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 struct tcpopt to; 868 struct inpcb *inp; 869 struct ip *ip; 870 struct mbuf *optm; 871 struct tcphdr *nth; 872 u_char *optp; 873 #ifdef INET6 874 struct ip6_hdr *ip6; 875 int isipv6; 876 #endif /* INET6 */ 877 int optlen, tlen, win; 878 bool incl_opts; 879 880 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 881 882 #ifdef INET6 883 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 884 ip6 = ipgen; 885 #endif /* INET6 */ 886 ip = ipgen; 887 888 if (tp != NULL) { 889 inp = tp->t_inpcb; 890 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 891 INP_WLOCK_ASSERT(inp); 892 } else 893 inp = NULL; 894 895 incl_opts = false; 896 win = 0; 897 if (tp != NULL) { 898 if (!(flags & TH_RST)) { 899 win = sbspace(&inp->inp_socket->so_rcv); 900 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 901 win = (long)TCP_MAXWIN << tp->rcv_scale; 902 } 903 if ((tp->t_flags & TF_NOOPT) == 0) 904 incl_opts = true; 905 } 906 if (m == NULL) { 907 m = m_gethdr(M_NOWAIT, MT_DATA); 908 if (m == NULL) 909 return; 910 m->m_data += max_linkhdr; 911 #ifdef INET6 912 if (isipv6) { 913 bcopy((caddr_t)ip6, mtod(m, caddr_t), 914 sizeof(struct ip6_hdr)); 915 ip6 = mtod(m, struct ip6_hdr *); 916 nth = (struct tcphdr *)(ip6 + 1); 917 } else 918 #endif /* INET6 */ 919 { 920 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 921 ip = mtod(m, struct ip *); 922 nth = (struct tcphdr *)(ip + 1); 923 } 924 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 925 flags = TH_ACK; 926 } else { 927 /* 928 * reuse the mbuf. 929 * XXX MRT We inherrit the FIB, which is lucky. 930 */ 931 m_freem(m->m_next); 932 m->m_next = NULL; 933 m->m_data = (caddr_t)ipgen; 934 /* m_len is set later */ 935 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 936 #ifdef INET6 937 if (isipv6) { 938 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 939 nth = (struct tcphdr *)(ip6 + 1); 940 } else 941 #endif /* INET6 */ 942 { 943 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 944 nth = (struct tcphdr *)(ip + 1); 945 } 946 if (th != nth) { 947 /* 948 * this is usually a case when an extension header 949 * exists between the IPv6 header and the 950 * TCP header. 951 */ 952 nth->th_sport = th->th_sport; 953 nth->th_dport = th->th_dport; 954 } 955 xchg(nth->th_dport, nth->th_sport, uint16_t); 956 #undef xchg 957 } 958 tlen = 0; 959 #ifdef INET6 960 if (isipv6) 961 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 962 #endif 963 #if defined(INET) && defined(INET6) 964 else 965 #endif 966 #ifdef INET 967 tlen = sizeof (struct tcpiphdr); 968 #endif 969 #ifdef INVARIANTS 970 m->m_len = 0; 971 KASSERT(M_TRAILINGSPACE(m) >= tlen, 972 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 973 m, tlen, (long)M_TRAILINGSPACE(m))); 974 #endif 975 m->m_len = tlen; 976 to.to_flags = 0; 977 if (incl_opts) { 978 /* Make sure we have room. */ 979 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 980 m->m_next = m_get(M_NOWAIT, MT_DATA); 981 if (m->m_next) { 982 optp = mtod(m->m_next, u_char *); 983 optm = m->m_next; 984 } else 985 incl_opts = false; 986 } else { 987 optp = (u_char *) (nth + 1); 988 optm = m; 989 } 990 } 991 if (incl_opts) { 992 /* Timestamps. */ 993 if (tp->t_flags & TF_RCVD_TSTMP) { 994 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 995 to.to_tsecr = tp->ts_recent; 996 to.to_flags |= TOF_TS; 997 } 998 #ifdef TCP_SIGNATURE 999 /* TCP-MD5 (RFC2385). */ 1000 if (tp->t_flags & TF_SIGNATURE) 1001 to.to_flags |= TOF_SIGNATURE; 1002 #endif 1003 1004 /* Add the options. */ 1005 tlen += optlen = tcp_addoptions(&to, optp); 1006 1007 /* Update m_len in the correct mbuf. */ 1008 optm->m_len += optlen; 1009 } else 1010 optlen = 0; 1011 #ifdef INET6 1012 if (isipv6) { 1013 ip6->ip6_flow = 0; 1014 ip6->ip6_vfc = IPV6_VERSION; 1015 ip6->ip6_nxt = IPPROTO_TCP; 1016 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1017 } 1018 #endif 1019 #if defined(INET) && defined(INET6) 1020 else 1021 #endif 1022 #ifdef INET 1023 { 1024 ip->ip_len = htons(tlen); 1025 ip->ip_ttl = V_ip_defttl; 1026 if (V_path_mtu_discovery) 1027 ip->ip_off |= htons(IP_DF); 1028 } 1029 #endif 1030 m->m_pkthdr.len = tlen; 1031 m->m_pkthdr.rcvif = NULL; 1032 #ifdef MAC 1033 if (inp != NULL) { 1034 /* 1035 * Packet is associated with a socket, so allow the 1036 * label of the response to reflect the socket label. 1037 */ 1038 INP_WLOCK_ASSERT(inp); 1039 mac_inpcb_create_mbuf(inp, m); 1040 } else { 1041 /* 1042 * Packet is not associated with a socket, so possibly 1043 * update the label in place. 1044 */ 1045 mac_netinet_tcp_reply(m); 1046 } 1047 #endif 1048 nth->th_seq = htonl(seq); 1049 nth->th_ack = htonl(ack); 1050 nth->th_x2 = 0; 1051 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1052 nth->th_flags = flags; 1053 if (tp != NULL) 1054 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1055 else 1056 nth->th_win = htons((u_short)win); 1057 nth->th_urp = 0; 1058 1059 #ifdef TCP_SIGNATURE 1060 if (to.to_flags & TOF_SIGNATURE) { 1061 tcp_signature_compute(m, 0, 0, optlen, to.to_signature, 1062 IPSEC_DIR_OUTBOUND); 1063 } 1064 #endif 1065 1066 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1067 #ifdef INET6 1068 if (isipv6) { 1069 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1070 nth->th_sum = in6_cksum_pseudo(ip6, 1071 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1072 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1073 NULL, NULL); 1074 } 1075 #endif /* INET6 */ 1076 #if defined(INET6) && defined(INET) 1077 else 1078 #endif 1079 #ifdef INET 1080 { 1081 m->m_pkthdr.csum_flags = CSUM_TCP; 1082 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1083 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1084 } 1085 #endif /* INET */ 1086 #ifdef TCPDEBUG 1087 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1088 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1089 #endif 1090 TCP_PROBE3(debug__output, tp, th, mtod(m, const char *)); 1091 if (flags & TH_RST) 1092 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), 1093 tp, nth); 1094 1095 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); 1096 #ifdef INET6 1097 if (isipv6) 1098 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1099 #endif /* INET6 */ 1100 #if defined(INET) && defined(INET6) 1101 else 1102 #endif 1103 #ifdef INET 1104 (void) ip_output(m, NULL, NULL, 0, NULL, inp); 1105 #endif 1106 } 1107 1108 /* 1109 * Create a new TCP control block, making an 1110 * empty reassembly queue and hooking it to the argument 1111 * protocol control block. The `inp' parameter must have 1112 * come from the zone allocator set up in tcp_init(). 1113 */ 1114 struct tcpcb * 1115 tcp_newtcpcb(struct inpcb *inp) 1116 { 1117 struct tcpcb_mem *tm; 1118 struct tcpcb *tp; 1119 #ifdef INET6 1120 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1121 #endif /* INET6 */ 1122 1123 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1124 if (tm == NULL) 1125 return (NULL); 1126 tp = &tm->tcb; 1127 1128 /* Initialise cc_var struct for this tcpcb. */ 1129 tp->ccv = &tm->ccv; 1130 tp->ccv->type = IPPROTO_TCP; 1131 tp->ccv->ccvc.tcp = tp; 1132 rw_rlock(&tcp_function_lock); 1133 tp->t_fb = tcp_func_set_ptr; 1134 refcount_acquire(&tp->t_fb->tfb_refcnt); 1135 rw_runlock(&tcp_function_lock); 1136 if (tp->t_fb->tfb_tcp_fb_init) { 1137 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1138 } 1139 /* 1140 * Use the current system default CC algorithm. 