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 #ifdef TCP_BLACKBOX 852 /* Initialize the TCP logging data. */ 853 tcp_log_init(); 854 #endif 855 856 if (tcp_soreceive_stream) { 857 #ifdef INET 858 tcp_usrreqs.pru_soreceive = soreceive_stream; 859 #endif 860 #ifdef INET6 861 tcp6_usrreqs.pru_soreceive = soreceive_stream; 862 #endif /* INET6 */ 863 } 864 865 #ifdef INET6 866 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 867 #else /* INET6 */ 868 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 869 #endif /* INET6 */ 870 if (max_protohdr < TCP_MINPROTOHDR) 871 max_protohdr = TCP_MINPROTOHDR; 872 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 873 panic("tcp_init"); 874 #undef TCP_MINPROTOHDR 875 876 ISN_LOCK_INIT(); 877 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 878 SHUTDOWN_PRI_DEFAULT); 879 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 880 EVENTHANDLER_PRI_ANY); 881 #ifdef TCPPCAP 882 tcp_pcap_init(); 883 #endif 884 } 885 886 #ifdef VIMAGE 887 static void 888 tcp_destroy(void *unused __unused) 889 { 890 int n; 891 #ifdef TCP_HHOOK 892 int error; 893 #endif 894 895 /* 896 * All our processes are gone, all our sockets should be cleaned 897 * up, which means, we should be past the tcp_discardcb() calls. 898 * Sleep to let all tcpcb timers really disappear and cleanup. 899 */ 900 for (;;) { 901 INP_LIST_RLOCK(&V_tcbinfo); 902 n = V_tcbinfo.ipi_count; 903 INP_LIST_RUNLOCK(&V_tcbinfo); 904 if (n == 0) 905 break; 906 pause("tcpdes", hz / 10); 907 } 908 tcp_hc_destroy(); 909 syncache_destroy(); 910 tcp_tw_destroy(); 911 in_pcbinfo_destroy(&V_tcbinfo); 912 /* tcp_discardcb() clears the sack_holes up. */ 913 uma_zdestroy(V_sack_hole_zone); 914 uma_zdestroy(V_tcpcb_zone); 915 916 /* 917 * Cannot free the zone until all tcpcbs are released as we attach 918 * the allocations to them. 919 */ 920 tcp_fastopen_destroy(); 921 922 #ifdef TCP_HHOOK 923 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 924 if (error != 0) { 925 printf("%s: WARNING: unable to deregister helper hook " 926 "type=%d, id=%d: error %d returned\n", __func__, 927 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 928 } 929 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 930 if (error != 0) { 931 printf("%s: WARNING: unable to deregister helper hook " 932 "type=%d, id=%d: error %d returned\n", __func__, 933 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 934 } 935 #endif 936 } 937 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 938 #endif 939 940 void 941 tcp_fini(void *xtp) 942 { 943 944 } 945 946 /* 947 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 948 * tcp_template used to store this data in mbufs, but we now recopy it out 949 * of the tcpcb each time to conserve mbufs. 950 */ 951 void 952 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 953 { 954 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 955 956 INP_WLOCK_ASSERT(inp); 957 958 #ifdef INET6 959 if ((inp->inp_vflag & INP_IPV6) != 0) { 960 struct ip6_hdr *ip6; 961 962 ip6 = (struct ip6_hdr *)ip_ptr; 963 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 964 (inp->inp_flow & IPV6_FLOWINFO_MASK); 965 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 966 (IPV6_VERSION & IPV6_VERSION_MASK); 967 ip6->ip6_nxt = IPPROTO_TCP; 968 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 969 ip6->ip6_src = inp->in6p_laddr; 970 ip6->ip6_dst = inp->in6p_faddr; 971 } 972 #endif /* INET6 */ 973 #if defined(INET6) && defined(INET) 974 else 975 #endif 976 #ifdef INET 977 { 978 struct ip *ip; 979 980 ip = (struct ip *)ip_ptr; 981 ip->ip_v = IPVERSION; 982 ip->ip_hl = 5; 983 ip->ip_tos = inp->inp_ip_tos; 984 ip->ip_len = 0; 985 ip->ip_id = 0; 986 ip->ip_off = 0; 987 ip->ip_ttl = inp->inp_ip_ttl; 988 ip->ip_sum = 0; 989 ip->ip_p = IPPROTO_TCP; 990 ip->ip_src = inp->inp_laddr; 991 ip->ip_dst = inp->inp_faddr; 992 } 993 #endif /* INET */ 994 th->th_sport = inp->inp_lport; 995 th->th_dport = inp->inp_fport; 996 th->th_seq = 0; 997 th->th_ack = 0; 998 th->th_x2 = 0; 999 th->th_off = 5; 1000 th->th_flags = 0; 1001 th->th_win = 0; 1002 th->th_urp = 0; 1003 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1004 } 1005 1006 /* 1007 * Create template to be used to send tcp packets on a connection. 1008 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1009 * use for this function is in keepalives, which use tcp_respond. 1010 */ 1011 struct tcptemp * 1012 tcpip_maketemplate(struct inpcb *inp) 1013 { 1014 struct tcptemp *t; 1015 1016 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1017 if (t == NULL) 1018 return (NULL); 1019 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1020 return (t); 1021 } 1022 1023 /* 1024 * Send a single message to the TCP at address specified by 1025 * the given TCP/IP header. If m == NULL, then we make a copy 1026 * of the tcpiphdr at th and send directly to the addressed host. 1027 * This is used to force keep alive messages out using the TCP 1028 * template for a connection. If flags are given then we send 1029 * a message back to the TCP which originated the segment th, 1030 * and discard the mbuf containing it and any other attached mbufs. 1031 * 1032 * In any case the ack and sequence number of the transmitted 1033 * segment are as specified by the parameters. 1034 * 1035 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1036 */ 1037 void 1038 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1039 tcp_seq ack, tcp_seq seq, int flags) 1040 { 1041 struct tcpopt to; 1042 struct inpcb *inp; 1043 struct ip *ip; 1044 struct mbuf *optm; 1045 struct tcphdr *nth; 1046 u_char *optp; 1047 #ifdef INET6 1048 struct ip6_hdr *ip6; 1049 int isipv6; 1050 #endif /* INET6 */ 1051 int optlen, tlen, win; 1052 bool incl_opts; 1053 1054 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1055 1056 #ifdef INET6 1057 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1058 ip6 = ipgen; 1059 #endif /* INET6 */ 1060 ip = ipgen; 1061 1062 if (tp != NULL) { 1063 inp = tp->t_inpcb; 1064 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1065 INP_WLOCK_ASSERT(inp); 1066 } else 1067 inp = NULL; 1068 1069 incl_opts = false; 1070 win = 0; 1071 if (tp != NULL) { 1072 if (!(flags & TH_RST)) { 1073 win = sbspace(&inp->inp_socket->so_rcv); 1074 if (win > TCP_MAXWIN << tp->rcv_scale) 1075 win = TCP_MAXWIN << tp->rcv_scale; 1076 } 1077 if ((tp->t_flags & TF_NOOPT) == 0) 1078 incl_opts = true; 1079 } 1080 if (m == NULL) { 1081 m = m_gethdr(M_NOWAIT, MT_DATA); 1082 if (m == NULL) 1083 return; 1084 m->m_data += max_linkhdr; 1085 #ifdef INET6 1086 if (isipv6) { 1087 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1088 sizeof(struct ip6_hdr)); 1089 ip6 = mtod(m, struct ip6_hdr *); 1090 nth = (struct tcphdr *)(ip6 + 1); 1091 } else 1092 #endif /* INET6 */ 1093 { 1094 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1095 ip = mtod(m, struct ip *); 1096 nth = (struct tcphdr *)(ip + 1); 1097 } 1098 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1099 flags = TH_ACK; 1100 } else if (!M_WRITABLE(m)) { 1101 struct mbuf *n; 1102 1103 /* Can't reuse 'm', allocate a new mbuf. */ 1104 n = m_gethdr(M_NOWAIT, MT_DATA); 1105 if (n == NULL) { 1106 m_freem(m); 1107 return; 1108 } 1109 1110 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1111 m_freem(m); 1112 m_freem(n); 1113 return; 1114 } 1115 1116 n->m_data += max_linkhdr; 1117 /* m_len is set later */ 1118 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1119 #ifdef INET6 1120 if (isipv6) { 1121 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1122 sizeof(struct ip6_hdr)); 1123 ip6 = mtod(n, struct ip6_hdr *); 1124 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1125 nth = (struct tcphdr *)(ip6 + 1); 1126 } else 1127 #endif /* INET6 */ 1128 { 1129 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1130 ip = mtod(n, struct ip *); 1131 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1132 nth = (struct tcphdr *)(ip + 1); 1133 } 1134 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1135 xchg(nth->th_dport, nth->th_sport, uint16_t); 1136 th = nth; 1137 m_freem(m); 1138 m = n; 1139 } else { 1140 /* 1141 * reuse the mbuf. 1142 * XXX MRT We inherit the FIB, which is lucky. 