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 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_syncache.h> 110 #include <netinet/tcp_hpts.h> 111 #include <netinet/cc/cc.h> 112 #ifdef INET6 113 #include <netinet6/tcp6_var.h> 114 #endif 115 #include <netinet/tcpip.h> 116 #include <netinet/tcp_fastopen.h> 117 #ifdef TCPPCAP 118 #include <netinet/tcp_pcap.h> 119 #endif 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif 123 #ifdef INET6 124 #include <netinet6/ip6protosw.h> 125 #endif 126 #ifdef TCP_OFFLOAD 127 #include <netinet/tcp_offload.h> 128 #endif 129 #include <netinet/udp.h> 130 #include <netinet/udp_var.h> 131 132 #include <netipsec/ipsec_support.h> 133 134 #include <machine/in_cksum.h> 135 #include <crypto/siphash/siphash.h> 136 137 #include <security/mac/mac_framework.h> 138 139 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 140 #ifdef INET6 141 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 142 #endif 143 144 #ifdef NETFLIX_EXP_DETECTION 145 /* Sack attack detection thresholds and such */ 146 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 147 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 148 "Sack Attack detection thresholds"); 149 int32_t tcp_force_detection = 0; 150 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 151 CTLFLAG_RW, 152 &tcp_force_detection, 0, 153 "Do we force detection even if the INP has it off?"); 154 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 155 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 156 CTLFLAG_RW, 157 &tcp_sack_to_ack_thresh, 700, 158 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 159 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 160 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 161 CTLFLAG_RW, 162 &tcp_sack_to_move_thresh, 600, 163 "Percentage of sack moves we must see above (10.1 percent is 101)"); 164 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 165 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 166 CTLFLAG_RW, 167 &tcp_restoral_thresh, 550, 168 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 169 int32_t tcp_sad_decay_val = 800; 170 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 171 CTLFLAG_RW, 172 &tcp_sad_decay_val, 800, 173 "The decay percentage (10.1 percent equals 101 )"); 174 int32_t tcp_map_minimum = 500; 175 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 176 CTLFLAG_RW, 177 &tcp_map_minimum, 500, 178 "Number of Map enteries before we start detection"); 179 int32_t tcp_attack_on_turns_on_logging = 0; 180 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 181 CTLFLAG_RW, 182 &tcp_attack_on_turns_on_logging, 0, 183 "When we have a positive hit on attack, do we turn on logging?"); 184 int32_t tcp_sad_pacing_interval = 2000; 185 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 186 CTLFLAG_RW, 187 &tcp_sad_pacing_interval, 2000, 188 "What is the minimum pacing interval for a classified attacker?"); 189 190 int32_t tcp_sad_low_pps = 100; 191 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 192 CTLFLAG_RW, 193 &tcp_sad_low_pps, 100, 194 "What is the input pps that below which we do not decay?"); 195 #endif 196 uint32_t tcp_ack_war_time_window = 1000; 197 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 198 CTLFLAG_RW, 199 &tcp_ack_war_time_window, 1000, 200 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 201 uint32_t tcp_ack_war_cnt = 5; 202 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 203 CTLFLAG_RW, 204 &tcp_ack_war_cnt, 5, 205 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 206 207 struct rwlock tcp_function_lock; 208 209 static int 210 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 211 { 212 int error, new; 213 214 new = V_tcp_mssdflt; 215 error = sysctl_handle_int(oidp, &new, 0, req); 216 if (error == 0 && req->newptr) { 217 if (new < TCP_MINMSS) 218 error = EINVAL; 219 else 220 V_tcp_mssdflt = new; 221 } 222 return (error); 223 } 224 225 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 226 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 227 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 228 "Default TCP Maximum Segment Size"); 229 230 #ifdef INET6 231 static int 232 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 233 { 234 int error, new; 235 236 new = V_tcp_v6mssdflt; 237 error = sysctl_handle_int(oidp, &new, 0, req); 238 if (error == 0 && req->newptr) { 239 if (new < TCP_MINMSS) 240 error = EINVAL; 241 else 242 V_tcp_v6mssdflt = new; 243 } 244 return (error); 245 } 246 247 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 248 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 249 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 250 "Default TCP Maximum Segment Size for IPv6"); 251 #endif /* INET6 */ 252 253 /* 254 * Minimum MSS we accept and use. This prevents DoS attacks where 255 * we are forced to a ridiculous low MSS like 20 and send hundreds 256 * of packets instead of one. The effect scales with the available 257 * bandwidth and quickly saturates the CPU and network interface 258 * with packet generation and sending. Set to zero to disable MINMSS 259 * checking. This setting prevents us from sending too small packets. 260 */ 261 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 263 &VNET_NAME(tcp_minmss), 0, 264 "Minimum TCP Maximum Segment Size"); 265 266 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 267 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 268 &VNET_NAME(tcp_do_rfc1323), 0, 269 "Enable rfc1323 (high performance TCP) extensions"); 270 271 /* 272 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 273 * Some stacks negotiate TS, but never send them after connection setup. Some 274 * stacks negotiate TS, but don't send them when sending keep-alive segments. 275 * These include modern widely deployed TCP stacks. 276 * Therefore tolerating violations for now... 277 */ 278 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 280 &VNET_NAME(tcp_tolerate_missing_ts), 0, 281 "Tolerate missing TCP timestamps"); 282 283 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 284 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 285 &VNET_NAME(tcp_ts_offset_per_conn), 0, 286 "Initialize TCP timestamps per connection instead of per host pair"); 287 288 /* How many connections are pacing */ 289 static volatile uint32_t number_of_tcp_connections_pacing = 0; 290 static uint32_t shadow_num_connections = 0; 291 292 static int tcp_pacing_limit = 10000; 293 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 294 &tcp_pacing_limit, 1000, 295 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 296 297 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 298 &shadow_num_connections, 0, "Number of TCP connections being paced"); 299 300 static int tcp_log_debug = 0; 301 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 302 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 303 304 static int tcp_tcbhashsize; 305 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 306 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 307 308 static int do_tcpdrain = 1; 309 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 310 "Enable tcp_drain routine for extra help when low on mbufs"); 311 312 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 313 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 314 315 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 316 #define V_icmp_may_rst VNET(icmp_may_rst) 317 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 318 &VNET_NAME(icmp_may_rst), 0, 319 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 320 321 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 322 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 323 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 324 &VNET_NAME(tcp_isn_reseed_interval), 0, 325 "Seconds between reseeding of ISN secret"); 326 327 static int tcp_soreceive_stream; 328 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 329 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 330 331 VNET_DEFINE(uma_zone_t, sack_hole_zone); 332 #define V_sack_hole_zone VNET(sack_hole_zone) 333 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 334 static int 335 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 336 { 337 int error; 338 uint32_t new; 339 340 new = V_tcp_map_entries_limit; 341 error = sysctl_handle_int(oidp, &new, 0, req); 342 if (error == 0 && req->newptr) { 343 /* only allow "0" and value > minimum */ 344 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 345 error = EINVAL; 346 else 347 V_tcp_map_entries_limit = new; 348 } 349 return (error); 350 } 351 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 352 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 353 &VNET_NAME(tcp_map_entries_limit), 0, 354 &sysctl_net_inet_tcp_map_limit_check, "IU", 355 "Total sendmap entries limit"); 356 357 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 358 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 359 &VNET_NAME(tcp_map_split_limit), 0, 360 "Total sendmap split entries limit"); 361 362 #ifdef TCP_HHOOK 363 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 364 #endif 365 366 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 367 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 368 #define V_ts_offset_secret VNET(ts_offset_secret) 369 370 static int tcp_default_fb_init(struct tcpcb *tp); 371 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 372 static int tcp_default_handoff_ok(struct tcpcb *tp); 373 static struct inpcb *tcp_notify(struct inpcb *, int); 374 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 375 static void tcp_mtudisc(struct inpcb *, int); 376 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 377 void *ip4hdr, const void *ip6hdr); 378 379 static struct tcp_function_block tcp_def_funcblk = { 380 .tfb_tcp_block_name = "freebsd", 381 .tfb_tcp_output = tcp_output, 382 .tfb_tcp_do_segment = tcp_do_segment, 383 .tfb_tcp_ctloutput = tcp_default_ctloutput, 384 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 385 .tfb_tcp_fb_init = tcp_default_fb_init, 386 .tfb_tcp_fb_fini = tcp_default_fb_fini, 387 }; 388 389 static int tcp_fb_cnt = 0; 390 struct tcp_funchead t_functions; 391 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 392 393 void 394 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 395 { 396 TCPSTAT_INC(tcps_dsack_count); 397 tp->t_dsack_pack++; 398 if (tlp == 0) { 399 if (SEQ_GT(end, start)) { 400 tp->t_dsack_bytes += (end - start); 401 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 402 } else { 403 tp->t_dsack_tlp_bytes += (start - end); 404 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 405 } 406 } else { 407 if (SEQ_GT(end, start)) { 408 tp->t_dsack_bytes += (end - start); 409 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 410 } else { 411 tp->t_dsack_tlp_bytes += (start - end); 412 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 413 } 414 } 415 } 416 417 static struct tcp_function_block * 418 find_tcp_functions_locked(struct tcp_function_set *fs) 419 { 420 struct tcp_function *f; 421 struct tcp_function_block *blk=NULL; 422 423 TAILQ_FOREACH(f, &t_functions, tf_next) { 424 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 425 blk = f->tf_fb; 426 break; 427 } 428 } 429 return(blk); 430 } 431 432 static struct tcp_function_block * 433 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 434 { 435 struct tcp_function_block *rblk=NULL; 436 struct tcp_function *f; 437 438 TAILQ_FOREACH(f, &t_functions, tf_next) { 439 if (f->tf_fb == blk) { 440 rblk = blk; 441 if (s) { 442 *s = f; 443 } 444 break; 445 } 446 } 447 return (rblk); 448 } 449 450 struct tcp_function_block * 451 find_and_ref_tcp_functions(struct tcp_function_set *fs) 452 { 453 struct tcp_function_block *blk; 454 455 rw_rlock(&tcp_function_lock); 456 blk = find_tcp_functions_locked(fs); 457 if (blk) 458 refcount_acquire(&blk->tfb_refcnt); 459 rw_runlock(&tcp_function_lock); 460 return(blk); 461 } 462 463 struct tcp_function_block * 464 find_and_ref_tcp_fb(struct tcp_function_block *blk) 465 { 466 struct tcp_function_block *rblk; 467 468 rw_rlock(&tcp_function_lock); 469 rblk = find_tcp_fb_locked(blk, NULL); 470 if (rblk) 471 refcount_acquire(&rblk->tfb_refcnt); 472 rw_runlock(&tcp_function_lock); 473 return(rblk); 474 } 475 476 /* Find a matching alias for the given tcp_function_block. */ 477 int 478 find_tcp_function_alias(struct tcp_function_block *blk, 479 struct tcp_function_set *fs) 480 { 481 struct tcp_function *f; 482 int found; 483 484 found = 0; 485 rw_rlock(&tcp_function_lock); 486 TAILQ_FOREACH(f, &t_functions, tf_next) { 487 if ((f->tf_fb == blk) && 488 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 489 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 490 /* Matching function block with different name. */ 491 strncpy(fs->function_set_name, f->tf_name, 492 TCP_FUNCTION_NAME_LEN_MAX); 493 found = 1; 494 break; 495 } 496 } 497 /* Null terminate the string appropriately. */ 498 if (found) { 499 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 500 } else { 501 fs->function_set_name[0] = '\0'; 502 } 503 rw_runlock(&tcp_function_lock); 504 return (found); 505 } 506 507 static struct tcp_function_block * 508 find_and_ref_tcp_default_fb(void) 509 { 510 struct tcp_function_block *rblk; 511 512 rw_rlock(&tcp_function_lock); 513 rblk = tcp_func_set_ptr; 514 refcount_acquire(&rblk->tfb_refcnt); 515 rw_runlock(&tcp_function_lock); 516 return (rblk); 517 } 518 519 void 520 tcp_switch_back_to_default(struct tcpcb *tp) 521 { 522 struct tcp_function_block *tfb; 523 524 KASSERT(tp->t_fb != &tcp_def_funcblk, 525 ("%s: called by the built-in default stack", __func__)); 526 527 /* 528 * Release the old stack. This function will either find a new one 529 * or panic. 530 */ 531 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 532 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 533 refcount_release(&tp->t_fb->tfb_refcnt); 534 535 /* 536 * Now, we'll find a new function block to use. 537 * Start by trying the current user-selected 538 * default, unless this stack is the user-selected 539 * default. 540 */ 541 tfb = find_and_ref_tcp_default_fb(); 542 if (tfb == tp->t_fb) { 543 refcount_release(&tfb->tfb_refcnt); 544 tfb = NULL; 545 } 546 /* Does the stack accept this connection? */ 547 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 548 (*tfb->tfb_tcp_handoff_ok)(tp)) { 549 refcount_release(&tfb->tfb_refcnt); 550 tfb = NULL; 551 } 552 /* Try to use that stack. */ 553 if (tfb != NULL) { 554 /* Initialize the new stack. If it succeeds, we are done. */ 555 tp->t_fb = tfb; 556 if (tp->t_fb->tfb_tcp_fb_init == NULL || 557 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 558 return; 559 560 /* 561 * Initialization failed. Release the reference count on 562 * the stack. 563 */ 564 refcount_release(&tfb->tfb_refcnt); 565 } 566 567 /* 568 * If that wasn't feasible, use the built-in default 569 * stack which is not allowed to reject anyone. 