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 void 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_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 /* 1150 * TCP initialization. 1151 */ 1152 static void 1153 tcp_zone_change(void *tag) 1154 { 1155 1156 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 1157 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1158 tcp_tw_zone_change(); 1159 } 1160 1161 static int 1162 tcp_inpcb_init(void *mem, int size, int flags) 1163 { 1164 struct inpcb *inp = mem; 1165 1166 INP_LOCK_INIT(inp, "inp", "tcpinp"); 1167 return (0); 1168 } 1169 1170 /* 1171 * Take a value and get the next power of 2 that doesn't overflow. 1172 * Used to size the tcp_inpcb hash buckets. 1173 */ 1174 static int 1175 maketcp_hashsize(int size) 1176 { 1177 int hashsize; 1178 1179 /* 1180 * auto tune. 1181 * get the next power of 2 higher than maxsockets. 1182 */ 1183 hashsize = 1 << fls(size); 1184 /* catch overflow, and just go one power of 2 smaller */ 1185 if (hashsize < size) { 1186 hashsize = 1 << (fls(size) - 1); 1187 } 1188 return (hashsize); 1189 } 1190 1191 static volatile int next_tcp_stack_id = 1; 1192 1193 /* 1194 * Register a TCP function block with the name provided in the names 1195 * array. (Note that this function does NOT automatically register 1196 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1197 * explicitly include blk->tfb_tcp_block_name in the list of names if 1198 * you wish to register the stack with that name.) 1199 * 1200 * Either all name registrations will succeed or all will fail. If 1201 * a name registration fails, the function will update the num_names 1202 * argument to point to the array index of the name that encountered 1203 * the failure. 1204 * 1205 * Returns 0 on success, or an error code on failure. 1206 */ 1207 int 1208 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1209 const char *names[], int *num_names) 1210 { 1211 struct tcp_function *n; 1212 struct tcp_function_set fs; 1213 int error, i; 1214 1215 KASSERT(names != NULL && *num_names > 0, 1216 ("%s: Called with 0-length name list", __func__)); 1217 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1218 KASSERT(rw_initialized(&tcp_function_lock), 1219 ("%s: called too early", __func__)); 1220 1221 if ((blk->tfb_tcp_output == NULL) || 1222 (blk->tfb_tcp_do_segment == NULL) || 1223 (blk->tfb_tcp_ctloutput == NULL) || 1224 (strlen(blk->tfb_tcp_block_name) == 0)) { 1225 /* 1226 * These functions are required and you 1227 * need a name. 1228 */ 1229 *num_names = 0; 1230 return (EINVAL); 1231 } 1232 if (blk->tfb_tcp_timer_stop_all || 1233 blk->tfb_tcp_timer_activate || 1234 blk->tfb_tcp_timer_active || 1235 blk->tfb_tcp_timer_stop) { 1236 /* 1237 * If you define one timer function you 1238 * must have them all. 1239 */ 1240 if ((blk->tfb_tcp_timer_stop_all == NULL) || 1241 (blk->tfb_tcp_timer_activate == NULL) || 1242 (blk->tfb_tcp_timer_active == NULL) || 1243 (blk->tfb_tcp_timer_stop == NULL)) { 1244 *num_names = 0; 1245 return (EINVAL); 1246 } 1247 } 1248 1249 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1250 *num_names = 0; 1251 return (EINVAL); 1252 } 1253 1254 refcount_init(&blk->tfb_refcnt, 0); 1255 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1256 for (i = 0; i < *num_names; i++) { 1257 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1258 if (n == NULL) { 1259 error = ENOMEM; 1260 goto cleanup; 1261 } 1262 n->tf_fb = blk; 1263 1264 (void)strlcpy(fs.function_set_name, names[i], 1265 sizeof(fs.function_set_name)); 1266 rw_wlock(&tcp_function_lock); 1267 if (find_tcp_functions_locked(&fs) != NULL) { 1268 /* Duplicate name space not allowed */ 1269 rw_wunlock(&tcp_function_lock); 1270 free(n, M_TCPFUNCTIONS); 1271 error = EALREADY; 1272 goto cleanup; 1273 } 1274 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1275 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1276 tcp_fb_cnt++; 1277 rw_wunlock(&tcp_function_lock); 1278 } 1279 return(0); 1280 1281 cleanup: 1282 /* 1283 * Deregister the names we just added. Because registration failed 1284 * for names[i], we don't need to deregister that name. 1285 */ 1286 *num_names = i; 1287 rw_wlock(&tcp_function_lock); 1288 while (--i >= 0) { 1289 TAILQ_FOREACH(n, &t_functions, tf_next) { 1290 if (!strncmp(n->tf_name, names[i], 1291 TCP_FUNCTION_NAME_LEN_MAX)) { 1292 TAILQ_REMOVE(&t_functions, n, tf_next); 1293 tcp_fb_cnt--; 1294 n->tf_fb = NULL; 1295 free(n, M_TCPFUNCTIONS); 1296 break; 1297 } 1298 } 1299 } 1300 rw_wunlock(&tcp_function_lock); 1301 return (error); 1302 } 1303 1304 /* 1305 * Register a TCP function block using the name provided in the name 1306 * argument. 1307 * 1308 * Returns 0 on success, or an error code on failure. 1309 */ 1310 int 1311 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1312 int wait) 1313 { 1314 const char *name_list[1]; 1315 int num_names, rv; 1316 1317 num_names = 1; 1318 if (name != NULL) 1319 name_list[0] = name; 1320 else 1321 name_list[0] = blk->tfb_tcp_block_name; 1322 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1323 return (rv); 1324 } 1325 1326 /* 1327 * Register a TCP function block using the name defined in 1328 * blk->tfb_tcp_block_name. 1329 * 1330 * Returns 0 on success, or an error code on failure. 1331 */ 1332 int 1333 register_tcp_functions(struct tcp_function_block *blk, int wait) 1334 { 1335 1336 return (register_tcp_functions_as_name(blk, NULL, wait)); 1337 } 1338 1339 /* 1340 * Deregister all names associated with a function block. This 1341 * functionally removes the function block from use within the system. 1342 * 1343 * When called with a true quiesce argument, mark the function block 1344 * as being removed so no more stacks will use it and determine 1345 * whether the removal would succeed. 1346 * 1347 * When called with a false quiesce argument, actually attempt the 1348 * removal. 1349 * 1350 * When called with a force argument, attempt to switch all TCBs to 1351 * use the default stack instead of returning EBUSY. 1352 * 1353 * Returns 0 on success (or if the removal would succeed, or an error 1354 * code on failure. 1355 */ 1356 int 1357 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1358 bool force) 1359 { 1360 struct tcp_function *f; 1361 1362 if (blk == &tcp_def_funcblk) { 1363 /* You can't un-register the default */ 1364 return (EPERM); 1365 } 1366 rw_wlock(&tcp_function_lock); 1367 if (blk == tcp_func_set_ptr) { 1368 /* You can't free the current default */ 1369 rw_wunlock(&tcp_function_lock); 1370 return (EBUSY); 1371 } 1372 /* Mark the block so no more stacks can use it. */ 1373 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1374 /* 1375 * If TCBs are still attached to the stack, attempt to switch them 1376 * to the default stack. 1377 */ 1378 if (force && blk->tfb_refcnt) { 1379 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1380 INPLOOKUP_WLOCKPCB); 1381 struct inpcb *inp; 1382 struct tcpcb *tp; 1383 VNET_ITERATOR_DECL(vnet_iter); 1384 1385 rw_wunlock(&tcp_function_lock); 1386 1387 VNET_LIST_RLOCK(); 1388 VNET_FOREACH(vnet_iter) { 1389 CURVNET_SET(vnet_iter); 1390 while ((inp = inp_next(&inpi)) != NULL) { 1391 if (inp->inp_flags & INP_TIMEWAIT) 1392 continue; 1393 tp = intotcpcb(inp); 1394 if (tp == NULL || tp->t_fb != blk) 1395 continue; 1396 tcp_switch_back_to_default(tp); 1397 } 1398 CURVNET_RESTORE(); 1399 } 1400 VNET_LIST_RUNLOCK(); 1401 1402 rw_wlock(&tcp_function_lock); 1403 } 1404 if (blk->tfb_refcnt) { 1405 /* TCBs still attached. */ 1406 rw_wunlock(&tcp_function_lock); 1407 return (EBUSY); 1408 } 1409 if (quiesce) { 1410 /* Skip removal. */ 1411 rw_wunlock(&tcp_function_lock); 1412 return (0); 1413 } 1414 /* Remove any function names that map to this function block. */ 1415 while (find_tcp_fb_locked(blk, &f) != NULL) { 1416 TAILQ_REMOVE(&t_functions, f, tf_next); 1417 tcp_fb_cnt--; 1418 f->tf_fb = NULL; 1419 free(f, M_TCPFUNCTIONS); 1420 } 1421 rw_wunlock(&tcp_function_lock); 1422 return (0); 1423 } 1424 1425 void 1426 tcp_init(void) 1427 { 1428 const char *tcbhash_tuneable; 1429 int hashsize; 1430 1431 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1432 1433 #ifdef TCP_HHOOK 1434 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1435 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1436 printf("%s: WARNING: unable to register helper hook\n", __func__); 1437 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1438 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1439 printf("%s: WARNING: unable to register helper hook\n", __func__); 1440 #endif 1441 #ifdef STATS 1442 if (tcp_stats_init()) 1443 printf("%s: WARNING: unable to initialise TCP stats\n", 1444 __func__); 1445 #endif 1446 hashsize = TCBHASHSIZE; 1447 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1448 if (hashsize == 0) { 1449 /* 1450 * Auto tune the hash size based on maxsockets. 1451 * A perfect hash would have a 1:1 mapping 1452 * (hashsize = maxsockets) however it's been 1453 * suggested that O(2) average is better. 1454 */ 1455 hashsize = maketcp_hashsize(maxsockets / 4); 1456 /* 1457 * Our historical default is 512, 1458 * do not autotune lower than this. 1459 */ 1460 if (hashsize < 512) 1461 hashsize = 512; 1462 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1463 printf("%s: %s auto tuned to %d\n", __func__, 1464 tcbhash_tuneable, hashsize); 1465 } 1466 /* 1467 * We require a hashsize to be a power of two. 1468 * Previously if it was not a power of two we would just reset it 1469 * back to 512, which could be a nasty surprise if you did not notice 1470 * the error message. 1471 * Instead what we do is clip it to the closest power of two lower 1472 * than the specified hash value. 1473 */ 1474 if (!powerof2(hashsize)) { 1475 int oldhashsize = hashsize; 1476 1477 hashsize = maketcp_hashsize(hashsize); 1478 /* prevent absurdly low value */ 1479 if (hashsize < 16) 1480 hashsize = 16; 1481 printf("%s: WARNING: TCB hash size not a power of 2, " 1482 "clipped from %d to %d.\n", __func__, oldhashsize, 1483 hashsize); 1484 } 1485 in_pcbinfo_init(&V_tcbinfo, "tcp", hashsize, hashsize, 1486 "tcp_inpcb", tcp_inpcb_init); 1487 1488 /* 1489 * These have to be type stable for the benefit of the timers. 