb_tcp_timer_activate == NULL) || 625 (blk->tfb_tcp_timer_active == NULL) || 626 (blk->tfb_tcp_timer_stop == NULL)) { 627 *num_names = 0; 628 return (EINVAL); 629 } 630 } 631 632 refcount_init(&blk->tfb_refcnt, 0); 633 blk->tfb_flags = 0; 634 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 635 for (i = 0; i < *num_names; i++) { 636 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 637 if (n == NULL) { 638 error = ENOMEM; 639 goto cleanup; 640 } 641 n->tf_fb = blk; 642 643 (void)strncpy(fs.function_set_name, names[i], 644 TCP_FUNCTION_NAME_LEN_MAX); 645 fs.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 646 rw_wlock(&tcp_function_lock); 647 if (find_tcp_functions_locked(&fs) != NULL) { 648 /* Duplicate name space not allowed */ 649 rw_wunlock(&tcp_function_lock); 650 free(n, M_TCPFUNCTIONS); 651 error = EALREADY; 652 goto cleanup; 653 } 654 (void)strncpy(n->tf_name, names[i], TCP_FUNCTION_NAME_LEN_MAX); 655 n->tf_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 656 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 657 rw_wunlock(&tcp_function_lock); 658 } 659 return(0); 660 661 cleanup: 662 /* 663 * Deregister the names we just added. Because registration failed 664 * for names[i], we don't need to deregister that name. 665 */ 666 *num_names = i; 667 rw_wlock(&tcp_function_lock); 668 while (--i >= 0) { 669 TAILQ_FOREACH(n, &t_functions, tf_next) { 670 if (!strncmp(n->tf_name, names[i], 671 TCP_FUNCTION_NAME_LEN_MAX)) { 672 TAILQ_REMOVE(&t_functions, n, tf_next); 673 n->tf_fb = NULL; 674 free(n, M_TCPFUNCTIONS); 675 break; 676 } 677 } 678 } 679 rw_wunlock(&tcp_function_lock); 680 return (error); 681 } 682 683 /* 684 * Register a TCP function block using the name provided in the name 685 * argument. 686 * 687 * Returns 0 on success, or an error code on failure. 688 */ 689 int 690 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 691 int wait) 692 { 693 const char *name_list[1]; 694 int num_names, rv; 695 696 num_names = 1; 697 if (name != NULL) 698 name_list[0] = name; 699 else 700 name_list[0] = blk->tfb_tcp_block_name; 701 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 702 return (rv); 703 } 704 705 /* 706 * Register a TCP function block using the name defined in 707 * blk->tfb_tcp_block_name. 708 * 709 * Returns 0 on success, or an error code on failure. 710 */ 711 int 712 register_tcp_functions(struct tcp_function_block *blk, int wait) 713 { 714 715 return (register_tcp_functions_as_name(blk, NULL, wait)); 716 } 717 718 int 719 deregister_tcp_functions(struct tcp_function_block *blk) 720 { 721 struct tcp_function *f; 722 int error=ENOENT; 723 724 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 725 /* You can't un-register the default */ 726 return (EPERM); 727 } 728 rw_wlock(&tcp_function_lock); 729 if (blk == tcp_func_set_ptr) { 730 /* You can't free the current default */ 731 rw_wunlock(&tcp_function_lock); 732 return (EBUSY); 733 } 734 if (blk->tfb_refcnt) { 735 /* Still tcb attached, mark it. */ 736 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 737 rw_wunlock(&tcp_function_lock); 738 return (EBUSY); 739 } 740 while (find_tcp_fb_locked(blk, &f) != NULL) { 741 /* Found */ 742 TAILQ_REMOVE(&t_functions, f, tf_next); 743 f->tf_fb = NULL; 744 free(f, M_TCPFUNCTIONS); 745 error = 0; 746 } 747 rw_wunlock(&tcp_function_lock); 748 return (error); 749 } 750 751 void 752 tcp_init(void) 753 { 754 const char *tcbhash_tuneable; 755 int hashsize; 756 757 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 758 759 #ifdef TCP_HHOOK 760 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 761 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 762 printf("%s: WARNING: unable to register helper hook\n", __func__); 763 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 764 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 765 printf("%s: WARNING: unable to register helper hook\n", __func__); 766 #endif 767 hashsize = TCBHASHSIZE; 768 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 769 if (hashsize == 0) { 770 /* 771 * Auto tune the hash size based on maxsockets. 772 * A perfect hash would have a 1:1 mapping 773 * (hashsize = maxsockets) however it's been 774 * suggested that O(2) average is better. 775 */ 776 hashsize = maketcp_hashsize(maxsockets / 4); 777 /* 778 * Our historical default is 512, 779 * do not autotune lower than this. 780 */ 781 if (hashsize < 512) 782 hashsize = 512; 783 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 784 printf("%s: %s auto tuned to %d\n", __func__, 785 tcbhash_tuneable, hashsize); 786 } 787 /* 788 * We require a hashsize to be a power of two. 789 * Previously if it was not a power of two we would just reset it 790 * back to 512, which could be a nasty surprise if you did not notice 791 * the error message. 792 * Instead what we do is clip it to the closest power of two lower 793 * than the specified hash value. 794 */ 795 if (!powerof2(hashsize)) { 796 int oldhashsize = hashsize; 797 798 hashsize = maketcp_hashsize(hashsize); 799 /* prevent absurdly low value */ 800 if (hashsize < 16) 801 hashsize = 16; 802 printf("%s: WARNING: TCB hash size not a power of 2, " 803 "clipped from %d to %d.\n", __func__, oldhashsize, 804 hashsize); 805 } 806 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 807 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 808 809 /* 810 * These have to be type stable for the benefit of the timers. 811 */ 812 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 813 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 814 uma_zone_set_max(V_tcpcb_zone, maxsockets); 815 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 816 817 tcp_tw_init(); 818 syncache_init(); 819 tcp_hc_init(); 820 821 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 822 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 823 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 824 825 tcp_fastopen_init(); 826 827 /* Skip initialization of globals for non-default instances. */ 828 if (!IS_DEFAULT_VNET(curvnet)) 829 return; 830 831 tcp_reass_global_init(); 832 833 /* XXX virtualize those bellow? */ 834 tcp_delacktime = TCPTV_DELACK; 835 tcp_keepinit = TCPTV_KEEP_INIT; 836 tcp_keepidle = TCPTV_KEEP_IDLE; 837 tcp_keepintvl = TCPTV_KEEPINTVL; 838 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 839 tcp_msl = TCPTV_MSL; 840 tcp_rexmit_min = TCPTV_MIN; 841 if (tcp_rexmit_min < 1) 842 tcp_rexmit_min = 1; 843 tcp_persmin = TCPTV_PERSMIN; 844 tcp_persmax = TCPTV_PERSMAX; 845 tcp_rexmit_slop = TCPTV_CPU_VAR; 846 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 847 tcp_tcbhashsize = hashsize; 848 /* Setup the tcp function block list */ 849 init_tcp_functions(); 850 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 851 /* Initialize the TCP logging data. */ 852 tcp_log_init(); 853 854 if (tcp_soreceive_stream) { 855 #ifdef INET 856 tcp_usrreqs.pru_soreceive = soreceive_stream; 857 #endif 858 #ifdef INET6 859 tcp6_usrreqs.pru_soreceive = soreceive_stream; 860 #endif /* INET6 */ 861 } 862 863 #ifdef INET6 864 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 865 #else /* INET6 */ 866 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 867 #endif /* INET6 */ 868 if (max_protohdr < TCP_MINPROTOHDR) 869 max_protohdr = TCP_MINPROTOHDR; 870 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 871 panic("tcp_init"); 872 #undef TCP_MINPROTOHDR 873 874 ISN_LOCK_INIT(); 875 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 876 SHUTDOWN_PRI_DEFAULT); 877 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 878 EVENTHANDLER_PRI_ANY); 879 #ifdef TCPPCAP 880 tcp_pcap_init(); 881 #endif 882 } 883 884 #ifdef VIMAGE 885 static void 886 tcp_destroy(void *unused __unused) 887 { 888 int n; 889 #ifdef TCP_HHOOK 890 int error; 891 #endif 892 893 /* 894 * All our processes are gone, all our sockets should be cleaned 895 * up, which means, we should be past the tcp_discardcb() calls. 896 * Sleep to let all tcpcb timers really disappear and cleanup. 