570 */ 571 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 572 if (tfb == NULL) { 573 /* there always should be a default */ 574 panic("Can't refer to tcp_def_funcblk"); 575 } 576 if (tfb->tfb_tcp_handoff_ok != NULL) { 577 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 578 /* The default stack cannot say no */ 579 panic("Default stack rejects a new session?"); 580 } 581 } 582 tp->t_fb = tfb; 583 if (tp->t_fb->tfb_tcp_fb_init != NULL && 584 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 585 /* The default stack cannot fail */ 586 panic("Default stack initialization failed"); 587 } 588 } 589 590 static void 591 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 592 const struct sockaddr *sa, void *ctx) 593 { 594 struct ip *iph; 595 #ifdef INET6 596 struct ip6_hdr *ip6; 597 #endif 598 struct udphdr *uh; 599 struct tcphdr *th; 600 int thlen; 601 uint16_t port; 602 603 TCPSTAT_INC(tcps_tunneled_pkts); 604 if ((m->m_flags & M_PKTHDR) == 0) { 605 /* Can't handle one that is not a pkt hdr */ 606 TCPSTAT_INC(tcps_tunneled_errs); 607 goto out; 608 } 609 thlen = sizeof(struct tcphdr); 610 if (m->m_len < off + sizeof(struct udphdr) + thlen && 611 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 612 TCPSTAT_INC(tcps_tunneled_errs); 613 goto out; 614 } 615 iph = mtod(m, struct ip *); 616 uh = (struct udphdr *)((caddr_t)iph + off); 617 th = (struct tcphdr *)(uh + 1); 618 thlen = th->th_off << 2; 619 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 620 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 621 if (m == NULL) { 622 TCPSTAT_INC(tcps_tunneled_errs); 623 goto out; 624 } else { 625 iph = mtod(m, struct ip *); 626 uh = (struct udphdr *)((caddr_t)iph + off); 627 th = (struct tcphdr *)(uh + 1); 628 } 629 } 630 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 631 bcopy(th, uh, m->m_len - off); 632 m->m_len -= sizeof(struct udphdr); 633 m->m_pkthdr.len -= sizeof(struct udphdr); 634 /* 635 * We use the same algorithm for 636 * both UDP and TCP for c-sum. So 637 * the code in tcp_input will skip 638 * the checksum. So we do nothing 639 * with the flag (m->m_pkthdr.csum_flags). 640 */ 641 switch (iph->ip_v) { 642 #ifdef INET 643 case IPVERSION: 644 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 645 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 646 break; 647 #endif 648 #ifdef INET6 649 case IPV6_VERSION >> 4: 650 ip6 = mtod(m, struct ip6_hdr *); 651 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 652 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 653 break; 654 #endif 655 default: 656 goto out; 657 break; 658 } 659 return; 660 out: 661 m_freem(m); 662 } 663 664 static int 665 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 666 { 667 int error=ENOENT; 668 struct tcp_function_set fs; 669 struct tcp_function_block *blk; 670 671 memset(&fs, 0, sizeof(fs)); 672 rw_rlock(&tcp_function_lock); 673 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 674 if (blk) { 675 /* Found him */ 676 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 677 fs.pcbcnt = blk->tfb_refcnt; 678 } 679 rw_runlock(&tcp_function_lock); 680 error = sysctl_handle_string(oidp, fs.function_set_name, 681 sizeof(fs.function_set_name), req); 682 683 /* Check for error or no change */ 684 if (error != 0 || req->newptr == NULL) 685 return(error); 686 687 rw_wlock(&tcp_function_lock); 688 blk = find_tcp_functions_locked(&fs); 689 if ((blk == NULL) || 690 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 691 error = ENOENT; 692 goto done; 693 } 694 tcp_func_set_ptr = blk; 695 done: 696 rw_wunlock(&tcp_function_lock); 697 return (error); 698 } 699 700 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 701 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 702 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 703 "Set/get the default TCP functions"); 704 705 static int 706 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 707 { 708 int error, cnt, linesz; 709 struct tcp_function *f; 710 char *buffer, *cp; 711 size_t bufsz, outsz; 712 bool alias; 713 714 cnt = 0; 715 rw_rlock(&tcp_function_lock); 716 TAILQ_FOREACH(f, &t_functions, tf_next) { 717 cnt++; 718 } 719 rw_runlock(&tcp_function_lock); 720 721 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 722 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 723 724 error = 0; 725 cp = buffer; 726 727 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 728 "Alias", "PCB count"); 729 cp += linesz; 730 bufsz -= linesz; 731 outsz = linesz; 732 733 rw_rlock(&tcp_function_lock); 734 TAILQ_FOREACH(f, &t_functions, tf_next) { 735 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 736 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 737 f->tf_fb->tfb_tcp_block_name, 738 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 739 alias ? f->tf_name : "-", 740 f->tf_fb->tfb_refcnt); 741 if (linesz >= bufsz) { 742 error = EOVERFLOW; 743 break; 744 } 745 cp += linesz; 746 bufsz -= linesz; 747 outsz += linesz; 748 } 749 rw_runlock(&tcp_function_lock); 750 if (error == 0) 751 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 752 free(buffer, M_TEMP); 753 return (error); 754 } 755 756 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 757 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 758 NULL, 0, sysctl_net_inet_list_available, "A", 759 "list available TCP Function sets"); 760 761 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 762 763 #ifdef INET 764 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 765 #define V_udp4_tun_socket VNET(udp4_tun_socket) 766 #endif 767 #ifdef INET6 768 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 769 #define V_udp6_tun_socket VNET(udp6_tun_socket) 770 #endif 771 772 static void 773 tcp_over_udp_stop(void) 774 { 775 /* 776 * This function assumes sysctl caller holds inp_rinfo_lock() 777 * for writing! 778 */ 779 #ifdef INET 780 if (V_udp4_tun_socket != NULL) { 781 soclose(V_udp4_tun_socket); 782 V_udp4_tun_socket = NULL; 783 } 784 #endif 785 #ifdef INET6 786 if (V_udp6_tun_socket != NULL) { 787 soclose(V_udp6_tun_socket); 788 V_udp6_tun_socket = NULL; 789 } 790 #endif 791 } 792 793 static int 794 tcp_over_udp_start(void) 795 { 796 uint16_t port; 797 int ret; 798 #ifdef INET 799 struct sockaddr_in sin; 800 #endif 801 #ifdef INET6 802 struct sockaddr_in6 sin6; 803 #endif 804 /* 805 * This function assumes sysctl caller holds inp_info_rlock() 806 * for writing! 807 */ 808 port = V_tcp_udp_tunneling_port; 809 if (ntohs(port) == 0) { 810 /* Must have a port set */ 811 return (EINVAL); 812 } 813 #ifdef INET 814 if (V_udp4_tun_socket != NULL) { 815 /* Already running -- must stop first */ 816 return (EALREADY); 817 } 818 #endif 819 #ifdef INET6 820 if (V_udp6_tun_socket != NULL) { 821 /* Already running -- must stop first */ 822 return (EALREADY); 823 } 824 #endif 825 #ifdef INET 826 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 827 SOCK_DGRAM, IPPROTO_UDP, 828 curthread->td_ucred, curthread))) { 829 tcp_over_udp_stop(); 830 return (ret); 831 } 832 /* Call the special UDP hook. */ 833 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 834 tcp_recv_udp_tunneled_packet, 835 tcp_ctlinput_viaudp, 836 NULL))) { 837 tcp_over_udp_stop(); 838 return (ret); 839 } 840 /* Ok, we have a socket, bind it to the port. */ 841 memset(&sin, 0, sizeof(struct sockaddr_in)); 842 sin.sin_len = sizeof(struct sockaddr_in); 843 sin.sin_family = AF_INET; 844 sin.sin_port = htons(port); 845 if ((ret = sobind(V_udp4_tun_socket, 846 (struct sockaddr *)&sin, curthread))) { 847 tcp_over_udp_stop(); 848 return (ret); 849 } 850 #endif 851 #ifdef INET6 852 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 853 SOCK_DGRAM, IPPROTO_UDP, 854 curthread->td_ucred, curthread))) { 855 tcp_over_udp_stop(); 856 return (ret); 857 } 858 /* Call the special UDP hook. */ 859 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 860 tcp_recv_udp_tunneled_packet, 861 tcp6_ctlinput_viaudp, 862 NULL))) { 863 tcp_over_udp_stop(); 864 return (ret); 865 } 866 /* Ok, we have a socket, bind it to the port. */ 867 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 868 sin6.sin6_len = sizeof(struct sockaddr_in6); 869 sin6.sin6_family = AF_INET6; 870 sin6.sin6_port = htons(port); 871 if ((ret = sobind(V_udp6_tun_socket, 872 (struct sockaddr *)&sin6, curthread))) { 873 tcp_over_udp_stop(); 874 return (ret); 875 } 876 #endif 877 return (0); 878 } 879 880 static int 881 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 882 { 883 int error; 884 uint32_t old, new; 885 886 old = V_tcp_udp_tunneling_port; 887 new = old; 888 error = sysctl_handle_int(oidp, &new, 0, req); 889 if ((error == 0) && 890 (req->newptr != NULL)) { 891 if ((new < TCP_TUNNELING_PORT_MIN) || 892 (new > TCP_TUNNELING_PORT_MAX)) { 893 error = EINVAL; 894 } else { 895 V_tcp_udp_tunneling_port = new; 896 if (old != 0) { 897 tcp_over_udp_stop(); 898 } 899 if (new != 0) { 900 error = tcp_over_udp_start(); 901 } 902 } 903 } 904 return (error); 905 } 906 907 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 908 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 909 &VNET_NAME(tcp_udp_tunneling_port), 910 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 911 "Tunneling port for tcp over udp"); 912 913 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 914 915 static int 916 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 917 { 918 int error, new; 919 920 new = V_tcp_udp_tunneling_overhead; 921 error = sysctl_handle_int(oidp, &new, 0, req); 922 if (error == 0 && req->newptr) { 923 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 924 (new > TCP_TUNNELING_OVERHEAD_MAX)) 925 error = EINVAL; 926 else 927 V_tcp_udp_tunneling_overhead = new; 928 } 929 return (error); 930 } 931 932 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 933 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 934 &VNET_NAME(tcp_udp_tunneling_overhead), 935 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 936 "MSS reduction when using tcp over udp"); 937 938 /* 939 * Exports one (struct tcp_function_info) for each alias/name. 940 */ 941 static int 942 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 943 { 944 int cnt, error; 945 struct tcp_function *f; 946 struct tcp_function_info tfi; 947 948 /* 949 * We don't allow writes. 950 */ 951 if (req->newptr != NULL) 952 return (EINVAL); 953 954 /* 955 * Wire the old buffer so we can directly copy the functions to 956 * user space without dropping the lock. 957 */ 958 if (req->oldptr != NULL) { 959 error = sysctl_wire_old_buffer(req, 0); 960 if (error) 961 return (error); 962 } 963 964 /* 965 * Walk the list and copy out matching entries. If INVARIANTS 966 * is compiled in, also walk the list to verify the length of 967 * the list matches what we have recorded. 968 */ 969 rw_rlock(&tcp_function_lock); 970 971 cnt = 0; 972 #ifndef INVARIANTS 973 if (req->oldptr == NULL) { 974 cnt = tcp_fb_cnt; 975 goto skip_loop; 976 } 977 #endif 978 TAILQ_FOREACH(f, &t_functions, tf_next) { 979 #ifdef INVARIANTS 980 cnt++; 981 #endif 982 if (req->oldptr != NULL) { 983 bzero(&tfi, sizeof(tfi)); 984 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 985 tfi.tfi_id = f->tf_fb->tfb_id; 986 (void)strlcpy(tfi.tfi_alias, f->tf_name, 987 sizeof(tfi.tfi_alias)); 988 (void)strlcpy(tfi.tfi_name, 989 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 990 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 991 /* 992 * Don't stop on error, as that is the 993 * mechanism we use to accumulate length 994 * information if the buffer was too short. 995 */ 996 } 997 } 998 KASSERT(cnt == tcp_fb_cnt, 999 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1000 #ifndef INVARIANTS 1001 skip_loop: 1002 #endif 1003 rw_runlock(&tcp_function_lock); 1004 if (req->oldptr == NULL) 1005 error = SYSCTL_OUT(req, NULL, 1006 (cnt + 1) * sizeof(struct tcp_function_info)); 1007 1008 return (error); 1009 } 1010 1011 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1012 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1013 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1014 "List TCP function block name-to-ID mappings"); 1015 1016 /* 1017 * tfb_tcp_handoff_ok() function for the default stack. 1018 * Note that we'll basically try to take all comers. 1019 */ 1020 static int 1021 tcp_default_handoff_ok(struct tcpcb *tp) 1022 { 1023 1024 return (0); 1025 } 1026 1027 /* 1028 * tfb_tcp_fb_init() function for the default stack. 1029 * 1030 * This handles making sure we have appropriate timers set if you are 1031 * transitioning a socket that has some amount of setup done. 1032 * 1033 * The init() fuction from the default can *never* return non-zero i.e. 1034 * it is required to always succeed since it is the stack of last resort! 1035 */ 1036 static int 1037 tcp_default_fb_init(struct tcpcb *tp) 1038 { 1039 1040 struct socket *so; 1041 1042 INP_WLOCK_ASSERT(tp->t_inpcb); 1043 1044 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 1045 ("%s: connection %p in unexpected state %d", __func__, tp, 1046 tp->t_state)); 1047 1048 /* 1049 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1050 * know what to do for unexpected states (which includes TIME_WAIT). 1051 */ 1052 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1053 return (0); 1054 1055 /* 1056 * Make sure some kind of transmission timer is set if there is 1057 * outstanding data. 1058 */ 1059 so = tp->t_inpcb->inp_socket; 1060 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1061 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1062 tcp_timer_active(tp, TT_PERSIST))) { 1063 /* 1064 * If the session has established and it looks like it should 1065 * be in the persist state, set the persist timer. Otherwise, 1066 * set the retransmit timer. 1067 */ 1068 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1069 (int32_t)(tp->snd_nxt - tp->snd_una) < 1070 (int32_t)sbavail(&so->so_snd)) 1071 tcp_setpersist(tp); 1072 else 1073 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1074 } 1075 1076 /* All non-embryonic sessions get a keepalive timer. */ 1077 if (!tcp_timer_active(tp, TT_KEEP)) 1078 tcp_timer_activate(tp, TT_KEEP, 1079 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1080 TP_KEEPINIT(tp)); 1081 1082 /* 1083 * Make sure critical variables are initialized 1084 * if transitioning while in Recovery. 1085 */ 1086 if IN_FASTRECOVERY(tp->t_flags) { 1087 if (tp->sackhint.recover_fs == 0) 1088 tp->sackhint.recover_fs = max(1, 1089 tp->snd_nxt - tp->snd_una); 1090 } 1091 1092 return (0); 1093 } 1094 1095 /* 1096 * tfb_tcp_fb_fini() function for the default stack. 1097 * 1098 * This changes state as necessary (or prudent) to prepare for another stack 1099 * to assume responsibility for the connection. 1100 */ 1101 static void 1102 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1103 { 1104 1105 INP_WLOCK_ASSERT(tp->t_inpcb); 1106 return; 1107 } 1108 1109 /* 1110 * Target size of TCP PCB hash tables. Must be a power of two. 1111 * 1112 * Note that this can be overridden by the kernel environment 1113 * variable net.inet.tcp.tcbhashsize 1114 */ 1115 #ifndef TCBHASHSIZE 1116 #define TCBHASHSIZE 0 1117 #endif 1118 1119 /* 1120 * XXX 1121 * Callouts should be moved into struct tcp directly. They are currently 1122 * separate because the tcpcb structure is exported to userland for sysctl 1123 * parsing purposes, which do not know about callouts. 