1141 */ 1142 CC_LIST_RLOCK(); 1143 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1144 CC_ALGO(tp) = CC_DEFAULT(); 1145 CC_LIST_RUNLOCK(); 1146 1147 if (CC_ALGO(tp)->cb_init != NULL) 1148 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1149 if (tp->t_fb->tfb_tcp_fb_fini) 1150 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1151 refcount_release(&tp->t_fb->tfb_refcnt); 1152 uma_zfree(V_tcpcb_zone, tm); 1153 return (NULL); 1154 } 1155 1156 tp->osd = &tm->osd; 1157 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1158 if (tp->t_fb->tfb_tcp_fb_fini) 1159 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1160 refcount_release(&tp->t_fb->tfb_refcnt); 1161 uma_zfree(V_tcpcb_zone, tm); 1162 return (NULL); 1163 } 1164 1165 #ifdef VIMAGE 1166 tp->t_vnet = inp->inp_vnet; 1167 #endif 1168 tp->t_timers = &tm->tt; 1169 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1170 tp->t_maxseg = 1171 #ifdef INET6 1172 isipv6 ? V_tcp_v6mssdflt : 1173 #endif /* INET6 */ 1174 V_tcp_mssdflt; 1175 1176 /* Set up our timeouts. */ 1177 callout_init(&tp->t_timers->tt_rexmt, 1); 1178 callout_init(&tp->t_timers->tt_persist, 1); 1179 callout_init(&tp->t_timers->tt_keep, 1); 1180 callout_init(&tp->t_timers->tt_2msl, 1); 1181 callout_init(&tp->t_timers->tt_delack, 1); 1182 1183 if (V_tcp_do_rfc1323) 1184 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1185 if (V_tcp_do_sack) 1186 tp->t_flags |= TF_SACK_PERMIT; 1187 TAILQ_INIT(&tp->snd_holes); 1188 /* 1189 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1190 * is called. 1191 */ 1192 in_pcbref(inp); /* Reference for tcpcb */ 1193 tp->t_inpcb = inp; 1194 1195 /* 1196 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1197 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1198 * reasonable initial retransmit time. 1199 */ 1200 tp->t_srtt = TCPTV_SRTTBASE; 1201 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1202 tp->t_rttmin = tcp_rexmit_min; 1203 tp->t_rxtcur = TCPTV_RTOBASE; 1204 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1205 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1206 tp->t_rcvtime = ticks; 1207 /* 1208 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1209 * because the socket may be bound to an IPv6 wildcard address, 1210 * which may match an IPv4-mapped IPv6 address. 1211 */ 1212 inp->inp_ip_ttl = V_ip_defttl; 1213 inp->inp_ppcb = tp; 1214 #ifdef TCPPCAP 1215 /* 1216 * Init the TCP PCAP queues. 1217 */ 1218 tcp_pcap_tcpcb_init(tp); 1219 #endif 1220 return (tp); /* XXX */ 1221 } 1222 1223 /* 1224 * Switch the congestion control algorithm back to NewReno for any active 1225 * control blocks using an algorithm which is about to go away. 1226 * This ensures the CC framework can allow the unload to proceed without leaving 1227 * any dangling pointers which would trigger a panic. 1228 * Returning non-zero would inform the CC framework that something went wrong 1229 * and it would be unsafe to allow the unload to proceed. However, there is no 1230 * way for this to occur with this implementation so we always return zero. 1231 */ 1232 int 1233 tcp_ccalgounload(struct cc_algo *unload_algo) 1234 { 1235 struct cc_algo *tmpalgo; 1236 struct inpcb *inp; 1237 struct tcpcb *tp; 1238 VNET_ITERATOR_DECL(vnet_iter); 1239 1240 /* 1241 * Check all active control blocks across all network stacks and change 1242 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1243 * requires cleanup code to be run, call it. 1244 */ 1245 VNET_LIST_RLOCK(); 1246 VNET_FOREACH(vnet_iter) { 1247 CURVNET_SET(vnet_iter); 1248 INP_INFO_WLOCK(&V_tcbinfo); 1249 /* 1250 * New connections already part way through being initialised 1251 * with the CC algo we're removing will not race with this code 1252 * because the INP_INFO_WLOCK is held during initialisation. We 1253 * therefore don't enter the loop below until the connection 1254 * list has stabilised. 1255 */ 1256 LIST_FOREACH(inp, &V_tcb, inp_list) { 1257 INP_WLOCK(inp); 1258 /* Important to skip tcptw structs. */ 1259 if (!(inp->inp_flags & INP_TIMEWAIT) && 1260 (tp = intotcpcb(inp)) != NULL) { 1261 /* 1262 * By holding INP_WLOCK here, we are assured 1263 * that the connection is not currently 1264 * executing inside the CC module's functions 1265 * i.e. it is safe to make the switch back to 1266 * NewReno. 1267 */ 1268 if (CC_ALGO(tp) == unload_algo) { 1269 tmpalgo = CC_ALGO(tp); 1270 /* NewReno does not require any init. */ 1271 CC_ALGO(tp) = &newreno_cc_algo; 1272 if (tmpalgo->cb_destroy != NULL) 1273 tmpalgo->cb_destroy(tp->ccv); 1274 } 1275 } 1276 INP_WUNLOCK(inp); 1277 } 1278 INP_INFO_WUNLOCK(&V_tcbinfo); 1279 CURVNET_RESTORE(); 1280 } 1281 VNET_LIST_RUNLOCK(); 1282 1283 return (0); 1284 } 1285 1286 /* 1287 * Drop a TCP connection, reporting 1288 * the specified error. If connection is synchronized, 1289 * then send a RST to peer. 1290 */ 1291 struct tcpcb * 1292 tcp_drop(struct tcpcb *tp, int errno) 1293 { 1294 struct socket *so = tp->t_inpcb->inp_socket; 1295 1296 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1297 INP_WLOCK_ASSERT(tp->t_inpcb); 1298 1299 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1300 tcp_state_change(tp, TCPS_CLOSED); 1301 (void) tp->t_fb->tfb_tcp_output(tp); 1302 TCPSTAT_INC(tcps_drops); 1303 } else 1304 TCPSTAT_INC(tcps_conndrops); 1305 if (errno == ETIMEDOUT && tp->t_softerror) 1306 errno = tp->t_softerror; 1307 so->so_error = errno; 1308 return (tcp_close(tp)); 1309 } 1310 1311 void 1312 tcp_discardcb(struct tcpcb *tp) 1313 { 1314 struct inpcb *inp = tp->t_inpcb; 1315 struct socket *so = inp->inp_socket; 1316 #ifdef INET6 1317 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1318 #endif /* INET6 */ 1319 int released; 1320 1321 INP_WLOCK_ASSERT(inp); 1322 1323 /* 1324 * Make sure that all of our timers are stopped before we delete the 1325 * PCB. 1326 * 1327 * If stopping a timer fails, we schedule a discard function in same 1328 * callout, and the last discard function called will take care of 1329 * deleting the tcpcb. 1330 */ 1331 tcp_timer_stop(tp, TT_REXMT); 1332 tcp_timer_stop(tp, TT_PERSIST); 1333 tcp_timer_stop(tp, TT_KEEP); 1334 tcp_timer_stop(tp, TT_2MSL); 1335 tcp_timer_stop(tp, TT_DELACK); 1336 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1337 /* Call the stop-all function of the methods */ 1338 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1339 } 1340 1341 /* 1342 * If we got enough samples through the srtt filter, 1343 * save the rtt and rttvar in the routing entry. 1344 * 'Enough' is arbitrarily defined as 4 rtt samples. 1345 * 4 samples is enough for the srtt filter to converge 1346 * to within enough % of the correct value; fewer samples 1347 * and we could save a bogus rtt. The danger is not high 1348 * as tcp quickly recovers from everything. 1349 * XXX: Works very well but needs some more statistics! 1350 */ 1351 if (tp->t_rttupdated >= 4) { 1352 struct hc_metrics_lite metrics; 1353 u_long ssthresh; 1354 1355 bzero(&metrics, sizeof(metrics)); 1356 /* 1357 * Update the ssthresh always when the conditions below 1358 * are satisfied. This gives us better new start value 1359 * for the congestion avoidance for new connections. 1360 * ssthresh is only set if packet loss occured on a session. 1361 * 1362 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1363 * being torn down. Ideally this code would not use 'so'. 1364 */ 1365 ssthresh = tp->snd_ssthresh; 1366 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1367 /* 1368 * convert the limit from user data bytes to 1369 * packets then to packet data bytes. 