1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/arb.h> 46 #include <sys/callout.h> 47 #include <sys/eventhandler.h> 48 #ifdef TCP_HHOOK 49 #include <sys/hhook.h> 50 #endif 51 #include <sys/kernel.h> 52 #ifdef TCP_HHOOK 53 #include <sys/khelp.h> 54 #endif 55 #ifdef KERN_TLS 56 #include <sys/ktls.h> 57 #endif 58 #include <sys/qmath.h> 59 #include <sys/stats.h> 60 #include <sys/sysctl.h> 61 #include <sys/jail.h> 62 #include <sys/malloc.h> 63 #include <sys/refcount.h> 64 #include <sys/mbuf.h> 65 #ifdef INET6 66 #include <sys/domain.h> 67 #endif 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/sdt.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/protosw.h> 74 #include <sys/random.h> 75 76 #include <vm/uma.h> 77 78 #include <net/route.h> 79 #include <net/route/nhop.h> 80 #include <net/if.h> 81 #include <net/if_var.h> 82 #include <net/vnet.h> 83 84 #include <netinet/in.h> 85 #include <netinet/in_fib.h> 86 #include <netinet/in_kdtrace.h> 87 #include <netinet/in_pcb.h> 88 #include <netinet/in_systm.h> 89 #include <netinet/in_var.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> 92 #include <netinet/ip_var.h> 93 #ifdef INET6 94 #include <netinet/icmp6.h> 95 #include <netinet/ip6.h> 96 #include <netinet6/in6_fib.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/ip6_var.h> 99 #include <netinet6/scope6_var.h> 100 #include <netinet6/nd6.h> 101 #endif 102 103 #include <netinet/tcp.h> 104 #ifdef INVARIANTS 105 #define TCPSTATES 106 #endif 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_seq.h> 109 #include <netinet/tcp_timer.h> 110 #include <netinet/tcp_var.h> 111 #include <netinet/tcp_log_buf.h> 112 #include <netinet/tcp_syncache.h> 113 #include <netinet/tcp_hpts.h> 114 #include <netinet/cc/cc.h> 115 #ifdef INET6 116 #include <netinet6/tcp6_var.h> 117 #endif 118 #include <netinet/tcpip.h> 119 #include <netinet/tcp_fastopen.h> 120 #ifdef TCPPCAP 121 #include <netinet/tcp_pcap.h> 122 #endif 123 #ifdef TCPDEBUG 124 #include <netinet/tcp_debug.h> 125 #endif 126 #ifdef INET6 127 #include <netinet6/ip6protosw.h> 128 #endif 129 #ifdef TCP_OFFLOAD 130 #include <netinet/tcp_offload.h> 131 #endif 132 #include <netinet/udp.h> 133 #include <netinet/udp_var.h> 134 135 #include <netipsec/ipsec_support.h> 136 137 #include <machine/in_cksum.h> 138 #include <crypto/siphash/siphash.h> 139 140 #include <security/mac/mac_framework.h> 141 142 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 143 #ifdef INET6 144 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 145 #endif 146 147 #ifdef NETFLIX_EXP_DETECTION 148 /* Sack attack detection thresholds and such */ 149 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 150 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 151 "Sack Attack detection thresholds"); 152 int32_t tcp_force_detection = 0; 153 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 154 CTLFLAG_RW, 155 &tcp_force_detection, 0, 156 "Do we force detection even if the INP has it off?"); 157 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 158 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 159 CTLFLAG_RW, 160 &tcp_sack_to_ack_thresh, 700, 161 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 162 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 163 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 164 CTLFLAG_RW, 165 &tcp_sack_to_move_thresh, 600, 166 "Percentage of sack moves we must see above (10.1 percent is 101)"); 167 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 168 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 169 CTLFLAG_RW, 170 &tcp_restoral_thresh, 550, 171 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 172 int32_t tcp_sad_decay_val = 800; 173 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 174 CTLFLAG_RW, 175 &tcp_sad_decay_val, 800, 176 "The decay percentage (10.1 percent equals 101 )"); 177 int32_t tcp_map_minimum = 500; 178 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 179 CTLFLAG_RW, 180 &tcp_map_minimum, 500, 181 "Number of Map enteries before we start detection"); 182 int32_t tcp_attack_on_turns_on_logging = 0; 183 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 184 CTLFLAG_RW, 185 &tcp_attack_on_turns_on_logging, 0, 186 "When we have a positive hit on attack, do we turn on logging?"); 187 int32_t tcp_sad_pacing_interval = 2000; 188 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 189 CTLFLAG_RW, 190 &tcp_sad_pacing_interval, 2000, 191 "What is the minimum pacing interval for a classified attacker?"); 192 193 int32_t tcp_sad_low_pps = 100; 194 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 195 CTLFLAG_RW, 196 &tcp_sad_low_pps, 100, 197 "What is the input pps that below which we do not decay?"); 198 #endif 199 uint32_t tcp_ack_war_time_window = 1000; 200 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 201 CTLFLAG_RW, 202 &tcp_ack_war_time_window, 1000, 203 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 204 uint32_t tcp_ack_war_cnt = 5; 205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 206 CTLFLAG_RW, 207 &tcp_ack_war_cnt, 5, 208 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 209 210 struct rwlock tcp_function_lock; 211 212 static int 213 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 214 { 215 int error, new; 216 217 new = V_tcp_mssdflt; 218 error = sysctl_handle_int(oidp, &new, 0, req); 219 if (error == 0 && req->newptr) { 220 if (new < TCP_MINMSS) 221 error = EINVAL; 222 else 223 V_tcp_mssdflt = new; 224 } 225 return (error); 226 } 227 228 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 229 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 230 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 231 "Default TCP Maximum Segment Size"); 232 233 #ifdef INET6 234 static int 235 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 236 { 237 int error, new; 238 239 new = V_tcp_v6mssdflt; 240 error = sysctl_handle_int(oidp, &new, 0, req); 241 if (error == 0 && req->newptr) { 242 if (new < TCP_MINMSS) 243 error = EINVAL; 244 else 245 V_tcp_v6mssdflt = new; 246 } 247 return (error); 248 } 249 250 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 251 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 252 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 253 "Default TCP Maximum Segment Size for IPv6"); 254 #endif /* INET6 */ 255 256 /* 257 * Minimum MSS we accept and use. This prevents DoS attacks where 258 * we are forced to a ridiculous low MSS like 20 and send hundreds 259 * of packets instead of one. The effect scales with the available 260 * bandwidth and quickly saturates the CPU and network interface 261 * with packet generation and sending. Set to zero to disable MINMSS 262 * checking. This setting prevents us from sending too small packets. 263 */ 264 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 265 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 266 &VNET_NAME(tcp_minmss), 0, 267 "Minimum TCP Maximum Segment Size"); 268 269 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 270 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 271 &VNET_NAME(tcp_do_rfc1323), 0, 272 "Enable rfc1323 (high performance TCP) extensions"); 273 274 /* 275 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 276 * Some stacks negotiate TS, but never send them after connection setup. Some 277 * stacks negotiate TS, but don't send them when sending keep-alive segments. 278 * These include modern widely deployed TCP stacks. 279 * Therefore tolerating violations for now... 280 */ 281 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 282 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 283 &VNET_NAME(tcp_tolerate_missing_ts), 0, 284 "Tolerate missing TCP timestamps"); 285 286 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 288 &VNET_NAME(tcp_ts_offset_per_conn), 0, 289 "Initialize TCP timestamps per connection instead of per host pair"); 290 291 /* How many connections are pacing */ 292 static volatile uint32_t number_of_tcp_connections_pacing = 0; 293 static uint32_t shadow_num_connections = 0; 294 295 static int tcp_pacing_limit = 10000; 296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 297 &tcp_pacing_limit, 1000, 298 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 299 300 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 301 &shadow_num_connections, 0, "Number of TCP connections being paced"); 302 303 static int tcp_log_debug = 0; 304 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 305 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 306 307 static int tcp_tcbhashsize; 308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 309 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 310 311 static int do_tcpdrain = 1; 312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 313 "Enable tcp_drain routine for extra help when low on mbufs"); 314 315 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 316 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 317 318 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 319 #define V_icmp_may_rst VNET(icmp_may_rst) 320 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 321 &VNET_NAME(icmp_may_rst), 0, 322 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 323 324 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 325 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 326 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 327 &VNET_NAME(tcp_isn_reseed_interval), 0, 328 "Seconds between reseeding of ISN secret"); 329 330 static int tcp_soreceive_stream; 331 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 332 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 333 334 VNET_DEFINE(uma_zone_t, sack_hole_zone); 335 #define V_sack_hole_zone VNET(sack_hole_zone) 336 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 337 static int 338 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 339 { 340 int error; 341 uint32_t new; 342 343 new = V_tcp_map_entries_limit; 344 error = sysctl_handle_int(oidp, &new, 0, req); 345 if (error == 0 && req->newptr) { 346 /* only allow "0" and value > minimum */ 347 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 348 error = EINVAL; 349 else 350 V_tcp_map_entries_limit = new; 351 } 352 return (error); 353 } 354 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 355 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 356 &VNET_NAME(tcp_map_entries_limit), 0, 357 &sysctl_net_inet_tcp_map_limit_check, "IU", 358 "Total sendmap entries limit"); 359 360 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 361 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 362 &VNET_NAME(tcp_map_split_limit), 0, 363 "Total sendmap split entries limit"); 364 365 #ifdef TCP_HHOOK 366 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 367 #endif 368 369 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 370 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 371 #define V_ts_offset_secret VNET(ts_offset_secret) 372 373 static int tcp_default_fb_init(struct tcpcb *tp); 374 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 375 static int tcp_default_handoff_ok(struct tcpcb *tp); 376 static struct inpcb *tcp_notify(struct inpcb *, int); 377 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 378 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 379 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 380 void *ip4hdr, const void *ip6hdr); 381 382 static struct tcp_function_block tcp_def_funcblk = { 383 .tfb_tcp_block_name = "freebsd", 384 .tfb_tcp_output = tcp_default_output, 385 .tfb_tcp_do_segment = tcp_do_segment, 386 .tfb_tcp_ctloutput = tcp_default_ctloutput, 387 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 388 .tfb_tcp_fb_init = tcp_default_fb_init, 389 .tfb_tcp_fb_fini = tcp_default_fb_fini, 390 }; 391 392 static int tcp_fb_cnt = 0; 393 struct tcp_funchead t_functions; 394 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 395 396 void 397 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 398 { 399 TCPSTAT_INC(tcps_dsack_count); 400 tp->t_dsack_pack++; 401 if (tlp == 0) { 402 if (SEQ_GT(end, start)) { 403 tp->t_dsack_bytes += (end - start); 404 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 405 } else { 406 tp->t_dsack_tlp_bytes += (start - end); 407 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 408 } 409 } else { 410 if (SEQ_GT(end, start)) { 411 tp->t_dsack_bytes += (end - start); 412 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 413 } else { 414 tp->t_dsack_tlp_bytes += (start - end); 415 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 416 } 417 } 418 } 419 420 static struct tcp_function_block * 421 find_tcp_functions_locked(struct tcp_function_set *fs) 422 { 423 struct tcp_function *f; 424 struct tcp_function_block *blk=NULL; 425 426 TAILQ_FOREACH(f, &t_functions, tf_next) { 427 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 428 blk = f->tf_fb; 429 break; 430 } 431 } 432 return(blk); 433 } 434 435 static struct tcp_function_block * 436 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 437 { 438 struct tcp_function_block *rblk=NULL; 439 struct tcp_function *f; 440 441 TAILQ_FOREACH(f, &t_functions, tf_next) { 442 if (f->tf_fb == blk) { 443 rblk = blk; 444 if (s) { 445 *s = f; 446 } 447 break; 448 } 449 } 450 return (rblk); 451 } 452 453 struct tcp_function_block * 454 find_and_ref_tcp_functions(struct tcp_function_set *fs) 455 { 456 struct tcp_function_block *blk; 457 458 rw_rlock(&tcp_function_lock); 459 blk = find_tcp_functions_locked(fs); 460 if (blk) 461 refcount_acquire(&blk->tfb_refcnt); 462 rw_runlock(&tcp_function_lock); 463 return(blk); 464 } 465 466 struct tcp_function_block * 467 find_and_ref_tcp_fb(struct tcp_function_block *blk) 468 { 469 struct tcp_function_block *rblk; 470 471 rw_rlock(&tcp_function_lock); 472 rblk = find_tcp_fb_locked(blk, NULL); 473 if (rblk) 474 refcount_acquire(&rblk->tfb_refcnt); 475 rw_runlock(&tcp_function_lock); 476 return(rblk); 477 } 478 479 /* Find a matching alias for the given tcp_function_block. */ 480 int 481 find_tcp_function_alias(struct tcp_function_block *blk, 482 struct tcp_function_set *fs) 483 { 484 struct tcp_function *f; 485 int found; 486 487 found = 0; 488 rw_rlock(&tcp_function_lock); 489 TAILQ_FOREACH(f, &t_functions, tf_next) { 490 if ((f->tf_fb == blk) && 491 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 492 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 493 /* Matching function block with different name. */ 494 strncpy(fs->function_set_name, f->tf_name, 495 TCP_FUNCTION_NAME_LEN_MAX); 496 found = 1; 497 break; 498 } 499 } 500 /* Null terminate the string appropriately. */ 501 if (found) { 502 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 503 } else { 504 fs->function_set_name[0] = '\0'; 505 } 506 rw_runlock(&tcp_function_lock); 507 return (found); 508 } 509 510 static struct tcp_function_block * 511 find_and_ref_tcp_default_fb(void) 512 { 513 struct tcp_function_block *rblk; 514 515 rw_rlock(&tcp_function_lock); 516 rblk = tcp_func_set_ptr; 517 refcount_acquire(&rblk->tfb_refcnt); 518 rw_runlock(&tcp_function_lock); 519 return (rblk); 520 } 521 522 void 523 tcp_switch_back_to_default(struct tcpcb *tp) 524 { 525 struct tcp_function_block *tfb; 526 527 KASSERT(tp->t_fb != &tcp_def_funcblk, 528 ("%s: called by the built-in default stack", __func__)); 529 530 /* 531 * Release the old stack. This function will either find a new one 532 * or panic. 