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); 247 248 249 static struct tcp_function_block tcp_def_funcblk = { 250 "default", 251 tcp_output, 252 tcp_do_segment, 253 tcp_default_ctloutput, 254 NULL, 255 NULL, 256 NULL, 257 NULL, 258 NULL, 259 NULL, 260 0, 261 0 262 }; 263 264 int t_functions_inited = 0; 265 struct tcp_funchead t_functions; 266 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 267 268 static void 269 init_tcp_functions(void) 270 { 271 if (t_functions_inited == 0) { 272 TAILQ_INIT(&t_functions); 273 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 274 t_functions_inited = 1; 275 } 276 } 277 278 static struct tcp_function_block * 279 find_tcp_functions_locked(struct tcp_function_set *fs) 280 { 281 struct tcp_function *f; 282 struct tcp_function_block *blk=NULL; 283 284 TAILQ_FOREACH(f, &t_functions, tf_next) { 285 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) { 286 blk = f->tf_fb; 287 break; 288 } 289 } 290 return(blk); 291 } 292 293 static struct tcp_function_block * 294 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 295 { 296 struct tcp_function_block *rblk=NULL; 297 struct tcp_function *f; 298 299 TAILQ_FOREACH(f, &t_functions, tf_next) { 300 if (f->tf_fb == blk) { 301 rblk = blk; 302 if (s) { 303 *s = f; 304 } 305 break; 306 } 307 } 308 return (rblk); 309 } 310 311 struct tcp_function_block * 312 find_and_ref_tcp_functions(struct tcp_function_set *fs) 313 { 314 struct tcp_function_block *blk; 315 316 rw_rlock(&tcp_function_lock); 317 blk = find_tcp_functions_locked(fs); 318 if (blk) 319 refcount_acquire(&blk->tfb_refcnt); 320 rw_runlock(&tcp_function_lock); 321 return(blk); 322 } 323 324 struct tcp_function_block * 325 find_and_ref_tcp_fb(struct tcp_function_block *blk) 326 { 327 struct tcp_function_block *rblk; 328 329 rw_rlock(&tcp_function_lock); 330 rblk = find_tcp_fb_locked(blk, NULL); 331 if (rblk) 332 refcount_acquire(&rblk->tfb_refcnt); 333 rw_runlock(&tcp_function_lock); 334 return(rblk); 335 } 336 337 338 static int 339 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 340 { 341 int error=ENOENT; 342 struct tcp_function_set fs; 343 struct tcp_function_block *blk; 344 345 memset(&fs, 0, sizeof(fs)); 346 rw_rlock(&tcp_function_lock); 347 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 348 if (blk) { 349 /* Found him */ 350 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 351 fs.pcbcnt = blk->tfb_refcnt; 352 } 353 rw_runlock(&tcp_function_lock); 354 error = sysctl_handle_string(oidp, fs.function_set_name, 355 sizeof(fs.function_set_name), req); 356 357 /* Check for error or no change */ 358 if (error != 0 || req->newptr == NULL) 359 return(error); 360 361 rw_wlock(&tcp_function_lock); 362 blk = find_tcp_functions_locked(&fs); 363 if ((blk == NULL) || 364 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 365 error = ENOENT; 366 goto done; 367 } 368 tcp_func_set_ptr = blk; 369 done: 370 rw_wunlock(&tcp_function_lock); 371 return (error); 372 } 373 374 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 375 CTLTYPE_STRING | CTLFLAG_RW, 376 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 377 "Set/get the default TCP functions"); 378 379 static int 380 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 381 { 382 int error, cnt, linesz; 383 struct tcp_function *f; 384 char *buffer, *cp; 385 size_t bufsz, outsz; 386 387 cnt = 0; 388 rw_rlock(&tcp_function_lock); 389 TAILQ_FOREACH(f, &t_functions, tf_next) { 390 cnt++; 391 } 392 rw_runlock(&tcp_function_lock); 393 394 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1; 395 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 396 397 error = 0; 398 cp = buffer; 399 400 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count"); 401 cp += linesz; 402 bufsz -= linesz; 403 outsz = linesz; 404 405 rw_rlock(&tcp_function_lock); 406 TAILQ_FOREACH(f, &t_functions, tf_next) { 407 linesz = snprintf(cp, bufsz, "%-32s%c %u\n", 408 f->tf_fb->tfb_tcp_block_name, 409 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 410 f->tf_fb->tfb_refcnt); 411 if (linesz >= bufsz) { 412 error = EOVERFLOW; 413 break; 414 } 415 cp += linesz; 416 bufsz -= linesz; 417 outsz += linesz; 418 } 419 rw_runlock(&tcp_function_lock); 420 if (error == 0) 421 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 422 free(buffer, M_TEMP); 423 return (error); 424 } 425 426 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 427 CTLTYPE_STRING|CTLFLAG_RD, 428 NULL, 0, sysctl_net_inet_list_available, "A", 429 "list available TCP Function sets"); 430 431 /* 432 * Target size of TCP PCB hash tables. Must be a power of two. 433 * 434 * Note that this can be overridden by the kernel environment 435 * variable net.inet.tcp.tcbhashsize 436 */ 437 #ifndef TCBHASHSIZE 438 #define TCBHASHSIZE 0 439 #endif 440 441 /* 442 * XXX 443 * Callouts should be moved into struct tcp directly. They are currently 444 * separate because the tcpcb structure is exported to userland for sysctl 445 * parsing purposes, which do not know about callouts. 446 */ 447 struct tcpcb_mem { 448 struct tcpcb tcb; 449 struct tcp_timer tt; 450 struct cc_var ccv; 451 struct osd osd; 452 }; 453 454 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 455 #define V_tcpcb_zone VNET(tcpcb_zone) 456 457 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 458 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 459 460 static struct mtx isn_mtx; 461 462 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 463 #define ISN_LOCK() mtx_lock(&isn_mtx) 464 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 465 466 /* 467 * TCP initialization. 468 */ 469 static void 470 tcp_zone_change(void *tag) 471 { 472 473 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 474 uma_zone_set_max(V_tcpcb_zone, maxsockets); 475 tcp_tw_zone_change(); 476 } 477 478 static int 479 tcp_inpcb_init(void *mem, int size, int flags) 480 { 481 struct inpcb *inp = mem; 482 483 INP_LOCK_INIT(inp, "inp", "tcpinp"); 484 return (0); 485 } 486 487 /* 488 * Take a value and get the next power of 2 that doesn't overflow. 489 * Used to size the tcp_inpcb hash buckets. 490 */ 491 static int 492 maketcp_hashsize(int size) 493 { 494 int hashsize; 495 496 /* 497 * auto tune. 498 * get the next power of 2 higher than maxsockets. 499 */ 500 hashsize = 1 << fls(size); 501 /* catch overflow, and just go one power of 2 smaller */ 502 if (hashsize < size) { 503 hashsize = 1 << (fls(size) - 1); 504 } 505 return (hashsize); 506 } 507 508 int 509 register_tcp_functions(struct tcp_function_block *blk, int wait) 510 { 511 struct tcp_function_block *lblk; 512 struct tcp_function *n; 513 struct tcp_function_set fs; 514 515 if (t_functions_inited == 0) { 516 init_tcp_functions(); 517 } 518 if ((blk->tfb_tcp_output == NULL) || 519 (blk->tfb_tcp_do_segment == NULL) || 520 (blk->tfb_tcp_ctloutput == NULL) || 521 (strlen(blk->tfb_tcp_block_name) == 0)) { 522 /* 523 * These functions are required and you 524 * need a name. 525 */ 526 return (EINVAL); 527 } 528 if (blk->tfb_tcp_timer_stop_all || 529 blk->tfb_tcp_timer_activate || 530 blk->tfb_tcp_timer_active || 531 blk->tfb_tcp_timer_stop) { 532 /* 533 * If you define one timer function you 534 * must have them all. 535 */ 536 if ((blk->tfb_tcp_timer_stop_all == NULL) || 537 (blk->tfb_tcp_timer_activate == NULL) || 538 (blk->tfb_tcp_timer_active == NULL) || 539 (blk->tfb_tcp_timer_stop == NULL)) { 540 return (EINVAL); 541 } 542 } 543 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 544 if (n == NULL) { 545 return (ENOMEM); 546 } 547 n->tf_fb = blk; 548 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 549 rw_wlock(&tcp_function_lock); 550 lblk = find_tcp_functions_locked(&fs); 551 if (lblk) { 552 /* Duplicate name space not allowed */ 553 rw_wunlock(&tcp_function_lock); 554 free(n, M_TCPFUNCTIONS); 555 return (EALREADY); 556 } 557 refcount_init(&blk->tfb_refcnt, 0); 558 blk->tfb_flags = 0; 559 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 560 rw_wunlock(&tcp_function_lock); 561 return(0); 562 } 563 564 int 565 deregister_tcp_functions(struct tcp_function_block *blk) 566 { 567 struct tcp_function_block *lblk; 568 struct tcp_function *f; 569 int error=ENOENT; 570 571 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 572 /* You can't un-register the default */ 573 return (EPERM); 574 } 575 rw_wlock(&tcp_function_lock); 576 if (blk == tcp_func_set_ptr) { 577 /* You can't free the current default */ 578 rw_wunlock(&tcp_function_lock); 579 return (EBUSY); 580 } 581 if (blk->tfb_refcnt) { 582 /* Still tcb attached, mark it. */ 583 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 584 rw_wunlock(&tcp_function_lock); 585 return (EBUSY); 586 } 587 lblk = find_tcp_fb_locked(blk, &f); 588 if (lblk) { 589 /* Found */ 590 TAILQ_REMOVE(&t_functions, f, tf_next); 591 f->tf_fb = NULL; 592 free(f, M_TCPFUNCTIONS); 593 error = 0; 594 } 595 rw_wunlock(&tcp_function_lock); 596 return (error); 597 } 598 599 void 600 tcp_init(void) 601 { 602 const char *tcbhash_tuneable; 603 int hashsize; 604 605 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 606 607 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 608 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 609 printf("%s: WARNING: unable to register helper hook\n", __func__); 610 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 611 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 612 printf("%s: WARNING: unable to register helper hook\n", __func__); 613 hashsize = TCBHASHSIZE; 614 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 615 if (hashsize == 0) { 616 /* 617 * Auto tune the hash size based on maxsockets. 618 * A perfect hash would have a 1:1 mapping 619 * (hashsize = maxsockets) however it's been 620 * suggested that O(2) average is better. 