1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_compat.h" 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_tcpdebug.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/eventhandler.h> 45 #include <sys/hhook.h> 46 #include <sys/kernel.h> 47 #include <sys/khelp.h> 48 #include <sys/sysctl.h> 49 #include <sys/jail.h> 50 #include <sys/malloc.h> 51 #include <sys/refcount.h> 52 #include <sys/mbuf.h> 53 #ifdef INET6 54 #include <sys/domain.h> 55 #endif 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/sdt.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/protosw.h> 62 #include <sys/random.h> 63 64 #include <vm/uma.h> 65 66 #include <net/route.h> 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/vnet.h> 70 71 #include <netinet/in.h> 72 #include <netinet/in_fib.h> 73 #include <netinet/in_kdtrace.h> 74 #include <netinet/in_pcb.h> 75 #include <netinet/in_systm.h> 76 #include <netinet/in_var.h> 77 #include <netinet/ip.h> 78 #include <netinet/ip_icmp.h> 79 #include <netinet/ip_var.h> 80 #ifdef INET6 81 #include <netinet/ip6.h> 82 #include <netinet6/in6_fib.h> 83 #include <netinet6/in6_pcb.h> 84 #include <netinet6/ip6_var.h> 85 #include <netinet6/scope6_var.h> 86 #include <netinet6/nd6.h> 87 #endif 88 89 #ifdef TCP_RFC7413 90 #include <netinet/tcp_fastopen.h> 91 #endif 92 #include <netinet/tcp.h> 93 #include <netinet/tcp_fsm.h> 94 #include <netinet/tcp_seq.h> 95 #include <netinet/tcp_timer.h> 96 #include <netinet/tcp_var.h> 97 #include <netinet/tcp_syncache.h> 98 #include <netinet/cc/cc.h> 99 #ifdef INET6 100 #include <netinet6/tcp6_var.h> 101 #endif 102 #include <netinet/tcpip.h> 103 #ifdef TCPPCAP 104 #include <netinet/tcp_pcap.h> 105 #endif 106 #ifdef TCPDEBUG 107 #include <netinet/tcp_debug.h> 108 #endif 109 #ifdef INET6 110 #include <netinet6/ip6protosw.h> 111 #endif 112 #ifdef TCP_OFFLOAD 113 #include <netinet/tcp_offload.h> 114 #endif 115 116 #ifdef IPSEC 117 #include <netipsec/ipsec.h> 118 #include <netipsec/xform.h> 119 #ifdef INET6 120 #include <netipsec/ipsec6.h> 121 #endif 122 #include <netipsec/key.h> 123 #include <sys/syslog.h> 124 #endif /*IPSEC*/ 125 126 #include <machine/in_cksum.h> 127 #include <sys/md5.h> 128 129 #include <security/mac/mac_framework.h> 130 131 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 132 #ifdef INET6 133 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 134 #endif 135 136 struct rwlock tcp_function_lock; 137 138 static int 139 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 140 { 141 int error, new; 142 143 new = V_tcp_mssdflt; 144 error = sysctl_handle_int(oidp, &new, 0, req); 145 if (error == 0 && req->newptr) { 146 if (new < TCP_MINMSS) 147 error = EINVAL; 148 else 149 V_tcp_mssdflt = new; 150 } 151 return (error); 152 } 153 154 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 155 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 156 &sysctl_net_inet_tcp_mss_check, "I", 157 "Default TCP Maximum Segment Size"); 158 159 #ifdef INET6 160 static int 161 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 162 { 163 int error, new; 164 165 new = V_tcp_v6mssdflt; 166 error = sysctl_handle_int(oidp, &new, 0, req); 167 if (error == 0 && req->newptr) { 168 if (new < TCP_MINMSS) 169 error = EINVAL; 170 else 171 V_tcp_v6mssdflt = new; 172 } 173 return (error); 174 } 175 176 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 177 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 178 &sysctl_net_inet_tcp_mss_v6_check, "I", 179 "Default TCP Maximum Segment Size for IPv6"); 180 #endif /* INET6 */ 181 182 /* 183 * Minimum MSS we accept and use. This prevents DoS attacks where 184 * we are forced to a ridiculous low MSS like 20 and send hundreds 185 * of packets instead of one. The effect scales with the available 186 * bandwidth and quickly saturates the CPU and network interface 187 * with packet generation and sending. Set to zero to disable MINMSS 188 * checking. This setting prevents us from sending too small packets. 189 */ 190 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_minmss), 0, 193 "Minimum TCP Maximum Segment Size"); 194 195 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_do_rfc1323), 0, 198 "Enable rfc1323 (high performance TCP) extensions"); 199 200 static int tcp_log_debug = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 202 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 203 204 static int tcp_tcbhashsize; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 206 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 207 208 static int do_tcpdrain = 1; 209 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 210 "Enable tcp_drain routine for extra help when low on mbufs"); 211 212 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 213 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 214 215 static VNET_DEFINE(int, icmp_may_rst) = 1; 216 #define V_icmp_may_rst VNET(icmp_may_rst) 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(icmp_may_rst), 0, 219 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 220 221 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 222 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_isn_reseed_interval), 0, 225 "Seconds between reseeding of ISN secret"); 226 227 static int tcp_soreceive_stream; 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 229 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 230 231 #ifdef TCP_SIGNATURE 232 static int tcp_sig_checksigs = 1; 233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, 234 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); 235 #endif 236 237 VNET_DEFINE(uma_zone_t, sack_hole_zone); 238 #define V_sack_hole_zone VNET(sack_hole_zone) 239 240 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 241 242 static struct inpcb *tcp_notify(struct inpcb *, int); 243 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 244 static void tcp_mtudisc(struct inpcb *, int); 245 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 246 void *ip4hdr, const void *ip6hdr); 247 static void tcp_timer_discard(struct tcpcb *, uint32_t); 248 249 250 static struct tcp_function_block tcp_def_funcblk = { 251 "default", 252 tcp_output, 253 tcp_do_segment, 254 tcp_default_ctloutput, 255 NULL, 256 NULL, 257 NULL, 258 NULL, 259 NULL, 260 NULL, 261 NULL, 262 0, 263 0 264 }; 265 266 int t_functions_inited = 0; 267 struct tcp_funchead t_functions; 268 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 269 270 static void 271 init_tcp_functions(void) 272 { 273 if (t_functions_inited == 0) { 274 TAILQ_INIT(&t_functions); 275 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 276 t_functions_inited = 1; 277 } 278 } 279 280 static struct tcp_function_block * 281 find_tcp_functions_locked(struct tcp_function_set *fs) 282 { 283 struct tcp_function *f; 284 struct tcp_function_block *blk=NULL; 285 286 TAILQ_FOREACH(f, &t_functions, tf_next) { 287 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) { 288 blk = f->tf_fb; 289 break; 290 } 291 } 292 return(blk); 293 } 294 295 static struct tcp_function_block * 296 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 297 { 298 struct tcp_function_block *rblk=NULL; 299 struct tcp_function *f; 300 301 TAILQ_FOREACH(f, &t_functions, tf_next) { 302 if (f->tf_fb == blk) { 303 rblk = blk; 304 if (s) { 305 *s = f; 306 } 307 break; 308 } 309 } 310 return (rblk); 311 } 312 313 struct tcp_function_block * 314 find_and_ref_tcp_functions(struct tcp_function_set *fs) 315 { 316 struct tcp_function_block *blk; 317 318 rw_rlock(&tcp_function_lock); 319 blk = find_tcp_functions_locked(fs); 320 if (blk) 321 refcount_acquire(&blk->tfb_refcnt); 322 rw_runlock(&tcp_function_lock); 323 return(blk); 324 } 325 326 struct tcp_function_block * 327 find_and_ref_tcp_fb(struct tcp_function_block *blk) 328 { 329 struct tcp_function_block *rblk; 330 331 rw_rlock(&tcp_function_lock); 332 rblk = find_tcp_fb_locked(blk, NULL); 333 if (rblk) 334 refcount_acquire(&rblk->tfb_refcnt); 335 rw_runlock(&tcp_function_lock); 336 return(rblk); 337 } 338 339 340 static int 341 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 342 { 343 int error=ENOENT; 344 struct tcp_function_set fs; 345 struct tcp_function_block *blk; 346 347 memset(&fs, 0, sizeof(fs)); 348 rw_rlock(&tcp_function_lock); 349 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 350 if (blk) { 351 /* Found him */ 352 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 353 fs.pcbcnt = blk->tfb_refcnt; 354 } 355 rw_runlock(&tcp_function_lock); 356 error = sysctl_handle_string(oidp, fs.function_set_name, 357 sizeof(fs.function_set_name), req); 358 359 /* Check for error or no change */ 360 if (error != 0 || req->newptr == NULL) 361 return(error); 362 363 rw_wlock(&tcp_function_lock); 364 blk = find_tcp_functions_locked(&fs); 365 if ((blk == NULL) || 366 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 367 error = ENOENT; 368 goto done; 369 } 370 tcp_func_set_ptr = blk; 371 done: 372 rw_wunlock(&tcp_function_lock); 373 return (error); 374 } 375 376 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 377 CTLTYPE_STRING | CTLFLAG_RW, 378 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 379 "Set/get the default TCP functions"); 380 381 static int 382 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 383 { 384 int error, cnt, linesz; 385 struct tcp_function *f; 386 char *buffer, *cp; 387 size_t bufsz, outsz; 388 389 cnt = 0; 390 rw_rlock(&tcp_function_lock); 391 TAILQ_FOREACH(f, &t_functions, tf_next) { 392 cnt++; 393 } 394 rw_runlock(&tcp_function_lock); 395 396 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1; 397 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 398 399 error = 0; 400 cp = buffer; 401 402 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count"); 403 cp += linesz; 404 bufsz -= linesz; 405 outsz = linesz; 406 407 rw_rlock(&tcp_function_lock); 408 TAILQ_FOREACH(f, &t_functions, tf_next) { 409 linesz = snprintf(cp, bufsz, "%-32s%c %u\n", 410 f->tf_fb->tfb_tcp_block_name, 411 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 412 f->tf_fb->tfb_refcnt); 413 if (linesz >= bufsz) { 414 error = EOVERFLOW; 415 break; 416 } 417 cp += linesz; 418 bufsz -= linesz; 419 outsz += linesz; 420 } 421 rw_runlock(&tcp_function_lock); 422 if (error == 0) 423 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 424 free(buffer, M_TEMP); 425 return (error); 426 } 427 428 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 429 CTLTYPE_STRING|CTLFLAG_RD, 430 NULL, 0, sysctl_net_inet_list_available, "A", 431 "list available TCP Function sets"); 432 433 /* 434 * Target size of TCP PCB hash tables. Must be a power of two. 435 * 436 * Note that this can be overridden by the kernel environment 437 * variable net.inet.tcp.tcbhashsize 438 */ 439 #ifndef TCBHASHSIZE 440 #define TCBHASHSIZE 0 441 #endif 442 443 /* 444 * XXX 445 * Callouts should be moved into struct tcp directly. They are currently 446 * separate because the tcpcb structure is exported to userland for sysctl 447 * parsing purposes, which do not know about callouts. 