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 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/sdt.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/protosw.h> 71 #include <sys/random.h> 72 73 #include <vm/uma.h> 74 75 #include <net/route.h> 76 #include <net/route/nhop.h> 77 #include <net/if.h> 78 #include <net/if_var.h> 79 #include <net/vnet.h> 80 81 #include <netinet/in.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/in_kdtrace.h> 84 #include <netinet/in_pcb.h> 85 #include <netinet/in_systm.h> 86 #include <netinet/in_var.h> 87 #include <netinet/ip.h> 88 #include <netinet/ip_icmp.h> 89 #include <netinet/ip_var.h> 90 #ifdef INET6 91 #include <netinet/icmp6.h> 92 #include <netinet/ip6.h> 93 #include <netinet6/in6_fib.h> 94 #include <netinet6/in6_pcb.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/scope6_var.h> 97 #include <netinet6/nd6.h> 98 #endif 99 100 #include <netinet/tcp.h> 101 #ifdef INVARIANTS 102 #define TCPSTATES 103 #endif 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 #include <netinet/tcpip.h> 113 #include <netinet/tcp_fastopen.h> 114 #ifdef TCPPCAP 115 #include <netinet/tcp_pcap.h> 116 #endif 117 #ifdef TCPDEBUG 118 #include <netinet/tcp_debug.h> 119 #endif 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 #include <netinet/udp.h> 124 #include <netinet/udp_var.h> 125 #ifdef INET6 126 #include <netinet6/tcp6_var.h> 127 #endif 128 129 #include <netipsec/ipsec_support.h> 130 131 #include <machine/in_cksum.h> 132 #include <crypto/siphash/siphash.h> 133 134 #include <security/mac/mac_framework.h> 135 136 #ifdef INET6 137 static ip6proto_ctlinput_t tcp6_ctlinput; 138 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 139 #endif 140 141 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 142 #ifdef INET6 143 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 144 #endif 145 146 #ifdef NETFLIX_EXP_DETECTION 147 /* Sack attack detection thresholds and such */ 148 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 149 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 150 "Sack Attack detection thresholds"); 151 int32_t tcp_force_detection = 0; 152 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 153 CTLFLAG_RW, 154 &tcp_force_detection, 0, 155 "Do we force detection even if the INP has it off?"); 156 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 157 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 158 CTLFLAG_RW, 159 &tcp_sack_to_ack_thresh, 700, 160 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 161 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 162 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 163 CTLFLAG_RW, 164 &tcp_sack_to_move_thresh, 600, 165 "Percentage of sack moves we must see above (10.1 percent is 101)"); 166 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 167 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 168 CTLFLAG_RW, 169 &tcp_restoral_thresh, 550, 170 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 171 int32_t tcp_sad_decay_val = 800; 172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 173 CTLFLAG_RW, 174 &tcp_sad_decay_val, 800, 175 "The decay percentage (10.1 percent equals 101 )"); 176 int32_t tcp_map_minimum = 500; 177 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 178 CTLFLAG_RW, 179 &tcp_map_minimum, 500, 180 "Number of Map enteries before we start detection"); 181 int32_t tcp_attack_on_turns_on_logging = 0; 182 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 183 CTLFLAG_RW, 184 &tcp_attack_on_turns_on_logging, 0, 185 "When we have a positive hit on attack, do we turn on logging?"); 186 int32_t tcp_sad_pacing_interval = 2000; 187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 188 CTLFLAG_RW, 189 &tcp_sad_pacing_interval, 2000, 190 "What is the minimum pacing interval for a classified attacker?"); 191 192 int32_t tcp_sad_low_pps = 100; 193 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 194 CTLFLAG_RW, 195 &tcp_sad_low_pps, 100, 196 "What is the input pps that below which we do not decay?"); 197 #endif 198 uint32_t tcp_ack_war_time_window = 1000; 199 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 200 CTLFLAG_RW, 201 &tcp_ack_war_time_window, 1000, 202 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 203 uint32_t tcp_ack_war_cnt = 5; 204 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 205 CTLFLAG_RW, 206 &tcp_ack_war_cnt, 5, 207 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 208 209 struct rwlock tcp_function_lock; 210 211 static int 212 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 213 { 214 int error, new; 215 216 new = V_tcp_mssdflt; 217 error = sysctl_handle_int(oidp, &new, 0, req); 218 if (error == 0 && req->newptr) { 219 if (new < TCP_MINMSS) 220 error = EINVAL; 221 else 222 V_tcp_mssdflt = new; 223 } 224 return (error); 225 } 226 227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 228 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 229 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 230 "Default TCP Maximum Segment Size"); 231 232 #ifdef INET6 233 static int 234 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 235 { 236 int error, new; 237 238 new = V_tcp_v6mssdflt; 239 error = sysctl_handle_int(oidp, &new, 0, req); 240 if (error == 0 && req->newptr) { 241 if (new < TCP_MINMSS) 242 error = EINVAL; 243 else 244 V_tcp_v6mssdflt = new; 245 } 246 return (error); 247 } 248 249 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 250 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 251 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 252 "Default TCP Maximum Segment Size for IPv6"); 253 #endif /* INET6 */ 254 255 /* 256 * Minimum MSS we accept and use. This prevents DoS attacks where 257 * we are forced to a ridiculous low MSS like 20 and send hundreds 258 * of packets instead of one. The effect scales with the available 259 * bandwidth and quickly saturates the CPU and network interface 260 * with packet generation and sending. Set to zero to disable MINMSS 261 * checking. This setting prevents us from sending too small packets. 262 */ 263 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 265 &VNET_NAME(tcp_minmss), 0, 266 "Minimum TCP Maximum Segment Size"); 267 268 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 269 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 270 &VNET_NAME(tcp_do_rfc1323), 0, 271 "Enable rfc1323 (high performance TCP) extensions"); 272 273 /* 274 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 275 * Some stacks negotiate TS, but never send them after connection setup. Some 276 * stacks negotiate TS, but don't send them when sending keep-alive segments. 277 * These include modern widely deployed TCP stacks. 278 * Therefore tolerating violations for now... 279 */ 280 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 282 &VNET_NAME(tcp_tolerate_missing_ts), 0, 283 "Tolerate missing TCP timestamps"); 284 285 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 287 &VNET_NAME(tcp_ts_offset_per_conn), 0, 288 "Initialize TCP timestamps per connection instead of per host pair"); 289 290 /* How many connections are pacing */ 291 static volatile uint32_t number_of_tcp_connections_pacing = 0; 292 static uint32_t shadow_num_connections = 0; 293 294 static int tcp_pacing_limit = 10000; 295 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 296 &tcp_pacing_limit, 1000, 297 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 298 299 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 300 &shadow_num_connections, 0, "Number of TCP connections being paced"); 301 302 static int tcp_log_debug = 0; 303 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 304 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 305 306 static int tcp_tcbhashsize; 307 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 308 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 309 310 static int do_tcpdrain = 1; 311 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 312 "Enable tcp_drain routine for extra help when low on mbufs"); 313 314 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 315 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 316 317 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 318 #define V_icmp_may_rst VNET(icmp_may_rst) 319 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 320 &VNET_NAME(icmp_may_rst), 0, 321 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 322 323 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 324 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 325 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 326 &VNET_NAME(tcp_isn_reseed_interval), 0, 327 "Seconds between reseeding of ISN secret"); 328 329 static int tcp_soreceive_stream; 330 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 331 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 332 333 VNET_DEFINE(uma_zone_t, sack_hole_zone); 334 #define V_sack_hole_zone VNET(sack_hole_zone) 335 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 336 static int 337 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 338 { 339 int error; 340 uint32_t new; 341 342 new = V_tcp_map_entries_limit; 343 error = sysctl_handle_int(oidp, &new, 0, req); 344 if (error == 0 && req->newptr) { 345 /* only allow "0" and value > minimum */ 346 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 347 error = EINVAL; 348 else 349 V_tcp_map_entries_limit = new; 350 } 351 return (error); 352 } 353 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 354 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 355 &VNET_NAME(tcp_map_entries_limit), 0, 356 &sysctl_net_inet_tcp_map_limit_check, "IU", 357 "Total sendmap entries limit"); 358 359 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 360 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 361 &VNET_NAME(tcp_map_split_limit), 0, 362 "Total sendmap split entries limit"); 363 364 #ifdef TCP_HHOOK 365 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 366 #endif 367 368 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 369 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 370 #define V_ts_offset_secret VNET(ts_offset_secret) 371 372 static int tcp_default_fb_init(struct tcpcb *tp); 373 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 374 static int tcp_default_handoff_ok(struct tcpcb *tp); 375 static struct inpcb *tcp_notify(struct inpcb *, int); 376 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 377 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 378 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 379 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 380 const void *ip4hdr, const void *ip6hdr); 381 static ipproto_ctlinput_t tcp_ctlinput; 382 static udp_tun_icmp_t tcp_ctlinput_viaudp; 383 384 static struct tcp_function_block tcp_def_funcblk = { 385 .tfb_tcp_block_name = "freebsd", 386 .tfb_tcp_output = tcp_default_output, 387 .tfb_tcp_do_segment = tcp_do_segment, 388 .tfb_tcp_ctloutput = tcp_default_ctloutput, 389 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 390 .tfb_tcp_fb_init = tcp_default_fb_init, 391 .tfb_tcp_fb_fini = tcp_default_fb_fini, 392 }; 393 394 static int tcp_fb_cnt = 0; 395 struct tcp_funchead t_functions; 396 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 397 398 void 399 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 400 { 401 TCPSTAT_INC(tcps_dsack_count); 402 tp->t_dsack_pack++; 403 if (tlp == 0) { 404 if (SEQ_GT(end, start)) { 405 tp->t_dsack_bytes += (end - start); 406 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 407 } else { 408 tp->t_dsack_tlp_bytes += (start - end); 409 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 410 } 411 } else { 412 if (SEQ_GT(end, start)) { 413 tp->t_dsack_bytes += (end - start); 414 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 415 } else { 416 tp->t_dsack_tlp_bytes += (start - end); 417 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 418 } 419 } 420 } 421 422 static struct tcp_function_block * 423 find_tcp_functions_locked(struct tcp_function_set *fs) 424 { 425 struct tcp_function *f; 426 struct tcp_function_block *blk=NULL; 427 428 TAILQ_FOREACH(f, &t_functions, tf_next) { 429 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 430 blk = f->tf_fb; 431 break; 432 } 433 } 434 return(blk); 435 } 436 437 static struct tcp_function_block * 438 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 439 { 440 struct tcp_function_block *rblk=NULL; 441 struct tcp_function *f; 442 443 TAILQ_FOREACH(f, &t_functions, tf_next) { 444 if (f->tf_fb == blk) { 445 rblk = blk; 446 if (s) { 447 *s = f; 448 } 449 break; 450 } 451 } 452 return (rblk); 453 } 454 455 struct tcp_function_block * 456 find_and_ref_tcp_functions(struct tcp_function_set *fs) 457 { 458 struct tcp_function_block *blk; 459 460 rw_rlock(&tcp_function_lock); 461 blk = find_tcp_functions_locked(fs); 462 if (blk) 463 refcount_acquire(&blk->tfb_refcnt); 464 rw_runlock(&tcp_function_lock); 465 return(blk); 466 } 467 468 struct tcp_function_block * 469 find_and_ref_tcp_fb(struct tcp_function_block *blk) 470 { 471 struct tcp_function_block *rblk; 472 473 rw_rlock(&tcp_function_lock); 474 rblk = find_tcp_fb_locked(blk, NULL); 475 if (rblk) 476 refcount_acquire(&rblk->tfb_refcnt); 477 rw_runlock(&tcp_function_lock); 478 return(rblk); 479 } 480 481 /* Find a matching alias for the given tcp_function_block. */ 482 int 483 find_tcp_function_alias(struct tcp_function_block *blk, 484 struct tcp_function_set *fs) 485 { 486 struct tcp_function *f; 487 int found; 488 489 found = 0; 490 rw_rlock(&tcp_function_lock); 491 TAILQ_FOREACH(f, &t_functions, tf_next) { 492 if ((f->tf_fb == blk) && 493 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 494 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 495 /* Matching function block with different name. */ 496 strncpy(fs->function_set_name, f->tf_name, 497 TCP_FUNCTION_NAME_LEN_MAX); 498 found = 1; 499 break; 500 } 501 } 502 /* Null terminate the string appropriately. */ 503 if (found) { 504 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 505 } else { 506 fs->function_set_name[0] = '\0'; 507 } 508 rw_runlock(&tcp_function_lock); 509 return (found); 510 } 511 512 static struct tcp_function_block * 513 find_and_ref_tcp_default_fb(void) 514 { 515 struct tcp_function_block *rblk; 516 517 rw_rlock(&tcp_function_lock); 518 rblk = tcp_func_set_ptr; 519 refcount_acquire(&rblk->tfb_refcnt); 520 rw_runlock(&tcp_function_lock); 521 return (rblk); 522 } 523 524 void 525 tcp_switch_back_to_default(struct tcpcb *tp) 526 { 527 struct tcp_function_block *tfb; 528 529 KASSERT(tp->t_fb != &tcp_def_funcblk, 530 ("%s: called by the built-in default stack", __func__)); 531 532 /* 533 * Release the old stack. This function will either find a new one 534 * or panic. 535 */ 536 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 537 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 538 refcount_release(&tp->t_fb->tfb_refcnt); 539 540 /* 541 * Now, we'll find a new function block to use. 542 * Start by trying the current user-selected 543 * default, unless this stack is the user-selected 544 * default. 