621 */ 622 hashsize = maketcp_hashsize(maxsockets / 4); 623 /* 624 * Our historical default is 512, 625 * do not autotune lower than this. 626 */ 627 if (hashsize < 512) 628 hashsize = 512; 629 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 630 printf("%s: %s auto tuned to %d\n", __func__, 631 tcbhash_tuneable, hashsize); 632 } 633 /* 634 * We require a hashsize to be a power of two. 635 * Previously if it was not a power of two we would just reset it 636 * back to 512, which could be a nasty surprise if you did not notice 637 * the error message. 638 * Instead what we do is clip it to the closest power of two lower 639 * than the specified hash value. 640 */ 641 if (!powerof2(hashsize)) { 642 int oldhashsize = hashsize; 643 644 hashsize = maketcp_hashsize(hashsize); 645 /* prevent absurdly low value */ 646 if (hashsize < 16) 647 hashsize = 16; 648 printf("%s: WARNING: TCB hash size not a power of 2, " 649 "clipped from %d to %d.\n", __func__, oldhashsize, 650 hashsize); 651 } 652 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 653 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE); 654 655 /* 656 * These have to be type stable for the benefit of the timers. 657 */ 658 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 659 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 660 uma_zone_set_max(V_tcpcb_zone, maxsockets); 661 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 662 663 tcp_tw_init(); 664 syncache_init(); 665 tcp_hc_init(); 666 667 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 668 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 669 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 670 671 /* Skip initialization of globals for non-default instances. */ 672 if (!IS_DEFAULT_VNET(curvnet)) 673 return; 674 675 tcp_reass_global_init(); 676 677 /* XXX virtualize those bellow? */ 678 tcp_delacktime = TCPTV_DELACK; 679 tcp_keepinit = TCPTV_KEEP_INIT; 680 tcp_keepidle = TCPTV_KEEP_IDLE; 681 tcp_keepintvl = TCPTV_KEEPINTVL; 682 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 683 tcp_msl = TCPTV_MSL; 684 tcp_rexmit_min = TCPTV_MIN; 685 if (tcp_rexmit_min < 1) 686 tcp_rexmit_min = 1; 687 tcp_persmin = TCPTV_PERSMIN; 688 tcp_persmax = TCPTV_PERSMAX; 689 tcp_rexmit_slop = TCPTV_CPU_VAR; 690 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 691 tcp_tcbhashsize = hashsize; 692 /* Setup the tcp function block list */ 693 init_tcp_functions(); 694 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 695 696 if (tcp_soreceive_stream) { 697 #ifdef INET 698 tcp_usrreqs.pru_soreceive = soreceive_stream; 699 #endif 700 #ifdef INET6 701 tcp6_usrreqs.pru_soreceive = soreceive_stream; 702 #endif /* INET6 */ 703 } 704 705 #ifdef INET6 706 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 707 #else /* INET6 */ 708 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 709 #endif /* INET6 */ 710 if (max_protohdr < TCP_MINPROTOHDR) 711 max_protohdr = TCP_MINPROTOHDR; 712 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 713 panic("tcp_init"); 714 #undef TCP_MINPROTOHDR 715 716 ISN_LOCK_INIT(); 717 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 718 SHUTDOWN_PRI_DEFAULT); 719 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 720 EVENTHANDLER_PRI_ANY); 721 #ifdef TCPPCAP 722 tcp_pcap_init(); 723 #endif 724 725 #ifdef TCP_RFC7413 726 tcp_fastopen_init(); 727 #endif 728 } 729 730 #ifdef VIMAGE 731 void 732 tcp_destroy(void) 733 { 734 int error; 735 736 /* 737 * All our processes are gone, all our sockets should be cleaned 738 * up, which means, we should be past the tcp_discardcb() calls. 739 * Sleep to let all tcpcb timers really disappear and then cleanup. 740 * Timewait will cleanup its queue and will be ready to go. 741 * XXX-BZ In theory a few ticks should be good enough to make sure 742 * the timers are all really gone. We should see if we could use a 743 * better metric here and, e.g., check a tcbcb count as an optimization? 744 */ 745 DELAY(1000000 / hz); 746 tcp_hc_destroy(); 747 syncache_destroy(); 748 tcp_tw_destroy(); 749 in_pcbinfo_destroy(&V_tcbinfo); 750 /* tcp_discardcb() clears the sack_holes up. */ 751 uma_zdestroy(V_sack_hole_zone); 752 uma_zdestroy(V_tcpcb_zone); 753 754 #ifdef TCP_RFC7413 755 /* 756 * Cannot free the zone until all tcpcbs are released as we attach 757 * the allocations to them. 758 */ 759 tcp_fastopen_destroy(); 760 #endif 761 762 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 763 if (error != 0) { 764 printf("%s: WARNING: unable to deregister helper hook " 765 "type=%d, id=%d: error %d returned\n", __func__, 766 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 767 } 768 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 769 if (error != 0) { 770 printf("%s: WARNING: unable to deregister helper hook " 771 "type=%d, id=%d: error %d returned\n", __func__, 772 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 773 } 774 } 775 #endif 776 777 void 778 tcp_fini(void *xtp) 779 { 780 781 } 782 783 /* 784 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 785 * tcp_template used to store this data in mbufs, but we now recopy it out 786 * of the tcpcb each time to conserve mbufs. 787 */ 788 void 789 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 790 { 791 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 792 793 INP_WLOCK_ASSERT(inp); 794 795 #ifdef INET6 796 if ((inp->inp_vflag & INP_IPV6) != 0) { 797 struct ip6_hdr *ip6; 798 799 ip6 = (struct ip6_hdr *)ip_ptr; 800 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 801 (inp->inp_flow & IPV6_FLOWINFO_MASK); 802 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 803 (IPV6_VERSION & IPV6_VERSION_MASK); 804 ip6->ip6_nxt = IPPROTO_TCP; 805 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 806 ip6->ip6_src = inp->in6p_laddr; 807 ip6->ip6_dst = inp->in6p_faddr; 808 } 809 #endif /* INET6 */ 810 #if defined(INET6) && defined(INET) 811 else 812 #endif 813 #ifdef INET 814 { 815 struct ip *ip; 816 817 ip = (struct ip *)ip_ptr; 818 ip->ip_v = IPVERSION; 819 ip->ip_hl = 5; 820 ip->ip_tos = inp->inp_ip_tos; 821 ip->ip_len = 0; 822 ip->ip_id = 0; 823 ip->ip_off = 0; 824 ip->ip_ttl = inp->inp_ip_ttl; 825 ip->ip_sum = 0; 826 ip->ip_p = IPPROTO_TCP; 827 ip->ip_src = inp->inp_laddr; 828 ip->ip_dst = inp->inp_faddr; 829 } 830 #endif /* INET */ 831 th->th_sport = inp->inp_lport; 832 th->th_dport = inp->inp_fport; 833 th->th_seq = 0; 834 th->th_ack = 0; 835 th->th_x2 = 0; 836 th->th_off = 5; 837 th->th_flags = 0; 838 th->th_win = 0; 839 th->th_urp = 0; 840 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 841 } 842 843 /* 844 * Create template to be used to send tcp packets on a connection. 845 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 846 * use for this function is in keepalives, which use tcp_respond. 847 */ 848 struct tcptemp * 849 tcpip_maketemplate(struct inpcb *inp) 850 { 851 struct tcptemp *t; 852 853 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 854 if (t == NULL) 855 return (NULL); 856 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 857 return (t); 858 } 859 860 /* 861 * Send a single message to the TCP at address specified by 862 * the given TCP/IP header. If m == NULL, then we make a copy 863 * of the tcpiphdr at th and send directly to the addressed host. 864 * This is used to force keep alive messages out using the TCP 865 * template for a connection. If flags are given then we send 866 * a message back to the TCP which originated the segment th, 867 * and discard the mbuf containing it and any other attached mbufs. 868 * 869 * In any case the ack and sequence number of the transmitted 870 * segment are as specified by the parameters. 871 * 872 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 873 */ 874 void 875 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 876 tcp_seq ack, tcp_seq seq, int flags) 877 { 878 struct tcpopt to; 879 struct inpcb *inp; 880 struct ip *ip; 881 struct mbuf *optm; 882 struct tcphdr *nth; 883 u_char *optp; 884 #ifdef INET6 885 struct ip6_hdr *ip6; 886 int isipv6; 887 #endif /* INET6 */ 888 int optlen, tlen, win; 889 bool incl_opts; 890 891 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 892 893 #ifdef INET6 894 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 895 ip6 = ipgen; 896 #endif /* INET6 */ 897 ip = ipgen; 898 899 if (tp != NULL) { 900 inp = tp->t_inpcb; 901 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 902 INP_WLOCK_ASSERT(inp); 903 } else 904 inp = NULL; 905 906 incl_opts = false; 907 win = 0; 908 if (tp != NULL) { 909 if (!(flags & TH_RST)) { 910 win = sbspace(&inp->inp_socket->so_rcv); 911 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 912 win = (long)TCP_MAXWIN << tp->rcv_scale; 913 } 914 if ((tp->t_flags & TF_NOOPT) == 0) 915 incl_opts = true; 916 } 917 if (m == NULL) { 918 m = m_gethdr(M_NOWAIT, MT_DATA); 919 if (m == NULL) 920 return; 921 m->m_data += max_linkhdr; 922 #ifdef INET6 923 if (isipv6) { 924 bcopy((caddr_t)ip6, mtod(m, caddr_t), 925 sizeof(struct ip6_hdr)); 926 ip6 = mtod(m, struct ip6_hdr *); 927 nth = (struct tcphdr *)(ip6 + 1); 928 } else 929 #endif /* INET6 */ 930 { 931 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 932 ip = mtod(m, struct ip *); 933 nth = (struct tcphdr *)(ip + 1); 934 } 935 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 936 flags = TH_ACK; 937 } else { 938 /* 939 * reuse the mbuf. 940 * XXX MRT We inherrit the FIB, which is lucky. 941 */ 942 m_freem(m->m_next); 943 m->m_next = NULL; 944 m->m_data = (caddr_t)ipgen; 945 /* m_len is set later */ 946 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 947 #ifdef INET6 948 if (isipv6) { 949 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 950 nth = (struct tcphdr *)(ip6 + 1); 951 } else 952 #endif /* INET6 */ 953 { 954 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 955 nth = (struct tcphdr *)(ip + 1); 956 } 957 if (th != nth) { 958 /* 959 * this is usually a case when an extension header 960 * exists between the IPv6 header and the 961 * TCP header. 