1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/arb.h> 46 #include <sys/callout.h> 47 #include <sys/eventhandler.h> 48 #ifdef TCP_HHOOK 49 #include <sys/hhook.h> 50 #endif 51 #include <sys/kernel.h> 52 #ifdef TCP_HHOOK 53 #include <sys/khelp.h> 54 #endif 55 #ifdef KERN_TLS 56 #include <sys/ktls.h> 57 #endif 58 #include <sys/qmath.h> 59 #include <sys/stats.h> 60 #include <sys/sysctl.h> 61 #include <sys/jail.h> 62 #include <sys/malloc.h> 63 #include <sys/refcount.h> 64 #include <sys/mbuf.h> 65 #ifdef INET6 66 #include <sys/domain.h> 67 #endif 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/sdt.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/protosw.h> 74 #include <sys/random.h> 75 76 #include <vm/uma.h> 77 78 #include <net/route.h> 79 #include <net/route/nhop.h> 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/vnet.h> 83 84 #include <netinet/in.h> 85 #include <netinet/in_fib.h> 86 #include <netinet/in_kdtrace.h> 87 #include <netinet/in_pcb.h> 88 #include <netinet/in_systm.h> 89 #include <netinet/in_var.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> 92 #include <netinet/ip_var.h> 93 #ifdef INET6 94 #include <netinet/icmp6.h> 95 #include <netinet/ip6.h> 96 #include <netinet6/in6_fib.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/ip6_var.h> 99 #include <netinet6/scope6_var.h> 100 #include <netinet6/nd6.h> 101 #endif 102 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_syncache.h> 110 #include <netinet/tcp_hpts.h> 111 #include <netinet/cc/cc.h> 112 #ifdef INET6 113 #include <netinet6/tcp6_var.h> 114 #endif 115 #include <netinet/tcpip.h> 116 #include <netinet/tcp_fastopen.h> 117 #ifdef TCPPCAP 118 #include <netinet/tcp_pcap.h> 119 #endif 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif 123 #ifdef INET6 124 #include <netinet6/ip6protosw.h> 125 #endif 126 #ifdef TCP_OFFLOAD 127 #include <netinet/tcp_offload.h> 128 #endif 129 130 #include <netipsec/ipsec_support.h> 131 132 #include <machine/in_cksum.h> 133 #include <crypto/siphash/siphash.h> 134 135 #include <security/mac/mac_framework.h> 136 137 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 138 #ifdef INET6 139 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 140 #endif 141 142 #ifdef NETFLIX_EXP_DETECTION 143 /* Sack attack detection thresholds and such */ 144 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 145 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 146 "Sack Attack detection thresholds"); 147 int32_t tcp_force_detection = 0; 148 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 149 CTLFLAG_RW, 150 &tcp_force_detection, 0, 151 "Do we force detection even if the INP has it off?"); 152 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 154 CTLFLAG_RW, 155 &tcp_sack_to_ack_thresh, 700, 156 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 157 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 159 CTLFLAG_RW, 160 &tcp_sack_to_move_thresh, 600, 161 "Percentage of sack moves we must see above (10.1 percent is 101)"); 162 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 164 CTLFLAG_RW, 165 &tcp_restoral_thresh, 550, 166 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 167 int32_t tcp_sad_decay_val = 800; 168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 169 CTLFLAG_RW, 170 &tcp_sad_decay_val, 800, 171 "The decay percentage (10.1 percent equals 101 )"); 172 int32_t tcp_map_minimum = 500; 173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 174 CTLFLAG_RW, 175 &tcp_map_minimum, 500, 176 "Number of Map enteries before we start detection"); 177 int32_t tcp_attack_on_turns_on_logging = 0; 178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 179 CTLFLAG_RW, 180 &tcp_attack_on_turns_on_logging, 0, 181 "When we have a positive hit on attack, do we turn on logging?"); 182 int32_t tcp_sad_pacing_interval = 2000; 183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 184 CTLFLAG_RW, 185 &tcp_sad_pacing_interval, 2000, 186 "What is the minimum pacing interval for a classified attacker?"); 187 188 int32_t tcp_sad_low_pps = 100; 189 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 190 CTLFLAG_RW, 191 &tcp_sad_low_pps, 100, 192 "What is the input pps that below which we do not decay?"); 193 #endif 194 195 struct rwlock tcp_function_lock; 196 197 static int 198 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 199 { 200 int error, new; 201 202 new = V_tcp_mssdflt; 203 error = sysctl_handle_int(oidp, &new, 0, req); 204 if (error == 0 && req->newptr) { 205 if (new < TCP_MINMSS) 206 error = EINVAL; 207 else 208 V_tcp_mssdflt = new; 209 } 210 return (error); 211 } 212 213 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 214 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 215 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 216 "Default TCP Maximum Segment Size"); 217 218 #ifdef INET6 219 static int 220 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 221 { 222 int error, new; 223 224 new = V_tcp_v6mssdflt; 225 error = sysctl_handle_int(oidp, &new, 0, req); 226 if (error == 0 && req->newptr) { 227 if (new < TCP_MINMSS) 228 error = EINVAL; 229 else 230 V_tcp_v6mssdflt = new; 231 } 232 return (error); 233 } 234 235 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 236 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 237 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 238 "Default TCP Maximum Segment Size for IPv6"); 239 #endif /* INET6 */ 240 241 /* 242 * Minimum MSS we accept and use. This prevents DoS attacks where 243 * we are forced to a ridiculous low MSS like 20 and send hundreds 244 * of packets instead of one. The effect scales with the available 245 * bandwidth and quickly saturates the CPU and network interface 246 * with packet generation and sending. Set to zero to disable MINMSS 247 * checking. This setting prevents us from sending too small packets. 248 */ 249 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 251 &VNET_NAME(tcp_minmss), 0, 252 "Minimum TCP Maximum Segment Size"); 253 254 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 255 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 256 &VNET_NAME(tcp_do_rfc1323), 0, 257 "Enable rfc1323 (high performance TCP) extensions"); 258 259 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 0; 260 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 261 &VNET_NAME(tcp_tolerate_missing_ts), 0, 262 "Tolerate missing TCP timestamps"); 263 264 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 266 &VNET_NAME(tcp_ts_offset_per_conn), 0, 267 "Initialize TCP timestamps per connection instead of per host pair"); 268 269 static int tcp_log_debug = 0; 270 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 271 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 272 273 static int tcp_tcbhashsize; 274 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 275 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 276 277 static int do_tcpdrain = 1; 278 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 279 "Enable tcp_drain routine for extra help when low on mbufs"); 280 281 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 282 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 283 284 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 285 #define V_icmp_may_rst VNET(icmp_may_rst) 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 287 &VNET_NAME(icmp_may_rst), 0, 288 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 289 290 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 291 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 293 &VNET_NAME(tcp_isn_reseed_interval), 0, 294 "Seconds between reseeding of ISN secret"); 295 296 static int tcp_soreceive_stream; 297 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 298 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 299 300 VNET_DEFINE(uma_zone_t, sack_hole_zone); 301 #define V_sack_hole_zone VNET(sack_hole_zone) 302 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 303 static int 304 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 305 { 306 int error; 307 uint32_t new; 308 309 new = V_tcp_map_entries_limit; 310 error = sysctl_handle_int(oidp, &new, 0, req); 311 if (error == 0 && req->newptr) { 312 /* only allow "0" and value > minimum */ 313 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 314 error = EINVAL; 315 else 316 V_tcp_map_entries_limit = new; 317 } 318 return (error); 319 } 320 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 321 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 322 &VNET_NAME(tcp_map_entries_limit), 0, 323 &sysctl_net_inet_tcp_map_limit_check, "IU", 324 "Total sendmap entries limit"); 325 326 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 327 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 328 &VNET_NAME(tcp_map_split_limit), 0, 329 "Total sendmap split entries limit"); 330 331 #ifdef TCP_HHOOK 332 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 333 #endif 334 335 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 336 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 337 #define V_ts_offset_secret VNET(ts_offset_secret) 338 339 static int tcp_default_fb_init(struct tcpcb *tp); 340 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 341 static int tcp_default_handoff_ok(struct tcpcb *tp); 342 static struct inpcb *tcp_notify(struct inpcb *, int); 343 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 344 static void tcp_mtudisc(struct inpcb *, int); 345 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 346 void *ip4hdr, const void *ip6hdr); 347 348 static struct tcp_function_block tcp_def_funcblk = { 349 .tfb_tcp_block_name = "freebsd", 350 .tfb_tcp_output = tcp_output, 351 .tfb_tcp_do_segment = tcp_do_segment, 352 .tfb_tcp_ctloutput = tcp_default_ctloutput, 353 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 354 .tfb_tcp_fb_init = tcp_default_fb_init, 355 .tfb_tcp_fb_fini = tcp_default_fb_fini, 356 }; 357 358 static int tcp_fb_cnt = 0; 359 struct tcp_funchead t_functions; 360 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 361 362 static struct tcp_function_block * 363 find_tcp_functions_locked(struct tcp_function_set *fs) 364 { 365 struct tcp_function *f; 366 struct tcp_function_block *blk=NULL; 367 368 TAILQ_FOREACH(f, &t_functions, tf_next) { 369 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 370 blk = f->tf_fb; 371 break; 372 } 373 } 374 return(blk); 375 } 376 377 static struct tcp_function_block * 378 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 379 { 380 struct tcp_function_block *rblk=NULL; 381 struct tcp_function *f; 382 383 TAILQ_FOREACH(f, &t_functions, tf_next) { 384 if (f->tf_fb == blk) { 385 rblk = blk; 386 if (s) { 387 *s = f; 388 } 389 break; 390 } 391 } 392 return (rblk); 393 } 394 395 struct tcp_function_block * 396 find_and_ref_tcp_functions(struct tcp_function_set *fs) 397 { 398 struct tcp_function_block *blk; 399 400 rw_rlock(&tcp_function_lock); 401 blk = find_tcp_functions_locked(fs); 402 if (blk) 403 refcount_acquire(&blk->tfb_refcnt); 404 rw_runlock(&tcp_function_lock); 405 return(blk); 406 } 407 408 struct tcp_function_block * 409 find_and_ref_tcp_fb(struct tcp_function_block *blk) 410 { 411 struct tcp_function_block *rblk; 412 413 rw_rlock(&tcp_function_lock); 414 rblk = find_tcp_fb_locked(blk, NULL); 415 if (rblk) 416 refcount_acquire(&rblk->tfb_refcnt); 417 rw_runlock(&tcp_function_lock); 418 return(rblk); 419 } 420 421 static struct tcp_function_block * 422 find_and_ref_tcp_default_fb(void) 423 { 424 struct tcp_function_block *rblk; 425 426 rw_rlock(&tcp_function_lock); 427 rblk = tcp_func_set_ptr; 428 refcount_acquire(&rblk->tfb_refcnt); 429 rw_runlock(&tcp_function_lock); 430 return (rblk); 431 } 432 433 void 434 tcp_switch_back_to_default(struct tcpcb *tp) 435 { 436 struct tcp_function_block *tfb; 437 438 KASSERT(tp->t_fb != &tcp_def_funcblk, 439 ("%s: called by the built-in default stack", __func__)); 440 441 /* 442 * Release the old stack. This function will either find a new one 443 * or panic. 444 */ 445 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 446 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 447 refcount_release(&tp->t_fb->tfb_refcnt); 448 449 /* 450 * Now, we'll find a new function block to use. 451 * Start by trying the current user-selected 452 * default, unless this stack is the user-selected 453 * default. 454 */ 455 tfb = find_and_ref_tcp_default_fb(); 456 if (tfb == tp->t_fb) { 457 refcount_release(&tfb->tfb_refcnt); 458 tfb = NULL; 459 } 460 /* Does the stack accept this connection? */ 461 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 462 (*tfb->tfb_tcp_handoff_ok)(tp)) { 463 refcount_release(&tfb->tfb_refcnt); 464 tfb = NULL; 465 } 466 /* Try to use that stack. */ 467 if (tfb != NULL) { 468 /* Initialize the new stack. If it succeeds, we are done. */ 469 tp->t_fb = tfb; 470 if (tp->t_fb->tfb_tcp_fb_init == NULL || 471 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 472 return; 473 474 /* 475 * Initialization failed. Release the reference count on 476 * the stack. 