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 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/arb.h> 45 #include <sys/callout.h> 46 #include <sys/eventhandler.h> 47 #ifdef TCP_HHOOK 48 #include <sys/hhook.h> 49 #endif 50 #include <sys/kernel.h> 51 #ifdef TCP_HHOOK 52 #include <sys/khelp.h> 53 #endif 54 #ifdef KERN_TLS 55 #include <sys/ktls.h> 56 #endif 57 #include <sys/qmath.h> 58 #include <sys/stats.h> 59 #include <sys/sysctl.h> 60 #include <sys/jail.h> 61 #include <sys/malloc.h> 62 #include <sys/refcount.h> 63 #include <sys/mbuf.h> 64 #include <sys/priv.h> 65 #include <sys/proc.h> 66 #include <sys/sdt.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/protosw.h> 70 #include <sys/random.h> 71 72 #include <vm/uma.h> 73 74 #include <net/route.h> 75 #include <net/route/nhop.h> 76 #include <net/if.h> 77 #include <net/if_var.h> 78 #include <net/if_private.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 TCP_OFFLOAD 119 #include <netinet/tcp_offload.h> 120 #endif 121 #include <netinet/udp.h> 122 #include <netinet/udp_var.h> 123 #ifdef INET6 124 #include <netinet6/tcp6_var.h> 125 #endif 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 #include <crypto/siphash/siphash.h> 131 132 #include <security/mac/mac_framework.h> 133 134 #ifdef INET6 135 static ip6proto_ctlinput_t tcp6_ctlinput; 136 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 137 #endif 138 139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 140 #ifdef INET6 141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 142 #endif 143 144 #ifdef NETFLIX_EXP_DETECTION 145 /* Sack attack detection thresholds and such */ 146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 147 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 148 "Sack Attack detection thresholds"); 149 int32_t tcp_force_detection = 0; 150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 151 CTLFLAG_RW, 152 &tcp_force_detection, 0, 153 "Do we force detection even if the INP has it off?"); 154 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 156 CTLFLAG_RW, 157 &tcp_sack_to_ack_thresh, 700, 158 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 159 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 161 CTLFLAG_RW, 162 &tcp_sack_to_move_thresh, 600, 163 "Percentage of sack moves we must see above (10.1 percent is 101)"); 164 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 166 CTLFLAG_RW, 167 &tcp_restoral_thresh, 550, 168 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 169 int32_t tcp_sad_decay_val = 800; 170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 171 CTLFLAG_RW, 172 &tcp_sad_decay_val, 800, 173 "The decay percentage (10.1 percent equals 101 )"); 174 int32_t tcp_map_minimum = 500; 175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 176 CTLFLAG_RW, 177 &tcp_map_minimum, 500, 178 "Number of Map enteries before we start detection"); 179 int32_t tcp_attack_on_turns_on_logging = 0; 180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 181 CTLFLAG_RW, 182 &tcp_attack_on_turns_on_logging, 0, 183 "When we have a positive hit on attack, do we turn on logging?"); 184 int32_t tcp_sad_pacing_interval = 2000; 185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 186 CTLFLAG_RW, 187 &tcp_sad_pacing_interval, 2000, 188 "What is the minimum pacing interval for a classified attacker?"); 189 190 int32_t tcp_sad_low_pps = 100; 191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 192 CTLFLAG_RW, 193 &tcp_sad_low_pps, 100, 194 "What is the input pps that below which we do not decay?"); 195 #endif 196 uint32_t tcp_ack_war_time_window = 1000; 197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 198 CTLFLAG_RW, 199 &tcp_ack_war_time_window, 1000, 200 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 201 uint32_t tcp_ack_war_cnt = 5; 202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 203 CTLFLAG_RW, 204 &tcp_ack_war_cnt, 5, 205 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 206 207 struct rwlock tcp_function_lock; 208 209 static int 210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 211 { 212 int error, new; 213 214 new = V_tcp_mssdflt; 215 error = sysctl_handle_int(oidp, &new, 0, req); 216 if (error == 0 && req->newptr) { 217 if (new < TCP_MINMSS) 218 error = EINVAL; 219 else 220 V_tcp_mssdflt = new; 221 } 222 return (error); 223 } 224 225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 226 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 227 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 228 "Default TCP Maximum Segment Size"); 229 230 #ifdef INET6 231 static int 232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 233 { 234 int error, new; 235 236 new = V_tcp_v6mssdflt; 237 error = sysctl_handle_int(oidp, &new, 0, req); 238 if (error == 0 && req->newptr) { 239 if (new < TCP_MINMSS) 240 error = EINVAL; 241 else 242 V_tcp_v6mssdflt = new; 243 } 244 return (error); 245 } 246 247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 248 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 249 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 250 "Default TCP Maximum Segment Size for IPv6"); 251 #endif /* INET6 */ 252 253 /* 254 * Minimum MSS we accept and use. This prevents DoS attacks where 255 * we are forced to a ridiculous low MSS like 20 and send hundreds 256 * of packets instead of one. The effect scales with the available 257 * bandwidth and quickly saturates the CPU and network interface 258 * with packet generation and sending. Set to zero to disable MINMSS 259 * checking. This setting prevents us from sending too small packets. 260 */ 261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 263 &VNET_NAME(tcp_minmss), 0, 264 "Minimum TCP Maximum Segment Size"); 265 266 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 268 &VNET_NAME(tcp_do_rfc1323), 0, 269 "Enable rfc1323 (high performance TCP) extensions"); 270 271 /* 272 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 273 * Some stacks negotiate TS, but never send them after connection setup. Some 274 * stacks negotiate TS, but don't send them when sending keep-alive segments. 275 * These include modern widely deployed TCP stacks. 276 * Therefore tolerating violations for now... 277 */ 278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 280 &VNET_NAME(tcp_tolerate_missing_ts), 0, 281 "Tolerate missing TCP timestamps"); 282 283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 285 &VNET_NAME(tcp_ts_offset_per_conn), 0, 286 "Initialize TCP timestamps per connection instead of per host pair"); 287 288 /* How many connections are pacing */ 289 static volatile uint32_t number_of_tcp_connections_pacing = 0; 290 static uint32_t shadow_num_connections = 0; 291 292 static int tcp_pacing_limit = 10000; 293 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 294 &tcp_pacing_limit, 1000, 295 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 296 297 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 298 &shadow_num_connections, 0, "Number of TCP connections being paced"); 299 300 static int tcp_log_debug = 0; 301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 302 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 303 304 static int tcp_tcbhashsize; 305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 306 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 307 308 static int do_tcpdrain = 1; 309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 310 "Enable tcp_drain routine for extra help when low on mbufs"); 311 312 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 313 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 314 315 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 316 #define V_icmp_may_rst VNET(icmp_may_rst) 317 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 318 &VNET_NAME(icmp_may_rst), 0, 319 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 320 321 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 322 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 324 &VNET_NAME(tcp_isn_reseed_interval), 0, 325 "Seconds between reseeding of ISN secret"); 326 327 static int tcp_soreceive_stream; 328 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 329 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 330 331 VNET_DEFINE(uma_zone_t, sack_hole_zone); 332 #define V_sack_hole_zone VNET(sack_hole_zone) 333 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 334 static int 335 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 336 { 337 int error; 338 uint32_t new; 339 340 new = V_tcp_map_entries_limit; 341 error = sysctl_handle_int(oidp, &new, 0, req); 342 if (error == 0 && req->newptr) { 343 /* only allow "0" and value > minimum */ 344 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 345 error = EINVAL; 346 else 347 V_tcp_map_entries_limit = new; 348 } 349 return (error); 350 } 351 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 352 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 353 &VNET_NAME(tcp_map_entries_limit), 0, 354 &sysctl_net_inet_tcp_map_limit_check, "IU", 355 "Total sendmap entries limit"); 356 357 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 358 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 359 &VNET_NAME(tcp_map_split_limit), 0, 360 "Total sendmap split entries limit"); 361 362 #ifdef TCP_HHOOK 363 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 364 #endif 365 366 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 367 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 368 #define V_ts_offset_secret VNET(ts_offset_secret) 369 370 static int tcp_default_fb_init(struct tcpcb *tp); 371 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 372 static int tcp_default_handoff_ok(struct tcpcb *tp); 373 static struct inpcb *tcp_notify(struct inpcb *, int); 374 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 375 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 376 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 377 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 378 const void *ip4hdr, const void *ip6hdr); 379 static ipproto_ctlinput_t tcp_ctlinput; 380 static udp_tun_icmp_t tcp_ctlinput_viaudp; 381 382 static struct tcp_function_block tcp_def_funcblk = { 383 .tfb_tcp_block_name = "freebsd", 384 .tfb_tcp_output = tcp_default_output, 385 .tfb_tcp_do_segment = tcp_do_segment, 386 .tfb_tcp_ctloutput = tcp_default_ctloutput, 387 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 388 .tfb_tcp_fb_init = tcp_default_fb_init, 389 .tfb_tcp_fb_fini = tcp_default_fb_fini, 390 }; 391 392 static int tcp_fb_cnt = 0; 393 struct tcp_funchead t_functions; 394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 395 396 void 397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 398 { 399 TCPSTAT_INC(tcps_dsack_count); 400 tp->t_dsack_pack++; 401 if (tlp == 0) { 402 if (SEQ_GT(end, start)) { 403 tp->t_dsack_bytes += (end - start); 404 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 405 } else { 406 tp->t_dsack_tlp_bytes += (start - end); 407 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 408 } 409 } else { 410 if (SEQ_GT(end, start)) { 411 tp->t_dsack_bytes += (end - start); 412 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 413 } else { 414 tp->t_dsack_tlp_bytes += (start - end); 415 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 416 } 417 } 418 } 419 420 static struct tcp_function_block * 421 find_tcp_functions_locked(struct tcp_function_set *fs) 422 { 423 struct tcp_function *f; 424 struct tcp_function_block *blk=NULL; 425 426 TAILQ_FOREACH(f, &t_functions, tf_next) { 427 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 428 blk = f->tf_fb; 429 break; 430 } 431 } 432 return(blk); 433 } 434 435 static struct tcp_function_block * 436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 437 { 438 struct tcp_function_block *rblk=NULL; 439 struct tcp_function *f; 440 441 TAILQ_FOREACH(f, &t_functions, tf_next) { 442 if (f->tf_fb == blk) { 443 rblk = blk; 444 if (s) { 445 *s = f; 446 } 447 break; 448 } 449 } 450 return (rblk); 451 } 452 453 struct tcp_function_block * 454 find_and_ref_tcp_functions(struct tcp_function_set *fs) 455 { 456 struct tcp_function_block *blk; 457 458 rw_rlock(&tcp_function_lock); 459 blk = find_tcp_functions_locked(fs); 460 if (blk) 461 refcount_acquire(&blk->tfb_refcnt); 462 rw_runlock(&tcp_function_lock); 463 return(blk); 464 } 465 466 struct tcp_function_block * 467 find_and_ref_tcp_fb(struct tcp_function_block *blk) 468 { 469 struct tcp_function_block *rblk; 470 471 rw_rlock(&tcp_function_lock); 472 rblk = find_tcp_fb_locked(blk, NULL); 473 if (rblk) 474 refcount_acquire(&rblk->tfb_refcnt); 475 rw_runlock(&tcp_function_lock); 476 return(rblk); 477 } 478 479 /* Find a matching alias for the given tcp_function_block. */ 480 int 481 find_tcp_function_alias(struct tcp_function_block *blk, 482 struct tcp_function_set *fs) 483 { 484 struct tcp_function *f; 485 int found; 486 487 found = 0; 488 rw_rlock(&tcp_function_lock); 489 TAILQ_FOREACH(f, &t_functions, tf_next) { 490 if ((f->tf_fb == blk) && 491 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 492 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 493 /* Matching function block with different name. */ 494 strncpy(fs->function_set_name, f->tf_name, 495 TCP_FUNCTION_NAME_LEN_MAX); 496 found = 1; 497 break; 498 } 499 } 500 /* Null terminate the string appropriately. */ 501 if (found) { 502 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 503 } else { 504 fs->function_set_name[0] = '\0'; 505 } 506 rw_runlock(&tcp_function_lock); 507 return (found); 508 } 509 510 static struct tcp_function_block * 511 find_and_ref_tcp_default_fb(void) 512 { 513 struct tcp_function_block *rblk; 514 515 rw_rlock(&tcp_function_lock); 516 rblk = tcp_func_set_ptr; 517 refcount_acquire(&rblk->tfb_refcnt); 518 rw_runlock(&tcp_function_lock); 519 return (rblk); 520 } 521 522 void 523 tcp_switch_back_to_default(struct tcpcb *tp) 524 { 525 struct tcp_function_block *tfb; 526 527 KASSERT(tp->t_fb != &tcp_def_funcblk, 528 ("%s: called by the built-in default stack", __func__)); 529 530 /* 531 * Release the old stack. This function will either find a new one 532 * or panic. 533 */ 534 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 535 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 536 refcount_release(&tp->t_fb->tfb_refcnt); 537 538 /* 539 * Now, we'll find a new function block to use. 540 * Start by trying the current user-selected 541 * default, unless this stack is the user-selected 542 * default. 