545 */ 546 tfb = find_and_ref_tcp_default_fb(); 547 if (tfb == tp->t_fb) { 548 refcount_release(&tfb->tfb_refcnt); 549 tfb = NULL; 550 } 551 /* Does the stack accept this connection? */ 552 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 553 (*tfb->tfb_tcp_handoff_ok)(tp)) { 554 refcount_release(&tfb->tfb_refcnt); 555 tfb = NULL; 556 } 557 /* Try to use that stack. */ 558 if (tfb != NULL) { 559 /* Initialize the new stack. If it succeeds, we are done. */ 560 tp->t_fb = tfb; 561 if (tp->t_fb->tfb_tcp_fb_init == NULL || 562 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 563 return; 564 565 /* 566 * Initialization failed. Release the reference count on 567 * the stack. 568 */ 569 refcount_release(&tfb->tfb_refcnt); 570 } 571 572 /* 573 * If that wasn't feasible, use the built-in default 574 * stack which is not allowed to reject anyone. 575 */ 576 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 577 if (tfb == NULL) { 578 /* there always should be a default */ 579 panic("Can't refer to tcp_def_funcblk"); 580 } 581 if (tfb->tfb_tcp_handoff_ok != NULL) { 582 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 583 /* The default stack cannot say no */ 584 panic("Default stack rejects a new session?"); 585 } 586 } 587 tp->t_fb = tfb; 588 if (tp->t_fb->tfb_tcp_fb_init != NULL && 589 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 590 /* The default stack cannot fail */ 591 panic("Default stack initialization failed"); 592 } 593 } 594 595 static bool 596 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 597 const struct sockaddr *sa, void *ctx) 598 { 599 struct ip *iph; 600 #ifdef INET6 601 struct ip6_hdr *ip6; 602 #endif 603 struct udphdr *uh; 604 struct tcphdr *th; 605 int thlen; 606 uint16_t port; 607 608 TCPSTAT_INC(tcps_tunneled_pkts); 609 if ((m->m_flags & M_PKTHDR) == 0) { 610 /* Can't handle one that is not a pkt hdr */ 611 TCPSTAT_INC(tcps_tunneled_errs); 612 goto out; 613 } 614 thlen = sizeof(struct tcphdr); 615 if (m->m_len < off + sizeof(struct udphdr) + thlen && 616 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 617 TCPSTAT_INC(tcps_tunneled_errs); 618 goto out; 619 } 620 iph = mtod(m, struct ip *); 621 uh = (struct udphdr *)((caddr_t)iph + off); 622 th = (struct tcphdr *)(uh + 1); 623 thlen = th->th_off << 2; 624 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 625 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 626 if (m == NULL) { 627 TCPSTAT_INC(tcps_tunneled_errs); 628 goto out; 629 } else { 630 iph = mtod(m, struct ip *); 631 uh = (struct udphdr *)((caddr_t)iph + off); 632 th = (struct tcphdr *)(uh + 1); 633 } 634 } 635 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 636 bcopy(th, uh, m->m_len - off); 637 m->m_len -= sizeof(struct udphdr); 638 m->m_pkthdr.len -= sizeof(struct udphdr); 639 /* 640 * We use the same algorithm for 641 * both UDP and TCP for c-sum. So 642 * the code in tcp_input will skip 643 * the checksum. So we do nothing 644 * with the flag (m->m_pkthdr.csum_flags). 645 */ 646 switch (iph->ip_v) { 647 #ifdef INET 648 case IPVERSION: 649 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 650 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 651 break; 652 #endif 653 #ifdef INET6 654 case IPV6_VERSION >> 4: 655 ip6 = mtod(m, struct ip6_hdr *); 656 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 657 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 658 break; 659 #endif 660 default: 661 goto out; 662 break; 663 } 664 return (true); 665 out: 666 m_freem(m); 667 668 return (true); 669 } 670 671 static int 672 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 673 { 674 int error=ENOENT; 675 struct tcp_function_set fs; 676 struct tcp_function_block *blk; 677 678 memset(&fs, 0, sizeof(fs)); 679 rw_rlock(&tcp_function_lock); 680 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 681 if (blk) { 682 /* Found him */ 683 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 684 fs.pcbcnt = blk->tfb_refcnt; 685 } 686 rw_runlock(&tcp_function_lock); 687 error = sysctl_handle_string(oidp, fs.function_set_name, 688 sizeof(fs.function_set_name), req); 689 690 /* Check for error or no change */ 691 if (error != 0 || req->newptr == NULL) 692 return(error); 693 694 rw_wlock(&tcp_function_lock); 695 blk = find_tcp_functions_locked(&fs); 696 if ((blk == NULL) || 697 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 698 error = ENOENT; 699 goto done; 700 } 701 tcp_func_set_ptr = blk; 702 done: 703 rw_wunlock(&tcp_function_lock); 704 return (error); 705 } 706 707 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 708 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 709 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 710 "Set/get the default TCP functions"); 711 712 static int 713 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 714 { 715 int error, cnt, linesz; 716 struct tcp_function *f; 717 char *buffer, *cp; 718 size_t bufsz, outsz; 719 bool alias; 720 721 cnt = 0; 722 rw_rlock(&tcp_function_lock); 723 TAILQ_FOREACH(f, &t_functions, tf_next) { 724 cnt++; 725 } 726 rw_runlock(&tcp_function_lock); 727 728 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 729 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 730 731 error = 0; 732 cp = buffer; 733 734 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 735 "Alias", "PCB count"); 736 cp += linesz; 737 bufsz -= linesz; 738 outsz = linesz; 739 740 rw_rlock(&tcp_function_lock); 741 TAILQ_FOREACH(f, &t_functions, tf_next) { 742 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 743 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 744 f->tf_fb->tfb_tcp_block_name, 745 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 746 alias ? f->tf_name : "-", 747 f->tf_fb->tfb_refcnt); 748 if (linesz >= bufsz) { 749 error = EOVERFLOW; 750 break; 751 } 752 cp += linesz; 753 bufsz -= linesz; 754 outsz += linesz; 755 } 756 rw_runlock(&tcp_function_lock); 757 if (error == 0) 758 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 759 free(buffer, M_TEMP); 760 return (error); 761 } 762 763 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 764 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 765 NULL, 0, sysctl_net_inet_list_available, "A", 766 "list available TCP Function sets"); 767 768 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 769 770 #ifdef INET 771 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 772 #define V_udp4_tun_socket VNET(udp4_tun_socket) 773 #endif 774 #ifdef INET6 775 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 776 #define V_udp6_tun_socket VNET(udp6_tun_socket) 777 #endif 778 779 static void 780 tcp_over_udp_stop(void) 781 { 782 /* 783 * This function assumes sysctl caller holds inp_rinfo_lock() 784 * for writing! 785 */ 786 #ifdef INET 787 if (V_udp4_tun_socket != NULL) { 788 soclose(V_udp4_tun_socket); 789 V_udp4_tun_socket = NULL; 790 } 791 #endif 792 #ifdef INET6 793 if (V_udp6_tun_socket != NULL) { 794 soclose(V_udp6_tun_socket); 795 V_udp6_tun_socket = NULL; 796 } 797 #endif 798 } 799 800 static int 801 tcp_over_udp_start(void) 802 { 803 uint16_t port; 804 int ret; 805 #ifdef INET 806 struct sockaddr_in sin; 807 #endif 808 #ifdef INET6 809 struct sockaddr_in6 sin6; 810 #endif 811 /* 812 * This function assumes sysctl caller holds inp_info_rlock() 813 * for writing! 814 */ 815 port = V_tcp_udp_tunneling_port; 816 if (ntohs(port) == 0) { 817 /* Must have a port set */ 818 return (EINVAL); 819 } 820 #ifdef INET 821 if (V_udp4_tun_socket != NULL) { 822 /* Already running -- must stop first */ 823 return (EALREADY); 824 } 825 #endif 826 #ifdef INET6 827 if (V_udp6_tun_socket != NULL) { 828 /* Already running -- must stop first */ 829 return (EALREADY); 830 } 831 #endif 832 #ifdef INET 833 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 834 SOCK_DGRAM, IPPROTO_UDP, 835 curthread->td_ucred, curthread))) { 836 tcp_over_udp_stop(); 837 return (ret); 838 } 839 /* Call the special UDP hook. */ 840 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 841 tcp_recv_udp_tunneled_packet, 842 tcp_ctlinput_viaudp, 843 NULL))) { 844 tcp_over_udp_stop(); 845 return (ret); 846 } 847 /* Ok, we have a socket, bind it to the port. */ 848 memset(&sin, 0, sizeof(struct sockaddr_in)); 849 sin.sin_len = sizeof(struct sockaddr_in); 850 sin.sin_family = AF_INET; 851 sin.sin_port = htons(port); 852 if ((ret = sobind(V_udp4_tun_socket, 853 (struct sockaddr *)&sin, curthread))) { 854 tcp_over_udp_stop(); 855 return (ret); 856 } 857 #endif 858 #ifdef INET6 859 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 860 SOCK_DGRAM, IPPROTO_UDP, 861 curthread->td_ucred, curthread))) { 862 tcp_over_udp_stop(); 863 return (ret); 864 } 865 /* Call the special UDP hook. */ 866 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 867 tcp_recv_udp_tunneled_packet, 868 tcp6_ctlinput_viaudp, 869 NULL))) { 870 tcp_over_udp_stop(); 871 return (ret); 872 } 873 /* Ok, we have a socket, bind it to the port. */ 874 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 875 sin6.sin6_len = sizeof(struct sockaddr_in6); 876 sin6.sin6_family = AF_INET6; 877 sin6.sin6_port = htons(port); 878 if ((ret = sobind(V_udp6_tun_socket, 879 (struct sockaddr *)&sin6, curthread))) { 880 tcp_over_udp_stop(); 881 return (ret); 882 } 883 #endif 884 return (0); 885 } 886 887 static int 888 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 889 { 890 int error; 891 uint32_t old, new; 892 893 old = V_tcp_udp_tunneling_port; 894 new = old; 895 error = sysctl_handle_int(oidp, &new, 0, req); 896 if ((error == 0) && 897 (req->newptr != NULL)) { 898 if ((new < TCP_TUNNELING_PORT_MIN) || 899 (new > TCP_TUNNELING_PORT_MAX)) { 900 error = EINVAL; 901 } else { 902 V_tcp_udp_tunneling_port = new; 903 if (old != 0) { 904 tcp_over_udp_stop(); 905 } 906 if (new != 0) { 907 error = tcp_over_udp_start(); 908 } 909 } 910 } 911 return (error); 912 } 913 914 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 915 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 916 &VNET_NAME(tcp_udp_tunneling_port), 917 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 918 "Tunneling port for tcp over udp"); 919 920 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 921 922 static int 923 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 924 { 925 int error, new; 926 927 new = V_tcp_udp_tunneling_overhead; 928 error = sysctl_handle_int(oidp, &new, 0, req); 929 if (error == 0 && req->newptr) { 930 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 931 (new > TCP_TUNNELING_OVERHEAD_MAX)) 932 error = EINVAL; 933 else 934 V_tcp_udp_tunneling_overhead = new; 935 } 936 return (error); 937 } 938 939 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 940 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 941 &VNET_NAME(tcp_udp_tunneling_overhead), 942 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 943 "MSS reduction when using tcp over udp"); 944 945 /* 946 * Exports one (struct tcp_function_info) for each alias/name. 947 */ 948 static int 949 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 950 { 951 int cnt, error; 952 struct tcp_function *f; 953 struct tcp_function_info tfi; 954 955 /* 956 * We don't allow writes. 957 */ 958 if (req->newptr != NULL) 959 return (EINVAL); 960 961 /* 962 * Wire the old buffer so we can directly copy the functions to 963 * user space without dropping the lock. 964 */ 965 if (req->oldptr != NULL) { 966 error = sysctl_wire_old_buffer(req, 0); 967 if (error) 968 return (error); 969 } 970 971 /* 972 * Walk the list and copy out matching entries. If INVARIANTS 973 * is compiled in, also walk the list to verify the length of 974 * the list matches what we have recorded. 975 */ 976 rw_rlock(&tcp_function_lock); 977 978 cnt = 0; 979 #ifndef INVARIANTS 980 if (req->oldptr == NULL) { 981 cnt = tcp_fb_cnt; 982 goto skip_loop; 983 } 984 #endif 985 TAILQ_FOREACH(f, &t_functions, tf_next) { 986 #ifdef INVARIANTS 987 cnt++; 988 #endif 989 if (req->oldptr != NULL) { 990 bzero(&tfi, sizeof(tfi)); 991 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 992 tfi.tfi_id = f->tf_fb->tfb_id; 993 (void)strlcpy(tfi.tfi_alias, f->tf_name, 994 sizeof(tfi.tfi_alias)); 995 (void)strlcpy(tfi.tfi_name, 996 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 997 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 998 /* 999 * Don't stop on error, as that is the 1000 * mechanism we use to accumulate length 1001 * information if the buffer was too short. 1002 */ 1003 } 1004 } 1005 KASSERT(cnt == tcp_fb_cnt, 1006 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1007 #ifndef INVARIANTS 1008 skip_loop: 1009 #endif 1010 rw_runlock(&tcp_function_lock); 1011 if (req->oldptr == NULL) 1012 error = SYSCTL_OUT(req, NULL, 1013 (cnt + 1) * sizeof(struct tcp_function_info)); 1014 1015 return (error); 1016 } 1017 1018 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1019 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1020 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1021 "List TCP function block name-to-ID mappings"); 1022 1023 /* 1024 * tfb_tcp_handoff_ok() function for the default stack. 1025 * Note that we'll basically try to take all comers. 1026 */ 1027 static int 1028 tcp_default_handoff_ok(struct tcpcb *tp) 1029 { 1030 1031 return (0); 1032 } 1033 1034 /* 1035 * tfb_tcp_fb_init() function for the default stack. 1036 * 1037 * This handles making sure we have appropriate timers set if you are 1038 * transitioning a socket that has some amount of setup done. 1039 * 1040 * The init() fuction from the default can *never* return non-zero i.e. 1041 * it is required to always succeed since it is the stack of last resort! 1042 */ 1043 static int 1044 tcp_default_fb_init(struct tcpcb *tp) 1045 { 1046 1047 struct socket *so; 1048 1049 INP_WLOCK_ASSERT(tp->t_inpcb); 1050 1051 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 1052 ("%s: connection %p in unexpected state %d", __func__, tp, 1053 tp->t_state)); 1054 1055 /* 1056 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1057 * know what to do for unexpected states (which includes TIME_WAIT). 1058 */ 1059 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1060 return (0); 1061 1062 /* 1063 * Make sure some kind of transmission timer is set if there is 1064 * outstanding data. 1065 */ 1066 so = tp->t_inpcb->inp_socket; 1067 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1068 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1069 tcp_timer_active(tp, TT_PERSIST))) { 1070 /* 1071 * If the session has established and it looks like it should 1072 * be in the persist state, set the persist timer. Otherwise, 1073 * set the retransmit timer. 1074 */ 1075 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1076 (int32_t)(tp->snd_nxt - tp->snd_una) < 1077 (int32_t)sbavail(&so->so_snd)) 1078 tcp_setpersist(tp); 1079 else 1080 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1081 } 1082 1083 /* All non-embryonic sessions get a keepalive timer. */ 1084 if (!tcp_timer_active(tp, TT_KEEP)) 1085 tcp_timer_activate(tp, TT_KEEP, 1086 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1087 TP_KEEPINIT(tp)); 1088 1089 /* 1090 * Make sure critical variables are initialized 1091 * if transitioning while in Recovery. 