1490 */ 1491 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1492 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1493 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1494 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1495 1496 tcp_tw_init(); 1497 syncache_init(); 1498 tcp_hc_init(); 1499 1500 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1501 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1502 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1503 1504 tcp_fastopen_init(); 1505 1506 /* Skip initialization of globals for non-default instances. */ 1507 if (!IS_DEFAULT_VNET(curvnet)) 1508 return; 1509 1510 tcp_reass_global_init(); 1511 1512 /* XXX virtualize those bellow? */ 1513 tcp_delacktime = TCPTV_DELACK; 1514 tcp_keepinit = TCPTV_KEEP_INIT; 1515 tcp_keepidle = TCPTV_KEEP_IDLE; 1516 tcp_keepintvl = TCPTV_KEEPINTVL; 1517 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1518 tcp_msl = TCPTV_MSL; 1519 tcp_rexmit_initial = TCPTV_RTOBASE; 1520 if (tcp_rexmit_initial < 1) 1521 tcp_rexmit_initial = 1; 1522 tcp_rexmit_min = TCPTV_MIN; 1523 if (tcp_rexmit_min < 1) 1524 tcp_rexmit_min = 1; 1525 tcp_persmin = TCPTV_PERSMIN; 1526 tcp_persmax = TCPTV_PERSMAX; 1527 tcp_rexmit_slop = TCPTV_CPU_VAR; 1528 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1529 tcp_tcbhashsize = hashsize; 1530 1531 /* Setup the tcp function block list */ 1532 TAILQ_INIT(&t_functions); 1533 rw_init(&tcp_function_lock, "tcp_func_lock"); 1534 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1535 #ifdef TCP_BLACKBOX 1536 /* Initialize the TCP logging data. */ 1537 tcp_log_init(); 1538 #endif 1539 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1540 1541 if (tcp_soreceive_stream) { 1542 #ifdef INET 1543 tcp_usrreqs.pru_soreceive = soreceive_stream; 1544 #endif 1545 #ifdef INET6 1546 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1547 #endif /* INET6 */ 1548 } 1549 1550 #ifdef INET6 1551 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1552 #else /* INET6 */ 1553 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1554 #endif /* INET6 */ 1555 if (max_protohdr < TCP_MINPROTOHDR) 1556 max_protohdr = TCP_MINPROTOHDR; 1557 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1558 panic("tcp_init"); 1559 #undef TCP_MINPROTOHDR 1560 1561 ISN_LOCK_INIT(); 1562 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1563 SHUTDOWN_PRI_DEFAULT); 1564 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1565 EVENTHANDLER_PRI_ANY); 1566 1567 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1568 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1569 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1570 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1571 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1572 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1573 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1574 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1575 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1576 #ifdef TCPPCAP 1577 tcp_pcap_init(); 1578 #endif 1579 } 1580 1581 #ifdef VIMAGE 1582 static void 1583 tcp_destroy(void *unused __unused) 1584 { 1585 int n; 1586 #ifdef TCP_HHOOK 1587 int error; 1588 #endif 1589 1590 /* 1591 * All our processes are gone, all our sockets should be cleaned 1592 * up, which means, we should be past the tcp_discardcb() calls. 1593 * Sleep to let all tcpcb timers really disappear and cleanup. 1594 */ 1595 for (;;) { 1596 INP_INFO_WLOCK(&V_tcbinfo); 1597 n = V_tcbinfo.ipi_count; 1598 INP_INFO_WUNLOCK(&V_tcbinfo); 1599 if (n == 0) 1600 break; 1601 pause("tcpdes", hz / 10); 1602 } 1603 tcp_hc_destroy(); 1604 syncache_destroy(); 1605 tcp_tw_destroy(); 1606 in_pcbinfo_destroy(&V_tcbinfo); 1607 /* tcp_discardcb() clears the sack_holes up. */ 1608 uma_zdestroy(V_sack_hole_zone); 1609 uma_zdestroy(V_tcpcb_zone); 1610 1611 /* 1612 * Cannot free the zone until all tcpcbs are released as we attach 1613 * the allocations to them. 1614 */ 1615 tcp_fastopen_destroy(); 1616 1617 #ifdef TCP_HHOOK 1618 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1619 if (error != 0) { 1620 printf("%s: WARNING: unable to deregister helper hook " 1621 "type=%d, id=%d: error %d returned\n", __func__, 1622 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1623 } 1624 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1625 if (error != 0) { 1626 printf("%s: WARNING: unable to deregister helper hook " 1627 "type=%d, id=%d: error %d returned\n", __func__, 1628 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1629 } 1630 #endif 1631 } 1632 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1633 #endif 1634 1635 void 1636 tcp_fini(void *xtp) 1637 { 1638 1639 } 1640 1641 /* 1642 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1643 * tcp_template used to store this data in mbufs, but we now recopy it out 1644 * of the tcpcb each time to conserve mbufs. 1645 */ 1646 void 1647 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1648 { 1649 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1650 1651 INP_WLOCK_ASSERT(inp); 1652 1653 #ifdef INET6 1654 if ((inp->inp_vflag & INP_IPV6) != 0) { 1655 struct ip6_hdr *ip6; 1656 1657 ip6 = (struct ip6_hdr *)ip_ptr; 1658 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1659 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1660 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1661 (IPV6_VERSION & IPV6_VERSION_MASK); 1662 if (port == 0) 1663 ip6->ip6_nxt = IPPROTO_TCP; 1664 else 1665 ip6->ip6_nxt = IPPROTO_UDP; 1666 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1667 ip6->ip6_src = inp->in6p_laddr; 1668 ip6->ip6_dst = inp->in6p_faddr; 1669 } 1670 #endif /* INET6 */ 1671 #if defined(INET6) && defined(INET) 1672 else 1673 #endif 1674 #ifdef INET 1675 { 1676 struct ip *ip; 1677 1678 ip = (struct ip *)ip_ptr; 1679 ip->ip_v = IPVERSION; 1680 ip->ip_hl = 5; 1681 ip->ip_tos = inp->inp_ip_tos; 1682 ip->ip_len = 0; 1683 ip->ip_id = 0; 1684 ip->ip_off = 0; 1685 ip->ip_ttl = inp->inp_ip_ttl; 1686 ip->ip_sum = 0; 1687 if (port == 0) 1688 ip->ip_p = IPPROTO_TCP; 1689 else 1690 ip->ip_p = IPPROTO_UDP; 1691 ip->ip_src = inp->inp_laddr; 1692 ip->ip_dst = inp->inp_faddr; 1693 } 1694 #endif /* INET */ 1695 th->th_sport = inp->inp_lport; 1696 th->th_dport = inp->inp_fport; 1697 th->th_seq = 0; 1698 th->th_ack = 0; 1699 th->th_x2 = 0; 1700 th->th_off = 5; 1701 th->th_flags = 0; 1702 th->th_win = 0; 1703 th->th_urp = 0; 1704 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1705 } 1706 1707 /* 1708 * Create template to be used to send tcp packets on a connection. 1709 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1710 * use for this function is in keepalives, which use tcp_respond. 1711 */ 1712 struct tcptemp * 1713 tcpip_maketemplate(struct inpcb *inp) 1714 { 1715 struct tcptemp *t; 1716 1717 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1718 if (t == NULL) 1719 return (NULL); 1720 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1721 return (t); 1722 } 1723 1724 /* 1725 * Send a single message to the TCP at address specified by 1726 * the given TCP/IP header. If m == NULL, then we make a copy 1727 * of the tcpiphdr at th and send directly to the addressed host. 1728 * This is used to force keep alive messages out using the TCP 1729 * template for a connection. If flags are given then we send 1730 * a message back to the TCP which originated the segment th, 1731 * and discard the mbuf containing it and any other attached mbufs. 1732 * 1733 * In any case the ack and sequence number of the transmitted 1734 * segment are as specified by the parameters. 1735 * 1736 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1737 */ 1738 void 1739 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1740 tcp_seq ack, tcp_seq seq, int flags) 1741 { 1742 struct tcpopt to; 1743 struct inpcb *inp; 1744 struct ip *ip; 1745 struct mbuf *optm; 1746 struct udphdr *uh = NULL; 1747 struct tcphdr *nth; 1748 struct tcp_log_buffer *lgb; 1749 u_char *optp; 1750 #ifdef INET6 1751 struct ip6_hdr *ip6; 1752 int isipv6; 1753 #endif /* INET6 */ 1754 int optlen, tlen, win, ulen; 1755 bool incl_opts; 1756 uint16_t port; 1757 int output_ret; 1758 #ifdef INVARIANTS 1759 int thflags = th->th_flags; 1760 #endif 1761 1762 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1763 NET_EPOCH_ASSERT(); 1764 1765 #ifdef INET6 1766 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1767 ip6 = ipgen; 1768 #endif /* INET6 */ 1769 ip = ipgen; 1770 1771 if (tp != NULL) { 1772 inp = tp->t_inpcb; 1773 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1774 INP_LOCK_ASSERT(inp); 1775 } else 1776 inp = NULL; 1777 1778 if (m != NULL) { 1779 #ifdef INET6 1780 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1781 port = m->m_pkthdr.tcp_tun_port; 1782 else 1783 #endif 1784 if (ip && (ip->ip_p == IPPROTO_UDP)) 1785 port = m->m_pkthdr.tcp_tun_port; 1786 else 1787 port = 0; 1788 } else 1789 port = tp->t_port; 1790 1791 incl_opts = false; 1792 win = 0; 1793 if (tp != NULL) { 1794 if (!(flags & TH_RST)) { 1795 win = sbspace(&inp->inp_socket->so_rcv); 1796 if (win > TCP_MAXWIN << tp->rcv_scale) 1797 win = TCP_MAXWIN << tp->rcv_scale; 1798 } 1799 if ((tp->t_flags & TF_NOOPT) == 0) 1800 incl_opts = true; 1801 } 1802 if (m == NULL) { 1803 m = m_gethdr(M_NOWAIT, MT_DATA); 1804 if (m == NULL) 1805 return; 1806 m->m_data += max_linkhdr; 1807 #ifdef INET6 1808 if (isipv6) { 1809 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1810 sizeof(struct ip6_hdr)); 1811 ip6 = mtod(m, struct ip6_hdr *); 1812 nth = (struct tcphdr *)(ip6 + 1); 1813 if (port) { 1814 /* Insert a UDP header */ 1815 uh = (struct udphdr *)nth; 1816 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1817 uh->uh_dport = port; 1818 nth = (struct tcphdr *)(uh + 1); 1819 } 1820 } else 1821 #endif /* INET6 */ 1822 { 1823 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1824 ip = mtod(m, struct ip *); 1825 nth = (struct tcphdr *)(ip + 1); 1826 if (port) { 1827 /* Insert a UDP header */ 1828 uh = (struct udphdr *)nth; 1829 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1830 uh->uh_dport = port; 1831 nth = (struct tcphdr *)(uh + 1); 1832 } 1833 } 1834 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1835 flags = TH_ACK; 1836 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1837 struct mbuf *n; 1838 1839 /* Can't reuse 'm', allocate a new mbuf. */ 1840 n = m_gethdr(M_NOWAIT, MT_DATA); 1841 if (n == NULL) { 1842 m_freem(m); 1843 return; 1844 } 1845 1846 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1847 m_freem(m); 1848 m_freem(n); 1849 return; 1850 } 1851 1852 n->m_data += max_linkhdr; 1853 /* m_len is set later */ 1854 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1855 #ifdef INET6 1856 if (isipv6) { 1857 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1858 sizeof(struct ip6_hdr)); 1859 ip6 = mtod(n, struct ip6_hdr *); 1860 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1861 nth = (struct tcphdr *)(ip6 + 1); 1862 if (port) { 1863 /* Insert a UDP header */ 1864 uh = (struct udphdr *)nth; 1865 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1866 uh->uh_dport = port; 1867 nth = (struct tcphdr *)(uh + 1); 1868 } 1869 } else 1870 #endif /* INET6 */ 1871 { 1872 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1873 ip = mtod(n, struct ip *); 1874 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1875 nth = (struct tcphdr *)(ip + 1); 1876 if (port) { 1877 /* Insert a UDP header */ 1878 uh = (struct udphdr *)nth; 1879 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1880 uh->uh_dport = port; 1881 nth = (struct tcphdr *)(uh + 1); 1882 } 1883 } 1884 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1885 xchg(nth->th_dport, nth->th_sport, uint16_t); 1886 th = nth; 1887 m_freem(m); 1888 m = n; 1889 } else { 1890 /* 1891 * reuse the mbuf. 1892 * XXX MRT We inherit the FIB, which is lucky. 1893 */ 1894 m_freem(m->m_next); 1895 m->m_next = NULL; 1896 m->m_data = (caddr_t)ipgen; 1897 /* m_len is set later */ 1898 #ifdef INET6 1899 if (isipv6) { 1900 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1901 nth = (struct tcphdr *)(ip6 + 1); 1902 } else 1903 #endif /* INET6 */ 1904 { 1905 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1906 nth = (struct tcphdr *)(ip + 1); 1907 } 1908 if (th != nth) { 1909 /* 1910 * this is usually a case when an extension header 1911 * exists between the IPv6 header and the 1912 * TCP header. 