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