1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/arb.h> 46 #include <sys/callout.h> 47 #include <sys/eventhandler.h> 48 #ifdef TCP_HHOOK 49 #include <sys/hhook.h> 50 #endif 51 #include <sys/kernel.h> 52 #ifdef TCP_HHOOK 53 #include <sys/khelp.h> 54 #endif 55 #ifdef KERN_TLS 56 #include <sys/ktls.h> 57 #endif 58 #include <sys/qmath.h> 59 #include <sys/stats.h> 60 #include <sys/sysctl.h> 61 #include <sys/jail.h> 62 #include <sys/malloc.h> 63 #include <sys/refcount.h> 64 #include <sys/mbuf.h> 65 #include <sys/priv.h> 66 #include <sys/proc.h> 67 #include <sys/sdt.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/protosw.h> 71 #include <sys/random.h> 72 73 #include <vm/uma.h> 74 75 #include <net/route.h> 76 #include <net/route/nhop.h> 77 #include <net/if.h> 78 #include <net/if_var.h> 79 #include <net/vnet.h> 80 81 #include <netinet/in.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/in_kdtrace.h> 84 #include <netinet/in_pcb.h> 85 #include <netinet/in_systm.h> 86 #include <netinet/in_var.h> 87 #include <netinet/ip.h> 88 #include <netinet/ip_icmp.h> 89 #include <netinet/ip_var.h> 90 #ifdef INET6 91 #include <netinet/icmp6.h> 92 #include <netinet/ip6.h> 93 #include <netinet6/in6_fib.h> 94 #include <netinet6/in6_pcb.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/scope6_var.h> 97 #include <netinet6/nd6.h> 98 #endif 99 100 #include <netinet/tcp.h> 101 #ifdef INVARIANTS 102 #define TCPSTATES 103 #endif 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_syncache.h> 110 #include <netinet/tcp_hpts.h> 111 #include <netinet/cc/cc.h> 112 #include <netinet/tcpip.h> 113 #include <netinet/tcp_fastopen.h> 114 #ifdef TCPPCAP 115 #include <netinet/tcp_pcap.h> 116 #endif 117 #ifdef TCPDEBUG 118 #include <netinet/tcp_debug.h> 119 #endif 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 #include <netinet/udp.h> 124 #include <netinet/udp_var.h> 125 #ifdef INET6 126 #include <netinet6/tcp6_var.h> 127 #endif 128 129 #include <netipsec/ipsec_support.h> 130 131 #include <machine/in_cksum.h> 132 #include <crypto/siphash/siphash.h> 133 134 #include <security/mac/mac_framework.h> 135 136 #ifdef INET6 137 static ip6proto_ctlinput_t tcp6_ctlinput; 138 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 139 #endif 140 141 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 142 #ifdef INET6 143 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 144 #endif 145 146 #ifdef NETFLIX_EXP_DETECTION 147 /* Sack attack detection thresholds and such */ 148 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 149 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 150 "Sack Attack detection thresholds"); 151 int32_t tcp_force_detection = 0; 152 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 153 CTLFLAG_RW, 154 &tcp_force_detection, 0, 155 "Do we force detection even if the INP has it off?"); 156 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 157 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 158 CTLFLAG_RW, 159 &tcp_sack_to_ack_thresh, 700, 160 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 161 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 162 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 163 CTLFLAG_RW, 164 &tcp_sack_to_move_thresh, 600, 165 "Percentage of sack moves we must see above (10.1 percent is 101)"); 166 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 167 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 168 CTLFLAG_RW, 169 &tcp_restoral_thresh, 550, 170 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 171 int32_t tcp_sad_decay_val = 800; 172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 173 CTLFLAG_RW, 174 &tcp_sad_decay_val, 800, 175 "The decay percentage (10.1 percent equals 101 )"); 176 int32_t tcp_map_minimum = 500; 177 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 178 CTLFLAG_RW, 179 &tcp_map_minimum, 500, 180 "Number of Map enteries before we start detection"); 181 int32_t tcp_attack_on_turns_on_logging = 0; 182 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 183 CTLFLAG_RW, 184 &tcp_attack_on_turns_on_logging, 0, 185 "When we have a positive hit on attack, do we turn on logging?"); 186 int32_t tcp_sad_pacing_interval = 2000; 187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 188 CTLFLAG_RW, 189 &tcp_sad_pacing_interval, 2000, 190 "What is the minimum pacing interval for a classified attacker?"); 191 192 int32_t tcp_sad_low_pps = 100; 193 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 194 CTLFLAG_RW, 195 &tcp_sad_low_pps, 100, 196 "What is the input pps that below which we do not decay?"); 197 #endif 198 uint32_t tcp_ack_war_time_window = 1000; 199 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 200 CTLFLAG_RW, 201 &tcp_ack_war_time_window, 1000, 202 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 203 uint32_t tcp_ack_war_cnt = 5; 204 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 205 CTLFLAG_RW, 206 &tcp_ack_war_cnt, 5, 207 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 208 209 struct rwlock tcp_function_lock; 210 211 static int 212 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 213 { 214 int error, new; 215 216 new = V_tcp_mssdflt; 217 error = sysctl_handle_int(oidp, &new, 0, req); 218 if (error == 0 && req->newptr) { 219 if (new < TCP_MINMSS) 220 error = EINVAL; 221 else 222 V_tcp_mssdflt = new; 223 } 224 return (error); 225 } 226 227 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 228 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 229 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 230 "Default TCP Maximum Segment Size"); 231 232 #ifdef INET6 233 static int 234 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 235 { 236 int error, new; 237 238 new = V_tcp_v6mssdflt; 239 error = sysctl_handle_int(oidp, &new, 0, req); 240 if (error == 0 && req->newptr) { 241 if (new < TCP_MINMSS) 242 error = EINVAL; 243 else 244 V_tcp_v6mssdflt = new; 245 } 246 return (error); 247 } 248 249 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 250 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 251 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 252 "Default TCP Maximum Segment Size for IPv6"); 253 #endif /* INET6 */ 254 255 /* 256 * Minimum MSS we accept and use. This prevents DoS attacks where 257 * we are forced to a ridiculous low MSS like 20 and send hundreds 258 * of packets instead of one. The effect scales with the available 259 * bandwidth and quickly saturates the CPU and network interface 260 * with packet generation and sending. Set to zero to disable MINMSS 261 * checking. This setting prevents us from sending too small packets. 262 */ 263 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 264 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 265 &VNET_NAME(tcp_minmss), 0, 266 "Minimum TCP Maximum Segment Size"); 267 268 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 269 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 270 &VNET_NAME(tcp_do_rfc1323), 0, 271 "Enable rfc1323 (high performance TCP) extensions"); 272 273 /* 274 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 275 * Some stacks negotiate TS, but never send them after connection setup. Some 276 * stacks negotiate TS, but don't send them when sending keep-alive segments. 277 * These include modern widely deployed TCP stacks. 278 * Therefore tolerating violations for now... 279 */ 280 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 282 &VNET_NAME(tcp_tolerate_missing_ts), 0, 283 "Tolerate missing TCP timestamps"); 284 285 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 287 &VNET_NAME(tcp_ts_offset_per_conn), 0, 288 "Initialize TCP timestamps per connection instead of per host pair"); 289 290 /* How many connections are pacing */ 291 static volatile uint32_t number_of_tcp_connections_pacing = 0; 292 static uint32_t shadow_num_connections = 0; 293 294 static int tcp_pacing_limit = 10000; 295 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 296 &tcp_pacing_limit, 1000, 297 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 298 299 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 300 &shadow_num_connections, 0, "Number of TCP connections being paced"); 301 302 static int tcp_log_debug = 0; 303 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 304 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 305 306 static int tcp_tcbhashsize; 307 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 308 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 309 310 static int do_tcpdrain = 1; 311 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 312 "Enable tcp_drain routine for extra help when low on mbufs"); 313 314 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 315 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 316 317 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 318 #define V_icmp_may_rst VNET(icmp_may_rst) 319 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 320 &VNET_NAME(icmp_may_rst), 0, 321 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 322 323 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 324 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 325 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 326 &VNET_NAME(tcp_isn_reseed_interval), 0, 327 "Seconds between reseeding of ISN secret"); 328 329 static int tcp_soreceive_stream; 330 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 331 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 332 333 VNET_DEFINE(uma_zone_t, sack_hole_zone); 334 #define V_sack_hole_zone VNET(sack_hole_zone) 335 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 336 static int 337 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 338 { 339 int error; 340 uint32_t new; 341 342 new = V_tcp_map_entries_limit; 343 error = sysctl_handle_int(oidp, &new, 0, req); 344 if (error == 0 && req->newptr) { 345 /* only allow "0" and value > minimum */ 346 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 347 error = EINVAL; 348 else 349 V_tcp_map_entries_limit = new; 350 } 351 return (error); 352 } 353 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 354 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 355 &VNET_NAME(tcp_map_entries_limit), 0, 356 &sysctl_net_inet_tcp_map_limit_check, "IU", 357 "Total sendmap entries limit"); 358 359 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 360 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 361 &VNET_NAME(tcp_map_split_limit), 0, 362 "Total sendmap split entries limit"); 363 364 #ifdef TCP_HHOOK 365 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 366 #endif 367 368 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 369 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 370 #define V_ts_offset_secret VNET(ts_offset_secret) 371 372 static int tcp_default_fb_init(struct tcpcb *tp); 373 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 374 static int tcp_default_handoff_ok(struct tcpcb *tp); 375 static struct inpcb *tcp_notify(struct inpcb *, int); 376 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 377 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 378 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 379 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 380 const void *ip4hdr, const void *ip6hdr); 381 static ipproto_ctlinput_t tcp_ctlinput; 382 static udp_tun_icmp_t tcp_ctlinput_viaudp; 383 384 static struct tcp_function_block tcp_def_funcblk = { 385 .tfb_tcp_block_name = "freebsd", 386 .tfb_tcp_output = tcp_default_output, 387 .tfb_tcp_do_segment = tcp_do_segment, 388 .tfb_tcp_ctloutput = tcp_default_ctloutput, 389 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 390 .tfb_tcp_fb_init = tcp_default_fb_init, 391 .tfb_tcp_fb_fini = tcp_default_fb_fini, 392 }; 393 394 static int tcp_fb_cnt = 0; 395 struct tcp_funchead t_functions; 396 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 397 398 void 399 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 400 { 401 TCPSTAT_INC(tcps_dsack_count); 402 tp->t_dsack_pack++; 403 if (tlp == 0) { 404 if (SEQ_GT(end, start)) { 405 tp->t_dsack_bytes += (end - start); 406 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 407 } else { 408 tp->t_dsack_tlp_bytes += (start - end); 409 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 410 } 411 } else { 412 if (SEQ_GT(end, start)) { 413 tp->t_dsack_bytes += (end - start); 414 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 415 } else { 416 tp->t_dsack_tlp_bytes += (start - end); 417 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 418 } 419 } 420 } 421 422 static struct tcp_function_block * 423 find_tcp_functions_locked(struct tcp_function_set *fs) 424 { 425 struct tcp_function *f; 426 struct tcp_function_block *blk=NULL; 427 428 TAILQ_FOREACH(f, &t_functions, tf_next) { 429 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 430 blk = f->tf_fb; 431 break; 432 } 433 } 434 return(blk); 435 } 436 437 static struct tcp_function_block * 438 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 439 { 440 struct tcp_function_block *rblk=NULL; 441 struct tcp_function *f; 442 443 TAILQ_FOREACH(f, &t_functions, tf_next) { 444 if (f->tf_fb == blk) { 445 rblk = blk; 446 if (s) { 447 *s = f; 448 } 449 break; 450 } 451 } 452 return (rblk); 453 } 454 455 struct tcp_function_block * 456 find_and_ref_tcp_functions(struct tcp_function_set *fs) 457 { 458 struct tcp_function_block *blk; 459 460 rw_rlock(&tcp_function_lock); 461 blk = find_tcp_functions_locked(fs); 462 if (blk) 463 refcount_acquire(&blk->tfb_refcnt); 464 rw_runlock(&tcp_function_lock); 465 return(blk); 466 } 467 468 struct tcp_function_block * 469 find_and_ref_tcp_fb(struct tcp_function_block *blk) 470 { 471 struct tcp_function_block *rblk; 472 473 rw_rlock(&tcp_function_lock); 474 rblk = find_tcp_fb_locked(blk, NULL); 475 if (rblk) 476 refcount_acquire(&rblk->tfb_refcnt); 477 rw_runlock(&tcp_function_lock); 478 return(rblk); 479 } 480 481 /* Find a matching alias for the given tcp_function_block. */ 482 int 483 find_tcp_function_alias(struct tcp_function_block *blk, 484 struct tcp_function_set *fs) 485 { 486 struct tcp_function *f; 487 int found; 488 489 found = 0; 490 rw_rlock(&tcp_function_lock); 491 TAILQ_FOREACH(f, &t_functions, tf_next) { 492 if ((f->tf_fb == blk) && 493 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 494 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 495 /* Matching function block with different name. */ 496 strncpy(fs->function_set_name, f->tf_name, 497 TCP_FUNCTION_NAME_LEN_MAX); 498 found = 1; 499 break; 500 } 501 } 502 /* Null terminate the string appropriately. */ 503 if (found) { 504 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 505 } else { 506 fs->function_set_name[0] = '\0'; 507 } 508 rw_runlock(&tcp_function_lock); 509 return (found); 510 } 511 512 static struct tcp_function_block * 513 find_and_ref_tcp_default_fb(void) 514 { 515 struct tcp_function_block *rblk; 516 517 rw_rlock(&tcp_function_lock); 518 rblk = tcp_func_set_ptr; 519 refcount_acquire(&rblk->tfb_refcnt); 520 rw_runlock(&tcp_function_lock); 521 return (rblk); 522 } 523 524 void 525 tcp_switch_back_to_default(struct tcpcb *tp) 526 { 527 struct tcp_function_block *tfb; 528 529 KASSERT(tp->t_fb != &tcp_def_funcblk, 530 ("%s: called by the built-in default stack", __func__)); 531 532 /* 533 * Release the old stack. This function will either find a new one 534 * or panic. 