897 */ 898 for (;;) { 899 INP_LIST_RLOCK(&V_tcbinfo); 900 n = V_tcbinfo.ipi_count; 901 INP_LIST_RUNLOCK(&V_tcbinfo); 902 if (n == 0) 903 break; 904 pause("tcpdes", hz / 10); 905 } 906 tcp_hc_destroy(); 907 syncache_destroy(); 908 tcp_tw_destroy(); 909 in_pcbinfo_destroy(&V_tcbinfo); 910 /* tcp_discardcb() clears the sack_holes up. */ 911 uma_zdestroy(V_sack_hole_zone); 912 uma_zdestroy(V_tcpcb_zone); 913 914 /* 915 * Cannot free the zone until all tcpcbs are released as we attach 916 * the allocations to them. 917 */ 918 tcp_fastopen_destroy(); 919 920 #ifdef TCP_HHOOK 921 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 922 if (error != 0) { 923 printf("%s: WARNING: unable to deregister helper hook " 924 "type=%d, id=%d: error %d returned\n", __func__, 925 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 926 } 927 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 928 if (error != 0) { 929 printf("%s: WARNING: unable to deregister helper hook " 930 "type=%d, id=%d: error %d returned\n", __func__, 931 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 932 } 933 #endif 934 } 935 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 936 #endif 937 938 void 939 tcp_fini(void *xtp) 940 { 941 942 } 943 944 /* 945 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 946 * tcp_template used to store this data in mbufs, but we now recopy it out 947 * of the tcpcb each time to conserve mbufs. 948 */ 949 void 950 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 951 { 952 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 953 954 INP_WLOCK_ASSERT(inp); 955 956 #ifdef INET6 957 if ((inp->inp_vflag & INP_IPV6) != 0) { 958 struct ip6_hdr *ip6; 959 960 ip6 = (struct ip6_hdr *)ip_ptr; 961 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 962 (inp->inp_flow & IPV6_FLOWINFO_MASK); 963 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 964 (IPV6_VERSION & IPV6_VERSION_MASK); 965 ip6->ip6_nxt = IPPROTO_TCP; 966 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 967 ip6->ip6_src = inp->in6p_laddr; 968 ip6->ip6_dst = inp->in6p_faddr; 969 } 970 #endif /* INET6 */ 971 #if defined(INET6) && defined(INET) 972 else 973 #endif 974 #ifdef INET 975 { 976 struct ip *ip; 977 978 ip = (struct ip *)ip_ptr; 979 ip->ip_v = IPVERSION; 980 ip->ip_hl = 5; 981 ip->ip_tos = inp->inp_ip_tos; 982 ip->ip_len = 0; 983 ip->ip_id = 0; 984 ip->ip_off = 0; 985 ip->ip_ttl = inp->inp_ip_ttl; 986 ip->ip_sum = 0; 987 ip->ip_p = IPPROTO_TCP; 988 ip->ip_src = inp->inp_laddr; 989 ip->ip_dst = inp->inp_faddr; 990 } 991 #endif /* INET */ 992 th->th_sport = inp->inp_lport; 993 th->th_dport = inp->inp_fport; 994 th->th_seq = 0; 995 th->th_ack = 0; 996 th->th_x2 = 0; 997 th->th_off = 5; 998 th->th_flags = 0; 999 th->th_win = 0; 1000 th->th_urp = 0; 1001 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1002 } 1003 1004 /* 1005 * Create template to be used to send tcp packets on a connection. 1006 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1007 * use for this function is in keepalives, which use tcp_respond. 1008 */ 1009 struct tcptemp * 1010 tcpip_maketemplate(struct inpcb *inp) 1011 { 1012 struct tcptemp *t; 1013 1014 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1015 if (t == NULL) 1016 return (NULL); 1017 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1018 return (t); 1019 } 1020 1021 /* 1022 * Send a single message to the TCP at address specified by 1023 * the given TCP/IP header. If m == NULL, then we make a copy 1024 * of the tcpiphdr at th and send directly to the addressed host. 1025 * This is used to force keep alive messages out using the TCP 1026 * template for a connection. If flags are given then we send 1027 * a message back to the TCP which originated the segment th, 1028 * and discard the mbuf containing it and any other attached mbufs. 1029 * 1030 * In any case the ack and sequence number of the transmitted 1031 * segment are as specified by the parameters. 1032 * 1033 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1034 */ 1035 void 1036 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1037 tcp_seq ack, tcp_seq seq, int flags) 1038 { 1039 struct tcpopt to; 1040 struct inpcb *inp; 1041 struct ip *ip; 1042 struct mbuf *optm; 1043 struct tcphdr *nth; 1044 u_char *optp; 1045 #ifdef INET6 1046 struct ip6_hdr *ip6; 1047 int isipv6; 1048 #endif /* INET6 */ 1049 int optlen, tlen, win; 1050 bool incl_opts; 1051 1052 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1053 1054 #ifdef INET6 1055 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1056 ip6 = ipgen; 1057 #endif /* INET6 */ 1058 ip = ipgen; 1059 1060 if (tp != NULL) { 1061 inp = tp->t_inpcb; 1062 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1063 INP_WLOCK_ASSERT(inp); 1064 } else 1065 inp = NULL; 1066 1067 incl_opts = false; 1068 win = 0; 1069 if (tp != NULL) { 1070 if (!(flags & TH_RST)) { 1071 win = sbspace(&inp->inp_socket->so_rcv); 1072 if (win > TCP_MAXWIN << tp->rcv_scale) 1073 win = TCP_MAXWIN << tp->rcv_scale; 1074 } 1075 if ((tp->t_flags & TF_NOOPT) == 0) 1076 incl_opts = true; 1077 } 1078 if (m == NULL) { 1079 m = m_gethdr(M_NOWAIT, MT_DATA); 1080 if (m == NULL) 1081 return; 1082 m->m_data += max_linkhdr; 1083 #ifdef INET6 1084 if (isipv6) { 1085 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1086 sizeof(struct ip6_hdr)); 1087 ip6 = mtod(m, struct ip6_hdr *); 1088 nth = (struct tcphdr *)(ip6 + 1); 1089 } else 1090 #endif /* INET6 */ 1091 { 1092 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1093 ip = mtod(m, struct ip *); 1094 nth = (struct tcphdr *)(ip + 1); 1095 } 1096 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1097 flags = TH_ACK; 1098 } else if (!M_WRITABLE(m)) { 1099 struct mbuf *n; 1100 1101 /* Can't reuse 'm', allocate a new mbuf. */ 1102 n = m_gethdr(M_NOWAIT, MT_DATA); 1103 if (n == NULL) { 1104 m_freem(m); 1105 return; 1106 } 1107 1108 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1109 m_freem(m); 1110 m_freem(n); 1111 return; 1112 } 1113 1114 n->m_data += max_linkhdr; 1115 /* m_len is set later */ 1116 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1117 #ifdef INET6 1118 if (isipv6) { 1119 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1120 sizeof(struct ip6_hdr)); 1121 ip6 = mtod(n, struct ip6_hdr *); 1122 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1123 nth = (struct tcphdr *)(ip6 + 1); 1124 } else 1125 #endif /* INET6 */ 1126 { 1127 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1128 ip = mtod(n, struct ip *); 1129 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1130 nth = (struct tcphdr *)(ip + 1); 1131 } 1132 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1133 xchg(nth->th_dport, nth->th_sport, uint16_t); 1134 th = nth; 1135 m_freem(m); 1136 m = n; 1137 } else { 1138 /* 1139 * reuse the mbuf. 1140 * XXX MRT We inherit the FIB, which is lucky. 1141 */ 1142 m_freem(m->m_next); 1143 m->m_next = NULL; 1144 m->m_data = (caddr_t)ipgen; 1145 /* m_len is set later */ 1146 #ifdef INET6 1147 if (isipv6) { 1148 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1149 nth = (struct tcphdr *)(ip6 + 1); 1150 } else 1151 #endif /* INET6 */ 1152 { 1153 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1154 nth = (struct tcphdr *)(ip + 1); 1155 } 1156 if (th != nth) { 1157 /* 1158 * this is usually a case when an extension header 1159 * exists between the IPv6 header and the 1160 * TCP header. 1161 */ 1162 nth->th_sport = th->th_sport; 1163 nth->th_dport = th->th_dport; 1164 } 1165 xchg(nth->th_dport, nth->th_sport, uint16_t); 1166 #undef xchg 1167 } 1168 tlen = 0; 1169 #ifdef INET6 1170 if (isipv6) 1171 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1172 #endif 1173 #if defined(INET) && defined(INET6) 1174 else 1175 #endif 1176 #ifdef INET 1177 tlen = sizeof (struct tcpiphdr); 1178 #endif 1179 #ifdef INVARIANTS 1180 m->m_len = 0; 1181 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1182 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1183 m, tlen, (long)M_TRAILINGSPACE(m))); 1184 #endif 1185 m->m_len = tlen; 1186 to.to_flags = 0; 1187 if (incl_opts) { 1188 /* Make sure we have room. */ 1189 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1190 m->m_next = m_get(M_NOWAIT, MT_DATA); 1191 if (m->m_next) { 1192 optp = mtod(m->m_next, u_char *); 1193 optm = m->m_next; 1194 } else 1195 incl_opts = false; 1196 } else { 1197 optp = (u_char *) (nth + 1); 1198 optm = m; 1199 } 1200 } 1201 if (incl_opts) { 1202 /* Timestamps. */ 1203 if (tp->t_flags & TF_RCVD_TSTMP) { 1204 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1205 to.to_tsecr = tp->ts_recent; 1206 to.to_flags |= TOF_TS; 1207 } 1208 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1209 /* TCP-MD5 (RFC2385). */ 1210 if (tp->t_flags & TF_SIGNATURE) 1211 to.to_flags |= TOF_SIGNATURE; 1212 #endif 1213 /* Add the options. */ 1214 tlen += optlen = tcp_addoptions(&to, optp); 1215 1216 /* Update m_len in the correct mbuf. */ 1217 optm->m_len += optlen; 1218 } else 1219 optlen = 0; 1220 #ifdef INET6 1221 if (isipv6) { 1222 ip6->ip6_flow = 0; 1223 ip6->ip6_vfc = IPV6_VERSION; 1224 ip6->ip6_nxt = IPPROTO_TCP; 1225 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1226 } 1227 #endif 1228 #if defined(INET) && defined(INET6) 1229 else 1230 #endif 1231 #ifdef INET 1232 { 1233 ip->ip_len = htons(tlen); 1234 ip->ip_ttl = V_ip_defttl; 1235 if (V_path_mtu_discovery) 1236 ip->ip_off |= htons(IP_DF); 1237 } 1238 #endif 1239 m->m_pkthdr.len = tlen; 1240 m->m_pkthdr.rcvif = NULL; 1241 #ifdef MAC 1242 if (inp != NULL) { 1243 /* 1244 * Packet is associated with a socket, so allow the 1245 * label of the response to reflect the socket label. 