1092 */ 1093 if IN_FASTRECOVERY(tp->t_flags) { 1094 if (tp->sackhint.recover_fs == 0) 1095 tp->sackhint.recover_fs = max(1, 1096 tp->snd_nxt - tp->snd_una); 1097 } 1098 1099 return (0); 1100 } 1101 1102 /* 1103 * tfb_tcp_fb_fini() function for the default stack. 1104 * 1105 * This changes state as necessary (or prudent) to prepare for another stack 1106 * to assume responsibility for the connection. 1107 */ 1108 static void 1109 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1110 { 1111 1112 INP_WLOCK_ASSERT(tp->t_inpcb); 1113 return; 1114 } 1115 1116 /* 1117 * Target size of TCP PCB hash tables. Must be a power of two. 1118 * 1119 * Note that this can be overridden by the kernel environment 1120 * variable net.inet.tcp.tcbhashsize 1121 */ 1122 #ifndef TCBHASHSIZE 1123 #define TCBHASHSIZE 0 1124 #endif 1125 1126 /* 1127 * XXX 1128 * Callouts should be moved into struct tcp directly. They are currently 1129 * separate because the tcpcb structure is exported to userland for sysctl 1130 * parsing purposes, which do not know about callouts. 1131 */ 1132 struct tcpcb_mem { 1133 struct tcpcb tcb; 1134 struct tcp_timer tt; 1135 struct cc_var ccv; 1136 #ifdef TCP_HHOOK 1137 struct osd osd; 1138 #endif 1139 }; 1140 1141 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 1142 #define V_tcpcb_zone VNET(tcpcb_zone) 1143 1144 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1145 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1146 1147 static struct mtx isn_mtx; 1148 1149 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1150 #define ISN_LOCK() mtx_lock(&isn_mtx) 1151 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1152 1153 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1154 1155 /* 1156 * Take a value and get the next power of 2 that doesn't overflow. 1157 * Used to size the tcp_inpcb hash buckets. 1158 */ 1159 static int 1160 maketcp_hashsize(int size) 1161 { 1162 int hashsize; 1163 1164 /* 1165 * auto tune. 1166 * get the next power of 2 higher than maxsockets. 1167 */ 1168 hashsize = 1 << fls(size); 1169 /* catch overflow, and just go one power of 2 smaller */ 1170 if (hashsize < size) { 1171 hashsize = 1 << (fls(size) - 1); 1172 } 1173 return (hashsize); 1174 } 1175 1176 static volatile int next_tcp_stack_id = 1; 1177 1178 /* 1179 * Register a TCP function block with the name provided in the names 1180 * array. (Note that this function does NOT automatically register 1181 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1182 * explicitly include blk->tfb_tcp_block_name in the list of names if 1183 * you wish to register the stack with that name.) 1184 * 1185 * Either all name registrations will succeed or all will fail. If 1186 * a name registration fails, the function will update the num_names 1187 * argument to point to the array index of the name that encountered 1188 * the failure. 1189 * 1190 * Returns 0 on success, or an error code on failure. 1191 */ 1192 int 1193 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1194 const char *names[], int *num_names) 1195 { 1196 struct tcp_function *n; 1197 struct tcp_function_set fs; 1198 int error, i; 1199 1200 KASSERT(names != NULL && *num_names > 0, 1201 ("%s: Called with 0-length name list", __func__)); 1202 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1203 KASSERT(rw_initialized(&tcp_function_lock), 1204 ("%s: called too early", __func__)); 1205 1206 if ((blk->tfb_tcp_output == NULL) || 1207 (blk->tfb_tcp_do_segment == NULL) || 1208 (blk->tfb_tcp_ctloutput == NULL) || 1209 (strlen(blk->tfb_tcp_block_name) == 0)) { 1210 /* 1211 * These functions are required and you 1212 * need a name. 1213 */ 1214 *num_names = 0; 1215 return (EINVAL); 1216 } 1217 if (blk->tfb_tcp_timer_stop_all || 1218 blk->tfb_tcp_timer_activate || 1219 blk->tfb_tcp_timer_active || 1220 blk->tfb_tcp_timer_stop) { 1221 /* 1222 * If you define one timer function you 1223 * must have them all. 1224 */ 1225 if ((blk->tfb_tcp_timer_stop_all == NULL) || 1226 (blk->tfb_tcp_timer_activate == NULL) || 1227 (blk->tfb_tcp_timer_active == NULL) || 1228 (blk->tfb_tcp_timer_stop == NULL)) { 1229 *num_names = 0; 1230 return (EINVAL); 1231 } 1232 } 1233 1234 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1235 *num_names = 0; 1236 return (EINVAL); 1237 } 1238 1239 refcount_init(&blk->tfb_refcnt, 0); 1240 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1241 for (i = 0; i < *num_names; i++) { 1242 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1243 if (n == NULL) { 1244 error = ENOMEM; 1245 goto cleanup; 1246 } 1247 n->tf_fb = blk; 1248 1249 (void)strlcpy(fs.function_set_name, names[i], 1250 sizeof(fs.function_set_name)); 1251 rw_wlock(&tcp_function_lock); 1252 if (find_tcp_functions_locked(&fs) != NULL) { 1253 /* Duplicate name space not allowed */ 1254 rw_wunlock(&tcp_function_lock); 1255 free(n, M_TCPFUNCTIONS); 1256 error = EALREADY; 1257 goto cleanup; 1258 } 1259 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1260 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1261 tcp_fb_cnt++; 1262 rw_wunlock(&tcp_function_lock); 1263 } 1264 return(0); 1265 1266 cleanup: 1267 /* 1268 * Deregister the names we just added. Because registration failed 1269 * for names[i], we don't need to deregister that name. 1270 */ 1271 *num_names = i; 1272 rw_wlock(&tcp_function_lock); 1273 while (--i >= 0) { 1274 TAILQ_FOREACH(n, &t_functions, tf_next) { 1275 if (!strncmp(n->tf_name, names[i], 1276 TCP_FUNCTION_NAME_LEN_MAX)) { 1277 TAILQ_REMOVE(&t_functions, n, tf_next); 1278 tcp_fb_cnt--; 1279 n->tf_fb = NULL; 1280 free(n, M_TCPFUNCTIONS); 1281 break; 1282 } 1283 } 1284 } 1285 rw_wunlock(&tcp_function_lock); 1286 return (error); 1287 } 1288 1289 /* 1290 * Register a TCP function block using the name provided in the name 1291 * argument. 1292 * 1293 * Returns 0 on success, or an error code on failure. 1294 */ 1295 int 1296 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1297 int wait) 1298 { 1299 const char *name_list[1]; 1300 int num_names, rv; 1301 1302 num_names = 1; 1303 if (name != NULL) 1304 name_list[0] = name; 1305 else 1306 name_list[0] = blk->tfb_tcp_block_name; 1307 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1308 return (rv); 1309 } 1310 1311 /* 1312 * Register a TCP function block using the name defined in 1313 * blk->tfb_tcp_block_name. 1314 * 1315 * Returns 0 on success, or an error code on failure. 1316 */ 1317 int 1318 register_tcp_functions(struct tcp_function_block *blk, int wait) 1319 { 1320 1321 return (register_tcp_functions_as_name(blk, NULL, wait)); 1322 } 1323 1324 /* 1325 * Deregister all names associated with a function block. This 1326 * functionally removes the function block from use within the system. 1327 * 1328 * When called with a true quiesce argument, mark the function block 1329 * as being removed so no more stacks will use it and determine 1330 * whether the removal would succeed. 1331 * 1332 * When called with a false quiesce argument, actually attempt the 1333 * removal. 1334 * 1335 * When called with a force argument, attempt to switch all TCBs to 1336 * use the default stack instead of returning EBUSY. 1337 * 1338 * Returns 0 on success (or if the removal would succeed, or an error 1339 * code on failure. 1340 */ 1341 int 1342 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1343 bool force) 1344 { 1345 struct tcp_function *f; 1346 1347 if (blk == &tcp_def_funcblk) { 1348 /* You can't un-register the default */ 1349 return (EPERM); 1350 } 1351 rw_wlock(&tcp_function_lock); 1352 if (blk == tcp_func_set_ptr) { 1353 /* You can't free the current default */ 1354 rw_wunlock(&tcp_function_lock); 1355 return (EBUSY); 1356 } 1357 /* Mark the block so no more stacks can use it. */ 1358 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1359 /* 1360 * If TCBs are still attached to the stack, attempt to switch them 1361 * to the default stack. 1362 */ 1363 if (force && blk->tfb_refcnt) { 1364 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1365 INPLOOKUP_WLOCKPCB); 1366 struct inpcb *inp; 1367 struct tcpcb *tp; 1368 VNET_ITERATOR_DECL(vnet_iter); 1369 1370 rw_wunlock(&tcp_function_lock); 1371 1372 VNET_LIST_RLOCK(); 1373 VNET_FOREACH(vnet_iter) { 1374 CURVNET_SET(vnet_iter); 1375 while ((inp = inp_next(&inpi)) != NULL) { 1376 if (inp->inp_flags & INP_TIMEWAIT) 1377 continue; 1378 tp = intotcpcb(inp); 1379 if (tp == NULL || tp->t_fb != blk) 1380 continue; 1381 tcp_switch_back_to_default(tp); 1382 } 1383 CURVNET_RESTORE(); 1384 } 1385 VNET_LIST_RUNLOCK(); 1386 1387 rw_wlock(&tcp_function_lock); 1388 } 1389 if (blk->tfb_refcnt) { 1390 /* TCBs still attached. */ 1391 rw_wunlock(&tcp_function_lock); 1392 return (EBUSY); 1393 } 1394 if (quiesce) { 1395 /* Skip removal. */ 1396 rw_wunlock(&tcp_function_lock); 1397 return (0); 1398 } 1399 /* Remove any function names that map to this function block. */ 1400 while (find_tcp_fb_locked(blk, &f) != NULL) { 1401 TAILQ_REMOVE(&t_functions, f, tf_next); 1402 tcp_fb_cnt--; 1403 f->tf_fb = NULL; 1404 free(f, M_TCPFUNCTIONS); 1405 } 1406 rw_wunlock(&tcp_function_lock); 1407 return (0); 1408 } 1409 1410 static void 1411 tcp_drain(void) 1412 { 1413 struct epoch_tracker et; 1414 VNET_ITERATOR_DECL(vnet_iter); 1415 1416 if (!do_tcpdrain) 1417 return; 1418 1419 NET_EPOCH_ENTER(et); 1420 VNET_LIST_RLOCK_NOSLEEP(); 1421 VNET_FOREACH(vnet_iter) { 1422 CURVNET_SET(vnet_iter); 1423 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1424 INPLOOKUP_WLOCKPCB); 1425 struct inpcb *inpb; 1426 struct tcpcb *tcpb; 1427 1428 /* 1429 * Walk the tcpbs, if existing, and flush the reassembly queue, 1430 * if there is one... 1431 * XXX: The "Net/3" implementation doesn't imply that the TCP 1432 * reassembly queue should be flushed, but in a situation 1433 * where we're really low on mbufs, this is potentially 1434 * useful. 1435 */ 1436 while ((inpb = inp_next(&inpi)) != NULL) { 1437 if (inpb->inp_flags & INP_TIMEWAIT) 1438 continue; 1439 if ((tcpb = intotcpcb(inpb)) != NULL) { 1440 tcp_reass_flush(tcpb); 1441 tcp_clean_sackreport(tcpb); 1442 #ifdef TCP_BLACKBOX 1443 tcp_log_drain(tcpb); 1444 #endif 1445 #ifdef TCPPCAP 1446 if (tcp_pcap_aggressive_free) { 1447 /* Free the TCP PCAP queues. */ 1448 tcp_pcap_drain(&(tcpb->t_inpkts)); 1449 tcp_pcap_drain(&(tcpb->t_outpkts)); 1450 } 1451 #endif 1452 } 1453 } 1454 CURVNET_RESTORE(); 1455 } 1456 VNET_LIST_RUNLOCK_NOSLEEP(); 1457 NET_EPOCH_EXIT(et); 1458 } 1459 1460 static void 1461 tcp_vnet_init(void *arg __unused) 1462 { 1463 1464 #ifdef TCP_HHOOK 1465 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1466 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1467 printf("%s: WARNING: unable to register helper hook\n", __func__); 1468 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1469 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1470 printf("%s: WARNING: unable to register helper hook\n", __func__); 1471 #endif 1472 #ifdef STATS 1473 if (tcp_stats_init()) 1474 printf("%s: WARNING: unable to initialise TCP stats\n", 1475 __func__); 1476 #endif 1477 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1478 tcp_tcbhashsize); 1479 1480 /* 1481 * These have to be type stable for the benefit of the timers. 1482 */ 1483 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1484 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1485 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1486 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1487 1488 tcp_tw_init(); 1489 syncache_init(); 1490 tcp_hc_init(); 1491 1492 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1493 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1494 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1495 1496 tcp_fastopen_init(); 1497 1498 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1499 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1500 1501 V_tcp_msl = TCPTV_MSL; 1502 } 1503 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1504 tcp_vnet_init, NULL); 1505 1506 static void 1507 tcp_init(void *arg __unused) 1508 { 1509 const char *tcbhash_tuneable; 1510 int hashsize; 1511 1512 tcp_reass_global_init(); 1513 1514 /* XXX virtualize those below? */ 1515 tcp_delacktime = TCPTV_DELACK; 1516 tcp_keepinit = TCPTV_KEEP_INIT; 1517 tcp_keepidle = TCPTV_KEEP_IDLE; 1518 tcp_keepintvl = TCPTV_KEEPINTVL; 1519 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1520 tcp_rexmit_initial = TCPTV_RTOBASE; 1521 if (tcp_rexmit_initial < 1) 1522 tcp_rexmit_initial = 1; 1523 tcp_rexmit_min = TCPTV_MIN; 1524 if (tcp_rexmit_min < 1) 1525 tcp_rexmit_min = 1; 1526 tcp_persmin = TCPTV_PERSMIN; 1527 tcp_persmax = TCPTV_PERSMAX; 1528 tcp_rexmit_slop = TCPTV_CPU_VAR; 1529 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1530 1531 /* Setup the tcp function block list */ 1532 TAILQ_INIT(&t_functions); 1533 rw_init(&tcp_function_lock, "tcp_func_lock"); 1534 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1535 #ifdef TCP_BLACKBOX 1536 /* Initialize the TCP logging data. */ 1537 tcp_log_init(); 1538 #endif 1539 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1540 1541 if (tcp_soreceive_stream) { 1542 #ifdef INET 1543 tcp_protosw.pr_soreceive = soreceive_stream; 1544 #endif 1545 #ifdef INET6 1546 tcp6_protosw.pr_soreceive = soreceive_stream; 1547 #endif /* INET6 */ 1548 } 1549 1550 #ifdef INET6 1551 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1552 #else /* INET6 */ 1553 max_protohdr_grow(sizeof(struct tcpiphdr)); 1554 #endif /* INET6 */ 1555 1556 ISN_LOCK_INIT(); 1557 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1558 SHUTDOWN_PRI_DEFAULT); 1559 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1560 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1561 1562 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1563 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1564 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1565 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1566 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1567 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1568 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1569 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1570 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1571 #ifdef TCPPCAP 1572 tcp_pcap_init(); 1573 #endif 1574 1575 hashsize = TCBHASHSIZE; 1576 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1577 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1578 if (hashsize == 0) { 1579 /* 1580 * Auto tune the hash size based on maxsockets. 1581 * A perfect hash would have a 1:1 mapping 1582 * (hashsize = maxsockets) however it's been 1583 * suggested that O(2) average is better. 