477 */ 478 refcount_release(&tfb->tfb_refcnt); 479 } 480 481 /* 482 * If that wasn't feasible, use the built-in default 483 * stack which is not allowed to reject anyone. 484 */ 485 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 486 if (tfb == NULL) { 487 /* there always should be a default */ 488 panic("Can't refer to tcp_def_funcblk"); 489 } 490 if (tfb->tfb_tcp_handoff_ok != NULL) { 491 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 492 /* The default stack cannot say no */ 493 panic("Default stack rejects a new session?"); 494 } 495 } 496 tp->t_fb = tfb; 497 if (tp->t_fb->tfb_tcp_fb_init != NULL && 498 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 499 /* The default stack cannot fail */ 500 panic("Default stack initialization failed"); 501 } 502 } 503 504 static int 505 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 506 { 507 int error=ENOENT; 508 struct tcp_function_set fs; 509 struct tcp_function_block *blk; 510 511 memset(&fs, 0, sizeof(fs)); 512 rw_rlock(&tcp_function_lock); 513 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 514 if (blk) { 515 /* Found him */ 516 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 517 fs.pcbcnt = blk->tfb_refcnt; 518 } 519 rw_runlock(&tcp_function_lock); 520 error = sysctl_handle_string(oidp, fs.function_set_name, 521 sizeof(fs.function_set_name), req); 522 523 /* Check for error or no change */ 524 if (error != 0 || req->newptr == NULL) 525 return(error); 526 527 rw_wlock(&tcp_function_lock); 528 blk = find_tcp_functions_locked(&fs); 529 if ((blk == NULL) || 530 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 531 error = ENOENT; 532 goto done; 533 } 534 tcp_func_set_ptr = blk; 535 done: 536 rw_wunlock(&tcp_function_lock); 537 return (error); 538 } 539 540 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 541 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 542 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 543 "Set/get the default TCP functions"); 544 545 static int 546 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 547 { 548 int error, cnt, linesz; 549 struct tcp_function *f; 550 char *buffer, *cp; 551 size_t bufsz, outsz; 552 bool alias; 553 554 cnt = 0; 555 rw_rlock(&tcp_function_lock); 556 TAILQ_FOREACH(f, &t_functions, tf_next) { 557 cnt++; 558 } 559 rw_runlock(&tcp_function_lock); 560 561 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 562 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 563 564 error = 0; 565 cp = buffer; 566 567 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 568 "Alias", "PCB count"); 569 cp += linesz; 570 bufsz -= linesz; 571 outsz = linesz; 572 573 rw_rlock(&tcp_function_lock); 574 TAILQ_FOREACH(f, &t_functions, tf_next) { 575 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 576 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 577 f->tf_fb->tfb_tcp_block_name, 578 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 579 alias ? f->tf_name : "-", 580 f->tf_fb->tfb_refcnt); 581 if (linesz >= bufsz) { 582 error = EOVERFLOW; 583 break; 584 } 585 cp += linesz; 586 bufsz -= linesz; 587 outsz += linesz; 588 } 589 rw_runlock(&tcp_function_lock); 590 if (error == 0) 591 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 592 free(buffer, M_TEMP); 593 return (error); 594 } 595 596 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 597 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 598 NULL, 0, sysctl_net_inet_list_available, "A", 599 "list available TCP Function sets"); 600 601 /* 602 * Exports one (struct tcp_function_info) for each alias/name. 603 */ 604 static int 605 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 606 { 607 int cnt, error; 608 struct tcp_function *f; 609 struct tcp_function_info tfi; 610 611 /* 612 * We don't allow writes. 613 */ 614 if (req->newptr != NULL) 615 return (EINVAL); 616 617 /* 618 * Wire the old buffer so we can directly copy the functions to 619 * user space without dropping the lock. 620 */ 621 if (req->oldptr != NULL) { 622 error = sysctl_wire_old_buffer(req, 0); 623 if (error) 624 return (error); 625 } 626 627 /* 628 * Walk the list and copy out matching entries. If INVARIANTS 629 * is compiled in, also walk the list to verify the length of 630 * the list matches what we have recorded. 631 */ 632 rw_rlock(&tcp_function_lock); 633 634 cnt = 0; 635 #ifndef INVARIANTS 636 if (req->oldptr == NULL) { 637 cnt = tcp_fb_cnt; 638 goto skip_loop; 639 } 640 #endif 641 TAILQ_FOREACH(f, &t_functions, tf_next) { 642 #ifdef INVARIANTS 643 cnt++; 644 #endif 645 if (req->oldptr != NULL) { 646 bzero(&tfi, sizeof(tfi)); 647 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 648 tfi.tfi_id = f->tf_fb->tfb_id; 649 (void)strlcpy(tfi.tfi_alias, f->tf_name, 650 sizeof(tfi.tfi_alias)); 651 (void)strlcpy(tfi.tfi_name, 652 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 653 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 654 /* 655 * Don't stop on error, as that is the 656 * mechanism we use to accumulate length 657 * information if the buffer was too short. 658 */ 659 } 660 } 661 KASSERT(cnt == tcp_fb_cnt, 662 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 663 #ifndef INVARIANTS 664 skip_loop: 665 #endif 666 rw_runlock(&tcp_function_lock); 667 if (req->oldptr == NULL) 668 error = SYSCTL_OUT(req, NULL, 669 (cnt + 1) * sizeof(struct tcp_function_info)); 670 671 return (error); 672 } 673 674 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 675 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 676 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 677 "List TCP function block name-to-ID mappings"); 678 679 /* 680 * tfb_tcp_handoff_ok() function for the default stack. 681 * Note that we'll basically try to take all comers. 682 */ 683 static int 684 tcp_default_handoff_ok(struct tcpcb *tp) 685 { 686 687 return (0); 688 } 689 690 /* 691 * tfb_tcp_fb_init() function for the default stack. 692 * 693 * This handles making sure we have appropriate timers set if you are 694 * transitioning a socket that has some amount of setup done. 695 * 696 * The init() fuction from the default can *never* return non-zero i.e. 697 * it is required to always succeed since it is the stack of last resort! 698 */ 699 static int 700 tcp_default_fb_init(struct tcpcb *tp) 701 { 702 703 struct socket *so; 704 705 INP_WLOCK_ASSERT(tp->t_inpcb); 706 707 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 708 ("%s: connection %p in unexpected state %d", __func__, tp, 709 tp->t_state)); 710 711 /* 712 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 713 * know what to do for unexpected states (which includes TIME_WAIT). 714 */ 715 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 716 return (0); 717 718 /* 719 * Make sure some kind of transmission timer is set if there is 720 * outstanding data. 721 */ 722 so = tp->t_inpcb->inp_socket; 723 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 724 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 725 tcp_timer_active(tp, TT_PERSIST))) { 726 /* 727 * If the session has established and it looks like it should 728 * be in the persist state, set the persist timer. Otherwise, 729 * set the retransmit timer. 730 */ 731 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 732 (int32_t)(tp->snd_nxt - tp->snd_una) < 733 (int32_t)sbavail(&so->so_snd)) 734 tcp_setpersist(tp); 735 else 736 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 737 } 738 739 /* All non-embryonic sessions get a keepalive timer. */ 740 if (!tcp_timer_active(tp, TT_KEEP)) 741 tcp_timer_activate(tp, TT_KEEP, 742 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 743 TP_KEEPINIT(tp)); 744 745 return (0); 746 } 747 748 /* 749 * tfb_tcp_fb_fini() function for the default stack. 750 * 751 * This changes state as necessary (or prudent) to prepare for another stack 752 * to assume responsibility for the connection. 753 */ 754 static void 755 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 756 { 757 758 INP_WLOCK_ASSERT(tp->t_inpcb); 759 return; 760 } 761 762 /* 763 * Target size of TCP PCB hash tables. Must be a power of two. 764 * 765 * Note that this can be overridden by the kernel environment 766 * variable net.inet.tcp.tcbhashsize 767 */ 768 #ifndef TCBHASHSIZE 769 #define TCBHASHSIZE 0 770 #endif 771 772 /* 773 * XXX 774 * Callouts should be moved into struct tcp directly. They are currently 775 * separate because the tcpcb structure is exported to userland for sysctl 776 * parsing purposes, which do not know about callouts. 777 */ 778 struct tcpcb_mem { 779 struct tcpcb tcb; 780 struct tcp_timer tt; 781 struct cc_var ccv; 782 #ifdef TCP_HHOOK 783 struct osd osd; 784 #endif 785 }; 786 787 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 788 #define V_tcpcb_zone VNET(tcpcb_zone) 789 790 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 791 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 792 793 static struct mtx isn_mtx; 794 795 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 796 #define ISN_LOCK() mtx_lock(&isn_mtx) 797 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 798 799 /* 800 * TCP initialization. 801 */ 802 static void 803 tcp_zone_change(void *tag) 804 { 805 806 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 807 uma_zone_set_max(V_tcpcb_zone, maxsockets); 808 tcp_tw_zone_change(); 809 } 810 811 static int 812 tcp_inpcb_init(void *mem, int size, int flags) 813 { 814 struct inpcb *inp = mem; 815 816 INP_LOCK_INIT(inp, "inp", "tcpinp"); 817 return (0); 818 } 819 820 /* 821 * Take a value and get the next power of 2 that doesn't overflow. 822 * Used to size the tcp_inpcb hash buckets. 823 */ 824 static int 825 maketcp_hashsize(int size) 826 { 827 int hashsize; 828 829 /* 830 * auto tune. 831 * get the next power of 2 higher than maxsockets. 832 */ 833 hashsize = 1 << fls(size); 834 /* catch overflow, and just go one power of 2 smaller */ 835 if (hashsize < size) { 836 hashsize = 1 << (fls(size) - 1); 837 } 838 return (hashsize); 839 } 840 841 static volatile int next_tcp_stack_id = 1; 842 843 /* 844 * Register a TCP function block with the name provided in the names 845 * array. (Note that this function does NOT automatically register 846 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 847 * explicitly include blk->tfb_tcp_block_name in the list of names if 848 * you wish to register the stack with that name.) 849 * 850 * Either all name registrations will succeed or all will fail. If 851 * a name registration fails, the function will update the num_names 852 * argument to point to the array index of the name that encountered 853 * the failure. 854 * 855 * Returns 0 on success, or an error code on failure. 856 */ 857 int 858 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 859 const char *names[], int *num_names) 860 { 861 struct tcp_function *n; 862 struct tcp_function_set fs; 863 int error, i; 864 865 KASSERT(names != NULL && *num_names > 0, 866 ("%s: Called with 0-length name list", __func__)); 867 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 868 KASSERT(rw_initialized(&tcp_function_lock), 869 ("%s: called too early", __func__)); 870 871 if ((blk->tfb_tcp_output == NULL) || 872 (blk->tfb_tcp_do_segment == NULL) || 873 (blk->tfb_tcp_ctloutput == NULL) || 874 (strlen(blk->tfb_tcp_block_name) == 0)) { 875 /* 876 * These functions are required and you 877 * need a name. 878 */ 879 *num_names = 0; 880 return (EINVAL); 881 } 882 if (blk->tfb_tcp_timer_stop_all || 883 blk->tfb_tcp_timer_activate || 884 blk->tfb_tcp_timer_active || 885 blk->tfb_tcp_timer_stop) { 886 /* 887 * If you define one timer function you 888 * must have them all. 889 */ 890 if ((blk->tfb_tcp_timer_stop_all == NULL) || 891 (blk->tfb_tcp_timer_activate == NULL) || 892 (blk->tfb_tcp_timer_active == NULL) || 893 (blk->tfb_tcp_timer_stop == NULL)) { 894 *num_names = 0; 895 return (EINVAL); 896 } 897 } 898 899 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 900 *num_names = 0; 901 return (EINVAL); 902 } 903 904 refcount_init(&blk->tfb_refcnt, 0); 905 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 906 for (i = 0; i < *num_names; i++) { 907 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 908 if (n == NULL) { 909 error = ENOMEM; 910 goto cleanup; 911 } 912 n->tf_fb = blk; 913 914 (void)strlcpy(fs.function_set_name, names[i], 915 sizeof(fs.function_set_name)); 916 rw_wlock(&tcp_function_lock); 917 if (find_tcp_functions_locked(&fs) != NULL) { 918 /* Duplicate name space not allowed */ 919 rw_wunlock(&tcp_function_lock); 920 free(n, M_TCPFUNCTIONS); 921 error = EALREADY; 922 goto cleanup; 923 } 924 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 925 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 926 tcp_fb_cnt++; 927 rw_wunlock(&tcp_function_lock); 928 } 929 return(0); 930 931 cleanup: 932 /* 933 * Deregister the names we just added. Because registration failed 934 * for names[i], we don't need to deregister that name. 935 */ 936 *num_names = i; 937 rw_wlock(&tcp_function_lock); 938 while (--i >= 0) { 939 TAILQ_FOREACH(n, &t_functions, tf_next) { 940 if (!strncmp(n->tf_name, names[i], 941 TCP_FUNCTION_NAME_LEN_MAX)) { 942 TAILQ_REMOVE(&t_functions, n, tf_next); 943 tcp_fb_cnt--; 944 n->tf_fb = NULL; 945 free(n, M_TCPFUNCTIONS); 946 break; 947 } 948 } 949 } 950 rw_wunlock(&tcp_function_lock); 951 return (error); 952 } 953 954 /* 955 * Register a TCP function block using the name provided in the name 956 * argument. 