448 */ 449 struct tcpcb_mem { 450 struct tcpcb tcb; 451 struct tcp_timer tt; 452 struct cc_var ccv; 453 struct osd osd; 454 }; 455 456 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 457 #define V_tcpcb_zone VNET(tcpcb_zone) 458 459 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 460 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 461 462 static struct mtx isn_mtx; 463 464 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 465 #define ISN_LOCK() mtx_lock(&isn_mtx) 466 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 467 468 /* 469 * TCP initialization. 470 */ 471 static void 472 tcp_zone_change(void *tag) 473 { 474 475 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 476 uma_zone_set_max(V_tcpcb_zone, maxsockets); 477 tcp_tw_zone_change(); 478 } 479 480 static int 481 tcp_inpcb_init(void *mem, int size, int flags) 482 { 483 struct inpcb *inp = mem; 484 485 INP_LOCK_INIT(inp, "inp", "tcpinp"); 486 return (0); 487 } 488 489 /* 490 * Take a value and get the next power of 2 that doesn't overflow. 491 * Used to size the tcp_inpcb hash buckets. 492 */ 493 static int 494 maketcp_hashsize(int size) 495 { 496 int hashsize; 497 498 /* 499 * auto tune. 500 * get the next power of 2 higher than maxsockets. 501 */ 502 hashsize = 1 << fls(size); 503 /* catch overflow, and just go one power of 2 smaller */ 504 if (hashsize < size) { 505 hashsize = 1 << (fls(size) - 1); 506 } 507 return (hashsize); 508 } 509 510 int 511 register_tcp_functions(struct tcp_function_block *blk, int wait) 512 { 513 struct tcp_function_block *lblk; 514 struct tcp_function *n; 515 struct tcp_function_set fs; 516 517 if (t_functions_inited == 0) { 518 init_tcp_functions(); 519 } 520 if ((blk->tfb_tcp_output == NULL) || 521 (blk->tfb_tcp_do_segment == NULL) || 522 (blk->tfb_tcp_ctloutput == NULL) || 523 (strlen(blk->tfb_tcp_block_name) == 0)) { 524 /* 525 * These functions are required and you 526 * need a name. 527 */ 528 return (EINVAL); 529 } 530 if (blk->tfb_tcp_timer_stop_all || 531 blk->tfb_tcp_timers_left || 532 blk->tfb_tcp_timer_activate || 533 blk->tfb_tcp_timer_active || 534 blk->tfb_tcp_timer_stop) { 535 /* 536 * If you define one timer function you 537 * must have them all. 538 */ 539 if ((blk->tfb_tcp_timer_stop_all == NULL) || 540 (blk->tfb_tcp_timers_left == NULL) || 541 (blk->tfb_tcp_timer_activate == NULL) || 542 (blk->tfb_tcp_timer_active == NULL) || 543 (blk->tfb_tcp_timer_stop == NULL)) { 544 return (EINVAL); 545 } 546 } 547 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 548 if (n == NULL) { 549 return (ENOMEM); 550 } 551 n->tf_fb = blk; 552 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 553 rw_wlock(&tcp_function_lock); 554 lblk = find_tcp_functions_locked(&fs); 555 if (lblk) { 556 /* Duplicate name space not allowed */ 557 rw_wunlock(&tcp_function_lock); 558 free(n, M_TCPFUNCTIONS); 559 return (EALREADY); 560 } 561 refcount_init(&blk->tfb_refcnt, 0); 562 blk->tfb_flags = 0; 563 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 564 rw_wunlock(&tcp_function_lock); 565 return(0); 566 } 567 568 int 569 deregister_tcp_functions(struct tcp_function_block *blk) 570 { 571 struct tcp_function_block *lblk; 572 struct tcp_function *f; 573 int error=ENOENT; 574 575 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 576 /* You can't un-register the default */ 577 return (EPERM); 578 } 579 rw_wlock(&tcp_function_lock); 580 if (blk == tcp_func_set_ptr) { 581 /* You can't free the current default */ 582 rw_wunlock(&tcp_function_lock); 583 return (EBUSY); 584 } 585 if (blk->tfb_refcnt) { 586 /* Still tcb attached, mark it. */ 587 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 588 rw_wunlock(&tcp_function_lock); 589 return (EBUSY); 590 } 591 lblk = find_tcp_fb_locked(blk, &f); 592 if (lblk) { 593 /* Found */ 594 TAILQ_REMOVE(&t_functions, f, tf_next); 595 f->tf_fb = NULL; 596 free(f, M_TCPFUNCTIONS); 597 error = 0; 598 } 599 rw_wunlock(&tcp_function_lock); 600 return (error); 601 } 602 603 void 604 tcp_init(void) 605 { 606 const char *tcbhash_tuneable; 607 int hashsize; 608 609 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 610 611 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 612 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 613 printf("%s: WARNING: unable to register helper hook\n", __func__); 614 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 615 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 616 printf("%s: WARNING: unable to register helper hook\n", __func__); 617 hashsize = TCBHASHSIZE; 618 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 619 if (hashsize == 0) { 620 /* 621 * Auto tune the hash size based on maxsockets. 622 * A perfect hash would have a 1:1 mapping 623 * (hashsize = maxsockets) however it's been 624 * suggested that O(2) average is better. 625 */ 626 hashsize = maketcp_hashsize(maxsockets / 4); 627 /* 628 * Our historical default is 512, 629 * do not autotune lower than this. 630 */ 631 if (hashsize < 512) 632 hashsize = 512; 633 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 634 printf("%s: %s auto tuned to %d\n", __func__, 635 tcbhash_tuneable, hashsize); 636 } 637 /* 638 * We require a hashsize to be a power of two. 639 * Previously if it was not a power of two we would just reset it 640 * back to 512, which could be a nasty surprise if you did not notice 641 * the error message. 642 * Instead what we do is clip it to the closest power of two lower 643 * than the specified hash value. 644 */ 645 if (!powerof2(hashsize)) { 646 int oldhashsize = hashsize; 647 648 hashsize = maketcp_hashsize(hashsize); 649 /* prevent absurdly low value */ 650 if (hashsize < 16) 651 hashsize = 16; 652 printf("%s: WARNING: TCB hash size not a power of 2, " 653 "clipped from %d to %d.\n", __func__, oldhashsize, 654 hashsize); 655 } 656 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 657 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE, 658 IPI_HASHFIELDS_4TUPLE); 659 660 /* 661 * These have to be type stable for the benefit of the timers. 662 */ 663 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 664 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 665 uma_zone_set_max(V_tcpcb_zone, maxsockets); 666 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 667 668 tcp_tw_init(); 669 syncache_init(); 670 tcp_hc_init(); 671 672 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 673 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 674 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 675 676 /* Skip initialization of globals for non-default instances. */ 677 if (!IS_DEFAULT_VNET(curvnet)) 678 return; 679 680 tcp_reass_global_init(); 681 682 /* XXX virtualize those bellow? */ 683 tcp_delacktime = TCPTV_DELACK; 684 tcp_keepinit = TCPTV_KEEP_INIT; 685 tcp_keepidle = TCPTV_KEEP_IDLE; 686 tcp_keepintvl = TCPTV_KEEPINTVL; 687 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 688 tcp_msl = TCPTV_MSL; 689 tcp_rexmit_min = TCPTV_MIN; 690 if (tcp_rexmit_min < 1) 691 tcp_rexmit_min = 1; 692 tcp_persmin = TCPTV_PERSMIN; 693 tcp_persmax = TCPTV_PERSMAX; 694 tcp_rexmit_slop = TCPTV_CPU_VAR; 695 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 696 tcp_tcbhashsize = hashsize; 697 /* Setup the tcp function block list */ 698 init_tcp_functions(); 699 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 700 701 if (tcp_soreceive_stream) { 702 #ifdef INET 703 tcp_usrreqs.pru_soreceive = soreceive_stream; 704 #endif 705 #ifdef INET6 706 tcp6_usrreqs.pru_soreceive = soreceive_stream; 707 #endif /* INET6 */ 708 } 709 710 #ifdef INET6 711 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 712 #else /* INET6 */ 713 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 714 #endif /* INET6 */ 715 if (max_protohdr < TCP_MINPROTOHDR) 716 max_protohdr = TCP_MINPROTOHDR; 717 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 718 panic("tcp_init"); 719 #undef TCP_MINPROTOHDR 720 721 ISN_LOCK_INIT(); 722 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 723 SHUTDOWN_PRI_DEFAULT); 724 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 725 EVENTHANDLER_PRI_ANY); 726 #ifdef TCPPCAP 727 tcp_pcap_init(); 728 #endif 729 730 #ifdef TCP_RFC7413 731 tcp_fastopen_init(); 732 #endif 733 } 734 735 #ifdef VIMAGE 736 void 737 tcp_destroy(void) 738 { 739 int error; 740 741 #ifdef TCP_RFC7413 742 tcp_fastopen_destroy(); 743 #endif 744 tcp_hc_destroy(); 745 syncache_destroy(); 746 tcp_tw_destroy(); 747 in_pcbinfo_destroy(&V_tcbinfo); 748 uma_zdestroy(V_sack_hole_zone); 749 uma_zdestroy(V_tcpcb_zone); 750 751 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 752 if (error != 0) { 753 printf("%s: WARNING: unable to deregister helper hook " 754 "type=%d, id=%d: error %d returned\n", __func__, 755 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 756 } 757 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 758 if (error != 0) { 759 printf("%s: WARNING: unable to deregister helper hook " 760 "type=%d, id=%d: error %d returned\n", __func__, 761 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 762 } 763 } 764 #endif 765 766 void 767 tcp_fini(void *xtp) 768 { 769 770 } 771 772 /* 773 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 774 * tcp_template used to store this data in mbufs, but we now recopy it out 775 * of the tcpcb each time to conserve mbufs. 