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 32#include <sys/cdefs.h> 33#include "opt_inet.h" 34#include "opt_inet6.h" 35#include "opt_ipsec.h" 36#include "opt_kern_tls.h" 37 38#include <sys/param.h> 39#include <sys/systm.h> 40#include <sys/arb.h> 41#include <sys/callout.h> 42#include <sys/eventhandler.h> 43#ifdef TCP_HHOOK 44#include <sys/hhook.h> 45#endif 46#include <sys/kernel.h> 47#ifdef TCP_HHOOK 48#include <sys/khelp.h> 49#endif 50#ifdef KERN_TLS 51#include <sys/ktls.h> 52#endif 53#include <sys/qmath.h> 54#include <sys/stats.h> 55#include <sys/sysctl.h> 56#include <sys/jail.h> 57#include <sys/malloc.h> 58#include <sys/refcount.h> 59#include <sys/mbuf.h> 60#include <sys/priv.h> 61#include <sys/proc.h> 62#include <sys/sdt.h> 63#include <sys/socket.h> 64#include <sys/socketvar.h> 65#include <sys/protosw.h> 66#include <sys/random.h> 67 68#include <vm/uma.h> 69 70#include <net/route.h> 71#include <net/route/nhop.h> 72#include <net/if.h> 73#include <net/if_var.h> 74#include <net/if_private.h> 75#include <net/vnet.h> 76 77#include <netinet/in.h> 78#include <netinet/in_fib.h> 79#include <netinet/in_kdtrace.h> 80#include <netinet/in_pcb.h> 81#include <netinet/in_systm.h> 82#include <netinet/in_var.h> 83#include <netinet/ip.h> 84#include <netinet/ip_icmp.h> 85#include <netinet/ip_var.h> 86#ifdef INET6 87#include <netinet/icmp6.h> 88#include <netinet/ip6.h> 89#include <netinet6/in6_fib.h> 90#include <netinet6/in6_pcb.h> 91#include <netinet6/ip6_var.h> 92#include <netinet6/scope6_var.h> 93#include <netinet6/nd6.h> 94#endif 95 96#include <netinet/tcp.h> 97#ifdef INVARIANTS 98#define TCPSTATES 99#endif 100#include <netinet/tcp_fsm.h> 101#include <netinet/tcp_seq.h> 102#include <netinet/tcp_timer.h> 103#include <netinet/tcp_var.h> 104#include <netinet/tcp_ecn.h> 105#include <netinet/tcp_log_buf.h> 106#include <netinet/tcp_syncache.h> 107#include <netinet/tcp_hpts.h> 108#include <netinet/tcp_lro.h> 109#include <netinet/cc/cc.h> 110#include <netinet/tcpip.h> 111#include <netinet/tcp_fastopen.h> 112#include <netinet/tcp_accounting.h> 113#ifdef TCPPCAP 114#include <netinet/tcp_pcap.h> 115#endif 116#ifdef TCP_OFFLOAD 117#include <netinet/tcp_offload.h> 118#endif 119#include <netinet/udp.h> 120#include <netinet/udp_var.h> 121#ifdef INET6 122#include <netinet6/tcp6_var.h> 123#endif 124 125#include <netipsec/ipsec_support.h> 126 127#include <machine/in_cksum.h> 128#include <crypto/siphash/siphash.h> 129 130#include <security/mac/mac_framework.h> 131 132#ifdef INET6 133static ip6proto_ctlinput_t tcp6_ctlinput; 134static udp_tun_icmp_t tcp6_ctlinput_viaudp; 135#endif 136 137VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 138#ifdef INET6 139VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 140#endif 141 142uint32_t tcp_ack_war_time_window = 1000; 143SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 144 CTLFLAG_RW, 145 &tcp_ack_war_time_window, 1000, 146 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 147uint32_t tcp_ack_war_cnt = 5; 148SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 149 CTLFLAG_RW, 150 &tcp_ack_war_cnt, 5, 151 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 152 153struct rwlock tcp_function_lock; 154 155static int 156sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 157{ 158 int error, new; 159 160 new = V_tcp_mssdflt; 161 error = sysctl_handle_int(oidp, &new, 0, req); 162 if (error == 0 && req->newptr) { 163 if (new < TCP_MINMSS) 164 error = EINVAL; 165 else 166 V_tcp_mssdflt = new; 167 } 168 return (error); 169} 170 171SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 172 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 173 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 174 "Default TCP Maximum Segment Size"); 175 176#ifdef INET6 177static int 178sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 179{ 180 int error, new; 181 182 new = V_tcp_v6mssdflt; 183 error = sysctl_handle_int(oidp, &new, 0, req); 184 if (error == 0 && req->newptr) { 185 if (new < TCP_MINMSS) 186 error = EINVAL; 187 else 188 V_tcp_v6mssdflt = new; 189 } 190 return (error); 191} 192 193SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 194 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 195 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 196 "Default TCP Maximum Segment Size for IPv6"); 197#endif /* INET6 */ 198 199/* 200 * Minimum MSS we accept and use. This prevents DoS attacks where 201 * we are forced to a ridiculous low MSS like 20 and send hundreds 202 * of packets instead of one. The effect scales with the available 203 * bandwidth and quickly saturates the CPU and network interface 204 * with packet generation and sending. Set to zero to disable MINMSS 205 * checking. This setting prevents us from sending too small packets. 206 */ 207VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 208SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_minmss), 0, 210 "Minimum TCP Maximum Segment Size"); 211 212VNET_DEFINE(int, tcp_do_rfc1323) = 1; 213SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_do_rfc1323), 0, 215 "Enable rfc1323 (high performance TCP) extensions"); 216 217/* 218 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 219 * Some stacks negotiate TS, but never send them after connection setup. Some 220 * stacks negotiate TS, but don't send them when sending keep-alive segments. 221 * These include modern widely deployed TCP stacks. 222 * Therefore tolerating violations for now... 223 */ 224VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 225SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 226 &VNET_NAME(tcp_tolerate_missing_ts), 0, 227 "Tolerate missing TCP timestamps"); 228 229VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 230SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 231 &VNET_NAME(tcp_ts_offset_per_conn), 0, 232 "Initialize TCP timestamps per connection instead of per host pair"); 233 234/* How many connections are pacing */ 235static volatile uint32_t number_of_tcp_connections_pacing = 0; 236static uint32_t shadow_num_connections = 0; 237static counter_u64_t tcp_pacing_failures; 238static counter_u64_t tcp_dgp_failures; 239static uint32_t shadow_tcp_pacing_dgp = 0; 240static volatile uint32_t number_of_dgp_connections = 0; 241 242static int tcp_pacing_limit = 10000; 243SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 244 &tcp_pacing_limit, 1000, 245 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 246 247static int tcp_dgp_limit = -1; 248SYSCTL_INT(_net_inet_tcp, OID_AUTO, dgp_limit, CTLFLAG_RW, 249 &tcp_dgp_limit, -1, 250 "If the TCP stack does DGP, is there a limit (-1 = no, 0 = no dgp N = number of connections)"); 251 252SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 253 &shadow_num_connections, 0, "Number of TCP connections being paced"); 254 255SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD, 256 &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit"); 257 258SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, dgp_failures, CTLFLAG_RD, 259 &tcp_dgp_failures, "Number of times we failed to enable dgp to avoid exceeding the limit"); 260 261static int tcp_log_debug = 0; 262SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 263 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 264 265/* 266 * Target size of TCP PCB hash tables. Must be a power of two. 267 * 268 * Note that this can be overridden by the kernel environment 269 * variable net.inet.tcp.tcbhashsize 270 */ 271#ifndef TCBHASHSIZE 272#define TCBHASHSIZE 0 273#endif 274static int tcp_tcbhashsize = TCBHASHSIZE; 275SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 276 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 277 278static int do_tcpdrain = 1; 279SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 280 "Enable tcp_drain routine for extra help when low on mbufs"); 281 282SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 283 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 284 285VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 286#define V_icmp_may_rst VNET(icmp_may_rst) 287SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 288 &VNET_NAME(icmp_may_rst), 0, 289 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 290 291VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 292#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 293SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 294 &VNET_NAME(tcp_isn_reseed_interval), 0, 295 "Seconds between reseeding of ISN secret"); 296 297static int tcp_soreceive_stream; 298SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 299 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 300 301VNET_DEFINE(uma_zone_t, sack_hole_zone); 302#define V_sack_hole_zone VNET(sack_hole_zone) 303VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 304static int 305sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 306{ 307 int error; 308 uint32_t new; 309 310 new = V_tcp_map_entries_limit; 311 error = sysctl_handle_int(oidp, &new, 0, req); 312 if (error == 0 && req->newptr) { 313 /* only allow "0" and value > minimum */ 314 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 315 error = EINVAL; 316 else 317 V_tcp_map_entries_limit = new; 318 } 319 return (error); 320} 321SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 322 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 323 &VNET_NAME(tcp_map_entries_limit), 0, 324 &sysctl_net_inet_tcp_map_limit_check, "IU", 325 "Total sendmap entries limit"); 326 327VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 328SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 329 &VNET_NAME(tcp_map_split_limit), 0, 330 "Total sendmap split entries limit"); 331 332#ifdef TCP_HHOOK 333VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 334#endif 335 336#define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 337VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 338#define V_ts_offset_secret VNET(ts_offset_secret) 