1584 */ 1585 hashsize = maketcp_hashsize(maxsockets / 4); 1586 /* 1587 * Our historical default is 512, 1588 * do not autotune lower than this. 1589 */ 1590 if (hashsize < 512) 1591 hashsize = 512; 1592 if (bootverbose) 1593 printf("%s: %s auto tuned to %d\n", __func__, 1594 tcbhash_tuneable, hashsize); 1595 } 1596 /* 1597 * We require a hashsize to be a power of two. 1598 * Previously if it was not a power of two we would just reset it 1599 * back to 512, which could be a nasty surprise if you did not notice 1600 * the error message. 1601 * Instead what we do is clip it to the closest power of two lower 1602 * than the specified hash value. 1603 */ 1604 if (!powerof2(hashsize)) { 1605 int oldhashsize = hashsize; 1606 1607 hashsize = maketcp_hashsize(hashsize); 1608 /* prevent absurdly low value */ 1609 if (hashsize < 16) 1610 hashsize = 16; 1611 printf("%s: WARNING: TCB hash size not a power of 2, " 1612 "clipped from %d to %d.\n", __func__, oldhashsize, 1613 hashsize); 1614 } 1615 tcp_tcbhashsize = hashsize; 1616 1617 #ifdef INET 1618 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1619 #endif 1620 #ifdef INET6 1621 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1622 #endif 1623 } 1624 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1625 1626 #ifdef VIMAGE 1627 static void 1628 tcp_destroy(void *unused __unused) 1629 { 1630 int n; 1631 #ifdef TCP_HHOOK 1632 int error; 1633 #endif 1634 1635 /* 1636 * All our processes are gone, all our sockets should be cleaned 1637 * up, which means, we should be past the tcp_discardcb() calls. 1638 * Sleep to let all tcpcb timers really disappear and cleanup. 1639 */ 1640 for (;;) { 1641 INP_INFO_WLOCK(&V_tcbinfo); 1642 n = V_tcbinfo.ipi_count; 1643 INP_INFO_WUNLOCK(&V_tcbinfo); 1644 if (n == 0) 1645 break; 1646 pause("tcpdes", hz / 10); 1647 } 1648 tcp_hc_destroy(); 1649 syncache_destroy(); 1650 tcp_tw_destroy(); 1651 in_pcbinfo_destroy(&V_tcbinfo); 1652 /* tcp_discardcb() clears the sack_holes up. */ 1653 uma_zdestroy(V_sack_hole_zone); 1654 uma_zdestroy(V_tcpcb_zone); 1655 1656 /* 1657 * Cannot free the zone until all tcpcbs are released as we attach 1658 * the allocations to them. 1659 */ 1660 tcp_fastopen_destroy(); 1661 1662 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1663 VNET_PCPUSTAT_FREE(tcpstat); 1664 1665 #ifdef TCP_HHOOK 1666 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1667 if (error != 0) { 1668 printf("%s: WARNING: unable to deregister helper hook " 1669 "type=%d, id=%d: error %d returned\n", __func__, 1670 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1671 } 1672 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1673 if (error != 0) { 1674 printf("%s: WARNING: unable to deregister helper hook " 1675 "type=%d, id=%d: error %d returned\n", __func__, 1676 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1677 } 1678 #endif 1679 } 1680 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1681 #endif 1682 1683 void 1684 tcp_fini(void *xtp) 1685 { 1686 1687 } 1688 1689 /* 1690 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1691 * tcp_template used to store this data in mbufs, but we now recopy it out 1692 * of the tcpcb each time to conserve mbufs. 1693 */ 1694 void 1695 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1696 { 1697 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1698 1699 INP_WLOCK_ASSERT(inp); 1700 1701 #ifdef INET6 1702 if ((inp->inp_vflag & INP_IPV6) != 0) { 1703 struct ip6_hdr *ip6; 1704 1705 ip6 = (struct ip6_hdr *)ip_ptr; 1706 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1707 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1708 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1709 (IPV6_VERSION & IPV6_VERSION_MASK); 1710 if (port == 0) 1711 ip6->ip6_nxt = IPPROTO_TCP; 1712 else 1713 ip6->ip6_nxt = IPPROTO_UDP; 1714 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1715 ip6->ip6_src = inp->in6p_laddr; 1716 ip6->ip6_dst = inp->in6p_faddr; 1717 } 1718 #endif /* INET6 */ 1719 #if defined(INET6) && defined(INET) 1720 else 1721 #endif 1722 #ifdef INET 1723 { 1724 struct ip *ip; 1725 1726 ip = (struct ip *)ip_ptr; 1727 ip->ip_v = IPVERSION; 1728 ip->ip_hl = 5; 1729 ip->ip_tos = inp->inp_ip_tos; 1730 ip->ip_len = 0; 1731 ip->ip_id = 0; 1732 ip->ip_off = 0; 1733 ip->ip_ttl = inp->inp_ip_ttl; 1734 ip->ip_sum = 0; 1735 if (port == 0) 1736 ip->ip_p = IPPROTO_TCP; 1737 else 1738 ip->ip_p = IPPROTO_UDP; 1739 ip->ip_src = inp->inp_laddr; 1740 ip->ip_dst = inp->inp_faddr; 1741 } 1742 #endif /* INET */ 1743 th->th_sport = inp->inp_lport; 1744 th->th_dport = inp->inp_fport; 1745 th->th_seq = 0; 1746 th->th_ack = 0; 1747 th->th_off = 5; 1748 tcp_set_flags(th, 0); 1749 th->th_win = 0; 1750 th->th_urp = 0; 1751 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1752 } 1753 1754 /* 1755 * Create template to be used to send tcp packets on a connection. 1756 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1757 * use for this function is in keepalives, which use tcp_respond. 1758 */ 1759 struct tcptemp * 1760 tcpip_maketemplate(struct inpcb *inp) 1761 { 1762 struct tcptemp *t; 1763 1764 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1765 if (t == NULL) 1766 return (NULL); 1767 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1768 return (t); 1769 } 1770 1771 /* 1772 * Send a single message to the TCP at address specified by 1773 * the given TCP/IP header. If m == NULL, then we make a copy 1774 * of the tcpiphdr at th and send directly to the addressed host. 1775 * This is used to force keep alive messages out using the TCP 1776 * template for a connection. If flags are given then we send 1777 * a message back to the TCP which originated the segment th, 1778 * and discard the mbuf containing it and any other attached mbufs. 1779 * 1780 * In any case the ack and sequence number of the transmitted 1781 * segment are as specified by the parameters. 1782 * 1783 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1784 */ 1785 void 1786 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1787 tcp_seq ack, tcp_seq seq, int flags) 1788 { 1789 struct tcpopt to; 1790 struct inpcb *inp; 1791 struct ip *ip; 1792 struct mbuf *optm; 1793 struct udphdr *uh = NULL; 1794 struct tcphdr *nth; 1795 struct tcp_log_buffer *lgb; 1796 u_char *optp; 1797 #ifdef INET6 1798 struct ip6_hdr *ip6; 1799 int isipv6; 1800 #endif /* INET6 */ 1801 int optlen, tlen, win, ulen; 1802 bool incl_opts; 1803 uint16_t port; 1804 int output_ret; 1805 #ifdef INVARIANTS 1806 int thflags = tcp_get_flags(th); 1807 #endif 1808 1809 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1810 NET_EPOCH_ASSERT(); 1811 1812 #ifdef INET6 1813 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1814 ip6 = ipgen; 1815 #endif /* INET6 */ 1816 ip = ipgen; 1817 1818 if (tp != NULL) { 1819 inp = tp->t_inpcb; 1820 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1821 INP_LOCK_ASSERT(inp); 1822 } else 1823 inp = NULL; 1824 1825 if (m != NULL) { 1826 #ifdef INET6 1827 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1828 port = m->m_pkthdr.tcp_tun_port; 1829 else 1830 #endif 1831 if (ip && (ip->ip_p == IPPROTO_UDP)) 1832 port = m->m_pkthdr.tcp_tun_port; 1833 else 1834 port = 0; 1835 } else 1836 port = tp->t_port; 1837 1838 incl_opts = false; 1839 win = 0; 1840 if (tp != NULL) { 1841 if (!(flags & TH_RST)) { 1842 win = sbspace(&inp->inp_socket->so_rcv); 1843 if (win > TCP_MAXWIN << tp->rcv_scale) 1844 win = TCP_MAXWIN << tp->rcv_scale; 1845 } 1846 if ((tp->t_flags & TF_NOOPT) == 0) 1847 incl_opts = true; 1848 } 1849 if (m == NULL) { 1850 m = m_gethdr(M_NOWAIT, MT_DATA); 1851 if (m == NULL) 1852 return; 1853 m->m_data += max_linkhdr; 1854 #ifdef INET6 1855 if (isipv6) { 1856 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1857 sizeof(struct ip6_hdr)); 1858 ip6 = mtod(m, struct ip6_hdr *); 1859 nth = (struct tcphdr *)(ip6 + 1); 1860 if (port) { 1861 /* Insert a UDP header */ 1862 uh = (struct udphdr *)nth; 1863 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1864 uh->uh_dport = port; 1865 nth = (struct tcphdr *)(uh + 1); 1866 } 1867 } else 1868 #endif /* INET6 */ 1869 { 1870 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1871 ip = mtod(m, struct ip *); 1872 nth = (struct tcphdr *)(ip + 1); 1873 if (port) { 1874 /* Insert a UDP header */ 1875 uh = (struct udphdr *)nth; 1876 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1877 uh->uh_dport = port; 1878 nth = (struct tcphdr *)(uh + 1); 1879 } 1880 } 1881 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1882 flags = TH_ACK; 1883 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1884 struct mbuf *n; 1885 1886 /* Can't reuse 'm', allocate a new mbuf. */ 1887 n = m_gethdr(M_NOWAIT, MT_DATA); 1888 if (n == NULL) { 1889 m_freem(m); 1890 return; 1891 } 1892 1893 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1894 m_freem(m); 1895 m_freem(n); 1896 return; 1897 } 1898 1899 n->m_data += max_linkhdr; 1900 /* m_len is set later */ 1901 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1902 #ifdef INET6 1903 if (isipv6) { 1904 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1905 sizeof(struct ip6_hdr)); 1906 ip6 = mtod(n, struct ip6_hdr *); 1907 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1908 nth = (struct tcphdr *)(ip6 + 1); 1909 if (port) { 1910 /* Insert a UDP header */ 1911 uh = (struct udphdr *)nth; 1912 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1913 uh->uh_dport = port; 1914 nth = (struct tcphdr *)(uh + 1); 1915 } 1916 } else 1917 #endif /* INET6 */ 1918 { 1919 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1920 ip = mtod(n, struct ip *); 1921 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1922 nth = (struct tcphdr *)(ip + 1); 1923 if (port) { 1924 /* Insert a UDP header */ 1925 uh = (struct udphdr *)nth; 1926 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1927 uh->uh_dport = port; 1928 nth = (struct tcphdr *)(uh + 1); 1929 } 1930 } 1931 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1932 xchg(nth->th_dport, nth->th_sport, uint16_t); 1933 th = nth; 1934 m_freem(m); 1935 m = n; 1936 } else { 1937 /* 1938 * reuse the mbuf. 1939 * XXX MRT We inherit the FIB, which is lucky. 1940 */ 1941 m_freem(m->m_next); 1942 m->m_next = NULL; 1943 m->m_data = (caddr_t)ipgen; 1944 /* m_len is set later */ 1945 #ifdef INET6 1946 if (isipv6) { 1947 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1948 nth = (struct tcphdr *)(ip6 + 1); 1949 } else 1950 #endif /* INET6 */ 1951 { 1952 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1953 nth = (struct tcphdr *)(ip + 1); 1954 } 1955 if (th != nth) { 1956 /* 1957 * this is usually a case when an extension header 1958 * exists between the IPv6 header and the 1959 * TCP header. 1960 */ 1961 nth->th_sport = th->th_sport; 1962 nth->th_dport = th->th_dport; 1963 } 1964 xchg(nth->th_dport, nth->th_sport, uint16_t); 1965 #undef xchg 1966 } 1967 tlen = 0; 1968 #ifdef INET6 1969 if (isipv6) 1970 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1971 #endif 1972 #if defined(INET) && defined(INET6) 1973 else 1974 #endif 1975 #ifdef INET 1976 tlen = sizeof (struct tcpiphdr); 1977 #endif 1978 if (port) 1979 tlen += sizeof (struct udphdr); 1980 #ifdef INVARIANTS 1981 m->m_len = 0; 1982 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1983 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1984 m, tlen, (long)M_TRAILINGSPACE(m))); 1985 #endif 1986 m->m_len = tlen; 1987 to.to_flags = 0; 1988 if (incl_opts) { 1989 /* Make sure we have room. */ 1990 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1991 m->m_next = m_get(M_NOWAIT, MT_DATA); 1992 if (m->m_next) { 1993 optp = mtod(m->m_next, u_char *); 1994 optm = m->m_next; 1995 } else 1996 incl_opts = false; 1997 } else { 1998 optp = (u_char *) (nth + 1); 1999 optm = m; 2000 } 2001 } 2002 if (incl_opts) { 2003 /* Timestamps. */ 2004 if (tp->t_flags & TF_RCVD_TSTMP) { 2005 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 2006 to.to_tsecr = tp->ts_recent; 2007 to.to_flags |= TOF_TS; 2008 } 2009 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2010 /* TCP-MD5 (RFC2385). */ 2011 if (tp->t_flags & TF_SIGNATURE) 2012 to.to_flags |= TOF_SIGNATURE; 2013 #endif 2014 /* Add the options. */ 2015 tlen += optlen = tcp_addoptions(&to, optp); 2016 2017 /* Update m_len in the correct mbuf. */ 2018 optm->m_len += optlen; 2019 } else 2020 optlen = 0; 2021 #ifdef INET6 2022 if (isipv6) { 2023 if (uh) { 2024 ulen = tlen - sizeof(struct ip6_hdr); 2025 uh->uh_ulen = htons(ulen); 2026 } 2027 ip6->ip6_flow = 0; 2028 ip6->ip6_vfc = IPV6_VERSION; 2029 if (port) 2030 ip6->ip6_nxt = IPPROTO_UDP; 2031 else 2032 ip6->ip6_nxt = IPPROTO_TCP; 2033 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 2034 } 2035 #endif 2036 #if defined(INET) && defined(INET6) 2037 else 2038 #endif 2039 #ifdef INET 2040 { 2041 if (uh) { 2042 ulen = tlen - sizeof(struct ip); 2043 uh->uh_ulen = htons(ulen); 2044 } 2045 ip->ip_len = htons(tlen); 2046 ip->ip_ttl = V_ip_defttl; 2047 if (port) { 2048 ip->ip_p = IPPROTO_UDP; 2049 } else { 2050 ip->ip_p = IPPROTO_TCP; 2051 } 2052 if (V_path_mtu_discovery) 2053 ip->ip_off |= htons(IP_DF); 2054 } 2055 #endif 2056 m->m_pkthdr.len = tlen; 2057 m->m_pkthdr.rcvif = NULL; 2058 #ifdef MAC 2059 if (inp != NULL) { 2060 /* 2061 * Packet is associated with a socket, so allow the 2062 * label of the response to reflect the socket label. 2063 */ 2064 INP_LOCK_ASSERT(inp); 2065 mac_inpcb_create_mbuf(inp, m); 2066 } else { 2067 /* 2068 * Packet is not associated with a socket, so possibly 2069 * update the label in place. 2070 */ 2071 mac_netinet_tcp_reply(m); 2072 } 2073 #endif 2074 nth->th_seq = htonl(seq); 2075 nth->th_ack = htonl(ack); 2076 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2077 tcp_set_flags(nth, flags); 2078 if (tp != NULL) 2079 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2080 else 2081 nth->th_win = htons((u_short)win); 2082 nth->th_urp = 0; 2083 2084 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2085 if (to.to_flags & TOF_SIGNATURE) { 2086 if (!TCPMD5_ENABLED() || 2087 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2088 m_freem(m); 2089 return; 2090 } 2091 } 2092 #endif 2093 2094 #ifdef INET6 2095 if (isipv6) { 2096 if (port) { 2097 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2098 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2099 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2100 nth->th_sum = 0; 2101 } else { 2102 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2103 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2104 nth->th_sum = in6_cksum_pseudo(ip6, 2105 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2106 } 2107 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 2108 NULL, NULL); 2109 } 2110 #endif /* INET6 */ 2111 #if defined(INET6) && defined(INET) 2112 else 2113 #endif 2114 #ifdef INET 2115 { 2116 if (port) { 2117 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2118 htons(ulen + IPPROTO_UDP)); 2119 m->m_pkthdr.csum_flags = CSUM_UDP; 2120 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2121 nth->th_sum = 0; 2122 } else { 2123 m->m_pkthdr.csum_flags = CSUM_TCP; 2124 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2125 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2126 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2127 } 2128 } 2129 #endif /* INET */ 2130 #ifdef TCPDEBUG 2131 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2132 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2133 #endif 2134 TCP_PROBE3(debug__output, tp, th, m); 2135 if (flags & TH_RST) 2136 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2137 lgb = NULL; 2138 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { 2139 if (INP_WLOCKED(inp)) { 2140 union tcp_log_stackspecific log; 2141 struct timeval tv; 2142 2143 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2144 log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts; 2145 log.u_bbr.flex8 = 4; 2146 log.u_bbr.pkts_out = tp->t_maxseg; 2147 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2148 log.u_bbr.delivered = 0; 2149 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, 2150 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2151 } else { 2152 /* 2153 * We can not log the packet, since we only own the 2154 * read lock, but a write lock is needed. The read lock 2155 * is not upgraded to a write lock, since only getting 2156 * the read lock was done intentionally to improve the 2157 * handling of SYN flooding attacks. 