957 * 958 * Returns 0 on success, or an error code on failure. 959 */ 960 int 961 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 962 int wait) 963 { 964 const char *name_list[1]; 965 int num_names, rv; 966 967 num_names = 1; 968 if (name != NULL) 969 name_list[0] = name; 970 else 971 name_list[0] = blk->tfb_tcp_block_name; 972 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 973 return (rv); 974 } 975 976 /* 977 * Register a TCP function block using the name defined in 978 * blk->tfb_tcp_block_name. 979 * 980 * Returns 0 on success, or an error code on failure. 981 */ 982 int 983 register_tcp_functions(struct tcp_function_block *blk, int wait) 984 { 985 986 return (register_tcp_functions_as_name(blk, NULL, wait)); 987 } 988 989 /* 990 * Deregister all names associated with a function block. This 991 * functionally removes the function block from use within the system. 992 * 993 * When called with a true quiesce argument, mark the function block 994 * as being removed so no more stacks will use it and determine 995 * whether the removal would succeed. 996 * 997 * When called with a false quiesce argument, actually attempt the 998 * removal. 999 * 1000 * When called with a force argument, attempt to switch all TCBs to 1001 * use the default stack instead of returning EBUSY. 1002 * 1003 * Returns 0 on success (or if the removal would succeed, or an error 1004 * code on failure. 1005 */ 1006 int 1007 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1008 bool force) 1009 { 1010 struct tcp_function *f; 1011 1012 if (blk == &tcp_def_funcblk) { 1013 /* You can't un-register the default */ 1014 return (EPERM); 1015 } 1016 rw_wlock(&tcp_function_lock); 1017 if (blk == tcp_func_set_ptr) { 1018 /* You can't free the current default */ 1019 rw_wunlock(&tcp_function_lock); 1020 return (EBUSY); 1021 } 1022 /* Mark the block so no more stacks can use it. */ 1023 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1024 /* 1025 * If TCBs are still attached to the stack, attempt to switch them 1026 * to the default stack. 1027 */ 1028 if (force && blk->tfb_refcnt) { 1029 struct inpcb *inp; 1030 struct tcpcb *tp; 1031 VNET_ITERATOR_DECL(vnet_iter); 1032 1033 rw_wunlock(&tcp_function_lock); 1034 1035 VNET_LIST_RLOCK(); 1036 VNET_FOREACH(vnet_iter) { 1037 CURVNET_SET(vnet_iter); 1038 INP_INFO_WLOCK(&V_tcbinfo); 1039 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 1040 INP_WLOCK(inp); 1041 if (inp->inp_flags & INP_TIMEWAIT) { 1042 INP_WUNLOCK(inp); 1043 continue; 1044 } 1045 tp = intotcpcb(inp); 1046 if (tp == NULL || tp->t_fb != blk) { 1047 INP_WUNLOCK(inp); 1048 continue; 1049 } 1050 tcp_switch_back_to_default(tp); 1051 INP_WUNLOCK(inp); 1052 } 1053 INP_INFO_WUNLOCK(&V_tcbinfo); 1054 CURVNET_RESTORE(); 1055 } 1056 VNET_LIST_RUNLOCK(); 1057 1058 rw_wlock(&tcp_function_lock); 1059 } 1060 if (blk->tfb_refcnt) { 1061 /* TCBs still attached. */ 1062 rw_wunlock(&tcp_function_lock); 1063 return (EBUSY); 1064 } 1065 if (quiesce) { 1066 /* Skip removal. */ 1067 rw_wunlock(&tcp_function_lock); 1068 return (0); 1069 } 1070 /* Remove any function names that map to this function block. */ 1071 while (find_tcp_fb_locked(blk, &f) != NULL) { 1072 TAILQ_REMOVE(&t_functions, f, tf_next); 1073 tcp_fb_cnt--; 1074 f->tf_fb = NULL; 1075 free(f, M_TCPFUNCTIONS); 1076 } 1077 rw_wunlock(&tcp_function_lock); 1078 return (0); 1079 } 1080 1081 void 1082 tcp_init(void) 1083 { 1084 const char *tcbhash_tuneable; 1085 int hashsize; 1086 1087 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1088 1089 #ifdef TCP_HHOOK 1090 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1091 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1092 printf("%s: WARNING: unable to register helper hook\n", __func__); 1093 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1094 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1095 printf("%s: WARNING: unable to register helper hook\n", __func__); 1096 #endif 1097 #ifdef STATS 1098 if (tcp_stats_init()) 1099 printf("%s: WARNING: unable to initialise TCP stats\n", 1100 __func__); 1101 #endif 1102 hashsize = TCBHASHSIZE; 1103 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1104 if (hashsize == 0) { 1105 /* 1106 * Auto tune the hash size based on maxsockets. 1107 * A perfect hash would have a 1:1 mapping 1108 * (hashsize = maxsockets) however it's been 1109 * suggested that O(2) average is better. 1110 */ 1111 hashsize = maketcp_hashsize(maxsockets / 4); 1112 /* 1113 * Our historical default is 512, 1114 * do not autotune lower than this. 1115 */ 1116 if (hashsize < 512) 1117 hashsize = 512; 1118 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1119 printf("%s: %s auto tuned to %d\n", __func__, 1120 tcbhash_tuneable, hashsize); 1121 } 1122 /* 1123 * We require a hashsize to be a power of two. 1124 * Previously if it was not a power of two we would just reset it 1125 * back to 512, which could be a nasty surprise if you did not notice 1126 * the error message. 1127 * Instead what we do is clip it to the closest power of two lower 1128 * than the specified hash value. 1129 */ 1130 if (!powerof2(hashsize)) { 1131 int oldhashsize = hashsize; 1132 1133 hashsize = maketcp_hashsize(hashsize); 1134 /* prevent absurdly low value */ 1135 if (hashsize < 16) 1136 hashsize = 16; 1137 printf("%s: WARNING: TCB hash size not a power of 2, " 1138 "clipped from %d to %d.\n", __func__, oldhashsize, 1139 hashsize); 1140 } 1141 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1142 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1143 1144 /* 1145 * These have to be type stable for the benefit of the timers. 1146 */ 1147 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1148 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1149 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1150 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1151 1152 tcp_tw_init(); 1153 syncache_init(); 1154 tcp_hc_init(); 1155 1156 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1157 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1158 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1159 1160 tcp_fastopen_init(); 1161 1162 /* Skip initialization of globals for non-default instances. */ 1163 if (!IS_DEFAULT_VNET(curvnet)) 1164 return; 1165 1166 tcp_reass_global_init(); 1167 1168 /* XXX virtualize those bellow? */ 1169 tcp_delacktime = TCPTV_DELACK; 1170 tcp_keepinit = TCPTV_KEEP_INIT; 1171 tcp_keepidle = TCPTV_KEEP_IDLE; 1172 tcp_keepintvl = TCPTV_KEEPINTVL; 1173 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1174 tcp_msl = TCPTV_MSL; 1175 tcp_rexmit_initial = TCPTV_RTOBASE; 1176 if (tcp_rexmit_initial < 1) 1177 tcp_rexmit_initial = 1; 1178 tcp_rexmit_min = TCPTV_MIN; 1179 if (tcp_rexmit_min < 1) 1180 tcp_rexmit_min = 1; 1181 tcp_persmin = TCPTV_PERSMIN; 1182 tcp_persmax = TCPTV_PERSMAX; 1183 tcp_rexmit_slop = TCPTV_CPU_VAR; 1184 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1185 tcp_tcbhashsize = hashsize; 1186 1187 /* Setup the tcp function block list */ 1188 TAILQ_INIT(&t_functions); 1189 rw_init(&tcp_function_lock, "tcp_func_lock"); 1190 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1191 #ifdef TCP_BLACKBOX 1192 /* Initialize the TCP logging data. */ 1193 tcp_log_init(); 1194 #endif 1195 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1196 1197 if (tcp_soreceive_stream) { 1198 #ifdef INET 1199 tcp_usrreqs.pru_soreceive = soreceive_stream; 1200 #endif 1201 #ifdef INET6 1202 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1203 #endif /* INET6 */ 1204 } 1205 1206 #ifdef INET6 1207 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1208 #else /* INET6 */ 1209 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1210 #endif /* INET6 */ 1211 if (max_protohdr < TCP_MINPROTOHDR) 1212 max_protohdr = TCP_MINPROTOHDR; 1213 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1214 panic("tcp_init"); 1215 #undef TCP_MINPROTOHDR 1216 1217 ISN_LOCK_INIT(); 1218 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1219 SHUTDOWN_PRI_DEFAULT); 1220 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1221 EVENTHANDLER_PRI_ANY); 1222 1223 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1224 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1225 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1226 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); 1227 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1228 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); 1229 #ifdef TCPPCAP 1230 tcp_pcap_init(); 1231 #endif 1232 } 1233 1234 #ifdef VIMAGE 1235 static void 1236 tcp_destroy(void *unused __unused) 1237 { 1238 int n; 1239 #ifdef TCP_HHOOK 1240 int error; 1241 #endif 1242 1243 /* 1244 * All our processes are gone, all our sockets should be cleaned 1245 * up, which means, we should be past the tcp_discardcb() calls. 1246 * Sleep to let all tcpcb timers really disappear and cleanup. 1247 */ 1248 for (;;) { 1249 INP_LIST_RLOCK(&V_tcbinfo); 1250 n = V_tcbinfo.ipi_count; 1251 INP_LIST_RUNLOCK(&V_tcbinfo); 1252 if (n == 0) 1253 break; 1254 pause("tcpdes", hz / 10); 1255 } 1256 tcp_hc_destroy(); 1257 syncache_destroy(); 1258 tcp_tw_destroy(); 1259 in_pcbinfo_destroy(&V_tcbinfo); 1260 /* tcp_discardcb() clears the sack_holes up. */ 1261 uma_zdestroy(V_sack_hole_zone); 1262 uma_zdestroy(V_tcpcb_zone); 1263 1264 /* 1265 * Cannot free the zone until all tcpcbs are released as we attach 1266 * the allocations to them. 1267 */ 1268 tcp_fastopen_destroy(); 1269 1270 #ifdef TCP_HHOOK 1271 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1272 if (error != 0) { 1273 printf("%s: WARNING: unable to deregister helper hook " 1274 "type=%d, id=%d: error %d returned\n", __func__, 1275 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1276 } 1277 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1278 if (error != 0) { 1279 printf("%s: WARNING: unable to deregister helper hook " 1280 "type=%d, id=%d: error %d returned\n", __func__, 1281 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1282 } 1283 #endif 1284 } 1285 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1286 #endif 1287 1288 void 1289 tcp_fini(void *xtp) 1290 { 1291 1292 } 1293 1294 /* 1295 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1296 * tcp_template used to store this data in mbufs, but we now recopy it out 1297 * of the tcpcb each time to conserve mbufs. 1298 */ 1299 void 1300 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1301 { 1302 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1303 1304 INP_WLOCK_ASSERT(inp); 1305 1306 #ifdef INET6 1307 if ((inp->inp_vflag & INP_IPV6) != 0) { 1308 struct ip6_hdr *ip6; 1309 1310 ip6 = (struct ip6_hdr *)ip_ptr; 1311 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1312 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1313 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1314 (IPV6_VERSION & IPV6_VERSION_MASK); 1315 ip6->ip6_nxt = IPPROTO_TCP; 1316 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1317 ip6->ip6_src = inp->in6p_laddr; 1318 ip6->ip6_dst = inp->in6p_faddr; 1319 } 1320 #endif /* INET6 */ 1321 #if defined(INET6) && defined(INET) 1322 else 1323 #endif 1324 #ifdef INET 1325 { 1326 struct ip *ip; 1327 1328 ip = (struct ip *)ip_ptr; 1329 ip->ip_v = IPVERSION; 1330 ip->ip_hl = 5; 1331 ip->ip_tos = inp->inp_ip_tos; 1332 ip->ip_len = 0; 1333 ip->ip_id = 0; 1334 ip->ip_off = 0; 1335 ip->ip_ttl = inp->inp_ip_ttl; 1336 ip->ip_sum = 0; 1337 ip->ip_p = IPPROTO_TCP; 1338 ip->ip_src = inp->inp_laddr; 1339 ip->ip_dst = inp->inp_faddr; 1340 } 1341 #endif /* INET */ 1342 th->th_sport = inp->inp_lport; 1343 th->th_dport = inp->inp_fport; 1344 th->th_seq = 0; 1345 th->th_ack = 0; 1346 th->th_x2 = 0; 1347 th->th_off = 5; 1348 th->th_flags = 0; 1349 th->th_win = 0; 1350 th->th_urp = 0; 1351 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1352 } 1353 1354 /* 1355 * Create template to be used to send tcp packets on a connection. 