776 */ 777 void 778 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 779 { 780 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 781 782 INP_WLOCK_ASSERT(inp); 783 784 #ifdef INET6 785 if ((inp->inp_vflag & INP_IPV6) != 0) { 786 struct ip6_hdr *ip6; 787 788 ip6 = (struct ip6_hdr *)ip_ptr; 789 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 790 (inp->inp_flow & IPV6_FLOWINFO_MASK); 791 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 792 (IPV6_VERSION & IPV6_VERSION_MASK); 793 ip6->ip6_nxt = IPPROTO_TCP; 794 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 795 ip6->ip6_src = inp->in6p_laddr; 796 ip6->ip6_dst = inp->in6p_faddr; 797 } 798 #endif /* INET6 */ 799 #if defined(INET6) && defined(INET) 800 else 801 #endif 802 #ifdef INET 803 { 804 struct ip *ip; 805 806 ip = (struct ip *)ip_ptr; 807 ip->ip_v = IPVERSION; 808 ip->ip_hl = 5; 809 ip->ip_tos = inp->inp_ip_tos; 810 ip->ip_len = 0; 811 ip->ip_id = 0; 812 ip->ip_off = 0; 813 ip->ip_ttl = inp->inp_ip_ttl; 814 ip->ip_sum = 0; 815 ip->ip_p = IPPROTO_TCP; 816 ip->ip_src = inp->inp_laddr; 817 ip->ip_dst = inp->inp_faddr; 818 } 819 #endif /* INET */ 820 th->th_sport = inp->inp_lport; 821 th->th_dport = inp->inp_fport; 822 th->th_seq = 0; 823 th->th_ack = 0; 824 th->th_x2 = 0; 825 th->th_off = 5; 826 th->th_flags = 0; 827 th->th_win = 0; 828 th->th_urp = 0; 829 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 830 } 831 832 /* 833 * Create template to be used to send tcp packets on a connection. 834 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 835 * use for this function is in keepalives, which use tcp_respond. 836 */ 837 struct tcptemp * 838 tcpip_maketemplate(struct inpcb *inp) 839 { 840 struct tcptemp *t; 841 842 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 843 if (t == NULL) 844 return (NULL); 845 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 846 return (t); 847 } 848 849 /* 850 * Send a single message to the TCP at address specified by 851 * the given TCP/IP header. If m == NULL, then we make a copy 852 * of the tcpiphdr at th and send directly to the addressed host. 853 * This is used to force keep alive messages out using the TCP 854 * template for a connection. If flags are given then we send 855 * a message back to the TCP which originated the segment th, 856 * and discard the mbuf containing it and any other attached mbufs. 857 * 858 * In any case the ack and sequence number of the transmitted 859 * segment are as specified by the parameters. 860 * 861 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 862 */ 863 void 864 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 865 tcp_seq ack, tcp_seq seq, int flags) 866 { 867 int tlen; 868 int win = 0; 869 struct ip *ip; 870 struct tcphdr *nth; 871 #ifdef INET6 872 struct ip6_hdr *ip6; 873 int isipv6; 874 #endif /* INET6 */ 875 int ipflags = 0; 876 struct inpcb *inp; 877 878 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 879 880 #ifdef INET6 881 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 882 ip6 = ipgen; 883 #endif /* INET6 */ 884 ip = ipgen; 885 886 if (tp != NULL) { 887 inp = tp->t_inpcb; 888 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 889 INP_WLOCK_ASSERT(inp); 890 } else 891 inp = NULL; 892 893 if (tp != NULL) { 894 if (!(flags & TH_RST)) { 895 win = sbspace(&inp->inp_socket->so_rcv); 896 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 897 win = (long)TCP_MAXWIN << tp->rcv_scale; 898 } 899 } 900 if (m == NULL) { 901 m = m_gethdr(M_NOWAIT, MT_DATA); 902 if (m == NULL) 903 return; 904 tlen = 0; 905 m->m_data += max_linkhdr; 906 #ifdef INET6 907 if (isipv6) { 908 bcopy((caddr_t)ip6, mtod(m, caddr_t), 909 sizeof(struct ip6_hdr)); 910 ip6 = mtod(m, struct ip6_hdr *); 911 nth = (struct tcphdr *)(ip6 + 1); 912 } else 913 #endif /* INET6 */ 914 { 915 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 916 ip = mtod(m, struct ip *); 917 nth = (struct tcphdr *)(ip + 1); 918 } 919 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 920 flags = TH_ACK; 921 } else { 922 /* 923 * reuse the mbuf. 924 * XXX MRT We inherrit the FIB, which is lucky. 925 */ 926 m_freem(m->m_next); 927 m->m_next = NULL; 928 m->m_data = (caddr_t)ipgen; 929 /* m_len is set later */ 930 tlen = 0; 931 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 932 #ifdef INET6 933 if (isipv6) { 934 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 935 nth = (struct tcphdr *)(ip6 + 1); 936 } else 937 #endif /* INET6 */ 938 { 939 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 940 nth = (struct tcphdr *)(ip + 1); 941 } 942 if (th != nth) { 943 /* 944 * this is usually a case when an extension header 945 * exists between the IPv6 header and the 946 * TCP header. 947 */ 948 nth->th_sport = th->th_sport; 949 nth->th_dport = th->th_dport; 950 } 951 xchg(nth->th_dport, nth->th_sport, uint16_t); 952 #undef xchg 953 } 954 #ifdef INET6 955 if (isipv6) { 956 ip6->ip6_flow = 0; 957 ip6->ip6_vfc = IPV6_VERSION; 958 ip6->ip6_nxt = IPPROTO_TCP; 959 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 960 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 961 } 962 #endif 963 #if defined(INET) && defined(INET6) 964 else 965 #endif 966 #ifdef INET 967 { 968 tlen += sizeof (struct tcpiphdr); 969 ip->ip_len = htons(tlen); 970 ip->ip_ttl = V_ip_defttl; 971 if (V_path_mtu_discovery) 972 ip->ip_off |= htons(IP_DF); 973 } 974 #endif 975 m->m_len = tlen; 976 m->m_pkthdr.len = tlen; 977 m->m_pkthdr.rcvif = NULL; 978 #ifdef MAC 979 if (inp != NULL) { 980 /* 981 * Packet is associated with a socket, so allow the 982 * label of the response to reflect the socket label. 983 */ 984 INP_WLOCK_ASSERT(inp); 985 mac_inpcb_create_mbuf(inp, m); 986 } else { 987 /* 988 * Packet is not associated with a socket, so possibly 989 * update the label in place. 990 */ 991 mac_netinet_tcp_reply(m); 992 } 993 #endif 994 nth->th_seq = htonl(seq); 995 nth->th_ack = htonl(ack); 996 nth->th_x2 = 0; 997 nth->th_off = sizeof (struct tcphdr) >> 2; 998 nth->th_flags = flags; 999 if (tp != NULL) 1000 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1001 else 1002 nth->th_win = htons((u_short)win); 1003 nth->th_urp = 0; 1004 1005 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1006 #ifdef INET6 1007 if (isipv6) { 1008 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1009 nth->th_sum = in6_cksum_pseudo(ip6, 1010 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1011 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1012 NULL, NULL); 1013 } 1014 #endif /* INET6 */ 1015 #if defined(INET6) && defined(INET) 1016 else 1017 #endif 1018 #ifdef INET 1019 { 1020 m->m_pkthdr.csum_flags = CSUM_TCP; 1021 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1022 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1023 } 1024 #endif /* INET */ 1025 #ifdef TCPDEBUG 1026 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1027 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1028 #endif 1029 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1030 if (flags & TH_RST) 1031 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), 1032 tp, nth); 1033 1034 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); 1035 #ifdef INET6 1036 if (isipv6) 1037 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 1038 #endif /* INET6 */ 1039 #if defined(INET) && defined(INET6) 1040 else 1041 #endif 1042 #ifdef INET 1043 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 1044 #endif 1045 } 1046 1047 /* 1048 * Create a new TCP control block, making an 1049 * empty reassembly queue and hooking it to the argument 1050 * protocol control block. The `inp' parameter must have 1051 * come from the zone allocator set up in tcp_init(). 1052 */ 1053 struct tcpcb * 1054 tcp_newtcpcb(struct inpcb *inp) 1055 { 1056 struct tcpcb_mem *tm; 1057 struct tcpcb *tp; 1058 #ifdef INET6 1059 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1060 #endif /* INET6 */ 1061 1062 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1063 if (tm == NULL) 1064 return (NULL); 1065 tp = &tm->tcb; 1066 1067 /* Initialise cc_var struct for this tcpcb. */ 1068 tp->ccv = &tm->ccv; 1069 tp->ccv->type = IPPROTO_TCP; 1070 tp->ccv->ccvc.tcp = tp; 1071 rw_rlock(&tcp_function_lock); 1072 tp->t_fb = tcp_func_set_ptr; 1073 refcount_acquire(&tp->t_fb->tfb_refcnt); 1074 rw_runlock(&tcp_function_lock); 1075 if (tp->t_fb->tfb_tcp_fb_init) { 1076 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1077 } 1078 /* 1079 * Use the current system default CC algorithm. 1080 */ 1081 CC_LIST_RLOCK(); 1082 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1083 CC_ALGO(tp) = CC_DEFAULT(); 1084 CC_LIST_RUNLOCK(); 1085 1086 if (CC_ALGO(tp)->cb_init != NULL) 1087 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1088 if (tp->t_fb->tfb_tcp_fb_fini) 1089 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1090 refcount_release(&tp->t_fb->tfb_refcnt); 1091 uma_zfree(V_tcpcb_zone, tm); 1092 return (NULL); 1093 } 1094 1095 tp->osd = &tm->osd; 1096 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1097 if (tp->t_fb->tfb_tcp_fb_fini) 1098 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1099 refcount_release(&tp->t_fb->tfb_refcnt); 1100 uma_zfree(V_tcpcb_zone, tm); 1101 return (NULL); 1102 } 1103 1104 #ifdef VIMAGE 1105 tp->t_vnet = inp->inp_vnet; 1106 #endif 1107 tp->t_timers = &tm->tt; 1108 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1109 tp->t_maxseg = 1110 #ifdef INET6 1111 isipv6 ? V_tcp_v6mssdflt : 1112 #endif /* INET6 */ 1113 V_tcp_mssdflt; 1114 1115 /* Set up our timeouts. */ 1116 callout_init(&tp->t_timers->tt_rexmt, 1); 1117 callout_init(&tp->t_timers->tt_persist, 1); 1118 callout_init(&tp->t_timers->tt_keep, 1); 1119 callout_init(&tp->t_timers->tt_2msl, 1); 1120 callout_init(&tp->t_timers->tt_delack, 1); 1121 1122 if (V_tcp_do_rfc1323) 1123 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1124 if (V_tcp_do_sack) 1125 tp->t_flags |= TF_SACK_PERMIT; 1126 TAILQ_INIT(&tp->snd_holes); 1127 /* 1128 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1129 * is called. 1130 */ 1131 in_pcbref(inp); /* Reference for tcpcb */ 1132 tp->t_inpcb = inp; 1133 1134 /* 1135 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1136 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1137 * reasonable initial retransmit time. 1138 */ 1139 tp->t_srtt = TCPTV_SRTTBASE; 1140 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1141 tp->t_rttmin = tcp_rexmit_min; 1142 tp->t_rxtcur = TCPTV_RTOBASE; 1143 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1144 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1145 tp->t_rcvtime = ticks; 1146 /* 1147 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1148 * because the socket may be bound to an IPv6 wildcard address, 1149 * which may match an IPv4-mapped IPv6 address. 1150 */ 1151 inp->inp_ip_ttl = V_ip_defttl; 1152 inp->inp_ppcb = tp; 1153 #ifdef TCPPCAP 1154 /* 1155 * Init the TCP PCAP queues. 1156 */ 1157 tcp_pcap_tcpcb_init(tp); 1158 #endif 1159 return (tp); /* XXX */ 1160 } 1161 1162 /* 1163 * Switch the congestion control algorithm back to NewReno for any active 1164 * control blocks using an algorithm which is about to go away. 1165 * This ensures the CC framework can allow the unload to proceed without leaving 1166 * any dangling pointers which would trigger a panic. 