533 */ 534 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 535 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 536 refcount_release(&tp->t_fb->tfb_refcnt); 537 538 /* 539 * Now, we'll find a new function block to use. 540 * Start by trying the current user-selected 541 * default, unless this stack is the user-selected 542 * default. 543 */ 544 tfb = find_and_ref_tcp_default_fb(); 545 if (tfb == tp->t_fb) { 546 refcount_release(&tfb->tfb_refcnt); 547 tfb = NULL; 548 } 549 /* Does the stack accept this connection? */ 550 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 551 (*tfb->tfb_tcp_handoff_ok)(tp)) { 552 refcount_release(&tfb->tfb_refcnt); 553 tfb = NULL; 554 } 555 /* Try to use that stack. */ 556 if (tfb != NULL) { 557 /* Initialize the new stack. If it succeeds, we are done. */ 558 tp->t_fb = tfb; 559 if (tp->t_fb->tfb_tcp_fb_init == NULL || 560 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 561 return; 562 563 /* 564 * Initialization failed. Release the reference count on 565 * the stack. 566 */ 567 refcount_release(&tfb->tfb_refcnt); 568 } 569 570 /* 571 * If that wasn't feasible, use the built-in default 572 * stack which is not allowed to reject anyone. 573 */ 574 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 575 if (tfb == NULL) { 576 /* there always should be a default */ 577 panic("Can't refer to tcp_def_funcblk"); 578 } 579 if (tfb->tfb_tcp_handoff_ok != NULL) { 580 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 581 /* The default stack cannot say no */ 582 panic("Default stack rejects a new session?"); 583 } 584 } 585 tp->t_fb = tfb; 586 if (tp->t_fb->tfb_tcp_fb_init != NULL && 587 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 588 /* The default stack cannot fail */ 589 panic("Default stack initialization failed"); 590 } 591 } 592 593 static void 594 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 595 const struct sockaddr *sa, void *ctx) 596 { 597 struct ip *iph; 598 #ifdef INET6 599 struct ip6_hdr *ip6; 600 #endif 601 struct udphdr *uh; 602 struct tcphdr *th; 603 int thlen; 604 uint16_t port; 605 606 TCPSTAT_INC(tcps_tunneled_pkts); 607 if ((m->m_flags & M_PKTHDR) == 0) { 608 /* Can't handle one that is not a pkt hdr */ 609 TCPSTAT_INC(tcps_tunneled_errs); 610 goto out; 611 } 612 thlen = sizeof(struct tcphdr); 613 if (m->m_len < off + sizeof(struct udphdr) + thlen && 614 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 615 TCPSTAT_INC(tcps_tunneled_errs); 616 goto out; 617 } 618 iph = mtod(m, struct ip *); 619 uh = (struct udphdr *)((caddr_t)iph + off); 620 th = (struct tcphdr *)(uh + 1); 621 thlen = th->th_off << 2; 622 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 623 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 624 if (m == NULL) { 625 TCPSTAT_INC(tcps_tunneled_errs); 626 goto out; 627 } else { 628 iph = mtod(m, struct ip *); 629 uh = (struct udphdr *)((caddr_t)iph + off); 630 th = (struct tcphdr *)(uh + 1); 631 } 632 } 633 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 634 bcopy(th, uh, m->m_len - off); 635 m->m_len -= sizeof(struct udphdr); 636 m->m_pkthdr.len -= sizeof(struct udphdr); 637 /* 638 * We use the same algorithm for 639 * both UDP and TCP for c-sum. So 640 * the code in tcp_input will skip 641 * the checksum. So we do nothing 642 * with the flag (m->m_pkthdr.csum_flags). 643 */ 644 switch (iph->ip_v) { 645 #ifdef INET 646 case IPVERSION: 647 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 648 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 649 break; 650 #endif 651 #ifdef INET6 652 case IPV6_VERSION >> 4: 653 ip6 = mtod(m, struct ip6_hdr *); 654 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 655 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 656 break; 657 #endif 658 default: 659 goto out; 660 break; 661 } 662 return; 663 out: 664 m_freem(m); 665 } 666 667 static int 668 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 669 { 670 int error=ENOENT; 671 struct tcp_function_set fs; 672 struct tcp_function_block *blk; 673 674 memset(&fs, 0, sizeof(fs)); 675 rw_rlock(&tcp_function_lock); 676 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 677 if (blk) { 678 /* Found him */ 679 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 680 fs.pcbcnt = blk->tfb_refcnt; 681 } 682 rw_runlock(&tcp_function_lock); 683 error = sysctl_handle_string(oidp, fs.function_set_name, 684 sizeof(fs.function_set_name), req); 685 686 /* Check for error or no change */ 687 if (error != 0 || req->newptr == NULL) 688 return(error); 689 690 rw_wlock(&tcp_function_lock); 691 blk = find_tcp_functions_locked(&fs); 692 if ((blk == NULL) || 693 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 694 error = ENOENT; 695 goto done; 696 } 697 tcp_func_set_ptr = blk; 698 done: 699 rw_wunlock(&tcp_function_lock); 700 return (error); 701 } 702 703 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 704 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 705 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 706 "Set/get the default TCP functions"); 707 708 static int 709 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 710 { 711 int error, cnt, linesz; 712 struct tcp_function *f; 713 char *buffer, *cp; 714 size_t bufsz, outsz; 715 bool alias; 716 717 cnt = 0; 718 rw_rlock(&tcp_function_lock); 719 TAILQ_FOREACH(f, &t_functions, tf_next) { 720 cnt++; 721 } 722 rw_runlock(&tcp_function_lock); 723 724 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 725 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 726 727 error = 0; 728 cp = buffer; 729 730 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 731 "Alias", "PCB count"); 732 cp += linesz; 733 bufsz -= linesz; 734 outsz = linesz; 735 736 rw_rlock(&tcp_function_lock); 737 TAILQ_FOREACH(f, &t_functions, tf_next) { 738 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 739 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 740 f->tf_fb->tfb_tcp_block_name, 741 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 742 alias ? f->tf_name : "-", 743 f->tf_fb->tfb_refcnt); 744 if (linesz >= bufsz) { 745 error = EOVERFLOW; 746 break; 747 } 748 cp += linesz; 749 bufsz -= linesz; 750 outsz += linesz; 751 } 752 rw_runlock(&tcp_function_lock); 753 if (error == 0) 754 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 755 free(buffer, M_TEMP); 756 return (error); 757 } 758 759 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 760 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 761 NULL, 0, sysctl_net_inet_list_available, "A", 762 "list available TCP Function sets"); 763 764 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 765 766 #ifdef INET 767 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 768 #define V_udp4_tun_socket VNET(udp4_tun_socket) 769 #endif 770 #ifdef INET6 771 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 772 #define V_udp6_tun_socket VNET(udp6_tun_socket) 773 #endif 774 775 static void 776 tcp_over_udp_stop(void) 777 { 778 /* 779 * This function assumes sysctl caller holds inp_rinfo_lock() 780 * for writing! 781 */ 782 #ifdef INET 783 if (V_udp4_tun_socket != NULL) { 784 soclose(V_udp4_tun_socket); 785 V_udp4_tun_socket = NULL; 786 } 787 #endif 788 #ifdef INET6 789 if (V_udp6_tun_socket != NULL) { 790 soclose(V_udp6_tun_socket); 791 V_udp6_tun_socket = NULL; 792 } 793 #endif 794 } 795 796 static int 797 tcp_over_udp_start(void) 798 { 799 uint16_t port; 800 int ret; 801 #ifdef INET 802 struct sockaddr_in sin; 803 #endif 804 #ifdef INET6 805 struct sockaddr_in6 sin6; 806 #endif 807 /* 808 * This function assumes sysctl caller holds inp_info_rlock() 809 * for writing! 810 */ 811 port = V_tcp_udp_tunneling_port; 812 if (ntohs(port) == 0) { 813 /* Must have a port set */ 814 return (EINVAL); 815 } 816 #ifdef INET 817 if (V_udp4_tun_socket != NULL) { 818 /* Already running -- must stop first */ 819 return (EALREADY); 820 } 821 #endif 822 #ifdef INET6 823 if (V_udp6_tun_socket != NULL) { 824 /* Already running -- must stop first */ 825 return (EALREADY); 826 } 827 #endif 828 #ifdef INET 829 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 830 SOCK_DGRAM, IPPROTO_UDP, 831 curthread->td_ucred, curthread))) { 832 tcp_over_udp_stop(); 833 return (ret); 834 } 835 /* Call the special UDP hook. */ 836 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 837 tcp_recv_udp_tunneled_packet, 838 tcp_ctlinput_viaudp, 839 NULL))) { 840 tcp_over_udp_stop(); 841 return (ret); 842 } 843 /* Ok, we have a socket, bind it to the port. */ 844 memset(&sin, 0, sizeof(struct sockaddr_in)); 845 sin.sin_len = sizeof(struct sockaddr_in); 846 sin.sin_family = AF_INET; 847 sin.sin_port = htons(port); 848 if ((ret = sobind(V_udp4_tun_socket, 849 (struct sockaddr *)&sin, curthread))) { 850 tcp_over_udp_stop(); 851 return (ret); 852 } 853 #endif 854 #ifdef INET6 855 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 856 SOCK_DGRAM, IPPROTO_UDP, 857 curthread->td_ucred, curthread))) { 858 tcp_over_udp_stop(); 859 return (ret); 860 } 861 /* Call the special UDP hook. */ 862 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 863 tcp_recv_udp_tunneled_packet, 864 tcp6_ctlinput_viaudp, 865 NULL))) { 866 tcp_over_udp_stop(); 867 return (ret); 868 } 869 /* Ok, we have a socket, bind it to the port. */ 870 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 871 sin6.sin6_len = sizeof(struct sockaddr_in6); 872 sin6.sin6_family = AF_INET6; 873 sin6.sin6_port = htons(port); 874 if ((ret = sobind(V_udp6_tun_socket, 875 (struct sockaddr *)&sin6, curthread))) { 876 tcp_over_udp_stop(); 877 return (ret); 878 } 879 #endif 880 return (0); 881 } 882 883 static int 884 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 885 { 886 int error; 887 uint32_t old, new; 888 889 old = V_tcp_udp_tunneling_port; 890 new = old; 891 error = sysctl_handle_int(oidp, &new, 0, req); 892 if ((error == 0) && 893 (req->newptr != NULL)) { 894 if ((new < TCP_TUNNELING_PORT_MIN) || 895 (new > TCP_TUNNELING_PORT_MAX)) { 896 error = EINVAL; 897 } else { 898 V_tcp_udp_tunneling_port = new; 899 if (old != 0) { 900 tcp_over_udp_stop(); 901 } 902 if (new != 0) { 903 error = tcp_over_udp_start(); 904 } 905 } 906 } 907 return (error); 908 } 909 910 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 911 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 912 &VNET_NAME(tcp_udp_tunneling_port), 913 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 914 "Tunneling port for tcp over udp"); 915 916 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 917 918 static int 919 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 920 { 921 int error, new; 922 923 new = V_tcp_udp_tunneling_overhead; 924 error = sysctl_handle_int(oidp, &new, 0, req); 925 if (error == 0 && req->newptr) { 926 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 927 (new > TCP_TUNNELING_OVERHEAD_MAX)) 928 error = EINVAL; 929 else 930 V_tcp_udp_tunneling_overhead = new; 931 } 932 return (error); 933 } 934 935 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 936 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 937 &VNET_NAME(tcp_udp_tunneling_overhead), 938 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 939 "MSS reduction when using tcp over udp"); 940 941 /* 942 * Exports one (struct tcp_function_info) for each alias/name. 943 */ 944 static int 945 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 946 { 947 int cnt, error; 948 struct tcp_function *f; 949 struct tcp_function_info tfi; 950 951 /* 952 * We don't allow writes. 953 */ 954 if (req->newptr != NULL) 955 return (EINVAL); 956 957 /* 958 * Wire the old buffer so we can directly copy the functions to 959 * user space without dropping the lock. 960 */ 961 if (req->oldptr != NULL) { 962 error = sysctl_wire_old_buffer(req, 0); 963 if (error) 964 return (error); 965 } 966 967 /* 968 * Walk the list and copy out matching entries. If INVARIANTS 969 * is compiled in, also walk the list to verify the length of 970 * the list matches what we have recorded. 971 */ 972 rw_rlock(&tcp_function_lock); 973 974 cnt = 0; 975 #ifndef INVARIANTS 976 if (req->oldptr == NULL) { 977 cnt = tcp_fb_cnt; 978 goto skip_loop; 979 } 980 #endif 981 TAILQ_FOREACH(f, &t_functions, tf_next) { 982 #ifdef INVARIANTS 983 cnt++; 984 #endif 985 if (req->oldptr != NULL) { 986 bzero(&tfi, sizeof(tfi)); 987 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 988 tfi.tfi_id = f->tf_fb->tfb_id; 989 (void)strlcpy(tfi.tfi_alias, f->tf_name, 990 sizeof(tfi.tfi_alias)); 991 (void)strlcpy(tfi.tfi_name, 992 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 993 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 994 /* 995 * Don't stop on error, as that is the 996 * mechanism we use to accumulate length 997 * information if the buffer was too short. 998 */ 999 } 1000 } 1001 KASSERT(cnt == tcp_fb_cnt, 1002 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1003 #ifndef INVARIANTS 1004 skip_loop: 1005 #endif 1006 rw_runlock(&tcp_function_lock); 1007 if (req->oldptr == NULL) 1008 error = SYSCTL_OUT(req, NULL, 1009 (cnt + 1) * sizeof(struct tcp_function_info)); 1010 1011 return (error); 1012 } 1013 1014 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1015 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1016 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1017 "List TCP function block name-to-ID mappings"); 1018 1019 /* 1020 * tfb_tcp_handoff_ok() function for the default stack. 1021 * Note that we'll basically try to take all comers. 1022 */ 1023 static int 1024 tcp_default_handoff_ok(struct tcpcb *tp) 1025 { 1026 1027 return (0); 1028 } 1029 1030 /* 1031 * tfb_tcp_fb_init() function for the default stack. 1032 * 1033 * This handles making sure we have appropriate timers set if you are 1034 * transitioning a socket that has some amount of setup done. 1035 * 1036 * The init() fuction from the default can *never* return non-zero i.e. 1037 * it is required to always succeed since it is the stack of last resort! 1038 */ 1039 static int 1040 tcp_default_fb_init(struct tcpcb *tp) 1041 { 1042 1043 struct socket *so; 1044 1045 INP_WLOCK_ASSERT(tp->t_inpcb); 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 so = tp->t_inpcb->inp_socket; 1063 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1064 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1065 tcp_timer_active(tp, TT_PERSIST))) { 1066 /* 1067 * If the session has established and it looks like it should 1068 * be in the persist state, set the persist timer. Otherwise, 1069 * set the retransmit timer. 1070 */ 1071 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1072 (int32_t)(tp->snd_nxt - tp->snd_una) < 1073 (int32_t)sbavail(&so->so_snd)) 1074 tcp_setpersist(tp); 1075 else 1076 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1077 } 1078 1079 /* All non-embryonic sessions get a keepalive timer. */ 1080 if (!tcp_timer_active(tp, TT_KEEP)) 1081 tcp_timer_activate(tp, TT_KEEP, 1082 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1083 TP_KEEPINIT(tp)); 1084 1085 /* 1086 * Make sure critical variables are initialized 1087 * if transitioning while in Recovery. 