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