1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/arb.h> 46 #include <sys/callout.h> 47 #include <sys/eventhandler.h> 48 #ifdef TCP_HHOOK 49 #include <sys/hhook.h> 50 #endif 51 #include <sys/kernel.h> 52 #ifdef TCP_HHOOK 53 #include <sys/khelp.h> 54 #endif 55 #ifdef KERN_TLS 56 #include <sys/ktls.h> 57 #endif 58 #include <sys/qmath.h> 59 #include <sys/stats.h> 60 #include <sys/sysctl.h> 61 #include <sys/jail.h> 62 #include <sys/malloc.h> 63 #include <sys/refcount.h> 64 #include <sys/mbuf.h> 65 #ifdef INET6 66 #include <sys/domain.h> 67 #endif 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/sdt.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/protosw.h> 74 #include <sys/random.h> 75 76 #include <vm/uma.h> 77 78 #include <net/route.h> 79 #include <net/route/nhop.h> 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/vnet.h> 83 84 #include <netinet/in.h> 85 #include <netinet/in_fib.h> 86 #include <netinet/in_kdtrace.h> 87 #include <netinet/in_pcb.h> 88 #include <netinet/in_systm.h> 89 #include <netinet/in_var.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> 92 #include <netinet/ip_var.h> 93 #ifdef INET6 94 #include <netinet/icmp6.h> 95 #include <netinet/ip6.h> 96 #include <netinet6/in6_fib.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/ip6_var.h> 99 #include <netinet6/scope6_var.h> 100 #include <netinet6/nd6.h> 101 #endif 102 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_syncache.h> 110 #include <netinet/tcp_hpts.h> 111 #include <netinet/cc/cc.h> 112 #ifdef INET6 113 #include <netinet6/tcp6_var.h> 114 #endif 115 #include <netinet/tcpip.h> 116 #include <netinet/tcp_fastopen.h> 117 #ifdef TCPPCAP 118 #include <netinet/tcp_pcap.h> 119 #endif 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif 123 #ifdef INET6 124 #include <netinet6/ip6protosw.h> 125 #endif 126 #ifdef TCP_OFFLOAD 127 #include <netinet/tcp_offload.h> 128 #endif 129 130 #include <netipsec/ipsec_support.h> 131 132 #include <machine/in_cksum.h> 133 #include <crypto/siphash/siphash.h> 134 135 #include <security/mac/mac_framework.h> 136 137 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 138 #ifdef INET6 139 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 140 #endif 141 142 #ifdef NETFLIX_EXP_DETECTION 143 /* Sack attack detection thresholds and such */ 144 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 145 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 146 "Sack Attack detection thresholds"); 147 int32_t tcp_force_detection = 0; 148 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 149 CTLFLAG_RW, 150 &tcp_force_detection, 0, 151 "Do we force detection even if the INP has it off?"); 152 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 154 CTLFLAG_RW, 155 &tcp_sack_to_ack_thresh, 700, 156 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 157 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 159 CTLFLAG_RW, 160 &tcp_sack_to_move_thresh, 600, 161 "Percentage of sack moves we must see above (10.1 percent is 101)"); 162 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 164 CTLFLAG_RW, 165 &tcp_restoral_thresh, 550, 166 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 167 int32_t tcp_sad_decay_val = 800; 168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 169 CTLFLAG_RW, 170 &tcp_sad_decay_val, 800, 171 "The decay percentage (10.1 percent equals 101 )"); 172 int32_t tcp_map_minimum = 500; 173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 174 CTLFLAG_RW, 175 &tcp_map_minimum, 500, 176 "Number of Map enteries before we start detection"); 177 int32_t tcp_attack_on_turns_on_logging = 0; 178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 179 CTLFLAG_RW, 180 &tcp_attack_on_turns_on_logging, 0, 181 "When we have a positive hit on attack, do we turn on logging?"); 182 int32_t tcp_sad_pacing_interval = 2000; 183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 184 CTLFLAG_RW, 185 &tcp_sad_pacing_interval, 2000, 186 "What is the minimum pacing interval for a classified attacker?"); 187 188 int32_t tcp_sad_low_pps = 100; 189 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 190 CTLFLAG_RW, 191 &tcp_sad_low_pps, 100, 192 "What is the input pps that below which we do not decay?"); 193 #endif 194 195 struct rwlock tcp_function_lock; 196 197 static int 198 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 199 { 200 int error, new; 201 202 new = V_tcp_mssdflt; 203 error = sysctl_handle_int(oidp, &new, 0, req); 204 if (error == 0 && req->newptr) { 205 if (new < TCP_MINMSS) 206 error = EINVAL; 207 else 208 V_tcp_mssdflt = new; 209 } 210 return (error); 211 } 212 213 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 214 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 215 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 216 "Default TCP Maximum Segment Size"); 217 218 #ifdef INET6 219 static int 220 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 221 { 222 int error, new; 223 224 new = V_tcp_v6mssdflt; 225 error = sysctl_handle_int(oidp, &new, 0, req); 226 if (error == 0 && req->newptr) { 227 if (new < TCP_MINMSS) 228 error = EINVAL; 229 else 230 V_tcp_v6mssdflt = new; 231 } 232 return (error); 233 } 234 235 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 236 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 237 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 238 "Default TCP Maximum Segment Size for IPv6"); 239 #endif /* INET6 */ 240 241 /* 242 * Minimum MSS we accept and use. This prevents DoS attacks where 243 * we are forced to a ridiculous low MSS like 20 and send hundreds 244 * of packets instead of one. The effect scales with the available 245 * bandwidth and quickly saturates the CPU and network interface 246 * with packet generation and sending. Set to zero to disable MINMSS 247 * checking. This setting prevents us from sending too small packets. 248 */ 249 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 251 &VNET_NAME(tcp_minmss), 0, 252 "Minimum TCP Maximum Segment Size"); 253 254 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 255 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 256 &VNET_NAME(tcp_do_rfc1323), 0, 257 "Enable rfc1323 (high performance TCP) extensions"); 258 259 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 0; 260 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 261 &VNET_NAME(tcp_tolerate_missing_ts), 0, 262 "Tolerate missing TCP timestamps"); 263 264 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 266 &VNET_NAME(tcp_ts_offset_per_conn), 0, 267 "Initialize TCP timestamps per connection instead of per host pair"); 268 269 static int tcp_log_debug = 0; 270 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 271 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 272 273 static int tcp_tcbhashsize; 274 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 275 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 276 277 static int do_tcpdrain = 1; 278 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 279 "Enable tcp_drain routine for extra help when low on mbufs"); 280 281 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 282 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 283 284 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 285 #define V_icmp_may_rst VNET(icmp_may_rst) 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 287 &VNET_NAME(icmp_may_rst), 0, 288 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 289 290 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 291 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 293 &VNET_NAME(tcp_isn_reseed_interval), 0, 294 "Seconds between reseeding of ISN secret"); 295 296 static int tcp_soreceive_stream; 297 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 298 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 299 300 VNET_DEFINE(uma_zone_t, sack_hole_zone); 301 #define V_sack_hole_zone VNET(sack_hole_zone) 302 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 303 static int 304 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 305 { 306 int error; 307 uint32_t new; 308 309 new = V_tcp_map_entries_limit; 310 error = sysctl_handle_int(oidp, &new, 0, req); 311 if (error == 0 && req->newptr) { 312 /* only allow "0" and value > minimum */ 313 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 314 error = EINVAL; 315 else 316 V_tcp_map_entries_limit = new; 317 } 318 return (error); 319 } 320 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 321 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 322 &VNET_NAME(tcp_map_entries_limit), 0, 323 &sysctl_net_inet_tcp_map_limit_check, "IU", 324 "Total sendmap entries limit"); 325 326 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 327 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 328 &VNET_NAME(tcp_map_split_limit), 0, 329 "Total sendmap split entries limit"); 330 331 #ifdef TCP_HHOOK 332 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 333 #endif 334 335 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 336 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 337 #define V_ts_offset_secret VNET(ts_offset_secret) 338 339 static int tcp_default_fb_init(struct tcpcb *tp); 340 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 341 static int tcp_default_handoff_ok(struct tcpcb *tp); 342 static struct inpcb *tcp_notify(struct inpcb *, int); 343 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 344 static void tcp_mtudisc(struct inpcb *, int); 345 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 346 void *ip4hdr, const void *ip6hdr); 347 348 static struct tcp_function_block tcp_def_funcblk = { 349 .tfb_tcp_block_name = "freebsd", 350 .tfb_tcp_output = tcp_output, 351 .tfb_tcp_do_segment = tcp_do_segment, 352 .tfb_tcp_ctloutput = tcp_default_ctloutput, 353 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 354 .tfb_tcp_fb_init = tcp_default_fb_init, 355 .tfb_tcp_fb_fini = tcp_default_fb_fini, 356 }; 357 358 static int tcp_fb_cnt = 0; 359 struct tcp_funchead t_functions; 360 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 361 362 static struct tcp_function_block * 363 find_tcp_functions_locked(struct tcp_function_set *fs) 364 { 365 struct tcp_function *f; 366 struct tcp_function_block *blk=NULL; 367 368 TAILQ_FOREACH(f, &t_functions, tf_next) { 369 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 370 blk = f->tf_fb; 371 break; 372 } 373 } 374 return(blk); 375 } 376 377 static struct tcp_function_block * 378 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 379 { 380 struct tcp_function_block *rblk=NULL; 381 struct tcp_function *f; 382 383 TAILQ_FOREACH(f, &t_functions, tf_next) { 384 if (f->tf_fb == blk) { 385 rblk = blk; 386 if (s) { 387 *s = f; 388 } 389 break; 390 } 391 } 392 return (rblk); 393 } 394 395 struct tcp_function_block * 396 find_and_ref_tcp_functions(struct tcp_function_set *fs) 397 { 398 struct tcp_function_block *blk; 399 400 rw_rlock(&tcp_function_lock); 401 blk = find_tcp_functions_locked(fs); 402 if (blk) 403 refcount_acquire(&blk->tfb_refcnt); 404 rw_runlock(&tcp_function_lock); 405 return(blk); 406 } 407 408 struct tcp_function_block * 409 find_and_ref_tcp_fb(struct tcp_function_block *blk) 410 { 411 struct tcp_function_block *rblk; 412 413 rw_rlock(&tcp_function_lock); 414 rblk = find_tcp_fb_locked(blk, NULL); 415 if (rblk) 416 refcount_acquire(&rblk->tfb_refcnt); 417 rw_runlock(&tcp_function_lock); 418 return(rblk); 419 } 420 421 static struct tcp_function_block * 422 find_and_ref_tcp_default_fb(void) 423 { 424 struct tcp_function_block *rblk; 425 426 rw_rlock(&tcp_function_lock); 427 rblk = tcp_func_set_ptr; 428 refcount_acquire(&rblk->tfb_refcnt); 429 rw_runlock(&tcp_function_lock); 430 return (rblk); 431 } 432 433 void 434 tcp_switch_back_to_default(struct tcpcb *tp) 435 { 436 struct tcp_function_block *tfb; 437 438 KASSERT(tp->t_fb != &tcp_def_funcblk, 439 ("%s: called by the built-in default stack", __func__)); 440 441 /* 442 * Release the old stack. This function will either find a new one 443 * or panic. 444 */ 445 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 446 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 447 refcount_release(&tp->t_fb->tfb_refcnt); 448 449 /* 450 * Now, we'll find a new function block to use. 451 * Start by trying the current user-selected 452 * default, unless this stack is the user-selected 453 * default. 454 */ 455 tfb = find_and_ref_tcp_default_fb(); 456 if (tfb == tp->t_fb) { 457 refcount_release(&tfb->tfb_refcnt); 458 tfb = NULL; 459 } 460 /* Does the stack accept this connection? */ 461 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 462 (*tfb->tfb_tcp_handoff_ok)(tp)) { 463 refcount_release(&tfb->tfb_refcnt); 464 tfb = NULL; 465 } 466 /* Try to use that stack. */ 467 if (tfb != NULL) { 468 /* Initialize the new stack. If it succeeds, we are done. */ 469 tp->t_fb = tfb; 470 if (tp->t_fb->tfb_tcp_fb_init == NULL || 471 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 472 return; 473 474 /* 475 * Initialization failed. Release the reference count on 476 * the stack. 