1124 */ 1125 struct tcpcb_mem { 1126 struct tcpcb tcb; 1127 struct tcp_timer tt; 1128 struct cc_var ccv; 1129 #ifdef TCP_HHOOK 1130 struct osd osd; 1131 #endif 1132 }; 1133 1134 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 1135 #define V_tcpcb_zone VNET(tcpcb_zone) 1136 1137 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1138 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1139 1140 static struct mtx isn_mtx; 1141 1142 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1143 #define ISN_LOCK() mtx_lock(&isn_mtx) 1144 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1145 1146 /* 1147 * TCP initialization. 1148 */ 1149 static void 1150 tcp_zone_change(void *tag) 1151 { 1152 1153 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 1154 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1155 tcp_tw_zone_change(); 1156 } 1157 1158 static int 1159 tcp_inpcb_init(void *mem, int size, int flags) 1160 { 1161 struct inpcb *inp = mem; 1162 1163 INP_LOCK_INIT(inp, "inp", "tcpinp"); 1164 return (0); 1165 } 1166 1167 /* 1168 * Take a value and get the next power of 2 that doesn't overflow. 1169 * Used to size the tcp_inpcb hash buckets. 1170 */ 1171 static int 1172 maketcp_hashsize(int size) 1173 { 1174 int hashsize; 1175 1176 /* 1177 * auto tune. 1178 * get the next power of 2 higher than maxsockets. 1179 */ 1180 hashsize = 1 << fls(size); 1181 /* catch overflow, and just go one power of 2 smaller */ 1182 if (hashsize < size) { 1183 hashsize = 1 << (fls(size) - 1); 1184 } 1185 return (hashsize); 1186 } 1187 1188 static volatile int next_tcp_stack_id = 1; 1189 1190 /* 1191 * Register a TCP function block with the name provided in the names 1192 * array. (Note that this function does NOT automatically register 1193 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1194 * explicitly include blk->tfb_tcp_block_name in the list of names if 1195 * you wish to register the stack with that name.) 1196 * 1197 * Either all name registrations will succeed or all will fail. If 1198 * a name registration fails, the function will update the num_names 1199 * argument to point to the array index of the name that encountered 1200 * the failure. 1201 * 1202 * Returns 0 on success, or an error code on failure. 1203 */ 1204 int 1205 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1206 const char *names[], int *num_names) 1207 { 1208 struct tcp_function *n; 1209 struct tcp_function_set fs; 1210 int error, i; 1211 1212 KASSERT(names != NULL && *num_names > 0, 1213 ("%s: Called with 0-length name list", __func__)); 1214 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1215 KASSERT(rw_initialized(&tcp_function_lock), 1216 ("%s: called too early", __func__)); 1217 1218 if ((blk->tfb_tcp_output == NULL) || 1219 (blk->tfb_tcp_do_segment == NULL) || 1220 (blk->tfb_tcp_ctloutput == NULL) || 1221 (strlen(blk->tfb_tcp_block_name) == 0)) { 1222 /* 1223 * These functions are required and you 1224 * need a name. 1225 */ 1226 *num_names = 0; 1227 return (EINVAL); 1228 } 1229 if (blk->tfb_tcp_timer_stop_all || 1230 blk->tfb_tcp_timer_activate || 1231 blk->tfb_tcp_timer_active || 1232 blk->tfb_tcp_timer_stop) { 1233 /* 1234 * If you define one timer function you 1235 * must have them all. 1236 */ 1237 if ((blk->tfb_tcp_timer_stop_all == NULL) || 1238 (blk->tfb_tcp_timer_activate == NULL) || 1239 (blk->tfb_tcp_timer_active == NULL) || 1240 (blk->tfb_tcp_timer_stop == NULL)) { 1241 *num_names = 0; 1242 return (EINVAL); 1243 } 1244 } 1245 1246 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1247 *num_names = 0; 1248 return (EINVAL); 1249 } 1250 1251 refcount_init(&blk->tfb_refcnt, 0); 1252 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1253 for (i = 0; i < *num_names; i++) { 1254 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1255 if (n == NULL) { 1256 error = ENOMEM; 1257 goto cleanup; 1258 } 1259 n->tf_fb = blk; 1260 1261 (void)strlcpy(fs.function_set_name, names[i], 1262 sizeof(fs.function_set_name)); 1263 rw_wlock(&tcp_function_lock); 1264 if (find_tcp_functions_locked(&fs) != NULL) { 1265 /* Duplicate name space not allowed */ 1266 rw_wunlock(&tcp_function_lock); 1267 free(n, M_TCPFUNCTIONS); 1268 error = EALREADY; 1269 goto cleanup; 1270 } 1271 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1272 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1273 tcp_fb_cnt++; 1274 rw_wunlock(&tcp_function_lock); 1275 } 1276 return(0); 1277 1278 cleanup: 1279 /* 1280 * Deregister the names we just added. Because registration failed 1281 * for names[i], we don't need to deregister that name. 1282 */ 1283 *num_names = i; 1284 rw_wlock(&tcp_function_lock); 1285 while (--i >= 0) { 1286 TAILQ_FOREACH(n, &t_functions, tf_next) { 1287 if (!strncmp(n->tf_name, names[i], 1288 TCP_FUNCTION_NAME_LEN_MAX)) { 1289 TAILQ_REMOVE(&t_functions, n, tf_next); 1290 tcp_fb_cnt--; 1291 n->tf_fb = NULL; 1292 free(n, M_TCPFUNCTIONS); 1293 break; 1294 } 1295 } 1296 } 1297 rw_wunlock(&tcp_function_lock); 1298 return (error); 1299 } 1300 1301 /* 1302 * Register a TCP function block using the name provided in the name 1303 * argument. 1304 * 1305 * Returns 0 on success, or an error code on failure. 1306 */ 1307 int 1308 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1309 int wait) 1310 { 1311 const char *name_list[1]; 1312 int num_names, rv; 1313 1314 num_names = 1; 1315 if (name != NULL) 1316 name_list[0] = name; 1317 else 1318 name_list[0] = blk->tfb_tcp_block_name; 1319 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1320 return (rv); 1321 } 1322 1323 /* 1324 * Register a TCP function block using the name defined in 1325 * blk->tfb_tcp_block_name. 1326 * 1327 * Returns 0 on success, or an error code on failure. 1328 */ 1329 int 1330 register_tcp_functions(struct tcp_function_block *blk, int wait) 1331 { 1332 1333 return (register_tcp_functions_as_name(blk, NULL, wait)); 1334 } 1335 1336 /* 1337 * Deregister all names associated with a function block. This 1338 * functionally removes the function block from use within the system. 1339 * 1340 * When called with a true quiesce argument, mark the function block 1341 * as being removed so no more stacks will use it and determine 1342 * whether the removal would succeed. 1343 * 1344 * When called with a false quiesce argument, actually attempt the 1345 * removal. 1346 * 1347 * When called with a force argument, attempt to switch all TCBs to 1348 * use the default stack instead of returning EBUSY. 1349 * 1350 * Returns 0 on success (or if the removal would succeed, or an error 1351 * code on failure. 1352 */ 1353 int 1354 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1355 bool force) 1356 { 1357 struct tcp_function *f; 1358 1359 if (blk == &tcp_def_funcblk) { 1360 /* You can't un-register the default */ 1361 return (EPERM); 1362 } 1363 rw_wlock(&tcp_function_lock); 1364 if (blk == tcp_func_set_ptr) { 1365 /* You can't free the current default */ 1366 rw_wunlock(&tcp_function_lock); 1367 return (EBUSY); 1368 } 1369 /* Mark the block so no more stacks can use it. */ 1370 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1371 /* 1372 * If TCBs are still attached to the stack, attempt to switch them 1373 * to the default stack. 1374 */ 1375 if (force && blk->tfb_refcnt) { 1376 struct inpcb *inp; 1377 struct tcpcb *tp; 1378 VNET_ITERATOR_DECL(vnet_iter); 1379 1380 rw_wunlock(&tcp_function_lock); 1381 1382 VNET_LIST_RLOCK(); 1383 VNET_FOREACH(vnet_iter) { 1384 CURVNET_SET(vnet_iter); 1385 INP_INFO_WLOCK(&V_tcbinfo); 1386 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 1387 INP_WLOCK(inp); 1388 if (inp->inp_flags & INP_TIMEWAIT) { 1389 INP_WUNLOCK(inp); 1390 continue; 1391 } 1392 tp = intotcpcb(inp); 1393 if (tp == NULL || tp->t_fb != blk) { 1394 INP_WUNLOCK(inp); 1395 continue; 1396 } 1397 tcp_switch_back_to_default(tp); 1398 INP_WUNLOCK(inp); 1399 } 1400 INP_INFO_WUNLOCK(&V_tcbinfo); 1401 CURVNET_RESTORE(); 1402 } 1403 VNET_LIST_RUNLOCK(); 1404 1405 rw_wlock(&tcp_function_lock); 1406 } 1407 if (blk->tfb_refcnt) { 1408 /* TCBs still attached. */ 1409 rw_wunlock(&tcp_function_lock); 1410 return (EBUSY); 1411 } 1412 if (quiesce) { 1413 /* Skip removal. */ 1414 rw_wunlock(&tcp_function_lock); 1415 return (0); 1416 } 1417 /* Remove any function names that map to this function block. */ 1418 while (find_tcp_fb_locked(blk, &f) != NULL) { 1419 TAILQ_REMOVE(&t_functions, f, tf_next); 1420 tcp_fb_cnt--; 1421 f->tf_fb = NULL; 1422 free(f, M_TCPFUNCTIONS); 1423 } 1424 rw_wunlock(&tcp_function_lock); 1425 return (0); 1426 } 1427 1428 void 1429 tcp_init(void) 1430 { 1431 const char *tcbhash_tuneable; 1432 int hashsize; 1433 1434 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1435 1436 #ifdef TCP_HHOOK 1437 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1438 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1439 printf("%s: WARNING: unable to register helper hook\n", __func__); 1440 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1441 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1442 printf("%s: WARNING: unable to register helper hook\n", __func__); 1443 #endif 1444 #ifdef STATS 1445 if (tcp_stats_init()) 1446 printf("%s: WARNING: unable to initialise TCP stats\n", 1447 __func__); 1448 #endif 1449 hashsize = TCBHASHSIZE; 1450 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1451 if (hashsize == 0) { 1452 /* 1453 * Auto tune the hash size based on maxsockets. 1454 * A perfect hash would have a 1:1 mapping 1455 * (hashsize = maxsockets) however it's been 1456 * suggested that O(2) average is better. 1457 */ 1458 hashsize = maketcp_hashsize(maxsockets / 4); 1459 /* 1460 * Our historical default is 512, 1461 * do not autotune lower than this. 1462 */ 1463 if (hashsize < 512) 1464 hashsize = 512; 1465 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1466 printf("%s: %s auto tuned to %d\n", __func__, 1467 tcbhash_tuneable, hashsize); 1468 } 1469 /* 1470 * We require a hashsize to be a power of two. 1471 * Previously if it was not a power of two we would just reset it 1472 * back to 512, which could be a nasty surprise if you did not notice 1473 * the error message. 1474 * Instead what we do is clip it to the closest power of two lower 1475 * than the specified hash value. 1476 */ 1477 if (!powerof2(hashsize)) { 1478 int oldhashsize = hashsize; 1479 1480 hashsize = maketcp_hashsize(hashsize); 1481 /* prevent absurdly low value */ 1482 if (hashsize < 16) 1483 hashsize = 16; 1484 printf("%s: WARNING: TCB hash size not a power of 2, " 1485 "clipped from %d to %d.\n", __func__, oldhashsize, 1486 hashsize); 1487 } 1488 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1489 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1490 1491 /* 1492 * These have to be type stable for the benefit of the timers. 1493 */ 1494 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1495 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1496 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1497 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1498 1499 tcp_tw_init(); 1500 syncache_init(); 1501 tcp_hc_init(); 1502 1503 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1504 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1505 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1506 1507 tcp_fastopen_init(); 1508 1509 /* Skip initialization of globals for non-default instances. */ 1510 if (!IS_DEFAULT_VNET(curvnet)) 1511 return; 1512 1513 tcp_reass_global_init(); 1514 1515 /* XXX virtualize those bellow? */ 1516 tcp_delacktime = TCPTV_DELACK; 1517 tcp_keepinit = TCPTV_KEEP_INIT; 1518 tcp_keepidle = TCPTV_KEEP_IDLE; 1519 tcp_keepintvl = TCPTV_KEEPINTVL; 1520 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1521 tcp_msl = TCPTV_MSL; 1522 tcp_rexmit_initial = TCPTV_RTOBASE; 1523 if (tcp_rexmit_initial < 1) 1524 tcp_rexmit_initial = 1; 1525 tcp_rexmit_min = TCPTV_MIN; 1526 if (tcp_rexmit_min < 1) 1527 tcp_rexmit_min = 1; 1528 tcp_persmin = TCPTV_PERSMIN; 1529 tcp_persmax = TCPTV_PERSMAX; 1530 tcp_rexmit_slop = TCPTV_CPU_VAR; 1531 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1532 tcp_tcbhashsize = hashsize; 1533 1534 /* Setup the tcp function block list */ 1535 TAILQ_INIT(&t_functions); 1536 rw_init(&tcp_function_lock, "tcp_func_lock"); 1537 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1538 #ifdef TCP_BLACKBOX 1539 /* Initialize the TCP logging data. */ 1540 tcp_log_init(); 1541 #endif 1542 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1543 1544 if (tcp_soreceive_stream) { 1545 #ifdef INET 1546 tcp_usrreqs.pru_soreceive = soreceive_stream; 1547 #endif 1548 #ifdef INET6 1549 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1550 #endif /* INET6 */ 1551 } 1552 1553 #ifdef INET6 1554 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1555 #else /* INET6 */ 1556 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1557 #endif /* INET6 */ 1558 if (max_protohdr < TCP_MINPROTOHDR) 1559 max_protohdr = TCP_MINPROTOHDR; 1560 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1561 panic("tcp_init"); 1562 #undef TCP_MINPROTOHDR 1563 1564 ISN_LOCK_INIT(); 1565 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1566 SHUTDOWN_PRI_DEFAULT); 1567 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1568 EVENTHANDLER_PRI_ANY); 1569 1570 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1571 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1572 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1573 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1574 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1575 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1576 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1577 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1578 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1579 #ifdef TCPPCAP 1580 tcp_pcap_init(); 1581 #endif 1582 } 1583 1584 #ifdef VIMAGE 1585 static void 1586 tcp_destroy(void *unused __unused) 1587 { 1588 int n; 1589 #ifdef TCP_HHOOK 1590 int error; 1591 #endif 1592 1593 /* 1594 * All our processes are gone, all our sockets should be cleaned 1595 * up, which means, we should be past the tcp_discardcb() calls. 1596 * Sleep to let all tcpcb timers really disappear and cleanup. 1597 */ 1598 for (;;) { 1599 INP_LIST_RLOCK(&V_tcbinfo); 1600 n = V_tcbinfo.ipi_count; 1601 INP_LIST_RUNLOCK(&V_tcbinfo); 1602 if (n == 0) 1603 break; 1604 pause("tcpdes", hz / 10); 1605 } 1606 tcp_hc_destroy(); 1607 syncache_destroy(); 1608 tcp_tw_destroy(); 1609 in_pcbinfo_destroy(&V_tcbinfo); 1610 /* tcp_discardcb() clears the sack_holes up. */ 1611 uma_zdestroy(V_sack_hole_zone); 1612 uma_zdestroy(V_tcpcb_zone); 1613 1614 /* 1615 * Cannot free the zone until all tcpcbs are released as we attach 1616 * the allocations to them. 1617 */ 1618 tcp_fastopen_destroy(); 1619 1620 #ifdef TCP_HHOOK 1621 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1622 if (error != 0) { 1623 printf("%s: WARNING: unable to deregister helper hook " 1624 "type=%d, id=%d: error %d returned\n", __func__, 1625 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1626 } 1627 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1628 if (error != 0) { 1629 printf("%s: WARNING: unable to deregister helper hook " 1630 "type=%d, id=%d: error %d returned\n", __func__, 1631 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1632 } 1633 #endif 1634 } 1635 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1636 #endif 1637 1638 void 1639 tcp_fini(void *xtp) 1640 { 1641 1642 } 1643 1644 /* 1645 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1646 * tcp_template used to store this data in mbufs, but we now recopy it out 1647 * of the tcpcb each time to conserve mbufs. 