1088 */ 1089 if IN_FASTRECOVERY(tp->t_flags) { 1090 if (tp->sackhint.recover_fs == 0) 1091 tp->sackhint.recover_fs = max(1, 1092 tp->snd_nxt - tp->snd_una); 1093 } 1094 1095 return (0); 1096 } 1097 1098 /* 1099 * tfb_tcp_fb_fini() function for the default stack. 1100 * 1101 * This changes state as necessary (or prudent) to prepare for another stack 1102 * to assume responsibility for the connection. 1103 */ 1104 static void 1105 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1106 { 1107 1108 INP_WLOCK_ASSERT(tp->t_inpcb); 1109 return; 1110 } 1111 1112 /* 1113 * Target size of TCP PCB hash tables. Must be a power of two. 1114 * 1115 * Note that this can be overridden by the kernel environment 1116 * variable net.inet.tcp.tcbhashsize 1117 */ 1118 #ifndef TCBHASHSIZE 1119 #define TCBHASHSIZE 0 1120 #endif 1121 1122 /* 1123 * XXX 1124 * Callouts should be moved into struct tcp directly. They are currently 1125 * separate because the tcpcb structure is exported to userland for sysctl 1126 * parsing purposes, which do not know about callouts. 1127 */ 1128 struct tcpcb_mem { 1129 struct tcpcb tcb; 1130 struct tcp_timer tt; 1131 struct cc_var ccv; 1132 #ifdef TCP_HHOOK 1133 struct osd osd; 1134 #endif 1135 }; 1136 1137 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 1138 #define V_tcpcb_zone VNET(tcpcb_zone) 1139 1140 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1141 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1142 1143 static struct mtx isn_mtx; 1144 1145 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1146 #define ISN_LOCK() mtx_lock(&isn_mtx) 1147 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1148 1149 INPCBSTORAGE_DEFINE(tcpcbstor, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1150 1151 /* 1152 * Take a value and get the next power of 2 that doesn't overflow. 1153 * Used to size the tcp_inpcb hash buckets. 1154 */ 1155 static int 1156 maketcp_hashsize(int size) 1157 { 1158 int hashsize; 1159 1160 /* 1161 * auto tune. 1162 * get the next power of 2 higher than maxsockets. 1163 */ 1164 hashsize = 1 << fls(size); 1165 /* catch overflow, and just go one power of 2 smaller */ 1166 if (hashsize < size) { 1167 hashsize = 1 << (fls(size) - 1); 1168 } 1169 return (hashsize); 1170 } 1171 1172 static volatile int next_tcp_stack_id = 1; 1173 1174 /* 1175 * Register a TCP function block with the name provided in the names 1176 * array. (Note that this function does NOT automatically register 1177 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1178 * explicitly include blk->tfb_tcp_block_name in the list of names if 1179 * you wish to register the stack with that name.) 1180 * 1181 * Either all name registrations will succeed or all will fail. If 1182 * a name registration fails, the function will update the num_names 1183 * argument to point to the array index of the name that encountered 1184 * the failure. 1185 * 1186 * Returns 0 on success, or an error code on failure. 1187 */ 1188 int 1189 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1190 const char *names[], int *num_names) 1191 { 1192 struct tcp_function *n; 1193 struct tcp_function_set fs; 1194 int error, i; 1195 1196 KASSERT(names != NULL && *num_names > 0, 1197 ("%s: Called with 0-length name list", __func__)); 1198 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1199 KASSERT(rw_initialized(&tcp_function_lock), 1200 ("%s: called too early", __func__)); 1201 1202 if ((blk->tfb_tcp_output == NULL) || 1203 (blk->tfb_tcp_do_segment == NULL) || 1204 (blk->tfb_tcp_ctloutput == NULL) || 1205 (strlen(blk->tfb_tcp_block_name) == 0)) { 1206 /* 1207 * These functions are required and you 1208 * need a name. 1209 */ 1210 *num_names = 0; 1211 return (EINVAL); 1212 } 1213 if (blk->tfb_tcp_timer_stop_all || 1214 blk->tfb_tcp_timer_activate || 1215 blk->tfb_tcp_timer_active || 1216 blk->tfb_tcp_timer_stop) { 1217 /* 1218 * If you define one timer function you 1219 * must have them all. 1220 */ 1221 if ((blk->tfb_tcp_timer_stop_all == NULL) || 1222 (blk->tfb_tcp_timer_activate == NULL) || 1223 (blk->tfb_tcp_timer_active == NULL) || 1224 (blk->tfb_tcp_timer_stop == NULL)) { 1225 *num_names = 0; 1226 return (EINVAL); 1227 } 1228 } 1229 1230 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1231 *num_names = 0; 1232 return (EINVAL); 1233 } 1234 1235 refcount_init(&blk->tfb_refcnt, 0); 1236 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1237 for (i = 0; i < *num_names; i++) { 1238 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1239 if (n == NULL) { 1240 error = ENOMEM; 1241 goto cleanup; 1242 } 1243 n->tf_fb = blk; 1244 1245 (void)strlcpy(fs.function_set_name, names[i], 1246 sizeof(fs.function_set_name)); 1247 rw_wlock(&tcp_function_lock); 1248 if (find_tcp_functions_locked(&fs) != NULL) { 1249 /* Duplicate name space not allowed */ 1250 rw_wunlock(&tcp_function_lock); 1251 free(n, M_TCPFUNCTIONS); 1252 error = EALREADY; 1253 goto cleanup; 1254 } 1255 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1256 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1257 tcp_fb_cnt++; 1258 rw_wunlock(&tcp_function_lock); 1259 } 1260 return(0); 1261 1262 cleanup: 1263 /* 1264 * Deregister the names we just added. Because registration failed 1265 * for names[i], we don't need to deregister that name. 1266 */ 1267 *num_names = i; 1268 rw_wlock(&tcp_function_lock); 1269 while (--i >= 0) { 1270 TAILQ_FOREACH(n, &t_functions, tf_next) { 1271 if (!strncmp(n->tf_name, names[i], 1272 TCP_FUNCTION_NAME_LEN_MAX)) { 1273 TAILQ_REMOVE(&t_functions, n, tf_next); 1274 tcp_fb_cnt--; 1275 n->tf_fb = NULL; 1276 free(n, M_TCPFUNCTIONS); 1277 break; 1278 } 1279 } 1280 } 1281 rw_wunlock(&tcp_function_lock); 1282 return (error); 1283 } 1284 1285 /* 1286 * Register a TCP function block using the name provided in the name 1287 * argument. 1288 * 1289 * Returns 0 on success, or an error code on failure. 1290 */ 1291 int 1292 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1293 int wait) 1294 { 1295 const char *name_list[1]; 1296 int num_names, rv; 1297 1298 num_names = 1; 1299 if (name != NULL) 1300 name_list[0] = name; 1301 else 1302 name_list[0] = blk->tfb_tcp_block_name; 1303 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1304 return (rv); 1305 } 1306 1307 /* 1308 * Register a TCP function block using the name defined in 1309 * blk->tfb_tcp_block_name. 1310 * 1311 * Returns 0 on success, or an error code on failure. 1312 */ 1313 int 1314 register_tcp_functions(struct tcp_function_block *blk, int wait) 1315 { 1316 1317 return (register_tcp_functions_as_name(blk, NULL, wait)); 1318 } 1319 1320 /* 1321 * Deregister all names associated with a function block. This 1322 * functionally removes the function block from use within the system. 1323 * 1324 * When called with a true quiesce argument, mark the function block 1325 * as being removed so no more stacks will use it and determine 1326 * whether the removal would succeed. 1327 * 1328 * When called with a false quiesce argument, actually attempt the 1329 * removal. 1330 * 1331 * When called with a force argument, attempt to switch all TCBs to 1332 * use the default stack instead of returning EBUSY. 1333 * 1334 * Returns 0 on success (or if the removal would succeed, or an error 1335 * code on failure. 1336 */ 1337 int 1338 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1339 bool force) 1340 { 1341 struct tcp_function *f; 1342 1343 if (blk == &tcp_def_funcblk) { 1344 /* You can't un-register the default */ 1345 return (EPERM); 1346 } 1347 rw_wlock(&tcp_function_lock); 1348 if (blk == tcp_func_set_ptr) { 1349 /* You can't free the current default */ 1350 rw_wunlock(&tcp_function_lock); 1351 return (EBUSY); 1352 } 1353 /* Mark the block so no more stacks can use it. */ 1354 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1355 /* 1356 * If TCBs are still attached to the stack, attempt to switch them 1357 * to the default stack. 1358 */ 1359 if (force && blk->tfb_refcnt) { 1360 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1361 INPLOOKUP_WLOCKPCB); 1362 struct inpcb *inp; 1363 struct tcpcb *tp; 1364 VNET_ITERATOR_DECL(vnet_iter); 1365 1366 rw_wunlock(&tcp_function_lock); 1367 1368 VNET_LIST_RLOCK(); 1369 VNET_FOREACH(vnet_iter) { 1370 CURVNET_SET(vnet_iter); 1371 while ((inp = inp_next(&inpi)) != NULL) { 1372 if (inp->inp_flags & INP_TIMEWAIT) 1373 continue; 1374 tp = intotcpcb(inp); 1375 if (tp == NULL || tp->t_fb != blk) 1376 continue; 1377 tcp_switch_back_to_default(tp); 1378 } 1379 CURVNET_RESTORE(); 1380 } 1381 VNET_LIST_RUNLOCK(); 1382 1383 rw_wlock(&tcp_function_lock); 1384 } 1385 if (blk->tfb_refcnt) { 1386 /* TCBs still attached. */ 1387 rw_wunlock(&tcp_function_lock); 1388 return (EBUSY); 1389 } 1390 if (quiesce) { 1391 /* Skip removal. */ 1392 rw_wunlock(&tcp_function_lock); 1393 return (0); 1394 } 1395 /* Remove any function names that map to this function block. */ 1396 while (find_tcp_fb_locked(blk, &f) != NULL) { 1397 TAILQ_REMOVE(&t_functions, f, tf_next); 1398 tcp_fb_cnt--; 1399 f->tf_fb = NULL; 1400 free(f, M_TCPFUNCTIONS); 1401 } 1402 rw_wunlock(&tcp_function_lock); 1403 return (0); 1404 } 1405 1406 static void 1407 tcp_vnet_init(void *arg __unused) 1408 { 1409 1410 #ifdef TCP_HHOOK 1411 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1412 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1413 printf("%s: WARNING: unable to register helper hook\n", __func__); 1414 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1415 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1416 printf("%s: WARNING: unable to register helper hook\n", __func__); 1417 #endif 1418 #ifdef STATS 1419 if (tcp_stats_init()) 1420 printf("%s: WARNING: unable to initialise TCP stats\n", 1421 __func__); 1422 #endif 1423 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1424 tcp_tcbhashsize); 1425 1426 /* 1427 * These have to be type stable for the benefit of the timers. 1428 */ 1429 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1430 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1431 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1432 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1433 1434 tcp_tw_init(); 1435 syncache_init(); 1436 tcp_hc_init(); 1437 1438 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1439 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1440 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1441 1442 tcp_fastopen_init(); 1443 1444 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1445 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1446 1447 V_tcp_msl = TCPTV_MSL; 1448 } 1449 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1450 tcp_vnet_init, NULL); 1451 1452 static void 1453 tcp_init(void *arg __unused) 1454 { 1455 const char *tcbhash_tuneable; 1456 int hashsize; 1457 1458 tcp_reass_global_init(); 1459 1460 /* XXX virtualize those below? */ 1461 tcp_delacktime = TCPTV_DELACK; 1462 tcp_keepinit = TCPTV_KEEP_INIT; 1463 tcp_keepidle = TCPTV_KEEP_IDLE; 1464 tcp_keepintvl = TCPTV_KEEPINTVL; 1465 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1466 tcp_rexmit_initial = TCPTV_RTOBASE; 1467 if (tcp_rexmit_initial < 1) 1468 tcp_rexmit_initial = 1; 1469 tcp_rexmit_min = TCPTV_MIN; 1470 if (tcp_rexmit_min < 1) 1471 tcp_rexmit_min = 1; 1472 tcp_persmin = TCPTV_PERSMIN; 1473 tcp_persmax = TCPTV_PERSMAX; 1474 tcp_rexmit_slop = TCPTV_CPU_VAR; 1475 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1476 1477 /* Setup the tcp function block list */ 1478 TAILQ_INIT(&t_functions); 1479 rw_init(&tcp_function_lock, "tcp_func_lock"); 1480 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1481 #ifdef TCP_BLACKBOX 1482 /* Initialize the TCP logging data. */ 1483 tcp_log_init(); 1484 #endif 1485 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1486 1487 if (tcp_soreceive_stream) { 1488 #ifdef INET 1489 tcp_usrreqs.pru_soreceive = soreceive_stream; 1490 #endif 1491 #ifdef INET6 1492 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1493 #endif /* INET6 */ 1494 } 1495 1496 #ifdef INET6 1497 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1498 #else /* INET6 */ 1499 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1500 #endif /* INET6 */ 1501 if (max_protohdr < TCP_MINPROTOHDR) 1502 max_protohdr = TCP_MINPROTOHDR; 1503 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1504 panic("tcp_init"); 1505 #undef TCP_MINPROTOHDR 1506 1507 ISN_LOCK_INIT(); 1508 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1509 SHUTDOWN_PRI_DEFAULT); 1510 1511 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1512 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1513 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1514 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1515 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1516 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1517 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1518 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1519 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1520 #ifdef TCPPCAP 1521 tcp_pcap_init(); 1522 #endif 1523 1524 hashsize = TCBHASHSIZE; 1525 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1526 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1527 if (hashsize == 0) { 1528 /* 1529 * Auto tune the hash size based on maxsockets. 1530 * A perfect hash would have a 1:1 mapping 1531 * (hashsize = maxsockets) however it's been 1532 * suggested that O(2) average is better. 1533 */ 1534 hashsize = maketcp_hashsize(maxsockets / 4); 1535 /* 1536 * Our historical default is 512, 1537 * do not autotune lower than this. 1538 */ 1539 if (hashsize < 512) 1540 hashsize = 512; 1541 if (bootverbose) 1542 printf("%s: %s auto tuned to %d\n", __func__, 1543 tcbhash_tuneable, hashsize); 1544 } 1545 /* 1546 * We require a hashsize to be a power of two. 1547 * Previously if it was not a power of two we would just reset it 1548 * back to 512, which could be a nasty surprise if you did not notice 1549 * the error message. 1550 * Instead what we do is clip it to the closest power of two lower 1551 * than the specified hash value. 1552 */ 1553 if (!powerof2(hashsize)) { 1554 int oldhashsize = hashsize; 1555 1556 hashsize = maketcp_hashsize(hashsize); 1557 /* prevent absurdly low value */ 1558 if (hashsize < 16) 1559 hashsize = 16; 1560 printf("%s: WARNING: TCB hash size not a power of 2, " 1561 "clipped from %d to %d.\n", __func__, oldhashsize, 1562 hashsize); 1563 } 1564 tcp_tcbhashsize = hashsize; 1565 } 1566 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1567 1568 #ifdef VIMAGE 1569 static void 1570 tcp_destroy(void *unused __unused) 1571 { 1572 int n; 1573 #ifdef TCP_HHOOK 1574 int error; 1575 #endif 1576 1577 /* 1578 * All our processes are gone, all our sockets should be cleaned 1579 * up, which means, we should be past the tcp_discardcb() calls. 1580 * Sleep to let all tcpcb timers really disappear and cleanup. 