962 */ 963 nth->th_sport = th->th_sport; 964 nth->th_dport = th->th_dport; 965 } 966 xchg(nth->th_dport, nth->th_sport, uint16_t); 967 #undef xchg 968 } 969 tlen = 0; 970 #ifdef INET6 971 if (isipv6) 972 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 973 #endif 974 #if defined(INET) && defined(INET6) 975 else 976 #endif 977 #ifdef INET 978 tlen = sizeof (struct tcpiphdr); 979 #endif 980 #ifdef INVARIANTS 981 m->m_len = 0; 982 KASSERT(M_TRAILINGSPACE(m) >= tlen, 983 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 984 m, tlen, (long)M_TRAILINGSPACE(m))); 985 #endif 986 m->m_len = tlen; 987 to.to_flags = 0; 988 if (incl_opts) { 989 /* Make sure we have room. */ 990 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 991 m->m_next = m_get(M_NOWAIT, MT_DATA); 992 if (m->m_next) { 993 optp = mtod(m->m_next, u_char *); 994 optm = m->m_next; 995 } else 996 incl_opts = false; 997 } else { 998 optp = (u_char *) (nth + 1); 999 optm = m; 1000 } 1001 } 1002 if (incl_opts) { 1003 /* Timestamps. */ 1004 if (tp->t_flags & TF_RCVD_TSTMP) { 1005 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1006 to.to_tsecr = tp->ts_recent; 1007 to.to_flags |= TOF_TS; 1008 } 1009 #ifdef TCP_SIGNATURE 1010 /* TCP-MD5 (RFC2385). */ 1011 if (tp->t_flags & TF_SIGNATURE) 1012 to.to_flags |= TOF_SIGNATURE; 1013 #endif 1014 1015 /* Add the options. */ 1016 tlen += optlen = tcp_addoptions(&to, optp); 1017 1018 /* Update m_len in the correct mbuf. */ 1019 optm->m_len += optlen; 1020 } else 1021 optlen = 0; 1022 #ifdef INET6 1023 if (isipv6) { 1024 ip6->ip6_flow = 0; 1025 ip6->ip6_vfc = IPV6_VERSION; 1026 ip6->ip6_nxt = IPPROTO_TCP; 1027 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1028 } 1029 #endif 1030 #if defined(INET) && defined(INET6) 1031 else 1032 #endif 1033 #ifdef INET 1034 { 1035 ip->ip_len = htons(tlen); 1036 ip->ip_ttl = V_ip_defttl; 1037 if (V_path_mtu_discovery) 1038 ip->ip_off |= htons(IP_DF); 1039 } 1040 #endif 1041 m->m_pkthdr.len = tlen; 1042 m->m_pkthdr.rcvif = NULL; 1043 #ifdef MAC 1044 if (inp != NULL) { 1045 /* 1046 * Packet is associated with a socket, so allow the 1047 * label of the response to reflect the socket label. 1048 */ 1049 INP_WLOCK_ASSERT(inp); 1050 mac_inpcb_create_mbuf(inp, m); 1051 } else { 1052 /* 1053 * Packet is not associated with a socket, so possibly 1054 * update the label in place. 1055 */ 1056 mac_netinet_tcp_reply(m); 1057 } 1058 #endif 1059 nth->th_seq = htonl(seq); 1060 nth->th_ack = htonl(ack); 1061 nth->th_x2 = 0; 1062 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1063 nth->th_flags = flags; 1064 if (tp != NULL) 1065 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1066 else 1067 nth->th_win = htons((u_short)win); 1068 nth->th_urp = 0; 1069 1070 #ifdef TCP_SIGNATURE 1071 if (to.to_flags & TOF_SIGNATURE) { 1072 tcp_signature_compute(m, 0, 0, optlen, to.to_signature, 1073 IPSEC_DIR_OUTBOUND); 1074 } 1075 #endif 1076 1077 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1078 #ifdef INET6 1079 if (isipv6) { 1080 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1081 nth->th_sum = in6_cksum_pseudo(ip6, 1082 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1083 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1084 NULL, NULL); 1085 } 1086 #endif /* INET6 */ 1087 #if defined(INET6) && defined(INET) 1088 else 1089 #endif 1090 #ifdef INET 1091 { 1092 m->m_pkthdr.csum_flags = CSUM_TCP; 1093 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1094 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1095 } 1096 #endif /* INET */ 1097 #ifdef TCPDEBUG 1098 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1099 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1100 #endif 1101 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *)); 1102 if (flags & TH_RST) 1103 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), 1104 tp, nth); 1105 1106 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); 1107 #ifdef INET6 1108 if (isipv6) 1109 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1110 #endif /* INET6 */ 1111 #if defined(INET) && defined(INET6) 1112 else 1113 #endif 1114 #ifdef INET 1115 (void) ip_output(m, NULL, NULL, 0, NULL, inp); 1116 #endif 1117 } 1118 1119 /* 1120 * Create a new TCP control block, making an 1121 * empty reassembly queue and hooking it to the argument 1122 * protocol control block. The `inp' parameter must have 1123 * come from the zone allocator set up in tcp_init(). 1124 */ 1125 struct tcpcb * 1126 tcp_newtcpcb(struct inpcb *inp) 1127 { 1128 struct tcpcb_mem *tm; 1129 struct tcpcb *tp; 1130 #ifdef INET6 1131 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1132 #endif /* INET6 */ 1133 1134 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1135 if (tm == NULL) 1136 return (NULL); 1137 tp = &tm->tcb; 1138 1139 /* Initialise cc_var struct for this tcpcb. */ 1140 tp->ccv = &tm->ccv; 1141 tp->ccv->type = IPPROTO_TCP; 1142 tp->ccv->ccvc.tcp = tp; 1143 rw_rlock(&tcp_function_lock); 1144 tp->t_fb = tcp_func_set_ptr; 1145 refcount_acquire(&tp->t_fb->tfb_refcnt); 1146 rw_runlock(&tcp_function_lock); 1147 if (tp->t_fb->tfb_tcp_fb_init) { 1148 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1149 } 1150 /* 1151 * Use the current system default CC algorithm. 1152 */ 1153 CC_LIST_RLOCK(); 1154 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1155 CC_ALGO(tp) = CC_DEFAULT(); 1156 CC_LIST_RUNLOCK(); 1157 1158 if (CC_ALGO(tp)->cb_init != NULL) 1159 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1160 if (tp->t_fb->tfb_tcp_fb_fini) 1161 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1162 refcount_release(&tp->t_fb->tfb_refcnt); 1163 uma_zfree(V_tcpcb_zone, tm); 1164 return (NULL); 1165 } 1166 1167 tp->osd = &tm->osd; 1168 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1169 if (tp->t_fb->tfb_tcp_fb_fini) 1170 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1171 refcount_release(&tp->t_fb->tfb_refcnt); 1172 uma_zfree(V_tcpcb_zone, tm); 1173 return (NULL); 1174 } 1175 1176 #ifdef VIMAGE 1177 tp->t_vnet = inp->inp_vnet; 1178 #endif 1179 tp->t_timers = &tm->tt; 1180 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1181 tp->t_maxseg = 1182 #ifdef INET6 1183 isipv6 ? V_tcp_v6mssdflt : 1184 #endif /* INET6 */ 1185 V_tcp_mssdflt; 1186 1187 /* Set up our timeouts. */ 1188 callout_init(&tp->t_timers->tt_rexmt, 1); 1189 callout_init(&tp->t_timers->tt_persist, 1); 1190 callout_init(&tp->t_timers->tt_keep, 1); 1191 callout_init(&tp->t_timers->tt_2msl, 1); 1192 callout_init(&tp->t_timers->tt_delack, 1); 1193 1194 if (V_tcp_do_rfc1323) 1195 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1196 if (V_tcp_do_sack) 1197 tp->t_flags |= TF_SACK_PERMIT; 1198 TAILQ_INIT(&tp->snd_holes); 1199 /* 1200 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1201 * is called. 1202 */ 1203 in_pcbref(inp); /* Reference for tcpcb */ 1204 tp->t_inpcb = inp; 1205 1206 /* 1207 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1208 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1209 * reasonable initial retransmit time. 1210 */ 1211 tp->t_srtt = TCPTV_SRTTBASE; 1212 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1213 tp->t_rttmin = tcp_rexmit_min; 1214 tp->t_rxtcur = TCPTV_RTOBASE; 1215 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1216 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1217 tp->t_rcvtime = ticks; 1218 /* 1219 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1220 * because the socket may be bound to an IPv6 wildcard address, 1221 * which may match an IPv4-mapped IPv6 address. 1222 */ 1223 inp->inp_ip_ttl = V_ip_defttl; 1224 inp->inp_ppcb = tp; 1225 #ifdef TCPPCAP 1226 /* 1227 * Init the TCP PCAP queues. 1228 */ 1229 tcp_pcap_tcpcb_init(tp); 1230 #endif 1231 return (tp); /* XXX */ 1232 } 1233 1234 /* 1235 * Switch the congestion control algorithm back to NewReno for any active 1236 * control blocks using an algorithm which is about to go away. 1237 * This ensures the CC framework can allow the unload to proceed without leaving 1238 * any dangling pointers which would trigger a panic. 1239 * Returning non-zero would inform the CC framework that something went wrong 1240 * and it would be unsafe to allow the unload to proceed. However, there is no 1241 * way for this to occur with this implementation so we always return zero. 1242 */ 1243 int 1244 tcp_ccalgounload(struct cc_algo *unload_algo) 1245 { 1246 struct cc_algo *tmpalgo; 1247 struct inpcb *inp; 1248 struct tcpcb *tp; 1249 VNET_ITERATOR_DECL(vnet_iter); 1250 1251 /* 1252 * Check all active control blocks across all network stacks and change 1253 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1254 * requires cleanup code to be run, call it. 1255 */ 1256 VNET_LIST_RLOCK(); 1257 VNET_FOREACH(vnet_iter) { 1258 CURVNET_SET(vnet_iter); 1259 INP_INFO_WLOCK(&V_tcbinfo); 1260 /* 1261 * New connections already part way through being initialised 1262 * with the CC algo we're removing will not race with this code 1263 * because the INP_INFO_WLOCK is held during initialisation. We 1264 * therefore don't enter the loop below until the connection 1265 * list has stabilised. 1266 */ 1267 LIST_FOREACH(inp, &V_tcb, inp_list) { 1268 INP_WLOCK(inp); 1269 /* Important to skip tcptw structs. */ 1270 if (!(inp->inp_flags & INP_TIMEWAIT) && 1271 (tp = intotcpcb(inp)) != NULL) { 1272 /* 1273 * By holding INP_WLOCK here, we are assured 1274 * that the connection is not currently 1275 * executing inside the CC module's functions 1276 * i.e. it is safe to make the switch back to 1277 * NewReno. 