1648 */ 1649 void 1650 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1651 { 1652 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1653 1654 INP_WLOCK_ASSERT(inp); 1655 1656 #ifdef INET6 1657 if ((inp->inp_vflag & INP_IPV6) != 0) { 1658 struct ip6_hdr *ip6; 1659 1660 ip6 = (struct ip6_hdr *)ip_ptr; 1661 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1662 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1663 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1664 (IPV6_VERSION & IPV6_VERSION_MASK); 1665 if (port == 0) 1666 ip6->ip6_nxt = IPPROTO_TCP; 1667 else 1668 ip6->ip6_nxt = IPPROTO_UDP; 1669 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1670 ip6->ip6_src = inp->in6p_laddr; 1671 ip6->ip6_dst = inp->in6p_faddr; 1672 } 1673 #endif /* INET6 */ 1674 #if defined(INET6) && defined(INET) 1675 else 1676 #endif 1677 #ifdef INET 1678 { 1679 struct ip *ip; 1680 1681 ip = (struct ip *)ip_ptr; 1682 ip->ip_v = IPVERSION; 1683 ip->ip_hl = 5; 1684 ip->ip_tos = inp->inp_ip_tos; 1685 ip->ip_len = 0; 1686 ip->ip_id = 0; 1687 ip->ip_off = 0; 1688 ip->ip_ttl = inp->inp_ip_ttl; 1689 ip->ip_sum = 0; 1690 if (port == 0) 1691 ip->ip_p = IPPROTO_TCP; 1692 else 1693 ip->ip_p = IPPROTO_UDP; 1694 ip->ip_src = inp->inp_laddr; 1695 ip->ip_dst = inp->inp_faddr; 1696 } 1697 #endif /* INET */ 1698 th->th_sport = inp->inp_lport; 1699 th->th_dport = inp->inp_fport; 1700 th->th_seq = 0; 1701 th->th_ack = 0; 1702 th->th_x2 = 0; 1703 th->th_off = 5; 1704 th->th_flags = 0; 1705 th->th_win = 0; 1706 th->th_urp = 0; 1707 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1708 } 1709 1710 /* 1711 * Create template to be used to send tcp packets on a connection. 1712 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1713 * use for this function is in keepalives, which use tcp_respond. 1714 */ 1715 struct tcptemp * 1716 tcpip_maketemplate(struct inpcb *inp) 1717 { 1718 struct tcptemp *t; 1719 1720 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1721 if (t == NULL) 1722 return (NULL); 1723 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1724 return (t); 1725 } 1726 1727 /* 1728 * Send a single message to the TCP at address specified by 1729 * the given TCP/IP header. If m == NULL, then we make a copy 1730 * of the tcpiphdr at th and send directly to the addressed host. 1731 * This is used to force keep alive messages out using the TCP 1732 * template for a connection. If flags are given then we send 1733 * a message back to the TCP which originated the segment th, 1734 * and discard the mbuf containing it and any other attached mbufs. 1735 * 1736 * In any case the ack and sequence number of the transmitted 1737 * segment are as specified by the parameters. 1738 * 1739 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1740 */ 1741 void 1742 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1743 tcp_seq ack, tcp_seq seq, int flags) 1744 { 1745 struct tcpopt to; 1746 struct inpcb *inp; 1747 struct ip *ip; 1748 struct mbuf *optm; 1749 struct udphdr *uh = NULL; 1750 struct tcphdr *nth; 1751 u_char *optp; 1752 #ifdef INET6 1753 struct ip6_hdr *ip6; 1754 int isipv6; 1755 #endif /* INET6 */ 1756 int optlen, tlen, win, ulen; 1757 bool incl_opts; 1758 uint16_t port; 1759 1760 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1761 NET_EPOCH_ASSERT(); 1762 1763 #ifdef INET6 1764 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1765 ip6 = ipgen; 1766 #endif /* INET6 */ 1767 ip = ipgen; 1768 1769 if (tp != NULL) { 1770 inp = tp->t_inpcb; 1771 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1772 INP_LOCK_ASSERT(inp); 1773 } else 1774 inp = NULL; 1775 1776 if (m != NULL) { 1777 #ifdef INET6 1778 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1779 port = m->m_pkthdr.tcp_tun_port; 1780 else 1781 #endif 1782 if (ip && (ip->ip_p == IPPROTO_UDP)) 1783 port = m->m_pkthdr.tcp_tun_port; 1784 else 1785 port = 0; 1786 } else 1787 port = tp->t_port; 1788 1789 incl_opts = false; 1790 win = 0; 1791 if (tp != NULL) { 1792 if (!(flags & TH_RST)) { 1793 win = sbspace(&inp->inp_socket->so_rcv); 1794 if (win > TCP_MAXWIN << tp->rcv_scale) 1795 win = TCP_MAXWIN << tp->rcv_scale; 1796 } 1797 if ((tp->t_flags & TF_NOOPT) == 0) 1798 incl_opts = true; 1799 } 1800 if (m == NULL) { 1801 m = m_gethdr(M_NOWAIT, MT_DATA); 1802 if (m == NULL) 1803 return; 1804 m->m_data += max_linkhdr; 1805 #ifdef INET6 1806 if (isipv6) { 1807 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1808 sizeof(struct ip6_hdr)); 1809 ip6 = mtod(m, struct ip6_hdr *); 1810 nth = (struct tcphdr *)(ip6 + 1); 1811 if (port) { 1812 /* Insert a UDP header */ 1813 uh = (struct udphdr *)nth; 1814 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1815 uh->uh_dport = port; 1816 nth = (struct tcphdr *)(uh + 1); 1817 } 1818 } else 1819 #endif /* INET6 */ 1820 { 1821 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1822 ip = mtod(m, struct ip *); 1823 nth = (struct tcphdr *)(ip + 1); 1824 if (port) { 1825 /* Insert a UDP header */ 1826 uh = (struct udphdr *)nth; 1827 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1828 uh->uh_dport = port; 1829 nth = (struct tcphdr *)(uh + 1); 1830 } 1831 } 1832 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1833 flags = TH_ACK; 1834 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1835 struct mbuf *n; 1836 1837 /* Can't reuse 'm', allocate a new mbuf. */ 1838 n = m_gethdr(M_NOWAIT, MT_DATA); 1839 if (n == NULL) { 1840 m_freem(m); 1841 return; 1842 } 1843 1844 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1845 m_freem(m); 1846 m_freem(n); 1847 return; 1848 } 1849 1850 n->m_data += max_linkhdr; 1851 /* m_len is set later */ 1852 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1853 #ifdef INET6 1854 if (isipv6) { 1855 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1856 sizeof(struct ip6_hdr)); 1857 ip6 = mtod(n, struct ip6_hdr *); 1858 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1859 nth = (struct tcphdr *)(ip6 + 1); 1860 if (port) { 1861 /* Insert a UDP header */ 1862 uh = (struct udphdr *)nth; 1863 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1864 uh->uh_dport = port; 1865 nth = (struct tcphdr *)(uh + 1); 1866 } 1867 } else 1868 #endif /* INET6 */ 1869 { 1870 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1871 ip = mtod(n, struct ip *); 1872 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1873 nth = (struct tcphdr *)(ip + 1); 1874 if (port) { 1875 /* Insert a UDP header */ 1876 uh = (struct udphdr *)nth; 1877 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1878 uh->uh_dport = port; 1879 nth = (struct tcphdr *)(uh + 1); 1880 } 1881 } 1882 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1883 xchg(nth->th_dport, nth->th_sport, uint16_t); 1884 th = nth; 1885 m_freem(m); 1886 m = n; 1887 } else { 1888 /* 1889 * reuse the mbuf. 1890 * XXX MRT We inherit the FIB, which is lucky. 1891 */ 1892 m_freem(m->m_next); 1893 m->m_next = NULL; 1894 m->m_data = (caddr_t)ipgen; 1895 /* m_len is set later */ 1896 #ifdef INET6 1897 if (isipv6) { 1898 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1899 nth = (struct tcphdr *)(ip6 + 1); 1900 } else 1901 #endif /* INET6 */ 1902 { 1903 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1904 nth = (struct tcphdr *)(ip + 1); 1905 } 1906 if (th != nth) { 1907 /* 1908 * this is usually a case when an extension header 1909 * exists between the IPv6 header and the 1910 * TCP header. 1911 */ 1912 nth->th_sport = th->th_sport; 1913 nth->th_dport = th->th_dport; 1914 } 1915 xchg(nth->th_dport, nth->th_sport, uint16_t); 1916 #undef xchg 1917 } 1918 tlen = 0; 1919 #ifdef INET6 1920 if (isipv6) 1921 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1922 #endif 1923 #if defined(INET) && defined(INET6) 1924 else 1925 #endif 1926 #ifdef INET 1927 tlen = sizeof (struct tcpiphdr); 1928 #endif 1929 if (port) 1930 tlen += sizeof (struct udphdr); 1931 #ifdef INVARIANTS 1932 m->m_len = 0; 1933 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1934 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1935 m, tlen, (long)M_TRAILINGSPACE(m))); 1936 #endif 1937 m->m_len = tlen; 1938 to.to_flags = 0; 1939 if (incl_opts) { 1940 /* Make sure we have room. */ 1941 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1942 m->m_next = m_get(M_NOWAIT, MT_DATA); 1943 if (m->m_next) { 1944 optp = mtod(m->m_next, u_char *); 1945 optm = m->m_next; 1946 } else 1947 incl_opts = false; 1948 } else { 1949 optp = (u_char *) (nth + 1); 1950 optm = m; 1951 } 1952 } 1953 if (incl_opts) { 1954 /* Timestamps. */ 1955 if (tp->t_flags & TF_RCVD_TSTMP) { 1956 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1957 to.to_tsecr = tp->ts_recent; 1958 to.to_flags |= TOF_TS; 1959 } 1960 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1961 /* TCP-MD5 (RFC2385). */ 1962 if (tp->t_flags & TF_SIGNATURE) 1963 to.to_flags |= TOF_SIGNATURE; 1964 #endif 1965 /* Add the options. */ 1966 tlen += optlen = tcp_addoptions(&to, optp); 1967 1968 /* Update m_len in the correct mbuf. */ 1969 optm->m_len += optlen; 1970 } else 1971 optlen = 0; 1972 #ifdef INET6 1973 if (isipv6) { 1974 if (uh) { 1975 ulen = tlen - sizeof(struct ip6_hdr); 1976 uh->uh_ulen = htons(ulen); 1977 } 1978 ip6->ip6_flow = 0; 1979 ip6->ip6_vfc = IPV6_VERSION; 1980 if (port) 1981 ip6->ip6_nxt = IPPROTO_UDP; 1982 else 1983 ip6->ip6_nxt = IPPROTO_TCP; 1984 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1985 } 1986 #endif 1987 #if defined(INET) && defined(INET6) 1988 else 1989 #endif 1990 #ifdef INET 1991 { 1992 if (uh) { 1993 ulen = tlen - sizeof(struct ip); 1994 uh->uh_ulen = htons(ulen); 1995 } 1996 ip->ip_len = htons(tlen); 1997 ip->ip_ttl = V_ip_defttl; 1998 if (port) { 1999 ip->ip_p = IPPROTO_UDP; 2000 } else { 2001 ip->ip_p = IPPROTO_TCP; 2002 } 2003 if (V_path_mtu_discovery) 2004 ip->ip_off |= htons(IP_DF); 2005 } 2006 #endif 2007 m->m_pkthdr.len = tlen; 2008 m->m_pkthdr.rcvif = NULL; 2009 #ifdef MAC 2010 if (inp != NULL) { 2011 /* 2012 * Packet is associated with a socket, so allow the 2013 * label of the response to reflect the socket label. 2014 */ 2015 INP_LOCK_ASSERT(inp); 2016 mac_inpcb_create_mbuf(inp, m); 2017 } else { 2018 /* 2019 * Packet is not associated with a socket, so possibly 2020 * update the label in place. 2021 */ 2022 mac_netinet_tcp_reply(m); 2023 } 2024 #endif 2025 nth->th_seq = htonl(seq); 2026 nth->th_ack = htonl(ack); 2027 nth->th_x2 = 0; 2028 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2029 nth->th_flags = flags; 2030 if (tp != NULL) 2031 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2032 else 2033 nth->th_win = htons((u_short)win); 2034 nth->th_urp = 0; 2035 2036 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2037 if (to.to_flags & TOF_SIGNATURE) { 2038 if (!TCPMD5_ENABLED() || 2039 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2040 m_freem(m); 2041 return; 2042 } 2043 } 2044 #endif 2045 2046 #ifdef INET6 2047 if (isipv6) { 2048 if (port) { 2049 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2050 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2051 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2052 nth->th_sum = 0; 2053 } else { 2054 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2055 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2056 nth->th_sum = in6_cksum_pseudo(ip6, 2057 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2058 } 2059 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 2060 NULL, NULL); 2061 } 2062 #endif /* INET6 */ 2063 #if defined(INET6) && defined(INET) 2064 else 2065 #endif 2066 #ifdef INET 2067 { 2068 if (port) { 2069 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2070 htons(ulen + IPPROTO_UDP)); 2071 m->m_pkthdr.csum_flags = CSUM_UDP; 2072 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2073 nth->th_sum = 0; 2074 } else { 2075 m->m_pkthdr.csum_flags = CSUM_TCP; 2076 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2077 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2078 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2079 } 2080 } 2081 #endif /* INET */ 2082 #ifdef TCPDEBUG 2083 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 2084 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 2085 #endif 2086 TCP_PROBE3(debug__output, tp, th, m); 2087 if (flags & TH_RST) 2088 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2089 2090 #ifdef INET6 2091 if (isipv6) { 2092 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2093 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2094 } 2095 #endif /* INET6 */ 2096 #if defined(INET) && defined(INET6) 2097 else 2098 #endif 2099 #ifdef INET 2100 { 2101 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2102 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 2103 } 2104 #endif 2105 } 2106 2107 /* 2108 * Create a new TCP control block, making an 2109 * empty reassembly queue and hooking it to the argument 2110 * protocol control block. The `inp' parameter must have 2111 * come from the zone allocator set up in tcp_init(). 2112 */ 2113 struct tcpcb * 2114 tcp_newtcpcb(struct inpcb *inp) 2115 { 2116 struct tcpcb_mem *tm; 2117 struct tcpcb *tp; 2118 #ifdef INET6 2119 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2120 #endif /* INET6 */ 2121 2122 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 2123 if (tm == NULL) 2124 return (NULL); 2125 tp = &tm->tcb; 2126 2127 /* Initialise cc_var struct for this tcpcb. */ 2128 tp->ccv = &tm->ccv; 2129 tp->ccv->type = IPPROTO_TCP; 2130 tp->ccv->ccvc.tcp = tp; 2131 rw_rlock(&tcp_function_lock); 2132 tp->t_fb = tcp_func_set_ptr; 2133 refcount_acquire(&tp->t_fb->tfb_refcnt); 2134 rw_runlock(&tcp_function_lock); 2135 /* 2136 * Use the current system default CC algorithm. 2137 */ 2138 CC_LIST_RLOCK(); 2139 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 2140 CC_ALGO(tp) = CC_DEFAULT(); 2141 CC_LIST_RUNLOCK(); 2142 /* 2143 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 2144 * is called. 