339 340static int tcp_default_fb_init(struct tcpcb *tp, void **ptr); 341static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 342static int tcp_default_handoff_ok(struct tcpcb *tp); 343static struct inpcb *tcp_notify(struct inpcb *, int); 344static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 345static struct inpcb *tcp_mtudisc(struct inpcb *, int); 346static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 347static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 348 const void *ip4hdr, const void *ip6hdr); 349static void tcp_default_switch_failed(struct tcpcb *tp); 350static ipproto_ctlinput_t tcp_ctlinput; 351static udp_tun_icmp_t tcp_ctlinput_viaudp; 352 353static struct tcp_function_block tcp_def_funcblk = { 354 .tfb_tcp_block_name = "freebsd", 355 .tfb_tcp_output = tcp_default_output, 356 .tfb_tcp_do_segment = tcp_do_segment, 357 .tfb_tcp_ctloutput = tcp_default_ctloutput, 358 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 359 .tfb_tcp_fb_init = tcp_default_fb_init, 360 .tfb_tcp_fb_fini = tcp_default_fb_fini, 361 .tfb_switch_failed = tcp_default_switch_failed, 362}; 363 364static int tcp_fb_cnt = 0; 365struct tcp_funchead t_functions; 366VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk; 367#define V_tcp_func_set_ptr VNET(tcp_func_set_ptr) 368 369void 370tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 371{ 372 TCPSTAT_INC(tcps_dsack_count); 373 tp->t_dsack_pack++; 374 if (tlp == 0) { 375 if (SEQ_GT(end, start)) { 376 tp->t_dsack_bytes += (end - start); 377 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 378 } else { 379 tp->t_dsack_tlp_bytes += (start - end); 380 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 381 } 382 } else { 383 if (SEQ_GT(end, start)) { 384 tp->t_dsack_bytes += (end - start); 385 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 386 } else { 387 tp->t_dsack_tlp_bytes += (start - end); 388 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 389 } 390 } 391} 392 393static struct tcp_function_block * 394find_tcp_functions_locked(struct tcp_function_set *fs) 395{ 396 struct tcp_function *f; 397 struct tcp_function_block *blk=NULL; 398 399 TAILQ_FOREACH(f, &t_functions, tf_next) { 400 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 401 blk = f->tf_fb; 402 break; 403 } 404 } 405 return(blk); 406} 407 408static struct tcp_function_block * 409find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 410{ 411 struct tcp_function_block *rblk=NULL; 412 struct tcp_function *f; 413 414 TAILQ_FOREACH(f, &t_functions, tf_next) { 415 if (f->tf_fb == blk) { 416 rblk = blk; 417 if (s) { 418 *s = f; 419 } 420 break; 421 } 422 } 423 return (rblk); 424} 425 426struct tcp_function_block * 427find_and_ref_tcp_functions(struct tcp_function_set *fs) 428{ 429 struct tcp_function_block *blk; 430 431 rw_rlock(&tcp_function_lock); 432 blk = find_tcp_functions_locked(fs); 433 if (blk) 434 refcount_acquire(&blk->tfb_refcnt); 435 rw_runlock(&tcp_function_lock); 436 return(blk); 437} 438 439struct tcp_function_block * 440find_and_ref_tcp_fb(struct tcp_function_block *blk) 441{ 442 struct tcp_function_block *rblk; 443 444 rw_rlock(&tcp_function_lock); 445 rblk = find_tcp_fb_locked(blk, NULL); 446 if (rblk) 447 refcount_acquire(&rblk->tfb_refcnt); 448 rw_runlock(&tcp_function_lock); 449 return(rblk); 450} 451 452/* Find a matching alias for the given tcp_function_block. */ 453int 454find_tcp_function_alias(struct tcp_function_block *blk, 455 struct tcp_function_set *fs) 456{ 457 struct tcp_function *f; 458 int found; 459 460 found = 0; 461 rw_rlock(&tcp_function_lock); 462 TAILQ_FOREACH(f, &t_functions, tf_next) { 463 if ((f->tf_fb == blk) && 464 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 465 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 466 /* Matching function block with different name. */ 467 strncpy(fs->function_set_name, f->tf_name, 468 TCP_FUNCTION_NAME_LEN_MAX); 469 found = 1; 470 break; 471 } 472 } 473 /* Null terminate the string appropriately. */ 474 if (found) { 475 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 476 } else { 477 fs->function_set_name[0] = '\0'; 478 } 479 rw_runlock(&tcp_function_lock); 480 return (found); 481} 482 483static struct tcp_function_block * 484find_and_ref_tcp_default_fb(void) 485{ 486 struct tcp_function_block *rblk; 487 488 rw_rlock(&tcp_function_lock); 489 rblk = V_tcp_func_set_ptr; 490 refcount_acquire(&rblk->tfb_refcnt); 491 rw_runlock(&tcp_function_lock); 492 return (rblk); 493} 494 495void 496tcp_switch_back_to_default(struct tcpcb *tp) 497{ 498 struct tcp_function_block *tfb; 499 void *ptr = NULL; 500 501 KASSERT(tp->t_fb != &tcp_def_funcblk, 502 ("%s: called by the built-in default stack", __func__)); 503 504 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 505 tp->t_fb->tfb_tcp_timer_stop_all(tp); 506 507 /* 508 * Now, we'll find a new function block to use. 509 * Start by trying the current user-selected 510 * default, unless this stack is the user-selected 511 * default. 512 */ 513 tfb = find_and_ref_tcp_default_fb(); 514 if (tfb == tp->t_fb) { 515 refcount_release(&tfb->tfb_refcnt); 516 tfb = NULL; 517 } 518 /* Does the stack accept this connection? */ 519 if (tfb != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { 520 refcount_release(&tfb->tfb_refcnt); 521 tfb = NULL; 522 } 523 /* Try to use that stack. */ 524 if (tfb != NULL) { 525 /* Initialize the new stack. If it succeeds, we are done. */ 526 if (tfb->tfb_tcp_fb_init == NULL || 527 (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) { 528 /* Release the old stack */ 529 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 530 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 531 refcount_release(&tp->t_fb->tfb_refcnt); 532 /* Now set in all the pointers */ 533 tp->t_fb = tfb; 534 tp->t_fb_ptr = ptr; 535 return; 536 } 537 /* 538 * Initialization failed. Release the reference count on 539 * the looked up default stack. 540 */ 541 refcount_release(&tfb->tfb_refcnt); 542 } 543 544 /* 545 * If that wasn't feasible, use the built-in default 546 * stack which is not allowed to reject anyone. 547 */ 548 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 549 if (tfb == NULL) { 550 /* there always should be a default */ 551 panic("Can't refer to tcp_def_funcblk"); 552 } 553 if ((*tfb->tfb_tcp_handoff_ok)(tp)) { 554 /* The default stack cannot say no */ 555 panic("Default stack rejects a new session?"); 556 } 557 if (tfb->tfb_tcp_fb_init != NULL && 558 (*tfb->tfb_tcp_fb_init)(tp, &ptr)) { 559 /* The default stack cannot fail */ 560 panic("Default stack initialization failed"); 561 } 562 /* Now release the old stack */ 563 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 564 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 565 refcount_release(&tp->t_fb->tfb_refcnt); 566 /* And set in the pointers to the new */ 567 tp->t_fb = tfb; 568 tp->t_fb_ptr = ptr; 569} 570 571static bool 572tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 573 const struct sockaddr *sa, void *ctx) 574{ 575 struct ip *iph; 576#ifdef INET6 577 struct ip6_hdr *ip6; 578#endif 579 struct udphdr *uh; 580 struct tcphdr *th; 581 int thlen; 582 uint16_t port; 583 584 TCPSTAT_INC(tcps_tunneled_pkts); 585 if ((m->m_flags & M_PKTHDR) == 0) { 586 /* Can't handle one that is not a pkt hdr */ 587 TCPSTAT_INC(tcps_tunneled_errs); 588 goto out; 589 } 590 thlen = sizeof(struct tcphdr); 591 if (m->m_len < off + sizeof(struct udphdr) + thlen && 592 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 593 TCPSTAT_INC(tcps_tunneled_errs); 594 goto out; 595 } 596 iph = mtod(m, struct ip *); 597 uh = (struct udphdr *)((caddr_t)iph + off); 598 th = (struct tcphdr *)(uh + 1); 599 thlen = th->th_off << 2; 600 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 601 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 602 if (m == NULL) { 603 TCPSTAT_INC(tcps_tunneled_errs); 604 goto out; 605 } else { 606 iph = mtod(m, struct ip *); 607 uh = (struct udphdr *)((caddr_t)iph + off); 608 th = (struct tcphdr *)(uh + 1); 609 } 610 } 611 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 612 bcopy(th, uh, m->m_len - off); 613 m->m_len -= sizeof(struct udphdr); 614 m->m_pkthdr.len -= sizeof(struct udphdr); 615 /* 616 * We use the same algorithm for 617 * both UDP and TCP for c-sum. So 618 * the code in tcp_input will skip 619 * the checksum. So we do nothing 620 * with the flag (m->m_pkthdr.csum_flags). 621 */ 622 switch (iph->ip_v) { 623#ifdef INET 624 case IPVERSION: 625 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 626 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 627 break; 628#endif 629#ifdef INET6 630 case IPV6_VERSION >> 4: 631 ip6 = mtod(m, struct ip6_hdr *); 632 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 633 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 634 break; 635#endif 636 default: 637 goto out; 638 break; 639 } 640 return (true); 641out: 642 m_freem(m); 643 644 return (true); 645} 646 647static int 648sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 649{ 650 int error=ENOENT; 651 struct tcp_function_set fs; 652 struct tcp_function_block *blk; 653 654 memset(&fs, 0, sizeof(fs)); 655 rw_rlock(&tcp_function_lock); 656 blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL); 657 if (blk) { 658 /* Found him */ 659 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 660 fs.pcbcnt = blk->tfb_refcnt; 661 } 662 rw_runlock(&tcp_function_lock); 663 error = sysctl_handle_string(oidp, fs.function_set_name, 664 sizeof(fs.function_set_name), req); 665 666 /* Check for error or no change */ 667 if (error != 0 || req->newptr == NULL) 668 return(error); 669 670 rw_wlock(&tcp_function_lock); 671 blk = find_tcp_functions_locked(&fs); 672 if ((blk == NULL) || 673 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 674 error = ENOENT; 675 goto done; 676 } 677 V_tcp_func_set_ptr = blk; 678done: 679 rw_wunlock(&tcp_function_lock); 680 return (error); 681} 682 683SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 684 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 685 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 686 "Set/get the default TCP functions"); 687 688static int 689sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 690{ 691 int error, cnt, linesz; 692 struct tcp_function *f; 693 char *buffer, *cp; 694 size_t bufsz, outsz; 695 bool alias; 696 697 cnt = 0; 698 rw_rlock(&tcp_function_lock); 699 TAILQ_FOREACH(f, &t_functions, tf_next) { 700 cnt++; 701 } 702 rw_runlock(&tcp_function_lock); 703 704 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 705 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 706 707 error = 0; 708 cp = buffer; 709 710 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 711 "Alias", "PCB count"); 712 cp += linesz; 713 bufsz -= linesz; 714 outsz = linesz; 715 716 rw_rlock(&tcp_function_lock); 717 TAILQ_FOREACH(f, &t_functions, tf_next) { 718 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 719 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 720 f->tf_fb->tfb_tcp_block_name, 721 (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ', 722 alias ? f->tf_name : "-", 723 f->tf_fb->tfb_refcnt); 724 if (linesz >= bufsz) { 725 error = EOVERFLOW; 726 break; 727 } 728 cp += linesz; 729 bufsz -= linesz; 730 outsz += linesz; 731 } 732 rw_runlock(&tcp_function_lock); 733 if (error == 0) 734 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 735 free(buffer, M_TEMP); 736 return (error); 737} 738 739SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 740 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 741 NULL, 0, sysctl_net_inet_list_available, "A", 742 "list available TCP Function sets"); 743 744VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 745 746#ifdef INET 747VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 748#define V_udp4_tun_socket VNET(udp4_tun_socket) 749#endif 750#ifdef INET6 751VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 752#define V_udp6_tun_socket VNET(udp6_tun_socket) 753#endif 754 755static struct sx tcpoudp_lock; 756 757static void 758tcp_over_udp_stop(void) 759{ 760 761 sx_assert(&tcpoudp_lock, SA_XLOCKED); 762 763#ifdef INET 764 if (V_udp4_tun_socket != NULL) { 765 soclose(V_udp4_tun_socket); 766 V_udp4_tun_socket = NULL; 767 } 768#endif 769#ifdef INET6 770 if (V_udp6_tun_socket != NULL) { 771 soclose(V_udp6_tun_socket); 772 V_udp6_tun_socket = NULL; 773 } 774#endif 775} 776 777static int 778tcp_over_udp_start(void) 779{ 780 uint16_t port; 781 int ret; 782#ifdef INET 783 struct sockaddr_in sin; 784#endif 785#ifdef INET6 786 struct sockaddr_in6 sin6; 787#endif 788 789 sx_assert(&tcpoudp_lock, SA_XLOCKED); 790 791 port = V_tcp_udp_tunneling_port; 792 if (ntohs(port) == 0) { 793 /* Must have a port set */ 794 return (EINVAL); 795 } 796#ifdef INET 797 if (V_udp4_tun_socket != NULL) { 798 /* Already running -- must stop first */ 799 return (EALREADY); 800 } 801#endif 802#ifdef INET6 803 if (V_udp6_tun_socket != NULL) { 804 /* Already running -- must stop first */ 805 return (EALREADY); 806 } 807#endif 808#ifdef INET 809 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 810 SOCK_DGRAM, IPPROTO_UDP, 811 curthread->td_ucred, curthread))) { 812 tcp_over_udp_stop(); 813 return (ret); 814 } 815 /* Call the special UDP hook. */ 816 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 817 tcp_recv_udp_tunneled_packet, 818 tcp_ctlinput_viaudp, 819 NULL))) { 820 tcp_over_udp_stop(); 821 return (ret); 822 } 823 /* Ok, we have a socket, bind it to the port. */ 824 memset(&sin, 0, sizeof(struct sockaddr_in)); 825 sin.sin_len = sizeof(struct sockaddr_in); 826 sin.sin_family = AF_INET; 827 sin.sin_port = htons(port); 828 if ((ret = sobind(V_udp4_tun_socket, 829 (struct sockaddr *)&sin, curthread))) { 830 tcp_over_udp_stop(); 831 return (ret); 832 } 833#endif 834#ifdef INET6 835 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 836 SOCK_DGRAM, IPPROTO_UDP, 837 curthread->td_ucred, curthread))) { 838 tcp_over_udp_stop(); 839 return (ret); 840 } 841 /* Call the special UDP hook. */ 842 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 843 tcp_recv_udp_tunneled_packet, 844 tcp6_ctlinput_viaudp, 845 NULL))) { 846 tcp_over_udp_stop(); 847 return (ret); 848 } 849 /* Ok, we have a socket, bind it to the port. */ 850 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 851 sin6.sin6_len = sizeof(struct sockaddr_in6); 852 sin6.sin6_family = AF_INET6; 853 sin6.sin6_port = htons(port); 854 if ((ret = sobind(V_udp6_tun_socket, 855 (struct sockaddr *)&sin6, curthread))) { 856 tcp_over_udp_stop(); 857 return (ret); 858 } 859#endif 860 return (0); 861} 862 863static int 864sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 865{ 866 int error; 867 uint32_t old, new; 868 869 old = V_tcp_udp_tunneling_port; 870 new = old; 871 error = sysctl_handle_int(oidp, &new, 0, req); 872 if ((error == 0) && 873 (req->newptr != NULL)) { 874 if ((new < TCP_TUNNELING_PORT_MIN) || 875 (new > TCP_TUNNELING_PORT_MAX)) { 876 error = EINVAL; 877 } else { 878 sx_xlock(&tcpoudp_lock); 879 V_tcp_udp_tunneling_port = new; 880 if (old != 0) { 881 tcp_over_udp_stop(); 882 } 883 if (new != 0) { 884 error = tcp_over_udp_start(); 885 if (error != 0) { 886 V_tcp_udp_tunneling_port = 0; 887 } 888 } 889 sx_xunlock(&tcpoudp_lock); 890 } 891 } 892 return (error); 893} 894 895SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 896 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 897 &VNET_NAME(tcp_udp_tunneling_port), 898 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 899 "Tunneling port for tcp over udp"); 900 901VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 902 903static int 904sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 905{ 906 int error, new; 907 908 new = V_tcp_udp_tunneling_overhead; 909 error = sysctl_handle_int(oidp, &new, 0, req); 910 if (error == 0 && req->newptr) { 911 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 912 (new > TCP_TUNNELING_OVERHEAD_MAX)) 913 error = EINVAL; 914 else 915 V_tcp_udp_tunneling_overhead = new; 916 } 917 return (error); 918} 919 920SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 921 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 922 &VNET_NAME(tcp_udp_tunneling_overhead), 923 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 924 "MSS reduction when using tcp over udp"); 925 926/* 927 * Exports one (struct tcp_function_info) for each alias/name. 928 */ 929static int 930sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 931{ 932 int cnt, error; 933 struct tcp_function *f; 934 struct tcp_function_info tfi; 935 936 /* 937 * We don't allow writes. 938 */ 939 if (req->newptr != NULL) 940 return (EINVAL); 941 942 /* 943 * Wire the old buffer so we can directly copy the functions to 944 * user space without dropping the lock. 945 */ 946 if (req->oldptr != NULL) { 947 error = sysctl_wire_old_buffer(req, 0); 948 if (error) 949 return (error); 950 } 951 952 /* 953 * Walk the list and copy out matching entries. If INVARIANTS 954 * is compiled in, also walk the list to verify the length of 955 * the list matches what we have recorded. 956 */ 957 rw_rlock(&tcp_function_lock); 958 959 cnt = 0; 960#ifndef INVARIANTS 961 if (req->oldptr == NULL) { 962 cnt = tcp_fb_cnt; 963 goto skip_loop; 964 } 965#endif 966 TAILQ_FOREACH(f, &t_functions, tf_next) { 967#ifdef INVARIANTS 968 cnt++; 969#endif 970 if (req->oldptr != NULL) { 971 bzero(&tfi, sizeof(tfi)); 972 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 973 tfi.tfi_id = f->tf_fb->tfb_id; 974 (void)strlcpy(tfi.tfi_alias, f->tf_name, 975 sizeof(tfi.tfi_alias)); 976 (void)strlcpy(tfi.tfi_name, 977 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 978 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 979 /* 980 * Don't stop on error, as that is the 981 * mechanism we use to accumulate length 982 * information if the buffer was too short. 983 */ 984 } 985 } 986 KASSERT(cnt == tcp_fb_cnt, 987 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 988#ifndef INVARIANTS 989skip_loop: 990#endif 991 rw_runlock(&tcp_function_lock); 992 if (req->oldptr == NULL) 993 error = SYSCTL_OUT(req, NULL, 994 (cnt + 1) * sizeof(struct tcp_function_info)); 995 996 return (error); 997} 998 999SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1000 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1001 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1002 "List TCP function block name-to-ID mappings"); 1003 1004/* 1005 * tfb_tcp_handoff_ok() function for the default stack. 1006 * Note that we'll basically try to take all comers. 1007 */ 1008static int 1009tcp_default_handoff_ok(struct tcpcb *tp) 1010{ 1011 1012 return (0); 1013} 1014 1015/* 1016 * tfb_tcp_fb_init() function for the default stack. 1017 * 1018 * This handles making sure we have appropriate timers set if you are 1019 * transitioning a socket that has some amount of setup done. 1020 * 1021 * The init() fuction from the default can *never* return non-zero i.e. 1022 * it is required to always succeed since it is the stack of last resort! 1023 */ 1024static int 1025tcp_default_fb_init(struct tcpcb *tp, void **ptr) 1026{ 1027 struct socket *so = tptosocket(tp); 1028 int rexmt; 1029 1030 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1031 /* We don't use the pointer */ 1032 *ptr = NULL; 1033 1034 KASSERT(tp->t_state < TCPS_TIME_WAIT, 1035 ("%s: connection %p in unexpected state %d", __func__, tp, 1036 tp->t_state)); 1037 1038 /* Make sure we get no interesting mbuf queuing behavior */ 1039 /* All mbuf queue/ack compress flags should be off */ 1040 tcp_lro_features_off(tp); 1041 1042 /* Cancel the GP measurement in progress */ 1043 tp->t_flags &= ~TF_GPUTINPROG; 1044 /* Validate the timers are not in usec, if they are convert */ 1045 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 1046 if ((tp->t_state == TCPS_SYN_SENT) || 1047 (tp->t_state == TCPS_SYN_RECEIVED)) 1048 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift]; 1049 else 1050 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 1051 if (tp->t_rxtshift == 0) 1052 tp->t_rxtcur = rexmt; 1053 else 1054 TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX); 1055 1056 /* 1057 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1058 * know what to do for unexpected states (which includes TIME_WAIT). 