1278 */ 1279 if (CC_ALGO(tp) == unload_algo) { 1280 tmpalgo = CC_ALGO(tp); 1281 /* NewReno does not require any init. */ 1282 CC_ALGO(tp) = &newreno_cc_algo; 1283 if (tmpalgo->cb_destroy != NULL) 1284 tmpalgo->cb_destroy(tp->ccv); 1285 } 1286 } 1287 INP_WUNLOCK(inp); 1288 } 1289 INP_INFO_WUNLOCK(&V_tcbinfo); 1290 CURVNET_RESTORE(); 1291 } 1292 VNET_LIST_RUNLOCK(); 1293 1294 return (0); 1295 } 1296 1297 /* 1298 * Drop a TCP connection, reporting 1299 * the specified error. If connection is synchronized, 1300 * then send a RST to peer. 1301 */ 1302 struct tcpcb * 1303 tcp_drop(struct tcpcb *tp, int errno) 1304 { 1305 struct socket *so = tp->t_inpcb->inp_socket; 1306 1307 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1308 INP_WLOCK_ASSERT(tp->t_inpcb); 1309 1310 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1311 tcp_state_change(tp, TCPS_CLOSED); 1312 (void) tp->t_fb->tfb_tcp_output(tp); 1313 TCPSTAT_INC(tcps_drops); 1314 } else 1315 TCPSTAT_INC(tcps_conndrops); 1316 if (errno == ETIMEDOUT && tp->t_softerror) 1317 errno = tp->t_softerror; 1318 so->so_error = errno; 1319 return (tcp_close(tp)); 1320 } 1321 1322 void 1323 tcp_discardcb(struct tcpcb *tp) 1324 { 1325 struct inpcb *inp = tp->t_inpcb; 1326 struct socket *so = inp->inp_socket; 1327 #ifdef INET6 1328 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1329 #endif /* INET6 */ 1330 int released; 1331 1332 INP_WLOCK_ASSERT(inp); 1333 1334 /* 1335 * Make sure that all of our timers are stopped before we delete the 1336 * PCB. 1337 * 1338 * If stopping a timer fails, we schedule a discard function in same 1339 * callout, and the last discard function called will take care of 1340 * deleting the tcpcb. 1341 */ 1342 tp->t_timers->tt_draincnt = 0; 1343 tcp_timer_stop(tp, TT_REXMT); 1344 tcp_timer_stop(tp, TT_PERSIST); 1345 tcp_timer_stop(tp, TT_KEEP); 1346 tcp_timer_stop(tp, TT_2MSL); 1347 tcp_timer_stop(tp, TT_DELACK); 1348 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1349 /* 1350 * Call the stop-all function of the methods, 1351 * this function should call the tcp_timer_stop() 1352 * method with each of the function specific timeouts. 1353 * That stop will be called via the tfb_tcp_timer_stop() 1354 * which should use the async drain function of the 1355 * callout system (see tcp_var.h). 1356 */ 1357 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1358 } 1359 1360 /* 1361 * If we got enough samples through the srtt filter, 1362 * save the rtt and rttvar in the routing entry. 1363 * 'Enough' is arbitrarily defined as 4 rtt samples. 1364 * 4 samples is enough for the srtt filter to converge 1365 * to within enough % of the correct value; fewer samples 1366 * and we could save a bogus rtt. The danger is not high 1367 * as tcp quickly recovers from everything. 1368 * XXX: Works very well but needs some more statistics! 1369 */ 1370 if (tp->t_rttupdated >= 4) { 1371 struct hc_metrics_lite metrics; 1372 u_long ssthresh; 1373 1374 bzero(&metrics, sizeof(metrics)); 1375 /* 1376 * Update the ssthresh always when the conditions below 1377 * are satisfied. This gives us better new start value 1378 * for the congestion avoidance for new connections. 1379 * ssthresh is only set if packet loss occured on a session. 1380 * 1381 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1382 * being torn down. Ideally this code would not use 'so'. 1383 */ 1384 ssthresh = tp->snd_ssthresh; 1385 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1386 /* 1387 * convert the limit from user data bytes to 1388 * packets then to packet data bytes. 1389 */ 1390 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1391 if (ssthresh < 2) 1392 ssthresh = 2; 1393 ssthresh *= (u_long)(tp->t_maxseg + 1394 #ifdef INET6 1395 (isipv6 ? sizeof (struct ip6_hdr) + 1396 sizeof (struct tcphdr) : 1397 #endif 1398 sizeof (struct tcpiphdr) 1399 #ifdef INET6 1400 ) 1401 #endif 1402 ); 1403 } else 1404 ssthresh = 0; 1405 metrics.rmx_ssthresh = ssthresh; 1406 1407 metrics.rmx_rtt = tp->t_srtt; 1408 metrics.rmx_rttvar = tp->t_rttvar; 1409 metrics.rmx_cwnd = tp->snd_cwnd; 1410 metrics.rmx_sendpipe = 0; 1411 metrics.rmx_recvpipe = 0; 1412 1413 tcp_hc_update(&inp->inp_inc, &metrics); 1414 } 1415 1416 /* free the reassembly queue, if any */ 1417 tcp_reass_flush(tp); 1418 1419 #ifdef TCP_OFFLOAD 1420 /* Disconnect offload device, if any. */ 1421 if (tp->t_flags & TF_TOE) 1422 tcp_offload_detach(tp); 1423 #endif 1424 1425 tcp_free_sackholes(tp); 1426 1427 #ifdef TCPPCAP 1428 /* Free the TCP PCAP queues. */ 1429 tcp_pcap_drain(&(tp->t_inpkts)); 1430 tcp_pcap_drain(&(tp->t_outpkts)); 1431 #endif 1432 1433 /* Allow the CC algorithm to clean up after itself. */ 1434 if (CC_ALGO(tp)->cb_destroy != NULL) 1435 CC_ALGO(tp)->cb_destroy(tp->ccv); 1436 1437 khelp_destroy_osd(tp->osd); 1438 1439 CC_ALGO(tp) = NULL; 1440 inp->inp_ppcb = NULL; 1441 if (tp->t_timers->tt_draincnt == 0) { 1442 /* We own the last reference on tcpcb, let's free it. */ 1443 if (tp->t_fb->tfb_tcp_fb_fini) 1444 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1445 refcount_release(&tp->t_fb->tfb_refcnt); 1446 tp->t_inpcb = NULL; 1447 uma_zfree(V_tcpcb_zone, tp); 1448 released = in_pcbrele_wlocked(inp); 1449 KASSERT(!released, ("%s: inp %p should not have been released " 1450 "here", __func__, inp)); 1451 } 1452 } 1453 1454 void 1455 tcp_timer_discard(void *ptp) 1456 { 1457 struct inpcb *inp; 1458 struct tcpcb *tp; 1459 1460 tp = (struct tcpcb *)ptp; 1461 CURVNET_SET(tp->t_vnet); 1462 INP_INFO_RLOCK(&V_tcbinfo); 1463 inp = tp->t_inpcb; 1464 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1465 __func__, tp)); 1466 INP_WLOCK(inp); 1467 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1468 ("%s: tcpcb has to be stopped here", __func__)); 1469 tp->t_timers->tt_draincnt--; 1470 if (tp->t_timers->tt_draincnt == 0) { 1471 /* We own the last reference on this tcpcb, let's free it. */ 1472 if (tp->t_fb->tfb_tcp_fb_fini) 1473 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1474 refcount_release(&tp->t_fb->tfb_refcnt); 1475 tp->t_inpcb = NULL; 1476 uma_zfree(V_tcpcb_zone, tp); 1477 if (in_pcbrele_wlocked(inp)) { 1478 INP_INFO_RUNLOCK(&V_tcbinfo); 1479 CURVNET_RESTORE(); 1480 return; 1481 } 1482 } 1483 INP_WUNLOCK(inp); 1484 INP_INFO_RUNLOCK(&V_tcbinfo); 1485 CURVNET_RESTORE(); 1486 } 1487 1488 /* 1489 * Attempt to close a TCP control block, marking it as dropped, and freeing 1490 * the socket if we hold the only reference. 1491 */ 1492 struct tcpcb * 1493 tcp_close(struct tcpcb *tp) 1494 { 1495 struct inpcb *inp = tp->t_inpcb; 1496 struct socket *so; 1497 1498 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1499 INP_WLOCK_ASSERT(inp); 1500 1501 #ifdef TCP_OFFLOAD 1502 if (tp->t_state == TCPS_LISTEN) 1503 tcp_offload_listen_stop(tp); 1504 #endif 1505 #ifdef TCP_RFC7413 1506 /* 1507 * This releases the TFO pending counter resource for TFO listen 1508 * sockets as well as passively-created TFO sockets that transition 1509 * from SYN_RECEIVED to CLOSED. 1510 */ 1511 if (tp->t_tfo_pending) { 1512 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1513 tp->t_tfo_pending = NULL; 1514 } 1515 #endif 1516 in_pcbdrop(inp); 1517 TCPSTAT_INC(tcps_closed); 1518 TCPSTATES_DEC(tp->t_state); 1519 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1520 so = inp->inp_socket; 1521 soisdisconnected(so); 1522 if (inp->inp_flags & INP_SOCKREF) { 1523 KASSERT(so->so_state & SS_PROTOREF, 1524 ("tcp_close: !SS_PROTOREF")); 1525 inp->inp_flags &= ~INP_SOCKREF; 1526 INP_WUNLOCK(inp); 1527 ACCEPT_LOCK(); 1528 SOCK_LOCK(so); 1529 so->so_state &= ~SS_PROTOREF; 1530 sofree(so); 1531 return (NULL); 1532 } 1533 return (tp); 1534 } 1535 1536 void 1537 tcp_drain(void) 1538 { 1539 VNET_ITERATOR_DECL(vnet_iter); 1540 1541 if (!do_tcpdrain) 1542 return; 1543 1544 VNET_LIST_RLOCK_NOSLEEP(); 1545 VNET_FOREACH(vnet_iter) { 1546 CURVNET_SET(vnet_iter); 1547 struct inpcb *inpb; 1548 struct tcpcb *tcpb; 1549 1550 /* 1551 * Walk the tcpbs, if existing, and flush the reassembly queue, 1552 * if there is one... 1553 * XXX: The "Net/3" implementation doesn't imply that the TCP 1554 * reassembly queue should be flushed, but in a situation 1555 * where we're really low on mbufs, this is potentially 1556 * useful. 1557 */ 1558 INP_INFO_WLOCK(&V_tcbinfo); 1559 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1560 if (inpb->inp_flags & INP_TIMEWAIT) 1561 continue; 1562 INP_WLOCK(inpb); 1563 if ((tcpb = intotcpcb(inpb)) != NULL) { 1564 tcp_reass_flush(tcpb); 1565 tcp_clean_sackreport(tcpb); 1566 } 1567 INP_WUNLOCK(inpb); 1568 } 1569 INP_INFO_WUNLOCK(&V_tcbinfo); 1570 CURVNET_RESTORE(); 1571 } 1572 VNET_LIST_RUNLOCK_NOSLEEP(); 1573 } 1574 1575 /* 1576 * Notify a tcp user of an asynchronous error; 1577 * store error as soft error, but wake up user 1578 * (for now, won't do anything until can select for soft error). 1579 * 1580 * Do not wake up user since there currently is no mechanism for 1581 * reporting soft errors (yet - a kqueue filter may be added). 1582 */ 1583 static struct inpcb * 1584 tcp_notify(struct inpcb *inp, int error) 1585 { 1586 struct tcpcb *tp; 1587 1588 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1589 INP_WLOCK_ASSERT(inp); 1590 1591 if ((inp->inp_flags & INP_TIMEWAIT) || 1592 (inp->inp_flags & INP_DROPPED)) 1593 return (inp); 1594 1595 tp = intotcpcb(inp); 1596 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1597 1598 /* 1599 * Ignore some errors if we are hooked up. 1600 * If connection hasn't completed, has retransmitted several times, 1601 * and receives a second error, give up now. This is better 1602 * than waiting a long time to establish a connection that 1603 * can never complete. 1604 */ 1605 if (tp->t_state == TCPS_ESTABLISHED && 1606 (error == EHOSTUNREACH || error == ENETUNREACH || 1607 error == EHOSTDOWN)) { 1608 if (inp->inp_route.ro_rt) { 1609 RTFREE(inp->inp_route.ro_rt); 1610 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1611 } 1612 return (inp); 1613 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1614 tp->t_softerror) { 1615 tp = tcp_drop(tp, error); 1616 if (tp != NULL) 1617 return (inp); 1618 else 1619 return (NULL); 1620 } else { 1621 tp->t_softerror = error; 1622 return (inp); 1623 } 1624 #if 0 1625 wakeup( &so->so_timeo); 1626 sorwakeup(so); 1627 sowwakeup(so); 1628 #endif 1629 } 1630 1631 static int 1632 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1633 { 1634 int error, i, m, n, pcb_count; 1635 struct inpcb *inp, **inp_list; 1636 inp_gen_t gencnt; 1637 struct xinpgen xig; 1638 1639 /* 1640 * The process of preparing the TCB list is too time-consuming and 1641 * resource-intensive to repeat twice on every request. 