2145 */ 2146 in_pcbref(inp); /* Reference for tcpcb */ 2147 tp->t_inpcb = inp; 2148 2149 if (CC_ALGO(tp)->cb_init != NULL) 2150 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 2151 if (tp->t_fb->tfb_tcp_fb_fini) 2152 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2153 in_pcbrele_wlocked(inp); 2154 refcount_release(&tp->t_fb->tfb_refcnt); 2155 uma_zfree(V_tcpcb_zone, tm); 2156 return (NULL); 2157 } 2158 2159 #ifdef TCP_HHOOK 2160 tp->osd = &tm->osd; 2161 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 2162 if (tp->t_fb->tfb_tcp_fb_fini) 2163 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2164 in_pcbrele_wlocked(inp); 2165 refcount_release(&tp->t_fb->tfb_refcnt); 2166 uma_zfree(V_tcpcb_zone, tm); 2167 return (NULL); 2168 } 2169 #endif 2170 2171 #ifdef VIMAGE 2172 tp->t_vnet = inp->inp_vnet; 2173 #endif 2174 tp->t_timers = &tm->tt; 2175 TAILQ_INIT(&tp->t_segq); 2176 tp->t_maxseg = 2177 #ifdef INET6 2178 isipv6 ? V_tcp_v6mssdflt : 2179 #endif /* INET6 */ 2180 V_tcp_mssdflt; 2181 2182 /* Set up our timeouts. */ 2183 callout_init(&tp->t_timers->tt_rexmt, 1); 2184 callout_init(&tp->t_timers->tt_persist, 1); 2185 callout_init(&tp->t_timers->tt_keep, 1); 2186 callout_init(&tp->t_timers->tt_2msl, 1); 2187 callout_init(&tp->t_timers->tt_delack, 1); 2188 2189 if (V_tcp_do_rfc1323) 2190 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2191 if (V_tcp_do_sack) 2192 tp->t_flags |= TF_SACK_PERMIT; 2193 TAILQ_INIT(&tp->snd_holes); 2194 2195 /* 2196 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2197 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2198 * reasonable initial retransmit time. 2199 */ 2200 tp->t_srtt = TCPTV_SRTTBASE; 2201 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2202 tp->t_rttmin = tcp_rexmit_min; 2203 tp->t_rxtcur = tcp_rexmit_initial; 2204 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2205 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2206 tp->t_rcvtime = ticks; 2207 /* 2208 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2209 * because the socket may be bound to an IPv6 wildcard address, 2210 * which may match an IPv4-mapped IPv6 address. 2211 */ 2212 inp->inp_ip_ttl = V_ip_defttl; 2213 inp->inp_ppcb = tp; 2214 #ifdef TCPPCAP 2215 /* 2216 * Init the TCP PCAP queues. 2217 */ 2218 tcp_pcap_tcpcb_init(tp); 2219 #endif 2220 #ifdef TCP_BLACKBOX 2221 /* Initialize the per-TCPCB log data. */ 2222 tcp_log_tcpcbinit(tp); 2223 #endif 2224 tp->t_pacing_rate = -1; 2225 if (tp->t_fb->tfb_tcp_fb_init) { 2226 if ((*tp->t_fb->tfb_tcp_fb_init)(tp)) { 2227 refcount_release(&tp->t_fb->tfb_refcnt); 2228 in_pcbrele_wlocked(inp); 2229 uma_zfree(V_tcpcb_zone, tm); 2230 return (NULL); 2231 } 2232 } 2233 #ifdef STATS 2234 if (V_tcp_perconn_stats_enable == 1) 2235 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2236 #endif 2237 if (V_tcp_do_lrd) 2238 tp->t_flags |= TF_LRD; 2239 return (tp); /* XXX */ 2240 } 2241 2242 /* 2243 * Switch the congestion control algorithm back to NewReno for any active 2244 * control blocks using an algorithm which is about to go away. 2245 * This ensures the CC framework can allow the unload to proceed without leaving 2246 * any dangling pointers which would trigger a panic. 2247 * Returning non-zero would inform the CC framework that something went wrong 2248 * and it would be unsafe to allow the unload to proceed. However, there is no 2249 * way for this to occur with this implementation so we always return zero. 2250 */ 2251 int 2252 tcp_ccalgounload(struct cc_algo *unload_algo) 2253 { 2254 struct cc_algo *tmpalgo; 2255 struct inpcb *inp; 2256 struct tcpcb *tp; 2257 VNET_ITERATOR_DECL(vnet_iter); 2258 2259 /* 2260 * Check all active control blocks across all network stacks and change 2261 * any that are using "unload_algo" back to NewReno. If "unload_algo" 2262 * requires cleanup code to be run, call it. 2263 */ 2264 VNET_LIST_RLOCK(); 2265 VNET_FOREACH(vnet_iter) { 2266 CURVNET_SET(vnet_iter); 2267 INP_INFO_WLOCK(&V_tcbinfo); 2268 /* 2269 * New connections already part way through being initialised 2270 * with the CC algo we're removing will not race with this code 2271 * because the INP_INFO_WLOCK is held during initialisation. We 2272 * therefore don't enter the loop below until the connection 2273 * list has stabilised. 2274 */ 2275 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 2276 INP_WLOCK(inp); 2277 /* Important to skip tcptw structs. */ 2278 if (!(inp->inp_flags & INP_TIMEWAIT) && 2279 (tp = intotcpcb(inp)) != NULL) { 2280 /* 2281 * By holding INP_WLOCK here, we are assured 2282 * that the connection is not currently 2283 * executing inside the CC module's functions 2284 * i.e. it is safe to make the switch back to 2285 * NewReno. 2286 */ 2287 if (CC_ALGO(tp) == unload_algo) { 2288 tmpalgo = CC_ALGO(tp); 2289 if (tmpalgo->cb_destroy != NULL) 2290 tmpalgo->cb_destroy(tp->ccv); 2291 CC_DATA(tp) = NULL; 2292 /* 2293 * NewReno may allocate memory on 2294 * demand for certain stateful 2295 * configuration as needed, but is 2296 * coded to never fail on memory 2297 * allocation failure so it is a safe 2298 * fallback. 2299 */ 2300 CC_ALGO(tp) = &newreno_cc_algo; 2301 } 2302 } 2303 INP_WUNLOCK(inp); 2304 } 2305 INP_INFO_WUNLOCK(&V_tcbinfo); 2306 CURVNET_RESTORE(); 2307 } 2308 VNET_LIST_RUNLOCK(); 2309 2310 return (0); 2311 } 2312 2313 /* 2314 * Drop a TCP connection, reporting 2315 * the specified error. If connection is synchronized, 2316 * then send a RST to peer. 2317 */ 2318 struct tcpcb * 2319 tcp_drop(struct tcpcb *tp, int errno) 2320 { 2321 struct socket *so = tp->t_inpcb->inp_socket; 2322 2323 NET_EPOCH_ASSERT(); 2324 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2325 INP_WLOCK_ASSERT(tp->t_inpcb); 2326 2327 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2328 tcp_state_change(tp, TCPS_CLOSED); 2329 (void) tp->t_fb->tfb_tcp_output(tp); 2330 TCPSTAT_INC(tcps_drops); 2331 } else 2332 TCPSTAT_INC(tcps_conndrops); 2333 if (errno == ETIMEDOUT && tp->t_softerror) 2334 errno = tp->t_softerror; 2335 so->so_error = errno; 2336 return (tcp_close(tp)); 2337 } 2338 2339 void 2340 tcp_discardcb(struct tcpcb *tp) 2341 { 2342 struct inpcb *inp = tp->t_inpcb; 2343 struct socket *so = inp->inp_socket; 2344 #ifdef INET6 2345 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2346 #endif /* INET6 */ 2347 int released __unused; 2348 2349 INP_WLOCK_ASSERT(inp); 2350 2351 /* 2352 * Make sure that all of our timers are stopped before we delete the 2353 * PCB. 2354 * 2355 * If stopping a timer fails, we schedule a discard function in same 2356 * callout, and the last discard function called will take care of 2357 * deleting the tcpcb. 2358 */ 2359 tp->t_timers->tt_draincnt = 0; 2360 tcp_timer_stop(tp, TT_REXMT); 2361 tcp_timer_stop(tp, TT_PERSIST); 2362 tcp_timer_stop(tp, TT_KEEP); 2363 tcp_timer_stop(tp, TT_2MSL); 2364 tcp_timer_stop(tp, TT_DELACK); 2365 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2366 /* 2367 * Call the stop-all function of the methods, 2368 * this function should call the tcp_timer_stop() 2369 * method with each of the function specific timeouts. 2370 * That stop will be called via the tfb_tcp_timer_stop() 2371 * which should use the async drain function of the 2372 * callout system (see tcp_var.h). 2373 */ 2374 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2375 } 2376 2377 /* free the reassembly queue, if any */ 2378 tcp_reass_flush(tp); 2379 2380 #ifdef TCP_OFFLOAD 2381 /* Disconnect offload device, if any. */ 2382 if (tp->t_flags & TF_TOE) 2383 tcp_offload_detach(tp); 2384 #endif 2385 2386 tcp_free_sackholes(tp); 2387 2388 #ifdef TCPPCAP 2389 /* Free the TCP PCAP queues. */ 2390 tcp_pcap_drain(&(tp->t_inpkts)); 2391 tcp_pcap_drain(&(tp->t_outpkts)); 2392 #endif 2393 2394 /* Allow the CC algorithm to clean up after itself. */ 2395 if (CC_ALGO(tp)->cb_destroy != NULL) 2396 CC_ALGO(tp)->cb_destroy(tp->ccv); 2397 CC_DATA(tp) = NULL; 2398 2399 #ifdef TCP_HHOOK 2400 khelp_destroy_osd(tp->osd); 2401 #endif 2402 #ifdef STATS 2403 stats_blob_destroy(tp->t_stats); 2404 #endif 2405 2406 CC_ALGO(tp) = NULL; 2407 inp->inp_ppcb = NULL; 2408 if (tp->t_timers->tt_draincnt == 0) { 2409 /* We own the last reference on tcpcb, let's free it. */ 2410 #ifdef TCP_BLACKBOX 2411 tcp_log_tcpcbfini(tp); 2412 #endif 2413 TCPSTATES_DEC(tp->t_state); 2414 if (tp->t_fb->tfb_tcp_fb_fini) 2415 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2416 2417 /* 2418 * If we got enough samples through the srtt filter, 2419 * save the rtt and rttvar in the routing entry. 2420 * 'Enough' is arbitrarily defined as 4 rtt samples. 2421 * 4 samples is enough for the srtt filter to converge 2422 * to within enough % of the correct value; fewer samples 2423 * and we could save a bogus rtt. The danger is not high 2424 * as tcp quickly recovers from everything. 2425 * XXX: Works very well but needs some more statistics! 2426 * 2427 * XXXRRS: Updating must be after the stack fini() since 2428 * that may be converting some internal representation of 2429 * say srtt etc into the general one used by other stacks. 2430 * Lets also at least protect against the so being NULL 2431 * as RW stated below. 2432 */ 2433 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2434 struct hc_metrics_lite metrics; 2435 uint32_t ssthresh; 2436 2437 bzero(&metrics, sizeof(metrics)); 2438 /* 2439 * Update the ssthresh always when the conditions below 2440 * are satisfied. This gives us better new start value 2441 * for the congestion avoidance for new connections. 2442 * ssthresh is only set if packet loss occurred on a session. 2443 * 2444 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2445 * being torn down. Ideally this code would not use 'so'. 2446 */ 2447 ssthresh = tp->snd_ssthresh; 2448 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2449 /* 2450 * convert the limit from user data bytes to 2451 * packets then to packet data bytes. 2452 */ 2453 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2454 if (ssthresh < 2) 2455 ssthresh = 2; 2456 ssthresh *= (tp->t_maxseg + 2457 #ifdef INET6 2458 (isipv6 ? sizeof (struct ip6_hdr) + 2459 sizeof (struct tcphdr) : 2460 #endif 2461 sizeof (struct tcpiphdr) 2462 #ifdef INET6 2463 ) 2464 #endif 2465 ); 2466 } else 2467 ssthresh = 0; 2468 metrics.rmx_ssthresh = ssthresh; 2469 2470 metrics.rmx_rtt = tp->t_srtt; 2471 metrics.rmx_rttvar = tp->t_rttvar; 2472 metrics.rmx_cwnd = tp->snd_cwnd; 2473 metrics.rmx_sendpipe = 0; 2474 metrics.rmx_recvpipe = 0; 2475 2476 tcp_hc_update(&inp->inp_inc, &metrics); 2477 } 2478 refcount_release(&tp->t_fb->tfb_refcnt); 2479 tp->t_inpcb = NULL; 2480 uma_zfree(V_tcpcb_zone, tp); 2481 released = in_pcbrele_wlocked(inp); 2482 KASSERT(!released, ("%s: inp %p should not have been released " 2483 "here", __func__, inp)); 2484 } 2485 } 2486 2487 void 2488 tcp_timer_discard(void *ptp) 2489 { 2490 struct inpcb *inp; 2491 struct tcpcb *tp; 2492 struct epoch_tracker et; 2493 2494 tp = (struct tcpcb *)ptp; 2495 CURVNET_SET(tp->t_vnet); 2496 NET_EPOCH_ENTER(et); 2497 inp = tp->t_inpcb; 2498 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 2499 __func__, tp)); 2500 INP_WLOCK(inp); 2501 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 2502 ("%s: tcpcb has to be stopped here", __func__)); 2503 tp->t_timers->tt_draincnt--; 2504 if (tp->t_timers->tt_draincnt == 0) { 2505 /* We own the last reference on this tcpcb, let's free it. */ 2506 #ifdef TCP_BLACKBOX 2507 tcp_log_tcpcbfini(tp); 2508 #endif 2509 TCPSTATES_DEC(tp->t_state); 2510 if (tp->t_fb->tfb_tcp_fb_fini) 2511 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2512 refcount_release(&tp->t_fb->tfb_refcnt); 2513 tp->t_inpcb = NULL; 2514 uma_zfree(V_tcpcb_zone, tp); 2515 if (in_pcbrele_wlocked(inp)) { 2516 NET_EPOCH_EXIT(et); 2517 CURVNET_RESTORE(); 2518 return; 2519 } 2520 } 2521 INP_WUNLOCK(inp); 2522 NET_EPOCH_EXIT(et); 2523 CURVNET_RESTORE(); 2524 } 2525 2526 /* 2527 * Attempt to close a TCP control block, marking it as dropped, and freeing 2528 * the socket if we hold the only reference. 2529 */ 2530 struct tcpcb * 2531 tcp_close(struct tcpcb *tp) 2532 { 2533 struct inpcb *inp = tp->t_inpcb; 2534 struct socket *so; 2535 2536 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2537 INP_WLOCK_ASSERT(inp); 2538 2539 #ifdef TCP_OFFLOAD 2540 if (tp->t_state == TCPS_LISTEN) 2541 tcp_offload_listen_stop(tp); 2542 #endif 2543 /* 2544 * This releases the TFO pending counter resource for TFO listen 2545 * sockets as well as passively-created TFO sockets that transition 2546 * from SYN_RECEIVED to CLOSED. 2547 */ 2548 if (tp->t_tfo_pending) { 2549 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2550 tp->t_tfo_pending = NULL; 2551 } 2552 in_pcbdrop(inp); 2553 TCPSTAT_INC(tcps_closed); 2554 if (tp->t_state != TCPS_CLOSED) 2555 tcp_state_change(tp, TCPS_CLOSED); 2556 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2557 so = inp->inp_socket; 2558 soisdisconnected(so); 2559 if (inp->inp_flags & INP_SOCKREF) { 2560 KASSERT(so->so_state & SS_PROTOREF, 2561 ("tcp_close: !SS_PROTOREF")); 2562 inp->inp_flags &= ~INP_SOCKREF; 2563 INP_WUNLOCK(inp); 2564 SOCK_LOCK(so); 2565 so->so_state &= ~SS_PROTOREF; 2566 sofree(so); 2567 return (NULL); 2568 } 2569 return (tp); 2570 } 2571 2572 void 2573 tcp_drain(void) 2574 { 2575 VNET_ITERATOR_DECL(vnet_iter); 2576 2577 if (!do_tcpdrain) 2578 return; 2579 2580 VNET_LIST_RLOCK_NOSLEEP(); 2581 VNET_FOREACH(vnet_iter) { 2582 CURVNET_SET(vnet_iter); 2583 struct inpcb *inpb; 2584 struct tcpcb *tcpb; 2585 2586 /* 2587 * Walk the tcpbs, if existing, and flush the reassembly queue, 2588 * if there is one... 2589 * XXX: The "Net/3" implementation doesn't imply that the TCP 2590 * reassembly queue should be flushed, but in a situation 2591 * where we're really low on mbufs, this is potentially 2592 * useful. 2593 */ 2594 INP_INFO_WLOCK(&V_tcbinfo); 2595 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2596 INP_WLOCK(inpb); 2597 if (inpb->inp_flags & INP_TIMEWAIT) { 2598 INP_WUNLOCK(inpb); 2599 continue; 2600 } 2601 if ((tcpb = intotcpcb(inpb)) != NULL) { 2602 tcp_reass_flush(tcpb); 2603 tcp_clean_sackreport(tcpb); 2604 #ifdef TCP_BLACKBOX 2605 tcp_log_drain(tcpb); 2606 #endif 2607 #ifdef TCPPCAP 2608 if (tcp_pcap_aggressive_free) { 2609 /* Free the TCP PCAP queues. */ 2610 tcp_pcap_drain(&(tcpb->t_inpkts)); 2611 tcp_pcap_drain(&(tcpb->t_outpkts)); 2612 } 2613 #endif 2614 } 2615 INP_WUNLOCK(inpb); 2616 } 2617 INP_INFO_WUNLOCK(&V_tcbinfo); 2618 CURVNET_RESTORE(); 2619 } 2620 VNET_LIST_RUNLOCK_NOSLEEP(); 2621 } 2622 2623 /* 2624 * Notify a tcp user of an asynchronous error; 2625 * store error as soft error, but wake up user 2626 * (for now, won't do anything until can select for soft error). 2627 * 2628 * Do not wake up user since there currently is no mechanism for 2629 * reporting soft errors (yet - a kqueue filter may be added). 2630 */ 2631 static struct inpcb * 2632 tcp_notify(struct inpcb *inp, int error) 2633 { 2634 struct tcpcb *tp; 2635 2636 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2637 INP_WLOCK_ASSERT(inp); 2638 2639 if ((inp->inp_flags & INP_TIMEWAIT) || 2640 (inp->inp_flags & INP_DROPPED)) 2641 return (inp); 2642 2643 tp = intotcpcb(inp); 2644 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2645 2646 /* 2647 * Ignore some errors if we are hooked up. 2648 * If connection hasn't completed, has retransmitted several times, 2649 * and receives a second error, give up now. This is better 2650 * than waiting a long time to establish a connection that 2651 * can never complete. 