1059 */ 1060 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1061 return (0); 1062 1063 /* 1064 * Make sure some kind of transmission timer is set if there is 1065 * outstanding data. 1066 */ 1067 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1068 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1069 tcp_timer_active(tp, TT_PERSIST))) { 1070 /* 1071 * If the session has established and it looks like it should 1072 * be in the persist state, set the persist timer. Otherwise, 1073 * set the retransmit timer. 1074 */ 1075 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1076 (int32_t)(tp->snd_nxt - tp->snd_una) < 1077 (int32_t)sbavail(&so->so_snd)) 1078 tcp_setpersist(tp); 1079 else 1080 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 1081 } 1082 1083 /* All non-embryonic sessions get a keepalive timer. */ 1084 if (!tcp_timer_active(tp, TT_KEEP)) 1085 tcp_timer_activate(tp, TT_KEEP, 1086 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1087 TP_KEEPINIT(tp)); 1088 1089 /* 1090 * Make sure critical variables are initialized 1091 * if transitioning while in Recovery. 1092 */ 1093 if IN_FASTRECOVERY(tp->t_flags) { 1094 if (tp->sackhint.recover_fs == 0) 1095 tp->sackhint.recover_fs = max(1, 1096 tp->snd_nxt - tp->snd_una); 1097 } 1098 1099 return (0); 1100} 1101 1102/* 1103 * tfb_tcp_fb_fini() function for the default stack. 1104 * 1105 * This changes state as necessary (or prudent) to prepare for another stack 1106 * to assume responsibility for the connection. 1107 */ 1108static void 1109tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1110{ 1111 1112 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1113 1114#ifdef TCP_BLACKBOX 1115 tcp_log_flowend(tp); 1116#endif 1117 tp->t_acktime = 0; 1118 return; 1119} 1120 1121MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1122MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1123 1124static struct mtx isn_mtx; 1125 1126#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1127#define ISN_LOCK() mtx_lock(&isn_mtx) 1128#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1129 1130INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1131 1132/* 1133 * Take a value and get the next power of 2 that doesn't overflow. 1134 * Used to size the tcp_inpcb hash buckets. 1135 */ 1136static int 1137maketcp_hashsize(int size) 1138{ 1139 int hashsize; 1140 1141 /* 1142 * auto tune. 1143 * get the next power of 2 higher than maxsockets. 1144 */ 1145 hashsize = 1 << fls(size); 1146 /* catch overflow, and just go one power of 2 smaller */ 1147 if (hashsize < size) { 1148 hashsize = 1 << (fls(size) - 1); 1149 } 1150 return (hashsize); 1151} 1152 1153static volatile int next_tcp_stack_id = 1; 1154 1155/* 1156 * Register a TCP function block with the name provided in the names 1157 * array. (Note that this function does NOT automatically register 1158 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1159 * explicitly include blk->tfb_tcp_block_name in the list of names if 1160 * you wish to register the stack with that name.) 1161 * 1162 * Either all name registrations will succeed or all will fail. If 1163 * a name registration fails, the function will update the num_names 1164 * argument to point to the array index of the name that encountered 1165 * the failure. 1166 * 1167 * Returns 0 on success, or an error code on failure. 1168 */ 1169int 1170register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1171 const char *names[], int *num_names) 1172{ 1173 struct tcp_function *n; 1174 struct tcp_function_set fs; 1175 int error, i; 1176 1177 KASSERT(names != NULL && *num_names > 0, 1178 ("%s: Called with 0-length name list", __func__)); 1179 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1180 KASSERT(rw_initialized(&tcp_function_lock), 1181 ("%s: called too early", __func__)); 1182 1183 if ((blk->tfb_tcp_output == NULL) || 1184 (blk->tfb_tcp_do_segment == NULL) || 1185 (blk->tfb_tcp_ctloutput == NULL) || 1186 (blk->tfb_tcp_handoff_ok == NULL) || 1187 (strlen(blk->tfb_tcp_block_name) == 0)) { 1188 /* 1189 * These functions are required and you 1190 * need a name. 1191 */ 1192 *num_names = 0; 1193 return (EINVAL); 1194 } 1195 1196 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1197 *num_names = 0; 1198 return (EINVAL); 1199 } 1200 1201 refcount_init(&blk->tfb_refcnt, 0); 1202 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1203 for (i = 0; i < *num_names; i++) { 1204 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1205 if (n == NULL) { 1206 error = ENOMEM; 1207 goto cleanup; 1208 } 1209 n->tf_fb = blk; 1210 1211 (void)strlcpy(fs.function_set_name, names[i], 1212 sizeof(fs.function_set_name)); 1213 rw_wlock(&tcp_function_lock); 1214 if (find_tcp_functions_locked(&fs) != NULL) { 1215 /* Duplicate name space not allowed */ 1216 rw_wunlock(&tcp_function_lock); 1217 free(n, M_TCPFUNCTIONS); 1218 error = EALREADY; 1219 goto cleanup; 1220 } 1221 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1222 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1223 tcp_fb_cnt++; 1224 rw_wunlock(&tcp_function_lock); 1225 } 1226 return(0); 1227 1228cleanup: 1229 /* 1230 * Deregister the names we just added. Because registration failed 1231 * for names[i], we don't need to deregister that name. 1232 */ 1233 *num_names = i; 1234 rw_wlock(&tcp_function_lock); 1235 while (--i >= 0) { 1236 TAILQ_FOREACH(n, &t_functions, tf_next) { 1237 if (!strncmp(n->tf_name, names[i], 1238 TCP_FUNCTION_NAME_LEN_MAX)) { 1239 TAILQ_REMOVE(&t_functions, n, tf_next); 1240 tcp_fb_cnt--; 1241 n->tf_fb = NULL; 1242 free(n, M_TCPFUNCTIONS); 1243 break; 1244 } 1245 } 1246 } 1247 rw_wunlock(&tcp_function_lock); 1248 return (error); 1249} 1250 1251/* 1252 * Register a TCP function block using the name provided in the name 1253 * argument. 1254 * 1255 * Returns 0 on success, or an error code on failure. 1256 */ 1257int 1258register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1259 int wait) 1260{ 1261 const char *name_list[1]; 1262 int num_names, rv; 1263 1264 num_names = 1; 1265 if (name != NULL) 1266 name_list[0] = name; 1267 else 1268 name_list[0] = blk->tfb_tcp_block_name; 1269 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1270 return (rv); 1271} 1272 1273/* 1274 * Register a TCP function block using the name defined in 1275 * blk->tfb_tcp_block_name. 1276 * 1277 * Returns 0 on success, or an error code on failure. 1278 */ 1279int 1280register_tcp_functions(struct tcp_function_block *blk, int wait) 1281{ 1282 1283 return (register_tcp_functions_as_name(blk, NULL, wait)); 1284} 1285 1286/* 1287 * Deregister all names associated with a function block. This 1288 * functionally removes the function block from use within the system. 1289 * 1290 * When called with a true quiesce argument, mark the function block 1291 * as being removed so no more stacks will use it and determine 1292 * whether the removal would succeed. 1293 * 1294 * When called with a false quiesce argument, actually attempt the 1295 * removal. 1296 * 1297 * When called with a force argument, attempt to switch all TCBs to 1298 * use the default stack instead of returning EBUSY. 1299 * 1300 * Returns 0 on success (or if the removal would succeed), or an error 1301 * code on failure. 1302 */ 1303int 1304deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1305 bool force) 1306{ 1307 struct tcp_function *f; 1308 VNET_ITERATOR_DECL(vnet_iter); 1309 1310 if (blk == &tcp_def_funcblk) { 1311 /* You can't un-register the default */ 1312 return (EPERM); 1313 } 1314 rw_wlock(&tcp_function_lock); 1315 VNET_LIST_RLOCK_NOSLEEP(); 1316 VNET_FOREACH(vnet_iter) { 1317 CURVNET_SET(vnet_iter); 1318 if (blk == V_tcp_func_set_ptr) { 1319 /* You can't free the current default in some vnet. */ 1320 CURVNET_RESTORE(); 1321 VNET_LIST_RUNLOCK_NOSLEEP(); 1322 rw_wunlock(&tcp_function_lock); 1323 return (EBUSY); 1324 } 1325 CURVNET_RESTORE(); 1326 } 1327 VNET_LIST_RUNLOCK_NOSLEEP(); 1328 /* Mark the block so no more stacks can use it. */ 1329 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1330 /* 1331 * If TCBs are still attached to the stack, attempt to switch them 1332 * to the default stack. 1333 */ 1334 if (force && blk->tfb_refcnt) { 1335 struct inpcb *inp; 1336 struct tcpcb *tp; 1337 VNET_ITERATOR_DECL(vnet_iter); 1338 1339 rw_wunlock(&tcp_function_lock); 1340 1341 VNET_LIST_RLOCK(); 1342 VNET_FOREACH(vnet_iter) { 1343 CURVNET_SET(vnet_iter); 1344 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1345 INPLOOKUP_WLOCKPCB); 1346 1347 while ((inp = inp_next(&inpi)) != NULL) { 1348 tp = intotcpcb(inp); 1349 if (tp == NULL || tp->t_fb != blk) 1350 continue; 1351 tcp_switch_back_to_default(tp); 1352 } 1353 CURVNET_RESTORE(); 1354 } 1355 VNET_LIST_RUNLOCK(); 1356 1357 rw_wlock(&tcp_function_lock); 1358 } 1359 if (blk->tfb_refcnt) { 1360 /* TCBs still attached. */ 1361 rw_wunlock(&tcp_function_lock); 1362 return (EBUSY); 1363 } 1364 if (quiesce) { 1365 /* Skip removal. */ 1366 rw_wunlock(&tcp_function_lock); 1367 return (0); 1368 } 1369 /* Remove any function names that map to this function block. */ 1370 while (find_tcp_fb_locked(blk, &f) != NULL) { 1371 TAILQ_REMOVE(&t_functions, f, tf_next); 1372 tcp_fb_cnt--; 1373 f->tf_fb = NULL; 1374 free(f, M_TCPFUNCTIONS); 1375 } 1376 rw_wunlock(&tcp_function_lock); 1377 return (0); 1378} 1379 1380static void 1381tcp_drain(void) 1382{ 1383 struct epoch_tracker et; 1384 VNET_ITERATOR_DECL(vnet_iter); 1385 1386 if (!do_tcpdrain) 1387 return; 1388 1389 NET_EPOCH_ENTER(et); 1390 VNET_LIST_RLOCK_NOSLEEP(); 1391 VNET_FOREACH(vnet_iter) { 1392 CURVNET_SET(vnet_iter); 1393 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1394 INPLOOKUP_WLOCKPCB); 1395 struct inpcb *inpb; 1396 struct tcpcb *tcpb; 1397 1398 /* 1399 * Walk the tcpbs, if existing, and flush the reassembly queue, 1400 * if there is one... 1401 * XXX: The "Net/3" implementation doesn't imply that the TCP 1402 * reassembly queue should be flushed, but in a situation 1403 * where we're really low on mbufs, this is potentially 1404 * useful. 