1246 */ 1247 INP_WLOCK_ASSERT(inp); 1248 mac_inpcb_create_mbuf(inp, m); 1249 } else { 1250 /* 1251 * Packet is not associated with a socket, so possibly 1252 * update the label in place. 1253 */ 1254 mac_netinet_tcp_reply(m); 1255 } 1256 #endif 1257 nth->th_seq = htonl(seq); 1258 nth->th_ack = htonl(ack); 1259 nth->th_x2 = 0; 1260 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1261 nth->th_flags = flags; 1262 if (tp != NULL) 1263 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1264 else 1265 nth->th_win = htons((u_short)win); 1266 nth->th_urp = 0; 1267 1268 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1269 if (to.to_flags & TOF_SIGNATURE) { 1270 if (!TCPMD5_ENABLED() || 1271 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1272 m_freem(m); 1273 return; 1274 } 1275 } 1276 #endif 1277 1278 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1279 #ifdef INET6 1280 if (isipv6) { 1281 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1282 nth->th_sum = in6_cksum_pseudo(ip6, 1283 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1284 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1285 NULL, NULL); 1286 } 1287 #endif /* INET6 */ 1288 #if defined(INET6) && defined(INET) 1289 else 1290 #endif 1291 #ifdef INET 1292 { 1293 m->m_pkthdr.csum_flags = CSUM_TCP; 1294 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1295 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1296 } 1297 #endif /* INET */ 1298 #ifdef TCPDEBUG 1299 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1300 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1301 #endif 1302 TCP_PROBE3(debug__output, tp, th, m); 1303 if (flags & TH_RST) 1304 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1305 1306 #ifdef INET6 1307 if (isipv6) { 1308 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1309 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1310 } 1311 #endif /* INET6 */ 1312 #if defined(INET) && defined(INET6) 1313 else 1314 #endif 1315 #ifdef INET 1316 { 1317 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1318 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1319 } 1320 #endif 1321 } 1322 1323 /* 1324 * Create a new TCP control block, making an 1325 * empty reassembly queue and hooking it to the argument 1326 * protocol control block. The `inp' parameter must have 1327 * come from the zone allocator set up in tcp_init(). 1328 */ 1329 struct tcpcb * 1330 tcp_newtcpcb(struct inpcb *inp) 1331 { 1332 struct tcpcb_mem *tm; 1333 struct tcpcb *tp; 1334 #ifdef INET6 1335 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1336 #endif /* INET6 */ 1337 1338 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1339 if (tm == NULL) 1340 return (NULL); 1341 tp = &tm->tcb; 1342 1343 /* Initialise cc_var struct for this tcpcb. */ 1344 tp->ccv = &tm->ccv; 1345 tp->ccv->type = IPPROTO_TCP; 1346 tp->ccv->ccvc.tcp = tp; 1347 rw_rlock(&tcp_function_lock); 1348 tp->t_fb = tcp_func_set_ptr; 1349 refcount_acquire(&tp->t_fb->tfb_refcnt); 1350 rw_runlock(&tcp_function_lock); 1351 /* 1352 * Use the current system default CC algorithm. 1353 */ 1354 CC_LIST_RLOCK(); 1355 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1356 CC_ALGO(tp) = CC_DEFAULT(); 1357 CC_LIST_RUNLOCK(); 1358 1359 if (CC_ALGO(tp)->cb_init != NULL) 1360 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1361 if (tp->t_fb->tfb_tcp_fb_fini) 1362 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1363 refcount_release(&tp->t_fb->tfb_refcnt); 1364 uma_zfree(V_tcpcb_zone, tm); 1365 return (NULL); 1366 } 1367 1368 #ifdef TCP_HHOOK 1369 tp->osd = &tm->osd; 1370 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1371 if (tp->t_fb->tfb_tcp_fb_fini) 1372 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1373 refcount_release(&tp->t_fb->tfb_refcnt); 1374 uma_zfree(V_tcpcb_zone, tm); 1375 return (NULL); 1376 } 1377 #endif 1378 1379 #ifdef VIMAGE 1380 tp->t_vnet = inp->inp_vnet; 1381 #endif 1382 tp->t_timers = &tm->tt; 1383 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1384 tp->t_maxseg = 1385 #ifdef INET6 1386 isipv6 ? V_tcp_v6mssdflt : 1387 #endif /* INET6 */ 1388 V_tcp_mssdflt; 1389 1390 /* Set up our timeouts. */ 1391 callout_init(&tp->t_timers->tt_rexmt, 1); 1392 callout_init(&tp->t_timers->tt_persist, 1); 1393 callout_init(&tp->t_timers->tt_keep, 1); 1394 callout_init(&tp->t_timers->tt_2msl, 1); 1395 callout_init(&tp->t_timers->tt_delack, 1); 1396 1397 if (V_tcp_do_rfc1323) 1398 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1399 if (V_tcp_do_sack) 1400 tp->t_flags |= TF_SACK_PERMIT; 1401 TAILQ_INIT(&tp->snd_holes); 1402 /* 1403 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1404 * is called. 1405 */ 1406 in_pcbref(inp); /* Reference for tcpcb */ 1407 tp->t_inpcb = inp; 1408 1409 /* 1410 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1411 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1412 * reasonable initial retransmit time. 1413 */ 1414 tp->t_srtt = TCPTV_SRTTBASE; 1415 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1416 tp->t_rttmin = tcp_rexmit_min; 1417 tp->t_rxtcur = TCPTV_RTOBASE; 1418 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1419 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1420 tp->t_rcvtime = ticks; 1421 /* 1422 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1423 * because the socket may be bound to an IPv6 wildcard address, 1424 * which may match an IPv4-mapped IPv6 address. 1425 */ 1426 inp->inp_ip_ttl = V_ip_defttl; 1427 inp->inp_ppcb = tp; 1428 #ifdef TCPPCAP 1429 /* 1430 * Init the TCP PCAP queues. 1431 */ 1432 tcp_pcap_tcpcb_init(tp); 1433 #endif 1434 /* Initialize the per-TCPCB log data. */ 1435 tcp_log_tcpcbinit(tp); 1436 if (tp->t_fb->tfb_tcp_fb_init) { 1437 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1438 } 1439 return (tp); /* XXX */ 1440 } 1441 1442 /* 1443 * Switch the congestion control algorithm back to NewReno for any active 1444 * control blocks using an algorithm which is about to go away. 1445 * This ensures the CC framework can allow the unload to proceed without leaving 1446 * any dangling pointers which would trigger a panic. 1447 * Returning non-zero would inform the CC framework that something went wrong 1448 * and it would be unsafe to allow the unload to proceed. However, there is no 1449 * way for this to occur with this implementation so we always return zero. 1450 */ 1451 int 1452 tcp_ccalgounload(struct cc_algo *unload_algo) 1453 { 1454 struct cc_algo *tmpalgo; 1455 struct inpcb *inp; 1456 struct tcpcb *tp; 1457 VNET_ITERATOR_DECL(vnet_iter); 1458 1459 /* 1460 * Check all active control blocks across all network stacks and change 1461 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1462 * requires cleanup code to be run, call it. 1463 */ 1464 VNET_LIST_RLOCK(); 1465 VNET_FOREACH(vnet_iter) { 1466 CURVNET_SET(vnet_iter); 1467 INP_INFO_WLOCK(&V_tcbinfo); 1468 /* 1469 * New connections already part way through being initialised 1470 * with the CC algo we're removing will not race with this code 1471 * because the INP_INFO_WLOCK is held during initialisation. We 1472 * therefore don't enter the loop below until the connection 1473 * list has stabilised. 1474 */ 1475 LIST_FOREACH(inp, &V_tcb, inp_list) { 1476 INP_WLOCK(inp); 1477 /* Important to skip tcptw structs. */ 1478 if (!(inp->inp_flags & INP_TIMEWAIT) && 1479 (tp = intotcpcb(inp)) != NULL) { 1480 /* 1481 * By holding INP_WLOCK here, we are assured 1482 * that the connection is not currently 1483 * executing inside the CC module's functions 1484 * i.e. it is safe to make the switch back to 1485 * NewReno. 1486 */ 1487 if (CC_ALGO(tp) == unload_algo) { 1488 tmpalgo = CC_ALGO(tp); 1489 /* NewReno does not require any init. */ 1490 CC_ALGO(tp) = &newreno_cc_algo; 1491 if (tmpalgo->cb_destroy != NULL) 1492 tmpalgo->cb_destroy(tp->ccv); 1493 } 1494 } 1495 INP_WUNLOCK(inp); 1496 } 1497 INP_INFO_WUNLOCK(&V_tcbinfo); 1498 CURVNET_RESTORE(); 1499 } 1500 VNET_LIST_RUNLOCK(); 1501 1502 return (0); 1503 } 1504 1505 /* 1506 * Drop a TCP connection, reporting 1507 * the specified error. If connection is synchronized, 1508 * then send a RST to peer. 1509 */ 1510 struct tcpcb * 1511 tcp_drop(struct tcpcb *tp, int errno) 1512 { 1513 struct socket *so = tp->t_inpcb->inp_socket; 1514 1515 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1516 INP_WLOCK_ASSERT(tp->t_inpcb); 1517 1518 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1519 tcp_state_change(tp, TCPS_CLOSED); 1520 (void) tp->t_fb->tfb_tcp_output(tp); 1521 TCPSTAT_INC(tcps_drops); 1522 } else 1523 TCPSTAT_INC(tcps_conndrops); 1524 if (errno == ETIMEDOUT && tp->t_softerror) 1525 errno = tp->t_softerror; 1526 so->so_error = errno; 1527 return (tcp_close(tp)); 1528 } 1529 1530 void 1531 tcp_discardcb(struct tcpcb *tp) 1532 { 1533 struct inpcb *inp = tp->t_inpcb; 1534 struct socket *so = inp->inp_socket; 1535 #ifdef INET6 1536 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1537 #endif /* INET6 */ 1538 int released; 1539 1540 INP_WLOCK_ASSERT(inp); 1541 1542 /* 1543 * Make sure that all of our timers are stopped before we delete the 1544 * PCB. 1545 * 1546 * If stopping a timer fails, we schedule a discard function in same 1547 * callout, and the last discard function called will take care of 1548 * deleting the tcpcb. 