1356 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1357 * use for this function is in keepalives, which use tcp_respond. 1358 */ 1359 struct tcptemp * 1360 tcpip_maketemplate(struct inpcb *inp) 1361 { 1362 struct tcptemp *t; 1363 1364 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1365 if (t == NULL) 1366 return (NULL); 1367 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1368 return (t); 1369 } 1370 1371 /* 1372 * Send a single message to the TCP at address specified by 1373 * the given TCP/IP header. If m == NULL, then we make a copy 1374 * of the tcpiphdr at th and send directly to the addressed host. 1375 * This is used to force keep alive messages out using the TCP 1376 * template for a connection. If flags are given then we send 1377 * a message back to the TCP which originated the segment th, 1378 * and discard the mbuf containing it and any other attached mbufs. 1379 * 1380 * In any case the ack and sequence number of the transmitted 1381 * segment are as specified by the parameters. 1382 * 1383 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1384 */ 1385 void 1386 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1387 tcp_seq ack, tcp_seq seq, int flags) 1388 { 1389 struct tcpopt to; 1390 struct inpcb *inp; 1391 struct ip *ip; 1392 struct mbuf *optm; 1393 struct tcphdr *nth; 1394 u_char *optp; 1395 #ifdef INET6 1396 struct ip6_hdr *ip6; 1397 int isipv6; 1398 #endif /* INET6 */ 1399 int optlen, tlen, win; 1400 bool incl_opts; 1401 1402 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1403 NET_EPOCH_ASSERT(); 1404 1405 #ifdef INET6 1406 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1407 ip6 = ipgen; 1408 #endif /* INET6 */ 1409 ip = ipgen; 1410 1411 if (tp != NULL) { 1412 inp = tp->t_inpcb; 1413 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1414 INP_WLOCK_ASSERT(inp); 1415 } else 1416 inp = NULL; 1417 1418 incl_opts = false; 1419 win = 0; 1420 if (tp != NULL) { 1421 if (!(flags & TH_RST)) { 1422 win = sbspace(&inp->inp_socket->so_rcv); 1423 if (win > TCP_MAXWIN << tp->rcv_scale) 1424 win = TCP_MAXWIN << tp->rcv_scale; 1425 } 1426 if ((tp->t_flags & TF_NOOPT) == 0) 1427 incl_opts = true; 1428 } 1429 if (m == NULL) { 1430 m = m_gethdr(M_NOWAIT, MT_DATA); 1431 if (m == NULL) 1432 return; 1433 m->m_data += max_linkhdr; 1434 #ifdef INET6 1435 if (isipv6) { 1436 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1437 sizeof(struct ip6_hdr)); 1438 ip6 = mtod(m, struct ip6_hdr *); 1439 nth = (struct tcphdr *)(ip6 + 1); 1440 } else 1441 #endif /* INET6 */ 1442 { 1443 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1444 ip = mtod(m, struct ip *); 1445 nth = (struct tcphdr *)(ip + 1); 1446 } 1447 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1448 flags = TH_ACK; 1449 } else if (!M_WRITABLE(m)) { 1450 struct mbuf *n; 1451 1452 /* Can't reuse 'm', allocate a new mbuf. */ 1453 n = m_gethdr(M_NOWAIT, MT_DATA); 1454 if (n == NULL) { 1455 m_freem(m); 1456 return; 1457 } 1458 1459 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1460 m_freem(m); 1461 m_freem(n); 1462 return; 1463 } 1464 1465 n->m_data += max_linkhdr; 1466 /* m_len is set later */ 1467 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1468 #ifdef INET6 1469 if (isipv6) { 1470 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1471 sizeof(struct ip6_hdr)); 1472 ip6 = mtod(n, struct ip6_hdr *); 1473 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1474 nth = (struct tcphdr *)(ip6 + 1); 1475 } else 1476 #endif /* INET6 */ 1477 { 1478 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1479 ip = mtod(n, struct ip *); 1480 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1481 nth = (struct tcphdr *)(ip + 1); 1482 } 1483 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1484 xchg(nth->th_dport, nth->th_sport, uint16_t); 1485 th = nth; 1486 m_freem(m); 1487 m = n; 1488 } else { 1489 /* 1490 * reuse the mbuf. 1491 * XXX MRT We inherit the FIB, which is lucky. 1492 */ 1493 m_freem(m->m_next); 1494 m->m_next = NULL; 1495 m->m_data = (caddr_t)ipgen; 1496 /* m_len is set later */ 1497 #ifdef INET6 1498 if (isipv6) { 1499 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1500 nth = (struct tcphdr *)(ip6 + 1); 1501 } else 1502 #endif /* INET6 */ 1503 { 1504 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1505 nth = (struct tcphdr *)(ip + 1); 1506 } 1507 if (th != nth) { 1508 /* 1509 * this is usually a case when an extension header 1510 * exists between the IPv6 header and the 1511 * TCP header. 1512 */ 1513 nth->th_sport = th->th_sport; 1514 nth->th_dport = th->th_dport; 1515 } 1516 xchg(nth->th_dport, nth->th_sport, uint16_t); 1517 #undef xchg 1518 } 1519 tlen = 0; 1520 #ifdef INET6 1521 if (isipv6) 1522 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1523 #endif 1524 #if defined(INET) && defined(INET6) 1525 else 1526 #endif 1527 #ifdef INET 1528 tlen = sizeof (struct tcpiphdr); 1529 #endif 1530 #ifdef INVARIANTS 1531 m->m_len = 0; 1532 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1533 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1534 m, tlen, (long)M_TRAILINGSPACE(m))); 1535 #endif 1536 m->m_len = tlen; 1537 to.to_flags = 0; 1538 if (incl_opts) { 1539 /* Make sure we have room. */ 1540 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1541 m->m_next = m_get(M_NOWAIT, MT_DATA); 1542 if (m->m_next) { 1543 optp = mtod(m->m_next, u_char *); 1544 optm = m->m_next; 1545 } else 1546 incl_opts = false; 1547 } else { 1548 optp = (u_char *) (nth + 1); 1549 optm = m; 1550 } 1551 } 1552 if (incl_opts) { 1553 /* Timestamps. */ 1554 if (tp->t_flags & TF_RCVD_TSTMP) { 1555 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1556 to.to_tsecr = tp->ts_recent; 1557 to.to_flags |= TOF_TS; 1558 } 1559 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1560 /* TCP-MD5 (RFC2385). */ 1561 if (tp->t_flags & TF_SIGNATURE) 1562 to.to_flags |= TOF_SIGNATURE; 1563 #endif 1564 /* Add the options. */ 1565 tlen += optlen = tcp_addoptions(&to, optp); 1566 1567 /* Update m_len in the correct mbuf. */ 1568 optm->m_len += optlen; 1569 } else 1570 optlen = 0; 1571 #ifdef INET6 1572 if (isipv6) { 1573 ip6->ip6_flow = 0; 1574 ip6->ip6_vfc = IPV6_VERSION; 1575 ip6->ip6_nxt = IPPROTO_TCP; 1576 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1577 } 1578 #endif 1579 #if defined(INET) && defined(INET6) 1580 else 1581 #endif 1582 #ifdef INET 1583 { 1584 ip->ip_len = htons(tlen); 1585 ip->ip_ttl = V_ip_defttl; 1586 if (V_path_mtu_discovery) 1587 ip->ip_off |= htons(IP_DF); 1588 } 1589 #endif 1590 m->m_pkthdr.len = tlen; 1591 m->m_pkthdr.rcvif = NULL; 1592 #ifdef MAC 1593 if (inp != NULL) { 1594 /* 1595 * Packet is associated with a socket, so allow the 1596 * label of the response to reflect the socket label. 1597 */ 1598 INP_WLOCK_ASSERT(inp); 1599 mac_inpcb_create_mbuf(inp, m); 1600 } else { 1601 /* 1602 * Packet is not associated with a socket, so possibly 1603 * update the label in place. 1604 */ 1605 mac_netinet_tcp_reply(m); 1606 } 1607 #endif 1608 nth->th_seq = htonl(seq); 1609 nth->th_ack = htonl(ack); 1610 nth->th_x2 = 0; 1611 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1612 nth->th_flags = flags; 1613 if (tp != NULL) 1614 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1615 else 1616 nth->th_win = htons((u_short)win); 1617 nth->th_urp = 0; 1618 1619 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1620 if (to.to_flags & TOF_SIGNATURE) { 1621 if (!TCPMD5_ENABLED() || 1622 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1623 m_freem(m); 1624 return; 1625 } 1626 } 1627 #endif 1628 1629 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1630 #ifdef INET6 1631 if (isipv6) { 1632 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1633 nth->th_sum = in6_cksum_pseudo(ip6, 1634 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1635 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1636 NULL, NULL); 1637 } 1638 #endif /* INET6 */ 1639 #if defined(INET6) && defined(INET) 1640 else 1641 #endif 1642 #ifdef INET 1643 { 1644 m->m_pkthdr.csum_flags = CSUM_TCP; 1645 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1646 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1647 } 1648 #endif /* INET */ 1649 #ifdef TCPDEBUG 1650 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1651 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1652 #endif 1653 TCP_PROBE3(debug__output, tp, th, m); 1654 if (flags & TH_RST) 1655 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1656 1657 #ifdef INET6 1658 if (isipv6) { 1659 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1660 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1661 } 1662 #endif /* INET6 */ 1663 #if defined(INET) && defined(INET6) 1664 else 1665 #endif 1666 #ifdef INET 1667 { 1668 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1669 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1670 } 1671 #endif 1672 } 1673 1674 /* 1675 * Create a new TCP control block, making an 1676 * empty reassembly queue and hooking it to the argument 1677 * protocol control block. The `inp' parameter must have 1678 * come from the zone allocator set up in tcp_init(). 1679 */ 1680 struct tcpcb * 1681 tcp_newtcpcb(struct inpcb *inp) 1682 { 1683 struct tcpcb_mem *tm; 1684 struct tcpcb *tp; 1685 #ifdef INET6 1686 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1687 #endif /* INET6 */ 1688 1689 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1690 if (tm == NULL) 1691 return (NULL); 1692 tp = &tm->tcb; 1693 1694 /* Initialise cc_var struct for this tcpcb. */ 1695 tp->ccv = &tm->ccv; 1696 tp->ccv->type = IPPROTO_TCP; 1697 tp->ccv->ccvc.tcp = tp; 1698 rw_rlock(&tcp_function_lock); 1699 tp->t_fb = tcp_func_set_ptr; 1700 refcount_acquire(&tp->t_fb->tfb_refcnt); 1701 rw_runlock(&tcp_function_lock); 1702 /* 1703 * Use the current system default CC algorithm. 1704 */ 1705 CC_LIST_RLOCK(); 1706 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1707 CC_ALGO(tp) = CC_DEFAULT(); 1708 CC_LIST_RUNLOCK(); 1709 /* 1710 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1711 * is called. 1712 */ 1713 in_pcbref(inp); /* Reference for tcpcb */ 1714 tp->t_inpcb = inp; 1715 1716 if (CC_ALGO(tp)->cb_init != NULL) 1717 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1718 if (tp->t_fb->tfb_tcp_fb_fini) 1719 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1720 in_pcbrele_wlocked(inp); 1721 refcount_release(&tp->t_fb->tfb_refcnt); 1722 uma_zfree(V_tcpcb_zone, tm); 1723 return (NULL); 1724 } 1725 1726 #ifdef TCP_HHOOK 1727 tp->osd = &tm->osd; 1728 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1729 if (tp->t_fb->tfb_tcp_fb_fini) 1730 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1731 in_pcbrele_wlocked(inp); 1732 refcount_release(&tp->t_fb->tfb_refcnt); 1733 uma_zfree(V_tcpcb_zone, tm); 1734 return (NULL); 1735 } 1736 #endif 1737 1738 #ifdef VIMAGE 1739 tp->t_vnet = inp->inp_vnet; 1740 #endif 1741 tp->t_timers = &tm->tt; 1742 TAILQ_INIT(&tp->t_segq); 1743 tp->t_maxseg = 1744 #ifdef INET6 1745 isipv6 ? V_tcp_v6mssdflt : 1746 #endif /* INET6 */ 1747 V_tcp_mssdflt; 1748 1749 /* Set up our timeouts. */ 1750 callout_init(&tp->t_timers->tt_rexmt, 1); 1751 callout_init(&tp->t_timers->tt_persist, 1); 1752 callout_init(&tp->t_timers->tt_keep, 1); 1753 callout_init(&tp->t_timers->tt_2msl, 1); 1754 callout_init(&tp->t_timers->tt_delack, 1); 1755 1756 if (V_tcp_do_rfc1323) 1757 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1758 if (V_tcp_do_sack) 1759 tp->t_flags |= TF_SACK_PERMIT; 1760 TAILQ_INIT(&tp->snd_holes); 1761 1762 /* 1763 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1764 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1765 * reasonable initial retransmit time. 1766 */ 1767 tp->t_srtt = TCPTV_SRTTBASE; 1768 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1769 tp->t_rttmin = tcp_rexmit_min; 1770 tp->t_rxtcur = tcp_rexmit_initial; 1771 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1772 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1773 tp->t_rcvtime = ticks; 1774 /* 1775 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1776 * because the socket may be bound to an IPv6 wildcard address, 1777 * which may match an IPv4-mapped IPv6 address. 1778 */ 1779 inp->inp_ip_ttl = V_ip_defttl; 1780 inp->inp_ppcb = tp; 1781 #ifdef TCPPCAP 1782 /* 1783 * Init the TCP PCAP queues. 