1405 */ 1406 while ((inpb = inp_next(&inpi)) != NULL) { 1407 if ((tcpb = intotcpcb(inpb)) != NULL) { 1408 tcp_reass_flush(tcpb); 1409 tcp_clean_sackreport(tcpb); 1410#ifdef TCP_BLACKBOX 1411 tcp_log_drain(tcpb); 1412#endif 1413#ifdef TCPPCAP 1414 if (tcp_pcap_aggressive_free) { 1415 /* Free the TCP PCAP queues. */ 1416 tcp_pcap_drain(&(tcpb->t_inpkts)); 1417 tcp_pcap_drain(&(tcpb->t_outpkts)); 1418 } 1419#endif 1420 } 1421 } 1422 CURVNET_RESTORE(); 1423 } 1424 VNET_LIST_RUNLOCK_NOSLEEP(); 1425 NET_EPOCH_EXIT(et); 1426} 1427 1428static void 1429tcp_vnet_init(void *arg __unused) 1430{ 1431 1432#ifdef TCP_HHOOK 1433 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1434 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1435 printf("%s: WARNING: unable to register helper hook\n", __func__); 1436 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1437 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1438 printf("%s: WARNING: unable to register helper hook\n", __func__); 1439#endif 1440#ifdef STATS 1441 if (tcp_stats_init()) 1442 printf("%s: WARNING: unable to initialise TCP stats\n", 1443 __func__); 1444#endif 1445 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1446 tcp_tcbhashsize); 1447 1448 syncache_init(); 1449 tcp_hc_init(); 1450 1451 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1452 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1453 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1454 1455 tcp_fastopen_init(); 1456 1457 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1458 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1459 1460 V_tcp_msl = TCPTV_MSL; 1461} 1462VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1463 tcp_vnet_init, NULL); 1464 1465static void 1466tcp_init(void *arg __unused) 1467{ 1468 int hashsize; 1469 1470 tcp_reass_global_init(); 1471 1472 /* XXX virtualize those below? */ 1473 tcp_delacktime = TCPTV_DELACK; 1474 tcp_keepinit = TCPTV_KEEP_INIT; 1475 tcp_keepidle = TCPTV_KEEP_IDLE; 1476 tcp_keepintvl = TCPTV_KEEPINTVL; 1477 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1478 tcp_rexmit_initial = TCPTV_RTOBASE; 1479 if (tcp_rexmit_initial < 1) 1480 tcp_rexmit_initial = 1; 1481 tcp_rexmit_min = TCPTV_MIN; 1482 if (tcp_rexmit_min < 1) 1483 tcp_rexmit_min = 1; 1484 tcp_persmin = TCPTV_PERSMIN; 1485 tcp_persmax = TCPTV_PERSMAX; 1486 tcp_rexmit_slop = TCPTV_CPU_VAR; 1487 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1488 1489 /* Setup the tcp function block list */ 1490 TAILQ_INIT(&t_functions); 1491 rw_init(&tcp_function_lock, "tcp_func_lock"); 1492 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1493 sx_init(&tcpoudp_lock, "TCP over UDP configuration"); 1494#ifdef TCP_BLACKBOX 1495 /* Initialize the TCP logging data. */ 1496 tcp_log_init(); 1497#endif 1498 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1499 1500 if (tcp_soreceive_stream) { 1501#ifdef INET 1502 tcp_protosw.pr_soreceive = soreceive_stream; 1503#endif 1504#ifdef INET6 1505 tcp6_protosw.pr_soreceive = soreceive_stream; 1506#endif /* INET6 */ 1507 } 1508 1509#ifdef INET6 1510 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1511#else /* INET6 */ 1512 max_protohdr_grow(sizeof(struct tcpiphdr)); 1513#endif /* INET6 */ 1514 1515 ISN_LOCK_INIT(); 1516 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1517 SHUTDOWN_PRI_DEFAULT); 1518 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1519 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1520 1521 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1522 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1523 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1524 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1525 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1526 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1527 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1528 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1529 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1530 tcp_pacing_failures = counter_u64_alloc(M_WAITOK); 1531 tcp_dgp_failures = counter_u64_alloc(M_WAITOK); 1532#ifdef TCPPCAP 1533 tcp_pcap_init(); 1534#endif 1535 1536 hashsize = tcp_tcbhashsize; 1537 if (hashsize == 0) { 1538 /* 1539 * Auto tune the hash size based on maxsockets. 1540 * A perfect hash would have a 1:1 mapping 1541 * (hashsize = maxsockets) however it's been 1542 * suggested that O(2) average is better. 1543 */ 1544 hashsize = maketcp_hashsize(maxsockets / 4); 1545 /* 1546 * Our historical default is 512, 1547 * do not autotune lower than this. 1548 */ 1549 if (hashsize < 512) 1550 hashsize = 512; 1551 if (bootverbose) 1552 printf("%s: %s auto tuned to %d\n", __func__, 1553 "net.inet.tcp.tcbhashsize", hashsize); 1554 } 1555 /* 1556 * We require a hashsize to be a power of two. 1557 * Previously if it was not a power of two we would just reset it 1558 * back to 512, which could be a nasty surprise if you did not notice 1559 * the error message. 1560 * Instead what we do is clip it to the closest power of two lower 1561 * than the specified hash value. 1562 */ 1563 if (!powerof2(hashsize)) { 1564 int oldhashsize = hashsize; 1565 1566 hashsize = maketcp_hashsize(hashsize); 1567 /* prevent absurdly low value */ 1568 if (hashsize < 16) 1569 hashsize = 16; 1570 printf("%s: WARNING: TCB hash size not a power of 2, " 1571 "clipped from %d to %d.\n", __func__, oldhashsize, 1572 hashsize); 1573 } 1574 tcp_tcbhashsize = hashsize; 1575 1576#ifdef INET 1577 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1578#endif 1579#ifdef INET6 1580 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1581#endif 1582} 1583SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1584 1585#ifdef VIMAGE 1586static void 1587tcp_destroy(void *unused __unused) 1588{ 1589 int n; 1590#ifdef TCP_HHOOK 1591 int error; 1592#endif 1593 1594 /* 1595 * All our processes are gone, all our sockets should be cleaned 1596 * up, which means, we should be past the tcp_discardcb() calls. 1597 * Sleep to let all tcpcb timers really disappear and cleanup. 1598 */ 1599 for (;;) { 1600 INP_INFO_WLOCK(&V_tcbinfo); 1601 n = V_tcbinfo.ipi_count; 1602 INP_INFO_WUNLOCK(&V_tcbinfo); 1603 if (n == 0) 1604 break; 1605 pause("tcpdes", hz / 10); 1606 } 1607 tcp_hc_destroy(); 1608 syncache_destroy(); 1609 in_pcbinfo_destroy(&V_tcbinfo); 1610 /* tcp_discardcb() clears the sack_holes up. */ 1611 uma_zdestroy(V_sack_hole_zone); 1612 1613 /* 1614 * Cannot free the zone until all tcpcbs are released as we attach 1615 * the allocations to them. 1616 */ 1617 tcp_fastopen_destroy(); 1618 1619 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1620 VNET_PCPUSTAT_FREE(tcpstat); 1621 1622#ifdef TCP_HHOOK 1623 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1624 if (error != 0) { 1625 printf("%s: WARNING: unable to deregister helper hook " 1626 "type=%d, id=%d: error %d returned\n", __func__, 1627 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1628 } 1629 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1630 if (error != 0) { 1631 printf("%s: WARNING: unable to deregister helper hook " 1632 "type=%d, id=%d: error %d returned\n", __func__, 1633 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1634 } 1635#endif 1636} 1637VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1638#endif 1639 1640void 1641tcp_fini(void *xtp) 1642{ 1643 1644} 1645 1646/* 1647 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1648 * tcp_template used to store this data in mbufs, but we now recopy it out 1649 * of the tcpcb each time to conserve mbufs. 1650 */ 1651void 1652tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1653{ 1654 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1655 1656 INP_WLOCK_ASSERT(inp); 1657 1658#ifdef INET6 1659 if ((inp->inp_vflag & INP_IPV6) != 0) { 1660 struct ip6_hdr *ip6; 1661 1662 ip6 = (struct ip6_hdr *)ip_ptr; 1663 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1664 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1665 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1666 (IPV6_VERSION & IPV6_VERSION_MASK); 1667 if (port == 0) 1668 ip6->ip6_nxt = IPPROTO_TCP; 1669 else 1670 ip6->ip6_nxt = IPPROTO_UDP; 1671 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1672 ip6->ip6_src = inp->in6p_laddr; 1673 ip6->ip6_dst = inp->in6p_faddr; 1674 } 1675#endif /* INET6 */ 1676#if defined(INET6) && defined(INET) 1677 else 1678#endif 1679#ifdef INET 1680 { 1681 struct ip *ip; 1682 1683 ip = (struct ip *)ip_ptr; 1684 ip->ip_v = IPVERSION; 1685 ip->ip_hl = 5; 1686 ip->ip_tos = inp->inp_ip_tos; 1687 ip->ip_len = 0; 1688 ip->ip_id = 0; 1689 ip->ip_off = 0; 1690 ip->ip_ttl = inp->inp_ip_ttl; 1691 ip->ip_sum = 0; 1692 if (port == 0) 1693 ip->ip_p = IPPROTO_TCP; 1694 else 1695 ip->ip_p = IPPROTO_UDP; 1696 ip->ip_src = inp->inp_laddr; 1697 ip->ip_dst = inp->inp_faddr; 1698 } 1699#endif /* INET */ 1700 th->th_sport = inp->inp_lport; 1701 th->th_dport = inp->inp_fport; 1702 th->th_seq = 0; 1703 th->th_ack = 0; 1704 th->th_off = 5; 1705 tcp_set_flags(th, 0); 1706 th->th_win = 0; 1707 th->th_urp = 0; 1708 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1709} 1710 1711/* 1712 * Create template to be used to send tcp packets on a connection. 1713 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1714 * use for this function is in keepalives, which use tcp_respond. 