1167 * Returning non-zero would inform the CC framework that something went wrong 1168 * and it would be unsafe to allow the unload to proceed. However, there is no 1169 * way for this to occur with this implementation so we always return zero. 1170 */ 1171 int 1172 tcp_ccalgounload(struct cc_algo *unload_algo) 1173 { 1174 struct cc_algo *tmpalgo; 1175 struct inpcb *inp; 1176 struct tcpcb *tp; 1177 VNET_ITERATOR_DECL(vnet_iter); 1178 1179 /* 1180 * Check all active control blocks across all network stacks and change 1181 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1182 * requires cleanup code to be run, call it. 1183 */ 1184 VNET_LIST_RLOCK(); 1185 VNET_FOREACH(vnet_iter) { 1186 CURVNET_SET(vnet_iter); 1187 INP_INFO_WLOCK(&V_tcbinfo); 1188 /* 1189 * New connections already part way through being initialised 1190 * with the CC algo we're removing will not race with this code 1191 * because the INP_INFO_WLOCK is held during initialisation. We 1192 * therefore don't enter the loop below until the connection 1193 * list has stabilised. 1194 */ 1195 LIST_FOREACH(inp, &V_tcb, inp_list) { 1196 INP_WLOCK(inp); 1197 /* Important to skip tcptw structs. */ 1198 if (!(inp->inp_flags & INP_TIMEWAIT) && 1199 (tp = intotcpcb(inp)) != NULL) { 1200 /* 1201 * By holding INP_WLOCK here, we are assured 1202 * that the connection is not currently 1203 * executing inside the CC module's functions 1204 * i.e. it is safe to make the switch back to 1205 * NewReno. 1206 */ 1207 if (CC_ALGO(tp) == unload_algo) { 1208 tmpalgo = CC_ALGO(tp); 1209 /* NewReno does not require any init. */ 1210 CC_ALGO(tp) = &newreno_cc_algo; 1211 if (tmpalgo->cb_destroy != NULL) 1212 tmpalgo->cb_destroy(tp->ccv); 1213 } 1214 } 1215 INP_WUNLOCK(inp); 1216 } 1217 INP_INFO_WUNLOCK(&V_tcbinfo); 1218 CURVNET_RESTORE(); 1219 } 1220 VNET_LIST_RUNLOCK(); 1221 1222 return (0); 1223 } 1224 1225 /* 1226 * Drop a TCP connection, reporting 1227 * the specified error. If connection is synchronized, 1228 * then send a RST to peer. 1229 */ 1230 struct tcpcb * 1231 tcp_drop(struct tcpcb *tp, int errno) 1232 { 1233 struct socket *so = tp->t_inpcb->inp_socket; 1234 1235 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1236 INP_WLOCK_ASSERT(tp->t_inpcb); 1237 1238 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1239 tcp_state_change(tp, TCPS_CLOSED); 1240 (void) tp->t_fb->tfb_tcp_output(tp); 1241 TCPSTAT_INC(tcps_drops); 1242 } else 1243 TCPSTAT_INC(tcps_conndrops); 1244 if (errno == ETIMEDOUT && tp->t_softerror) 1245 errno = tp->t_softerror; 1246 so->so_error = errno; 1247 return (tcp_close(tp)); 1248 } 1249 1250 void 1251 tcp_discardcb(struct tcpcb *tp) 1252 { 1253 struct inpcb *inp = tp->t_inpcb; 1254 struct socket *so = inp->inp_socket; 1255 #ifdef INET6 1256 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1257 #endif /* INET6 */ 1258 int released; 1259 1260 INP_WLOCK_ASSERT(inp); 1261 1262 /* 1263 * Make sure that all of our timers are stopped before we delete the 1264 * PCB. 1265 * 1266 * If stopping a timer fails, we schedule a discard function in same 1267 * callout, and the last discard function called will take care of 1268 * deleting the tcpcb. 1269 */ 1270 tcp_timer_stop(tp, TT_REXMT); 1271 tcp_timer_stop(tp, TT_PERSIST); 1272 tcp_timer_stop(tp, TT_KEEP); 1273 tcp_timer_stop(tp, TT_2MSL); 1274 tcp_timer_stop(tp, TT_DELACK); 1275 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1276 /* Call the stop-all function of the methods */ 1277 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1278 } 1279 1280 /* 1281 * If we got enough samples through the srtt filter, 1282 * save the rtt and rttvar in the routing entry. 1283 * 'Enough' is arbitrarily defined as 4 rtt samples. 1284 * 4 samples is enough for the srtt filter to converge 1285 * to within enough % of the correct value; fewer samples 1286 * and we could save a bogus rtt. The danger is not high 1287 * as tcp quickly recovers from everything. 1288 * XXX: Works very well but needs some more statistics! 1289 */ 1290 if (tp->t_rttupdated >= 4) { 1291 struct hc_metrics_lite metrics; 1292 u_long ssthresh; 1293 1294 bzero(&metrics, sizeof(metrics)); 1295 /* 1296 * Update the ssthresh always when the conditions below 1297 * are satisfied. This gives us better new start value 1298 * for the congestion avoidance for new connections. 1299 * ssthresh is only set if packet loss occured on a session. 1300 * 1301 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1302 * being torn down. Ideally this code would not use 'so'. 1303 */ 1304 ssthresh = tp->snd_ssthresh; 1305 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1306 /* 1307 * convert the limit from user data bytes to 1308 * packets then to packet data bytes. 1309 */ 1310 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1311 if (ssthresh < 2) 1312 ssthresh = 2; 1313 ssthresh *= (u_long)(tp->t_maxseg + 1314 #ifdef INET6 1315 (isipv6 ? sizeof (struct ip6_hdr) + 1316 sizeof (struct tcphdr) : 1317 #endif 1318 sizeof (struct tcpiphdr) 1319 #ifdef INET6 1320 ) 1321 #endif 1322 ); 1323 } else 1324 ssthresh = 0; 1325 metrics.rmx_ssthresh = ssthresh; 1326 1327 metrics.rmx_rtt = tp->t_srtt; 1328 metrics.rmx_rttvar = tp->t_rttvar; 1329 metrics.rmx_cwnd = tp->snd_cwnd; 1330 metrics.rmx_sendpipe = 0; 1331 metrics.rmx_recvpipe = 0; 1332 1333 tcp_hc_update(&inp->inp_inc, &metrics); 1334 } 1335 1336 /* free the reassembly queue, if any */ 1337 tcp_reass_flush(tp); 1338 1339 #ifdef TCP_OFFLOAD 1340 /* Disconnect offload device, if any. */ 1341 if (tp->t_flags & TF_TOE) 1342 tcp_offload_detach(tp); 1343 #endif 1344 1345 tcp_free_sackholes(tp); 1346 1347 #ifdef TCPPCAP 1348 /* Free the TCP PCAP queues. */ 1349 tcp_pcap_drain(&(tp->t_inpkts)); 1350 tcp_pcap_drain(&(tp->t_outpkts)); 1351 #endif 1352 1353 /* Allow the CC algorithm to clean up after itself. */ 1354 if (CC_ALGO(tp)->cb_destroy != NULL) 1355 CC_ALGO(tp)->cb_destroy(tp->ccv); 1356 1357 khelp_destroy_osd(tp->osd); 1358 1359 CC_ALGO(tp) = NULL; 1360 inp->inp_ppcb = NULL; 1361 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1362 /* We own the last reference on tcpcb, let's free it. */ 1363 if ((tp->t_fb->tfb_tcp_timers_left) && 1364 (tp->t_fb->tfb_tcp_timers_left(tp))) { 1365 /* Some fb timers left running! */ 1366 return; 1367 } 1368 if (tp->t_fb->tfb_tcp_fb_fini) 1369 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1370 refcount_release(&tp->t_fb->tfb_refcnt); 1371 tp->t_inpcb = NULL; 1372 uma_zfree(V_tcpcb_zone, tp); 1373 released = in_pcbrele_wlocked(inp); 1374 KASSERT(!released, ("%s: inp %p should not have been released " 1375 "here", __func__, inp)); 1376 } 1377 } 1378 1379 void 1380 tcp_timer_2msl_discard(void *xtp) 1381 { 1382 1383 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL); 1384 } 1385 1386 void 1387 tcp_timer_keep_discard(void *xtp) 1388 { 1389 1390 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP); 1391 } 1392 1393 void 1394 tcp_timer_persist_discard(void *xtp) 1395 { 1396 1397 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST); 1398 } 1399 1400 void 1401 tcp_timer_rexmt_discard(void *xtp) 1402 { 1403 1404 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT); 1405 } 1406 1407 void 1408 tcp_timer_delack_discard(void *xtp) 1409 { 1410 1411 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK); 1412 } 1413 1414 void 1415 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type) 1416 { 1417 struct inpcb *inp; 1418 1419 CURVNET_SET(tp->t_vnet); 1420 INP_INFO_RLOCK(&V_tcbinfo); 1421 inp = tp->t_inpcb; 1422 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1423 __func__, tp)); 1424 INP_WLOCK(inp); 1425 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1426 ("%s: tcpcb has to be stopped here", __func__)); 1427 KASSERT((tp->t_timers->tt_flags & timer_type) != 0, 1428 ("%s: discard callout should be running", __func__)); 1429 tp->t_timers->tt_flags &= ~timer_type; 1430 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1431 /* We own the last reference on this tcpcb, let's free it. */ 1432 if ((tp->t_fb->tfb_tcp_timers_left) && 1433 (tp->t_fb->tfb_tcp_timers_left(tp))) { 1434 /* Some fb timers left running! */ 1435 goto leave; 1436 } 1437 if (tp->t_fb->tfb_tcp_fb_fini) 1438 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1439 refcount_release(&tp->t_fb->tfb_refcnt); 1440 tp->t_inpcb = NULL; 1441 uma_zfree(V_tcpcb_zone, tp); 1442 if (in_pcbrele_wlocked(inp)) { 1443 INP_INFO_RUNLOCK(&V_tcbinfo); 1444 CURVNET_RESTORE(); 1445 return; 1446 } 1447 } 1448 leave: 1449 INP_WUNLOCK(inp); 1450 INP_INFO_RUNLOCK(&V_tcbinfo); 1451 CURVNET_RESTORE(); 1452 } 1453 1454 /* 1455 * Attempt to close a TCP control block, marking it as dropped, and freeing 1456 * the socket if we hold the only reference. 1457 */ 1458 struct tcpcb * 1459 tcp_close(struct tcpcb *tp) 1460 { 1461 struct inpcb *inp = tp->t_inpcb; 1462 struct socket *so; 1463 1464 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1465 INP_WLOCK_ASSERT(inp); 1466 1467 #ifdef TCP_OFFLOAD 1468 if (tp->t_state == TCPS_LISTEN) 1469 tcp_offload_listen_stop(tp); 1470 #endif 1471 #ifdef TCP_RFC7413 1472 /* 1473 * This releases the TFO pending counter resource for TFO listen 1474 * sockets as well as passively-created TFO sockets that transition 1475 * from SYN_RECEIVED to CLOSED. 1476 */ 1477 if (tp->t_tfo_pending) { 1478 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1479 tp->t_tfo_pending = NULL; 1480 } 1481 #endif 1482 in_pcbdrop(inp); 1483 TCPSTAT_INC(tcps_closed); 1484 TCPSTAT_DEC(tcps_states[tp->t_state]); 1485 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1486 so = inp->inp_socket; 1487 soisdisconnected(so); 1488 if (inp->inp_flags & INP_SOCKREF) { 1489 KASSERT(so->so_state & SS_PROTOREF, 1490 ("tcp_close: !SS_PROTOREF")); 1491 inp->inp_flags &= ~INP_SOCKREF; 1492 INP_WUNLOCK(inp); 1493 ACCEPT_LOCK(); 1494 SOCK_LOCK(so); 1495 so->so_state &= ~SS_PROTOREF; 1496 sofree(so); 1497 return (NULL); 1498 } 1499 return (tp); 1500 } 1501 1502 void 1503 tcp_drain(void) 1504 { 1505 VNET_ITERATOR_DECL(vnet_iter); 1506 1507 if (!do_tcpdrain) 1508 return; 1509 1510 VNET_LIST_RLOCK_NOSLEEP(); 1511 VNET_FOREACH(vnet_iter) { 1512 CURVNET_SET(vnet_iter); 1513 struct inpcb *inpb; 1514 struct tcpcb *tcpb; 1515 1516 /* 1517 * Walk the tcpbs, if existing, and flush the reassembly queue, 1518 * if there is one... 1519 * XXX: The "Net/3" implementation doesn't imply that the TCP 1520 * reassembly queue should be flushed, but in a situation 1521 * where we're really low on mbufs, this is potentially 1522 * useful. 