543 */ 544 tfb = find_and_ref_tcp_default_fb(); 545 if (tfb == tp->t_fb) { 546 refcount_release(&tfb->tfb_refcnt); 547 tfb = NULL; 548 } 549 /* Does the stack accept this connection? */ 550 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 551 (*tfb->tfb_tcp_handoff_ok)(tp)) { 552 refcount_release(&tfb->tfb_refcnt); 553 tfb = NULL; 554 } 555 /* Try to use that stack. */ 556 if (tfb != NULL) { 557 /* Initialize the new stack. If it succeeds, we are done. */ 558 tp->t_fb = tfb; 559 if (tp->t_fb->tfb_tcp_fb_init == NULL || 560 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 561 return; 562 563 /* 564 * Initialization failed. Release the reference count on 565 * the stack. 566 */ 567 refcount_release(&tfb->tfb_refcnt); 568 } 569 570 /* 571 * If that wasn't feasible, use the built-in default 572 * stack which is not allowed to reject anyone. 573 */ 574 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 575 if (tfb == NULL) { 576 /* there always should be a default */ 577 panic("Can't refer to tcp_def_funcblk"); 578 } 579 if (tfb->tfb_tcp_handoff_ok != NULL) { 580 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 581 /* The default stack cannot say no */ 582 panic("Default stack rejects a new session?"); 583 } 584 } 585 tp->t_fb = tfb; 586 if (tp->t_fb->tfb_tcp_fb_init != NULL && 587 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 588 /* The default stack cannot fail */ 589 panic("Default stack initialization failed"); 590 } 591 } 592 593 static bool 594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 595 const struct sockaddr *sa, void *ctx) 596 { 597 struct ip *iph; 598 #ifdef INET6 599 struct ip6_hdr *ip6; 600 #endif 601 struct udphdr *uh; 602 struct tcphdr *th; 603 int thlen; 604 uint16_t port; 605 606 TCPSTAT_INC(tcps_tunneled_pkts); 607 if ((m->m_flags & M_PKTHDR) == 0) { 608 /* Can't handle one that is not a pkt hdr */ 609 TCPSTAT_INC(tcps_tunneled_errs); 610 goto out; 611 } 612 thlen = sizeof(struct tcphdr); 613 if (m->m_len < off + sizeof(struct udphdr) + thlen && 614 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 615 TCPSTAT_INC(tcps_tunneled_errs); 616 goto out; 617 } 618 iph = mtod(m, struct ip *); 619 uh = (struct udphdr *)((caddr_t)iph + off); 620 th = (struct tcphdr *)(uh + 1); 621 thlen = th->th_off << 2; 622 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 623 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 624 if (m == NULL) { 625 TCPSTAT_INC(tcps_tunneled_errs); 626 goto out; 627 } else { 628 iph = mtod(m, struct ip *); 629 uh = (struct udphdr *)((caddr_t)iph + off); 630 th = (struct tcphdr *)(uh + 1); 631 } 632 } 633 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 634 bcopy(th, uh, m->m_len - off); 635 m->m_len -= sizeof(struct udphdr); 636 m->m_pkthdr.len -= sizeof(struct udphdr); 637 /* 638 * We use the same algorithm for 639 * both UDP and TCP for c-sum. So 640 * the code in tcp_input will skip 641 * the checksum. So we do nothing 642 * with the flag (m->m_pkthdr.csum_flags). 643 */ 644 switch (iph->ip_v) { 645 #ifdef INET 646 case IPVERSION: 647 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 648 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 649 break; 650 #endif 651 #ifdef INET6 652 case IPV6_VERSION >> 4: 653 ip6 = mtod(m, struct ip6_hdr *); 654 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 655 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 656 break; 657 #endif 658 default: 659 goto out; 660 break; 661 } 662 return (true); 663 out: 664 m_freem(m); 665 666 return (true); 667 } 668 669 static int 670 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 671 { 672 int error=ENOENT; 673 struct tcp_function_set fs; 674 struct tcp_function_block *blk; 675 676 memset(&fs, 0, sizeof(fs)); 677 rw_rlock(&tcp_function_lock); 678 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 679 if (blk) { 680 /* Found him */ 681 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 682 fs.pcbcnt = blk->tfb_refcnt; 683 } 684 rw_runlock(&tcp_function_lock); 685 error = sysctl_handle_string(oidp, fs.function_set_name, 686 sizeof(fs.function_set_name), req); 687 688 /* Check for error or no change */ 689 if (error != 0 || req->newptr == NULL) 690 return(error); 691 692 rw_wlock(&tcp_function_lock); 693 blk = find_tcp_functions_locked(&fs); 694 if ((blk == NULL) || 695 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 696 error = ENOENT; 697 goto done; 698 } 699 tcp_func_set_ptr = blk; 700 done: 701 rw_wunlock(&tcp_function_lock); 702 return (error); 703 } 704 705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 706 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 707 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 708 "Set/get the default TCP functions"); 709 710 static int 711 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 712 { 713 int error, cnt, linesz; 714 struct tcp_function *f; 715 char *buffer, *cp; 716 size_t bufsz, outsz; 717 bool alias; 718 719 cnt = 0; 720 rw_rlock(&tcp_function_lock); 721 TAILQ_FOREACH(f, &t_functions, tf_next) { 722 cnt++; 723 } 724 rw_runlock(&tcp_function_lock); 725 726 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 727 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 728 729 error = 0; 730 cp = buffer; 731 732 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 733 "Alias", "PCB count"); 734 cp += linesz; 735 bufsz -= linesz; 736 outsz = linesz; 737 738 rw_rlock(&tcp_function_lock); 739 TAILQ_FOREACH(f, &t_functions, tf_next) { 740 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 741 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 742 f->tf_fb->tfb_tcp_block_name, 743 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 744 alias ? f->tf_name : "-", 745 f->tf_fb->tfb_refcnt); 746 if (linesz >= bufsz) { 747 error = EOVERFLOW; 748 break; 749 } 750 cp += linesz; 751 bufsz -= linesz; 752 outsz += linesz; 753 } 754 rw_runlock(&tcp_function_lock); 755 if (error == 0) 756 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 757 free(buffer, M_TEMP); 758 return (error); 759 } 760 761 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 762 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 763 NULL, 0, sysctl_net_inet_list_available, "A", 764 "list available TCP Function sets"); 765 766 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 767 768 #ifdef INET 769 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 770 #define V_udp4_tun_socket VNET(udp4_tun_socket) 771 #endif 772 #ifdef INET6 773 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 774 #define V_udp6_tun_socket VNET(udp6_tun_socket) 775 #endif 776 777 static struct sx tcpoudp_lock; 778 779 static void 780 tcp_over_udp_stop(void) 781 { 782 783 sx_assert(&tcpoudp_lock, SA_XLOCKED); 784 785 #ifdef INET 786 if (V_udp4_tun_socket != NULL) { 787 soclose(V_udp4_tun_socket); 788 V_udp4_tun_socket = NULL; 789 } 790 #endif 791 #ifdef INET6 792 if (V_udp6_tun_socket != NULL) { 793 soclose(V_udp6_tun_socket); 794 V_udp6_tun_socket = NULL; 795 } 796 #endif 797 } 798 799 static int 800 tcp_over_udp_start(void) 801 { 802 uint16_t port; 803 int ret; 804 #ifdef INET 805 struct sockaddr_in sin; 806 #endif 807 #ifdef INET6 808 struct sockaddr_in6 sin6; 809 #endif 810 811 sx_assert(&tcpoudp_lock, SA_XLOCKED); 812 813 port = V_tcp_udp_tunneling_port; 814 if (ntohs(port) == 0) { 815 /* Must have a port set */ 816 return (EINVAL); 817 } 818 #ifdef INET 819 if (V_udp4_tun_socket != NULL) { 820 /* Already running -- must stop first */ 821 return (EALREADY); 822 } 823 #endif 824 #ifdef INET6 825 if (V_udp6_tun_socket != NULL) { 826 /* Already running -- must stop first */ 827 return (EALREADY); 828 } 829 #endif 830 #ifdef INET 831 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 832 SOCK_DGRAM, IPPROTO_UDP, 833 curthread->td_ucred, curthread))) { 834 tcp_over_udp_stop(); 835 return (ret); 836 } 837 /* Call the special UDP hook. */ 838 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 839 tcp_recv_udp_tunneled_packet, 840 tcp_ctlinput_viaudp, 841 NULL))) { 842 tcp_over_udp_stop(); 843 return (ret); 844 } 845 /* Ok, we have a socket, bind it to the port. */ 846 memset(&sin, 0, sizeof(struct sockaddr_in)); 847 sin.sin_len = sizeof(struct sockaddr_in); 848 sin.sin_family = AF_INET; 849 sin.sin_port = htons(port); 850 if ((ret = sobind(V_udp4_tun_socket, 851 (struct sockaddr *)&sin, curthread))) { 852 tcp_over_udp_stop(); 853 return (ret); 854 } 855 #endif 856 #ifdef INET6 857 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 858 SOCK_DGRAM, IPPROTO_UDP, 859 curthread->td_ucred, curthread))) { 860 tcp_over_udp_stop(); 861 return (ret); 862 } 863 /* Call the special UDP hook. */ 864 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 865 tcp_recv_udp_tunneled_packet, 866 tcp6_ctlinput_viaudp, 867 NULL))) { 868 tcp_over_udp_stop(); 869 return (ret); 870 } 871 /* Ok, we have a socket, bind it to the port. */ 872 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 873 sin6.sin6_len = sizeof(struct sockaddr_in6); 874 sin6.sin6_family = AF_INET6; 875 sin6.sin6_port = htons(port); 876 if ((ret = sobind(V_udp6_tun_socket, 877 (struct sockaddr *)&sin6, curthread))) { 878 tcp_over_udp_stop(); 879 return (ret); 880 } 881 #endif 882 return (0); 883 } 884 885 static int 886 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 887 { 888 int error; 889 uint32_t old, new; 890 891 old = V_tcp_udp_tunneling_port; 892 new = old; 893 error = sysctl_handle_int(oidp, &new, 0, req); 894 if ((error == 0) && 895 (req->newptr != NULL)) { 896 if ((new < TCP_TUNNELING_PORT_MIN) || 897 (new > TCP_TUNNELING_PORT_MAX)) { 898 error = EINVAL; 899 } else { 900 sx_xlock(&tcpoudp_lock); 901 V_tcp_udp_tunneling_port = new; 902 if (old != 0) { 903 tcp_over_udp_stop(); 904 } 905 if (new != 0) { 906 error = tcp_over_udp_start(); 907 if (error != 0) { 908 V_tcp_udp_tunneling_port = 0; 909 } 910 } 911 sx_xunlock(&tcpoudp_lock); 912 } 913 } 914 return (error); 915 } 916 917 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 918 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 919 &VNET_NAME(tcp_udp_tunneling_port), 920 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 921 "Tunneling port for tcp over udp"); 922 923 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 924 925 static int 926 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 927 { 928 int error, new; 929 930 new = V_tcp_udp_tunneling_overhead; 931 error = sysctl_handle_int(oidp, &new, 0, req); 932 if (error == 0 && req->newptr) { 933 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 934 (new > TCP_TUNNELING_OVERHEAD_MAX)) 935 error = EINVAL; 936 else 937 V_tcp_udp_tunneling_overhead = new; 938 } 939 return (error); 940 } 941 942 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 943 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 944 &VNET_NAME(tcp_udp_tunneling_overhead), 945 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 946 "MSS reduction when using tcp over udp"); 947 948 /* 949 * Exports one (struct tcp_function_info) for each alias/name. 950 */ 951 static int 952 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 953 { 954 int cnt, error; 955 struct tcp_function *f; 956 struct tcp_function_info tfi; 957 958 /* 959 * We don't allow writes. 960 */ 961 if (req->newptr != NULL) 962 return (EINVAL); 963 964 /* 965 * Wire the old buffer so we can directly copy the functions to 966 * user space without dropping the lock. 967 */ 968 if (req->oldptr != NULL) { 969 error = sysctl_wire_old_buffer(req, 0); 970 if (error) 971 return (error); 972 } 973 974 /* 975 * Walk the list and copy out matching entries. If INVARIANTS 976 * is compiled in, also walk the list to verify the length of 977 * the list matches what we have recorded. 978 */ 979 rw_rlock(&tcp_function_lock); 980 981 cnt = 0; 982 #ifndef INVARIANTS 983 if (req->oldptr == NULL) { 984 cnt = tcp_fb_cnt; 985 goto skip_loop; 986 } 987 #endif 988 TAILQ_FOREACH(f, &t_functions, tf_next) { 989 #ifdef INVARIANTS 990 cnt++; 991 #endif 992 if (req->oldptr != NULL) { 993 bzero(&tfi, sizeof(tfi)); 994 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 995 tfi.tfi_id = f->tf_fb->tfb_id; 996 (void)strlcpy(tfi.tfi_alias, f->tf_name, 997 sizeof(tfi.tfi_alias)); 998 (void)strlcpy(tfi.tfi_name, 999 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 1000 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 1001 /* 1002 * Don't stop on error, as that is the 1003 * mechanism we use to accumulate length 1004 * information if the buffer was too short. 1005 */ 1006 } 1007 } 1008 KASSERT(cnt == tcp_fb_cnt, 1009 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1010 #ifndef INVARIANTS 1011 skip_loop: 1012 #endif 1013 rw_runlock(&tcp_function_lock); 1014 if (req->oldptr == NULL) 1015 error = SYSCTL_OUT(req, NULL, 1016 (cnt + 1) * sizeof(struct tcp_function_info)); 1017 1018 return (error); 1019 } 1020 1021 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1022 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1023 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1024 "List TCP function block name-to-ID mappings"); 1025 1026 /* 1027 * tfb_tcp_handoff_ok() function for the default stack. 1028 * Note that we'll basically try to take all comers. 1029 */ 1030 static int 1031 tcp_default_handoff_ok(struct tcpcb *tp) 1032 { 1033 1034 return (0); 1035 } 1036 1037 /* 1038 * tfb_tcp_fb_init() function for the default stack. 1039 * 1040 * This handles making sure we have appropriate timers set if you are 1041 * transitioning a socket that has some amount of setup done. 1042 * 1043 * The init() fuction from the default can *never* return non-zero i.e. 1044 * it is required to always succeed since it is the stack of last resort! 1045 */ 1046 static int 1047 tcp_default_fb_init(struct tcpcb *tp) 1048 { 1049 struct socket *so = tptosocket(tp); 1050 1051 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1052 1053 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 1054 ("%s: connection %p in unexpected state %d", __func__, tp, 1055 tp->t_state)); 1056 1057 /* 1058 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1059 * know what to do for unexpected states (which includes TIME_WAIT). 