1581 */ 1582 for (;;) { 1583 INP_INFO_WLOCK(&V_tcbinfo); 1584 n = V_tcbinfo.ipi_count; 1585 INP_INFO_WUNLOCK(&V_tcbinfo); 1586 if (n == 0) 1587 break; 1588 pause("tcpdes", hz / 10); 1589 } 1590 tcp_hc_destroy(); 1591 syncache_destroy(); 1592 tcp_tw_destroy(); 1593 in_pcbinfo_destroy(&V_tcbinfo); 1594 /* tcp_discardcb() clears the sack_holes up. */ 1595 uma_zdestroy(V_sack_hole_zone); 1596 uma_zdestroy(V_tcpcb_zone); 1597 1598 /* 1599 * Cannot free the zone until all tcpcbs are released as we attach 1600 * the allocations to them. 1601 */ 1602 tcp_fastopen_destroy(); 1603 1604 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1605 VNET_PCPUSTAT_FREE(tcpstat); 1606 1607 #ifdef TCP_HHOOK 1608 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1609 if (error != 0) { 1610 printf("%s: WARNING: unable to deregister helper hook " 1611 "type=%d, id=%d: error %d returned\n", __func__, 1612 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1613 } 1614 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1615 if (error != 0) { 1616 printf("%s: WARNING: unable to deregister helper hook " 1617 "type=%d, id=%d: error %d returned\n", __func__, 1618 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1619 } 1620 #endif 1621 } 1622 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1623 #endif 1624 1625 void 1626 tcp_fini(void *xtp) 1627 { 1628 1629 } 1630 1631 /* 1632 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1633 * tcp_template used to store this data in mbufs, but we now recopy it out 1634 * of the tcpcb each time to conserve mbufs. 1635 */ 1636 void 1637 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1638 { 1639 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1640 1641 INP_WLOCK_ASSERT(inp); 1642 1643 #ifdef INET6 1644 if ((inp->inp_vflag & INP_IPV6) != 0) { 1645 struct ip6_hdr *ip6; 1646 1647 ip6 = (struct ip6_hdr *)ip_ptr; 1648 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1649 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1650 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1651 (IPV6_VERSION & IPV6_VERSION_MASK); 1652 if (port == 0) 1653 ip6->ip6_nxt = IPPROTO_TCP; 1654 else 1655 ip6->ip6_nxt = IPPROTO_UDP; 1656 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1657 ip6->ip6_src = inp->in6p_laddr; 1658 ip6->ip6_dst = inp->in6p_faddr; 1659 } 1660 #endif /* INET6 */ 1661 #if defined(INET6) && defined(INET) 1662 else 1663 #endif 1664 #ifdef INET 1665 { 1666 struct ip *ip; 1667 1668 ip = (struct ip *)ip_ptr; 1669 ip->ip_v = IPVERSION; 1670 ip->ip_hl = 5; 1671 ip->ip_tos = inp->inp_ip_tos; 1672 ip->ip_len = 0; 1673 ip->ip_id = 0; 1674 ip->ip_off = 0; 1675 ip->ip_ttl = inp->inp_ip_ttl; 1676 ip->ip_sum = 0; 1677 if (port == 0) 1678 ip->ip_p = IPPROTO_TCP; 1679 else 1680 ip->ip_p = IPPROTO_UDP; 1681 ip->ip_src = inp->inp_laddr; 1682 ip->ip_dst = inp->inp_faddr; 1683 } 1684 #endif /* INET */ 1685 th->th_sport = inp->inp_lport; 1686 th->th_dport = inp->inp_fport; 1687 th->th_seq = 0; 1688 th->th_ack = 0; 1689 th->th_off = 5; 1690 tcp_set_flags(th, 0); 1691 th->th_win = 0; 1692 th->th_urp = 0; 1693 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1694 } 1695 1696 /* 1697 * Create template to be used to send tcp packets on a connection. 1698 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1699 * use for this function is in keepalives, which use tcp_respond. 1700 */ 1701 struct tcptemp * 1702 tcpip_maketemplate(struct inpcb *inp) 1703 { 1704 struct tcptemp *t; 1705 1706 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1707 if (t == NULL) 1708 return (NULL); 1709 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1710 return (t); 1711 } 1712 1713 /* 1714 * Send a single message to the TCP at address specified by 1715 * the given TCP/IP header. If m == NULL, then we make a copy 1716 * of the tcpiphdr at th and send directly to the addressed host. 1717 * This is used to force keep alive messages out using the TCP 1718 * template for a connection. If flags are given then we send 1719 * a message back to the TCP which originated the segment th, 1720 * and discard the mbuf containing it and any other attached mbufs. 1721 * 1722 * In any case the ack and sequence number of the transmitted 1723 * segment are as specified by the parameters. 1724 * 1725 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1726 */ 1727 void 1728 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1729 tcp_seq ack, tcp_seq seq, int flags) 1730 { 1731 struct tcpopt to; 1732 struct inpcb *inp; 1733 struct ip *ip; 1734 struct mbuf *optm; 1735 struct udphdr *uh = NULL; 1736 struct tcphdr *nth; 1737 struct tcp_log_buffer *lgb; 1738 u_char *optp; 1739 #ifdef INET6 1740 struct ip6_hdr *ip6; 1741 int isipv6; 1742 #endif /* INET6 */ 1743 int optlen, tlen, win, ulen; 1744 bool incl_opts; 1745 uint16_t port; 1746 int output_ret; 1747 #ifdef INVARIANTS 1748 int thflags = tcp_get_flags(th); 1749 #endif 1750 1751 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1752 NET_EPOCH_ASSERT(); 1753 1754 #ifdef INET6 1755 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1756 ip6 = ipgen; 1757 #endif /* INET6 */ 1758 ip = ipgen; 1759 1760 if (tp != NULL) { 1761 inp = tp->t_inpcb; 1762 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1763 INP_LOCK_ASSERT(inp); 1764 } else 1765 inp = NULL; 1766 1767 if (m != NULL) { 1768 #ifdef INET6 1769 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1770 port = m->m_pkthdr.tcp_tun_port; 1771 else 1772 #endif 1773 if (ip && (ip->ip_p == IPPROTO_UDP)) 1774 port = m->m_pkthdr.tcp_tun_port; 1775 else 1776 port = 0; 1777 } else 1778 port = tp->t_port; 1779 1780 incl_opts = false; 1781 win = 0; 1782 if (tp != NULL) { 1783 if (!(flags & TH_RST)) { 1784 win = sbspace(&inp->inp_socket->so_rcv); 1785 if (win > TCP_MAXWIN << tp->rcv_scale) 1786 win = TCP_MAXWIN << tp->rcv_scale; 1787 } 1788 if ((tp->t_flags & TF_NOOPT) == 0) 1789 incl_opts = true; 1790 } 1791 if (m == NULL) { 1792 m = m_gethdr(M_NOWAIT, MT_DATA); 1793 if (m == NULL) 1794 return; 1795 m->m_data += max_linkhdr; 1796 #ifdef INET6 1797 if (isipv6) { 1798 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1799 sizeof(struct ip6_hdr)); 1800 ip6 = mtod(m, struct ip6_hdr *); 1801 nth = (struct tcphdr *)(ip6 + 1); 1802 if (port) { 1803 /* Insert a UDP header */ 1804 uh = (struct udphdr *)nth; 1805 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1806 uh->uh_dport = port; 1807 nth = (struct tcphdr *)(uh + 1); 1808 } 1809 } else 1810 #endif /* INET6 */ 1811 { 1812 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1813 ip = mtod(m, struct ip *); 1814 nth = (struct tcphdr *)(ip + 1); 1815 if (port) { 1816 /* Insert a UDP header */ 1817 uh = (struct udphdr *)nth; 1818 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1819 uh->uh_dport = port; 1820 nth = (struct tcphdr *)(uh + 1); 1821 } 1822 } 1823 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1824 flags = TH_ACK; 1825 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1826 struct mbuf *n; 1827 1828 /* Can't reuse 'm', allocate a new mbuf. */ 1829 n = m_gethdr(M_NOWAIT, MT_DATA); 1830 if (n == NULL) { 1831 m_freem(m); 1832 return; 1833 } 1834 1835 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1836 m_freem(m); 1837 m_freem(n); 1838 return; 1839 } 1840 1841 n->m_data += max_linkhdr; 1842 /* m_len is set later */ 1843 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1844 #ifdef INET6 1845 if (isipv6) { 1846 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1847 sizeof(struct ip6_hdr)); 1848 ip6 = mtod(n, struct ip6_hdr *); 1849 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1850 nth = (struct tcphdr *)(ip6 + 1); 1851 if (port) { 1852 /* Insert a UDP header */ 1853 uh = (struct udphdr *)nth; 1854 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1855 uh->uh_dport = port; 1856 nth = (struct tcphdr *)(uh + 1); 1857 } 1858 } else 1859 #endif /* INET6 */ 1860 { 1861 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1862 ip = mtod(n, struct ip *); 1863 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1864 nth = (struct tcphdr *)(ip + 1); 1865 if (port) { 1866 /* Insert a UDP header */ 1867 uh = (struct udphdr *)nth; 1868 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1869 uh->uh_dport = port; 1870 nth = (struct tcphdr *)(uh + 1); 1871 } 1872 } 1873 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1874 xchg(nth->th_dport, nth->th_sport, uint16_t); 1875 th = nth; 1876 m_freem(m); 1877 m = n; 1878 } else { 1879 /* 1880 * reuse the mbuf. 1881 * XXX MRT We inherit the FIB, which is lucky. 1882 */ 1883 m_freem(m->m_next); 1884 m->m_next = NULL; 1885 m->m_data = (caddr_t)ipgen; 1886 /* m_len is set later */ 1887 #ifdef INET6 1888 if (isipv6) { 1889 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1890 nth = (struct tcphdr *)(ip6 + 1); 1891 } else 1892 #endif /* INET6 */ 1893 { 1894 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1895 nth = (struct tcphdr *)(ip + 1); 1896 } 1897 if (th != nth) { 1898 /* 1899 * this is usually a case when an extension header 1900 * exists between the IPv6 header and the 1901 * TCP header. 1902 */ 1903 nth->th_sport = th->th_sport; 1904 nth->th_dport = th->th_dport; 1905 } 1906 xchg(nth->th_dport, nth->th_sport, uint16_t); 1907 #undef xchg 1908 } 1909 tlen = 0; 1910 #ifdef INET6 1911 if (isipv6) 1912 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1913 #endif 1914 #if defined(INET) && defined(INET6) 1915 else 1916 #endif 1917 #ifdef INET 1918 tlen = sizeof (struct tcpiphdr); 1919 #endif 1920 if (port) 1921 tlen += sizeof (struct udphdr); 1922 #ifdef INVARIANTS 1923 m->m_len = 0; 1924 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1925 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1926 m, tlen, (long)M_TRAILINGSPACE(m))); 1927 #endif 1928 m->m_len = tlen; 1929 to.to_flags = 0; 1930 if (incl_opts) { 1931 /* Make sure we have room. */ 1932 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1933 m->m_next = m_get(M_NOWAIT, MT_DATA); 1934 if (m->m_next) { 1935 optp = mtod(m->m_next, u_char *); 1936 optm = m->m_next; 1937 } else 1938 incl_opts = false; 1939 } else { 1940 optp = (u_char *) (nth + 1); 1941 optm = m; 1942 } 1943 } 1944 if (incl_opts) { 1945 /* Timestamps. */ 1946 if (tp->t_flags & TF_RCVD_TSTMP) { 1947 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1948 to.to_tsecr = tp->ts_recent; 1949 to.to_flags |= TOF_TS; 1950 } 1951 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1952 /* TCP-MD5 (RFC2385). */ 1953 if (tp->t_flags & TF_SIGNATURE) 1954 to.to_flags |= TOF_SIGNATURE; 1955 #endif 1956 /* Add the options. */ 1957 tlen += optlen = tcp_addoptions(&to, optp); 1958 1959 /* Update m_len in the correct mbuf. */ 1960 optm->m_len += optlen; 1961 } else 1962 optlen = 0; 1963 #ifdef INET6 1964 if (isipv6) { 1965 if (uh) { 1966 ulen = tlen - sizeof(struct ip6_hdr); 1967 uh->uh_ulen = htons(ulen); 1968 } 1969 ip6->ip6_flow = 0; 1970 ip6->ip6_vfc = IPV6_VERSION; 1971 if (port) 1972 ip6->ip6_nxt = IPPROTO_UDP; 1973 else 1974 ip6->ip6_nxt = IPPROTO_TCP; 1975 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1976 } 1977 #endif 1978 #if defined(INET) && defined(INET6) 1979 else 1980 #endif 1981 #ifdef INET 1982 { 1983 if (uh) { 1984 ulen = tlen - sizeof(struct ip); 1985 uh->uh_ulen = htons(ulen); 1986 } 1987 ip->ip_len = htons(tlen); 1988 ip->ip_ttl = V_ip_defttl; 1989 if (port) { 1990 ip->ip_p = IPPROTO_UDP; 1991 } else { 1992 ip->ip_p = IPPROTO_TCP; 1993 } 1994 if (V_path_mtu_discovery) 1995 ip->ip_off |= htons(IP_DF); 1996 } 1997 #endif 1998 m->m_pkthdr.len = tlen; 1999 m->m_pkthdr.rcvif = NULL; 2000 #ifdef MAC 2001 if (inp != NULL) { 2002 /* 2003 * Packet is associated with a socket, so allow the 2004 * label of the response to reflect the socket label. 2005 */ 2006 INP_LOCK_ASSERT(inp); 2007 mac_inpcb_create_mbuf(inp, m); 2008 } else { 2009 /* 2010 * Packet is not associated with a socket, so possibly 2011 * update the label in place. 2012 */ 2013 mac_netinet_tcp_reply(m); 2014 } 2015 #endif 2016 nth->th_seq = htonl(seq); 2017 nth->th_ack = htonl(ack); 2018 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2019 tcp_set_flags(nth, flags); 2020 if (tp != NULL) 2021 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2022 else 2023 nth->th_win = htons((u_short)win); 2024 nth->th_urp = 0; 2025 2026 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2027 if (to.to_flags & TOF_SIGNATURE) { 2028 if (!TCPMD5_ENABLED() || 2029 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2030 m_freem(m); 2031 return; 2032 } 2033 } 2034 #endif 2035 2036 #ifdef INET6 2037 if (isipv6) { 2038 if (port) { 2039 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2040 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2041 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2042 nth->th_sum = 0; 2043 } else { 2044 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2045 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2046 nth->th_sum = in6_cksum_pseudo(ip6, 2047 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2048 } 2049 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 2050 NULL, NULL); 2051 } 2052 #endif /* INET6 */ 2053 #if defined(INET6) && defined(INET) 2054 else 2055 #endif 2056 #ifdef INET 2057 { 2058 if (port) { 2059 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2060 htons(ulen + IPPROTO_UDP)); 2061 m->m_pkthdr.csum_flags = CSUM_UDP; 2062 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2063 nth->th_sum = 0; 2064 } else { 2065 m->m_pkthdr.csum_flags = CSUM_TCP; 2066 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2067 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2068 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2069 } 2070 } 2071 #endif /* INET */ 2072 #ifdef TCPDEBUG 2073 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2074 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2075 #endif 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 = tp->t_inpcb->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 #ifdef INET6 2125 if (isipv6) { 2126 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2127 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2128 } 2129 #endif /* INET6 */ 2130 #if defined(INET) && defined(INET6) 2131 else 2132 #endif 2133 #ifdef INET 2134 { 2135 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2136 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2137 } 2138 #endif 2139 if (lgb != NULL) 2140 lgb->tlb_errno = output_ret; 2141 } 2142 2143 /* 2144 * Create a new TCP control block, making an 2145 * empty reassembly queue and hooking it to the argument 2146 * protocol control block. The `inp' parameter must have 2147 * come from the zone allocator set up in tcp_init(). 2148 */ 2149 struct tcpcb * 2150 tcp_newtcpcb(struct inpcb *inp) 2151 { 2152 struct tcpcb_mem *tm; 2153 struct tcpcb *tp; 2154 #ifdef INET6 2155 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2156 #endif /* INET6 */ 2157 2158 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 2159 if (tm == NULL) 2160 return (NULL); 2161 tp = &tm->tcb; 2162 2163 /* Initialise cc_var struct for this tcpcb. */ 2164 tp->ccv = &tm->ccv; 2165 tp->ccv->type = IPPROTO_TCP; 2166 tp->ccv->ccvc.tcp = tp; 2167 rw_rlock(&tcp_function_lock); 2168 tp->t_fb = tcp_func_set_ptr; 2169 refcount_acquire(&tp->t_fb->tfb_refcnt); 2170 rw_runlock(&tcp_function_lock); 2171 /* 2172 * Use the current system default CC algorithm. 2173 */ 2174 CC_LIST_RLOCK(); 2175 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 2176 CC_ALGO(tp) = CC_DEFAULT_ALGO(); 2177 CC_LIST_RUNLOCK(); 2178 2179 /* 2180 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 2181 * is called. 