1784 */ 1785 tcp_pcap_tcpcb_init(tp); 1786 #endif 1787 #ifdef TCP_BLACKBOX 1788 /* Initialize the per-TCPCB log data. */ 1789 tcp_log_tcpcbinit(tp); 1790 #endif 1791 tp->t_pacing_rate = -1; 1792 if (tp->t_fb->tfb_tcp_fb_init) { 1793 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 1794 refcount_release(&tp->t_fb->tfb_refcnt); 1795 in_pcbrele_wlocked(inp); 1796 uma_zfree(V_tcpcb_zone, tm); 1797 return (NULL); 1798 } 1799 } 1800 #ifdef STATS 1801 if (V_tcp_perconn_stats_enable == 1) 1802 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 1803 #endif 1804 return (tp); /* XXX */ 1805 } 1806 1807 /* 1808 * Switch the congestion control algorithm back to NewReno for any active 1809 * control blocks using an algorithm which is about to go away. 1810 * This ensures the CC framework can allow the unload to proceed without leaving 1811 * any dangling pointers which would trigger a panic. 1812 * Returning non-zero would inform the CC framework that something went wrong 1813 * and it would be unsafe to allow the unload to proceed. However, there is no 1814 * way for this to occur with this implementation so we always return zero. 1815 */ 1816 int 1817 tcp_ccalgounload(struct cc_algo *unload_algo) 1818 { 1819 struct cc_algo *tmpalgo; 1820 struct inpcb *inp; 1821 struct tcpcb *tp; 1822 VNET_ITERATOR_DECL(vnet_iter); 1823 1824 /* 1825 * Check all active control blocks across all network stacks and change 1826 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1827 * requires cleanup code to be run, call it. 1828 */ 1829 VNET_LIST_RLOCK(); 1830 VNET_FOREACH(vnet_iter) { 1831 CURVNET_SET(vnet_iter); 1832 INP_INFO_WLOCK(&V_tcbinfo); 1833 /* 1834 * New connections already part way through being initialised 1835 * with the CC algo we're removing will not race with this code 1836 * because the INP_INFO_WLOCK is held during initialisation. We 1837 * therefore don't enter the loop below until the connection 1838 * list has stabilised. 1839 */ 1840 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 1841 INP_WLOCK(inp); 1842 /* Important to skip tcptw structs. */ 1843 if (!(inp->inp_flags & INP_TIMEWAIT) && 1844 (tp = intotcpcb(inp)) != NULL) { 1845 /* 1846 * By holding INP_WLOCK here, we are assured 1847 * that the connection is not currently 1848 * executing inside the CC module's functions 1849 * i.e. it is safe to make the switch back to 1850 * NewReno. 1851 */ 1852 if (CC_ALGO(tp) == unload_algo) { 1853 tmpalgo = CC_ALGO(tp); 1854 if (tmpalgo->cb_destroy != NULL) 1855 tmpalgo->cb_destroy(tp->ccv); 1856 CC_DATA(tp) = NULL; 1857 /* 1858 * NewReno may allocate memory on 1859 * demand for certain stateful 1860 * configuration as needed, but is 1861 * coded to never fail on memory 1862 * allocation failure so it is a safe 1863 * fallback. 1864 */ 1865 CC_ALGO(tp) = &newreno_cc_algo; 1866 } 1867 } 1868 INP_WUNLOCK(inp); 1869 } 1870 INP_INFO_WUNLOCK(&V_tcbinfo); 1871 CURVNET_RESTORE(); 1872 } 1873 VNET_LIST_RUNLOCK(); 1874 1875 return (0); 1876 } 1877 1878 /* 1879 * Drop a TCP connection, reporting 1880 * the specified error. If connection is synchronized, 1881 * then send a RST to peer. 1882 */ 1883 struct tcpcb * 1884 tcp_drop(struct tcpcb *tp, int errno) 1885 { 1886 struct socket *so = tp->t_inpcb->inp_socket; 1887 1888 NET_EPOCH_ASSERT(); 1889 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1890 INP_WLOCK_ASSERT(tp->t_inpcb); 1891 1892 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1893 tcp_state_change(tp, TCPS_CLOSED); 1894 (void) tp->t_fb->tfb_tcp_output(tp); 1895 TCPSTAT_INC(tcps_drops); 1896 } else 1897 TCPSTAT_INC(tcps_conndrops); 1898 if (errno == ETIMEDOUT && tp->t_softerror) 1899 errno = tp->t_softerror; 1900 so->so_error = errno; 1901 return (tcp_close(tp)); 1902 } 1903 1904 void 1905 tcp_discardcb(struct tcpcb *tp) 1906 { 1907 struct inpcb *inp = tp->t_inpcb; 1908 struct socket *so = inp->inp_socket; 1909 #ifdef INET6 1910 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1911 #endif /* INET6 */ 1912 int released __unused; 1913 1914 INP_WLOCK_ASSERT(inp); 1915 1916 /* 1917 * Make sure that all of our timers are stopped before we delete the 1918 * PCB. 1919 * 1920 * If stopping a timer fails, we schedule a discard function in same 1921 * callout, and the last discard function called will take care of 1922 * deleting the tcpcb. 1923 */ 1924 tp->t_timers->tt_draincnt = 0; 1925 tcp_timer_stop(tp, TT_REXMT); 1926 tcp_timer_stop(tp, TT_PERSIST); 1927 tcp_timer_stop(tp, TT_KEEP); 1928 tcp_timer_stop(tp, TT_2MSL); 1929 tcp_timer_stop(tp, TT_DELACK); 1930 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1931 /* 1932 * Call the stop-all function of the methods, 1933 * this function should call the tcp_timer_stop() 1934 * method with each of the function specific timeouts. 1935 * That stop will be called via the tfb_tcp_timer_stop() 1936 * which should use the async drain function of the 1937 * callout system (see tcp_var.h). 1938 */ 1939 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1940 } 1941 1942 /* 1943 * If we got enough samples through the srtt filter, 1944 * save the rtt and rttvar in the routing entry. 1945 * 'Enough' is arbitrarily defined as 4 rtt samples. 1946 * 4 samples is enough for the srtt filter to converge 1947 * to within enough % of the correct value; fewer samples 1948 * and we could save a bogus rtt. The danger is not high 1949 * as tcp quickly recovers from everything. 1950 * XXX: Works very well but needs some more statistics! 1951 */ 1952 if (tp->t_rttupdated >= 4) { 1953 struct hc_metrics_lite metrics; 1954 uint32_t ssthresh; 1955 1956 bzero(&metrics, sizeof(metrics)); 1957 /* 1958 * Update the ssthresh always when the conditions below 1959 * are satisfied. This gives us better new start value 1960 * for the congestion avoidance for new connections. 1961 * ssthresh is only set if packet loss occurred on a session. 1962 * 1963 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1964 * being torn down. Ideally this code would not use 'so'. 1965 */ 1966 ssthresh = tp->snd_ssthresh; 1967 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1968 /* 1969 * convert the limit from user data bytes to 1970 * packets then to packet data bytes. 1971 */ 1972 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1973 if (ssthresh < 2) 1974 ssthresh = 2; 1975 ssthresh *= (tp->t_maxseg + 1976 #ifdef INET6 1977 (isipv6 ? sizeof (struct ip6_hdr) + 1978 sizeof (struct tcphdr) : 1979 #endif 1980 sizeof (struct tcpiphdr) 1981 #ifdef INET6 1982 ) 1983 #endif 1984 ); 1985 } else 1986 ssthresh = 0; 1987 metrics.rmx_ssthresh = ssthresh; 1988 1989 metrics.rmx_rtt = tp->t_srtt; 1990 metrics.rmx_rttvar = tp->t_rttvar; 1991 metrics.rmx_cwnd = tp->snd_cwnd; 1992 metrics.rmx_sendpipe = 0; 1993 metrics.rmx_recvpipe = 0; 1994 1995 tcp_hc_update(&inp->inp_inc, &metrics); 1996 } 1997 1998 /* free the reassembly queue, if any */ 1999 tcp_reass_flush(tp); 2000 2001 #ifdef TCP_OFFLOAD 2002 /* Disconnect offload device, if any. */ 2003 if (tp->t_flags & TF_TOE) 2004 tcp_offload_detach(tp); 2005 #endif 2006 2007 tcp_free_sackholes(tp); 2008 2009 #ifdef TCPPCAP 2010 /* Free the TCP PCAP queues. */ 2011 tcp_pcap_drain(&(tp->t_inpkts)); 2012 tcp_pcap_drain(&(tp->t_outpkts)); 2013 #endif 2014 2015 /* Allow the CC algorithm to clean up after itself. */ 2016 if (CC_ALGO(tp)->cb_destroy != NULL) 2017 CC_ALGO(tp)->cb_destroy(tp->ccv); 2018 CC_DATA(tp) = NULL; 2019 2020 #ifdef TCP_HHOOK 2021 khelp_destroy_osd(tp->osd); 2022 #endif 2023 #ifdef STATS 2024 stats_blob_destroy(tp->t_stats); 2025 #endif 2026 2027 CC_ALGO(tp) = NULL; 2028 inp->inp_ppcb = NULL; 2029 if (tp->t_timers->tt_draincnt == 0) { 2030 /* We own the last reference on tcpcb, let's free it. */ 2031 #ifdef TCP_BLACKBOX 2032 tcp_log_tcpcbfini(tp); 2033 #endif 2034 TCPSTATES_DEC(tp->t_state); 2035 if (tp->t_fb->tfb_tcp_fb_fini) 2036 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2037 refcount_release(&tp->t_fb->tfb_refcnt); 2038 tp->t_inpcb = NULL; 2039 uma_zfree(V_tcpcb_zone, tp); 2040 released = in_pcbrele_wlocked(inp); 2041 KASSERT(!released, ("%s: inp %p should not have been released " 2042 "here", __func__, inp)); 2043 } 2044 } 2045 2046 void 2047 tcp_timer_discard(void *ptp) 2048 { 2049 struct inpcb *inp; 2050 struct tcpcb *tp; 2051 struct epoch_tracker et; 2052 2053 tp = (struct tcpcb *)ptp; 2054 CURVNET_SET(tp->t_vnet); 2055 NET_EPOCH_ENTER(et); 2056 inp = tp->t_inpcb; 2057 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 2058 __func__, tp)); 2059 INP_WLOCK(inp); 2060 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 2061 ("%s: tcpcb has to be stopped here", __func__)); 2062 tp->t_timers->tt_draincnt--; 2063 if (tp->t_timers->tt_draincnt == 0) { 2064 /* We own the last reference on this tcpcb, let's free it. */ 2065 #ifdef TCP_BLACKBOX 2066 tcp_log_tcpcbfini(tp); 2067 #endif 2068 TCPSTATES_DEC(tp->t_state); 2069 if (tp->t_fb->tfb_tcp_fb_fini) 2070 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2071 refcount_release(&tp->t_fb->tfb_refcnt); 2072 tp->t_inpcb = NULL; 2073 uma_zfree(V_tcpcb_zone, tp); 2074 if (in_pcbrele_wlocked(inp)) { 2075 NET_EPOCH_EXIT(et); 2076 CURVNET_RESTORE(); 2077 return; 2078 } 2079 } 2080 INP_WUNLOCK(inp); 2081 NET_EPOCH_EXIT(et); 2082 CURVNET_RESTORE(); 2083 } 2084 2085 /* 2086 * Attempt to close a TCP control block, marking it as dropped, and freeing 2087 * the socket if we hold the only reference. 2088 */ 2089 struct tcpcb * 2090 tcp_close(struct tcpcb *tp) 2091 { 2092 struct inpcb *inp = tp->t_inpcb; 2093 struct socket *so; 2094 2095 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2096 INP_WLOCK_ASSERT(inp); 2097 2098 #ifdef TCP_OFFLOAD 2099 if (tp->t_state == TCPS_LISTEN) 2100 tcp_offload_listen_stop(tp); 2101 #endif 2102 /* 2103 * This releases the TFO pending counter resource for TFO listen 2104 * sockets as well as passively-created TFO sockets that transition 2105 * from SYN_RECEIVED to CLOSED. 2106 */ 2107 if (tp->t_tfo_pending) { 2108 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2109 tp->t_tfo_pending = NULL; 2110 } 2111 in_pcbdrop(inp); 2112 TCPSTAT_INC(tcps_closed); 2113 if (tp->t_state != TCPS_CLOSED) 2114 tcp_state_change(tp, TCPS_CLOSED); 2115 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2116 so = inp->inp_socket; 2117 soisdisconnected(so); 2118 if (inp->inp_flags & INP_SOCKREF) { 2119 KASSERT(so->so_state & SS_PROTOREF, 2120 ("tcp_close: !SS_PROTOREF")); 2121 inp->inp_flags &= ~INP_SOCKREF; 2122 INP_WUNLOCK(inp); 2123 SOCK_LOCK(so); 2124 so->so_state &= ~SS_PROTOREF; 2125 sofree(so); 2126 return (NULL); 2127 } 2128 return (tp); 2129 } 2130 2131 void 2132 tcp_drain(void) 2133 { 2134 VNET_ITERATOR_DECL(vnet_iter); 2135 2136 if (!do_tcpdrain) 2137 return; 2138 2139 VNET_LIST_RLOCK_NOSLEEP(); 2140 VNET_FOREACH(vnet_iter) { 2141 CURVNET_SET(vnet_iter); 2142 struct inpcb *inpb; 2143 struct tcpcb *tcpb; 2144 2145 /* 2146 * Walk the tcpbs, if existing, and flush the reassembly queue, 2147 * if there is one... 2148 * XXX: The "Net/3" implementation doesn't imply that the TCP 2149 * reassembly queue should be flushed, but in a situation 2150 * where we're really low on mbufs, this is potentially 2151 * useful. 2152 */ 2153 INP_INFO_WLOCK(&V_tcbinfo); 2154 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2155 INP_WLOCK(inpb); 2156 if (inpb->inp_flags & INP_TIMEWAIT) { 2157 INP_WUNLOCK(inpb); 2158 continue; 2159 } 2160 if ((tcpb = intotcpcb(inpb)) != NULL) { 2161 tcp_reass_flush(tcpb); 2162 tcp_clean_sackreport(tcpb); 2163 #ifdef TCP_BLACKBOX 2164 tcp_log_drain(tcpb); 2165 #endif 2166 #ifdef TCPPCAP 2167 if (tcp_pcap_aggressive_free) { 2168 /* Free the TCP PCAP queues. */ 2169 tcp_pcap_drain(&(tcpb->t_inpkts)); 2170 tcp_pcap_drain(&(tcpb->t_outpkts)); 2171 } 2172 #endif 2173 } 2174 INP_WUNLOCK(inpb); 2175 } 2176 INP_INFO_WUNLOCK(&V_tcbinfo); 2177 CURVNET_RESTORE(); 2178 } 2179 VNET_LIST_RUNLOCK_NOSLEEP(); 2180 } 2181 2182 /* 2183 * Notify a tcp user of an asynchronous error; 2184 * store error as soft error, but wake up user 2185 * (for now, won't do anything until can select for soft error). 2186 * 2187 * Do not wake up user since there currently is no mechanism for 2188 * reporting soft errors (yet - a kqueue filter may be added). 2189 */ 2190 static struct inpcb * 2191 tcp_notify(struct inpcb *inp, int error) 2192 { 2193 struct tcpcb *tp; 2194 2195 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2196 INP_WLOCK_ASSERT(inp); 2197 2198 if ((inp->inp_flags & INP_TIMEWAIT) || 2199 (inp->inp_flags & INP_DROPPED)) 2200 return (inp); 2201 2202 tp = intotcpcb(inp); 2203 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2204 2205 /* 2206 * Ignore some errors if we are hooked up. 2207 * If connection hasn't completed, has retransmitted several times, 2208 * and receives a second error, give up now. This is better 2209 * than waiting a long time to establish a connection that 2210 * can never complete. 