535 */ 536 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 537 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 538 refcount_release(&tp->t_fb->tfb_refcnt); 539 540 /* 541 * Now, we'll find a new function block to use. 542 * Start by trying the current user-selected 543 * default, unless this stack is the user-selected 544 * default. 545 */ 546 tfb = find_and_ref_tcp_default_fb(); 547 if (tfb == tp->t_fb) { 548 refcount_release(&tfb->tfb_refcnt); 549 tfb = NULL; 550 } 551 /* Does the stack accept this connection? */ 552 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 553 (*tfb->tfb_tcp_handoff_ok)(tp)) { 554 refcount_release(&tfb->tfb_refcnt); 555 tfb = NULL; 556 } 557 /* Try to use that stack. */ 558 if (tfb != NULL) { 559 /* Initialize the new stack. If it succeeds, we are done. */ 560 tp->t_fb = tfb; 561 if (tp->t_fb->tfb_tcp_fb_init == NULL || 562 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 563 return; 564 565 /* 566 * Initialization failed. Release the reference count on 567 * the stack. 568 */ 569 refcount_release(&tfb->tfb_refcnt); 570 } 571 572 /* 573 * If that wasn't feasible, use the built-in default 574 * stack which is not allowed to reject anyone. 575 */ 576 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 577 if (tfb == NULL) { 578 /* there always should be a default */ 579 panic("Can't refer to tcp_def_funcblk"); 580 } 581 if (tfb->tfb_tcp_handoff_ok != NULL) { 582 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 583 /* The default stack cannot say no */ 584 panic("Default stack rejects a new session?"); 585 } 586 } 587 tp->t_fb = tfb; 588 if (tp->t_fb->tfb_tcp_fb_init != NULL && 589 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 590 /* The default stack cannot fail */ 591 panic("Default stack initialization failed"); 592 } 593 } 594 595 static bool 596 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 597 const struct sockaddr *sa, void *ctx) 598 { 599 struct ip *iph; 600 #ifdef INET6 601 struct ip6_hdr *ip6; 602 #endif 603 struct udphdr *uh; 604 struct tcphdr *th; 605 int thlen; 606 uint16_t port; 607 608 TCPSTAT_INC(tcps_tunneled_pkts); 609 if ((m->m_flags & M_PKTHDR) == 0) { 610 /* Can't handle one that is not a pkt hdr */ 611 TCPSTAT_INC(tcps_tunneled_errs); 612 goto out; 613 } 614 thlen = sizeof(struct tcphdr); 615 if (m->m_len < off + sizeof(struct udphdr) + thlen && 616 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 617 TCPSTAT_INC(tcps_tunneled_errs); 618 goto out; 619 } 620 iph = mtod(m, struct ip *); 621 uh = (struct udphdr *)((caddr_t)iph + off); 622 th = (struct tcphdr *)(uh + 1); 623 thlen = th->th_off << 2; 624 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 625 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 626 if (m == NULL) { 627 TCPSTAT_INC(tcps_tunneled_errs); 628 goto out; 629 } else { 630 iph = mtod(m, struct ip *); 631 uh = (struct udphdr *)((caddr_t)iph + off); 632 th = (struct tcphdr *)(uh + 1); 633 } 634 } 635 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 636 bcopy(th, uh, m->m_len - off); 637 m->m_len -= sizeof(struct udphdr); 638 m->m_pkthdr.len -= sizeof(struct udphdr); 639 /* 640 * We use the same algorithm for 641 * both UDP and TCP for c-sum. So 642 * the code in tcp_input will skip 643 * the checksum. So we do nothing 644 * with the flag (m->m_pkthdr.csum_flags). 645 */ 646 switch (iph->ip_v) { 647 #ifdef INET 648 case IPVERSION: 649 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 650 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 651 break; 652 #endif 653 #ifdef INET6 654 case IPV6_VERSION >> 4: 655 ip6 = mtod(m, struct ip6_hdr *); 656 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 657 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 658 break; 659 #endif 660 default: 661 goto out; 662 break; 663 } 664 return (true); 665 out: 666 m_freem(m); 667 668 return (true); 669 } 670 671 static int 672 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 673 { 674 int error=ENOENT; 675 struct tcp_function_set fs; 676 struct tcp_function_block *blk; 677 678 memset(&fs, 0, sizeof(fs)); 679 rw_rlock(&tcp_function_lock); 680 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 681 if (blk) { 682 /* Found him */ 683 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 684 fs.pcbcnt = blk->tfb_refcnt; 685 } 686 rw_runlock(&tcp_function_lock); 687 error = sysctl_handle_string(oidp, fs.function_set_name, 688 sizeof(fs.function_set_name), req); 689 690 /* Check for error or no change */ 691 if (error != 0 || req->newptr == NULL) 692 return(error); 693 694 rw_wlock(&tcp_function_lock); 695 blk = find_tcp_functions_locked(&fs); 696 if ((blk == NULL) || 697 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 698 error = ENOENT; 699 goto done; 700 } 701 tcp_func_set_ptr = blk; 702 done: 703 rw_wunlock(&tcp_function_lock); 704 return (error); 705 } 706 707 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 708 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 709 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 710 "Set/get the default TCP functions"); 711 712 static int 713 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 714 { 715 int error, cnt, linesz; 716 struct tcp_function *f; 717 char *buffer, *cp; 718 size_t bufsz, outsz; 719 bool alias; 720 721 cnt = 0; 722 rw_rlock(&tcp_function_lock); 723 TAILQ_FOREACH(f, &t_functions, tf_next) { 724 cnt++; 725 } 726 rw_runlock(&tcp_function_lock); 727 728 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 729 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 730 731 error = 0; 732 cp = buffer; 733 734 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 735 "Alias", "PCB count"); 736 cp += linesz; 737 bufsz -= linesz; 738 outsz = linesz; 739 740 rw_rlock(&tcp_function_lock); 741 TAILQ_FOREACH(f, &t_functions, tf_next) { 742 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 743 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 744 f->tf_fb->tfb_tcp_block_name, 745 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 746 alias ? f->tf_name : "-", 747 f->tf_fb->tfb_refcnt); 748 if (linesz >= bufsz) { 749 error = EOVERFLOW; 750 break; 751 } 752 cp += linesz; 753 bufsz -= linesz; 754 outsz += linesz; 755 } 756 rw_runlock(&tcp_function_lock); 757 if (error == 0) 758 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 759 free(buffer, M_TEMP); 760 return (error); 761 } 762 763 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 764 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 765 NULL, 0, sysctl_net_inet_list_available, "A", 766 "list available TCP Function sets"); 767 768 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 769 770 #ifdef INET 771 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 772 #define V_udp4_tun_socket VNET(udp4_tun_socket) 773 #endif 774 #ifdef INET6 775 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 776 #define V_udp6_tun_socket VNET(udp6_tun_socket) 777 #endif 778 779 static void 780 tcp_over_udp_stop(void) 781 { 782 /* 783 * This function assumes sysctl caller holds inp_rinfo_lock() 784 * for writing! 785 */ 786 #ifdef INET 787 if (V_udp4_tun_socket != NULL) { 788 soclose(V_udp4_tun_socket); 789 V_udp4_tun_socket = NULL; 790 } 791 #endif 792 #ifdef INET6 793 if (V_udp6_tun_socket != NULL) { 794 soclose(V_udp6_tun_socket); 795 V_udp6_tun_socket = NULL; 796 } 797 #endif 798 } 799 800 static int 801 tcp_over_udp_start(void) 802 { 803 uint16_t port; 804 int ret; 805 #ifdef INET 806 struct sockaddr_in sin; 807 #endif 808 #ifdef INET6 809 struct sockaddr_in6 sin6; 810 #endif 811 /* 812 * This function assumes sysctl caller holds inp_info_rlock() 813 * for writing! 814 */ 815 port = V_tcp_udp_tunneling_port; 816 if (ntohs(port) == 0) { 817 /* Must have a port set */ 818 return (EINVAL); 819 } 820 #ifdef INET 821 if (V_udp4_tun_socket != NULL) { 822 /* Already running -- must stop first */ 823 return (EALREADY); 824 } 825 #endif 826 #ifdef INET6 827 if (V_udp6_tun_socket != NULL) { 828 /* Already running -- must stop first */ 829 return (EALREADY); 830 } 831 #endif 832 #ifdef INET 833 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 834 SOCK_DGRAM, IPPROTO_UDP, 835 curthread->td_ucred, curthread))) { 836 tcp_over_udp_stop(); 837 return (ret); 838 } 839 /* Call the special UDP hook. */ 840 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 841 tcp_recv_udp_tunneled_packet, 842 tcp_ctlinput_viaudp, 843 NULL))) { 844 tcp_over_udp_stop(); 845 return (ret); 846 } 847 /* Ok, we have a socket, bind it to the port. */ 848 memset(&sin, 0, sizeof(struct sockaddr_in)); 849 sin.sin_len = sizeof(struct sockaddr_in); 850 sin.sin_family = AF_INET; 851 sin.sin_port = htons(port); 852 if ((ret = sobind(V_udp4_tun_socket, 853 (struct sockaddr *)&sin, curthread))) { 854 tcp_over_udp_stop(); 855 return (ret); 856 } 857 #endif 858 #ifdef INET6 859 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 860 SOCK_DGRAM, IPPROTO_UDP, 861 curthread->td_ucred, curthread))) { 862 tcp_over_udp_stop(); 863 return (ret); 864 } 865 /* Call the special UDP hook. */ 866 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 867 tcp_recv_udp_tunneled_packet, 868 tcp6_ctlinput_viaudp, 869 NULL))) { 870 tcp_over_udp_stop(); 871 return (ret); 872 } 873 /* Ok, we have a socket, bind it to the port. */ 874 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 875 sin6.sin6_len = sizeof(struct sockaddr_in6); 876 sin6.sin6_family = AF_INET6; 877 sin6.sin6_port = htons(port); 878 if ((ret = sobind(V_udp6_tun_socket, 879 (struct sockaddr *)&sin6, curthread))) { 880 tcp_over_udp_stop(); 881 return (ret); 882 } 883 #endif 884 return (0); 885 } 886 887 static int 888 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 889 { 890 int error; 891 uint32_t old, new; 892 893 old = V_tcp_udp_tunneling_port; 894 new = old; 895 error = sysctl_handle_int(oidp, &new, 0, req); 896 if ((error == 0) && 897 (req->newptr != NULL)) { 898 if ((new < TCP_TUNNELING_PORT_MIN) || 899 (new > TCP_TUNNELING_PORT_MAX)) { 900 error = EINVAL; 901 } else { 902 V_tcp_udp_tunneling_port = new; 903 if (old != 0) { 904 tcp_over_udp_stop(); 905 } 906 if (new != 0) { 907 error = tcp_over_udp_start(); 908 } 909 } 910 } 911 return (error); 912 } 913 914 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 915 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 916 &VNET_NAME(tcp_udp_tunneling_port), 917 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 918 "Tunneling port for tcp over udp"); 919 920 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 921 922 static int 923 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 924 { 925 int error, new; 926 927 new = V_tcp_udp_tunneling_overhead; 928 error = sysctl_handle_int(oidp, &new, 0, req); 929 if (error == 0 && req->newptr) { 930 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 931 (new > TCP_TUNNELING_OVERHEAD_MAX)) 932 error = EINVAL; 933 else 934 V_tcp_udp_tunneling_overhead = new; 935 } 936 return (error); 937 } 938 939 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 940 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 941 &VNET_NAME(tcp_udp_tunneling_overhead), 942 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 943 "MSS reduction when using tcp over udp"); 944 945 /* 946 * Exports one (struct tcp_function_info) for each alias/name. 947 */ 948 static int 949 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 950 { 951 int cnt, error; 952 struct tcp_function *f; 953 struct tcp_function_info tfi; 954 955 /* 956 * We don't allow writes. 957 */ 958 if (req->newptr != NULL) 959 return (EINVAL); 960 961 /* 962 * Wire the old buffer so we can directly copy the functions to 963 * user space without dropping the lock. 964 */ 965 if (req->oldptr != NULL) { 966 error = sysctl_wire_old_buffer(req, 0); 967 if (error) 968 return (error); 969 } 970 971 /* 972 * Walk the list and copy out matching entries. If INVARIANTS 973 * is compiled in, also walk the list to verify the length of 974 * the list matches what we have recorded. 975 */ 976 rw_rlock(&tcp_function_lock); 977 978 cnt = 0; 979 #ifndef INVARIANTS 980 if (req->oldptr == NULL) { 981 cnt = tcp_fb_cnt; 982 goto skip_loop; 983 } 984 #endif 985 TAILQ_FOREACH(f, &t_functions, tf_next) { 986 #ifdef INVARIANTS 987 cnt++; 988 #endif 989 if (req->oldptr != NULL) { 990 bzero(&tfi, sizeof(tfi)); 991 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 992 tfi.tfi_id = f->tf_fb->tfb_id; 993 (void)strlcpy(tfi.tfi_alias, f->tf_name, 994 sizeof(tfi.tfi_alias)); 995 (void)strlcpy(tfi.tfi_name, 996 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 997 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 998 /* 999 * Don't stop on error, as that is the 1000 * mechanism we use to accumulate length 1001 * information if the buffer was too short. 1002 */ 1003 } 1004 } 1005 KASSERT(cnt == tcp_fb_cnt, 1006 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1007 #ifndef INVARIANTS 1008 skip_loop: 1009 #endif 1010 rw_runlock(&tcp_function_lock); 1011 if (req->oldptr == NULL) 1012 error = SYSCTL_OUT(req, NULL, 1013 (cnt + 1) * sizeof(struct tcp_function_info)); 1014 1015 return (error); 1016 } 1017 1018 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1019 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1020 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1021 "List TCP function block name-to-ID mappings"); 1022 1023 /* 1024 * tfb_tcp_handoff_ok() function for the default stack. 1025 * Note that we'll basically try to take all comers. 1026 */ 1027 static int 1028 tcp_default_handoff_ok(struct tcpcb *tp) 1029 { 1030 1031 return (0); 1032 } 1033 1034 /* 1035 * tfb_tcp_fb_init() function for the default stack. 1036 * 1037 * This handles making sure we have appropriate timers set if you are 1038 * transitioning a socket that has some amount of setup done. 1039 * 1040 * The init() fuction from the default can *never* return non-zero i.e. 1041 * it is required to always succeed since it is the stack of last resort! 1042 */ 1043 static int 1044 tcp_default_fb_init(struct tcpcb *tp) 1045 { 1046 1047 struct socket *so; 1048 1049 INP_WLOCK_ASSERT(tp->t_inpcb); 1050 1051 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 1052 ("%s: connection %p in unexpected state %d", __func__, tp, 1053 tp->t_state)); 1054 1055 /* 1056 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1057 * know what to do for unexpected states (which includes TIME_WAIT). 