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_ecn.h> 109 #include <netinet/tcp_log_buf.h> 110 #include <netinet/tcp_syncache.h> 111 #include <netinet/tcp_hpts.h> 112 #include <netinet/cc/cc.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/tcp_fastopen.h> 115 #ifdef TCPPCAP 116 #include <netinet/tcp_pcap.h> 117 #endif 118 #ifdef TCPDEBUG 119 #include <netinet/tcp_debug.h> 120 #endif 121 #ifdef TCP_OFFLOAD 122 #include <netinet/tcp_offload.h> 123 #endif 124 #include <netinet/udp.h> 125 #include <netinet/udp_var.h> 126 #ifdef INET6 127 #include <netinet6/tcp6_var.h> 128 #endif 129 130 #include <netipsec/ipsec_support.h> 131 132 #include <machine/in_cksum.h> 133 #include <crypto/siphash/siphash.h> 134 135 #include <security/mac/mac_framework.h> 136 137 #ifdef INET6 138 static ip6proto_ctlinput_t tcp6_ctlinput; 139 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 140 #endif 141 142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 143 #ifdef INET6 144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 145 #endif 146 147 #ifdef NETFLIX_EXP_DETECTION 148 /* Sack attack detection thresholds and such */ 149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 150 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 151 "Sack Attack detection thresholds"); 152 int32_t tcp_force_detection = 0; 153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 154 CTLFLAG_RW, 155 &tcp_force_detection, 0, 156 "Do we force detection even if the INP has it off?"); 157 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 159 CTLFLAG_RW, 160 &tcp_sack_to_ack_thresh, 700, 161 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 162 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 164 CTLFLAG_RW, 165 &tcp_sack_to_move_thresh, 600, 166 "Percentage of sack moves we must see above (10.1 percent is 101)"); 167 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 169 CTLFLAG_RW, 170 &tcp_restoral_thresh, 550, 171 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 172 int32_t tcp_sad_decay_val = 800; 173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 174 CTLFLAG_RW, 175 &tcp_sad_decay_val, 800, 176 "The decay percentage (10.1 percent equals 101 )"); 177 int32_t tcp_map_minimum = 500; 178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 179 CTLFLAG_RW, 180 &tcp_map_minimum, 500, 181 "Number of Map enteries before we start detection"); 182 int32_t tcp_attack_on_turns_on_logging = 0; 183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 184 CTLFLAG_RW, 185 &tcp_attack_on_turns_on_logging, 0, 186 "When we have a positive hit on attack, do we turn on logging?"); 187 int32_t tcp_sad_pacing_interval = 2000; 188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 189 CTLFLAG_RW, 190 &tcp_sad_pacing_interval, 2000, 191 "What is the minimum pacing interval for a classified attacker?"); 192 193 int32_t tcp_sad_low_pps = 100; 194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 195 CTLFLAG_RW, 196 &tcp_sad_low_pps, 100, 197 "What is the input pps that below which we do not decay?"); 198 #endif 199 uint32_t tcp_ack_war_time_window = 1000; 200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 201 CTLFLAG_RW, 202 &tcp_ack_war_time_window, 1000, 203 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 204 uint32_t tcp_ack_war_cnt = 5; 205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 206 CTLFLAG_RW, 207 &tcp_ack_war_cnt, 5, 208 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 209 210 struct rwlock tcp_function_lock; 211 212 static int 213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 214 { 215 int error, new; 216 217 new = V_tcp_mssdflt; 218 error = sysctl_handle_int(oidp, &new, 0, req); 219 if (error == 0 && req->newptr) { 220 if (new < TCP_MINMSS) 221 error = EINVAL; 222 else 223 V_tcp_mssdflt = new; 224 } 225 return (error); 226 } 227 228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 229 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 230 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 231 "Default TCP Maximum Segment Size"); 232 233 #ifdef INET6 234 static int 235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 236 { 237 int error, new; 238 239 new = V_tcp_v6mssdflt; 240 error = sysctl_handle_int(oidp, &new, 0, req); 241 if (error == 0 && req->newptr) { 242 if (new < TCP_MINMSS) 243 error = EINVAL; 244 else 245 V_tcp_v6mssdflt = new; 246 } 247 return (error); 248 } 249 250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 251 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 252 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 253 "Default TCP Maximum Segment Size for IPv6"); 254 #endif /* INET6 */ 255 256 /* 257 * Minimum MSS we accept and use. This prevents DoS attacks where 258 * we are forced to a ridiculous low MSS like 20 and send hundreds 259 * of packets instead of one. The effect scales with the available 260 * bandwidth and quickly saturates the CPU and network interface 261 * with packet generation and sending. Set to zero to disable MINMSS 262 * checking. This setting prevents us from sending too small packets. 263 */ 264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 266 &VNET_NAME(tcp_minmss), 0, 267 "Minimum TCP Maximum Segment Size"); 268 269 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 271 &VNET_NAME(tcp_do_rfc1323), 0, 272 "Enable rfc1323 (high performance TCP) extensions"); 273 274 /* 275 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 276 * Some stacks negotiate TS, but never send them after connection setup. Some 277 * stacks negotiate TS, but don't send them when sending keep-alive segments. 278 * These include modern widely deployed TCP stacks. 279 * Therefore tolerating violations for now... 280 */ 281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 283 &VNET_NAME(tcp_tolerate_missing_ts), 0, 284 "Tolerate missing TCP timestamps"); 285 286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 288 &VNET_NAME(tcp_ts_offset_per_conn), 0, 289 "Initialize TCP timestamps per connection instead of per host pair"); 290 291 /* How many connections are pacing */ 292 static volatile uint32_t number_of_tcp_connections_pacing = 0; 293 static uint32_t shadow_num_connections = 0; 294 295 static int tcp_pacing_limit = 10000; 296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 297 &tcp_pacing_limit, 1000, 298 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 299 300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 301 &shadow_num_connections, 0, "Number of TCP connections being paced"); 302 303 static int tcp_log_debug = 0; 304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 305 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 306 307 static int tcp_tcbhashsize; 308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 309 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 310 311 static int do_tcpdrain = 1; 312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 313 "Enable tcp_drain routine for extra help when low on mbufs"); 314 315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 316 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 317 318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 319 #define V_icmp_may_rst VNET(icmp_may_rst) 320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 321 &VNET_NAME(icmp_may_rst), 0, 322 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 323 324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 325 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 327 &VNET_NAME(tcp_isn_reseed_interval), 0, 328 "Seconds between reseeding of ISN secret"); 329 330 static int tcp_soreceive_stream; 331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 332 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 333 334 VNET_DEFINE(uma_zone_t, sack_hole_zone); 335 #define V_sack_hole_zone VNET(sack_hole_zone) 336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 337 static int 338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 339 { 340 int error; 341 uint32_t new; 342 343 new = V_tcp_map_entries_limit; 344 error = sysctl_handle_int(oidp, &new, 0, req); 345 if (error == 0 && req->newptr) { 346 /* only allow "0" and value > minimum */ 347 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 348 error = EINVAL; 349 else 350 V_tcp_map_entries_limit = new; 351 } 352 return (error); 353 } 354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 355 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 356 &VNET_NAME(tcp_map_entries_limit), 0, 357 &sysctl_net_inet_tcp_map_limit_check, "IU", 358 "Total sendmap entries limit"); 359 360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 362 &VNET_NAME(tcp_map_split_limit), 0, 363 "Total sendmap split entries limit"); 364 365 #ifdef TCP_HHOOK 366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 367 #endif 368 369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 371 #define V_ts_offset_secret VNET(ts_offset_secret) 372 373 static int tcp_default_fb_init(struct tcpcb *tp); 374 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 375 static int tcp_default_handoff_ok(struct tcpcb *tp); 376 static struct inpcb *tcp_notify(struct inpcb *, int); 377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 378 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 379 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 380 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 381 const void *ip4hdr, const void *ip6hdr); 382 static ipproto_ctlinput_t tcp_ctlinput; 383 static udp_tun_icmp_t tcp_ctlinput_viaudp; 384 385 static struct tcp_function_block tcp_def_funcblk = { 386 .tfb_tcp_block_name = "freebsd", 387 .tfb_tcp_output = tcp_default_output, 388 .tfb_tcp_do_segment = tcp_do_segment, 389 .tfb_tcp_ctloutput = tcp_default_ctloutput, 390 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 391 .tfb_tcp_fb_init = tcp_default_fb_init, 392 .tfb_tcp_fb_fini = tcp_default_fb_fini, 393 }; 394 395 static int tcp_fb_cnt = 0; 396 struct tcp_funchead t_functions; 397 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 398 399 void 400 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 401 { 402 TCPSTAT_INC(tcps_dsack_count); 403 tp->t_dsack_pack++; 404 if (tlp == 0) { 405 if (SEQ_GT(end, start)) { 406 tp->t_dsack_bytes += (end - start); 407 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 408 } else { 409 tp->t_dsack_tlp_bytes += (start - end); 410 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 411 } 412 } else { 413 if (SEQ_GT(end, start)) { 414 tp->t_dsack_bytes += (end - start); 415 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 416 } else { 417 tp->t_dsack_tlp_bytes += (start - end); 418 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 419 } 420 } 421 } 422 423 static struct tcp_function_block * 424 find_tcp_functions_locked(struct tcp_function_set *fs) 425 { 426 struct tcp_function *f; 427 struct tcp_function_block *blk=NULL; 428 429 TAILQ_FOREACH(f, &t_functions, tf_next) { 430 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 431 blk = f->tf_fb; 432 break; 433 } 434 } 435 return(blk); 436 } 437 438 static struct tcp_function_block * 439 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 440 { 441 struct tcp_function_block *rblk=NULL; 442 struct tcp_function *f; 443 444 TAILQ_FOREACH(f, &t_functions, tf_next) { 445 if (f->tf_fb == blk) { 446 rblk = blk; 447 if (s) { 448 *s = f; 449 } 450 break; 451 } 452 } 453 return (rblk); 454 } 455 456 struct tcp_function_block * 457 find_and_ref_tcp_functions(struct tcp_function_set *fs) 458 { 459 struct tcp_function_block *blk; 460 461 rw_rlock(&tcp_function_lock); 462 blk = find_tcp_functions_locked(fs); 463 if (blk) 464 refcount_acquire(&blk->tfb_refcnt); 465 rw_runlock(&tcp_function_lock); 466 return(blk); 467 } 468 469 struct tcp_function_block * 470 find_and_ref_tcp_fb(struct tcp_function_block *blk) 471 { 472 struct tcp_function_block *rblk; 473 474 rw_rlock(&tcp_function_lock); 475 rblk = find_tcp_fb_locked(blk, NULL); 476 if (rblk) 477 refcount_acquire(&rblk->tfb_refcnt); 478 rw_runlock(&tcp_function_lock); 479 return(rblk); 480 } 481 482 /* Find a matching alias for the given tcp_function_block. */ 483 int 484 find_tcp_function_alias(struct tcp_function_block *blk, 485 struct tcp_function_set *fs) 486 { 487 struct tcp_function *f; 488 int found; 489 490 found = 0; 491 rw_rlock(&tcp_function_lock); 492 TAILQ_FOREACH(f, &t_functions, tf_next) { 493 if ((f->tf_fb == blk) && 494 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 495 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 496 /* Matching function block with different name. */ 497 strncpy(fs->function_set_name, f->tf_name, 498 TCP_FUNCTION_NAME_LEN_MAX); 499 found = 1; 500 break; 501 } 502 } 503 /* Null terminate the string appropriately. */ 504 if (found) { 505 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 506 } else { 507 fs->function_set_name[0] = '\0'; 508 } 509 rw_runlock(&tcp_function_lock); 510 return (found); 511 } 512 513 static struct tcp_function_block * 514 find_and_ref_tcp_default_fb(void) 515 { 516 struct tcp_function_block *rblk; 517 518 rw_rlock(&tcp_function_lock); 519 rblk = tcp_func_set_ptr; 520 refcount_acquire(&rblk->tfb_refcnt); 521 rw_runlock(&tcp_function_lock); 522 return (rblk); 523 } 524 525 void 526 tcp_switch_back_to_default(struct tcpcb *tp) 527 { 528 struct tcp_function_block *tfb; 529 530 KASSERT(tp->t_fb != &tcp_def_funcblk, 531 ("%s: called by the built-in default stack", __func__)); 532 533 /* 534 * Release the old stack. This function will either find a new one 535 * or panic. 536 */ 537 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 538 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 539 refcount_release(&tp->t_fb->tfb_refcnt); 540 541 /* 542 * Now, we'll find a new function block to use. 543 * Start by trying the current user-selected 544 * default, unless this stack is the user-selected 545 * default. 