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