1058 */ 1059 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1060 return (0); 1061 1062 /* 1063 * Make sure some kind of transmission timer is set if there is 1064 * outstanding data. 1065 */ 1066 so = tp->t_inpcb->inp_socket; 1067 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1068 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1069 tcp_timer_active(tp, TT_PERSIST))) { 1070 /* 1071 * If the session has established and it looks like it should 1072 * be in the persist state, set the persist timer. Otherwise, 1073 * set the retransmit timer. 1074 */ 1075 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1076 (int32_t)(tp->snd_nxt - tp->snd_una) < 1077 (int32_t)sbavail(&so->so_snd)) 1078 tcp_setpersist(tp); 1079 else 1080 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1081 } 1082 1083 /* All non-embryonic sessions get a keepalive timer. */ 1084 if (!tcp_timer_active(tp, TT_KEEP)) 1085 tcp_timer_activate(tp, TT_KEEP, 1086 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1087 TP_KEEPINIT(tp)); 1088 1089 /* 1090 * Make sure critical variables are initialized 1091 * if transitioning while in Recovery. 1092 */ 1093 if IN_FASTRECOVERY(tp->t_flags) { 1094 if (tp->sackhint.recover_fs == 0) 1095 tp->sackhint.recover_fs = max(1, 1096 tp->snd_nxt - tp->snd_una); 1097 } 1098 1099 return (0); 1100 } 1101 1102 /* 1103 * tfb_tcp_fb_fini() function for the default stack. 1104 * 1105 * This changes state as necessary (or prudent) to prepare for another stack 1106 * to assume responsibility for the connection. 1107 */ 1108 static void 1109 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1110 { 1111 1112 INP_WLOCK_ASSERT(tp->t_inpcb); 1113 return; 1114 } 1115 1116 /* 1117 * Target size of TCP PCB hash tables. Must be a power of two. 1118 * 1119 * Note that this can be overridden by the kernel environment 1120 * variable net.inet.tcp.tcbhashsize 1121 */ 1122 #ifndef TCBHASHSIZE 1123 #define TCBHASHSIZE 0 1124 #endif 1125 1126 /* 1127 * XXX 1128 * Callouts should be moved into struct tcp directly. They are currently 1129 * separate because the tcpcb structure is exported to userland for sysctl 1130 * parsing purposes, which do not know about callouts. 1131 */ 1132 struct tcpcb_mem { 1133 struct tcpcb tcb; 1134 struct tcp_timer tt; 1135 struct cc_var ccv; 1136 #ifdef TCP_HHOOK 1137 struct osd osd; 1138 #endif 1139 }; 1140 1141 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 1142 #define V_tcpcb_zone VNET(tcpcb_zone) 1143 1144 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1145 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1146 1147 static struct mtx isn_mtx; 1148 1149 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1150 #define ISN_LOCK() mtx_lock(&isn_mtx) 1151 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1152 1153 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1154 1155 /* 1156 * Take a value and get the next power of 2 that doesn't overflow. 1157 * Used to size the tcp_inpcb hash buckets. 1158 */ 1159 static int 1160 maketcp_hashsize(int size) 1161 { 1162 int hashsize; 1163 1164 /* 1165 * auto tune. 1166 * get the next power of 2 higher than maxsockets. 1167 */ 1168 hashsize = 1 << fls(size); 1169 /* catch overflow, and just go one power of 2 smaller */ 1170 if (hashsize < size) { 1171 hashsize = 1 << (fls(size) - 1); 1172 } 1173 return (hashsize); 1174 } 1175 1176 static volatile int next_tcp_stack_id = 1; 1177 1178 /* 1179 * Register a TCP function block with the name provided in the names 1180 * array. (Note that this function does NOT automatically register 1181 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1182 * explicitly include blk->tfb_tcp_block_name in the list of names if 1183 * you wish to register the stack with that name.) 1184 * 1185 * Either all name registrations will succeed or all will fail. If 1186 * a name registration fails, the function will update the num_names 1187 * argument to point to the array index of the name that encountered 1188 * the failure. 1189 * 1190 * Returns 0 on success, or an error code on failure. 1191 */ 1192 int 1193 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1194 const char *names[], int *num_names) 1195 { 1196 struct tcp_function *n; 1197 struct tcp_function_set fs; 1198 int error, i; 1199 1200 KASSERT(names != NULL && *num_names > 0, 1201 ("%s: Called with 0-length name list", __func__)); 1202 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1203 KASSERT(rw_initialized(&tcp_function_lock), 1204 ("%s: called too early", __func__)); 1205 1206 if ((blk->tfb_tcp_output == NULL) || 1207 (blk->tfb_tcp_do_segment == NULL) || 1208 (blk->tfb_tcp_ctloutput == NULL) || 1209 (strlen(blk->tfb_tcp_block_name) == 0)) { 1210 /* 1211 * These functions are required and you 1212 * need a name. 1213 */ 1214 *num_names = 0; 1215 return (EINVAL); 1216 } 1217 if (blk->tfb_tcp_timer_stop_all || 1218 blk->tfb_tcp_timer_activate || 1219 blk->tfb_tcp_timer_active || 1220 blk->tfb_tcp_timer_stop) { 1221 /* 1222 * If you define one timer function you 1223 * must have them all. 1224 */ 1225 if ((blk->tfb_tcp_timer_stop_all == NULL) || 1226 (blk->tfb_tcp_timer_activate == NULL) || 1227 (blk->tfb_tcp_timer_active == NULL) || 1228 (blk->tfb_tcp_timer_stop == NULL)) { 1229 *num_names = 0; 1230 return (EINVAL); 1231 } 1232 } 1233 1234 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1235 *num_names = 0; 1236 return (EINVAL); 1237 } 1238 1239 refcount_init(&blk->tfb_refcnt, 0); 1240 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1241 for (i = 0; i < *num_names; i++) { 1242 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1243 if (n == NULL) { 1244 error = ENOMEM; 1245 goto cleanup; 1246 } 1247 n->tf_fb = blk; 1248 1249 (void)strlcpy(fs.function_set_name, names[i], 1250 sizeof(fs.function_set_name)); 1251 rw_wlock(&tcp_function_lock); 1252 if (find_tcp_functions_locked(&fs) != NULL) { 1253 /* Duplicate name space not allowed */ 1254 rw_wunlock(&tcp_function_lock); 1255 free(n, M_TCPFUNCTIONS); 1256 error = EALREADY; 1257 goto cleanup; 1258 } 1259 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1260 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1261 tcp_fb_cnt++; 1262 rw_wunlock(&tcp_function_lock); 1263 } 1264 return(0); 1265 1266 cleanup: 1267 /* 1268 * Deregister the names we just added. Because registration failed 1269 * for names[i], we don't need to deregister that name. 1270 */ 1271 *num_names = i; 1272 rw_wlock(&tcp_function_lock); 1273 while (--i >= 0) { 1274 TAILQ_FOREACH(n, &t_functions, tf_next) { 1275 if (!strncmp(n->tf_name, names[i], 1276 TCP_FUNCTION_NAME_LEN_MAX)) { 1277 TAILQ_REMOVE(&t_functions, n, tf_next); 1278 tcp_fb_cnt--; 1279 n->tf_fb = NULL; 1280 free(n, M_TCPFUNCTIONS); 1281 break; 1282 } 1283 } 1284 } 1285 rw_wunlock(&tcp_function_lock); 1286 return (error); 1287 } 1288 1289 /* 1290 * Register a TCP function block using the name provided in the name 1291 * argument. 1292 * 1293 * Returns 0 on success, or an error code on failure. 1294 */ 1295 int 1296 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1297 int wait) 1298 { 1299 const char *name_list[1]; 1300 int num_names, rv; 1301 1302 num_names = 1; 1303 if (name != NULL) 1304 name_list[0] = name; 1305 else 1306 name_list[0] = blk->tfb_tcp_block_name; 1307 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1308 return (rv); 1309 } 1310 1311 /* 1312 * Register a TCP function block using the name defined in 1313 * blk->tfb_tcp_block_name. 1314 * 1315 * Returns 0 on success, or an error code on failure. 1316 */ 1317 int 1318 register_tcp_functions(struct tcp_function_block *blk, int wait) 1319 { 1320 1321 return (register_tcp_functions_as_name(blk, NULL, wait)); 1322 } 1323 1324 /* 1325 * Deregister all names associated with a function block. This 1326 * functionally removes the function block from use within the system. 1327 * 1328 * When called with a true quiesce argument, mark the function block 1329 * as being removed so no more stacks will use it and determine 1330 * whether the removal would succeed. 1331 * 1332 * When called with a false quiesce argument, actually attempt the 1333 * removal. 1334 * 1335 * When called with a force argument, attempt to switch all TCBs to 1336 * use the default stack instead of returning EBUSY. 1337 * 1338 * Returns 0 on success (or if the removal would succeed, or an error 1339 * code on failure. 1340 */ 1341 int 1342 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1343 bool force) 1344 { 1345 struct tcp_function *f; 1346 1347 if (blk == &tcp_def_funcblk) { 1348 /* You can't un-register the default */ 1349 return (EPERM); 1350 } 1351 rw_wlock(&tcp_function_lock); 1352 if (blk == tcp_func_set_ptr) { 1353 /* You can't free the current default */ 1354 rw_wunlock(&tcp_function_lock); 1355 return (EBUSY); 1356 } 1357 /* Mark the block so no more stacks can use it. */ 1358 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1359 /* 1360 * If TCBs are still attached to the stack, attempt to switch them 1361 * to the default stack. 1362 */ 1363 if (force && blk->tfb_refcnt) { 1364 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1365 INPLOOKUP_WLOCKPCB); 1366 struct inpcb *inp; 1367 struct tcpcb *tp; 1368 VNET_ITERATOR_DECL(vnet_iter); 1369 1370 rw_wunlock(&tcp_function_lock); 1371 1372 VNET_LIST_RLOCK(); 1373 VNET_FOREACH(vnet_iter) { 1374 CURVNET_SET(vnet_iter); 1375 while ((inp = inp_next(&inpi)) != NULL) { 1376 tp = intotcpcb(inp); 1377 if (tp == NULL || tp->t_fb != blk) 1378 continue; 1379 tcp_switch_back_to_default(tp); 1380 } 1381 CURVNET_RESTORE(); 1382 } 1383 VNET_LIST_RUNLOCK(); 1384 1385 rw_wlock(&tcp_function_lock); 1386 } 1387 if (blk->tfb_refcnt) { 1388 /* TCBs still attached. */ 1389 rw_wunlock(&tcp_function_lock); 1390 return (EBUSY); 1391 } 1392 if (quiesce) { 1393 /* Skip removal. */ 1394 rw_wunlock(&tcp_function_lock); 1395 return (0); 1396 } 1397 /* Remove any function names that map to this function block. */ 1398 while (find_tcp_fb_locked(blk, &f) != NULL) { 1399 TAILQ_REMOVE(&t_functions, f, tf_next); 1400 tcp_fb_cnt--; 1401 f->tf_fb = NULL; 1402 free(f, M_TCPFUNCTIONS); 1403 } 1404 rw_wunlock(&tcp_function_lock); 1405 return (0); 1406 } 1407 1408 static void 1409 tcp_drain(void) 1410 { 1411 struct epoch_tracker et; 1412 VNET_ITERATOR_DECL(vnet_iter); 1413 1414 if (!do_tcpdrain) 1415 return; 1416 1417 NET_EPOCH_ENTER(et); 1418 VNET_LIST_RLOCK_NOSLEEP(); 1419 VNET_FOREACH(vnet_iter) { 1420 CURVNET_SET(vnet_iter); 1421 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1422 INPLOOKUP_WLOCKPCB); 1423 struct inpcb *inpb; 1424 struct tcpcb *tcpb; 1425 1426 /* 1427 * Walk the tcpbs, if existing, and flush the reassembly queue, 1428 * if there is one... 1429 * XXX: The "Net/3" implementation doesn't imply that the TCP 1430 * reassembly queue should be flushed, but in a situation 1431 * where we're really low on mbufs, this is potentially 1432 * useful. 1433 */ 1434 while ((inpb = inp_next(&inpi)) != NULL) { 1435 if ((tcpb = intotcpcb(inpb)) != NULL) { 1436 tcp_reass_flush(tcpb); 1437 tcp_clean_sackreport(tcpb); 1438 #ifdef TCP_BLACKBOX 1439 tcp_log_drain(tcpb); 1440 #endif 1441 #ifdef TCPPCAP 1442 if (tcp_pcap_aggressive_free) { 1443 /* Free the TCP PCAP queues. */ 1444 tcp_pcap_drain(&(tcpb->t_inpkts)); 1445 tcp_pcap_drain(&(tcpb->t_outpkts)); 1446 } 1447 #endif 1448 } 1449 } 1450 CURVNET_RESTORE(); 1451 } 1452 VNET_LIST_RUNLOCK_NOSLEEP(); 1453 NET_EPOCH_EXIT(et); 1454 } 1455 1456 static void 1457 tcp_vnet_init(void *arg __unused) 1458 { 1459 1460 #ifdef TCP_HHOOK 1461 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1462 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1463 printf("%s: WARNING: unable to register helper hook\n", __func__); 1464 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1465 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1466 printf("%s: WARNING: unable to register helper hook\n", __func__); 1467 #endif 1468 #ifdef STATS 1469 if (tcp_stats_init()) 1470 printf("%s: WARNING: unable to initialise TCP stats\n", 1471 __func__); 1472 #endif 1473 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1474 tcp_tcbhashsize); 1475 1476 /* 1477 * These have to be type stable for the benefit of the timers. 1478 */ 1479 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1480 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1481 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1482 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1483 1484 syncache_init(); 1485 tcp_hc_init(); 1486 1487 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1488 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1489 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1490 1491 tcp_fastopen_init(); 1492 1493 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1494 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1495 1496 V_tcp_msl = TCPTV_MSL; 1497 } 1498 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1499 tcp_vnet_init, NULL); 1500 1501 static void 1502 tcp_init(void *arg __unused) 1503 { 1504 const char *tcbhash_tuneable; 1505 int hashsize; 1506 1507 tcp_reass_global_init(); 1508 1509 /* XXX virtualize those below? */ 1510 tcp_delacktime = TCPTV_DELACK; 1511 tcp_keepinit = TCPTV_KEEP_INIT; 1512 tcp_keepidle = TCPTV_KEEP_IDLE; 1513 tcp_keepintvl = TCPTV_KEEPINTVL; 1514 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1515 tcp_rexmit_initial = TCPTV_RTOBASE; 1516 if (tcp_rexmit_initial < 1) 1517 tcp_rexmit_initial = 1; 1518 tcp_rexmit_min = TCPTV_MIN; 1519 if (tcp_rexmit_min < 1) 1520 tcp_rexmit_min = 1; 1521 tcp_persmin = TCPTV_PERSMIN; 1522 tcp_persmax = TCPTV_PERSMAX; 1523 tcp_rexmit_slop = TCPTV_CPU_VAR; 1524 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1525 1526 /* Setup the tcp function block list */ 1527 TAILQ_INIT(&t_functions); 1528 rw_init(&tcp_function_lock, "tcp_func_lock"); 1529 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1530 #ifdef TCP_BLACKBOX 1531 /* Initialize the TCP logging data. */ 1532 tcp_log_init(); 1533 #endif 1534 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1535 1536 if (tcp_soreceive_stream) { 1537 #ifdef INET 1538 tcp_protosw.pr_soreceive = soreceive_stream; 1539 #endif 1540 #ifdef INET6 1541 tcp6_protosw.pr_soreceive = soreceive_stream; 1542 #endif /* INET6 */ 1543 } 1544 1545 #ifdef INET6 1546 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1547 #else /* INET6 */ 1548 max_protohdr_grow(sizeof(struct tcpiphdr)); 1549 #endif /* INET6 */ 1550 1551 ISN_LOCK_INIT(); 1552 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1553 SHUTDOWN_PRI_DEFAULT); 1554 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1555 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1556 1557 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1558 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1559 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1560 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1561 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1562 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1563 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1564 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1565 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1566 #ifdef TCPPCAP 1567 tcp_pcap_init(); 1568 #endif 1569 1570 hashsize = TCBHASHSIZE; 1571 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1572 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1573 if (hashsize == 0) { 1574 /* 1575 * Auto tune the hash size based on maxsockets. 