1913 */ 1914 nth->th_sport = th->th_sport; 1915 nth->th_dport = th->th_dport; 1916 } 1917 xchg(nth->th_dport, nth->th_sport, uint16_t); 1918 #undef xchg 1919 } 1920 tlen = 0; 1921 #ifdef INET6 1922 if (isipv6) 1923 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1924 #endif 1925 #if defined(INET) && defined(INET6) 1926 else 1927 #endif 1928 #ifdef INET 1929 tlen = sizeof (struct tcpiphdr); 1930 #endif 1931 if (port) 1932 tlen += sizeof (struct udphdr); 1933 #ifdef INVARIANTS 1934 m->m_len = 0; 1935 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1936 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1937 m, tlen, (long)M_TRAILINGSPACE(m))); 1938 #endif 1939 m->m_len = tlen; 1940 to.to_flags = 0; 1941 if (incl_opts) { 1942 /* Make sure we have room. */ 1943 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1944 m->m_next = m_get(M_NOWAIT, MT_DATA); 1945 if (m->m_next) { 1946 optp = mtod(m->m_next, u_char *); 1947 optm = m->m_next; 1948 } else 1949 incl_opts = false; 1950 } else { 1951 optp = (u_char *) (nth + 1); 1952 optm = m; 1953 } 1954 } 1955 if (incl_opts) { 1956 /* Timestamps. */ 1957 if (tp->t_flags & TF_RCVD_TSTMP) { 1958 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1959 to.to_tsecr = tp->ts_recent; 1960 to.to_flags |= TOF_TS; 1961 } 1962 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1963 /* TCP-MD5 (RFC2385). */ 1964 if (tp->t_flags & TF_SIGNATURE) 1965 to.to_flags |= TOF_SIGNATURE; 1966 #endif 1967 /* Add the options. */ 1968 tlen += optlen = tcp_addoptions(&to, optp); 1969 1970 /* Update m_len in the correct mbuf. */ 1971 optm->m_len += optlen; 1972 } else 1973 optlen = 0; 1974 #ifdef INET6 1975 if (isipv6) { 1976 if (uh) { 1977 ulen = tlen - sizeof(struct ip6_hdr); 1978 uh->uh_ulen = htons(ulen); 1979 } 1980 ip6->ip6_flow = 0; 1981 ip6->ip6_vfc = IPV6_VERSION; 1982 if (port) 1983 ip6->ip6_nxt = IPPROTO_UDP; 1984 else 1985 ip6->ip6_nxt = IPPROTO_TCP; 1986 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1987 } 1988 #endif 1989 #if defined(INET) && defined(INET6) 1990 else 1991 #endif 1992 #ifdef INET 1993 { 1994 if (uh) { 1995 ulen = tlen - sizeof(struct ip); 1996 uh->uh_ulen = htons(ulen); 1997 } 1998 ip->ip_len = htons(tlen); 1999 ip->ip_ttl = V_ip_defttl; 2000 if (port) { 2001 ip->ip_p = IPPROTO_UDP; 2002 } else { 2003 ip->ip_p = IPPROTO_TCP; 2004 } 2005 if (V_path_mtu_discovery) 2006 ip->ip_off |= htons(IP_DF); 2007 } 2008 #endif 2009 m->m_pkthdr.len = tlen; 2010 m->m_pkthdr.rcvif = NULL; 2011 #ifdef MAC 2012 if (inp != NULL) { 2013 /* 2014 * Packet is associated with a socket, so allow the 2015 * label of the response to reflect the socket label. 2016 */ 2017 INP_LOCK_ASSERT(inp); 2018 mac_inpcb_create_mbuf(inp, m); 2019 } else { 2020 /* 2021 * Packet is not associated with a socket, so possibly 2022 * update the label in place. 2023 */ 2024 mac_netinet_tcp_reply(m); 2025 } 2026 #endif 2027 nth->th_seq = htonl(seq); 2028 nth->th_ack = htonl(ack); 2029 nth->th_x2 = 0; 2030 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2031 nth->th_flags = flags; 2032 if (tp != NULL) 2033 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2034 else 2035 nth->th_win = htons((u_short)win); 2036 nth->th_urp = 0; 2037 2038 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2039 if (to.to_flags & TOF_SIGNATURE) { 2040 if (!TCPMD5_ENABLED() || 2041 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2042 m_freem(m); 2043 return; 2044 } 2045 } 2046 #endif 2047 2048 #ifdef INET6 2049 if (isipv6) { 2050 if (port) { 2051 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2052 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2053 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2054 nth->th_sum = 0; 2055 } else { 2056 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2057 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2058 nth->th_sum = in6_cksum_pseudo(ip6, 2059 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2060 } 2061 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 2062 NULL, NULL); 2063 } 2064 #endif /* INET6 */ 2065 #if defined(INET6) && defined(INET) 2066 else 2067 #endif 2068 #ifdef INET 2069 { 2070 if (port) { 2071 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2072 htons(ulen + IPPROTO_UDP)); 2073 m->m_pkthdr.csum_flags = CSUM_UDP; 2074 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2075 nth->th_sum = 0; 2076 } else { 2077 m->m_pkthdr.csum_flags = CSUM_TCP; 2078 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2079 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2080 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2081 } 2082 } 2083 #endif /* INET */ 2084 #ifdef TCPDEBUG 2085 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2086 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2087 #endif 2088 TCP_PROBE3(debug__output, tp, th, m); 2089 if (flags & TH_RST) 2090 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2091 lgb = NULL; 2092 if ((tp != NULL) && (tp->t_logstate != TCP_LOG_STATE_OFF)) { 2093 if (INP_WLOCKED(inp)) { 2094 union tcp_log_stackspecific log; 2095 struct timeval tv; 2096 2097 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2098 log.u_bbr.inhpts = tp->t_inpcb->inp_in_hpts; 2099 log.u_bbr.ininput = tp->t_inpcb->inp_in_dropq; 2100 log.u_bbr.flex8 = 4; 2101 log.u_bbr.pkts_out = tp->t_maxseg; 2102 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2103 log.u_bbr.delivered = 0; 2104 lgb = tcp_log_event_(tp, nth, NULL, NULL, TCP_LOG_OUT, 2105 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2106 } else { 2107 /* 2108 * We can not log the packet, since we only own the 2109 * read lock, but a write lock is needed. The read lock 2110 * is not upgraded to a write lock, since only getting 2111 * the read lock was done intentionally to improve the 2112 * handling of SYN flooding attacks. 2113 * This happens only for pure SYN segments received in 2114 * the initial CLOSED state, or received in a more 2115 * advanced state than listen and the UDP encapsulation 2116 * port is unexpected. 2117 * The incoming SYN segments do not really belong to 2118 * the TCP connection and the handling does not change 2119 * the state of the TCP connection. Therefore, the 2120 * sending of the RST segments is not logged. Please 2121 * note that also the incoming SYN segments are not 2122 * logged. 2123 * 2124 * The following code ensures that the above description 2125 * is and stays correct. 2126 */ 2127 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2128 (tp->t_state == TCPS_CLOSED || 2129 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2130 ("%s: Logging of TCP segment with flags 0x%b and " 2131 "UDP encapsulation port %u skipped in state %s", 2132 __func__, thflags, PRINT_TH_FLAGS, 2133 ntohs(port), tcpstates[tp->t_state])); 2134 } 2135 } 2136 2137 #ifdef INET6 2138 if (isipv6) { 2139 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2140 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2141 } 2142 #endif /* INET6 */ 2143 #if defined(INET) && defined(INET6) 2144 else 2145 #endif 2146 #ifdef INET 2147 { 2148 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2149 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2150 } 2151 #endif 2152 if (lgb != NULL) 2153 lgb->tlb_errno = output_ret; 2154 } 2155 2156 /* 2157 * Create a new TCP control block, making an 2158 * empty reassembly queue and hooking it to the argument 2159 * protocol control block. The `inp' parameter must have 2160 * come from the zone allocator set up in tcp_init(). 2161 */ 2162 struct tcpcb * 2163 tcp_newtcpcb(struct inpcb *inp) 2164 { 2165 struct tcpcb_mem *tm; 2166 struct tcpcb *tp; 2167 #ifdef INET6 2168 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2169 #endif /* INET6 */ 2170 2171 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 2172 if (tm == NULL) 2173 return (NULL); 2174 tp = &tm->tcb; 2175 2176 /* Initialise cc_var struct for this tcpcb. */ 2177 tp->ccv = &tm->ccv; 2178 tp->ccv->type = IPPROTO_TCP; 2179 tp->ccv->ccvc.tcp = tp; 2180 rw_rlock(&tcp_function_lock); 2181 tp->t_fb = tcp_func_set_ptr; 2182 refcount_acquire(&tp->t_fb->tfb_refcnt); 2183 rw_runlock(&tcp_function_lock); 2184 /* 2185 * Use the current system default CC algorithm. 2186 */ 2187 CC_LIST_RLOCK(); 2188 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 2189 CC_ALGO(tp) = CC_DEFAULT_ALGO(); 2190 CC_LIST_RUNLOCK(); 2191 2192 /* 2193 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 2194 * is called. 2195 */ 2196 in_pcbref(inp); /* Reference for tcpcb */ 2197 tp->t_inpcb = inp; 2198 2199 if (CC_ALGO(tp)->cb_init != NULL) 2200 if (CC_ALGO(tp)->cb_init(tp->ccv, NULL) > 0) { 2201 if (tp->t_fb->tfb_tcp_fb_fini) 2202 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2203 in_pcbrele_wlocked(inp); 2204 refcount_release(&tp->t_fb->tfb_refcnt); 2205 uma_zfree(V_tcpcb_zone, tm); 2206 return (NULL); 2207 } 2208 2209 #ifdef TCP_HHOOK 2210 tp->osd = &tm->osd; 2211 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 2212 if (tp->t_fb->tfb_tcp_fb_fini) 2213 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2214 in_pcbrele_wlocked(inp); 2215 refcount_release(&tp->t_fb->tfb_refcnt); 2216 uma_zfree(V_tcpcb_zone, tm); 2217 return (NULL); 2218 } 2219 #endif 2220 2221 #ifdef VIMAGE 2222 tp->t_vnet = inp->inp_vnet; 2223 #endif 2224 tp->t_timers = &tm->tt; 2225 TAILQ_INIT(&tp->t_segq); 2226 tp->t_maxseg = 2227 #ifdef INET6 2228 isipv6 ? V_tcp_v6mssdflt : 2229 #endif /* INET6 */ 2230 V_tcp_mssdflt; 2231 2232 /* Set up our timeouts. */ 2233 callout_init(&tp->t_timers->tt_rexmt, 1); 2234 callout_init(&tp->t_timers->tt_persist, 1); 2235 callout_init(&tp->t_timers->tt_keep, 1); 2236 callout_init(&tp->t_timers->tt_2msl, 1); 2237 callout_init(&tp->t_timers->tt_delack, 1); 2238 2239 if (V_tcp_do_rfc1323) 2240 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2241 if (V_tcp_do_sack) 2242 tp->t_flags |= TF_SACK_PERMIT; 2243 TAILQ_INIT(&tp->snd_holes); 2244 2245 /* 2246 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2247 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2248 * reasonable initial retransmit time. 2249 */ 2250 tp->t_srtt = TCPTV_SRTTBASE; 2251 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2252 tp->t_rttmin = tcp_rexmit_min; 2253 tp->t_rxtcur = tcp_rexmit_initial; 2254 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2255 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2256 tp->t_rcvtime = ticks; 2257 /* 2258 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2259 * because the socket may be bound to an IPv6 wildcard address, 2260 * which may match an IPv4-mapped IPv6 address. 2261 */ 2262 inp->inp_ip_ttl = V_ip_defttl; 2263 inp->inp_ppcb = tp; 2264 #ifdef TCPPCAP 2265 /* 2266 * Init the TCP PCAP queues. 2267 */ 2268 tcp_pcap_tcpcb_init(tp); 2269 #endif 2270 #ifdef TCP_BLACKBOX 2271 /* Initialize the per-TCPCB log data. */ 2272 tcp_log_tcpcbinit(tp); 2273 #endif 2274 tp->t_pacing_rate = -1; 2275 if (tp->t_fb->tfb_tcp_fb_init) { 2276 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2277 refcount_release(&tp->t_fb->tfb_refcnt); 2278 in_pcbrele_wlocked(inp); 2279 uma_zfree(V_tcpcb_zone, tm); 2280 return (NULL); 2281 } 2282 } 2283 #ifdef STATS 2284 if (V_tcp_perconn_stats_enable == 1) 2285 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2286 #endif 2287 if (V_tcp_do_lrd) 2288 tp->t_flags |= TF_LRD; 2289 return (tp); /* XXX */ 2290 } 2291 2292 /* 2293 * Switch the congestion control algorithm back to Vnet default for any active 2294 * control blocks using an algorithm which is about to go away. If the algorithm 2295 * has a cb_init function and it fails (no memory) then the operation fails and 2296 * the unload will not succeed. 