1549 */ 1550 tp->t_timers->tt_draincnt = 0; 1551 tcp_timer_stop(tp, TT_REXMT); 1552 tcp_timer_stop(tp, TT_PERSIST); 1553 tcp_timer_stop(tp, TT_KEEP); 1554 tcp_timer_stop(tp, TT_2MSL); 1555 tcp_timer_stop(tp, TT_DELACK); 1556 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1557 /* 1558 * Call the stop-all function of the methods, 1559 * this function should call the tcp_timer_stop() 1560 * method with each of the function specific timeouts. 1561 * That stop will be called via the tfb_tcp_timer_stop() 1562 * which should use the async drain function of the 1563 * callout system (see tcp_var.h). 1564 */ 1565 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1566 } 1567 1568 /* 1569 * If we got enough samples through the srtt filter, 1570 * save the rtt and rttvar in the routing entry. 1571 * 'Enough' is arbitrarily defined as 4 rtt samples. 1572 * 4 samples is enough for the srtt filter to converge 1573 * to within enough % of the correct value; fewer samples 1574 * and we could save a bogus rtt. The danger is not high 1575 * as tcp quickly recovers from everything. 1576 * XXX: Works very well but needs some more statistics! 1577 */ 1578 if (tp->t_rttupdated >= 4) { 1579 struct hc_metrics_lite metrics; 1580 uint32_t ssthresh; 1581 1582 bzero(&metrics, sizeof(metrics)); 1583 /* 1584 * Update the ssthresh always when the conditions below 1585 * are satisfied. This gives us better new start value 1586 * for the congestion avoidance for new connections. 1587 * ssthresh is only set if packet loss occurred on a session. 1588 * 1589 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1590 * being torn down. Ideally this code would not use 'so'. 1591 */ 1592 ssthresh = tp->snd_ssthresh; 1593 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1594 /* 1595 * convert the limit from user data bytes to 1596 * packets then to packet data bytes. 1597 */ 1598 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1599 if (ssthresh < 2) 1600 ssthresh = 2; 1601 ssthresh *= (tp->t_maxseg + 1602 #ifdef INET6 1603 (isipv6 ? sizeof (struct ip6_hdr) + 1604 sizeof (struct tcphdr) : 1605 #endif 1606 sizeof (struct tcpiphdr) 1607 #ifdef INET6 1608 ) 1609 #endif 1610 ); 1611 } else 1612 ssthresh = 0; 1613 metrics.rmx_ssthresh = ssthresh; 1614 1615 metrics.rmx_rtt = tp->t_srtt; 1616 metrics.rmx_rttvar = tp->t_rttvar; 1617 metrics.rmx_cwnd = tp->snd_cwnd; 1618 metrics.rmx_sendpipe = 0; 1619 metrics.rmx_recvpipe = 0; 1620 1621 tcp_hc_update(&inp->inp_inc, &metrics); 1622 } 1623 1624 /* free the reassembly queue, if any */ 1625 tcp_reass_flush(tp); 1626 1627 #ifdef TCP_OFFLOAD 1628 /* Disconnect offload device, if any. */ 1629 if (tp->t_flags & TF_TOE) 1630 tcp_offload_detach(tp); 1631 #endif 1632 1633 tcp_free_sackholes(tp); 1634 1635 #ifdef TCPPCAP 1636 /* Free the TCP PCAP queues. */ 1637 tcp_pcap_drain(&(tp->t_inpkts)); 1638 tcp_pcap_drain(&(tp->t_outpkts)); 1639 #endif 1640 1641 /* Allow the CC algorithm to clean up after itself. */ 1642 if (CC_ALGO(tp)->cb_destroy != NULL) 1643 CC_ALGO(tp)->cb_destroy(tp->ccv); 1644 1645 #ifdef TCP_HHOOK 1646 khelp_destroy_osd(tp->osd); 1647 #endif 1648 1649 CC_ALGO(tp) = NULL; 1650 inp->inp_ppcb = NULL; 1651 if (tp->t_timers->tt_draincnt == 0) { 1652 /* We own the last reference on tcpcb, let's free it. */ 1653 tcp_log_tcpcbfini(tp); 1654 TCPSTATES_DEC(tp->t_state); 1655 if (tp->t_fb->tfb_tcp_fb_fini) 1656 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1657 refcount_release(&tp->t_fb->tfb_refcnt); 1658 tp->t_inpcb = NULL; 1659 uma_zfree(V_tcpcb_zone, tp); 1660 released = in_pcbrele_wlocked(inp); 1661 KASSERT(!released, ("%s: inp %p should not have been released " 1662 "here", __func__, inp)); 1663 } 1664 } 1665 1666 void 1667 tcp_timer_discard(void *ptp) 1668 { 1669 struct inpcb *inp; 1670 struct tcpcb *tp; 1671 1672 tp = (struct tcpcb *)ptp; 1673 CURVNET_SET(tp->t_vnet); 1674 INP_INFO_RLOCK(&V_tcbinfo); 1675 inp = tp->t_inpcb; 1676 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1677 __func__, tp)); 1678 INP_WLOCK(inp); 1679 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1680 ("%s: tcpcb has to be stopped here", __func__)); 1681 tp->t_timers->tt_draincnt--; 1682 if (tp->t_timers->tt_draincnt == 0) { 1683 /* We own the last reference on this tcpcb, let's free it. */ 1684 tcp_log_tcpcbfini(tp); 1685 TCPSTATES_DEC(tp->t_state); 1686 if (tp->t_fb->tfb_tcp_fb_fini) 1687 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1688 refcount_release(&tp->t_fb->tfb_refcnt); 1689 tp->t_inpcb = NULL; 1690 uma_zfree(V_tcpcb_zone, tp); 1691 if (in_pcbrele_wlocked(inp)) { 1692 INP_INFO_RUNLOCK(&V_tcbinfo); 1693 CURVNET_RESTORE(); 1694 return; 1695 } 1696 } 1697 INP_WUNLOCK(inp); 1698 INP_INFO_RUNLOCK(&V_tcbinfo); 1699 CURVNET_RESTORE(); 1700 } 1701 1702 /* 1703 * Attempt to close a TCP control block, marking it as dropped, and freeing 1704 * the socket if we hold the only reference. 1705 */ 1706 struct tcpcb * 1707 tcp_close(struct tcpcb *tp) 1708 { 1709 struct inpcb *inp = tp->t_inpcb; 1710 struct socket *so; 1711 1712 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1713 INP_WLOCK_ASSERT(inp); 1714 1715 #ifdef TCP_OFFLOAD 1716 if (tp->t_state == TCPS_LISTEN) 1717 tcp_offload_listen_stop(tp); 1718 #endif 1719 /* 1720 * This releases the TFO pending counter resource for TFO listen 1721 * sockets as well as passively-created TFO sockets that transition 1722 * from SYN_RECEIVED to CLOSED. 1723 */ 1724 if (tp->t_tfo_pending) { 1725 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1726 tp->t_tfo_pending = NULL; 1727 } 1728 in_pcbdrop(inp); 1729 TCPSTAT_INC(tcps_closed); 1730 if (tp->t_state != TCPS_CLOSED) 1731 tcp_state_change(tp, TCPS_CLOSED); 1732 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1733 so = inp->inp_socket; 1734 soisdisconnected(so); 1735 if (inp->inp_flags & INP_SOCKREF) { 1736 KASSERT(so->so_state & SS_PROTOREF, 1737 ("tcp_close: !SS_PROTOREF")); 1738 inp->inp_flags &= ~INP_SOCKREF; 1739 INP_WUNLOCK(inp); 1740 SOCK_LOCK(so); 1741 so->so_state &= ~SS_PROTOREF; 1742 sofree(so); 1743 return (NULL); 1744 } 1745 return (tp); 1746 } 1747 1748 void 1749 tcp_drain(void) 1750 { 1751 VNET_ITERATOR_DECL(vnet_iter); 1752 1753 if (!do_tcpdrain) 1754 return; 1755 1756 VNET_LIST_RLOCK_NOSLEEP(); 1757 VNET_FOREACH(vnet_iter) { 1758 CURVNET_SET(vnet_iter); 1759 struct inpcb *inpb; 1760 struct tcpcb *tcpb; 1761 1762 /* 1763 * Walk the tcpbs, if existing, and flush the reassembly queue, 1764 * if there is one... 1765 * XXX: The "Net/3" implementation doesn't imply that the TCP 1766 * reassembly queue should be flushed, but in a situation 1767 * where we're really low on mbufs, this is potentially 1768 * useful. 1769 */ 1770 INP_INFO_WLOCK(&V_tcbinfo); 1771 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1772 if (inpb->inp_flags & INP_TIMEWAIT) 1773 continue; 1774 INP_WLOCK(inpb); 1775 if ((tcpb = intotcpcb(inpb)) != NULL) { 1776 tcp_reass_flush(tcpb); 1777 tcp_clean_sackreport(tcpb); 1778 tcp_log_drain(tcpb); 1779 #ifdef TCPPCAP 1780 if (tcp_pcap_aggressive_free) { 1781 /* Free the TCP PCAP queues. */ 1782 tcp_pcap_drain(&(tcpb->t_inpkts)); 1783 tcp_pcap_drain(&(tcpb->t_outpkts)); 1784 } 1785 #endif 1786 } 1787 INP_WUNLOCK(inpb); 1788 } 1789 INP_INFO_WUNLOCK(&V_tcbinfo); 1790 CURVNET_RESTORE(); 1791 } 1792 VNET_LIST_RUNLOCK_NOSLEEP(); 1793 } 1794 1795 /* 1796 * Notify a tcp user of an asynchronous error; 1797 * store error as soft error, but wake up user 1798 * (for now, won't do anything until can select for soft error). 1799 * 1800 * Do not wake up user since there currently is no mechanism for 1801 * reporting soft errors (yet - a kqueue filter may be added). 