546 */ 547 tfb = find_and_ref_tcp_default_fb(); 548 if (tfb == tp->t_fb) { 549 refcount_release(&tfb->tfb_refcnt); 550 tfb = NULL; 551 } 552 /* Does the stack accept this connection? */ 553 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 554 (*tfb->tfb_tcp_handoff_ok)(tp)) { 555 refcount_release(&tfb->tfb_refcnt); 556 tfb = NULL; 557 } 558 /* Try to use that stack. */ 559 if (tfb != NULL) { 560 /* Initialize the new stack. If it succeeds, we are done. */ 561 tp->t_fb = tfb; 562 if (tp->t_fb->tfb_tcp_fb_init == NULL || 563 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 564 return; 565 566 /* 567 * Initialization failed. Release the reference count on 568 * the stack. 569 */ 570 refcount_release(&tfb->tfb_refcnt); 571 } 572 573 /* 574 * If that wasn't feasible, use the built-in default 575 * stack which is not allowed to reject anyone. 576 */ 577 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 578 if (tfb == NULL) { 579 /* there always should be a default */ 580 panic("Can't refer to tcp_def_funcblk"); 581 } 582 if (tfb->tfb_tcp_handoff_ok != NULL) { 583 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 584 /* The default stack cannot say no */ 585 panic("Default stack rejects a new session?"); 586 } 587 } 588 tp->t_fb = tfb; 589 if (tp->t_fb->tfb_tcp_fb_init != NULL && 590 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 591 /* The default stack cannot fail */ 592 panic("Default stack initialization failed"); 593 } 594 } 595 596 static bool 597 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 598 const struct sockaddr *sa, void *ctx) 599 { 600 struct ip *iph; 601 #ifdef INET6 602 struct ip6_hdr *ip6; 603 #endif 604 struct udphdr *uh; 605 struct tcphdr *th; 606 int thlen; 607 uint16_t port; 608 609 TCPSTAT_INC(tcps_tunneled_pkts); 610 if ((m->m_flags & M_PKTHDR) == 0) { 611 /* Can't handle one that is not a pkt hdr */ 612 TCPSTAT_INC(tcps_tunneled_errs); 613 goto out; 614 } 615 thlen = sizeof(struct tcphdr); 616 if (m->m_len < off + sizeof(struct udphdr) + thlen && 617 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 618 TCPSTAT_INC(tcps_tunneled_errs); 619 goto out; 620 } 621 iph = mtod(m, struct ip *); 622 uh = (struct udphdr *)((caddr_t)iph + off); 623 th = (struct tcphdr *)(uh + 1); 624 thlen = th->th_off << 2; 625 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 626 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 627 if (m == NULL) { 628 TCPSTAT_INC(tcps_tunneled_errs); 629 goto out; 630 } else { 631 iph = mtod(m, struct ip *); 632 uh = (struct udphdr *)((caddr_t)iph + off); 633 th = (struct tcphdr *)(uh + 1); 634 } 635 } 636 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 637 bcopy(th, uh, m->m_len - off); 638 m->m_len -= sizeof(struct udphdr); 639 m->m_pkthdr.len -= sizeof(struct udphdr); 640 /* 641 * We use the same algorithm for 642 * both UDP and TCP for c-sum. So 643 * the code in tcp_input will skip 644 * the checksum. So we do nothing 645 * with the flag (m->m_pkthdr.csum_flags). 646 */ 647 switch (iph->ip_v) { 648 #ifdef INET 649 case IPVERSION: 650 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 651 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 652 break; 653 #endif 654 #ifdef INET6 655 case IPV6_VERSION >> 4: 656 ip6 = mtod(m, struct ip6_hdr *); 657 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 658 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 659 break; 660 #endif 661 default: 662 goto out; 663 break; 664 } 665 return (true); 666 out: 667 m_freem(m); 668 669 return (true); 670 } 671 672 static int 673 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 674 { 675 int error=ENOENT; 676 struct tcp_function_set fs; 677 struct tcp_function_block *blk; 678 679 memset(&fs, 0, sizeof(fs)); 680 rw_rlock(&tcp_function_lock); 681 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 682 if (blk) { 683 /* Found him */ 684 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 685 fs.pcbcnt = blk->tfb_refcnt; 686 } 687 rw_runlock(&tcp_function_lock); 688 error = sysctl_handle_string(oidp, fs.function_set_name, 689 sizeof(fs.function_set_name), req); 690 691 /* Check for error or no change */ 692 if (error != 0 || req->newptr == NULL) 693 return(error); 694 695 rw_wlock(&tcp_function_lock); 696 blk = find_tcp_functions_locked(&fs); 697 if ((blk == NULL) || 698 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 699 error = ENOENT; 700 goto done; 701 } 702 tcp_func_set_ptr = blk; 703 done: 704 rw_wunlock(&tcp_function_lock); 705 return (error); 706 } 707 708 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 709 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 710 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 711 "Set/get the default TCP functions"); 712 713 static int 714 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 715 { 716 int error, cnt, linesz; 717 struct tcp_function *f; 718 char *buffer, *cp; 719 size_t bufsz, outsz; 720 bool alias; 721 722 cnt = 0; 723 rw_rlock(&tcp_function_lock); 724 TAILQ_FOREACH(f, &t_functions, tf_next) { 725 cnt++; 726 } 727 rw_runlock(&tcp_function_lock); 728 729 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 730 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 731 732 error = 0; 733 cp = buffer; 734 735 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 736 "Alias", "PCB count"); 737 cp += linesz; 738 bufsz -= linesz; 739 outsz = linesz; 740 741 rw_rlock(&tcp_function_lock); 742 TAILQ_FOREACH(f, &t_functions, tf_next) { 743 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 744 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 745 f->tf_fb->tfb_tcp_block_name, 746 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 747 alias ? f->tf_name : "-", 748 f->tf_fb->tfb_refcnt); 749 if (linesz >= bufsz) { 750 error = EOVERFLOW; 751 break; 752 } 753 cp += linesz; 754 bufsz -= linesz; 755 outsz += linesz; 756 } 757 rw_runlock(&tcp_function_lock); 758 if (error == 0) 759 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 760 free(buffer, M_TEMP); 761 return (error); 762 } 763 764 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 765 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 766 NULL, 0, sysctl_net_inet_list_available, "A", 767 "list available TCP Function sets"); 768 769 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 770 771 #ifdef INET 772 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 773 #define V_udp4_tun_socket VNET(udp4_tun_socket) 774 #endif 775 #ifdef INET6 776 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 777 #define V_udp6_tun_socket VNET(udp6_tun_socket) 778 #endif 779 780 static void 781 tcp_over_udp_stop(void) 782 { 783 /* 784 * This function assumes sysctl caller holds inp_rinfo_lock() 785 * for writing! 786 */ 787 #ifdef INET 788 if (V_udp4_tun_socket != NULL) { 789 soclose(V_udp4_tun_socket); 790 V_udp4_tun_socket = NULL; 791 } 792 #endif 793 #ifdef INET6 794 if (V_udp6_tun_socket != NULL) { 795 soclose(V_udp6_tun_socket); 796 V_udp6_tun_socket = NULL; 797 } 798 #endif 799 } 800 801 static int 802 tcp_over_udp_start(void) 803 { 804 uint16_t port; 805 int ret; 806 #ifdef INET 807 struct sockaddr_in sin; 808 #endif 809 #ifdef INET6 810 struct sockaddr_in6 sin6; 811 #endif 812 /* 813 * This function assumes sysctl caller holds inp_info_rlock() 814 * for writing! 815 */ 816 port = V_tcp_udp_tunneling_port; 817 if (ntohs(port) == 0) { 818 /* Must have a port set */ 819 return (EINVAL); 820 } 821 #ifdef INET 822 if (V_udp4_tun_socket != NULL) { 823 /* Already running -- must stop first */ 824 return (EALREADY); 825 } 826 #endif 827 #ifdef INET6 828 if (V_udp6_tun_socket != NULL) { 829 /* Already running -- must stop first */ 830 return (EALREADY); 831 } 832 #endif 833 #ifdef INET 834 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 835 SOCK_DGRAM, IPPROTO_UDP, 836 curthread->td_ucred, curthread))) { 837 tcp_over_udp_stop(); 838 return (ret); 839 } 840 /* Call the special UDP hook. */ 841 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 842 tcp_recv_udp_tunneled_packet, 843 tcp_ctlinput_viaudp, 844 NULL))) { 845 tcp_over_udp_stop(); 846 return (ret); 847 } 848 /* Ok, we have a socket, bind it to the port. */ 849 memset(&sin, 0, sizeof(struct sockaddr_in)); 850 sin.sin_len = sizeof(struct sockaddr_in); 851 sin.sin_family = AF_INET; 852 sin.sin_port = htons(port); 853 if ((ret = sobind(V_udp4_tun_socket, 854 (struct sockaddr *)&sin, curthread))) { 855 tcp_over_udp_stop(); 856 return (ret); 857 } 858 #endif 859 #ifdef INET6 860 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 861 SOCK_DGRAM, IPPROTO_UDP, 862 curthread->td_ucred, curthread))) { 863 tcp_over_udp_stop(); 864 return (ret); 865 } 866 /* Call the special UDP hook. */ 867 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 868 tcp_recv_udp_tunneled_packet, 869 tcp6_ctlinput_viaudp, 870 NULL))) { 871 tcp_over_udp_stop(); 872 return (ret); 873 } 874 /* Ok, we have a socket, bind it to the port. */ 875 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 876 sin6.sin6_len = sizeof(struct sockaddr_in6); 877 sin6.sin6_family = AF_INET6; 878 sin6.sin6_port = htons(port); 879 if ((ret = sobind(V_udp6_tun_socket, 880 (struct sockaddr *)&sin6, curthread))) { 881 tcp_over_udp_stop(); 882 return (ret); 883 } 884 #endif 885 return (0); 886 } 887 888 static int 889 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 890 { 891 int error; 892 uint32_t old, new; 893 894 old = V_tcp_udp_tunneling_port; 895 new = old; 896 error = sysctl_handle_int(oidp, &new, 0, req); 897 if ((error == 0) && 898 (req->newptr != NULL)) { 899 if ((new < TCP_TUNNELING_PORT_MIN) || 900 (new > TCP_TUNNELING_PORT_MAX)) { 901 error = EINVAL; 902 } else { 903 V_tcp_udp_tunneling_port = new; 904 if (old != 0) { 905 tcp_over_udp_stop(); 906 } 907 if (new != 0) { 908 error = tcp_over_udp_start(); 909 } 910 } 911 } 912 return (error); 913 } 914 915 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 916 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 917 &VNET_NAME(tcp_udp_tunneling_port), 918 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 919 "Tunneling port for tcp over udp"); 920 921 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 922 923 static int 924 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 925 { 926 int error, new; 927 928 new = V_tcp_udp_tunneling_overhead; 929 error = sysctl_handle_int(oidp, &new, 0, req); 930 if (error == 0 && req->newptr) { 931 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 932 (new > TCP_TUNNELING_OVERHEAD_MAX)) 933 error = EINVAL; 934 else 935 V_tcp_udp_tunneling_overhead = new; 936 } 937 return (error); 938 } 939 940 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 941 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 942 &VNET_NAME(tcp_udp_tunneling_overhead), 943 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 944 "MSS reduction when using tcp over udp"); 945 946 /* 947 * Exports one (struct tcp_function_info) for each alias/name. 948 */ 949 static int 950 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 951 { 952 int cnt, error; 953 struct tcp_function *f; 954 struct tcp_function_info tfi; 955 956 /* 957 * We don't allow writes. 958 */ 959 if (req->newptr != NULL) 960 return (EINVAL); 961 962 /* 963 * Wire the old buffer so we can directly copy the functions to 964 * user space without dropping the lock. 965 */ 966 if (req->oldptr != NULL) { 967 error = sysctl_wire_old_buffer(req, 0); 968 if (error) 969 return (error); 970 } 971 972 /* 973 * Walk the list and copy out matching entries. If INVARIANTS 974 * is compiled in, also walk the list to verify the length of 975 * the list matches what we have recorded. 976 */ 977 rw_rlock(&tcp_function_lock); 978 979 cnt = 0; 980 #ifndef INVARIANTS 981 if (req->oldptr == NULL) { 982 cnt = tcp_fb_cnt; 983 goto skip_loop; 984 } 985 #endif 986 TAILQ_FOREACH(f, &t_functions, tf_next) { 987 #ifdef INVARIANTS 988 cnt++; 989 #endif 990 if (req->oldptr != NULL) { 991 bzero(&tfi, sizeof(tfi)); 992 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 993 tfi.tfi_id = f->tf_fb->tfb_id; 994 (void)strlcpy(tfi.tfi_alias, f->tf_name, 995 sizeof(tfi.tfi_alias)); 996 (void)strlcpy(tfi.tfi_name, 997 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 998 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 999 /* 1000 * Don't stop on error, as that is the 1001 * mechanism we use to accumulate length 1002 * information if the buffer was too short. 1003 */ 1004 } 1005 } 1006 KASSERT(cnt == tcp_fb_cnt, 1007 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1008 #ifndef INVARIANTS 1009 skip_loop: 1010 #endif 1011 rw_runlock(&tcp_function_lock); 1012 if (req->oldptr == NULL) 1013 error = SYSCTL_OUT(req, NULL, 1014 (cnt + 1) * sizeof(struct tcp_function_info)); 1015 1016 return (error); 1017 } 1018 1019 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1020 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1021 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1022 "List TCP function block name-to-ID mappings"); 1023 1024 /* 1025 * tfb_tcp_handoff_ok() function for the default stack. 1026 * Note that we'll basically try to take all comers. 1027 */ 1028 static int 1029 tcp_default_handoff_ok(struct tcpcb *tp) 1030 { 1031 1032 return (0); 1033 } 1034 1035 /* 1036 * tfb_tcp_fb_init() function for the default stack. 1037 * 1038 * This handles making sure we have appropriate timers set if you are 1039 * transitioning a socket that has some amount of setup done. 1040 * 1041 * The init() fuction from the default can *never* return non-zero i.e. 1042 * it is required to always succeed since it is the stack of last resort! 1043 */ 1044 static int 1045 tcp_default_fb_init(struct tcpcb *tp) 1046 { 1047 struct socket *so = tptosocket(tp); 1048 1049 INP_WLOCK_ASSERT(tptoinpcb(tp)); 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 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1067 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1068 tcp_timer_active(tp, TT_PERSIST))) { 1069 /* 1070 * If the session has established and it looks like it should 1071 * be in the persist state, set the persist timer. Otherwise, 1072 * set the retransmit timer. 1073 */ 1074 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1075 (int32_t)(tp->snd_nxt - tp->snd_una) < 1076 (int32_t)sbavail(&so->so_snd)) 1077 tcp_setpersist(tp); 1078 else 1079 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1080 } 1081 1082 /* All non-embryonic sessions get a keepalive timer. */ 1083 if (!tcp_timer_active(tp, TT_KEEP)) 1084 tcp_timer_activate(tp, TT_KEEP, 1085 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1086 TP_KEEPINIT(tp)); 1087 1088 /* 1089 * Make sure critical variables are initialized 1090 * if transitioning while in Recovery. 