1370 */ 1371 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1372 if (ssthresh < 2) 1373 ssthresh = 2; 1374 ssthresh *= (u_long)(tp->t_maxseg + 1375 #ifdef INET6 1376 (isipv6 ? sizeof (struct ip6_hdr) + 1377 sizeof (struct tcphdr) : 1378 #endif 1379 sizeof (struct tcpiphdr) 1380 #ifdef INET6 1381 ) 1382 #endif 1383 ); 1384 } else 1385 ssthresh = 0; 1386 metrics.rmx_ssthresh = ssthresh; 1387 1388 metrics.rmx_rtt = tp->t_srtt; 1389 metrics.rmx_rttvar = tp->t_rttvar; 1390 metrics.rmx_cwnd = tp->snd_cwnd; 1391 metrics.rmx_sendpipe = 0; 1392 metrics.rmx_recvpipe = 0; 1393 1394 tcp_hc_update(&inp->inp_inc, &metrics); 1395 } 1396 1397 /* free the reassembly queue, if any */ 1398 tcp_reass_flush(tp); 1399 1400 #ifdef TCP_OFFLOAD 1401 /* Disconnect offload device, if any. */ 1402 if (tp->t_flags & TF_TOE) 1403 tcp_offload_detach(tp); 1404 #endif 1405 1406 tcp_free_sackholes(tp); 1407 1408 #ifdef TCPPCAP 1409 /* Free the TCP PCAP queues. */ 1410 tcp_pcap_drain(&(tp->t_inpkts)); 1411 tcp_pcap_drain(&(tp->t_outpkts)); 1412 #endif 1413 1414 /* Allow the CC algorithm to clean up after itself. */ 1415 if (CC_ALGO(tp)->cb_destroy != NULL) 1416 CC_ALGO(tp)->cb_destroy(tp->ccv); 1417 1418 khelp_destroy_osd(tp->osd); 1419 1420 CC_ALGO(tp) = NULL; 1421 inp->inp_ppcb = NULL; 1422 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1423 /* We own the last reference on tcpcb, let's free it. */ 1424 if ((tp->t_fb->tfb_tcp_timers_left) && 1425 (tp->t_fb->tfb_tcp_timers_left(tp))) { 1426 /* Some fb timers left running! */ 1427 return; 1428 } 1429 if (tp->t_fb->tfb_tcp_fb_fini) 1430 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1431 refcount_release(&tp->t_fb->tfb_refcnt); 1432 tp->t_inpcb = NULL; 1433 uma_zfree(V_tcpcb_zone, tp); 1434 released = in_pcbrele_wlocked(inp); 1435 KASSERT(!released, ("%s: inp %p should not have been released " 1436 "here", __func__, inp)); 1437 } 1438 } 1439 1440 void 1441 tcp_timer_2msl_discard(void *xtp) 1442 { 1443 1444 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL); 1445 } 1446 1447 void 1448 tcp_timer_keep_discard(void *xtp) 1449 { 1450 1451 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP); 1452 } 1453 1454 void 1455 tcp_timer_persist_discard(void *xtp) 1456 { 1457 1458 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST); 1459 } 1460 1461 void 1462 tcp_timer_rexmt_discard(void *xtp) 1463 { 1464 1465 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT); 1466 } 1467 1468 void 1469 tcp_timer_delack_discard(void *xtp) 1470 { 1471 1472 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK); 1473 } 1474 1475 void 1476 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type) 1477 { 1478 struct inpcb *inp; 1479 1480 CURVNET_SET(tp->t_vnet); 1481 INP_INFO_RLOCK(&V_tcbinfo); 1482 inp = tp->t_inpcb; 1483 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1484 __func__, tp)); 1485 INP_WLOCK(inp); 1486 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1487 ("%s: tcpcb has to be stopped here", __func__)); 1488 KASSERT((tp->t_timers->tt_flags & timer_type) != 0, 1489 ("%s: discard callout should be running", __func__)); 1490 tp->t_timers->tt_flags &= ~timer_type; 1491 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1492 /* We own the last reference on this tcpcb, let's free it. */ 1493 if ((tp->t_fb->tfb_tcp_timers_left) && 1494 (tp->t_fb->tfb_tcp_timers_left(tp))) { 1495 /* Some fb timers left running! */ 1496 goto leave; 1497 } 1498 if (tp->t_fb->tfb_tcp_fb_fini) 1499 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1500 refcount_release(&tp->t_fb->tfb_refcnt); 1501 tp->t_inpcb = NULL; 1502 uma_zfree(V_tcpcb_zone, tp); 1503 if (in_pcbrele_wlocked(inp)) { 1504 INP_INFO_RUNLOCK(&V_tcbinfo); 1505 CURVNET_RESTORE(); 1506 return; 1507 } 1508 } 1509 leave: 1510 INP_WUNLOCK(inp); 1511 INP_INFO_RUNLOCK(&V_tcbinfo); 1512 CURVNET_RESTORE(); 1513 } 1514 1515 /* 1516 * Attempt to close a TCP control block, marking it as dropped, and freeing 1517 * the socket if we hold the only reference. 1518 */ 1519 struct tcpcb * 1520 tcp_close(struct tcpcb *tp) 1521 { 1522 struct inpcb *inp = tp->t_inpcb; 1523 struct socket *so; 1524 1525 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1526 INP_WLOCK_ASSERT(inp); 1527 1528 #ifdef TCP_OFFLOAD 1529 if (tp->t_state == TCPS_LISTEN) 1530 tcp_offload_listen_stop(tp); 1531 #endif 1532 #ifdef TCP_RFC7413 1533 /* 1534 * This releases the TFO pending counter resource for TFO listen 1535 * sockets as well as passively-created TFO sockets that transition 1536 * from SYN_RECEIVED to CLOSED. 1537 */ 1538 if (tp->t_tfo_pending) { 1539 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1540 tp->t_tfo_pending = NULL; 1541 } 1542 #endif 1543 in_pcbdrop(inp); 1544 TCPSTAT_INC(tcps_closed); 1545 TCPSTATES_DEC(tp->t_state); 1546 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1547 so = inp->inp_socket; 1548 soisdisconnected(so); 1549 if (inp->inp_flags & INP_SOCKREF) { 1550 KASSERT(so->so_state & SS_PROTOREF, 1551 ("tcp_close: !SS_PROTOREF")); 1552 inp->inp_flags &= ~INP_SOCKREF; 1553 INP_WUNLOCK(inp); 1554 ACCEPT_LOCK(); 1555 SOCK_LOCK(so); 1556 so->so_state &= ~SS_PROTOREF; 1557 sofree(so); 1558 return (NULL); 1559 } 1560 return (tp); 1561 } 1562 1563 void 1564 tcp_drain(void) 1565 { 1566 VNET_ITERATOR_DECL(vnet_iter); 1567 1568 if (!do_tcpdrain) 1569 return; 1570 1571 VNET_LIST_RLOCK_NOSLEEP(); 1572 VNET_FOREACH(vnet_iter) { 1573 CURVNET_SET(vnet_iter); 1574 struct inpcb *inpb; 1575 struct tcpcb *tcpb; 1576 1577 /* 1578 * Walk the tcpbs, if existing, and flush the reassembly queue, 1579 * if there is one... 1580 * XXX: The "Net/3" implementation doesn't imply that the TCP 1581 * reassembly queue should be flushed, but in a situation 1582 * where we're really low on mbufs, this is potentially 1583 * useful. 1584 */ 1585 INP_INFO_WLOCK(&V_tcbinfo); 1586 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1587 if (inpb->inp_flags & INP_TIMEWAIT) 1588 continue; 1589 INP_WLOCK(inpb); 1590 if ((tcpb = intotcpcb(inpb)) != NULL) { 1591 tcp_reass_flush(tcpb); 1592 tcp_clean_sackreport(tcpb); 1593 } 1594 INP_WUNLOCK(inpb); 1595 } 1596 INP_INFO_WUNLOCK(&V_tcbinfo); 1597 CURVNET_RESTORE(); 1598 } 1599 VNET_LIST_RUNLOCK_NOSLEEP(); 1600 } 1601 1602 /* 1603 * Notify a tcp user of an asynchronous error; 1604 * store error as soft error, but wake up user 1605 * (for now, won't do anything until can select for soft error). 1606 * 1607 * Do not wake up user since there currently is no mechanism for 1608 * reporting soft errors (yet - a kqueue filter may be added). 1609 */ 1610 static struct inpcb * 1611 tcp_notify(struct inpcb *inp, int error) 1612 { 1613 struct tcpcb *tp; 1614 1615 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1616 INP_WLOCK_ASSERT(inp); 1617 1618 if ((inp->inp_flags & INP_TIMEWAIT) || 1619 (inp->inp_flags & INP_DROPPED)) 1620 return (inp); 1621 1622 tp = intotcpcb(inp); 1623 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1624 1625 /* 1626 * Ignore some errors if we are hooked up. 1627 * If connection hasn't completed, has retransmitted several times, 1628 * and receives a second error, give up now. This is better 1629 * than waiting a long time to establish a connection that 1630 * can never complete. 1631 */ 1632 if (tp->t_state == TCPS_ESTABLISHED && 1633 (error == EHOSTUNREACH || error == ENETUNREACH || 1634 error == EHOSTDOWN)) { 1635 if (inp->inp_route.ro_rt) { 1636 RTFREE(inp->inp_route.ro_rt); 1637 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1638 } 1639 return (inp); 1640 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1641 tp->t_softerror) { 1642 tp = tcp_drop(tp, error); 1643 if (tp != NULL) 1644 return (inp); 1645 else 1646 return (NULL); 1647 } else { 1648 tp->t_softerror = error; 1649 return (inp); 1650 } 1651 #if 0 1652 wakeup( &so->so_timeo); 1653 sorwakeup(so); 1654 sowwakeup(so); 1655 #endif 1656 } 1657 1658 static int 1659 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1660 { 1661 int error, i, m, n, pcb_count; 1662 struct inpcb *inp, **inp_list; 1663 inp_gen_t gencnt; 1664 struct xinpgen xig; 1665 1666 /* 1667 * The process of preparing the TCB list is too time-consuming and 1668 * resource-intensive to repeat twice on every request. 1669 */ 1670 if (req->oldptr == NULL) { 1671 n = V_tcbinfo.ipi_count + 1672 counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]); 1673 n += imax(n / 8, 10); 1674 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1675 return (0); 1676 } 1677 1678 if (req->newptr != NULL) 1679 return (EPERM); 1680 1681 /* 1682 * OK, now we're committed to doing something. 