1576 * A perfect hash would have a 1:1 mapping 1577 * (hashsize = maxsockets) however it's been 1578 * suggested that O(2) average is better. 1579 */ 1580 hashsize = maketcp_hashsize(maxsockets / 4); 1581 /* 1582 * Our historical default is 512, 1583 * do not autotune lower than this. 1584 */ 1585 if (hashsize < 512) 1586 hashsize = 512; 1587 if (bootverbose) 1588 printf("%s: %s auto tuned to %d\n", __func__, 1589 tcbhash_tuneable, hashsize); 1590 } 1591 /* 1592 * We require a hashsize to be a power of two. 1593 * Previously if it was not a power of two we would just reset it 1594 * back to 512, which could be a nasty surprise if you did not notice 1595 * the error message. 1596 * Instead what we do is clip it to the closest power of two lower 1597 * than the specified hash value. 1598 */ 1599 if (!powerof2(hashsize)) { 1600 int oldhashsize = hashsize; 1601 1602 hashsize = maketcp_hashsize(hashsize); 1603 /* prevent absurdly low value */ 1604 if (hashsize < 16) 1605 hashsize = 16; 1606 printf("%s: WARNING: TCB hash size not a power of 2, " 1607 "clipped from %d to %d.\n", __func__, oldhashsize, 1608 hashsize); 1609 } 1610 tcp_tcbhashsize = hashsize; 1611 1612 #ifdef INET 1613 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1614 #endif 1615 #ifdef INET6 1616 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1617 #endif 1618 } 1619 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1620 1621 #ifdef VIMAGE 1622 static void 1623 tcp_destroy(void *unused __unused) 1624 { 1625 int n; 1626 #ifdef TCP_HHOOK 1627 int error; 1628 #endif 1629 1630 /* 1631 * All our processes are gone, all our sockets should be cleaned 1632 * up, which means, we should be past the tcp_discardcb() calls. 1633 * Sleep to let all tcpcb timers really disappear and cleanup. 1634 */ 1635 for (;;) { 1636 INP_INFO_WLOCK(&V_tcbinfo); 1637 n = V_tcbinfo.ipi_count; 1638 INP_INFO_WUNLOCK(&V_tcbinfo); 1639 if (n == 0) 1640 break; 1641 pause("tcpdes", hz / 10); 1642 } 1643 tcp_hc_destroy(); 1644 syncache_destroy(); 1645 in_pcbinfo_destroy(&V_tcbinfo); 1646 /* tcp_discardcb() clears the sack_holes up. */ 1647 uma_zdestroy(V_sack_hole_zone); 1648 uma_zdestroy(V_tcpcb_zone); 1649 1650 /* 1651 * Cannot free the zone until all tcpcbs are released as we attach 1652 * the allocations to them. 1653 */ 1654 tcp_fastopen_destroy(); 1655 1656 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1657 VNET_PCPUSTAT_FREE(tcpstat); 1658 1659 #ifdef TCP_HHOOK 1660 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1661 if (error != 0) { 1662 printf("%s: WARNING: unable to deregister helper hook " 1663 "type=%d, id=%d: error %d returned\n", __func__, 1664 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1665 } 1666 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1667 if (error != 0) { 1668 printf("%s: WARNING: unable to deregister helper hook " 1669 "type=%d, id=%d: error %d returned\n", __func__, 1670 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1671 } 1672 #endif 1673 } 1674 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1675 #endif 1676 1677 void 1678 tcp_fini(void *xtp) 1679 { 1680 1681 } 1682 1683 /* 1684 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1685 * tcp_template used to store this data in mbufs, but we now recopy it out 1686 * of the tcpcb each time to conserve mbufs. 1687 */ 1688 void 1689 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1690 { 1691 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1692 1693 INP_WLOCK_ASSERT(inp); 1694 1695 #ifdef INET6 1696 if ((inp->inp_vflag & INP_IPV6) != 0) { 1697 struct ip6_hdr *ip6; 1698 1699 ip6 = (struct ip6_hdr *)ip_ptr; 1700 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1701 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1702 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1703 (IPV6_VERSION & IPV6_VERSION_MASK); 1704 if (port == 0) 1705 ip6->ip6_nxt = IPPROTO_TCP; 1706 else 1707 ip6->ip6_nxt = IPPROTO_UDP; 1708 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1709 ip6->ip6_src = inp->in6p_laddr; 1710 ip6->ip6_dst = inp->in6p_faddr; 1711 } 1712 #endif /* INET6 */ 1713 #if defined(INET6) && defined(INET) 1714 else 1715 #endif 1716 #ifdef INET 1717 { 1718 struct ip *ip; 1719 1720 ip = (struct ip *)ip_ptr; 1721 ip->ip_v = IPVERSION; 1722 ip->ip_hl = 5; 1723 ip->ip_tos = inp->inp_ip_tos; 1724 ip->ip_len = 0; 1725 ip->ip_id = 0; 1726 ip->ip_off = 0; 1727 ip->ip_ttl = inp->inp_ip_ttl; 1728 ip->ip_sum = 0; 1729 if (port == 0) 1730 ip->ip_p = IPPROTO_TCP; 1731 else 1732 ip->ip_p = IPPROTO_UDP; 1733 ip->ip_src = inp->inp_laddr; 1734 ip->ip_dst = inp->inp_faddr; 1735 } 1736 #endif /* INET */ 1737 th->th_sport = inp->inp_lport; 1738 th->th_dport = inp->inp_fport; 1739 th->th_seq = 0; 1740 th->th_ack = 0; 1741 th->th_off = 5; 1742 tcp_set_flags(th, 0); 1743 th->th_win = 0; 1744 th->th_urp = 0; 1745 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1746 } 1747 1748 /* 1749 * Create template to be used to send tcp packets on a connection. 1750 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1751 * use for this function is in keepalives, which use tcp_respond. 1752 */ 1753 struct tcptemp * 1754 tcpip_maketemplate(struct inpcb *inp) 1755 { 1756 struct tcptemp *t; 1757 1758 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1759 if (t == NULL) 1760 return (NULL); 1761 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1762 return (t); 1763 } 1764 1765 /* 1766 * Send a single message to the TCP at address specified by 1767 * the given TCP/IP header. If m == NULL, then we make a copy 1768 * of the tcpiphdr at th and send directly to the addressed host. 1769 * This is used to force keep alive messages out using the TCP 1770 * template for a connection. If flags are given then we send 1771 * a message back to the TCP which originated the segment th, 1772 * and discard the mbuf containing it and any other attached mbufs. 1773 * 1774 * In any case the ack and sequence number of the transmitted 1775 * segment are as specified by the parameters. 1776 * 1777 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1778 */ 1779 void 1780 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1781 tcp_seq ack, tcp_seq seq, int flags) 1782 { 1783 struct tcpopt to; 1784 struct inpcb *inp; 1785 struct ip *ip; 1786 struct mbuf *optm; 1787 struct udphdr *uh = NULL; 1788 struct tcphdr *nth; 1789 struct tcp_log_buffer *lgb; 1790 u_char *optp; 1791 #ifdef INET6 1792 struct ip6_hdr *ip6; 1793 int isipv6; 1794 #endif /* INET6 */ 1795 int optlen, tlen, win, ulen; 1796 bool incl_opts; 1797 uint16_t port; 1798 int output_ret; 1799 #ifdef INVARIANTS 1800 int thflags = tcp_get_flags(th); 1801 #endif 1802 1803 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1804 NET_EPOCH_ASSERT(); 1805 1806 #ifdef INET6 1807 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1808 ip6 = ipgen; 1809 #endif /* INET6 */ 1810 ip = ipgen; 1811 1812 if (tp != NULL) { 1813 inp = tp->t_inpcb; 1814 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1815 INP_LOCK_ASSERT(inp); 1816 } else 1817 inp = NULL; 1818 1819 if (m != NULL) { 1820 #ifdef INET6 1821 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1822 port = m->m_pkthdr.tcp_tun_port; 1823 else 1824 #endif 1825 if (ip && (ip->ip_p == IPPROTO_UDP)) 1826 port = m->m_pkthdr.tcp_tun_port; 1827 else 1828 port = 0; 1829 } else 1830 port = tp->t_port; 1831 1832 incl_opts = false; 1833 win = 0; 1834 if (tp != NULL) { 1835 if (!(flags & TH_RST)) { 1836 win = sbspace(&inp->inp_socket->so_rcv); 1837 if (win > TCP_MAXWIN << tp->rcv_scale) 1838 win = TCP_MAXWIN << tp->rcv_scale; 1839 } 1840 if ((tp->t_flags & TF_NOOPT) == 0) 1841 incl_opts = true; 1842 } 1843 if (m == NULL) { 1844 m = m_gethdr(M_NOWAIT, MT_DATA); 1845 if (m == NULL) 1846 return; 1847 m->m_data += max_linkhdr; 1848 #ifdef INET6 1849 if (isipv6) { 1850 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1851 sizeof(struct ip6_hdr)); 1852 ip6 = mtod(m, struct ip6_hdr *); 1853 nth = (struct tcphdr *)(ip6 + 1); 1854 if (port) { 1855 /* Insert a UDP header */ 1856 uh = (struct udphdr *)nth; 1857 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1858 uh->uh_dport = port; 1859 nth = (struct tcphdr *)(uh + 1); 1860 } 1861 } else 1862 #endif /* INET6 */ 1863 { 1864 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1865 ip = mtod(m, struct ip *); 1866 nth = (struct tcphdr *)(ip + 1); 1867 if (port) { 1868 /* Insert a UDP header */ 1869 uh = (struct udphdr *)nth; 1870 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1871 uh->uh_dport = port; 1872 nth = (struct tcphdr *)(uh + 1); 1873 } 1874 } 1875 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1876 flags = TH_ACK; 1877 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1878 struct mbuf *n; 1879 1880 /* Can't reuse 'm', allocate a new mbuf. */ 1881 n = m_gethdr(M_NOWAIT, MT_DATA); 1882 if (n == NULL) { 1883 m_freem(m); 1884 return; 1885 } 1886 1887 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1888 m_freem(m); 1889 m_freem(n); 1890 return; 1891 } 1892 1893 n->m_data += max_linkhdr; 1894 /* m_len is set later */ 1895 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1896 #ifdef INET6 1897 if (isipv6) { 1898 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1899 sizeof(struct ip6_hdr)); 1900 ip6 = mtod(n, struct ip6_hdr *); 1901 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1902 nth = (struct tcphdr *)(ip6 + 1); 1903 if (port) { 1904 /* Insert a UDP header */ 1905 uh = (struct udphdr *)nth; 1906 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1907 uh->uh_dport = port; 1908 nth = (struct tcphdr *)(uh + 1); 1909 } 1910 } else 1911 #endif /* INET6 */ 1912 { 1913 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1914 ip = mtod(n, struct ip *); 1915 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1916 nth = (struct tcphdr *)(ip + 1); 1917 if (port) { 1918 /* Insert a UDP header */ 1919 uh = (struct udphdr *)nth; 1920 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1921 uh->uh_dport = port; 1922 nth = (struct tcphdr *)(uh + 1); 1923 } 1924 } 1925 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1926 xchg(nth->th_dport, nth->th_sport, uint16_t); 1927 th = nth; 1928 m_freem(m); 1929 m = n; 1930 } else { 1931 /* 1932 * reuse the mbuf. 1933 * XXX MRT We inherit the FIB, which is lucky. 1934 */ 1935 m_freem(m->m_next); 1936 m->m_next = NULL; 1937 m->m_data = (caddr_t)ipgen; 1938 /* m_len is set later */ 1939 #ifdef INET6 1940 if (isipv6) { 1941 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1942 nth = (struct tcphdr *)(ip6 + 1); 1943 } else 1944 #endif /* INET6 */ 1945 { 1946 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1947 nth = (struct tcphdr *)(ip + 1); 1948 } 1949 if (th != nth) { 1950 /* 1951 * this is usually a case when an extension header 1952 * exists between the IPv6 header and the 1953 * TCP header. 1954 */ 1955 nth->th_sport = th->th_sport; 1956 nth->th_dport = th->th_dport; 1957 } 1958 xchg(nth->th_dport, nth->th_sport, uint16_t); 1959 #undef xchg 1960 } 1961 tlen = 0; 1962 #ifdef INET6 1963 if (isipv6) 1964 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1965 #endif 1966 #if defined(INET) && defined(INET6) 1967 else 1968 #endif 1969 #ifdef INET 1970 tlen = sizeof (struct tcpiphdr); 1971 #endif 1972 if (port) 1973 tlen += sizeof (struct udphdr); 1974 #ifdef INVARIANTS 1975 m->m_len = 0; 1976 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1977 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1978 m, tlen, (long)M_TRAILINGSPACE(m))); 1979 #endif 1980 m->m_len = tlen; 1981 to.to_flags = 0; 1982 if (incl_opts) { 1983 /* Make sure we have room. */ 1984 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1985 m->m_next = m_get(M_NOWAIT, MT_DATA); 1986 if (m->m_next) { 1987 optp = mtod(m->m_next, u_char *); 1988 optm = m->m_next; 1989 } else 1990 incl_opts = false; 1991 } else { 1992 optp = (u_char *) (nth + 1); 1993 optm = m; 1994 } 1995 } 1996 if (incl_opts) { 1997 /* Timestamps. */ 1998 if (tp->t_flags & TF_RCVD_TSTMP) { 1999 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 2000 to.to_tsecr = tp->ts_recent; 2001 to.to_flags |= TOF_TS; 2002 } 2003 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2004 /* TCP-MD5 (RFC2385). */ 2005 if (tp->t_flags & TF_SIGNATURE) 2006 to.to_flags |= TOF_SIGNATURE; 2007 #endif 2008 /* Add the options. */ 2009 tlen += optlen = tcp_addoptions(&to, optp); 2010 2011 /* Update m_len in the correct mbuf. */ 2012 optm->m_len += optlen; 2013 } else 2014 optlen = 0; 2015 #ifdef INET6 2016 if (isipv6) { 2017 if (uh) { 2018 ulen = tlen - sizeof(struct ip6_hdr); 2019 uh->uh_ulen = htons(ulen); 2020 } 2021 ip6->ip6_flow = 0; 2022 ip6->ip6_vfc = IPV6_VERSION; 2023 if (port) 2024 ip6->ip6_nxt = IPPROTO_UDP; 2025 else 2026 ip6->ip6_nxt = IPPROTO_TCP; 2027 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 2028 } 2029 #endif 2030 #if defined(INET) && defined(INET6) 2031 else 2032 #endif 2033 #ifdef INET 2034 { 2035 if (uh) { 2036 ulen = tlen - sizeof(struct ip); 2037 uh->uh_ulen = htons(ulen); 2038 } 2039 ip->ip_len = htons(tlen); 2040 ip->ip_ttl = V_ip_defttl; 2041 if (port) { 2042 ip->ip_p = IPPROTO_UDP; 2043 } else { 2044 ip->ip_p = IPPROTO_TCP; 2045 } 2046 if (V_path_mtu_discovery) 2047 ip->ip_off |= htons(IP_DF); 2048 } 2049 #endif 2050 m->m_pkthdr.len = tlen; 2051 m->m_pkthdr.rcvif = NULL; 2052 #ifdef MAC 2053 if (inp != NULL) { 2054 /* 2055 * Packet is associated with a socket, so allow the 2056 * label of the response to reflect the socket label. 2057 */ 2058 INP_LOCK_ASSERT(inp); 2059 mac_inpcb_create_mbuf(inp, m); 2060 } else { 2061 /* 2062 * Packet is not associated with a socket, so possibly 2063 * update the label in place. 