1642 */ 1643 if (req->oldptr == NULL) { 1644 n = V_tcbinfo.ipi_count + 1645 counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]); 1646 n += imax(n / 8, 10); 1647 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1648 return (0); 1649 } 1650 1651 if (req->newptr != NULL) 1652 return (EPERM); 1653 1654 /* 1655 * OK, now we're committed to doing something. 1656 */ 1657 INP_LIST_RLOCK(&V_tcbinfo); 1658 gencnt = V_tcbinfo.ipi_gencnt; 1659 n = V_tcbinfo.ipi_count; 1660 INP_LIST_RUNLOCK(&V_tcbinfo); 1661 1662 m = counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]); 1663 1664 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1665 + (n + m) * sizeof(struct xtcpcb)); 1666 if (error != 0) 1667 return (error); 1668 1669 xig.xig_len = sizeof xig; 1670 xig.xig_count = n + m; 1671 xig.xig_gen = gencnt; 1672 xig.xig_sogen = so_gencnt; 1673 error = SYSCTL_OUT(req, &xig, sizeof xig); 1674 if (error) 1675 return (error); 1676 1677 error = syncache_pcblist(req, m, &pcb_count); 1678 if (error) 1679 return (error); 1680 1681 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1682 1683 INP_INFO_WLOCK(&V_tcbinfo); 1684 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1685 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1686 INP_WLOCK(inp); 1687 if (inp->inp_gencnt <= gencnt) { 1688 /* 1689 * XXX: This use of cr_cansee(), introduced with 1690 * TCP state changes, is not quite right, but for 1691 * now, better than nothing. 1692 */ 1693 if (inp->inp_flags & INP_TIMEWAIT) { 1694 if (intotw(inp) != NULL) 1695 error = cr_cansee(req->td->td_ucred, 1696 intotw(inp)->tw_cred); 1697 else 1698 error = EINVAL; /* Skip this inp. */ 1699 } else 1700 error = cr_canseeinpcb(req->td->td_ucred, inp); 1701 if (error == 0) { 1702 in_pcbref(inp); 1703 inp_list[i++] = inp; 1704 } 1705 } 1706 INP_WUNLOCK(inp); 1707 } 1708 INP_INFO_WUNLOCK(&V_tcbinfo); 1709 n = i; 1710 1711 error = 0; 1712 for (i = 0; i < n; i++) { 1713 inp = inp_list[i]; 1714 INP_RLOCK(inp); 1715 if (inp->inp_gencnt <= gencnt) { 1716 struct xtcpcb xt; 1717 void *inp_ppcb; 1718 1719 bzero(&xt, sizeof(xt)); 1720 xt.xt_len = sizeof xt; 1721 /* XXX should avoid extra copy */ 1722 bcopy(inp, &xt.xt_inp, sizeof *inp); 1723 inp_ppcb = inp->inp_ppcb; 1724 if (inp_ppcb == NULL) 1725 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1726 else if (inp->inp_flags & INP_TIMEWAIT) { 1727 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1728 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1729 } else { 1730 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1731 if (xt.xt_tp.t_timers) 1732 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1733 } 1734 if (inp->inp_socket != NULL) 1735 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1736 else { 1737 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1738 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1739 } 1740 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1741 INP_RUNLOCK(inp); 1742 error = SYSCTL_OUT(req, &xt, sizeof xt); 1743 } else 1744 INP_RUNLOCK(inp); 1745 } 1746 INP_INFO_RLOCK(&V_tcbinfo); 1747 for (i = 0; i < n; i++) { 1748 inp = inp_list[i]; 1749 INP_RLOCK(inp); 1750 if (!in_pcbrele_rlocked(inp)) 1751 INP_RUNLOCK(inp); 1752 } 1753 INP_INFO_RUNLOCK(&V_tcbinfo); 1754 1755 if (!error) { 1756 /* 1757 * Give the user an updated idea of our state. 1758 * If the generation differs from what we told 1759 * her before, she knows that something happened 1760 * while we were processing this request, and it 1761 * might be necessary to retry. 1762 */ 1763 INP_LIST_RLOCK(&V_tcbinfo); 1764 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1765 xig.xig_sogen = so_gencnt; 1766 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1767 INP_LIST_RUNLOCK(&V_tcbinfo); 1768 error = SYSCTL_OUT(req, &xig, sizeof xig); 1769 } 1770 free(inp_list, M_TEMP); 1771 return (error); 1772 } 1773 1774 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1775 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1776 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1777 1778 #ifdef INET 1779 static int 1780 tcp_getcred(SYSCTL_HANDLER_ARGS) 1781 { 1782 struct xucred xuc; 1783 struct sockaddr_in addrs[2]; 1784 struct inpcb *inp; 1785 int error; 1786 1787 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1788 if (error) 1789 return (error); 1790 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1791 if (error) 1792 return (error); 1793 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1794 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1795 if (inp != NULL) { 1796 if (inp->inp_socket == NULL) 1797 error = ENOENT; 1798 if (error == 0) 1799 error = cr_canseeinpcb(req->td->td_ucred, inp); 1800 if (error == 0) 1801 cru2x(inp->inp_cred, &xuc); 1802 INP_RUNLOCK(inp); 1803 } else 1804 error = ENOENT; 1805 if (error == 0) 1806 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1807 return (error); 1808 } 1809 1810 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1811 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1812 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1813 #endif /* INET */ 1814 1815 #ifdef INET6 1816 static int 1817 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1818 { 1819 struct xucred xuc; 1820 struct sockaddr_in6 addrs[2]; 1821 struct inpcb *inp; 1822 int error; 1823 #ifdef INET 1824 int mapped = 0; 1825 #endif 1826 1827 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1828 if (error) 1829 return (error); 1830 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1831 if (error) 1832 return (error); 1833 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1834 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1835 return (error); 1836 } 1837 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1838 #ifdef INET 1839 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1840 mapped = 1; 1841 else 1842 #endif 1843 return (EINVAL); 1844 } 1845 1846 #ifdef INET 1847 if (mapped == 1) 1848 inp = in_pcblookup(&V_tcbinfo, 1849 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1850 addrs[1].sin6_port, 1851 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1852 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1853 else 1854 #endif 1855 inp = in6_pcblookup(&V_tcbinfo, 1856 &addrs[1].sin6_addr, addrs[1].sin6_port, 1857 &addrs[0].sin6_addr, addrs[0].sin6_port, 1858 INPLOOKUP_RLOCKPCB, NULL); 1859 if (inp != NULL) { 1860 if (inp->inp_socket == NULL) 1861 error = ENOENT; 1862 if (error == 0) 1863 error = cr_canseeinpcb(req->td->td_ucred, inp); 1864 if (error == 0) 1865 cru2x(inp->inp_cred, &xuc); 1866 INP_RUNLOCK(inp); 1867 } else 1868 error = ENOENT; 1869 if (error == 0) 1870 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1871 return (error); 1872 } 1873 1874 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1875 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1876 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1877 #endif /* INET6 */ 1878 1879 1880 #ifdef INET 1881 void 1882 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1883 { 1884 struct ip *ip = vip; 1885 struct tcphdr *th; 1886 struct in_addr faddr; 1887 struct inpcb *inp; 1888 struct tcpcb *tp; 1889 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1890 struct icmp *icp; 1891 struct in_conninfo inc; 1892 tcp_seq icmp_tcp_seq; 1893 int mtu; 1894 1895 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1896 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1897 return; 1898 1899 if (cmd == PRC_MSGSIZE) 1900 notify = tcp_mtudisc_notify; 1901 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1902 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1903 notify = tcp_drop_syn_sent; 1904 else if (PRC_IS_REDIRECT(cmd)) { 1905 /* signal EHOSTDOWN, as it flushes the cached route */ 1906 in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify); 1907 return; 1908 } 1909 /* 1910 * Hostdead is ugly because it goes linearly through all PCBs. 1911 * XXX: We never get this from ICMP, otherwise it makes an 1912 * excellent DoS attack on machines with many connections. 1913 */ 1914 else if (cmd == PRC_HOSTDEAD) 1915 ip = NULL; 1916 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1917 return; 1918 1919 if (ip == NULL) { 1920 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1921 return; 1922 } 1923 1924 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1925 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1926 INP_INFO_RLOCK(&V_tcbinfo); 1927 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 1928 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1929 if (inp != NULL) { 1930 if (!(inp->inp_flags & INP_TIMEWAIT) && 1931 !(inp->inp_flags & INP_DROPPED) && 1932 !(inp->inp_socket == NULL)) { 1933 icmp_tcp_seq = ntohl(th->th_seq); 1934 tp = intotcpcb(inp); 1935 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1936 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1937 if (cmd == PRC_MSGSIZE) { 1938 /* 1939 * MTU discovery: 1940 * If we got a needfrag set the MTU 1941 * in the route to the suggested new 1942 * value (if given) and then notify. 