1523 */ 1524 INP_INFO_WLOCK(&V_tcbinfo); 1525 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1526 if (inpb->inp_flags & INP_TIMEWAIT) 1527 continue; 1528 INP_WLOCK(inpb); 1529 if ((tcpb = intotcpcb(inpb)) != NULL) { 1530 tcp_reass_flush(tcpb); 1531 tcp_clean_sackreport(tcpb); 1532 } 1533 INP_WUNLOCK(inpb); 1534 } 1535 INP_INFO_WUNLOCK(&V_tcbinfo); 1536 CURVNET_RESTORE(); 1537 } 1538 VNET_LIST_RUNLOCK_NOSLEEP(); 1539 } 1540 1541 /* 1542 * Notify a tcp user of an asynchronous error; 1543 * store error as soft error, but wake up user 1544 * (for now, won't do anything until can select for soft error). 1545 * 1546 * Do not wake up user since there currently is no mechanism for 1547 * reporting soft errors (yet - a kqueue filter may be added). 1548 */ 1549 static struct inpcb * 1550 tcp_notify(struct inpcb *inp, int error) 1551 { 1552 struct tcpcb *tp; 1553 1554 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1555 INP_WLOCK_ASSERT(inp); 1556 1557 if ((inp->inp_flags & INP_TIMEWAIT) || 1558 (inp->inp_flags & INP_DROPPED)) 1559 return (inp); 1560 1561 tp = intotcpcb(inp); 1562 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1563 1564 /* 1565 * Ignore some errors if we are hooked up. 1566 * If connection hasn't completed, has retransmitted several times, 1567 * and receives a second error, give up now. This is better 1568 * than waiting a long time to establish a connection that 1569 * can never complete. 1570 */ 1571 if (tp->t_state == TCPS_ESTABLISHED && 1572 (error == EHOSTUNREACH || error == ENETUNREACH || 1573 error == EHOSTDOWN)) { 1574 return (inp); 1575 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1576 tp->t_softerror) { 1577 tp = tcp_drop(tp, error); 1578 if (tp != NULL) 1579 return (inp); 1580 else 1581 return (NULL); 1582 } else { 1583 tp->t_softerror = error; 1584 return (inp); 1585 } 1586 #if 0 1587 wakeup( &so->so_timeo); 1588 sorwakeup(so); 1589 sowwakeup(so); 1590 #endif 1591 } 1592 1593 static int 1594 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1595 { 1596 int error, i, m, n, pcb_count; 1597 struct inpcb *inp, **inp_list; 1598 inp_gen_t gencnt; 1599 struct xinpgen xig; 1600 1601 /* 1602 * The process of preparing the TCB list is too time-consuming and 1603 * resource-intensive to repeat twice on every request. 1604 */ 1605 if (req->oldptr == NULL) { 1606 n = V_tcbinfo.ipi_count + 1607 TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]); 1608 n += imax(n / 8, 10); 1609 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1610 return (0); 1611 } 1612 1613 if (req->newptr != NULL) 1614 return (EPERM); 1615 1616 /* 1617 * OK, now we're committed to doing something. 1618 */ 1619 INP_LIST_RLOCK(&V_tcbinfo); 1620 gencnt = V_tcbinfo.ipi_gencnt; 1621 n = V_tcbinfo.ipi_count; 1622 INP_LIST_RUNLOCK(&V_tcbinfo); 1623 1624 m = TCPSTAT_FETCH(tcps_states[TCPS_SYN_RECEIVED]); 1625 1626 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1627 + (n + m) * sizeof(struct xtcpcb)); 1628 if (error != 0) 1629 return (error); 1630 1631 xig.xig_len = sizeof xig; 1632 xig.xig_count = n + m; 1633 xig.xig_gen = gencnt; 1634 xig.xig_sogen = so_gencnt; 1635 error = SYSCTL_OUT(req, &xig, sizeof xig); 1636 if (error) 1637 return (error); 1638 1639 error = syncache_pcblist(req, m, &pcb_count); 1640 if (error) 1641 return (error); 1642 1643 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1644 if (inp_list == NULL) 1645 return (ENOMEM); 1646 1647 INP_INFO_WLOCK(&V_tcbinfo); 1648 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1649 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1650 INP_WLOCK(inp); 1651 if (inp->inp_gencnt <= gencnt) { 1652 /* 1653 * XXX: This use of cr_cansee(), introduced with 1654 * TCP state changes, is not quite right, but for 1655 * now, better than nothing. 1656 */ 1657 if (inp->inp_flags & INP_TIMEWAIT) { 1658 if (intotw(inp) != NULL) 1659 error = cr_cansee(req->td->td_ucred, 1660 intotw(inp)->tw_cred); 1661 else 1662 error = EINVAL; /* Skip this inp. */ 1663 } else 1664 error = cr_canseeinpcb(req->td->td_ucred, inp); 1665 if (error == 0) { 1666 in_pcbref(inp); 1667 inp_list[i++] = inp; 1668 } 1669 } 1670 INP_WUNLOCK(inp); 1671 } 1672 INP_INFO_WUNLOCK(&V_tcbinfo); 1673 n = i; 1674 1675 error = 0; 1676 for (i = 0; i < n; i++) { 1677 inp = inp_list[i]; 1678 INP_RLOCK(inp); 1679 if (inp->inp_gencnt <= gencnt) { 1680 struct xtcpcb xt; 1681 void *inp_ppcb; 1682 1683 bzero(&xt, sizeof(xt)); 1684 xt.xt_len = sizeof xt; 1685 /* XXX should avoid extra copy */ 1686 bcopy(inp, &xt.xt_inp, sizeof *inp); 1687 inp_ppcb = inp->inp_ppcb; 1688 if (inp_ppcb == NULL) 1689 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1690 else if (inp->inp_flags & INP_TIMEWAIT) { 1691 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1692 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1693 } else { 1694 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1695 if (xt.xt_tp.t_timers) 1696 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1697 } 1698 if (inp->inp_socket != NULL) 1699 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1700 else { 1701 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1702 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1703 } 1704 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1705 INP_RUNLOCK(inp); 1706 error = SYSCTL_OUT(req, &xt, sizeof xt); 1707 } else 1708 INP_RUNLOCK(inp); 1709 } 1710 INP_INFO_RLOCK(&V_tcbinfo); 1711 for (i = 0; i < n; i++) { 1712 inp = inp_list[i]; 1713 INP_RLOCK(inp); 1714 if (!in_pcbrele_rlocked(inp)) 1715 INP_RUNLOCK(inp); 1716 } 1717 INP_INFO_RUNLOCK(&V_tcbinfo); 1718 1719 if (!error) { 1720 /* 1721 * Give the user an updated idea of our state. 1722 * If the generation differs from what we told 1723 * her before, she knows that something happened 1724 * while we were processing this request, and it 1725 * might be necessary to retry. 1726 */ 1727 INP_LIST_RLOCK(&V_tcbinfo); 1728 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1729 xig.xig_sogen = so_gencnt; 1730 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1731 INP_LIST_RUNLOCK(&V_tcbinfo); 1732 error = SYSCTL_OUT(req, &xig, sizeof xig); 1733 } 1734 free(inp_list, M_TEMP); 1735 return (error); 1736 } 1737 1738 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1739 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1740 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1741 1742 #ifdef INET 1743 static int 1744 tcp_getcred(SYSCTL_HANDLER_ARGS) 1745 { 1746 struct xucred xuc; 1747 struct sockaddr_in addrs[2]; 1748 struct inpcb *inp; 1749 int error; 1750 1751 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1752 if (error) 1753 return (error); 1754 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1755 if (error) 1756 return (error); 1757 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1758 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1759 if (inp != NULL) { 1760 if (inp->inp_socket == NULL) 1761 error = ENOENT; 1762 if (error == 0) 1763 error = cr_canseeinpcb(req->td->td_ucred, inp); 1764 if (error == 0) 1765 cru2x(inp->inp_cred, &xuc); 1766 INP_RUNLOCK(inp); 1767 } else 1768 error = ENOENT; 1769 if (error == 0) 1770 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1771 return (error); 1772 } 1773 1774 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1775 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1776 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1777 #endif /* INET */ 1778 1779 #ifdef INET6 1780 static int 1781 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1782 { 1783 struct xucred xuc; 1784 struct sockaddr_in6 addrs[2]; 1785 struct inpcb *inp; 1786 int error; 1787 #ifdef INET 1788 int mapped = 0; 1789 #endif 1790 1791 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1792 if (error) 1793 return (error); 1794 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1795 if (error) 1796 return (error); 1797 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1798 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1799 return (error); 1800 } 1801 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1802 #ifdef INET 1803 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1804 mapped = 1; 1805 else 1806 #endif 1807 return (EINVAL); 1808 } 1809 1810 #ifdef INET 1811 if (mapped == 1) 1812 inp = in_pcblookup(&V_tcbinfo, 1813 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1814 addrs[1].sin6_port, 1815 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1816 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1817 else 1818 #endif 1819 inp = in6_pcblookup(&V_tcbinfo, 1820 &addrs[1].sin6_addr, addrs[1].sin6_port, 1821 &addrs[0].sin6_addr, addrs[0].sin6_port, 1822 INPLOOKUP_RLOCKPCB, NULL); 1823 if (inp != NULL) { 1824 if (inp->inp_socket == NULL) 1825 error = ENOENT; 1826 if (error == 0) 1827 error = cr_canseeinpcb(req->td->td_ucred, inp); 1828 if (error == 0) 1829 cru2x(inp->inp_cred, &xuc); 1830 INP_RUNLOCK(inp); 1831 } else 1832 error = ENOENT; 1833 if (error == 0) 1834 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1835 return (error); 1836 } 1837 1838 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1839 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1840 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1841 #endif /* INET6 */ 1842 1843 1844 #ifdef INET 1845 void 1846 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1847 { 1848 struct ip *ip = vip; 1849 struct tcphdr *th; 1850 struct in_addr faddr; 1851 struct inpcb *inp; 1852 struct tcpcb *tp; 1853 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1854 struct icmp *icp; 1855 struct in_conninfo inc; 1856 tcp_seq icmp_tcp_seq; 1857 int mtu; 1858 1859 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1860 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1861 return; 1862 1863 if (cmd == PRC_MSGSIZE) 1864 notify = tcp_mtudisc_notify; 1865 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1866 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1867 notify = tcp_drop_syn_sent; 1868 /* 1869 * Redirects don't need to be handled up here. 1870 */ 1871 else if (PRC_IS_REDIRECT(cmd)) 1872 return; 1873 /* 1874 * Hostdead is ugly because it goes linearly through all PCBs. 1875 * XXX: We never get this from ICMP, otherwise it makes an 1876 * excellent DoS attack on machines with many connections. 