1143 */ 1144 m_freem(m->m_next); 1145 m->m_next = NULL; 1146 m->m_data = (caddr_t)ipgen; 1147 /* m_len is set later */ 1148 #ifdef INET6 1149 if (isipv6) { 1150 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1151 nth = (struct tcphdr *)(ip6 + 1); 1152 } else 1153 #endif /* INET6 */ 1154 { 1155 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1156 nth = (struct tcphdr *)(ip + 1); 1157 } 1158 if (th != nth) { 1159 /* 1160 * this is usually a case when an extension header 1161 * exists between the IPv6 header and the 1162 * TCP header. 1163 */ 1164 nth->th_sport = th->th_sport; 1165 nth->th_dport = th->th_dport; 1166 } 1167 xchg(nth->th_dport, nth->th_sport, uint16_t); 1168 #undef xchg 1169 } 1170 tlen = 0; 1171 #ifdef INET6 1172 if (isipv6) 1173 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1174 #endif 1175 #if defined(INET) && defined(INET6) 1176 else 1177 #endif 1178 #ifdef INET 1179 tlen = sizeof (struct tcpiphdr); 1180 #endif 1181 #ifdef INVARIANTS 1182 m->m_len = 0; 1183 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1184 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1185 m, tlen, (long)M_TRAILINGSPACE(m))); 1186 #endif 1187 m->m_len = tlen; 1188 to.to_flags = 0; 1189 if (incl_opts) { 1190 /* Make sure we have room. */ 1191 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1192 m->m_next = m_get(M_NOWAIT, MT_DATA); 1193 if (m->m_next) { 1194 optp = mtod(m->m_next, u_char *); 1195 optm = m->m_next; 1196 } else 1197 incl_opts = false; 1198 } else { 1199 optp = (u_char *) (nth + 1); 1200 optm = m; 1201 } 1202 } 1203 if (incl_opts) { 1204 /* Timestamps. */ 1205 if (tp->t_flags & TF_RCVD_TSTMP) { 1206 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1207 to.to_tsecr = tp->ts_recent; 1208 to.to_flags |= TOF_TS; 1209 } 1210 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1211 /* TCP-MD5 (RFC2385). */ 1212 if (tp->t_flags & TF_SIGNATURE) 1213 to.to_flags |= TOF_SIGNATURE; 1214 #endif 1215 /* Add the options. */ 1216 tlen += optlen = tcp_addoptions(&to, optp); 1217 1218 /* Update m_len in the correct mbuf. */ 1219 optm->m_len += optlen; 1220 } else 1221 optlen = 0; 1222 #ifdef INET6 1223 if (isipv6) { 1224 ip6->ip6_flow = 0; 1225 ip6->ip6_vfc = IPV6_VERSION; 1226 ip6->ip6_nxt = IPPROTO_TCP; 1227 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1228 } 1229 #endif 1230 #if defined(INET) && defined(INET6) 1231 else 1232 #endif 1233 #ifdef INET 1234 { 1235 ip->ip_len = htons(tlen); 1236 ip->ip_ttl = V_ip_defttl; 1237 if (V_path_mtu_discovery) 1238 ip->ip_off |= htons(IP_DF); 1239 } 1240 #endif 1241 m->m_pkthdr.len = tlen; 1242 m->m_pkthdr.rcvif = NULL; 1243 #ifdef MAC 1244 if (inp != NULL) { 1245 /* 1246 * Packet is associated with a socket, so allow the 1247 * label of the response to reflect the socket label. 1248 */ 1249 INP_WLOCK_ASSERT(inp); 1250 mac_inpcb_create_mbuf(inp, m); 1251 } else { 1252 /* 1253 * Packet is not associated with a socket, so possibly 1254 * update the label in place. 1255 */ 1256 mac_netinet_tcp_reply(m); 1257 } 1258 #endif 1259 nth->th_seq = htonl(seq); 1260 nth->th_ack = htonl(ack); 1261 nth->th_x2 = 0; 1262 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1263 nth->th_flags = flags; 1264 if (tp != NULL) 1265 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1266 else 1267 nth->th_win = htons((u_short)win); 1268 nth->th_urp = 0; 1269 1270 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1271 if (to.to_flags & TOF_SIGNATURE) { 1272 if (!TCPMD5_ENABLED() || 1273 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1274 m_freem(m); 1275 return; 1276 } 1277 } 1278 #endif 1279 1280 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1281 #ifdef INET6 1282 if (isipv6) { 1283 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1284 nth->th_sum = in6_cksum_pseudo(ip6, 1285 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1286 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1287 NULL, NULL); 1288 } 1289 #endif /* INET6 */ 1290 #if defined(INET6) && defined(INET) 1291 else 1292 #endif 1293 #ifdef INET 1294 { 1295 m->m_pkthdr.csum_flags = CSUM_TCP; 1296 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1297 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1298 } 1299 #endif /* INET */ 1300 #ifdef TCPDEBUG 1301 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1302 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1303 #endif 1304 TCP_PROBE3(debug__output, tp, th, m); 1305 if (flags & TH_RST) 1306 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1307 1308 #ifdef INET6 1309 if (isipv6) { 1310 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1311 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1312 } 1313 #endif /* INET6 */ 1314 #if defined(INET) && defined(INET6) 1315 else 1316 #endif 1317 #ifdef INET 1318 { 1319 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1320 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1321 } 1322 #endif 1323 } 1324 1325 /* 1326 * Create a new TCP control block, making an 1327 * empty reassembly queue and hooking it to the argument 1328 * protocol control block. The `inp' parameter must have 1329 * come from the zone allocator set up in tcp_init(). 1330 */ 1331 struct tcpcb * 1332 tcp_newtcpcb(struct inpcb *inp) 1333 { 1334 struct tcpcb_mem *tm; 1335 struct tcpcb *tp; 1336 #ifdef INET6 1337 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1338 #endif /* INET6 */ 1339 1340 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1341 if (tm == NULL) 1342 return (NULL); 1343 tp = &tm->tcb; 1344 1345 /* Initialise cc_var struct for this tcpcb. */ 1346 tp->ccv = &tm->ccv; 1347 tp->ccv->type = IPPROTO_TCP; 1348 tp->ccv->ccvc.tcp = tp; 1349 rw_rlock(&tcp_function_lock); 1350 tp->t_fb = tcp_func_set_ptr; 1351 refcount_acquire(&tp->t_fb->tfb_refcnt); 1352 rw_runlock(&tcp_function_lock); 1353 /* 1354 * Use the current system default CC algorithm. 1355 */ 1356 CC_LIST_RLOCK(); 1357 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1358 CC_ALGO(tp) = CC_DEFAULT(); 1359 CC_LIST_RUNLOCK(); 1360 1361 if (CC_ALGO(tp)->cb_init != NULL) 1362 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1363 if (tp->t_fb->tfb_tcp_fb_fini) 1364 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1365 refcount_release(&tp->t_fb->tfb_refcnt); 1366 uma_zfree(V_tcpcb_zone, tm); 1367 return (NULL); 1368 } 1369 1370 #ifdef TCP_HHOOK 1371 tp->osd = &tm->osd; 1372 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1373 if (tp->t_fb->tfb_tcp_fb_fini) 1374 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1375 refcount_release(&tp->t_fb->tfb_refcnt); 1376 uma_zfree(V_tcpcb_zone, tm); 1377 return (NULL); 1378 } 1379 #endif 1380 1381 #ifdef VIMAGE 1382 tp->t_vnet = inp->inp_vnet; 1383 #endif 1384 tp->t_timers = &tm->tt; 1385 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1386 tp->t_maxseg = 1387 #ifdef INET6 1388 isipv6 ? V_tcp_v6mssdflt : 1389 #endif /* INET6 */ 1390 V_tcp_mssdflt; 1391 1392 /* Set up our timeouts. */ 1393 callout_init(&tp->t_timers->tt_rexmt, 1); 1394 callout_init(&tp->t_timers->tt_persist, 1); 1395 callout_init(&tp->t_timers->tt_keep, 1); 1396 callout_init(&tp->t_timers->tt_2msl, 1); 1397 callout_init(&tp->t_timers->tt_delack, 1); 1398 1399 if (V_tcp_do_rfc1323) 1400 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1401 if (V_tcp_do_sack) 1402 tp->t_flags |= TF_SACK_PERMIT; 1403 TAILQ_INIT(&tp->snd_holes); 1404 /* 1405 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1406 * is called. 