477 */ 478 refcount_release(&tfb->tfb_refcnt); 479 } 480 481 /* 482 * If that wasn't feasible, use the built-in default 483 * stack which is not allowed to reject anyone. 484 */ 485 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 486 if (tfb == NULL) { 487 /* there always should be a default */ 488 panic("Can't refer to tcp_def_funcblk"); 489 } 490 if (tfb->tfb_tcp_handoff_ok != NULL) { 491 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 492 /* The default stack cannot say no */ 493 panic("Default stack rejects a new session?"); 494 } 495 } 496 tp->t_fb = tfb; 497 if (tp->t_fb->tfb_tcp_fb_init != NULL && 498 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 499 /* The default stack cannot fail */ 500 panic("Default stack initialization failed"); 501 } 502 } 503 504 static int 505 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 506 { 507 int error=ENOENT; 508 struct tcp_function_set fs; 509 struct tcp_function_block *blk; 510 511 memset(&fs, 0, sizeof(fs)); 512 rw_rlock(&tcp_function_lock); 513 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 514 if (blk) { 515 /* Found him */ 516 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 517 fs.pcbcnt = blk->tfb_refcnt; 518 } 519 rw_runlock(&tcp_function_lock); 520 error = sysctl_handle_string(oidp, fs.function_set_name, 521 sizeof(fs.function_set_name), req); 522 523 /* Check for error or no change */ 524 if (error != 0 || req->newptr == NULL) 525 return(error); 526 527 rw_wlock(&tcp_function_lock); 528 blk = find_tcp_functions_locked(&fs); 529 if ((blk == NULL) || 530 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 531 error = ENOENT; 532 goto done; 533 } 534 tcp_func_set_ptr = blk; 535 done: 536 rw_wunlock(&tcp_function_lock); 537 return (error); 538 } 539 540 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 541 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 542 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 543 "Set/get the default TCP functions"); 544 545 static int 546 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 547 { 548 int error, cnt, linesz; 549 struct tcp_function *f; 550 char *buffer, *cp; 551 size_t bufsz, outsz; 552 bool alias; 553 554 cnt = 0; 555 rw_rlock(&tcp_function_lock); 556 TAILQ_FOREACH(f, &t_functions, tf_next) { 557 cnt++; 558 } 559 rw_runlock(&tcp_function_lock); 560 561 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 562 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 563 564 error = 0; 565 cp = buffer; 566 567 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 568 "Alias", "PCB count"); 569 cp += linesz; 570 bufsz -= linesz; 571 outsz = linesz; 572 573 rw_rlock(&tcp_function_lock); 574 TAILQ_FOREACH(f, &t_functions, tf_next) { 575 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 576 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 577 f->tf_fb->tfb_tcp_block_name, 578 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 579 alias ? f->tf_name : "-", 580 f->tf_fb->tfb_refcnt); 581 if (linesz >= bufsz) { 582 error = EOVERFLOW; 583 break; 584 } 585 cp += linesz; 586 bufsz -= linesz; 587 outsz += linesz; 588 } 589 rw_runlock(&tcp_function_lock); 590 if (error == 0) 591 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 592 free(buffer, M_TEMP); 593 return (error); 594 } 595 596 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 597 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 598 NULL, 0, sysctl_net_inet_list_available, "A", 599 "list available TCP Function sets"); 600 601 /* 602 * Exports one (struct tcp_function_info) for each alias/name. 603 */ 604 static int 605 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 606 { 607 int cnt, error; 608 struct tcp_function *f; 609 struct tcp_function_info tfi; 610 611 /* 612 * We don't allow writes. 613 */ 614 if (req->newptr != NULL) 615 return (EINVAL); 616 617 /* 618 * Wire the old buffer so we can directly copy the functions to 619 * user space without dropping the lock. 620 */ 621 if (req->oldptr != NULL) { 622 error = sysctl_wire_old_buffer(req, 0); 623 if (error) 624 return (error); 625 } 626 627 /* 628 * Walk the list and copy out matching entries. If INVARIANTS 629 * is compiled in, also walk the list to verify the length of 630 * the list matches what we have recorded. 631 */ 632 rw_rlock(&tcp_function_lock); 633 634 cnt = 0; 635 #ifndef INVARIANTS 636 if (req->oldptr == NULL) { 637 cnt = tcp_fb_cnt; 638 goto skip_loop; 639 } 640 #endif 641 TAILQ_FOREACH(f, &t_functions, tf_next) { 642 #ifdef INVARIANTS 643 cnt++; 644 #endif 645 if (req->oldptr != NULL) { 646 bzero(&tfi, sizeof(tfi)); 647 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 648 tfi.tfi_id = f->tf_fb->tfb_id; 649 (void)strlcpy(tfi.tfi_alias, f->tf_name, 650 sizeof(tfi.tfi_alias)); 651 (void)strlcpy(tfi.tfi_name, 652 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 653 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 654 /* 655 * Don't stop on error, as that is the 656 * mechanism we use to accumulate length 657 * information if the buffer was too short. 658 */ 659 } 660 } 661 KASSERT(cnt == tcp_fb_cnt, 662 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 663 #ifndef INVARIANTS 664 skip_loop: 665 #endif 666 rw_runlock(&tcp_function_lock); 667 if (req->oldptr == NULL) 668 error = SYSCTL_OUT(req, NULL, 669 (cnt + 1) * sizeof(struct tcp_function_info)); 670 671 return (error); 672 } 673 674 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 675 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 676 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 677 "List TCP function block name-to-ID mappings"); 678 679 /* 680 * tfb_tcp_handoff_ok() function for the default stack. 681 * Note that we'll basically try to take all comers. 682 */ 683 static int 684 tcp_default_handoff_ok(struct tcpcb *tp) 685 { 686 687 return (0); 688 } 689 690 /* 691 * tfb_tcp_fb_init() function for the default stack. 692 * 693 * This handles making sure we have appropriate timers set if you are 694 * transitioning a socket that has some amount of setup done. 695 * 696 * The init() fuction from the default can *never* return non-zero i.e. 697 * it is required to always succeed since it is the stack of last resort! 698 */ 699 static int 700 tcp_default_fb_init(struct tcpcb *tp) 701 { 702 703 struct socket *so; 704 705 INP_WLOCK_ASSERT(tp->t_inpcb); 706 707 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 708 ("%s: connection %p in unexpected state %d", __func__, tp, 709 tp->t_state)); 710 711 /* 712 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 713 * know what to do for unexpected states (which includes TIME_WAIT). 714 */ 715 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 716 return (0); 717 718 /* 719 * Make sure some kind of transmission timer is set if there is 720 * outstanding data. 721 */ 722 so = tp->t_inpcb->inp_socket; 723 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 724 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 725 tcp_timer_active(tp, TT_PERSIST))) { 726 /* 727 * If the session has established and it looks like it should 728 * be in the persist state, set the persist timer. Otherwise, 729 * set the retransmit timer. 730 */ 731 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 732 (int32_t)(tp->snd_nxt - tp->snd_una) < 733 (int32_t)sbavail(&so->so_snd)) 734 tcp_setpersist(tp); 735 else 736 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 737 } 738 739 /* All non-embryonic sessions get a keepalive timer. */ 740 if (!tcp_timer_active(tp, TT_KEEP)) 741 tcp_timer_activate(tp, TT_KEEP, 742 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 743 TP_KEEPINIT(tp)); 744 745 /* 746 * Make sure critical variables are initialized 747 * if transitioning while in Recovery. 748 */ 749 if IN_FASTRECOVERY(tp->t_flags) { 750 if (tp->sackhint.recover_fs == 0) 751 tp->sackhint.recover_fs = max(1, 752 tp->snd_nxt - tp->snd_una); 753 } 754 755 return (0); 756 } 757 758 /* 759 * tfb_tcp_fb_fini() function for the default stack. 760 * 761 * This changes state as necessary (or prudent) to prepare for another stack 762 * to assume responsibility for the connection. 763 */ 764 static void 765 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 766 { 767 768 INP_WLOCK_ASSERT(tp->t_inpcb); 769 return; 770 } 771 772 /* 773 * Target size of TCP PCB hash tables. Must be a power of two. 774 * 775 * Note that this can be overridden by the kernel environment 776 * variable net.inet.tcp.tcbhashsize 777 */ 778 #ifndef TCBHASHSIZE 779 #define TCBHASHSIZE 0 780 #endif 781 782 /* 783 * XXX 784 * Callouts should be moved into struct tcp directly. They are currently 785 * separate because the tcpcb structure is exported to userland for sysctl 786 * parsing purposes, which do not know about callouts. 787 */ 788 struct tcpcb_mem { 789 struct tcpcb tcb; 790 struct tcp_timer tt; 791 struct cc_var ccv; 792 #ifdef TCP_HHOOK 793 struct osd osd; 794 #endif 795 }; 796 797 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 798 #define V_tcpcb_zone VNET(tcpcb_zone) 799 800 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 801 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 802 803 static struct mtx isn_mtx; 804 805 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 806 #define ISN_LOCK() mtx_lock(&isn_mtx) 807 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 808 809 /* 810 * TCP initialization. 811 */ 812 static void 813 tcp_zone_change(void *tag) 814 { 815 816 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 817 uma_zone_set_max(V_tcpcb_zone, maxsockets); 818 tcp_tw_zone_change(); 819 } 820 821 static int 822 tcp_inpcb_init(void *mem, int size, int flags) 823 { 824 struct inpcb *inp = mem; 825 826 INP_LOCK_INIT(inp, "inp", "tcpinp"); 827 return (0); 828 } 829 830 /* 831 * Take a value and get the next power of 2 that doesn't overflow. 832 * Used to size the tcp_inpcb hash buckets. 833 */ 834 static int 835 maketcp_hashsize(int size) 836 { 837 int hashsize; 838 839 /* 840 * auto tune. 841 * get the next power of 2 higher than maxsockets. 842 */ 843 hashsize = 1 << fls(size); 844 /* catch overflow, and just go one power of 2 smaller */ 845 if (hashsize < size) { 846 hashsize = 1 << (fls(size) - 1); 847 } 848 return (hashsize); 849 } 850 851 static volatile int next_tcp_stack_id = 1; 852 853 /* 854 * Register a TCP function block with the name provided in the names 855 * array. (Note that this function does NOT automatically register 856 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 857 * explicitly include blk->tfb_tcp_block_name in the list of names if 858 * you wish to register the stack with that name.) 859 * 860 * Either all name registrations will succeed or all will fail. If 861 * a name registration fails, the function will update the num_names 862 * argument to point to the array index of the name that encountered 863 * the failure. 864 * 865 * Returns 0 on success, or an error code on failure. 866 */ 867 int 868 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 869 const char *names[], int *num_names) 870 { 871 struct tcp_function *n; 872 struct tcp_function_set fs; 873 int error, i; 874 875 KASSERT(names != NULL && *num_names > 0, 876 ("%s: Called with 0-length name list", __func__)); 877 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 878 KASSERT(rw_initialized(&tcp_function_lock), 879 ("%s: called too early", __func__)); 880 881 if ((blk->tfb_tcp_output == NULL) || 882 (blk->tfb_tcp_do_segment == NULL) || 883 (blk->tfb_tcp_ctloutput == NULL) || 884 (strlen(blk->tfb_tcp_block_name) == 0)) { 885 /* 886 * These functions are required and you 887 * need a name. 888 */ 889 *num_names = 0; 890 return (EINVAL); 891 } 892 if (blk->tfb_tcp_timer_stop_all || 893 blk->tfb_tcp_timer_activate || 894 blk->tfb_tcp_timer_active || 895 blk->tfb_tcp_timer_stop) { 896 /* 897 * If you define one timer function you 898 * must have them all. 899 */ 900 if ((blk->tfb_tcp_timer_stop_all == NULL) || 901 (blk->tfb_tcp_timer_activate == NULL) || 902 (blk->tfb_tcp_timer_active == NULL) || 903 (blk->tfb_tcp_timer_stop == NULL)) { 904 *num_names = 0; 905 return (EINVAL); 906 } 907 } 908 909 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 910 *num_names = 0; 911 return (EINVAL); 912 } 913 914 refcount_init(&blk->tfb_refcnt, 0); 915 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 916 for (i = 0; i < *num_names; i++) { 917 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 918 if (n == NULL) { 919 error = ENOMEM; 920 goto cleanup; 921 } 922 n->tf_fb = blk; 923 924 (void)strlcpy(fs.function_set_name, names[i], 925 sizeof(fs.function_set_name)); 926 rw_wlock(&tcp_function_lock); 927 if (find_tcp_functions_locked(&fs) != NULL) { 928 /* Duplicate name space not allowed */ 929 rw_wunlock(&tcp_function_lock); 930 free(n, M_TCPFUNCTIONS); 931 error = EALREADY; 932 goto cleanup; 933 } 934 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 935 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 936 tcp_fb_cnt++; 937 rw_wunlock(&tcp_function_lock); 938 } 939 return(0); 940 941 cleanup: 942 /* 943 * Deregister the names we just added. Because registration failed 944 * for names[i], we don't need to deregister that name. 945 */ 946 *num_names = i; 947 rw_wlock(&tcp_function_lock); 948 while (--i >= 0) { 949 TAILQ_FOREACH(n, &t_functions, tf_next) { 950 if (!strncmp(n->tf_name, names[i], 951 TCP_FUNCTION_NAME_LEN_MAX)) { 952 TAILQ_REMOVE(&t_functions, n, tf_next); 953 tcp_fb_cnt--; 954 n->tf_fb = NULL; 955 free(n, M_TCPFUNCTIONS); 956 break; 957 } 958 } 959 } 960 rw_wunlock(&tcp_function_lock); 961 return (error); 962 } 963 964 /* 965 * Register a TCP function block using the name provided in the name 966 * argument. 