1943 */ 1944 mtu = ntohs(icp->icmp_nextmtu); 1945 /* 1946 * If no alternative MTU was 1947 * proposed, try the next smaller 1948 * one. 1949 */ 1950 if (!mtu) 1951 mtu = ip_next_mtu( 1952 ntohs(ip->ip_len), 1); 1953 if (mtu < V_tcp_minmss + 1954 sizeof(struct tcpiphdr)) 1955 mtu = V_tcp_minmss + 1956 sizeof(struct tcpiphdr); 1957 /* 1958 * Only process the offered MTU if it 1959 * is smaller than the current one. 1960 */ 1961 if (mtu < tp->t_maxseg + 1962 sizeof(struct tcpiphdr)) { 1963 bzero(&inc, sizeof(inc)); 1964 inc.inc_faddr = faddr; 1965 inc.inc_fibnum = 1966 inp->inp_inc.inc_fibnum; 1967 tcp_hc_updatemtu(&inc, mtu); 1968 tcp_mtudisc(inp, mtu); 1969 } 1970 } else 1971 inp = (*notify)(inp, 1972 inetctlerrmap[cmd]); 1973 } 1974 } 1975 if (inp != NULL) 1976 INP_WUNLOCK(inp); 1977 } else { 1978 bzero(&inc, sizeof(inc)); 1979 inc.inc_fport = th->th_dport; 1980 inc.inc_lport = th->th_sport; 1981 inc.inc_faddr = faddr; 1982 inc.inc_laddr = ip->ip_src; 1983 syncache_unreach(&inc, th); 1984 } 1985 INP_INFO_RUNLOCK(&V_tcbinfo); 1986 } 1987 #endif /* INET */ 1988 1989 #ifdef INET6 1990 void 1991 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1992 { 1993 struct tcphdr th; 1994 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1995 struct ip6_hdr *ip6; 1996 struct mbuf *m; 1997 struct ip6ctlparam *ip6cp = NULL; 1998 const struct sockaddr_in6 *sa6_src = NULL; 1999 int off; 2000 struct tcp_portonly { 2001 u_int16_t th_sport; 2002 u_int16_t th_dport; 2003 } *thp; 2004 2005 if (sa->sa_family != AF_INET6 || 2006 sa->sa_len != sizeof(struct sockaddr_in6)) 2007 return; 2008 2009 if (cmd == PRC_MSGSIZE) 2010 notify = tcp_mtudisc_notify; 2011 else if (!PRC_IS_REDIRECT(cmd) && 2012 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 2013 return; 2014 2015 /* if the parameter is from icmp6, decode it. */ 2016 if (d != NULL) { 2017 ip6cp = (struct ip6ctlparam *)d; 2018 m = ip6cp->ip6c_m; 2019 ip6 = ip6cp->ip6c_ip6; 2020 off = ip6cp->ip6c_off; 2021 sa6_src = ip6cp->ip6c_src; 2022 } else { 2023 m = NULL; 2024 ip6 = NULL; 2025 off = 0; /* fool gcc */ 2026 sa6_src = &sa6_any; 2027 } 2028 2029 if (ip6 != NULL) { 2030 struct in_conninfo inc; 2031 /* 2032 * XXX: We assume that when IPV6 is non NULL, 2033 * M and OFF are valid. 2034 */ 2035 2036 /* check if we can safely examine src and dst ports */ 2037 if (m->m_pkthdr.len < off + sizeof(*thp)) 2038 return; 2039 2040 bzero(&th, sizeof(th)); 2041 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 2042 2043 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 2044 (struct sockaddr *)ip6cp->ip6c_src, 2045 th.th_sport, cmd, NULL, notify); 2046 2047 bzero(&inc, sizeof(inc)); 2048 inc.inc_fport = th.th_dport; 2049 inc.inc_lport = th.th_sport; 2050 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 2051 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 2052 inc.inc_flags |= INC_ISIPV6; 2053 INP_INFO_RLOCK(&V_tcbinfo); 2054 syncache_unreach(&inc, &th); 2055 INP_INFO_RUNLOCK(&V_tcbinfo); 2056 } else 2057 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 2058 0, cmd, NULL, notify); 2059 } 2060 #endif /* INET6 */ 2061 2062 2063 /* 2064 * Following is where TCP initial sequence number generation occurs. 2065 * 2066 * There are two places where we must use initial sequence numbers: 2067 * 1. In SYN-ACK packets. 2068 * 2. In SYN packets. 2069 * 2070 * All ISNs for SYN-ACK packets are generated by the syncache. See 2071 * tcp_syncache.c for details. 2072 * 2073 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2074 * depends on this property. In addition, these ISNs should be 2075 * unguessable so as to prevent connection hijacking. To satisfy 2076 * the requirements of this situation, the algorithm outlined in 2077 * RFC 1948 is used, with only small modifications. 2078 * 2079 * Implementation details: 2080 * 2081 * Time is based off the system timer, and is corrected so that it 2082 * increases by one megabyte per second. This allows for proper 2083 * recycling on high speed LANs while still leaving over an hour 2084 * before rollover. 2085 * 2086 * As reading the *exact* system time is too expensive to be done 2087 * whenever setting up a TCP connection, we increment the time 2088 * offset in two ways. First, a small random positive increment 2089 * is added to isn_offset for each connection that is set up. 2090 * Second, the function tcp_isn_tick fires once per clock tick 2091 * and increments isn_offset as necessary so that sequence numbers 2092 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2093 * random positive increments serve only to ensure that the same 2094 * exact sequence number is never sent out twice (as could otherwise 2095 * happen when a port is recycled in less than the system tick 2096 * interval.) 2097 * 2098 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2099 * between seeding of isn_secret. This is normally set to zero, 2100 * as reseeding should not be necessary. 2101 * 2102 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2103 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2104 * general, this means holding an exclusive (write) lock. 2105 */ 2106 2107 #define ISN_BYTES_PER_SECOND 1048576 2108 #define ISN_STATIC_INCREMENT 4096 2109 #define ISN_RANDOM_INCREMENT (4096 - 1) 2110 2111 static VNET_DEFINE(u_char, isn_secret[32]); 2112 static VNET_DEFINE(int, isn_last); 2113 static VNET_DEFINE(int, isn_last_reseed); 2114 static VNET_DEFINE(u_int32_t, isn_offset); 2115 static VNET_DEFINE(u_int32_t, isn_offset_old); 2116 2117 #define V_isn_secret VNET(isn_secret) 2118 #define V_isn_last VNET(isn_last) 2119 #define V_isn_last_reseed VNET(isn_last_reseed) 2120 #define V_isn_offset VNET(isn_offset) 2121 #define V_isn_offset_old VNET(isn_offset_old) 2122 2123 tcp_seq 2124 tcp_new_isn(struct tcpcb *tp) 2125 { 2126 MD5_CTX isn_ctx; 2127 u_int32_t md5_buffer[4]; 2128 tcp_seq new_isn; 2129 u_int32_t projected_offset; 2130 2131 INP_WLOCK_ASSERT(tp->t_inpcb); 2132 2133 ISN_LOCK(); 2134 /* Seed if this is the first use, reseed if requested. */ 2135 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2136 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2137 < (u_int)ticks))) { 2138 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2139 V_isn_last_reseed = ticks; 2140 } 2141 2142 /* Compute the md5 hash and return the ISN. */ 2143 MD5Init(&isn_ctx); 2144 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2145 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2146 #ifdef INET6 2147 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2148 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2149 sizeof(struct in6_addr)); 2150 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2151 sizeof(struct in6_addr)); 2152 } else 2153 #endif 2154 { 2155 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2156 sizeof(struct in_addr)); 2157 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2158 sizeof(struct in_addr)); 2159 } 2160 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2161 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2162 new_isn = (tcp_seq) md5_buffer[0]; 2163 V_isn_offset += ISN_STATIC_INCREMENT + 2164 (arc4random() & ISN_RANDOM_INCREMENT); 2165 if (ticks != V_isn_last) { 2166 projected_offset = V_isn_offset_old + 2167 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2168 if (SEQ_GT(projected_offset, V_isn_offset)) 2169 V_isn_offset = projected_offset; 2170 V_isn_offset_old = V_isn_offset; 2171 V_isn_last = ticks; 2172 } 2173 new_isn += V_isn_offset; 2174 ISN_UNLOCK(); 2175 return (new_isn); 2176 } 2177 2178 /* 2179 * When a specific ICMP unreachable message is received and the 2180 * connection state is SYN-SENT, drop the connection. This behavior 2181 * is controlled by the icmp_may_rst sysctl. 2182 */ 2183 struct inpcb * 2184 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2185 { 2186 struct tcpcb *tp; 2187 2188 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2189 INP_WLOCK_ASSERT(inp); 2190 2191 if ((inp->inp_flags & INP_TIMEWAIT) || 2192 (inp->inp_flags & INP_DROPPED)) 2193 return (inp); 2194 2195 tp = intotcpcb(inp); 2196 if (tp->t_state != TCPS_SYN_SENT) 2197 return (inp); 2198 2199 tp = tcp_drop(tp, errno); 2200 if (tp != NULL) 2201 return (inp); 2202 else 2203 return (NULL); 2204 } 2205 2206 /* 2207 * When `need fragmentation' ICMP is received, update our idea of the MSS 2208 * based on the new value. Also nudge TCP to send something, since we 2209 * know the packet we just sent was dropped. 2210 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2211 */ 2212 static struct inpcb * 2213 tcp_mtudisc_notify(struct inpcb *inp, int error) 2214 { 2215 2216 tcp_mtudisc(inp, -1); 2217 return (inp); 2218 } 2219 2220 static void 2221 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2222 { 2223 struct tcpcb *tp; 2224 struct socket *so; 2225 2226 INP_WLOCK_ASSERT(inp); 2227 if ((inp->inp_flags & INP_TIMEWAIT) || 2228 (inp->inp_flags & INP_DROPPED)) 2229 return; 2230 2231 tp = intotcpcb(inp); 2232 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2233 2234 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2235 2236 so = inp->inp_socket; 2237 SOCKBUF_LOCK(&so->so_snd); 2238 /* If the mss is larger than the socket buffer, decrease the mss. */ 2239 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2240 tp->t_maxseg = so->so_snd.sb_hiwat; 2241 SOCKBUF_UNLOCK(&so->so_snd); 2242 2243 TCPSTAT_INC(tcps_mturesent); 2244 tp->t_rtttime = 0; 2245 tp->snd_nxt = tp->snd_una; 2246 tcp_free_sackholes(tp); 2247 tp->snd_recover = tp->snd_max; 2248 if (tp->t_flags & TF_SACK_PERMIT) 2249 EXIT_FASTRECOVERY(tp->t_flags); 2250 tp->t_fb->tfb_tcp_output(tp); 2251 } 2252 2253 #ifdef INET 2254 /* 2255 * Look-up the routing entry to the peer of this inpcb. If no route 2256 * is found and it cannot be allocated, then return 0. This routine 2257 * is called by TCP routines that access the rmx structure and by 2258 * tcp_mss_update to get the peer/interface MTU. 2259 */ 2260 u_long 2261 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2262 { 2263 struct nhop4_extended nh4; 2264 struct ifnet *ifp; 2265 u_long maxmtu = 0; 2266 2267 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2268 2269 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2270 2271 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2272 NHR_REF, 0, &nh4) != 0) 2273 return (0); 2274 2275 ifp = nh4.nh_ifp; 2276 maxmtu = nh4.nh_mtu; 2277 2278 /* Report additional interface capabilities. */ 2279 if (cap != NULL) { 2280 if (ifp->if_capenable & IFCAP_TSO4 && 2281 ifp->if_hwassist & CSUM_TSO) { 2282 cap->ifcap |= CSUM_TSO; 2283 cap->tsomax = ifp->if_hw_tsomax; 2284 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2285 