1715 */ 1716struct tcptemp * 1717tcpip_maketemplate(struct inpcb *inp) 1718{ 1719 struct tcptemp *t; 1720 1721 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1722 if (t == NULL) 1723 return (NULL); 1724 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1725 return (t); 1726} 1727 1728/* 1729 * Send a single message to the TCP at address specified by 1730 * the given TCP/IP header. If m == NULL, then we make a copy 1731 * of the tcpiphdr at th and send directly to the addressed host. 1732 * This is used to force keep alive messages out using the TCP 1733 * template for a connection. If flags are given then we send 1734 * a message back to the TCP which originated the segment th, 1735 * and discard the mbuf containing it and any other attached mbufs. 1736 * 1737 * In any case the ack and sequence number of the transmitted 1738 * segment are as specified by the parameters. 1739 * 1740 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1741 */ 1742 1743void 1744tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1745 tcp_seq ack, tcp_seq seq, uint16_t flags) 1746{ 1747 struct tcpopt to; 1748 struct inpcb *inp; 1749 struct ip *ip; 1750 struct mbuf *optm; 1751 struct udphdr *uh = NULL; 1752 struct tcphdr *nth; 1753 struct tcp_log_buffer *lgb; 1754 u_char *optp; 1755#ifdef INET6 1756 struct ip6_hdr *ip6; 1757 int isipv6; 1758#endif /* INET6 */ 1759 int optlen, tlen, win, ulen; 1760 int ect = 0; 1761 bool incl_opts; 1762 uint16_t port; 1763 int output_ret; 1764#ifdef INVARIANTS 1765 int thflags = tcp_get_flags(th); 1766#endif 1767 1768 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1769 NET_EPOCH_ASSERT(); 1770 1771#ifdef INET6 1772 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1773 ip6 = ipgen; 1774#endif /* INET6 */ 1775 ip = ipgen; 1776 1777 if (tp != NULL) { 1778 inp = tptoinpcb(tp); 1779 INP_LOCK_ASSERT(inp); 1780 } else 1781 inp = NULL; 1782 1783 if (m != NULL) { 1784#ifdef INET6 1785 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1786 port = m->m_pkthdr.tcp_tun_port; 1787 else 1788#endif 1789 if (ip && (ip->ip_p == IPPROTO_UDP)) 1790 port = m->m_pkthdr.tcp_tun_port; 1791 else 1792 port = 0; 1793 } else 1794 port = tp->t_port; 1795 1796 incl_opts = false; 1797 win = 0; 1798 if (tp != NULL) { 1799 if (!(flags & TH_RST)) { 1800 win = sbspace(&inp->inp_socket->so_rcv); 1801 if (win > TCP_MAXWIN << tp->rcv_scale) 1802 win = TCP_MAXWIN << tp->rcv_scale; 1803 } 1804 if ((tp->t_flags & TF_NOOPT) == 0) 1805 incl_opts = true; 1806 } 1807 if (m == NULL) { 1808 m = m_gethdr(M_NOWAIT, MT_DATA); 1809 if (m == NULL) 1810 return; 1811 m->m_data += max_linkhdr; 1812#ifdef INET6 1813 if (isipv6) { 1814 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1815 sizeof(struct ip6_hdr)); 1816 ip6 = mtod(m, struct ip6_hdr *); 1817 nth = (struct tcphdr *)(ip6 + 1); 1818 if (port) { 1819 /* Insert a UDP header */ 1820 uh = (struct udphdr *)nth; 1821 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1822 uh->uh_dport = port; 1823 nth = (struct tcphdr *)(uh + 1); 1824 } 1825 } else 1826#endif /* INET6 */ 1827 { 1828 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1829 ip = mtod(m, struct ip *); 1830 nth = (struct tcphdr *)(ip + 1); 1831 if (port) { 1832 /* Insert a UDP header */ 1833 uh = (struct udphdr *)nth; 1834 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1835 uh->uh_dport = port; 1836 nth = (struct tcphdr *)(uh + 1); 1837 } 1838 } 1839 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1840 flags = TH_ACK; 1841 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1842 struct mbuf *n; 1843 1844 /* Can't reuse 'm', allocate a new mbuf. */ 1845 n = m_gethdr(M_NOWAIT, MT_DATA); 1846 if (n == NULL) { 1847 m_freem(m); 1848 return; 1849 } 1850 1851 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1852 m_freem(m); 1853 m_freem(n); 1854 return; 1855 } 1856 1857 n->m_data += max_linkhdr; 1858 /* m_len is set later */ 1859#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1860#ifdef INET6 1861 if (isipv6) { 1862 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1863 sizeof(struct ip6_hdr)); 1864 ip6 = mtod(n, struct ip6_hdr *); 1865 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1866 nth = (struct tcphdr *)(ip6 + 1); 1867 if (port) { 1868 /* Insert a UDP header */ 1869 uh = (struct udphdr *)nth; 1870 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1871 uh->uh_dport = port; 1872 nth = (struct tcphdr *)(uh + 1); 1873 } 1874 } else 1875#endif /* INET6 */ 1876 { 1877 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1878 ip = mtod(n, struct ip *); 1879 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1880 nth = (struct tcphdr *)(ip + 1); 1881 if (port) { 1882 /* Insert a UDP header */ 1883 uh = (struct udphdr *)nth; 1884 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1885 uh->uh_dport = port; 1886 nth = (struct tcphdr *)(uh + 1); 1887 } 1888 } 1889 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1890 xchg(nth->th_dport, nth->th_sport, uint16_t); 1891 th = nth; 1892 m_freem(m); 1893 m = n; 1894 } else { 1895 /* 1896 * reuse the mbuf. 1897 * XXX MRT We inherit the FIB, which is lucky. 1898 */ 1899 m_freem(m->m_next); 1900 m->m_next = NULL; 1901 m->m_data = (caddr_t)ipgen; 1902 /* clear any receive flags for proper bpf timestamping */ 1903 m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO); 1904 /* m_len is set later */ 1905#ifdef INET6 1906 if (isipv6) { 1907 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1908 nth = (struct tcphdr *)(ip6 + 1); 1909 } else 1910#endif /* INET6 */ 1911 { 1912 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1913 nth = (struct tcphdr *)(ip + 1); 1914 } 1915 if (th != nth) { 1916 /* 1917 * this is usually a case when an extension header 1918 * exists between the IPv6 header and the 1919 * TCP header. 1920 */ 1921 nth->th_sport = th->th_sport; 1922 nth->th_dport = th->th_dport; 1923 } 1924 xchg(nth->th_dport, nth->th_sport, uint16_t); 1925#undef xchg 1926 } 1927 tlen = 0; 1928#ifdef INET6 1929 if (isipv6) 1930 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1931#endif 1932#if defined(INET) && defined(INET6) 1933 else 1934#endif 1935#ifdef INET 1936 tlen = sizeof (struct tcpiphdr); 1937#endif 1938 if (port) 1939 tlen += sizeof (struct udphdr); 1940#ifdef INVARIANTS 1941 m->m_len = 0; 1942 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1943 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1944 m, tlen, (long)M_TRAILINGSPACE(m))); 1945#endif 1946 m->m_len = tlen; 1947 to.to_flags = 0; 1948 if (incl_opts) { 1949 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1950 /* Make sure we have room. */ 1951 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1952 m->m_next = m_get(M_NOWAIT, MT_DATA); 1953 if (m->m_next) { 1954 optp = mtod(m->m_next, u_char *); 1955 optm = m->m_next; 1956 } else 1957 incl_opts = false; 1958 } else { 1959 optp = (u_char *) (nth + 1); 1960 optm = m; 1961 } 1962 } 1963 if (incl_opts) { 1964 /* Timestamps. */ 1965 if (tp->t_flags & TF_RCVD_TSTMP) { 1966 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1967 to.to_tsecr = tp->ts_recent; 1968 to.to_flags |= TOF_TS; 1969 } 1970#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1971 /* TCP-MD5 (RFC2385). */ 1972 if (tp->t_flags & TF_SIGNATURE) 1973 to.to_flags |= TOF_SIGNATURE; 1974#endif 1975 /* Add the options. */ 1976 tlen += optlen = tcp_addoptions(&to, optp); 1977 1978 /* Update m_len in the correct mbuf. */ 1979 optm->m_len += optlen; 1980 } else 1981 optlen = 0; 1982#ifdef INET6 1983 if (isipv6) { 1984 if (uh) { 1985 ulen = tlen - sizeof(struct ip6_hdr); 1986 uh->uh_ulen = htons(ulen); 1987 } 1988 ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN); 1989 ip6->ip6_vfc = IPV6_VERSION; 1990 if (port) 1991 ip6->ip6_nxt = IPPROTO_UDP; 1992 else 1993 ip6->ip6_nxt = IPPROTO_TCP; 1994 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1995 } 1996#endif 1997#if defined(INET) && defined(INET6) 1998 else 1999#endif 2000#ifdef INET 2001 { 2002 if (uh) { 2003 ulen = tlen - sizeof(struct ip); 2004 uh->uh_ulen = htons(ulen); 2005 } 2006 ip->ip_len = htons(tlen); 2007 if (inp != NULL) { 2008 ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK; 2009 ip->ip_ttl = inp->inp_ip_ttl; 2010 } else { 2011 ip->ip_tos = 0; 2012 ip->ip_ttl = V_ip_defttl; 2013 } 2014 ip->ip_tos |= ect; 2015 if (port) { 2016 ip->ip_p = IPPROTO_UDP; 2017 } else { 2018 ip->ip_p = IPPROTO_TCP; 2019 } 2020 if (V_path_mtu_discovery) 2021 ip->ip_off |= htons(IP_DF); 2022 } 2023#endif 2024 m->m_pkthdr.len = tlen; 2025 m->m_pkthdr.rcvif = NULL; 2026#ifdef MAC 2027 if (inp != NULL) { 2028 /* 2029 * Packet is associated with a socket, so allow the 2030 * label of the response to reflect the socket label. 2031 */ 2032 INP_LOCK_ASSERT(inp); 2033 mac_inpcb_create_mbuf(inp, m); 2034 } else { 2035 /* 2036 * Packet is not associated with a socket, so possibly 2037 * update the label in place. 