967 * 968 * Returns 0 on success, or an error code on failure. 969 */ 970 int 971 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 972 int wait) 973 { 974 const char *name_list[1]; 975 int num_names, rv; 976 977 num_names = 1; 978 if (name != NULL) 979 name_list[0] = name; 980 else 981 name_list[0] = blk->tfb_tcp_block_name; 982 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 983 return (rv); 984 } 985 986 /* 987 * Register a TCP function block using the name defined in 988 * blk->tfb_tcp_block_name. 989 * 990 * Returns 0 on success, or an error code on failure. 991 */ 992 int 993 register_tcp_functions(struct tcp_function_block *blk, int wait) 994 { 995 996 return (register_tcp_functions_as_name(blk, NULL, wait)); 997 } 998 999 /* 1000 * Deregister all names associated with a function block. This 1001 * functionally removes the function block from use within the system. 1002 * 1003 * When called with a true quiesce argument, mark the function block 1004 * as being removed so no more stacks will use it and determine 1005 * whether the removal would succeed. 1006 * 1007 * When called with a false quiesce argument, actually attempt the 1008 * removal. 1009 * 1010 * When called with a force argument, attempt to switch all TCBs to 1011 * use the default stack instead of returning EBUSY. 1012 * 1013 * Returns 0 on success (or if the removal would succeed, or an error 1014 * code on failure. 1015 */ 1016 int 1017 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1018 bool force) 1019 { 1020 struct tcp_function *f; 1021 1022 if (blk == &tcp_def_funcblk) { 1023 /* You can't un-register the default */ 1024 return (EPERM); 1025 } 1026 rw_wlock(&tcp_function_lock); 1027 if (blk == tcp_func_set_ptr) { 1028 /* You can't free the current default */ 1029 rw_wunlock(&tcp_function_lock); 1030 return (EBUSY); 1031 } 1032 /* Mark the block so no more stacks can use it. */ 1033 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1034 /* 1035 * If TCBs are still attached to the stack, attempt to switch them 1036 * to the default stack. 1037 */ 1038 if (force && blk->tfb_refcnt) { 1039 struct inpcb *inp; 1040 struct tcpcb *tp; 1041 VNET_ITERATOR_DECL(vnet_iter); 1042 1043 rw_wunlock(&tcp_function_lock); 1044 1045 VNET_LIST_RLOCK(); 1046 VNET_FOREACH(vnet_iter) { 1047 CURVNET_SET(vnet_iter); 1048 INP_INFO_WLOCK(&V_tcbinfo); 1049 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 1050 INP_WLOCK(inp); 1051 if (inp->inp_flags & INP_TIMEWAIT) { 1052 INP_WUNLOCK(inp); 1053 continue; 1054 } 1055 tp = intotcpcb(inp); 1056 if (tp == NULL || tp->t_fb != blk) { 1057 INP_WUNLOCK(inp); 1058 continue; 1059 } 1060 tcp_switch_back_to_default(tp); 1061 INP_WUNLOCK(inp); 1062 } 1063 INP_INFO_WUNLOCK(&V_tcbinfo); 1064 CURVNET_RESTORE(); 1065 } 1066 VNET_LIST_RUNLOCK(); 1067 1068 rw_wlock(&tcp_function_lock); 1069 } 1070 if (blk->tfb_refcnt) { 1071 /* TCBs still attached. */ 1072 rw_wunlock(&tcp_function_lock); 1073 return (EBUSY); 1074 } 1075 if (quiesce) { 1076 /* Skip removal. */ 1077 rw_wunlock(&tcp_function_lock); 1078 return (0); 1079 } 1080 /* Remove any function names that map to this function block. */ 1081 while (find_tcp_fb_locked(blk, &f) != NULL) { 1082 TAILQ_REMOVE(&t_functions, f, tf_next); 1083 tcp_fb_cnt--; 1084 f->tf_fb = NULL; 1085 free(f, M_TCPFUNCTIONS); 1086 } 1087 rw_wunlock(&tcp_function_lock); 1088 return (0); 1089 } 1090 1091 void 1092 tcp_init(void) 1093 { 1094 const char *tcbhash_tuneable; 1095 int hashsize; 1096 1097 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1098 1099 #ifdef TCP_HHOOK 1100 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1101 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1102 printf("%s: WARNING: unable to register helper hook\n", __func__); 1103 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1104 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1105 printf("%s: WARNING: unable to register helper hook\n", __func__); 1106 #endif 1107 #ifdef STATS 1108 if (tcp_stats_init()) 1109 printf("%s: WARNING: unable to initialise TCP stats\n", 1110 __func__); 1111 #endif 1112 hashsize = TCBHASHSIZE; 1113 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1114 if (hashsize == 0) { 1115 /* 1116 * Auto tune the hash size based on maxsockets. 1117 * A perfect hash would have a 1:1 mapping 1118 * (hashsize = maxsockets) however it's been 1119 * suggested that O(2) average is better. 1120 */ 1121 hashsize = maketcp_hashsize(maxsockets / 4); 1122 /* 1123 * Our historical default is 512, 1124 * do not autotune lower than this. 1125 */ 1126 if (hashsize < 512) 1127 hashsize = 512; 1128 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1129 printf("%s: %s auto tuned to %d\n", __func__, 1130 tcbhash_tuneable, hashsize); 1131 } 1132 /* 1133 * We require a hashsize to be a power of two. 1134 * Previously if it was not a power of two we would just reset it 1135 * back to 512, which could be a nasty surprise if you did not notice 1136 * the error message. 1137 * Instead what we do is clip it to the closest power of two lower 1138 * than the specified hash value. 1139 */ 1140 if (!powerof2(hashsize)) { 1141 int oldhashsize = hashsize; 1142 1143 hashsize = maketcp_hashsize(hashsize); 1144 /* prevent absurdly low value */ 1145 if (hashsize < 16) 1146 hashsize = 16; 1147 printf("%s: WARNING: TCB hash size not a power of 2, " 1148 "clipped from %d to %d.\n", __func__, oldhashsize, 1149 hashsize); 1150 } 1151 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1152 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1153 1154 /* 1155 * These have to be type stable for the benefit of the timers. 1156 */ 1157 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1158 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1159 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1160 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1161 1162 tcp_tw_init(); 1163 syncache_init(); 1164 tcp_hc_init(); 1165 1166 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1167 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1168 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1169 1170 tcp_fastopen_init(); 1171 1172 /* Skip initialization of globals for non-default instances. */ 1173 if (!IS_DEFAULT_VNET(curvnet)) 1174 return; 1175 1176 tcp_reass_global_init(); 1177 1178 /* XXX virtualize those bellow? */ 1179 tcp_delacktime = TCPTV_DELACK; 1180 tcp_keepinit = TCPTV_KEEP_INIT; 1181 tcp_keepidle = TCPTV_KEEP_IDLE; 1182 tcp_keepintvl = TCPTV_KEEPINTVL; 1183 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1184 tcp_msl = TCPTV_MSL; 1185 tcp_rexmit_initial = TCPTV_RTOBASE; 1186 if (tcp_rexmit_initial < 1) 1187 tcp_rexmit_initial = 1; 1188 tcp_rexmit_min = TCPTV_MIN; 1189 if (tcp_rexmit_min < 1) 1190 tcp_rexmit_min = 1; 1191 tcp_persmin = TCPTV_PERSMIN; 1192 tcp_persmax = TCPTV_PERSMAX; 1193 tcp_rexmit_slop = TCPTV_CPU_VAR; 1194 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1195 tcp_tcbhashsize = hashsize; 1196 1197 /* Setup the tcp function block list */ 1198 TAILQ_INIT(&t_functions); 1199 rw_init(&tcp_function_lock, "tcp_func_lock"); 1200 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1201 #ifdef TCP_BLACKBOX 1202 /* Initialize the TCP logging data. */ 1203 tcp_log_init(); 1204 #endif 1205 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1206 1207 if (tcp_soreceive_stream) { 1208 #ifdef INET 1209 tcp_usrreqs.pru_soreceive = soreceive_stream; 1210 #endif 1211 #ifdef INET6 1212 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1213 #endif /* INET6 */ 1214 } 1215 1216 #ifdef INET6 1217 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1218 #else /* INET6 */ 1219 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1220 #endif /* INET6 */ 1221 if (max_protohdr < TCP_MINPROTOHDR) 1222 max_protohdr = TCP_MINPROTOHDR; 1223 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1224 panic("tcp_init"); 1225 #undef TCP_MINPROTOHDR 1226 1227 ISN_LOCK_INIT(); 1228 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1229 SHUTDOWN_PRI_DEFAULT); 1230 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1231 EVENTHANDLER_PRI_ANY); 1232 1233 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1234 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1235 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1236 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); 1237 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1238 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); 1239 #ifdef TCPPCAP 1240 tcp_pcap_init(); 1241 #endif 1242 } 1243 1244 #ifdef VIMAGE 1245 static void 1246 tcp_destroy(void *unused __unused) 1247 { 1248 int n; 1249 #ifdef TCP_HHOOK 1250 int error; 1251 #endif 1252 1253 /* 1254 * All our processes are gone, all our sockets should be cleaned 1255 * up, which means, we should be past the tcp_discardcb() calls. 1256 * Sleep to let all tcpcb timers really disappear and cleanup. 1257 */ 1258 for (;;) { 1259 INP_LIST_RLOCK(&V_tcbinfo); 1260 n = V_tcbinfo.ipi_count; 1261 INP_LIST_RUNLOCK(&V_tcbinfo); 1262 if (n == 0) 1263 break; 1264 pause("tcpdes", hz / 10); 1265 } 1266 tcp_hc_destroy(); 1267 syncache_destroy(); 1268 tcp_tw_destroy(); 1269 in_pcbinfo_destroy(&V_tcbinfo); 1270 /* tcp_discardcb() clears the sack_holes up. */ 1271 uma_zdestroy(V_sack_hole_zone); 1272 uma_zdestroy(V_tcpcb_zone); 1273 1274 /* 1275 * Cannot free the zone until all tcpcbs are released as we attach 1276 * the allocations to them. 1277 */ 1278 tcp_fastopen_destroy(); 1279 1280 #ifdef TCP_HHOOK 1281 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1282 if (error != 0) { 1283 printf("%s: WARNING: unable to deregister helper hook " 1284 "type=%d, id=%d: error %d returned\n", __func__, 1285 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1286 } 1287 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1288 if (error != 0) { 1289 printf("%s: WARNING: unable to deregister helper hook " 1290 "type=%d, id=%d: error %d returned\n", __func__, 1291 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1292 } 1293 #endif 1294 } 1295 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1296 #endif 1297 1298 void 1299 tcp_fini(void *xtp) 1300 { 1301 1302 } 1303 1304 /* 1305 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1306 * tcp_template used to store this data in mbufs, but we now recopy it out 1307 * of the tcpcb each time to conserve mbufs. 1308 */ 1309 void 1310 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1311 { 1312 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1313 1314 INP_WLOCK_ASSERT(inp); 1315 1316 #ifdef INET6 1317 if ((inp->inp_vflag & INP_IPV6) != 0) { 1318 struct ip6_hdr *ip6; 1319 1320 ip6 = (struct ip6_hdr *)ip_ptr; 1321 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1322 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1323 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1324 (IPV6_VERSION & IPV6_VERSION_MASK); 1325 ip6->ip6_nxt = IPPROTO_TCP; 1326 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1327 ip6->ip6_src = inp->in6p_laddr; 1328 ip6->ip6_dst = inp->in6p_faddr; 1329 } 1330 #endif /* INET6 */ 1331 #if defined(INET6) && defined(INET) 1332 else 1333 #endif 1334 #ifdef INET 1335 { 1336 struct ip *ip; 1337 1338 ip = (struct ip *)ip_ptr; 1339 ip->ip_v = IPVERSION; 1340 ip->ip_hl = 5; 1341 ip->ip_tos = inp->inp_ip_tos; 1342 ip->ip_len = 0; 1343 ip->ip_id = 0; 1344 ip->ip_off = 0; 1345 ip->ip_ttl = inp->inp_ip_ttl; 1346 ip->ip_sum = 0; 1347 ip->ip_p = IPPROTO_TCP; 1348 ip->ip_src = inp->inp_laddr; 1349 ip->ip_dst = inp->inp_faddr; 1350 } 1351 #endif /* INET */ 1352 th->th_sport = inp->inp_lport; 1353 th->th_dport = inp->inp_fport; 1354 th->th_seq = 0; 1355 th->th_ack = 0; 1356 th->th_x2 = 0; 1357 th->th_off = 5; 1358 th->th_flags = 0; 1359 th->th_win = 0; 1360 th->th_urp = 0; 1361 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1362 } 1363 1364 /* 1365 * Create template to be used to send tcp packets on a connection. 