1877 */ 1878 else if (cmd == PRC_HOSTDEAD) 1879 ip = NULL; 1880 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1881 return; 1882 1883 if (ip == NULL) { 1884 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1885 return; 1886 } 1887 1888 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1889 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1890 INP_INFO_RLOCK(&V_tcbinfo); 1891 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 1892 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1893 if (inp != NULL) { 1894 if (!(inp->inp_flags & INP_TIMEWAIT) && 1895 !(inp->inp_flags & INP_DROPPED) && 1896 !(inp->inp_socket == NULL)) { 1897 icmp_tcp_seq = ntohl(th->th_seq); 1898 tp = intotcpcb(inp); 1899 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1900 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1901 if (cmd == PRC_MSGSIZE) { 1902 /* 1903 * MTU discovery: 1904 * If we got a needfrag set the MTU 1905 * in the route to the suggested new 1906 * value (if given) and then notify. 1907 */ 1908 mtu = ntohs(icp->icmp_nextmtu); 1909 /* 1910 * If no alternative MTU was 1911 * proposed, try the next smaller 1912 * one. 1913 */ 1914 if (!mtu) 1915 mtu = ip_next_mtu( 1916 ntohs(ip->ip_len), 1); 1917 if (mtu < V_tcp_minmss + 1918 sizeof(struct tcpiphdr)) 1919 mtu = V_tcp_minmss + 1920 sizeof(struct tcpiphdr); 1921 /* 1922 * Only process the offered MTU if it 1923 * is smaller than the current one. 1924 */ 1925 if (mtu < tp->t_maxseg + 1926 sizeof(struct tcpiphdr)) { 1927 bzero(&inc, sizeof(inc)); 1928 inc.inc_faddr = faddr; 1929 inc.inc_fibnum = 1930 inp->inp_inc.inc_fibnum; 1931 tcp_hc_updatemtu(&inc, mtu); 1932 tcp_mtudisc(inp, mtu); 1933 } 1934 } else 1935 inp = (*notify)(inp, 1936 inetctlerrmap[cmd]); 1937 } 1938 } 1939 if (inp != NULL) 1940 INP_WUNLOCK(inp); 1941 } else { 1942 bzero(&inc, sizeof(inc)); 1943 inc.inc_fport = th->th_dport; 1944 inc.inc_lport = th->th_sport; 1945 inc.inc_faddr = faddr; 1946 inc.inc_laddr = ip->ip_src; 1947 syncache_unreach(&inc, th); 1948 } 1949 INP_INFO_RUNLOCK(&V_tcbinfo); 1950 } 1951 #endif /* INET */ 1952 1953 #ifdef INET6 1954 void 1955 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1956 { 1957 struct tcphdr th; 1958 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1959 struct ip6_hdr *ip6; 1960 struct mbuf *m; 1961 struct ip6ctlparam *ip6cp = NULL; 1962 const struct sockaddr_in6 *sa6_src = NULL; 1963 int off; 1964 struct tcp_portonly { 1965 u_int16_t th_sport; 1966 u_int16_t th_dport; 1967 } *thp; 1968 1969 if (sa->sa_family != AF_INET6 || 1970 sa->sa_len != sizeof(struct sockaddr_in6)) 1971 return; 1972 1973 if (cmd == PRC_MSGSIZE) 1974 notify = tcp_mtudisc_notify; 1975 else if (!PRC_IS_REDIRECT(cmd) && 1976 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1977 return; 1978 1979 /* if the parameter is from icmp6, decode it. */ 1980 if (d != NULL) { 1981 ip6cp = (struct ip6ctlparam *)d; 1982 m = ip6cp->ip6c_m; 1983 ip6 = ip6cp->ip6c_ip6; 1984 off = ip6cp->ip6c_off; 1985 sa6_src = ip6cp->ip6c_src; 1986 } else { 1987 m = NULL; 1988 ip6 = NULL; 1989 off = 0; /* fool gcc */ 1990 sa6_src = &sa6_any; 1991 } 1992 1993 if (ip6 != NULL) { 1994 struct in_conninfo inc; 1995 /* 1996 * XXX: We assume that when IPV6 is non NULL, 1997 * M and OFF are valid. 1998 */ 1999 2000 /* check if we can safely examine src and dst ports */ 2001 if (m->m_pkthdr.len < off + sizeof(*thp)) 2002 return; 2003 2004 bzero(&th, sizeof(th)); 2005 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 2006 2007 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 2008 (struct sockaddr *)ip6cp->ip6c_src, 2009 th.th_sport, cmd, NULL, notify); 2010 2011 bzero(&inc, sizeof(inc)); 2012 inc.inc_fport = th.th_dport; 2013 inc.inc_lport = th.th_sport; 2014 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 2015 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 2016 inc.inc_flags |= INC_ISIPV6; 2017 INP_INFO_RLOCK(&V_tcbinfo); 2018 syncache_unreach(&inc, &th); 2019 INP_INFO_RUNLOCK(&V_tcbinfo); 2020 } else 2021 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 2022 0, cmd, NULL, notify); 2023 } 2024 #endif /* INET6 */ 2025 2026 2027 /* 2028 * Following is where TCP initial sequence number generation occurs. 2029 * 2030 * There are two places where we must use initial sequence numbers: 2031 * 1. In SYN-ACK packets. 2032 * 2. In SYN packets. 2033 * 2034 * All ISNs for SYN-ACK packets are generated by the syncache. See 2035 * tcp_syncache.c for details. 2036 * 2037 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2038 * depends on this property. In addition, these ISNs should be 2039 * unguessable so as to prevent connection hijacking. To satisfy 2040 * the requirements of this situation, the algorithm outlined in 2041 * RFC 1948 is used, with only small modifications. 2042 * 2043 * Implementation details: 2044 * 2045 * Time is based off the system timer, and is corrected so that it 2046 * increases by one megabyte per second. This allows for proper 2047 * recycling on high speed LANs while still leaving over an hour 2048 * before rollover. 2049 * 2050 * As reading the *exact* system time is too expensive to be done 2051 * whenever setting up a TCP connection, we increment the time 2052 * offset in two ways. First, a small random positive increment 2053 * is added to isn_offset for each connection that is set up. 2054 * Second, the function tcp_isn_tick fires once per clock tick 2055 * and increments isn_offset as necessary so that sequence numbers 2056 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2057 * random positive increments serve only to ensure that the same 2058 * exact sequence number is never sent out twice (as could otherwise 2059 * happen when a port is recycled in less than the system tick 2060 * interval.) 2061 * 2062 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2063 * between seeding of isn_secret. This is normally set to zero, 2064 * as reseeding should not be necessary. 2065 * 2066 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2067 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2068 * general, this means holding an exclusive (write) lock. 2069 */ 2070 2071 #define ISN_BYTES_PER_SECOND 1048576 2072 #define ISN_STATIC_INCREMENT 4096 2073 #define ISN_RANDOM_INCREMENT (4096 - 1) 2074 2075 static VNET_DEFINE(u_char, isn_secret[32]); 2076 static VNET_DEFINE(int, isn_last); 2077 static VNET_DEFINE(int, isn_last_reseed); 2078 static VNET_DEFINE(u_int32_t, isn_offset); 2079 static VNET_DEFINE(u_int32_t, isn_offset_old); 2080 2081 #define V_isn_secret VNET(isn_secret) 2082 #define V_isn_last VNET(isn_last) 2083 #define V_isn_last_reseed VNET(isn_last_reseed) 2084 #define V_isn_offset VNET(isn_offset) 2085 #define V_isn_offset_old VNET(isn_offset_old) 2086 2087 tcp_seq 2088 tcp_new_isn(struct tcpcb *tp) 2089 { 2090 MD5_CTX isn_ctx; 2091 u_int32_t md5_buffer[4]; 2092 tcp_seq new_isn; 2093 u_int32_t projected_offset; 2094 2095 INP_WLOCK_ASSERT(tp->t_inpcb); 2096 2097 ISN_LOCK(); 2098 /* Seed if this is the first use, reseed if requested. */ 2099 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2100 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2101 < (u_int)ticks))) { 2102 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2103 V_isn_last_reseed = ticks; 2104 } 2105 2106 /* Compute the md5 hash and return the ISN. */ 2107 MD5Init(&isn_ctx); 2108 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2109 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2110 #ifdef INET6 2111 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2112 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2113 sizeof(struct in6_addr)); 2114 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2115 sizeof(struct in6_addr)); 2116 } else 2117 #endif 2118 { 2119 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2120 sizeof(struct in_addr)); 2121 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2122 sizeof(struct in_addr)); 2123 } 2124 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2125 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2126 new_isn = (tcp_seq) md5_buffer[0]; 2127 V_isn_offset += ISN_STATIC_INCREMENT + 2128 (arc4random() & ISN_RANDOM_INCREMENT); 2129 if (ticks != V_isn_last) { 2130 projected_offset = V_isn_offset_old + 2131 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2132 if (SEQ_GT(projected_offset, V_isn_offset)) 2133 V_isn_offset = projected_offset; 2134 V_isn_offset_old = V_isn_offset; 2135 V_isn_last = ticks; 2136 } 2137 new_isn += V_isn_offset; 2138 ISN_UNLOCK(); 2139 return (new_isn); 2140 } 2141 2142 /* 2143 * When a specific ICMP unreachable message is received and the 2144 * connection state is SYN-SENT, drop the connection. This behavior 2145 * is controlled by the icmp_may_rst sysctl. 2146 */ 2147 struct inpcb * 2148 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2149 { 2150 struct tcpcb *tp; 2151 2152 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2153 INP_WLOCK_ASSERT(inp); 2154 2155 if ((inp->inp_flags & INP_TIMEWAIT) || 2156 (inp->inp_flags & INP_DROPPED)) 2157 return (inp); 2158 2159 tp = intotcpcb(inp); 2160 if (tp->t_state != TCPS_SYN_SENT) 2161 return (inp); 2162 2163 tp = tcp_drop(tp, errno); 2164 if (tp != NULL) 2165 return (inp); 2166 else 2167 return (NULL); 2168 } 2169 2170 /* 2171 * When `need fragmentation' ICMP is received, update our idea of the MSS 2172 * based on the new value. Also nudge TCP to send something, since we 2173 * know the packet we just sent was dropped. 