1683 */ 1684 INP_LIST_RLOCK(&V_tcbinfo); 1685 gencnt = V_tcbinfo.ipi_gencnt; 1686 n = V_tcbinfo.ipi_count; 1687 INP_LIST_RUNLOCK(&V_tcbinfo); 1688 1689 m = counter_u64_fetch(VNET(tcps_states)[TCPS_SYN_RECEIVED]); 1690 1691 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1692 + (n + m) * sizeof(struct xtcpcb)); 1693 if (error != 0) 1694 return (error); 1695 1696 xig.xig_len = sizeof xig; 1697 xig.xig_count = n + m; 1698 xig.xig_gen = gencnt; 1699 xig.xig_sogen = so_gencnt; 1700 error = SYSCTL_OUT(req, &xig, sizeof xig); 1701 if (error) 1702 return (error); 1703 1704 error = syncache_pcblist(req, m, &pcb_count); 1705 if (error) 1706 return (error); 1707 1708 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1709 1710 INP_INFO_WLOCK(&V_tcbinfo); 1711 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1712 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1713 INP_WLOCK(inp); 1714 if (inp->inp_gencnt <= gencnt) { 1715 /* 1716 * XXX: This use of cr_cansee(), introduced with 1717 * TCP state changes, is not quite right, but for 1718 * now, better than nothing. 1719 */ 1720 if (inp->inp_flags & INP_TIMEWAIT) { 1721 if (intotw(inp) != NULL) 1722 error = cr_cansee(req->td->td_ucred, 1723 intotw(inp)->tw_cred); 1724 else 1725 error = EINVAL; /* Skip this inp. */ 1726 } else 1727 error = cr_canseeinpcb(req->td->td_ucred, inp); 1728 if (error == 0) { 1729 in_pcbref(inp); 1730 inp_list[i++] = inp; 1731 } 1732 } 1733 INP_WUNLOCK(inp); 1734 } 1735 INP_INFO_WUNLOCK(&V_tcbinfo); 1736 n = i; 1737 1738 error = 0; 1739 for (i = 0; i < n; i++) { 1740 inp = inp_list[i]; 1741 INP_RLOCK(inp); 1742 if (inp->inp_gencnt <= gencnt) { 1743 struct xtcpcb xt; 1744 void *inp_ppcb; 1745 1746 bzero(&xt, sizeof(xt)); 1747 xt.xt_len = sizeof xt; 1748 /* XXX should avoid extra copy */ 1749 bcopy(inp, &xt.xt_inp, sizeof *inp); 1750 inp_ppcb = inp->inp_ppcb; 1751 if (inp_ppcb == NULL) 1752 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1753 else if (inp->inp_flags & INP_TIMEWAIT) { 1754 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1755 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1756 } else { 1757 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1758 if (xt.xt_tp.t_timers) 1759 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1760 } 1761 if (inp->inp_socket != NULL) 1762 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1763 else { 1764 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1765 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1766 } 1767 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1768 INP_RUNLOCK(inp); 1769 error = SYSCTL_OUT(req, &xt, sizeof xt); 1770 } else 1771 INP_RUNLOCK(inp); 1772 } 1773 INP_INFO_RLOCK(&V_tcbinfo); 1774 for (i = 0; i < n; i++) { 1775 inp = inp_list[i]; 1776 INP_RLOCK(inp); 1777 if (!in_pcbrele_rlocked(inp)) 1778 INP_RUNLOCK(inp); 1779 } 1780 INP_INFO_RUNLOCK(&V_tcbinfo); 1781 1782 if (!error) { 1783 /* 1784 * Give the user an updated idea of our state. 1785 * If the generation differs from what we told 1786 * her before, she knows that something happened 1787 * while we were processing this request, and it 1788 * might be necessary to retry. 1789 */ 1790 INP_LIST_RLOCK(&V_tcbinfo); 1791 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1792 xig.xig_sogen = so_gencnt; 1793 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1794 INP_LIST_RUNLOCK(&V_tcbinfo); 1795 error = SYSCTL_OUT(req, &xig, sizeof xig); 1796 } 1797 free(inp_list, M_TEMP); 1798 return (error); 1799 } 1800 1801 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1802 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1803 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1804 1805 #ifdef INET 1806 static int 1807 tcp_getcred(SYSCTL_HANDLER_ARGS) 1808 { 1809 struct xucred xuc; 1810 struct sockaddr_in addrs[2]; 1811 struct inpcb *inp; 1812 int error; 1813 1814 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1815 if (error) 1816 return (error); 1817 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1818 if (error) 1819 return (error); 1820 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1821 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1822 if (inp != NULL) { 1823 if (inp->inp_socket == NULL) 1824 error = ENOENT; 1825 if (error == 0) 1826 error = cr_canseeinpcb(req->td->td_ucred, inp); 1827 if (error == 0) 1828 cru2x(inp->inp_cred, &xuc); 1829 INP_RUNLOCK(inp); 1830 } else 1831 error = ENOENT; 1832 if (error == 0) 1833 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1834 return (error); 1835 } 1836 1837 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1838 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1839 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1840 #endif /* INET */ 1841 1842 #ifdef INET6 1843 static int 1844 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1845 { 1846 struct xucred xuc; 1847 struct sockaddr_in6 addrs[2]; 1848 struct inpcb *inp; 1849 int error; 1850 #ifdef INET 1851 int mapped = 0; 1852 #endif 1853 1854 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1855 if (error) 1856 return (error); 1857 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1858 if (error) 1859 return (error); 1860 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1861 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1862 return (error); 1863 } 1864 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1865 #ifdef INET 1866 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1867 mapped = 1; 1868 else 1869 #endif 1870 return (EINVAL); 1871 } 1872 1873 #ifdef INET 1874 if (mapped == 1) 1875 inp = in_pcblookup(&V_tcbinfo, 1876 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1877 addrs[1].sin6_port, 1878 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1879 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1880 else 1881 #endif 1882 inp = in6_pcblookup(&V_tcbinfo, 1883 &addrs[1].sin6_addr, addrs[1].sin6_port, 1884 &addrs[0].sin6_addr, addrs[0].sin6_port, 1885 INPLOOKUP_RLOCKPCB, NULL); 1886 if (inp != NULL) { 1887 if (inp->inp_socket == NULL) 1888 error = ENOENT; 1889 if (error == 0) 1890 error = cr_canseeinpcb(req->td->td_ucred, inp); 1891 if (error == 0) 1892 cru2x(inp->inp_cred, &xuc); 1893 INP_RUNLOCK(inp); 1894 } else 1895 error = ENOENT; 1896 if (error == 0) 1897 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1898 return (error); 1899 } 1900 1901 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1902 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1903 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1904 #endif /* INET6 */ 1905 1906 1907 #ifdef INET 1908 void 1909 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1910 { 1911 struct ip *ip = vip; 1912 struct tcphdr *th; 1913 struct in_addr faddr; 1914 struct inpcb *inp; 1915 struct tcpcb *tp; 1916 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1917 struct icmp *icp; 1918 struct in_conninfo inc; 1919 tcp_seq icmp_tcp_seq; 1920 int mtu; 1921 1922 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1923 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1924 return; 1925 1926 if (cmd == PRC_MSGSIZE) 1927 notify = tcp_mtudisc_notify; 1928 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1929 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1930 notify = tcp_drop_syn_sent; 1931 else if (PRC_IS_REDIRECT(cmd)) { 1932 /* signal EHOSTDOWN, as it flushes the cached route */ 1933 in_pcbnotifyall(&V_tcbinfo, faddr, EHOSTDOWN, notify); 1934 return; 1935 } 1936 /* 1937 * Hostdead is ugly because it goes linearly through all PCBs. 1938 * XXX: We never get this from ICMP, otherwise it makes an 1939 * excellent DoS attack on machines with many connections. 