1366 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1367 * use for this function is in keepalives, which use tcp_respond. 1368 */ 1369 struct tcptemp * 1370 tcpip_maketemplate(struct inpcb *inp) 1371 { 1372 struct tcptemp *t; 1373 1374 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1375 if (t == NULL) 1376 return (NULL); 1377 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1378 return (t); 1379 } 1380 1381 /* 1382 * Send a single message to the TCP at address specified by 1383 * the given TCP/IP header. If m == NULL, then we make a copy 1384 * of the tcpiphdr at th and send directly to the addressed host. 1385 * This is used to force keep alive messages out using the TCP 1386 * template for a connection. If flags are given then we send 1387 * a message back to the TCP which originated the segment th, 1388 * and discard the mbuf containing it and any other attached mbufs. 1389 * 1390 * In any case the ack and sequence number of the transmitted 1391 * segment are as specified by the parameters. 1392 * 1393 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1394 */ 1395 void 1396 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1397 tcp_seq ack, tcp_seq seq, int flags) 1398 { 1399 struct tcpopt to; 1400 struct inpcb *inp; 1401 struct ip *ip; 1402 struct mbuf *optm; 1403 struct tcphdr *nth; 1404 u_char *optp; 1405 #ifdef INET6 1406 struct ip6_hdr *ip6; 1407 int isipv6; 1408 #endif /* INET6 */ 1409 int optlen, tlen, win; 1410 bool incl_opts; 1411 1412 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1413 NET_EPOCH_ASSERT(); 1414 1415 #ifdef INET6 1416 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1417 ip6 = ipgen; 1418 #endif /* INET6 */ 1419 ip = ipgen; 1420 1421 if (tp != NULL) { 1422 inp = tp->t_inpcb; 1423 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1424 INP_WLOCK_ASSERT(inp); 1425 } else 1426 inp = NULL; 1427 1428 incl_opts = false; 1429 win = 0; 1430 if (tp != NULL) { 1431 if (!(flags & TH_RST)) { 1432 win = sbspace(&inp->inp_socket->so_rcv); 1433 if (win > TCP_MAXWIN << tp->rcv_scale) 1434 win = TCP_MAXWIN << tp->rcv_scale; 1435 } 1436 if ((tp->t_flags & TF_NOOPT) == 0) 1437 incl_opts = true; 1438 } 1439 if (m == NULL) { 1440 m = m_gethdr(M_NOWAIT, MT_DATA); 1441 if (m == NULL) 1442 return; 1443 m->m_data += max_linkhdr; 1444 #ifdef INET6 1445 if (isipv6) { 1446 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1447 sizeof(struct ip6_hdr)); 1448 ip6 = mtod(m, struct ip6_hdr *); 1449 nth = (struct tcphdr *)(ip6 + 1); 1450 } else 1451 #endif /* INET6 */ 1452 { 1453 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1454 ip = mtod(m, struct ip *); 1455 nth = (struct tcphdr *)(ip + 1); 1456 } 1457 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1458 flags = TH_ACK; 1459 } else if (!M_WRITABLE(m)) { 1460 struct mbuf *n; 1461 1462 /* Can't reuse 'm', allocate a new mbuf. */ 1463 n = m_gethdr(M_NOWAIT, MT_DATA); 1464 if (n == NULL) { 1465 m_freem(m); 1466 return; 1467 } 1468 1469 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1470 m_freem(m); 1471 m_freem(n); 1472 return; 1473 } 1474 1475 n->m_data += max_linkhdr; 1476 /* m_len is set later */ 1477 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1478 #ifdef INET6 1479 if (isipv6) { 1480 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1481 sizeof(struct ip6_hdr)); 1482 ip6 = mtod(n, struct ip6_hdr *); 1483 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1484 nth = (struct tcphdr *)(ip6 + 1); 1485 } else 1486 #endif /* INET6 */ 1487 { 1488 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1489 ip = mtod(n, struct ip *); 1490 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1491 nth = (struct tcphdr *)(ip + 1); 1492 } 1493 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1494 xchg(nth->th_dport, nth->th_sport, uint16_t); 1495 th = nth; 1496 m_freem(m); 1497 m = n; 1498 } else { 1499 /* 1500 * reuse the mbuf. 1501 * XXX MRT We inherit the FIB, which is lucky. 1502 */ 1503 m_freem(m->m_next); 1504 m->m_next = NULL; 1505 m->m_data = (caddr_t)ipgen; 1506 /* m_len is set later */ 1507 #ifdef INET6 1508 if (isipv6) { 1509 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1510 nth = (struct tcphdr *)(ip6 + 1); 1511 } else 1512 #endif /* INET6 */ 1513 { 1514 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1515 nth = (struct tcphdr *)(ip + 1); 1516 } 1517 if (th != nth) { 1518 /* 1519 * this is usually a case when an extension header 1520 * exists between the IPv6 header and the 1521 * TCP header. 1522 */ 1523 nth->th_sport = th->th_sport; 1524 nth->th_dport = th->th_dport; 1525 } 1526 xchg(nth->th_dport, nth->th_sport, uint16_t); 1527 #undef xchg 1528 } 1529 tlen = 0; 1530 #ifdef INET6 1531 if (isipv6) 1532 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1533 #endif 1534 #if defined(INET) && defined(INET6) 1535 else 1536 #endif 1537 #ifdef INET 1538 tlen = sizeof (struct tcpiphdr); 1539 #endif 1540 #ifdef INVARIANTS 1541 m->m_len = 0; 1542 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1543 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1544 m, tlen, (long)M_TRAILINGSPACE(m))); 1545 #endif 1546 m->m_len = tlen; 1547 to.to_flags = 0; 1548 if (incl_opts) { 1549 /* Make sure we have room. */ 1550 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1551 m->m_next = m_get(M_NOWAIT, MT_DATA); 1552 if (m->m_next) { 1553 optp = mtod(m->m_next, u_char *); 1554 optm = m->m_next; 1555 } else 1556 incl_opts = false; 1557 } else { 1558 optp = (u_char *) (nth + 1); 1559 optm = m; 1560 } 1561 } 1562 if (incl_opts) { 1563 /* Timestamps. */ 1564 if (tp->t_flags & TF_RCVD_TSTMP) { 1565 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1566 to.to_tsecr = tp->ts_recent; 1567 to.to_flags |= TOF_TS; 1568 } 1569 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1570 /* TCP-MD5 (RFC2385). */ 1571 if (tp->t_flags & TF_SIGNATURE) 1572 to.to_flags |= TOF_SIGNATURE; 1573 #endif 1574 /* Add the options. */ 1575 tlen += optlen = tcp_addoptions(&to, optp); 1576 1577 /* Update m_len in the correct mbuf. */ 1578 optm->m_len += optlen; 1579 } else 1580 optlen = 0; 1581 #ifdef INET6 1582 if (isipv6) { 1583 ip6->ip6_flow = 0; 1584 ip6->ip6_vfc = IPV6_VERSION; 1585 ip6->ip6_nxt = IPPROTO_TCP; 1586 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1587 } 1588 #endif 1589 #if defined(INET) && defined(INET6) 1590 else 1591 #endif 1592 #ifdef INET 1593 { 1594 ip->ip_len = htons(tlen); 1595 ip->ip_ttl = V_ip_defttl; 1596 if (V_path_mtu_discovery) 1597 ip->ip_off |= htons(IP_DF); 1598 } 1599 #endif 1600 m->m_pkthdr.len = tlen; 1601 m->m_pkthdr.rcvif = NULL; 1602 #ifdef MAC 1603 if (inp != NULL) { 1604 /* 1605 * Packet is associated with a socket, so allow the 1606 * label of the response to reflect the socket label. 1607 */ 1608 INP_WLOCK_ASSERT(inp); 1609 mac_inpcb_create_mbuf(inp, m); 1610 } else { 1611 /* 1612 * Packet is not associated with a socket, so possibly 1613 * update the label in place. 1614 */ 1615 mac_netinet_tcp_reply(m); 1616 } 1617 #endif 1618 nth->th_seq = htonl(seq); 1619 nth->th_ack = htonl(ack); 1620 nth->th_x2 = 0; 1621 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1622 nth->th_flags = flags; 1623 if (tp != NULL) 1624 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1625 else 1626 nth->th_win = htons((u_short)win); 1627 nth->th_urp = 0; 1628 1629 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1630 if (to.to_flags & TOF_SIGNATURE) { 1631 if (!TCPMD5_ENABLED() || 1632 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1633 m_freem(m); 1634 return; 1635 } 1636 } 1637 #endif 1638 1639 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1640 #ifdef INET6 1641 if (isipv6) { 1642 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1643 nth->th_sum = in6_cksum_pseudo(ip6, 1644 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1645 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1646 NULL, NULL); 1647 } 1648 #endif /* INET6 */ 1649 #if defined(INET6) && defined(INET) 1650 else 1651 #endif 1652 #ifdef INET 1653 { 1654 m->m_pkthdr.csum_flags = CSUM_TCP; 1655 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1656 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1657 } 1658 #endif /* INET */ 1659 #ifdef TCPDEBUG 1660 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1661 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1662 #endif 1663 TCP_PROBE3(debug__output, tp, th, m); 1664 if (flags & TH_RST) 1665 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1666 1667 #ifdef INET6 1668 if (isipv6) { 1669 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1670 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1671 } 1672 #endif /* INET6 */ 1673 #if defined(INET) && defined(INET6) 1674 else 1675 #endif 1676 #ifdef INET 1677 { 1678 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1679 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1680 } 1681 #endif 1682 } 1683 1684 /* 1685 * Create a new TCP control block, making an 1686 * empty reassembly queue and hooking it to the argument 1687 * protocol control block. The `inp' parameter must have 1688 * come from the zone allocator set up in tcp_init(). 1689 */ 1690 struct tcpcb * 1691 tcp_newtcpcb(struct inpcb *inp) 1692 { 1693 struct tcpcb_mem *tm; 1694 struct tcpcb *tp; 1695 #ifdef INET6 1696 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1697 #endif /* INET6 */ 1698 1699 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1700 if (tm == NULL) 1701 return (NULL); 1702 tp = &tm->tcb; 1703 1704 /* Initialise cc_var struct for this tcpcb. */ 1705 tp->ccv = &tm->ccv; 1706 tp->ccv->type = IPPROTO_TCP; 1707 tp->ccv->ccvc.tcp = tp; 1708 rw_rlock(&tcp_function_lock); 1709 tp->t_fb = tcp_func_set_ptr; 1710 refcount_acquire(&tp->t_fb->tfb_refcnt); 1711 rw_runlock(&tcp_function_lock); 1712 /* 1713 * Use the current system default CC algorithm. 1714 */ 1715 CC_LIST_RLOCK(); 1716 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1717 CC_ALGO(tp) = CC_DEFAULT(); 1718 CC_LIST_RUNLOCK(); 1719 /* 1720 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1721 * is called. 1722 */ 1723 in_pcbref(inp); /* Reference for tcpcb */ 1724 tp->t_inpcb = inp; 1725 1726 if (CC_ALGO(tp)->cb_init != NULL) 1727 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1728 if (tp->t_fb->tfb_tcp_fb_fini) 1729 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1730 in_pcbrele_wlocked(inp); 1731 refcount_release(&tp->t_fb->tfb_refcnt); 1732 uma_zfree(V_tcpcb_zone, tm); 1733 return (NULL); 1734 } 1735 1736 #ifdef TCP_HHOOK 1737 tp->osd = &tm->osd; 1738 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1739 if (tp->t_fb->tfb_tcp_fb_fini) 1740 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1741 in_pcbrele_wlocked(inp); 1742 refcount_release(&tp->t_fb->tfb_refcnt); 1743 uma_zfree(V_tcpcb_zone, tm); 1744 return (NULL); 1745 } 1746 #endif 1747 1748 #ifdef VIMAGE 1749 tp->t_vnet = inp->inp_vnet; 1750 #endif 1751 tp->t_timers = &tm->tt; 1752 TAILQ_INIT(&tp->t_segq); 1753 tp->t_maxseg = 1754 #ifdef INET6 1755 isipv6 ? V_tcp_v6mssdflt : 1756 #endif /* INET6 */ 1757 V_tcp_mssdflt; 1758 1759 /* Set up our timeouts. */ 1760 callout_init(&tp->t_timers->tt_rexmt, 1); 1761 callout_init(&tp->t_timers->tt_persist, 1); 1762 callout_init(&tp->t_timers->tt_keep, 1); 1763 callout_init(&tp->t_timers->tt_2msl, 1); 1764 callout_init(&tp->t_timers->tt_delack, 1); 1765 1766 if (V_tcp_do_rfc1323) 1767 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1768 if (V_tcp_do_sack) 1769 tp->t_flags |= TF_SACK_PERMIT; 1770 TAILQ_INIT(&tp->snd_holes); 1771 1772 /* 1773 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1774 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1775 * reasonable initial retransmit time. 1776 */ 1777 tp->t_srtt = TCPTV_SRTTBASE; 1778 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1779 tp->t_rttmin = tcp_rexmit_min; 1780 tp->t_rxtcur = tcp_rexmit_initial; 1781 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1782 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1783 tp->t_rcvtime = ticks; 1784 /* 1785 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1786 * because the socket may be bound to an IPv6 wildcard address, 1787 * which may match an IPv4-mapped IPv6 address. 1788 */ 1789 inp->inp_ip_ttl = V_ip_defttl; 1790 inp->inp_ppcb = tp; 1791 #ifdef TCPPCAP 1792 /* 1793 * Init the TCP PCAP queues. 