1060 */ 1061 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1062 return (0); 1063 1064 /* 1065 * Make sure some kind of transmission timer is set if there is 1066 * outstanding data. 1067 */ 1068 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1069 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1070 tcp_timer_active(tp, TT_PERSIST))) { 1071 /* 1072 * If the session has established and it looks like it should 1073 * be in the persist state, set the persist timer. Otherwise, 1074 * set the retransmit timer. 1075 */ 1076 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1077 (int32_t)(tp->snd_nxt - tp->snd_una) < 1078 (int32_t)sbavail(&so->so_snd)) 1079 tcp_setpersist(tp); 1080 else 1081 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1082 } 1083 1084 /* All non-embryonic sessions get a keepalive timer. */ 1085 if (!tcp_timer_active(tp, TT_KEEP)) 1086 tcp_timer_activate(tp, TT_KEEP, 1087 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1088 TP_KEEPINIT(tp)); 1089 1090 /* 1091 * Make sure critical variables are initialized 1092 * if transitioning while in Recovery. 1093 */ 1094 if IN_FASTRECOVERY(tp->t_flags) { 1095 if (tp->sackhint.recover_fs == 0) 1096 tp->sackhint.recover_fs = max(1, 1097 tp->snd_nxt - tp->snd_una); 1098 } 1099 1100 return (0); 1101 } 1102 1103 /* 1104 * tfb_tcp_fb_fini() function for the default stack. 1105 * 1106 * This changes state as necessary (or prudent) to prepare for another stack 1107 * to assume responsibility for the connection. 1108 */ 1109 static void 1110 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1111 { 1112 1113 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1114 } 1115 1116 /* 1117 * Target size of TCP PCB hash tables. Must be a power of two. 1118 * 1119 * Note that this can be overridden by the kernel environment 1120 * variable net.inet.tcp.tcbhashsize 1121 */ 1122 #ifndef TCBHASHSIZE 1123 #define TCBHASHSIZE 0 1124 #endif 1125 1126 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1127 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1128 1129 static struct mtx isn_mtx; 1130 1131 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1132 #define ISN_LOCK() mtx_lock(&isn_mtx) 1133 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1134 1135 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1136 1137 /* 1138 * Take a value and get the next power of 2 that doesn't overflow. 1139 * Used to size the tcp_inpcb hash buckets. 1140 */ 1141 static int 1142 maketcp_hashsize(int size) 1143 { 1144 int hashsize; 1145 1146 /* 1147 * auto tune. 1148 * get the next power of 2 higher than maxsockets. 1149 */ 1150 hashsize = 1 << fls(size); 1151 /* catch overflow, and just go one power of 2 smaller */ 1152 if (hashsize < size) { 1153 hashsize = 1 << (fls(size) - 1); 1154 } 1155 return (hashsize); 1156 } 1157 1158 static volatile int next_tcp_stack_id = 1; 1159 1160 /* 1161 * Register a TCP function block with the name provided in the names 1162 * array. (Note that this function does NOT automatically register 1163 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1164 * explicitly include blk->tfb_tcp_block_name in the list of names if 1165 * you wish to register the stack with that name.) 1166 * 1167 * Either all name registrations will succeed or all will fail. If 1168 * a name registration fails, the function will update the num_names 1169 * argument to point to the array index of the name that encountered 1170 * the failure. 1171 * 1172 * Returns 0 on success, or an error code on failure. 1173 */ 1174 int 1175 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1176 const char *names[], int *num_names) 1177 { 1178 struct tcp_function *n; 1179 struct tcp_function_set fs; 1180 int error, i; 1181 1182 KASSERT(names != NULL && *num_names > 0, 1183 ("%s: Called with 0-length name list", __func__)); 1184 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1185 KASSERT(rw_initialized(&tcp_function_lock), 1186 ("%s: called too early", __func__)); 1187 1188 if ((blk->tfb_tcp_output == NULL) || 1189 (blk->tfb_tcp_do_segment == NULL) || 1190 (blk->tfb_tcp_ctloutput == NULL) || 1191 (strlen(blk->tfb_tcp_block_name) == 0)) { 1192 /* 1193 * These functions are required and you 1194 * need a name. 1195 */ 1196 *num_names = 0; 1197 return (EINVAL); 1198 } 1199 1200 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1201 *num_names = 0; 1202 return (EINVAL); 1203 } 1204 1205 refcount_init(&blk->tfb_refcnt, 0); 1206 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1207 for (i = 0; i < *num_names; i++) { 1208 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1209 if (n == NULL) { 1210 error = ENOMEM; 1211 goto cleanup; 1212 } 1213 n->tf_fb = blk; 1214 1215 (void)strlcpy(fs.function_set_name, names[i], 1216 sizeof(fs.function_set_name)); 1217 rw_wlock(&tcp_function_lock); 1218 if (find_tcp_functions_locked(&fs) != NULL) { 1219 /* Duplicate name space not allowed */ 1220 rw_wunlock(&tcp_function_lock); 1221 free(n, M_TCPFUNCTIONS); 1222 error = EALREADY; 1223 goto cleanup; 1224 } 1225 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1226 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1227 tcp_fb_cnt++; 1228 rw_wunlock(&tcp_function_lock); 1229 } 1230 return(0); 1231 1232 cleanup: 1233 /* 1234 * Deregister the names we just added. Because registration failed 1235 * for names[i], we don't need to deregister that name. 1236 */ 1237 *num_names = i; 1238 rw_wlock(&tcp_function_lock); 1239 while (--i >= 0) { 1240 TAILQ_FOREACH(n, &t_functions, tf_next) { 1241 if (!strncmp(n->tf_name, names[i], 1242 TCP_FUNCTION_NAME_LEN_MAX)) { 1243 TAILQ_REMOVE(&t_functions, n, tf_next); 1244 tcp_fb_cnt--; 1245 n->tf_fb = NULL; 1246 free(n, M_TCPFUNCTIONS); 1247 break; 1248 } 1249 } 1250 } 1251 rw_wunlock(&tcp_function_lock); 1252 return (error); 1253 } 1254 1255 /* 1256 * Register a TCP function block using the name provided in the name 1257 * argument. 1258 * 1259 * Returns 0 on success, or an error code on failure. 1260 */ 1261 int 1262 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1263 int wait) 1264 { 1265 const char *name_list[1]; 1266 int num_names, rv; 1267 1268 num_names = 1; 1269 if (name != NULL) 1270 name_list[0] = name; 1271 else 1272 name_list[0] = blk->tfb_tcp_block_name; 1273 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1274 return (rv); 1275 } 1276 1277 /* 1278 * Register a TCP function block using the name defined in 1279 * blk->tfb_tcp_block_name. 1280 * 1281 * Returns 0 on success, or an error code on failure. 1282 */ 1283 int 1284 register_tcp_functions(struct tcp_function_block *blk, int wait) 1285 { 1286 1287 return (register_tcp_functions_as_name(blk, NULL, wait)); 1288 } 1289 1290 /* 1291 * Deregister all names associated with a function block. This 1292 * functionally removes the function block from use within the system. 1293 * 1294 * When called with a true quiesce argument, mark the function block 1295 * as being removed so no more stacks will use it and determine 1296 * whether the removal would succeed. 1297 * 1298 * When called with a false quiesce argument, actually attempt the 1299 * removal. 1300 * 1301 * When called with a force argument, attempt to switch all TCBs to 1302 * use the default stack instead of returning EBUSY. 1303 * 1304 * Returns 0 on success (or if the removal would succeed, or an error 1305 * code on failure. 1306 */ 1307 int 1308 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1309 bool force) 1310 { 1311 struct tcp_function *f; 1312 1313 if (blk == &tcp_def_funcblk) { 1314 /* You can't un-register the default */ 1315 return (EPERM); 1316 } 1317 rw_wlock(&tcp_function_lock); 1318 if (blk == tcp_func_set_ptr) { 1319 /* You can't free the current default */ 1320 rw_wunlock(&tcp_function_lock); 1321 return (EBUSY); 1322 } 1323 /* Mark the block so no more stacks can use it. */ 1324 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1325 /* 1326 * If TCBs are still attached to the stack, attempt to switch them 1327 * to the default stack. 1328 */ 1329 if (force && blk->tfb_refcnt) { 1330 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1331 INPLOOKUP_WLOCKPCB); 1332 struct inpcb *inp; 1333 struct tcpcb *tp; 1334 VNET_ITERATOR_DECL(vnet_iter); 1335 1336 rw_wunlock(&tcp_function_lock); 1337 1338 VNET_LIST_RLOCK(); 1339 VNET_FOREACH(vnet_iter) { 1340 CURVNET_SET(vnet_iter); 1341 while ((inp = inp_next(&inpi)) != NULL) { 1342 tp = intotcpcb(inp); 1343 if (tp == NULL || tp->t_fb != blk) 1344 continue; 1345 tcp_switch_back_to_default(tp); 1346 } 1347 CURVNET_RESTORE(); 1348 } 1349 VNET_LIST_RUNLOCK(); 1350 1351 rw_wlock(&tcp_function_lock); 1352 } 1353 if (blk->tfb_refcnt) { 1354 /* TCBs still attached. */ 1355 rw_wunlock(&tcp_function_lock); 1356 return (EBUSY); 1357 } 1358 if (quiesce) { 1359 /* Skip removal. */ 1360 rw_wunlock(&tcp_function_lock); 1361 return (0); 1362 } 1363 /* Remove any function names that map to this function block. */ 1364 while (find_tcp_fb_locked(blk, &f) != NULL) { 1365 TAILQ_REMOVE(&t_functions, f, tf_next); 1366 tcp_fb_cnt--; 1367 f->tf_fb = NULL; 1368 free(f, M_TCPFUNCTIONS); 1369 } 1370 rw_wunlock(&tcp_function_lock); 1371 return (0); 1372 } 1373 1374 static void 1375 tcp_drain(void) 1376 { 1377 struct epoch_tracker et; 1378 VNET_ITERATOR_DECL(vnet_iter); 1379 1380 if (!do_tcpdrain) 1381 return; 1382 1383 NET_EPOCH_ENTER(et); 1384 VNET_LIST_RLOCK_NOSLEEP(); 1385 VNET_FOREACH(vnet_iter) { 1386 CURVNET_SET(vnet_iter); 1387 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1388 INPLOOKUP_WLOCKPCB); 1389 struct inpcb *inpb; 1390 struct tcpcb *tcpb; 1391 1392 /* 1393 * Walk the tcpbs, if existing, and flush the reassembly queue, 1394 * if there is one... 1395 * XXX: The "Net/3" implementation doesn't imply that the TCP 1396 * reassembly queue should be flushed, but in a situation 1397 * where we're really low on mbufs, this is potentially 1398 * useful. 1399 */ 1400 while ((inpb = inp_next(&inpi)) != NULL) { 1401 if ((tcpb = intotcpcb(inpb)) != NULL) { 1402 tcp_reass_flush(tcpb); 1403 tcp_clean_sackreport(tcpb); 1404 #ifdef TCP_BLACKBOX 1405 tcp_log_drain(tcpb); 1406 #endif 1407 #ifdef TCPPCAP 1408 if (tcp_pcap_aggressive_free) { 1409 /* Free the TCP PCAP queues. */ 1410 tcp_pcap_drain(&(tcpb->t_inpkts)); 1411 tcp_pcap_drain(&(tcpb->t_outpkts)); 1412 } 1413 #endif 1414 } 1415 } 1416 CURVNET_RESTORE(); 1417 } 1418 VNET_LIST_RUNLOCK_NOSLEEP(); 1419 NET_EPOCH_EXIT(et); 1420 } 1421 1422 static void 1423 tcp_vnet_init(void *arg __unused) 1424 { 1425 1426 #ifdef TCP_HHOOK 1427 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1428 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1429 printf("%s: WARNING: unable to register helper hook\n", __func__); 1430 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1431 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1432 printf("%s: WARNING: unable to register helper hook\n", __func__); 1433 #endif 1434 #ifdef STATS 1435 if (tcp_stats_init()) 1436 printf("%s: WARNING: unable to initialise TCP stats\n", 1437 __func__); 1438 #endif 1439 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1440 tcp_tcbhashsize); 1441 1442 syncache_init(); 1443 tcp_hc_init(); 1444 1445 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1446 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1447 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1448 1449 tcp_fastopen_init(); 1450 1451 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1452 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1453 1454 V_tcp_msl = TCPTV_MSL; 1455 } 1456 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1457 tcp_vnet_init, NULL); 1458 1459 static void 1460 tcp_init(void *arg __unused) 1461 { 1462 const char *tcbhash_tuneable; 1463 int hashsize; 1464 1465 tcp_reass_global_init(); 1466 1467 /* XXX virtualize those below? */ 1468 tcp_delacktime = TCPTV_DELACK; 1469 tcp_keepinit = TCPTV_KEEP_INIT; 1470 tcp_keepidle = TCPTV_KEEP_IDLE; 1471 tcp_keepintvl = TCPTV_KEEPINTVL; 1472 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1473 tcp_rexmit_initial = TCPTV_RTOBASE; 1474 if (tcp_rexmit_initial < 1) 1475 tcp_rexmit_initial = 1; 1476 tcp_rexmit_min = TCPTV_MIN; 1477 if (tcp_rexmit_min < 1) 1478 tcp_rexmit_min = 1; 1479 tcp_persmin = TCPTV_PERSMIN; 1480 tcp_persmax = TCPTV_PERSMAX; 1481 tcp_rexmit_slop = TCPTV_CPU_VAR; 1482 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1483 1484 /* Setup the tcp function block list */ 1485 TAILQ_INIT(&t_functions); 1486 rw_init(&tcp_function_lock, "tcp_func_lock"); 1487 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1488 sx_init(&tcpoudp_lock, "TCP over UDP configuration"); 1489 #ifdef TCP_BLACKBOX 1490 /* Initialize the TCP logging data. */ 1491 tcp_log_init(); 1492 #endif 1493 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1494 1495 if (tcp_soreceive_stream) { 1496 #ifdef INET 1497 tcp_protosw.pr_soreceive = soreceive_stream; 1498 #endif 1499 #ifdef INET6 1500 tcp6_protosw.pr_soreceive = soreceive_stream; 1501 #endif /* INET6 */ 1502 } 1503 1504 #ifdef INET6 1505 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1506 #else /* INET6 */ 1507 max_protohdr_grow(sizeof(struct tcpiphdr)); 1508 #endif /* INET6 */ 1509 1510 ISN_LOCK_INIT(); 1511 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1512 SHUTDOWN_PRI_DEFAULT); 1513 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1514 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1515 1516 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1517 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1518 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1519 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1520 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1521 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1522 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1523 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1524 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1525 #ifdef TCPPCAP 1526 tcp_pcap_init(); 1527 #endif 1528 1529 hashsize = TCBHASHSIZE; 1530 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1531 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1532 if (hashsize == 0) { 1533 /* 1534 * Auto tune the hash size based on maxsockets. 1535 * A perfect hash would have a 1:1 mapping 1536 * (hashsize = maxsockets) however it's been 1537 * suggested that O(2) average is better. 