2652 */ 2653 if (tp->t_state == TCPS_ESTABLISHED && 2654 (error == EHOSTUNREACH || error == ENETUNREACH || 2655 error == EHOSTDOWN)) { 2656 if (inp->inp_route.ro_nh) { 2657 NH_FREE(inp->inp_route.ro_nh); 2658 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2659 } 2660 return (inp); 2661 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2662 tp->t_softerror) { 2663 tp = tcp_drop(tp, error); 2664 if (tp != NULL) 2665 return (inp); 2666 else 2667 return (NULL); 2668 } else { 2669 tp->t_softerror = error; 2670 return (inp); 2671 } 2672 #if 0 2673 wakeup( &so->so_timeo); 2674 sorwakeup(so); 2675 sowwakeup(so); 2676 #endif 2677 } 2678 2679 static int 2680 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2681 { 2682 struct epoch_tracker et; 2683 struct inpcb *inp; 2684 struct xinpgen xig; 2685 int error; 2686 2687 if (req->newptr != NULL) 2688 return (EPERM); 2689 2690 if (req->oldptr == NULL) { 2691 int n; 2692 2693 n = V_tcbinfo.ipi_count + 2694 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2695 n += imax(n / 8, 10); 2696 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2697 return (0); 2698 } 2699 2700 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2701 return (error); 2702 2703 bzero(&xig, sizeof(xig)); 2704 xig.xig_len = sizeof xig; 2705 xig.xig_count = V_tcbinfo.ipi_count + 2706 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2707 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2708 xig.xig_sogen = so_gencnt; 2709 error = SYSCTL_OUT(req, &xig, sizeof xig); 2710 if (error) 2711 return (error); 2712 2713 error = syncache_pcblist(req); 2714 if (error) 2715 return (error); 2716 2717 NET_EPOCH_ENTER(et); 2718 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); 2719 inp != NULL; 2720 inp = CK_LIST_NEXT(inp, inp_list)) { 2721 INP_RLOCK(inp); 2722 if (inp->inp_gencnt <= xig.xig_gen) { 2723 int crerr; 2724 2725 /* 2726 * XXX: This use of cr_cansee(), introduced with 2727 * TCP state changes, is not quite right, but for 2728 * now, better than nothing. 2729 */ 2730 if (inp->inp_flags & INP_TIMEWAIT) { 2731 if (intotw(inp) != NULL) 2732 crerr = cr_cansee(req->td->td_ucred, 2733 intotw(inp)->tw_cred); 2734 else 2735 crerr = EINVAL; /* Skip this inp. */ 2736 } else 2737 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2738 if (crerr == 0) { 2739 struct xtcpcb xt; 2740 2741 tcp_inptoxtp(inp, &xt); 2742 INP_RUNLOCK(inp); 2743 error = SYSCTL_OUT(req, &xt, sizeof xt); 2744 if (error) 2745 break; 2746 else 2747 continue; 2748 } 2749 } 2750 INP_RUNLOCK(inp); 2751 } 2752 NET_EPOCH_EXIT(et); 2753 2754 if (!error) { 2755 /* 2756 * Give the user an updated idea of our state. 2757 * If the generation differs from what we told 2758 * her before, she knows that something happened 2759 * while we were processing this request, and it 2760 * might be necessary to retry. 2761 */ 2762 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2763 xig.xig_sogen = so_gencnt; 2764 xig.xig_count = V_tcbinfo.ipi_count + 2765 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2766 error = SYSCTL_OUT(req, &xig, sizeof xig); 2767 } 2768 2769 return (error); 2770 } 2771 2772 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2773 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2774 NULL, 0, tcp_pcblist, "S,xtcpcb", 2775 "List of active TCP connections"); 2776 2777 #ifdef INET 2778 static int 2779 tcp_getcred(SYSCTL_HANDLER_ARGS) 2780 { 2781 struct xucred xuc; 2782 struct sockaddr_in addrs[2]; 2783 struct epoch_tracker et; 2784 struct inpcb *inp; 2785 int error; 2786 2787 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2788 if (error) 2789 return (error); 2790 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2791 if (error) 2792 return (error); 2793 NET_EPOCH_ENTER(et); 2794 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2795 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2796 NET_EPOCH_EXIT(et); 2797 if (inp != NULL) { 2798 if (inp->inp_socket == NULL) 2799 error = ENOENT; 2800 if (error == 0) 2801 error = cr_canseeinpcb(req->td->td_ucred, inp); 2802 if (error == 0) 2803 cru2x(inp->inp_cred, &xuc); 2804 INP_RUNLOCK(inp); 2805 } else 2806 error = ENOENT; 2807 if (error == 0) 2808 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2809 return (error); 2810 } 2811 2812 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2813 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2814 0, 0, tcp_getcred, "S,xucred", 2815 "Get the xucred of a TCP connection"); 2816 #endif /* INET */ 2817 2818 #ifdef INET6 2819 static int 2820 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2821 { 2822 struct epoch_tracker et; 2823 struct xucred xuc; 2824 struct sockaddr_in6 addrs[2]; 2825 struct inpcb *inp; 2826 int error; 2827 #ifdef INET 2828 int mapped = 0; 2829 #endif 2830 2831 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2832 if (error) 2833 return (error); 2834 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2835 if (error) 2836 return (error); 2837 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2838 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2839 return (error); 2840 } 2841 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2842 #ifdef INET 2843 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2844 mapped = 1; 2845 else 2846 #endif 2847 return (EINVAL); 2848 } 2849 2850 NET_EPOCH_ENTER(et); 2851 #ifdef INET 2852 if (mapped == 1) 2853 inp = in_pcblookup(&V_tcbinfo, 2854 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2855 addrs[1].sin6_port, 2856 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2857 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2858 else 2859 #endif 2860 inp = in6_pcblookup(&V_tcbinfo, 2861 &addrs[1].sin6_addr, addrs[1].sin6_port, 2862 &addrs[0].sin6_addr, addrs[0].sin6_port, 2863 INPLOOKUP_RLOCKPCB, NULL); 2864 NET_EPOCH_EXIT(et); 2865 if (inp != NULL) { 2866 if (inp->inp_socket == NULL) 2867 error = ENOENT; 2868 if (error == 0) 2869 error = cr_canseeinpcb(req->td->td_ucred, inp); 2870 if (error == 0) 2871 cru2x(inp->inp_cred, &xuc); 2872 INP_RUNLOCK(inp); 2873 } else 2874 error = ENOENT; 2875 if (error == 0) 2876 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2877 return (error); 2878 } 2879 2880 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2881 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2882 0, 0, tcp6_getcred, "S,xucred", 2883 "Get the xucred of a TCP6 connection"); 2884 #endif /* INET6 */ 2885 2886 #ifdef INET 2887 /* Path MTU to try next when a fragmentation-needed message is received. */ 2888 static inline int 2889 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2890 { 2891 int mtu = ntohs(icp->icmp_nextmtu); 2892 2893 /* If no alternative MTU was proposed, try the next smaller one. */ 2894 if (!mtu) 2895 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2896 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2897 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2898 2899 return (mtu); 2900 } 2901 2902 static void 2903 tcp_ctlinput_with_port(int cmd, struct sockaddr *sa, void *vip, uint16_t port) 2904 { 2905 struct ip *ip = vip; 2906 struct tcphdr *th; 2907 struct in_addr faddr; 2908 struct inpcb *inp; 2909 struct tcpcb *tp; 2910 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2911 struct icmp *icp; 2912 struct in_conninfo inc; 2913 tcp_seq icmp_tcp_seq; 2914 int mtu; 2915 2916 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2917 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2918 return; 2919 2920 if (cmd == PRC_MSGSIZE) 2921 notify = tcp_mtudisc_notify; 2922 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2923 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2924 cmd == PRC_TIMXCEED_INTRANS) && ip) 2925 notify = tcp_drop_syn_sent; 2926 2927 /* 2928 * Hostdead is ugly because it goes linearly through all PCBs. 2929 * XXX: We never get this from ICMP, otherwise it makes an 2930 * excellent DoS attack on machines with many connections. 2931 */ 2932 else if (cmd == PRC_HOSTDEAD) 2933 ip = NULL; 2934 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2935 return; 2936 2937 if (ip == NULL) { 2938 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2939 return; 2940 } 2941 2942 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2943 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2944 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2945 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2946 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2947 /* signal EHOSTDOWN, as it flushes the cached route */ 2948 inp = (*notify)(inp, EHOSTDOWN); 2949 goto out; 2950 } 2951 icmp_tcp_seq = th->th_seq; 2952 if (inp != NULL) { 2953 if (!(inp->inp_flags & INP_TIMEWAIT) && 2954 !(inp->inp_flags & INP_DROPPED) && 2955 !(inp->inp_socket == NULL)) { 2956 tp = intotcpcb(inp); 2957 #ifdef TCP_OFFLOAD 2958 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 2959 /* 2960 * MTU discovery for offloaded connections. Let 2961 * the TOE driver verify seq# and process it. 2962 */ 2963 mtu = tcp_next_pmtu(icp, ip); 2964 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2965 goto out; 2966 } 2967 #endif 2968 if (tp->t_port != port) { 2969 goto out; 2970 } 2971 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2972 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2973 if (cmd == PRC_MSGSIZE) { 2974 /* 2975 * MTU discovery: we got a needfrag and 2976 * will potentially try a lower MTU. 2977 */ 2978 mtu = tcp_next_pmtu(icp, ip); 2979 2980 /* 2981 * Only process the offered MTU if it 2982 * is smaller than the current one. 2983 */ 2984 if (mtu < tp->t_maxseg + 2985 sizeof(struct tcpiphdr)) { 2986 bzero(&inc, sizeof(inc)); 2987 inc.inc_faddr = faddr; 2988 inc.inc_fibnum = 2989 inp->inp_inc.inc_fibnum; 2990 tcp_hc_updatemtu(&inc, mtu); 2991 tcp_mtudisc(inp, mtu); 2992 } 2993 } else 2994 inp = (*notify)(inp, 2995 inetctlerrmap[cmd]); 2996 } 2997 } 2998 } else { 2999 bzero(&inc, sizeof(inc)); 3000 inc.inc_fport = th->th_dport; 3001 inc.inc_lport = th->th_sport; 3002 inc.inc_faddr = faddr; 3003 inc.inc_laddr = ip->ip_src; 3004 syncache_unreach(&inc, icmp_tcp_seq, port); 3005 } 3006 out: 3007 if (inp != NULL) 3008 INP_WUNLOCK(inp); 3009 } 3010 3011 void 3012 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 3013 { 3014 tcp_ctlinput_with_port(cmd, sa, vip, htons(0)); 3015 } 3016 3017 void 3018 tcp_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *vip, void *unused) 3019 { 3020 /* Its a tunneled TCP over UDP icmp */ 3021 struct ip *outer_ip, *inner_ip; 3022 struct icmp *icmp; 3023 struct udphdr *udp; 3024 struct tcphdr *th, ttemp; 3025 int i_hlen, o_len; 3026 uint16_t port; 3027 3028 inner_ip = (struct ip *)vip; 3029 icmp = (struct icmp *)((caddr_t)inner_ip - 3030 (sizeof(struct icmp) - sizeof(struct ip))); 3031 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 3032 i_hlen = inner_ip->ip_hl << 2; 3033 o_len = ntohs(outer_ip->ip_len); 3034 if (o_len < 3035 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 3036 /* Not enough data present */ 3037 return; 3038 } 3039 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 3040 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 3041 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3042 return; 3043 } 3044 port = udp->uh_dport; 3045 th = (struct tcphdr *)(udp + 1); 3046 memcpy(&ttemp, th, sizeof(struct tcphdr)); 3047 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 3048 /* Now adjust down the size of the outer IP header */ 3049 o_len -= sizeof(struct udphdr); 3050 outer_ip->ip_len = htons(o_len); 3051 /* Now call in to the normal handling code */ 3052 tcp_ctlinput_with_port(cmd, sa, vip, port); 3053 } 3054 #endif /* INET */ 3055 3056 #ifdef INET6 3057 static inline int 3058 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3059 { 3060 int mtu = ntohl(icmp6->icmp6_mtu); 3061 3062 /* 3063 * If no alternative MTU was proposed, or the proposed MTU was too 3064 * small, set to the min. 3065 */ 3066 if (mtu < IPV6_MMTU) 3067 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 3068 return (mtu); 3069 } 3070 3071 static void 3072 tcp6_ctlinput_with_port(int cmd, struct sockaddr *sa, void *d, uint16_t port) 3073 { 3074 struct in6_addr *dst; 3075 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 3076 struct ip6_hdr *ip6; 3077 struct mbuf *m; 3078 struct inpcb *inp; 3079 struct tcpcb *tp; 3080 struct icmp6_hdr *icmp6; 3081 struct ip6ctlparam *ip6cp = NULL; 3082 const struct sockaddr_in6 *sa6_src = NULL; 3083 struct in_conninfo inc; 3084 struct tcp_ports { 3085 uint16_t th_sport; 3086 uint16_t th_dport; 3087 } t_ports; 3088 tcp_seq icmp_tcp_seq; 3089 unsigned int mtu; 3090 unsigned int off; 3091 3092 if (sa->sa_family != AF_INET6 || 3093 sa->sa_len != sizeof(struct sockaddr_in6)) 3094 return; 3095 3096 /* if the parameter is from icmp6, decode it. */ 3097 if (d != NULL) { 3098 ip6cp = (struct ip6ctlparam *)d; 3099 icmp6 = ip6cp->ip6c_icmp6; 3100 m = ip6cp->ip6c_m; 3101 ip6 = ip6cp->ip6c_ip6; 3102 off = ip6cp->ip6c_off; 3103 sa6_src = ip6cp->ip6c_src; 3104 dst = ip6cp->ip6c_finaldst; 3105 } else { 3106 m = NULL; 3107 ip6 = NULL; 3108 off = 0; /* fool gcc */ 3109 sa6_src = &sa6_any; 3110 dst = NULL; 3111 } 3112 3113 if (cmd == PRC_MSGSIZE) 3114 notify = tcp_mtudisc_notify; 3115 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 3116 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 3117 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 3118 notify = tcp_drop_syn_sent; 3119 3120 /* 3121 * Hostdead is ugly because it goes linearly through all PCBs. 3122 * XXX: We never get this from ICMP, otherwise it makes an 3123 * excellent DoS attack on machines with many connections. 3124 */ 3125 else if (cmd == PRC_HOSTDEAD) 3126 ip6 = NULL; 3127 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 3128 return; 3129 3130 if (ip6 == NULL) { 3131 in6_pcbnotify(&V_tcbinfo, sa, 0, 3132 (const struct sockaddr *)sa6_src, 3133 0, cmd, NULL, notify); 3134 return; 3135 } 3136 3137 /* Check if we can safely get the ports from the tcp hdr */ 3138 if (m == NULL || 3139 (m->m_pkthdr.len < 3140 (int32_t) (off + sizeof(struct tcp_ports)))) { 3141 return; 3142 } 3143 bzero(&t_ports, sizeof(struct tcp_ports)); 3144 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3145 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3146 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3147 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 3148 /* signal EHOSTDOWN, as it flushes the cached route */ 3149 inp = (*notify)(inp, EHOSTDOWN); 3150 goto out; 3151 } 3152 off += sizeof(struct tcp_ports); 3153 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3154 goto out; 3155 } 3156 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3157 if (inp != NULL) { 3158 if (!(inp->inp_flags & INP_TIMEWAIT) && 3159 !(inp->inp_flags & INP_DROPPED) && 3160 !(inp->inp_socket == NULL)) { 3161 tp = intotcpcb(inp); 3162 #ifdef TCP_OFFLOAD 3163 if (tp->t_flags & TF_TOE && cmd == PRC_MSGSIZE) { 3164 /* MTU discovery for offloaded connections. */ 3165 mtu = tcp6_next_pmtu(icmp6); 3166 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3167 goto out; 3168 } 3169 #endif 3170 if (tp->t_port != port) { 3171 goto out; 3172 } 3173 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3174 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3175 if (cmd == PRC_MSGSIZE) { 3176 /* 3177 * MTU discovery: 3178 * If we got a needfrag set the MTU 3179 * in the route to the suggested new 3180 * value (if given) and then notify. 