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2286 } 2287 } 2288 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2289 } 2290 return (maxmtu); 2291 } 2292 #endif /* INET */ 2293 2294 #ifdef INET6 2295 u_long 2296 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2297 { 2298 struct nhop6_extended nh6; 2299 struct in6_addr dst6; 2300 uint32_t scopeid; 2301 struct ifnet *ifp; 2302 u_long maxmtu = 0; 2303 2304 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2305 2306 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2307 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2308 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2309 0, &nh6) != 0) 2310 return (0); 2311 2312 ifp = nh6.nh_ifp; 2313 maxmtu = nh6.nh_mtu; 2314 2315 /* Report additional interface capabilities. */ 2316 if (cap != NULL) { 2317 if (ifp->if_capenable & IFCAP_TSO6 && 2318 ifp->if_hwassist & CSUM_TSO) { 2319 cap->ifcap |= CSUM_TSO; 2320 cap->tsomax = ifp->if_hw_tsomax; 2321 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2322 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2323 } 2324 } 2325 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2326 } 2327 2328 return (maxmtu); 2329 } 2330 #endif /* INET6 */ 2331 2332 /* 2333 * Calculate effective SMSS per RFC5681 definition for a given TCP 2334 * connection at its current state, taking into account SACK and etc. 2335 */ 2336 u_int 2337 tcp_maxseg(const struct tcpcb *tp) 2338 { 2339 u_int optlen; 2340 2341 if (tp->t_flags & TF_NOOPT) 2342 return (tp->t_maxseg); 2343 2344 /* 2345 * Here we have a simplified code from tcp_addoptions(), 2346 * without a proper loop, and having most of paddings hardcoded. 2347 * We might make mistakes with padding here in some edge cases, 2348 * but this is harmless, since result of tcp_maxseg() is used 2349 * only in cwnd and ssthresh estimations. 2350 */ 2351 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2352 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2353 if (tp->t_flags & TF_RCVD_TSTMP) 2354 optlen = TCPOLEN_TSTAMP_APPA; 2355 else 2356 optlen = 0; 2357 #ifdef TCP_SIGNATURE 2358 if (tp->t_flags & TF_SIGNATURE) 2359 optlen += PAD(TCPOLEN_SIGNATURE); 2360 #endif 2361 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2362 optlen += TCPOLEN_SACKHDR; 2363 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2364 optlen = PAD(optlen); 2365 } 2366 } else { 2367 if (tp->t_flags & TF_REQ_TSTMP) 2368 optlen = TCPOLEN_TSTAMP_APPA; 2369 else 2370 optlen = PAD(TCPOLEN_MAXSEG); 2371 if (tp->t_flags & TF_REQ_SCALE) 2372 optlen += PAD(TCPOLEN_WINDOW); 2373 #ifdef TCP_SIGNATURE 2374 if (tp->t_flags & TF_SIGNATURE) 2375 optlen += PAD(TCPOLEN_SIGNATURE); 2376 #endif 2377 if (tp->t_flags & TF_SACK_PERMIT) 2378 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2379 } 2380 #undef PAD 2381 optlen = min(optlen, TCP_MAXOLEN); 2382 return (tp->t_maxseg - optlen); 2383 } 2384 2385 #ifdef IPSEC 2386 /* compute ESP/AH header size for TCP, including outer IP header. */ 2387 size_t 2388 ipsec_hdrsiz_tcp(struct tcpcb *tp) 2389 { 2390 struct inpcb *inp; 2391 struct mbuf *m; 2392 size_t hdrsiz; 2393 struct ip *ip; 2394 #ifdef INET6 2395 struct ip6_hdr *ip6; 2396 #endif 2397 struct tcphdr *th; 2398 2399 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || 2400 (!key_havesp(IPSEC_DIR_OUTBOUND))) 2401 return (0); 2402 m = m_gethdr(M_NOWAIT, MT_DATA); 2403 if (!m) 2404 return (0); 2405 2406 #ifdef INET6 2407 if ((inp->inp_vflag & INP_IPV6) != 0) { 2408 ip6 = mtod(m, struct ip6_hdr *); 2409 th = (struct tcphdr *)(ip6 + 1); 2410 m->m_pkthdr.len = m->m_len = 2411 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2412 tcpip_fillheaders(inp, ip6, th); 2413 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2414 } else 2415 #endif /* INET6 */ 2416 { 2417 ip = mtod(m, struct ip *); 2418 th = (struct tcphdr *)(ip + 1); 2419 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 2420 tcpip_fillheaders(inp, ip, th); 2421 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2422 } 2423 2424 m_free(m); 2425 return (hdrsiz); 2426 } 2427 #endif /* IPSEC */ 2428 2429 #ifdef TCP_SIGNATURE 2430 /* 2431 * Callback function invoked by m_apply() to digest TCP segment data 2432 * contained within an mbuf chain. 2433 */ 2434 static int 2435 tcp_signature_apply(void *fstate, void *data, u_int len) 2436 { 2437 2438 MD5Update(fstate, (u_char *)data, len); 2439 return (0); 2440 } 2441 2442 /* 2443 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2444 * search with the destination IP address, and a 'magic SPI' to be 2445 * determined by the application. This is hardcoded elsewhere to 1179 2446 */ 2447 struct secasvar * 2448 tcp_get_sav(struct mbuf *m, u_int direction) 2449 { 2450 union sockaddr_union dst; 2451 struct secasvar *sav; 2452 struct ip *ip; 2453 #ifdef INET6 2454 struct ip6_hdr *ip6; 2455 char ip6buf[INET6_ADDRSTRLEN]; 2456 #endif 2457 2458 /* Extract the destination from the IP header in the mbuf. */ 2459 bzero(&dst, sizeof(union sockaddr_union)); 2460 ip = mtod(m, struct ip *); 2461 #ifdef INET6 2462 ip6 = NULL; /* Make the compiler happy. */ 2463 #endif 2464 switch (ip->ip_v) { 2465 #ifdef INET 2466 case IPVERSION: 2467 dst.sa.sa_len = sizeof(struct sockaddr_in); 2468 dst.sa.sa_family = AF_INET; 2469 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2470 ip->ip_src : ip->ip_dst; 2471 break; 2472 #endif 2473 #ifdef INET6 2474 case (IPV6_VERSION >> 4): 2475 ip6 = mtod(m, struct ip6_hdr *); 2476 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2477 dst.sa.sa_family = AF_INET6; 2478 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 2479 ip6->ip6_src : ip6->ip6_dst; 2480 break; 2481 #endif 2482 default: 2483 return (NULL); 2484 /* NOTREACHED */ 2485 break; 2486 } 2487 2488 /* Look up an SADB entry which matches the address of the peer. */ 2489 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2490 if (sav == NULL) { 2491 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 2492 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 2493 #ifdef INET6 2494 (ip->ip_v == (IPV6_VERSION >> 4)) ? 2495 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 2496 #endif 2497 "(unsupported)")); 2498 } 2499 2500 return (sav); 2501 } 2502 2503 /* 2504 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2505 * 2506 * Parameters: 2507 * m pointer to head of mbuf chain 2508 * len length of TCP segment data, excluding options 2509 * optlen length of TCP segment options 2510 * buf pointer to storage for computed MD5 digest 2511 * sav pointer to security assosiation 2512 * 2513 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2514 * When called from tcp_input(), we can be sure that th_sum has been 2515 * zeroed out and verified already. 2516 * 2517 * Releases reference to SADB key before return. 2518 * 2519 * Return 0 if successful, otherwise return -1. 2520 * 2521 */ 2522 int 2523 tcp_signature_do_compute(struct mbuf *m, int len, int optlen, 2524 u_char *buf, struct secasvar *sav) 2525 { 2526 #ifdef INET 2527 struct ippseudo ippseudo; 2528 #endif 2529 MD5_CTX ctx; 2530 int doff; 2531 struct ip *ip; 2532 #ifdef INET 2533 struct ipovly *ipovly; 2534 #endif 2535 struct tcphdr *th; 2536 #ifdef INET6 2537 struct ip6_hdr *ip6; 2538 struct in6_addr in6; 2539 uint32_t plen; 2540 uint16_t nhdr; 2541 #endif 2542 u_short savecsum; 2543 2544 KASSERT(m != NULL, ("NULL mbuf chain")); 2545 KASSERT(buf != NULL, ("NULL signature pointer")); 2546 2547 /* Extract the destination from the IP header in the mbuf. */ 2548 ip = mtod(m, struct ip *); 2549 #ifdef INET6 2550 ip6 = NULL; /* Make the compiler happy. */ 2551 #endif 2552 2553 MD5Init(&ctx); 2554 /* 2555 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 2556 * 2557 * XXX The ippseudo header MUST be digested in network byte order, 2558 * or else we'll fail the regression test. Assume all fields we've 2559 * been doing arithmetic on have been in host byte order. 2560 * XXX One cannot depend on ipovly->ih_len here. When called from 2561 * tcp_output(), the underlying ip_len member has not yet been set. 2562 */ 2563 switch (ip->ip_v) { 2564 #ifdef INET 2565 case IPVERSION: 2566 ipovly = (struct ipovly *)ip; 2567 ippseudo.ippseudo_src = ipovly->ih_src; 2568 ippseudo.ippseudo_dst = ipovly->ih_dst; 2569 ippseudo.ippseudo_pad = 0; 2570 ippseudo.ippseudo_p = IPPROTO_TCP; 2571 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 2572 optlen); 2573 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2574 2575 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2576 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2577 break; 2578 #endif 2579 #ifdef INET6 2580 /* 2581 * RFC 2385, 2.0 Proposal 2582 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2583 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2584 * extended next header value (to form 32 bits), and 32-bit segment 2585 * length. 2586 * Note: Upper-Layer Packet Length comes before Next Header. 2587 */ 2588 case (IPV6_VERSION >> 4): 2589 in6 = ip6->ip6_src; 2590 in6_clearscope(&in6); 2591 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2592 in6 = ip6->ip6_dst; 2593 in6_clearscope(&in6); 2594 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2595 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2596 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2597 nhdr = 0; 2598 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2599 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2600 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2601 nhdr = IPPROTO_TCP; 2602 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2603 2604 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2605 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2606 break; 2607 #endif 2608 default: 2609 KEY_FREESAV(&sav); 2610 return (-1); 2611 /* NOTREACHED */ 2612 break; 2613 } 2614 2615 2616 /* 2617 * Step 2: Update MD5 hash with TCP header, excluding options. 