1802 */ 1803 static struct inpcb * 1804 tcp_notify(struct inpcb *inp, int error) 1805 { 1806 struct tcpcb *tp; 1807 1808 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1809 INP_WLOCK_ASSERT(inp); 1810 1811 if ((inp->inp_flags & INP_TIMEWAIT) || 1812 (inp->inp_flags & INP_DROPPED)) 1813 return (inp); 1814 1815 tp = intotcpcb(inp); 1816 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1817 1818 /* 1819 * Ignore some errors if we are hooked up. 1820 * If connection hasn't completed, has retransmitted several times, 1821 * and receives a second error, give up now. This is better 1822 * than waiting a long time to establish a connection that 1823 * can never complete. 1824 */ 1825 if (tp->t_state == TCPS_ESTABLISHED && 1826 (error == EHOSTUNREACH || error == ENETUNREACH || 1827 error == EHOSTDOWN)) { 1828 if (inp->inp_route.ro_rt) { 1829 RTFREE(inp->inp_route.ro_rt); 1830 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1831 } 1832 return (inp); 1833 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1834 tp->t_softerror) { 1835 tp = tcp_drop(tp, error); 1836 if (tp != NULL) 1837 return (inp); 1838 else 1839 return (NULL); 1840 } else { 1841 tp->t_softerror = error; 1842 return (inp); 1843 } 1844 #if 0 1845 wakeup( &so->so_timeo); 1846 sorwakeup(so); 1847 sowwakeup(so); 1848 #endif 1849 } 1850 1851 static int 1852 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1853 { 1854 int error, i, m, n, pcb_count; 1855 struct inpcb *inp, **inp_list; 1856 inp_gen_t gencnt; 1857 struct xinpgen xig; 1858 1859 /* 1860 * The process of preparing the TCB list is too time-consuming and 1861 * resource-intensive to repeat twice on every request. 1862 */ 1863 if (req->oldptr == NULL) { 1864 n = V_tcbinfo.ipi_count + 1865 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1866 n += imax(n / 8, 10); 1867 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1868 return (0); 1869 } 1870 1871 if (req->newptr != NULL) 1872 return (EPERM); 1873 1874 /* 1875 * OK, now we're committed to doing something. 1876 */ 1877 INP_LIST_RLOCK(&V_tcbinfo); 1878 gencnt = V_tcbinfo.ipi_gencnt; 1879 n = V_tcbinfo.ipi_count; 1880 INP_LIST_RUNLOCK(&V_tcbinfo); 1881 1882 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1883 1884 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1885 + (n + m) * sizeof(struct xtcpcb)); 1886 if (error != 0) 1887 return (error); 1888 1889 xig.xig_len = sizeof xig; 1890 xig.xig_count = n + m; 1891 xig.xig_gen = gencnt; 1892 xig.xig_sogen = so_gencnt; 1893 error = SYSCTL_OUT(req, &xig, sizeof xig); 1894 if (error) 1895 return (error); 1896 1897 error = syncache_pcblist(req, m, &pcb_count); 1898 if (error) 1899 return (error); 1900 1901 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1902 1903 INP_INFO_WLOCK(&V_tcbinfo); 1904 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1905 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1906 INP_WLOCK(inp); 1907 if (inp->inp_gencnt <= gencnt) { 1908 /* 1909 * XXX: This use of cr_cansee(), introduced with 1910 * TCP state changes, is not quite right, but for 1911 * now, better than nothing. 1912 */ 1913 if (inp->inp_flags & INP_TIMEWAIT) { 1914 if (intotw(inp) != NULL) 1915 error = cr_cansee(req->td->td_ucred, 1916 intotw(inp)->tw_cred); 1917 else 1918 error = EINVAL; /* Skip this inp. */ 1919 } else 1920 error = cr_canseeinpcb(req->td->td_ucred, inp); 1921 if (error == 0) { 1922 in_pcbref(inp); 1923 inp_list[i++] = inp; 1924 } 1925 } 1926 INP_WUNLOCK(inp); 1927 } 1928 INP_INFO_WUNLOCK(&V_tcbinfo); 1929 n = i; 1930 1931 error = 0; 1932 for (i = 0; i < n; i++) { 1933 inp = inp_list[i]; 1934 INP_RLOCK(inp); 1935 if (inp->inp_gencnt <= gencnt) { 1936 struct xtcpcb xt; 1937 1938 tcp_inptoxtp(inp, &xt); 1939 INP_RUNLOCK(inp); 1940 error = SYSCTL_OUT(req, &xt, sizeof xt); 1941 } else 1942 INP_RUNLOCK(inp); 1943 } 1944 INP_INFO_RLOCK(&V_tcbinfo); 1945 for (i = 0; i < n; i++) { 1946 inp = inp_list[i]; 1947 INP_RLOCK(inp); 1948 if (!in_pcbrele_rlocked(inp)) 1949 INP_RUNLOCK(inp); 1950 } 1951 INP_INFO_RUNLOCK(&V_tcbinfo); 1952 1953 if (!error) { 1954 /* 1955 * Give the user an updated idea of our state. 1956 * If the generation differs from what we told 1957 * her before, she knows that something happened 1958 * while we were processing this request, and it 1959 * might be necessary to retry. 1960 */ 1961 INP_LIST_RLOCK(&V_tcbinfo); 1962 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1963 xig.xig_sogen = so_gencnt; 1964 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1965 INP_LIST_RUNLOCK(&V_tcbinfo); 1966 error = SYSCTL_OUT(req, &xig, sizeof xig); 1967 } 1968 free(inp_list, M_TEMP); 1969 return (error); 1970 } 1971 1972 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1973 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1974 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1975 1976 #ifdef INET 1977 static int 1978 tcp_getcred(SYSCTL_HANDLER_ARGS) 1979 { 1980 struct xucred xuc; 1981 struct sockaddr_in addrs[2]; 1982 struct inpcb *inp; 1983 int error; 1984 1985 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1986 if (error) 1987 return (error); 1988 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1989 if (error) 1990 return (error); 1991 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1992 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1993 if (inp != NULL) { 1994 if (inp->inp_socket == NULL) 1995 error = ENOENT; 1996 if (error == 0) 1997 error = cr_canseeinpcb(req->td->td_ucred, inp); 1998 if (error == 0) 1999 cru2x(inp->inp_cred, &xuc); 2000 INP_RUNLOCK(inp); 2001 } else 2002 error = ENOENT; 2003 if (error == 0) 2004 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2005 return (error); 2006 } 2007 2008 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2009 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2010 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2011 #endif /* INET */ 2012 2013 #ifdef INET6 2014 static int 2015 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2016 { 2017 struct xucred xuc; 2018 struct sockaddr_in6 addrs[2]; 2019 struct inpcb *inp; 2020 int error; 2021 #ifdef INET 2022 int mapped = 0; 2023 #endif 2024 2025 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2026 if (error) 2027 return (error); 2028 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2029 if (error) 2030 return (error); 2031 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2032 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2033 return (error); 2034 } 2035 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2036 #ifdef INET 2037 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2038 mapped = 1; 2039 else 2040 #endif 2041 return (EINVAL); 2042 } 2043 2044 #ifdef INET 2045 if (mapped == 1) 2046 inp = in_pcblookup(&V_tcbinfo, 2047 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2048 addrs[1].sin6_port, 2049 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2050 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2051 else 2052 #endif 2053 inp = in6_pcblookup(&V_tcbinfo, 2054 &addrs[1].sin6_addr, addrs[1].sin6_port, 2055 &addrs[0].sin6_addr, addrs[0].sin6_port, 2056 INPLOOKUP_RLOCKPCB, NULL); 2057 if (inp != NULL) { 2058 if (inp->inp_socket == NULL) 2059 error = ENOENT; 2060 if (error == 0) 2061 error = cr_canseeinpcb(req->td->td_ucred, inp); 2062 if (error == 0) 2063 cru2x(inp->inp_cred, &xuc); 2064 INP_RUNLOCK(inp); 2065 } else 2066 error = ENOENT; 2067 if (error == 0) 2068 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2069 return (error); 2070 } 2071 2072 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2073 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2074 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2075 #endif /* INET6 */ 2076 2077 2078 #ifdef INET 2079 void 2080 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2081 { 2082 struct ip *ip = vip; 2083 struct tcphdr *th; 2084 struct in_addr faddr; 2085 struct inpcb *inp; 2086 struct tcpcb *tp; 2087 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2088 struct icmp *icp; 2089 struct in_conninfo inc; 2090 tcp_seq icmp_tcp_seq; 2091 int mtu; 2092 2093 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2094 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2095 return; 2096 2097 if (cmd == PRC_MSGSIZE) 2098 notify = tcp_mtudisc_notify; 2099 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2100 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2101 cmd == PRC_TIMXCEED_INTRANS) && ip) 2102 notify = tcp_drop_syn_sent; 2103 2104 /* 2105 * Hostdead is ugly because it goes linearly through all PCBs. 2106 * XXX: We never get this from ICMP, otherwise it makes an 2107 * excellent DoS attack on machines with many connections. 