2182 */ 2183 in_pcbref(inp); /* Reference for tcpcb */ 2184 tp->t_inpcb = inp; 2185 2186 if (CC_ALGO(tp)->cb_init != NULL) 2187 if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) { 2188 if (tp->t_fb->tfb_tcp_fb_fini) 2189 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2190 in_pcbrele_wlocked(inp); 2191 refcount_release(&tp->t_fb->tfb_refcnt); 2192 uma_zfree(V_tcpcb_zone, tm); 2193 return (NULL); 2194 } 2195 2196 #ifdef TCP_HHOOK 2197 tp->osd = &tm->osd; 2198 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 2199 if (tp->t_fb->tfb_tcp_fb_fini) 2200 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2201 in_pcbrele_wlocked(inp); 2202 refcount_release(&tp->t_fb->tfb_refcnt); 2203 uma_zfree(V_tcpcb_zone, tm); 2204 return (NULL); 2205 } 2206 #endif 2207 2208 #ifdef VIMAGE 2209 tp->t_vnet = inp->inp_vnet; 2210 #endif 2211 tp->t_timers = &tm->tt; 2212 TAILQ_INIT(&tp->t_segq); 2213 tp->t_maxseg = 2214 #ifdef INET6 2215 isipv6 ? V_tcp_v6mssdflt : 2216 #endif /* INET6 */ 2217 V_tcp_mssdflt; 2218 2219 /* Set up our timeouts. */ 2220 callout_init(&tp->t_timers->tt_rexmt, 1); 2221 callout_init(&tp->t_timers->tt_persist, 1); 2222 callout_init(&tp->t_timers->tt_keep, 1); 2223 callout_init(&tp->t_timers->tt_2msl, 1); 2224 callout_init(&tp->t_timers->tt_delack, 1); 2225 2226 if (V_tcp_do_rfc1323) 2227 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2228 if (V_tcp_do_sack) 2229 tp->t_flags |= TF_SACK_PERMIT; 2230 TAILQ_INIT(&tp->snd_holes); 2231 2232 /* 2233 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2234 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2235 * reasonable initial retransmit time. 2236 */ 2237 tp->t_srtt = TCPTV_SRTTBASE; 2238 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2239 tp->t_rttmin = tcp_rexmit_min; 2240 tp->t_rxtcur = tcp_rexmit_initial; 2241 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2242 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2243 tp->t_rcvtime = ticks; 2244 /* 2245 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2246 * because the socket may be bound to an IPv6 wildcard address, 2247 * which may match an IPv4-mapped IPv6 address. 2248 */ 2249 inp->inp_ip_ttl = V_ip_defttl; 2250 inp->inp_ppcb = tp; 2251 #ifdef TCPPCAP 2252 /* 2253 * Init the TCP PCAP queues. 2254 */ 2255 tcp_pcap_tcpcb_init(tp); 2256 #endif 2257 #ifdef TCP_BLACKBOX 2258 /* Initialize the per-TCPCB log data. */ 2259 tcp_log_tcpcbinit(tp); 2260 #endif 2261 tp->t_pacing_rate = -1; 2262 if (tp->t_fb->tfb_tcp_fb_init) { 2263 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2264 refcount_release(&tp->t_fb->tfb_refcnt); 2265 in_pcbrele_wlocked(inp); 2266 uma_zfree(V_tcpcb_zone, tm); 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 return (tp); /* XXX */ 2277 } 2278 2279 /* 2280 * Switch the congestion control algorithm back to Vnet default for any active 2281 * control blocks using an algorithm which is about to go away. If the algorithm 2282 * has a cb_init function and it fails (no memory) then the operation fails and 2283 * the unload will not succeed. 2284 * 2285 */ 2286 int 2287 tcp_ccalgounload(struct cc_algo *unload_algo) 2288 { 2289 struct cc_algo *oldalgo, *newalgo; 2290 struct inpcb *inp; 2291 struct tcpcb *tp; 2292 VNET_ITERATOR_DECL(vnet_iter); 2293 2294 /* 2295 * Check all active control blocks across all network stacks and change 2296 * any that are using "unload_algo" back to its default. If "unload_algo" 2297 * requires cleanup code to be run, call it. 2298 */ 2299 VNET_LIST_RLOCK(); 2300 VNET_FOREACH(vnet_iter) { 2301 CURVNET_SET(vnet_iter); 2302 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2303 INPLOOKUP_WLOCKPCB); 2304 /* 2305 * XXXGL: would new accept(2)d connections use algo being 2306 * unloaded? 2307 */ 2308 newalgo = CC_DEFAULT_ALGO(); 2309 while ((inp = inp_next(&inpi)) != NULL) { 2310 /* Important to skip tcptw structs. */ 2311 if (!(inp->inp_flags & INP_TIMEWAIT) && 2312 (tp = intotcpcb(inp)) != NULL) { 2313 /* 2314 * By holding INP_WLOCK here, we are assured 2315 * that the connection is not currently 2316 * executing inside the CC module's functions. 2317 * We attempt to switch to the Vnets default, 2318 * if the init fails then we fail the whole 2319 * operation and the module unload will fail. 2320 */ 2321 if (CC_ALGO(tp) == unload_algo) { 2322 struct cc_var cc_mem; 2323 int err; 2324 2325 oldalgo = CC_ALGO(tp); 2326 memset(&cc_mem, 0, sizeof(cc_mem)); 2327 cc_mem.ccvc.tcp = tp; 2328 if (newalgo->cb_init == NULL) { 2329 /* 2330 * No init we can skip the 2331 * dance around a possible failure. 2332 */ 2333 CC_DATA(tp) = NULL; 2334 goto proceed; 2335 } 2336 err = (newalgo->cb_init)(&cc_mem, NULL); 2337 if (err) { 2338 /* 2339 * Presumably no memory the caller will 2340 * need to try again. 2341 */ 2342 INP_WUNLOCK(inp); 2343 CURVNET_RESTORE(); 2344 VNET_LIST_RUNLOCK(); 2345 return (err); 2346 } 2347 proceed: 2348 if (oldalgo->cb_destroy != NULL) 2349 oldalgo->cb_destroy(tp->ccv); 2350 CC_ALGO(tp) = newalgo; 2351 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2352 if (TCPS_HAVEESTABLISHED(tp->t_state) && 2353 (CC_ALGO(tp)->conn_init != NULL)) { 2354 /* Yep run the connection init for the new CC */ 2355 CC_ALGO(tp)->conn_init(tp->ccv); 2356 } 2357 } 2358 } 2359 } 2360 CURVNET_RESTORE(); 2361 } 2362 VNET_LIST_RUNLOCK(); 2363 return (0); 2364 } 2365 2366 /* 2367 * Drop a TCP connection, reporting 2368 * the specified error. If connection is synchronized, 2369 * then send a RST to peer. 2370 */ 2371 struct tcpcb * 2372 tcp_drop(struct tcpcb *tp, int errno) 2373 { 2374 struct socket *so = tp->t_inpcb->inp_socket; 2375 2376 NET_EPOCH_ASSERT(); 2377 INP_WLOCK_ASSERT(tp->t_inpcb); 2378 2379 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2380 tcp_state_change(tp, TCPS_CLOSED); 2381 /* Don't use tcp_output() here due to possible recursion. */ 2382 (void)tcp_output_nodrop(tp); 2383 TCPSTAT_INC(tcps_drops); 2384 } else 2385 TCPSTAT_INC(tcps_conndrops); 2386 if (errno == ETIMEDOUT && tp->t_softerror) 2387 errno = tp->t_softerror; 2388 so->so_error = errno; 2389 return (tcp_close(tp)); 2390 } 2391 2392 void 2393 tcp_discardcb(struct tcpcb *tp) 2394 { 2395 struct inpcb *inp = tp->t_inpcb; 2396 2397 INP_WLOCK_ASSERT(inp); 2398 2399 /* 2400 * Make sure that all of our timers are stopped before we delete the 2401 * PCB. 2402 * 2403 * If stopping a timer fails, we schedule a discard function in same 2404 * callout, and the last discard function called will take care of 2405 * deleting the tcpcb. 2406 */ 2407 tp->t_timers->tt_draincnt = 0; 2408 tcp_timer_stop(tp, TT_REXMT); 2409 tcp_timer_stop(tp, TT_PERSIST); 2410 tcp_timer_stop(tp, TT_KEEP); 2411 tcp_timer_stop(tp, TT_2MSL); 2412 tcp_timer_stop(tp, TT_DELACK); 2413 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2414 /* 2415 * Call the stop-all function of the methods, 2416 * this function should call the tcp_timer_stop() 2417 * method with each of the function specific timeouts. 2418 * That stop will be called via the tfb_tcp_timer_stop() 2419 * which should use the async drain function of the 2420 * callout system (see tcp_var.h). 2421 */ 2422 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2423 } 2424 2425 /* free the reassembly queue, if any */ 2426 tcp_reass_flush(tp); 2427 2428 #ifdef TCP_OFFLOAD 2429 /* Disconnect offload device, if any. */ 2430 if (tp->t_flags & TF_TOE) 2431 tcp_offload_detach(tp); 2432 #endif 2433 2434 tcp_free_sackholes(tp); 2435 2436 #ifdef TCPPCAP 2437 /* Free the TCP PCAP queues. */ 2438 tcp_pcap_drain(&(tp->t_inpkts)); 2439 tcp_pcap_drain(&(tp->t_outpkts)); 2440 #endif 2441 2442 /* Allow the CC algorithm to clean up after itself. */ 2443 if (CC_ALGO(tp)->cb_destroy != NULL) 2444 CC_ALGO(tp)->cb_destroy(tp->ccv); 2445 CC_DATA(tp) = NULL; 2446 2447 #ifdef TCP_HHOOK 2448 khelp_destroy_osd(tp->osd); 2449 #endif 2450 #ifdef STATS 2451 stats_blob_destroy(tp->t_stats); 2452 #endif 2453 2454 CC_ALGO(tp) = NULL; 2455 inp->inp_ppcb = NULL; 2456 if (tp->t_timers->tt_draincnt == 0) { 2457 bool released __diagused; 2458 2459 released = tcp_freecb(tp); 2460 KASSERT(!released, ("%s: inp %p should not have been released " 2461 "here", __func__, inp)); 2462 } 2463 } 2464 2465 bool 2466 tcp_freecb(struct tcpcb *tp) 2467 { 2468 struct inpcb *inp = tp->t_inpcb; 2469 struct socket *so = inp->inp_socket; 2470 #ifdef INET6 2471 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2472 #endif 2473 2474 INP_WLOCK_ASSERT(inp); 2475 MPASS(tp->t_timers->tt_draincnt == 0); 2476 2477 /* We own the last reference on tcpcb, let's free it. */ 2478 #ifdef TCP_BLACKBOX 2479 tcp_log_tcpcbfini(tp); 2480 #endif 2481 TCPSTATES_DEC(tp->t_state); 2482 if (tp->t_fb->tfb_tcp_fb_fini) 2483 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2484 2485 /* 2486 * If we got enough samples through the srtt filter, 2487 * save the rtt and rttvar in the routing entry. 2488 * 'Enough' is arbitrarily defined as 4 rtt samples. 2489 * 4 samples is enough for the srtt filter to converge 2490 * to within enough % of the correct value; fewer samples 2491 * and we could save a bogus rtt. The danger is not high 2492 * as tcp quickly recovers from everything. 2493 * XXX: Works very well but needs some more statistics! 2494 * 2495 * XXXRRS: Updating must be after the stack fini() since 2496 * that may be converting some internal representation of 2497 * say srtt etc into the general one used by other stacks. 2498 * Lets also at least protect against the so being NULL 2499 * as RW stated below. 2500 */ 2501 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2502 struct hc_metrics_lite metrics; 2503 uint32_t ssthresh; 2504 2505 bzero(&metrics, sizeof(metrics)); 2506 /* 2507 * Update the ssthresh always when the conditions below 2508 * are satisfied. This gives us better new start value 2509 * for the congestion avoidance for new connections. 2510 * ssthresh is only set if packet loss occurred on a session. 2511 * 2512 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2513 * being torn down. Ideally this code would not use 'so'. 2514 */ 2515 ssthresh = tp->snd_ssthresh; 2516 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2517 /* 2518 * convert the limit from user data bytes to 2519 * packets then to packet data bytes. 2520 */ 2521 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2522 if (ssthresh < 2) 2523 ssthresh = 2; 2524 ssthresh *= (tp->t_maxseg + 2525 #ifdef INET6 2526 (isipv6 ? sizeof (struct ip6_hdr) + 2527 sizeof (struct tcphdr) : 2528 #endif 2529 sizeof (struct tcpiphdr) 2530 #ifdef INET6 2531 ) 2532 #endif 2533 ); 2534 } else 2535 ssthresh = 0; 2536 metrics.rmx_ssthresh = ssthresh; 2537 2538 metrics.rmx_rtt = tp->t_srtt; 2539 metrics.rmx_rttvar = tp->t_rttvar; 2540 metrics.rmx_cwnd = tp->snd_cwnd; 2541 metrics.rmx_sendpipe = 0; 2542 metrics.rmx_recvpipe = 0; 2543 2544 tcp_hc_update(&inp->inp_inc, &metrics); 2545 } 2546 2547 refcount_release(&tp->t_fb->tfb_refcnt); 2548 uma_zfree(V_tcpcb_zone, tp); 2549 2550 return (in_pcbrele_wlocked(inp)); 2551 } 2552 2553 /* 2554 * Attempt to close a TCP control block, marking it as dropped, and freeing 2555 * the socket if we hold the only reference. 2556 */ 2557 struct tcpcb * 2558 tcp_close(struct tcpcb *tp) 2559 { 2560 struct inpcb *inp = tp->t_inpcb; 2561 struct socket *so; 2562 2563 INP_WLOCK_ASSERT(inp); 2564 2565 #ifdef TCP_OFFLOAD 2566 if (tp->t_state == TCPS_LISTEN) 2567 tcp_offload_listen_stop(tp); 2568 #endif 2569 /* 2570 * This releases the TFO pending counter resource for TFO listen 2571 * sockets as well as passively-created TFO sockets that transition 2572 * from SYN_RECEIVED to CLOSED. 2573 */ 2574 if (tp->t_tfo_pending) { 2575 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2576 tp->t_tfo_pending = NULL; 2577 } 2578 #ifdef TCPHPTS 2579 tcp_hpts_remove(inp); 2580 #endif 2581 in_pcbdrop(inp); 2582 TCPSTAT_INC(tcps_closed); 2583 if (tp->t_state != TCPS_CLOSED) 2584 tcp_state_change(tp, TCPS_CLOSED); 2585 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2586 so = inp->inp_socket; 2587 soisdisconnected(so); 2588 if (inp->inp_flags & INP_SOCKREF) { 2589 KASSERT(so->so_state & SS_PROTOREF, 2590 ("tcp_close: !SS_PROTOREF")); 2591 inp->inp_flags &= ~INP_SOCKREF; 2592 INP_WUNLOCK(inp); 2593 SOCK_LOCK(so); 2594 so->so_state &= ~SS_PROTOREF; 2595 sofree(so); 2596 return (NULL); 2597 } 2598 return (tp); 2599 } 2600 2601 void 2602 tcp_drain(void) 2603 { 2604 VNET_ITERATOR_DECL(vnet_iter); 2605 2606 if (!do_tcpdrain) 2607 return; 2608 2609 VNET_LIST_RLOCK_NOSLEEP(); 2610 VNET_FOREACH(vnet_iter) { 2611 CURVNET_SET(vnet_iter); 2612 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2613 INPLOOKUP_WLOCKPCB); 2614 struct inpcb *inpb; 2615 struct tcpcb *tcpb; 2616 2617 /* 2618 * Walk the tcpbs, if existing, and flush the reassembly queue, 2619 * if there is one... 2620 * XXX: The "Net/3" implementation doesn't imply that the TCP 2621 * reassembly queue should be flushed, but in a situation 2622 * where we're really low on mbufs, this is potentially 2623 * useful. 2624 */ 2625 while ((inpb = inp_next(&inpi)) != NULL) { 2626 if (inpb->inp_flags & INP_TIMEWAIT) 2627 continue; 2628 if ((tcpb = intotcpcb(inpb)) != NULL) { 2629 tcp_reass_flush(tcpb); 2630 tcp_clean_sackreport(tcpb); 2631 #ifdef TCP_BLACKBOX 2632 tcp_log_drain(tcpb); 2633 #endif 2634 #ifdef TCPPCAP 2635 if (tcp_pcap_aggressive_free) { 2636 /* Free the TCP PCAP queues. */ 2637 tcp_pcap_drain(&(tcpb->t_inpkts)); 2638 tcp_pcap_drain(&(tcpb->t_outpkts)); 2639 } 2640 #endif 2641 } 2642 } 2643 CURVNET_RESTORE(); 2644 } 2645 VNET_LIST_RUNLOCK_NOSLEEP(); 2646 } 2647 2648 /* 2649 * Notify a tcp user of an asynchronous error; 2650 * store error as soft error, but wake up user 2651 * (for now, won't do anything until can select for soft error). 2652 * 2653 * Do not wake up user since there currently is no mechanism for 2654 * reporting soft errors (yet - a kqueue filter may be added). 2655 */ 2656 static struct inpcb * 2657 tcp_notify(struct inpcb *inp, int error) 2658 { 2659 struct tcpcb *tp; 2660 2661 INP_WLOCK_ASSERT(inp); 2662 2663 if ((inp->inp_flags & INP_TIMEWAIT) || 2664 (inp->inp_flags & INP_DROPPED)) 2665 return (inp); 2666 2667 tp = intotcpcb(inp); 2668 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2669 2670 /* 2671 * Ignore some errors if we are hooked up. 2672 * If connection hasn't completed, has retransmitted several times, 2673 * and receives a second error, give up now. This is better 2674 * than waiting a long time to establish a connection that 2675 * can never complete. 