2297 * 2298 */ 2299 int 2300 tcp_ccalgounload(struct cc_algo *unload_algo) 2301 { 2302 struct cc_algo *oldalgo, *newalgo; 2303 struct inpcb *inp; 2304 struct tcpcb *tp; 2305 VNET_ITERATOR_DECL(vnet_iter); 2306 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2307 INPLOOKUP_WLOCKPCB); 2308 2309 /* 2310 * Check all active control blocks across all network stacks and change 2311 * any that are using "unload_algo" back to its default. If "unload_algo" 2312 * requires cleanup code to be run, call it. 2313 */ 2314 VNET_LIST_RLOCK(); 2315 VNET_FOREACH(vnet_iter) { 2316 CURVNET_SET(vnet_iter); 2317 /* 2318 * XXXGL: would new accept(2)d connections use algo being 2319 * unloaded? 2320 */ 2321 newalgo = CC_DEFAULT_ALGO(); 2322 while ((inp = inp_next(&inpi)) != NULL) { 2323 /* Important to skip tcptw structs. */ 2324 if (!(inp->inp_flags & INP_TIMEWAIT) && 2325 (tp = intotcpcb(inp)) != NULL) { 2326 /* 2327 * By holding INP_WLOCK here, we are assured 2328 * that the connection is not currently 2329 * executing inside the CC module's functions. 2330 * We attempt to switch to the Vnets default, 2331 * if the init fails then we fail the whole 2332 * operation and the module unload will fail. 2333 */ 2334 if (CC_ALGO(tp) == unload_algo) { 2335 struct cc_var cc_mem; 2336 int err; 2337 2338 oldalgo = CC_ALGO(tp); 2339 memset(&cc_mem, 0, sizeof(cc_mem)); 2340 cc_mem.ccvc.tcp = tp; 2341 if (newalgo->cb_init == NULL) { 2342 /* 2343 * No init we can skip the 2344 * dance around a possible failure. 2345 */ 2346 CC_DATA(tp) = NULL; 2347 goto proceed; 2348 } 2349 err = (newalgo->cb_init)(&cc_mem, NULL); 2350 if (err) { 2351 /* 2352 * Presumably no memory the caller will 2353 * need to try again. 2354 */ 2355 INP_WUNLOCK(inp); 2356 CURVNET_RESTORE(); 2357 VNET_LIST_RUNLOCK(); 2358 return (err); 2359 } 2360 proceed: 2361 if (oldalgo->cb_destroy != NULL) 2362 oldalgo->cb_destroy(tp->ccv); 2363 CC_ALGO(tp) = newalgo; 2364 memcpy(tp->ccv, &cc_mem, sizeof(struct cc_var)); 2365 if (TCPS_HAVEESTABLISHED(tp->t_state) && 2366 (CC_ALGO(tp)->conn_init != NULL)) { 2367 /* Yep run the connection init for the new CC */ 2368 CC_ALGO(tp)->conn_init(tp->ccv); 2369 } 2370 } 2371 } 2372 } 2373 CURVNET_RESTORE(); 2374 } 2375 VNET_LIST_RUNLOCK(); 2376 return (0); 2377 } 2378 2379 /* 2380 * Drop a TCP connection, reporting 2381 * the specified error. If connection is synchronized, 2382 * then send a RST to peer. 2383 */ 2384 struct tcpcb * 2385 tcp_drop(struct tcpcb *tp, int errno) 2386 { 2387 struct socket *so = tp->t_inpcb->inp_socket; 2388 2389 NET_EPOCH_ASSERT(); 2390 INP_WLOCK_ASSERT(tp->t_inpcb); 2391 2392 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2393 tcp_state_change(tp, TCPS_CLOSED); 2394 (void) tp->t_fb->tfb_tcp_output(tp); 2395 TCPSTAT_INC(tcps_drops); 2396 } else 2397 TCPSTAT_INC(tcps_conndrops); 2398 if (errno == ETIMEDOUT && tp->t_softerror) 2399 errno = tp->t_softerror; 2400 so->so_error = errno; 2401 return (tcp_close(tp)); 2402 } 2403 2404 void 2405 tcp_discardcb(struct tcpcb *tp) 2406 { 2407 struct inpcb *inp = tp->t_inpcb; 2408 2409 INP_WLOCK_ASSERT(inp); 2410 2411 /* 2412 * Make sure that all of our timers are stopped before we delete the 2413 * PCB. 2414 * 2415 * If stopping a timer fails, we schedule a discard function in same 2416 * callout, and the last discard function called will take care of 2417 * deleting the tcpcb. 2418 */ 2419 tp->t_timers->tt_draincnt = 0; 2420 tcp_timer_stop(tp, TT_REXMT); 2421 tcp_timer_stop(tp, TT_PERSIST); 2422 tcp_timer_stop(tp, TT_KEEP); 2423 tcp_timer_stop(tp, TT_2MSL); 2424 tcp_timer_stop(tp, TT_DELACK); 2425 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2426 /* 2427 * Call the stop-all function of the methods, 2428 * this function should call the tcp_timer_stop() 2429 * method with each of the function specific timeouts. 2430 * That stop will be called via the tfb_tcp_timer_stop() 2431 * which should use the async drain function of the 2432 * callout system (see tcp_var.h). 2433 */ 2434 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2435 } 2436 2437 /* free the reassembly queue, if any */ 2438 tcp_reass_flush(tp); 2439 2440 #ifdef TCP_OFFLOAD 2441 /* Disconnect offload device, if any. */ 2442 if (tp->t_flags & TF_TOE) 2443 tcp_offload_detach(tp); 2444 #endif 2445 2446 tcp_free_sackholes(tp); 2447 2448 #ifdef TCPPCAP 2449 /* Free the TCP PCAP queues. */ 2450 tcp_pcap_drain(&(tp->t_inpkts)); 2451 tcp_pcap_drain(&(tp->t_outpkts)); 2452 #endif 2453 2454 /* Allow the CC algorithm to clean up after itself. */ 2455 if (CC_ALGO(tp)->cb_destroy != NULL) 2456 CC_ALGO(tp)->cb_destroy(tp->ccv); 2457 CC_DATA(tp) = NULL; 2458 2459 #ifdef TCP_HHOOK 2460 khelp_destroy_osd(tp->osd); 2461 #endif 2462 #ifdef STATS 2463 stats_blob_destroy(tp->t_stats); 2464 #endif 2465 2466 CC_ALGO(tp) = NULL; 2467 inp->inp_ppcb = NULL; 2468 if (tp->t_timers->tt_draincnt == 0) { 2469 bool released __diagused; 2470 2471 released = tcp_freecb(tp); 2472 KASSERT(!released, ("%s: inp %p should not have been released " 2473 "here", __func__, inp)); 2474 } 2475 } 2476 2477 bool 2478 tcp_freecb(struct tcpcb *tp) 2479 { 2480 struct inpcb *inp = tp->t_inpcb; 2481 struct socket *so = inp->inp_socket; 2482 #ifdef INET6 2483 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2484 #endif 2485 2486 INP_WLOCK_ASSERT(inp); 2487 MPASS(tp->t_timers->tt_draincnt == 0); 2488 2489 /* We own the last reference on tcpcb, let's free it. */ 2490 #ifdef TCP_BLACKBOX 2491 tcp_log_tcpcbfini(tp); 2492 #endif 2493 TCPSTATES_DEC(tp->t_state); 2494 if (tp->t_fb->tfb_tcp_fb_fini) 2495 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2496 2497 /* 2498 * If we got enough samples through the srtt filter, 2499 * save the rtt and rttvar in the routing entry. 2500 * 'Enough' is arbitrarily defined as 4 rtt samples. 2501 * 4 samples is enough for the srtt filter to converge 2502 * to within enough % of the correct value; fewer samples 2503 * and we could save a bogus rtt. The danger is not high 2504 * as tcp quickly recovers from everything. 2505 * XXX: Works very well but needs some more statistics! 2506 * 2507 * XXXRRS: Updating must be after the stack fini() since 2508 * that may be converting some internal representation of 2509 * say srtt etc into the general one used by other stacks. 2510 * Lets also at least protect against the so being NULL 2511 * as RW stated below. 2512 */ 2513 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2514 struct hc_metrics_lite metrics; 2515 uint32_t ssthresh; 2516 2517 bzero(&metrics, sizeof(metrics)); 2518 /* 2519 * Update the ssthresh always when the conditions below 2520 * are satisfied. This gives us better new start value 2521 * for the congestion avoidance for new connections. 2522 * ssthresh is only set if packet loss occurred on a session. 2523 * 2524 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2525 * being torn down. Ideally this code would not use 'so'. 2526 */ 2527 ssthresh = tp->snd_ssthresh; 2528 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2529 /* 2530 * convert the limit from user data bytes to 2531 * packets then to packet data bytes. 2532 */ 2533 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2534 if (ssthresh < 2) 2535 ssthresh = 2; 2536 ssthresh *= (tp->t_maxseg + 2537 #ifdef INET6 2538 (isipv6 ? sizeof (struct ip6_hdr) + 2539 sizeof (struct tcphdr) : 2540 #endif 2541 sizeof (struct tcpiphdr) 2542 #ifdef INET6 2543 ) 2544 #endif 2545 ); 2546 } else 2547 ssthresh = 0; 2548 metrics.rmx_ssthresh = ssthresh; 2549 2550 metrics.rmx_rtt = tp->t_srtt; 2551 metrics.rmx_rttvar = tp->t_rttvar; 2552 metrics.rmx_cwnd = tp->snd_cwnd; 2553 metrics.rmx_sendpipe = 0; 2554 metrics.rmx_recvpipe = 0; 2555 2556 tcp_hc_update(&inp->inp_inc, &metrics); 2557 } 2558 2559 refcount_release(&tp->t_fb->tfb_refcnt); 2560 uma_zfree(V_tcpcb_zone, tp); 2561 2562 return (in_pcbrele_wlocked(inp)); 2563 } 2564 2565 /* 2566 * Attempt to close a TCP control block, marking it as dropped, and freeing 2567 * the socket if we hold the only reference. 2568 */ 2569 struct tcpcb * 2570 tcp_close(struct tcpcb *tp) 2571 { 2572 struct inpcb *inp = tp->t_inpcb; 2573 struct socket *so; 2574 2575 INP_WLOCK_ASSERT(inp); 2576 2577 #ifdef TCP_OFFLOAD 2578 if (tp->t_state == TCPS_LISTEN) 2579 tcp_offload_listen_stop(tp); 2580 #endif 2581 /* 2582 * This releases the TFO pending counter resource for TFO listen 2583 * sockets as well as passively-created TFO sockets that transition 2584 * from SYN_RECEIVED to CLOSED. 2585 */ 2586 if (tp->t_tfo_pending) { 2587 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2588 tp->t_tfo_pending = NULL; 2589 } 2590 #ifdef TCPHPTS 2591 tcp_hpts_remove(inp, HPTS_REMOVE_ALL); 2592 #endif 2593 in_pcbdrop(inp); 2594 TCPSTAT_INC(tcps_closed); 2595 if (tp->t_state != TCPS_CLOSED) 2596 tcp_state_change(tp, TCPS_CLOSED); 2597 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2598 so = inp->inp_socket; 2599 soisdisconnected(so); 2600 if (inp->inp_flags & INP_SOCKREF) { 2601 KASSERT(so->so_state & SS_PROTOREF, 2602 ("tcp_close: !SS_PROTOREF")); 2603 inp->inp_flags &= ~INP_SOCKREF; 2604 INP_WUNLOCK(inp); 2605 SOCK_LOCK(so); 2606 so->so_state &= ~SS_PROTOREF; 2607 sofree(so); 2608 return (NULL); 2609 } 2610 return (tp); 2611 } 2612 2613 void 2614 tcp_drain(void) 2615 { 2616 VNET_ITERATOR_DECL(vnet_iter); 2617 2618 if (!do_tcpdrain) 2619 return; 2620 2621 VNET_LIST_RLOCK_NOSLEEP(); 2622 VNET_FOREACH(vnet_iter) { 2623 CURVNET_SET(vnet_iter); 2624 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2625 INPLOOKUP_WLOCKPCB); 2626 struct inpcb *inpb; 2627 struct tcpcb *tcpb; 2628 2629 /* 2630 * Walk the tcpbs, if existing, and flush the reassembly queue, 2631 * if there is one... 2632 * XXX: The "Net/3" implementation doesn't imply that the TCP 2633 * reassembly queue should be flushed, but in a situation 2634 * where we're really low on mbufs, this is potentially 2635 * useful. 2636 */ 2637 while ((inpb = inp_next(&inpi)) != NULL) { 2638 if (inpb->inp_flags & INP_TIMEWAIT) 2639 continue; 2640 if ((tcpb = intotcpcb(inpb)) != NULL) { 2641 tcp_reass_flush(tcpb); 2642 tcp_clean_sackreport(tcpb); 2643 #ifdef TCP_BLACKBOX 2644 tcp_log_drain(tcpb); 2645 #endif 2646 #ifdef TCPPCAP 2647 if (tcp_pcap_aggressive_free) { 2648 /* Free the TCP PCAP queues. */ 2649 tcp_pcap_drain(&(tcpb->t_inpkts)); 2650 tcp_pcap_drain(&(tcpb->t_outpkts)); 2651 } 2652 #endif 2653 } 2654 } 2655 CURVNET_RESTORE(); 2656 } 2657 VNET_LIST_RUNLOCK_NOSLEEP(); 2658 } 2659 2660 /* 2661 * Notify a tcp user of an asynchronous error; 2662 * store error as soft error, but wake up user 2663 * (for now, won't do anything until can select for soft error). 2664 * 2665 * Do not wake up user since there currently is no mechanism for 2666 * reporting soft errors (yet - a kqueue filter may be added). 2667 */ 2668 static struct inpcb * 2669 tcp_notify(struct inpcb *inp, int error) 2670 { 2671 struct tcpcb *tp; 2672 2673 INP_WLOCK_ASSERT(inp); 2674 2675 if ((inp->inp_flags & INP_TIMEWAIT) || 2676 (inp->inp_flags & INP_DROPPED)) 2677 return (inp); 2678 2679 tp = intotcpcb(inp); 2680 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2681 2682 /* 2683 * Ignore some errors if we are hooked up. 2684 * If connection hasn't completed, has retransmitted several times, 2685 * and receives a second error, give up now. This is better 2686 * than waiting a long time to establish a connection that 2687 * can never complete. 