1407 */ 1408 in_pcbref(inp); /* Reference for tcpcb */ 1409 tp->t_inpcb = inp; 1410 1411 /* 1412 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1413 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1414 * reasonable initial retransmit time. 1415 */ 1416 tp->t_srtt = TCPTV_SRTTBASE; 1417 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1418 tp->t_rttmin = tcp_rexmit_min; 1419 tp->t_rxtcur = TCPTV_RTOBASE; 1420 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1421 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1422 tp->t_rcvtime = ticks; 1423 /* 1424 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1425 * because the socket may be bound to an IPv6 wildcard address, 1426 * which may match an IPv4-mapped IPv6 address. 1427 */ 1428 inp->inp_ip_ttl = V_ip_defttl; 1429 inp->inp_ppcb = tp; 1430 #ifdef TCPPCAP 1431 /* 1432 * Init the TCP PCAP queues. 1433 */ 1434 tcp_pcap_tcpcb_init(tp); 1435 #endif 1436 #ifdef TCP_BLACKBOX 1437 /* Initialize the per-TCPCB log data. */ 1438 tcp_log_tcpcbinit(tp); 1439 #endif 1440 if (tp->t_fb->tfb_tcp_fb_init) { 1441 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1442 } 1443 return (tp); /* XXX */ 1444 } 1445 1446 /* 1447 * Switch the congestion control algorithm back to NewReno for any active 1448 * control blocks using an algorithm which is about to go away. 1449 * This ensures the CC framework can allow the unload to proceed without leaving 1450 * any dangling pointers which would trigger a panic. 1451 * Returning non-zero would inform the CC framework that something went wrong 1452 * and it would be unsafe to allow the unload to proceed. However, there is no 1453 * way for this to occur with this implementation so we always return zero. 1454 */ 1455 int 1456 tcp_ccalgounload(struct cc_algo *unload_algo) 1457 { 1458 struct cc_algo *tmpalgo; 1459 struct inpcb *inp; 1460 struct tcpcb *tp; 1461 VNET_ITERATOR_DECL(vnet_iter); 1462 1463 /* 1464 * Check all active control blocks across all network stacks and change 1465 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1466 * requires cleanup code to be run, call it. 1467 */ 1468 VNET_LIST_RLOCK(); 1469 VNET_FOREACH(vnet_iter) { 1470 CURVNET_SET(vnet_iter); 1471 INP_INFO_WLOCK(&V_tcbinfo); 1472 /* 1473 * New connections already part way through being initialised 1474 * with the CC algo we're removing will not race with this code 1475 * because the INP_INFO_WLOCK is held during initialisation. We 1476 * therefore don't enter the loop below until the connection 1477 * list has stabilised. 1478 */ 1479 LIST_FOREACH(inp, &V_tcb, inp_list) { 1480 INP_WLOCK(inp); 1481 /* Important to skip tcptw structs. */ 1482 if (!(inp->inp_flags & INP_TIMEWAIT) && 1483 (tp = intotcpcb(inp)) != NULL) { 1484 /* 1485 * By holding INP_WLOCK here, we are assured 1486 * that the connection is not currently 1487 * executing inside the CC module's functions 1488 * i.e. it is safe to make the switch back to 1489 * NewReno. 1490 */ 1491 if (CC_ALGO(tp) == unload_algo) { 1492 tmpalgo = CC_ALGO(tp); 1493 /* NewReno does not require any init. */ 1494 CC_ALGO(tp) = &newreno_cc_algo; 1495 if (tmpalgo->cb_destroy != NULL) 1496 tmpalgo->cb_destroy(tp->ccv); 1497 } 1498 } 1499 INP_WUNLOCK(inp); 1500 } 1501 INP_INFO_WUNLOCK(&V_tcbinfo); 1502 CURVNET_RESTORE(); 1503 } 1504 VNET_LIST_RUNLOCK(); 1505 1506 return (0); 1507 } 1508 1509 /* 1510 * Drop a TCP connection, reporting 1511 * the specified error. If connection is synchronized, 1512 * then send a RST to peer. 1513 */ 1514 struct tcpcb * 1515 tcp_drop(struct tcpcb *tp, int errno) 1516 { 1517 struct socket *so = tp->t_inpcb->inp_socket; 1518 1519 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1520 INP_WLOCK_ASSERT(tp->t_inpcb); 1521 1522 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1523 tcp_state_change(tp, TCPS_CLOSED); 1524 (void) tp->t_fb->tfb_tcp_output(tp); 1525 TCPSTAT_INC(tcps_drops); 1526 } else 1527 TCPSTAT_INC(tcps_conndrops); 1528 if (errno == ETIMEDOUT && tp->t_softerror) 1529 errno = tp->t_softerror; 1530 so->so_error = errno; 1531 return (tcp_close(tp)); 1532 } 1533 1534 void 1535 tcp_discardcb(struct tcpcb *tp) 1536 { 1537 struct inpcb *inp = tp->t_inpcb; 1538 struct socket *so = inp->inp_socket; 1539 #ifdef INET6 1540 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1541 #endif /* INET6 */ 1542 int released; 1543 1544 INP_WLOCK_ASSERT(inp); 1545 1546 /* 1547 * Make sure that all of our timers are stopped before we delete the 1548 * PCB. 1549 * 1550 * If stopping a timer fails, we schedule a discard function in same 1551 * callout, and the last discard function called will take care of 1552 * deleting the tcpcb. 1553 */ 1554 tp->t_timers->tt_draincnt = 0; 1555 tcp_timer_stop(tp, TT_REXMT); 1556 tcp_timer_stop(tp, TT_PERSIST); 1557 tcp_timer_stop(tp, TT_KEEP); 1558 tcp_timer_stop(tp, TT_2MSL); 1559 tcp_timer_stop(tp, TT_DELACK); 1560 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1561 /* 1562 * Call the stop-all function of the methods, 1563 * this function should call the tcp_timer_stop() 1564 * method with each of the function specific timeouts. 1565 * That stop will be called via the tfb_tcp_timer_stop() 1566 * which should use the async drain function of the 1567 * callout system (see tcp_var.h). 1568 */ 1569 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1570 } 1571 1572 /* 1573 * If we got enough samples through the srtt filter, 1574 * save the rtt and rttvar in the routing entry. 1575 * 'Enough' is arbitrarily defined as 4 rtt samples. 1576 * 4 samples is enough for the srtt filter to converge 1577 * to within enough % of the correct value; fewer samples 1578 * and we could save a bogus rtt. The danger is not high 1579 * as tcp quickly recovers from everything. 1580 * XXX: Works very well but needs some more statistics! 1581 */ 1582 if (tp->t_rttupdated >= 4) { 1583 struct hc_metrics_lite metrics; 1584 uint32_t ssthresh; 1585 1586 bzero(&metrics, sizeof(metrics)); 1587 /* 1588 * Update the ssthresh always when the conditions below 1589 * are satisfied. This gives us better new start value 1590 * for the congestion avoidance for new connections. 1591 * ssthresh is only set if packet loss occurred on a session. 1592 * 1593 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1594 * being torn down. Ideally this code would not use 'so'. 1595 */ 1596 ssthresh = tp->snd_ssthresh; 1597 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1598 /* 1599 * convert the limit from user data bytes to 1600 * packets then to packet data bytes. 1601 */ 1602 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1603 if (ssthresh < 2) 1604 ssthresh = 2; 1605 ssthresh *= (tp->t_maxseg + 1606 #ifdef INET6 1607 (isipv6 ? sizeof (struct ip6_hdr) + 1608 sizeof (struct tcphdr) : 1609 #endif 1610 sizeof (struct tcpiphdr) 1611 #ifdef INET6 1612 ) 1613 #endif 1614 ); 1615 } else 1616 ssthresh = 0; 1617 metrics.rmx_ssthresh = ssthresh; 1618 1619 metrics.rmx_rtt = tp->t_srtt; 1620 metrics.rmx_rttvar = tp->t_rttvar; 1621 metrics.rmx_cwnd = tp->snd_cwnd; 1622 metrics.rmx_sendpipe = 0; 1623 metrics.rmx_recvpipe = 0; 1624 1625 tcp_hc_update(&inp->inp_inc, &metrics); 1626 } 1627 1628 /* free the reassembly queue, if any */ 1629 tcp_reass_flush(tp); 1630 1631 #ifdef TCP_OFFLOAD 1632 /* Disconnect offload device, if any. */ 1633 if (tp->t_flags & TF_TOE) 1634 tcp_offload_detach(tp); 1635 #endif 1636 1637 tcp_free_sackholes(tp); 1638 1639 #ifdef TCPPCAP 1640 /* Free the TCP PCAP queues. */ 1641 tcp_pcap_drain(&(tp->t_inpkts)); 1642 tcp_pcap_drain(&(tp->t_outpkts)); 1643 #endif 1644 1645 /* Allow the CC algorithm to clean up after itself. */ 1646 if (CC_ALGO(tp)->cb_destroy != NULL) 1647 CC_ALGO(tp)->cb_destroy(tp->ccv); 1648 1649 #ifdef TCP_HHOOK 1650 khelp_destroy_osd(tp->osd); 1651 #endif 1652 1653 CC_ALGO(tp) = NULL; 1654 inp->inp_ppcb = NULL; 1655 if (tp->t_timers->tt_draincnt == 0) { 1656 /* We own the last reference on tcpcb, let's free it. */ 1657 #ifdef TCP_BLACKBOX 1658 tcp_log_tcpcbfini(tp); 1659 #endif 