2158 * This happens only for pure SYN segments received in 2159 * the initial CLOSED state, or received in a more 2160 * advanced state than listen and the UDP encapsulation 2161 * port is unexpected. 2162 * The incoming SYN segments do not really belong to 2163 * the TCP connection and the handling does not change 2164 * the state of the TCP connection. Therefore, the 2165 * sending of the RST segments is not logged. Please 2166 * note that also the incoming SYN segments are not 2167 * logged. 2168 * 2169 * The following code ensures that the above description 2170 * is and stays correct. 2171 */ 2172 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2173 (tp->t_state == TCPS_CLOSED || 2174 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2175 ("%s: Logging of TCP segment with flags 0x%b and " 2176 "UDP encapsulation port %u skipped in state %s", 2177 __func__, thflags, PRINT_TH_FLAGS, 2178 ntohs(port), tcpstates[tp->t_state])); 2179 } 2180 } 2181 2182 if (flags & TH_ACK) 2183 TCPSTAT_INC(tcps_sndacks); 2184 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2185 TCPSTAT_INC(tcps_sndctrl); 2186 TCPSTAT_INC(tcps_sndtotal); 2187 2188 #ifdef INET6 2189 if (isipv6) { 2190 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2191 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2192 } 2193 #endif /* INET6 */ 2194 #if defined(INET) && defined(INET6) 2195 else 2196 #endif 2197 #ifdef INET 2198 { 2199 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2200 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2201 } 2202 #endif 2203 if (lgb != NULL) 2204 lgb->tlb_errno = output_ret; 2205 } 2206 2207 /* 2208 * Create a new TCP control block, making an 2209 * empty reassembly queue and hooking it to the argument 2210 * protocol control block. The `inp' parameter must have 2211 * come from the zone allocator set up in tcp_init(). 2212 */ 2213 struct tcpcb * 2214 tcp_newtcpcb(struct inpcb *inp) 2215 { 2216 struct tcpcb_mem *tm; 2217 struct tcpcb *tp; 2218 #ifdef INET6 2219 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2220 #endif /* INET6 */ 2221 2222 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 2223 if (tm == NULL) 2224 return (NULL); 2225 tp = &tm->tcb; 2226 2227 /* Initialise cc_var struct for this tcpcb. */ 2228 tp->ccv = &tm->ccv; 2229 tp->ccv->type = IPPROTO_TCP; 2230 tp->ccv->ccvc.tcp = tp; 2231 rw_rlock(&tcp_function_lock); 2232 tp->t_fb = tcp_func_set_ptr; 2233 refcount_acquire(&tp->t_fb->tfb_refcnt); 2234 rw_runlock(&tcp_function_lock); 2235 /* 2236 * Use the current system default CC algorithm. 2237 */ 2238 cc_attach(tp, CC_DEFAULT_ALGO()); 2239 2240 /* 2241 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 2242 * is called. 2243 */ 2244 in_pcbref(inp); /* Reference for tcpcb */ 2245 tp->t_inpcb = inp; 2246 2247 if (CC_ALGO(tp)->cb_init != NULL) 2248 if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) { 2249 cc_detach(tp); 2250 if (tp->t_fb->tfb_tcp_fb_fini) 2251 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2252 in_pcbrele_wlocked(inp); 2253 refcount_release(&tp->t_fb->tfb_refcnt); 2254 uma_zfree(V_tcpcb_zone, tm); 2255 return (NULL); 2256 } 2257 2258 #ifdef TCP_HHOOK 2259 tp->osd = &tm->osd; 2260 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 2261 if (tp->t_fb->tfb_tcp_fb_fini) 2262 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2263 in_pcbrele_wlocked(inp); 2264 refcount_release(&tp->t_fb->tfb_refcnt); 2265 uma_zfree(V_tcpcb_zone, tm); 2266 return (NULL); 2267 } 2268 #endif 2269 2270 #ifdef VIMAGE 2271 tp->t_vnet = inp->inp_vnet; 2272 #endif 2273 tp->t_timers = &tm->tt; 2274 TAILQ_INIT(&tp->t_segq); 2275 tp->t_maxseg = 2276 #ifdef INET6 2277 isipv6 ? V_tcp_v6mssdflt : 2278 #endif /* INET6 */ 2279 V_tcp_mssdflt; 2280 2281 /* Set up our timeouts. */ 2282 callout_init(&tp->t_timers->tt_rexmt, 1); 2283 callout_init(&tp->t_timers->tt_persist, 1); 2284 callout_init(&tp->t_timers->tt_keep, 1); 2285 callout_init(&tp->t_timers->tt_2msl, 1); 2286 callout_init(&tp->t_timers->tt_delack, 1); 2287 2288 switch (V_tcp_do_rfc1323) { 2289 case 0: 2290 break; 2291 default: 2292 case 1: 2293 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2294 break; 2295 case 2: 2296 tp->t_flags = TF_REQ_SCALE; 2297 break; 2298 case 3: 2299 tp->t_flags = TF_REQ_TSTMP; 2300 break; 2301 } 2302 if (V_tcp_do_sack) 2303 tp->t_flags |= TF_SACK_PERMIT; 2304 TAILQ_INIT(&tp->snd_holes); 2305 2306 /* 2307 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2308 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2309 * reasonable initial retransmit time. 2310 */ 2311 tp->t_srtt = TCPTV_SRTTBASE; 2312 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2313 tp->t_rttmin = tcp_rexmit_min; 2314 tp->t_rxtcur = tcp_rexmit_initial; 2315 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2316 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2317 tp->t_rcvtime = ticks; 2318 /* 2319 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2320 * because the socket may be bound to an IPv6 wildcard address, 2321 * which may match an IPv4-mapped IPv6 address. 2322 */ 2323 inp->inp_ip_ttl = V_ip_defttl; 2324 inp->inp_ppcb = tp; 2325 #ifdef TCPPCAP 2326 /* 2327 * Init the TCP PCAP queues. 2328 */ 2329 tcp_pcap_tcpcb_init(tp); 2330 #endif 2331 #ifdef TCP_BLACKBOX 2332 /* Initialize the per-TCPCB log data. */ 2333 tcp_log_tcpcbinit(tp); 2334 #endif 2335 tp->t_pacing_rate = -1; 2336 if (tp->t_fb->tfb_tcp_fb_init) { 2337 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2338 refcount_release(&tp->t_fb->tfb_refcnt); 2339 in_pcbrele_wlocked(inp); 2340 uma_zfree(V_tcpcb_zone, tm); 2341 return (NULL); 2342 } 2343 } 2344 #ifdef STATS 2345 if (V_tcp_perconn_stats_enable == 1) 2346 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2347 #endif 2348 if (V_tcp_do_lrd) 2349 tp->t_flags |= TF_LRD; 2350 return (tp); /* XXX */ 2351 } 2352 2353 /* 2354 * Drop a TCP connection, reporting 2355 * the specified error. If connection is synchronized, 2356 * then send a RST to peer. 2357 */ 2358 struct tcpcb * 2359 tcp_drop(struct tcpcb *tp, int errno) 2360 { 2361 struct socket *so = tp->t_inpcb->inp_socket; 2362 2363 NET_EPOCH_ASSERT(); 2364 INP_WLOCK_ASSERT(tp->t_inpcb); 2365 2366 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2367 tcp_state_change(tp, TCPS_CLOSED); 2368 /* Don't use tcp_output() here due to possible recursion. */ 2369 (void)tcp_output_nodrop(tp); 2370 TCPSTAT_INC(tcps_drops); 2371 } else 2372 TCPSTAT_INC(tcps_conndrops); 2373 if (errno == ETIMEDOUT && tp->t_softerror) 2374 errno = tp->t_softerror; 2375 so->so_error = errno; 2376 return (tcp_close(tp)); 2377 } 2378 2379 void 2380 tcp_discardcb(struct tcpcb *tp) 2381 { 2382 struct inpcb *inp = tp->t_inpcb; 2383 2384 INP_WLOCK_ASSERT(inp); 2385 2386 /* 2387 * Make sure that all of our timers are stopped before we delete the 2388 * PCB. 2389 * 2390 * If stopping a timer fails, we schedule a discard function in same 2391 * callout, and the last discard function called will take care of 2392 * deleting the tcpcb. 2393 */ 2394 tp->t_timers->tt_draincnt = 0; 2395 tcp_timer_stop(tp, TT_REXMT); 2396 tcp_timer_stop(tp, TT_PERSIST); 2397 tcp_timer_stop(tp, TT_KEEP); 2398 tcp_timer_stop(tp, TT_2MSL); 2399 tcp_timer_stop(tp, TT_DELACK); 2400 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2401 /* 2402 * Call the stop-all function of the methods, 2403 * this function should call the tcp_timer_stop() 2404 * method with each of the function specific timeouts. 2405 * That stop will be called via the tfb_tcp_timer_stop() 2406 * which should use the async drain function of the 2407 * callout system (see tcp_var.h). 2408 */ 2409 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2410 } 2411 2412 /* free the reassembly queue, if any */ 2413 tcp_reass_flush(tp); 2414 2415 #ifdef TCP_OFFLOAD 2416 /* Disconnect offload device, if any. */ 2417 if (tp->t_flags & TF_TOE) 2418 tcp_offload_detach(tp); 2419 #endif 2420 2421 tcp_free_sackholes(tp); 2422 2423 #ifdef TCPPCAP 2424 /* Free the TCP PCAP queues. */ 2425 tcp_pcap_drain(&(tp->t_inpkts)); 2426 tcp_pcap_drain(&(tp->t_outpkts)); 2427 #endif 2428 2429 /* Allow the CC algorithm to clean up after itself. */ 2430 if (CC_ALGO(tp)->cb_destroy != NULL) 2431 CC_ALGO(tp)->cb_destroy(tp->ccv); 2432 CC_DATA(tp) = NULL; 2433 /* Detach from the CC algorithm */ 2434 cc_detach(tp); 2435 2436 #ifdef TCP_HHOOK 2437 khelp_destroy_osd(tp->osd); 2438 #endif 2439 #ifdef STATS 2440 stats_blob_destroy(tp->t_stats); 2441 #endif 2442 2443 CC_ALGO(tp) = NULL; 2444 inp->inp_ppcb = NULL; 2445 if (tp->t_timers->tt_draincnt == 0) { 2446 bool released __diagused; 2447 2448 released = tcp_freecb(tp); 2449 KASSERT(!released, ("%s: inp %p should not have been released " 2450 "here", __func__, inp)); 2451 } 2452 } 2453 2454 bool 2455 tcp_freecb(struct tcpcb *tp) 2456 { 2457 struct inpcb *inp = tp->t_inpcb; 2458 struct socket *so = inp->inp_socket; 2459 #ifdef INET6 2460 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2461 #endif 2462 2463 INP_WLOCK_ASSERT(inp); 2464 MPASS(tp->t_timers->tt_draincnt == 0); 2465 2466 /* We own the last reference on tcpcb, let's free it. */ 2467 #ifdef TCP_BLACKBOX 2468 tcp_log_tcpcbfini(tp); 2469 #endif 2470 TCPSTATES_DEC(tp->t_state); 2471 if (tp->t_fb->tfb_tcp_fb_fini) 2472 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2473 2474 /* 2475 * If we got enough samples through the srtt filter, 2476 * save the rtt and rttvar in the routing entry. 2477 * 'Enough' is arbitrarily defined as 4 rtt samples. 2478 * 4 samples is enough for the srtt filter to converge 2479 * to within enough % of the correct value; fewer samples 2480 * and we could save a bogus rtt. The danger is not high 2481 * as tcp quickly recovers from everything. 2482 * XXX: Works very well but needs some more statistics! 2483 * 2484 * XXXRRS: Updating must be after the stack fini() since 2485 * that may be converting some internal representation of 2486 * say srtt etc into the general one used by other stacks. 2487 * Lets also at least protect against the so being NULL 2488 * as RW stated below. 2489 */ 2490 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2491 struct hc_metrics_lite metrics; 2492 uint32_t ssthresh; 2493 2494 bzero(&metrics, sizeof(metrics)); 2495 /* 2496 * Update the ssthresh always when the conditions below 2497 * are satisfied. This gives us better new start value 2498 * for the congestion avoidance for new connections. 2499 * ssthresh is only set if packet loss occurred on a session. 2500 * 2501 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2502 * being torn down. Ideally this code would not use 'so'. 2503 */ 2504 ssthresh = tp->snd_ssthresh; 2505 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2506 /* 2507 * convert the limit from user data bytes to 2508 * packets then to packet data bytes. 2509 */ 2510 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2511 if (ssthresh < 2) 2512 ssthresh = 2; 2513 ssthresh *= (tp->t_maxseg + 2514 #ifdef INET6 2515 (isipv6 ? sizeof (struct ip6_hdr) + 2516 sizeof (struct tcphdr) : 2517 #endif 2518 sizeof (struct tcpiphdr) 2519 #ifdef INET6 2520 ) 2521 #endif 2522 ); 2523 } else 2524 ssthresh = 0; 2525 metrics.rmx_ssthresh = ssthresh; 2526 2527 metrics.rmx_rtt = tp->t_srtt; 2528 metrics.rmx_rttvar = tp->t_rttvar; 2529 metrics.rmx_cwnd = tp->snd_cwnd; 2530 metrics.rmx_sendpipe = 0; 2531 metrics.rmx_recvpipe = 0; 2532 2533 tcp_hc_update(&inp->inp_inc, &metrics); 2534 } 2535 2536 refcount_release(&tp->t_fb->tfb_refcnt); 2537 uma_zfree(V_tcpcb_zone, tp); 2538 2539 return (in_pcbrele_wlocked(inp)); 2540 } 2541 2542 /* 2543 * Attempt to close a TCP control block, marking it as dropped, and freeing 2544 * the socket if we hold the only reference. 2545 */ 2546 struct tcpcb * 2547 tcp_close(struct tcpcb *tp) 2548 { 2549 struct inpcb *inp = tp->t_inpcb; 2550 struct socket *so; 2551 2552 INP_WLOCK_ASSERT(inp); 2553 2554 #ifdef TCP_OFFLOAD 2555 if (tp->t_state == TCPS_LISTEN) 2556 tcp_offload_listen_stop(tp); 2557 #endif 2558 /* 2559 * This releases the TFO pending counter resource for TFO listen 2560 * sockets as well as passively-created TFO sockets that transition 2561 * from SYN_RECEIVED to CLOSED. 2562 */ 2563 if (tp->t_tfo_pending) { 2564 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2565 tp->t_tfo_pending = NULL; 2566 } 2567 #ifdef TCPHPTS 2568 tcp_hpts_remove(inp); 2569 #endif 2570 in_pcbdrop(inp); 2571 TCPSTAT_INC(tcps_closed); 2572 if (tp->t_state != TCPS_CLOSED) 2573 tcp_state_change(tp, TCPS_CLOSED); 2574 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2575 so = inp->inp_socket; 2576 soisdisconnected(so); 2577 if (inp->inp_flags & INP_SOCKREF) { 2578 inp->inp_flags &= ~INP_SOCKREF; 2579 INP_WUNLOCK(inp); 2580 sorele(so); 2581 return (NULL); 2582 } 2583 return (tp); 2584 } 2585 2586 /* 2587 * Notify a tcp user of an asynchronous error; 2588 * store error as soft error, but wake up user 2589 * (for now, won't do anything until can select for soft error). 2590 * 2591 * Do not wake up user since there currently is no mechanism for 2592 * reporting soft errors (yet - a kqueue filter may be added). 2593 */ 2594 static struct inpcb * 2595 tcp_notify(struct inpcb *inp, int error) 2596 { 2597 struct tcpcb *tp; 2598 2599 INP_WLOCK_ASSERT(inp); 2600 2601 if ((inp->inp_flags & INP_TIMEWAIT) || 2602 (inp->inp_flags & INP_DROPPED)) 2603 return (inp); 2604 2605 tp = intotcpcb(inp); 2606 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2607 2608 /* 2609 * Ignore some errors if we are hooked up. 2610 * If connection hasn't completed, has retransmitted several times, 2611 * and receives a second error, give up now. This is better 2612 * than waiting a long time to establish a connection that 2613 * can never complete. 