2688 */ 2689 if (tp->t_state == TCPS_ESTABLISHED && 2690 (error == EHOSTUNREACH || error == ENETUNREACH || 2691 error == EHOSTDOWN)) { 2692 if (inp->inp_route.ro_nh) { 2693 NH_FREE(inp->inp_route.ro_nh); 2694 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2695 } 2696 return (inp); 2697 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2698 tp->t_softerror) { 2699 tp = tcp_drop(tp, error); 2700 if (tp != NULL) 2701 return (inp); 2702 else 2703 return (NULL); 2704 } else { 2705 tp->t_softerror = error; 2706 return (inp); 2707 } 2708 #if 0 2709 wakeup( &so->so_timeo); 2710 sorwakeup(so); 2711 sowwakeup(so); 2712 #endif 2713 } 2714 2715 static int 2716 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2717 { 2718 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2719 INPLOOKUP_RLOCKPCB); 2720 struct xinpgen xig; 2721 struct inpcb *inp; 2722 int error; 2723 2724 if (req->newptr != NULL) 2725 return (EPERM); 2726 2727 if (req->oldptr == NULL) { 2728 int n; 2729 2730 n = V_tcbinfo.ipi_count + 2731 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2732 n += imax(n / 8, 10); 2733 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2734 return (0); 2735 } 2736 2737 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2738 return (error); 2739 2740 bzero(&xig, sizeof(xig)); 2741 xig.xig_len = sizeof xig; 2742 xig.xig_count = V_tcbinfo.ipi_count + 2743 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2744 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2745 xig.xig_sogen = so_gencnt; 2746 error = SYSCTL_OUT(req, &xig, sizeof xig); 2747 if (error) 2748 return (error); 2749 2750 error = syncache_pcblist(req); 2751 if (error) 2752 return (error); 2753 2754 while ((inp = inp_next(&inpi)) != NULL) { 2755 if (inp->inp_gencnt <= xig.xig_gen) { 2756 int crerr; 2757 2758 /* 2759 * XXX: This use of cr_cansee(), introduced with 2760 * TCP state changes, is not quite right, but for 2761 * now, better than nothing. 2762 */ 2763 if (inp->inp_flags & INP_TIMEWAIT) { 2764 if (intotw(inp) != NULL) 2765 crerr = cr_cansee(req->td->td_ucred, 2766 intotw(inp)->tw_cred); 2767 else 2768 crerr = EINVAL; /* Skip this inp. */ 2769 } else 2770 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2771 if (crerr == 0) { 2772 struct xtcpcb xt; 2773 2774 tcp_inptoxtp(inp, &xt); 2775 error = SYSCTL_OUT(req, &xt, sizeof xt); 2776 if (error) { 2777 INP_RUNLOCK(inp); 2778 break; 2779 } else 2780 continue; 2781 } 2782 } 2783 } 2784 2785 if (!error) { 2786 /* 2787 * Give the user an updated idea of our state. 2788 * If the generation differs from what we told 2789 * her before, she knows that something happened 2790 * while we were processing this request, and it 2791 * might be necessary to retry. 2792 */ 2793 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2794 xig.xig_sogen = so_gencnt; 2795 xig.xig_count = V_tcbinfo.ipi_count + 2796 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2797 error = SYSCTL_OUT(req, &xig, sizeof xig); 2798 } 2799 2800 return (error); 2801 } 2802 2803 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2804 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2805 NULL, 0, tcp_pcblist, "S,xtcpcb", 2806 "List of active TCP connections"); 2807 2808 #ifdef INET 2809 static int 2810 tcp_getcred(SYSCTL_HANDLER_ARGS) 2811 { 2812 struct xucred xuc; 2813 struct sockaddr_in addrs[2]; 2814 struct epoch_tracker et; 2815 struct inpcb *inp; 2816 int error; 2817 2818 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2819 if (error) 2820 return (error); 2821 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2822 if (error) 2823 return (error); 2824 NET_EPOCH_ENTER(et); 2825 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2826 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2827 NET_EPOCH_EXIT(et); 2828 if (inp != NULL) { 2829 if (inp->inp_socket == NULL) 2830 error = ENOENT; 2831 if (error == 0) 2832 error = cr_canseeinpcb(req->td->td_ucred, inp); 2833 if (error == 0) 2834 cru2x(inp->inp_cred, &xuc); 2835 INP_RUNLOCK(inp); 2836 } else 2837 error = ENOENT; 2838 if (error == 0) 2839 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2840 return (error); 2841 } 2842 2843 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2844 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2845 0, 0, tcp_getcred, "S,xucred", 2846 "Get the xucred of a TCP connection"); 2847 #endif /* INET */ 2848 2849 #ifdef INET6 2850 static int 2851 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2852 { 2853 struct epoch_tracker et; 2854 struct xucred xuc; 2855 struct sockaddr_in6 addrs[2]; 2856 struct inpcb *inp; 2857 int error; 2858 #ifdef INET 2859 int mapped = 0; 2860 #endif 2861 2862 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2863 if (error) 2864 return (error); 2865 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2866 if (error) 2867 return (error); 2868 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2869 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2870 return (error); 2871 } 2872 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2873 #ifdef INET 2874 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2875 mapped = 1; 2876 else 2877 #endif 2878 return (EINVAL); 2879 } 2880 2881 NET_EPOCH_ENTER(et); 2882 #ifdef INET 2883 if (mapped == 1) 2884 inp = in_pcblookup(&V_tcbinfo, 2885 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2886 addrs[1].sin6_port, 2887 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2888 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2889 else 2890 #endif 2891 inp = in6_pcblookup(&V_tcbinfo, 2892 &addrs[1].sin6_addr, addrs[1].sin6_port, 2893 &addrs[0].sin6_addr, addrs[0].sin6_port, 2894 INPLOOKUP_RLOCKPCB, NULL); 2895 NET_EPOCH_EXIT(et); 2896 if (inp != NULL) { 2897 if (inp->inp_socket == NULL) 2898 error = ENOENT; 2899 if (error == 0) 2900 error = cr_canseeinpcb(req->td->td_ucred, inp); 2901 if (error == 0) 2902 cru2x(inp->inp_cred, &xuc); 2903 INP_RUNLOCK(inp); 2904 } else 2905 error = ENOENT; 2906 if (error == 0) 2907 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2908 return (error); 2909 } 2910 2911 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2912 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2913 0, 0, tcp6_getcred, "S,xucred", 2914 "Get the xucred of a TCP6 connection"); 2915 #endif /* INET6 */ 2916 2917 #ifdef INET 2918 /* Path MTU to try next when a fragmentation-needed message is received. */ 2919 static inline int 2920 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2921 { 2922 int mtu = ntohs(icp->icmp_nextmtu); 2923 2924 /* If no alternative MTU was proposed, try the next smaller one. */ 2925 if (!mtu) 2926 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2927 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2928 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2929 2930 return (mtu); 2931 } 2932 2933 static void 2934 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port) 2935 { 2936 struct ip *ip = vip; 2937 struct tcphdr *th; 2938 struct in_addr faddr; 2939 struct inpcb *inp; 2940 struct tcpcb *tp; 2941 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2942 struct icmp *icp; 2943 struct in_conninfo inc; 2944 tcp_seq icmp_tcp_seq; 2945 int mtu; 2946 2947 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2948 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2949 return; 2950 2951 if (cmd == PRC_MSGSIZE) 2952 notify = tcp_mtudisc_notify; 2953 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2954 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2955 cmd == PRC_TIMXCEED_INTRANS) && ip) 2956 notify = tcp_drop_syn_sent; 2957 2958 /* 2959 * Hostdead is ugly because it goes linearly through all PCBs. 2960 * XXX: We never get this from ICMP, otherwise it makes an 2961 * excellent DoS attack on machines with many connections. 2962 */ 2963 else if (cmd == PRC_HOSTDEAD) 2964 ip = NULL; 2965 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2966 return; 2967 2968 if (ip == NULL) { 2969 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2970 return; 2971 } 2972 2973 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2974 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2975 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2976 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2977 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2978 /* signal EHOSTDOWN, as it flushes the cached route */ 2979 inp = (*notify)(inp, EHOSTDOWN); 2980 goto out; 2981 } 2982 icmp_tcp_seq = th->th_seq; 2983 if (inp != NULL) { 2984 if (!(inp->inp_flags & INP_TIMEWAIT) && 2985 !(inp->inp_flags & INP_DROPPED) && 2986 !(inp->inp_socket == NULL)) { 2987 tp = intotcpcb(inp); 2988 #ifdef TCP_OFFLOAD 2989 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 2990 /* 2991 * MTU discovery for offloaded connections. Let 2992 * the TOE driver verify seq# and process it. 2993 */ 2994 mtu = tcp_next_pmtu(icp, ip); 2995 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2996 goto out; 2997 } 2998 #endif 2999 if (tp->t_port != port) { 3000 goto out; 3001 } 3002 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3003 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3004 if (cmd == PRC_MSGSIZE) { 3005 /* 3006 * MTU discovery: we got a needfrag and 3007 * will potentially try a lower MTU. 3008 */ 3009 mtu = tcp_next_pmtu(icp, ip); 3010 3011 /* 3012 * Only process the offered MTU if it 3013 * is smaller than the current one. 3014 */ 3015 if (mtu < tp->t_maxseg + 3016 sizeof(struct tcpiphdr)) { 3017 bzero(&inc, sizeof(inc)); 3018 inc.inc_faddr = faddr; 3019 inc.inc_fibnum = 3020 inp->inp_inc.inc_fibnum; 3021 tcp_hc_updatemtu(&inc, mtu); 3022 tcp_mtudisc(inp, mtu); 3023 } 3024 } else 3025 inp = (*notify)(inp, 3026 inetctlerrmap[cmd]); 3027 } 3028 } 3029 } else { 3030 bzero(&inc, sizeof(inc)); 3031 inc.inc_fport = th->th_dport; 3032 inc.inc_lport = th->th_sport; 3033 inc.inc_faddr = faddr; 3034 inc.inc_laddr = ip->ip_src; 3035 syncache_unreach(&inc, icmp_tcp_seq, port); 3036 } 3037 out: 3038 if (inp != NULL) 3039 INP_WUNLOCK(inp); 3040 } 3041 3042 void 3043 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 3044 { 3045 tcp_ctlinput_with_port(cmd, sa, vip, htons(0)); 3046 } 3047 3048 void 3049 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused) 3050 { 3051 /* Its a tunneled TCP over UDP icmp */ 3052 struct ip *outer_ip, *inner_ip; 3053 struct icmp *icmp; 3054 struct udphdr *udp; 3055 struct tcphdr *th, ttemp; 3056 int i_hlen, o_len; 3057 uint16_t port; 3058 3059 inner_ip = (struct ip *)vip; 3060 icmp = (struct icmp *)((caddr_t)inner_ip - 3061 (sizeof(struct icmp) - sizeof(struct ip))); 3062 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 3063 i_hlen = inner_ip->ip_hl << 2; 3064 o_len = ntohs(outer_ip->ip_len); 3065 if (o_len < 3066 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 3067 /* Not enough data present */ 3068 return; 3069 } 3070 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 3071 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 3072 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3073 return; 3074 } 3075 port = udp->uh_dport; 3076 th = (struct tcphdr *)(udp + 1); 3077 memcpy(&ttemp, th, sizeof(struct tcphdr)); 3078 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 3079 /* Now adjust down the size of the outer IP header */ 3080 o_len -= sizeof(struct udphdr); 3081 outer_ip->ip_len = htons(o_len); 3082 /* Now call in to the normal handling code */ 3083 tcp_ctlinput_with_port(cmd, sa, vip, port); 3084 } 3085 #endif /* INET */ 3086 3087 #ifdef INET6 3088 static inline int 3089 