2064 */ 2065 mac_netinet_tcp_reply(m); 2066 } 2067 #endif 2068 nth->th_seq = htonl(seq); 2069 nth->th_ack = htonl(ack); 2070 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2071 tcp_set_flags(nth, flags); 2072 if (tp != NULL) 2073 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2074 else 2075 nth->th_win = htons((u_short)win); 2076 nth->th_urp = 0; 2077 2078 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2079 if (to.to_flags & TOF_SIGNATURE) { 2080 if (!TCPMD5_ENABLED() || 2081 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2082 m_freem(m); 2083 return; 2084 } 2085 } 2086 #endif 2087 2088 #ifdef INET6 2089 if (isipv6) { 2090 if (port) { 2091 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2092 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2093 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2094 nth->th_sum = 0; 2095 } else { 2096 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2097 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2098 nth->th_sum = in6_cksum_pseudo(ip6, 2099 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2100 } 2101 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 2102 NULL, NULL); 2103 } 2104 #endif /* INET6 */ 2105 #if defined(INET6) && defined(INET) 2106 else 2107 #endif 2108 #ifdef INET 2109 { 2110 if (port) { 2111 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2112 htons(ulen + IPPROTO_UDP)); 2113 m->m_pkthdr.csum_flags = CSUM_UDP; 2114 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2115 nth->th_sum = 0; 2116 } else { 2117 m->m_pkthdr.csum_flags = CSUM_TCP; 2118 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2119 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2120 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2121 } 2122 } 2123 #endif /* INET */ 2124 #ifdef TCPDEBUG 2125 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2126 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2127 #endif 2128 TCP_PROBE3(debug__output, tp, th, m); 2129 if (flags & TH_RST) 2130 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2131 lgb = NULL; 2132 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { 2133 if (INP_WLOCKED(inp)) { 2134 union tcp_log_stackspecific log; 2135 struct timeval tv; 2136 2137 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2138 log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts; 2139 log.u_bbr.flex8 = 4; 2140 log.u_bbr.pkts_out = tp->t_maxseg; 2141 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2142 log.u_bbr.delivered = 0; 2143 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, 2144 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2145 } else { 2146 /* 2147 * We can not log the packet, since we only own the 2148 * read lock, but a write lock is needed. The read lock 2149 * is not upgraded to a write lock, since only getting 2150 * the read lock was done intentionally to improve the 2151 * handling of SYN flooding attacks. 2152 * This happens only for pure SYN segments received in 2153 * the initial CLOSED state, or received in a more 2154 * advanced state than listen and the UDP encapsulation 2155 * port is unexpected. 2156 * The incoming SYN segments do not really belong to 2157 * the TCP connection and the handling does not change 2158 * the state of the TCP connection. Therefore, the 2159 * sending of the RST segments is not logged. Please 2160 * note that also the incoming SYN segments are not 2161 * logged. 2162 * 2163 * The following code ensures that the above description 2164 * is and stays correct. 2165 */ 2166 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2167 (tp->t_state == TCPS_CLOSED || 2168 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2169 ("%s: Logging of TCP segment with flags 0x%b and " 2170 "UDP encapsulation port %u skipped in state %s", 2171 __func__, thflags, PRINT_TH_FLAGS, 2172 ntohs(port), tcpstates[tp->t_state])); 2173 } 2174 } 2175 2176 if (flags & TH_ACK) 2177 TCPSTAT_INC(tcps_sndacks); 2178 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2179 TCPSTAT_INC(tcps_sndctrl); 2180 TCPSTAT_INC(tcps_sndtotal); 2181 2182 #ifdef INET6 2183 if (isipv6) { 2184 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2185 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2186 } 2187 #endif /* INET6 */ 2188 #if defined(INET) && defined(INET6) 2189 else 2190 #endif 2191 #ifdef INET 2192 { 2193 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2194 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2195 } 2196 #endif 2197 if (lgb != NULL) 2198 lgb->tlb_errno = output_ret; 2199 } 2200 2201 /* 2202 * Create a new TCP control block, making an 2203 * empty reassembly queue and hooking it to the argument 2204 * protocol control block. The `inp' parameter must have 2205 * come from the zone allocator set up in tcp_init(). 2206 */ 2207 struct tcpcb * 2208 tcp_newtcpcb(struct inpcb *inp) 2209 { 2210 struct tcpcb_mem *tm; 2211 struct tcpcb *tp; 2212 #ifdef INET6 2213 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2214 #endif /* INET6 */ 2215 2216 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 2217 if (tm == NULL) 2218 return (NULL); 2219 tp = &tm->tcb; 2220 2221 /* Initialise cc_var struct for this tcpcb. */ 2222 tp->ccv = &tm->ccv; 2223 tp->ccv->type = IPPROTO_TCP; 2224 tp->ccv->ccvc.tcp = tp; 2225 rw_rlock(&tcp_function_lock); 2226 tp->t_fb = tcp_func_set_ptr; 2227 refcount_acquire(&tp->t_fb->tfb_refcnt); 2228 rw_runlock(&tcp_function_lock); 2229 /* 2230 * Use the current system default CC algorithm. 2231 */ 2232 cc_attach(tp, CC_DEFAULT_ALGO()); 2233 2234 /* 2235 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 2236 * is called. 2237 */ 2238 in_pcbref(inp); /* Reference for tcpcb */ 2239 tp->t_inpcb = inp; 2240 2241 if (CC_ALGO(tp)->cb_init != NULL) 2242 if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) { 2243 cc_detach(tp); 2244 if (tp->t_fb->tfb_tcp_fb_fini) 2245 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2246 in_pcbrele_wlocked(inp); 2247 refcount_release(&tp->t_fb->tfb_refcnt); 2248 uma_zfree(V_tcpcb_zone, tm); 2249 return (NULL); 2250 } 2251 2252 #ifdef TCP_HHOOK 2253 tp->osd = &tm->osd; 2254 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 2255 if (tp->t_fb->tfb_tcp_fb_fini) 2256 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2257 in_pcbrele_wlocked(inp); 2258 refcount_release(&tp->t_fb->tfb_refcnt); 2259 uma_zfree(V_tcpcb_zone, tm); 2260 return (NULL); 2261 } 2262 #endif 2263 2264 #ifdef VIMAGE 2265 tp->t_vnet = inp->inp_vnet; 2266 #endif 2267 tp->t_timers = &tm->tt; 2268 TAILQ_INIT(&tp->t_segq); 2269 tp->t_maxseg = 2270 #ifdef INET6 2271 isipv6 ? V_tcp_v6mssdflt : 2272 #endif /* INET6 */ 2273 V_tcp_mssdflt; 2274 2275 /* Set up our timeouts. */ 2276 callout_init(&tp->t_timers->tt_rexmt, 1); 2277 callout_init(&tp->t_timers->tt_persist, 1); 2278 callout_init(&tp->t_timers->tt_keep, 1); 2279 callout_init(&tp->t_timers->tt_2msl, 1); 2280 callout_init(&tp->t_timers->tt_delack, 1); 2281 2282 switch (V_tcp_do_rfc1323) { 2283 case 0: 2284 break; 2285 default: 2286 case 1: 2287 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2288 break; 2289 case 2: 2290 tp->t_flags = TF_REQ_SCALE; 2291 break; 2292 case 3: 2293 tp->t_flags = TF_REQ_TSTMP; 2294 break; 2295 } 2296 if (V_tcp_do_sack) 2297 tp->t_flags |= TF_SACK_PERMIT; 2298 TAILQ_INIT(&tp->snd_holes); 2299 2300 /* 2301 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2302 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2303 * reasonable initial retransmit time. 2304 */ 2305 tp->t_srtt = TCPTV_SRTTBASE; 2306 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2307 tp->t_rttmin = tcp_rexmit_min; 2308 tp->t_rxtcur = tcp_rexmit_initial; 2309 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2310 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2311 tp->t_rcvtime = ticks; 2312 /* 2313 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2314 * because the socket may be bound to an IPv6 wildcard address, 2315 * which may match an IPv4-mapped IPv6 address. 2316 */ 2317 inp->inp_ip_ttl = V_ip_defttl; 2318 inp->inp_ppcb = tp; 2319 #ifdef TCPPCAP 2320 /* 2321 * Init the TCP PCAP queues. 2322 */ 2323 tcp_pcap_tcpcb_init(tp); 2324 #endif 2325 #ifdef TCP_BLACKBOX 2326 /* Initialize the per-TCPCB log data. */ 2327 tcp_log_tcpcbinit(tp); 2328 #endif 2329 tp->t_pacing_rate = -1; 2330 if (tp->t_fb->tfb_tcp_fb_init) { 2331 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2332 refcount_release(&tp->t_fb->tfb_refcnt); 2333 in_pcbrele_wlocked(inp); 2334 uma_zfree(V_tcpcb_zone, tm); 2335 return (NULL); 2336 } 2337 } 2338 #ifdef STATS 2339 if (V_tcp_perconn_stats_enable == 1) 2340 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2341 #endif 2342 if (V_tcp_do_lrd) 2343 tp->t_flags |= TF_LRD; 2344 return (tp); /* XXX */ 2345 } 2346 2347 /* 2348 * Drop a TCP connection, reporting 2349 * the specified error. If connection is synchronized, 2350 * then send a RST to peer. 2351 */ 2352 struct tcpcb * 2353 tcp_drop(struct tcpcb *tp, int errno) 2354 { 2355 struct socket *so = tp->t_inpcb->inp_socket; 2356 2357 NET_EPOCH_ASSERT(); 2358 INP_WLOCK_ASSERT(tp->t_inpcb); 2359 2360 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2361 tcp_state_change(tp, TCPS_CLOSED); 2362 /* Don't use tcp_output() here due to possible recursion. */ 2363 (void)tcp_output_nodrop(tp); 2364 TCPSTAT_INC(tcps_drops); 2365 } else 2366 TCPSTAT_INC(tcps_conndrops); 2367 if (errno == ETIMEDOUT && tp->t_softerror) 2368 errno = tp->t_softerror; 2369 so->so_error = errno; 2370 return (tcp_close(tp)); 2371 } 2372 2373 void 2374 tcp_discardcb(struct tcpcb *tp) 2375 { 2376 struct inpcb *inp = tp->t_inpcb; 2377 2378 INP_WLOCK_ASSERT(inp); 2379 2380 /* 2381 * Make sure that all of our timers are stopped before we delete the 2382 * PCB. 2383 * 2384 * If stopping a timer fails, we schedule a discard function in same 2385 * callout, and the last discard function called will take care of 2386 * deleting the tcpcb. 2387 */ 2388 tp->t_timers->tt_draincnt = 0; 2389 tcp_timer_stop(tp, TT_REXMT); 2390 tcp_timer_stop(tp, TT_PERSIST); 2391 tcp_timer_stop(tp, TT_KEEP); 2392 tcp_timer_stop(tp, TT_2MSL); 2393 tcp_timer_stop(tp, TT_DELACK); 2394 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2395 /* 2396 * Call the stop-all function of the methods, 2397 * this function should call the tcp_timer_stop() 2398 * method with each of the function specific timeouts. 2399 * That stop will be called via the tfb_tcp_timer_stop() 2400 * which should use the async drain function of the 2401 * callout system (see tcp_var.h). 2402 */ 2403 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2404 } 2405 2406 /* free the reassembly queue, if any */ 2407 tcp_reass_flush(tp); 2408 2409 #ifdef TCP_OFFLOAD 2410 /* Disconnect offload device, if any. */ 2411 if (tp->t_flags & TF_TOE) 2412 tcp_offload_detach(tp); 2413 #endif 2414 2415 tcp_free_sackholes(tp); 2416 2417 #ifdef TCPPCAP 2418 /* Free the TCP PCAP queues. */ 2419 tcp_pcap_drain(&(tp->t_inpkts)); 2420 tcp_pcap_drain(&(tp->t_outpkts)); 2421 #endif 2422 2423 /* Allow the CC algorithm to clean up after itself. */ 2424 if (CC_ALGO(tp)->cb_destroy != NULL) 2425 CC_ALGO(tp)->cb_destroy(tp->ccv); 2426 CC_DATA(tp) = NULL; 2427 /* Detach from the CC algorithm */ 2428 cc_detach(tp); 2429 2430 #ifdef TCP_HHOOK 2431 khelp_destroy_osd(tp->osd); 2432 #endif 2433 #ifdef STATS 2434 stats_blob_destroy(tp->t_stats); 2435 #endif 2436 2437 CC_ALGO(tp) = NULL; 2438 inp->inp_ppcb = NULL; 2439 if (tp->t_timers->tt_draincnt == 0) { 2440 bool released __diagused; 2441 2442 released = tcp_freecb(tp); 2443 KASSERT(!released, ("%s: inp %p should not have been released " 2444 "here", __func__, inp)); 2445 } 2446 } 2447 2448 bool 2449 tcp_freecb(struct tcpcb *tp) 2450 { 2451 struct inpcb *inp = tp->t_inpcb; 2452 struct socket *so = inp->inp_socket; 2453 #ifdef INET6 2454 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2455 #endif 2456 2457 INP_WLOCK_ASSERT(inp); 2458 MPASS(tp->t_timers->tt_draincnt == 0); 2459 2460 /* We own the last reference on tcpcb, let's free it. */ 2461 #ifdef TCP_BLACKBOX 2462 tcp_log_tcpcbfini(tp); 2463 #endif 2464 TCPSTATES_DEC(tp->t_state); 2465 if (tp->t_fb->tfb_tcp_fb_fini) 2466 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2467 2468 /* 2469 * If we got enough samples through the srtt filter, 2470 * save the rtt and rttvar in the routing entry. 2471 * 'Enough' is arbitrarily defined as 4 rtt samples. 2472 * 4 samples is enough for the srtt filter to converge 2473 * to within enough % of the correct value; fewer samples 2474 * and we could save a bogus rtt. The danger is not high 2475 * as tcp quickly recovers from everything. 2476 * XXX: Works very well but needs some more statistics! 2477 * 2478 * XXXRRS: Updating must be after the stack fini() since 2479 * that may be converting some internal representation of 2480 * say srtt etc into the general one used by other stacks. 2481 * Lets also at least protect against the so being NULL 2482 * as RW stated below. 2483 */ 2484 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2485 struct hc_metrics_lite metrics; 2486 uint32_t ssthresh; 2487 2488 bzero(&metrics, sizeof(metrics)); 2489 /* 2490 * Update the ssthresh always when the conditions below 2491 * are satisfied. This gives us better new start value 2492 * for the congestion avoidance for new connections. 2493 * ssthresh is only set if packet loss occurred on a session. 2494 * 2495 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2496 * being torn down. Ideally this code would not use 'so'. 2497 */ 2498 ssthresh = tp->snd_ssthresh; 2499 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2500 /* 2501 * convert the limit from user data bytes to 2502 * packets then to packet data bytes. 2503 */ 2504 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2505 if (ssthresh < 2) 2506 ssthresh = 2; 2507 ssthresh *= (tp->t_maxseg + 2508 #ifdef INET6 2509 (isipv6 ? sizeof (struct ip6_hdr) + 2510 sizeof (struct tcphdr) : 2511 #endif 2512 sizeof (struct tcpiphdr) 2513 #ifdef INET6 2514 ) 2515 #endif 2516 ); 2517 } else 2518 ssthresh = 0; 2519 metrics.rmx_ssthresh = ssthresh; 2520 2521 metrics.rmx_rtt = tp->t_srtt; 2522 metrics.rmx_rttvar = tp->t_rttvar; 2523 metrics.rmx_cwnd = tp->snd_cwnd; 2524 metrics.rmx_sendpipe = 0; 2525 metrics.rmx_recvpipe = 0; 2526 2527 tcp_hc_update(&inp->inp_inc, &metrics); 2528 } 2529 2530 refcount_release(&tp->t_fb->tfb_refcnt); 2531 uma_zfree(V_tcpcb_zone, tp); 2532 2533 return (in_pcbrele_wlocked(inp)); 2534 } 2535 2536 /* 2537 * Attempt to close a TCP control block, marking it as dropped, and freeing 2538 * the socket if we hold the only reference. 2539 */ 2540 struct tcpcb * 2541 tcp_close(struct tcpcb *tp) 2542 { 2543 struct inpcb *inp = tp->t_inpcb; 2544 struct socket *so; 2545 2546 INP_WLOCK_ASSERT(inp); 2547 2548 #ifdef TCP_OFFLOAD 2549 if (tp->t_state == TCPS_LISTEN) 2550 tcp_offload_listen_stop(tp); 2551 #endif 2552 /* 2553 * This releases the TFO pending counter resource for TFO listen 2554 * sockets as well as passively-created TFO sockets that transition 2555 * from SYN_RECEIVED to CLOSED. 2556 */ 2557 if (tp->t_tfo_pending) { 2558 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2559 tp->t_tfo_pending = NULL; 2560 } 2561 #ifdef TCPHPTS 2562 tcp_hpts_remove(inp); 2563 #endif 2564 in_pcbdrop(inp); 2565 TCPSTAT_INC(tcps_closed); 2566 if (tp->t_state != TCPS_CLOSED) 2567 tcp_state_change(tp, TCPS_CLOSED); 2568 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2569 so = inp->inp_socket; 2570 soisdisconnected(so); 2571 if (inp->inp_flags & INP_SOCKREF) { 2572 inp->inp_flags &= ~INP_SOCKREF; 2573 INP_WUNLOCK(inp); 2574 sorele(so); 2575 return (NULL); 2576 } 2577 return (tp); 2578 } 2579 2580 /* 2581 * Notify a tcp user of an asynchronous error; 2582 * store error as soft error, but wake up user 2583 * (for now, won't do anything until can select for soft error). 