1091 */ 1092 if IN_FASTRECOVERY(tp->t_flags) { 1093 if (tp->sackhint.recover_fs == 0) 1094 tp->sackhint.recover_fs = max(1, 1095 tp->snd_nxt - tp->snd_una); 1096 } 1097 1098 return (0); 1099 } 1100 1101 /* 1102 * tfb_tcp_fb_fini() function for the default stack. 1103 * 1104 * This changes state as necessary (or prudent) to prepare for another stack 1105 * to assume responsibility for the connection. 1106 */ 1107 static void 1108 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1109 { 1110 1111 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1112 } 1113 1114 /* 1115 * Target size of TCP PCB hash tables. Must be a power of two. 1116 * 1117 * Note that this can be overridden by the kernel environment 1118 * variable net.inet.tcp.tcbhashsize 1119 */ 1120 #ifndef TCBHASHSIZE 1121 #define TCBHASHSIZE 0 1122 #endif 1123 1124 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1125 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1126 1127 static struct mtx isn_mtx; 1128 1129 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1130 #define ISN_LOCK() mtx_lock(&isn_mtx) 1131 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1132 1133 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1134 1135 /* 1136 * Take a value and get the next power of 2 that doesn't overflow. 1137 * Used to size the tcp_inpcb hash buckets. 1138 */ 1139 static int 1140 maketcp_hashsize(int size) 1141 { 1142 int hashsize; 1143 1144 /* 1145 * auto tune. 1146 * get the next power of 2 higher than maxsockets. 1147 */ 1148 hashsize = 1 << fls(size); 1149 /* catch overflow, and just go one power of 2 smaller */ 1150 if (hashsize < size) { 1151 hashsize = 1 << (fls(size) - 1); 1152 } 1153 return (hashsize); 1154 } 1155 1156 static volatile int next_tcp_stack_id = 1; 1157 1158 /* 1159 * Register a TCP function block with the name provided in the names 1160 * array. (Note that this function does NOT automatically register 1161 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1162 * explicitly include blk->tfb_tcp_block_name in the list of names if 1163 * you wish to register the stack with that name.) 1164 * 1165 * Either all name registrations will succeed or all will fail. If 1166 * a name registration fails, the function will update the num_names 1167 * argument to point to the array index of the name that encountered 1168 * the failure. 1169 * 1170 * Returns 0 on success, or an error code on failure. 1171 */ 1172 int 1173 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1174 const char *names[], int *num_names) 1175 { 1176 struct tcp_function *n; 1177 struct tcp_function_set fs; 1178 int error, i; 1179 1180 KASSERT(names != NULL && *num_names > 0, 1181 ("%s: Called with 0-length name list", __func__)); 1182 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1183 KASSERT(rw_initialized(&tcp_function_lock), 1184 ("%s: called too early", __func__)); 1185 1186 if ((blk->tfb_tcp_output == NULL) || 1187 (blk->tfb_tcp_do_segment == NULL) || 1188 (blk->tfb_tcp_ctloutput == NULL) || 1189 (strlen(blk->tfb_tcp_block_name) == 0)) { 1190 /* 1191 * These functions are required and you 1192 * need a name. 1193 */ 1194 *num_names = 0; 1195 return (EINVAL); 1196 } 1197 1198 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1199 *num_names = 0; 1200 return (EINVAL); 1201 } 1202 1203 refcount_init(&blk->tfb_refcnt, 0); 1204 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1205 for (i = 0; i < *num_names; i++) { 1206 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1207 if (n == NULL) { 1208 error = ENOMEM; 1209 goto cleanup; 1210 } 1211 n->tf_fb = blk; 1212 1213 (void)strlcpy(fs.function_set_name, names[i], 1214 sizeof(fs.function_set_name)); 1215 rw_wlock(&tcp_function_lock); 1216 if (find_tcp_functions_locked(&fs) != NULL) { 1217 /* Duplicate name space not allowed */ 1218 rw_wunlock(&tcp_function_lock); 1219 free(n, M_TCPFUNCTIONS); 1220 error = EALREADY; 1221 goto cleanup; 1222 } 1223 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1224 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1225 tcp_fb_cnt++; 1226 rw_wunlock(&tcp_function_lock); 1227 } 1228 return(0); 1229 1230 cleanup: 1231 /* 1232 * Deregister the names we just added. Because registration failed 1233 * for names[i], we don't need to deregister that name. 1234 */ 1235 *num_names = i; 1236 rw_wlock(&tcp_function_lock); 1237 while (--i >= 0) { 1238 TAILQ_FOREACH(n, &t_functions, tf_next) { 1239 if (!strncmp(n->tf_name, names[i], 1240 TCP_FUNCTION_NAME_LEN_MAX)) { 1241 TAILQ_REMOVE(&t_functions, n, tf_next); 1242 tcp_fb_cnt--; 1243 n->tf_fb = NULL; 1244 free(n, M_TCPFUNCTIONS); 1245 break; 1246 } 1247 } 1248 } 1249 rw_wunlock(&tcp_function_lock); 1250 return (error); 1251 } 1252 1253 /* 1254 * Register a TCP function block using the name provided in the name 1255 * argument. 1256 * 1257 * Returns 0 on success, or an error code on failure. 1258 */ 1259 int 1260 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1261 int wait) 1262 { 1263 const char *name_list[1]; 1264 int num_names, rv; 1265 1266 num_names = 1; 1267 if (name != NULL) 1268 name_list[0] = name; 1269 else 1270 name_list[0] = blk->tfb_tcp_block_name; 1271 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1272 return (rv); 1273 } 1274 1275 /* 1276 * Register a TCP function block using the name defined in 1277 * blk->tfb_tcp_block_name. 1278 * 1279 * Returns 0 on success, or an error code on failure. 1280 */ 1281 int 1282 register_tcp_functions(struct tcp_function_block *blk, int wait) 1283 { 1284 1285 return (register_tcp_functions_as_name(blk, NULL, wait)); 1286 } 1287 1288 /* 1289 * Deregister all names associated with a function block. This 1290 * functionally removes the function block from use within the system. 1291 * 1292 * When called with a true quiesce argument, mark the function block 1293 * as being removed so no more stacks will use it and determine 1294 * whether the removal would succeed. 1295 * 1296 * When called with a false quiesce argument, actually attempt the 1297 * removal. 1298 * 1299 * When called with a force argument, attempt to switch all TCBs to 1300 * use the default stack instead of returning EBUSY. 1301 * 1302 * Returns 0 on success (or if the removal would succeed, or an error 1303 * code on failure. 1304 */ 1305 int 1306 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1307 bool force) 1308 { 1309 struct tcp_function *f; 1310 1311 if (blk == &tcp_def_funcblk) { 1312 /* You can't un-register the default */ 1313 return (EPERM); 1314 } 1315 rw_wlock(&tcp_function_lock); 1316 if (blk == tcp_func_set_ptr) { 1317 /* You can't free the current default */ 1318 rw_wunlock(&tcp_function_lock); 1319 return (EBUSY); 1320 } 1321 /* Mark the block so no more stacks can use it. */ 1322 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1323 /* 1324 * If TCBs are still attached to the stack, attempt to switch them 1325 * to the default stack. 1326 */ 1327 if (force && blk->tfb_refcnt) { 1328 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1329 INPLOOKUP_WLOCKPCB); 1330 struct inpcb *inp; 1331 struct tcpcb *tp; 1332 VNET_ITERATOR_DECL(vnet_iter); 1333 1334 rw_wunlock(&tcp_function_lock); 1335 1336 VNET_LIST_RLOCK(); 1337 VNET_FOREACH(vnet_iter) { 1338 CURVNET_SET(vnet_iter); 1339 while ((inp = inp_next(&inpi)) != NULL) { 1340 tp = intotcpcb(inp); 1341 if (tp == NULL || tp->t_fb != blk) 1342 continue; 1343 tcp_switch_back_to_default(tp); 1344 } 1345 CURVNET_RESTORE(); 1346 } 1347 VNET_LIST_RUNLOCK(); 1348 1349 rw_wlock(&tcp_function_lock); 1350 } 1351 if (blk->tfb_refcnt) { 1352 /* TCBs still attached. */ 1353 rw_wunlock(&tcp_function_lock); 1354 return (EBUSY); 1355 } 1356 if (quiesce) { 1357 /* Skip removal. */ 1358 rw_wunlock(&tcp_function_lock); 1359 return (0); 1360 } 1361 /* Remove any function names that map to this function block. */ 1362 while (find_tcp_fb_locked(blk, &f) != NULL) { 1363 TAILQ_REMOVE(&t_functions, f, tf_next); 1364 tcp_fb_cnt--; 1365 f->tf_fb = NULL; 1366 free(f, M_TCPFUNCTIONS); 1367 } 1368 rw_wunlock(&tcp_function_lock); 1369 return (0); 1370 } 1371 1372 static void 1373 tcp_drain(void) 1374 { 1375 struct epoch_tracker et; 1376 VNET_ITERATOR_DECL(vnet_iter); 1377 1378 if (!do_tcpdrain) 1379 return; 1380 1381 NET_EPOCH_ENTER(et); 1382 VNET_LIST_RLOCK_NOSLEEP(); 1383 VNET_FOREACH(vnet_iter) { 1384 CURVNET_SET(vnet_iter); 1385 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1386 INPLOOKUP_WLOCKPCB); 1387 struct inpcb *inpb; 1388 struct tcpcb *tcpb; 1389 1390 /* 1391 * Walk the tcpbs, if existing, and flush the reassembly queue, 1392 * if there is one... 1393 * XXX: The "Net/3" implementation doesn't imply that the TCP 1394 * reassembly queue should be flushed, but in a situation 1395 * where we're really low on mbufs, this is potentially 1396 * useful. 1397 */ 1398 while ((inpb = inp_next(&inpi)) != NULL) { 1399 if ((tcpb = intotcpcb(inpb)) != NULL) { 1400 tcp_reass_flush(tcpb); 1401 tcp_clean_sackreport(tcpb); 1402 #ifdef TCP_BLACKBOX 1403 tcp_log_drain(tcpb); 1404 #endif 1405 #ifdef TCPPCAP 1406 if (tcp_pcap_aggressive_free) { 1407 /* Free the TCP PCAP queues. */ 1408 tcp_pcap_drain(&(tcpb->t_inpkts)); 1409 tcp_pcap_drain(&(tcpb->t_outpkts)); 1410 } 1411 #endif 1412 } 1413 } 1414 CURVNET_RESTORE(); 1415 } 1416 VNET_LIST_RUNLOCK_NOSLEEP(); 1417 NET_EPOCH_EXIT(et); 1418 } 1419 1420 static void 1421 tcp_vnet_init(void *arg __unused) 1422 { 1423 1424 #ifdef TCP_HHOOK 1425 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1426 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1427 printf("%s: WARNING: unable to register helper hook\n", __func__); 1428 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1429 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1430 printf("%s: WARNING: unable to register helper hook\n", __func__); 1431 #endif 1432 #ifdef STATS 1433 if (tcp_stats_init()) 1434 printf("%s: WARNING: unable to initialise TCP stats\n", 1435 __func__); 1436 #endif 1437 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1438 tcp_tcbhashsize); 1439 1440 syncache_init(); 1441 tcp_hc_init(); 1442 1443 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1444 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1445 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1446 1447 tcp_fastopen_init(); 1448 1449 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1450 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1451 1452 V_tcp_msl = TCPTV_MSL; 1453 } 1454 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1455 tcp_vnet_init, NULL); 1456 1457 static void 1458 tcp_init(void *arg __unused) 1459 { 1460 const char *tcbhash_tuneable; 1461 int hashsize; 1462 1463 tcp_reass_global_init(); 1464 1465 /* XXX virtualize those below? */ 1466 tcp_delacktime = TCPTV_DELACK; 1467 tcp_keepinit = TCPTV_KEEP_INIT; 1468 tcp_keepidle = TCPTV_KEEP_IDLE; 1469 tcp_keepintvl = TCPTV_KEEPINTVL; 1470 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1471 tcp_rexmit_initial = TCPTV_RTOBASE; 1472 if (tcp_rexmit_initial < 1) 1473 tcp_rexmit_initial = 1; 1474 tcp_rexmit_min = TCPTV_MIN; 1475 if (tcp_rexmit_min < 1) 1476 tcp_rexmit_min = 1; 1477 tcp_persmin = TCPTV_PERSMIN; 1478 tcp_persmax = TCPTV_PERSMAX; 1479 tcp_rexmit_slop = TCPTV_CPU_VAR; 1480 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1481 1482 /* Setup the tcp function block list */ 1483 TAILQ_INIT(&t_functions); 1484 rw_init(&tcp_function_lock, "tcp_func_lock"); 1485 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1486 #ifdef TCP_BLACKBOX 1487 /* Initialize the TCP logging data. */ 1488 tcp_log_init(); 1489 #endif 1490 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1491 1492 if (tcp_soreceive_stream) { 1493 #ifdef INET 1494 tcp_protosw.pr_soreceive = soreceive_stream; 1495 #endif 1496 #ifdef INET6 1497 tcp6_protosw.pr_soreceive = soreceive_stream; 1498 #endif /* INET6 */ 1499 } 1500 1501 #ifdef INET6 1502 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1503 #else /* INET6 */ 1504 max_protohdr_grow(sizeof(struct tcpiphdr)); 1505 #endif /* INET6 */ 1506 1507 ISN_LOCK_INIT(); 1508 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1509 SHUTDOWN_PRI_DEFAULT); 1510 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1511 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1512 1513 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1514 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1515 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1516 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1517 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1518 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1519 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1520 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1521 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1522 #ifdef TCPPCAP 1523 tcp_pcap_init(); 1524 #endif 1525 1526 hashsize = TCBHASHSIZE; 1527 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1528 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1529 if (hashsize == 0) { 1530 /* 1531 * Auto tune the hash size based on maxsockets. 1532 * A perfect hash would have a 1:1 mapping 1533 * (hashsize = maxsockets) however it's been 1534 * suggested that O(2) average is better. 1535 */ 1536 hashsize = maketcp_hashsize(maxsockets / 4); 1537 /* 1538 * Our historical default is 512, 1539 * do not autotune lower than this. 1540 */ 1541 if (hashsize < 512) 1542 hashsize = 512; 1543 if (bootverbose) 1544 printf("%s: %s auto tuned to %d\n", __func__, 1545 tcbhash_tuneable, hashsize); 1546 } 1547 /* 1548 * We require a hashsize to be a power of two. 1549 * Previously if it was not a power of two we would just reset it 1550 * back to 512, which could be a nasty surprise if you did not notice 1551 * the error message. 1552 * Instead what we do is clip it to the closest power of two lower 1553 * than the specified hash value. 1554 */ 1555 if (!powerof2(hashsize)) { 1556 int oldhashsize = hashsize; 1557 1558 hashsize = maketcp_hashsize(hashsize); 1559 /* prevent absurdly low value */ 1560 if (hashsize < 16) 1561 hashsize = 16; 1562 printf("%s: WARNING: TCB hash size not a power of 2, " 1563 "clipped from %d to %d.