1538 */ 1539 hashsize = maketcp_hashsize(maxsockets / 4); 1540 /* 1541 * Our historical default is 512, 1542 * do not autotune lower than this. 1543 */ 1544 if (hashsize < 512) 1545 hashsize = 512; 1546 if (bootverbose) 1547 printf("%s: %s auto tuned to %d\n", __func__, 1548 tcbhash_tuneable, hashsize); 1549 } 1550 /* 1551 * We require a hashsize to be a power of two. 1552 * Previously if it was not a power of two we would just reset it 1553 * back to 512, which could be a nasty surprise if you did not notice 1554 * the error message. 1555 * Instead what we do is clip it to the closest power of two lower 1556 * than the specified hash value. 1557 */ 1558 if (!powerof2(hashsize)) { 1559 int oldhashsize = hashsize; 1560 1561 hashsize = maketcp_hashsize(hashsize); 1562 /* prevent absurdly low value */ 1563 if (hashsize < 16) 1564 hashsize = 16; 1565 printf("%s: WARNING: TCB hash size not a power of 2, " 1566 "clipped from %d to %d.\n", __func__, oldhashsize, 1567 hashsize); 1568 } 1569 tcp_tcbhashsize = hashsize; 1570 1571 #ifdef INET 1572 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1573 #endif 1574 #ifdef INET6 1575 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1576 #endif 1577 } 1578 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1579 1580 #ifdef VIMAGE 1581 static void 1582 tcp_destroy(void *unused __unused) 1583 { 1584 int n; 1585 #ifdef TCP_HHOOK 1586 int error; 1587 #endif 1588 1589 /* 1590 * All our processes are gone, all our sockets should be cleaned 1591 * up, which means, we should be past the tcp_discardcb() calls. 1592 * Sleep to let all tcpcb timers really disappear and cleanup. 1593 */ 1594 for (;;) { 1595 INP_INFO_WLOCK(&V_tcbinfo); 1596 n = V_tcbinfo.ipi_count; 1597 INP_INFO_WUNLOCK(&V_tcbinfo); 1598 if (n == 0) 1599 break; 1600 pause("tcpdes", hz / 10); 1601 } 1602 tcp_hc_destroy(); 1603 syncache_destroy(); 1604 in_pcbinfo_destroy(&V_tcbinfo); 1605 /* tcp_discardcb() clears the sack_holes up. */ 1606 uma_zdestroy(V_sack_hole_zone); 1607 1608 /* 1609 * Cannot free the zone until all tcpcbs are released as we attach 1610 * the allocations to them. 1611 */ 1612 tcp_fastopen_destroy(); 1613 1614 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1615 VNET_PCPUSTAT_FREE(tcpstat); 1616 1617 #ifdef TCP_HHOOK 1618 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1619 if (error != 0) { 1620 printf("%s: WARNING: unable to deregister helper hook " 1621 "type=%d, id=%d: error %d returned\n", __func__, 1622 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1623 } 1624 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1625 if (error != 0) { 1626 printf("%s: WARNING: unable to deregister helper hook " 1627 "type=%d, id=%d: error %d returned\n", __func__, 1628 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1629 } 1630 #endif 1631 } 1632 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1633 #endif 1634 1635 void 1636 tcp_fini(void *xtp) 1637 { 1638 1639 } 1640 1641 /* 1642 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1643 * tcp_template used to store this data in mbufs, but we now recopy it out 1644 * of the tcpcb each time to conserve mbufs. 1645 */ 1646 void 1647 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1648 { 1649 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1650 1651 INP_WLOCK_ASSERT(inp); 1652 1653 #ifdef INET6 1654 if ((inp->inp_vflag & INP_IPV6) != 0) { 1655 struct ip6_hdr *ip6; 1656 1657 ip6 = (struct ip6_hdr *)ip_ptr; 1658 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1659 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1660 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1661 (IPV6_VERSION & IPV6_VERSION_MASK); 1662 if (port == 0) 1663 ip6->ip6_nxt = IPPROTO_TCP; 1664 else 1665 ip6->ip6_nxt = IPPROTO_UDP; 1666 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1667 ip6->ip6_src = inp->in6p_laddr; 1668 ip6->ip6_dst = inp->in6p_faddr; 1669 } 1670 #endif /* INET6 */ 1671 #if defined(INET6) && defined(INET) 1672 else 1673 #endif 1674 #ifdef INET 1675 { 1676 struct ip *ip; 1677 1678 ip = (struct ip *)ip_ptr; 1679 ip->ip_v = IPVERSION; 1680 ip->ip_hl = 5; 1681 ip->ip_tos = inp->inp_ip_tos; 1682 ip->ip_len = 0; 1683 ip->ip_id = 0; 1684 ip->ip_off = 0; 1685 ip->ip_ttl = inp->inp_ip_ttl; 1686 ip->ip_sum = 0; 1687 if (port == 0) 1688 ip->ip_p = IPPROTO_TCP; 1689 else 1690 ip->ip_p = IPPROTO_UDP; 1691 ip->ip_src = inp->inp_laddr; 1692 ip->ip_dst = inp->inp_faddr; 1693 } 1694 #endif /* INET */ 1695 th->th_sport = inp->inp_lport; 1696 th->th_dport = inp->inp_fport; 1697 th->th_seq = 0; 1698 th->th_ack = 0; 1699 th->th_off = 5; 1700 tcp_set_flags(th, 0); 1701 th->th_win = 0; 1702 th->th_urp = 0; 1703 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1704 } 1705 1706 /* 1707 * Create template to be used to send tcp packets on a connection. 1708 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1709 * use for this function is in keepalives, which use tcp_respond. 1710 */ 1711 struct tcptemp * 1712 tcpip_maketemplate(struct inpcb *inp) 1713 { 1714 struct tcptemp *t; 1715 1716 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1717 if (t == NULL) 1718 return (NULL); 1719 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1720 return (t); 1721 } 1722 1723 /* 1724 * Send a single message to the TCP at address specified by 1725 * the given TCP/IP header. If m == NULL, then we make a copy 1726 * of the tcpiphdr at th and send directly to the addressed host. 1727 * This is used to force keep alive messages out using the TCP 1728 * template for a connection. If flags are given then we send 1729 * a message back to the TCP which originated the segment th, 1730 * and discard the mbuf containing it and any other attached mbufs. 1731 * 1732 * In any case the ack and sequence number of the transmitted 1733 * segment are as specified by the parameters. 1734 * 1735 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1736 */ 1737 void 1738 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1739 tcp_seq ack, tcp_seq seq, uint16_t flags) 1740 { 1741 struct tcpopt to; 1742 struct inpcb *inp; 1743 struct ip *ip; 1744 struct mbuf *optm; 1745 struct udphdr *uh = NULL; 1746 struct tcphdr *nth; 1747 struct tcp_log_buffer *lgb; 1748 u_char *optp; 1749 #ifdef INET6 1750 struct ip6_hdr *ip6; 1751 int isipv6; 1752 #endif /* INET6 */ 1753 int optlen, tlen, win, ulen; 1754 int ect = 0; 1755 bool incl_opts; 1756 uint16_t port; 1757 int output_ret; 1758 #ifdef INVARIANTS 1759 int thflags = tcp_get_flags(th); 1760 #endif 1761 1762 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1763 NET_EPOCH_ASSERT(); 1764 1765 #ifdef INET6 1766 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1767 ip6 = ipgen; 1768 #endif /* INET6 */ 1769 ip = ipgen; 1770 1771 if (tp != NULL) { 1772 inp = tptoinpcb(tp); 1773 INP_LOCK_ASSERT(inp); 1774 } else 1775 inp = NULL; 1776 1777 if (m != NULL) { 1778 #ifdef INET6 1779 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1780 port = m->m_pkthdr.tcp_tun_port; 1781 else 1782 #endif 1783 if (ip && (ip->ip_p == IPPROTO_UDP)) 1784 port = m->m_pkthdr.tcp_tun_port; 1785 else 1786 port = 0; 1787 } else 1788 port = tp->t_port; 1789 1790 incl_opts = false; 1791 win = 0; 1792 if (tp != NULL) { 1793 if (!(flags & TH_RST)) { 1794 win = sbspace(&inp->inp_socket->so_rcv); 1795 if (win > TCP_MAXWIN << tp->rcv_scale) 1796 win = TCP_MAXWIN << tp->rcv_scale; 1797 } 1798 if ((tp->t_flags & TF_NOOPT) == 0) 1799 incl_opts = true; 1800 } 1801 if (m == NULL) { 1802 m = m_gethdr(M_NOWAIT, MT_DATA); 1803 if (m == NULL) 1804 return; 1805 m->m_data += max_linkhdr; 1806 #ifdef INET6 1807 if (isipv6) { 1808 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1809 sizeof(struct ip6_hdr)); 1810 ip6 = mtod(m, struct ip6_hdr *); 1811 nth = (struct tcphdr *)(ip6 + 1); 1812 if (port) { 1813 /* Insert a UDP header */ 1814 uh = (struct udphdr *)nth; 1815 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1816 uh->uh_dport = port; 1817 nth = (struct tcphdr *)(uh + 1); 1818 } 1819 } else 1820 #endif /* INET6 */ 1821 { 1822 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1823 ip = mtod(m, struct ip *); 1824 nth = (struct tcphdr *)(ip + 1); 1825 if (port) { 1826 /* Insert a UDP header */ 1827 uh = (struct udphdr *)nth; 1828 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1829 uh->uh_dport = port; 1830 nth = (struct tcphdr *)(uh + 1); 1831 } 1832 } 1833 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1834 flags = TH_ACK; 1835 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1836 struct mbuf *n; 1837 1838 /* Can't reuse 'm', allocate a new mbuf. */ 1839 n = m_gethdr(M_NOWAIT, MT_DATA); 1840 if (n == NULL) { 1841 m_freem(m); 1842 return; 1843 } 1844 1845 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1846 m_freem(m); 1847 m_freem(n); 1848 return; 1849 } 1850 1851 n->m_data += max_linkhdr; 1852 /* m_len is set later */ 1853 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1854 #ifdef INET6 1855 if (isipv6) { 1856 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1857 sizeof(struct ip6_hdr)); 1858 ip6 = mtod(n, struct ip6_hdr *); 1859 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1860 nth = (struct tcphdr *)(ip6 + 1); 1861 if (port) { 1862 /* Insert a UDP header */ 1863 uh = (struct udphdr *)nth; 1864 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1865 uh->uh_dport = port; 1866 nth = (struct tcphdr *)(uh + 1); 1867 } 1868 } else 1869 #endif /* INET6 */ 1870 { 1871 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1872 ip = mtod(n, struct ip *); 1873 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1874 nth = (struct tcphdr *)(ip + 1); 1875 if (port) { 1876 /* Insert a UDP header */ 1877 uh = (struct udphdr *)nth; 1878 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1879 uh->uh_dport = port; 1880 nth = (struct tcphdr *)(uh + 1); 1881 } 1882 } 1883 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1884 xchg(nth->th_dport, nth->th_sport, uint16_t); 1885 th = nth; 1886 m_freem(m); 1887 m = n; 1888 } else { 1889 /* 1890 * reuse the mbuf. 1891 * XXX MRT We inherit the FIB, which is lucky. 1892 */ 1893 m_freem(m->m_next); 1894 m->m_next = NULL; 1895 m->m_data = (caddr_t)ipgen; 1896 /* m_len is set later */ 1897 #ifdef INET6 1898 if (isipv6) { 1899 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1900 nth = (struct tcphdr *)(ip6 + 1); 1901 } else 1902 #endif /* INET6 */ 1903 { 1904 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1905 nth = (struct tcphdr *)(ip + 1); 1906 } 1907 if (th != nth) { 1908 /* 1909 * this is usually a case when an extension header 1910 * exists between the IPv6 header and the 1911 * TCP header. 1912 */ 1913 nth->th_sport = th->th_sport; 1914 nth->th_dport = th->th_dport; 1915 } 1916 xchg(nth->th_dport, nth->th_sport, uint16_t); 1917 #undef xchg 1918 } 1919 tlen = 0; 1920 #ifdef INET6 1921 if (isipv6) 1922 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1923 #endif 1924 #if defined(INET) && defined(INET6) 1925 else 1926 #endif 1927 #ifdef INET 1928 tlen = sizeof (struct tcpiphdr); 1929 #endif 1930 if (port) 1931 tlen += sizeof (struct udphdr); 1932 #ifdef INVARIANTS 1933 m->m_len = 0; 1934 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1935 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1936 m, tlen, (long)M_TRAILINGSPACE(m))); 1937 #endif 1938 m->m_len = tlen; 1939 to.to_flags = 0; 1940 if (incl_opts) { 1941 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1942 /* Make sure we have room. */ 1943 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1944 m->m_next = m_get(M_NOWAIT, MT_DATA); 1945 if (m->m_next) { 1946 optp = mtod(m->m_next, u_char *); 1947 optm = m->m_next; 1948 } else 1949 incl_opts = false; 1950 } else { 1951 optp = (u_char *) (nth + 1); 1952 optm = m; 1953 } 1954 } 1955 if (incl_opts) { 1956 /* Timestamps. */ 1957 if (tp->t_flags & TF_RCVD_TSTMP) { 1958 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1959 to.to_tsecr = tp->ts_recent; 1960 to.to_flags |= TOF_TS; 1961 } 1962 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1963 /* TCP-MD5 (RFC2385). */ 1964 if (tp->t_flags & TF_SIGNATURE) 1965 to.to_flags |= TOF_SIGNATURE; 1966 #endif 1967 /* Add the options. */ 1968 tlen += optlen = tcp_addoptions(&to, optp); 1969 1970 /* Update m_len in the correct mbuf. */ 1971 optm->m_len += optlen; 1972 } else 1973 optlen = 0; 1974 #ifdef INET6 1975 if (isipv6) { 1976 if (uh) { 1977 ulen = tlen - sizeof(struct ip6_hdr); 1978 uh->uh_ulen = htons(ulen); 1979 } 1980 ip6->ip6_flow = htonl(ect << 20); 1981 ip6->ip6_vfc = IPV6_VERSION; 1982 if (port) 1983 ip6->ip6_nxt = IPPROTO_UDP; 1984 else 1985 ip6->ip6_nxt = IPPROTO_TCP; 1986 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1987 } 1988 #endif 1989 #if defined(INET) && defined(INET6) 1990 else 1991 #endif 1992 #ifdef INET 1993 { 1994 if (uh) { 1995 ulen = tlen - sizeof(struct ip); 1996 uh->uh_ulen = htons(ulen); 1997 } 1998 ip->ip_tos = ect; 1999 ip->ip_len = htons(tlen); 2000 ip->ip_ttl = V_ip_defttl; 2001 if (port) { 2002 ip->ip_p = IPPROTO_UDP; 2003 } else { 2004 ip->ip_p = IPPROTO_TCP; 2005 } 2006 if (V_path_mtu_discovery) 2007 ip->ip_off |= htons(IP_DF); 2008 } 2009 #endif 2010 m->m_pkthdr.len = tlen; 2011 m->m_pkthdr.rcvif = NULL; 2012 #ifdef MAC 2013 if (inp != NULL) { 2014 /* 2015 * Packet is associated with a socket, so allow the 2016 * label of the response to reflect the socket label. 2017 */ 2018 INP_LOCK_ASSERT(inp); 2019 mac_inpcb_create_mbuf(inp, m); 2020 } else { 2021 /* 2022 * Packet is not associated with a socket, so possibly 2023 * update the label in place. 