2584 * 2585 * Do not wake up user since there currently is no mechanism for 2586 * reporting soft errors (yet - a kqueue filter may be added). 2587 */ 2588 static struct inpcb * 2589 tcp_notify(struct inpcb *inp, int error) 2590 { 2591 struct tcpcb *tp; 2592 2593 INP_WLOCK_ASSERT(inp); 2594 2595 if (inp->inp_flags & INP_DROPPED) 2596 return (inp); 2597 2598 tp = intotcpcb(inp); 2599 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2600 2601 /* 2602 * Ignore some errors if we are hooked up. 2603 * If connection hasn't completed, has retransmitted several times, 2604 * and receives a second error, give up now. This is better 2605 * than waiting a long time to establish a connection that 2606 * can never complete. 2607 */ 2608 if (tp->t_state == TCPS_ESTABLISHED && 2609 (error == EHOSTUNREACH || error == ENETUNREACH || 2610 error == EHOSTDOWN)) { 2611 if (inp->inp_route.ro_nh) { 2612 NH_FREE(inp->inp_route.ro_nh); 2613 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2614 } 2615 return (inp); 2616 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2617 tp->t_softerror) { 2618 tp = tcp_drop(tp, error); 2619 if (tp != NULL) 2620 return (inp); 2621 else 2622 return (NULL); 2623 } else { 2624 tp->t_softerror = error; 2625 return (inp); 2626 } 2627 #if 0 2628 wakeup( &so->so_timeo); 2629 sorwakeup(so); 2630 sowwakeup(so); 2631 #endif 2632 } 2633 2634 static int 2635 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2636 { 2637 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2638 INPLOOKUP_RLOCKPCB); 2639 struct xinpgen xig; 2640 struct inpcb *inp; 2641 int error; 2642 2643 if (req->newptr != NULL) 2644 return (EPERM); 2645 2646 if (req->oldptr == NULL) { 2647 int n; 2648 2649 n = V_tcbinfo.ipi_count + 2650 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2651 n += imax(n / 8, 10); 2652 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2653 return (0); 2654 } 2655 2656 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2657 return (error); 2658 2659 bzero(&xig, sizeof(xig)); 2660 xig.xig_len = sizeof xig; 2661 xig.xig_count = V_tcbinfo.ipi_count + 2662 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2663 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2664 xig.xig_sogen = so_gencnt; 2665 error = SYSCTL_OUT(req, &xig, sizeof xig); 2666 if (error) 2667 return (error); 2668 2669 error = syncache_pcblist(req); 2670 if (error) 2671 return (error); 2672 2673 while ((inp = inp_next(&inpi)) != NULL) { 2674 if (inp->inp_gencnt <= xig.xig_gen && 2675 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2676 struct xtcpcb xt; 2677 2678 tcp_inptoxtp(inp, &xt); 2679 error = SYSCTL_OUT(req, &xt, sizeof xt); 2680 if (error) { 2681 INP_RUNLOCK(inp); 2682 break; 2683 } else 2684 continue; 2685 } 2686 } 2687 2688 if (!error) { 2689 /* 2690 * Give the user an updated idea of our state. 2691 * If the generation differs from what we told 2692 * her before, she knows that something happened 2693 * while we were processing this request, and it 2694 * might be necessary to retry. 2695 */ 2696 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2697 xig.xig_sogen = so_gencnt; 2698 xig.xig_count = V_tcbinfo.ipi_count + 2699 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2700 error = SYSCTL_OUT(req, &xig, sizeof xig); 2701 } 2702 2703 return (error); 2704 } 2705 2706 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2707 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2708 NULL, 0, tcp_pcblist, "S,xtcpcb", 2709 "List of active TCP connections"); 2710 2711 #ifdef INET 2712 static int 2713 tcp_getcred(SYSCTL_HANDLER_ARGS) 2714 { 2715 struct xucred xuc; 2716 struct sockaddr_in addrs[2]; 2717 struct epoch_tracker et; 2718 struct inpcb *inp; 2719 int error; 2720 2721 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2722 if (error) 2723 return (error); 2724 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2725 if (error) 2726 return (error); 2727 NET_EPOCH_ENTER(et); 2728 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2729 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2730 NET_EPOCH_EXIT(et); 2731 if (inp != NULL) { 2732 if (inp->inp_socket == NULL) 2733 error = ENOENT; 2734 if (error == 0) 2735 error = cr_canseeinpcb(req->td->td_ucred, inp); 2736 if (error == 0) 2737 cru2x(inp->inp_cred, &xuc); 2738 INP_RUNLOCK(inp); 2739 } else 2740 error = ENOENT; 2741 if (error == 0) 2742 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2743 return (error); 2744 } 2745 2746 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2747 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2748 0, 0, tcp_getcred, "S,xucred", 2749 "Get the xucred of a TCP connection"); 2750 #endif /* INET */ 2751 2752 #ifdef INET6 2753 static int 2754 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2755 { 2756 struct epoch_tracker et; 2757 struct xucred xuc; 2758 struct sockaddr_in6 addrs[2]; 2759 struct inpcb *inp; 2760 int error; 2761 #ifdef INET 2762 int mapped = 0; 2763 #endif 2764 2765 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2766 if (error) 2767 return (error); 2768 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2769 if (error) 2770 return (error); 2771 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2772 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2773 return (error); 2774 } 2775 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2776 #ifdef INET 2777 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2778 mapped = 1; 2779 else 2780 #endif 2781 return (EINVAL); 2782 } 2783 2784 NET_EPOCH_ENTER(et); 2785 #ifdef INET 2786 if (mapped == 1) 2787 inp = in_pcblookup(&V_tcbinfo, 2788 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2789 addrs[1].sin6_port, 2790 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2791 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2792 else 2793 #endif 2794 inp = in6_pcblookup(&V_tcbinfo, 2795 &addrs[1].sin6_addr, addrs[1].sin6_port, 2796 &addrs[0].sin6_addr, addrs[0].sin6_port, 2797 INPLOOKUP_RLOCKPCB, NULL); 2798 NET_EPOCH_EXIT(et); 2799 if (inp != NULL) { 2800 if (inp->inp_socket == NULL) 2801 error = ENOENT; 2802 if (error == 0) 2803 error = cr_canseeinpcb(req->td->td_ucred, inp); 2804 if (error == 0) 2805 cru2x(inp->inp_cred, &xuc); 2806 INP_RUNLOCK(inp); 2807 } else 2808 error = ENOENT; 2809 if (error == 0) 2810 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2811 return (error); 2812 } 2813 2814 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2815 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2816 0, 0, tcp6_getcred, "S,xucred", 2817 "Get the xucred of a TCP6 connection"); 2818 #endif /* INET6 */ 2819 2820 #ifdef INET 2821 /* Path MTU to try next when a fragmentation-needed message is received. */ 2822 static inline int 2823 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2824 { 2825 int mtu = ntohs(icp->icmp_nextmtu); 2826 2827 /* If no alternative MTU was proposed, try the next smaller one. */ 2828 if (!mtu) 2829 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2830 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2831 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2832 2833 return (mtu); 2834 } 2835 2836 static void 2837 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2838 { 2839 struct ip *ip; 2840 struct tcphdr *th; 2841 struct inpcb *inp; 2842 struct tcpcb *tp; 2843 struct inpcb *(*notify)(struct inpcb *, int); 2844 struct in_conninfo inc; 2845 tcp_seq icmp_tcp_seq; 2846 int errno, mtu; 2847 2848 errno = icmp_errmap(icp); 2849 switch (errno) { 2850 case 0: 2851 return; 2852 case EMSGSIZE: 2853 notify = tcp_mtudisc_notify; 2854 break; 2855 case ECONNREFUSED: 2856 if (V_icmp_may_rst) 2857 notify = tcp_drop_syn_sent; 2858 else 2859 notify = tcp_notify; 2860 break; 2861 case EHOSTUNREACH: 2862 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2863 notify = tcp_drop_syn_sent; 2864 else 2865 notify = tcp_notify; 2866 break; 2867 default: 2868 notify = tcp_notify; 2869 } 2870 2871 ip = &icp->icmp_ip; 2872 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2873 icmp_tcp_seq = th->th_seq; 2874 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2875 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2876 if (inp != NULL) { 2877 if (!(inp->inp_flags & INP_DROPPED) && 2878 !(inp->inp_socket == NULL)) { 2879 tp = intotcpcb(inp); 2880 #ifdef TCP_OFFLOAD 2881 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2882 /* 2883 * MTU discovery for offloaded connections. Let 2884 * the TOE driver verify seq# and process it. 2885 */ 2886 mtu = tcp_next_pmtu(icp, ip); 2887 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2888 goto out; 2889 } 2890 #endif 2891 if (tp->t_port != port) { 2892 goto out; 2893 } 2894 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2895 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2896 if (errno == EMSGSIZE) { 2897 /* 2898 * MTU discovery: we got a needfrag and 2899 * will potentially try a lower MTU. 2900 */ 2901 mtu = tcp_next_pmtu(icp, ip); 2902 2903 /* 2904 * Only process the offered MTU if it 2905 * is smaller than the current one. 2906 */ 2907 if (mtu < tp->t_maxseg + 2908 sizeof(struct tcpiphdr)) { 2909 bzero(&inc, sizeof(inc)); 2910 inc.inc_faddr = ip->ip_dst; 2911 inc.inc_fibnum = 2912 inp->inp_inc.inc_fibnum; 2913 tcp_hc_updatemtu(&inc, mtu); 2914 inp = tcp_mtudisc(inp, mtu); 2915 } 2916 } else 2917 inp = (*notify)(inp, errno); 2918 } 2919 } 2920 } else { 2921 bzero(&inc, sizeof(inc)); 2922 inc.inc_fport = th->th_dport; 2923 inc.inc_lport = th->th_sport; 2924 inc.inc_faddr = ip->ip_dst; 2925 inc.inc_laddr = ip->ip_src; 2926 syncache_unreach(&inc, icmp_tcp_seq, port); 2927 } 2928 out: 2929 if (inp != NULL) 2930 INP_WUNLOCK(inp); 2931 } 2932 2933 static void 2934 tcp_ctlinput(struct icmp *icmp) 2935 { 2936 tcp_ctlinput_with_port(icmp, htons(0)); 2937 } 2938 2939 static void 2940 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2941 { 2942 /* Its a tunneled TCP over UDP icmp */ 2943 struct icmp *icmp = param.icmp; 2944 struct ip *outer_ip, *inner_ip; 2945 struct udphdr *udp; 2946 struct tcphdr *th, ttemp; 2947 int i_hlen, o_len; 2948 uint16_t port; 2949 2950 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2951 inner_ip = &icmp->icmp_ip; 2952 i_hlen = inner_ip->ip_hl << 2; 2953 o_len = ntohs(outer_ip->ip_len); 2954 if (o_len < 2955 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2956 /* Not enough data present */ 2957 return; 2958 } 2959 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2960 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2961 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2962 return; 2963 } 2964 port = udp->uh_dport; 2965 th = (struct tcphdr *)(udp + 1); 2966 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2967 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2968 /* Now adjust down the size of the outer IP header */ 2969 o_len -= sizeof(struct udphdr); 2970 outer_ip->ip_len = htons(o_len); 2971 /* Now call in to the normal handling code */ 2972 tcp_ctlinput_with_port(icmp, port); 2973 } 2974 #endif /* INET */ 2975 2976 #ifdef INET6 2977 static inline int 2978 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2979 { 2980 int mtu = ntohl(icmp6->icmp6_mtu); 2981 2982 /* 2983 * If no alternative MTU was proposed, or the proposed MTU was too 2984 * small, set to the min. 2985 */ 2986 if (mtu < IPV6_MMTU) 2987 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2988 return (mtu); 2989 } 2990 2991 static void 2992 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2993 { 2994 struct in6_addr *dst; 2995 struct inpcb *(*notify)(struct inpcb *, int); 2996 struct ip6_hdr *ip6; 2997 struct mbuf *m; 2998 struct inpcb *inp; 2999 struct tcpcb *tp; 3000 struct icmp6_hdr *icmp6; 3001 struct in_conninfo inc; 3002 struct tcp_ports { 3003 uint16_t th_sport; 3004 uint16_t th_dport; 3005 } t_ports; 3006 tcp_seq icmp_tcp_seq; 3007 unsigned int mtu; 3008 unsigned int off; 3009 int errno; 3010 3011 icmp6 = ip6cp->ip6c_icmp6; 3012 m = ip6cp->ip6c_m; 3013 ip6 = ip6cp->ip6c_ip6; 3014 off = ip6cp->ip6c_off; 3015 dst = &ip6cp->ip6c_finaldst->sin6_addr; 3016 3017 errno = icmp6_errmap(icmp6); 3018 switch (errno) { 3019 case 0: 3020 return; 3021 case EMSGSIZE: 3022 notify = tcp_mtudisc_notify; 3023 break; 3024 case ECONNREFUSED: 3025 if (V_icmp_may_rst) 3026 notify = tcp_drop_syn_sent; 3027 else 3028 notify = tcp_notify; 3029 break; 3030 case EHOSTUNREACH: 3031 /* 3032 * There are only four ICMPs that may reset connection: 3033 * - administratively prohibited 3034 * - port unreachable 3035 * - time exceeded in transit 3036 * - unknown next header 3037 */ 3038 if (V_icmp_may_rst && 3039 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 3040 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 3041 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 3042 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 3043 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 3044 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 3045 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 3046 notify = tcp_drop_syn_sent; 3047 else 3048 notify = tcp_notify; 3049 break; 3050 default: 3051 notify = tcp_notify; 3052 } 3053 3054 /* Check if we can safely get the ports from the tcp hdr */ 3055 if (m == NULL || 3056 (m->m_pkthdr.len < 3057 (int32_t) (off + sizeof(struct tcp_ports)))) { 3058 return; 3059 } 3060 bzero(&t_ports, sizeof(struct tcp_ports)); 3061 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3062 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3063 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3064 off += sizeof(struct tcp_ports); 3065 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3066 goto out; 3067 } 3068 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3069 if (inp != NULL) { 3070 if (!(inp->inp_flags & INP_DROPPED) && 3071 !(inp->inp_socket == NULL)) { 3072 tp = intotcpcb(inp); 3073 #ifdef TCP_OFFLOAD 3074 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3075 /* MTU discovery for offloaded connections. */ 3076 mtu = tcp6_next_pmtu(icmp6); 3077 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3078 goto out; 3079 } 3080 #endif 3081 if (tp->t_port != port) { 3082 goto out; 3083 } 3084 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3085 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3086 if (errno == EMSGSIZE) { 3087 /* 3088 * MTU discovery: 3089 * If we got a needfrag set the MTU 3090 * in the route to the suggested new 3091 * value (if given) and then notify. 