2108 */ 2109 else if (cmd == PRC_HOSTDEAD) 2110 ip = NULL; 2111 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2112 return; 2113 2114 if (ip == NULL) { 2115 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2116 return; 2117 } 2118 2119 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2120 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2121 INP_INFO_RLOCK(&V_tcbinfo); 2122 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2123 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2124 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2125 /* signal EHOSTDOWN, as it flushes the cached route */ 2126 inp = (*notify)(inp, EHOSTDOWN); 2127 goto out; 2128 } 2129 icmp_tcp_seq = th->th_seq; 2130 if (inp != NULL) { 2131 if (!(inp->inp_flags & INP_TIMEWAIT) && 2132 !(inp->inp_flags & INP_DROPPED) && 2133 !(inp->inp_socket == NULL)) { 2134 tp = intotcpcb(inp); 2135 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2136 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2137 if (cmd == PRC_MSGSIZE) { 2138 /* 2139 * MTU discovery: 2140 * If we got a needfrag set the MTU 2141 * in the route to the suggested new 2142 * value (if given) and then notify. 2143 */ 2144 mtu = ntohs(icp->icmp_nextmtu); 2145 /* 2146 * If no alternative MTU was 2147 * proposed, try the next smaller 2148 * one. 2149 */ 2150 if (!mtu) 2151 mtu = ip_next_mtu( 2152 ntohs(ip->ip_len), 1); 2153 if (mtu < V_tcp_minmss + 2154 sizeof(struct tcpiphdr)) 2155 mtu = V_tcp_minmss + 2156 sizeof(struct tcpiphdr); 2157 /* 2158 * Only process the offered MTU if it 2159 * is smaller than the current one. 2160 */ 2161 if (mtu < tp->t_maxseg + 2162 sizeof(struct tcpiphdr)) { 2163 bzero(&inc, sizeof(inc)); 2164 inc.inc_faddr = faddr; 2165 inc.inc_fibnum = 2166 inp->inp_inc.inc_fibnum; 2167 tcp_hc_updatemtu(&inc, mtu); 2168 tcp_mtudisc(inp, mtu); 2169 } 2170 } else 2171 inp = (*notify)(inp, 2172 inetctlerrmap[cmd]); 2173 } 2174 } 2175 } else { 2176 bzero(&inc, sizeof(inc)); 2177 inc.inc_fport = th->th_dport; 2178 inc.inc_lport = th->th_sport; 2179 inc.inc_faddr = faddr; 2180 inc.inc_laddr = ip->ip_src; 2181 syncache_unreach(&inc, icmp_tcp_seq); 2182 } 2183 out: 2184 if (inp != NULL) 2185 INP_WUNLOCK(inp); 2186 INP_INFO_RUNLOCK(&V_tcbinfo); 2187 } 2188 #endif /* INET */ 2189 2190 #ifdef INET6 2191 void 2192 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2193 { 2194 struct in6_addr *dst; 2195 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2196 struct ip6_hdr *ip6; 2197 struct mbuf *m; 2198 struct inpcb *inp; 2199 struct tcpcb *tp; 2200 struct icmp6_hdr *icmp6; 2201 struct ip6ctlparam *ip6cp = NULL; 2202 const struct sockaddr_in6 *sa6_src = NULL; 2203 struct in_conninfo inc; 2204 struct tcp_ports { 2205 uint16_t th_sport; 2206 uint16_t th_dport; 2207 } t_ports; 2208 tcp_seq icmp_tcp_seq; 2209 unsigned int mtu; 2210 unsigned int off; 2211 2212 if (sa->sa_family != AF_INET6 || 2213 sa->sa_len != sizeof(struct sockaddr_in6)) 2214 return; 2215 2216 /* if the parameter is from icmp6, decode it. */ 2217 if (d != NULL) { 2218 ip6cp = (struct ip6ctlparam *)d; 2219 icmp6 = ip6cp->ip6c_icmp6; 2220 m = ip6cp->ip6c_m; 2221 ip6 = ip6cp->ip6c_ip6; 2222 off = ip6cp->ip6c_off; 2223 sa6_src = ip6cp->ip6c_src; 2224 dst = ip6cp->ip6c_finaldst; 2225 } else { 2226 m = NULL; 2227 ip6 = NULL; 2228 off = 0; /* fool gcc */ 2229 sa6_src = &sa6_any; 2230 dst = NULL; 2231 } 2232 2233 if (cmd == PRC_MSGSIZE) 2234 notify = tcp_mtudisc_notify; 2235 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2236 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2237 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2238 notify = tcp_drop_syn_sent; 2239 2240 /* 2241 * Hostdead is ugly because it goes linearly through all PCBs. 2242 * XXX: We never get this from ICMP, otherwise it makes an 2243 * excellent DoS attack on machines with many connections. 2244 */ 2245 else if (cmd == PRC_HOSTDEAD) 2246 ip6 = NULL; 2247 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2248 return; 2249 2250 if (ip6 == NULL) { 2251 in6_pcbnotify(&V_tcbinfo, sa, 0, 2252 (const struct sockaddr *)sa6_src, 2253 0, cmd, NULL, notify); 2254 return; 2255 } 2256 2257 /* Check if we can safely get the ports from the tcp hdr */ 2258 if (m == NULL || 2259 (m->m_pkthdr.len < 2260 (int32_t) (off + sizeof(struct tcp_ports)))) { 2261 return; 2262 } 2263 bzero(&t_ports, sizeof(struct tcp_ports)); 2264 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2265 INP_INFO_RLOCK(&V_tcbinfo); 2266 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2267 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2268 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2269 /* signal EHOSTDOWN, as it flushes the cached route */ 2270 inp = (*notify)(inp, EHOSTDOWN); 2271 goto out; 2272 } 2273 off += sizeof(struct tcp_ports); 2274 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2275 goto out; 2276 } 2277 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2278 if (inp != NULL) { 2279 if (!(inp->inp_flags & INP_TIMEWAIT) && 2280 !(inp->inp_flags & INP_DROPPED) && 2281 !(inp->inp_socket == NULL)) { 2282 tp = intotcpcb(inp); 2283 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2284 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2285 if (cmd == PRC_MSGSIZE) { 2286 /* 2287 * MTU discovery: 2288 * If we got a needfrag set the MTU 2289 * in the route to the suggested new 2290 * value (if given) and then notify. 2291 */ 2292 mtu = ntohl(icmp6->icmp6_mtu); 2293 /* 2294 * If no alternative MTU was 2295 * proposed, or the proposed 2296 * MTU was too small, set to 2297 * the min. 2298 */ 2299 if (mtu < IPV6_MMTU) 2300 mtu = IPV6_MMTU - 8; 2301 bzero(&inc, sizeof(inc)); 2302 inc.inc_fibnum = M_GETFIB(m); 2303 inc.inc_flags |= INC_ISIPV6; 2304 inc.inc6_faddr = *dst; 2305 if (in6_setscope(&inc.inc6_faddr, 2306 m->m_pkthdr.rcvif, NULL)) 2307 goto out; 2308 /* 2309 * Only process the offered MTU if it 2310 * is smaller than the current one. 2311 */ 2312 if (mtu < tp->t_maxseg + 2313 sizeof (struct tcphdr) + 2314 sizeof (struct ip6_hdr)) { 2315 tcp_hc_updatemtu(&inc, mtu); 2316 tcp_mtudisc(inp, mtu); 2317 ICMP6STAT_INC(icp6s_pmtuchg); 2318 } 2319 } else 2320 inp = (*notify)(inp, 2321 inet6ctlerrmap[cmd]); 2322 } 2323 } 2324 } else { 2325 bzero(&inc, sizeof(inc)); 2326 inc.inc_fibnum = M_GETFIB(m); 2327 inc.inc_flags |= INC_ISIPV6; 2328 inc.inc_fport = t_ports.th_dport; 2329 inc.inc_lport = t_ports.th_sport; 2330 inc.inc6_faddr = *dst; 2331 inc.inc6_laddr = ip6->ip6_src; 2332 syncache_unreach(&inc, icmp_tcp_seq); 2333 } 2334 out: 2335 if (inp != NULL) 2336 INP_WUNLOCK(inp); 2337 INP_INFO_RUNLOCK(&V_tcbinfo); 2338 } 2339 #endif /* INET6 */ 2340 2341 2342 /* 2343 * Following is where TCP initial sequence number generation occurs. 2344 * 2345 * There are two places where we must use initial sequence numbers: 2346 * 1. In SYN-ACK packets. 2347 * 2. In SYN packets. 2348 * 2349 * All ISNs for SYN-ACK packets are generated by the syncache. See 2350 * tcp_syncache.c for details. 2351 * 2352 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2353 * depends on this property. In addition, these ISNs should be 2354 * unguessable so as to prevent connection hijacking. To satisfy 2355 * the requirements of this situation, the algorithm outlined in 2356 * RFC 1948 is used, with only small modifications. 2357 * 2358 * Implementation details: 2359 * 2360 * Time is based off the system timer, and is corrected so that it 2361 * increases by one megabyte per second. This allows for proper 2362 * recycling on high speed LANs while still leaving over an hour 2363 * before rollover. 2364 * 2365 * As reading the *exact* system time is too expensive to be done 2366 * whenever setting up a TCP connection, we increment the time 2367 * offset in two ways. First, a small random positive increment 2368 * is added to isn_offset for each connection that is set up. 2369 * Second, the function tcp_isn_tick fires once per clock tick 2370 * and increments isn_offset as necessary so that sequence numbers 2371 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2372 * random positive increments serve only to ensure that the same 2373 * exact sequence number is never sent out twice (as could otherwise 2374 * happen when a port is recycled in less than the system tick 2375 * interval.) 2376 * 2377 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2378 * between seeding of isn_secret. This is normally set to zero, 2379 * as reseeding should not be necessary. 