\n", __func__, oldhashsize, 1564 hashsize); 1565 } 1566 tcp_tcbhashsize = hashsize; 1567 1568 #ifdef INET 1569 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1570 #endif 1571 #ifdef INET6 1572 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1573 #endif 1574 } 1575 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1576 1577 #ifdef VIMAGE 1578 static void 1579 tcp_destroy(void *unused __unused) 1580 { 1581 int n; 1582 #ifdef TCP_HHOOK 1583 int error; 1584 #endif 1585 1586 /* 1587 * All our processes are gone, all our sockets should be cleaned 1588 * up, which means, we should be past the tcp_discardcb() calls. 1589 * Sleep to let all tcpcb timers really disappear and cleanup. 1590 */ 1591 for (;;) { 1592 INP_INFO_WLOCK(&V_tcbinfo); 1593 n = V_tcbinfo.ipi_count; 1594 INP_INFO_WUNLOCK(&V_tcbinfo); 1595 if (n == 0) 1596 break; 1597 pause("tcpdes", hz / 10); 1598 } 1599 tcp_hc_destroy(); 1600 syncache_destroy(); 1601 in_pcbinfo_destroy(&V_tcbinfo); 1602 /* tcp_discardcb() clears the sack_holes up. */ 1603 uma_zdestroy(V_sack_hole_zone); 1604 1605 /* 1606 * Cannot free the zone until all tcpcbs are released as we attach 1607 * the allocations to them. 1608 */ 1609 tcp_fastopen_destroy(); 1610 1611 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1612 VNET_PCPUSTAT_FREE(tcpstat); 1613 1614 #ifdef TCP_HHOOK 1615 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1616 if (error != 0) { 1617 printf("%s: WARNING: unable to deregister helper hook " 1618 "type=%d, id=%d: error %d returned\n", __func__, 1619 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1620 } 1621 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1622 if (error != 0) { 1623 printf("%s: WARNING: unable to deregister helper hook " 1624 "type=%d, id=%d: error %d returned\n", __func__, 1625 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1626 } 1627 #endif 1628 } 1629 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1630 #endif 1631 1632 void 1633 tcp_fini(void *xtp) 1634 { 1635 1636 } 1637 1638 /* 1639 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1640 * tcp_template used to store this data in mbufs, but we now recopy it out 1641 * of the tcpcb each time to conserve mbufs. 1642 */ 1643 void 1644 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1645 { 1646 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1647 1648 INP_WLOCK_ASSERT(inp); 1649 1650 #ifdef INET6 1651 if ((inp->inp_vflag & INP_IPV6) != 0) { 1652 struct ip6_hdr *ip6; 1653 1654 ip6 = (struct ip6_hdr *)ip_ptr; 1655 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1656 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1657 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1658 (IPV6_VERSION & IPV6_VERSION_MASK); 1659 if (port == 0) 1660 ip6->ip6_nxt = IPPROTO_TCP; 1661 else 1662 ip6->ip6_nxt = IPPROTO_UDP; 1663 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1664 ip6->ip6_src = inp->in6p_laddr; 1665 ip6->ip6_dst = inp->in6p_faddr; 1666 } 1667 #endif /* INET6 */ 1668 #if defined(INET6) && defined(INET) 1669 else 1670 #endif 1671 #ifdef INET 1672 { 1673 struct ip *ip; 1674 1675 ip = (struct ip *)ip_ptr; 1676 ip->ip_v = IPVERSION; 1677 ip->ip_hl = 5; 1678 ip->ip_tos = inp->inp_ip_tos; 1679 ip->ip_len = 0; 1680 ip->ip_id = 0; 1681 ip->ip_off = 0; 1682 ip->ip_ttl = inp->inp_ip_ttl; 1683 ip->ip_sum = 0; 1684 if (port == 0) 1685 ip->ip_p = IPPROTO_TCP; 1686 else 1687 ip->ip_p = IPPROTO_UDP; 1688 ip->ip_src = inp->inp_laddr; 1689 ip->ip_dst = inp->inp_faddr; 1690 } 1691 #endif /* INET */ 1692 th->th_sport = inp->inp_lport; 1693 th->th_dport = inp->inp_fport; 1694 th->th_seq = 0; 1695 th->th_ack = 0; 1696 th->th_off = 5; 1697 tcp_set_flags(th, 0); 1698 th->th_win = 0; 1699 th->th_urp = 0; 1700 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1701 } 1702 1703 /* 1704 * Create template to be used to send tcp packets on a connection. 1705 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1706 * use for this function is in keepalives, which use tcp_respond. 1707 */ 1708 struct tcptemp * 1709 tcpip_maketemplate(struct inpcb *inp) 1710 { 1711 struct tcptemp *t; 1712 1713 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1714 if (t == NULL) 1715 return (NULL); 1716 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1717 return (t); 1718 } 1719 1720 /* 1721 * Send a single message to the TCP at address specified by 1722 * the given TCP/IP header. If m == NULL, then we make a copy 1723 * of the tcpiphdr at th and send directly to the addressed host. 1724 * This is used to force keep alive messages out using the TCP 1725 * template for a connection. If flags are given then we send 1726 * a message back to the TCP which originated the segment th, 1727 * and discard the mbuf containing it and any other attached mbufs. 1728 * 1729 * In any case the ack and sequence number of the transmitted 1730 * segment are as specified by the parameters. 1731 * 1732 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1733 */ 1734 void 1735 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1736 tcp_seq ack, tcp_seq seq, uint16_t flags) 1737 { 1738 struct tcpopt to; 1739 struct inpcb *inp; 1740 struct ip *ip; 1741 struct mbuf *optm; 1742 struct udphdr *uh = NULL; 1743 struct tcphdr *nth; 1744 struct tcp_log_buffer *lgb; 1745 u_char *optp; 1746 #ifdef INET6 1747 struct ip6_hdr *ip6; 1748 int isipv6; 1749 #endif /* INET6 */ 1750 int optlen, tlen, win, ulen; 1751 int ect = 0; 1752 bool incl_opts; 1753 uint16_t port; 1754 int output_ret; 1755 #ifdef INVARIANTS 1756 int thflags = tcp_get_flags(th); 1757 #endif 1758 1759 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1760 NET_EPOCH_ASSERT(); 1761 1762 #ifdef INET6 1763 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1764 ip6 = ipgen; 1765 #endif /* INET6 */ 1766 ip = ipgen; 1767 1768 if (tp != NULL) { 1769 inp = tptoinpcb(tp); 1770 INP_LOCK_ASSERT(inp); 1771 } else 1772 inp = NULL; 1773 1774 if (m != NULL) { 1775 #ifdef INET6 1776 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1777 port = m->m_pkthdr.tcp_tun_port; 1778 else 1779 #endif 1780 if (ip && (ip->ip_p == IPPROTO_UDP)) 1781 port = m->m_pkthdr.tcp_tun_port; 1782 else 1783 port = 0; 1784 } else 1785 port = tp->t_port; 1786 1787 incl_opts = false; 1788 win = 0; 1789 if (tp != NULL) { 1790 if (!(flags & TH_RST)) { 1791 win = sbspace(&inp->inp_socket->so_rcv); 1792 if (win > TCP_MAXWIN << tp->rcv_scale) 1793 win = TCP_MAXWIN << tp->rcv_scale; 1794 } 1795 if ((tp->t_flags & TF_NOOPT) == 0) 1796 incl_opts = true; 1797 } 1798 if (m == NULL) { 1799 m = m_gethdr(M_NOWAIT, MT_DATA); 1800 if (m == NULL) 1801 return; 1802 m->m_data += max_linkhdr; 1803 #ifdef INET6 1804 if (isipv6) { 1805 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1806 sizeof(struct ip6_hdr)); 1807 ip6 = mtod(m, struct ip6_hdr *); 1808 nth = (struct tcphdr *)(ip6 + 1); 1809 if (port) { 1810 /* Insert a UDP header */ 1811 uh = (struct udphdr *)nth; 1812 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1813 uh->uh_dport = port; 1814 nth = (struct tcphdr *)(uh + 1); 1815 } 1816 } else 1817 #endif /* INET6 */ 1818 { 1819 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1820 ip = mtod(m, struct ip *); 1821 nth = (struct tcphdr *)(ip + 1); 1822 if (port) { 1823 /* Insert a UDP header */ 1824 uh = (struct udphdr *)nth; 1825 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1826 uh->uh_dport = port; 1827 nth = (struct tcphdr *)(uh + 1); 1828 } 1829 } 1830 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1831 flags = TH_ACK; 1832 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1833 struct mbuf *n; 1834 1835 /* Can't reuse 'm', allocate a new mbuf. */ 1836 n = m_gethdr(M_NOWAIT, MT_DATA); 1837 if (n == NULL) { 1838 m_freem(m); 1839 return; 1840 } 1841 1842 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1843 m_freem(m); 1844 m_freem(n); 1845 return; 1846 } 1847 1848 n->m_data += max_linkhdr; 1849 /* m_len is set later */ 1850 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1851 #ifdef INET6 1852 if (isipv6) { 1853 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1854 sizeof(struct ip6_hdr)); 1855 ip6 = mtod(n, struct ip6_hdr *); 1856 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1857 nth = (struct tcphdr *)(ip6 + 1); 1858 if (port) { 1859 /* Insert a UDP header */ 1860 uh = (struct udphdr *)nth; 1861 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1862 uh->uh_dport = port; 1863 nth = (struct tcphdr *)(uh + 1); 1864 } 1865 } else 1866 #endif /* INET6 */ 1867 { 1868 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1869 ip = mtod(n, struct ip *); 1870 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1871 nth = (struct tcphdr *)(ip + 1); 1872 if (port) { 1873 /* Insert a UDP header */ 1874 uh = (struct udphdr *)nth; 1875 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1876 uh->uh_dport = port; 1877 nth = (struct tcphdr *)(uh + 1); 1878 } 1879 } 1880 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1881 xchg(nth->th_dport, nth->th_sport, uint16_t); 1882 th = nth; 1883 m_freem(m); 1884 m = n; 1885 } else { 1886 /* 1887 * reuse the mbuf. 1888 * XXX MRT We inherit the FIB, which is lucky. 1889 */ 1890 m_freem(m->m_next); 1891 m->m_next = NULL; 1892 m->m_data = (caddr_t)ipgen; 1893 /* m_len is set later */ 1894 #ifdef INET6 1895 if (isipv6) { 1896 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1897 nth = (struct tcphdr *)(ip6 + 1); 1898 } else 1899 #endif /* INET6 */ 1900 { 1901 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1902 nth = (struct tcphdr *)(ip + 1); 1903 } 1904 if (th != nth) { 1905 /* 1906 * this is usually a case when an extension header 1907 * exists between the IPv6 header and the 1908 * TCP header. 1909 */ 1910 nth->th_sport = th->th_sport; 1911 nth->th_dport = th->th_dport; 1912 } 1913 xchg(nth->th_dport, nth->th_sport, uint16_t); 1914 #undef xchg 1915 } 1916 tlen = 0; 1917 #ifdef INET6 1918 if (isipv6) 1919 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1920 #endif 1921 #if defined(INET) && defined(INET6) 1922 else 1923 #endif 1924 #ifdef INET 1925 tlen = sizeof (struct tcpiphdr); 1926 #endif 1927 if (port) 1928 tlen += sizeof (struct udphdr); 1929 #ifdef INVARIANTS 1930 m->m_len = 0; 1931 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1932 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1933 m, tlen, (long)M_TRAILINGSPACE(m))); 1934 #endif 1935 m->m_len = tlen; 1936 to.to_flags = 0; 1937 if (incl_opts) { 1938 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1939 /* Make sure we have room. */ 1940 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1941 m->m_next = m_get(M_NOWAIT, MT_DATA); 1942 if (m->m_next) { 1943 optp = mtod(m->m_next, u_char *); 1944 optm = m->m_next; 1945 } else 1946 incl_opts = false; 1947 } else { 1948 optp = (u_char *) (nth + 1); 1949 optm = m; 1950 } 1951 } 1952 if (incl_opts) { 1953 /* Timestamps. */ 1954 if (tp->t_flags & TF_RCVD_TSTMP) { 1955 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1956 to.to_tsecr = tp->ts_recent; 1957 to.to_flags |= TOF_TS; 1958 } 1959 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1960 /* TCP-MD5 (RFC2385). */ 1961 if (tp->t_flags & TF_SIGNATURE) 1962 to.to_flags |= TOF_SIGNATURE; 1963 #endif 1964 /* Add the options. */ 1965 tlen += optlen = tcp_addoptions(&to, optp); 1966 1967 /* Update m_len in the correct mbuf. */ 1968 optm->m_len += optlen; 1969 } else 1970 optlen = 0; 1971 #ifdef INET6 1972 if (isipv6) { 1973 if (uh) { 1974 ulen = tlen - sizeof(struct ip6_hdr); 1975 uh->uh_ulen = htons(ulen); 1976 } 1977 ip6->ip6_flow = htonl(ect << 20); 1978 ip6->ip6_vfc = IPV6_VERSION; 1979 if (port) 1980 ip6->ip6_nxt = IPPROTO_UDP; 1981 else 1982 ip6->ip6_nxt = IPPROTO_TCP; 1983 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1984 } 1985 #endif 1986 #if defined(INET) && defined(INET6) 1987 else 1988 #endif 1989 #ifdef INET 1990 { 1991 if (uh) { 1992 ulen = tlen - sizeof(struct ip); 1993 uh->uh_ulen = htons(ulen); 1994 } 1995 ip->ip_tos = ect; 1996 ip->ip_len = htons(tlen); 1997 ip->ip_ttl = V_ip_defttl; 1998 if (port) { 1999 ip->ip_p = IPPROTO_UDP; 2000 } else { 2001 ip->ip_p = IPPROTO_TCP; 2002 } 2003 if (V_path_mtu_discovery) 2004 ip->ip_off |= htons(IP_DF); 2005 } 2006 #endif 2007 m->m_pkthdr.len = tlen; 2008 m->m_pkthdr.rcvif = NULL; 2009 #ifdef MAC 2010 if (inp != NULL) { 2011 /* 2012 * Packet is associated with a socket, so allow the 2013 * label of the response to reflect the socket label. 2014 */ 2015 INP_LOCK_ASSERT(inp); 2016 mac_inpcb_create_mbuf(inp, m); 2017 } else { 2018 /* 2019 * Packet is not associated with a socket, so possibly 2020 * update the label in place. 2021 */ 2022 mac_netinet_tcp_reply(m); 2023 } 2024 #endif 2025 nth->th_seq = htonl(seq); 2026 nth->th_ack = htonl(ack); 2027 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2028 tcp_set_flags(nth, flags); 2029 if (tp != NULL) 2030 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2031 else 2032 nth->th_win = htons((u_short)win); 2033 nth->th_urp = 0; 2034 2035 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2036 if (to.to_flags & TOF_SIGNATURE) { 2037 if (!TCPMD5_ENABLED() || 2038 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2039 m_freem(m); 2040 return; 2041 } 2042 } 2043 #endif 2044 2045 #ifdef INET6 2046 if (isipv6) { 2047 if (port) { 2048 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2049 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2050 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2051 nth->th_sum = 0; 2052 } else { 2053 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2054 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2055 nth->th_sum = in6_cksum_pseudo(ip6, 2056 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2057 } 2058 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2059 } 2060 #endif /* INET6 */ 2061 #if defined(INET6) && defined(INET) 2062 else 2063 #endif 2064 #ifdef INET 2065 { 2066 if (port) { 2067 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2068 htons(ulen + IPPROTO_UDP)); 2069 m->m_pkthdr.csum_flags = CSUM_UDP; 2070 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2071 nth->th_sum = 0; 2072 } else { 2073 m->m_pkthdr.csum_flags = CSUM_TCP; 2074 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2075 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2076 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2077 } 2078 } 2079 #endif /* INET */ 2080 #ifdef TCPDEBUG 2081 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2082 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2083 #endif 2084 TCP_PROBE3(debug__output, tp, th, m); 2085 if (flags & TH_RST) 2086 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2087 lgb = NULL; 2088 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { 2089 if (INP_WLOCKED(inp)) { 2090 union tcp_log_stackspecific log; 2091 struct timeval tv; 2092 2093 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2094 log.u_bbr.inhpts = inp->inp_in_hpts; 2095 log.u_bbr.flex8 = 4; 2096 log.u_bbr.pkts_out = tp->t_maxseg; 2097 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2098 log.u_bbr.delivered = 0; 2099 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, 2100 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2101 } else { 2102 /* 2103 * We can not log the packet, since we only own the 2104 * read lock, but a write lock is needed. The read lock 2105 * is not upgraded to a write lock, since only getting 2106 * the read lock was done intentionally to improve the 2107 * handling of SYN flooding attacks. 