2024 */ 2025 mac_netinet_tcp_reply(m); 2026 } 2027 #endif 2028 nth->th_seq = htonl(seq); 2029 nth->th_ack = htonl(ack); 2030 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2031 tcp_set_flags(nth, flags); 2032 if (tp != NULL) 2033 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2034 else 2035 nth->th_win = htons((u_short)win); 2036 nth->th_urp = 0; 2037 2038 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2039 if (to.to_flags & TOF_SIGNATURE) { 2040 if (!TCPMD5_ENABLED() || 2041 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2042 m_freem(m); 2043 return; 2044 } 2045 } 2046 #endif 2047 2048 #ifdef INET6 2049 if (isipv6) { 2050 if (port) { 2051 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2052 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2053 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2054 nth->th_sum = 0; 2055 } else { 2056 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2057 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2058 nth->th_sum = in6_cksum_pseudo(ip6, 2059 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2060 } 2061 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2062 } 2063 #endif /* INET6 */ 2064 #if defined(INET6) && defined(INET) 2065 else 2066 #endif 2067 #ifdef INET 2068 { 2069 if (port) { 2070 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2071 htons(ulen + IPPROTO_UDP)); 2072 m->m_pkthdr.csum_flags = CSUM_UDP; 2073 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2074 nth->th_sum = 0; 2075 } else { 2076 m->m_pkthdr.csum_flags = CSUM_TCP; 2077 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2078 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2079 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2080 } 2081 } 2082 #endif /* INET */ 2083 TCP_PROBE3(debug__output, tp, th, m); 2084 if (flags & TH_RST) 2085 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2086 lgb = NULL; 2087 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { 2088 if (INP_WLOCKED(inp)) { 2089 union tcp_log_stackspecific log; 2090 struct timeval tv; 2091 2092 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2093 log.u_bbr.inhpts = inp->inp_in_hpts; 2094 log.u_bbr.flex8 = 4; 2095 log.u_bbr.pkts_out = tp->t_maxseg; 2096 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2097 log.u_bbr.delivered = 0; 2098 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, 2099 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2100 } else { 2101 /* 2102 * We can not log the packet, since we only own the 2103 * read lock, but a write lock is needed. The read lock 2104 * is not upgraded to a write lock, since only getting 2105 * the read lock was done intentionally to improve the 2106 * handling of SYN flooding attacks. 2107 * This happens only for pure SYN segments received in 2108 * the initial CLOSED state, or received in a more 2109 * advanced state than listen and the UDP encapsulation 2110 * port is unexpected. 2111 * The incoming SYN segments do not really belong to 2112 * the TCP connection and the handling does not change 2113 * the state of the TCP connection. Therefore, the 2114 * sending of the RST segments is not logged. Please 2115 * note that also the incoming SYN segments are not 2116 * logged. 2117 * 2118 * The following code ensures that the above description 2119 * is and stays correct. 2120 */ 2121 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2122 (tp->t_state == TCPS_CLOSED || 2123 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2124 ("%s: Logging of TCP segment with flags 0x%b and " 2125 "UDP encapsulation port %u skipped in state %s", 2126 __func__, thflags, PRINT_TH_FLAGS, 2127 ntohs(port), tcpstates[tp->t_state])); 2128 } 2129 } 2130 2131 if (flags & TH_ACK) 2132 TCPSTAT_INC(tcps_sndacks); 2133 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2134 TCPSTAT_INC(tcps_sndctrl); 2135 TCPSTAT_INC(tcps_sndtotal); 2136 2137 #ifdef INET6 2138 if (isipv6) { 2139 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2140 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2141 } 2142 #endif /* INET6 */ 2143 #if defined(INET) && defined(INET6) 2144 else 2145 #endif 2146 #ifdef INET 2147 { 2148 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2149 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2150 } 2151 #endif 2152 if (lgb != NULL) 2153 lgb->tlb_errno = output_ret; 2154 } 2155 2156 /* 2157 * Create a new TCP control block, making an empty reassembly queue and hooking 2158 * it to the argument protocol control block. The `inp' parameter must have 2159 * come from the zone allocator set up by tcpcbstor declaration. 2160 */ 2161 struct tcpcb * 2162 tcp_newtcpcb(struct inpcb *inp) 2163 { 2164 struct tcpcb *tp = intotcpcb(inp); 2165 #ifdef INET6 2166 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2167 #endif /* INET6 */ 2168 2169 /* 2170 * Historically allocation was done with M_ZERO. There is a lot of 2171 * code that rely on that. For now take safe approach and zero whole 2172 * tcpcb. This definitely can be optimized. 2173 */ 2174 bzero(&tp->t_start_zero, t_zero_size); 2175 2176 /* Initialise cc_var struct for this tcpcb. */ 2177 tp->t_ccv.type = IPPROTO_TCP; 2178 tp->t_ccv.ccvc.tcp = tp; 2179 rw_rlock(&tcp_function_lock); 2180 tp->t_fb = tcp_func_set_ptr; 2181 refcount_acquire(&tp->t_fb->tfb_refcnt); 2182 rw_runlock(&tcp_function_lock); 2183 /* 2184 * Use the current system default CC algorithm. 2185 */ 2186 cc_attach(tp, CC_DEFAULT_ALGO()); 2187 2188 if (CC_ALGO(tp)->cb_init != NULL) 2189 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2190 cc_detach(tp); 2191 if (tp->t_fb->tfb_tcp_fb_fini) 2192 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2193 refcount_release(&tp->t_fb->tfb_refcnt); 2194 return (NULL); 2195 } 2196 2197 #ifdef TCP_HHOOK 2198 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2199 if (tp->t_fb->tfb_tcp_fb_fini) 2200 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2201 refcount_release(&tp->t_fb->tfb_refcnt); 2202 return (NULL); 2203 } 2204 #endif 2205 2206 TAILQ_INIT(&tp->t_segq); 2207 tp->t_maxseg = 2208 #ifdef INET6 2209 isipv6 ? V_tcp_v6mssdflt : 2210 #endif /* INET6 */ 2211 V_tcp_mssdflt; 2212 2213 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED); 2214 for (int i = 0; i < TT_N; i++) 2215 tp->t_timers[i] = SBT_MAX; 2216 2217 switch (V_tcp_do_rfc1323) { 2218 case 0: 2219 break; 2220 default: 2221 case 1: 2222 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2223 break; 2224 case 2: 2225 tp->t_flags = TF_REQ_SCALE; 2226 break; 2227 case 3: 2228 tp->t_flags = TF_REQ_TSTMP; 2229 break; 2230 } 2231 if (V_tcp_do_sack) 2232 tp->t_flags |= TF_SACK_PERMIT; 2233 TAILQ_INIT(&tp->snd_holes); 2234 2235 /* 2236 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2237 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2238 * reasonable initial retransmit time. 2239 */ 2240 tp->t_srtt = TCPTV_SRTTBASE; 2241 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2242 tp->t_rttmin = tcp_rexmit_min; 2243 tp->t_rxtcur = tcp_rexmit_initial; 2244 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2245 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2246 tp->t_rcvtime = ticks; 2247 /* 2248 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2249 * because the socket may be bound to an IPv6 wildcard address, 2250 * which may match an IPv4-mapped IPv6 address. 2251 */ 2252 inp->inp_ip_ttl = V_ip_defttl; 2253 #ifdef TCPHPTS 2254 /* 2255 * If using hpts lets drop a random number in so 2256 * not all new connections fall on the same CPU. 2257 */ 2258 inp->inp_hpts_cpu = hpts_random_cpu(inp); 2259 #endif 2260 #ifdef TCPPCAP 2261 /* 2262 * Init the TCP PCAP queues. 2263 */ 2264 tcp_pcap_tcpcb_init(tp); 2265 #endif 2266 #ifdef TCP_BLACKBOX 2267 /* Initialize the per-TCPCB log data. */ 2268 tcp_log_tcpcbinit(tp); 2269 #endif 2270 tp->t_pacing_rate = -1; 2271 if (tp->t_fb->tfb_tcp_fb_init) { 2272 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2273 refcount_release(&tp->t_fb->tfb_refcnt); 2274 return (NULL); 2275 } 2276 } 2277 #ifdef STATS 2278 if (V_tcp_perconn_stats_enable == 1) 2279 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2280 #endif 2281 if (V_tcp_do_lrd) 2282 tp->t_flags |= TF_LRD; 2283 2284 return (tp); 2285 } 2286 2287 /* 2288 * Drop a TCP connection, reporting 2289 * the specified error. If connection is synchronized, 2290 * then send a RST to peer. 2291 */ 2292 struct tcpcb * 2293 tcp_drop(struct tcpcb *tp, int errno) 2294 { 2295 struct socket *so = tptosocket(tp); 2296 2297 NET_EPOCH_ASSERT(); 2298 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2299 2300 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2301 tcp_state_change(tp, TCPS_CLOSED); 2302 /* Don't use tcp_output() here due to possible recursion. */ 2303 (void)tcp_output_nodrop(tp); 2304 TCPSTAT_INC(tcps_drops); 2305 } else 2306 TCPSTAT_INC(tcps_conndrops); 2307 if (errno == ETIMEDOUT && tp->t_softerror) 2308 errno = tp->t_softerror; 2309 so->so_error = errno; 2310 return (tcp_close(tp)); 2311 } 2312 2313 void 2314 tcp_discardcb(struct tcpcb *tp) 2315 { 2316 struct inpcb *inp = tptoinpcb(tp); 2317 struct socket *so = tptosocket(tp); 2318 #ifdef INET6 2319 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2320 #endif 2321 2322 INP_WLOCK_ASSERT(inp); 2323 2324 tcp_timer_stop(tp); 2325 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2326 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2327 } 2328 2329 /* free the reassembly queue, if any */ 2330 tcp_reass_flush(tp); 2331 2332 #ifdef TCP_OFFLOAD 2333 /* Disconnect offload device, if any. */ 2334 if (tp->t_flags & TF_TOE) 2335 tcp_offload_detach(tp); 2336 #endif 2337 2338 tcp_free_sackholes(tp); 2339 2340 #ifdef TCPPCAP 2341 /* Free the TCP PCAP queues. */ 2342 tcp_pcap_drain(&(tp->t_inpkts)); 2343 tcp_pcap_drain(&(tp->t_outpkts)); 2344 #endif 2345 2346 /* Allow the CC algorithm to clean up after itself. */ 2347 if (CC_ALGO(tp)->cb_destroy != NULL) 2348 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2349 CC_DATA(tp) = NULL; 2350 /* Detach from the CC algorithm */ 2351 cc_detach(tp); 2352 2353 #ifdef TCP_HHOOK 2354 khelp_destroy_osd(&tp->t_osd); 2355 #endif 2356 #ifdef STATS 2357 stats_blob_destroy(tp->t_stats); 2358 #endif 2359 2360 CC_ALGO(tp) = NULL; 2361 2362 #ifdef TCP_BLACKBOX 2363 tcp_log_tcpcbfini(tp); 2364 #endif 2365 TCPSTATES_DEC(tp->t_state); 2366 if (tp->t_fb->tfb_tcp_fb_fini) 2367 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2368 2369 /* 2370 * If we got enough samples through the srtt filter, 2371 * save the rtt and rttvar in the routing entry. 2372 * 'Enough' is arbitrarily defined as 4 rtt samples. 2373 * 4 samples is enough for the srtt filter to converge 2374 * to within enough % of the correct value; fewer samples 2375 * and we could save a bogus rtt. The danger is not high 2376 * as tcp quickly recovers from everything. 2377 * XXX: Works very well but needs some more statistics! 2378 * 2379 * XXXRRS: Updating must be after the stack fini() since 2380 * that may be converting some internal representation of 2381 * say srtt etc into the general one used by other stacks. 2382 * Lets also at least protect against the so being NULL 2383 * as RW stated below. 2384 */ 2385 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2386 struct hc_metrics_lite metrics; 2387 uint32_t ssthresh; 2388 2389 bzero(&metrics, sizeof(metrics)); 2390 /* 2391 * Update the ssthresh always when the conditions below 2392 * are satisfied. This gives us better new start value 2393 * for the congestion avoidance for new connections. 2394 * ssthresh is only set if packet loss occurred on a session. 2395 * 2396 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2397 * being torn down. Ideally this code would not use 'so'. 2398 */ 2399 ssthresh = tp->snd_ssthresh; 2400 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2401 /* 2402 * convert the limit from user data bytes to 2403 * packets then to packet data bytes. 2404 */ 2405 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2406 if (ssthresh < 2) 2407 ssthresh = 2; 2408 ssthresh *= (tp->t_maxseg + 2409 #ifdef INET6 2410 (isipv6 ? sizeof (struct ip6_hdr) + 2411 sizeof (struct tcphdr) : 2412 #endif 2413 sizeof (struct tcpiphdr) 2414 #ifdef INET6 2415 ) 2416 #endif 2417 ); 2418 } else 2419 ssthresh = 0; 2420 metrics.rmx_ssthresh = ssthresh; 2421 2422 metrics.rmx_rtt = tp->t_srtt; 2423 metrics.rmx_rttvar = tp->t_rttvar; 2424 metrics.rmx_cwnd = tp->snd_cwnd; 2425 metrics.rmx_sendpipe = 0; 2426 metrics.rmx_recvpipe = 0; 2427 2428 tcp_hc_update(&inp->inp_inc, &metrics); 2429 } 2430 2431 refcount_release(&tp->t_fb->tfb_refcnt); 2432 } 2433 2434 /* 2435 * Attempt to close a TCP control block, marking it as dropped, and freeing 2436 * the socket if we hold the only reference. 2437 */ 2438 struct tcpcb * 2439 tcp_close(struct tcpcb *tp) 2440 { 2441 struct inpcb *inp = tptoinpcb(tp); 2442 struct socket *so = tptosocket(tp); 2443 2444 INP_WLOCK_ASSERT(inp); 2445 2446 #ifdef TCP_OFFLOAD 2447 if (tp->t_state == TCPS_LISTEN) 2448 tcp_offload_listen_stop(tp); 2449 #endif 2450 /* 2451 * This releases the TFO pending counter resource for TFO listen 2452 * sockets as well as passively-created TFO sockets that transition 2453 * from SYN_RECEIVED to CLOSED. 2454 */ 2455 if (tp->t_tfo_pending) { 2456 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2457 tp->t_tfo_pending = NULL; 2458 } 2459 #ifdef TCPHPTS 2460 tcp_hpts_remove(inp); 2461 #endif 2462 in_pcbdrop(inp); 2463 TCPSTAT_INC(tcps_closed); 2464 if (tp->t_state != TCPS_CLOSED) 2465 tcp_state_change(tp, TCPS_CLOSED); 2466 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2467 soisdisconnected(so); 2468 if (inp->inp_flags & INP_SOCKREF) { 2469 inp->inp_flags &= ~INP_SOCKREF; 2470 INP_WUNLOCK(inp); 2471 sorele(so); 2472 return (NULL); 2473 } 2474 return (tp); 2475 } 2476 2477 /* 2478 * Notify a tcp user of an asynchronous error; 2479 * store error as soft error, but wake up user 2480 * (for now, won't do anything until can select for soft error). 2481 * 2482 * Do not wake up user since there currently is no mechanism for 2483 * reporting soft errors (yet - a kqueue filter may be added). 2484 */ 2485 static struct inpcb * 2486 tcp_notify(struct inpcb *inp, int error) 2487 { 2488 struct tcpcb *tp; 2489 2490 INP_WLOCK_ASSERT(inp); 2491 2492 tp = intotcpcb(inp); 2493 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2494 2495 /* 2496 * Ignore some errors if we are hooked up. 2497 * If connection hasn't completed, has retransmitted several times, 2498 * and receives a second error, give up now. This is better 2499 * than waiting a long time to establish a connection that 2500 * can never complete. 