1794 */ 1795 tcp_pcap_tcpcb_init(tp); 1796 #endif 1797 #ifdef TCP_BLACKBOX 1798 /* Initialize the per-TCPCB log data. */ 1799 tcp_log_tcpcbinit(tp); 1800 #endif 1801 tp->t_pacing_rate = -1; 1802 if (tp->t_fb->tfb_tcp_fb_init) { 1803 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 1804 refcount_release(&tp->t_fb->tfb_refcnt); 1805 in_pcbrele_wlocked(inp); 1806 uma_zfree(V_tcpcb_zone, tm); 1807 return (NULL); 1808 } 1809 } 1810 #ifdef STATS 1811 if (V_tcp_perconn_stats_enable == 1) 1812 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 1813 #endif 1814 return (tp); /* XXX */ 1815 } 1816 1817 /* 1818 * Switch the congestion control algorithm back to NewReno for any active 1819 * control blocks using an algorithm which is about to go away. 1820 * This ensures the CC framework can allow the unload to proceed without leaving 1821 * any dangling pointers which would trigger a panic. 1822 * Returning non-zero would inform the CC framework that something went wrong 1823 * and it would be unsafe to allow the unload to proceed. However, there is no 1824 * way for this to occur with this implementation so we always return zero. 1825 */ 1826 int 1827 tcp_ccalgounload(struct cc_algo *unload_algo) 1828 { 1829 struct cc_algo *tmpalgo; 1830 struct inpcb *inp; 1831 struct tcpcb *tp; 1832 VNET_ITERATOR_DECL(vnet_iter); 1833 1834 /* 1835 * Check all active control blocks across all network stacks and change 1836 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1837 * requires cleanup code to be run, call it. 1838 */ 1839 VNET_LIST_RLOCK(); 1840 VNET_FOREACH(vnet_iter) { 1841 CURVNET_SET(vnet_iter); 1842 INP_INFO_WLOCK(&V_tcbinfo); 1843 /* 1844 * New connections already part way through being initialised 1845 * with the CC algo we're removing will not race with this code 1846 * because the INP_INFO_WLOCK is held during initialisation. We 1847 * therefore don't enter the loop below until the connection 1848 * list has stabilised. 1849 */ 1850 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 1851 INP_WLOCK(inp); 1852 /* Important to skip tcptw structs. */ 1853 if (!(inp->inp_flags & INP_TIMEWAIT) && 1854 (tp = intotcpcb(inp)) != NULL) { 1855 /* 1856 * By holding INP_WLOCK here, we are assured 1857 * that the connection is not currently 1858 * executing inside the CC module's functions 1859 * i.e. it is safe to make the switch back to 1860 * NewReno. 1861 */ 1862 if (CC_ALGO(tp) == unload_algo) { 1863 tmpalgo = CC_ALGO(tp); 1864 if (tmpalgo->cb_destroy != NULL) 1865 tmpalgo->cb_destroy(tp->ccv); 1866 CC_DATA(tp) = NULL; 1867 /* 1868 * NewReno may allocate memory on 1869 * demand for certain stateful 1870 * configuration as needed, but is 1871 * coded to never fail on memory 1872 * allocation failure so it is a safe 1873 * fallback. 1874 */ 1875 CC_ALGO(tp) = &newreno_cc_algo; 1876 } 1877 } 1878 INP_WUNLOCK(inp); 1879 } 1880 INP_INFO_WUNLOCK(&V_tcbinfo); 1881 CURVNET_RESTORE(); 1882 } 1883 VNET_LIST_RUNLOCK(); 1884 1885 return (0); 1886 } 1887 1888 /* 1889 * Drop a TCP connection, reporting 1890 * the specified error. If connection is synchronized, 1891 * then send a RST to peer. 1892 */ 1893 struct tcpcb * 1894 tcp_drop(struct tcpcb *tp, int errno) 1895 { 1896 struct socket *so = tp->t_inpcb->inp_socket; 1897 1898 NET_EPOCH_ASSERT(); 1899 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1900 INP_WLOCK_ASSERT(tp->t_inpcb); 1901 1902 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1903 tcp_state_change(tp, TCPS_CLOSED); 1904 (void) tp->t_fb->tfb_tcp_output(tp); 1905 TCPSTAT_INC(tcps_drops); 1906 } else 1907 TCPSTAT_INC(tcps_conndrops); 1908 if (errno == ETIMEDOUT && tp->t_softerror) 1909 errno = tp->t_softerror; 1910 so->so_error = errno; 1911 return (tcp_close(tp)); 1912 } 1913 1914 void 1915 tcp_discardcb(struct tcpcb *tp) 1916 { 1917 struct inpcb *inp = tp->t_inpcb; 1918 struct socket *so = inp->inp_socket; 1919 #ifdef INET6 1920 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1921 #endif /* INET6 */ 1922 int released __unused; 1923 1924 INP_WLOCK_ASSERT(inp); 1925 1926 /* 1927 * Make sure that all of our timers are stopped before we delete the 1928 * PCB. 1929 * 1930 * If stopping a timer fails, we schedule a discard function in same 1931 * callout, and the last discard function called will take care of 1932 * deleting the tcpcb. 1933 */ 1934 tp->t_timers->tt_draincnt = 0; 1935 tcp_timer_stop(tp, TT_REXMT); 1936 tcp_timer_stop(tp, TT_PERSIST); 1937 tcp_timer_stop(tp, TT_KEEP); 1938 tcp_timer_stop(tp, TT_2MSL); 1939 tcp_timer_stop(tp, TT_DELACK); 1940 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1941 /* 1942 * Call the stop-all function of the methods, 1943 * this function should call the tcp_timer_stop() 1944 * method with each of the function specific timeouts. 1945 * That stop will be called via the tfb_tcp_timer_stop() 1946 * which should use the async drain function of the 1947 * callout system (see tcp_var.h). 1948 */ 1949 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1950 } 1951 1952 /* 1953 * If we got enough samples through the srtt filter, 1954 * save the rtt and rttvar in the routing entry. 1955 * 'Enough' is arbitrarily defined as 4 rtt samples. 1956 * 4 samples is enough for the srtt filter to converge 1957 * to within enough % of the correct value; fewer samples 1958 * and we could save a bogus rtt. The danger is not high 1959 * as tcp quickly recovers from everything. 1960 * XXX: Works very well but needs some more statistics! 1961 */ 1962 if (tp->t_rttupdated >= 4) { 1963 struct hc_metrics_lite metrics; 1964 uint32_t ssthresh; 1965 1966 bzero(&metrics, sizeof(metrics)); 1967 /* 1968 * Update the ssthresh always when the conditions below 1969 * are satisfied. This gives us better new start value 1970 * for the congestion avoidance for new connections. 1971 * ssthresh is only set if packet loss occurred on a session. 1972 * 1973 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1974 * being torn down. Ideally this code would not use 'so'. 1975 */ 1976 ssthresh = tp->snd_ssthresh; 1977 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1978 /* 1979 * convert the limit from user data bytes to 1980 * packets then to packet data bytes. 1981 */ 1982 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1983 if (ssthresh < 2) 1984 ssthresh = 2; 1985 ssthresh *= (tp->t_maxseg + 1986 #ifdef INET6 1987 (isipv6 ? sizeof (struct ip6_hdr) + 1988 sizeof (struct tcphdr) : 1989 #endif 1990 sizeof (struct tcpiphdr) 1991 #ifdef INET6 1992 ) 1993 #endif 1994 ); 1995 } else 1996 ssthresh = 0; 1997 metrics.rmx_ssthresh = ssthresh; 1998 1999 metrics.rmx_rtt = tp->t_srtt; 2000 metrics.rmx_rttvar = tp->t_rttvar; 2001 metrics.rmx_cwnd = tp->snd_cwnd; 2002 metrics.rmx_sendpipe = 0; 2003 metrics.rmx_recvpipe = 0; 2004 2005 tcp_hc_update(&inp->inp_inc, &metrics); 2006 } 2007 2008 /* free the reassembly queue, if any */ 2009 tcp_reass_flush(tp); 2010 2011 #ifdef TCP_OFFLOAD 2012 /* Disconnect offload device, if any. */ 2013 if (tp->t_flags & TF_TOE) 2014 tcp_offload_detach(tp); 2015 #endif 2016 2017 tcp_free_sackholes(tp); 2018 2019 #ifdef TCPPCAP 2020 /* Free the TCP PCAP queues. */ 2021 tcp_pcap_drain(&(tp->t_inpkts)); 2022 tcp_pcap_drain(&(tp->t_outpkts)); 2023 #endif 2024 2025 /* Allow the CC algorithm to clean up after itself. */ 2026 if (CC_ALGO(tp)->cb_destroy != NULL) 2027 CC_ALGO(tp)->cb_destroy(tp->ccv); 2028 CC_DATA(tp) = NULL; 2029 2030 #ifdef TCP_HHOOK 2031 khelp_destroy_osd(tp->osd); 2032 #endif 2033 #ifdef STATS 2034 stats_blob_destroy(tp->t_stats); 2035 #endif 2036 2037 CC_ALGO(tp) = NULL; 2038 inp->inp_ppcb = NULL; 2039 if (tp->t_timers->tt_draincnt == 0) { 2040 /* We own the last reference on tcpcb, let's free it. */ 2041 #ifdef TCP_BLACKBOX 2042 tcp_log_tcpcbfini(tp); 2043 #endif 2044 TCPSTATES_DEC(tp->t_state); 2045 if (tp->t_fb->tfb_tcp_fb_fini) 2046 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2047 refcount_release(&tp->t_fb->tfb_refcnt); 2048 tp->t_inpcb = NULL; 2049 uma_zfree(V_tcpcb_zone, tp); 2050 released = in_pcbrele_wlocked(inp); 2051 KASSERT(!released, ("%s: inp %p should not have been released " 2052 "here", __func__, inp)); 2053 } 2054 } 2055 2056 void 2057 tcp_timer_discard(void *ptp) 2058 { 2059 struct inpcb *inp; 2060 struct tcpcb *tp; 2061 struct epoch_tracker et; 2062 2063 tp = (struct tcpcb *)ptp; 2064 CURVNET_SET(tp->t_vnet); 2065 NET_EPOCH_ENTER(et); 2066 inp = tp->t_inpcb; 2067 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 2068 __func__, tp)); 2069 INP_WLOCK(inp); 2070 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 2071 ("%s: tcpcb has to be stopped here", __func__)); 2072 tp->t_timers->tt_draincnt--; 2073 if (tp->t_timers->tt_draincnt == 0) { 2074 /* We own the last reference on this tcpcb, let's free it. */ 2075 #ifdef TCP_BLACKBOX 2076 tcp_log_tcpcbfini(tp); 2077 #endif 2078 TCPSTATES_DEC(tp->t_state); 2079 if (tp->t_fb->tfb_tcp_fb_fini) 2080 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2081 refcount_release(&tp->t_fb->tfb_refcnt); 2082 tp->t_inpcb = NULL; 2083 uma_zfree(V_tcpcb_zone, tp); 2084 if (in_pcbrele_wlocked(inp)) { 2085 NET_EPOCH_EXIT(et); 2086 CURVNET_RESTORE(); 2087 return; 2088 } 2089 } 2090 INP_WUNLOCK(inp); 2091 NET_EPOCH_EXIT(et); 2092 CURVNET_RESTORE(); 2093 } 2094 2095 /* 2096 * Attempt to close a TCP control block, marking it as dropped, and freeing 2097 * the socket if we hold the only reference. 2098 */ 2099 struct tcpcb * 2100 tcp_close(struct tcpcb *tp) 2101 { 2102 struct inpcb *inp = tp->t_inpcb; 2103 struct socket *so; 2104 2105 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2106 INP_WLOCK_ASSERT(inp); 2107 2108 #ifdef TCP_OFFLOAD 2109 if (tp->t_state == TCPS_LISTEN) 2110 tcp_offload_listen_stop(tp); 2111 #endif 2112 /* 2113 * This releases the TFO pending counter resource for TFO listen 2114 * sockets as well as passively-created TFO sockets that transition 2115 * from SYN_RECEIVED to CLOSED. 2116 */ 2117 if (tp->t_tfo_pending) { 2118 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2119 tp->t_tfo_pending = NULL; 2120 } 2121 in_pcbdrop(inp); 2122 TCPSTAT_INC(tcps_closed); 2123 if (tp->t_state != TCPS_CLOSED) 2124 tcp_state_change(tp, TCPS_CLOSED); 2125 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2126 so = inp->inp_socket; 2127 soisdisconnected(so); 2128 if (inp->inp_flags & INP_SOCKREF) { 2129 KASSERT(so->so_state & SS_PROTOREF, 2130 ("tcp_close: !SS_PROTOREF")); 2131 inp->inp_flags &= ~INP_SOCKREF; 2132 INP_WUNLOCK(inp); 2133 SOCK_LOCK(so); 2134 so->so_state &= ~SS_PROTOREF; 2135 sofree(so); 2136 return (NULL); 2137 } 2138 return (tp); 2139 } 2140 2141 void 2142 tcp_drain(void) 2143 { 2144 VNET_ITERATOR_DECL(vnet_iter); 2145 2146 if (!do_tcpdrain) 2147 return; 2148 2149 VNET_LIST_RLOCK_NOSLEEP(); 2150 VNET_FOREACH(vnet_iter) { 2151 CURVNET_SET(vnet_iter); 2152 struct inpcb *inpb; 2153 struct tcpcb *tcpb; 2154 2155 /* 2156 * Walk the tcpbs, if existing, and flush the reassembly queue, 2157 * if there is one... 2158 * XXX: The "Net/3" implementation doesn't imply that the TCP 2159 * reassembly queue should be flushed, but in a situation 2160 * where we're really low on mbufs, this is potentially 2161 * useful. 2162 */ 2163 INP_INFO_WLOCK(&V_tcbinfo); 2164 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2165 INP_WLOCK(inpb); 2166 if (inpb->inp_flags & INP_TIMEWAIT) { 2167 INP_WUNLOCK(inpb); 2168 continue; 2169 } 2170 if ((tcpb = intotcpcb(inpb)) != NULL) { 2171 tcp_reass_flush(tcpb); 2172 tcp_clean_sackreport(tcpb); 2173 #ifdef TCP_BLACKBOX 2174 tcp_log_drain(tcpb); 2175 #endif 2176 #ifdef TCPPCAP 2177 if (tcp_pcap_aggressive_free) { 2178 /* Free the TCP PCAP queues. */ 2179 tcp_pcap_drain(&(tcpb->t_inpkts)); 2180 tcp_pcap_drain(&(tcpb->t_outpkts)); 2181 } 2182 #endif 2183 } 2184 INP_WUNLOCK(inpb); 2185 } 2186 INP_INFO_WUNLOCK(&V_tcbinfo); 2187 CURVNET_RESTORE(); 2188 } 2189 VNET_LIST_RUNLOCK_NOSLEEP(); 2190 } 2191 2192 /* 2193 * Notify a tcp user of an asynchronous error; 2194 * store error as soft error, but wake up user 2195 * (for now, won't do anything until can select for soft error). 2196 * 2197 * Do not wake up user since there currently is no mechanism for 2198 * reporting soft errors (yet - a kqueue filter may be added). 2199 */ 2200 static struct inpcb * 2201 tcp_notify(struct inpcb *inp, int error) 2202 { 2203 struct tcpcb *tp; 2204 2205 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2206 INP_WLOCK_ASSERT(inp); 2207 2208 if ((inp->inp_flags & INP_TIMEWAIT) || 2209 (inp->inp_flags & INP_DROPPED)) 2210 return (inp); 2211 2212 tp = intotcpcb(inp); 2213 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2214 2215 /* 2216 * Ignore some errors if we are hooked up. 2217 * If connection hasn't completed, has retransmitted several times, 2218 * and receives a second error, give up now. This is better 2219 * than waiting a long time to establish a connection that 2220 * can never complete. 