2108 * This happens only for pure SYN segments received in 2109 * the initial CLOSED state, or received in a more 2110 * advanced state than listen and the UDP encapsulation 2111 * port is unexpected. 2112 * The incoming SYN segments do not really belong to 2113 * the TCP connection and the handling does not change 2114 * the state of the TCP connection. Therefore, the 2115 * sending of the RST segments is not logged. Please 2116 * note that also the incoming SYN segments are not 2117 * logged. 2118 * 2119 * The following code ensures that the above description 2120 * is and stays correct. 2121 */ 2122 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2123 (tp->t_state == TCPS_CLOSED || 2124 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2125 ("%s: Logging of TCP segment with flags 0x%b and " 2126 "UDP encapsulation port %u skipped in state %s", 2127 __func__, thflags, PRINT_TH_FLAGS, 2128 ntohs(port), tcpstates[tp->t_state])); 2129 } 2130 } 2131 2132 if (flags & TH_ACK) 2133 TCPSTAT_INC(tcps_sndacks); 2134 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2135 TCPSTAT_INC(tcps_sndctrl); 2136 TCPSTAT_INC(tcps_sndtotal); 2137 2138 #ifdef INET6 2139 if (isipv6) { 2140 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2141 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2142 } 2143 #endif /* INET6 */ 2144 #if defined(INET) && defined(INET6) 2145 else 2146 #endif 2147 #ifdef INET 2148 { 2149 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2150 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2151 } 2152 #endif 2153 if (lgb != NULL) 2154 lgb->tlb_errno = output_ret; 2155 } 2156 2157 /* 2158 * Create a new TCP control block, making an empty reassembly queue and hooking 2159 * it to the argument protocol control block. The `inp' parameter must have 2160 * come from the zone allocator set up by tcpcbstor declaration. 2161 */ 2162 struct tcpcb * 2163 tcp_newtcpcb(struct inpcb *inp) 2164 { 2165 struct tcpcb *tp = intotcpcb(inp); 2166 #ifdef INET6 2167 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2168 #endif /* INET6 */ 2169 2170 /* 2171 * Historically allocation was done with M_ZERO. There is a lot of 2172 * code that rely on that. For now take safe approach and zero whole 2173 * tcpcb. This definitely can be optimized. 2174 */ 2175 bzero(&tp->t_start_zero, t_zero_size); 2176 2177 /* Initialise cc_var struct for this tcpcb. */ 2178 tp->t_ccv.type = IPPROTO_TCP; 2179 tp->t_ccv.ccvc.tcp = tp; 2180 rw_rlock(&tcp_function_lock); 2181 tp->t_fb = tcp_func_set_ptr; 2182 refcount_acquire(&tp->t_fb->tfb_refcnt); 2183 rw_runlock(&tcp_function_lock); 2184 /* 2185 * Use the current system default CC algorithm. 2186 */ 2187 cc_attach(tp, CC_DEFAULT_ALGO()); 2188 2189 if (CC_ALGO(tp)->cb_init != NULL) 2190 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2191 cc_detach(tp); 2192 if (tp->t_fb->tfb_tcp_fb_fini) 2193 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2194 refcount_release(&tp->t_fb->tfb_refcnt); 2195 return (NULL); 2196 } 2197 2198 #ifdef TCP_HHOOK 2199 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2200 if (tp->t_fb->tfb_tcp_fb_fini) 2201 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2202 refcount_release(&tp->t_fb->tfb_refcnt); 2203 return (NULL); 2204 } 2205 #endif 2206 2207 TAILQ_INIT(&tp->t_segq); 2208 tp->t_maxseg = 2209 #ifdef INET6 2210 isipv6 ? V_tcp_v6mssdflt : 2211 #endif /* INET6 */ 2212 V_tcp_mssdflt; 2213 2214 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED); 2215 for (int i = 0; i < TT_N; i++) 2216 tp->t_timers[i] = SBT_MAX; 2217 2218 switch (V_tcp_do_rfc1323) { 2219 case 0: 2220 break; 2221 default: 2222 case 1: 2223 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2224 break; 2225 case 2: 2226 tp->t_flags = TF_REQ_SCALE; 2227 break; 2228 case 3: 2229 tp->t_flags = TF_REQ_TSTMP; 2230 break; 2231 } 2232 if (V_tcp_do_sack) 2233 tp->t_flags |= TF_SACK_PERMIT; 2234 TAILQ_INIT(&tp->snd_holes); 2235 2236 /* 2237 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2238 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2239 * reasonable initial retransmit time. 2240 */ 2241 tp->t_srtt = TCPTV_SRTTBASE; 2242 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2243 tp->t_rttmin = tcp_rexmit_min; 2244 tp->t_rxtcur = tcp_rexmit_initial; 2245 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2246 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2247 tp->t_rcvtime = ticks; 2248 /* 2249 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2250 * because the socket may be bound to an IPv6 wildcard address, 2251 * which may match an IPv4-mapped IPv6 address. 2252 */ 2253 inp->inp_ip_ttl = V_ip_defttl; 2254 #ifdef TCPHPTS 2255 /* 2256 * If using hpts lets drop a random number in so 2257 * not all new connections fall on the same CPU. 2258 */ 2259 inp->inp_hpts_cpu = hpts_random_cpu(inp); 2260 #endif 2261 #ifdef TCPPCAP 2262 /* 2263 * Init the TCP PCAP queues. 2264 */ 2265 tcp_pcap_tcpcb_init(tp); 2266 #endif 2267 #ifdef TCP_BLACKBOX 2268 /* Initialize the per-TCPCB log data. */ 2269 tcp_log_tcpcbinit(tp); 2270 #endif 2271 tp->t_pacing_rate = -1; 2272 if (tp->t_fb->tfb_tcp_fb_init) { 2273 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2274 refcount_release(&tp->t_fb->tfb_refcnt); 2275 return (NULL); 2276 } 2277 } 2278 #ifdef STATS 2279 if (V_tcp_perconn_stats_enable == 1) 2280 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2281 #endif 2282 if (V_tcp_do_lrd) 2283 tp->t_flags |= TF_LRD; 2284 2285 return (tp); 2286 } 2287 2288 /* 2289 * Drop a TCP connection, reporting 2290 * the specified error. If connection is synchronized, 2291 * then send a RST to peer. 2292 */ 2293 struct tcpcb * 2294 tcp_drop(struct tcpcb *tp, int errno) 2295 { 2296 struct socket *so = tptosocket(tp); 2297 2298 NET_EPOCH_ASSERT(); 2299 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2300 2301 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2302 tcp_state_change(tp, TCPS_CLOSED); 2303 /* Don't use tcp_output() here due to possible recursion. */ 2304 (void)tcp_output_nodrop(tp); 2305 TCPSTAT_INC(tcps_drops); 2306 } else 2307 TCPSTAT_INC(tcps_conndrops); 2308 if (errno == ETIMEDOUT && tp->t_softerror) 2309 errno = tp->t_softerror; 2310 so->so_error = errno; 2311 return (tcp_close(tp)); 2312 } 2313 2314 void 2315 tcp_discardcb(struct tcpcb *tp) 2316 { 2317 struct inpcb *inp = tptoinpcb(tp); 2318 struct socket *so = tptosocket(tp); 2319 #ifdef INET6 2320 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2321 #endif 2322 2323 INP_WLOCK_ASSERT(inp); 2324 2325 tcp_timer_stop(tp); 2326 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2327 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2328 } 2329 2330 /* free the reassembly queue, if any */ 2331 tcp_reass_flush(tp); 2332 2333 #ifdef TCP_OFFLOAD 2334 /* Disconnect offload device, if any. */ 2335 if (tp->t_flags & TF_TOE) 2336 tcp_offload_detach(tp); 2337 #endif 2338 2339 tcp_free_sackholes(tp); 2340 2341 #ifdef TCPPCAP 2342 /* Free the TCP PCAP queues. */ 2343 tcp_pcap_drain(&(tp->t_inpkts)); 2344 tcp_pcap_drain(&(tp->t_outpkts)); 2345 #endif 2346 2347 /* Allow the CC algorithm to clean up after itself. */ 2348 if (CC_ALGO(tp)->cb_destroy != NULL) 2349 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2350 CC_DATA(tp) = NULL; 2351 /* Detach from the CC algorithm */ 2352 cc_detach(tp); 2353 2354 #ifdef TCP_HHOOK 2355 khelp_destroy_osd(&tp->t_osd); 2356 #endif 2357 #ifdef STATS 2358 stats_blob_destroy(tp->t_stats); 2359 #endif 2360 2361 CC_ALGO(tp) = NULL; 2362 2363 #ifdef TCP_BLACKBOX 2364 tcp_log_tcpcbfini(tp); 2365 #endif 2366 TCPSTATES_DEC(tp->t_state); 2367 if (tp->t_fb->tfb_tcp_fb_fini) 2368 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2369 2370 /* 2371 * If we got enough samples through the srtt filter, 2372 * save the rtt and rttvar in the routing entry. 2373 * 'Enough' is arbitrarily defined as 4 rtt samples. 2374 * 4 samples is enough for the srtt filter to converge 2375 * to within enough % of the correct value; fewer samples 2376 * and we could save a bogus rtt. The danger is not high 2377 * as tcp quickly recovers from everything. 2378 * XXX: Works very well but needs some more statistics! 2379 * 2380 * XXXRRS: Updating must be after the stack fini() since 2381 * that may be converting some internal representation of 2382 * say srtt etc into the general one used by other stacks. 2383 * Lets also at least protect against the so being NULL 2384 * as RW stated below. 2385 */ 2386 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2387 struct hc_metrics_lite metrics; 2388 uint32_t ssthresh; 2389 2390 bzero(&metrics, sizeof(metrics)); 2391 /* 2392 * Update the ssthresh always when the conditions below 2393 * are satisfied. This gives us better new start value 2394 * for the congestion avoidance for new connections. 2395 * ssthresh is only set if packet loss occurred on a session. 2396 * 2397 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2398 * being torn down. Ideally this code would not use 'so'. 2399 */ 2400 ssthresh = tp->snd_ssthresh; 2401 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2402 /* 2403 * convert the limit from user data bytes to 2404 * packets then to packet data bytes. 2405 */ 2406 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2407 if (ssthresh < 2) 2408 ssthresh = 2; 2409 ssthresh *= (tp->t_maxseg + 2410 #ifdef INET6 2411 (isipv6 ? sizeof (struct ip6_hdr) + 2412 sizeof (struct tcphdr) : 2413 #endif 2414 sizeof (struct tcpiphdr) 2415 #ifdef INET6 2416 ) 2417 #endif 2418 ); 2419 } else 2420 ssthresh = 0; 2421 metrics.rmx_ssthresh = ssthresh; 2422 2423 metrics.rmx_rtt = tp->t_srtt; 2424 metrics.rmx_rttvar = tp->t_rttvar; 2425 metrics.rmx_cwnd = tp->snd_cwnd; 2426 metrics.rmx_sendpipe = 0; 2427 metrics.rmx_recvpipe = 0; 2428 2429 tcp_hc_update(&inp->inp_inc, &metrics); 2430 } 2431 2432 refcount_release(&tp->t_fb->tfb_refcnt); 2433 } 2434 2435 /* 2436 * Attempt to close a TCP control block, marking it as dropped, and freeing 2437 * the socket if we hold the only reference. 2438 */ 2439 struct tcpcb * 2440 tcp_close(struct tcpcb *tp) 2441 { 2442 struct inpcb *inp = tptoinpcb(tp); 2443 struct socket *so = tptosocket(tp); 2444 2445 INP_WLOCK_ASSERT(inp); 2446 2447 #ifdef TCP_OFFLOAD 2448 if (tp->t_state == TCPS_LISTEN) 2449 tcp_offload_listen_stop(tp); 2450 #endif 2451 /* 2452 * This releases the TFO pending counter resource for TFO listen 2453 * sockets as well as passively-created TFO sockets that transition 2454 * from SYN_RECEIVED to CLOSED. 2455 */ 2456 if (tp->t_tfo_pending) { 2457 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2458 tp->t_tfo_pending = NULL; 2459 } 2460 #ifdef TCPHPTS 2461 tcp_hpts_remove(inp); 2462 #endif 2463 in_pcbdrop(inp); 2464 TCPSTAT_INC(tcps_closed); 2465 if (tp->t_state != TCPS_CLOSED) 2466 tcp_state_change(tp, TCPS_CLOSED); 2467 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2468 soisdisconnected(so); 2469 if (inp->inp_flags & INP_SOCKREF) { 2470 inp->inp_flags &= ~INP_SOCKREF; 2471 INP_WUNLOCK(inp); 2472 sorele(so); 2473 return (NULL); 2474 } 2475 return (tp); 2476 } 2477 2478 /* 2479 * Notify a tcp user of an asynchronous error; 2480 * store error as soft error, but wake up user 2481 * (for now, won't do anything until can select for soft error). 2482 * 2483 * Do not wake up user since there currently is no mechanism for 2484 * reporting soft errors (yet - a kqueue filter may be added). 2485 */ 2486 static struct inpcb * 2487 tcp_notify(struct inpcb *inp, int error) 2488 { 2489 struct tcpcb *tp; 2490 2491 INP_WLOCK_ASSERT(inp); 2492 2493 tp = intotcpcb(inp); 2494 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2495 2496 /* 2497 * Ignore some errors if we are hooked up. 2498 * If connection hasn't completed, has retransmitted several times, 2499 * and receives a second error, give up now. This is better 2500 * than waiting a long time to establish a connection that 2501 * can never complete. 