2380 * 2381 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2382 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2383 * general, this means holding an exclusive (write) lock. 2384 */ 2385 2386 #define ISN_BYTES_PER_SECOND 1048576 2387 #define ISN_STATIC_INCREMENT 4096 2388 #define ISN_RANDOM_INCREMENT (4096 - 1) 2389 2390 static VNET_DEFINE(u_char, isn_secret[32]); 2391 static VNET_DEFINE(int, isn_last); 2392 static VNET_DEFINE(int, isn_last_reseed); 2393 static VNET_DEFINE(u_int32_t, isn_offset); 2394 static VNET_DEFINE(u_int32_t, isn_offset_old); 2395 2396 #define V_isn_secret VNET(isn_secret) 2397 #define V_isn_last VNET(isn_last) 2398 #define V_isn_last_reseed VNET(isn_last_reseed) 2399 #define V_isn_offset VNET(isn_offset) 2400 #define V_isn_offset_old VNET(isn_offset_old) 2401 2402 tcp_seq 2403 tcp_new_isn(struct tcpcb *tp) 2404 { 2405 MD5_CTX isn_ctx; 2406 u_int32_t md5_buffer[4]; 2407 tcp_seq new_isn; 2408 u_int32_t projected_offset; 2409 2410 INP_WLOCK_ASSERT(tp->t_inpcb); 2411 2412 ISN_LOCK(); 2413 /* Seed if this is the first use, reseed if requested. */ 2414 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2415 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2416 < (u_int)ticks))) { 2417 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2418 V_isn_last_reseed = ticks; 2419 } 2420 2421 /* Compute the md5 hash and return the ISN. */ 2422 MD5Init(&isn_ctx); 2423 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2424 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2425 #ifdef INET6 2426 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2427 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2428 sizeof(struct in6_addr)); 2429 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2430 sizeof(struct in6_addr)); 2431 } else 2432 #endif 2433 { 2434 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2435 sizeof(struct in_addr)); 2436 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2437 sizeof(struct in_addr)); 2438 } 2439 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2440 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2441 new_isn = (tcp_seq) md5_buffer[0]; 2442 V_isn_offset += ISN_STATIC_INCREMENT + 2443 (arc4random() & ISN_RANDOM_INCREMENT); 2444 if (ticks != V_isn_last) { 2445 projected_offset = V_isn_offset_old + 2446 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2447 if (SEQ_GT(projected_offset, V_isn_offset)) 2448 V_isn_offset = projected_offset; 2449 V_isn_offset_old = V_isn_offset; 2450 V_isn_last = ticks; 2451 } 2452 new_isn += V_isn_offset; 2453 ISN_UNLOCK(); 2454 return (new_isn); 2455 } 2456 2457 /* 2458 * When a specific ICMP unreachable message is received and the 2459 * connection state is SYN-SENT, drop the connection. This behavior 2460 * is controlled by the icmp_may_rst sysctl. 2461 */ 2462 struct inpcb * 2463 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2464 { 2465 struct tcpcb *tp; 2466 2467 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2468 INP_WLOCK_ASSERT(inp); 2469 2470 if ((inp->inp_flags & INP_TIMEWAIT) || 2471 (inp->inp_flags & INP_DROPPED)) 2472 return (inp); 2473 2474 tp = intotcpcb(inp); 2475 if (tp->t_state != TCPS_SYN_SENT) 2476 return (inp); 2477 2478 if (IS_FASTOPEN(tp->t_flags)) 2479 tcp_fastopen_disable_path(tp); 2480 2481 tp = tcp_drop(tp, errno); 2482 if (tp != NULL) 2483 return (inp); 2484 else 2485 return (NULL); 2486 } 2487 2488 /* 2489 * When `need fragmentation' ICMP is received, update our idea of the MSS 2490 * based on the new value. Also nudge TCP to send something, since we 2491 * know the packet we just sent was dropped. 2492 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2493 */ 2494 static struct inpcb * 2495 tcp_mtudisc_notify(struct inpcb *inp, int error) 2496 { 2497 2498 tcp_mtudisc(inp, -1); 2499 return (inp); 2500 } 2501 2502 static void 2503 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2504 { 2505 struct tcpcb *tp; 2506 struct socket *so; 2507 2508 INP_WLOCK_ASSERT(inp); 2509 if ((inp->inp_flags & INP_TIMEWAIT) || 2510 (inp->inp_flags & INP_DROPPED)) 2511 return; 2512 2513 tp = intotcpcb(inp); 2514 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2515 2516 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2517 2518 so = inp->inp_socket; 2519 SOCKBUF_LOCK(&so->so_snd); 2520 /* If the mss is larger than the socket buffer, decrease the mss. */ 2521 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2522 tp->t_maxseg = so->so_snd.sb_hiwat; 2523 SOCKBUF_UNLOCK(&so->so_snd); 2524 2525 TCPSTAT_INC(tcps_mturesent); 2526 tp->t_rtttime = 0; 2527 tp->snd_nxt = tp->snd_una; 2528 tcp_free_sackholes(tp); 2529 tp->snd_recover = tp->snd_max; 2530 if (tp->t_flags & TF_SACK_PERMIT) 2531 EXIT_FASTRECOVERY(tp->t_flags); 2532 tp->t_fb->tfb_tcp_output(tp); 2533 } 2534 2535 #ifdef INET 2536 /* 2537 * Look-up the routing entry to the peer of this inpcb. If no route 2538 * is found and it cannot be allocated, then return 0. This routine 2539 * is called by TCP routines that access the rmx structure and by 2540 * tcp_mss_update to get the peer/interface MTU. 2541 */ 2542 uint32_t 2543 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2544 { 2545 struct nhop4_extended nh4; 2546 struct ifnet *ifp; 2547 uint32_t maxmtu = 0; 2548 2549 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2550 2551 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2552 2553 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2554 NHR_REF, 0, &nh4) != 0) 2555 return (0); 2556 2557 ifp = nh4.nh_ifp; 2558 maxmtu = nh4.nh_mtu; 2559 2560 /* Report additional interface capabilities. */ 2561 if (cap != NULL) { 2562 if (ifp->if_capenable & IFCAP_TSO4 && 2563 ifp->if_hwassist & CSUM_TSO) { 2564 cap->ifcap |= CSUM_TSO; 2565 cap->tsomax = ifp->if_hw_tsomax; 2566 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2567 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2568 } 2569 } 2570 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2571 } 2572 return (maxmtu); 2573 } 2574 #endif /* INET */ 2575 2576 #ifdef INET6 2577 uint32_t 2578 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2579 { 2580 struct nhop6_extended nh6; 2581 struct in6_addr dst6; 2582 uint32_t scopeid; 2583 struct ifnet *ifp; 2584 uint32_t maxmtu = 0; 2585 2586 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2587 2588 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2589 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2590 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2591 0, &nh6) != 0) 2592 return (0); 2593 2594 ifp = nh6.nh_ifp; 2595 maxmtu = nh6.nh_mtu; 2596 2597 /* Report additional interface capabilities. */ 2598 if (cap != NULL) { 2599 if (ifp->if_capenable & IFCAP_TSO6 && 2600 ifp->if_hwassist & CSUM_TSO) { 2601 cap->ifcap |= CSUM_TSO; 2602 cap->tsomax = ifp->if_hw_tsomax; 2603 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2604 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2605 } 2606 } 2607 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2608 } 2609 2610 return (maxmtu); 2611 } 2612 #endif /* INET6 */ 2613 2614 /* 2615 * Calculate effective SMSS per RFC5681 definition for a given TCP 2616 * connection at its current state, taking into account SACK and etc. 2617 */ 2618 u_int 2619 tcp_maxseg(const struct tcpcb *tp) 2620 { 2621 u_int optlen; 2622 2623 if (tp->t_flags & TF_NOOPT) 2624 return (tp->t_maxseg); 2625 2626 /* 2627 * Here we have a simplified code from tcp_addoptions(), 2628 * without a proper loop, and having most of paddings hardcoded. 2629 * We might make mistakes with padding here in some edge cases, 2630 * but this is harmless, since result of tcp_maxseg() is used 2631 * only in cwnd and ssthresh estimations. 2632 */ 2633 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2634 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2635 if (tp->t_flags & TF_RCVD_TSTMP) 2636 optlen = TCPOLEN_TSTAMP_APPA; 2637 else 2638 optlen = 0; 2639 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2640 if (tp->t_flags & TF_SIGNATURE) 2641 optlen += PAD(TCPOLEN_SIGNATURE); 2642 #endif 2643 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2644 optlen += TCPOLEN_SACKHDR; 2645 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2646 optlen = PAD(optlen); 2647 } 2648 } else { 2649 if (tp->t_flags & TF_REQ_TSTMP) 2650 optlen = TCPOLEN_TSTAMP_APPA; 2651 else 2652 optlen = PAD(TCPOLEN_MAXSEG); 2653 if (tp->t_flags & TF_REQ_SCALE) 2654 optlen += PAD(TCPOLEN_WINDOW); 2655 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2656 if (tp->t_flags & TF_SIGNATURE) 2657 optlen += PAD(TCPOLEN_SIGNATURE); 2658 #endif 2659 if (tp->t_flags & TF_SACK_PERMIT) 2660 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2661 } 2662 #undef PAD 2663 optlen = min(optlen, TCP_MAXOLEN); 2664 return (tp->t_maxseg - optlen); 2665 } 2666 2667 static int 2668 sysctl_drop(SYSCTL_HANDLER_ARGS) 2669 { 2670 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2671 struct sockaddr_storage addrs[2]; 2672 struct inpcb *inp; 2673 struct tcpcb *tp; 2674 struct tcptw *tw; 2675 struct sockaddr_in *fin, *lin; 2676 #ifdef INET6 2677 struct