1660 TCPSTATES_DEC(tp->t_state); 1661 if (tp->t_fb->tfb_tcp_fb_fini) 1662 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1663 refcount_release(&tp->t_fb->tfb_refcnt); 1664 tp->t_inpcb = NULL; 1665 uma_zfree(V_tcpcb_zone, tp); 1666 released = in_pcbrele_wlocked(inp); 1667 KASSERT(!released, ("%s: inp %p should not have been released " 1668 "here", __func__, inp)); 1669 } 1670 } 1671 1672 void 1673 tcp_timer_discard(void *ptp) 1674 { 1675 struct inpcb *inp; 1676 struct tcpcb *tp; 1677 1678 tp = (struct tcpcb *)ptp; 1679 CURVNET_SET(tp->t_vnet); 1680 INP_INFO_RLOCK(&V_tcbinfo); 1681 inp = tp->t_inpcb; 1682 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1683 __func__, tp)); 1684 INP_WLOCK(inp); 1685 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1686 ("%s: tcpcb has to be stopped here", __func__)); 1687 tp->t_timers->tt_draincnt--; 1688 if (tp->t_timers->tt_draincnt == 0) { 1689 /* We own the last reference on this tcpcb, let's free it. */ 1690 #ifdef TCP_BLACKBOX 1691 tcp_log_tcpcbfini(tp); 1692 #endif 1693 TCPSTATES_DEC(tp->t_state); 1694 if (tp->t_fb->tfb_tcp_fb_fini) 1695 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1696 refcount_release(&tp->t_fb->tfb_refcnt); 1697 tp->t_inpcb = NULL; 1698 uma_zfree(V_tcpcb_zone, tp); 1699 if (in_pcbrele_wlocked(inp)) { 1700 INP_INFO_RUNLOCK(&V_tcbinfo); 1701 CURVNET_RESTORE(); 1702 return; 1703 } 1704 } 1705 INP_WUNLOCK(inp); 1706 INP_INFO_RUNLOCK(&V_tcbinfo); 1707 CURVNET_RESTORE(); 1708 } 1709 1710 /* 1711 * Attempt to close a TCP control block, marking it as dropped, and freeing 1712 * the socket if we hold the only reference. 1713 */ 1714 struct tcpcb * 1715 tcp_close(struct tcpcb *tp) 1716 { 1717 struct inpcb *inp = tp->t_inpcb; 1718 struct socket *so; 1719 1720 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1721 INP_WLOCK_ASSERT(inp); 1722 1723 #ifdef TCP_OFFLOAD 1724 if (tp->t_state == TCPS_LISTEN) 1725 tcp_offload_listen_stop(tp); 1726 #endif 1727 /* 1728 * This releases the TFO pending counter resource for TFO listen 1729 * sockets as well as passively-created TFO sockets that transition 1730 * from SYN_RECEIVED to CLOSED. 1731 */ 1732 if (tp->t_tfo_pending) { 1733 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1734 tp->t_tfo_pending = NULL; 1735 } 1736 in_pcbdrop(inp); 1737 TCPSTAT_INC(tcps_closed); 1738 if (tp->t_state != TCPS_CLOSED) 1739 tcp_state_change(tp, TCPS_CLOSED); 1740 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1741 so = inp->inp_socket; 1742 soisdisconnected(so); 1743 if (inp->inp_flags & INP_SOCKREF) { 1744 KASSERT(so->so_state & SS_PROTOREF, 1745 ("tcp_close: !SS_PROTOREF")); 1746 inp->inp_flags &= ~INP_SOCKREF; 1747 INP_WUNLOCK(inp); 1748 SOCK_LOCK(so); 1749 so->so_state &= ~SS_PROTOREF; 1750 sofree(so); 1751 return (NULL); 1752 } 1753 return (tp); 1754 } 1755 1756 void 1757 tcp_drain(void) 1758 { 1759 VNET_ITERATOR_DECL(vnet_iter); 1760 1761 if (!do_tcpdrain) 1762 return; 1763 1764 VNET_LIST_RLOCK_NOSLEEP(); 1765 VNET_FOREACH(vnet_iter) { 1766 CURVNET_SET(vnet_iter); 1767 struct inpcb *inpb; 1768 struct tcpcb *tcpb; 1769 1770 /* 1771 * Walk the tcpbs, if existing, and flush the reassembly queue, 1772 * if there is one... 1773 * XXX: The "Net/3" implementation doesn't imply that the TCP 1774 * reassembly queue should be flushed, but in a situation 1775 * where we're really low on mbufs, this is potentially 1776 * useful. 1777 */ 1778 INP_INFO_WLOCK(&V_tcbinfo); 1779 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1780 if (inpb->inp_flags & INP_TIMEWAIT) 1781 continue; 1782 INP_WLOCK(inpb); 1783 if ((tcpb = intotcpcb(inpb)) != NULL) { 1784 tcp_reass_flush(tcpb); 1785 tcp_clean_sackreport(tcpb); 1786 #ifdef TCP_BLACKBOX 1787 tcp_log_drain(tcpb); 1788 #endif 1789 #ifdef TCPPCAP 1790 if (tcp_pcap_aggressive_free) { 1791 /* Free the TCP PCAP queues. */ 1792 tcp_pcap_drain(&(tcpb->t_inpkts)); 1793 tcp_pcap_drain(&(tcpb->t_outpkts)); 1794 } 1795 #endif 1796 } 1797 INP_WUNLOCK(inpb); 1798 } 1799 INP_INFO_WUNLOCK(&V_tcbinfo); 1800 CURVNET_RESTORE(); 1801 } 1802 VNET_LIST_RUNLOCK_NOSLEEP(); 1803 } 1804 1805 /* 1806 * Notify a tcp user of an asynchronous error; 1807 * store error as soft error, but wake up user 1808 * (for now, won't do anything until can select for soft error). 1809 * 1810 * Do not wake up user since there currently is no mechanism for 1811 * reporting soft errors (yet - a kqueue filter may be added). 1812 */ 1813 static struct inpcb * 1814 tcp_notify(struct inpcb *inp, int error) 1815 { 1816 struct tcpcb *tp; 1817 1818 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1819 INP_WLOCK_ASSERT(inp); 1820 1821 if ((inp->inp_flags & INP_TIMEWAIT) || 1822 (inp->inp_flags & INP_DROPPED)) 1823 return (inp); 1824 1825 tp = intotcpcb(inp); 1826 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1827 1828 /* 1829 * Ignore some errors if we are hooked up. 1830 * If connection hasn't completed, has retransmitted several times, 1831 * and receives a second error, give up now. This is better 1832 * than waiting a long time to establish a connection that 1833 * can never complete. 1834 */ 1835 if (tp->t_state == TCPS_ESTABLISHED && 1836 (error == EHOSTUNREACH || error == ENETUNREACH || 1837 error == EHOSTDOWN)) { 1838 if (inp->inp_route.ro_rt) { 1839 RTFREE(inp->inp_route.ro_rt); 1840 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1841 } 1842 return (inp); 1843 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1844 tp->t_softerror) { 1845 tp = tcp_drop(tp, error); 1846 if (tp != NULL) 1847 return (inp); 1848 else 1849 return (NULL); 1850 } else { 1851 tp->t_softerror = error; 1852 return (inp); 1853 } 1854 #if 0 1855 wakeup( &so->so_timeo); 1856 sorwakeup(so); 1857 sowwakeup(so); 1858 #endif 1859 } 1860 1861 static int 1862 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1863 { 1864 int error, i, m, n, pcb_count; 1865 struct inpcb *inp, **inp_list; 1866 inp_gen_t gencnt; 1867 struct xinpgen xig; 1868 1869 /* 1870 * The process of preparing the TCB list is too time-consuming and 1871 * resource-intensive to repeat twice on every request. 1872 */ 1873 if (req->oldptr == NULL) { 1874 n = V_tcbinfo.ipi_count + 1875 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1876 n += imax(n / 8, 10); 1877 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1878 return (0); 1879 } 1880 1881 if (req->newptr != NULL) 1882 return (EPERM); 1883 1884 /* 1885 * OK, now we're committed to doing something. 1886 */ 1887 INP_LIST_RLOCK(&V_tcbinfo); 1888 gencnt = V_tcbinfo.ipi_gencnt; 1889 n = V_tcbinfo.ipi_count; 1890 INP_LIST_RUNLOCK(&V_tcbinfo); 1891 1892 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1893 1894 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1895 + (n + m) * sizeof(struct xtcpcb)); 1896 if (error != 0) 1897 return (error); 1898 1899 xig.xig_len = sizeof xig; 1900 xig.xig_count = n + m; 1901 xig.xig_gen = gencnt; 1902 xig.xig_sogen = so_gencnt; 1903 error = SYSCTL_OUT(req, &xig, sizeof xig); 1904 if (error) 1905 return (error); 1906 1907 error = syncache_pcblist(req, m, &pcb_count); 1908 if (error) 1909 return (error); 1910 1911 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1912 1913 INP_INFO_WLOCK(&V_tcbinfo); 1914 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1915 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1916 INP_WLOCK(inp); 1917 if (inp->inp_gencnt <= gencnt) { 1918 /* 1919 * XXX: This use of cr_cansee(), introduced with 1920 * TCP state changes, is not quite right, but for 1921 * now, better than nothing. 