2038 */ 2039 mac_netinet_tcp_reply(m); 2040 } 2041#endif 2042 nth->th_seq = htonl(seq); 2043 nth->th_ack = htonl(ack); 2044 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2045 tcp_set_flags(nth, flags); 2046 if (tp && (flags & TH_RST)) { 2047 /* Log the reset */ 2048 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2049 } 2050 if (tp != NULL) 2051 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2052 else 2053 nth->th_win = htons((u_short)win); 2054 nth->th_urp = 0; 2055 2056#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2057 if (to.to_flags & TOF_SIGNATURE) { 2058 if (!TCPMD5_ENABLED() || 2059 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2060 m_freem(m); 2061 return; 2062 } 2063 } 2064#endif 2065 2066#ifdef INET6 2067 if (isipv6) { 2068 if (port) { 2069 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2070 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2071 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2072 nth->th_sum = 0; 2073 } else { 2074 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2075 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2076 nth->th_sum = in6_cksum_pseudo(ip6, 2077 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2078 } 2079 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2080 } 2081#endif /* INET6 */ 2082#if defined(INET6) && defined(INET) 2083 else 2084#endif 2085#ifdef INET 2086 { 2087 if (port) { 2088 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2089 htons(ulen + IPPROTO_UDP)); 2090 m->m_pkthdr.csum_flags = CSUM_UDP; 2091 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2092 nth->th_sum = 0; 2093 } else { 2094 m->m_pkthdr.csum_flags = CSUM_TCP; 2095 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2096 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2097 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2098 } 2099 } 2100#endif /* INET */ 2101 TCP_PROBE3(debug__output, tp, th, m); 2102 if (flags & TH_RST) 2103 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2104 lgb = NULL; 2105 if ((tp != NULL) && tcp_bblogging_on(tp)) { 2106 if (INP_WLOCKED(inp)) { 2107 union tcp_log_stackspecific log; 2108 struct timeval tv; 2109 2110 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2111 log.u_bbr.inhpts = tcp_in_hpts(tp); 2112 log.u_bbr.flex8 = 4; 2113 log.u_bbr.pkts_out = tp->t_maxseg; 2114 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2115 log.u_bbr.delivered = 0; 2116 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT, 2117 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2118 } else { 2119 /* 2120 * We can not log the packet, since we only own the 2121 * read lock, but a write lock is needed. The read lock 2122 * is not upgraded to a write lock, since only getting 2123 * the read lock was done intentionally to improve the 2124 * handling of SYN flooding attacks. 2125 * This happens only for pure SYN segments received in 2126 * the initial CLOSED state, or received in a more 2127 * advanced state than listen and the UDP encapsulation 2128 * port is unexpected. 2129 * The incoming SYN segments do not really belong to 2130 * the TCP connection and the handling does not change 2131 * the state of the TCP connection. Therefore, the 2132 * sending of the RST segments is not logged. Please 2133 * note that also the incoming SYN segments are not 2134 * logged. 2135 * 2136 * The following code ensures that the above description 2137 * is and stays correct. 2138 */ 2139 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2140 (tp->t_state == TCPS_CLOSED || 2141 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2142 ("%s: Logging of TCP segment with flags 0x%b and " 2143 "UDP encapsulation port %u skipped in state %s", 2144 __func__, thflags, PRINT_TH_FLAGS, 2145 ntohs(port), tcpstates[tp->t_state])); 2146 } 2147 } 2148 2149 if (flags & TH_ACK) 2150 TCPSTAT_INC(tcps_sndacks); 2151 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2152 TCPSTAT_INC(tcps_sndctrl); 2153 TCPSTAT_INC(tcps_sndtotal); 2154 2155#ifdef INET6 2156 if (isipv6) { 2157 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2158 output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL, 2159 NULL, 0, NULL, NULL, inp); 2160 } 2161#endif /* INET6 */ 2162#if defined(INET) && defined(INET6) 2163 else 2164#endif 2165#ifdef INET 2166 { 2167 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2168 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2169 } 2170#endif 2171 if (lgb != NULL) 2172 lgb->tlb_errno = output_ret; 2173} 2174 2175/* 2176 * Create a new TCP control block, making an empty reassembly queue and hooking 2177 * it to the argument protocol control block. The `inp' parameter must have 2178 * come from the zone allocator set up by tcpcbstor declaration. 2179 */ 2180struct tcpcb * 2181tcp_newtcpcb(struct inpcb *inp) 2182{ 2183 struct tcpcb *tp = intotcpcb(inp); 2184#ifdef INET6 2185 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2186#endif /* INET6 */ 2187 2188 /* 2189 * Historically allocation was done with M_ZERO. There is a lot of 2190 * code that rely on that. For now take safe approach and zero whole 2191 * tcpcb. This definitely can be optimized. 2192 */ 2193 bzero(&tp->t_start_zero, t_zero_size); 2194 2195 /* Initialise cc_var struct for this tcpcb. */ 2196 tp->t_ccv.type = IPPROTO_TCP; 2197 tp->t_ccv.ccvc.tcp = tp; 2198 rw_rlock(&tcp_function_lock); 2199 tp->t_fb = V_tcp_func_set_ptr; 2200 refcount_acquire(&tp->t_fb->tfb_refcnt); 2201 rw_runlock(&tcp_function_lock); 2202 /* 2203 * Use the current system default CC algorithm. 2204 */ 2205 cc_attach(tp, CC_DEFAULT_ALGO()); 2206 2207 if (CC_ALGO(tp)->cb_init != NULL) 2208 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2209 cc_detach(tp); 2210 if (tp->t_fb->tfb_tcp_fb_fini) 2211 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2212 refcount_release(&tp->t_fb->tfb_refcnt); 2213 return (NULL); 2214 } 2215 2216#ifdef TCP_HHOOK 2217 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2218 if (tp->t_fb->tfb_tcp_fb_fini) 2219 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2220 refcount_release(&tp->t_fb->tfb_refcnt); 2221 return (NULL); 2222 } 2223#endif 2224 2225 TAILQ_INIT(&tp->t_segq); 2226 STAILQ_INIT(&tp->t_inqueue); 2227 tp->t_maxseg = 2228#ifdef INET6 2229 isipv6 ? V_tcp_v6mssdflt : 2230#endif /* INET6 */ 2231 V_tcp_mssdflt; 2232 2233 /* All mbuf queue/ack compress flags should be off */ 2234 tcp_lro_features_off(tp); 2235 2236 tp->t_hpts_cpu = HPTS_CPU_NONE; 2237 tp->t_lro_cpu = HPTS_CPU_NONE; 2238 2239 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED); 2240 for (int i = 0; i < TT_N; i++) 2241 tp->t_timers[i] = SBT_MAX; 2242 2243 switch (V_tcp_do_rfc1323) { 2244 case 0: 2245 break; 2246 default: 2247 case 1: 2248 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2249 break; 2250 case 2: 2251 tp->t_flags = TF_REQ_SCALE; 2252 break; 2253 case 3: 2254 tp->t_flags = TF_REQ_TSTMP; 2255 break; 2256 } 2257 if (V_tcp_do_sack) 2258 tp->t_flags |= TF_SACK_PERMIT; 2259 TAILQ_INIT(&tp->snd_holes); 2260 2261 /* 2262 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2263 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2264 * reasonable initial retransmit time. 2265 */ 2266 tp->t_srtt = TCPTV_SRTTBASE; 2267 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2268 tp->t_rttmin = tcp_rexmit_min; 2269 tp->t_rxtcur = tcp_rexmit_initial; 2270 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2271 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2272 tp->t_rcvtime = ticks; 2273 /* We always start with ticks granularity */ 2274 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 2275 /* 2276 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2277 * because the socket may be bound to an IPv6 wildcard address, 2278 * which may match an IPv4-mapped IPv6 address. 2279 */ 2280 inp->inp_ip_ttl = V_ip_defttl; 2281#ifdef TCPPCAP 2282 /* 2283 * Init the TCP PCAP queues. 2284 */ 2285 tcp_pcap_tcpcb_init(tp); 2286#endif 2287#ifdef TCP_BLACKBOX 2288 /* Initialize the per-TCPCB log data. */ 2289 tcp_log_tcpcbinit(tp); 2290#endif 2291 tp->t_pacing_rate = -1; 2292 if (tp->t_fb->tfb_tcp_fb_init) { 2293 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) { 2294 refcount_release(&tp->t_fb->tfb_refcnt); 2295 return (NULL); 2296 } 2297 } 2298#ifdef STATS 2299 if (V_tcp_perconn_stats_enable == 1) 2300 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2301#endif 2302 if (V_tcp_do_lrd) 2303 tp->t_flags |= TF_LRD; 2304 2305 return (tp); 2306} 2307 2308/* 2309 * Drop a TCP connection, reporting 2310 * the specified error. If connection is synchronized, 2311 * then send a RST to peer. 2312 */ 2313struct tcpcb * 2314tcp_drop(struct tcpcb *tp, int errno) 2315{ 2316 struct socket *so = tptosocket(tp); 2317 2318 NET_EPOCH_ASSERT(); 2319 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2320 2321 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2322 tcp_state_change(tp, TCPS_CLOSED); 2323 /* Don't use tcp_output() here due to possible recursion. */ 2324 (void)tcp_output_nodrop(tp); 2325 TCPSTAT_INC(tcps_drops); 2326 } else 2327 TCPSTAT_INC(tcps_conndrops); 2328 if (errno == ETIMEDOUT && tp->t_softerror) 2329 errno = tp->t_softerror; 2330 so->so_error = errno; 2331 return (tcp_close(tp)); 2332} 2333 2334void 2335tcp_discardcb(struct tcpcb *tp) 2336{ 2337 struct inpcb *inp = tptoinpcb(tp); 2338 struct socket *so = tptosocket(tp); 2339 struct mbuf *m; 2340#ifdef INET6 2341 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2342#endif 2343 2344 INP_WLOCK_ASSERT(inp); 2345 MPASS(!callout_active(&tp->t_callout)); 2346 MPASS(TAILQ_EMPTY(&tp->snd_holes)); 2347 2348 /* free the reassembly queue, if any */ 2349 tcp_reass_flush(tp); 2350 2351#ifdef TCP_OFFLOAD 2352 /* Disconnect offload device, if any. */ 2353 if (tp->t_flags & TF_TOE) 2354 tcp_offload_detach(tp); 2355#endif 2356#ifdef TCPPCAP 2357 /* Free the TCP PCAP queues. */ 2358 tcp_pcap_drain(&(tp->t_inpkts)); 2359 tcp_pcap_drain(&(tp->t_outpkts)); 2360#endif 2361 2362 /* Allow the CC algorithm to clean up after itself. */ 2363 if (CC_ALGO(tp)->cb_destroy != NULL) 2364 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2365 CC_DATA(tp) = NULL; 2366 /* Detach from the CC algorithm */ 2367 cc_detach(tp); 2368 2369#ifdef TCP_HHOOK 2370 khelp_destroy_osd(&tp->t_osd); 2371#endif 2372#ifdef STATS 2373 stats_blob_destroy(tp->t_stats); 2374#endif 2375 2376 CC_ALGO(tp) = NULL; 2377 if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) { 2378 struct mbuf *prev; 2379 2380 STAILQ_INIT(&tp->t_inqueue); 2381 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev) 2382 m_freem(m); 2383 } 2384 TCPSTATES_DEC(tp->t_state); 2385 2386 if (tp->t_fb->tfb_tcp_fb_fini) 2387 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2388 MPASS(!tcp_in_hpts(tp)); 2389#ifdef TCP_BLACKBOX 2390 tcp_log_tcpcbfini(tp); 2391#endif 2392 2393 /* 2394 * If we got enough samples through the srtt filter, 2395 * save the rtt and rttvar in the routing entry. 2396 * 'Enough' is arbitrarily defined as 4 rtt samples. 2397 * 4 samples is enough for the srtt filter to converge 2398 * to within enough % of the correct value; fewer samples 2399 * and we could save a bogus rtt. The danger is not high 2400 * as tcp quickly recovers from everything. 