2614 */ 2615 if (tp->t_state == TCPS_ESTABLISHED && 2616 (error == EHOSTUNREACH || error == ENETUNREACH || 2617 error == EHOSTDOWN)) { 2618 if (inp->inp_route.ro_nh) { 2619 NH_FREE(inp->inp_route.ro_nh); 2620 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2621 } 2622 return (inp); 2623 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2624 tp->t_softerror) { 2625 tp = tcp_drop(tp, error); 2626 if (tp != NULL) 2627 return (inp); 2628 else 2629 return (NULL); 2630 } else { 2631 tp->t_softerror = error; 2632 return (inp); 2633 } 2634 #if 0 2635 wakeup( &so->so_timeo); 2636 sorwakeup(so); 2637 sowwakeup(so); 2638 #endif 2639 } 2640 2641 static int 2642 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2643 { 2644 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2645 INPLOOKUP_RLOCKPCB); 2646 struct xinpgen xig; 2647 struct inpcb *inp; 2648 int error; 2649 2650 if (req->newptr != NULL) 2651 return (EPERM); 2652 2653 if (req->oldptr == NULL) { 2654 int n; 2655 2656 n = V_tcbinfo.ipi_count + 2657 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2658 n += imax(n / 8, 10); 2659 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2660 return (0); 2661 } 2662 2663 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2664 return (error); 2665 2666 bzero(&xig, sizeof(xig)); 2667 xig.xig_len = sizeof xig; 2668 xig.xig_count = V_tcbinfo.ipi_count + 2669 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2670 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2671 xig.xig_sogen = so_gencnt; 2672 error = SYSCTL_OUT(req, &xig, sizeof xig); 2673 if (error) 2674 return (error); 2675 2676 error = syncache_pcblist(req); 2677 if (error) 2678 return (error); 2679 2680 while ((inp = inp_next(&inpi)) != NULL) { 2681 if (inp->inp_gencnt <= xig.xig_gen) { 2682 int crerr; 2683 2684 /* 2685 * XXX: This use of cr_cansee(), introduced with 2686 * TCP state changes, is not quite right, but for 2687 * now, better than nothing. 2688 */ 2689 if (inp->inp_flags & INP_TIMEWAIT) { 2690 if (intotw(inp) != NULL) 2691 crerr = cr_cansee(req->td->td_ucred, 2692 intotw(inp)->tw_cred); 2693 else 2694 crerr = EINVAL; /* Skip this inp. */ 2695 } else 2696 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2697 if (crerr == 0) { 2698 struct xtcpcb xt; 2699 2700 tcp_inptoxtp(inp, &xt); 2701 error = SYSCTL_OUT(req, &xt, sizeof xt); 2702 if (error) { 2703 INP_RUNLOCK(inp); 2704 break; 2705 } else 2706 continue; 2707 } 2708 } 2709 } 2710 2711 if (!error) { 2712 /* 2713 * Give the user an updated idea of our state. 2714 * If the generation differs from what we told 2715 * her before, she knows that something happened 2716 * while we were processing this request, and it 2717 * might be necessary to retry. 2718 */ 2719 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2720 xig.xig_sogen = so_gencnt; 2721 xig.xig_count = V_tcbinfo.ipi_count + 2722 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2723 error = SYSCTL_OUT(req, &xig, sizeof xig); 2724 } 2725 2726 return (error); 2727 } 2728 2729 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2730 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2731 NULL, 0, tcp_pcblist, "S,xtcpcb", 2732 "List of active TCP connections"); 2733 2734 #ifdef INET 2735 static int 2736 tcp_getcred(SYSCTL_HANDLER_ARGS) 2737 { 2738 struct xucred xuc; 2739 struct sockaddr_in addrs[2]; 2740 struct epoch_tracker et; 2741 struct inpcb *inp; 2742 int error; 2743 2744 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2745 if (error) 2746 return (error); 2747 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2748 if (error) 2749 return (error); 2750 NET_EPOCH_ENTER(et); 2751 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2752 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2753 NET_EPOCH_EXIT(et); 2754 if (inp != NULL) { 2755 if (inp->inp_socket == NULL) 2756 error = ENOENT; 2757 if (error == 0) 2758 error = cr_canseeinpcb(req->td->td_ucred, inp); 2759 if (error == 0) 2760 cru2x(inp->inp_cred, &xuc); 2761 INP_RUNLOCK(inp); 2762 } else 2763 error = ENOENT; 2764 if (error == 0) 2765 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2766 return (error); 2767 } 2768 2769 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2770 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2771 0, 0, tcp_getcred, "S,xucred", 2772 "Get the xucred of a TCP connection"); 2773 #endif /* INET */ 2774 2775 #ifdef INET6 2776 static int 2777 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2778 { 2779 struct epoch_tracker et; 2780 struct xucred xuc; 2781 struct sockaddr_in6 addrs[2]; 2782 struct inpcb *inp; 2783 int error; 2784 #ifdef INET 2785 int mapped = 0; 2786 #endif 2787 2788 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2789 if (error) 2790 return (error); 2791 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2792 if (error) 2793 return (error); 2794 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2795 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2796 return (error); 2797 } 2798 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2799 #ifdef INET 2800 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2801 mapped = 1; 2802 else 2803 #endif 2804 return (EINVAL); 2805 } 2806 2807 NET_EPOCH_ENTER(et); 2808 #ifdef INET 2809 if (mapped == 1) 2810 inp = in_pcblookup(&V_tcbinfo, 2811 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2812 addrs[1].sin6_port, 2813 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2814 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2815 else 2816 #endif 2817 inp = in6_pcblookup(&V_tcbinfo, 2818 &addrs[1].sin6_addr, addrs[1].sin6_port, 2819 &addrs[0].sin6_addr, addrs[0].sin6_port, 2820 INPLOOKUP_RLOCKPCB, NULL); 2821 NET_EPOCH_EXIT(et); 2822 if (inp != NULL) { 2823 if (inp->inp_socket == NULL) 2824 error = ENOENT; 2825 if (error == 0) 2826 error = cr_canseeinpcb(req->td->td_ucred, inp); 2827 if (error == 0) 2828 cru2x(inp->inp_cred, &xuc); 2829 INP_RUNLOCK(inp); 2830 } else 2831 error = ENOENT; 2832 if (error == 0) 2833 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2834 return (error); 2835 } 2836 2837 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2838 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2839 0, 0, tcp6_getcred, "S,xucred", 2840 "Get the xucred of a TCP6 connection"); 2841 #endif /* INET6 */ 2842 2843 #ifdef INET 2844 /* Path MTU to try next when a fragmentation-needed message is received. */ 2845 static inline int 2846 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2847 { 2848 int mtu = ntohs(icp->icmp_nextmtu); 2849 2850 /* If no alternative MTU was proposed, try the next smaller one. */ 2851 if (!mtu) 2852 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2853 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2854 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2855 2856 return (mtu); 2857 } 2858 2859 static void 2860 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2861 { 2862 struct ip *ip; 2863 struct tcphdr *th; 2864 struct inpcb *inp; 2865 struct tcpcb *tp; 2866 struct inpcb *(*notify)(struct inpcb *, int); 2867 struct in_conninfo inc; 2868 tcp_seq icmp_tcp_seq; 2869 int errno, mtu; 2870 2871 errno = icmp_errmap(icp); 2872 switch (errno) { 2873 case 0: 2874 return; 2875 case EMSGSIZE: 2876 notify = tcp_mtudisc_notify; 2877 break; 2878 case ECONNREFUSED: 2879 if (V_icmp_may_rst) 2880 notify = tcp_drop_syn_sent; 2881 else 2882 notify = tcp_notify; 2883 break; 2884 case EHOSTUNREACH: 2885 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2886 notify = tcp_drop_syn_sent; 2887 else 2888 notify = tcp_notify; 2889 break; 2890 default: 2891 notify = tcp_notify; 2892 } 2893 2894 ip = &icp->icmp_ip; 2895 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2896 icmp_tcp_seq = th->th_seq; 2897 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2898 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2899 if (inp != NULL) { 2900 if (!(inp->inp_flags & INP_TIMEWAIT) && 2901 !(inp->inp_flags & INP_DROPPED) && 2902 !(inp->inp_socket == NULL)) { 2903 tp = intotcpcb(inp); 2904 #ifdef TCP_OFFLOAD 2905 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2906 /* 2907 * MTU discovery for offloaded connections. Let 2908 * the TOE driver verify seq# and process it. 2909 */ 2910 mtu = tcp_next_pmtu(icp, ip); 2911 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2912 goto out; 2913 } 2914 #endif 2915 if (tp->t_port != port) { 2916 goto out; 2917 } 2918 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2919 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2920 if (errno == EMSGSIZE) { 2921 /* 2922 * MTU discovery: we got a needfrag and 2923 * will potentially try a lower MTU. 2924 */ 2925 mtu = tcp_next_pmtu(icp, ip); 2926 2927 /* 2928 * Only process the offered MTU if it 2929 * is smaller than the current one. 2930 */ 2931 if (mtu < tp->t_maxseg + 2932 sizeof(struct tcpiphdr)) { 2933 bzero(&inc, sizeof(inc)); 2934 inc.inc_faddr = ip->ip_dst; 2935 inc.inc_fibnum = 2936 inp->inp_inc.inc_fibnum; 2937 tcp_hc_updatemtu(&inc, mtu); 2938 inp = tcp_mtudisc(inp, mtu); 2939 } 2940 } else 2941 inp = (*notify)(inp, errno); 2942 } 2943 } 2944 } else { 2945 bzero(&inc, sizeof(inc)); 2946 inc.inc_fport = th->th_dport; 2947 inc.inc_lport = th->th_sport; 2948 inc.inc_faddr = ip->ip_dst; 2949 inc.inc_laddr = ip->ip_src; 2950 syncache_unreach(&inc, icmp_tcp_seq, port); 2951 } 2952 out: 2953 if (inp != NULL) 2954 INP_WUNLOCK(inp); 2955 } 2956 2957 static void 2958 tcp_ctlinput(struct icmp *icmp) 2959 { 2960 tcp_ctlinput_with_port(icmp, htons(0)); 2961 } 2962 2963 static void 2964 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2965 { 2966 /* Its a tunneled TCP over UDP icmp */ 2967 struct icmp *icmp = param.icmp; 2968 struct ip *outer_ip, *inner_ip; 2969 struct udphdr *udp; 2970 struct tcphdr *th, ttemp; 2971 int i_hlen, o_len; 2972 uint16_t port; 2973 2974 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2975 inner_ip = &icmp->icmp_ip; 2976 i_hlen = inner_ip->ip_hl << 2; 2977 o_len = ntohs(outer_ip->ip_len); 2978 if (o_len < 2979 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2980 /* Not enough data present */ 2981 return; 2982 } 2983 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2984 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2985 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2986 return; 2987 } 2988 port = udp->uh_dport; 2989 th = (struct tcphdr *)(udp + 1); 2990 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2991 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2992 /* Now adjust down the size of the outer IP header */ 2993 o_len -= sizeof(struct udphdr); 2994 outer_ip->ip_len = htons(o_len); 2995 /* Now call in to the normal handling code */ 2996 tcp_ctlinput_with_port(icmp, port); 2997 } 2998 #endif /* INET */ 2999 3000 #ifdef INET6 3001 static inline int 3002 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3003 { 3004 int mtu = ntohl(icmp6->icmp6_mtu); 3005 3006 /* 3007 * If no alternative MTU was proposed, or the proposed MTU was too 3008 * small, set to the min. 3009 */ 3010 if (mtu < IPV6_MMTU) 3011 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 3012 return (mtu); 3013 } 3014 3015 static void 3016 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 3017 { 3018 struct in6_addr *dst; 3019 struct inpcb *(*notify)(struct inpcb *, int); 3020 struct ip6_hdr *ip6; 3021 struct mbuf *m; 3022 struct inpcb *inp; 3023 struct tcpcb *tp; 3024 struct icmp6_hdr *icmp6; 3025 struct in_conninfo inc; 3026 struct tcp_ports { 3027 uint16_t th_sport; 3028 uint16_t th_dport; 3029 } t_ports; 3030 tcp_seq icmp_tcp_seq; 3031 unsigned int mtu; 3032 unsigned int off; 3033 int errno; 3034 3035 icmp6 = ip6cp->ip6c_icmp6; 3036 m = ip6cp->ip6c_m; 3037 ip6 = ip6cp->ip6c_ip6; 3038 off = ip6cp->ip6c_off; 3039 dst = &ip6cp->ip6c_finaldst->sin6_addr; 3040 3041 errno = icmp6_errmap(icmp6); 3042 switch (errno) { 3043 case 0: 3044 return; 3045 case EMSGSIZE: 3046 notify = tcp_mtudisc_notify; 3047 break; 3048 case ECONNREFUSED: 3049 if (V_icmp_may_rst) 3050 notify = tcp_drop_syn_sent; 3051 else 3052 notify = tcp_notify; 3053 break; 3054 case EHOSTUNREACH: 3055 /* 3056 * There are only four ICMPs that may reset connection: 3057 * - administratively prohibited 3058 * - port unreachable 3059 * - time exceeded in transit 3060 * - unknown next header 3061 */ 3062 if (V_icmp_may_rst && 3063 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 3064 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 3065 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 3066 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 3067 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 3068 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 3069 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 3070 notify = tcp_drop_syn_sent; 3071 else 3072 notify = tcp_notify; 3073 break; 3074 default: 3075 notify = tcp_notify; 3076 } 3077 3078 /* Check if we can safely get the ports from the tcp hdr */ 3079 if (m == NULL || 3080 (m->m_pkthdr.len < 3081 (int32_t) (off + sizeof(struct tcp_ports)))) { 3082 return; 3083 } 3084 bzero(&t_ports, sizeof(struct tcp_ports)); 3085 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3086 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3087 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3088 off += sizeof(struct tcp_ports); 3089 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3090 goto out; 3091 } 3092 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3093 if (inp != NULL) { 3094 if (!(inp->inp_flags & INP_TIMEWAIT) && 3095 !(inp->inp_flags & INP_DROPPED) && 3096 !(inp->inp_socket == NULL)) { 3097 tp = intotcpcb(inp); 3098 #ifdef TCP_OFFLOAD 3099 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3100 /* MTU discovery for offloaded connections. */ 3101 mtu = tcp6_next_pmtu(icmp6); 3102 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3103 goto out; 3104 } 3105 #endif 3106 if (tp->t_port != port) { 3107 goto out; 3108 } 3109 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3110 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3111 if (errno == EMSGSIZE) { 3112 /* 3113 * MTU discovery: 3114 * If we got a needfrag set the MTU 3115 * in the route to the suggested new 3116 * value (if given) and then notify. 