1922 */ 1923 if (inp->inp_flags & INP_TIMEWAIT) { 1924 if (intotw(inp) != NULL) 1925 error = cr_cansee(req->td->td_ucred, 1926 intotw(inp)->tw_cred); 1927 else 1928 error = EINVAL; /* Skip this inp. */ 1929 } else 1930 error = cr_canseeinpcb(req->td->td_ucred, inp); 1931 if (error == 0) { 1932 in_pcbref(inp); 1933 inp_list[i++] = inp; 1934 } 1935 } 1936 INP_WUNLOCK(inp); 1937 } 1938 INP_INFO_WUNLOCK(&V_tcbinfo); 1939 n = i; 1940 1941 error = 0; 1942 for (i = 0; i < n; i++) { 1943 inp = inp_list[i]; 1944 INP_RLOCK(inp); 1945 if (inp->inp_gencnt <= gencnt) { 1946 struct xtcpcb xt; 1947 1948 tcp_inptoxtp(inp, &xt); 1949 INP_RUNLOCK(inp); 1950 error = SYSCTL_OUT(req, &xt, sizeof xt); 1951 } else 1952 INP_RUNLOCK(inp); 1953 } 1954 INP_INFO_RLOCK(&V_tcbinfo); 1955 for (i = 0; i < n; i++) { 1956 inp = inp_list[i]; 1957 INP_RLOCK(inp); 1958 if (!in_pcbrele_rlocked(inp)) 1959 INP_RUNLOCK(inp); 1960 } 1961 INP_INFO_RUNLOCK(&V_tcbinfo); 1962 1963 if (!error) { 1964 /* 1965 * Give the user an updated idea of our state. 1966 * If the generation differs from what we told 1967 * her before, she knows that something happened 1968 * while we were processing this request, and it 1969 * might be necessary to retry. 1970 */ 1971 INP_LIST_RLOCK(&V_tcbinfo); 1972 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1973 xig.xig_sogen = so_gencnt; 1974 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1975 INP_LIST_RUNLOCK(&V_tcbinfo); 1976 error = SYSCTL_OUT(req, &xig, sizeof xig); 1977 } 1978 free(inp_list, M_TEMP); 1979 return (error); 1980 } 1981 1982 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1983 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1984 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1985 1986 #ifdef INET 1987 static int 1988 tcp_getcred(SYSCTL_HANDLER_ARGS) 1989 { 1990 struct xucred xuc; 1991 struct sockaddr_in addrs[2]; 1992 struct inpcb *inp; 1993 int error; 1994 1995 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1996 if (error) 1997 return (error); 1998 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1999 if (error) 2000 return (error); 2001 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2002 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2003 if (inp != NULL) { 2004 if (inp->inp_socket == NULL) 2005 error = ENOENT; 2006 if (error == 0) 2007 error = cr_canseeinpcb(req->td->td_ucred, inp); 2008 if (error == 0) 2009 cru2x(inp->inp_cred, &xuc); 2010 INP_RUNLOCK(inp); 2011 } else 2012 error = ENOENT; 2013 if (error == 0) 2014 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2015 return (error); 2016 } 2017 2018 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2019 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2020 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2021 #endif /* INET */ 2022 2023 #ifdef INET6 2024 static int 2025 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2026 { 2027 struct xucred xuc; 2028 struct sockaddr_in6 addrs[2]; 2029 struct inpcb *inp; 2030 int error; 2031 #ifdef INET 2032 int mapped = 0; 2033 #endif 2034 2035 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2036 if (error) 2037 return (error); 2038 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2039 if (error) 2040 return (error); 2041 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2042 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2043 return (error); 2044 } 2045 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2046 #ifdef INET 2047 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2048 mapped = 1; 2049 else 2050 #endif 2051 return (EINVAL); 2052 } 2053 2054 #ifdef INET 2055 if (mapped == 1) 2056 inp = in_pcblookup(&V_tcbinfo, 2057 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2058 addrs[1].sin6_port, 2059 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2060 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2061 else 2062 #endif 2063 inp = in6_pcblookup(&V_tcbinfo, 2064 &addrs[1].sin6_addr, addrs[1].sin6_port, 2065 &addrs[0].sin6_addr, addrs[0].sin6_port, 2066 INPLOOKUP_RLOCKPCB, NULL); 2067 if (inp != NULL) { 2068 if (inp->inp_socket == NULL) 2069 error = ENOENT; 2070 if (error == 0) 2071 error = cr_canseeinpcb(req->td->td_ucred, inp); 2072 if (error == 0) 2073 cru2x(inp->inp_cred, &xuc); 2074 INP_RUNLOCK(inp); 2075 } else 2076 error = ENOENT; 2077 if (error == 0) 2078 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2079 return (error); 2080 } 2081 2082 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2083 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2084 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2085 #endif /* INET6 */ 2086 2087 2088 #ifdef INET 2089 void 2090 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2091 { 2092 struct ip *ip = vip; 2093 struct tcphdr *th; 2094 struct in_addr faddr; 2095 struct inpcb *inp; 2096 struct tcpcb *tp; 2097 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2098 struct icmp *icp; 2099 struct in_conninfo inc; 2100 tcp_seq icmp_tcp_seq; 2101 int mtu; 2102 2103 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2104 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2105 return; 2106 2107 if (cmd == PRC_MSGSIZE) 2108 notify = tcp_mtudisc_notify; 2109 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2110 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2111 cmd == PRC_TIMXCEED_INTRANS) && ip) 2112 notify = tcp_drop_syn_sent; 2113 2114 /* 2115 * Hostdead is ugly because it goes linearly through all PCBs. 2116 * XXX: We never get this from ICMP, otherwise it makes an 2117 * excellent DoS attack on machines with many connections. 2118 */ 2119 else if (cmd == PRC_HOSTDEAD) 2120 ip = NULL; 2121 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2122 return; 2123 2124 if (ip == NULL) { 2125 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2126 return; 2127 } 2128 2129 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2130 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2131 INP_INFO_RLOCK(&V_tcbinfo); 2132 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2133 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2134 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2135 /* signal EHOSTDOWN, as it flushes the cached route */ 2136 inp = (*notify)(inp, EHOSTDOWN); 2137 goto out; 2138 } 2139 icmp_tcp_seq = th->th_seq; 2140 if (inp != NULL) { 2141 if (!(inp->inp_flags & INP_TIMEWAIT) && 2142 !(inp->inp_flags & INP_DROPPED) && 2143 !(inp->inp_socket == NULL)) { 2144 tp = intotcpcb(inp); 2145 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2146 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2147 if (cmd == PRC_MSGSIZE) { 2148 /* 2149 * MTU discovery: 2150 * If we got a needfrag set the MTU 2151 * in the route to the suggested new 2152 * value (if given) and then notify. 2153 */ 2154 mtu = ntohs(icp->icmp_nextmtu); 2155 /* 2156 * If no alternative MTU was 2157 * proposed, try the next smaller 2158 * one. 2159 */ 2160 if (!mtu) 2161 mtu = ip_next_mtu( 2162 ntohs(ip->ip_len), 1); 2163 if (mtu < V_tcp_minmss + 2164 sizeof(struct tcpiphdr)) 2165 mtu = V_tcp_minmss + 2166 sizeof(struct tcpiphdr); 2167 /* 2168 * Only process the offered MTU if it 2169 * is smaller than the current one. 2170 */ 2171 if (mtu < tp->t_maxseg + 2172 sizeof(struct tcpiphdr)) { 2173 bzero(&inc, sizeof(inc)); 2174 inc.inc_faddr = faddr; 2175 inc.inc_fibnum = 2176 inp->inp_inc.inc_fibnum; 2177 tcp_hc_updatemtu(&inc, mtu); 2178 tcp_mtudisc(inp, mtu); 2179 } 2180 } else 2181 inp = (*notify)(inp, 2182 inetctlerrmap[cmd]); 2183 } 2184 } 2185 } else { 2186 bzero(&inc, sizeof(inc)); 2187 inc.inc_fport = th->th_dport; 2188 inc.inc_lport = th->th_sport; 2189 inc.inc_faddr = faddr; 2190 inc.inc_laddr = ip->ip_src; 2191 syncache_unreach(&inc, icmp_tcp_seq); 2192 } 2193 out: 2194 if (inp != NULL) 2195 INP_WUNLOCK(inp); 2196 INP_INFO_RUNLOCK(&V_tcbinfo); 2197 } 2198 #endif /* INET */ 2199 2200 #ifdef INET6 2201 void 2202 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2203 { 2204 struct in6_addr *dst; 2205 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2206 struct ip6_hdr *ip6; 2207 struct mbuf *m; 2208 struct inpcb *inp; 2209 struct tcpcb *tp; 2210 struct icmp6_hdr *icmp6; 2211 struct ip6ctlparam *ip6cp = NULL; 2212 const struct sockaddr_in6 *sa6_src = NULL; 2213 struct in_conninfo inc; 2214 struct tcp_ports { 2215 uint16_t th_sport; 2216 uint16_t th_dport; 2217 } t_ports; 2218 tcp_seq icmp_tcp_seq; 2219 unsigned int mtu; 2220 unsigned int off; 2221 2222 if (sa->sa_family != AF_INET6 || 2223 sa->sa_len != sizeof(struct sockaddr_in6)) 2224 return; 2225 2226 /* if the parameter is from icmp6, decode it. */ 2227 if (d != NULL) { 2228 ip6cp = (struct ip6ctlparam *)d; 2229 icmp6 = ip6cp->ip6c_icmp6; 2230 m = ip6cp->ip6c_m; 2231 ip6 = ip6cp->ip6c_ip6; 2232 off = ip6cp->ip6c_off; 2233 sa6_src = ip6cp->ip6c_src; 2234 dst = ip6cp->ip6c_finaldst; 2235 } else { 2236 m = NULL; 2237 ip6 = NULL; 2238 off = 0; /* fool gcc */ 2239 sa6_src = &sa6_any; 2240 dst = NULL; 2241 } 2242 2243 if (cmd == PRC_MSGSIZE) 2244 notify = tcp_mtudisc_notify; 2245 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2246 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2247 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2248 notify = tcp_drop_syn_sent; 2249 2250 /* 2251 * Hostdead is ugly because it goes linearly through all PCBs. 2252 * XXX: We never get this from ICMP, otherwise it makes an 2253 * excellent DoS attack on machines with many connections. 2254 */ 2255 else if (cmd == PRC_HOSTDEAD) 2256 ip6 = NULL; 2257 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2258 return; 2259 2260 if (ip6 == NULL) { 2261 in6_pcbnotify(&V_tcbinfo, sa, 0, 2262 (const struct sockaddr *)sa6_src, 2263 0, cmd, NULL, notify); 2264 return; 2265 } 2266 2267 /* Check if we can safely get the ports from the tcp hdr */ 2268 if (m == NULL || 2269 (m->m_pkthdr.len < 2270 (int32_t) (off + sizeof(struct tcp_ports)))) { 2271 return; 2272 } 2273 bzero(&t_ports, sizeof(struct tcp_ports)); 2274 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2275 INP_INFO_RLOCK(&V_tcbinfo); 2276 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2277 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2278 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2279 /* signal EHOSTDOWN, as it flushes the cached route */ 2280 inp = (*notify)(inp, EHOSTDOWN); 2281 goto out; 2282 } 2283 off += sizeof(struct tcp_ports); 2284 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2285 goto out; 2286 } 2287 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2288 if (inp != NULL) { 2289 if (!(inp->inp_flags & INP_TIMEWAIT) && 2290 !(inp->inp_flags & INP_DROPPED) && 2291 !(inp->inp_socket == NULL)) { 2292 tp = intotcpcb(inp); 2293 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2294 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2295 if (cmd == PRC_MSGSIZE) { 2296 /* 2297 * MTU discovery: 2298 * If we got a needfrag set the MTU 2299 * in the route to the suggested new 2300 * value (if given) and then notify. 2301 */ 2302 mtu = ntohl(icmp6->icmp6_mtu); 2303 /* 2304 * If no alternative MTU was 2305 * proposed, or the proposed 2306 * MTU was too small, set to 2307 * the min. 2308 */ 2309 if (mtu < IPV6_MMTU) 2310 mtu = IPV6_MMTU - 8; 2311 bzero(&inc, sizeof(inc)); 2312 inc.inc_fibnum = M_GETFIB(m); 2313 inc.inc_flags |= INC_ISIPV6; 2314 inc.inc6_faddr = *dst; 2315 if (in6_setscope(&inc.inc6_faddr, 2316 m->m_pkthdr.rcvif, NULL)) 2317 goto out; 2318 /* 2319 * Only process the offered MTU if it 2320 * is smaller than the current one. 2321 */ 2322 if (mtu < tp->t_maxseg + 2323 sizeof (struct tcphdr) + 2324 sizeof (struct ip6_hdr)) { 2325 tcp_hc_updatemtu(&inc, mtu); 2326 tcp_mtudisc(inp, mtu); 2327 ICMP6STAT_INC(icp6s_pmtuchg); 2328 } 2329 } else 2330 inp = (*notify)(inp, 2331 inet6ctlerrmap[cmd]); 2332 } 2333 } 2334 } else { 2335 bzero(&inc, sizeof(inc)); 2336 inc.inc_fibnum = M_GETFIB(m); 2337 inc.inc_flags |= INC_ISIPV6; 2338 inc.inc_fport = t_ports.th_dport; 2339 inc.inc_lport = t_ports.th_sport; 2340 inc.inc6_faddr = *dst; 2341 inc.inc6_laddr = ip6->ip6_src; 2342 syncache_unreach(&inc, icmp_tcp_seq); 2343 } 2344 out: 2345 if (inp != NULL) 2346 INP_WUNLOCK(inp); 2347 INP_INFO_RUNLOCK(&V_tcbinfo); 2348 } 2349 #endif /* INET6 */ 2350 2351 2352 /* 2353 * Following is where TCP initial sequence number generation occurs. 2354 * 2355 * There are two places where we must use initial sequence numbers: 2356 * 1. In SYN-ACK packets. 2357 * 2. In SYN packets. 2358 * 2359 * All ISNs for SYN-ACK packets are generated by the syncache. See 2360 * tcp_syncache.c for details. 2361 * 2362 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2363 * depends on this property. In addition, these ISNs should be 2364 * unguessable so as to prevent connection hijacking. To satisfy 2365 * the requirements of this situation, the algorithm outlined in 2366 * RFC 1948 is used, with only small modifications. 2367 * 2368 * Implementation details: 2369 * 2370 * Time is based off the system timer, and is corrected so that it 2371 * increases by one megabyte per second. This allows for proper 2372 * recycling on high speed LANs while still leaving over an hour 2373 * before rollover. 2374 * 2375 * As reading the *exact* system time is too expensive to be done 2376 * whenever setting up a TCP connection, we increment the time 2377 * offset in two ways. First, a small random positive increment 2378 * is added to isn_offset for each connection that is set up. 2379 * Second, the function tcp_isn_tick fires once per clock tick 2380 * and increments isn_offset as necessary so that sequence numbers 2381 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2382 * random positive increments serve only to ensure that the same 2383 * exact sequence number is never sent out twice (as could otherwise 2384 * happen when a port is recycled in less than the system tick 2385 * interval.) 2386 * 2387 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2388 * between seeding of isn_secret. This is normally set to zero, 2389 * as reseeding should not be necessary. 2390 * 2391 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2392 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2393 * general, this means holding an exclusive (write) lock. 2394 */ 2395 2396 #define ISN_BYTES_PER_SECOND 1048576 2397 #define ISN_STATIC_INCREMENT 4096 2398 #define ISN_RANDOM_INCREMENT (4096 - 1) 2399 2400 static VNET_DEFINE(u_char, isn_secret[32]); 2401 static VNET_DEFINE(int, isn_last); 2402 static VNET_DEFINE(int, isn_last_reseed); 2403 static VNET_DEFINE(u_int32_t, isn_offset); 2404 static VNET_DEFINE(u_int32_t, isn_offset_old); 2405 2406 #define V_isn_secret VNET(isn_secret) 2407 #define V_isn_last VNET(isn_last) 2408 #define V_isn_last_reseed VNET(isn_last_reseed) 2409 #define V_isn_offset VNET(isn_offset) 2410 #define V_isn_offset_old VNET(isn_offset_old) 2411 2412 tcp_seq 2413 tcp_new_isn(struct tcpcb *tp) 2414 { 2415 MD5_CTX isn_ctx; 2416 u_int32_t md5_buffer[4]; 2417 tcp_seq new_isn; 2418 u_int32_t projected_offset; 2419 2420 INP_WLOCK_ASSERT(tp->t_inpcb); 2421 2422 ISN_LOCK(); 2423 /* Seed if this is the first use, reseed if requested. */ 2424 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2425 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2426 < (u_int)ticks))) { 2427 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2428 V_isn_last_reseed = ticks; 2429 } 2430 2431 /* Compute the md5 hash and return the ISN. */ 2432 MD5Init(&isn_ctx); 2433 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2434 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2435 #ifdef INET6 2436 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2437 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2438 sizeof(struct in6_addr)); 2439 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2440 sizeof(struct in6_addr)); 2441 } else 2442 #endif 2443 { 2444 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2445 sizeof(struct in_addr)); 2446 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2447 sizeof(struct in_addr)); 2448 } 2449 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2450 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2451 new_isn = (tcp_seq) md5_buffer[0]; 2452 V_isn_offset += ISN_STATIC_INCREMENT + 2453 (arc4random() & ISN_RANDOM_INCREMENT); 2454 if (ticks != V_isn_last) { 2455 projected_offset = V_isn_offset_old + 2456 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2457 if (SEQ_GT(projected_offset, V_isn_offset)) 2458 V_isn_offset = projected_offset; 2459 V_isn_offset_old = V_isn_offset; 2460 V_isn_last = ticks; 2461 } 2462 new_isn += V_isn_offset; 2463 ISN_UNLOCK(); 2464 return (new_isn); 2465 } 2466 2467 /* 2468 * When a specific ICMP unreachable message is received and the 2469 * connection state is SYN-SENT, drop the connection. This behavior 2470 * is controlled by the icmp_may_rst sysctl. 