2501 */ 2502 if (tp->t_state == TCPS_ESTABLISHED && 2503 (error == EHOSTUNREACH || error == ENETUNREACH || 2504 error == EHOSTDOWN)) { 2505 if (inp->inp_route.ro_nh) { 2506 NH_FREE(inp->inp_route.ro_nh); 2507 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2508 } 2509 return (inp); 2510 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2511 tp->t_softerror) { 2512 tp = tcp_drop(tp, error); 2513 if (tp != NULL) 2514 return (inp); 2515 else 2516 return (NULL); 2517 } else { 2518 tp->t_softerror = error; 2519 return (inp); 2520 } 2521 #if 0 2522 wakeup( &so->so_timeo); 2523 sorwakeup(so); 2524 sowwakeup(so); 2525 #endif 2526 } 2527 2528 static int 2529 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2530 { 2531 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2532 INPLOOKUP_RLOCKPCB); 2533 struct xinpgen xig; 2534 struct inpcb *inp; 2535 int error; 2536 2537 if (req->newptr != NULL) 2538 return (EPERM); 2539 2540 if (req->oldptr == NULL) { 2541 int n; 2542 2543 n = V_tcbinfo.ipi_count + 2544 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2545 n += imax(n / 8, 10); 2546 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2547 return (0); 2548 } 2549 2550 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2551 return (error); 2552 2553 bzero(&xig, sizeof(xig)); 2554 xig.xig_len = sizeof xig; 2555 xig.xig_count = V_tcbinfo.ipi_count + 2556 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2557 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2558 xig.xig_sogen = so_gencnt; 2559 error = SYSCTL_OUT(req, &xig, sizeof xig); 2560 if (error) 2561 return (error); 2562 2563 error = syncache_pcblist(req); 2564 if (error) 2565 return (error); 2566 2567 while ((inp = inp_next(&inpi)) != NULL) { 2568 if (inp->inp_gencnt <= xig.xig_gen && 2569 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2570 struct xtcpcb xt; 2571 2572 tcp_inptoxtp(inp, &xt); 2573 error = SYSCTL_OUT(req, &xt, sizeof xt); 2574 if (error) { 2575 INP_RUNLOCK(inp); 2576 break; 2577 } else 2578 continue; 2579 } 2580 } 2581 2582 if (!error) { 2583 /* 2584 * Give the user an updated idea of our state. 2585 * If the generation differs from what we told 2586 * her before, she knows that something happened 2587 * while we were processing this request, and it 2588 * might be necessary to retry. 2589 */ 2590 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2591 xig.xig_sogen = so_gencnt; 2592 xig.xig_count = V_tcbinfo.ipi_count + 2593 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2594 error = SYSCTL_OUT(req, &xig, sizeof xig); 2595 } 2596 2597 return (error); 2598 } 2599 2600 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2601 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2602 NULL, 0, tcp_pcblist, "S,xtcpcb", 2603 "List of active TCP connections"); 2604 2605 #ifdef INET 2606 static int 2607 tcp_getcred(SYSCTL_HANDLER_ARGS) 2608 { 2609 struct xucred xuc; 2610 struct sockaddr_in addrs[2]; 2611 struct epoch_tracker et; 2612 struct inpcb *inp; 2613 int error; 2614 2615 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2616 if (error) 2617 return (error); 2618 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2619 if (error) 2620 return (error); 2621 NET_EPOCH_ENTER(et); 2622 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2623 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2624 NET_EPOCH_EXIT(et); 2625 if (inp != NULL) { 2626 if (error == 0) 2627 error = cr_canseeinpcb(req->td->td_ucred, inp); 2628 if (error == 0) 2629 cru2x(inp->inp_cred, &xuc); 2630 INP_RUNLOCK(inp); 2631 } else 2632 error = ENOENT; 2633 if (error == 0) 2634 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2635 return (error); 2636 } 2637 2638 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2639 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2640 0, 0, tcp_getcred, "S,xucred", 2641 "Get the xucred of a TCP connection"); 2642 #endif /* INET */ 2643 2644 #ifdef INET6 2645 static int 2646 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2647 { 2648 struct epoch_tracker et; 2649 struct xucred xuc; 2650 struct sockaddr_in6 addrs[2]; 2651 struct inpcb *inp; 2652 int error; 2653 #ifdef INET 2654 int mapped = 0; 2655 #endif 2656 2657 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2658 if (error) 2659 return (error); 2660 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2661 if (error) 2662 return (error); 2663 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2664 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2665 return (error); 2666 } 2667 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2668 #ifdef INET 2669 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2670 mapped = 1; 2671 else 2672 #endif 2673 return (EINVAL); 2674 } 2675 2676 NET_EPOCH_ENTER(et); 2677 #ifdef INET 2678 if (mapped == 1) 2679 inp = in_pcblookup(&V_tcbinfo, 2680 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2681 addrs[1].sin6_port, 2682 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2683 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2684 else 2685 #endif 2686 inp = in6_pcblookup(&V_tcbinfo, 2687 &addrs[1].sin6_addr, addrs[1].sin6_port, 2688 &addrs[0].sin6_addr, addrs[0].sin6_port, 2689 INPLOOKUP_RLOCKPCB, NULL); 2690 NET_EPOCH_EXIT(et); 2691 if (inp != NULL) { 2692 if (error == 0) 2693 error = cr_canseeinpcb(req->td->td_ucred, inp); 2694 if (error == 0) 2695 cru2x(inp->inp_cred, &xuc); 2696 INP_RUNLOCK(inp); 2697 } else 2698 error = ENOENT; 2699 if (error == 0) 2700 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2701 return (error); 2702 } 2703 2704 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2705 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2706 0, 0, tcp6_getcred, "S,xucred", 2707 "Get the xucred of a TCP6 connection"); 2708 #endif /* INET6 */ 2709 2710 #ifdef INET 2711 /* Path MTU to try next when a fragmentation-needed message is received. */ 2712 static inline int 2713 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2714 { 2715 int mtu = ntohs(icp->icmp_nextmtu); 2716 2717 /* If no alternative MTU was proposed, try the next smaller one. */ 2718 if (!mtu) 2719 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2720 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2721 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2722 2723 return (mtu); 2724 } 2725 2726 static void 2727 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2728 { 2729 struct ip *ip; 2730 struct tcphdr *th; 2731 struct inpcb *inp; 2732 struct tcpcb *tp; 2733 struct inpcb *(*notify)(struct inpcb *, int); 2734 struct in_conninfo inc; 2735 tcp_seq icmp_tcp_seq; 2736 int errno, mtu; 2737 2738 errno = icmp_errmap(icp); 2739 switch (errno) { 2740 case 0: 2741 return; 2742 case EMSGSIZE: 2743 notify = tcp_mtudisc_notify; 2744 break; 2745 case ECONNREFUSED: 2746 if (V_icmp_may_rst) 2747 notify = tcp_drop_syn_sent; 2748 else 2749 notify = tcp_notify; 2750 break; 2751 case EHOSTUNREACH: 2752 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2753 notify = tcp_drop_syn_sent; 2754 else 2755 notify = tcp_notify; 2756 break; 2757 default: 2758 notify = tcp_notify; 2759 } 2760 2761 ip = &icp->icmp_ip; 2762 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2763 icmp_tcp_seq = th->th_seq; 2764 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2765 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2766 if (inp != NULL) { 2767 tp = intotcpcb(inp); 2768 #ifdef TCP_OFFLOAD 2769 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2770 /* 2771 * MTU discovery for offloaded connections. Let 2772 * the TOE driver verify seq# and process it. 2773 */ 2774 mtu = tcp_next_pmtu(icp, ip); 2775 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2776 goto out; 2777 } 2778 #endif 2779 if (tp->t_port != port) 2780 goto out; 2781 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2782 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2783 if (errno == EMSGSIZE) { 2784 /* 2785 * MTU discovery: we got a needfrag and 2786 * will potentially try a lower MTU. 2787 */ 2788 mtu = tcp_next_pmtu(icp, ip); 2789 2790 /* 2791 * Only process the offered MTU if it 2792 * is smaller than the current one. 2793 */ 2794 if (mtu < tp->t_maxseg + 2795 sizeof(struct tcpiphdr)) { 2796 bzero(&inc, sizeof(inc)); 2797 inc.inc_faddr = ip->ip_dst; 2798 inc.inc_fibnum = 2799 inp->inp_inc.inc_fibnum; 2800 tcp_hc_updatemtu(&inc, mtu); 2801 inp = tcp_mtudisc(inp, mtu); 2802 } 2803 } else 2804 inp = (*notify)(inp, errno); 2805 } 2806 } else { 2807 bzero(&inc, sizeof(inc)); 2808 inc.inc_fport = th->th_dport; 2809 inc.inc_lport = th->th_sport; 2810 inc.inc_faddr = ip->ip_dst; 2811 inc.inc_laddr = ip->ip_src; 2812 syncache_unreach(&inc, icmp_tcp_seq, port); 2813 } 2814 out: 2815 if (inp != NULL) 2816 INP_WUNLOCK(inp); 2817 } 2818 2819 static void 2820 tcp_ctlinput(struct icmp *icmp) 2821 { 2822 tcp_ctlinput_with_port(icmp, htons(0)); 2823 } 2824 2825 static void 2826 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2827 { 2828 /* Its a tunneled TCP over UDP icmp */ 2829 struct icmp *icmp = param.icmp; 2830 struct ip *outer_ip, *inner_ip; 2831 struct udphdr *udp; 2832 struct tcphdr *th, ttemp; 2833 int i_hlen, o_len; 2834 uint16_t port; 2835 2836 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2837 inner_ip = &icmp->icmp_ip; 2838 i_hlen = inner_ip->ip_hl << 2; 2839 o_len = ntohs(outer_ip->ip_len); 2840 if (o_len < 2841 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2842 /* Not enough data present */ 2843 return; 2844 } 2845 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2846 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2847 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2848 return; 2849 } 2850 port = udp->uh_dport; 2851 th = (struct tcphdr *)(udp + 1); 2852 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2853 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2854 /* Now adjust down the size of the outer IP header */ 2855 o_len -= sizeof(struct udphdr); 2856 outer_ip->ip_len = htons(o_len); 2857 /* Now call in to the normal handling code */ 2858 tcp_ctlinput_with_port(icmp, port); 2859 } 2860 #endif /* INET */ 2861 2862 #ifdef INET6 2863 static inline int 2864 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2865 { 2866 int mtu = ntohl(icmp6->icmp6_mtu); 2867 2868 /* 2869 * If no alternative MTU was proposed, or the proposed MTU was too 2870 * small, set to the min. 2871 */ 2872 if (mtu < IPV6_MMTU) 2873 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2874 return (mtu); 2875 } 2876 2877 static void 2878 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2879 { 2880 struct in6_addr *dst; 2881 struct inpcb *(*notify)(struct inpcb *, int); 2882 struct ip6_hdr *ip6; 2883 struct mbuf *m; 2884 struct inpcb *inp; 2885 struct tcpcb *tp; 2886 struct icmp6_hdr *icmp6; 2887 struct in_conninfo inc; 2888 struct tcp_ports { 2889 uint16_t th_sport; 2890 uint16_t th_dport; 2891 } t_ports; 2892 tcp_seq icmp_tcp_seq; 2893 unsigned int mtu; 2894 unsigned int off; 2895 int errno; 2896 2897 icmp6 = ip6cp->ip6c_icmp6; 2898 m = ip6cp->ip6c_m; 2899 ip6 = ip6cp->ip6c_ip6; 2900 off = ip6cp->ip6c_off; 2901 dst = &ip6cp->ip6c_finaldst->sin6_addr; 2902 2903 errno = icmp6_errmap(icmp6); 2904 switch (errno) { 2905 case 0: 2906 return; 2907 case EMSGSIZE: 2908 notify = tcp_mtudisc_notify; 2909 break; 2910 case ECONNREFUSED: 2911 if (V_icmp_may_rst) 2912 notify = tcp_drop_syn_sent; 2913 else 2914 notify = tcp_notify; 2915 break; 2916 case EHOSTUNREACH: 2917 /* 2918 * There are only four ICMPs that may reset connection: 2919 * - administratively prohibited 2920 * - port unreachable 2921 * - time exceeded in transit 2922 * - unknown next header 2923 */ 2924 if (V_icmp_may_rst && 2925 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 2926 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 2927 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 2928 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 2929 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 2930 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 2931 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 2932 notify = tcp_drop_syn_sent; 2933 else 2934 notify = tcp_notify; 2935 break; 2936 default: 2937 notify = tcp_notify; 2938 } 2939 2940 /* Check if we can safely get the ports from the tcp hdr */ 2941 if (m == NULL || 2942 (m->m_pkthdr.len < 2943 (int32_t) (off + sizeof(struct tcp_ports)))) { 2944 return; 2945 } 2946 bzero(&t_ports, sizeof(struct tcp_ports)); 2947 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2948 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2949 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2950 off += sizeof(struct tcp_ports); 2951 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2952 goto out; 2953 } 2954 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2955 if (inp != NULL) { 2956 tp = intotcpcb(inp); 2957 #ifdef TCP_OFFLOAD 2958 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2959 /* MTU discovery for offloaded connections. */ 2960 mtu = tcp6_next_pmtu(icmp6); 2961 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2962 goto out; 2963 } 2964 #endif 2965 if (tp->t_port != port) 2966 goto out; 2967 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2968 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2969 if (errno == EMSGSIZE) { 2970 /* 2971 * MTU discovery: 2972 * If we got a needfrag set the MTU 2973 * in the route to the suggested new 2974 * value (if given) and then notify. 