2221 */ 2222 if (tp->t_state == TCPS_ESTABLISHED && 2223 (error == EHOSTUNREACH || error == ENETUNREACH || 2224 error == EHOSTDOWN)) { 2225 if (inp->inp_route.ro_nh) { 2226 NH_FREE(inp->inp_route.ro_nh); 2227 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2228 } 2229 return (inp); 2230 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2231 tp->t_softerror) { 2232 tp = tcp_drop(tp, error); 2233 if (tp != NULL) 2234 return (inp); 2235 else 2236 return (NULL); 2237 } else { 2238 tp->t_softerror = error; 2239 return (inp); 2240 } 2241 #if 0 2242 wakeup( &so->so_timeo); 2243 sorwakeup(so); 2244 sowwakeup(so); 2245 #endif 2246 } 2247 2248 static int 2249 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2250 { 2251 struct epoch_tracker et; 2252 struct inpcb *inp; 2253 struct xinpgen xig; 2254 int error; 2255 2256 if (req->newptr != NULL) 2257 return (EPERM); 2258 2259 if (req->oldptr == NULL) { 2260 int n; 2261 2262 n = V_tcbinfo.ipi_count + 2263 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2264 n += imax(n / 8, 10); 2265 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2266 return (0); 2267 } 2268 2269 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2270 return (error); 2271 2272 bzero(&xig, sizeof(xig)); 2273 xig.xig_len = sizeof xig; 2274 xig.xig_count = V_tcbinfo.ipi_count + 2275 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2276 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2277 xig.xig_sogen = so_gencnt; 2278 error = SYSCTL_OUT(req, &xig, sizeof xig); 2279 if (error) 2280 return (error); 2281 2282 error = syncache_pcblist(req); 2283 if (error) 2284 return (error); 2285 2286 NET_EPOCH_ENTER(et); 2287 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); 2288 inp != NULL; 2289 inp = CK_LIST_NEXT(inp, inp_list)) { 2290 INP_RLOCK(inp); 2291 if (inp->inp_gencnt <= xig.xig_gen) { 2292 int crerr; 2293 2294 /* 2295 * XXX: This use of cr_cansee(), introduced with 2296 * TCP state changes, is not quite right, but for 2297 * now, better than nothing. 2298 */ 2299 if (inp->inp_flags & INP_TIMEWAIT) { 2300 if (intotw(inp) != NULL) 2301 crerr = cr_cansee(req->td->td_ucred, 2302 intotw(inp)->tw_cred); 2303 else 2304 crerr = EINVAL; /* Skip this inp. */ 2305 } else 2306 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2307 if (crerr == 0) { 2308 struct xtcpcb xt; 2309 2310 tcp_inptoxtp(inp, &xt); 2311 INP_RUNLOCK(inp); 2312 error = SYSCTL_OUT(req, &xt, sizeof xt); 2313 if (error) 2314 break; 2315 else 2316 continue; 2317 } 2318 } 2319 INP_RUNLOCK(inp); 2320 } 2321 NET_EPOCH_EXIT(et); 2322 2323 if (!error) { 2324 /* 2325 * Give the user an updated idea of our state. 2326 * If the generation differs from what we told 2327 * her before, she knows that something happened 2328 * while we were processing this request, and it 2329 * might be necessary to retry. 2330 */ 2331 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2332 xig.xig_sogen = so_gencnt; 2333 xig.xig_count = V_tcbinfo.ipi_count + 2334 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2335 error = SYSCTL_OUT(req, &xig, sizeof xig); 2336 } 2337 2338 return (error); 2339 } 2340 2341 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2342 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2343 NULL, 0, tcp_pcblist, "S,xtcpcb", 2344 "List of active TCP connections"); 2345 2346 #ifdef INET 2347 static int 2348 tcp_getcred(SYSCTL_HANDLER_ARGS) 2349 { 2350 struct xucred xuc; 2351 struct sockaddr_in addrs[2]; 2352 struct epoch_tracker et; 2353 struct inpcb *inp; 2354 int error; 2355 2356 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2357 if (error) 2358 return (error); 2359 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2360 if (error) 2361 return (error); 2362 NET_EPOCH_ENTER(et); 2363 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2364 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2365 NET_EPOCH_EXIT(et); 2366 if (inp != NULL) { 2367 if (inp->inp_socket == NULL) 2368 error = ENOENT; 2369 if (error == 0) 2370 error = cr_canseeinpcb(req->td->td_ucred, inp); 2371 if (error == 0) 2372 cru2x(inp->inp_cred, &xuc); 2373 INP_RUNLOCK(inp); 2374 } else 2375 error = ENOENT; 2376 if (error == 0) 2377 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2378 return (error); 2379 } 2380 2381 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2382 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2383 0, 0, tcp_getcred, "S,xucred", 2384 "Get the xucred of a TCP connection"); 2385 #endif /* INET */ 2386 2387 #ifdef INET6 2388 static int 2389 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2390 { 2391 struct epoch_tracker et; 2392 struct xucred xuc; 2393 struct sockaddr_in6 addrs[2]; 2394 struct inpcb *inp; 2395 int error; 2396 #ifdef INET 2397 int mapped = 0; 2398 #endif 2399 2400 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2401 if (error) 2402 return (error); 2403 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2404 if (error) 2405 return (error); 2406 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2407 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2408 return (error); 2409 } 2410 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2411 #ifdef INET 2412 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2413 mapped = 1; 2414 else 2415 #endif 2416 return (EINVAL); 2417 } 2418 2419 NET_EPOCH_ENTER(et); 2420 #ifdef INET 2421 if (mapped == 1) 2422 inp = in_pcblookup(&V_tcbinfo, 2423 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2424 addrs[1].sin6_port, 2425 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2426 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2427 else 2428 #endif 2429 inp = in6_pcblookup(&V_tcbinfo, 2430 &addrs[1].sin6_addr, addrs[1].sin6_port, 2431 &addrs[0].sin6_addr, addrs[0].sin6_port, 2432 INPLOOKUP_RLOCKPCB, NULL); 2433 NET_EPOCH_EXIT(et); 2434 if (inp != NULL) { 2435 if (inp->inp_socket == NULL) 2436 error = ENOENT; 2437 if (error == 0) 2438 error = cr_canseeinpcb(req->td->td_ucred, inp); 2439 if (error == 0) 2440 cru2x(inp->inp_cred, &xuc); 2441 INP_RUNLOCK(inp); 2442 } else 2443 error = ENOENT; 2444 if (error == 0) 2445 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2446 return (error); 2447 } 2448 2449 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2450 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2451 0, 0, tcp6_getcred, "S,xucred", 2452 "Get the xucred of a TCP6 connection"); 2453 #endif /* INET6 */ 2454 2455 #ifdef INET 2456 void 2457 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2458 { 2459 struct ip *ip = vip; 2460 struct tcphdr *th; 2461 struct in_addr faddr; 2462 struct inpcb *inp; 2463 struct tcpcb *tp; 2464 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2465 struct icmp *icp; 2466 struct in_conninfo inc; 2467 tcp_seq icmp_tcp_seq; 2468 int mtu; 2469 2470 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2471 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2472 return; 2473 2474 if (cmd == PRC_MSGSIZE) 2475 notify = tcp_mtudisc_notify; 2476 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2477 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2478 cmd == PRC_TIMXCEED_INTRANS) && ip) 2479 notify = tcp_drop_syn_sent; 2480 2481 /* 2482 * Hostdead is ugly because it goes linearly through all PCBs. 2483 * XXX: We never get this from ICMP, otherwise it makes an 2484 * excellent DoS attack on machines with many connections. 2485 */ 2486 else if (cmd == PRC_HOSTDEAD) 2487 ip = NULL; 2488 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2489 return; 2490 2491 if (ip == NULL) { 2492 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2493 return; 2494 } 2495 2496 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2497 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2498 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2499 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2500 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2501 /* signal EHOSTDOWN, as it flushes the cached route */ 2502 inp = (*notify)(inp, EHOSTDOWN); 2503 goto out; 2504 } 2505 icmp_tcp_seq = th->th_seq; 2506 if (inp != NULL) { 2507 if (!(inp->inp_flags & INP_TIMEWAIT) && 2508 !(inp->inp_flags & INP_DROPPED) && 2509 !(inp->inp_socket == NULL)) { 2510 tp = intotcpcb(inp); 2511 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2512 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2513 if (cmd == PRC_MSGSIZE) { 2514 /* 2515 * MTU discovery: 2516 * If we got a needfrag set the MTU 2517 * in the route to the suggested new 2518 * value (if given) and then notify. 2519 */ 2520 mtu = ntohs(icp->icmp_nextmtu); 2521 /* 2522 * If no alternative MTU was 2523 * proposed, try the next smaller 2524 * one. 2525 */ 2526 if (!mtu) 2527 mtu = ip_next_mtu( 2528 ntohs(ip->ip_len), 1); 2529 if (mtu < V_tcp_minmss + 2530 sizeof(struct tcpiphdr)) 2531 mtu = V_tcp_minmss + 2532 sizeof(struct tcpiphdr); 2533 /* 2534 * Only process the offered MTU if it 2535 * is smaller than the current one. 2536 */ 2537 if (mtu < tp->t_maxseg + 2538 sizeof(struct tcpiphdr)) { 2539 bzero(&inc, sizeof(inc)); 2540 inc.inc_faddr = faddr; 2541 inc.inc_fibnum = 2542 inp->inp_inc.inc_fibnum; 2543 tcp_hc_updatemtu(&inc, mtu); 2544 tcp_mtudisc(inp, mtu); 2545 } 2546 } else 2547 inp = (*notify)(inp, 2548 inetctlerrmap[cmd]); 2549 } 2550 } 2551 } else { 2552 bzero(&inc, sizeof(inc)); 2553 inc.inc_fport = th->th_dport; 2554 inc.inc_lport = th->th_sport; 2555 inc.inc_faddr = faddr; 2556 inc.inc_laddr = ip->ip_src; 2557 syncache_unreach(&inc, icmp_tcp_seq); 2558 } 2559 out: 2560 if (inp != NULL) 2561 INP_WUNLOCK(inp); 2562 } 2563 #endif /* INET */ 2564 2565 #ifdef INET6 2566 void 2567 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2568 { 2569 struct in6_addr *dst; 2570 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2571 struct ip6_hdr *ip6; 2572 struct mbuf *m; 2573 struct inpcb *inp; 2574 struct tcpcb *tp; 2575 struct icmp6_hdr *icmp6; 2576 struct ip6ctlparam *ip6cp = NULL; 2577 const struct sockaddr_in6 *sa6_src = NULL; 2578 struct in_conninfo inc; 2579 struct tcp_ports { 2580 uint16_t th_sport; 2581 uint16_t th_dport; 2582 } t_ports; 2583 tcp_seq icmp_tcp_seq; 2584 unsigned int mtu; 2585 unsigned int off; 2586 2587 if (sa->sa_family != AF_INET6 || 2588 sa->sa_len != sizeof(struct sockaddr_in6)) 2589 return; 2590 2591 /* if the parameter is from icmp6, decode it. */ 2592 if (d != NULL) { 2593 ip6cp = (struct ip6ctlparam *)d; 2594 icmp6 = ip6cp->ip6c_icmp6; 2595 m = ip6cp->ip6c_m; 2596 ip6 = ip6cp->ip6c_ip6; 2597 off = ip6cp->ip6c_off; 2598 sa6_src = ip6cp->ip6c_src; 2599 dst = ip6cp->ip6c_finaldst; 2600 } else { 2601 m = NULL; 2602 ip6 = NULL; 2603 off = 0; /* fool gcc */ 2604 sa6_src = &sa6_any; 2605 dst = NULL; 2606 } 2607 2608 if (cmd == PRC_MSGSIZE) 2609 notify = tcp_mtudisc_notify; 2610 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2611 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2612 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2613 notify = tcp_drop_syn_sent; 2614 2615 /* 2616 * Hostdead is ugly because it goes linearly through all PCBs. 2617 * XXX: We never get this from ICMP, otherwise it makes an 2618 * excellent DoS attack on machines with many connections. 2619 */ 2620 else if (cmd == PRC_HOSTDEAD) 2621 ip6 = NULL; 2622 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2623 return; 2624 2625 if (ip6 == NULL) { 2626 in6_pcbnotify(&V_tcbinfo, sa, 0, 2627 (const struct sockaddr *)sa6_src, 2628 0, cmd, NULL, notify); 2629 return; 2630 } 2631 2632 /* Check if we can safely get the ports from the tcp hdr */ 2633 if (m == NULL || 2634 (m->m_pkthdr.len < 2635 (int32_t) (off + sizeof(struct tcp_ports)))) { 2636 return; 2637 } 2638 bzero(&t_ports, sizeof(struct tcp_ports)); 2639 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2640 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2641 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2642 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2643 /* signal EHOSTDOWN, as it flushes the cached route */ 2644 inp = (*notify)(inp, EHOSTDOWN); 2645 goto out; 2646 } 2647 off += sizeof(struct tcp_ports); 2648 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2649 goto out; 2650 } 2651 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2652 if (inp != NULL) { 2653 if (!(inp->inp_flags & INP_TIMEWAIT) && 2654 !(inp->inp_flags & INP_DROPPED) && 2655 !(inp->inp_socket == NULL)) { 2656 tp = intotcpcb(inp); 2657 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2658 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2659 if (cmd == PRC_MSGSIZE) { 2660 /* 2661 * MTU discovery: 2662 * If we got a needfrag set the MTU 2663 * in the route to the suggested new 2664 * value (if given) and then notify. 