2502 */ 2503 if (tp->t_state == TCPS_ESTABLISHED && 2504 (error == EHOSTUNREACH || error == ENETUNREACH || 2505 error == EHOSTDOWN)) { 2506 if (inp->inp_route.ro_nh) { 2507 NH_FREE(inp->inp_route.ro_nh); 2508 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2509 } 2510 return (inp); 2511 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2512 tp->t_softerror) { 2513 tp = tcp_drop(tp, error); 2514 if (tp != NULL) 2515 return (inp); 2516 else 2517 return (NULL); 2518 } else { 2519 tp->t_softerror = error; 2520 return (inp); 2521 } 2522 #if 0 2523 wakeup( &so->so_timeo); 2524 sorwakeup(so); 2525 sowwakeup(so); 2526 #endif 2527 } 2528 2529 static int 2530 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2531 { 2532 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2533 INPLOOKUP_RLOCKPCB); 2534 struct xinpgen xig; 2535 struct inpcb *inp; 2536 int error; 2537 2538 if (req->newptr != NULL) 2539 return (EPERM); 2540 2541 if (req->oldptr == NULL) { 2542 int n; 2543 2544 n = V_tcbinfo.ipi_count + 2545 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2546 n += imax(n / 8, 10); 2547 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2548 return (0); 2549 } 2550 2551 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2552 return (error); 2553 2554 bzero(&xig, sizeof(xig)); 2555 xig.xig_len = sizeof xig; 2556 xig.xig_count = V_tcbinfo.ipi_count + 2557 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2558 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2559 xig.xig_sogen = so_gencnt; 2560 error = SYSCTL_OUT(req, &xig, sizeof xig); 2561 if (error) 2562 return (error); 2563 2564 error = syncache_pcblist(req); 2565 if (error) 2566 return (error); 2567 2568 while ((inp = inp_next(&inpi)) != NULL) { 2569 if (inp->inp_gencnt <= xig.xig_gen && 2570 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2571 struct xtcpcb xt; 2572 2573 tcp_inptoxtp(inp, &xt); 2574 error = SYSCTL_OUT(req, &xt, sizeof xt); 2575 if (error) { 2576 INP_RUNLOCK(inp); 2577 break; 2578 } else 2579 continue; 2580 } 2581 } 2582 2583 if (!error) { 2584 /* 2585 * Give the user an updated idea of our state. 2586 * If the generation differs from what we told 2587 * her before, she knows that something happened 2588 * while we were processing this request, and it 2589 * might be necessary to retry. 2590 */ 2591 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2592 xig.xig_sogen = so_gencnt; 2593 xig.xig_count = V_tcbinfo.ipi_count + 2594 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2595 error = SYSCTL_OUT(req, &xig, sizeof xig); 2596 } 2597 2598 return (error); 2599 } 2600 2601 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2602 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2603 NULL, 0, tcp_pcblist, "S,xtcpcb", 2604 "List of active TCP connections"); 2605 2606 #ifdef INET 2607 static int 2608 tcp_getcred(SYSCTL_HANDLER_ARGS) 2609 { 2610 struct xucred xuc; 2611 struct sockaddr_in addrs[2]; 2612 struct epoch_tracker et; 2613 struct inpcb *inp; 2614 int error; 2615 2616 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2617 if (error) 2618 return (error); 2619 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2620 if (error) 2621 return (error); 2622 NET_EPOCH_ENTER(et); 2623 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2624 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2625 NET_EPOCH_EXIT(et); 2626 if (inp != NULL) { 2627 if (error == 0) 2628 error = cr_canseeinpcb(req->td->td_ucred, inp); 2629 if (error == 0) 2630 cru2x(inp->inp_cred, &xuc); 2631 INP_RUNLOCK(inp); 2632 } else 2633 error = ENOENT; 2634 if (error == 0) 2635 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2636 return (error); 2637 } 2638 2639 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2640 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2641 0, 0, tcp_getcred, "S,xucred", 2642 "Get the xucred of a TCP connection"); 2643 #endif /* INET */ 2644 2645 #ifdef INET6 2646 static int 2647 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2648 { 2649 struct epoch_tracker et; 2650 struct xucred xuc; 2651 struct sockaddr_in6 addrs[2]; 2652 struct inpcb *inp; 2653 int error; 2654 #ifdef INET 2655 int mapped = 0; 2656 #endif 2657 2658 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2659 if (error) 2660 return (error); 2661 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2662 if (error) 2663 return (error); 2664 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2665 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2666 return (error); 2667 } 2668 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2669 #ifdef INET 2670 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2671 mapped = 1; 2672 else 2673 #endif 2674 return (EINVAL); 2675 } 2676 2677 NET_EPOCH_ENTER(et); 2678 #ifdef INET 2679 if (mapped == 1) 2680 inp = in_pcblookup(&V_tcbinfo, 2681 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2682 addrs[1].sin6_port, 2683 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2684 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2685 else 2686 #endif 2687 inp = in6_pcblookup(&V_tcbinfo, 2688 &addrs[1].sin6_addr, addrs[1].sin6_port, 2689 &addrs[0].sin6_addr, addrs[0].sin6_port, 2690 INPLOOKUP_RLOCKPCB, NULL); 2691 NET_EPOCH_EXIT(et); 2692 if (inp != NULL) { 2693 if (error == 0) 2694 error = cr_canseeinpcb(req->td->td_ucred, inp); 2695 if (error == 0) 2696 cru2x(inp->inp_cred, &xuc); 2697 INP_RUNLOCK(inp); 2698 } else 2699 error = ENOENT; 2700 if (error == 0) 2701 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2702 return (error); 2703 } 2704 2705 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2706 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2707 0, 0, tcp6_getcred, "S,xucred", 2708 "Get the xucred of a TCP6 connection"); 2709 #endif /* INET6 */ 2710 2711 #ifdef INET 2712 /* Path MTU to try next when a fragmentation-needed message is received. */ 2713 static inline int 2714 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2715 { 2716 int mtu = ntohs(icp->icmp_nextmtu); 2717 2718 /* If no alternative MTU was proposed, try the next smaller one. */ 2719 if (!mtu) 2720 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2721 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2722 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2723 2724 return (mtu); 2725 } 2726 2727 static void 2728 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2729 { 2730 struct ip *ip; 2731 struct tcphdr *th; 2732 struct inpcb *inp; 2733 struct tcpcb *tp; 2734 struct inpcb *(*notify)(struct inpcb *, int); 2735 struct in_conninfo inc; 2736 tcp_seq icmp_tcp_seq; 2737 int errno, mtu; 2738 2739 errno = icmp_errmap(icp); 2740 switch (errno) { 2741 case 0: 2742 return; 2743 case EMSGSIZE: 2744 notify = tcp_mtudisc_notify; 2745 break; 2746 case ECONNREFUSED: 2747 if (V_icmp_may_rst) 2748 notify = tcp_drop_syn_sent; 2749 else 2750 notify = tcp_notify; 2751 break; 2752 case EHOSTUNREACH: 2753 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2754 notify = tcp_drop_syn_sent; 2755 else 2756 notify = tcp_notify; 2757 break; 2758 default: 2759 notify = tcp_notify; 2760 } 2761 2762 ip = &icp->icmp_ip; 2763 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2764 icmp_tcp_seq = th->th_seq; 2765 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2766 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2767 if (inp != NULL) { 2768 tp = intotcpcb(inp); 2769 #ifdef TCP_OFFLOAD 2770 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2771 /* 2772 * MTU discovery for offloaded connections. Let 2773 * the TOE driver verify seq# and process it. 2774 */ 2775 mtu = tcp_next_pmtu(icp, ip); 2776 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2777 goto out; 2778 } 2779 #endif 2780 if (tp->t_port != port) 2781 goto out; 2782 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2783 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2784 if (errno == EMSGSIZE) { 2785 /* 2786 * MTU discovery: we got a needfrag and 2787 * will potentially try a lower MTU. 2788 */ 2789 mtu = tcp_next_pmtu(icp, ip); 2790 2791 /* 2792 * Only process the offered MTU if it 2793 * is smaller than the current one. 2794 */ 2795 if (mtu < tp->t_maxseg + 2796 sizeof(struct tcpiphdr)) { 2797 bzero(&inc, sizeof(inc)); 2798 inc.inc_faddr = ip->ip_dst; 2799 inc.inc_fibnum = 2800 inp->inp_inc.inc_fibnum; 2801 tcp_hc_updatemtu(&inc, mtu); 2802 inp = tcp_mtudisc(inp, mtu); 2803 } 2804 } else 2805 inp = (*notify)(inp, errno); 2806 } 2807 } else { 2808 bzero(&inc, sizeof(inc)); 2809 inc.inc_fport = th->th_dport; 2810 inc.inc_lport = th->th_sport; 2811 inc.inc_faddr = ip->ip_dst; 2812 inc.inc_laddr = ip->ip_src; 2813 syncache_unreach(&inc, icmp_tcp_seq, port); 2814 } 2815 out: 2816 if (inp != NULL) 2817 INP_WUNLOCK(inp); 2818 } 2819 2820 static void 2821 tcp_ctlinput(struct icmp *icmp) 2822 { 2823 tcp_ctlinput_with_port(icmp, htons(0)); 2824 } 2825 2826 static void 2827 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2828 { 2829 /* Its a tunneled TCP over UDP icmp */ 2830 struct icmp *icmp = param.icmp; 2831 struct ip *outer_ip, *inner_ip; 2832 struct udphdr *udp; 2833 struct tcphdr *th, ttemp; 2834 int i_hlen, o_len; 2835 uint16_t port; 2836 2837 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2838 inner_ip = &icmp->icmp_ip; 2839 i_hlen = inner_ip->ip_hl << 2; 2840 o_len = ntohs(outer_ip->ip_len); 2841 if (o_len < 2842 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2843 /* Not enough data present */ 2844 return; 2845 } 2846 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2847 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2848 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2849 return; 2850 } 2851 port = udp->uh_dport; 2852 th = (struct tcphdr *)(udp + 1); 2853 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2854 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2855 /* Now adjust down the size of the outer IP header */ 2856 o_len -= sizeof(struct udphdr); 2857 outer_ip->ip_len = htons(o_len); 2858 /* Now call in to the normal handling code */ 2859 tcp_ctlinput_with_port(icmp, port); 2860 } 2861 #endif /* INET */ 2862 2863 #ifdef INET6 2864 static inline int 2865 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2866 { 2867 int mtu = ntohl(icmp6->icmp6_mtu); 2868 2869 /* 2870 * If no alternative MTU was proposed, or the proposed MTU was too 2871 * small, set to the min. 2872 */ 2873 if (mtu < IPV6_MMTU) 2874 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2875 return (mtu); 2876 } 2877 2878 static void 2879 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2880 { 2881 struct in6_addr *dst; 2882 struct inpcb *(*notify)(struct inpcb *, int); 2883 struct ip6_hdr *ip6; 2884 struct mbuf *m; 2885 struct inpcb *inp; 2886 struct tcpcb *tp; 2887 struct icmp6_hdr *icmp6; 2888 struct in_conninfo inc; 2889 struct tcp_ports { 2890 uint16_t th_sport; 2891 uint16_t th_dport; 2892 } t_ports; 2893 tcp_seq icmp_tcp_seq; 2894 unsigned int mtu; 2895 unsigned int off; 2896 int errno; 2897 2898 icmp6 = ip6cp->ip6c_icmp6; 2899 m = ip6cp->ip6c_m; 2900 ip6 = ip6cp->ip6c_ip6; 2901 off = ip6cp->ip6c_off; 2902 dst = &ip6cp->ip6c_finaldst->sin6_addr; 2903 2904 errno = icmp6_errmap(icmp6); 2905 switch (errno) { 2906 case 0: 2907 return; 2908 case EMSGSIZE: 2909 notify = tcp_mtudisc_notify; 2910 break; 2911 case ECONNREFUSED: 2912 if (V_icmp_may_rst) 2913 notify = tcp_drop_syn_sent; 2914 else 2915 notify = tcp_notify; 2916 break; 2917 case EHOSTUNREACH: 2918 /* 2919 * There are only four ICMPs that may reset connection: 2920 * - administratively prohibited 2921 * - port unreachable 2922 * - time exceeded in transit 2923 * - unknown next header 2924 */ 2925 if (V_icmp_may_rst && 2926 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 2927 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 2928 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 2929 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 2930 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 2931 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 2932 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 2933 notify = tcp_drop_syn_sent; 2934 else 2935 notify = tcp_notify; 2936 break; 2937 default: 2938 notify = tcp_notify; 2939 } 2940 2941 /* Check if we can safely get the ports from the tcp hdr */ 2942 if (m == NULL || 2943 (m->m_pkthdr.len < 2944 (int32_t) (off + sizeof(struct tcp_ports)))) { 2945 return; 2946 } 2947 bzero(&t_ports, sizeof(struct tcp_ports)); 2948 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2949 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2950 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2951 off += sizeof(struct tcp_ports); 2952 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2953 goto out; 2954 } 2955 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2956 if (inp != NULL) { 2957 tp = intotcpcb(inp); 2958 #ifdef TCP_OFFLOAD 2959 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2960 /* MTU discovery for offloaded connections. */ 2961 mtu = tcp6_next_pmtu(icmp6); 2962 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2963 goto out; 2964 } 2965 #endif 2966 if (tp->t_port != port) 2967 goto out; 2968 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2969 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2970 if (errno == EMSGSIZE) { 2971 /* 2972 * MTU discovery: 2973 * If we got a needfrag set the MTU 2974 * in the route to the suggested new 2975 * value (if given) and then notify. 2976 */ 2977 mtu = tcp6_next_pmtu(icmp6); 2978 2979 bzero(&inc, sizeof(inc)); 2980 inc.inc_fibnum = M_GETFIB(m); 2981 inc.inc_flags |= INC_ISIPV6; 2982 inc.inc6_faddr = *dst; 2983 if (in6_setscope(&inc.inc6_faddr, 2984 m->m_pkthdr.rcvif, NULL)) 2985 goto out; 2986 /* 2987 * Only process the offered MTU if it 2988 * is smaller than the current one. 2989 */ 2990 if (mtu < tp->t_maxseg + 2991 sizeof (struct tcphdr) + 2992 sizeof (struct ip6_hdr)) { 2993 tcp_hc_updatemtu(&inc, mtu); 2994 tcp_mtudisc(inp, mtu); 2995 ICMP6STAT_INC(icp6s_pmtuchg); 2996 } 2997 } else 2998 inp = (*notify)(inp, errno); 2999 } 3000 } else { 3001 bzero(&inc, sizeof(inc)); 3002 inc.inc_fibnum = M_GETFIB(m); 3003 inc.inc_flags |= INC_ISIPV6; 3004 inc.inc_fport = t_ports.th_dport; 3005 inc.inc_lport = t_ports.th_sport; 3006 inc.inc6_faddr = *dst; 3007 inc.inc6_laddr = ip6->ip6_src; 3008 syncache_unreach(&inc, icmp_tcp_seq, port); 3009 } 3010 out: 3011 if (inp != NULL) 3012 INP_WUNLOCK(inp); 3013 } 3014 3015 static void 3016 tcp6_ctlinput(struct ip6ctlparam *ctl) 3017 { 3018 tcp6_ctlinput_with_port(ctl, htons(0)); 3019 } 3020 3021 static void 3022 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3023 { 3024 struct ip6ctlparam *ip6cp = param.ip6cp; 3025 struct mbuf *m; 3026 struct udphdr *udp; 3027 uint16_t port; 3028 3029 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3030 if (m == NULL) { 3031 return; 3032 } 3033 udp = mtod(m, struct udphdr *); 3034 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3035 return; 3036 } 3037 port = udp->uh_dport; 3038 m_adj(m, sizeof(struct udphdr)); 3039 if ((m->m_flags & M_PKTHDR) == 0) { 3040 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3041 } 3042 /* Now call in to the normal handling code */ 3043 tcp6_ctlinput_with_port(ip6cp, port); 3044 } 3045 3046 #endif /* INET6 */ 3047 3048 static uint32_t 3049 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3050 { 3051 SIPHASH_CTX ctx; 3052 uint32_t hash[2]; 3053 3054 KASSERT(len >= SIPHASH_KEY_LENGTH, 3055 ("%s: keylen %u too short ", __func__, len)); 3056 SipHash24_Init(&ctx); 3057 SipHash_SetKey(&ctx, (uint8_t *)key); 3058 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3059 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3060 switch (inc->inc_flags & INC_ISIPV6) { 3061 #ifdef INET 3062 case 0: 3063 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3064 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3065 break; 3066 #endif 3067 #ifdef INET6 3068 case INC_ISIPV6: 3069 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3070 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3071 break; 3072 #endif 3073 } 3074 SipHash_Final((uint8_t *)hash, &ctx); 3075 3076 return (hash[0] ^ hash[1]); 3077 } 3078 3079 uint32_t 3080 tcp_new_ts_offset(struct in_conninfo *inc) 3081 { 3082 struct in_conninfo inc_store, *local_inc; 3083 3084 if (!V_tcp_ts_offset_per_conn) { 3085 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3086 inc_store.inc_lport = 0; 3087 inc_store.inc_fport = 0; 3088 local_inc = &inc_store; 3089 } else { 3090 local_inc = inc; 3091 } 3092 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3093 sizeof(V_ts_offset_secret))); 3094 } 3095 3096 /* 3097 * Following is where TCP initial sequence number generation occurs. 