2676 */ 2677 if (tp->t_state == TCPS_ESTABLISHED && 2678 (error == EHOSTUNREACH || error == ENETUNREACH || 2679 error == EHOSTDOWN)) { 2680 if (inp->inp_route.ro_nh) { 2681 NH_FREE(inp->inp_route.ro_nh); 2682 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2683 } 2684 return (inp); 2685 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2686 tp->t_softerror) { 2687 tp = tcp_drop(tp, error); 2688 if (tp != NULL) 2689 return (inp); 2690 else 2691 return (NULL); 2692 } else { 2693 tp->t_softerror = error; 2694 return (inp); 2695 } 2696 #if 0 2697 wakeup( &so->so_timeo); 2698 sorwakeup(so); 2699 sowwakeup(so); 2700 #endif 2701 } 2702 2703 static int 2704 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2705 { 2706 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2707 INPLOOKUP_RLOCKPCB); 2708 struct xinpgen xig; 2709 struct inpcb *inp; 2710 int error; 2711 2712 if (req->newptr != NULL) 2713 return (EPERM); 2714 2715 if (req->oldptr == NULL) { 2716 int n; 2717 2718 n = V_tcbinfo.ipi_count + 2719 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2720 n += imax(n / 8, 10); 2721 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2722 return (0); 2723 } 2724 2725 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2726 return (error); 2727 2728 bzero(&xig, sizeof(xig)); 2729 xig.xig_len = sizeof xig; 2730 xig.xig_count = V_tcbinfo.ipi_count + 2731 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2732 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2733 xig.xig_sogen = so_gencnt; 2734 error = SYSCTL_OUT(req, &xig, sizeof xig); 2735 if (error) 2736 return (error); 2737 2738 error = syncache_pcblist(req); 2739 if (error) 2740 return (error); 2741 2742 while ((inp = inp_next(&inpi)) != NULL) { 2743 if (inp->inp_gencnt <= xig.xig_gen) { 2744 int crerr; 2745 2746 /* 2747 * XXX: This use of cr_cansee(), introduced with 2748 * TCP state changes, is not quite right, but for 2749 * now, better than nothing. 2750 */ 2751 if (inp->inp_flags & INP_TIMEWAIT) { 2752 if (intotw(inp) != NULL) 2753 crerr = cr_cansee(req->td->td_ucred, 2754 intotw(inp)->tw_cred); 2755 else 2756 crerr = EINVAL; /* Skip this inp. */ 2757 } else 2758 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2759 if (crerr == 0) { 2760 struct xtcpcb xt; 2761 2762 tcp_inptoxtp(inp, &xt); 2763 error = SYSCTL_OUT(req, &xt, sizeof xt); 2764 if (error) { 2765 INP_RUNLOCK(inp); 2766 break; 2767 } else 2768 continue; 2769 } 2770 } 2771 } 2772 2773 if (!error) { 2774 /* 2775 * Give the user an updated idea of our state. 2776 * If the generation differs from what we told 2777 * her before, she knows that something happened 2778 * while we were processing this request, and it 2779 * might be necessary to retry. 2780 */ 2781 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2782 xig.xig_sogen = so_gencnt; 2783 xig.xig_count = V_tcbinfo.ipi_count + 2784 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2785 error = SYSCTL_OUT(req, &xig, sizeof xig); 2786 } 2787 2788 return (error); 2789 } 2790 2791 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2792 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2793 NULL, 0, tcp_pcblist, "S,xtcpcb", 2794 "List of active TCP connections"); 2795 2796 #ifdef INET 2797 static int 2798 tcp_getcred(SYSCTL_HANDLER_ARGS) 2799 { 2800 struct xucred xuc; 2801 struct sockaddr_in addrs[2]; 2802 struct epoch_tracker et; 2803 struct inpcb *inp; 2804 int error; 2805 2806 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2807 if (error) 2808 return (error); 2809 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2810 if (error) 2811 return (error); 2812 NET_EPOCH_ENTER(et); 2813 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2814 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2815 NET_EPOCH_EXIT(et); 2816 if (inp != NULL) { 2817 if (inp->inp_socket == NULL) 2818 error = ENOENT; 2819 if (error == 0) 2820 error = cr_canseeinpcb(req->td->td_ucred, inp); 2821 if (error == 0) 2822 cru2x(inp->inp_cred, &xuc); 2823 INP_RUNLOCK(inp); 2824 } else 2825 error = ENOENT; 2826 if (error == 0) 2827 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2828 return (error); 2829 } 2830 2831 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2832 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2833 0, 0, tcp_getcred, "S,xucred", 2834 "Get the xucred of a TCP connection"); 2835 #endif /* INET */ 2836 2837 #ifdef INET6 2838 static int 2839 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2840 { 2841 struct epoch_tracker et; 2842 struct xucred xuc; 2843 struct sockaddr_in6 addrs[2]; 2844 struct inpcb *inp; 2845 int error; 2846 #ifdef INET 2847 int mapped = 0; 2848 #endif 2849 2850 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2851 if (error) 2852 return (error); 2853 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2854 if (error) 2855 return (error); 2856 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2857 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2858 return (error); 2859 } 2860 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2861 #ifdef INET 2862 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2863 mapped = 1; 2864 else 2865 #endif 2866 return (EINVAL); 2867 } 2868 2869 NET_EPOCH_ENTER(et); 2870 #ifdef INET 2871 if (mapped == 1) 2872 inp = in_pcblookup(&V_tcbinfo, 2873 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2874 addrs[1].sin6_port, 2875 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2876 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2877 else 2878 #endif 2879 inp = in6_pcblookup(&V_tcbinfo, 2880 &addrs[1].sin6_addr, addrs[1].sin6_port, 2881 &addrs[0].sin6_addr, addrs[0].sin6_port, 2882 INPLOOKUP_RLOCKPCB, NULL); 2883 NET_EPOCH_EXIT(et); 2884 if (inp != NULL) { 2885 if (inp->inp_socket == NULL) 2886 error = ENOENT; 2887 if (error == 0) 2888 error = cr_canseeinpcb(req->td->td_ucred, inp); 2889 if (error == 0) 2890 cru2x(inp->inp_cred, &xuc); 2891 INP_RUNLOCK(inp); 2892 } else 2893 error = ENOENT; 2894 if (error == 0) 2895 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2896 return (error); 2897 } 2898 2899 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2900 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2901 0, 0, tcp6_getcred, "S,xucred", 2902 "Get the xucred of a TCP6 connection"); 2903 #endif /* INET6 */ 2904 2905 #ifdef INET 2906 /* Path MTU to try next when a fragmentation-needed message is received. */ 2907 static inline int 2908 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2909 { 2910 int mtu = ntohs(icp->icmp_nextmtu); 2911 2912 /* If no alternative MTU was proposed, try the next smaller one. */ 2913 if (!mtu) 2914 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2915 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2916 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2917 2918 return (mtu); 2919 } 2920 2921 static void 2922 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port) 2923 { 2924 struct ip *ip = vip; 2925 struct tcphdr *th; 2926 struct in_addr faddr; 2927 struct inpcb *inp; 2928 struct tcpcb *tp; 2929 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2930 struct icmp *icp; 2931 struct in_conninfo inc; 2932 tcp_seq icmp_tcp_seq; 2933 int mtu; 2934 2935 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2936 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2937 return; 2938 2939 if (cmd == PRC_MSGSIZE) 2940 notify = tcp_mtudisc_notify; 2941 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2942 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2943 cmd == PRC_TIMXCEED_INTRANS) && ip) 2944 notify = tcp_drop_syn_sent; 2945 2946 /* 2947 * Hostdead is ugly because it goes linearly through all PCBs. 2948 * XXX: We never get this from ICMP, otherwise it makes an 2949 * excellent DoS attack on machines with many connections. 2950 */ 2951 else if (cmd == PRC_HOSTDEAD) 2952 ip = NULL; 2953 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2954 return; 2955 2956 if (ip == NULL) { 2957 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2958 return; 2959 } 2960 2961 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2962 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2963 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2964 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2965 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2966 /* signal EHOSTDOWN, as it flushes the cached route */ 2967 inp = (*notify)(inp, EHOSTDOWN); 2968 goto out; 2969 } 2970 icmp_tcp_seq = th->th_seq; 2971 if (inp != NULL) { 2972 if (!(inp->inp_flags & INP_TIMEWAIT) && 2973 !(inp->inp_flags & INP_DROPPED) && 2974 !(inp->inp_socket == NULL)) { 2975 tp = intotcpcb(inp); 2976 #ifdef TCP_OFFLOAD 2977 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 2978 /* 2979 * MTU discovery for offloaded connections. Let 2980 * the TOE driver verify seq# and process it. 2981 */ 2982 mtu = tcp_next_pmtu(icp, ip); 2983 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2984 goto out; 2985 } 2986 #endif 2987 if (tp->t_port != port) { 2988 goto out; 2989 } 2990 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2991 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2992 if (cmd == PRC_MSGSIZE) { 2993 /* 2994 * MTU discovery: we got a needfrag and 2995 * will potentially try a lower MTU. 2996 */ 2997 mtu = tcp_next_pmtu(icp, ip); 2998 2999 /* 3000 * Only process the offered MTU if it 3001 * is smaller than the current one. 3002 */ 3003 if (mtu < tp->t_maxseg + 3004 sizeof(struct tcpiphdr)) { 3005 bzero(&inc, sizeof(inc)); 3006 inc.inc_faddr = faddr; 3007 inc.inc_fibnum = 3008 inp->inp_inc.inc_fibnum; 3009 tcp_hc_updatemtu(&inc, mtu); 3010 inp = tcp_mtudisc(inp, mtu); 3011 } 3012 } else 3013 inp = (*notify)(inp, 3014 inetctlerrmap[cmd]); 3015 } 3016 } 3017 } else { 3018 bzero(&inc, sizeof(inc)); 3019 inc.inc_fport = th->th_dport; 3020 inc.inc_lport = th->th_sport; 3021 inc.inc_faddr = faddr; 3022 inc.inc_laddr = ip->ip_src; 3023 syncache_unreach(&inc, icmp_tcp_seq, port); 3024 } 3025 out: 3026 if (inp != NULL) 3027 INP_WUNLOCK(inp); 3028 } 3029 3030 void 3031 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 3032 { 3033 tcp_ctlinput_with_port(cmd, sa, vip, htons(0)); 3034 } 3035 3036 void 3037 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused) 3038 { 3039 /* Its a tunneled TCP over UDP icmp */ 3040 struct ip *outer_ip, *inner_ip; 3041 struct icmp *icmp; 3042 struct udphdr *udp; 3043 struct tcphdr *th, ttemp; 3044 int i_hlen, o_len; 3045 uint16_t port; 3046 3047 inner_ip = (struct ip *)vip; 3048 icmp = (struct icmp *)((caddr_t)inner_ip - 3049 (sizeof(struct icmp) - sizeof(struct ip))); 3050 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 3051 i_hlen = inner_ip->ip_hl << 2; 3052 o_len = ntohs(outer_ip->ip_len); 3053 if (o_len < 3054 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 3055 /* Not enough data present */ 3056 return; 3057 } 3058 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 3059 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 3060 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3061 return; 3062 } 3063 port = udp->uh_dport; 3064 th = (struct tcphdr *)(udp + 1); 3065 memcpy(&ttemp, th, sizeof(struct tcphdr)); 3066 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 3067 /* Now adjust down the size of the outer IP header */ 3068 o_len -= sizeof(struct udphdr); 3069 outer_ip->ip_len = htons(o_len); 3070 /* Now call in to the normal handling code */ 3071 tcp_ctlinput_with_port(cmd, sa, vip, port); 3072 } 3073 #endif /* INET */ 3074 3075 #ifdef INET6 3076 static inline int 3077 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3078 { 3079 int mtu = ntohl(icmp6->icmp6_mtu); 3080 3081 /* 3082 * If no alternative MTU was proposed, or the proposed MTU was too 3083 * small, set to the min. 3084 */ 3085 if (mtu < IPV6_MMTU) 3086 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 3087 return (mtu); 3088 } 3089 3090 static void 3091 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port) 3092 { 3093 struct in6_addr *dst; 3094 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 3095 struct ip6_hdr *ip6; 3096 struct mbuf *m; 3097 struct inpcb *inp; 3098 struct tcpcb *tp; 3099 struct icmp6_hdr *icmp6; 3100 struct ip6ctlparam *ip6cp = NULL; 3101 const struct sockaddr_in6 *sa6_src = NULL; 3102 struct in_conninfo inc; 3103 struct tcp_ports { 3104 uint16_t th_sport; 3105 uint16_t th_dport; 3106 } t_ports; 3107 tcp_seq icmp_tcp_seq; 3108 unsigned int mtu; 3109 unsigned int off; 3110 3111 if (sa->sa_family != AF_INET6 || 3112 sa->sa_len != sizeof(struct sockaddr_in6)) 3113 return; 3114 3115 /* if the parameter is from icmp6, decode it. */ 3116 if (d != NULL) { 3117 ip6cp = (struct ip6ctlparam *)d; 3118 icmp6 = ip6cp->ip6c_icmp6; 3119 m = ip6cp->ip6c_m; 3120 ip6 = ip6cp->ip6c_ip6; 3121 off = ip6cp->ip6c_off; 3122 sa6_src = ip6cp->ip6c_src; 3123 dst = ip6cp->ip6c_finaldst; 3124 } else { 3125 m = NULL; 3126 ip6 = NULL; 3127 off = 0; /* fool gcc */ 3128 sa6_src = &sa6_any; 3129 dst = NULL; 3130 } 3131 3132 if (cmd == PRC_MSGSIZE) 3133 notify = tcp_mtudisc_notify; 3134 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 3135 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 3136 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 3137 notify = tcp_drop_syn_sent; 3138 3139 /* 3140 * Hostdead is ugly because it goes linearly through all PCBs. 3141 * XXX: We never get this from ICMP, otherwise it makes an 3142 * excellent DoS attack on machines with many connections. 3143 */ 3144 else if (cmd == PRC_HOSTDEAD) 3145 ip6 = NULL; 3146 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 3147 return; 3148 3149 if (ip6 == NULL) { 3150 in6_pcbnotify(&V_tcbinfo, sa, 0, 3151 (const struct sockaddr *)sa6_src, 3152 0, cmd, NULL, notify); 3153 return; 3154 } 3155 3156 /* Check if we can safely get the ports from the tcp hdr */ 3157 if (m == NULL || 3158 (m->m_pkthdr.len < 3159 (int32_t) (off + sizeof(struct tcp_ports)))) { 3160 return; 3161 } 3162 bzero(&t_ports, sizeof(struct tcp_ports)); 3163 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3164 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3165 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3166 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 3167 /* signal EHOSTDOWN, as it flushes the cached route */ 3168 inp = (*notify)(inp, EHOSTDOWN); 3169 goto out; 3170 } 3171 off += sizeof(struct tcp_ports); 3172 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3173 goto out; 3174 } 3175 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3176 if (inp != NULL) { 3177 if (!(inp->inp_flags & INP_TIMEWAIT) && 3178 !(inp->inp_flags & INP_DROPPED) && 3179 !(inp->inp_socket == NULL)) { 3180 tp = intotcpcb(inp); 3181 #ifdef TCP_OFFLOAD 3182 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 3183 /* MTU discovery for offloaded connections. */ 3184 mtu = tcp6_next_pmtu(icmp6); 3185 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3186 goto out; 3187 } 3188 #endif 3189 if (tp->t_port != port) { 3190 goto out; 3191 } 3192 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3193 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3194 if (cmd == PRC_MSGSIZE) { 3195 /* 3196 * MTU discovery: 3197 * If we got a needfrag set the MTU 3198 * in the route to the suggested new 3199 * value (if given) and then notify. 