2975 */ 2976 mtu = tcp6_next_pmtu(icmp6); 2977 2978 bzero(&inc, sizeof(inc)); 2979 inc.inc_fibnum = M_GETFIB(m); 2980 inc.inc_flags |= INC_ISIPV6; 2981 inc.inc6_faddr = *dst; 2982 if (in6_setscope(&inc.inc6_faddr, 2983 m->m_pkthdr.rcvif, NULL)) 2984 goto out; 2985 /* 2986 * Only process the offered MTU if it 2987 * is smaller than the current one. 2988 */ 2989 if (mtu < tp->t_maxseg + 2990 sizeof (struct tcphdr) + 2991 sizeof (struct ip6_hdr)) { 2992 tcp_hc_updatemtu(&inc, mtu); 2993 tcp_mtudisc(inp, mtu); 2994 ICMP6STAT_INC(icp6s_pmtuchg); 2995 } 2996 } else 2997 inp = (*notify)(inp, errno); 2998 } 2999 } else { 3000 bzero(&inc, sizeof(inc)); 3001 inc.inc_fibnum = M_GETFIB(m); 3002 inc.inc_flags |= INC_ISIPV6; 3003 inc.inc_fport = t_ports.th_dport; 3004 inc.inc_lport = t_ports.th_sport; 3005 inc.inc6_faddr = *dst; 3006 inc.inc6_laddr = ip6->ip6_src; 3007 syncache_unreach(&inc, icmp_tcp_seq, port); 3008 } 3009 out: 3010 if (inp != NULL) 3011 INP_WUNLOCK(inp); 3012 } 3013 3014 static void 3015 tcp6_ctlinput(struct ip6ctlparam *ctl) 3016 { 3017 tcp6_ctlinput_with_port(ctl, htons(0)); 3018 } 3019 3020 static void 3021 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3022 { 3023 struct ip6ctlparam *ip6cp = param.ip6cp; 3024 struct mbuf *m; 3025 struct udphdr *udp; 3026 uint16_t port; 3027 3028 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3029 if (m == NULL) { 3030 return; 3031 } 3032 udp = mtod(m, struct udphdr *); 3033 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3034 return; 3035 } 3036 port = udp->uh_dport; 3037 m_adj(m, sizeof(struct udphdr)); 3038 if ((m->m_flags & M_PKTHDR) == 0) { 3039 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3040 } 3041 /* Now call in to the normal handling code */ 3042 tcp6_ctlinput_with_port(ip6cp, port); 3043 } 3044 3045 #endif /* INET6 */ 3046 3047 static uint32_t 3048 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3049 { 3050 SIPHASH_CTX ctx; 3051 uint32_t hash[2]; 3052 3053 KASSERT(len >= SIPHASH_KEY_LENGTH, 3054 ("%s: keylen %u too short ", __func__, len)); 3055 SipHash24_Init(&ctx); 3056 SipHash_SetKey(&ctx, (uint8_t *)key); 3057 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3058 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3059 switch (inc->inc_flags & INC_ISIPV6) { 3060 #ifdef INET 3061 case 0: 3062 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3063 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3064 break; 3065 #endif 3066 #ifdef INET6 3067 case INC_ISIPV6: 3068 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3069 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3070 break; 3071 #endif 3072 } 3073 SipHash_Final((uint8_t *)hash, &ctx); 3074 3075 return (hash[0] ^ hash[1]); 3076 } 3077 3078 uint32_t 3079 tcp_new_ts_offset(struct in_conninfo *inc) 3080 { 3081 struct in_conninfo inc_store, *local_inc; 3082 3083 if (!V_tcp_ts_offset_per_conn) { 3084 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3085 inc_store.inc_lport = 0; 3086 inc_store.inc_fport = 0; 3087 local_inc = &inc_store; 3088 } else { 3089 local_inc = inc; 3090 } 3091 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3092 sizeof(V_ts_offset_secret))); 3093 } 3094 3095 /* 3096 * Following is where TCP initial sequence number generation occurs. 3097 * 3098 * There are two places where we must use initial sequence numbers: 3099 * 1. In SYN-ACK packets. 3100 * 2. In SYN packets. 3101 * 3102 * All ISNs for SYN-ACK packets are generated by the syncache. See 3103 * tcp_syncache.c for details. 3104 * 3105 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3106 * depends on this property. In addition, these ISNs should be 3107 * unguessable so as to prevent connection hijacking. To satisfy 3108 * the requirements of this situation, the algorithm outlined in 3109 * RFC 1948 is used, with only small modifications. 3110 * 3111 * Implementation details: 3112 * 3113 * Time is based off the system timer, and is corrected so that it 3114 * increases by one megabyte per second. This allows for proper 3115 * recycling on high speed LANs while still leaving over an hour 3116 * before rollover. 3117 * 3118 * As reading the *exact* system time is too expensive to be done 3119 * whenever setting up a TCP connection, we increment the time 3120 * offset in two ways. First, a small random positive increment 3121 * is added to isn_offset for each connection that is set up. 3122 * Second, the function tcp_isn_tick fires once per clock tick 3123 * and increments isn_offset as necessary so that sequence numbers 3124 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3125 * random positive increments serve only to ensure that the same 3126 * exact sequence number is never sent out twice (as could otherwise 3127 * happen when a port is recycled in less than the system tick 3128 * interval.) 3129 * 3130 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3131 * between seeding of isn_secret. This is normally set to zero, 3132 * as reseeding should not be necessary. 3133 * 3134 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3135 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3136 * general, this means holding an exclusive (write) lock. 3137 */ 3138 3139 #define ISN_BYTES_PER_SECOND 1048576 3140 #define ISN_STATIC_INCREMENT 4096 3141 #define ISN_RANDOM_INCREMENT (4096 - 1) 3142 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3143 3144 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3145 VNET_DEFINE_STATIC(int, isn_last); 3146 VNET_DEFINE_STATIC(int, isn_last_reseed); 3147 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3148 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3149 3150 #define V_isn_secret VNET(isn_secret) 3151 #define V_isn_last VNET(isn_last) 3152 #define V_isn_last_reseed VNET(isn_last_reseed) 3153 #define V_isn_offset VNET(isn_offset) 3154 #define V_isn_offset_old VNET(isn_offset_old) 3155 3156 tcp_seq 3157 tcp_new_isn(struct in_conninfo *inc) 3158 { 3159 tcp_seq new_isn; 3160 u_int32_t projected_offset; 3161 3162 ISN_LOCK(); 3163 /* Seed if this is the first use, reseed if requested. */ 3164 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3165 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3166 < (u_int)ticks))) { 3167 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3168 V_isn_last_reseed = ticks; 3169 } 3170 3171 /* Compute the hash and return the ISN. */ 3172 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3173 sizeof(V_isn_secret)); 3174 V_isn_offset += ISN_STATIC_INCREMENT + 3175 (arc4random() & ISN_RANDOM_INCREMENT); 3176 if (ticks != V_isn_last) { 3177 projected_offset = V_isn_offset_old + 3178 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3179 if (SEQ_GT(projected_offset, V_isn_offset)) 3180 V_isn_offset = projected_offset; 3181 V_isn_offset_old = V_isn_offset; 3182 V_isn_last = ticks; 3183 } 3184 new_isn += V_isn_offset; 3185 ISN_UNLOCK(); 3186 return (new_isn); 3187 } 3188 3189 /* 3190 * When a specific ICMP unreachable message is received and the 3191 * connection state is SYN-SENT, drop the connection. This behavior 3192 * is controlled by the icmp_may_rst sysctl. 3193 */ 3194 static struct inpcb * 3195 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3196 { 3197 struct tcpcb *tp; 3198 3199 NET_EPOCH_ASSERT(); 3200 INP_WLOCK_ASSERT(inp); 3201 3202 tp = intotcpcb(inp); 3203 if (tp->t_state != TCPS_SYN_SENT) 3204 return (inp); 3205 3206 if (IS_FASTOPEN(tp->t_flags)) 3207 tcp_fastopen_disable_path(tp); 3208 3209 tp = tcp_drop(tp, errno); 3210 if (tp != NULL) 3211 return (inp); 3212 else 3213 return (NULL); 3214 } 3215 3216 /* 3217 * When `need fragmentation' ICMP is received, update our idea of the MSS 3218 * based on the new value. Also nudge TCP to send something, since we 3219 * know the packet we just sent was dropped. 3220 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3221 */ 3222 static struct inpcb * 3223 tcp_mtudisc_notify(struct inpcb *inp, int error) 3224 { 3225 3226 return (tcp_mtudisc(inp, -1)); 3227 } 3228 3229 static struct inpcb * 3230 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3231 { 3232 struct tcpcb *tp; 3233 struct socket *so; 3234 3235 INP_WLOCK_ASSERT(inp); 3236 3237 tp = intotcpcb(inp); 3238 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3239 3240 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3241 3242 so = inp->inp_socket; 3243 SOCKBUF_LOCK(&so->so_snd); 3244 /* If the mss is larger than the socket buffer, decrease the mss. */ 3245 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3246 tp->t_maxseg = so->so_snd.sb_hiwat; 3247 SOCKBUF_UNLOCK(&so->so_snd); 3248 3249 TCPSTAT_INC(tcps_mturesent); 3250 tp->t_rtttime = 0; 3251 tp->snd_nxt = tp->snd_una; 3252 tcp_free_sackholes(tp); 3253 tp->snd_recover = tp->snd_max; 3254 if (tp->t_flags & TF_SACK_PERMIT) 3255 EXIT_FASTRECOVERY(tp->t_flags); 3256 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3257 /* 3258 * Conceptually the snd_nxt setting 3259 * and freeing sack holes should 3260 * be done by the default stacks 3261 * own tfb_tcp_mtu_chg(). 3262 */ 3263 tp->t_fb->tfb_tcp_mtu_chg(tp); 3264 } 3265 if (tcp_output(tp) < 0) 3266 return (NULL); 3267 else 3268 return (inp); 3269 } 3270 3271 #ifdef INET 3272 /* 3273 * Look-up the routing entry to the peer of this inpcb. If no route 3274 * is found and it cannot be allocated, then return 0. This routine 3275 * is called by TCP routines that access the rmx structure and by 3276 * tcp_mss_update to get the peer/interface MTU. 3277 */ 3278 uint32_t 3279 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3280 { 3281 struct nhop_object *nh; 3282 struct ifnet *ifp; 3283 uint32_t maxmtu = 0; 3284 3285 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3286 3287 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3288 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3289 if (nh == NULL) 3290 return (0); 3291 3292 ifp = nh->nh_ifp; 3293 maxmtu = nh->nh_mtu; 3294 3295 /* Report additional interface capabilities. */ 3296 if (cap != NULL) { 3297 if (ifp->if_capenable & IFCAP_TSO4 && 3298 ifp->if_hwassist & CSUM_TSO) { 3299 cap->ifcap |= CSUM_TSO; 3300 cap->tsomax = ifp->if_hw_tsomax; 3301 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3302 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3303 } 3304 } 3305 } 3306 return (maxmtu); 3307 } 3308 #endif /* INET */ 3309 3310 #ifdef INET6 3311 uint32_t 3312 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3313 { 3314 struct nhop_object *nh; 3315 struct in6_addr dst6; 3316 uint32_t scopeid; 3317 struct ifnet *ifp; 3318 uint32_t maxmtu = 0; 3319 3320 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3321 3322 if (inc->inc_flags & INC_IPV6MINMTU) 3323 return (IPV6_MMTU); 3324 3325 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3326 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3327 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3328 if (nh == NULL) 3329 return (0); 3330 3331 ifp = nh->nh_ifp; 3332 maxmtu = nh->nh_mtu; 3333 3334 /* Report additional interface capabilities. */ 3335 if (cap != NULL) { 3336 if (ifp->if_capenable & IFCAP_TSO6 && 3337 ifp->if_hwassist & CSUM_TSO) { 3338 cap->ifcap |= CSUM_TSO; 3339 cap->tsomax = ifp->if_hw_tsomax; 3340 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3341 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3342 } 3343 } 3344 } 3345 3346 return (maxmtu); 3347 } 3348 3349 /* 3350 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3351 * 3352 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3353 * The right place to do that is ip6_setpktopt() that has just been 3354 * executed. By the way it just filled ip6po_minmtu for us. 3355 */ 3356 void 3357 tcp6_use_min_mtu(struct tcpcb *tp) 3358 { 3359 struct inpcb *inp = tptoinpcb(tp); 3360 3361 INP_WLOCK_ASSERT(inp); 3362 /* 3363 * In case of the IPV6_USE_MIN_MTU socket 3364 * option, the INC_IPV6MINMTU flag to announce 3365 * a corresponding MSS during the initial 3366 * handshake. If the TCP connection is not in 3367 * the front states, just reduce the MSS being 3368 * used. This avoids the sending of TCP 3369 * segments which will be fragmented at the 3370 * IPv6 layer. 3371 */ 3372 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3373 if ((tp->t_state >= TCPS_SYN_SENT) && 3374 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3375 struct ip6_pktopts *opt; 3376 3377 opt = inp->in6p_outputopts; 3378 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3379 tp->t_maxseg > TCP6_MSS) 3380 tp->t_maxseg = TCP6_MSS; 3381 } 3382 } 3383 #endif /* INET6 */ 3384 3385 /* 3386 * Calculate effective SMSS per RFC5681 definition for a given TCP 3387 * connection at its current state, taking into account SACK and etc. 3388 */ 3389 u_int 3390 tcp_maxseg(const struct tcpcb *tp) 3391 { 3392 u_int optlen; 3393 3394 if (tp->t_flags & TF_NOOPT) 3395 return (tp->t_maxseg); 3396 3397 /* 3398 * Here we have a simplified code from tcp_addoptions(), 3399 * without a proper loop, and having most of paddings hardcoded. 3400 * We might make mistakes with padding here in some edge cases, 3401 * but this is harmless, since result of tcp_maxseg() is used 3402 * only in cwnd and ssthresh estimations. 3403 */ 3404 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3405 if (tp->t_flags & TF_RCVD_TSTMP) 3406 optlen = TCPOLEN_TSTAMP_APPA; 3407 else 3408 optlen = 0; 3409 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3410 if (tp->t_flags & TF_SIGNATURE) 3411 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3412 #endif 3413 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3414 optlen += TCPOLEN_SACKHDR; 3415 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3416 optlen = PADTCPOLEN(optlen); 3417 } 3418 } else { 3419 if (tp->t_flags & TF_REQ_TSTMP) 3420 optlen = TCPOLEN_TSTAMP_APPA; 3421 else 3422 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3423 if (tp->t_flags & TF_REQ_SCALE) 3424 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3425 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3426 if (tp->t_flags & TF_SIGNATURE) 3427 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3428 #endif 3429 if (tp->t_flags & TF_SACK_PERMIT) 3430 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3431 } 3432 #undef PAD 3433 optlen = min(optlen, TCP_MAXOLEN); 3434 return (tp->t_maxseg - optlen); 3435 } 3436 3437 3438 u_int 3439 tcp_fixed_maxseg(const struct tcpcb *tp) 3440 { 3441 int optlen; 3442 3443 if (tp->t_flags & TF_NOOPT) 3444 return (tp->t_maxseg); 3445 3446 /* 3447 * Here we have a simplified code from tcp_addoptions(), 3448 * without a proper loop, and having most of paddings hardcoded. 