2618 * The TCP checksum must be set to zero. 2619 */ 2620 savecsum = th->th_sum; 2621 th->th_sum = 0; 2622 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2623 th->th_sum = savecsum; 2624 2625 /* 2626 * Step 3: Update MD5 hash with TCP segment data. 2627 * Use m_apply() to avoid an early m_pullup(). 2628 */ 2629 if (len > 0) 2630 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2631 2632 /* 2633 * Step 4: Update MD5 hash with shared secret. 2634 */ 2635 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2636 MD5Final(buf, &ctx); 2637 2638 key_sa_recordxfer(sav, m); 2639 KEY_FREESAV(&sav); 2640 return (0); 2641 } 2642 2643 /* 2644 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2645 * 2646 * Return 0 if successful, otherwise return -1. 2647 */ 2648 int 2649 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 2650 u_char *buf, u_int direction) 2651 { 2652 struct secasvar *sav; 2653 2654 if ((sav = tcp_get_sav(m, direction)) == NULL) 2655 return (-1); 2656 2657 return (tcp_signature_do_compute(m, len, optlen, buf, sav)); 2658 } 2659 2660 /* 2661 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2662 * 2663 * Parameters: 2664 * m pointer to head of mbuf chain 2665 * len length of TCP segment data, excluding options 2666 * optlen length of TCP segment options 2667 * buf pointer to storage for computed MD5 digest 2668 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2669 * 2670 * Return 1 if successful, otherwise return 0. 2671 */ 2672 int 2673 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2674 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2675 { 2676 char tmpdigest[TCP_SIGLEN]; 2677 2678 if (tcp_sig_checksigs == 0) 2679 return (1); 2680 if ((tcpbflag & TF_SIGNATURE) == 0) { 2681 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2682 2683 /* 2684 * If this socket is not expecting signature but 2685 * the segment contains signature just fail. 2686 */ 2687 TCPSTAT_INC(tcps_sig_err_sigopt); 2688 TCPSTAT_INC(tcps_sig_rcvbadsig); 2689 return (0); 2690 } 2691 2692 /* Signature is not expected, and not present in segment. */ 2693 return (1); 2694 } 2695 2696 /* 2697 * If this socket is expecting signature but the segment does not 2698 * contain any just fail. 2699 */ 2700 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2701 TCPSTAT_INC(tcps_sig_err_nosigopt); 2702 TCPSTAT_INC(tcps_sig_rcvbadsig); 2703 return (0); 2704 } 2705 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2706 IPSEC_DIR_INBOUND) == -1) { 2707 TCPSTAT_INC(tcps_sig_err_buildsig); 2708 TCPSTAT_INC(tcps_sig_rcvbadsig); 2709 return (0); 2710 } 2711 2712 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2713 TCPSTAT_INC(tcps_sig_rcvbadsig); 2714 return (0); 2715 } 2716 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2717 return (1); 2718 } 2719 #endif /* TCP_SIGNATURE */ 2720 2721 static int 2722 sysctl_drop(SYSCTL_HANDLER_ARGS) 2723 { 2724 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2725 struct sockaddr_storage addrs[2]; 2726 struct inpcb *inp; 2727 struct tcpcb *tp; 2728 struct tcptw *tw; 2729 struct sockaddr_in *fin, *lin; 2730 #ifdef INET6 2731 struct sockaddr_in6 *fin6, *lin6; 2732 #endif 2733 int error; 2734 2735 inp = NULL; 2736 fin = lin = NULL; 2737 #ifdef INET6 2738 fin6 = lin6 = NULL; 2739 #endif 2740 error = 0; 2741 2742 if (req->oldptr != NULL || req->oldlen != 0) 2743 return (EINVAL); 2744 if (req->newptr == NULL) 2745 return (EPERM); 2746 if (req->newlen < sizeof(addrs)) 2747 return (ENOMEM); 2748 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2749 if (error) 2750 return (error); 2751 2752 switch (addrs[0].ss_family) { 2753 #ifdef INET6 2754 case AF_INET6: 2755 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2756 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2757 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2758 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2759 return (EINVAL); 2760 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2761 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2762 return (EINVAL); 2763 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2764 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2765 fin = (struct sockaddr_in *)&addrs[0]; 2766 lin = (struct sockaddr_in *)&addrs[1]; 2767 break; 2768 } 2769 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2770 if (error) 2771 return (error); 2772 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2773 if (error) 2774 return (error); 2775 break; 2776 #endif 2777 #ifdef INET 2778 case AF_INET: 2779 fin = (struct sockaddr_in *)&addrs[0]; 2780 lin = (struct sockaddr_in *)&addrs[1]; 2781 if (fin->sin_len != sizeof(struct sockaddr_in) || 2782 lin->sin_len != sizeof(struct sockaddr_in)) 2783 return (EINVAL); 2784 break; 2785 #endif 2786 default: 2787 return (EINVAL); 2788 } 2789 INP_INFO_RLOCK(&V_tcbinfo); 2790 switch (addrs[0].ss_family) { 2791 #ifdef INET6 2792 case AF_INET6: 2793 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2794 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2795 INPLOOKUP_WLOCKPCB, NULL); 2796 break; 2797 #endif 2798 #ifdef INET 2799 case AF_INET: 2800 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2801 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2802 break; 2803 #endif 2804 } 2805 if (inp != NULL) { 2806 if (inp->inp_flags & INP_TIMEWAIT) { 2807 /* 2808 * XXXRW: There currently exists a state where an 2809 * inpcb is present, but its timewait state has been 2810 * discarded. For now, don't allow dropping of this 2811 * type of inpcb. 2812 */ 2813 tw = intotw(inp); 2814 if (tw != NULL) 2815 tcp_twclose(tw, 0); 2816 else 2817 INP_WUNLOCK(inp); 2818 } else if (!(inp->inp_flags & INP_DROPPED) && 2819 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2820 tp = intotcpcb(inp); 2821 tp = tcp_drop(tp, ECONNABORTED); 2822 if (tp != NULL) 2823 INP_WUNLOCK(inp); 2824 } else 2825 INP_WUNLOCK(inp); 2826 } else 2827 error = ESRCH; 2828 INP_INFO_RUNLOCK(&V_tcbinfo); 2829 return (error); 2830 } 2831 2832 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2833 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2834 0, sysctl_drop, "", "Drop TCP connection"); 2835 2836 /* 2837 * Generate a standardized TCP log line for use throughout the 2838 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2839 * allow use in the interrupt context. 2840 * 2841 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2842 * NB: The function may return NULL if memory allocation failed. 2843 * 2844 * Due to header inclusion and ordering limitations the struct ip 2845 * and ip6_hdr pointers have to be passed as void pointers. 2846 */ 2847 char * 2848 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2849 const void *ip6hdr) 2850 { 2851 2852 /* Is logging enabled? */ 2853 if (tcp_log_in_vain == 0) 2854 return (NULL); 2855 2856 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2857 } 2858 2859 char * 2860 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2861 const void *ip6hdr) 2862 { 2863 2864 /* Is logging enabled? */ 2865 if (tcp_log_debug == 0) 2866 return (NULL); 2867 2868 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2869 } 2870 2871 static char * 2872 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2873 const void *ip6hdr) 2874 { 2875 char *s, *sp; 2876 size_t size; 2877 struct ip *ip; 2878 #ifdef INET6 2879 const struct ip6_hdr *ip6; 2880 2881 ip6 = (const struct ip6_hdr *)ip6hdr; 2882 #endif /* INET6 */ 2883 ip = (struct ip *)ip4hdr; 2884 2885 /* 2886 * The log line looks like this: 2887 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2888 */ 2889 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2890 sizeof(PRINT_TH_FLAGS) + 1 + 2891 #ifdef INET6 2892 2 * INET6_ADDRSTRLEN; 2893 #else 2894 2 * INET_ADDRSTRLEN; 2895 #endif /* INET6 */ 2896 2897 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2898 if (s == NULL) 2899 return (NULL); 2900 2901 strcat(s, "TCP: ["); 2902 sp = s + strlen(s); 2903 2904 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2905 inet_ntoa_r(inc->inc_faddr, sp); 2906 sp = s + strlen(s); 2907 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2908 sp = s + strlen(s); 2909 inet_ntoa_r(inc->inc_laddr, sp); 2910 sp = s + strlen(s); 2911 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2912 #ifdef INET6 2913 } else if (inc) { 2914 ip6_sprintf(sp, &inc->inc6_faddr); 2915 sp = s + strlen(s); 2916 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2917 sp = s + strlen(s); 2918 ip6_sprintf(sp, &inc->inc6_laddr); 2919 sp = s + strlen(s); 2920 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2921 } else if (ip6 && th) { 2922 ip6_sprintf(sp, &ip6->ip6_src); 2923 sp = s + strlen(s); 2924 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2925 sp = s + strlen(s); 2926 ip6_sprintf(sp, &ip6->ip6_dst); 2927 sp = s + strlen(s); 2928 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2929 #endif /* INET6 */ 2930 #ifdef INET 2931 } else if (ip && th) { 2932 inet_ntoa_r(ip->ip_src, sp); 2933 sp = s + strlen(s); 2934 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2935 sp = s + strlen(s); 2936 inet_ntoa_r(ip->ip_dst, sp); 2937 sp = s + strlen(s); 2938 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2939 #endif /* INET */ 2940 } else { 2941 free(s, M_TCPLOG); 2942 return (NULL); 2943 } 2944 sp = s + strlen(s); 2945 if (th) 2946 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2947 if (*(s + size - 1) != '\0') 2948 panic("%s: string too long", __func__); 2949 return (s); 2950 } 2951 2952 /* 2953 * A subroutine which makes it easy to track TCP state changes with DTrace. 2954 * This function shouldn't be called for t_state initializations that don't 2955 * correspond to actual TCP state transitions. 2956 */ 2957 void 2958 tcp_state_change(struct tcpcb *tp, int newstate) 2959 { 2960 #if defined(KDTRACE_HOOKS) 2961 int pstate = tp->t_state; 2962 #endif 2963 2964 TCPSTATES_DEC(tp->t_state); 2965 TCPSTATES_INC(newstate); 2966 tp->t_state = newstate; 2967 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2968 } 2969