1940 */ 1941 else if (cmd == PRC_HOSTDEAD) 1942 ip = NULL; 1943 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1944 return; 1945 1946 if (ip == NULL) { 1947 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1948 return; 1949 } 1950 1951 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1952 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1953 INP_INFO_RLOCK(&V_tcbinfo); 1954 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 1955 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1956 if (inp != NULL) { 1957 if (!(inp->inp_flags & INP_TIMEWAIT) && 1958 !(inp->inp_flags & INP_DROPPED) && 1959 !(inp->inp_socket == NULL)) { 1960 icmp_tcp_seq = ntohl(th->th_seq); 1961 tp = intotcpcb(inp); 1962 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1963 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1964 if (cmd == PRC_MSGSIZE) { 1965 /* 1966 * MTU discovery: 1967 * If we got a needfrag set the MTU 1968 * in the route to the suggested new 1969 * value (if given) and then notify. 1970 */ 1971 mtu = ntohs(icp->icmp_nextmtu); 1972 /* 1973 * If no alternative MTU was 1974 * proposed, try the next smaller 1975 * one. 1976 */ 1977 if (!mtu) 1978 mtu = ip_next_mtu( 1979 ntohs(ip->ip_len), 1); 1980 if (mtu < V_tcp_minmss + 1981 sizeof(struct tcpiphdr)) 1982 mtu = V_tcp_minmss + 1983 sizeof(struct tcpiphdr); 1984 /* 1985 * Only process the offered MTU if it 1986 * is smaller than the current one. 1987 */ 1988 if (mtu < tp->t_maxseg + 1989 sizeof(struct tcpiphdr)) { 1990 bzero(&inc, sizeof(inc)); 1991 inc.inc_faddr = faddr; 1992 inc.inc_fibnum = 1993 inp->inp_inc.inc_fibnum; 1994 tcp_hc_updatemtu(&inc, mtu); 1995 tcp_mtudisc(inp, mtu); 1996 } 1997 } else 1998 inp = (*notify)(inp, 1999 inetctlerrmap[cmd]); 2000 } 2001 } 2002 if (inp != NULL) 2003 INP_WUNLOCK(inp); 2004 } else { 2005 bzero(&inc, sizeof(inc)); 2006 inc.inc_fport = th->th_dport; 2007 inc.inc_lport = th->th_sport; 2008 inc.inc_faddr = faddr; 2009 inc.inc_laddr = ip->ip_src; 2010 syncache_unreach(&inc, th); 2011 } 2012 INP_INFO_RUNLOCK(&V_tcbinfo); 2013 } 2014 #endif /* INET */ 2015 2016 #ifdef INET6 2017 void 2018 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2019 { 2020 struct tcphdr th; 2021 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2022 struct ip6_hdr *ip6; 2023 struct mbuf *m; 2024 struct ip6ctlparam *ip6cp = NULL; 2025 const struct sockaddr_in6 *sa6_src = NULL; 2026 int off; 2027 struct tcp_portonly { 2028 u_int16_t th_sport; 2029 u_int16_t th_dport; 2030 } *thp; 2031 2032 if (sa->sa_family != AF_INET6 || 2033 sa->sa_len != sizeof(struct sockaddr_in6)) 2034 return; 2035 2036 if (cmd == PRC_MSGSIZE) 2037 notify = tcp_mtudisc_notify; 2038 else if (!PRC_IS_REDIRECT(cmd) && 2039 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 2040 return; 2041 2042 /* if the parameter is from icmp6, decode it. */ 2043 if (d != NULL) { 2044 ip6cp = (struct ip6ctlparam *)d; 2045 m = ip6cp->ip6c_m; 2046 ip6 = ip6cp->ip6c_ip6; 2047 off = ip6cp->ip6c_off; 2048 sa6_src = ip6cp->ip6c_src; 2049 } else { 2050 m = NULL; 2051 ip6 = NULL; 2052 off = 0; /* fool gcc */ 2053 sa6_src = &sa6_any; 2054 } 2055 2056 if (ip6 != NULL) { 2057 struct in_conninfo inc; 2058 /* 2059 * XXX: We assume that when IPV6 is non NULL, 2060 * M and OFF are valid. 2061 */ 2062 2063 /* check if we can safely examine src and dst ports */ 2064 if (m->m_pkthdr.len < off + sizeof(*thp)) 2065 return; 2066 2067 bzero(&th, sizeof(th)); 2068 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 2069 2070 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 2071 (struct sockaddr *)ip6cp->ip6c_src, 2072 th.th_sport, cmd, NULL, notify); 2073 2074 bzero(&inc, sizeof(inc)); 2075 inc.inc_fport = th.th_dport; 2076 inc.inc_lport = th.th_sport; 2077 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 2078 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 2079 inc.inc_flags |= INC_ISIPV6; 2080 INP_INFO_RLOCK(&V_tcbinfo); 2081 syncache_unreach(&inc, &th); 2082 INP_INFO_RUNLOCK(&V_tcbinfo); 2083 } else 2084 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 2085 0, cmd, NULL, notify); 2086 } 2087 #endif /* INET6 */ 2088 2089 2090 /* 2091 * Following is where TCP initial sequence number generation occurs. 2092 * 2093 * There are two places where we must use initial sequence numbers: 2094 * 1. In SYN-ACK packets. 2095 * 2. In SYN packets. 2096 * 2097 * All ISNs for SYN-ACK packets are generated by the syncache. See 2098 * tcp_syncache.c for details. 2099 * 2100 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2101 * depends on this property. In addition, these ISNs should be 2102 * unguessable so as to prevent connection hijacking. To satisfy 2103 * the requirements of this situation, the algorithm outlined in 2104 * RFC 1948 is used, with only small modifications. 2105 * 2106 * Implementation details: 2107 * 2108 * Time is based off the system timer, and is corrected so that it 2109 * increases by one megabyte per second. This allows for proper 2110 * recycling on high speed LANs while still leaving over an hour 2111 * before rollover. 2112 * 2113 * As reading the *exact* system time is too expensive to be done 2114 * whenever setting up a TCP connection, we increment the time 2115 * offset in two ways. First, a small random positive increment 2116 * is added to isn_offset for each connection that is set up. 2117 * Second, the function tcp_isn_tick fires once per clock tick 2118 * and increments isn_offset as necessary so that sequence numbers 2119 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2120 * random positive increments serve only to ensure that the same 2121 * exact sequence number is never sent out twice (as could otherwise 2122 * happen when a port is recycled in less than the system tick 2123 * interval.) 2124 * 2125 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2126 * between seeding of isn_secret. This is normally set to zero, 2127 * as reseeding should not be necessary. 2128 * 2129 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2130 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2131 * general, this means holding an exclusive (write) lock. 2132 */ 2133 2134 #define ISN_BYTES_PER_SECOND 1048576 2135 #define ISN_STATIC_INCREMENT 4096 2136 #define ISN_RANDOM_INCREMENT (4096 - 1) 2137 2138 static VNET_DEFINE(u_char, isn_secret[32]); 2139 static VNET_DEFINE(int, isn_last); 2140 static VNET_DEFINE(int, isn_last_reseed); 2141 static VNET_DEFINE(u_int32_t, isn_offset); 2142 static VNET_DEFINE(u_int32_t, isn_offset_old); 2143 2144 #define V_isn_secret VNET(isn_secret) 2145 #define V_isn_last VNET(isn_last) 2146 #define V_isn_last_reseed VNET(isn_last_reseed) 2147 #define V_isn_offset VNET(isn_offset) 2148 #define V_isn_offset_old VNET(isn_offset_old) 2149 2150 tcp_seq 2151 tcp_new_isn(struct tcpcb *tp) 2152 { 2153 MD5_CTX isn_ctx; 2154 u_int32_t md5_buffer[4]; 2155 tcp_seq new_isn; 2156 u_int32_t projected_offset; 2157 2158 INP_WLOCK_ASSERT(tp->t_inpcb); 2159 2160 ISN_LOCK(); 2161 /* Seed if this is the first use, reseed if requested. */ 2162 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2163 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2164 < (u_int)ticks))) { 2165 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2166 V_isn_last_reseed = ticks; 2167 } 2168 2169 /* Compute the md5 hash and return the ISN. */ 2170 MD5Init(&isn_ctx); 2171 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2172 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2173 #ifdef INET6 2174 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2175 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2176 sizeof(struct in6_addr)); 2177 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2178 sizeof(struct in6_addr)); 2179 } else 2180 #endif 2181 { 2182 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2183 sizeof(struct in_addr)); 2184 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2185 sizeof(struct in_addr)); 2186 } 2187 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2188 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2189 new_isn = (tcp_seq) md5_buffer[0]; 2190 V_isn_offset += ISN_STATIC_INCREMENT + 2191 (arc4random() & ISN_RANDOM_INCREMENT); 2192 if (ticks != V_isn_last) { 2193 projected_offset = V_isn_offset_old + 2194 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2195 if (SEQ_GT(projected_offset, V_isn_offset)) 2196 V_isn_offset = projected_offset; 2197 V_isn_offset_old = V_isn_offset; 2198 V_isn_last = ticks; 2199 } 2200 new_isn += V_isn_offset; 2201 ISN_UNLOCK(); 2202 return (new_isn); 2203 } 2204 2205 /* 2206 * When a specific ICMP unreachable message is received and the 2207 * connection state is SYN-SENT, drop the connection. This behavior 2208 * is controlled by the icmp_may_rst sysctl. 