3098 * 3099 * There are two places where we must use initial sequence numbers: 3100 * 1. In SYN-ACK packets. 3101 * 2. In SYN packets. 3102 * 3103 * All ISNs for SYN-ACK packets are generated by the syncache. See 3104 * tcp_syncache.c for details. 3105 * 3106 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3107 * depends on this property. In addition, these ISNs should be 3108 * unguessable so as to prevent connection hijacking. To satisfy 3109 * the requirements of this situation, the algorithm outlined in 3110 * RFC 1948 is used, with only small modifications. 3111 * 3112 * Implementation details: 3113 * 3114 * Time is based off the system timer, and is corrected so that it 3115 * increases by one megabyte per second. This allows for proper 3116 * recycling on high speed LANs while still leaving over an hour 3117 * before rollover. 3118 * 3119 * As reading the *exact* system time is too expensive to be done 3120 * whenever setting up a TCP connection, we increment the time 3121 * offset in two ways. First, a small random positive increment 3122 * is added to isn_offset for each connection that is set up. 3123 * Second, the function tcp_isn_tick fires once per clock tick 3124 * and increments isn_offset as necessary so that sequence numbers 3125 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3126 * random positive increments serve only to ensure that the same 3127 * exact sequence number is never sent out twice (as could otherwise 3128 * happen when a port is recycled in less than the system tick 3129 * interval.) 3130 * 3131 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3132 * between seeding of isn_secret. This is normally set to zero, 3133 * as reseeding should not be necessary. 3134 * 3135 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3136 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3137 * general, this means holding an exclusive (write) lock. 3138 */ 3139 3140 #define ISN_BYTES_PER_SECOND 1048576 3141 #define ISN_STATIC_INCREMENT 4096 3142 #define ISN_RANDOM_INCREMENT (4096 - 1) 3143 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3144 3145 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3146 VNET_DEFINE_STATIC(int, isn_last); 3147 VNET_DEFINE_STATIC(int, isn_last_reseed); 3148 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3149 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3150 3151 #define V_isn_secret VNET(isn_secret) 3152 #define V_isn_last VNET(isn_last) 3153 #define V_isn_last_reseed VNET(isn_last_reseed) 3154 #define V_isn_offset VNET(isn_offset) 3155 #define V_isn_offset_old VNET(isn_offset_old) 3156 3157 tcp_seq 3158 tcp_new_isn(struct in_conninfo *inc) 3159 { 3160 tcp_seq new_isn; 3161 u_int32_t projected_offset; 3162 3163 ISN_LOCK(); 3164 /* Seed if this is the first use, reseed if requested. */ 3165 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3166 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3167 < (u_int)ticks))) { 3168 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3169 V_isn_last_reseed = ticks; 3170 } 3171 3172 /* Compute the hash and return the ISN. */ 3173 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3174 sizeof(V_isn_secret)); 3175 V_isn_offset += ISN_STATIC_INCREMENT + 3176 (arc4random() & ISN_RANDOM_INCREMENT); 3177 if (ticks != V_isn_last) { 3178 projected_offset = V_isn_offset_old + 3179 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3180 if (SEQ_GT(projected_offset, V_isn_offset)) 3181 V_isn_offset = projected_offset; 3182 V_isn_offset_old = V_isn_offset; 3183 V_isn_last = ticks; 3184 } 3185 new_isn += V_isn_offset; 3186 ISN_UNLOCK(); 3187 return (new_isn); 3188 } 3189 3190 /* 3191 * When a specific ICMP unreachable message is received and the 3192 * connection state is SYN-SENT, drop the connection. This behavior 3193 * is controlled by the icmp_may_rst sysctl. 3194 */ 3195 static struct inpcb * 3196 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3197 { 3198 struct tcpcb *tp; 3199 3200 NET_EPOCH_ASSERT(); 3201 INP_WLOCK_ASSERT(inp); 3202 3203 tp = intotcpcb(inp); 3204 if (tp->t_state != TCPS_SYN_SENT) 3205 return (inp); 3206 3207 if (IS_FASTOPEN(tp->t_flags)) 3208 tcp_fastopen_disable_path(tp); 3209 3210 tp = tcp_drop(tp, errno); 3211 if (tp != NULL) 3212 return (inp); 3213 else 3214 return (NULL); 3215 } 3216 3217 /* 3218 * When `need fragmentation' ICMP is received, update our idea of the MSS 3219 * based on the new value. Also nudge TCP to send something, since we 3220 * know the packet we just sent was dropped. 3221 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3222 */ 3223 static struct inpcb * 3224 tcp_mtudisc_notify(struct inpcb *inp, int error) 3225 { 3226 3227 return (tcp_mtudisc(inp, -1)); 3228 } 3229 3230 static struct inpcb * 3231 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3232 { 3233 struct tcpcb *tp; 3234 struct socket *so; 3235 3236 INP_WLOCK_ASSERT(inp); 3237 3238 tp = intotcpcb(inp); 3239 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3240 3241 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3242 3243 so = inp->inp_socket; 3244 SOCKBUF_LOCK(&so->so_snd); 3245 /* If the mss is larger than the socket buffer, decrease the mss. */ 3246 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3247 tp->t_maxseg = so->so_snd.sb_hiwat; 3248 SOCKBUF_UNLOCK(&so->so_snd); 3249 3250 TCPSTAT_INC(tcps_mturesent); 3251 tp->t_rtttime = 0; 3252 tp->snd_nxt = tp->snd_una; 3253 tcp_free_sackholes(tp); 3254 tp->snd_recover = tp->snd_max; 3255 if (tp->t_flags & TF_SACK_PERMIT) 3256 EXIT_FASTRECOVERY(tp->t_flags); 3257 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3258 /* 3259 * Conceptually the snd_nxt setting 3260 * and freeing sack holes should 3261 * be done by the default stacks 3262 * own tfb_tcp_mtu_chg(). 3263 */ 3264 tp->t_fb->tfb_tcp_mtu_chg(tp); 3265 } 3266 if (tcp_output(tp) < 0) 3267 return (NULL); 3268 else 3269 return (inp); 3270 } 3271 3272 #ifdef INET 3273 /* 3274 * Look-up the routing entry to the peer of this inpcb. If no route 3275 * is found and it cannot be allocated, then return 0. This routine 3276 * is called by TCP routines that access the rmx structure and by 3277 * tcp_mss_update to get the peer/interface MTU. 3278 */ 3279 uint32_t 3280 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3281 { 3282 struct nhop_object *nh; 3283 struct ifnet *ifp; 3284 uint32_t maxmtu = 0; 3285 3286 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3287 3288 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3289 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3290 if (nh == NULL) 3291 return (0); 3292 3293 ifp = nh->nh_ifp; 3294 maxmtu = nh->nh_mtu; 3295 3296 /* Report additional interface capabilities. */ 3297 if (cap != NULL) { 3298 if (ifp->if_capenable & IFCAP_TSO4 && 3299 ifp->if_hwassist & CSUM_TSO) { 3300 cap->ifcap |= CSUM_TSO; 3301 cap->tsomax = ifp->if_hw_tsomax; 3302 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3303 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3304 } 3305 } 3306 } 3307 return (maxmtu); 3308 } 3309 #endif /* INET */ 3310 3311 #ifdef INET6 3312 uint32_t 3313 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3314 { 3315 struct nhop_object *nh; 3316 struct in6_addr dst6; 3317 uint32_t scopeid; 3318 struct ifnet *ifp; 3319 uint32_t maxmtu = 0; 3320 3321 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3322 3323 if (inc->inc_flags & INC_IPV6MINMTU) 3324 return (IPV6_MMTU); 3325 3326 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3327 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3328 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3329 if (nh == NULL) 3330 return (0); 3331 3332 ifp = nh->nh_ifp; 3333 maxmtu = nh->nh_mtu; 3334 3335 /* Report additional interface capabilities. */ 3336 if (cap != NULL) { 3337 if (ifp->if_capenable & IFCAP_TSO6 && 3338 ifp->if_hwassist & CSUM_TSO) { 3339 cap->ifcap |= CSUM_TSO; 3340 cap->tsomax = ifp->if_hw_tsomax; 3341 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3342 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3343 } 3344 } 3345 } 3346 3347 return (maxmtu); 3348 } 3349 3350 /* 3351 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3352 * 3353 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3354 * The right place to do that is ip6_setpktopt() that has just been 3355 * executed. By the way it just filled ip6po_minmtu for us. 3356 */ 3357 void 3358 tcp6_use_min_mtu(struct tcpcb *tp) 3359 { 3360 struct inpcb *inp = tptoinpcb(tp); 3361 3362 INP_WLOCK_ASSERT(inp); 3363 /* 3364 * In case of the IPV6_USE_MIN_MTU socket 3365 * option, the INC_IPV6MINMTU flag to announce 3366 * a corresponding MSS during the initial 3367 * handshake. If the TCP connection is not in 3368 * the front states, just reduce the MSS being 3369 * used. This avoids the sending of TCP 3370 * segments which will be fragmented at the 3371 * IPv6 layer. 3372 */ 3373 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3374 if ((tp->t_state >= TCPS_SYN_SENT) && 3375 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3376 struct ip6_pktopts *opt; 3377 3378 opt = inp->in6p_outputopts; 3379 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3380 tp->t_maxseg > TCP6_MSS) 3381 tp->t_maxseg = TCP6_MSS; 3382 } 3383 } 3384 #endif /* INET6 */ 3385 3386 /* 3387 * Calculate effective SMSS per RFC5681 definition for a given TCP 3388 * connection at its current state, taking into account SACK and etc. 3389 */ 3390 u_int 3391 tcp_maxseg(const struct tcpcb *tp) 3392 { 3393 u_int optlen; 3394 3395 if (tp->t_flags & TF_NOOPT) 3396 return (tp->t_maxseg); 3397 3398 /* 3399 * Here we have a simplified code from tcp_addoptions(), 3400 * without a proper loop, and having most of paddings hardcoded. 3401 * We might make mistakes with padding here in some edge cases, 3402 * but this is harmless, since result of tcp_maxseg() is used 3403 * only in cwnd and ssthresh estimations. 3404 */ 3405 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3406 if (tp->t_flags & TF_RCVD_TSTMP) 3407 optlen = TCPOLEN_TSTAMP_APPA; 3408 else 3409 optlen = 0; 3410 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3411 if (tp->t_flags & TF_SIGNATURE) 3412 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3413 #endif 3414 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3415 optlen += TCPOLEN_SACKHDR; 3416 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3417 optlen = PADTCPOLEN(optlen); 3418 } 3419 } else { 3420 if (tp->t_flags & TF_REQ_TSTMP) 3421 optlen = TCPOLEN_TSTAMP_APPA; 3422 else 3423 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3424 if (tp->t_flags & TF_REQ_SCALE) 3425 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3426 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3427 if (tp->t_flags & TF_SIGNATURE) 3428 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3429 #endif 3430 if (tp->t_flags & TF_SACK_PERMIT) 3431 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3432 } 3433 #undef PAD 3434 optlen = min(optlen, TCP_MAXOLEN); 3435 return (tp->t_maxseg - optlen); 3436 } 3437 3438 3439 u_int 3440 tcp_fixed_maxseg(const struct tcpcb *tp) 3441 { 3442 int optlen; 3443 3444 if (tp->t_flags & TF_NOOPT) 3445 return (tp->t_maxseg); 3446 3447 /* 3448 * Here we have a simplified code from tcp_addoptions(), 3449 * without a proper loop, and having most of paddings hardcoded. 