2174 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2175 */ 2176 static struct inpcb * 2177 tcp_mtudisc_notify(struct inpcb *inp, int error) 2178 { 2179 2180 tcp_mtudisc(inp, -1); 2181 return (inp); 2182 } 2183 2184 static void 2185 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2186 { 2187 struct tcpcb *tp; 2188 struct socket *so; 2189 2190 INP_WLOCK_ASSERT(inp); 2191 if ((inp->inp_flags & INP_TIMEWAIT) || 2192 (inp->inp_flags & INP_DROPPED)) 2193 return; 2194 2195 tp = intotcpcb(inp); 2196 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2197 2198 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2199 2200 so = inp->inp_socket; 2201 SOCKBUF_LOCK(&so->so_snd); 2202 /* If the mss is larger than the socket buffer, decrease the mss. */ 2203 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2204 tp->t_maxseg = so->so_snd.sb_hiwat; 2205 SOCKBUF_UNLOCK(&so->so_snd); 2206 2207 TCPSTAT_INC(tcps_mturesent); 2208 tp->t_rtttime = 0; 2209 tp->snd_nxt = tp->snd_una; 2210 tcp_free_sackholes(tp); 2211 tp->snd_recover = tp->snd_max; 2212 if (tp->t_flags & TF_SACK_PERMIT) 2213 EXIT_FASTRECOVERY(tp->t_flags); 2214 tp->t_fb->tfb_tcp_output(tp); 2215 } 2216 2217 #ifdef INET 2218 /* 2219 * Look-up the routing entry to the peer of this inpcb. If no route 2220 * is found and it cannot be allocated, then return 0. This routine 2221 * is called by TCP routines that access the rmx structure and by 2222 * tcp_mss_update to get the peer/interface MTU. 2223 */ 2224 u_long 2225 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2226 { 2227 struct nhop4_extended nh4; 2228 struct ifnet *ifp; 2229 u_long maxmtu = 0; 2230 2231 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2232 2233 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2234 2235 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2236 NHR_REF, 0, &nh4) != 0) 2237 return (0); 2238 2239 ifp = nh4.nh_ifp; 2240 maxmtu = nh4.nh_mtu; 2241 2242 /* Report additional interface capabilities. */ 2243 if (cap != NULL) { 2244 if (ifp->if_capenable & IFCAP_TSO4 && 2245 ifp->if_hwassist & CSUM_TSO) { 2246 cap->ifcap |= CSUM_TSO; 2247 cap->tsomax = ifp->if_hw_tsomax; 2248 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2249 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2250 } 2251 } 2252 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2253 } 2254 return (maxmtu); 2255 } 2256 #endif /* INET */ 2257 2258 #ifdef INET6 2259 u_long 2260 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2261 { 2262 struct nhop6_extended nh6; 2263 struct in6_addr dst6; 2264 uint32_t scopeid; 2265 struct ifnet *ifp; 2266 u_long maxmtu = 0; 2267 2268 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2269 2270 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2271 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2272 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2273 0, &nh6) != 0) 2274 return (0); 2275 2276 ifp = nh6.nh_ifp; 2277 maxmtu = nh6.nh_mtu; 2278 2279 /* Report additional interface capabilities. */ 2280 if (cap != NULL) { 2281 if (ifp->if_capenable & IFCAP_TSO6 && 2282 ifp->if_hwassist & CSUM_TSO) { 2283 cap->ifcap |= CSUM_TSO; 2284 cap->tsomax = ifp->if_hw_tsomax; 2285 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2286 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2287 } 2288 } 2289 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2290 } 2291 2292 return (maxmtu); 2293 } 2294 #endif /* INET6 */ 2295 2296 /* 2297 * Calculate effective SMSS per RFC5681 definition for a given TCP 2298 * connection at its current state, taking into account SACK and etc. 2299 */ 2300 u_int 2301 tcp_maxseg(const struct tcpcb *tp) 2302 { 2303 u_int optlen; 2304 2305 if (tp->t_flags & TF_NOOPT) 2306 return (tp->t_maxseg); 2307 2308 /* 2309 * Here we have a simplified code from tcp_addoptions(), 2310 * without a proper loop, and having most of paddings hardcoded. 2311 * We might make mistakes with padding here in some edge cases, 2312 * but this is harmless, since result of tcp_maxseg() is used 2313 * only in cwnd and ssthresh estimations. 2314 */ 2315 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2316 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2317 if (tp->t_flags & TF_RCVD_TSTMP) 2318 optlen = TCPOLEN_TSTAMP_APPA; 2319 else 2320 optlen = 0; 2321 #ifdef TCP_SIGNATURE 2322 if (tp->t_flags & TF_SIGNATURE) 2323 optlen += PAD(TCPOLEN_SIGNATURE); 2324 #endif 2325 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2326 optlen += TCPOLEN_SACKHDR; 2327 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2328 optlen = PAD(optlen); 2329 } 2330 } else { 2331 if (tp->t_flags & TF_REQ_TSTMP) 2332 optlen = TCPOLEN_TSTAMP_APPA; 2333 else 2334 optlen = PAD(TCPOLEN_MAXSEG); 2335 if (tp->t_flags & TF_REQ_SCALE) 2336 optlen += PAD(TCPOLEN_WINDOW); 2337 #ifdef TCP_SIGNATURE 2338 if (tp->t_flags & TF_SIGNATURE) 2339 optlen += PAD(TCPOLEN_SIGNATURE); 2340 #endif 2341 if (tp->t_flags & TF_SACK_PERMIT) 2342 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2343 } 2344 #undef PAD 2345 optlen = min(optlen, TCP_MAXOLEN); 2346 return (tp->t_maxseg - optlen); 2347 } 2348 2349 #ifdef IPSEC 2350 /* compute ESP/AH header size for TCP, including outer IP header. */ 2351 size_t 2352 ipsec_hdrsiz_tcp(struct tcpcb *tp) 2353 { 2354 struct inpcb *inp; 2355 struct mbuf *m; 2356 size_t hdrsiz; 2357 struct ip *ip; 2358 #ifdef INET6 2359 struct ip6_hdr *ip6; 2360 #endif 2361 struct tcphdr *th; 2362 2363 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || 2364 (!key_havesp(IPSEC_DIR_OUTBOUND))) 2365 return (0); 2366 m = m_gethdr(M_NOWAIT, MT_DATA); 2367 if (!m) 2368 return (0); 2369 2370 #ifdef INET6 2371 if ((inp->inp_vflag & INP_IPV6) != 0) { 2372 ip6 = mtod(m, struct ip6_hdr *); 2373 th = (struct tcphdr *)(ip6 + 1); 2374 m->m_pkthdr.len = m->m_len = 2375 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2376 tcpip_fillheaders(inp, ip6, th); 2377 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2378 } else 2379 #endif /* INET6 */ 2380 { 2381 ip = mtod(m, struct ip *); 2382 th = (struct tcphdr *)(ip + 1); 2383 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 2384 tcpip_fillheaders(inp, ip, th); 2385 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2386 } 2387 2388 m_free(m); 2389 return (hdrsiz); 2390 } 2391 #endif /* IPSEC */ 2392 2393 #ifdef TCP_SIGNATURE 2394 /* 2395 * Callback function invoked by m_apply() to digest TCP segment data 2396 * contained within an mbuf chain. 2397 */ 2398 static int 2399 tcp_signature_apply(void *fstate, void *data, u_int len) 2400 { 2401 2402 MD5Update(fstate, (u_char *)data, len); 2403 return (0); 2404 } 2405 2406 /* 2407 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2408 * search with the destination IP address, and a 'magic SPI' to be 2409 * determined by the application. This is hardcoded elsewhere to 1179 2410 */ 2411 struct secasvar * 2412 tcp_get_sav(struct mbuf *m, u_int direction) 2413 { 2414 union sockaddr_union dst; 2415 struct secasvar *sav; 2416 struct ip *ip; 2417 #ifdef INET6 2418 struct ip6_hdr *ip6; 2419 char ip6buf[INET6_ADDRSTRLEN]; 2420 #endif 2421 2422 /* Extract the destination from the IP header in the mbuf. */ 2423 bzero(&dst, sizeof(union sockaddr_union)); 2424 ip = mtod(m, struct ip *); 2425 #ifdef INET6 2426 ip6 = NULL; /* Make the compiler happy. */ 2427 #endif 2428 switch (ip->ip_v) { 2429 #ifdef INET 2430 case IPVERSION: 2431 dst.sa.sa_len = sizeof(struct sockaddr_in); 2432 dst.sa.sa_family = AF_INET; 2433 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2434 ip->ip_src : ip->ip_dst; 2435 break; 2436 #endif 2437 #ifdef INET6 2438 case (IPV6_VERSION >> 4): 2439 ip6 = mtod(m, struct ip6_hdr *); 2440 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2441 dst.sa.sa_family = AF_INET6; 2442 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 2443 ip6->ip6_src : ip6->ip6_dst; 2444 break; 2445 #endif 2446 default: 2447 return (NULL); 2448 /* NOTREACHED */ 2449 break; 2450 } 2451 2452 /* Look up an SADB entry which matches the address of the peer. */ 2453 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2454 if (sav == NULL) { 2455 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 2456 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 2457 #ifdef INET6 2458 (ip->ip_v == (IPV6_VERSION >> 4)) ? 2459 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 2460 #endif 2461 "(unsupported)")); 2462 } 2463 2464 return (sav); 2465 } 2466 2467 /* 2468 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2469 * 2470 * Parameters: 2471 * m pointer to head of mbuf chain 2472 * len length of TCP segment data, excluding options 2473 * optlen length of TCP segment options 2474 * buf pointer to storage for computed MD5 digest 2475 * sav pointer to security assosiation 2476 * 2477 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2478 * When called from tcp_input(), we can be sure that th_sum has been 2479 * zeroed out and verified already. 2480 * 2481 * Releases reference to SADB key before return. 2482 * 2483 * Return 0 if successful, otherwise return -1. 2484 * 2485 */ 2486 int 2487 tcp_signature_do_compute(struct mbuf *m, int len, int optlen, 2488 u_char *buf, struct secasvar *sav) 2489 { 2490 #ifdef INET 2491 struct ippseudo ippseudo; 2492 #endif 2493 MD5_CTX ctx; 2494 int doff; 2495 struct ip *ip; 2496 #ifdef INET 2497 struct ipovly *ipovly; 2498 #endif 2499 struct tcphdr *th; 2500 #ifdef INET6 2501 struct ip6_hdr *ip6; 2502 struct in6_addr in6; 2503 uint32_t plen; 2504 uint16_t nhdr; 2505 #endif 2506 u_short savecsum; 2507 2508 KASSERT(m != NULL, ("NULL mbuf chain")); 2509 KASSERT(buf != NULL, ("NULL signature pointer")); 2510 2511 /* Extract the destination from the IP header in the mbuf. */ 2512 ip = mtod(m, struct ip *); 2513 #ifdef INET6 2514 ip6 = NULL; /* Make the compiler happy. */ 2515 #endif 2516 2517 MD5Init(&ctx); 2518 /* 2519 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 2520 * 2521 * XXX The ippseudo header MUST be digested in network byte order, 2522 * or else we'll fail the regression test. Assume all fields we've 2523 * been doing arithmetic on have been in host byte order. 2524 * XXX One cannot depend on ipovly->ih_len here. When called from 2525 * tcp_output(), the underlying ip_len member has not yet been set. 2526 */ 2527 switch (ip->ip_v) { 2528 #ifdef INET 2529 case IPVERSION: 2530 ipovly = (struct ipovly *)ip; 2531 ippseudo.ippseudo_src = ipovly->ih_src; 2532 ippseudo.ippseudo_dst = ipovly->ih_dst; 2533 ippseudo.ippseudo_pad = 0; 2534 ippseudo.ippseudo_p = IPPROTO_TCP; 2535 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 2536 optlen); 2537 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2538 2539 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2540 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2541 break; 2542 #endif 2543 #ifdef INET6 2544 /* 2545 * RFC 2385, 2.0 Proposal 2546 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2547 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2548 * extended next header value (to form 32 bits), and 32-bit segment 2549 * length. 2550 * Note: Upper-Layer Packet Length comes before Next Header. 2551 */ 2552 case (IPV6_VERSION >> 4): 2553 in6 = ip6->ip6_src; 2554 in6_clearscope(&in6); 2555 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2556 in6 = ip6->ip6_dst; 2557 in6_clearscope(&in6); 2558 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2559 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2560 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2561 nhdr = 0; 2562 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2563 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2564 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2565 nhdr = IPPROTO_TCP; 2566 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2567 2568 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2569 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2570 break; 2571 #endif 2572 default: 2573 KEY_FREESAV(&sav); 2574 return (-1); 2575 /* NOTREACHED */ 2576 break; 2577 } 2578 2579 2580 /* 2581 * Step 2: Update MD5 hash with TCP header, excluding options. 