2209 */ 2210 struct inpcb * 2211 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2212 { 2213 struct tcpcb *tp; 2214 2215 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2216 INP_WLOCK_ASSERT(inp); 2217 2218 if ((inp->inp_flags & INP_TIMEWAIT) || 2219 (inp->inp_flags & INP_DROPPED)) 2220 return (inp); 2221 2222 tp = intotcpcb(inp); 2223 if (tp->t_state != TCPS_SYN_SENT) 2224 return (inp); 2225 2226 tp = tcp_drop(tp, errno); 2227 if (tp != NULL) 2228 return (inp); 2229 else 2230 return (NULL); 2231 } 2232 2233 /* 2234 * When `need fragmentation' ICMP is received, update our idea of the MSS 2235 * based on the new value. Also nudge TCP to send something, since we 2236 * know the packet we just sent was dropped. 2237 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2238 */ 2239 static struct inpcb * 2240 tcp_mtudisc_notify(struct inpcb *inp, int error) 2241 { 2242 2243 tcp_mtudisc(inp, -1); 2244 return (inp); 2245 } 2246 2247 static void 2248 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2249 { 2250 struct tcpcb *tp; 2251 struct socket *so; 2252 2253 INP_WLOCK_ASSERT(inp); 2254 if ((inp->inp_flags & INP_TIMEWAIT) || 2255 (inp->inp_flags & INP_DROPPED)) 2256 return; 2257 2258 tp = intotcpcb(inp); 2259 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2260 2261 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2262 2263 so = inp->inp_socket; 2264 SOCKBUF_LOCK(&so->so_snd); 2265 /* If the mss is larger than the socket buffer, decrease the mss. */ 2266 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2267 tp->t_maxseg = so->so_snd.sb_hiwat; 2268 SOCKBUF_UNLOCK(&so->so_snd); 2269 2270 TCPSTAT_INC(tcps_mturesent); 2271 tp->t_rtttime = 0; 2272 tp->snd_nxt = tp->snd_una; 2273 tcp_free_sackholes(tp); 2274 tp->snd_recover = tp->snd_max; 2275 if (tp->t_flags & TF_SACK_PERMIT) 2276 EXIT_FASTRECOVERY(tp->t_flags); 2277 tp->t_fb->tfb_tcp_output(tp); 2278 } 2279 2280 #ifdef INET 2281 /* 2282 * Look-up the routing entry to the peer of this inpcb. If no route 2283 * is found and it cannot be allocated, then return 0. This routine 2284 * is called by TCP routines that access the rmx structure and by 2285 * tcp_mss_update to get the peer/interface MTU. 2286 */ 2287 u_long 2288 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2289 { 2290 struct nhop4_extended nh4; 2291 struct ifnet *ifp; 2292 u_long maxmtu = 0; 2293 2294 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2295 2296 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2297 2298 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2299 NHR_REF, 0, &nh4) != 0) 2300 return (0); 2301 2302 ifp = nh4.nh_ifp; 2303 maxmtu = nh4.nh_mtu; 2304 2305 /* Report additional interface capabilities. */ 2306 if (cap != NULL) { 2307 if (ifp->if_capenable & IFCAP_TSO4 && 2308 ifp->if_hwassist & CSUM_TSO) { 2309 cap->ifcap |= CSUM_TSO; 2310 cap->tsomax = ifp->if_hw_tsomax; 2311 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2312 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2313 } 2314 } 2315 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2316 } 2317 return (maxmtu); 2318 } 2319 #endif /* INET */ 2320 2321 #ifdef INET6 2322 u_long 2323 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2324 { 2325 struct nhop6_extended nh6; 2326 struct in6_addr dst6; 2327 uint32_t scopeid; 2328 struct ifnet *ifp; 2329 u_long maxmtu = 0; 2330 2331 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2332 2333 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2334 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2335 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2336 0, &nh6) != 0) 2337 return (0); 2338 2339 ifp = nh6.nh_ifp; 2340 maxmtu = nh6.nh_mtu; 2341 2342 /* Report additional interface capabilities. */ 2343 if (cap != NULL) { 2344 if (ifp->if_capenable & IFCAP_TSO6 && 2345 ifp->if_hwassist & CSUM_TSO) { 2346 cap->ifcap |= CSUM_TSO; 2347 cap->tsomax = ifp->if_hw_tsomax; 2348 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2349 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2350 } 2351 } 2352 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2353 } 2354 2355 return (maxmtu); 2356 } 2357 #endif /* INET6 */ 2358 2359 /* 2360 * Calculate effective SMSS per RFC5681 definition for a given TCP 2361 * connection at its current state, taking into account SACK and etc. 2362 */ 2363 u_int 2364 tcp_maxseg(const struct tcpcb *tp) 2365 { 2366 u_int optlen; 2367 2368 if (tp->t_flags & TF_NOOPT) 2369 return (tp->t_maxseg); 2370 2371 /* 2372 * Here we have a simplified code from tcp_addoptions(), 2373 * without a proper loop, and having most of paddings hardcoded. 2374 * We might make mistakes with padding here in some edge cases, 2375 * but this is harmless, since result of tcp_maxseg() is used 2376 * only in cwnd and ssthresh estimations. 2377 */ 2378 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2379 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2380 if (tp->t_flags & TF_RCVD_TSTMP) 2381 optlen = TCPOLEN_TSTAMP_APPA; 2382 else 2383 optlen = 0; 2384 #ifdef TCP_SIGNATURE 2385 if (tp->t_flags & TF_SIGNATURE) 2386 optlen += PAD(TCPOLEN_SIGNATURE); 2387 #endif 2388 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2389 optlen += TCPOLEN_SACKHDR; 2390 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2391 optlen = PAD(optlen); 2392 } 2393 } else { 2394 if (tp->t_flags & TF_REQ_TSTMP) 2395 optlen = TCPOLEN_TSTAMP_APPA; 2396 else 2397 optlen = PAD(TCPOLEN_MAXSEG); 2398 if (tp->t_flags & TF_REQ_SCALE) 2399 optlen += PAD(TCPOLEN_WINDOW); 2400 #ifdef TCP_SIGNATURE 2401 if (tp->t_flags & TF_SIGNATURE) 2402 optlen += PAD(TCPOLEN_SIGNATURE); 2403 #endif 2404 if (tp->t_flags & TF_SACK_PERMIT) 2405 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2406 } 2407 #undef PAD 2408 optlen = min(optlen, TCP_MAXOLEN); 2409 return (tp->t_maxseg - optlen); 2410 } 2411 2412 #ifdef IPSEC 2413 /* compute ESP/AH header size for TCP, including outer IP header. */ 2414 size_t 2415 ipsec_hdrsiz_tcp(struct tcpcb *tp) 2416 { 2417 struct inpcb *inp; 2418 struct mbuf *m; 2419 size_t hdrsiz; 2420 struct ip *ip; 2421 #ifdef INET6 2422 struct ip6_hdr *ip6; 2423 #endif 2424 struct tcphdr *th; 2425 2426 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || 2427 (!key_havesp(IPSEC_DIR_OUTBOUND))) 2428 return (0); 2429 m = m_gethdr(M_NOWAIT, MT_DATA); 2430 if (!m) 2431 return (0); 2432 2433 #ifdef INET6 2434 if ((inp->inp_vflag & INP_IPV6) != 0) { 2435 ip6 = mtod(m, struct ip6_hdr *); 2436 th = (struct tcphdr *)(ip6 + 1); 2437 m->m_pkthdr.len = m->m_len = 2438 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2439 tcpip_fillheaders(inp, ip6, th); 2440 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2441 } else 2442 #endif /* INET6 */ 2443 { 2444 ip = mtod(m, struct ip *); 2445 th = (struct tcphdr *)(ip + 1); 2446 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 2447 tcpip_fillheaders(inp, ip, th); 2448 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 2449 } 2450 2451 m_free(m); 2452 return (hdrsiz); 2453 } 2454 #endif /* IPSEC */ 2455 2456 #ifdef TCP_SIGNATURE 2457 /* 2458 * Callback function invoked by m_apply() to digest TCP segment data 2459 * contained within an mbuf chain. 