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3090 { 3091 int mtu = ntohl(icmp6->icmp6_mtu); 3092 3093 /* 3094 * If no alternative MTU was proposed, or the proposed MTU was too 3095 * small, set to the min. 3096 */ 3097 if (mtu < IPV6_MMTU) 3098 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 3099 return (mtu); 3100 } 3101 3102 static void 3103 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port) 3104 { 3105 struct in6_addr *dst; 3106 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 3107 struct ip6_hdr *ip6; 3108 struct mbuf *m; 3109 struct inpcb *inp; 3110 struct tcpcb *tp; 3111 struct icmp6_hdr *icmp6; 3112 struct ip6ctlparam *ip6cp = NULL; 3113 const struct sockaddr_in6 *sa6_src = NULL; 3114 struct in_conninfo inc; 3115 struct tcp_ports { 3116 uint16_t th_sport; 3117 uint16_t th_dport; 3118 } t_ports; 3119 tcp_seq icmp_tcp_seq; 3120 unsigned int mtu; 3121 unsigned int off; 3122 3123 if (sa->sa_family != AF_INET6 || 3124 sa->sa_len != sizeof(struct sockaddr_in6)) 3125 return; 3126 3127 /* if the parameter is from icmp6, decode it. */ 3128 if (d != NULL) { 3129 ip6cp = (struct ip6ctlparam *)d; 3130 icmp6 = ip6cp->ip6c_icmp6; 3131 m = ip6cp->ip6c_m; 3132 ip6 = ip6cp->ip6c_ip6; 3133 off = ip6cp->ip6c_off; 3134 sa6_src = ip6cp->ip6c_src; 3135 dst = ip6cp->ip6c_finaldst; 3136 } else { 3137 m = NULL; 3138 ip6 = NULL; 3139 off = 0; /* fool gcc */ 3140 sa6_src = &sa6_any; 3141 dst = NULL; 3142 } 3143 3144 if (cmd == PRC_MSGSIZE) 3145 notify = tcp_mtudisc_notify; 3146 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 3147 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 3148 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 3149 notify = tcp_drop_syn_sent; 3150 3151 /* 3152 * Hostdead is ugly because it goes linearly through all PCBs. 3153 * XXX: We never get this from ICMP, otherwise it makes an 3154 * excellent DoS attack on machines with many connections. 3155 */ 3156 else if (cmd == PRC_HOSTDEAD) 3157 ip6 = NULL; 3158 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 3159 return; 3160 3161 if (ip6 == NULL) { 3162 in6_pcbnotify(&V_tcbinfo, sa, 0, 3163 (const struct sockaddr *)sa6_src, 3164 0, cmd, NULL, notify); 3165 return; 3166 } 3167 3168 /* Check if we can safely get the ports from the tcp hdr */ 3169 if (m == NULL || 3170 (m->m_pkthdr.len < 3171 (int32_t) (off + sizeof(struct tcp_ports)))) { 3172 return; 3173 } 3174 bzero(&t_ports, sizeof(struct tcp_ports)); 3175 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3176 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3177 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3178 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 3179 /* signal EHOSTDOWN, as it flushes the cached route */ 3180 inp = (*notify)(inp, EHOSTDOWN); 3181 goto out; 3182 } 3183 off += sizeof(struct tcp_ports); 3184 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3185 goto out; 3186 } 3187 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3188 if (inp != NULL) { 3189 if (!(inp->inp_flags & INP_TIMEWAIT) && 3190 !(inp->inp_flags & INP_DROPPED) && 3191 !(inp->inp_socket == NULL)) { 3192 tp = intotcpcb(inp); 3193 #ifdef TCP_OFFLOAD 3194 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 3195 /* MTU discovery for offloaded connections. */ 3196 mtu = tcp6_next_pmtu(icmp6); 3197 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3198 goto out; 3199 } 3200 #endif 3201 if (tp->t_port != port) { 3202 goto out; 3203 } 3204 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3205 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3206 if (cmd == PRC_MSGSIZE) { 3207 /* 3208 * MTU discovery: 3209 * If we got a needfrag set the MTU 3210 * in the route to the suggested new 3211 * value (if given) and then notify. 3212 */ 3213 mtu = tcp6_next_pmtu(icmp6); 3214 3215 bzero(&inc, sizeof(inc)); 3216 inc.inc_fibnum = M_GETFIB(m); 3217 inc.inc_flags |= INC_ISIPV6; 3218 inc.inc6_faddr = *dst; 3219 if (in6_setscope(&inc.inc6_faddr, 3220 m->m_pkthdr.rcvif, NULL)) 3221 goto out; 3222 /* 3223 * Only process the offered MTU if it 3224 * is smaller than the current one. 3225 */ 3226 if (mtu < tp->t_maxseg + 3227 sizeof (struct tcphdr) + 3228 sizeof (struct ip6_hdr)) { 3229 tcp_hc_updatemtu(&inc, mtu); 3230 tcp_mtudisc(inp, mtu); 3231 ICMP6STAT_INC(icp6s_pmtuchg); 3232 } 3233 } else 3234 inp = (*notify)(inp, 3235 inet6ctlerrmap[cmd]); 3236 } 3237 } 3238 } else { 3239 bzero(&inc, sizeof(inc)); 3240 inc.inc_fibnum = M_GETFIB(m); 3241 inc.inc_flags |= INC_ISIPV6; 3242 inc.inc_fport = t_ports.th_dport; 3243 inc.inc_lport = t_ports.th_sport; 3244 inc.inc6_faddr = *dst; 3245 inc.inc6_laddr = ip6->ip6_src; 3246 syncache_unreach(&inc, icmp_tcp_seq, port); 3247 } 3248 out: 3249 if (inp != NULL) 3250 INP_WUNLOCK(inp); 3251 } 3252 3253 void 3254 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 3255 { 3256 tcp6_ctlinput_with_port(cmd, sa, d, htons(0)); 3257 } 3258 3259 void 3260 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused) 3261 { 3262 struct ip6ctlparam *ip6cp; 3263 struct mbuf *m; 3264 struct udphdr *udp; 3265 uint16_t port; 3266 3267 ip6cp = (struct ip6ctlparam *)d; 3268 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3269 if (m == NULL) { 3270 return; 3271 } 3272 udp = mtod(m, struct udphdr *); 3273 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3274 return; 3275 } 3276 port = udp->uh_dport; 3277 m_adj(m, sizeof(struct udphdr)); 3278 if ((m->m_flags & M_PKTHDR) == 0) { 3279 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3280 } 3281 /* Now call in to the normal handling code */ 3282 tcp6_ctlinput_with_port(cmd, sa, d, port); 3283 } 3284 3285 #endif /* INET6 */ 3286 3287 static uint32_t 3288 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3289 { 3290 SIPHASH_CTX ctx; 3291 uint32_t hash[2]; 3292 3293 KASSERT(len >= SIPHASH_KEY_LENGTH, 3294 ("%s: keylen %u too short ", __func__, len)); 3295 SipHash24_Init(&ctx); 3296 SipHash_SetKey(&ctx, (uint8_t *)key); 3297 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3298 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3299 switch (inc->inc_flags & INC_ISIPV6) { 3300 #ifdef INET 3301 case 0: 3302 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3303 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3304 break; 3305 #endif 3306 #ifdef INET6 3307 case INC_ISIPV6: 3308 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3309 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3310 break; 3311 #endif 3312 } 3313 SipHash_Final((uint8_t *)hash, &ctx); 3314 3315 return (hash[0] ^ hash[1]); 3316 } 3317 3318 uint32_t 3319 tcp_new_ts_offset(struct in_conninfo *inc) 3320 { 3321 struct in_conninfo inc_store, *local_inc; 3322 3323 if (!V_tcp_ts_offset_per_conn) { 3324 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3325 inc_store.inc_lport = 0; 3326 inc_store.inc_fport = 0; 3327 local_inc = &inc_store; 3328 } else { 3329 local_inc = inc; 3330 } 3331 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3332 sizeof(V_ts_offset_secret))); 3333 } 3334 3335 /* 3336 * Following is where TCP initial sequence number generation occurs. 3337 * 3338 * There are two places where we must use initial sequence numbers: 3339 * 1. In SYN-ACK packets. 3340 * 2. In SYN packets. 3341 * 3342 * All ISNs for SYN-ACK packets are generated by the syncache. See 3343 * tcp_syncache.c for details. 3344 * 3345 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3346 * depends on this property. In addition, these ISNs should be 3347 * unguessable so as to prevent connection hijacking. To satisfy 3348 * the requirements of this situation, the algorithm outlined in 3349 * RFC 1948 is used, with only small modifications. 3350 * 3351 * Implementation details: 3352 * 3353 * Time is based off the system timer, and is corrected so that it 3354 * increases by one megabyte per second. This allows for proper 3355 * recycling on high speed LANs while still leaving over an hour 3356 * before rollover. 3357 * 3358 * As reading the *exact* system time is too expensive to be done 3359 * whenever setting up a TCP connection, we increment the time 3360 * offset in two ways. First, a small random positive increment 3361 * is added to isn_offset for each connection that is set up. 3362 * Second, the function tcp_isn_tick fires once per clock tick 3363 * and increments isn_offset as necessary so that sequence numbers 3364 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3365 * random positive increments serve only to ensure that the same 3366 * exact sequence number is never sent out twice (as could otherwise 3367 * happen when a port is recycled in less than the system tick 3368 * interval.) 3369 * 3370 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3371 * between seeding of isn_secret. This is normally set to zero, 3372 * as reseeding should not be necessary. 3373 * 3374 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3375 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3376 * general, this means holding an exclusive (write) lock. 3377 */ 3378 3379 #define ISN_BYTES_PER_SECOND 1048576 3380 #define ISN_STATIC_INCREMENT 4096 3381 #define ISN_RANDOM_INCREMENT (4096 - 1) 3382 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3383 3384 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3385 VNET_DEFINE_STATIC(int, isn_last); 3386 VNET_DEFINE_STATIC(int, isn_last_reseed); 3387 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3388 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3389 3390 #define V_isn_secret VNET(isn_secret) 3391 #define V_isn_last VNET(isn_last) 3392 #define V_isn_last_reseed VNET(isn_last_reseed) 3393 #define V_isn_offset VNET(isn_offset) 3394 #define V_isn_offset_old VNET(isn_offset_old) 3395 3396 tcp_seq 3397 tcp_new_isn(struct in_conninfo *inc) 3398 { 3399 tcp_seq new_isn; 3400 u_int32_t projected_offset; 3401 3402 ISN_LOCK(); 3403 /* Seed if this is the first use, reseed if requested. */ 3404 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3405 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3406 < (u_int)ticks))) { 3407 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3408 V_isn_last_reseed = ticks; 3409 } 3410 3411 /* Compute the hash and return the ISN. */ 3412 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3413 sizeof(V_isn_secret)); 3414 V_isn_offset += ISN_STATIC_INCREMENT + 3415 (arc4random() & ISN_RANDOM_INCREMENT); 3416 if (ticks != V_isn_last) { 3417 projected_offset = V_isn_offset_old + 3418 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3419 if (SEQ_GT(projected_offset, V_isn_offset)) 3420 V_isn_offset = projected_offset; 3421 V_isn_offset_old = V_isn_offset; 3422 V_isn_last = ticks; 3423 } 3424 new_isn += V_isn_offset; 3425 ISN_UNLOCK(); 3426 return (new_isn); 3427 } 3428 3429 /* 3430 * When a specific ICMP unreachable message is received and the 3431 * connection state is SYN-SENT, drop the connection. This behavior 3432 * is controlled by the icmp_may_rst sysctl. 3433 */ 3434 struct inpcb * 3435 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3436 { 3437 struct tcpcb *tp; 3438 3439 NET_EPOCH_ASSERT(); 3440 INP_WLOCK_ASSERT(inp); 3441 3442 if ((inp->inp_flags & INP_TIMEWAIT) || 3443 (inp->inp_flags & INP_DROPPED)) 3444 return (inp); 3445 3446 tp = intotcpcb(inp); 3447 if (tp->t_state != TCPS_SYN_SENT) 3448 return (inp); 3449 3450 if (IS_FASTOPEN(tp->t_flags)) 3451 tcp_fastopen_disable_path(tp); 3452 3453 tp = tcp_drop(tp, errno); 3454 if (tp != NULL) 3455 return (inp); 3456 else 3457 return (NULL); 3458 } 3459 3460 /* 3461 * When `need fragmentation' ICMP is received, update our idea of the MSS 3462 * based on the new value. Also nudge TCP to send something, since we 3463 * know the packet we just sent was dropped. 