3200 */ 3201 mtu = tcp6_next_pmtu(icmp6); 3202 3203 bzero(&inc, sizeof(inc)); 3204 inc.inc_fibnum = M_GETFIB(m); 3205 inc.inc_flags |= INC_ISIPV6; 3206 inc.inc6_faddr = *dst; 3207 if (in6_setscope(&inc.inc6_faddr, 3208 m->m_pkthdr.rcvif, NULL)) 3209 goto out; 3210 /* 3211 * Only process the offered MTU if it 3212 * is smaller than the current one. 3213 */ 3214 if (mtu < tp->t_maxseg + 3215 sizeof (struct tcphdr) + 3216 sizeof (struct ip6_hdr)) { 3217 tcp_hc_updatemtu(&inc, mtu); 3218 tcp_mtudisc(inp, mtu); 3219 ICMP6STAT_INC(icp6s_pmtuchg); 3220 } 3221 } else 3222 inp = (*notify)(inp, 3223 inet6ctlerrmap[cmd]); 3224 } 3225 } 3226 } else { 3227 bzero(&inc, sizeof(inc)); 3228 inc.inc_fibnum = M_GETFIB(m); 3229 inc.inc_flags |= INC_ISIPV6; 3230 inc.inc_fport = t_ports.th_dport; 3231 inc.inc_lport = t_ports.th_sport; 3232 inc.inc6_faddr = *dst; 3233 inc.inc6_laddr = ip6->ip6_src; 3234 syncache_unreach(&inc, icmp_tcp_seq, port); 3235 } 3236 out: 3237 if (inp != NULL) 3238 INP_WUNLOCK(inp); 3239 } 3240 3241 void 3242 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 3243 { 3244 tcp6_ctlinput_with_port(cmd, sa, d, htons(0)); 3245 } 3246 3247 void 3248 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused) 3249 { 3250 struct ip6ctlparam *ip6cp; 3251 struct mbuf *m; 3252 struct udphdr *udp; 3253 uint16_t port; 3254 3255 ip6cp = (struct ip6ctlparam *)d; 3256 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3257 if (m == NULL) { 3258 return; 3259 } 3260 udp = mtod(m, struct udphdr *); 3261 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3262 return; 3263 } 3264 port = udp->uh_dport; 3265 m_adj(m, sizeof(struct udphdr)); 3266 if ((m->m_flags & M_PKTHDR) == 0) { 3267 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3268 } 3269 /* Now call in to the normal handling code */ 3270 tcp6_ctlinput_with_port(cmd, sa, d, port); 3271 } 3272 3273 #endif /* INET6 */ 3274 3275 static uint32_t 3276 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3277 { 3278 SIPHASH_CTX ctx; 3279 uint32_t hash[2]; 3280 3281 KASSERT(len >= SIPHASH_KEY_LENGTH, 3282 ("%s: keylen %u too short ", __func__, len)); 3283 SipHash24_Init(&ctx); 3284 SipHash_SetKey(&ctx, (uint8_t *)key); 3285 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3286 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3287 switch (inc->inc_flags & INC_ISIPV6) { 3288 #ifdef INET 3289 case 0: 3290 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3291 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3292 break; 3293 #endif 3294 #ifdef INET6 3295 case INC_ISIPV6: 3296 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3297 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3298 break; 3299 #endif 3300 } 3301 SipHash_Final((uint8_t *)hash, &ctx); 3302 3303 return (hash[0] ^ hash[1]); 3304 } 3305 3306 uint32_t 3307 tcp_new_ts_offset(struct in_conninfo *inc) 3308 { 3309 struct in_conninfo inc_store, *local_inc; 3310 3311 if (!V_tcp_ts_offset_per_conn) { 3312 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3313 inc_store.inc_lport = 0; 3314 inc_store.inc_fport = 0; 3315 local_inc = &inc_store; 3316 } else { 3317 local_inc = inc; 3318 } 3319 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3320 sizeof(V_ts_offset_secret))); 3321 } 3322 3323 /* 3324 * Following is where TCP initial sequence number generation occurs. 3325 * 3326 * There are two places where we must use initial sequence numbers: 3327 * 1. In SYN-ACK packets. 3328 * 2. In SYN packets. 3329 * 3330 * All ISNs for SYN-ACK packets are generated by the syncache. See 3331 * tcp_syncache.c for details. 3332 * 3333 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3334 * depends on this property. In addition, these ISNs should be 3335 * unguessable so as to prevent connection hijacking. To satisfy 3336 * the requirements of this situation, the algorithm outlined in 3337 * RFC 1948 is used, with only small modifications. 3338 * 3339 * Implementation details: 3340 * 3341 * Time is based off the system timer, and is corrected so that it 3342 * increases by one megabyte per second. This allows for proper 3343 * recycling on high speed LANs while still leaving over an hour 3344 * before rollover. 3345 * 3346 * As reading the *exact* system time is too expensive to be done 3347 * whenever setting up a TCP connection, we increment the time 3348 * offset in two ways. First, a small random positive increment 3349 * is added to isn_offset for each connection that is set up. 3350 * Second, the function tcp_isn_tick fires once per clock tick 3351 * and increments isn_offset as necessary so that sequence numbers 3352 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3353 * random positive increments serve only to ensure that the same 3354 * exact sequence number is never sent out twice (as could otherwise 3355 * happen when a port is recycled in less than the system tick 3356 * interval.) 3357 * 3358 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3359 * between seeding of isn_secret. This is normally set to zero, 3360 * as reseeding should not be necessary. 3361 * 3362 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3363 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3364 * general, this means holding an exclusive (write) lock. 3365 */ 3366 3367 #define ISN_BYTES_PER_SECOND 1048576 3368 #define ISN_STATIC_INCREMENT 4096 3369 #define ISN_RANDOM_INCREMENT (4096 - 1) 3370 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3371 3372 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3373 VNET_DEFINE_STATIC(int, isn_last); 3374 VNET_DEFINE_STATIC(int, isn_last_reseed); 3375 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3376 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3377 3378 #define V_isn_secret VNET(isn_secret) 3379 #define V_isn_last VNET(isn_last) 3380 #define V_isn_last_reseed VNET(isn_last_reseed) 3381 #define V_isn_offset VNET(isn_offset) 3382 #define V_isn_offset_old VNET(isn_offset_old) 3383 3384 tcp_seq 3385 tcp_new_isn(struct in_conninfo *inc) 3386 { 3387 tcp_seq new_isn; 3388 u_int32_t projected_offset; 3389 3390 ISN_LOCK(); 3391 /* Seed if this is the first use, reseed if requested. */ 3392 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3393 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3394 < (u_int)ticks))) { 3395 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3396 V_isn_last_reseed = ticks; 3397 } 3398 3399 /* Compute the hash and return the ISN. */ 3400 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3401 sizeof(V_isn_secret)); 3402 V_isn_offset += ISN_STATIC_INCREMENT + 3403 (arc4random() & ISN_RANDOM_INCREMENT); 3404 if (ticks != V_isn_last) { 3405 projected_offset = V_isn_offset_old + 3406 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3407 if (SEQ_GT(projected_offset, V_isn_offset)) 3408 V_isn_offset = projected_offset; 3409 V_isn_offset_old = V_isn_offset; 3410 V_isn_last = ticks; 3411 } 3412 new_isn += V_isn_offset; 3413 ISN_UNLOCK(); 3414 return (new_isn); 3415 } 3416 3417 /* 3418 * When a specific ICMP unreachable message is received and the 3419 * connection state is SYN-SENT, drop the connection. This behavior 3420 * is controlled by the icmp_may_rst sysctl. 3421 */ 3422 struct inpcb * 3423 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3424 { 3425 struct tcpcb *tp; 3426 3427 NET_EPOCH_ASSERT(); 3428 INP_WLOCK_ASSERT(inp); 3429 3430 if ((inp->inp_flags & INP_TIMEWAIT) || 3431 (inp->inp_flags & INP_DROPPED)) 3432 return (inp); 3433 3434 tp = intotcpcb(inp); 3435 if (tp->t_state != TCPS_SYN_SENT) 3436 return (inp); 3437 3438 if (IS_FASTOPEN(tp->t_flags)) 3439 tcp_fastopen_disable_path(tp); 3440 3441 tp = tcp_drop(tp, errno); 3442 if (tp != NULL) 3443 return (inp); 3444 else 3445 return (NULL); 3446 } 3447 3448 /* 3449 * When `need fragmentation' ICMP is received, update our idea of the MSS 3450 * based on the new value. Also nudge TCP to send something, since we 3451 * know the packet we just sent was dropped. 3452 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3453 */ 3454 static struct inpcb * 3455 tcp_mtudisc_notify(struct inpcb *inp, int error) 3456 { 3457 3458 return (tcp_mtudisc(inp, -1)); 3459 } 3460 3461 static struct inpcb * 3462 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3463 { 3464 struct tcpcb *tp; 3465 struct socket *so; 3466 3467 INP_WLOCK_ASSERT(inp); 3468 if ((inp->inp_flags & INP_TIMEWAIT) || 3469 (inp->inp_flags & INP_DROPPED)) 3470 return (inp); 3471 3472 tp = intotcpcb(inp); 3473 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3474 3475 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3476 3477 so = inp->inp_socket; 3478 SOCKBUF_LOCK(&so->so_snd); 3479 /* If the mss is larger than the socket buffer, decrease the mss. */ 3480 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3481 tp->t_maxseg = so->so_snd.sb_hiwat; 3482 SOCKBUF_UNLOCK(&so->so_snd); 3483 3484 TCPSTAT_INC(tcps_mturesent); 3485 tp->t_rtttime = 0; 3486 tp->snd_nxt = tp->snd_una; 3487 tcp_free_sackholes(tp); 3488 tp->snd_recover = tp->snd_max; 3489 if (tp->t_flags & TF_SACK_PERMIT) 3490 EXIT_FASTRECOVERY(tp->t_flags); 3491 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3492 /* 3493 * Conceptually the snd_nxt setting 3494 * and freeing sack holes should 3495 * be done by the default stacks 3496 * own tfb_tcp_mtu_chg(). 3497 */ 3498 tp->t_fb->tfb_tcp_mtu_chg(tp); 3499 } 3500 if (tcp_output(tp) < 0) 3501 return (NULL); 3502 else 3503 return (inp); 3504 } 3505 3506 #ifdef INET 3507 /* 3508 * Look-up the routing entry to the peer of this inpcb. If no route 3509 * is found and it cannot be allocated, then return 0. This routine 3510 * is called by TCP routines that access the rmx structure and by 3511 * tcp_mss_update to get the peer/interface MTU. 3512 */ 3513 uint32_t 3514 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3515 { 3516 struct nhop_object *nh; 3517 struct ifnet *ifp; 3518 uint32_t maxmtu = 0; 3519 3520 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3521 3522 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3523 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3524 if (nh == NULL) 3525 return (0); 3526 3527 ifp = nh->nh_ifp; 3528 maxmtu = nh->nh_mtu; 3529 3530 /* Report additional interface capabilities. */ 3531 if (cap != NULL) { 3532 if (ifp->if_capenable & IFCAP_TSO4 && 3533 ifp->if_hwassist & CSUM_TSO) { 3534 cap->ifcap |= CSUM_TSO; 3535 cap->tsomax = ifp->if_hw_tsomax; 3536 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3537 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3538 } 3539 } 3540 } 3541 return (maxmtu); 3542 } 3543 #endif /* INET */ 3544 3545 #ifdef INET6 3546 uint32_t 3547 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3548 { 3549 struct nhop_object *nh; 3550 struct in6_addr dst6; 3551 uint32_t scopeid; 3552 struct ifnet *ifp; 3553 uint32_t maxmtu = 0; 3554 3555 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3556 3557 if (inc->inc_flags & INC_IPV6MINMTU) 3558 return (IPV6_MMTU); 3559 3560 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3561 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3562 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3563 if (nh == NULL) 3564 return (0); 3565 3566 ifp = nh->nh_ifp; 3567 maxmtu = nh->nh_mtu; 3568 3569 /* Report additional interface capabilities. */ 3570 if (cap != NULL) { 3571 if (ifp->if_capenable & IFCAP_TSO6 && 3572 ifp->if_hwassist & CSUM_TSO) { 3573 cap->ifcap |= CSUM_TSO; 3574 cap->tsomax = ifp->if_hw_tsomax; 3575 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3576 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3577 } 3578 } 3579 } 3580 3581 return (maxmtu); 3582 } 3583 3584 /* 3585 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3586 * 3587 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3588 * The right place to do that is ip6_setpktopt() that has just been 3589 * executed. By the way it just filled ip6po_minmtu for us. 3590 */ 3591 void 3592 tcp6_use_min_mtu(struct tcpcb *tp) 3593 { 3594 struct inpcb *inp = tp->t_inpcb; 3595 3596 INP_WLOCK_ASSERT(inp); 3597 /* 3598 * In case of the IPV6_USE_MIN_MTU socket 3599 * option, the INC_IPV6MINMTU flag to announce 3600 * a corresponding MSS during the initial 3601 * handshake. If the TCP connection is not in 3602 * the front states, just reduce the MSS being 3603 * used. This avoids the sending of TCP 3604 * segments which will be fragmented at the 3605 * IPv6 layer. 3606 */ 3607 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3608 if ((tp->t_state >= TCPS_SYN_SENT) && 3609 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3610 struct ip6_pktopts *opt; 3611 3612 opt = inp->in6p_outputopts; 3613 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3614 tp->t_maxseg > TCP6_MSS) 3615 tp->t_maxseg = TCP6_MSS; 3616 } 3617 } 3618 #endif /* INET6 */ 3619 3620 /* 3621 * Calculate effective SMSS per RFC5681 definition for a given TCP 3622 * connection at its current state, taking into account SACK and etc. 3623 */ 3624 u_int 3625 tcp_maxseg(const struct tcpcb *tp) 3626 { 3627 u_int optlen; 3628 3629 if (tp->t_flags & TF_NOOPT) 3630 return (tp->t_maxseg); 3631 3632 /* 3633 * Here we have a simplified code from tcp_addoptions(), 3634 * without a proper loop, and having most of paddings hardcoded. 3635 * We might make mistakes with padding here in some edge cases, 3636 * but this is harmless, since result of tcp_maxseg() is used 3637 * only in cwnd and ssthresh estimations. 3638 */ 3639 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3640 if (tp->t_flags & TF_RCVD_TSTMP) 3641 optlen = TCPOLEN_TSTAMP_APPA; 3642 else 3643 optlen = 0; 3644 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3645 if (tp->t_flags & TF_SIGNATURE) 3646 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3647 #endif 3648 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3649 optlen += TCPOLEN_SACKHDR; 3650 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3651 optlen = PADTCPOLEN(optlen); 3652 } 3653 } else { 3654 if (tp->t_flags & TF_REQ_TSTMP) 3655 optlen = TCPOLEN_TSTAMP_APPA; 3656 else 3657 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3658 if (tp->t_flags & TF_REQ_SCALE) 3659 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3660 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3661 if (tp->t_flags & TF_SIGNATURE) 3662 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3663 #endif 3664 if (tp->t_flags & TF_SACK_PERMIT) 3665 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3666 } 3667 #undef PAD 3668 optlen = min(optlen, TCP_MAXOLEN); 3669 return (tp->t_maxseg - optlen); 3670 } 3671 3672 3673 u_int 3674 tcp_fixed_maxseg(const struct tcpcb *tp) 3675 { 3676 int optlen; 3677 3678 if (tp->t_flags & TF_NOOPT) 3679 return (tp->t_maxseg); 3680 3681 /* 3682 * Here we have a simplified code from tcp_addoptions(), 3683 * without a proper loop, and having most of paddings hardcoded. 