2460 */ 2461 static int 2462 tcp_signature_apply(void *fstate, void *data, u_int len) 2463 { 2464 2465 MD5Update(fstate, (u_char *)data, len); 2466 return (0); 2467 } 2468 2469 /* 2470 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2471 * search with the destination IP address, and a 'magic SPI' to be 2472 * determined by the application. This is hardcoded elsewhere to 1179 2473 */ 2474 struct secasvar * 2475 tcp_get_sav(struct mbuf *m, u_int direction) 2476 { 2477 union sockaddr_union dst; 2478 struct secasvar *sav; 2479 struct ip *ip; 2480 #ifdef INET6 2481 struct ip6_hdr *ip6; 2482 char ip6buf[INET6_ADDRSTRLEN]; 2483 #endif 2484 2485 /* Extract the destination from the IP header in the mbuf. */ 2486 bzero(&dst, sizeof(union sockaddr_union)); 2487 ip = mtod(m, struct ip *); 2488 #ifdef INET6 2489 ip6 = NULL; /* Make the compiler happy. */ 2490 #endif 2491 switch (ip->ip_v) { 2492 #ifdef INET 2493 case IPVERSION: 2494 dst.sa.sa_len = sizeof(struct sockaddr_in); 2495 dst.sa.sa_family = AF_INET; 2496 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2497 ip->ip_src : ip->ip_dst; 2498 break; 2499 #endif 2500 #ifdef INET6 2501 case (IPV6_VERSION >> 4): 2502 ip6 = mtod(m, struct ip6_hdr *); 2503 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2504 dst.sa.sa_family = AF_INET6; 2505 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 2506 ip6->ip6_src : ip6->ip6_dst; 2507 break; 2508 #endif 2509 default: 2510 return (NULL); 2511 /* NOTREACHED */ 2512 break; 2513 } 2514 2515 /* Look up an SADB entry which matches the address of the peer. */ 2516 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2517 if (sav == NULL) { 2518 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 2519 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 2520 #ifdef INET6 2521 (ip->ip_v == (IPV6_VERSION >> 4)) ? 2522 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 2523 #endif 2524 "(unsupported)")); 2525 } 2526 2527 return (sav); 2528 } 2529 2530 /* 2531 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2532 * 2533 * Parameters: 2534 * m pointer to head of mbuf chain 2535 * len length of TCP segment data, excluding options 2536 * optlen length of TCP segment options 2537 * buf pointer to storage for computed MD5 digest 2538 * sav pointer to security assosiation 2539 * 2540 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2541 * When called from tcp_input(), we can be sure that th_sum has been 2542 * zeroed out and verified already. 2543 * 2544 * Releases reference to SADB key before return. 2545 * 2546 * Return 0 if successful, otherwise return -1. 2547 * 2548 */ 2549 int 2550 tcp_signature_do_compute(struct mbuf *m, int len, int optlen, 2551 u_char *buf, struct secasvar *sav) 2552 { 2553 #ifdef INET 2554 struct ippseudo ippseudo; 2555 #endif 2556 MD5_CTX ctx; 2557 int doff; 2558 struct ip *ip; 2559 #ifdef INET 2560 struct ipovly *ipovly; 2561 #endif 2562 struct tcphdr *th; 2563 #ifdef INET6 2564 struct ip6_hdr *ip6; 2565 struct in6_addr in6; 2566 uint32_t plen; 2567 uint16_t nhdr; 2568 #endif 2569 u_short savecsum; 2570 2571 KASSERT(m != NULL, ("NULL mbuf chain")); 2572 KASSERT(buf != NULL, ("NULL signature pointer")); 2573 2574 /* Extract the destination from the IP header in the mbuf. */ 2575 ip = mtod(m, struct ip *); 2576 #ifdef INET6 2577 ip6 = NULL; /* Make the compiler happy. */ 2578 #endif 2579 2580 MD5Init(&ctx); 2581 /* 2582 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 2583 * 2584 * XXX The ippseudo header MUST be digested in network byte order, 2585 * or else we'll fail the regression test. Assume all fields we've 2586 * been doing arithmetic on have been in host byte order. 2587 * XXX One cannot depend on ipovly->ih_len here. When called from 2588 * tcp_output(), the underlying ip_len member has not yet been set. 2589 */ 2590 switch (ip->ip_v) { 2591 #ifdef INET 2592 case IPVERSION: 2593 ipovly = (struct ipovly *)ip; 2594 ippseudo.ippseudo_src = ipovly->ih_src; 2595 ippseudo.ippseudo_dst = ipovly->ih_dst; 2596 ippseudo.ippseudo_pad = 0; 2597 ippseudo.ippseudo_p = IPPROTO_TCP; 2598 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 2599 optlen); 2600 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2601 2602 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2603 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2604 break; 2605 #endif 2606 #ifdef INET6 2607 /* 2608 * RFC 2385, 2.0 Proposal 2609 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2610 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2611 * extended next header value (to form 32 bits), and 32-bit segment 2612 * length. 2613 * Note: Upper-Layer Packet Length comes before Next Header. 2614 */ 2615 case (IPV6_VERSION >> 4): 2616 in6 = ip6->ip6_src; 2617 in6_clearscope(&in6); 2618 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2619 in6 = ip6->ip6_dst; 2620 in6_clearscope(&in6); 2621 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2622 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2623 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2624 nhdr = 0; 2625 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2626 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2627 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2628 nhdr = IPPROTO_TCP; 2629 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2630 2631 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2632 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2633 break; 2634 #endif 2635 default: 2636 KEY_FREESAV(&sav); 2637 return (-1); 2638 /* NOTREACHED */ 2639 break; 2640 } 2641 2642 2643 /* 2644 * Step 2: Update MD5 hash with TCP header, excluding options. 2645 * The TCP checksum must be set to zero. 2646 */ 2647 savecsum = th->th_sum; 2648 th->th_sum = 0; 2649 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2650 th->th_sum = savecsum; 2651 2652 /* 2653 * Step 3: Update MD5 hash with TCP segment data. 2654 * Use m_apply() to avoid an early m_pullup(). 2655 */ 2656 if (len > 0) 2657 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2658 2659 /* 2660 * Step 4: Update MD5 hash with shared secret. 2661 */ 2662 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2663 MD5Final(buf, &ctx); 2664 2665 key_sa_recordxfer(sav, m); 2666 KEY_FREESAV(&sav); 2667 return (0); 2668 } 2669 2670 /* 2671 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2672 * 2673 * Return 0 if successful, otherwise return -1. 2674 */ 2675 int 2676 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 2677 u_char *buf, u_int direction) 2678 { 2679 struct secasvar *sav; 2680 2681 if ((sav = tcp_get_sav(m, direction)) == NULL) 2682 return (-1); 2683 2684 return (tcp_signature_do_compute(m, len, optlen, buf, sav)); 2685 } 2686 2687 /* 2688 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2689 * 2690 * Parameters: 2691 * m pointer to head of mbuf chain 2692 * len length of TCP segment data, excluding options 2693 * optlen length of TCP segment options 2694 * buf pointer to storage for computed MD5 digest 2695 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2696 * 2697 * Return 1 if successful, otherwise return 0. 