2211 */ 2212 if (tp->t_state == TCPS_ESTABLISHED && 2213 (error == EHOSTUNREACH || error == ENETUNREACH || 2214 error == EHOSTDOWN)) { 2215 if (inp->inp_route.ro_nh) { 2216 NH_FREE(inp->inp_route.ro_nh); 2217 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2218 } 2219 return (inp); 2220 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2221 tp->t_softerror) { 2222 tp = tcp_drop(tp, error); 2223 if (tp != NULL) 2224 return (inp); 2225 else 2226 return (NULL); 2227 } else { 2228 tp->t_softerror = error; 2229 return (inp); 2230 } 2231 #if 0 2232 wakeup( &so->so_timeo); 2233 sorwakeup(so); 2234 sowwakeup(so); 2235 #endif 2236 } 2237 2238 static int 2239 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2240 { 2241 struct epoch_tracker et; 2242 struct inpcb *inp; 2243 struct xinpgen xig; 2244 int error; 2245 2246 if (req->newptr != NULL) 2247 return (EPERM); 2248 2249 if (req->oldptr == NULL) { 2250 int n; 2251 2252 n = V_tcbinfo.ipi_count + 2253 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2254 n += imax(n / 8, 10); 2255 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2256 return (0); 2257 } 2258 2259 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2260 return (error); 2261 2262 bzero(&xig, sizeof(xig)); 2263 xig.xig_len = sizeof xig; 2264 xig.xig_count = V_tcbinfo.ipi_count + 2265 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2266 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2267 xig.xig_sogen = so_gencnt; 2268 error = SYSCTL_OUT(req, &xig, sizeof xig); 2269 if (error) 2270 return (error); 2271 2272 error = syncache_pcblist(req); 2273 if (error) 2274 return (error); 2275 2276 NET_EPOCH_ENTER(et); 2277 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); 2278 inp != NULL; 2279 inp = CK_LIST_NEXT(inp, inp_list)) { 2280 INP_RLOCK(inp); 2281 if (inp->inp_gencnt <= xig.xig_gen) { 2282 int crerr; 2283 2284 /* 2285 * XXX: This use of cr_cansee(), introduced with 2286 * TCP state changes, is not quite right, but for 2287 * now, better than nothing. 2288 */ 2289 if (inp->inp_flags & INP_TIMEWAIT) { 2290 if (intotw(inp) != NULL) 2291 crerr = cr_cansee(req->td->td_ucred, 2292 intotw(inp)->tw_cred); 2293 else 2294 crerr = EINVAL; /* Skip this inp. */ 2295 } else 2296 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2297 if (crerr == 0) { 2298 struct xtcpcb xt; 2299 2300 tcp_inptoxtp(inp, &xt); 2301 INP_RUNLOCK(inp); 2302 error = SYSCTL_OUT(req, &xt, sizeof xt); 2303 if (error) 2304 break; 2305 else 2306 continue; 2307 } 2308 } 2309 INP_RUNLOCK(inp); 2310 } 2311 NET_EPOCH_EXIT(et); 2312 2313 if (!error) { 2314 /* 2315 * Give the user an updated idea of our state. 2316 * If the generation differs from what we told 2317 * her before, she knows that something happened 2318 * while we were processing this request, and it 2319 * might be necessary to retry. 2320 */ 2321 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2322 xig.xig_sogen = so_gencnt; 2323 xig.xig_count = V_tcbinfo.ipi_count + 2324 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2325 error = SYSCTL_OUT(req, &xig, sizeof xig); 2326 } 2327 2328 return (error); 2329 } 2330 2331 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2332 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2333 NULL, 0, tcp_pcblist, "S,xtcpcb", 2334 "List of active TCP connections"); 2335 2336 #ifdef INET 2337 static int 2338 tcp_getcred(SYSCTL_HANDLER_ARGS) 2339 { 2340 struct xucred xuc; 2341 struct sockaddr_in addrs[2]; 2342 struct epoch_tracker et; 2343 struct inpcb *inp; 2344 int error; 2345 2346 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2347 if (error) 2348 return (error); 2349 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2350 if (error) 2351 return (error); 2352 NET_EPOCH_ENTER(et); 2353 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2354 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2355 NET_EPOCH_EXIT(et); 2356 if (inp != NULL) { 2357 if (inp->inp_socket == NULL) 2358 error = ENOENT; 2359 if (error == 0) 2360 error = cr_canseeinpcb(req->td->td_ucred, inp); 2361 if (error == 0) 2362 cru2x(inp->inp_cred, &xuc); 2363 INP_RUNLOCK(inp); 2364 } else 2365 error = ENOENT; 2366 if (error == 0) 2367 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2368 return (error); 2369 } 2370 2371 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2372 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2373 0, 0, tcp_getcred, "S,xucred", 2374 "Get the xucred of a TCP connection"); 2375 #endif /* INET */ 2376 2377 #ifdef INET6 2378 static int 2379 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2380 { 2381 struct epoch_tracker et; 2382 struct xucred xuc; 2383 struct sockaddr_in6 addrs[2]; 2384 struct inpcb *inp; 2385 int error; 2386 #ifdef INET 2387 int mapped = 0; 2388 #endif 2389 2390 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2391 if (error) 2392 return (error); 2393 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2394 if (error) 2395 return (error); 2396 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2397 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2398 return (error); 2399 } 2400 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2401 #ifdef INET 2402 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2403 mapped = 1; 2404 else 2405 #endif 2406 return (EINVAL); 2407 } 2408 2409 NET_EPOCH_ENTER(et); 2410 #ifdef INET 2411 if (mapped == 1) 2412 inp = in_pcblookup(&V_tcbinfo, 2413 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2414 addrs[1].sin6_port, 2415 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2416 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2417 else 2418 #endif 2419 inp = in6_pcblookup(&V_tcbinfo, 2420 &addrs[1].sin6_addr, addrs[1].sin6_port, 2421 &addrs[0].sin6_addr, addrs[0].sin6_port, 2422 INPLOOKUP_RLOCKPCB, NULL); 2423 NET_EPOCH_EXIT(et); 2424 if (inp != NULL) { 2425 if (inp->inp_socket == NULL) 2426 error = ENOENT; 2427 if (error == 0) 2428 error = cr_canseeinpcb(req->td->td_ucred, inp); 2429 if (error == 0) 2430 cru2x(inp->inp_cred, &xuc); 2431 INP_RUNLOCK(inp); 2432 } else 2433 error = ENOENT; 2434 if (error == 0) 2435 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2436 return (error); 2437 } 2438 2439 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2440 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2441 0, 0, tcp6_getcred, "S,xucred", 2442 "Get the xucred of a TCP6 connection"); 2443 #endif /* INET6 */ 2444 2445 #ifdef INET 2446 void 2447 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2448 { 2449 struct ip *ip = vip; 2450 struct tcphdr *th; 2451 struct in_addr faddr; 2452 struct inpcb *inp; 2453 struct tcpcb *tp; 2454 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2455 struct icmp *icp; 2456 struct in_conninfo inc; 2457 tcp_seq icmp_tcp_seq; 2458 int mtu; 2459 2460 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2461 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2462 return; 2463 2464 if (cmd == PRC_MSGSIZE) 2465 notify = tcp_mtudisc_notify; 2466 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2467 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2468 cmd == PRC_TIMXCEED_INTRANS) && ip) 2469 notify = tcp_drop_syn_sent; 2470 2471 /* 2472 * Hostdead is ugly because it goes linearly through all PCBs. 2473 * XXX: We never get this from ICMP, otherwise it makes an 2474 * excellent DoS attack on machines with many connections. 2475 */ 2476 else if (cmd == PRC_HOSTDEAD) 2477 ip = NULL; 2478 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2479 return; 2480 2481 if (ip == NULL) { 2482 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2483 return; 2484 } 2485 2486 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2487 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2488 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2489 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2490 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2491 /* signal EHOSTDOWN, as it flushes the cached route */ 2492 inp = (*notify)(inp, EHOSTDOWN); 2493 goto out; 2494 } 2495 icmp_tcp_seq = th->th_seq; 2496 if (inp != NULL) { 2497 if (!(inp->inp_flags & INP_TIMEWAIT) && 2498 !(inp->inp_flags & INP_DROPPED) && 2499 !(inp->inp_socket == NULL)) { 2500 tp = intotcpcb(inp); 2501 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2502 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2503 if (cmd == PRC_MSGSIZE) { 2504 /* 2505 * MTU discovery: 2506 * If we got a needfrag set the MTU 2507 * in the route to the suggested new 2508 * value (if given) and then notify. 2509 */ 2510 mtu = ntohs(icp->icmp_nextmtu); 2511 /* 2512 * If no alternative MTU was 2513 * proposed, try the next smaller 2514 * one. 2515 */ 2516 if (!mtu) 2517 mtu = ip_next_mtu( 2518 ntohs(ip->ip_len), 1); 2519 if (mtu < V_tcp_minmss + 2520 sizeof(struct tcpiphdr)) 2521 mtu = V_tcp_minmss + 2522 sizeof(struct tcpiphdr); 2523 /* 2524 * Only process the offered MTU if it 2525 * is smaller than the current one. 2526 */ 2527 if (mtu < tp->t_maxseg + 2528 sizeof(struct tcpiphdr)) { 2529 bzero(&inc, sizeof(inc)); 2530 inc.inc_faddr = faddr; 2531 inc.inc_fibnum = 2532 inp->inp_inc.inc_fibnum; 2533 tcp_hc_updatemtu(&inc, mtu); 2534 tcp_mtudisc(inp, mtu); 2535 } 2536 } else 2537 inp = (*notify)(inp, 2538 inetctlerrmap[cmd]); 2539 } 2540 } 2541 } else { 2542 bzero(&inc, sizeof(inc)); 2543 inc.inc_fport = th->th_dport; 2544 inc.inc_lport = th->th_sport; 2545 inc.inc_faddr = faddr; 2546 inc.inc_laddr = ip->ip_src; 2547 syncache_unreach(&inc, icmp_tcp_seq); 2548 } 2549 out: 2550 if (inp != NULL) 2551 INP_WUNLOCK(inp); 2552 } 2553 #endif /* INET */ 2554 2555 #ifdef INET6 2556 void 2557 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2558 { 2559 struct in6_addr *dst; 2560 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2561 struct ip6_hdr *ip6; 2562 struct mbuf *m; 2563 struct inpcb *inp; 2564 struct tcpcb *tp; 2565 struct icmp6_hdr *icmp6; 2566 struct ip6ctlparam *ip6cp = NULL; 2567 const struct sockaddr_in6 *sa6_src = NULL; 2568 struct in_conninfo inc; 2569 struct tcp_ports { 2570 uint16_t th_sport; 2571 uint16_t th_dport; 2572 } t_ports; 2573 tcp_seq icmp_tcp_seq; 2574 unsigned int mtu; 2575 unsigned int off; 2576 2577 if (sa->sa_family != AF_INET6 || 2578 sa->sa_len != sizeof(struct sockaddr_in6)) 2579 return; 2580 2581 /* if the parameter is from icmp6, decode it. */ 2582 if (d != NULL) { 2583 ip6cp = (struct ip6ctlparam *)d; 2584 icmp6 = ip6cp->ip6c_icmp6; 2585 m = ip6cp->ip6c_m; 2586 ip6 = ip6cp->ip6c_ip6; 2587 off = ip6cp->ip6c_off; 2588 sa6_src = ip6cp->ip6c_src; 2589 dst = ip6cp->ip6c_finaldst; 2590 } else { 2591 m = NULL; 2592 ip6 = NULL; 2593 off = 0; /* fool gcc */ 2594 sa6_src = &sa6_any; 2595 dst = NULL; 2596 } 2597 2598 if (cmd == PRC_MSGSIZE) 2599 notify = tcp_mtudisc_notify; 2600 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2601 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2602 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2603 notify = tcp_drop_syn_sent; 2604 2605 /* 2606 * Hostdead is ugly because it goes linearly through all PCBs. 2607 * XXX: We never get this from ICMP, otherwise it makes an 2608 * excellent DoS attack on machines with many connections. 2609 */ 2610 else if (cmd == PRC_HOSTDEAD) 2611 ip6 = NULL; 2612 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2613 return; 2614 2615 if (ip6 == NULL) { 2616 in6_pcbnotify(&V_tcbinfo, sa, 0, 2617 (const struct sockaddr *)sa6_src, 2618 0, cmd, NULL, notify); 2619 return; 2620 } 2621 2622 /* Check if we can safely get the ports from the tcp hdr */ 2623 if (m == NULL || 2624 (m->m_pkthdr.len < 2625 (int32_t) (off + sizeof(struct tcp_ports)))) { 2626 return; 2627 } 2628 bzero(&t_ports, sizeof(struct tcp_ports)); 2629 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2630 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2631 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2632 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2633 /* signal EHOSTDOWN, as it flushes the cached route */ 2634 inp = (*notify)(inp, EHOSTDOWN); 2635 goto out; 2636 } 2637 off += sizeof(struct tcp_ports); 2638 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2639 goto out; 2640 } 2641 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2642 if (inp != NULL) { 2643 if (!(inp->inp_flags & INP_TIMEWAIT) && 2644 !(inp->inp_flags & INP_DROPPED) && 2645 !(inp->inp_socket == NULL)) { 2646 tp = intotcpcb(inp); 2647 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2648 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2649 if (cmd == PRC_MSGSIZE) { 2650 /* 2651 * MTU discovery: 2652 * If we got a needfrag set the MTU 2653 * in the route to the suggested new 2654 * value (if given) and then notify. 