2582 * The TCP checksum must be set to zero. 2583 */ 2584 savecsum = th->th_sum; 2585 th->th_sum = 0; 2586 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2587 th->th_sum = savecsum; 2588 2589 /* 2590 * Step 3: Update MD5 hash with TCP segment data. 2591 * Use m_apply() to avoid an early m_pullup(). 2592 */ 2593 if (len > 0) 2594 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2595 2596 /* 2597 * Step 4: Update MD5 hash with shared secret. 2598 */ 2599 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2600 MD5Final(buf, &ctx); 2601 2602 key_sa_recordxfer(sav, m); 2603 KEY_FREESAV(&sav); 2604 return (0); 2605 } 2606 2607 /* 2608 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2609 * 2610 * Return 0 if successful, otherwise return -1. 2611 */ 2612 int 2613 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 2614 u_char *buf, u_int direction) 2615 { 2616 struct secasvar *sav; 2617 2618 if ((sav = tcp_get_sav(m, direction)) == NULL) 2619 return (-1); 2620 2621 return (tcp_signature_do_compute(m, len, optlen, buf, sav)); 2622 } 2623 2624 /* 2625 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2626 * 2627 * Parameters: 2628 * m pointer to head of mbuf chain 2629 * len length of TCP segment data, excluding options 2630 * optlen length of TCP segment options 2631 * buf pointer to storage for computed MD5 digest 2632 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2633 * 2634 * Return 1 if successful, otherwise return 0. 2635 */ 2636 int 2637 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2638 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2639 { 2640 char tmpdigest[TCP_SIGLEN]; 2641 2642 if (tcp_sig_checksigs == 0) 2643 return (1); 2644 if ((tcpbflag & TF_SIGNATURE) == 0) { 2645 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2646 2647 /* 2648 * If this socket is not expecting signature but 2649 * the segment contains signature just fail. 2650 */ 2651 TCPSTAT_INC(tcps_sig_err_sigopt); 2652 TCPSTAT_INC(tcps_sig_rcvbadsig); 2653 return (0); 2654 } 2655 2656 /* Signature is not expected, and not present in segment. */ 2657 return (1); 2658 } 2659 2660 /* 2661 * If this socket is expecting signature but the segment does not 2662 * contain any just fail. 2663 */ 2664 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2665 TCPSTAT_INC(tcps_sig_err_nosigopt); 2666 TCPSTAT_INC(tcps_sig_rcvbadsig); 2667 return (0); 2668 } 2669 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2670 IPSEC_DIR_INBOUND) == -1) { 2671 TCPSTAT_INC(tcps_sig_err_buildsig); 2672 TCPSTAT_INC(tcps_sig_rcvbadsig); 2673 return (0); 2674 } 2675 2676 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2677 TCPSTAT_INC(tcps_sig_rcvbadsig); 2678 return (0); 2679 } 2680 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2681 return (1); 2682 } 2683 #endif /* TCP_SIGNATURE */ 2684 2685 static int 2686 sysctl_drop(SYSCTL_HANDLER_ARGS) 2687 { 2688 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2689 struct sockaddr_storage addrs[2]; 2690 struct inpcb *inp; 2691 struct tcpcb *tp; 2692 struct tcptw *tw; 2693 struct sockaddr_in *fin, *lin; 2694 #ifdef INET6 2695 struct sockaddr_in6 *fin6, *lin6; 2696 #endif 2697 int error; 2698 2699 inp = NULL; 2700 fin = lin = NULL; 2701 #ifdef INET6 2702 fin6 = lin6 = NULL; 2703 #endif 2704 error = 0; 2705 2706 if (req->oldptr != NULL || req->oldlen != 0) 2707 return (EINVAL); 2708 if (req->newptr == NULL) 2709 return (EPERM); 2710 if (req->newlen < sizeof(addrs)) 2711 return (ENOMEM); 2712 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2713 if (error) 2714 return (error); 2715 2716 switch (addrs[0].ss_family) { 2717 #ifdef INET6 2718 case AF_INET6: 2719 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2720 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2721 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2722 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2723 return (EINVAL); 2724 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2725 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2726 return (EINVAL); 2727 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2728 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2729 fin = (struct sockaddr_in *)&addrs[0]; 2730 lin = (struct sockaddr_in *)&addrs[1]; 2731 break; 2732 } 2733 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2734 if (error) 2735 return (error); 2736 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2737 if (error) 2738 return (error); 2739 break; 2740 #endif 2741 #ifdef INET 2742 case AF_INET: 2743 fin = (struct sockaddr_in *)&addrs[0]; 2744 lin = (struct sockaddr_in *)&addrs[1]; 2745 if (fin->sin_len != sizeof(struct sockaddr_in) || 2746 lin->sin_len != sizeof(struct sockaddr_in)) 2747 return (EINVAL); 2748 break; 2749 #endif 2750 default: 2751 return (EINVAL); 2752 } 2753 INP_INFO_RLOCK(&V_tcbinfo); 2754 switch (addrs[0].ss_family) { 2755 #ifdef INET6 2756 case AF_INET6: 2757 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2758 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2759 INPLOOKUP_WLOCKPCB, NULL); 2760 break; 2761 #endif 2762 #ifdef INET 2763 case AF_INET: 2764 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2765 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2766 break; 2767 #endif 2768 } 2769 if (inp != NULL) { 2770 if (inp->inp_flags & INP_TIMEWAIT) { 2771 /* 2772 * XXXRW: There currently exists a state where an 2773 * inpcb is present, but its timewait state has been 2774 * discarded. For now, don't allow dropping of this 2775 * type of inpcb. 2776 */ 2777 tw = intotw(inp); 2778 if (tw != NULL) 2779 tcp_twclose(tw, 0); 2780 else 2781 INP_WUNLOCK(inp); 2782 } else if (!(inp->inp_flags & INP_DROPPED) && 2783 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2784 tp = intotcpcb(inp); 2785 tp = tcp_drop(tp, ECONNABORTED); 2786 if (tp != NULL) 2787 INP_WUNLOCK(inp); 2788 } else 2789 INP_WUNLOCK(inp); 2790 } else 2791 error = ESRCH; 2792 INP_INFO_RUNLOCK(&V_tcbinfo); 2793 return (error); 2794 } 2795 2796 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2797 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2798 0, sysctl_drop, "", "Drop TCP connection"); 2799 2800 /* 2801 * Generate a standardized TCP log line for use throughout the 2802 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2803 * allow use in the interrupt context. 2804 * 2805 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2806 * NB: The function may return NULL if memory allocation failed. 2807 * 2808 * Due to header inclusion and ordering limitations the struct ip 2809 * and ip6_hdr pointers have to be passed as void pointers. 2810 */ 2811 char * 2812 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2813 const void *ip6hdr) 2814 { 2815 2816 /* Is logging enabled? */ 2817 if (tcp_log_in_vain == 0) 2818 return (NULL); 2819 2820 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2821 } 2822 2823 char * 2824 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2825 const void *ip6hdr) 2826 { 2827 2828 /* Is logging enabled? */ 2829 if (tcp_log_debug == 0) 2830 return (NULL); 2831 2832 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2833 } 2834 2835 static char * 2836 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2837 const void *ip6hdr) 2838 { 2839 char *s, *sp; 2840 size_t size; 2841 struct ip *ip; 2842 #ifdef INET6 2843 const struct ip6_hdr *ip6; 2844 2845 ip6 = (const struct ip6_hdr *)ip6hdr; 2846 #endif /* INET6 */ 2847 ip = (struct ip *)ip4hdr; 2848 2849 /* 2850 * The log line looks like this: 2851 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2852 */ 2853 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2854 sizeof(PRINT_TH_FLAGS) + 1 + 2855 #ifdef INET6 2856 2 * INET6_ADDRSTRLEN; 2857 #else 2858 2 * INET_ADDRSTRLEN; 2859 #endif /* INET6 */ 2860 2861 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2862 if (s == NULL) 2863 return (NULL); 2864 2865 strcat(s, "TCP: ["); 2866 sp = s + strlen(s); 2867 2868 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2869 inet_ntoa_r(inc->inc_faddr, sp); 2870 sp = s + strlen(s); 2871 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2872 sp = s + strlen(s); 2873 inet_ntoa_r(inc->inc_laddr, sp); 2874 sp = s + strlen(s); 2875 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2876 #ifdef INET6 2877 } else if (inc) { 2878 ip6_sprintf(sp, &inc->inc6_faddr); 2879 sp = s + strlen(s); 2880 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2881 sp = s + strlen(s); 2882 ip6_sprintf(sp, &inc->inc6_laddr); 2883 sp = s + strlen(s); 2884 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2885 } else if (ip6 && th) { 2886 ip6_sprintf(sp, &ip6->ip6_src); 2887 sp = s + strlen(s); 2888 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2889 sp = s + strlen(s); 2890 ip6_sprintf(sp, &ip6->ip6_dst); 2891 sp = s + strlen(s); 2892 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2893 #endif /* INET6 */ 2894 #ifdef INET 2895 } else if (ip && th) { 2896 inet_ntoa_r(ip->ip_src, sp); 2897 sp = s + strlen(s); 2898 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2899 sp = s + strlen(s); 2900 inet_ntoa_r(ip->ip_dst, sp); 2901 sp = s + strlen(s); 2902 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2903 #endif /* INET */ 2904 } else { 2905 free(s, M_TCPLOG); 2906 return (NULL); 2907 } 2908 sp = s + strlen(s); 2909 if (th) 2910 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2911 if (*(s + size - 1) != '\0') 2912 panic("%s: string too long", __func__); 2913 return (s); 2914 } 2915 2916 /* 2917 * A subroutine which makes it easy to track TCP state changes with DTrace. 2918 * This function shouldn't be called for t_state initializations that don't 2919 * correspond to actual TCP state transitions. 2920 */ 2921 void 2922 tcp_state_change(struct tcpcb *tp, int newstate) 2923 { 2924 #if defined(KDTRACE_HOOKS) 2925 int pstate = tp->t_state; 2926 #endif 2927 2928 TCPSTAT_DEC(tcps_states[tp->t_state]); 2929 TCPSTAT_INC(tcps_states[newstate]); 2930 tp->t_state = newstate; 2931 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2932 } 2933