3464 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3465 */ 3466 static struct inpcb * 3467 tcp_mtudisc_notify(struct inpcb *inp, int error) 3468 { 3469 3470 tcp_mtudisc(inp, -1); 3471 return (inp); 3472 } 3473 3474 static void 3475 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3476 { 3477 struct tcpcb *tp; 3478 struct socket *so; 3479 3480 INP_WLOCK_ASSERT(inp); 3481 if ((inp->inp_flags & INP_TIMEWAIT) || 3482 (inp->inp_flags & INP_DROPPED)) 3483 return; 3484 3485 tp = intotcpcb(inp); 3486 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3487 3488 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3489 3490 so = inp->inp_socket; 3491 SOCKBUF_LOCK(&so->so_snd); 3492 /* If the mss is larger than the socket buffer, decrease the mss. */ 3493 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3494 tp->t_maxseg = so->so_snd.sb_hiwat; 3495 SOCKBUF_UNLOCK(&so->so_snd); 3496 3497 TCPSTAT_INC(tcps_mturesent); 3498 tp->t_rtttime = 0; 3499 tp->snd_nxt = tp->snd_una; 3500 tcp_free_sackholes(tp); 3501 tp->snd_recover = tp->snd_max; 3502 if (tp->t_flags & TF_SACK_PERMIT) 3503 EXIT_FASTRECOVERY(tp->t_flags); 3504 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3505 /* 3506 * Conceptually the snd_nxt setting 3507 * and freeing sack holes should 3508 * be done by the default stacks 3509 * own tfb_tcp_mtu_chg(). 3510 */ 3511 tp->t_fb->tfb_tcp_mtu_chg(tp); 3512 } 3513 tp->t_fb->tfb_tcp_output(tp); 3514 } 3515 3516 #ifdef INET 3517 /* 3518 * Look-up the routing entry to the peer of this inpcb. If no route 3519 * is found and it cannot be allocated, then return 0. This routine 3520 * is called by TCP routines that access the rmx structure and by 3521 * tcp_mss_update to get the peer/interface MTU. 3522 */ 3523 uint32_t 3524 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3525 { 3526 struct nhop_object *nh; 3527 struct ifnet *ifp; 3528 uint32_t maxmtu = 0; 3529 3530 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3531 3532 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3533 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3534 if (nh == NULL) 3535 return (0); 3536 3537 ifp = nh->nh_ifp; 3538 maxmtu = nh->nh_mtu; 3539 3540 /* Report additional interface capabilities. */ 3541 if (cap != NULL) { 3542 if (ifp->if_capenable & IFCAP_TSO4 && 3543 ifp->if_hwassist & CSUM_TSO) { 3544 cap->ifcap |= CSUM_TSO; 3545 cap->tsomax = ifp->if_hw_tsomax; 3546 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3547 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3548 } 3549 } 3550 } 3551 return (maxmtu); 3552 } 3553 #endif /* INET */ 3554 3555 #ifdef INET6 3556 uint32_t 3557 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3558 { 3559 struct nhop_object *nh; 3560 struct in6_addr dst6; 3561 uint32_t scopeid; 3562 struct ifnet *ifp; 3563 uint32_t maxmtu = 0; 3564 3565 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3566 3567 if (inc->inc_flags & INC_IPV6MINMTU) 3568 return (IPV6_MMTU); 3569 3570 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3571 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3572 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3573 if (nh == NULL) 3574 return (0); 3575 3576 ifp = nh->nh_ifp; 3577 maxmtu = nh->nh_mtu; 3578 3579 /* Report additional interface capabilities. */ 3580 if (cap != NULL) { 3581 if (ifp->if_capenable & IFCAP_TSO6 && 3582 ifp->if_hwassist & CSUM_TSO) { 3583 cap->ifcap |= CSUM_TSO; 3584 cap->tsomax = ifp->if_hw_tsomax; 3585 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3586 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3587 } 3588 } 3589 } 3590 3591 return (maxmtu); 3592 } 3593 3594 /* 3595 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3596 * 3597 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3598 * The right place to do that is ip6_setpktopt() that has just been 3599 * executed. By the way it just filled ip6po_minmtu for us. 3600 */ 3601 void 3602 tcp6_use_min_mtu(struct tcpcb *tp) 3603 { 3604 struct inpcb *inp = tp->t_inpcb; 3605 3606 INP_WLOCK_ASSERT(inp); 3607 /* 3608 * In case of the IPV6_USE_MIN_MTU socket 3609 * option, the INC_IPV6MINMTU flag to announce 3610 * a corresponding MSS during the initial 3611 * handshake. If the TCP connection is not in 3612 * the front states, just reduce the MSS being 3613 * used. This avoids the sending of TCP 3614 * segments which will be fragmented at the 3615 * IPv6 layer. 3616 */ 3617 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3618 if ((tp->t_state >= TCPS_SYN_SENT) && 3619 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3620 struct ip6_pktopts *opt; 3621 3622 opt = inp->in6p_outputopts; 3623 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3624 tp->t_maxseg > TCP6_MSS) 3625 tp->t_maxseg = TCP6_MSS; 3626 } 3627 } 3628 #endif /* INET6 */ 3629 3630 /* 3631 * Calculate effective SMSS per RFC5681 definition for a given TCP 3632 * connection at its current state, taking into account SACK and etc. 3633 */ 3634 u_int 3635 tcp_maxseg(const struct tcpcb *tp) 3636 { 3637 u_int optlen; 3638 3639 if (tp->t_flags & TF_NOOPT) 3640 return (tp->t_maxseg); 3641 3642 /* 3643 * Here we have a simplified code from tcp_addoptions(), 3644 * without a proper loop, and having most of paddings hardcoded. 3645 * We might make mistakes with padding here in some edge cases, 3646 * but this is harmless, since result of tcp_maxseg() is used 3647 * only in cwnd and ssthresh estimations. 3648 */ 3649 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3650 if (tp->t_flags & TF_RCVD_TSTMP) 3651 optlen = TCPOLEN_TSTAMP_APPA; 3652 else 3653 optlen = 0; 3654 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3655 if (tp->t_flags & TF_SIGNATURE) 3656 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3657 #endif 3658 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3659 optlen += TCPOLEN_SACKHDR; 3660 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3661 optlen = PADTCPOLEN(optlen); 3662 } 3663 } else { 3664 if (tp->t_flags & TF_REQ_TSTMP) 3665 optlen = TCPOLEN_TSTAMP_APPA; 3666 else 3667 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3668 if (tp->t_flags & TF_REQ_SCALE) 3669 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3670 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3671 if (tp->t_flags & TF_SIGNATURE) 3672 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3673 #endif 3674 if (tp->t_flags & TF_SACK_PERMIT) 3675 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3676 } 3677 #undef PAD 3678 optlen = min(optlen, TCP_MAXOLEN); 3679 return (tp->t_maxseg - optlen); 3680 } 3681 3682 3683 u_int 3684 tcp_fixed_maxseg(const struct tcpcb *tp) 3685 { 3686 int optlen; 3687 3688 if (tp->t_flags & TF_NOOPT) 3689 return (tp->t_maxseg); 3690 3691 /* 3692 * Here we have a simplified code from tcp_addoptions(), 3693 * without a proper loop, and having most of paddings hardcoded. 3694 * We only consider fixed options that we would send every 3695 * time I.e. SACK is not considered. This is important 3696 * for cc modules to figure out what the modulo of the 3697 * cwnd should be. 3698 */ 3699 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3700 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3701 if (tp->t_flags & TF_RCVD_TSTMP) 3702 optlen = TCPOLEN_TSTAMP_APPA; 3703 else 3704 optlen = 0; 3705 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3706 if (tp->t_flags & TF_SIGNATURE) 3707 optlen += PAD(TCPOLEN_SIGNATURE); 3708 #endif 3709 } else { 3710 if (tp->t_flags & TF_REQ_TSTMP) 3711 optlen = TCPOLEN_TSTAMP_APPA; 3712 else 3713 optlen = PAD(TCPOLEN_MAXSEG); 3714 if (tp->t_flags & TF_REQ_SCALE) 3715 optlen += PAD(TCPOLEN_WINDOW); 3716 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3717 if (tp->t_flags & TF_SIGNATURE) 3718 optlen += PAD(TCPOLEN_SIGNATURE); 3719 #endif 3720 if (tp->t_flags & TF_SACK_PERMIT) 3721 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3722 } 3723 #undef PAD 3724 optlen = min(optlen, TCP_MAXOLEN); 3725 return (tp->t_maxseg - optlen); 3726 } 3727 3728 3729 3730 static int 3731 sysctl_drop(SYSCTL_HANDLER_ARGS) 3732 { 3733 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3734 struct sockaddr_storage addrs[2]; 3735 struct inpcb *inp; 3736 struct tcpcb *tp; 3737 struct tcptw *tw; 3738 struct sockaddr_in *fin, *lin; 3739 struct epoch_tracker et; 3740 #ifdef INET6 3741 struct sockaddr_in6 *fin6, *lin6; 3742 #endif 3743 int error; 3744 3745 inp = NULL; 3746 fin = lin = NULL; 3747 #ifdef INET6 3748 fin6 = lin6 = NULL; 3749 #endif 3750 error = 0; 3751 3752 if (req->oldptr != NULL || req->oldlen != 0) 3753 return (EINVAL); 3754 if (req->newptr == NULL) 3755 return (EPERM); 3756 if (req->newlen < sizeof(addrs)) 3757 return (ENOMEM); 3758 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3759 if (error) 3760 return (error); 3761 3762 switch (addrs[0].ss_family) { 3763 #ifdef INET6 3764 case AF_INET6: 3765 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3766 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3767 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3768 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3769 return (EINVAL); 3770 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3771 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3772 return (EINVAL); 3773 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3774 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3775 fin = (struct sockaddr_in *)&addrs[0]; 3776 lin = (struct sockaddr_in *)&addrs[1]; 3777 break; 3778 } 3779 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3780 if (error) 3781 return (error); 3782 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3783 if (error) 3784 return (error); 3785 break; 3786 #endif 3787 #ifdef INET 3788 case AF_INET: 3789 fin = (struct sockaddr_in *)&addrs[0]; 3790 lin = (struct sockaddr_in *)&addrs[1]; 3791 if (fin->sin_len != sizeof(struct sockaddr_in) || 3792 lin->sin_len != sizeof(struct sockaddr_in)) 3793 return (EINVAL); 3794 break; 3795 #endif 3796 default: 3797 return (EINVAL); 3798 } 3799 NET_EPOCH_ENTER(et); 3800 switch (addrs[0].ss_family) { 3801 #ifdef INET6 3802 case AF_INET6: 3803 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3804 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3805 INPLOOKUP_WLOCKPCB, NULL); 3806 break; 3807 #endif 3808 #ifdef INET 3809 case AF_INET: 3810 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3811 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3812 break; 3813 #endif 3814 } 3815 if (inp != NULL) { 3816 if (inp->inp_flags & INP_TIMEWAIT) { 3817 /* 3818 * XXXRW: There currently exists a state where an 3819 * inpcb is present, but its timewait state has been 3820 * discarded. For now, don't allow dropping of this 3821 * type of inpcb. 