2655 */ 2656 mtu = ntohl(icmp6->icmp6_mtu); 2657 /* 2658 * If no alternative MTU was 2659 * proposed, or the proposed 2660 * MTU was too small, set to 2661 * the min. 2662 */ 2663 if (mtu < IPV6_MMTU) 2664 mtu = IPV6_MMTU - 8; 2665 bzero(&inc, sizeof(inc)); 2666 inc.inc_fibnum = M_GETFIB(m); 2667 inc.inc_flags |= INC_ISIPV6; 2668 inc.inc6_faddr = *dst; 2669 if (in6_setscope(&inc.inc6_faddr, 2670 m->m_pkthdr.rcvif, NULL)) 2671 goto out; 2672 /* 2673 * Only process the offered MTU if it 2674 * is smaller than the current one. 2675 */ 2676 if (mtu < tp->t_maxseg + 2677 sizeof (struct tcphdr) + 2678 sizeof (struct ip6_hdr)) { 2679 tcp_hc_updatemtu(&inc, mtu); 2680 tcp_mtudisc(inp, mtu); 2681 ICMP6STAT_INC(icp6s_pmtuchg); 2682 } 2683 } else 2684 inp = (*notify)(inp, 2685 inet6ctlerrmap[cmd]); 2686 } 2687 } 2688 } else { 2689 bzero(&inc, sizeof(inc)); 2690 inc.inc_fibnum = M_GETFIB(m); 2691 inc.inc_flags |= INC_ISIPV6; 2692 inc.inc_fport = t_ports.th_dport; 2693 inc.inc_lport = t_ports.th_sport; 2694 inc.inc6_faddr = *dst; 2695 inc.inc6_laddr = ip6->ip6_src; 2696 syncache_unreach(&inc, icmp_tcp_seq); 2697 } 2698 out: 2699 if (inp != NULL) 2700 INP_WUNLOCK(inp); 2701 } 2702 #endif /* INET6 */ 2703 2704 static uint32_t 2705 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 2706 { 2707 SIPHASH_CTX ctx; 2708 uint32_t hash[2]; 2709 2710 KASSERT(len >= SIPHASH_KEY_LENGTH, 2711 ("%s: keylen %u too short ", __func__, len)); 2712 SipHash24_Init(&ctx); 2713 SipHash_SetKey(&ctx, (uint8_t *)key); 2714 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 2715 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 2716 switch (inc->inc_flags & INC_ISIPV6) { 2717 #ifdef INET 2718 case 0: 2719 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 2720 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 2721 break; 2722 #endif 2723 #ifdef INET6 2724 case INC_ISIPV6: 2725 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 2726 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 2727 break; 2728 #endif 2729 } 2730 SipHash_Final((uint8_t *)hash, &ctx); 2731 2732 return (hash[0] ^ hash[1]); 2733 } 2734 2735 uint32_t 2736 tcp_new_ts_offset(struct in_conninfo *inc) 2737 { 2738 struct in_conninfo inc_store, *local_inc; 2739 2740 if (!V_tcp_ts_offset_per_conn) { 2741 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 2742 inc_store.inc_lport = 0; 2743 inc_store.inc_fport = 0; 2744 local_inc = &inc_store; 2745 } else { 2746 local_inc = inc; 2747 } 2748 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 2749 sizeof(V_ts_offset_secret))); 2750 } 2751 2752 /* 2753 * Following is where TCP initial sequence number generation occurs. 2754 * 2755 * There are two places where we must use initial sequence numbers: 2756 * 1. In SYN-ACK packets. 2757 * 2. In SYN packets. 2758 * 2759 * All ISNs for SYN-ACK packets are generated by the syncache. See 2760 * tcp_syncache.c for details. 2761 * 2762 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2763 * depends on this property. In addition, these ISNs should be 2764 * unguessable so as to prevent connection hijacking. To satisfy 2765 * the requirements of this situation, the algorithm outlined in 2766 * RFC 1948 is used, with only small modifications. 2767 * 2768 * Implementation details: 2769 * 2770 * Time is based off the system timer, and is corrected so that it 2771 * increases by one megabyte per second. This allows for proper 2772 * recycling on high speed LANs while still leaving over an hour 2773 * before rollover. 2774 * 2775 * As reading the *exact* system time is too expensive to be done 2776 * whenever setting up a TCP connection, we increment the time 2777 * offset in two ways. First, a small random positive increment 2778 * is added to isn_offset for each connection that is set up. 2779 * Second, the function tcp_isn_tick fires once per clock tick 2780 * and increments isn_offset as necessary so that sequence numbers 2781 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2782 * random positive increments serve only to ensure that the same 2783 * exact sequence number is never sent out twice (as could otherwise 2784 * happen when a port is recycled in less than the system tick 2785 * interval.) 2786 * 2787 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2788 * between seeding of isn_secret. This is normally set to zero, 2789 * as reseeding should not be necessary. 2790 * 2791 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2792 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 2793 * general, this means holding an exclusive (write) lock. 2794 */ 2795 2796 #define ISN_BYTES_PER_SECOND 1048576 2797 #define ISN_STATIC_INCREMENT 4096 2798 #define ISN_RANDOM_INCREMENT (4096 - 1) 2799 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 2800 2801 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 2802 VNET_DEFINE_STATIC(int, isn_last); 2803 VNET_DEFINE_STATIC(int, isn_last_reseed); 2804 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 2805 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 2806 2807 #define V_isn_secret VNET(isn_secret) 2808 #define V_isn_last VNET(isn_last) 2809 #define V_isn_last_reseed VNET(isn_last_reseed) 2810 #define V_isn_offset VNET(isn_offset) 2811 #define V_isn_offset_old VNET(isn_offset_old) 2812 2813 tcp_seq 2814 tcp_new_isn(struct in_conninfo *inc) 2815 { 2816 tcp_seq new_isn; 2817 u_int32_t projected_offset; 2818 2819 ISN_LOCK(); 2820 /* Seed if this is the first use, reseed if requested. */ 2821 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2822 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2823 < (u_int)ticks))) { 2824 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 2825 V_isn_last_reseed = ticks; 2826 } 2827 2828 /* Compute the hash and return the ISN. */ 2829 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 2830 sizeof(V_isn_secret)); 2831 V_isn_offset += ISN_STATIC_INCREMENT + 2832 (arc4random() & ISN_RANDOM_INCREMENT); 2833 if (ticks != V_isn_last) { 2834 projected_offset = V_isn_offset_old + 2835 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2836 if (SEQ_GT(projected_offset, V_isn_offset)) 2837 V_isn_offset = projected_offset; 2838 V_isn_offset_old = V_isn_offset; 2839 V_isn_last = ticks; 2840 } 2841 new_isn += V_isn_offset; 2842 ISN_UNLOCK(); 2843 return (new_isn); 2844 } 2845 2846 /* 2847 * When a specific ICMP unreachable message is received and the 2848 * connection state is SYN-SENT, drop the connection. This behavior 2849 * is controlled by the icmp_may_rst sysctl. 2850 */ 2851 struct inpcb * 2852 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2853 { 2854 struct tcpcb *tp; 2855 2856 NET_EPOCH_ASSERT(); 2857 INP_WLOCK_ASSERT(inp); 2858 2859 if ((inp->inp_flags & INP_TIMEWAIT) || 2860 (inp->inp_flags & INP_DROPPED)) 2861 return (inp); 2862 2863 tp = intotcpcb(inp); 2864 if (tp->t_state != TCPS_SYN_SENT) 2865 return (inp); 2866 2867 if (IS_FASTOPEN(tp->t_flags)) 2868 tcp_fastopen_disable_path(tp); 2869 2870 tp = tcp_drop(tp, errno); 2871 if (tp != NULL) 2872 return (inp); 2873 else 2874 return (NULL); 2875 } 2876 2877 /* 2878 * When `need fragmentation' ICMP is received, update our idea of the MSS 2879 * based on the new value. Also nudge TCP to send something, since we 2880 * know the packet we just sent was dropped. 2881 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2882 */ 2883 static struct inpcb * 2884 tcp_mtudisc_notify(struct inpcb *inp, int error) 2885 { 2886 2887 tcp_mtudisc(inp, -1); 2888 return (inp); 2889 } 2890 2891 static void 2892 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2893 { 2894 struct tcpcb *tp; 2895 struct socket *so; 2896 2897 INP_WLOCK_ASSERT(inp); 2898 if ((inp->inp_flags & INP_TIMEWAIT) || 2899 (inp->inp_flags & INP_DROPPED)) 2900 return; 2901 2902 tp = intotcpcb(inp); 2903 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2904 2905 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2906 2907 so = inp->inp_socket; 2908 SOCKBUF_LOCK(&so->so_snd); 2909 /* If the mss is larger than the socket buffer, decrease the mss. */ 2910 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2911 tp->t_maxseg = so->so_snd.sb_hiwat; 2912 SOCKBUF_UNLOCK(&so->so_snd); 2913 2914 TCPSTAT_INC(tcps_mturesent); 2915 tp->t_rtttime = 0; 2916 tp->snd_nxt = tp->snd_una; 2917 tcp_free_sackholes(tp); 2918 tp->snd_recover = tp->snd_max; 2919 if (tp->t_flags & TF_SACK_PERMIT) 2920 EXIT_FASTRECOVERY(tp->t_flags); 2921 tp->t_fb->tfb_tcp_output(tp); 2922 } 2923 2924 #ifdef INET 2925 /* 2926 * Look-up the routing entry to the peer of this inpcb. If no route 2927 * is found and it cannot be allocated, then return 0. This routine 2928 * is called by TCP routines that access the rmx structure and by 2929 * tcp_mss_update to get the peer/interface MTU. 2930 */ 2931 uint32_t 2932 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2933 { 2934 struct nhop_object *nh; 2935 struct ifnet *ifp; 2936 uint32_t maxmtu = 0; 2937 2938 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2939 2940 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2941 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 2942 if (nh == NULL) 2943 return (0); 2944 2945 ifp = nh->nh_ifp; 2946 maxmtu = nh->nh_mtu; 2947 2948 /* Report additional interface capabilities. */ 2949 if (cap != NULL) { 2950 if (ifp->if_capenable & IFCAP_TSO4 && 2951 ifp->if_hwassist & CSUM_TSO) { 2952 cap->ifcap |= CSUM_TSO; 2953 cap->tsomax = ifp->if_hw_tsomax; 2954 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2955 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2956 } 2957 } 2958 } 2959 return (maxmtu); 2960 } 2961 #endif /* INET */ 2962 2963 #ifdef INET6 2964 uint32_t 2965 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2966 { 2967 struct nhop_object *nh; 2968 struct in6_addr dst6; 2969 uint32_t scopeid; 2970 struct ifnet *ifp; 2971 uint32_t maxmtu = 0; 2972 2973 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2974 2975 if (inc->inc_flags & INC_IPV6MINMTU) 2976 return (IPV6_MMTU); 2977 2978 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2979 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2980 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 2981 if (nh == NULL) 2982 return (0); 2983 2984 ifp = nh->nh_ifp; 2985 maxmtu = nh->nh_mtu; 2986 2987 /* Report additional interface capabilities. */ 2988 if (cap != NULL) { 2989 if (ifp->if_capenable & IFCAP_TSO6 && 2990 ifp->if_hwassist & CSUM_TSO) { 2991 cap->ifcap |= CSUM_TSO; 2992 cap->tsomax = ifp->if_hw_tsomax; 2993 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2994 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2995 } 2996 } 2997 } 2998 2999 return (maxmtu); 3000 } 3001 #endif /* INET6 */ 3002 3003 /* 3004 * Calculate effective SMSS per RFC5681 definition for a given TCP 3005 * connection at its current state, taking into account SACK and etc. 3006 */ 3007 u_int 3008 tcp_maxseg(const struct tcpcb *tp) 3009 { 3010 u_int optlen; 3011 3012 if (tp->t_flags & TF_NOOPT) 3013 return (tp->t_maxseg); 3014 3015 /* 3016 * Here we have a simplified code from tcp_addoptions(), 3017 * without a proper loop, and having most of paddings hardcoded. 3018 * We might make mistakes with padding here in some edge cases, 3019 * but this is harmless, since result of tcp_maxseg() is used 3020 * only in cwnd and ssthresh estimations. 