2665 */ 2666 mtu = ntohl(icmp6->icmp6_mtu); 2667 /* 2668 * If no alternative MTU was 2669 * proposed, or the proposed 2670 * MTU was too small, set to 2671 * the min. 2672 */ 2673 if (mtu < IPV6_MMTU) 2674 mtu = IPV6_MMTU - 8; 2675 bzero(&inc, sizeof(inc)); 2676 inc.inc_fibnum = M_GETFIB(m); 2677 inc.inc_flags |= INC_ISIPV6; 2678 inc.inc6_faddr = *dst; 2679 if (in6_setscope(&inc.inc6_faddr, 2680 m->m_pkthdr.rcvif, NULL)) 2681 goto out; 2682 /* 2683 * Only process the offered MTU if it 2684 * is smaller than the current one. 2685 */ 2686 if (mtu < tp->t_maxseg + 2687 sizeof (struct tcphdr) + 2688 sizeof (struct ip6_hdr)) { 2689 tcp_hc_updatemtu(&inc, mtu); 2690 tcp_mtudisc(inp, mtu); 2691 ICMP6STAT_INC(icp6s_pmtuchg); 2692 } 2693 } else 2694 inp = (*notify)(inp, 2695 inet6ctlerrmap[cmd]); 2696 } 2697 } 2698 } else { 2699 bzero(&inc, sizeof(inc)); 2700 inc.inc_fibnum = M_GETFIB(m); 2701 inc.inc_flags |= INC_ISIPV6; 2702 inc.inc_fport = t_ports.th_dport; 2703 inc.inc_lport = t_ports.th_sport; 2704 inc.inc6_faddr = *dst; 2705 inc.inc6_laddr = ip6->ip6_src; 2706 syncache_unreach(&inc, icmp_tcp_seq); 2707 } 2708 out: 2709 if (inp != NULL) 2710 INP_WUNLOCK(inp); 2711 } 2712 #endif /* INET6 */ 2713 2714 static uint32_t 2715 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 2716 { 2717 SIPHASH_CTX ctx; 2718 uint32_t hash[2]; 2719 2720 KASSERT(len >= SIPHASH_KEY_LENGTH, 2721 ("%s: keylen %u too short ", __func__, len)); 2722 SipHash24_Init(&ctx); 2723 SipHash_SetKey(&ctx, (uint8_t *)key); 2724 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 2725 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 2726 switch (inc->inc_flags & INC_ISIPV6) { 2727 #ifdef INET 2728 case 0: 2729 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 2730 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 2731 break; 2732 #endif 2733 #ifdef INET6 2734 case INC_ISIPV6: 2735 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 2736 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 2737 break; 2738 #endif 2739 } 2740 SipHash_Final((uint8_t *)hash, &ctx); 2741 2742 return (hash[0] ^ hash[1]); 2743 } 2744 2745 uint32_t 2746 tcp_new_ts_offset(struct in_conninfo *inc) 2747 { 2748 struct in_conninfo inc_store, *local_inc; 2749 2750 if (!V_tcp_ts_offset_per_conn) { 2751 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 2752 inc_store.inc_lport = 0; 2753 inc_store.inc_fport = 0; 2754 local_inc = &inc_store; 2755 } else { 2756 local_inc = inc; 2757 } 2758 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 2759 sizeof(V_ts_offset_secret))); 2760 } 2761 2762 /* 2763 * Following is where TCP initial sequence number generation occurs. 2764 * 2765 * There are two places where we must use initial sequence numbers: 2766 * 1. In SYN-ACK packets. 2767 * 2. In SYN packets. 2768 * 2769 * All ISNs for SYN-ACK packets are generated by the syncache. See 2770 * tcp_syncache.c for details. 2771 * 2772 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2773 * depends on this property. In addition, these ISNs should be 2774 * unguessable so as to prevent connection hijacking. To satisfy 2775 * the requirements of this situation, the algorithm outlined in 2776 * RFC 1948 is used, with only small modifications. 2777 * 2778 * Implementation details: 2779 * 2780 * Time is based off the system timer, and is corrected so that it 2781 * increases by one megabyte per second. This allows for proper 2782 * recycling on high speed LANs while still leaving over an hour 2783 * before rollover. 2784 * 2785 * As reading the *exact* system time is too expensive to be done 2786 * whenever setting up a TCP connection, we increment the time 2787 * offset in two ways. First, a small random positive increment 2788 * is added to isn_offset for each connection that is set up. 2789 * Second, the function tcp_isn_tick fires once per clock tick 2790 * and increments isn_offset as necessary so that sequence numbers 2791 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2792 * random positive increments serve only to ensure that the same 2793 * exact sequence number is never sent out twice (as could otherwise 2794 * happen when a port is recycled in less than the system tick 2795 * interval.) 2796 * 2797 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2798 * between seeding of isn_secret. This is normally set to zero, 2799 * as reseeding should not be necessary. 2800 * 2801 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2802 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 2803 * general, this means holding an exclusive (write) lock. 2804 */ 2805 2806 #define ISN_BYTES_PER_SECOND 1048576 2807 #define ISN_STATIC_INCREMENT 4096 2808 #define ISN_RANDOM_INCREMENT (4096 - 1) 2809 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 2810 2811 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 2812 VNET_DEFINE_STATIC(int, isn_last); 2813 VNET_DEFINE_STATIC(int, isn_last_reseed); 2814 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 2815 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 2816 2817 #define V_isn_secret VNET(isn_secret) 2818 #define V_isn_last VNET(isn_last) 2819 #define V_isn_last_reseed VNET(isn_last_reseed) 2820 #define V_isn_offset VNET(isn_offset) 2821 #define V_isn_offset_old VNET(isn_offset_old) 2822 2823 tcp_seq 2824 tcp_new_isn(struct in_conninfo *inc) 2825 { 2826 tcp_seq new_isn; 2827 u_int32_t projected_offset; 2828 2829 ISN_LOCK(); 2830 /* Seed if this is the first use, reseed if requested. */ 2831 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2832 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2833 < (u_int)ticks))) { 2834 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 2835 V_isn_last_reseed = ticks; 2836 } 2837 2838 /* Compute the hash and return the ISN. */ 2839 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 2840 sizeof(V_isn_secret)); 2841 V_isn_offset += ISN_STATIC_INCREMENT + 2842 (arc4random() & ISN_RANDOM_INCREMENT); 2843 if (ticks != V_isn_last) { 2844 projected_offset = V_isn_offset_old + 2845 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2846 if (SEQ_GT(projected_offset, V_isn_offset)) 2847 V_isn_offset = projected_offset; 2848 V_isn_offset_old = V_isn_offset; 2849 V_isn_last = ticks; 2850 } 2851 new_isn += V_isn_offset; 2852 ISN_UNLOCK(); 2853 return (new_isn); 2854 } 2855 2856 /* 2857 * When a specific ICMP unreachable message is received and the 2858 * connection state is SYN-SENT, drop the connection. This behavior 2859 * is controlled by the icmp_may_rst sysctl. 2860 */ 2861 struct inpcb * 2862 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2863 { 2864 struct tcpcb *tp; 2865 2866 NET_EPOCH_ASSERT(); 2867 INP_WLOCK_ASSERT(inp); 2868 2869 if ((inp->inp_flags & INP_TIMEWAIT) || 2870 (inp->inp_flags & INP_DROPPED)) 2871 return (inp); 2872 2873 tp = intotcpcb(inp); 2874 if (tp->t_state != TCPS_SYN_SENT) 2875 return (inp); 2876 2877 if (IS_FASTOPEN(tp->t_flags)) 2878 tcp_fastopen_disable_path(tp); 2879 2880 tp = tcp_drop(tp, errno); 2881 if (tp != NULL) 2882 return (inp); 2883 else 2884 return (NULL); 2885 } 2886 2887 /* 2888 * When `need fragmentation' ICMP is received, update our idea of the MSS 2889 * based on the new value. Also nudge TCP to send something, since we 2890 * know the packet we just sent was dropped. 2891 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2892 */ 2893 static struct inpcb * 2894 tcp_mtudisc_notify(struct inpcb *inp, int error) 2895 { 2896 2897 tcp_mtudisc(inp, -1); 2898 return (inp); 2899 } 2900 2901 static void 2902 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2903 { 2904 struct tcpcb *tp; 2905 struct socket *so; 2906 2907 INP_WLOCK_ASSERT(inp); 2908 if ((inp->inp_flags & INP_TIMEWAIT) || 2909 (inp->inp_flags & INP_DROPPED)) 2910 return; 2911 2912 tp = intotcpcb(inp); 2913 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2914 2915 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2916 2917 so = inp->inp_socket; 2918 SOCKBUF_LOCK(&so->so_snd); 2919 /* If the mss is larger than the socket buffer, decrease the mss. */ 2920 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2921 tp->t_maxseg = so->so_snd.sb_hiwat; 2922 SOCKBUF_UNLOCK(&so->so_snd); 2923 2924 TCPSTAT_INC(tcps_mturesent); 2925 tp->t_rtttime = 0; 2926 tp->snd_nxt = tp->snd_una; 2927 tcp_free_sackholes(tp); 2928 tp->snd_recover = tp->snd_max; 2929 if (tp->t_flags & TF_SACK_PERMIT) 2930 EXIT_FASTRECOVERY(tp->t_flags); 2931 tp->t_fb->tfb_tcp_output(tp); 2932 } 2933 2934 #ifdef INET 2935 /* 2936 * Look-up the routing entry to the peer of this inpcb. If no route 2937 * is found and it cannot be allocated, then return 0. This routine 2938 * is called by TCP routines that access the rmx structure and by 2939 * tcp_mss_update to get the peer/interface MTU. 2940 */ 2941 uint32_t 2942 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2943 { 2944 struct nhop_object *nh; 2945 struct ifnet *ifp; 2946 uint32_t maxmtu = 0; 2947 2948 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2949 2950 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2951 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 2952 if (nh == NULL) 2953 return (0); 2954 2955 ifp = nh->nh_ifp; 2956 maxmtu = nh->nh_mtu; 2957 2958 /* Report additional interface capabilities. */ 2959 if (cap != NULL) { 2960 if (ifp->if_capenable & IFCAP_TSO4 && 2961 ifp->if_hwassist & CSUM_TSO) { 2962 cap->ifcap |= CSUM_TSO; 2963 cap->tsomax = ifp->if_hw_tsomax; 2964 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2965 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2966 } 2967 } 2968 } 2969 return (maxmtu); 2970 } 2971 #endif /* INET */ 2972 2973 #ifdef INET6 2974 uint32_t 2975 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2976 { 2977 struct nhop_object *nh; 2978 struct in6_addr dst6; 2979 uint32_t scopeid; 2980 struct ifnet *ifp; 2981 uint32_t maxmtu = 0; 2982 2983 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2984 2985 if (inc->inc_flags & INC_IPV6MINMTU) 2986 return (IPV6_MMTU); 2987 2988 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2989 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2990 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 2991 if (nh == NULL) 2992 return (0); 2993 2994 ifp = nh->nh_ifp; 2995 maxmtu = nh->nh_mtu; 2996 2997 /* Report additional interface capabilities. */ 2998 if (cap != NULL) { 2999 if (ifp->if_capenable & IFCAP_TSO6 && 3000 ifp->if_hwassist & CSUM_TSO) { 3001 cap->ifcap |= CSUM_TSO; 3002 cap->tsomax = ifp->if_hw_tsomax; 3003 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3004 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3005 } 3006 } 3007 } 3008 3009 return (maxmtu); 3010 } 3011 #endif /* INET6 */ 3012 3013 /* 3014 * Calculate effective SMSS per RFC5681 definition for a given TCP 3015 * connection at its current state, taking into account SACK and etc. 3016 */ 3017 u_int 3018 tcp_maxseg(const struct tcpcb *tp) 3019 { 3020 u_int optlen; 3021 3022 if (tp->t_flags & TF_NOOPT) 3023 return (tp->t_maxseg); 3024 3025 /* 3026 * Here we have a simplified code from tcp_addoptions(), 3027 * without a proper loop, and having most of paddings hardcoded. 3028 * We might make mistakes with padding here in some edge cases, 3029 * but this is harmless, since result of tcp_maxseg() is used 3030 * only in cwnd and ssthresh estimations. 