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