3181 */ 3182 mtu = tcp6_next_pmtu(icmp6); 3183 3184 bzero(&inc, sizeof(inc)); 3185 inc.inc_fibnum = M_GETFIB(m); 3186 inc.inc_flags |= INC_ISIPV6; 3187 inc.inc6_faddr = *dst; 3188 if (in6_setscope(&inc.inc6_faddr, 3189 m->m_pkthdr.rcvif, NULL)) 3190 goto out; 3191 /* 3192 * Only process the offered MTU if it 3193 * is smaller than the current one. 3194 */ 3195 if (mtu < tp->t_maxseg + 3196 sizeof (struct tcphdr) + 3197 sizeof (struct ip6_hdr)) { 3198 tcp_hc_updatemtu(&inc, mtu); 3199 tcp_mtudisc(inp, mtu); 3200 ICMP6STAT_INC(icp6s_pmtuchg); 3201 } 3202 } else 3203 inp = (*notify)(inp, 3204 inet6ctlerrmap[cmd]); 3205 } 3206 } 3207 } else { 3208 bzero(&inc, sizeof(inc)); 3209 inc.inc_fibnum = M_GETFIB(m); 3210 inc.inc_flags |= INC_ISIPV6; 3211 inc.inc_fport = t_ports.th_dport; 3212 inc.inc_lport = t_ports.th_sport; 3213 inc.inc6_faddr = *dst; 3214 inc.inc6_laddr = ip6->ip6_src; 3215 syncache_unreach(&inc, icmp_tcp_seq, port); 3216 } 3217 out: 3218 if (inp != NULL) 3219 INP_WUNLOCK(inp); 3220 } 3221 3222 void 3223 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 3224 { 3225 tcp6_ctlinput_with_port(cmd, sa, d, htons(0)); 3226 } 3227 3228 void 3229 tcp6_ctlinput_viaudp(int cmd, struct sockaddr *sa, void *d, void *unused) 3230 { 3231 struct ip6ctlparam *ip6cp; 3232 struct mbuf *m; 3233 struct udphdr *udp; 3234 uint16_t port; 3235 3236 ip6cp = (struct ip6ctlparam *)d; 3237 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3238 if (m == NULL) { 3239 return; 3240 } 3241 udp = mtod(m, struct udphdr *); 3242 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3243 return; 3244 } 3245 port = udp->uh_dport; 3246 m_adj(m, sizeof(struct udphdr)); 3247 if ((m->m_flags & M_PKTHDR) == 0) { 3248 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3249 } 3250 /* Now call in to the normal handling code */ 3251 tcp6_ctlinput_with_port(cmd, sa, d, port); 3252 } 3253 3254 #endif /* INET6 */ 3255 3256 static uint32_t 3257 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3258 { 3259 SIPHASH_CTX ctx; 3260 uint32_t hash[2]; 3261 3262 KASSERT(len >= SIPHASH_KEY_LENGTH, 3263 ("%s: keylen %u too short ", __func__, len)); 3264 SipHash24_Init(&ctx); 3265 SipHash_SetKey(&ctx, (uint8_t *)key); 3266 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3267 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3268 switch (inc->inc_flags & INC_ISIPV6) { 3269 #ifdef INET 3270 case 0: 3271 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3272 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3273 break; 3274 #endif 3275 #ifdef INET6 3276 case INC_ISIPV6: 3277 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3278 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3279 break; 3280 #endif 3281 } 3282 SipHash_Final((uint8_t *)hash, &ctx); 3283 3284 return (hash[0] ^ hash[1]); 3285 } 3286 3287 uint32_t 3288 tcp_new_ts_offset(struct in_conninfo *inc) 3289 { 3290 struct in_conninfo inc_store, *local_inc; 3291 3292 if (!V_tcp_ts_offset_per_conn) { 3293 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3294 inc_store.inc_lport = 0; 3295 inc_store.inc_fport = 0; 3296 local_inc = &inc_store; 3297 } else { 3298 local_inc = inc; 3299 } 3300 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3301 sizeof(V_ts_offset_secret))); 3302 } 3303 3304 /* 3305 * Following is where TCP initial sequence number generation occurs. 3306 * 3307 * There are two places where we must use initial sequence numbers: 3308 * 1. In SYN-ACK packets. 3309 * 2. In SYN packets. 3310 * 3311 * All ISNs for SYN-ACK packets are generated by the syncache. See 3312 * tcp_syncache.c for details. 3313 * 3314 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3315 * depends on this property. In addition, these ISNs should be 3316 * unguessable so as to prevent connection hijacking. To satisfy 3317 * the requirements of this situation, the algorithm outlined in 3318 * RFC 1948 is used, with only small modifications. 3319 * 3320 * Implementation details: 3321 * 3322 * Time is based off the system timer, and is corrected so that it 3323 * increases by one megabyte per second. This allows for proper 3324 * recycling on high speed LANs while still leaving over an hour 3325 * before rollover. 3326 * 3327 * As reading the *exact* system time is too expensive to be done 3328 * whenever setting up a TCP connection, we increment the time 3329 * offset in two ways. First, a small random positive increment 3330 * is added to isn_offset for each connection that is set up. 3331 * Second, the function tcp_isn_tick fires once per clock tick 3332 * and increments isn_offset as necessary so that sequence numbers 3333 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3334 * random positive increments serve only to ensure that the same 3335 * exact sequence number is never sent out twice (as could otherwise 3336 * happen when a port is recycled in less than the system tick 3337 * interval.) 3338 * 3339 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3340 * between seeding of isn_secret. This is normally set to zero, 3341 * as reseeding should not be necessary. 3342 * 3343 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3344 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3345 * general, this means holding an exclusive (write) lock. 3346 */ 3347 3348 #define ISN_BYTES_PER_SECOND 1048576 3349 #define ISN_STATIC_INCREMENT 4096 3350 #define ISN_RANDOM_INCREMENT (4096 - 1) 3351 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3352 3353 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3354 VNET_DEFINE_STATIC(int, isn_last); 3355 VNET_DEFINE_STATIC(int, isn_last_reseed); 3356 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3357 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3358 3359 #define V_isn_secret VNET(isn_secret) 3360 #define V_isn_last VNET(isn_last) 3361 #define V_isn_last_reseed VNET(isn_last_reseed) 3362 #define V_isn_offset VNET(isn_offset) 3363 #define V_isn_offset_old VNET(isn_offset_old) 3364 3365 tcp_seq 3366 tcp_new_isn(struct in_conninfo *inc) 3367 { 3368 tcp_seq new_isn; 3369 u_int32_t projected_offset; 3370 3371 ISN_LOCK(); 3372 /* Seed if this is the first use, reseed if requested. */ 3373 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3374 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3375 < (u_int)ticks))) { 3376 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3377 V_isn_last_reseed = ticks; 3378 } 3379 3380 /* Compute the hash and return the ISN. */ 3381 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3382 sizeof(V_isn_secret)); 3383 V_isn_offset += ISN_STATIC_INCREMENT + 3384 (arc4random() & ISN_RANDOM_INCREMENT); 3385 if (ticks != V_isn_last) { 3386 projected_offset = V_isn_offset_old + 3387 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3388 if (SEQ_GT(projected_offset, V_isn_offset)) 3389 V_isn_offset = projected_offset; 3390 V_isn_offset_old = V_isn_offset; 3391 V_isn_last = ticks; 3392 } 3393 new_isn += V_isn_offset; 3394 ISN_UNLOCK(); 3395 return (new_isn); 3396 } 3397 3398 /* 3399 * When a specific ICMP unreachable message is received and the 3400 * connection state is SYN-SENT, drop the connection. This behavior 3401 * is controlled by the icmp_may_rst sysctl. 3402 */ 3403 struct inpcb * 3404 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3405 { 3406 struct tcpcb *tp; 3407 3408 NET_EPOCH_ASSERT(); 3409 INP_WLOCK_ASSERT(inp); 3410 3411 if ((inp->inp_flags & INP_TIMEWAIT) || 3412 (inp->inp_flags & INP_DROPPED)) 3413 return (inp); 3414 3415 tp = intotcpcb(inp); 3416 if (tp->t_state != TCPS_SYN_SENT) 3417 return (inp); 3418 3419 if (IS_FASTOPEN(tp->t_flags)) 3420 tcp_fastopen_disable_path(tp); 3421 3422 tp = tcp_drop(tp, errno); 3423 if (tp != NULL) 3424 return (inp); 3425 else 3426 return (NULL); 3427 } 3428 3429 /* 3430 * When `need fragmentation' ICMP is received, update our idea of the MSS 3431 * based on the new value. Also nudge TCP to send something, since we 3432 * know the packet we just sent was dropped. 3433 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3434 */ 3435 static struct inpcb * 3436 tcp_mtudisc_notify(struct inpcb *inp, int error) 3437 { 3438 3439 tcp_mtudisc(inp, -1); 3440 return (inp); 3441 } 3442 3443 static void 3444 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3445 { 3446 struct tcpcb *tp; 3447 struct socket *so; 3448 3449 INP_WLOCK_ASSERT(inp); 3450 if ((inp->inp_flags & INP_TIMEWAIT) || 3451 (inp->inp_flags & INP_DROPPED)) 3452 return; 3453 3454 tp = intotcpcb(inp); 3455 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3456 3457 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3458 3459 so = inp->inp_socket; 3460 SOCKBUF_LOCK(&so->so_snd); 3461 /* If the mss is larger than the socket buffer, decrease the mss. */ 3462 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3463 tp->t_maxseg = so->so_snd.sb_hiwat; 3464 SOCKBUF_UNLOCK(&so->so_snd); 3465 3466 TCPSTAT_INC(tcps_mturesent); 3467 tp->t_rtttime = 0; 3468 tp->snd_nxt = tp->snd_una; 3469 tcp_free_sackholes(tp); 3470 tp->snd_recover = tp->snd_max; 3471 if (tp->t_flags & TF_SACK_PERMIT) 3472 EXIT_FASTRECOVERY(tp->t_flags); 3473 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3474 /* 3475 * Conceptually the snd_nxt setting 3476 * and freeing sack holes should 3477 * be done by the default stacks 3478 * own tfb_tcp_mtu_chg(). 3479 */ 3480 tp->t_fb->tfb_tcp_mtu_chg(tp); 3481 } 3482 tp->t_fb->tfb_tcp_output(tp); 3483 } 3484 3485 #ifdef INET 3486 /* 3487 * Look-up the routing entry to the peer of this inpcb. If no route 3488 * is found and it cannot be allocated, then return 0. This routine 3489 * is called by TCP routines that access the rmx structure and by 3490 * tcp_mss_update to get the peer/interface MTU. 3491 */ 3492 uint32_t 3493 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3494 { 3495 struct nhop_object *nh; 3496 struct ifnet *ifp; 3497 uint32_t maxmtu = 0; 3498 3499 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3500 3501 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3502 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3503 if (nh == NULL) 3504 return (0); 3505 3506 ifp = nh->nh_ifp; 3507 maxmtu = nh->nh_mtu; 3508 3509 /* Report additional interface capabilities. */ 3510 if (cap != NULL) { 3511 if (ifp->if_capenable & IFCAP_TSO4 && 3512 ifp->if_hwassist & CSUM_TSO) { 3513 cap->ifcap |= CSUM_TSO; 3514 cap->tsomax = ifp->if_hw_tsomax; 3515 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3516 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3517 } 3518 } 3519 } 3520 return (maxmtu); 3521 } 3522 #endif /* INET */ 3523 3524 #ifdef INET6 3525 uint32_t 3526 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3527 { 3528 struct nhop_object *nh; 3529 struct in6_addr dst6; 3530 uint32_t scopeid; 3531 struct ifnet *ifp; 3532 uint32_t maxmtu = 0; 3533 3534 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3535 3536 if (inc->inc_flags & INC_IPV6MINMTU) 3537 return (IPV6_MMTU); 3538 3539 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3540 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3541 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3542 if (nh == NULL) 3543 return (0); 3544 3545 ifp = nh->nh_ifp; 3546 maxmtu = nh->nh_mtu; 3547 3548 /* Report additional interface capabilities. */ 3549 if (cap != NULL) { 3550 if (ifp->if_capenable & IFCAP_TSO6 && 3551 ifp->if_hwassist & CSUM_TSO) { 3552 cap->ifcap |= CSUM_TSO; 3553 cap->tsomax = ifp->if_hw_tsomax; 3554 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3555 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3556 } 3557 } 3558 } 3559 3560 return (maxmtu); 3561 } 3562 3563 /* 3564 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3565 * 3566 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3567 * The right place to do that is ip6_setpktopt() that has just been 3568 * executed. By the way it just filled ip6po_minmtu for us. 3569 */ 3570 void 3571 tcp6_use_min_mtu(struct tcpcb *tp) 3572 { 3573 struct inpcb *inp = tp->t_inpcb; 3574 3575 INP_WLOCK_ASSERT(inp); 3576 /* 3577 * In case of the IPV6_USE_MIN_MTU socket 3578 * option, the INC_IPV6MINMTU flag to announce 3579 * a corresponding MSS during the initial 3580 * handshake. If the TCP connection is not in 3581 * the front states, just reduce the MSS being 3582 * used. This avoids the sending of TCP 3583 * segments which will be fragmented at the 3584 * IPv6 layer. 3585 */ 3586 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3587 if ((tp->t_state >= TCPS_SYN_SENT) && 3588 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3589 struct ip6_pktopts *opt; 3590 3591 opt = inp->in6p_outputopts; 3592 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3593 tp->t_maxseg > TCP6_MSS) 3594 tp->t_maxseg = TCP6_MSS; 3595 } 3596 } 3597 #endif /* INET6 */ 3598 3599 /* 3600 * Calculate effective SMSS per RFC5681 definition for a given TCP 3601 * connection at its current state, taking into account SACK and etc. 3602 */ 3603 u_int 3604 tcp_maxseg(const struct tcpcb *tp) 3605 { 3606 u_int optlen; 3607 3608 if (tp->t_flags & TF_NOOPT) 3609 return (tp->t_maxseg); 3610 3611 /* 3612 * Here we have a simplified code from tcp_addoptions(), 3613 * without a proper loop, and having most of paddings hardcoded. 3614 * We might make mistakes with padding here in some edge cases, 3615 * but this is harmless, since result of tcp_maxseg() is used 3616 * only in cwnd and ssthresh estimations. 3617 */ 3618 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3619 if (tp->t_flags & TF_RCVD_TSTMP) 3620 optlen = TCPOLEN_TSTAMP_APPA; 3621 else 3622 optlen = 0; 3623 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3624 if (tp->t_flags & TF_SIGNATURE) 3625 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3626 #endif 3627 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3628 optlen += TCPOLEN_SACKHDR; 3629 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3630 optlen = PADTCPOLEN(optlen); 3631 } 3632 } else { 3633 if (tp->t_flags & TF_REQ_TSTMP) 3634 optlen = TCPOLEN_TSTAMP_APPA; 3635 else 3636 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3637 if (tp->t_flags & TF_REQ_SCALE) 3638 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3639 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3640 if (tp->t_flags & TF_SIGNATURE) 3641 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3642 #endif 3643 if (tp->t_flags & TF_SACK_PERMIT) 3644 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3645 } 3646 #undef PAD 3647 optlen = min(optlen, TCP_MAXOLEN); 3648 return (tp->t_maxseg - optlen); 3649 } 3650 3651 3652 u_int 3653 tcp_fixed_maxseg(const struct tcpcb *tp) 3654 { 3655 int optlen; 3656 3657 if (tp->t_flags & TF_NOOPT) 3658 return (tp->t_maxseg); 3659 3660 /* 3661 * Here we have a simplified code from tcp_addoptions(), 3662 * without a proper loop, and having most of paddings hardcoded. 