2471 */ 2472 struct inpcb * 2473 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2474 { 2475 struct tcpcb *tp; 2476 2477 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2478 INP_WLOCK_ASSERT(inp); 2479 2480 if ((inp->inp_flags & INP_TIMEWAIT) || 2481 (inp->inp_flags & INP_DROPPED)) 2482 return (inp); 2483 2484 tp = intotcpcb(inp); 2485 if (tp->t_state != TCPS_SYN_SENT) 2486 return (inp); 2487 2488 if (IS_FASTOPEN(tp->t_flags)) 2489 tcp_fastopen_disable_path(tp); 2490 2491 tp = tcp_drop(tp, errno); 2492 if (tp != NULL) 2493 return (inp); 2494 else 2495 return (NULL); 2496 } 2497 2498 /* 2499 * When `need fragmentation' ICMP is received, update our idea of the MSS 2500 * based on the new value. Also nudge TCP to send something, since we 2501 * know the packet we just sent was dropped. 2502 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2503 */ 2504 static struct inpcb * 2505 tcp_mtudisc_notify(struct inpcb *inp, int error) 2506 { 2507 2508 tcp_mtudisc(inp, -1); 2509 return (inp); 2510 } 2511 2512 static void 2513 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2514 { 2515 struct tcpcb *tp; 2516 struct socket *so; 2517 2518 INP_WLOCK_ASSERT(inp); 2519 if ((inp->inp_flags & INP_TIMEWAIT) || 2520 (inp->inp_flags & INP_DROPPED)) 2521 return; 2522 2523 tp = intotcpcb(inp); 2524 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2525 2526 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2527 2528 so = inp->inp_socket; 2529 SOCKBUF_LOCK(&so->so_snd); 2530 /* If the mss is larger than the socket buffer, decrease the mss. */ 2531 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2532 tp->t_maxseg = so->so_snd.sb_hiwat; 2533 SOCKBUF_UNLOCK(&so->so_snd); 2534 2535 TCPSTAT_INC(tcps_mturesent); 2536 tp->t_rtttime = 0; 2537 tp->snd_nxt = tp->snd_una; 2538 tcp_free_sackholes(tp); 2539 tp->snd_recover = tp->snd_max; 2540 if (tp->t_flags & TF_SACK_PERMIT) 2541 EXIT_FASTRECOVERY(tp->t_flags); 2542 tp->t_fb->tfb_tcp_output(tp); 2543 } 2544 2545 #ifdef INET 2546 /* 2547 * Look-up the routing entry to the peer of this inpcb. If no route 2548 * is found and it cannot be allocated, then return 0. This routine 2549 * is called by TCP routines that access the rmx structure and by 2550 * tcp_mss_update to get the peer/interface MTU. 2551 */ 2552 uint32_t 2553 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2554 { 2555 struct nhop4_extended nh4; 2556 struct ifnet *ifp; 2557 uint32_t maxmtu = 0; 2558 2559 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2560 2561 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2562 2563 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2564 NHR_REF, 0, &nh4) != 0) 2565 return (0); 2566 2567 ifp = nh4.nh_ifp; 2568 maxmtu = nh4.nh_mtu; 2569 2570 /* Report additional interface capabilities. */ 2571 if (cap != NULL) { 2572 if (ifp->if_capenable & IFCAP_TSO4 && 2573 ifp->if_hwassist & CSUM_TSO) { 2574 cap->ifcap |= CSUM_TSO; 2575 cap->tsomax = ifp->if_hw_tsomax; 2576 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2577 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2578 } 2579 } 2580 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2581 } 2582 return (maxmtu); 2583 } 2584 #endif /* INET */ 2585 2586 #ifdef INET6 2587 uint32_t 2588 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2589 { 2590 struct nhop6_extended nh6; 2591 struct in6_addr dst6; 2592 uint32_t scopeid; 2593 struct ifnet *ifp; 2594 uint32_t maxmtu = 0; 2595 2596 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2597 2598 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2599 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2600 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2601 0, &nh6) != 0) 2602 return (0); 2603 2604 ifp = nh6.nh_ifp; 2605 maxmtu = nh6.nh_mtu; 2606 2607 /* Report additional interface capabilities. */ 2608 if (cap != NULL) { 2609 if (ifp->if_capenable & IFCAP_TSO6 && 2610 ifp->if_hwassist & CSUM_TSO) { 2611 cap->ifcap |= CSUM_TSO; 2612 cap->tsomax = ifp->if_hw_tsomax; 2613 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2614 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2615 } 2616 } 2617 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2618 } 2619 2620 return (maxmtu); 2621 } 2622 #endif /* INET6 */ 2623 2624 /* 2625 * Calculate effective SMSS per RFC5681 definition for a given TCP 2626 * connection at its current state, taking into account SACK and etc. 2627 */ 2628 u_int 2629 tcp_maxseg(const struct tcpcb *tp) 2630 { 2631 u_int optlen; 2632 2633 if (tp->t_flags & TF_NOOPT) 2634 return (tp->t_maxseg); 2635 2636 /* 2637 * Here we have a simplified code from tcp_addoptions(), 2638 * without a proper loop, and having most of paddings hardcoded. 2639 * We might make mistakes with padding here in some edge cases, 2640 * but this is harmless, since result of tcp_maxseg() is used 2641 * only in cwnd and ssthresh estimations. 2642 */ 2643 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2644 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2645 if (tp->t_flags & TF_RCVD_TSTMP) 2646 optlen = TCPOLEN_TSTAMP_APPA; 2647 else 2648 optlen = 0; 2649 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2650 if (tp->t_flags & TF_SIGNATURE) 2651 optlen += PAD(TCPOLEN_SIGNATURE); 2652 #endif 2653 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2654 optlen += TCPOLEN_SACKHDR; 2655 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2656 optlen = PAD(optlen); 2657 } 2658 } else { 2659 if (tp->t_flags & TF_REQ_TSTMP) 2660 optlen = TCPOLEN_TSTAMP_APPA; 2661 else 2662 optlen = PAD(TCPOLEN_MAXSEG); 2663 if (tp->t_flags & TF_REQ_SCALE) 2664 optlen += PAD(TCPOLEN_WINDOW); 2665 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2666 if (tp->t_flags & TF_SIGNATURE) 2667 optlen += PAD(TCPOLEN_SIGNATURE); 2668 #endif 2669 if (tp->t_flags & TF_SACK_PERMIT) 2670 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2671 } 2672 #undef PAD 2673 optlen = min(optlen, TCP_MAXOLEN); 2674 return (tp->t_maxseg - optlen); 2675 } 2676 2677 static int 2678 sysctl_drop(SYSCTL_HANDLER_ARGS) 2679 { 2680 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2681 struct sockaddr_storage addrs[2]; 2682 struct inpcb *inp; 2683 struct tcpcb *tp; 2684 struct tcptw *tw; 2685 struct sockaddr_in *fin, *lin; 2686 #ifdef INET6 2687 struct sockaddr_in6 *fin6, *lin6; 2688 #endif 2689 int error; 2690 2691 inp = NULL; 2692 fin = lin = NULL; 2693 #ifdef INET6 2694 fin6 = lin6 = NULL; 2695 #endif 2696 error = 0; 2697 2698 if (req->oldptr != NULL || req->oldlen != 0) 2699 return (EINVAL); 2700 if (req->newptr == NULL) 2701 return (EPERM); 2702 if (req->newlen < sizeof(addrs)) 2703 return (ENOMEM); 2704 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2705 if (error) 2706 return (error); 2707 2708 switch (addrs[0].ss_family) { 2709 #ifdef INET6 2710 case AF_INET6: 2711 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2712 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2713 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2714 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2715 return (EINVAL); 2716 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2717 