3117 */ 3118 mtu = tcp6_next_pmtu(icmp6); 3119 3120 bzero(&inc, sizeof(inc)); 3121 inc.inc_fibnum = M_GETFIB(m); 3122 inc.inc_flags |= INC_ISIPV6; 3123 inc.inc6_faddr = *dst; 3124 if (in6_setscope(&inc.inc6_faddr, 3125 m->m_pkthdr.rcvif, NULL)) 3126 goto out; 3127 /* 3128 * Only process the offered MTU if it 3129 * is smaller than the current one. 3130 */ 3131 if (mtu < tp->t_maxseg + 3132 sizeof (struct tcphdr) + 3133 sizeof (struct ip6_hdr)) { 3134 tcp_hc_updatemtu(&inc, mtu); 3135 tcp_mtudisc(inp, mtu); 3136 ICMP6STAT_INC(icp6s_pmtuchg); 3137 } 3138 } else 3139 inp = (*notify)(inp, errno); 3140 } 3141 } 3142 } else { 3143 bzero(&inc, sizeof(inc)); 3144 inc.inc_fibnum = M_GETFIB(m); 3145 inc.inc_flags |= INC_ISIPV6; 3146 inc.inc_fport = t_ports.th_dport; 3147 inc.inc_lport = t_ports.th_sport; 3148 inc.inc6_faddr = *dst; 3149 inc.inc6_laddr = ip6->ip6_src; 3150 syncache_unreach(&inc, icmp_tcp_seq, port); 3151 } 3152 out: 3153 if (inp != NULL) 3154 INP_WUNLOCK(inp); 3155 } 3156 3157 static void 3158 tcp6_ctlinput(struct ip6ctlparam *ctl) 3159 { 3160 tcp6_ctlinput_with_port(ctl, htons(0)); 3161 } 3162 3163 static void 3164 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3165 { 3166 struct ip6ctlparam *ip6cp = param.ip6cp; 3167 struct mbuf *m; 3168 struct udphdr *udp; 3169 uint16_t port; 3170 3171 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3172 if (m == NULL) { 3173 return; 3174 } 3175 udp = mtod(m, struct udphdr *); 3176 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3177 return; 3178 } 3179 port = udp->uh_dport; 3180 m_adj(m, sizeof(struct udphdr)); 3181 if ((m->m_flags & M_PKTHDR) == 0) { 3182 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3183 } 3184 /* Now call in to the normal handling code */ 3185 tcp6_ctlinput_with_port(ip6cp, port); 3186 } 3187 3188 #endif /* INET6 */ 3189 3190 static uint32_t 3191 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3192 { 3193 SIPHASH_CTX ctx; 3194 uint32_t hash[2]; 3195 3196 KASSERT(len >= SIPHASH_KEY_LENGTH, 3197 ("%s: keylen %u too short ", __func__, len)); 3198 SipHash24_Init(&ctx); 3199 SipHash_SetKey(&ctx, (uint8_t *)key); 3200 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3201 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3202 switch (inc->inc_flags & INC_ISIPV6) { 3203 #ifdef INET 3204 case 0: 3205 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3206 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3207 break; 3208 #endif 3209 #ifdef INET6 3210 case INC_ISIPV6: 3211 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3212 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3213 break; 3214 #endif 3215 } 3216 SipHash_Final((uint8_t *)hash, &ctx); 3217 3218 return (hash[0] ^ hash[1]); 3219 } 3220 3221 uint32_t 3222 tcp_new_ts_offset(struct in_conninfo *inc) 3223 { 3224 struct in_conninfo inc_store, *local_inc; 3225 3226 if (!V_tcp_ts_offset_per_conn) { 3227 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3228 inc_store.inc_lport = 0; 3229 inc_store.inc_fport = 0; 3230 local_inc = &inc_store; 3231 } else { 3232 local_inc = inc; 3233 } 3234 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3235 sizeof(V_ts_offset_secret))); 3236 } 3237 3238 /* 3239 * Following is where TCP initial sequence number generation occurs. 3240 * 3241 * There are two places where we must use initial sequence numbers: 3242 * 1. In SYN-ACK packets. 3243 * 2. In SYN packets. 3244 * 3245 * All ISNs for SYN-ACK packets are generated by the syncache. See 3246 * tcp_syncache.c for details. 3247 * 3248 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3249 * depends on this property. In addition, these ISNs should be 3250 * unguessable so as to prevent connection hijacking. To satisfy 3251 * the requirements of this situation, the algorithm outlined in 3252 * RFC 1948 is used, with only small modifications. 3253 * 3254 * Implementation details: 3255 * 3256 * Time is based off the system timer, and is corrected so that it 3257 * increases by one megabyte per second. This allows for proper 3258 * recycling on high speed LANs while still leaving over an hour 3259 * before rollover. 3260 * 3261 * As reading the *exact* system time is too expensive to be done 3262 * whenever setting up a TCP connection, we increment the time 3263 * offset in two ways. First, a small random positive increment 3264 * is added to isn_offset for each connection that is set up. 3265 * Second, the function tcp_isn_tick fires once per clock tick 3266 * and increments isn_offset as necessary so that sequence numbers 3267 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3268 * random positive increments serve only to ensure that the same 3269 * exact sequence number is never sent out twice (as could otherwise 3270 * happen when a port is recycled in less than the system tick 3271 * interval.) 3272 * 3273 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3274 * between seeding of isn_secret. This is normally set to zero, 3275 * as reseeding should not be necessary. 3276 * 3277 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3278 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3279 * general, this means holding an exclusive (write) lock. 3280 */ 3281 3282 #define ISN_BYTES_PER_SECOND 1048576 3283 #define ISN_STATIC_INCREMENT 4096 3284 #define ISN_RANDOM_INCREMENT (4096 - 1) 3285 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3286 3287 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3288 VNET_DEFINE_STATIC(int, isn_last); 3289 VNET_DEFINE_STATIC(int, isn_last_reseed); 3290 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3291 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3292 3293 #define V_isn_secret VNET(isn_secret) 3294 #define V_isn_last VNET(isn_last) 3295 #define V_isn_last_reseed VNET(isn_last_reseed) 3296 #define V_isn_offset VNET(isn_offset) 3297 #define V_isn_offset_old VNET(isn_offset_old) 3298 3299 tcp_seq 3300 tcp_new_isn(struct in_conninfo *inc) 3301 { 3302 tcp_seq new_isn; 3303 u_int32_t projected_offset; 3304 3305 ISN_LOCK(); 3306 /* Seed if this is the first use, reseed if requested. */ 3307 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3308 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3309 < (u_int)ticks))) { 3310 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3311 V_isn_last_reseed = ticks; 3312 } 3313 3314 /* Compute the hash and return the ISN. */ 3315 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3316 sizeof(V_isn_secret)); 3317 V_isn_offset += ISN_STATIC_INCREMENT + 3318 (arc4random() & ISN_RANDOM_INCREMENT); 3319 if (ticks != V_isn_last) { 3320 projected_offset = V_isn_offset_old + 3321 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3322 if (SEQ_GT(projected_offset, V_isn_offset)) 3323 V_isn_offset = projected_offset; 3324 V_isn_offset_old = V_isn_offset; 3325 V_isn_last = ticks; 3326 } 3327 new_isn += V_isn_offset; 3328 ISN_UNLOCK(); 3329 return (new_isn); 3330 } 3331 3332 /* 3333 * When a specific ICMP unreachable message is received and the 3334 * connection state is SYN-SENT, drop the connection. This behavior 3335 * is controlled by the icmp_may_rst sysctl. 3336 */ 3337 static struct inpcb * 3338 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3339 { 3340 struct tcpcb *tp; 3341 3342 NET_EPOCH_ASSERT(); 3343 INP_WLOCK_ASSERT(inp); 3344 3345 if ((inp->inp_flags & INP_TIMEWAIT) || 3346 (inp->inp_flags & INP_DROPPED)) 3347 return (inp); 3348 3349 tp = intotcpcb(inp); 3350 if (tp->t_state != TCPS_SYN_SENT) 3351 return (inp); 3352 3353 if (IS_FASTOPEN(tp->t_flags)) 3354 tcp_fastopen_disable_path(tp); 3355 3356 tp = tcp_drop(tp, errno); 3357 if (tp != NULL) 3358 return (inp); 3359 else 3360 return (NULL); 3361 } 3362 3363 /* 3364 * When `need fragmentation' ICMP is received, update our idea of the MSS 3365 * based on the new value. Also nudge TCP to send something, since we 3366 * know the packet we just sent was dropped. 3367 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3368 */ 3369 static struct inpcb * 3370 tcp_mtudisc_notify(struct inpcb *inp, int error) 3371 { 3372 3373 return (tcp_mtudisc(inp, -1)); 3374 } 3375 3376 static struct inpcb * 3377 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3378 { 3379 struct tcpcb *tp; 3380 struct socket *so; 3381 3382 INP_WLOCK_ASSERT(inp); 3383 if ((inp->inp_flags & INP_TIMEWAIT) || 3384 (inp->inp_flags & INP_DROPPED)) 3385 return (inp); 3386 3387 tp = intotcpcb(inp); 3388 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3389 3390 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3391 3392 so = inp->inp_socket; 3393 SOCKBUF_LOCK(&so->so_snd); 3394 /* If the mss is larger than the socket buffer, decrease the mss. */ 3395 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3396 tp->t_maxseg = so->so_snd.sb_hiwat; 3397 SOCKBUF_UNLOCK(&so->so_snd); 3398 3399 TCPSTAT_INC(tcps_mturesent); 3400 tp->t_rtttime = 0; 3401 tp->snd_nxt = tp->snd_una; 3402 tcp_free_sackholes(tp); 3403 tp->snd_recover = tp->snd_max; 3404 if (tp->t_flags & TF_SACK_PERMIT) 3405 EXIT_FASTRECOVERY(tp->t_flags); 3406 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3407 /* 3408 * Conceptually the snd_nxt setting 3409 * and freeing sack holes should 3410 * be done by the default stacks 3411 * own tfb_tcp_mtu_chg(). 3412 */ 3413 tp->t_fb->tfb_tcp_mtu_chg(tp); 3414 } 3415 if (tcp_output(tp) < 0) 3416 return (NULL); 3417 else 3418 return (inp); 3419 } 3420 3421 #ifdef INET 3422 /* 3423 * Look-up the routing entry to the peer of this inpcb. If no route 3424 * is found and it cannot be allocated, then return 0. This routine 3425 * is called by TCP routines that access the rmx structure and by 3426 * tcp_mss_update to get the peer/interface MTU. 3427 */ 3428 uint32_t 3429 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3430 { 3431 struct nhop_object *nh; 3432 struct ifnet *ifp; 3433 uint32_t maxmtu = 0; 3434 3435 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3436 3437 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3438 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3439 if (nh == NULL) 3440 return (0); 3441 3442 ifp = nh->nh_ifp; 3443 maxmtu = nh->nh_mtu; 3444 3445 /* Report additional interface capabilities. */ 3446 if (cap != NULL) { 3447 if (ifp->if_capenable & IFCAP_TSO4 && 3448 ifp->if_hwassist & CSUM_TSO) { 3449 cap->ifcap |= CSUM_TSO; 3450 cap->tsomax = ifp->if_hw_tsomax; 3451 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3452 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3453 } 3454 } 3455 } 3456 return (maxmtu); 3457 } 3458 #endif /* INET */ 3459 3460 #ifdef INET6 3461 uint32_t 3462 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3463 { 3464 struct nhop_object *nh; 3465 struct in6_addr dst6; 3466 uint32_t scopeid; 3467 struct ifnet *ifp; 3468 uint32_t maxmtu = 0; 3469 3470 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3471 3472 if (inc->inc_flags & INC_IPV6MINMTU) 3473 return (IPV6_MMTU); 3474 3475 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3476 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3477 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3478 if (nh == NULL) 3479 return (0); 3480 3481 ifp = nh->nh_ifp; 3482 maxmtu = nh->nh_mtu; 3483 3484 /* Report additional interface capabilities. */ 3485 if (cap != NULL) { 3486 if (ifp->if_capenable & IFCAP_TSO6 && 3487 ifp->if_hwassist & CSUM_TSO) { 3488 cap->ifcap |= CSUM_TSO; 3489 cap->tsomax = ifp->if_hw_tsomax; 3490 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3491 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3492 } 3493 } 3494 } 3495 3496 return (maxmtu); 3497 } 3498 3499 /* 3500 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3501 * 3502 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3503 * The right place to do that is ip6_setpktopt() that has just been 3504 * executed. By the way it just filled ip6po_minmtu for us. 3505 */ 3506 void 3507 tcp6_use_min_mtu(struct tcpcb *tp) 3508 { 3509 struct inpcb *inp = tp->t_inpcb; 3510 3511 INP_WLOCK_ASSERT(inp); 3512 /* 3513 * In case of the IPV6_USE_MIN_MTU socket 3514 * option, the INC_IPV6MINMTU flag to announce 3515 * a corresponding MSS during the initial 3516 * handshake. If the TCP connection is not in 3517 * the front states, just reduce the MSS being 3518 * used. This avoids the sending of TCP 3519 * segments which will be fragmented at the 3520 * IPv6 layer. 3521 */ 3522 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3523 if ((tp->t_state >= TCPS_SYN_SENT) && 3524 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3525 struct ip6_pktopts *opt; 3526 3527 opt = inp->in6p_outputopts; 3528 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3529 tp->t_maxseg > TCP6_MSS) 3530 tp->t_maxseg = TCP6_MSS; 3531 } 3532 } 3533 #endif /* INET6 */ 3534 3535 /* 3536 * Calculate effective SMSS per RFC5681 definition for a given TCP 3537 * connection at its current state, taking into account SACK and etc. 3538 */ 3539 u_int 3540 tcp_maxseg(const struct tcpcb *tp) 3541 { 3542 u_int optlen; 3543 3544 if (tp->t_flags & TF_NOOPT) 3545 return (tp->t_maxseg); 3546 3547 /* 3548 * Here we have a simplified code from tcp_addoptions(), 3549 * without a proper loop, and having most of paddings hardcoded. 3550 * We might make mistakes with padding here in some edge cases, 3551 * but this is harmless, since result of tcp_maxseg() is used 3552 * only in cwnd and ssthresh estimations. 3553 */ 3554 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3555 if (tp->t_flags & TF_RCVD_TSTMP) 3556 optlen = TCPOLEN_TSTAMP_APPA; 3557 else 3558 optlen = 0; 3559 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3560 if (tp->t_flags & TF_SIGNATURE) 3561 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3562 #endif 3563 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3564 optlen += TCPOLEN_SACKHDR; 3565 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3566 optlen = PADTCPOLEN(optlen); 3567 } 3568 } else { 3569 if (tp->t_flags & TF_REQ_TSTMP) 3570 optlen = TCPOLEN_TSTAMP_APPA; 3571 else 3572 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3573 if (tp->t_flags & TF_REQ_SCALE) 3574 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3575 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3576 if (tp->t_flags & TF_SIGNATURE) 3577 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3578 #endif 3579 if (tp->t_flags & TF_SACK_PERMIT) 3580 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3581 } 3582 #undef PAD 3583 optlen = min(optlen, TCP_MAXOLEN); 3584 return (tp->t_maxseg - optlen); 3585 } 3586 3587 3588 u_int 3589 tcp_fixed_maxseg(const struct tcpcb *tp) 3590 { 3591 int optlen; 3592 3593 if (tp->t_flags & TF_NOOPT) 3594 return (tp->t_maxseg); 3595 3596 /* 3597 * Here we have a simplified code from tcp_addoptions(), 3598 * without a proper loop, and having most of paddings hardcoded. 