3684 * We only consider fixed options that we would send every 3685 * time I.e. SACK is not considered. This is important 3686 * for cc modules to figure out what the modulo of the 3687 * cwnd should be. 3688 */ 3689 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3690 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3691 if (tp->t_flags & TF_RCVD_TSTMP) 3692 optlen = TCPOLEN_TSTAMP_APPA; 3693 else 3694 optlen = 0; 3695 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3696 if (tp->t_flags & TF_SIGNATURE) 3697 optlen += PAD(TCPOLEN_SIGNATURE); 3698 #endif 3699 } else { 3700 if (tp->t_flags & TF_REQ_TSTMP) 3701 optlen = TCPOLEN_TSTAMP_APPA; 3702 else 3703 optlen = PAD(TCPOLEN_MAXSEG); 3704 if (tp->t_flags & TF_REQ_SCALE) 3705 optlen += PAD(TCPOLEN_WINDOW); 3706 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3707 if (tp->t_flags & TF_SIGNATURE) 3708 optlen += PAD(TCPOLEN_SIGNATURE); 3709 #endif 3710 if (tp->t_flags & TF_SACK_PERMIT) 3711 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3712 } 3713 #undef PAD 3714 optlen = min(optlen, TCP_MAXOLEN); 3715 return (tp->t_maxseg - optlen); 3716 } 3717 3718 3719 3720 static int 3721 sysctl_drop(SYSCTL_HANDLER_ARGS) 3722 { 3723 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3724 struct sockaddr_storage addrs[2]; 3725 struct inpcb *inp; 3726 struct tcpcb *tp; 3727 struct tcptw *tw; 3728 struct sockaddr_in *fin, *lin; 3729 struct epoch_tracker et; 3730 #ifdef INET6 3731 struct sockaddr_in6 *fin6, *lin6; 3732 #endif 3733 int error; 3734 3735 inp = NULL; 3736 fin = lin = NULL; 3737 #ifdef INET6 3738 fin6 = lin6 = NULL; 3739 #endif 3740 error = 0; 3741 3742 if (req->oldptr != NULL || req->oldlen != 0) 3743 return (EINVAL); 3744 if (req->newptr == NULL) 3745 return (EPERM); 3746 if (req->newlen < sizeof(addrs)) 3747 return (ENOMEM); 3748 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3749 if (error) 3750 return (error); 3751 3752 switch (addrs[0].ss_family) { 3753 #ifdef INET6 3754 case AF_INET6: 3755 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3756 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3757 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3758 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3759 return (EINVAL); 3760 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3761 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3762 return (EINVAL); 3763 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3764 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3765 fin = (struct sockaddr_in *)&addrs[0]; 3766 lin = (struct sockaddr_in *)&addrs[1]; 3767 break; 3768 } 3769 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3770 if (error) 3771 return (error); 3772 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3773 if (error) 3774 return (error); 3775 break; 3776 #endif 3777 #ifdef INET 3778 case AF_INET: 3779 fin = (struct sockaddr_in *)&addrs[0]; 3780 lin = (struct sockaddr_in *)&addrs[1]; 3781 if (fin->sin_len != sizeof(struct sockaddr_in) || 3782 lin->sin_len != sizeof(struct sockaddr_in)) 3783 return (EINVAL); 3784 break; 3785 #endif 3786 default: 3787 return (EINVAL); 3788 } 3789 NET_EPOCH_ENTER(et); 3790 switch (addrs[0].ss_family) { 3791 #ifdef INET6 3792 case AF_INET6: 3793 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3794 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3795 INPLOOKUP_WLOCKPCB, NULL); 3796 break; 3797 #endif 3798 #ifdef INET 3799 case AF_INET: 3800 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3801 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3802 break; 3803 #endif 3804 } 3805 if (inp != NULL) { 3806 if (inp->inp_flags & INP_TIMEWAIT) { 3807 /* 3808 * XXXRW: There currently exists a state where an 3809 * inpcb is present, but its timewait state has been 3810 * discarded. For now, don't allow dropping of this 3811 * type of inpcb. 3812 */ 3813 tw = intotw(inp); 3814 if (tw != NULL) 3815 tcp_twclose(tw, 0); 3816 else 3817 INP_WUNLOCK(inp); 3818 } else if ((inp->inp_flags & INP_DROPPED) == 0 && 3819 !SOLISTENING(inp->inp_socket)) { 3820 tp = intotcpcb(inp); 3821 tp = tcp_drop(tp, ECONNABORTED); 3822 if (tp != NULL) 3823 INP_WUNLOCK(inp); 3824 } else 3825 INP_WUNLOCK(inp); 3826 } else 3827 error = ESRCH; 3828 NET_EPOCH_EXIT(et); 3829 return (error); 3830 } 3831 3832 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3833 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3834 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3835 "Drop TCP connection"); 3836 3837 static int 3838 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3839 { 3840 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3841 &tcp_ctloutput_set)); 3842 } 3843 3844 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3845 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3846 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3847 "Set socket option for TCP endpoint"); 3848 3849 #ifdef KERN_TLS 3850 static int 3851 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3852 { 3853 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3854 struct sockaddr_storage addrs[2]; 3855 struct inpcb *inp; 3856 struct sockaddr_in *fin, *lin; 3857 struct epoch_tracker et; 3858 #ifdef INET6 3859 struct sockaddr_in6 *fin6, *lin6; 3860 #endif 3861 int error; 3862 3863 inp = NULL; 3864 fin = lin = NULL; 3865 #ifdef INET6 3866 fin6 = lin6 = NULL; 3867 #endif 3868 error = 0; 3869 3870 if (req->oldptr != NULL || req->oldlen != 0) 3871 return (EINVAL); 3872 if (req->newptr == NULL) 3873 return (EPERM); 3874 if (req->newlen < sizeof(addrs)) 3875 return (ENOMEM); 3876 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3877 if (error) 3878 return (error); 3879 3880 switch (addrs[0].ss_family) { 3881 #ifdef INET6 3882 case AF_INET6: 3883 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3884 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3885 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3886 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3887 return (EINVAL); 3888 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3889 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3890 return (EINVAL); 3891 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3892 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3893 fin = (struct sockaddr_in *)&addrs[0]; 3894 lin = (struct sockaddr_in *)&addrs[1]; 3895 break; 3896 } 3897 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3898 if (error) 3899 return (error); 3900 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3901 if (error) 3902 return (error); 3903 break; 3904 #endif 3905 #ifdef INET 3906 case AF_INET: 3907 fin = (struct sockaddr_in *)&addrs[0]; 3908 lin = (struct sockaddr_in *)&addrs[1]; 3909 if (fin->sin_len != sizeof(struct sockaddr_in) || 3910 lin->sin_len != sizeof(struct sockaddr_in)) 3911 return (EINVAL); 3912 break; 3913 #endif 3914 default: 3915 return (EINVAL); 3916 } 3917 NET_EPOCH_ENTER(et); 3918 switch (addrs[0].ss_family) { 3919 #ifdef INET6 3920 case AF_INET6: 3921 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3922 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3923 INPLOOKUP_WLOCKPCB, NULL); 3924 break; 3925 #endif 3926 #ifdef INET 3927 case AF_INET: 3928 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3929 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3930 break; 3931 #endif 3932 } 3933 NET_EPOCH_EXIT(et); 3934 if (inp != NULL) { 3935 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3936 inp->inp_socket == NULL) { 3937 error = ECONNRESET; 3938 INP_WUNLOCK(inp); 3939 } else { 3940 struct socket *so; 3941 3942 so = inp->inp_socket; 3943 soref(so); 3944 error = ktls_set_tx_mode(so, 3945 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3946 INP_WUNLOCK(inp); 3947 sorele(so); 3948 } 3949 } else 3950 error = ESRCH; 3951 return (error); 3952 } 3953 3954 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3955 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3956 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3957 "Switch TCP connection to SW TLS"); 3958 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3959 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3960 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3961 "Switch TCP connection to ifnet TLS"); 3962 #endif 3963 3964 /* 3965 * Generate a standardized TCP log line for use throughout the 3966 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3967 * allow use in the interrupt context. 3968 * 3969 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3970 * NB: The function may return NULL if memory allocation failed. 3971 * 3972 * Due to header inclusion and ordering limitations the struct ip 3973 * and ip6_hdr pointers have to be passed as void pointers. 3974 */ 3975 char * 3976 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3977 const void *ip6hdr) 3978 { 3979 3980 /* Is logging enabled? */ 3981 if (V_tcp_log_in_vain == 0) 3982 return (NULL); 3983 3984 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3985 } 3986 3987 char * 3988 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3989 const void *ip6hdr) 3990 { 3991 3992 /* Is logging enabled? */ 3993 if (tcp_log_debug == 0) 3994 return (NULL); 3995 3996 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3997 } 3998 3999 static char * 4000 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 4001 const void *ip6hdr) 4002 { 4003 char *s, *sp; 4004 size_t size; 4005 struct ip *ip; 4006 #ifdef INET6 4007 const struct ip6_hdr *ip6; 4008 4009 ip6 = (const struct ip6_hdr *)ip6hdr; 4010 #endif /* INET6 */ 4011 ip = (struct ip *)ip4hdr; 4012 4013 /* 4014 * The log line looks like this: 4015 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 4016 */ 4017 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 4018 sizeof(PRINT_TH_FLAGS) + 1 + 4019 #ifdef INET6 4020 2 * INET6_ADDRSTRLEN; 4021 #else 4022 2 * INET_ADDRSTRLEN; 4023 #endif /* INET6 */ 4024 4025 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 4026 if (s == NULL) 4027 return (NULL); 4028 4029 strcat(s, "TCP: ["); 4030 sp = s + strlen(s); 4031 4032 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 4033 inet_ntoa_r(inc->inc_faddr, sp); 4034 sp = s + strlen(s); 4035 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4036 sp = s + strlen(s); 4037 inet_ntoa_r(inc->inc_laddr, sp); 4038 sp = s + strlen(s); 4039 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4040 #ifdef INET6 4041 } else if (inc) { 4042 ip6_sprintf(sp, &inc->inc6_faddr); 4043 sp = s + strlen(s); 4044 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4045 sp = s + strlen(s); 4046 ip6_sprintf(sp, &inc->inc6_laddr); 4047 sp = s + strlen(s); 4048 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4049 } else if (ip6 && th) { 4050 ip6_sprintf(sp, &ip6->ip6_src); 4051 sp = s + strlen(s); 4052 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4053 sp = s + strlen(s); 4054 ip6_sprintf(sp, &ip6->ip6_dst); 4055 sp = s + strlen(s); 4056 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4057 #endif /* INET6 */ 4058 #ifdef INET 4059 } else if (ip && th) { 4060 inet_ntoa_r(ip->ip_src, sp); 4061 sp = s + strlen(s); 4062 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4063 sp = s + strlen(s); 4064 inet_ntoa_r(ip->ip_dst, sp); 4065 sp = s + strlen(s); 4066 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4067 #endif /* INET */ 4068 } else { 4069 free(s, M_TCPLOG); 4070 return (NULL); 4071 } 4072 sp = s + strlen(s); 4073 if (th) 4074 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 4075 if (*(s + size - 1) != '\0') 4076 panic("%s: string too long", __func__); 4077 return (s); 4078 } 4079 4080 /* 4081 * A subroutine which makes it easy to track TCP state changes with DTrace. 4082 * This function shouldn't be called for t_state initializations that don't 4083 * correspond to actual TCP state transitions. 4084 */ 4085 void 4086 tcp_state_change(struct tcpcb *tp, int newstate) 4087 { 4088 #if defined(KDTRACE_HOOKS) 4089 int pstate = tp->t_state; 4090 #endif 4091 4092 TCPSTATES_DEC(tp->t_state); 4093 TCPSTATES_INC(newstate); 4094 tp->t_state = newstate; 4095 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4096 } 4097 4098 /* 4099 * Create an external-format (``xtcpcb'') structure using the information in 4100 * the kernel-format tcpcb structure pointed to by tp. This is done to 4101 * reduce the spew of irrelevant information over this interface, to isolate 4102 * user code from changes in the kernel structure, and potentially to provide 4103 * information-hiding if we decide that some of this information should be 4104 * hidden from users. 4105 */ 4106 void 4107 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4108 { 4109 struct tcpcb *tp = intotcpcb(inp); 4110 struct tcptw *tw = intotw(inp); 4111 sbintime_t now; 4112 4113 bzero(xt, sizeof(*xt)); 4114 if (inp->inp_flags & INP_TIMEWAIT) { 4115 xt->t_state = TCPS_TIME_WAIT; 4116 xt->xt_encaps_port = tw->t_port; 4117 } else { 4118 xt->t_state = tp->t_state; 4119 xt->t_logstate = tp->t_logstate; 4120 xt->t_flags = tp->t_flags; 4121 xt->t_sndzerowin = tp->t_sndzerowin; 4122 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4123 xt->t_rcvoopack = tp->t_rcvoopack; 4124 xt->t_rcv_wnd = tp->rcv_wnd; 4125 xt->t_snd_wnd = tp->snd_wnd; 4126 xt->t_snd_cwnd = tp->snd_cwnd; 4127 xt->t_snd_ssthresh = tp->snd_ssthresh; 4128 xt->t_dsack_bytes = tp->t_dsack_bytes; 4129 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4130 xt->t_dsack_pack = tp->t_dsack_pack; 4131 xt->t_maxseg = tp->t_maxseg; 4132 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4133 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4134 4135 now = getsbinuptime(); 4136 #define COPYTIMER(ttt) do { \ 4137 if (callout_active(&tp->t_timers->ttt)) \ 4138 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 4139 SBT_1MS; \ 4140 else \ 4141 xt->ttt = 0; \ 4142 } while (0) 4143 COPYTIMER(tt_delack); 4144 COPYTIMER(tt_rexmt); 4145 COPYTIMER(tt_persist); 4146 COPYTIMER(tt_keep); 4147 COPYTIMER(tt_2msl); 4148 #undef COPYTIMER 4149 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4150 4151 xt->xt_encaps_port = tp->t_port; 4152 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4153 TCP_FUNCTION_NAME_LEN_MAX); 4154 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 4155 TCP_CA_NAME_MAX); 4156 #ifdef TCP_BLACKBOX 4157 (void)tcp_log_get_id(tp, xt->xt_logid); 4158 #endif 4159 } 4160 4161 xt->xt_len = sizeof(struct xtcpcb); 4162 in_pcbtoxinpcb(inp, &xt->xt_inp); 4163 if (inp->inp_socket == NULL) 4164 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 4165 } 4166 4167 void 4168 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4169 { 4170 uint32_t bit, i; 4171 4172 if ((tp == NULL) || 4173 (status > TCP_EI_STATUS_MAX_VALUE) || 4174 (status == 0)) { 4175 /* Invalid */ 4176 return; 4177 } 4178 if (status > (sizeof(uint32_t) * 8)) { 4179 /* Should this be a KASSERT? */ 4180 return; 4181 } 4182 bit = 1U << (status - 1); 4183 if (bit & tp->t_end_info_status) { 4184 /* already logged */ 4185 return; 4186 } 4187 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4188 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4189 tp->t_end_info_bytes[i] = status; 4190 tp->t_end_info_status |= bit; 4191 break; 4192 } 4193 } 4194 } 4195 4196 int 4197 tcp_can_enable_pacing(void) 4198 { 4199 4200 if ((tcp_pacing_limit == -1) || 4201 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4202 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4203 shadow_num_connections = number_of_tcp_connections_pacing; 4204 return (1); 4205 } else { 4206 return (0); 4207 } 4208 } 4209 4210 static uint8_t tcp_pacing_warning = 0; 4211 4212 void 4213 tcp_decrement_paced_conn(void) 4214 { 4215 uint32_t ret; 4216 4217 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4218 shadow_num_connections = number_of_tcp_connections_pacing; 4219 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4220 if (ret == 0) { 4221 if (tcp_pacing_limit != -1) { 4222 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4223 tcp_pacing_limit = 0; 4224 } else if (tcp_pacing_warning == 0) { 4225 printf("Warning pacing count is invalid, invalid decrement\n"); 4226 tcp_pacing_warning = 1; 4227 } 4228 } 4229 } 4230