3021 */ 3022 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3023 if (tp->t_flags & TF_RCVD_TSTMP) 3024 optlen = TCPOLEN_TSTAMP_APPA; 3025 else 3026 optlen = 0; 3027 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3028 if (tp->t_flags & TF_SIGNATURE) 3029 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3030 #endif 3031 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3032 optlen += TCPOLEN_SACKHDR; 3033 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3034 optlen = PADTCPOLEN(optlen); 3035 } 3036 } else { 3037 if (tp->t_flags & TF_REQ_TSTMP) 3038 optlen = TCPOLEN_TSTAMP_APPA; 3039 else 3040 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3041 if (tp->t_flags & TF_REQ_SCALE) 3042 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3043 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3044 if (tp->t_flags & TF_SIGNATURE) 3045 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3046 #endif 3047 if (tp->t_flags & TF_SACK_PERMIT) 3048 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3049 } 3050 #undef PAD 3051 optlen = min(optlen, TCP_MAXOLEN); 3052 return (tp->t_maxseg - optlen); 3053 } 3054 3055 static int 3056 sysctl_drop(SYSCTL_HANDLER_ARGS) 3057 { 3058 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3059 struct sockaddr_storage addrs[2]; 3060 struct inpcb *inp; 3061 struct tcpcb *tp; 3062 struct tcptw *tw; 3063 struct sockaddr_in *fin, *lin; 3064 struct epoch_tracker et; 3065 #ifdef INET6 3066 struct sockaddr_in6 *fin6, *lin6; 3067 #endif 3068 int error; 3069 3070 inp = NULL; 3071 fin = lin = NULL; 3072 #ifdef INET6 3073 fin6 = lin6 = NULL; 3074 #endif 3075 error = 0; 3076 3077 if (req->oldptr != NULL || req->oldlen != 0) 3078 return (EINVAL); 3079 if (req->newptr == NULL) 3080 return (EPERM); 3081 if (req->newlen < sizeof(addrs)) 3082 return (ENOMEM); 3083 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3084 if (error) 3085 return (error); 3086 3087 switch (addrs[0].ss_family) { 3088 #ifdef INET6 3089 case AF_INET6: 3090 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3091 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3092 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3093 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3094 return (EINVAL); 3095 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3096 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3097 return (EINVAL); 3098 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3099 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3100 fin = (struct sockaddr_in *)&addrs[0]; 3101 lin = (struct sockaddr_in *)&addrs[1]; 3102 break; 3103 } 3104 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3105 if (error) 3106 return (error); 3107 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3108 if (error) 3109 return (error); 3110 break; 3111 #endif 3112 #ifdef INET 3113 case AF_INET: 3114 fin = (struct sockaddr_in *)&addrs[0]; 3115 lin = (struct sockaddr_in *)&addrs[1]; 3116 if (fin->sin_len != sizeof(struct sockaddr_in) || 3117 lin->sin_len != sizeof(struct sockaddr_in)) 3118 return (EINVAL); 3119 break; 3120 #endif 3121 default: 3122 return (EINVAL); 3123 } 3124 NET_EPOCH_ENTER(et); 3125 switch (addrs[0].ss_family) { 3126 #ifdef INET6 3127 case AF_INET6: 3128 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3129 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3130 INPLOOKUP_WLOCKPCB, NULL); 3131 break; 3132 #endif 3133 #ifdef INET 3134 case AF_INET: 3135 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3136 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3137 break; 3138 #endif 3139 } 3140 if (inp != NULL) { 3141 if (inp->inp_flags & INP_TIMEWAIT) { 3142 /* 3143 * XXXRW: There currently exists a state where an 3144 * inpcb is present, but its timewait state has been 3145 * discarded. For now, don't allow dropping of this 3146 * type of inpcb. 3147 */ 3148 tw = intotw(inp); 3149 if (tw != NULL) 3150 tcp_twclose(tw, 0); 3151 else 3152 INP_WUNLOCK(inp); 3153 } else if (!(inp->inp_flags & INP_DROPPED) && 3154 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3155 tp = intotcpcb(inp); 3156 tp = tcp_drop(tp, ECONNABORTED); 3157 if (tp != NULL) 3158 INP_WUNLOCK(inp); 3159 } else 3160 INP_WUNLOCK(inp); 3161 } else 3162 error = ESRCH; 3163 NET_EPOCH_EXIT(et); 3164 return (error); 3165 } 3166 3167 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3168 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3169 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3170 "Drop TCP connection"); 3171 3172 #ifdef KERN_TLS 3173 static int 3174 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3175 { 3176 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3177 struct sockaddr_storage addrs[2]; 3178 struct inpcb *inp; 3179 struct sockaddr_in *fin, *lin; 3180 struct epoch_tracker et; 3181 #ifdef INET6 3182 struct sockaddr_in6 *fin6, *lin6; 3183 #endif 3184 int error; 3185 3186 inp = NULL; 3187 fin = lin = NULL; 3188 #ifdef INET6 3189 fin6 = lin6 = NULL; 3190 #endif 3191 error = 0; 3192 3193 if (req->oldptr != NULL || req->oldlen != 0) 3194 return (EINVAL); 3195 if (req->newptr == NULL) 3196 return (EPERM); 3197 if (req->newlen < sizeof(addrs)) 3198 return (ENOMEM); 3199 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3200 if (error) 3201 return (error); 3202 3203 switch (addrs[0].ss_family) { 3204 #ifdef INET6 3205 case AF_INET6: 3206 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3207 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3208 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3209 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3210 return (EINVAL); 3211 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3212 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3213 return (EINVAL); 3214 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3215 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3216 fin = (struct sockaddr_in *)&addrs[0]; 3217 lin = (struct sockaddr_in *)&addrs[1]; 3218 break; 3219 } 3220 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3221 if (error) 3222 return (error); 3223 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3224 if (error) 3225 return (error); 3226 break; 3227 #endif 3228 #ifdef INET 3229 case AF_INET: 3230 fin = (struct sockaddr_in *)&addrs[0]; 3231 lin = (struct sockaddr_in *)&addrs[1]; 3232 if (fin->sin_len != sizeof(struct sockaddr_in) || 3233 lin->sin_len != sizeof(struct sockaddr_in)) 3234 return (EINVAL); 3235 break; 3236 #endif 3237 default: 3238 return (EINVAL); 3239 } 3240 NET_EPOCH_ENTER(et); 3241 switch (addrs[0].ss_family) { 3242 #ifdef INET6 3243 case AF_INET6: 3244 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3245 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3246 INPLOOKUP_WLOCKPCB, NULL); 3247 break; 3248 #endif 3249 #ifdef INET 3250 case AF_INET: 3251 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3252 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3253 break; 3254 #endif 3255 } 3256 NET_EPOCH_EXIT(et); 3257 if (inp != NULL) { 3258 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3259 inp->inp_socket == NULL) { 3260 error = ECONNRESET; 3261 INP_WUNLOCK(inp); 3262 } else { 3263 struct socket *so; 3264 3265 so = inp->inp_socket; 3266 soref(so); 3267 error = ktls_set_tx_mode(so, 3268 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3269 INP_WUNLOCK(inp); 3270 SOCK_LOCK(so); 3271 sorele(so); 3272 } 3273 } else 3274 error = ESRCH; 3275 return (error); 3276 } 3277 3278 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3279 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3280 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3281 "Switch TCP connection to SW TLS"); 3282 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3283 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3284 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3285 "Switch TCP connection to ifnet TLS"); 3286 #endif 3287 3288 /* 3289 * Generate a standardized TCP log line for use throughout the 3290 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3291 * allow use in the interrupt context. 3292 * 3293 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3294 * NB: The function may return NULL if memory allocation failed. 3295 * 3296 * Due to header inclusion and ordering limitations the struct ip 3297 * and ip6_hdr pointers have to be passed as void pointers. 3298 */ 3299 char * 3300 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3301 const void *ip6hdr) 3302 { 3303 3304 /* Is logging enabled? */ 3305 if (V_tcp_log_in_vain == 0) 3306 return (NULL); 3307 3308 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3309 } 3310 3311 char * 3312 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3313 const void *ip6hdr) 3314 { 3315 3316 /* Is logging enabled? */ 3317 if (tcp_log_debug == 0) 3318 return (NULL); 3319 3320 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3321 } 3322 3323 static char * 3324 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3325 const void *ip6hdr) 3326 { 3327 char *s, *sp; 3328 size_t size; 3329 struct ip *ip; 3330 #ifdef INET6 3331 const struct ip6_hdr *ip6; 3332 3333 ip6 = (const struct ip6_hdr *)ip6hdr; 3334 #endif /* INET6 */ 3335 ip = (struct ip *)ip4hdr; 3336 3337 /* 3338 * The log line looks like this: 3339 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3340 */ 3341 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3342 sizeof(PRINT_TH_FLAGS) + 1 + 3343 #ifdef INET6 3344 2 * INET6_ADDRSTRLEN; 3345 #else 3346 2 * INET_ADDRSTRLEN; 3347 #endif /* INET6 */ 3348 3349 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3350 if (s == NULL) 3351 return (NULL); 3352 3353 strcat(s, "TCP: ["); 3354 sp = s + strlen(s); 3355 3356 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3357 inet_ntoa_r(inc->inc_faddr, sp); 3358 sp = s + strlen(s); 3359 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3360 sp = s + strlen(s); 3361 inet_ntoa_r(inc->inc_laddr, sp); 3362 sp = s + strlen(s); 3363 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3364 #ifdef INET6 3365 } else if (inc) { 3366 ip6_sprintf(sp, &inc->inc6_faddr); 3367 sp = s + strlen(s); 3368 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3369 sp = s + strlen(s); 3370 ip6_sprintf(sp, &inc->inc6_laddr); 3371 sp = s + strlen(s); 3372 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3373 } else if (ip6 && th) { 3374 ip6_sprintf(sp, &ip6->ip6_src); 3375 sp = s + strlen(s); 3376 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3377 sp = s + strlen(s); 3378 ip6_sprintf(sp, &ip6->ip6_dst); 3379 sp = s + strlen(s); 3380 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3381 #endif /* INET6 */ 3382 #ifdef INET 3383 } else if (ip && th) { 3384 inet_ntoa_r(ip->ip_src, sp); 3385 sp = s + strlen(s); 3386 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3387 sp = s + strlen(s); 3388 inet_ntoa_r(ip->ip_dst, sp); 3389 sp = s + strlen(s); 3390 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3391 #endif /* INET */ 3392 } else { 3393 free(s, M_TCPLOG); 3394 return (NULL); 3395 } 3396 sp = s + strlen(s); 3397 if (th) 3398 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3399 if (*(s + size - 1) != '\0') 3400 panic("%s: string too long", __func__); 3401 return (s); 3402 } 3403 3404 /* 3405 * A subroutine which makes it easy to track TCP state changes with DTrace. 3406 * This function shouldn't be called for t_state initializations that don't 3407 * correspond to actual TCP state transitions. 3408 */ 3409 void 3410 tcp_state_change(struct tcpcb *tp, int newstate) 3411 { 3412 #if defined(KDTRACE_HOOKS) 3413 int pstate = tp->t_state; 3414 #endif 3415 3416 TCPSTATES_DEC(tp->t_state); 3417 TCPSTATES_INC(newstate); 3418 tp->t_state = newstate; 3419 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3420 } 3421 3422 /* 3423 * Create an external-format (``xtcpcb'') structure using the information in 3424 * the kernel-format tcpcb structure pointed to by tp. This is done to 3425 * reduce the spew of irrelevant information over this interface, to isolate 3426 * user code from changes in the kernel structure, and potentially to provide 3427 * information-hiding if we decide that some of this information should be 3428 * hidden from users. 3429 */ 3430 void 3431 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3432 { 3433 struct tcpcb *tp = intotcpcb(inp); 3434 sbintime_t now; 3435 3436 bzero(xt, sizeof(*xt)); 3437 if (inp->inp_flags & INP_TIMEWAIT) { 3438 xt->t_state = TCPS_TIME_WAIT; 3439 } else { 3440 xt->t_state = tp->t_state; 3441 xt->t_logstate = tp->t_logstate; 3442 xt->t_flags = tp->t_flags; 3443 xt->t_sndzerowin = tp->t_sndzerowin; 3444 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3445 xt->t_rcvoopack = tp->t_rcvoopack; 3446 xt->t_rcv_wnd = tp->rcv_wnd; 3447 xt->t_snd_wnd = tp->snd_wnd; 3448 xt->t_snd_cwnd = tp->snd_cwnd; 3449 xt->t_snd_ssthresh = tp->snd_ssthresh; 3450 xt->t_maxseg = tp->t_maxseg; 3451 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3452 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3453 3454 now = getsbinuptime(); 3455 #define COPYTIMER(ttt) do { \ 3456 if (callout_active(&tp->t_timers->ttt)) \ 3457 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3458 SBT_1MS; \ 3459 else \ 3460 xt->ttt = 0; \ 3461 } while (0) 3462 COPYTIMER(tt_delack); 3463 COPYTIMER(tt_rexmt); 3464 COPYTIMER(tt_persist); 3465 COPYTIMER(tt_keep); 3466 COPYTIMER(tt_2msl); 3467 #undef COPYTIMER 3468 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3469 3470 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3471 TCP_FUNCTION_NAME_LEN_MAX); 3472 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 3473 TCP_CA_NAME_MAX); 3474 #ifdef TCP_BLACKBOX 3475 (void)tcp_log_get_id(tp, xt->xt_logid); 3476 #endif 3477 } 3478 3479 xt->xt_len = sizeof(struct xtcpcb); 3480 in_pcbtoxinpcb(inp, &xt->xt_inp); 3481 if (inp->inp_socket == NULL) 3482 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3483 } 3484 3485 void 3486 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3487 { 3488 uint32_t bit, i; 3489 3490 if ((tp == NULL) || 3491 (status > TCP_EI_STATUS_MAX_VALUE) || 3492 (status == 0)) { 3493 /* Invalid */ 3494 return; 3495 } 3496 if (status > (sizeof(uint32_t) * 8)) { 3497 /* Should this be a KASSERT? */ 3498 return; 3499 } 3500 bit = 1U << (status - 1); 3501 if (bit & tp->t_end_info_status) { 3502 /* already logged */ 3503 return; 3504 } 3505 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3506 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3507 tp->t_end_info_bytes[i] = status; 3508 tp->t_end_info_status |= bit; 3509 break; 3510 } 3511 } 3512 } 3513