3599 * We only consider fixed options that we would send every 3600 * time I.e. SACK is not considered. This is important 3601 * for cc modules to figure out what the modulo of the 3602 * cwnd should be. 3603 */ 3604 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3605 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3606 if (tp->t_flags & TF_RCVD_TSTMP) 3607 optlen = TCPOLEN_TSTAMP_APPA; 3608 else 3609 optlen = 0; 3610 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3611 if (tp->t_flags & TF_SIGNATURE) 3612 optlen += PAD(TCPOLEN_SIGNATURE); 3613 #endif 3614 } else { 3615 if (tp->t_flags & TF_REQ_TSTMP) 3616 optlen = TCPOLEN_TSTAMP_APPA; 3617 else 3618 optlen = PAD(TCPOLEN_MAXSEG); 3619 if (tp->t_flags & TF_REQ_SCALE) 3620 optlen += PAD(TCPOLEN_WINDOW); 3621 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3622 if (tp->t_flags & TF_SIGNATURE) 3623 optlen += PAD(TCPOLEN_SIGNATURE); 3624 #endif 3625 if (tp->t_flags & TF_SACK_PERMIT) 3626 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3627 } 3628 #undef PAD 3629 optlen = min(optlen, TCP_MAXOLEN); 3630 return (tp->t_maxseg - optlen); 3631 } 3632 3633 3634 3635 static int 3636 sysctl_drop(SYSCTL_HANDLER_ARGS) 3637 { 3638 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3639 struct sockaddr_storage addrs[2]; 3640 struct inpcb *inp; 3641 struct tcpcb *tp; 3642 struct tcptw *tw; 3643 #ifdef INET 3644 struct sockaddr_in *fin = NULL, *lin = NULL; 3645 #endif 3646 struct epoch_tracker et; 3647 #ifdef INET6 3648 struct sockaddr_in6 *fin6, *lin6; 3649 #endif 3650 int error; 3651 3652 inp = NULL; 3653 #ifdef INET6 3654 fin6 = lin6 = NULL; 3655 #endif 3656 error = 0; 3657 3658 if (req->oldptr != NULL || req->oldlen != 0) 3659 return (EINVAL); 3660 if (req->newptr == NULL) 3661 return (EPERM); 3662 if (req->newlen < sizeof(addrs)) 3663 return (ENOMEM); 3664 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3665 if (error) 3666 return (error); 3667 3668 switch (addrs[0].ss_family) { 3669 #ifdef INET6 3670 case AF_INET6: 3671 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3672 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3673 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3674 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3675 return (EINVAL); 3676 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3677 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3678 return (EINVAL); 3679 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3680 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3681 #ifdef INET 3682 fin = (struct sockaddr_in *)&addrs[0]; 3683 lin = (struct sockaddr_in *)&addrs[1]; 3684 #endif 3685 break; 3686 } 3687 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3688 if (error) 3689 return (error); 3690 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3691 if (error) 3692 return (error); 3693 break; 3694 #endif 3695 #ifdef INET 3696 case AF_INET: 3697 fin = (struct sockaddr_in *)&addrs[0]; 3698 lin = (struct sockaddr_in *)&addrs[1]; 3699 if (fin->sin_len != sizeof(struct sockaddr_in) || 3700 lin->sin_len != sizeof(struct sockaddr_in)) 3701 return (EINVAL); 3702 break; 3703 #endif 3704 default: 3705 return (EINVAL); 3706 } 3707 NET_EPOCH_ENTER(et); 3708 switch (addrs[0].ss_family) { 3709 #ifdef INET6 3710 case AF_INET6: 3711 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3712 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3713 INPLOOKUP_WLOCKPCB, NULL); 3714 break; 3715 #endif 3716 #ifdef INET 3717 case AF_INET: 3718 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3719 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3720 break; 3721 #endif 3722 } 3723 if (inp != NULL) { 3724 if (inp->inp_flags & INP_TIMEWAIT) { 3725 /* 3726 * XXXRW: There currently exists a state where an 3727 * inpcb is present, but its timewait state has been 3728 * discarded. For now, don't allow dropping of this 3729 * type of inpcb. 3730 */ 3731 tw = intotw(inp); 3732 if (tw != NULL) 3733 tcp_twclose(tw, 0); 3734 else 3735 INP_WUNLOCK(inp); 3736 } else if ((inp->inp_flags & INP_DROPPED) == 0 && 3737 !SOLISTENING(inp->inp_socket)) { 3738 tp = intotcpcb(inp); 3739 tp = tcp_drop(tp, ECONNABORTED); 3740 if (tp != NULL) 3741 INP_WUNLOCK(inp); 3742 } else 3743 INP_WUNLOCK(inp); 3744 } else 3745 error = ESRCH; 3746 NET_EPOCH_EXIT(et); 3747 return (error); 3748 } 3749 3750 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3751 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3752 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3753 "Drop TCP connection"); 3754 3755 static int 3756 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3757 { 3758 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3759 &tcp_ctloutput_set)); 3760 } 3761 3762 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3763 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3764 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3765 "Set socket option for TCP endpoint"); 3766 3767 #ifdef KERN_TLS 3768 static int 3769 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3770 { 3771 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3772 struct sockaddr_storage addrs[2]; 3773 struct inpcb *inp; 3774 #ifdef INET 3775 struct sockaddr_in *fin = NULL, *lin = NULL; 3776 #endif 3777 struct epoch_tracker et; 3778 #ifdef INET6 3779 struct sockaddr_in6 *fin6, *lin6; 3780 #endif 3781 int error; 3782 3783 inp = NULL; 3784 #ifdef INET6 3785 fin6 = lin6 = NULL; 3786 #endif 3787 error = 0; 3788 3789 if (req->oldptr != NULL || req->oldlen != 0) 3790 return (EINVAL); 3791 if (req->newptr == NULL) 3792 return (EPERM); 3793 if (req->newlen < sizeof(addrs)) 3794 return (ENOMEM); 3795 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3796 if (error) 3797 return (error); 3798 3799 switch (addrs[0].ss_family) { 3800 #ifdef INET6 3801 case AF_INET6: 3802 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3803 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3804 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3805 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3806 return (EINVAL); 3807 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3808 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3809 return (EINVAL); 3810 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3811 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3812 #ifdef INET 3813 fin = (struct sockaddr_in *)&addrs[0]; 3814 lin = (struct sockaddr_in *)&addrs[1]; 3815 #endif 3816 break; 3817 } 3818 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3819 if (error) 3820 return (error); 3821 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3822 if (error) 3823 return (error); 3824 break; 3825 #endif 3826 #ifdef INET 3827 case AF_INET: 3828 fin = (struct sockaddr_in *)&addrs[0]; 3829 lin = (struct sockaddr_in *)&addrs[1]; 3830 if (fin->sin_len != sizeof(struct sockaddr_in) || 3831 lin->sin_len != sizeof(struct sockaddr_in)) 3832 return (EINVAL); 3833 break; 3834 #endif 3835 default: 3836 return (EINVAL); 3837 } 3838 NET_EPOCH_ENTER(et); 3839 switch (addrs[0].ss_family) { 3840 #ifdef INET6 3841 case AF_INET6: 3842 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3843 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3844 INPLOOKUP_WLOCKPCB, NULL); 3845 break; 3846 #endif 3847 #ifdef INET 3848 case AF_INET: 3849 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3850 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3851 break; 3852 #endif 3853 } 3854 NET_EPOCH_EXIT(et); 3855 if (inp != NULL) { 3856 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3857 inp->inp_socket == NULL) { 3858 error = ECONNRESET; 3859 INP_WUNLOCK(inp); 3860 } else { 3861 struct socket *so; 3862 3863 so = inp->inp_socket; 3864 soref(so); 3865 error = ktls_set_tx_mode(so, 3866 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3867 INP_WUNLOCK(inp); 3868 sorele(so); 3869 } 3870 } else 3871 error = ESRCH; 3872 return (error); 3873 } 3874 3875 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3876 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3877 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3878 "Switch TCP connection to SW TLS"); 3879 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3880 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3881 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3882 "Switch TCP connection to ifnet TLS"); 3883 #endif 3884 3885 /* 3886 * Generate a standardized TCP log line for use throughout the 3887 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3888 * allow use in the interrupt context. 3889 * 3890 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3891 * NB: The function may return NULL if memory allocation failed. 3892 * 3893 * Due to header inclusion and ordering limitations the struct ip 3894 * and ip6_hdr pointers have to be passed as void pointers. 3895 */ 3896 char * 3897 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3898 const void *ip6hdr) 3899 { 3900 3901 /* Is logging enabled? */ 3902 if (V_tcp_log_in_vain == 0) 3903 return (NULL); 3904 3905 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3906 } 3907 3908 char * 3909 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3910 const void *ip6hdr) 3911 { 3912 3913 /* Is logging enabled? */ 3914 if (tcp_log_debug == 0) 3915 return (NULL); 3916 3917 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3918 } 3919 3920 static char * 3921 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3922 const void *ip6hdr) 3923 { 3924 char *s, *sp; 3925 size_t size; 3926 #ifdef INET 3927 const struct ip *ip = (const struct ip *)ip4hdr; 3928 #endif 3929 #ifdef INET6 3930 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3931 #endif /* INET6 */ 3932 3933 /* 3934 * The log line looks like this: 3935 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3936 */ 3937 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3938 sizeof(PRINT_TH_FLAGS) + 1 + 3939 #ifdef INET6 3940 2 * INET6_ADDRSTRLEN; 3941 #else 3942 2 * INET_ADDRSTRLEN; 3943 #endif /* INET6 */ 3944 3945 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3946 if (s == NULL) 3947 return (NULL); 3948 3949 strcat(s, "TCP: ["); 3950 sp = s + strlen(s); 3951 3952 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3953 inet_ntoa_r(inc->inc_faddr, sp); 3954 sp = s + strlen(s); 3955 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3956 sp = s + strlen(s); 3957 inet_ntoa_r(inc->inc_laddr, sp); 3958 sp = s + strlen(s); 3959 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3960 #ifdef INET6 3961 } else if (inc) { 3962 ip6_sprintf(sp, &inc->inc6_faddr); 3963 sp = s + strlen(s); 3964 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3965 sp = s + strlen(s); 3966 ip6_sprintf(sp, &inc->inc6_laddr); 3967 sp = s + strlen(s); 3968 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3969 } else if (ip6 && th) { 3970 ip6_sprintf(sp, &ip6->ip6_src); 3971 sp = s + strlen(s); 3972 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3973 sp = s + strlen(s); 3974 ip6_sprintf(sp, &ip6->ip6_dst); 3975 sp = s + strlen(s); 3976 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3977 #endif /* INET6 */ 3978 #ifdef INET 3979 } else if (ip && th) { 3980 inet_ntoa_r(ip->ip_src, sp); 3981 sp = s + strlen(s); 3982 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3983 sp = s + strlen(s); 3984 inet_ntoa_r(ip->ip_dst, sp); 3985 sp = s + strlen(s); 3986 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3987 #endif /* INET */ 3988 } else { 3989 free(s, M_TCPLOG); 3990 return (NULL); 3991 } 3992 sp = s + strlen(s); 3993 if (th) 3994 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3995 if (*(s + size - 1) != '\0') 3996 panic("%s: string too long", __func__); 3997 return (s); 3998 } 3999 4000 /* 4001 * A subroutine which makes it easy to track TCP state changes with DTrace. 4002 * This function shouldn't be called for t_state initializations that don't 4003 * correspond to actual TCP state transitions. 4004 */ 4005 void 4006 tcp_state_change(struct tcpcb *tp, int newstate) 4007 { 4008 #if defined(KDTRACE_HOOKS) 4009 int pstate = tp->t_state; 4010 #endif 4011 4012 TCPSTATES_DEC(tp->t_state); 4013 TCPSTATES_INC(newstate); 4014 tp->t_state = newstate; 4015 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4016 } 4017 4018 /* 4019 * Create an external-format (``xtcpcb'') structure using the information in 4020 * the kernel-format tcpcb structure pointed to by tp. This is done to 4021 * reduce the spew of irrelevant information over this interface, to isolate 4022 * user code from changes in the kernel structure, and potentially to provide 4023 * information-hiding if we decide that some of this information should be 4024 * hidden from users. 4025 */ 4026 void 4027 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4028 { 4029 struct tcpcb *tp = intotcpcb(inp); 4030 struct tcptw *tw = intotw(inp); 4031 sbintime_t now; 4032 4033 bzero(xt, sizeof(*xt)); 4034 if (inp->inp_flags & INP_TIMEWAIT) { 4035 xt->t_state = TCPS_TIME_WAIT; 4036 xt->xt_encaps_port = tw->t_port; 4037 } else { 4038 xt->t_state = tp->t_state; 4039 xt->t_logstate = tp->t_logstate; 4040 xt->t_flags = tp->t_flags; 4041 xt->t_sndzerowin = tp->t_sndzerowin; 4042 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4043 xt->t_rcvoopack = tp->t_rcvoopack; 4044 xt->t_rcv_wnd = tp->rcv_wnd; 4045 xt->t_snd_wnd = tp->snd_wnd; 4046 xt->t_snd_cwnd = tp->snd_cwnd; 4047 xt->t_snd_ssthresh = tp->snd_ssthresh; 4048 xt->t_dsack_bytes = tp->t_dsack_bytes; 4049 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4050 xt->t_dsack_pack = tp->t_dsack_pack; 4051 xt->t_maxseg = tp->t_maxseg; 4052 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4053 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4054 4055 now = getsbinuptime(); 4056 #define COPYTIMER(ttt) do { \ 4057 if (callout_active(&tp->t_timers->ttt)) \ 4058 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 4059 SBT_1MS; \ 4060 else \ 4061 xt->ttt = 0; \ 4062 } while (0) 4063 COPYTIMER(tt_delack); 4064 COPYTIMER(tt_rexmt); 4065 COPYTIMER(tt_persist); 4066 COPYTIMER(tt_keep); 4067 COPYTIMER(tt_2msl); 4068 #undef COPYTIMER 4069 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4070 4071 xt->xt_encaps_port = tp->t_port; 4072 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4073 TCP_FUNCTION_NAME_LEN_MAX); 4074 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 4075 TCP_CA_NAME_MAX); 4076 #ifdef TCP_BLACKBOX 4077 (void)tcp_log_get_id(tp, xt->xt_logid); 4078 #endif 4079 } 4080 4081 xt->xt_len = sizeof(struct xtcpcb); 4082 in_pcbtoxinpcb(inp, &xt->xt_inp); 4083 if (inp->inp_socket == NULL) 4084 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 4085 } 4086 4087 void 4088 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4089 { 4090 uint32_t bit, i; 4091 4092 if ((tp == NULL) || 4093 (status > TCP_EI_STATUS_MAX_VALUE) || 4094 (status == 0)) { 4095 /* Invalid */ 4096 return; 4097 } 4098 if (status > (sizeof(uint32_t) * 8)) { 4099 /* Should this be a KASSERT? */ 4100 return; 4101 } 4102 bit = 1U << (status - 1); 4103 if (bit & tp->t_end_info_status) { 4104 /* already logged */ 4105 return; 4106 } 4107 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4108 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4109 tp->t_end_info_bytes[i] = status; 4110 tp->t_end_info_status |= bit; 4111 break; 4112 } 4113 } 4114 } 4115 4116 int 4117 tcp_can_enable_pacing(void) 4118 { 4119 4120 if ((tcp_pacing_limit == -1) || 4121 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4122 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4123 shadow_num_connections = number_of_tcp_connections_pacing; 4124 return (1); 4125 } else { 4126 return (0); 4127 } 4128 } 4129 4130 static uint8_t tcp_pacing_warning = 0; 4131 4132 void 4133 tcp_decrement_paced_conn(void) 4134 { 4135 uint32_t ret; 4136 4137 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4138 shadow_num_connections = number_of_tcp_connections_pacing; 4139 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4140 if (ret == 0) { 4141 if (tcp_pacing_limit != -1) { 4142 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4143 tcp_pacing_limit = 0; 4144 } else if (tcp_pacing_warning == 0) { 4145 printf("Warning pacing count is invalid, invalid decrement\n"); 4146 tcp_pacing_warning = 1; 4147 } 4148 } 4149 } 4150