2698 */ 2699 int 2700 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2701 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2702 { 2703 char tmpdigest[TCP_SIGLEN]; 2704 2705 if (tcp_sig_checksigs == 0) 2706 return (1); 2707 if ((tcpbflag & TF_SIGNATURE) == 0) { 2708 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2709 2710 /* 2711 * If this socket is not expecting signature but 2712 * the segment contains signature just fail. 2713 */ 2714 TCPSTAT_INC(tcps_sig_err_sigopt); 2715 TCPSTAT_INC(tcps_sig_rcvbadsig); 2716 return (0); 2717 } 2718 2719 /* Signature is not expected, and not present in segment. */ 2720 return (1); 2721 } 2722 2723 /* 2724 * If this socket is expecting signature but the segment does not 2725 * contain any just fail. 2726 */ 2727 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2728 TCPSTAT_INC(tcps_sig_err_nosigopt); 2729 TCPSTAT_INC(tcps_sig_rcvbadsig); 2730 return (0); 2731 } 2732 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2733 IPSEC_DIR_INBOUND) == -1) { 2734 TCPSTAT_INC(tcps_sig_err_buildsig); 2735 TCPSTAT_INC(tcps_sig_rcvbadsig); 2736 return (0); 2737 } 2738 2739 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2740 TCPSTAT_INC(tcps_sig_rcvbadsig); 2741 return (0); 2742 } 2743 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2744 return (1); 2745 } 2746 #endif /* TCP_SIGNATURE */ 2747 2748 static int 2749 sysctl_drop(SYSCTL_HANDLER_ARGS) 2750 { 2751 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2752 struct sockaddr_storage addrs[2]; 2753 struct inpcb *inp; 2754 struct tcpcb *tp; 2755 struct tcptw *tw; 2756 struct sockaddr_in *fin, *lin; 2757 #ifdef INET6 2758 struct sockaddr_in6 *fin6, *lin6; 2759 #endif 2760 int error; 2761 2762 inp = NULL; 2763 fin = lin = NULL; 2764 #ifdef INET6 2765 fin6 = lin6 = NULL; 2766 #endif 2767 error = 0; 2768 2769 if (req->oldptr != NULL || req->oldlen != 0) 2770 return (EINVAL); 2771 if (req->newptr == NULL) 2772 return (EPERM); 2773 if (req->newlen < sizeof(addrs)) 2774 return (ENOMEM); 2775 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2776 if (error) 2777 return (error); 2778 2779 switch (addrs[0].ss_family) { 2780 #ifdef INET6 2781 case AF_INET6: 2782 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2783 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2784 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2785 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2786 return (EINVAL); 2787 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2788 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2789 return (EINVAL); 2790 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2791 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2792 fin = (struct sockaddr_in *)&addrs[0]; 2793 lin = (struct sockaddr_in *)&addrs[1]; 2794 break; 2795 } 2796 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2797 if (error) 2798 return (error); 2799 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2800 if (error) 2801 return (error); 2802 break; 2803 #endif 2804 #ifdef INET 2805 case AF_INET: 2806 fin = (struct sockaddr_in *)&addrs[0]; 2807 lin = (struct sockaddr_in *)&addrs[1]; 2808 if (fin->sin_len != sizeof(struct sockaddr_in) || 2809 lin->sin_len != sizeof(struct sockaddr_in)) 2810 return (EINVAL); 2811 break; 2812 #endif 2813 default: 2814 return (EINVAL); 2815 } 2816 INP_INFO_RLOCK(&V_tcbinfo); 2817 switch (addrs[0].ss_family) { 2818 #ifdef INET6 2819 case AF_INET6: 2820 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2821 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2822 INPLOOKUP_WLOCKPCB, NULL); 2823 break; 2824 #endif 2825 #ifdef INET 2826 case AF_INET: 2827 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2828 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2829 break; 2830 #endif 2831 } 2832 if (inp != NULL) { 2833 if (inp->inp_flags & INP_TIMEWAIT) { 2834 /* 2835 * XXXRW: There currently exists a state where an 2836 * inpcb is present, but its timewait state has been 2837 * discarded. For now, don't allow dropping of this 2838 * type of inpcb. 2839 */ 2840 tw = intotw(inp); 2841 if (tw != NULL) 2842 tcp_twclose(tw, 0); 2843 else 2844 INP_WUNLOCK(inp); 2845 } else if (!(inp->inp_flags & INP_DROPPED) && 2846 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2847 tp = intotcpcb(inp); 2848 tp = tcp_drop(tp, ECONNABORTED); 2849 if (tp != NULL) 2850 INP_WUNLOCK(inp); 2851 } else 2852 INP_WUNLOCK(inp); 2853 } else 2854 error = ESRCH; 2855 INP_INFO_RUNLOCK(&V_tcbinfo); 2856 return (error); 2857 } 2858 2859 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2860 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2861 0, sysctl_drop, "", "Drop TCP connection"); 2862 2863 /* 2864 * Generate a standardized TCP log line for use throughout the 2865 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2866 * allow use in the interrupt context. 2867 * 2868 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2869 * NB: The function may return NULL if memory allocation failed. 2870 * 2871 * Due to header inclusion and ordering limitations the struct ip 2872 * and ip6_hdr pointers have to be passed as void pointers. 2873 */ 2874 char * 2875 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2876 const void *ip6hdr) 2877 { 2878 2879 /* Is logging enabled? */ 2880 if (tcp_log_in_vain == 0) 2881 return (NULL); 2882 2883 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2884 } 2885 2886 char * 2887 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2888 const void *ip6hdr) 2889 { 2890 2891 /* Is logging enabled? */ 2892 if (tcp_log_debug == 0) 2893 return (NULL); 2894 2895 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2896 } 2897 2898 static char * 2899 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2900 const void *ip6hdr) 2901 { 2902 char *s, *sp; 2903 size_t size; 2904 struct ip *ip; 2905 #ifdef INET6 2906 const struct ip6_hdr *ip6; 2907 2908 ip6 = (const struct ip6_hdr *)ip6hdr; 2909 #endif /* INET6 */ 2910 ip = (struct ip *)ip4hdr; 2911 2912 /* 2913 * The log line looks like this: 2914 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2915 */ 2916 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2917 sizeof(PRINT_TH_FLAGS) + 1 + 2918 #ifdef INET6 2919 2 * INET6_ADDRSTRLEN; 2920 #else 2921 2 * INET_ADDRSTRLEN; 2922 #endif /* INET6 */ 2923 2924 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2925 if (s == NULL) 2926 return (NULL); 2927 2928 strcat(s, "TCP: ["); 2929 sp = s + strlen(s); 2930 2931 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2932 inet_ntoa_r(inc->inc_faddr, sp); 2933 sp = s + strlen(s); 2934 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2935 sp = s + strlen(s); 2936 inet_ntoa_r(inc->inc_laddr, sp); 2937 sp = s + strlen(s); 2938 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2939 #ifdef INET6 2940 } else if (inc) { 2941 ip6_sprintf(sp, &inc->inc6_faddr); 2942 sp = s + strlen(s); 2943 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2944 sp = s + strlen(s); 2945 ip6_sprintf(sp, &inc->inc6_laddr); 2946 sp = s + strlen(s); 2947 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2948 } else if (ip6 && th) { 2949 ip6_sprintf(sp, &ip6->ip6_src); 2950 sp = s + strlen(s); 2951 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2952 sp = s + strlen(s); 2953 ip6_sprintf(sp, &ip6->ip6_dst); 2954 sp = s + strlen(s); 2955 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2956 #endif /* INET6 */ 2957 #ifdef INET 2958 } else if (ip && th) { 2959 inet_ntoa_r(ip->ip_src, sp); 2960 sp = s + strlen(s); 2961 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2962 sp = s + strlen(s); 2963 inet_ntoa_r(ip->ip_dst, sp); 2964 sp = s + strlen(s); 2965 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2966 #endif /* INET */ 2967 } else { 2968 free(s, M_TCPLOG); 2969 return (NULL); 2970 } 2971 sp = s + strlen(s); 2972 if (th) 2973 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2974 if (*(s + size - 1) != '\0') 2975 panic("%s: string too long", __func__); 2976 return (s); 2977 } 2978 2979 /* 2980 * A subroutine which makes it easy to track TCP state changes with DTrace. 2981 * This function shouldn't be called for t_state initializations that don't 2982 * correspond to actual TCP state transitions. 2983 */ 2984 void 2985 tcp_state_change(struct tcpcb *tp, int newstate) 2986 { 2987 #if defined(KDTRACE_HOOKS) 2988 int pstate = tp->t_state; 2989 #endif 2990 2991 TCPSTATES_DEC(tp->t_state); 2992 TCPSTATES_INC(newstate); 2993 tp->t_state = newstate; 2994 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2995 } 2996