3822 */ 3823 tw = intotw(inp); 3824 if (tw != NULL) 3825 tcp_twclose(tw, 0); 3826 else 3827 INP_WUNLOCK(inp); 3828 } else if ((inp->inp_flags & INP_DROPPED) == 0 && 3829 !SOLISTENING(inp->inp_socket)) { 3830 tp = intotcpcb(inp); 3831 tp = tcp_drop(tp, ECONNABORTED); 3832 if (tp != NULL) 3833 INP_WUNLOCK(inp); 3834 } else 3835 INP_WUNLOCK(inp); 3836 } else 3837 error = ESRCH; 3838 NET_EPOCH_EXIT(et); 3839 return (error); 3840 } 3841 3842 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3843 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3844 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3845 "Drop TCP connection"); 3846 3847 #ifdef KERN_TLS 3848 static int 3849 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3850 { 3851 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3852 struct sockaddr_storage addrs[2]; 3853 struct inpcb *inp; 3854 struct sockaddr_in *fin, *lin; 3855 struct epoch_tracker et; 3856 #ifdef INET6 3857 struct sockaddr_in6 *fin6, *lin6; 3858 #endif 3859 int error; 3860 3861 inp = NULL; 3862 fin = lin = NULL; 3863 #ifdef INET6 3864 fin6 = lin6 = NULL; 3865 #endif 3866 error = 0; 3867 3868 if (req->oldptr != NULL || req->oldlen != 0) 3869 return (EINVAL); 3870 if (req->newptr == NULL) 3871 return (EPERM); 3872 if (req->newlen < sizeof(addrs)) 3873 return (ENOMEM); 3874 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3875 if (error) 3876 return (error); 3877 3878 switch (addrs[0].ss_family) { 3879 #ifdef INET6 3880 case AF_INET6: 3881 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3882 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3883 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3884 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3885 return (EINVAL); 3886 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3887 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3888 return (EINVAL); 3889 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3890 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3891 fin = (struct sockaddr_in *)&addrs[0]; 3892 lin = (struct sockaddr_in *)&addrs[1]; 3893 break; 3894 } 3895 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3896 if (error) 3897 return (error); 3898 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3899 if (error) 3900 return (error); 3901 break; 3902 #endif 3903 #ifdef INET 3904 case AF_INET: 3905 fin = (struct sockaddr_in *)&addrs[0]; 3906 lin = (struct sockaddr_in *)&addrs[1]; 3907 if (fin->sin_len != sizeof(struct sockaddr_in) || 3908 lin->sin_len != sizeof(struct sockaddr_in)) 3909 return (EINVAL); 3910 break; 3911 #endif 3912 default: 3913 return (EINVAL); 3914 } 3915 NET_EPOCH_ENTER(et); 3916 switch (addrs[0].ss_family) { 3917 #ifdef INET6 3918 case AF_INET6: 3919 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3920 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3921 INPLOOKUP_WLOCKPCB, NULL); 3922 break; 3923 #endif 3924 #ifdef INET 3925 case AF_INET: 3926 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3927 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3928 break; 3929 #endif 3930 } 3931 NET_EPOCH_EXIT(et); 3932 if (inp != NULL) { 3933 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3934 inp->inp_socket == NULL) { 3935 error = ECONNRESET; 3936 INP_WUNLOCK(inp); 3937 } else { 3938 struct socket *so; 3939 3940 so = inp->inp_socket; 3941 soref(so); 3942 error = ktls_set_tx_mode(so, 3943 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3944 INP_WUNLOCK(inp); 3945 sorele(so); 3946 } 3947 } else 3948 error = ESRCH; 3949 return (error); 3950 } 3951 3952 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3953 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3954 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3955 "Switch TCP connection to SW TLS"); 3956 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3957 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3958 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3959 "Switch TCP connection to ifnet TLS"); 3960 #endif 3961 3962 /* 3963 * Generate a standardized TCP log line for use throughout the 3964 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3965 * allow use in the interrupt context. 3966 * 3967 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3968 * NB: The function may return NULL if memory allocation failed. 3969 * 3970 * Due to header inclusion and ordering limitations the struct ip 3971 * and ip6_hdr pointers have to be passed as void pointers. 3972 */ 3973 char * 3974 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3975 const void *ip6hdr) 3976 { 3977 3978 /* Is logging enabled? */ 3979 if (V_tcp_log_in_vain == 0) 3980 return (NULL); 3981 3982 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3983 } 3984 3985 char * 3986 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3987 const void *ip6hdr) 3988 { 3989 3990 /* Is logging enabled? */ 3991 if (tcp_log_debug == 0) 3992 return (NULL); 3993 3994 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3995 } 3996 3997 static char * 3998 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3999 const void *ip6hdr) 4000 { 4001 char *s, *sp; 4002 size_t size; 4003 struct ip *ip; 4004 #ifdef INET6 4005 const struct ip6_hdr *ip6; 4006 4007 ip6 = (const struct ip6_hdr *)ip6hdr; 4008 #endif /* INET6 */ 4009 ip = (struct ip *)ip4hdr; 4010 4011 /* 4012 * The log line looks like this: 4013 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 4014 */ 4015 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 4016 sizeof(PRINT_TH_FLAGS) + 1 + 4017 #ifdef INET6 4018 2 * INET6_ADDRSTRLEN; 4019 #else 4020 2 * INET_ADDRSTRLEN; 4021 #endif /* INET6 */ 4022 4023 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 4024 if (s == NULL) 4025 return (NULL); 4026 4027 strcat(s, "TCP: ["); 4028 sp = s + strlen(s); 4029 4030 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 4031 inet_ntoa_r(inc->inc_faddr, sp); 4032 sp = s + strlen(s); 4033 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4034 sp = s + strlen(s); 4035 inet_ntoa_r(inc->inc_laddr, sp); 4036 sp = s + strlen(s); 4037 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4038 #ifdef INET6 4039 } else if (inc) { 4040 ip6_sprintf(sp, &inc->inc6_faddr); 4041 sp = s + strlen(s); 4042 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4043 sp = s + strlen(s); 4044 ip6_sprintf(sp, &inc->inc6_laddr); 4045 sp = s + strlen(s); 4046 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4047 } else if (ip6 && th) { 4048 ip6_sprintf(sp, &ip6->ip6_src); 4049 sp = s + strlen(s); 4050 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4051 sp = s + strlen(s); 4052 ip6_sprintf(sp, &ip6->ip6_dst); 4053 sp = s + strlen(s); 4054 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4055 #endif /* INET6 */ 4056 #ifdef INET 4057 } else if (ip && th) { 4058 inet_ntoa_r(ip->ip_src, sp); 4059 sp = s + strlen(s); 4060 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4061 sp = s + strlen(s); 4062 inet_ntoa_r(ip->ip_dst, sp); 4063 sp = s + strlen(s); 4064 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4065 #endif /* INET */ 4066 } else { 4067 free(s, M_TCPLOG); 4068 return (NULL); 4069 } 4070 sp = s + strlen(s); 4071 if (th) 4072 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 4073 if (*(s + size - 1) != '\0') 4074 panic("%s: string too long", __func__); 4075 return (s); 4076 } 4077 4078 /* 4079 * A subroutine which makes it easy to track TCP state changes with DTrace. 4080 * This function shouldn't be called for t_state initializations that don't 4081 * correspond to actual TCP state transitions. 4082 */ 4083 void 4084 tcp_state_change(struct tcpcb *tp, int newstate) 4085 { 4086 #if defined(KDTRACE_HOOKS) 4087 int pstate = tp->t_state; 4088 #endif 4089 4090 TCPSTATES_DEC(tp->t_state); 4091 TCPSTATES_INC(newstate); 4092 tp->t_state = newstate; 4093 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4094 } 4095 4096 /* 4097 * Create an external-format (``xtcpcb'') structure using the information in 4098 * the kernel-format tcpcb structure pointed to by tp. This is done to 4099 * reduce the spew of irrelevant information over this interface, to isolate 4100 * user code from changes in the kernel structure, and potentially to provide 4101 * information-hiding if we decide that some of this information should be 4102 * hidden from users. 4103 */ 4104 void 4105 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4106 { 4107 struct tcpcb *tp = intotcpcb(inp); 4108 struct tcptw *tw = intotw(inp); 4109 sbintime_t now; 4110 4111 bzero(xt, sizeof(*xt)); 4112 if (inp->inp_flags & INP_TIMEWAIT) { 4113 xt->t_state = TCPS_TIME_WAIT; 4114 xt->xt_encaps_port = tw->t_port; 4115 } else { 4116 xt->t_state = tp->t_state; 4117 xt->t_logstate = tp->t_logstate; 4118 xt->t_flags = tp->t_flags; 4119 xt->t_sndzerowin = tp->t_sndzerowin; 4120 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4121 xt->t_rcvoopack = tp->t_rcvoopack; 4122 xt->t_rcv_wnd = tp->rcv_wnd; 4123 xt->t_snd_wnd = tp->snd_wnd; 4124 xt->t_snd_cwnd = tp->snd_cwnd; 4125 xt->t_snd_ssthresh = tp->snd_ssthresh; 4126 xt->t_dsack_bytes = tp->t_dsack_bytes; 4127 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4128 xt->t_dsack_pack = tp->t_dsack_pack; 4129 xt->t_maxseg = tp->t_maxseg; 4130 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4131 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4132 4133 now = getsbinuptime(); 4134 #define COPYTIMER(ttt) do { \ 4135 if (callout_active(&tp->t_timers->ttt)) \ 4136 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 4137 SBT_1MS; \ 4138 else \ 4139 xt->ttt = 0; \ 4140 } while (0) 4141 COPYTIMER(tt_delack); 4142 COPYTIMER(tt_rexmt); 4143 COPYTIMER(tt_persist); 4144 COPYTIMER(tt_keep); 4145 COPYTIMER(tt_2msl); 4146 #undef COPYTIMER 4147 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4148 4149 xt->xt_encaps_port = tp->t_port; 4150 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4151 TCP_FUNCTION_NAME_LEN_MAX); 4152 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 4153 TCP_CA_NAME_MAX); 4154 #ifdef TCP_BLACKBOX 4155 (void)tcp_log_get_id(tp, xt->xt_logid); 4156 #endif 4157 } 4158 4159 xt->xt_len = sizeof(struct xtcpcb); 4160 in_pcbtoxinpcb(inp, &xt->xt_inp); 4161 if (inp->inp_socket == NULL) 4162 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 4163 } 4164 4165 void 4166 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4167 { 4168 uint32_t bit, i; 4169 4170 if ((tp == NULL) || 4171 (status > TCP_EI_STATUS_MAX_VALUE) || 4172 (status == 0)) { 4173 /* Invalid */ 4174 return; 4175 } 4176 if (status > (sizeof(uint32_t) * 8)) { 4177 /* Should this be a KASSERT? */ 4178 return; 4179 } 4180 bit = 1U << (status - 1); 4181 if (bit & tp->t_end_info_status) { 4182 /* already logged */ 4183 return; 4184 } 4185 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4186 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4187 tp->t_end_info_bytes[i] = status; 4188 tp->t_end_info_status |= bit; 4189 break; 4190 } 4191 } 4192 } 4193 4194 int 4195 tcp_can_enable_pacing(void) 4196 { 4197 4198 if ((tcp_pacing_limit == -1) || 4199 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4200 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4201 shadow_num_connections = number_of_tcp_connections_pacing; 4202 return (1); 4203 } else { 4204 return (0); 4205 } 4206 } 4207 4208 static uint8_t tcp_pacing_warning = 0; 4209 4210 void 4211 tcp_decrement_paced_conn(void) 4212 { 4213 uint32_t ret; 4214 4215 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4216 shadow_num_connections = number_of_tcp_connections_pacing; 4217 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4218 if (ret == 0) { 4219 if (tcp_pacing_limit != -1) { 4220 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4221 tcp_pacing_limit = 0; 4222 } else if (tcp_pacing_warning == 0) { 4223 printf("Warning pacing count is invalid, invalid decrement\n"); 4224 tcp_pacing_warning = 1; 4225 } 4226 } 4227 } 4228