3450 * We only consider fixed options that we would send every 3451 * time I.e. SACK is not considered. This is important 3452 * for cc modules to figure out what the modulo of the 3453 * cwnd should be. 3454 */ 3455 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3456 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3457 if (tp->t_flags & TF_RCVD_TSTMP) 3458 optlen = TCPOLEN_TSTAMP_APPA; 3459 else 3460 optlen = 0; 3461 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3462 if (tp->t_flags & TF_SIGNATURE) 3463 optlen += PAD(TCPOLEN_SIGNATURE); 3464 #endif 3465 } else { 3466 if (tp->t_flags & TF_REQ_TSTMP) 3467 optlen = TCPOLEN_TSTAMP_APPA; 3468 else 3469 optlen = PAD(TCPOLEN_MAXSEG); 3470 if (tp->t_flags & TF_REQ_SCALE) 3471 optlen += PAD(TCPOLEN_WINDOW); 3472 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3473 if (tp->t_flags & TF_SIGNATURE) 3474 optlen += PAD(TCPOLEN_SIGNATURE); 3475 #endif 3476 if (tp->t_flags & TF_SACK_PERMIT) 3477 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3478 } 3479 #undef PAD 3480 optlen = min(optlen, TCP_MAXOLEN); 3481 return (tp->t_maxseg - optlen); 3482 } 3483 3484 3485 3486 static int 3487 sysctl_drop(SYSCTL_HANDLER_ARGS) 3488 { 3489 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3490 struct sockaddr_storage addrs[2]; 3491 struct inpcb *inp; 3492 struct tcpcb *tp; 3493 #ifdef INET 3494 struct sockaddr_in *fin = NULL, *lin = NULL; 3495 #endif 3496 struct epoch_tracker et; 3497 #ifdef INET6 3498 struct sockaddr_in6 *fin6, *lin6; 3499 #endif 3500 int error; 3501 3502 inp = NULL; 3503 #ifdef INET6 3504 fin6 = lin6 = NULL; 3505 #endif 3506 error = 0; 3507 3508 if (req->oldptr != NULL || req->oldlen != 0) 3509 return (EINVAL); 3510 if (req->newptr == NULL) 3511 return (EPERM); 3512 if (req->newlen < sizeof(addrs)) 3513 return (ENOMEM); 3514 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3515 if (error) 3516 return (error); 3517 3518 switch (addrs[0].ss_family) { 3519 #ifdef INET6 3520 case AF_INET6: 3521 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3522 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3523 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3524 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3525 return (EINVAL); 3526 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3527 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3528 return (EINVAL); 3529 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3530 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3531 #ifdef INET 3532 fin = (struct sockaddr_in *)&addrs[0]; 3533 lin = (struct sockaddr_in *)&addrs[1]; 3534 #endif 3535 break; 3536 } 3537 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3538 if (error) 3539 return (error); 3540 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3541 if (error) 3542 return (error); 3543 break; 3544 #endif 3545 #ifdef INET 3546 case AF_INET: 3547 fin = (struct sockaddr_in *)&addrs[0]; 3548 lin = (struct sockaddr_in *)&addrs[1]; 3549 if (fin->sin_len != sizeof(struct sockaddr_in) || 3550 lin->sin_len != sizeof(struct sockaddr_in)) 3551 return (EINVAL); 3552 break; 3553 #endif 3554 default: 3555 return (EINVAL); 3556 } 3557 NET_EPOCH_ENTER(et); 3558 switch (addrs[0].ss_family) { 3559 #ifdef INET6 3560 case AF_INET6: 3561 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3562 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3563 INPLOOKUP_WLOCKPCB, NULL); 3564 break; 3565 #endif 3566 #ifdef INET 3567 case AF_INET: 3568 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3569 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3570 break; 3571 #endif 3572 } 3573 if (inp != NULL) { 3574 if (!SOLISTENING(inp->inp_socket)) { 3575 tp = intotcpcb(inp); 3576 tp = tcp_drop(tp, ECONNABORTED); 3577 if (tp != NULL) 3578 INP_WUNLOCK(inp); 3579 } else 3580 INP_WUNLOCK(inp); 3581 } else 3582 error = ESRCH; 3583 NET_EPOCH_EXIT(et); 3584 return (error); 3585 } 3586 3587 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3588 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3589 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3590 "Drop TCP connection"); 3591 3592 static int 3593 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3594 { 3595 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3596 &tcp_ctloutput_set)); 3597 } 3598 3599 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3600 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3601 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3602 "Set socket option for TCP endpoint"); 3603 3604 #ifdef KERN_TLS 3605 static int 3606 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3607 { 3608 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3609 struct sockaddr_storage addrs[2]; 3610 struct inpcb *inp; 3611 #ifdef INET 3612 struct sockaddr_in *fin = NULL, *lin = NULL; 3613 #endif 3614 struct epoch_tracker et; 3615 #ifdef INET6 3616 struct sockaddr_in6 *fin6, *lin6; 3617 #endif 3618 int error; 3619 3620 inp = NULL; 3621 #ifdef INET6 3622 fin6 = lin6 = NULL; 3623 #endif 3624 error = 0; 3625 3626 if (req->oldptr != NULL || req->oldlen != 0) 3627 return (EINVAL); 3628 if (req->newptr == NULL) 3629 return (EPERM); 3630 if (req->newlen < sizeof(addrs)) 3631 return (ENOMEM); 3632 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3633 if (error) 3634 return (error); 3635 3636 switch (addrs[0].ss_family) { 3637 #ifdef INET6 3638 case AF_INET6: 3639 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3640 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3641 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3642 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3643 return (EINVAL); 3644 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3645 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3646 return (EINVAL); 3647 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3648 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3649 #ifdef INET 3650 fin = (struct sockaddr_in *)&addrs[0]; 3651 lin = (struct sockaddr_in *)&addrs[1]; 3652 #endif 3653 break; 3654 } 3655 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3656 if (error) 3657 return (error); 3658 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3659 if (error) 3660 return (error); 3661 break; 3662 #endif 3663 #ifdef INET 3664 case AF_INET: 3665 fin = (struct sockaddr_in *)&addrs[0]; 3666 lin = (struct sockaddr_in *)&addrs[1]; 3667 if (fin->sin_len != sizeof(struct sockaddr_in) || 3668 lin->sin_len != sizeof(struct sockaddr_in)) 3669 return (EINVAL); 3670 break; 3671 #endif 3672 default: 3673 return (EINVAL); 3674 } 3675 NET_EPOCH_ENTER(et); 3676 switch (addrs[0].ss_family) { 3677 #ifdef INET6 3678 case AF_INET6: 3679 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3680 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3681 INPLOOKUP_WLOCKPCB, NULL); 3682 break; 3683 #endif 3684 #ifdef INET 3685 case AF_INET: 3686 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3687 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3688 break; 3689 #endif 3690 } 3691 NET_EPOCH_EXIT(et); 3692 if (inp != NULL) { 3693 struct socket *so; 3694 3695 so = inp->inp_socket; 3696 soref(so); 3697 error = ktls_set_tx_mode(so, 3698 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3699 INP_WUNLOCK(inp); 3700 sorele(so); 3701 } else 3702 error = ESRCH; 3703 return (error); 3704 } 3705 3706 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3707 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3708 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3709 "Switch TCP connection to SW TLS"); 3710 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3711 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3712 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3713 "Switch TCP connection to ifnet TLS"); 3714 #endif 3715 3716 /* 3717 * Generate a standardized TCP log line for use throughout the 3718 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3719 * allow use in the interrupt context. 3720 * 3721 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3722 * NB: The function may return NULL if memory allocation failed. 3723 * 3724 * Due to header inclusion and ordering limitations the struct ip 3725 * and ip6_hdr pointers have to be passed as void pointers. 3726 */ 3727 char * 3728 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3729 const void *ip6hdr) 3730 { 3731 3732 /* Is logging enabled? */ 3733 if (V_tcp_log_in_vain == 0) 3734 return (NULL); 3735 3736 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3737 } 3738 3739 char * 3740 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3741 const void *ip6hdr) 3742 { 3743 3744 /* Is logging enabled? */ 3745 if (tcp_log_debug == 0) 3746 return (NULL); 3747 3748 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3749 } 3750 3751 static char * 3752 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3753 const void *ip6hdr) 3754 { 3755 char *s, *sp; 3756 size_t size; 3757 #ifdef INET 3758 const struct ip *ip = (const struct ip *)ip4hdr; 3759 #endif 3760 #ifdef INET6 3761 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3762 #endif /* INET6 */ 3763 3764 /* 3765 * The log line looks like this: 3766 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3767 */ 3768 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3769 sizeof(PRINT_TH_FLAGS) + 1 + 3770 #ifdef INET6 3771 2 * INET6_ADDRSTRLEN; 3772 #else 3773 2 * INET_ADDRSTRLEN; 3774 #endif /* INET6 */ 3775 3776 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3777 if (s == NULL) 3778 return (NULL); 3779 3780 strcat(s, "TCP: ["); 3781 sp = s + strlen(s); 3782 3783 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3784 inet_ntoa_r(inc->inc_faddr, sp); 3785 sp = s + strlen(s); 3786 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3787 sp = s + strlen(s); 3788 inet_ntoa_r(inc->inc_laddr, sp); 3789 sp = s + strlen(s); 3790 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3791 #ifdef INET6 3792 } else if (inc) { 3793 ip6_sprintf(sp, &inc->inc6_faddr); 3794 sp = s + strlen(s); 3795 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3796 sp = s + strlen(s); 3797 ip6_sprintf(sp, &inc->inc6_laddr); 3798 sp = s + strlen(s); 3799 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3800 } else if (ip6 && th) { 3801 ip6_sprintf(sp, &ip6->ip6_src); 3802 sp = s + strlen(s); 3803 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3804 sp = s + strlen(s); 3805 ip6_sprintf(sp, &ip6->ip6_dst); 3806 sp = s + strlen(s); 3807 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3808 #endif /* INET6 */ 3809 #ifdef INET 3810 } else if (ip && th) { 3811 inet_ntoa_r(ip->ip_src, sp); 3812 sp = s + strlen(s); 3813 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3814 sp = s + strlen(s); 3815 inet_ntoa_r(ip->ip_dst, sp); 3816 sp = s + strlen(s); 3817 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3818 #endif /* INET */ 3819 } else { 3820 free(s, M_TCPLOG); 3821 return (NULL); 3822 } 3823 sp = s + strlen(s); 3824 if (th) 3825 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3826 if (*(s + size - 1) != '\0') 3827 panic("%s: string too long", __func__); 3828 return (s); 3829 } 3830 3831 /* 3832 * A subroutine which makes it easy to track TCP state changes with DTrace. 3833 * This function shouldn't be called for t_state initializations that don't 3834 * correspond to actual TCP state transitions. 3835 */ 3836 void 3837 tcp_state_change(struct tcpcb *tp, int newstate) 3838 { 3839 #if defined(KDTRACE_HOOKS) 3840 int pstate = tp->t_state; 3841 #endif 3842 3843 TCPSTATES_DEC(tp->t_state); 3844 TCPSTATES_INC(newstate); 3845 tp->t_state = newstate; 3846 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3847 } 3848 3849 /* 3850 * Create an external-format (``xtcpcb'') structure using the information in 3851 * the kernel-format tcpcb structure pointed to by tp. This is done to 3852 * reduce the spew of irrelevant information over this interface, to isolate 3853 * user code from changes in the kernel structure, and potentially to provide 3854 * information-hiding if we decide that some of this information should be 3855 * hidden from users. 3856 */ 3857 void 3858 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3859 { 3860 struct tcpcb *tp = intotcpcb(inp); 3861 sbintime_t now; 3862 3863 bzero(xt, sizeof(*xt)); 3864 xt->t_state = tp->t_state; 3865 xt->t_logstate = tp->t_logstate; 3866 xt->t_flags = tp->t_flags; 3867 xt->t_sndzerowin = tp->t_sndzerowin; 3868 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3869 xt->t_rcvoopack = tp->t_rcvoopack; 3870 xt->t_rcv_wnd = tp->rcv_wnd; 3871 xt->t_snd_wnd = tp->snd_wnd; 3872 xt->t_snd_cwnd = tp->snd_cwnd; 3873 xt->t_snd_ssthresh = tp->snd_ssthresh; 3874 xt->t_dsack_bytes = tp->t_dsack_bytes; 3875 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 3876 xt->t_dsack_pack = tp->t_dsack_pack; 3877 xt->t_maxseg = tp->t_maxseg; 3878 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3879 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3880 3881 now = getsbinuptime(); 3882 #define COPYTIMER(which,where) do { \ 3883 if (tp->t_timers[which] != SBT_MAX) \ 3884 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 3885 else \ 3886 xt->where = 0; \ 3887 } while (0) 3888 COPYTIMER(TT_DELACK, tt_delack); 3889 COPYTIMER(TT_REXMT, tt_rexmt); 3890 COPYTIMER(TT_PERSIST, tt_persist); 3891 COPYTIMER(TT_KEEP, tt_keep); 3892 COPYTIMER(TT_2MSL, tt_2msl); 3893 #undef COPYTIMER 3894 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3895 3896 xt->xt_encaps_port = tp->t_port; 3897 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3898 TCP_FUNCTION_NAME_LEN_MAX); 3899 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 3900 #ifdef TCP_BLACKBOX 3901 (void)tcp_log_get_id(tp, xt->xt_logid); 3902 #endif 3903 3904 xt->xt_len = sizeof(struct xtcpcb); 3905 in_pcbtoxinpcb(inp, &xt->xt_inp); 3906 } 3907 3908 void 3909 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3910 { 3911 uint32_t bit, i; 3912 3913 if ((tp == NULL) || 3914 (status > TCP_EI_STATUS_MAX_VALUE) || 3915 (status == 0)) { 3916 /* Invalid */ 3917 return; 3918 } 3919 if (status > (sizeof(uint32_t) * 8)) { 3920 /* Should this be a KASSERT? */ 3921 return; 3922 } 3923 bit = 1U << (status - 1); 3924 if (bit & tp->t_end_info_status) { 3925 /* already logged */ 3926 return; 3927 } 3928 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3929 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3930 tp->t_end_info_bytes[i] = status; 3931 tp->t_end_info_status |= bit; 3932 break; 3933 } 3934 } 3935 } 3936 3937 int 3938 tcp_can_enable_pacing(void) 3939 { 3940 3941 if ((tcp_pacing_limit == -1) || 3942 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 3943 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 3944 shadow_num_connections = number_of_tcp_connections_pacing; 3945 return (1); 3946 } else { 3947 return (0); 3948 } 3949 } 3950 3951 static uint8_t tcp_pacing_warning = 0; 3952 3953 void 3954 tcp_decrement_paced_conn(void) 3955 { 3956 uint32_t ret; 3957 3958 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 3959 shadow_num_connections = number_of_tcp_connections_pacing; 3960 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 3961 if (ret == 0) { 3962 if (tcp_pacing_limit != -1) { 3963 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 3964 tcp_pacing_limit = 0; 3965 } else if (tcp_pacing_warning == 0) { 3966 printf("Warning pacing count is invalid, invalid decrement\n"); 3967 tcp_pacing_warning = 1; 3968 } 3969 } 3970 } 3971