3449 * We only consider fixed options that we would send every 3450 * time I.e. SACK is not considered. This is important 3451 * for cc modules to figure out what the modulo of the 3452 * cwnd should be. 3453 */ 3454 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3455 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3456 if (tp->t_flags & TF_RCVD_TSTMP) 3457 optlen = TCPOLEN_TSTAMP_APPA; 3458 else 3459 optlen = 0; 3460 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3461 if (tp->t_flags & TF_SIGNATURE) 3462 optlen += PAD(TCPOLEN_SIGNATURE); 3463 #endif 3464 } else { 3465 if (tp->t_flags & TF_REQ_TSTMP) 3466 optlen = TCPOLEN_TSTAMP_APPA; 3467 else 3468 optlen = PAD(TCPOLEN_MAXSEG); 3469 if (tp->t_flags & TF_REQ_SCALE) 3470 optlen += PAD(TCPOLEN_WINDOW); 3471 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3472 if (tp->t_flags & TF_SIGNATURE) 3473 optlen += PAD(TCPOLEN_SIGNATURE); 3474 #endif 3475 if (tp->t_flags & TF_SACK_PERMIT) 3476 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3477 } 3478 #undef PAD 3479 optlen = min(optlen, TCP_MAXOLEN); 3480 return (tp->t_maxseg - optlen); 3481 } 3482 3483 3484 3485 static int 3486 sysctl_drop(SYSCTL_HANDLER_ARGS) 3487 { 3488 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3489 struct sockaddr_storage addrs[2]; 3490 struct inpcb *inp; 3491 struct tcpcb *tp; 3492 #ifdef INET 3493 struct sockaddr_in *fin = NULL, *lin = NULL; 3494 #endif 3495 struct epoch_tracker et; 3496 #ifdef INET6 3497 struct sockaddr_in6 *fin6, *lin6; 3498 #endif 3499 int error; 3500 3501 inp = NULL; 3502 #ifdef INET6 3503 fin6 = lin6 = NULL; 3504 #endif 3505 error = 0; 3506 3507 if (req->oldptr != NULL || req->oldlen != 0) 3508 return (EINVAL); 3509 if (req->newptr == NULL) 3510 return (EPERM); 3511 if (req->newlen < sizeof(addrs)) 3512 return (ENOMEM); 3513 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3514 if (error) 3515 return (error); 3516 3517 switch (addrs[0].ss_family) { 3518 #ifdef INET6 3519 case AF_INET6: 3520 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3521 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3522 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3523 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3524 return (EINVAL); 3525 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3526 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3527 return (EINVAL); 3528 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3529 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3530 #ifdef INET 3531 fin = (struct sockaddr_in *)&addrs[0]; 3532 lin = (struct sockaddr_in *)&addrs[1]; 3533 #endif 3534 break; 3535 } 3536 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3537 if (error) 3538 return (error); 3539 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3540 if (error) 3541 return (error); 3542 break; 3543 #endif 3544 #ifdef INET 3545 case AF_INET: 3546 fin = (struct sockaddr_in *)&addrs[0]; 3547 lin = (struct sockaddr_in *)&addrs[1]; 3548 if (fin->sin_len != sizeof(struct sockaddr_in) || 3549 lin->sin_len != sizeof(struct sockaddr_in)) 3550 return (EINVAL); 3551 break; 3552 #endif 3553 default: 3554 return (EINVAL); 3555 } 3556 NET_EPOCH_ENTER(et); 3557 switch (addrs[0].ss_family) { 3558 #ifdef INET6 3559 case AF_INET6: 3560 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3561 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3562 INPLOOKUP_WLOCKPCB, NULL); 3563 break; 3564 #endif 3565 #ifdef INET 3566 case AF_INET: 3567 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3568 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3569 break; 3570 #endif 3571 } 3572 if (inp != NULL) { 3573 if (!SOLISTENING(inp->inp_socket)) { 3574 tp = intotcpcb(inp); 3575 tp = tcp_drop(tp, ECONNABORTED); 3576 if (tp != NULL) 3577 INP_WUNLOCK(inp); 3578 } else 3579 INP_WUNLOCK(inp); 3580 } else 3581 error = ESRCH; 3582 NET_EPOCH_EXIT(et); 3583 return (error); 3584 } 3585 3586 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3587 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3588 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3589 "Drop TCP connection"); 3590 3591 static int 3592 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3593 { 3594 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3595 &tcp_ctloutput_set)); 3596 } 3597 3598 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3599 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3600 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3601 "Set socket option for TCP endpoint"); 3602 3603 #ifdef KERN_TLS 3604 static int 3605 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3606 { 3607 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3608 struct sockaddr_storage addrs[2]; 3609 struct inpcb *inp; 3610 #ifdef INET 3611 struct sockaddr_in *fin = NULL, *lin = NULL; 3612 #endif 3613 struct epoch_tracker et; 3614 #ifdef INET6 3615 struct sockaddr_in6 *fin6, *lin6; 3616 #endif 3617 int error; 3618 3619 inp = NULL; 3620 #ifdef INET6 3621 fin6 = lin6 = NULL; 3622 #endif 3623 error = 0; 3624 3625 if (req->oldptr != NULL || req->oldlen != 0) 3626 return (EINVAL); 3627 if (req->newptr == NULL) 3628 return (EPERM); 3629 if (req->newlen < sizeof(addrs)) 3630 return (ENOMEM); 3631 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3632 if (error) 3633 return (error); 3634 3635 switch (addrs[0].ss_family) { 3636 #ifdef INET6 3637 case AF_INET6: 3638 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3639 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3640 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3641 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3642 return (EINVAL); 3643 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3644 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3645 return (EINVAL); 3646 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3647 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3648 #ifdef INET 3649 fin = (struct sockaddr_in *)&addrs[0]; 3650 lin = (struct sockaddr_in *)&addrs[1]; 3651 #endif 3652 break; 3653 } 3654 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3655 if (error) 3656 return (error); 3657 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3658 if (error) 3659 return (error); 3660 break; 3661 #endif 3662 #ifdef INET 3663 case AF_INET: 3664 fin = (struct sockaddr_in *)&addrs[0]; 3665 lin = (struct sockaddr_in *)&addrs[1]; 3666 if (fin->sin_len != sizeof(struct sockaddr_in) || 3667 lin->sin_len != sizeof(struct sockaddr_in)) 3668 return (EINVAL); 3669 break; 3670 #endif 3671 default: 3672 return (EINVAL); 3673 } 3674 NET_EPOCH_ENTER(et); 3675 switch (addrs[0].ss_family) { 3676 #ifdef INET6 3677 case AF_INET6: 3678 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3679 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3680 INPLOOKUP_WLOCKPCB, NULL); 3681 break; 3682 #endif 3683 #ifdef INET 3684 case AF_INET: 3685 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3686 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3687 break; 3688 #endif 3689 } 3690 NET_EPOCH_EXIT(et); 3691 if (inp != NULL) { 3692 struct socket *so; 3693 3694 so = inp->inp_socket; 3695 soref(so); 3696 error = ktls_set_tx_mode(so, 3697 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3698 INP_WUNLOCK(inp); 3699 sorele(so); 3700 } else 3701 error = ESRCH; 3702 return (error); 3703 } 3704 3705 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3706 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3707 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3708 "Switch TCP connection to SW TLS"); 3709 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3710 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3711 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3712 "Switch TCP connection to ifnet TLS"); 3713 #endif 3714 3715 /* 3716 * Generate a standardized TCP log line for use throughout the 3717 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3718 * allow use in the interrupt context. 3719 * 3720 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3721 * NB: The function may return NULL if memory allocation failed. 3722 * 3723 * Due to header inclusion and ordering limitations the struct ip 3724 * and ip6_hdr pointers have to be passed as void pointers. 3725 */ 3726 char * 3727 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3728 const void *ip6hdr) 3729 { 3730 3731 /* Is logging enabled? */ 3732 if (V_tcp_log_in_vain == 0) 3733 return (NULL); 3734 3735 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3736 } 3737 3738 char * 3739 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3740 const void *ip6hdr) 3741 { 3742 3743 /* Is logging enabled? */ 3744 if (tcp_log_debug == 0) 3745 return (NULL); 3746 3747 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3748 } 3749 3750 static char * 3751 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3752 const void *ip6hdr) 3753 { 3754 char *s, *sp; 3755 size_t size; 3756 #ifdef INET 3757 const struct ip *ip = (const struct ip *)ip4hdr; 3758 #endif 3759 #ifdef INET6 3760 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3761 #endif /* INET6 */ 3762 3763 /* 3764 * The log line looks like this: 3765 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3766 */ 3767 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3768 sizeof(PRINT_TH_FLAGS) + 1 + 3769 #ifdef INET6 3770 2 * INET6_ADDRSTRLEN; 3771 #else 3772 2 * INET_ADDRSTRLEN; 3773 #endif /* INET6 */ 3774 3775 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3776 if (s == NULL) 3777 return (NULL); 3778 3779 strcat(s, "TCP: ["); 3780 sp = s + strlen(s); 3781 3782 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3783 inet_ntoa_r(inc->inc_faddr, sp); 3784 sp = s + strlen(s); 3785 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3786 sp = s + strlen(s); 3787 inet_ntoa_r(inc->inc_laddr, sp); 3788 sp = s + strlen(s); 3789 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3790 #ifdef INET6 3791 } else if (inc) { 3792 ip6_sprintf(sp, &inc->inc6_faddr); 3793 sp = s + strlen(s); 3794 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3795 sp = s + strlen(s); 3796 ip6_sprintf(sp, &inc->inc6_laddr); 3797 sp = s + strlen(s); 3798 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3799 } else if (ip6 && th) { 3800 ip6_sprintf(sp, &ip6->ip6_src); 3801 sp = s + strlen(s); 3802 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3803 sp = s + strlen(s); 3804 ip6_sprintf(sp, &ip6->ip6_dst); 3805 sp = s + strlen(s); 3806 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3807 #endif /* INET6 */ 3808 #ifdef INET 3809 } else if (ip && th) { 3810 inet_ntoa_r(ip->ip_src, sp); 3811 sp = s + strlen(s); 3812 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3813 sp = s + strlen(s); 3814 inet_ntoa_r(ip->ip_dst, sp); 3815 sp = s + strlen(s); 3816 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3817 #endif /* INET */ 3818 } else { 3819 free(s, M_TCPLOG); 3820 return (NULL); 3821 } 3822 sp = s + strlen(s); 3823 if (th) 3824 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3825 if (*(s + size - 1) != '\0') 3826 panic("%s: string too long", __func__); 3827 return (s); 3828 } 3829 3830 /* 3831 * A subroutine which makes it easy to track TCP state changes with DTrace. 3832 * This function shouldn't be called for t_state initializations that don't 3833 * correspond to actual TCP state transitions. 3834 */ 3835 void 3836 tcp_state_change(struct tcpcb *tp, int newstate) 3837 { 3838 #if defined(KDTRACE_HOOKS) 3839 int pstate = tp->t_state; 3840 #endif 3841 3842 TCPSTATES_DEC(tp->t_state); 3843 TCPSTATES_INC(newstate); 3844 tp->t_state = newstate; 3845 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3846 } 3847 3848 /* 3849 * Create an external-format (``xtcpcb'') structure using the information in 3850 * the kernel-format tcpcb structure pointed to by tp. This is done to 3851 * reduce the spew of irrelevant information over this interface, to isolate 3852 * user code from changes in the kernel structure, and potentially to provide 3853 * information-hiding if we decide that some of this information should be 3854 * hidden from users. 3855 */ 3856 void 3857 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3858 { 3859 struct tcpcb *tp = intotcpcb(inp); 3860 sbintime_t now; 3861 3862 bzero(xt, sizeof(*xt)); 3863 xt->t_state = tp->t_state; 3864 xt->t_logstate = tp->t_logstate; 3865 xt->t_flags = tp->t_flags; 3866 xt->t_sndzerowin = tp->t_sndzerowin; 3867 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3868 xt->t_rcvoopack = tp->t_rcvoopack; 3869 xt->t_rcv_wnd = tp->rcv_wnd; 3870 xt->t_snd_wnd = tp->snd_wnd; 3871 xt->t_snd_cwnd = tp->snd_cwnd; 3872 xt->t_snd_ssthresh = tp->snd_ssthresh; 3873 xt->t_dsack_bytes = tp->t_dsack_bytes; 3874 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 3875 xt->t_dsack_pack = tp->t_dsack_pack; 3876 xt->t_maxseg = tp->t_maxseg; 3877 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3878 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3879 3880 now = getsbinuptime(); 3881 #define COPYTIMER(which,where) do { \ 3882 if (tp->t_timers[which] != SBT_MAX) \ 3883 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 3884 else \ 3885 xt->where = 0; \ 3886 } while (0) 3887 COPYTIMER(TT_DELACK, tt_delack); 3888 COPYTIMER(TT_REXMT, tt_rexmt); 3889 COPYTIMER(TT_PERSIST, tt_persist); 3890 COPYTIMER(TT_KEEP, tt_keep); 3891 COPYTIMER(TT_2MSL, tt_2msl); 3892 #undef COPYTIMER 3893 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3894 3895 xt->xt_encaps_port = tp->t_port; 3896 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3897 TCP_FUNCTION_NAME_LEN_MAX); 3898 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 3899 #ifdef TCP_BLACKBOX 3900 (void)tcp_log_get_id(tp, xt->xt_logid); 3901 #endif 3902 3903 xt->xt_len = sizeof(struct xtcpcb); 3904 in_pcbtoxinpcb(inp, &xt->xt_inp); 3905 } 3906 3907 void 3908 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3909 { 3910 uint32_t bit, i; 3911 3912 if ((tp == NULL) || 3913 (status > TCP_EI_STATUS_MAX_VALUE) || 3914 (status == 0)) { 3915 /* Invalid */ 3916 return; 3917 } 3918 if (status > (sizeof(uint32_t) * 8)) { 3919 /* Should this be a KASSERT? */ 3920 return; 3921 } 3922 bit = 1U << (status - 1); 3923 if (bit & tp->t_end_info_status) { 3924 /* already logged */ 3925 return; 3926 } 3927 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3928 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3929 tp->t_end_info_bytes[i] = status; 3930 tp->t_end_info_status |= bit; 3931 break; 3932 } 3933 } 3934 } 3935 3936 int 3937 tcp_can_enable_pacing(void) 3938 { 3939 3940 if ((tcp_pacing_limit == -1) || 3941 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 3942 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 3943 shadow_num_connections = number_of_tcp_connections_pacing; 3944 return (1); 3945 } else { 3946 return (0); 3947 } 3948 } 3949 3950 static uint8_t tcp_pacing_warning = 0; 3951 3952 void 3953 tcp_decrement_paced_conn(void) 3954 { 3955 uint32_t ret; 3956 3957 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 3958 shadow_num_connections = number_of_tcp_connections_pacing; 3959 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 3960 if (ret == 0) { 3961 if (tcp_pacing_limit != -1) { 3962 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 3963 tcp_pacing_limit = 0; 3964 } else if (tcp_pacing_warning == 0) { 3965 printf("Warning pacing count is invalid, invalid decrement\n"); 3966 tcp_pacing_warning = 1; 3967 } 3968 } 3969 } 3970