3031 */ 3032 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3033 if (tp->t_flags & TF_RCVD_TSTMP) 3034 optlen = TCPOLEN_TSTAMP_APPA; 3035 else 3036 optlen = 0; 3037 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3038 if (tp->t_flags & TF_SIGNATURE) 3039 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3040 #endif 3041 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3042 optlen += TCPOLEN_SACKHDR; 3043 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3044 optlen = PADTCPOLEN(optlen); 3045 } 3046 } else { 3047 if (tp->t_flags & TF_REQ_TSTMP) 3048 optlen = TCPOLEN_TSTAMP_APPA; 3049 else 3050 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3051 if (tp->t_flags & TF_REQ_SCALE) 3052 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3053 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3054 if (tp->t_flags & TF_SIGNATURE) 3055 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3056 #endif 3057 if (tp->t_flags & TF_SACK_PERMIT) 3058 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3059 } 3060 #undef PAD 3061 optlen = min(optlen, TCP_MAXOLEN); 3062 return (tp->t_maxseg - optlen); 3063 } 3064 3065 static int 3066 sysctl_drop(SYSCTL_HANDLER_ARGS) 3067 { 3068 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3069 struct sockaddr_storage addrs[2]; 3070 struct inpcb *inp; 3071 struct tcpcb *tp; 3072 struct tcptw *tw; 3073 struct sockaddr_in *fin, *lin; 3074 struct epoch_tracker et; 3075 #ifdef INET6 3076 struct sockaddr_in6 *fin6, *lin6; 3077 #endif 3078 int error; 3079 3080 inp = NULL; 3081 fin = lin = NULL; 3082 #ifdef INET6 3083 fin6 = lin6 = NULL; 3084 #endif 3085 error = 0; 3086 3087 if (req->oldptr != NULL || req->oldlen != 0) 3088 return (EINVAL); 3089 if (req->newptr == NULL) 3090 return (EPERM); 3091 if (req->newlen < sizeof(addrs)) 3092 return (ENOMEM); 3093 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3094 if (error) 3095 return (error); 3096 3097 switch (addrs[0].ss_family) { 3098 #ifdef INET6 3099 case AF_INET6: 3100 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3101 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3102 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3103 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3104 return (EINVAL); 3105 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3106 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3107 return (EINVAL); 3108 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3109 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3110 fin = (struct sockaddr_in *)&addrs[0]; 3111 lin = (struct sockaddr_in *)&addrs[1]; 3112 break; 3113 } 3114 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3115 if (error) 3116 return (error); 3117 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3118 if (error) 3119 return (error); 3120 break; 3121 #endif 3122 #ifdef INET 3123 case AF_INET: 3124 fin = (struct sockaddr_in *)&addrs[0]; 3125 lin = (struct sockaddr_in *)&addrs[1]; 3126 if (fin->sin_len != sizeof(struct sockaddr_in) || 3127 lin->sin_len != sizeof(struct sockaddr_in)) 3128 return (EINVAL); 3129 break; 3130 #endif 3131 default: 3132 return (EINVAL); 3133 } 3134 NET_EPOCH_ENTER(et); 3135 switch (addrs[0].ss_family) { 3136 #ifdef INET6 3137 case AF_INET6: 3138 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3139 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3140 INPLOOKUP_WLOCKPCB, NULL); 3141 break; 3142 #endif 3143 #ifdef INET 3144 case AF_INET: 3145 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3146 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3147 break; 3148 #endif 3149 } 3150 if (inp != NULL) { 3151 if (inp->inp_flags & INP_TIMEWAIT) { 3152 /* 3153 * XXXRW: There currently exists a state where an 3154 * inpcb is present, but its timewait state has been 3155 * discarded. For now, don't allow dropping of this 3156 * type of inpcb. 3157 */ 3158 tw = intotw(inp); 3159 if (tw != NULL) 3160 tcp_twclose(tw, 0); 3161 else 3162 INP_WUNLOCK(inp); 3163 } else if (!(inp->inp_flags & INP_DROPPED) && 3164 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3165 tp = intotcpcb(inp); 3166 tp = tcp_drop(tp, ECONNABORTED); 3167 if (tp != NULL) 3168 INP_WUNLOCK(inp); 3169 } else 3170 INP_WUNLOCK(inp); 3171 } else 3172 error = ESRCH; 3173 NET_EPOCH_EXIT(et); 3174 return (error); 3175 } 3176 3177 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3178 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3179 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3180 "Drop TCP connection"); 3181 3182 #ifdef KERN_TLS 3183 static int 3184 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3185 { 3186 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3187 struct sockaddr_storage addrs[2]; 3188 struct inpcb *inp; 3189 struct sockaddr_in *fin, *lin; 3190 struct epoch_tracker et; 3191 #ifdef INET6 3192 struct sockaddr_in6 *fin6, *lin6; 3193 #endif 3194 int error; 3195 3196 inp = NULL; 3197 fin = lin = NULL; 3198 #ifdef INET6 3199 fin6 = lin6 = NULL; 3200 #endif 3201 error = 0; 3202 3203 if (req->oldptr != NULL || req->oldlen != 0) 3204 return (EINVAL); 3205 if (req->newptr == NULL) 3206 return (EPERM); 3207 if (req->newlen < sizeof(addrs)) 3208 return (ENOMEM); 3209 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3210 if (error) 3211 return (error); 3212 3213 switch (addrs[0].ss_family) { 3214 #ifdef INET6 3215 case AF_INET6: 3216 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3217 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3218 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3219 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3220 return (EINVAL); 3221 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3222 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3223 return (EINVAL); 3224 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3225 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3226 fin = (struct sockaddr_in *)&addrs[0]; 3227 lin = (struct sockaddr_in *)&addrs[1]; 3228 break; 3229 } 3230 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3231 if (error) 3232 return (error); 3233 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3234 if (error) 3235 return (error); 3236 break; 3237 #endif 3238 #ifdef INET 3239 case AF_INET: 3240 fin = (struct sockaddr_in *)&addrs[0]; 3241 lin = (struct sockaddr_in *)&addrs[1]; 3242 if (fin->sin_len != sizeof(struct sockaddr_in) || 3243 lin->sin_len != sizeof(struct sockaddr_in)) 3244 return (EINVAL); 3245 break; 3246 #endif 3247 default: 3248 return (EINVAL); 3249 } 3250 NET_EPOCH_ENTER(et); 3251 switch (addrs[0].ss_family) { 3252 #ifdef INET6 3253 case AF_INET6: 3254 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3255 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3256 INPLOOKUP_WLOCKPCB, NULL); 3257 break; 3258 #endif 3259 #ifdef INET 3260 case AF_INET: 3261 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3262 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3263 break; 3264 #endif 3265 } 3266 NET_EPOCH_EXIT(et); 3267 if (inp != NULL) { 3268 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3269 inp->inp_socket == NULL) { 3270 error = ECONNRESET; 3271 INP_WUNLOCK(inp); 3272 } else { 3273 struct socket *so; 3274 3275 so = inp->inp_socket; 3276 soref(so); 3277 error = ktls_set_tx_mode(so, 3278 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3279 INP_WUNLOCK(inp); 3280 SOCK_LOCK(so); 3281 sorele(so); 3282 } 3283 } else 3284 error = ESRCH; 3285 return (error); 3286 } 3287 3288 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3289 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3290 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3291 "Switch TCP connection to SW TLS"); 3292 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3293 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3294 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3295 "Switch TCP connection to ifnet TLS"); 3296 #endif 3297 3298 /* 3299 * Generate a standardized TCP log line for use throughout the 3300 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3301 * allow use in the interrupt context. 3302 * 3303 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3304 * NB: The function may return NULL if memory allocation failed. 3305 * 3306 * Due to header inclusion and ordering limitations the struct ip 3307 * and ip6_hdr pointers have to be passed as void pointers. 3308 */ 3309 char * 3310 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3311 const void *ip6hdr) 3312 { 3313 3314 /* Is logging enabled? */ 3315 if (V_tcp_log_in_vain == 0) 3316 return (NULL); 3317 3318 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3319 } 3320 3321 char * 3322 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3323 const void *ip6hdr) 3324 { 3325 3326 /* Is logging enabled? */ 3327 if (tcp_log_debug == 0) 3328 return (NULL); 3329 3330 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3331 } 3332 3333 static char * 3334 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3335 const void *ip6hdr) 3336 { 3337 char *s, *sp; 3338 size_t size; 3339 struct ip *ip; 3340 #ifdef INET6 3341 const struct ip6_hdr *ip6; 3342 3343 ip6 = (const struct ip6_hdr *)ip6hdr; 3344 #endif /* INET6 */ 3345 ip = (struct ip *)ip4hdr; 3346 3347 /* 3348 * The log line looks like this: 3349 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3350 */ 3351 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3352 sizeof(PRINT_TH_FLAGS) + 1 + 3353 #ifdef INET6 3354 2 * INET6_ADDRSTRLEN; 3355 #else 3356 2 * INET_ADDRSTRLEN; 3357 #endif /* INET6 */ 3358 3359 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3360 if (s == NULL) 3361 return (NULL); 3362 3363 strcat(s, "TCP: ["); 3364 sp = s + strlen(s); 3365 3366 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3367 inet_ntoa_r(inc->inc_faddr, sp); 3368 sp = s + strlen(s); 3369 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3370 sp = s + strlen(s); 3371 inet_ntoa_r(inc->inc_laddr, sp); 3372 sp = s + strlen(s); 3373 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3374 #ifdef INET6 3375 } else if (inc) { 3376 ip6_sprintf(sp, &inc->inc6_faddr); 3377 sp = s + strlen(s); 3378 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3379 sp = s + strlen(s); 3380 ip6_sprintf(sp, &inc->inc6_laddr); 3381 sp = s + strlen(s); 3382 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3383 } else if (ip6 && th) { 3384 ip6_sprintf(sp, &ip6->ip6_src); 3385 sp = s + strlen(s); 3386 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3387 sp = s + strlen(s); 3388 ip6_sprintf(sp, &ip6->ip6_dst); 3389 sp = s + strlen(s); 3390 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3391 #endif /* INET6 */ 3392 #ifdef INET 3393 } else if (ip && th) { 3394 inet_ntoa_r(ip->ip_src, sp); 3395 sp = s + strlen(s); 3396 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3397 sp = s + strlen(s); 3398 inet_ntoa_r(ip->ip_dst, sp); 3399 sp = s + strlen(s); 3400 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3401 #endif /* INET */ 3402 } else { 3403 free(s, M_TCPLOG); 3404 return (NULL); 3405 } 3406 sp = s + strlen(s); 3407 if (th) 3408 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3409 if (*(s + size - 1) != '\0') 3410 panic("%s: string too long", __func__); 3411 return (s); 3412 } 3413 3414 /* 3415 * A subroutine which makes it easy to track TCP state changes with DTrace. 3416 * This function shouldn't be called for t_state initializations that don't 3417 * correspond to actual TCP state transitions. 3418 */ 3419 void 3420 tcp_state_change(struct tcpcb *tp, int newstate) 3421 { 3422 #if defined(KDTRACE_HOOKS) 3423 int pstate = tp->t_state; 3424 #endif 3425 3426 TCPSTATES_DEC(tp->t_state); 3427 TCPSTATES_INC(newstate); 3428 tp->t_state = newstate; 3429 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3430 } 3431 3432 /* 3433 * Create an external-format (``xtcpcb'') structure using the information in 3434 * the kernel-format tcpcb structure pointed to by tp. This is done to 3435 * reduce the spew of irrelevant information over this interface, to isolate 3436 * user code from changes in the kernel structure, and potentially to provide 3437 * information-hiding if we decide that some of this information should be 3438 * hidden from users. 3439 */ 3440 void 3441 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3442 { 3443 struct tcpcb *tp = intotcpcb(inp); 3444 sbintime_t now; 3445 3446 bzero(xt, sizeof(*xt)); 3447 if (inp->inp_flags & INP_TIMEWAIT) { 3448 xt->t_state = TCPS_TIME_WAIT; 3449 } else { 3450 xt->t_state = tp->t_state; 3451 xt->t_logstate = tp->t_logstate; 3452 xt->t_flags = tp->t_flags; 3453 xt->t_sndzerowin = tp->t_sndzerowin; 3454 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3455 xt->t_rcvoopack = tp->t_rcvoopack; 3456 xt->t_rcv_wnd = tp->rcv_wnd; 3457 xt->t_snd_wnd = tp->snd_wnd; 3458 xt->t_snd_cwnd = tp->snd_cwnd; 3459 xt->t_snd_ssthresh = tp->snd_ssthresh; 3460 xt->t_maxseg = tp->t_maxseg; 3461 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3462 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3463 3464 now = getsbinuptime(); 3465 #define COPYTIMER(ttt) do { \ 3466 if (callout_active(&tp->t_timers->ttt)) \ 3467 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3468 SBT_1MS; \ 3469 else \ 3470 xt->ttt = 0; \ 3471 } while (0) 3472 COPYTIMER(tt_delack); 3473 COPYTIMER(tt_rexmt); 3474 COPYTIMER(tt_persist); 3475 COPYTIMER(tt_keep); 3476 COPYTIMER(tt_2msl); 3477 #undef COPYTIMER 3478 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3479 3480 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3481 TCP_FUNCTION_NAME_LEN_MAX); 3482 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 3483 TCP_CA_NAME_MAX); 3484 #ifdef TCP_BLACKBOX 3485 (void)tcp_log_get_id(tp, xt->xt_logid); 3486 #endif 3487 } 3488 3489 xt->xt_len = sizeof(struct xtcpcb); 3490 in_pcbtoxinpcb(inp, &xt->xt_inp); 3491 if (inp->inp_socket == NULL) 3492 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3493 } 3494 3495 void 3496 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3497 { 3498 uint32_t bit, i; 3499 3500 if ((tp == NULL) || 3501 (status > TCP_EI_STATUS_MAX_VALUE) || 3502 (status == 0)) { 3503 /* Invalid */ 3504 return; 3505 } 3506 if (status > (sizeof(uint32_t) * 8)) { 3507 /* Should this be a KASSERT? */ 3508 return; 3509 } 3510 bit = 1U << (status - 1); 3511 if (bit & tp->t_end_info_status) { 3512 /* already logged */ 3513 return; 3514 } 3515 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3516 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3517 tp->t_end_info_bytes[i] = status; 3518 tp->t_end_info_status |= bit; 3519 break; 3520 } 3521 } 3522 } 3523