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2718 return (EINVAL); 2719 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2720 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2721 fin = (struct sockaddr_in *)&addrs[0]; 2722 lin = (struct sockaddr_in *)&addrs[1]; 2723 break; 2724 } 2725 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2726 if (error) 2727 return (error); 2728 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2729 if (error) 2730 return (error); 2731 break; 2732 #endif 2733 #ifdef INET 2734 case AF_INET: 2735 fin = (struct sockaddr_in *)&addrs[0]; 2736 lin = (struct sockaddr_in *)&addrs[1]; 2737 if (fin->sin_len != sizeof(struct sockaddr_in) || 2738 lin->sin_len != sizeof(struct sockaddr_in)) 2739 return (EINVAL); 2740 break; 2741 #endif 2742 default: 2743 return (EINVAL); 2744 } 2745 INP_INFO_RLOCK(&V_tcbinfo); 2746 switch (addrs[0].ss_family) { 2747 #ifdef INET6 2748 case AF_INET6: 2749 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2750 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2751 INPLOOKUP_WLOCKPCB, NULL); 2752 break; 2753 #endif 2754 #ifdef INET 2755 case AF_INET: 2756 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2757 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2758 break; 2759 #endif 2760 } 2761 if (inp != NULL) { 2762 if (inp->inp_flags & INP_TIMEWAIT) { 2763 /* 2764 * XXXRW: There currently exists a state where an 2765 * inpcb is present, but its timewait state has been 2766 * discarded. For now, don't allow dropping of this 2767 * type of inpcb. 2768 */ 2769 tw = intotw(inp); 2770 if (tw != NULL) 2771 tcp_twclose(tw, 0); 2772 else 2773 INP_WUNLOCK(inp); 2774 } else if (!(inp->inp_flags & INP_DROPPED) && 2775 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2776 tp = intotcpcb(inp); 2777 tp = tcp_drop(tp, ECONNABORTED); 2778 if (tp != NULL) 2779 INP_WUNLOCK(inp); 2780 } else 2781 INP_WUNLOCK(inp); 2782 } else 2783 error = ESRCH; 2784 INP_INFO_RUNLOCK(&V_tcbinfo); 2785 return (error); 2786 } 2787 2788 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2789 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2790 0, sysctl_drop, "", "Drop TCP connection"); 2791 2792 /* 2793 * Generate a standardized TCP log line for use throughout the 2794 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2795 * allow use in the interrupt context. 2796 * 2797 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2798 * NB: The function may return NULL if memory allocation failed. 2799 * 2800 * Due to header inclusion and ordering limitations the struct ip 2801 * and ip6_hdr pointers have to be passed as void pointers. 2802 */ 2803 char * 2804 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2805 const void *ip6hdr) 2806 { 2807 2808 /* Is logging enabled? */ 2809 if (tcp_log_in_vain == 0) 2810 return (NULL); 2811 2812 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2813 } 2814 2815 char * 2816 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2817 const void *ip6hdr) 2818 { 2819 2820 /* Is logging enabled? */ 2821 if (tcp_log_debug == 0) 2822 return (NULL); 2823 2824 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2825 } 2826 2827 static char * 2828 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2829 const void *ip6hdr) 2830 { 2831 char *s, *sp; 2832 size_t size; 2833 struct ip *ip; 2834 #ifdef INET6 2835 const struct ip6_hdr *ip6; 2836 2837 ip6 = (const struct ip6_hdr *)ip6hdr; 2838 #endif /* INET6 */ 2839 ip = (struct ip *)ip4hdr; 2840 2841 /* 2842 * The log line looks like this: 2843 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2844 */ 2845 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2846 sizeof(PRINT_TH_FLAGS) + 1 + 2847 #ifdef INET6 2848 2 * INET6_ADDRSTRLEN; 2849 #else 2850 2 * INET_ADDRSTRLEN; 2851 #endif /* INET6 */ 2852 2853 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2854 if (s == NULL) 2855 return (NULL); 2856 2857 strcat(s, "TCP: ["); 2858 sp = s + strlen(s); 2859 2860 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2861 inet_ntoa_r(inc->inc_faddr, sp); 2862 sp = s + strlen(s); 2863 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2864 sp = s + strlen(s); 2865 inet_ntoa_r(inc->inc_laddr, sp); 2866 sp = s + strlen(s); 2867 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2868 #ifdef INET6 2869 } else if (inc) { 2870 ip6_sprintf(sp, &inc->inc6_faddr); 2871 sp = s + strlen(s); 2872 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2873 sp = s + strlen(s); 2874 ip6_sprintf(sp, &inc->inc6_laddr); 2875 sp = s + strlen(s); 2876 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2877 } else if (ip6 && th) { 2878 ip6_sprintf(sp, &ip6->ip6_src); 2879 sp = s + strlen(s); 2880 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2881 sp = s + strlen(s); 2882 ip6_sprintf(sp, &ip6->ip6_dst); 2883 sp = s + strlen(s); 2884 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2885 #endif /* INET6 */ 2886 #ifdef INET 2887 } else if (ip && th) { 2888 inet_ntoa_r(ip->ip_src, sp); 2889 sp = s + strlen(s); 2890 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2891 sp = s + strlen(s); 2892 inet_ntoa_r(ip->ip_dst, sp); 2893 sp = s + strlen(s); 2894 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2895 #endif /* INET */ 2896 } else { 2897 free(s, M_TCPLOG); 2898 return (NULL); 2899 } 2900 sp = s + strlen(s); 2901 if (th) 2902 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2903 if (*(s + size - 1) != '\0') 2904 panic("%s: string too long", __func__); 2905 return (s); 2906 } 2907 2908 /* 2909 * A subroutine which makes it easy to track TCP state changes with DTrace. 2910 * This function shouldn't be called for t_state initializations that don't 2911 * correspond to actual TCP state transitions. 2912 */ 2913 void 2914 tcp_state_change(struct tcpcb *tp, int newstate) 2915 { 2916 #if defined(KDTRACE_HOOKS) 2917 int pstate = tp->t_state; 2918 #endif 2919 2920 TCPSTATES_DEC(tp->t_state); 2921 TCPSTATES_INC(newstate); 2922 tp->t_state = newstate; 2923 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2924 } 2925 2926 /* 2927 * Create an external-format (``xtcpcb'') structure using the information in 2928 * the kernel-format tcpcb structure pointed to by tp. This is done to 2929 * reduce the spew of irrelevant information over this interface, to isolate 2930 * user code from changes in the kernel structure, and potentially to provide 2931 * information-hiding if we decide that some of this information should be 2932 * hidden from users. 2933 */ 2934 void 2935 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 2936 { 2937 struct tcpcb *tp = intotcpcb(inp); 2938 sbintime_t now; 2939 2940 if (inp->inp_flags & INP_TIMEWAIT) { 2941 bzero(xt, sizeof(struct xtcpcb)); 2942 xt->t_state = TCPS_TIME_WAIT; 2943 } else { 2944 xt->t_state = tp->t_state; 2945 xt->t_logstate = tp->t_logstate; 2946 xt->t_flags = tp->t_flags; 2947 xt->t_sndzerowin = tp->t_sndzerowin; 2948 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 2949 xt->t_rcvoopack = tp->t_rcvoopack; 2950 2951 now = getsbinuptime(); 2952 #define COPYTIMER(ttt) do { \ 2953 if (callout_active(&tp->t_timers->ttt)) \ 2954 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 2955 SBT_1MS; \ 2956 else \ 2957 xt->ttt = 0; \ 2958 } while (0) 2959 COPYTIMER(tt_delack); 2960 COPYTIMER(tt_rexmt); 2961 COPYTIMER(tt_persist); 2962 COPYTIMER(tt_keep); 2963 COPYTIMER(tt_2msl); 2964 #undef COPYTIMER 2965 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 2966 2967 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 2968 TCP_FUNCTION_NAME_LEN_MAX); 2969 bzero(xt->xt_logid, TCP_LOG_ID_LEN); 2970 #ifdef TCP_BLACKBOX 2971 (void)tcp_log_get_id(tp, xt->xt_logid); 2972 #endif 2973 } 2974 2975 xt->xt_len = sizeof(struct xtcpcb); 2976 in_pcbtoxinpcb(inp, &xt->xt_inp); 2977 if (inp->inp_socket == NULL) 2978 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 2979 } 2980