3663 * We only consider fixed options that we would send every 3664 * time I.e. SACK is not considered. This is important 3665 * for cc modules to figure out what the modulo of the 3666 * cwnd should be. 3667 */ 3668 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3669 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3670 if (tp->t_flags & TF_RCVD_TSTMP) 3671 optlen = TCPOLEN_TSTAMP_APPA; 3672 else 3673 optlen = 0; 3674 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3675 if (tp->t_flags & TF_SIGNATURE) 3676 optlen += PAD(TCPOLEN_SIGNATURE); 3677 #endif 3678 } else { 3679 if (tp->t_flags & TF_REQ_TSTMP) 3680 optlen = TCPOLEN_TSTAMP_APPA; 3681 else 3682 optlen = PAD(TCPOLEN_MAXSEG); 3683 if (tp->t_flags & TF_REQ_SCALE) 3684 optlen += PAD(TCPOLEN_WINDOW); 3685 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3686 if (tp->t_flags & TF_SIGNATURE) 3687 optlen += PAD(TCPOLEN_SIGNATURE); 3688 #endif 3689 if (tp->t_flags & TF_SACK_PERMIT) 3690 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3691 } 3692 #undef PAD 3693 optlen = min(optlen, TCP_MAXOLEN); 3694 return (tp->t_maxseg - optlen); 3695 } 3696 3697 3698 3699 static int 3700 sysctl_drop(SYSCTL_HANDLER_ARGS) 3701 { 3702 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3703 struct sockaddr_storage addrs[2]; 3704 struct inpcb *inp; 3705 struct tcpcb *tp; 3706 struct tcptw *tw; 3707 struct sockaddr_in *fin, *lin; 3708 struct epoch_tracker et; 3709 #ifdef INET6 3710 struct sockaddr_in6 *fin6, *lin6; 3711 #endif 3712 int error; 3713 3714 inp = NULL; 3715 fin = lin = NULL; 3716 #ifdef INET6 3717 fin6 = lin6 = NULL; 3718 #endif 3719 error = 0; 3720 3721 if (req->oldptr != NULL || req->oldlen != 0) 3722 return (EINVAL); 3723 if (req->newptr == NULL) 3724 return (EPERM); 3725 if (req->newlen < sizeof(addrs)) 3726 return (ENOMEM); 3727 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3728 if (error) 3729 return (error); 3730 3731 switch (addrs[0].ss_family) { 3732 #ifdef INET6 3733 case AF_INET6: 3734 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3735 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3736 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3737 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3738 return (EINVAL); 3739 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3740 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3741 return (EINVAL); 3742 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3743 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3744 fin = (struct sockaddr_in *)&addrs[0]; 3745 lin = (struct sockaddr_in *)&addrs[1]; 3746 break; 3747 } 3748 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3749 if (error) 3750 return (error); 3751 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3752 if (error) 3753 return (error); 3754 break; 3755 #endif 3756 #ifdef INET 3757 case AF_INET: 3758 fin = (struct sockaddr_in *)&addrs[0]; 3759 lin = (struct sockaddr_in *)&addrs[1]; 3760 if (fin->sin_len != sizeof(struct sockaddr_in) || 3761 lin->sin_len != sizeof(struct sockaddr_in)) 3762 return (EINVAL); 3763 break; 3764 #endif 3765 default: 3766 return (EINVAL); 3767 } 3768 NET_EPOCH_ENTER(et); 3769 switch (addrs[0].ss_family) { 3770 #ifdef INET6 3771 case AF_INET6: 3772 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3773 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3774 INPLOOKUP_WLOCKPCB, NULL); 3775 break; 3776 #endif 3777 #ifdef INET 3778 case AF_INET: 3779 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3780 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3781 break; 3782 #endif 3783 } 3784 if (inp != NULL) { 3785 if (inp->inp_flags & INP_TIMEWAIT) { 3786 /* 3787 * XXXRW: There currently exists a state where an 3788 * inpcb is present, but its timewait state has been 3789 * discarded. For now, don't allow dropping of this 3790 * type of inpcb. 3791 */ 3792 tw = intotw(inp); 3793 if (tw != NULL) 3794 tcp_twclose(tw, 0); 3795 else 3796 INP_WUNLOCK(inp); 3797 } else if ((inp->inp_flags & INP_DROPPED) == 0 && 3798 !SOLISTENING(inp->inp_socket)) { 3799 tp = intotcpcb(inp); 3800 tp = tcp_drop(tp, ECONNABORTED); 3801 if (tp != NULL) 3802 INP_WUNLOCK(inp); 3803 } else 3804 INP_WUNLOCK(inp); 3805 } else 3806 error = ESRCH; 3807 NET_EPOCH_EXIT(et); 3808 return (error); 3809 } 3810 3811 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3812 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3813 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3814 "Drop TCP connection"); 3815 3816 #ifdef KERN_TLS 3817 static int 3818 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3819 { 3820 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3821 struct sockaddr_storage addrs[2]; 3822 struct inpcb *inp; 3823 struct sockaddr_in *fin, *lin; 3824 struct epoch_tracker et; 3825 #ifdef INET6 3826 struct sockaddr_in6 *fin6, *lin6; 3827 #endif 3828 int error; 3829 3830 inp = NULL; 3831 fin = lin = NULL; 3832 #ifdef INET6 3833 fin6 = lin6 = NULL; 3834 #endif 3835 error = 0; 3836 3837 if (req->oldptr != NULL || req->oldlen != 0) 3838 return (EINVAL); 3839 if (req->newptr == NULL) 3840 return (EPERM); 3841 if (req->newlen < sizeof(addrs)) 3842 return (ENOMEM); 3843 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3844 if (error) 3845 return (error); 3846 3847 switch (addrs[0].ss_family) { 3848 #ifdef INET6 3849 case AF_INET6: 3850 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3851 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3852 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3853 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3854 return (EINVAL); 3855 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3856 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3857 return (EINVAL); 3858 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3859 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3860 fin = (struct sockaddr_in *)&addrs[0]; 3861 lin = (struct sockaddr_in *)&addrs[1]; 3862 break; 3863 } 3864 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3865 if (error) 3866 return (error); 3867 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3868 if (error) 3869 return (error); 3870 break; 3871 #endif 3872 #ifdef INET 3873 case AF_INET: 3874 fin = (struct sockaddr_in *)&addrs[0]; 3875 lin = (struct sockaddr_in *)&addrs[1]; 3876 if (fin->sin_len != sizeof(struct sockaddr_in) || 3877 lin->sin_len != sizeof(struct sockaddr_in)) 3878 return (EINVAL); 3879 break; 3880 #endif 3881 default: 3882 return (EINVAL); 3883 } 3884 NET_EPOCH_ENTER(et); 3885 switch (addrs[0].ss_family) { 3886 #ifdef INET6 3887 case AF_INET6: 3888 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3889 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3890 INPLOOKUP_WLOCKPCB, NULL); 3891 break; 3892 #endif 3893 #ifdef INET 3894 case AF_INET: 3895 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3896 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3897 break; 3898 #endif 3899 } 3900 NET_EPOCH_EXIT(et); 3901 if (inp != NULL) { 3902 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3903 inp->inp_socket == NULL) { 3904 error = ECONNRESET; 3905 INP_WUNLOCK(inp); 3906 } else { 3907 struct socket *so; 3908 3909 so = inp->inp_socket; 3910 soref(so); 3911 error = ktls_set_tx_mode(so, 3912 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3913 INP_WUNLOCK(inp); 3914 sorele(so); 3915 } 3916 } else 3917 error = ESRCH; 3918 return (error); 3919 } 3920 3921 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3922 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3923 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3924 "Switch TCP connection to SW TLS"); 3925 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3926 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3927 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3928 "Switch TCP connection to ifnet TLS"); 3929 #endif 3930 3931 /* 3932 * Generate a standardized TCP log line for use throughout the 3933 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3934 * allow use in the interrupt context. 3935 * 3936 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3937 * NB: The function may return NULL if memory allocation failed. 3938 * 3939 * Due to header inclusion and ordering limitations the struct ip 3940 * and ip6_hdr pointers have to be passed as void pointers. 3941 */ 3942 char * 3943 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3944 const void *ip6hdr) 3945 { 3946 3947 /* Is logging enabled? */ 3948 if (V_tcp_log_in_vain == 0) 3949 return (NULL); 3950 3951 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3952 } 3953 3954 char * 3955 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3956 const void *ip6hdr) 3957 { 3958 3959 /* Is logging enabled? */ 3960 if (tcp_log_debug == 0) 3961 return (NULL); 3962 3963 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3964 } 3965 3966 static char * 3967 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3968 const void *ip6hdr) 3969 { 3970 char *s, *sp; 3971 size_t size; 3972 struct ip *ip; 3973 #ifdef INET6 3974 const struct ip6_hdr *ip6; 3975 3976 ip6 = (const struct ip6_hdr *)ip6hdr; 3977 #endif /* INET6 */ 3978 ip = (struct ip *)ip4hdr; 3979 3980 /* 3981 * The log line looks like this: 3982 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3983 */ 3984 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3985 sizeof(PRINT_TH_FLAGS) + 1 + 3986 #ifdef INET6 3987 2 * INET6_ADDRSTRLEN; 3988 #else 3989 2 * INET_ADDRSTRLEN; 3990 #endif /* INET6 */ 3991 3992 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3993 if (s == NULL) 3994 return (NULL); 3995 3996 strcat(s, "TCP: ["); 3997 sp = s + strlen(s); 3998 3999 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 4000 inet_ntoa_r(inc->inc_faddr, sp); 4001 sp = s + strlen(s); 4002 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4003 sp = s + strlen(s); 4004 inet_ntoa_r(inc->inc_laddr, sp); 4005 sp = s + strlen(s); 4006 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4007 #ifdef INET6 4008 } else if (inc) { 4009 ip6_sprintf(sp, &inc->inc6_faddr); 4010 sp = s + strlen(s); 4011 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4012 sp = s + strlen(s); 4013 ip6_sprintf(sp, &inc->inc6_laddr); 4014 sp = s + strlen(s); 4015 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4016 } else if (ip6 && th) { 4017 ip6_sprintf(sp, &ip6->ip6_src); 4018 sp = s + strlen(s); 4019 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4020 sp = s + strlen(s); 4021 ip6_sprintf(sp, &ip6->ip6_dst); 4022 sp = s + strlen(s); 4023 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4024 #endif /* INET6 */ 4025 #ifdef INET 4026 } else if (ip && th) { 4027 inet_ntoa_r(ip->ip_src, sp); 4028 sp = s + strlen(s); 4029 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4030 sp = s + strlen(s); 4031 inet_ntoa_r(ip->ip_dst, sp); 4032 sp = s + strlen(s); 4033 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4034 #endif /* INET */ 4035 } else { 4036 free(s, M_TCPLOG); 4037 return (NULL); 4038 } 4039 sp = s + strlen(s); 4040 if (th) 4041 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 4042 if (*(s + size - 1) != '\0') 4043 panic("%s: string too long", __func__); 4044 return (s); 4045 } 4046 4047 /* 4048 * A subroutine which makes it easy to track TCP state changes with DTrace. 4049 * This function shouldn't be called for t_state initializations that don't 4050 * correspond to actual TCP state transitions. 4051 */ 4052 void 4053 tcp_state_change(struct tcpcb *tp, int newstate) 4054 { 4055 #if defined(KDTRACE_HOOKS) 4056 int pstate = tp->t_state; 4057 #endif 4058 4059 TCPSTATES_DEC(tp->t_state); 4060 TCPSTATES_INC(newstate); 4061 tp->t_state = newstate; 4062 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4063 } 4064 4065 /* 4066 * Create an external-format (``xtcpcb'') structure using the information in 4067 * the kernel-format tcpcb structure pointed to by tp. This is done to 4068 * reduce the spew of irrelevant information over this interface, to isolate 4069 * user code from changes in the kernel structure, and potentially to provide 4070 * information-hiding if we decide that some of this information should be 4071 * hidden from users. 4072 */ 4073 void 4074 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4075 { 4076 struct tcpcb *tp = intotcpcb(inp); 4077 struct tcptw *tw = intotw(inp); 4078 sbintime_t now; 4079 4080 bzero(xt, sizeof(*xt)); 4081 if (inp->inp_flags & INP_TIMEWAIT) { 4082 xt->t_state = TCPS_TIME_WAIT; 4083 xt->xt_encaps_port = tw->t_port; 4084 } else { 4085 xt->t_state = tp->t_state; 4086 xt->t_logstate = tp->t_logstate; 4087 xt->t_flags = tp->t_flags; 4088 xt->t_sndzerowin = tp->t_sndzerowin; 4089 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4090 xt->t_rcvoopack = tp->t_rcvoopack; 4091 xt->t_rcv_wnd = tp->rcv_wnd; 4092 xt->t_snd_wnd = tp->snd_wnd; 4093 xt->t_snd_cwnd = tp->snd_cwnd; 4094 xt->t_snd_ssthresh = tp->snd_ssthresh; 4095 xt->t_dsack_bytes = tp->t_dsack_bytes; 4096 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4097 xt->t_dsack_pack = tp->t_dsack_pack; 4098 xt->t_maxseg = tp->t_maxseg; 4099 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4100 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4101 4102 now = getsbinuptime(); 4103 #define COPYTIMER(ttt) do { \ 4104 if (callout_active(&tp->t_timers->ttt)) \ 4105 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 4106 SBT_1MS; \ 4107 else \ 4108 xt->ttt = 0; \ 4109 } while (0) 4110 COPYTIMER(tt_delack); 4111 COPYTIMER(tt_rexmt); 4112 COPYTIMER(tt_persist); 4113 COPYTIMER(tt_keep); 4114 COPYTIMER(tt_2msl); 4115 #undef COPYTIMER 4116 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4117 4118 xt->xt_encaps_port = tp->t_port; 4119 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4120 TCP_FUNCTION_NAME_LEN_MAX); 4121 bcopy(CC_ALGO(tp)->name, xt->xt_cc, 4122 TCP_CA_NAME_MAX); 4123 #ifdef TCP_BLACKBOX 4124 (void)tcp_log_get_id(tp, xt->xt_logid); 4125 #endif 4126 } 4127 4128 xt->xt_len = sizeof(struct xtcpcb); 4129 in_pcbtoxinpcb(inp, &xt->xt_inp); 4130 if (inp->inp_socket == NULL) 4131 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 4132 } 4133 4134 void 4135 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4136 { 4137 uint32_t bit, i; 4138 4139 if ((tp == NULL) || 4140 (status > TCP_EI_STATUS_MAX_VALUE) || 4141 (status == 0)) { 4142 /* Invalid */ 4143 return; 4144 } 4145 if (status > (sizeof(uint32_t) * 8)) { 4146 /* Should this be a KASSERT? */ 4147 return; 4148 } 4149 bit = 1U << (status - 1); 4150 if (bit & tp->t_end_info_status) { 4151 /* already logged */ 4152 return; 4153 } 4154 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4155 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4156 tp->t_end_info_bytes[i] = status; 4157 tp->t_end_info_status |= bit; 4158 break; 4159 } 4160 } 4161 } 4162 4163 int 4164 tcp_can_enable_pacing(void) 4165 { 4166 4167 if ((tcp_pacing_limit == -1) || 4168 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4169 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4170 shadow_num_connections = number_of_tcp_connections_pacing; 4171 return (1); 4172 } else { 4173 return (0); 4174 } 4175 } 4176 4177 static uint8_t tcp_pacing_warning = 0; 4178 4179 void 4180 tcp_decrement_paced_conn(void) 4181 { 4182 uint32_t ret; 4183 4184 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4185 shadow_num_connections = number_of_tcp_connections_pacing; 4186 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4187 if (ret == 0) { 4188 if (tcp_pacing_limit != -1) { 4189 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4190 tcp_pacing_limit = 0; 4191 } else if (tcp_pacing_warning == 0) { 4192 printf("Warning pacing count is invalid, invalid decrement\n"); 4193 tcp_pacing_warning = 1; 4194 } 4195 } 4196 } 4197