3092 */ 3093 mtu = tcp6_next_pmtu(icmp6); 3094 3095 bzero(&inc, sizeof(inc)); 3096 inc.inc_fibnum = M_GETFIB(m); 3097 inc.inc_flags |= INC_ISIPV6; 3098 inc.inc6_faddr = *dst; 3099 if (in6_setscope(&inc.inc6_faddr, 3100 m->m_pkthdr.rcvif, NULL)) 3101 goto out; 3102 /* 3103 * Only process the offered MTU if it 3104 * is smaller than the current one. 3105 */ 3106 if (mtu < tp->t_maxseg + 3107 sizeof (struct tcphdr) + 3108 sizeof (struct ip6_hdr)) { 3109 tcp_hc_updatemtu(&inc, mtu); 3110 tcp_mtudisc(inp, mtu); 3111 ICMP6STAT_INC(icp6s_pmtuchg); 3112 } 3113 } else 3114 inp = (*notify)(inp, errno); 3115 } 3116 } 3117 } else { 3118 bzero(&inc, sizeof(inc)); 3119 inc.inc_fibnum = M_GETFIB(m); 3120 inc.inc_flags |= INC_ISIPV6; 3121 inc.inc_fport = t_ports.th_dport; 3122 inc.inc_lport = t_ports.th_sport; 3123 inc.inc6_faddr = *dst; 3124 inc.inc6_laddr = ip6->ip6_src; 3125 syncache_unreach(&inc, icmp_tcp_seq, port); 3126 } 3127 out: 3128 if (inp != NULL) 3129 INP_WUNLOCK(inp); 3130 } 3131 3132 static void 3133 tcp6_ctlinput(struct ip6ctlparam *ctl) 3134 { 3135 tcp6_ctlinput_with_port(ctl, htons(0)); 3136 } 3137 3138 static void 3139 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3140 { 3141 struct ip6ctlparam *ip6cp = param.ip6cp; 3142 struct mbuf *m; 3143 struct udphdr *udp; 3144 uint16_t port; 3145 3146 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3147 if (m == NULL) { 3148 return; 3149 } 3150 udp = mtod(m, struct udphdr *); 3151 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3152 return; 3153 } 3154 port = udp->uh_dport; 3155 m_adj(m, sizeof(struct udphdr)); 3156 if ((m->m_flags & M_PKTHDR) == 0) { 3157 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3158 } 3159 /* Now call in to the normal handling code */ 3160 tcp6_ctlinput_with_port(ip6cp, port); 3161 } 3162 3163 #endif /* INET6 */ 3164 3165 static uint32_t 3166 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3167 { 3168 SIPHASH_CTX ctx; 3169 uint32_t hash[2]; 3170 3171 KASSERT(len >= SIPHASH_KEY_LENGTH, 3172 ("%s: keylen %u too short ", __func__, len)); 3173 SipHash24_Init(&ctx); 3174 SipHash_SetKey(&ctx, (uint8_t *)key); 3175 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3176 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3177 switch (inc->inc_flags & INC_ISIPV6) { 3178 #ifdef INET 3179 case 0: 3180 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3181 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3182 break; 3183 #endif 3184 #ifdef INET6 3185 case INC_ISIPV6: 3186 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3187 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3188 break; 3189 #endif 3190 } 3191 SipHash_Final((uint8_t *)hash, &ctx); 3192 3193 return (hash[0] ^ hash[1]); 3194 } 3195 3196 uint32_t 3197 tcp_new_ts_offset(struct in_conninfo *inc) 3198 { 3199 struct in_conninfo inc_store, *local_inc; 3200 3201 if (!V_tcp_ts_offset_per_conn) { 3202 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3203 inc_store.inc_lport = 0; 3204 inc_store.inc_fport = 0; 3205 local_inc = &inc_store; 3206 } else { 3207 local_inc = inc; 3208 } 3209 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3210 sizeof(V_ts_offset_secret))); 3211 } 3212 3213 /* 3214 * Following is where TCP initial sequence number generation occurs. 3215 * 3216 * There are two places where we must use initial sequence numbers: 3217 * 1. In SYN-ACK packets. 3218 * 2. In SYN packets. 3219 * 3220 * All ISNs for SYN-ACK packets are generated by the syncache. See 3221 * tcp_syncache.c for details. 3222 * 3223 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3224 * depends on this property. In addition, these ISNs should be 3225 * unguessable so as to prevent connection hijacking. To satisfy 3226 * the requirements of this situation, the algorithm outlined in 3227 * RFC 1948 is used, with only small modifications. 3228 * 3229 * Implementation details: 3230 * 3231 * Time is based off the system timer, and is corrected so that it 3232 * increases by one megabyte per second. This allows for proper 3233 * recycling on high speed LANs while still leaving over an hour 3234 * before rollover. 3235 * 3236 * As reading the *exact* system time is too expensive to be done 3237 * whenever setting up a TCP connection, we increment the time 3238 * offset in two ways. First, a small random positive increment 3239 * is added to isn_offset for each connection that is set up. 3240 * Second, the function tcp_isn_tick fires once per clock tick 3241 * and increments isn_offset as necessary so that sequence numbers 3242 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3243 * random positive increments serve only to ensure that the same 3244 * exact sequence number is never sent out twice (as could otherwise 3245 * happen when a port is recycled in less than the system tick 3246 * interval.) 3247 * 3248 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3249 * between seeding of isn_secret. This is normally set to zero, 3250 * as reseeding should not be necessary. 3251 * 3252 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3253 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3254 * general, this means holding an exclusive (write) lock. 3255 */ 3256 3257 #define ISN_BYTES_PER_SECOND 1048576 3258 #define ISN_STATIC_INCREMENT 4096 3259 #define ISN_RANDOM_INCREMENT (4096 - 1) 3260 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3261 3262 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3263 VNET_DEFINE_STATIC(int, isn_last); 3264 VNET_DEFINE_STATIC(int, isn_last_reseed); 3265 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3266 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3267 3268 #define V_isn_secret VNET(isn_secret) 3269 #define V_isn_last VNET(isn_last) 3270 #define V_isn_last_reseed VNET(isn_last_reseed) 3271 #define V_isn_offset VNET(isn_offset) 3272 #define V_isn_offset_old VNET(isn_offset_old) 3273 3274 tcp_seq 3275 tcp_new_isn(struct in_conninfo *inc) 3276 { 3277 tcp_seq new_isn; 3278 u_int32_t projected_offset; 3279 3280 ISN_LOCK(); 3281 /* Seed if this is the first use, reseed if requested. */ 3282 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3283 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3284 < (u_int)ticks))) { 3285 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3286 V_isn_last_reseed = ticks; 3287 } 3288 3289 /* Compute the hash and return the ISN. */ 3290 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3291 sizeof(V_isn_secret)); 3292 V_isn_offset += ISN_STATIC_INCREMENT + 3293 (arc4random() & ISN_RANDOM_INCREMENT); 3294 if (ticks != V_isn_last) { 3295 projected_offset = V_isn_offset_old + 3296 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3297 if (SEQ_GT(projected_offset, V_isn_offset)) 3298 V_isn_offset = projected_offset; 3299 V_isn_offset_old = V_isn_offset; 3300 V_isn_last = ticks; 3301 } 3302 new_isn += V_isn_offset; 3303 ISN_UNLOCK(); 3304 return (new_isn); 3305 } 3306 3307 /* 3308 * When a specific ICMP unreachable message is received and the 3309 * connection state is SYN-SENT, drop the connection. This behavior 3310 * is controlled by the icmp_may_rst sysctl. 3311 */ 3312 static struct inpcb * 3313 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3314 { 3315 struct tcpcb *tp; 3316 3317 NET_EPOCH_ASSERT(); 3318 INP_WLOCK_ASSERT(inp); 3319 3320 if (inp->inp_flags & INP_DROPPED) 3321 return (inp); 3322 3323 tp = intotcpcb(inp); 3324 if (tp->t_state != TCPS_SYN_SENT) 3325 return (inp); 3326 3327 if (IS_FASTOPEN(tp->t_flags)) 3328 tcp_fastopen_disable_path(tp); 3329 3330 tp = tcp_drop(tp, errno); 3331 if (tp != NULL) 3332 return (inp); 3333 else 3334 return (NULL); 3335 } 3336 3337 /* 3338 * When `need fragmentation' ICMP is received, update our idea of the MSS 3339 * based on the new value. Also nudge TCP to send something, since we 3340 * know the packet we just sent was dropped. 3341 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3342 */ 3343 static struct inpcb * 3344 tcp_mtudisc_notify(struct inpcb *inp, int error) 3345 { 3346 3347 return (tcp_mtudisc(inp, -1)); 3348 } 3349 3350 static struct inpcb * 3351 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3352 { 3353 struct tcpcb *tp; 3354 struct socket *so; 3355 3356 INP_WLOCK_ASSERT(inp); 3357 if (inp->inp_flags & INP_DROPPED) 3358 return (inp); 3359 3360 tp = intotcpcb(inp); 3361 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3362 3363 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3364 3365 so = inp->inp_socket; 3366 SOCKBUF_LOCK(&so->so_snd); 3367 /* If the mss is larger than the socket buffer, decrease the mss. */ 3368 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3369 tp->t_maxseg = so->so_snd.sb_hiwat; 3370 SOCKBUF_UNLOCK(&so->so_snd); 3371 3372 TCPSTAT_INC(tcps_mturesent); 3373 tp->t_rtttime = 0; 3374 tp->snd_nxt = tp->snd_una; 3375 tcp_free_sackholes(tp); 3376 tp->snd_recover = tp->snd_max; 3377 if (tp->t_flags & TF_SACK_PERMIT) 3378 EXIT_FASTRECOVERY(tp->t_flags); 3379 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3380 /* 3381 * Conceptually the snd_nxt setting 3382 * and freeing sack holes should 3383 * be done by the default stacks 3384 * own tfb_tcp_mtu_chg(). 3385 */ 3386 tp->t_fb->tfb_tcp_mtu_chg(tp); 3387 } 3388 if (tcp_output(tp) < 0) 3389 return (NULL); 3390 else 3391 return (inp); 3392 } 3393 3394 #ifdef INET 3395 /* 3396 * Look-up the routing entry to the peer of this inpcb. If no route 3397 * is found and it cannot be allocated, then return 0. This routine 3398 * is called by TCP routines that access the rmx structure and by 3399 * tcp_mss_update to get the peer/interface MTU. 3400 */ 3401 uint32_t 3402 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3403 { 3404 struct nhop_object *nh; 3405 struct ifnet *ifp; 3406 uint32_t maxmtu = 0; 3407 3408 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3409 3410 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3411 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3412 if (nh == NULL) 3413 return (0); 3414 3415 ifp = nh->nh_ifp; 3416 maxmtu = nh->nh_mtu; 3417 3418 /* Report additional interface capabilities. */ 3419 if (cap != NULL) { 3420 if (ifp->if_capenable & IFCAP_TSO4 && 3421 ifp->if_hwassist & CSUM_TSO) { 3422 cap->ifcap |= CSUM_TSO; 3423 cap->tsomax = ifp->if_hw_tsomax; 3424 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3425 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3426 } 3427 } 3428 } 3429 return (maxmtu); 3430 } 3431 #endif /* INET */ 3432 3433 #ifdef INET6 3434 uint32_t 3435 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3436 { 3437 struct nhop_object *nh; 3438 struct in6_addr dst6; 3439 uint32_t scopeid; 3440 struct ifnet *ifp; 3441 uint32_t maxmtu = 0; 3442 3443 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3444 3445 if (inc->inc_flags & INC_IPV6MINMTU) 3446 return (IPV6_MMTU); 3447 3448 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3449 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3450 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3451 if (nh == NULL) 3452 return (0); 3453 3454 ifp = nh->nh_ifp; 3455 maxmtu = nh->nh_mtu; 3456 3457 /* Report additional interface capabilities. */ 3458 if (cap != NULL) { 3459 if (ifp->if_capenable & IFCAP_TSO6 && 3460 ifp->if_hwassist & CSUM_TSO) { 3461 cap->ifcap |= CSUM_TSO; 3462 cap->tsomax = ifp->if_hw_tsomax; 3463 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3464 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3465 } 3466 } 3467 } 3468 3469 return (maxmtu); 3470 } 3471 3472 /* 3473 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3474 * 3475 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3476 * The right place to do that is ip6_setpktopt() that has just been 3477 * executed. By the way it just filled ip6po_minmtu for us. 3478 */ 3479 void 3480 tcp6_use_min_mtu(struct tcpcb *tp) 3481 { 3482 struct inpcb *inp = tp->t_inpcb; 3483 3484 INP_WLOCK_ASSERT(inp); 3485 /* 3486 * In case of the IPV6_USE_MIN_MTU socket 3487 * option, the INC_IPV6MINMTU flag to announce 3488 * a corresponding MSS during the initial 3489 * handshake. If the TCP connection is not in 3490 * the front states, just reduce the MSS being 3491 * used. This avoids the sending of TCP 3492 * segments which will be fragmented at the 3493 * IPv6 layer. 3494 */ 3495 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3496 if ((tp->t_state >= TCPS_SYN_SENT) && 3497 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3498 struct ip6_pktopts *opt; 3499 3500 opt = inp->in6p_outputopts; 3501 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3502 tp->t_maxseg > TCP6_MSS) 3503 tp->t_maxseg = TCP6_MSS; 3504 } 3505 } 3506 #endif /* INET6 */ 3507 3508 /* 3509 * Calculate effective SMSS per RFC5681 definition for a given TCP 3510 * connection at its current state, taking into account SACK and etc. 3511 */ 3512 u_int 3513 tcp_maxseg(const struct tcpcb *tp) 3514 { 3515 u_int optlen; 3516 3517 if (tp->t_flags & TF_NOOPT) 3518 return (tp->t_maxseg); 3519 3520 /* 3521 * Here we have a simplified code from tcp_addoptions(), 3522 * without a proper loop, and having most of paddings hardcoded. 3523 * We might make mistakes with padding here in some edge cases, 3524 * but this is harmless, since result of tcp_maxseg() is used 3525 * only in cwnd and ssthresh estimations. 3526 */ 3527 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3528 if (tp->t_flags & TF_RCVD_TSTMP) 3529 optlen = TCPOLEN_TSTAMP_APPA; 3530 else 3531 optlen = 0; 3532 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3533 if (tp->t_flags & TF_SIGNATURE) 3534 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3535 #endif 3536 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3537 optlen += TCPOLEN_SACKHDR; 3538 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3539 optlen = PADTCPOLEN(optlen); 3540 } 3541 } else { 3542 if (tp->t_flags & TF_REQ_TSTMP) 3543 optlen = TCPOLEN_TSTAMP_APPA; 3544 else 3545 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3546 if (tp->t_flags & TF_REQ_SCALE) 3547 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3548 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3549 if (tp->t_flags & TF_SIGNATURE) 3550 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3551 #endif 3552 if (tp->t_flags & TF_SACK_PERMIT) 3553 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3554 } 3555 #undef PAD 3556 optlen = min(optlen, TCP_MAXOLEN); 3557 return (tp->t_maxseg - optlen); 3558 } 3559 3560 3561 u_int 3562 tcp_fixed_maxseg(const struct tcpcb *tp) 3563 { 3564 int optlen; 3565 3566 if (tp->t_flags & TF_NOOPT) 3567 return (tp->t_maxseg); 3568 3569 /* 3570 * Here we have a simplified code from tcp_addoptions(), 3571 * without a proper loop, and having most of paddings hardcoded. 