sockaddr_in6 *fin6, *lin6; 2678 #endif 2679 int error; 2680 2681 inp = NULL; 2682 fin = lin = NULL; 2683 #ifdef INET6 2684 fin6 = lin6 = NULL; 2685 #endif 2686 error = 0; 2687 2688 if (req->oldptr != NULL || req->oldlen != 0) 2689 return (EINVAL); 2690 if (req->newptr == NULL) 2691 return (EPERM); 2692 if (req->newlen < sizeof(addrs)) 2693 return (ENOMEM); 2694 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2695 if (error) 2696 return (error); 2697 2698 switch (addrs[0].ss_family) { 2699 #ifdef INET6 2700 case AF_INET6: 2701 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2702 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2703 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2704 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2705 return (EINVAL); 2706 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2707 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2708 return (EINVAL); 2709 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2710 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2711 fin = (struct sockaddr_in *)&addrs[0]; 2712 lin = (struct sockaddr_in *)&addrs[1]; 2713 break; 2714 } 2715 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2716 if (error) 2717 return (error); 2718 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2719 if (error) 2720 return (error); 2721 break; 2722 #endif 2723 #ifdef INET 2724 case AF_INET: 2725 fin = (struct sockaddr_in *)&addrs[0]; 2726 lin = (struct sockaddr_in *)&addrs[1]; 2727 if (fin->sin_len != sizeof(struct sockaddr_in) || 2728 lin->sin_len != sizeof(struct sockaddr_in)) 2729 return (EINVAL); 2730 break; 2731 #endif 2732 default: 2733 return (EINVAL); 2734 } 2735 INP_INFO_RLOCK(&V_tcbinfo); 2736 switch (addrs[0].ss_family) { 2737 #ifdef INET6 2738 case AF_INET6: 2739 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2740 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2741 INPLOOKUP_WLOCKPCB, NULL); 2742 break; 2743 #endif 2744 #ifdef INET 2745 case AF_INET: 2746 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2747 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2748 break; 2749 #endif 2750 } 2751 if (inp != NULL) { 2752 if (inp->inp_flags & INP_TIMEWAIT) { 2753 /* 2754 * XXXRW: There currently exists a state where an 2755 * inpcb is present, but its timewait state has been 2756 * discarded. For now, don't allow dropping of this 2757 * type of inpcb. 2758 */ 2759 tw = intotw(inp); 2760 if (tw != NULL) 2761 tcp_twclose(tw, 0); 2762 else 2763 INP_WUNLOCK(inp); 2764 } else if (!(inp->inp_flags & INP_DROPPED) && 2765 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2766 tp = intotcpcb(inp); 2767 tp = tcp_drop(tp, ECONNABORTED); 2768 if (tp != NULL) 2769 INP_WUNLOCK(inp); 2770 } else 2771 INP_WUNLOCK(inp); 2772 } else 2773 error = ESRCH; 2774 INP_INFO_RUNLOCK(&V_tcbinfo); 2775 return (error); 2776 } 2777 2778 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2779 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2780 0, sysctl_drop, "", "Drop TCP connection"); 2781 2782 /* 2783 * Generate a standardized TCP log line for use throughout the 2784 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2785 * allow use in the interrupt context. 2786 * 2787 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2788 * NB: The function may return NULL if memory allocation failed. 2789 * 2790 * Due to header inclusion and ordering limitations the struct ip 2791 * and ip6_hdr pointers have to be passed as void pointers. 2792 */ 2793 char * 2794 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2795 const void *ip6hdr) 2796 { 2797 2798 /* Is logging enabled? */ 2799 if (tcp_log_in_vain == 0) 2800 return (NULL); 2801 2802 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2803 } 2804 2805 char * 2806 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2807 const void *ip6hdr) 2808 { 2809 2810 /* Is logging enabled? */ 2811 if (tcp_log_debug == 0) 2812 return (NULL); 2813 2814 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2815 } 2816 2817 static char * 2818 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2819 const void *ip6hdr) 2820 { 2821 char *s, *sp; 2822 size_t size; 2823 struct ip *ip; 2824 #ifdef INET6 2825 const struct ip6_hdr *ip6; 2826 2827 ip6 = (const struct ip6_hdr *)ip6hdr; 2828 #endif /* INET6 */ 2829 ip = (struct ip *)ip4hdr; 2830 2831 /* 2832 * The log line looks like this: 2833 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2834 */ 2835 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2836 sizeof(PRINT_TH_FLAGS) + 1 + 2837 #ifdef INET6 2838 2 * INET6_ADDRSTRLEN; 2839 #else 2840 2 * INET_ADDRSTRLEN; 2841 #endif /* INET6 */ 2842 2843 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2844 if (s == NULL) 2845 return (NULL); 2846 2847 strcat(s, "TCP: ["); 2848 sp = s + strlen(s); 2849 2850 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2851 inet_ntoa_r(inc->inc_faddr, sp); 2852 sp = s + strlen(s); 2853 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2854 sp = s + strlen(s); 2855 inet_ntoa_r(inc->inc_laddr, sp); 2856 sp = s + strlen(s); 2857 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2858 #ifdef INET6 2859 } else if (inc) { 2860 ip6_sprintf(sp, &inc->inc6_faddr); 2861 sp = s + strlen(s); 2862 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2863 sp = s + strlen(s); 2864 ip6_sprintf(sp, &inc->inc6_laddr); 2865 sp = s + strlen(s); 2866 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2867 } else if (ip6 && th) { 2868 ip6_sprintf(sp, &ip6->ip6_src); 2869 sp = s + strlen(s); 2870 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2871 sp = s + strlen(s); 2872 ip6_sprintf(sp, &ip6->ip6_dst); 2873 sp = s + strlen(s); 2874 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2875 #endif /* INET6 */ 2876 #ifdef INET 2877 } else if (ip && th) { 2878 inet_ntoa_r(ip->ip_src, sp); 2879 sp = s + strlen(s); 2880 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2881 sp = s + strlen(s); 2882 inet_ntoa_r(ip->ip_dst, sp); 2883 sp = s + strlen(s); 2884 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2885 #endif /* INET */ 2886 } else { 2887 free(s, M_TCPLOG); 2888 return (NULL); 2889 } 2890 sp = s + strlen(s); 2891 if (th) 2892 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2893 if (*(s + size - 1) != '\0') 2894 panic("%s: string too long", __func__); 2895 return (s); 2896 } 2897 2898 /* 2899 * A subroutine which makes it easy to track TCP state changes with DTrace. 2900 * This function shouldn't be called for t_state initializations that don't 2901 * correspond to actual TCP state transitions. 2902 */ 2903 void 2904 tcp_state_change(struct tcpcb *tp, int newstate) 2905 { 2906 #if defined(KDTRACE_HOOKS) 2907 int pstate = tp->t_state; 2908 #endif 2909 2910 TCPSTATES_DEC(tp->t_state); 2911 TCPSTATES_INC(newstate); 2912 tp->t_state = newstate; 2913 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2914 } 2915 2916 /* 2917 * Create an external-format (``xtcpcb'') structure using the information in 2918 * the kernel-format tcpcb structure pointed to by tp. This is done to 2919 * reduce the spew of irrelevant information over this interface, to isolate 2920 * user code from changes in the kernel structure, and potentially to provide 2921 * information-hiding if we decide that some of this information should be 2922 * hidden from users. 2923 */ 2924 void 2925 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 2926 { 2927 struct tcpcb *tp = intotcpcb(inp); 2928 sbintime_t now; 2929 2930 if (inp->inp_flags & INP_TIMEWAIT) { 2931 bzero(xt, sizeof(struct xtcpcb)); 2932 xt->t_state = TCPS_TIME_WAIT; 2933 } else { 2934 xt->t_state = tp->t_state; 2935 xt->t_logstate = tp->t_logstate; 2936 xt->t_flags = tp->t_flags; 2937 xt->t_sndzerowin = tp->t_sndzerowin; 2938 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 2939 xt->t_rcvoopack = tp->t_rcvoopack; 2940 2941 now = getsbinuptime(); 2942 #define COPYTIMER(ttt) do { \ 2943 if (callout_active(&tp->t_timers->ttt)) \ 2944 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 2945 SBT_1MS; \ 2946 else \ 2947 xt->ttt = 0; \ 2948 } while (0) 2949 COPYTIMER(tt_delack); 2950 COPYTIMER(tt_rexmt); 2951 COPYTIMER(tt_persist); 2952 COPYTIMER(tt_keep); 2953 COPYTIMER(tt_2msl); 2954 #undef COPYTIMER 2955 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 2956 2957 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 2958 TCP_FUNCTION_NAME_LEN_MAX); 2959 bzero(xt->xt_logid, TCP_LOG_ID_LEN); 2960 (void)tcp_log_get_id(tp, xt->xt_logid); 2961 } 2962 2963 xt->xt_len = sizeof(struct xtcpcb); 2964 in_pcbtoxinpcb(inp, &xt->xt_inp); 2965 if (inp->inp_socket == NULL) 2966 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 2967 } 2968