3572 * We only consider fixed options that we would send every 3573 * time I.e. SACK is not considered. This is important 3574 * for cc modules to figure out what the modulo of the 3575 * cwnd should be. 3576 */ 3577 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3578 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3579 if (tp->t_flags & TF_RCVD_TSTMP) 3580 optlen = TCPOLEN_TSTAMP_APPA; 3581 else 3582 optlen = 0; 3583 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3584 if (tp->t_flags & TF_SIGNATURE) 3585 optlen += PAD(TCPOLEN_SIGNATURE); 3586 #endif 3587 } else { 3588 if (tp->t_flags & TF_REQ_TSTMP) 3589 optlen = TCPOLEN_TSTAMP_APPA; 3590 else 3591 optlen = PAD(TCPOLEN_MAXSEG); 3592 if (tp->t_flags & TF_REQ_SCALE) 3593 optlen += PAD(TCPOLEN_WINDOW); 3594 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3595 if (tp->t_flags & TF_SIGNATURE) 3596 optlen += PAD(TCPOLEN_SIGNATURE); 3597 #endif 3598 if (tp->t_flags & TF_SACK_PERMIT) 3599 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3600 } 3601 #undef PAD 3602 optlen = min(optlen, TCP_MAXOLEN); 3603 return (tp->t_maxseg - optlen); 3604 } 3605 3606 3607 3608 static int 3609 sysctl_drop(SYSCTL_HANDLER_ARGS) 3610 { 3611 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3612 struct sockaddr_storage addrs[2]; 3613 struct inpcb *inp; 3614 struct tcpcb *tp; 3615 #ifdef INET 3616 struct sockaddr_in *fin = NULL, *lin = NULL; 3617 #endif 3618 struct epoch_tracker et; 3619 #ifdef INET6 3620 struct sockaddr_in6 *fin6, *lin6; 3621 #endif 3622 int error; 3623 3624 inp = NULL; 3625 #ifdef INET6 3626 fin6 = lin6 = NULL; 3627 #endif 3628 error = 0; 3629 3630 if (req->oldptr != NULL || req->oldlen != 0) 3631 return (EINVAL); 3632 if (req->newptr == NULL) 3633 return (EPERM); 3634 if (req->newlen < sizeof(addrs)) 3635 return (ENOMEM); 3636 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3637 if (error) 3638 return (error); 3639 3640 switch (addrs[0].ss_family) { 3641 #ifdef INET6 3642 case AF_INET6: 3643 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3644 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3645 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3646 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3647 return (EINVAL); 3648 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3649 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3650 return (EINVAL); 3651 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3652 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3653 #ifdef INET 3654 fin = (struct sockaddr_in *)&addrs[0]; 3655 lin = (struct sockaddr_in *)&addrs[1]; 3656 #endif 3657 break; 3658 } 3659 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3660 if (error) 3661 return (error); 3662 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3663 if (error) 3664 return (error); 3665 break; 3666 #endif 3667 #ifdef INET 3668 case AF_INET: 3669 fin = (struct sockaddr_in *)&addrs[0]; 3670 lin = (struct sockaddr_in *)&addrs[1]; 3671 if (fin->sin_len != sizeof(struct sockaddr_in) || 3672 lin->sin_len != sizeof(struct sockaddr_in)) 3673 return (EINVAL); 3674 break; 3675 #endif 3676 default: 3677 return (EINVAL); 3678 } 3679 NET_EPOCH_ENTER(et); 3680 switch (addrs[0].ss_family) { 3681 #ifdef INET6 3682 case AF_INET6: 3683 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3684 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3685 INPLOOKUP_WLOCKPCB, NULL); 3686 break; 3687 #endif 3688 #ifdef INET 3689 case AF_INET: 3690 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3691 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3692 break; 3693 #endif 3694 } 3695 if (inp != NULL) { 3696 if ((inp->inp_flags & INP_DROPPED) == 0 && 3697 !SOLISTENING(inp->inp_socket)) { 3698 tp = intotcpcb(inp); 3699 tp = tcp_drop(tp, ECONNABORTED); 3700 if (tp != NULL) 3701 INP_WUNLOCK(inp); 3702 } else 3703 INP_WUNLOCK(inp); 3704 } else 3705 error = ESRCH; 3706 NET_EPOCH_EXIT(et); 3707 return (error); 3708 } 3709 3710 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3711 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3712 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3713 "Drop TCP connection"); 3714 3715 static int 3716 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3717 { 3718 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3719 &tcp_ctloutput_set)); 3720 } 3721 3722 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3723 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3724 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3725 "Set socket option for TCP endpoint"); 3726 3727 #ifdef KERN_TLS 3728 static int 3729 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3730 { 3731 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3732 struct sockaddr_storage addrs[2]; 3733 struct inpcb *inp; 3734 #ifdef INET 3735 struct sockaddr_in *fin = NULL, *lin = NULL; 3736 #endif 3737 struct epoch_tracker et; 3738 #ifdef INET6 3739 struct sockaddr_in6 *fin6, *lin6; 3740 #endif 3741 int error; 3742 3743 inp = NULL; 3744 #ifdef INET6 3745 fin6 = lin6 = NULL; 3746 #endif 3747 error = 0; 3748 3749 if (req->oldptr != NULL || req->oldlen != 0) 3750 return (EINVAL); 3751 if (req->newptr == NULL) 3752 return (EPERM); 3753 if (req->newlen < sizeof(addrs)) 3754 return (ENOMEM); 3755 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3756 if (error) 3757 return (error); 3758 3759 switch (addrs[0].ss_family) { 3760 #ifdef INET6 3761 case AF_INET6: 3762 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3763 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3764 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3765 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3766 return (EINVAL); 3767 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3768 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3769 return (EINVAL); 3770 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3771 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3772 #ifdef INET 3773 fin = (struct sockaddr_in *)&addrs[0]; 3774 lin = (struct sockaddr_in *)&addrs[1]; 3775 #endif 3776 break; 3777 } 3778 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3779 if (error) 3780 return (error); 3781 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3782 if (error) 3783 return (error); 3784 break; 3785 #endif 3786 #ifdef INET 3787 case AF_INET: 3788 fin = (struct sockaddr_in *)&addrs[0]; 3789 lin = (struct sockaddr_in *)&addrs[1]; 3790 if (fin->sin_len != sizeof(struct sockaddr_in) || 3791 lin->sin_len != sizeof(struct sockaddr_in)) 3792 return (EINVAL); 3793 break; 3794 #endif 3795 default: 3796 return (EINVAL); 3797 } 3798 NET_EPOCH_ENTER(et); 3799 switch (addrs[0].ss_family) { 3800 #ifdef INET6 3801 case AF_INET6: 3802 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3803 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3804 INPLOOKUP_WLOCKPCB, NULL); 3805 break; 3806 #endif 3807 #ifdef INET 3808 case AF_INET: 3809 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3810 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3811 break; 3812 #endif 3813 } 3814 NET_EPOCH_EXIT(et); 3815 if (inp != NULL) { 3816 if ((inp->inp_flags & INP_DROPPED) != 0 || 3817 inp->inp_socket == NULL) { 3818 error = ECONNRESET; 3819 INP_WUNLOCK(inp); 3820 } else { 3821 struct socket *so; 3822 3823 so = inp->inp_socket; 3824 soref(so); 3825 error = ktls_set_tx_mode(so, 3826 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3827 INP_WUNLOCK(inp); 3828 sorele(so); 3829 } 3830 } else 3831 error = ESRCH; 3832 return (error); 3833 } 3834 3835 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3836 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3837 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3838 "Switch TCP connection to SW TLS"); 3839 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3840 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3841 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3842 "Switch TCP connection to ifnet TLS"); 3843 #endif 3844 3845 /* 3846 * Generate a standardized TCP log line for use throughout the 3847 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3848 * allow use in the interrupt context. 3849 * 3850 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3851 * NB: The function may return NULL if memory allocation failed. 3852 * 3853 * Due to header inclusion and ordering limitations the struct ip 3854 * and ip6_hdr pointers have to be passed as void pointers. 3855 */ 3856 char * 3857 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3858 const void *ip6hdr) 3859 { 3860 3861 /* Is logging enabled? */ 3862 if (V_tcp_log_in_vain == 0) 3863 return (NULL); 3864 3865 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3866 } 3867 3868 char * 3869 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3870 const void *ip6hdr) 3871 { 3872 3873 /* Is logging enabled? */ 3874 if (tcp_log_debug == 0) 3875 return (NULL); 3876 3877 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3878 } 3879 3880 static char * 3881 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3882 const void *ip6hdr) 3883 { 3884 char *s, *sp; 3885 size_t size; 3886 #ifdef INET 3887 const struct ip *ip = (const struct ip *)ip4hdr; 3888 #endif 3889 #ifdef INET6 3890 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3891 #endif /* INET6 */ 3892 3893 /* 3894 * The log line looks like this: 3895 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3896 */ 3897 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3898 sizeof(PRINT_TH_FLAGS) + 1 + 3899 #ifdef INET6 3900 2 * INET6_ADDRSTRLEN; 3901 #else 3902 2 * INET_ADDRSTRLEN; 3903 #endif /* INET6 */ 3904 3905 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3906 if (s == NULL) 3907 return (NULL); 3908 3909 strcat(s, "TCP: ["); 3910 sp = s + strlen(s); 3911 3912 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3913 inet_ntoa_r(inc->inc_faddr, sp); 3914 sp = s + strlen(s); 3915 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3916 sp = s + strlen(s); 3917 inet_ntoa_r(inc->inc_laddr, sp); 3918 sp = s + strlen(s); 3919 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3920 #ifdef INET6 3921 } else if (inc) { 3922 ip6_sprintf(sp, &inc->inc6_faddr); 3923 sp = s + strlen(s); 3924 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3925 sp = s + strlen(s); 3926 ip6_sprintf(sp, &inc->inc6_laddr); 3927 sp = s + strlen(s); 3928 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3929 } else if (ip6 && th) { 3930 ip6_sprintf(sp, &ip6->ip6_src); 3931 sp = s + strlen(s); 3932 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3933 sp = s + strlen(s); 3934 ip6_sprintf(sp, &ip6->ip6_dst); 3935 sp = s + strlen(s); 3936 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3937 #endif /* INET6 */ 3938 #ifdef INET 3939 } else if (ip && th) { 3940 inet_ntoa_r(ip->ip_src, sp); 3941 sp = s + strlen(s); 3942 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3943 sp = s + strlen(s); 3944 inet_ntoa_r(ip->ip_dst, sp); 3945 sp = s + strlen(s); 3946 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3947 #endif /* INET */ 3948 } else { 3949 free(s, M_TCPLOG); 3950 return (NULL); 3951 } 3952 sp = s + strlen(s); 3953 if (th) 3954 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3955 if (*(s + size - 1) != '\0') 3956 panic("%s: string too long", __func__); 3957 return (s); 3958 } 3959 3960 /* 3961 * A subroutine which makes it easy to track TCP state changes with DTrace. 3962 * This function shouldn't be called for t_state initializations that don't 3963 * correspond to actual TCP state transitions. 3964 */ 3965 void 3966 tcp_state_change(struct tcpcb *tp, int newstate) 3967 { 3968 #if defined(KDTRACE_HOOKS) 3969 int pstate = tp->t_state; 3970 #endif 3971 3972 TCPSTATES_DEC(tp->t_state); 3973 TCPSTATES_INC(newstate); 3974 tp->t_state = newstate; 3975 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3976 } 3977 3978 /* 3979 * Create an external-format (``xtcpcb'') structure using the information in 3980 * the kernel-format tcpcb structure pointed to by tp. This is done to 3981 * reduce the spew of irrelevant information over this interface, to isolate 3982 * user code from changes in the kernel structure, and potentially to provide 3983 * information-hiding if we decide that some of this information should be 3984 * hidden from users. 3985 */ 3986 void 3987 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3988 { 3989 struct tcpcb *tp = intotcpcb(inp); 3990 sbintime_t now; 3991 3992 bzero(xt, sizeof(*xt)); 3993 xt->t_state = tp->t_state; 3994 xt->t_logstate = tp->t_logstate; 3995 xt->t_flags = tp->t_flags; 3996 xt->t_sndzerowin = tp->t_sndzerowin; 3997 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3998 xt->t_rcvoopack = tp->t_rcvoopack; 3999 xt->t_rcv_wnd = tp->rcv_wnd; 4000 xt->t_snd_wnd = tp->snd_wnd; 4001 xt->t_snd_cwnd = tp->snd_cwnd; 4002 xt->t_snd_ssthresh = tp->snd_ssthresh; 4003 xt->t_dsack_bytes = tp->t_dsack_bytes; 4004 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4005 xt->t_dsack_pack = tp->t_dsack_pack; 4006 xt->t_maxseg = tp->t_maxseg; 4007 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4008 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4009 4010 now = getsbinuptime(); 4011 #define COPYTIMER(ttt) do { \ 4012 if (callout_active(&tp->t_timers->ttt)) \ 4013 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 4014 SBT_1MS; \ 4015 else \ 4016 xt->ttt = 0; \ 4017 } while (0) 4018 COPYTIMER(tt_delack); 4019 COPYTIMER(tt_rexmt); 4020 COPYTIMER(tt_persist); 4021 COPYTIMER(tt_keep); 4022 COPYTIMER(tt_2msl); 4023 #undef COPYTIMER 4024 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4025 4026 xt->xt_encaps_port = tp->t_port; 4027 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4028 TCP_FUNCTION_NAME_LEN_MAX); 4029 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 4030 #ifdef TCP_BLACKBOX 4031 (void)tcp_log_get_id(tp, xt->xt_logid); 4032 #endif 4033 4034 xt->xt_len = sizeof(struct xtcpcb); 4035 in_pcbtoxinpcb(inp, &xt->xt_inp); 4036 if (inp->inp_socket == NULL) 4037 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 4038 } 4039 4040 void 4041 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4042 { 4043 uint32_t bit, i; 4044 4045 if ((tp == NULL) || 4046 (status > TCP_EI_STATUS_MAX_VALUE) || 4047 (status == 0)) { 4048 /* Invalid */ 4049 return; 4050 } 4051 if (status > (sizeof(uint32_t) * 8)) { 4052 /* Should this be a KASSERT? */ 4053 return; 4054 } 4055 bit = 1U << (status - 1); 4056 if (bit & tp->t_end_info_status) { 4057 /* already logged */ 4058 return; 4059 } 4060 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4061 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4062 tp->t_end_info_bytes[i] = status; 4063 tp->t_end_info_status |= bit; 4064 break; 4065 } 4066 } 4067 } 4068 4069 int 4070 tcp_can_enable_pacing(void) 4071 { 4072 4073 if ((tcp_pacing_limit == -1) || 4074 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4075 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4076 shadow_num_connections = number_of_tcp_connections_pacing; 4077 return (1); 4078 } else { 4079 return (0); 4080 } 4081 } 4082 4083 static uint8_t tcp_pacing_warning = 0; 4084 4085 void 4086 tcp_decrement_paced_conn(void) 4087 { 4088 uint32_t ret; 4089 4090 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4091 shadow_num_connections = number_of_tcp_connections_pacing; 4092 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4093 if (ret == 0) { 4094 if (tcp_pacing_limit != -1) { 4095 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4096 tcp_pacing_limit = 0; 4097 } else if (tcp_pacing_warning == 0) { 4098 printf("Warning pacing count is invalid, invalid decrement\n"); 4099 tcp_pacing_warning = 1; 4100 } 4101 } 4102 } 4103