2401 * XXX: Works very well but needs some more statistics! 2402 * 2403 * XXXRRS: Updating must be after the stack fini() since 2404 * that may be converting some internal representation of 2405 * say srtt etc into the general one used by other stacks. 2406 * Lets also at least protect against the so being NULL 2407 * as RW stated below. 2408 */ 2409 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2410 struct hc_metrics_lite metrics; 2411 uint32_t ssthresh; 2412 2413 bzero(&metrics, sizeof(metrics)); 2414 /* 2415 * Update the ssthresh always when the conditions below 2416 * are satisfied. This gives us better new start value 2417 * for the congestion avoidance for new connections. 2418 * ssthresh is only set if packet loss occurred on a session. 2419 * 2420 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2421 * being torn down. Ideally this code would not use 'so'. 2422 */ 2423 ssthresh = tp->snd_ssthresh; 2424 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2425 /* 2426 * convert the limit from user data bytes to 2427 * packets then to packet data bytes. 2428 */ 2429 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2430 if (ssthresh < 2) 2431 ssthresh = 2; 2432 ssthresh *= (tp->t_maxseg + 2433#ifdef INET6 2434 (isipv6 ? sizeof (struct ip6_hdr) + 2435 sizeof (struct tcphdr) : 2436#endif 2437 sizeof (struct tcpiphdr) 2438#ifdef INET6 2439 ) 2440#endif 2441 ); 2442 } else 2443 ssthresh = 0; 2444 metrics.rmx_ssthresh = ssthresh; 2445 2446 metrics.rmx_rtt = tp->t_srtt; 2447 metrics.rmx_rttvar = tp->t_rttvar; 2448 metrics.rmx_cwnd = tp->snd_cwnd; 2449 metrics.rmx_sendpipe = 0; 2450 metrics.rmx_recvpipe = 0; 2451 2452 tcp_hc_update(&inp->inp_inc, &metrics); 2453 } 2454 2455 refcount_release(&tp->t_fb->tfb_refcnt); 2456} 2457 2458/* 2459 * Attempt to close a TCP control block, marking it as dropped, and freeing 2460 * the socket if we hold the only reference. 2461 */ 2462struct tcpcb * 2463tcp_close(struct tcpcb *tp) 2464{ 2465 struct inpcb *inp = tptoinpcb(tp); 2466 struct socket *so = tptosocket(tp); 2467 2468 INP_WLOCK_ASSERT(inp); 2469 2470#ifdef TCP_OFFLOAD 2471 if (tp->t_state == TCPS_LISTEN) 2472 tcp_offload_listen_stop(tp); 2473#endif 2474 /* 2475 * This releases the TFO pending counter resource for TFO listen 2476 * sockets as well as passively-created TFO sockets that transition 2477 * from SYN_RECEIVED to CLOSED. 2478 */ 2479 if (tp->t_tfo_pending) { 2480 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2481 tp->t_tfo_pending = NULL; 2482 } 2483 tcp_timer_stop(tp); 2484 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 2485 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2486 in_pcbdrop(inp); 2487 TCPSTAT_INC(tcps_closed); 2488 if (tp->t_state != TCPS_CLOSED) 2489 tcp_state_change(tp, TCPS_CLOSED); 2490 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2491 tcp_free_sackholes(tp); 2492 soisdisconnected(so); 2493 if (inp->inp_flags & INP_SOCKREF) { 2494 inp->inp_flags &= ~INP_SOCKREF; 2495 INP_WUNLOCK(inp); 2496 sorele(so); 2497 return (NULL); 2498 } 2499 return (tp); 2500} 2501 2502/* 2503 * Notify a tcp user of an asynchronous error; 2504 * store error as soft error, but wake up user 2505 * (for now, won't do anything until can select for soft error). 2506 * 2507 * Do not wake up user since there currently is no mechanism for 2508 * reporting soft errors (yet - a kqueue filter may be added). 2509 */ 2510static struct inpcb * 2511tcp_notify(struct inpcb *inp, int error) 2512{ 2513 struct tcpcb *tp; 2514 2515 INP_WLOCK_ASSERT(inp); 2516 2517 tp = intotcpcb(inp); 2518 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2519 2520 /* 2521 * Ignore some errors if we are hooked up. 2522 * If connection hasn't completed, has retransmitted several times, 2523 * and receives a second error, give up now. This is better 2524 * than waiting a long time to establish a connection that 2525 * can never complete. 2526 */ 2527 if (tp->t_state == TCPS_ESTABLISHED && 2528 (error == EHOSTUNREACH || error == ENETUNREACH || 2529 error == EHOSTDOWN)) { 2530 if (inp->inp_route.ro_nh) { 2531 NH_FREE(inp->inp_route.ro_nh); 2532 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2533 } 2534 return (inp); 2535 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2536 tp->t_softerror) { 2537 tp = tcp_drop(tp, error); 2538 if (tp != NULL) 2539 return (inp); 2540 else 2541 return (NULL); 2542 } else { 2543 tp->t_softerror = error; 2544 return (inp); 2545 } 2546#if 0 2547 wakeup( &so->so_timeo); 2548 sorwakeup(so); 2549 sowwakeup(so); 2550#endif 2551} 2552 2553static int 2554tcp_pcblist(SYSCTL_HANDLER_ARGS) 2555{ 2556 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2557 INPLOOKUP_RLOCKPCB); 2558 struct xinpgen xig; 2559 struct inpcb *inp; 2560 int error; 2561 2562 if (req->newptr != NULL) 2563 return (EPERM); 2564 2565 if (req->oldptr == NULL) { 2566 int n; 2567 2568 n = V_tcbinfo.ipi_count + 2569 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2570 n += imax(n / 8, 10); 2571 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2572 return (0); 2573 } 2574 2575 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2576 return (error); 2577 2578 bzero(&xig, sizeof(xig)); 2579 xig.xig_len = sizeof xig; 2580 xig.xig_count = V_tcbinfo.ipi_count + 2581 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2582 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2583 xig.xig_sogen = so_gencnt; 2584 error = SYSCTL_OUT(req, &xig, sizeof xig); 2585 if (error) 2586 return (error); 2587 2588 error = syncache_pcblist(req); 2589 if (error) 2590 return (error); 2591 2592 while ((inp = inp_next(&inpi)) != NULL) { 2593 if (inp->inp_gencnt <= xig.xig_gen && 2594 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2595 struct xtcpcb xt; 2596 2597 tcp_inptoxtp(inp, &xt); 2598 error = SYSCTL_OUT(req, &xt, sizeof xt); 2599 if (error) { 2600 INP_RUNLOCK(inp); 2601 break; 2602 } else 2603 continue; 2604 } 2605 } 2606 2607 if (!error) { 2608 /* 2609 * Give the user an updated idea of our state. 2610 * If the generation differs from what we told 2611 * her before, she knows that something happened 2612 * while we were processing this request, and it 2613 * might be necessary to retry. 2614 */ 2615 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2616 xig.xig_sogen = so_gencnt; 2617 xig.xig_count = V_tcbinfo.ipi_count + 2618 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2619 error = SYSCTL_OUT(req, &xig, sizeof xig); 2620 } 2621 2622 return (error); 2623} 2624 2625SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2626 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2627 NULL, 0, tcp_pcblist, "S,xtcpcb", 2628 "List of active TCP connections"); 2629 2630#ifdef INET 2631static int 2632tcp_getcred(SYSCTL_HANDLER_ARGS) 2633{ 2634 struct xucred xuc; 2635 struct sockaddr_in addrs[2]; 2636 struct epoch_tracker et; 2637 struct inpcb *inp; 2638 int error; 2639 2640 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2641 if (error) 2642 return (error); 2643 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2644 if (error) 2645 return (error); 2646 NET_EPOCH_ENTER(et); 2647 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2648 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2649 NET_EPOCH_EXIT(et); 2650 if (inp != NULL) { 2651 if (error == 0) 2652 error = cr_canseeinpcb(req->td->td_ucred, inp); 2653 if (error == 0) 2654 cru2x(inp->inp_cred, &xuc); 2655 INP_RUNLOCK(inp); 2656 } else 2657 error = ENOENT; 2658 if (error == 0) 2659 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2660 return (error); 2661} 2662 2663SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2664 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2665 0, 0, tcp_getcred, "S,xucred", 2666 "Get the xucred of a TCP connection"); 2667#endif /* INET */ 2668 2669#ifdef INET6 2670static int 2671tcp6_getcred(SYSCTL_HANDLER_ARGS) 2672{ 2673 struct epoch_tracker et; 2674 struct xucred xuc; 2675 struct sockaddr_in6 addrs[2]; 2676 struct inpcb *inp; 2677 int error; 2678#ifdef INET 2679 int mapped = 0; 2680#endif 2681 2682 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2683 if (error) 2684 return (error); 2685 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2686 if (error) 2687 return (error); 2688 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2689 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2690 return (error); 2691 } 2692 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2693#ifdef INET 2694 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2695 mapped = 1; 2696 else 2697#endif 2698 return (EINVAL); 2699 } 2700 2701 NET_EPOCH_ENTER(et); 2702#ifdef INET 2703 if (mapped == 1) 2704 inp = in_pcblookup(&V_tcbinfo, 2705 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2706 addrs[1].sin6_port, 2707 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2708 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2709 else 2710#endif 2711 inp = in6_pcblookup(&V_tcbinfo, 2712 &addrs[1].sin6_addr, addrs[1].sin6_port, 2713 &addrs[0].sin6_addr, addrs[0].sin6_port, 2714 INPLOOKUP_RLOCKPCB, NULL); 2715 NET_EPOCH_EXIT(et); 2716 if (inp != NULL) { 2717 if (error == 0) 2718 error = cr_canseeinpcb(req->td->td_ucred, inp); 2719 if (error == 0) 2720 cru2x(inp->inp_cred, &xuc); 2721 INP_RUNLOCK(inp); 2722 } else 2723 error = ENOENT; 2724 if (error == 0) 2725 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2726 return (error); 2727} 2728 2729SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2730 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2731 0, 0, tcp6_getcred, "S,xucred", 2732 "Get the xucred of a TCP6 connection"); 2733#endif /* INET6 */ 2734 2735#ifdef INET 2736/* Path MTU to try next when a fragmentation-needed message is received. */ 2737static inline int 2738tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2739{ 2740 int mtu = ntohs(icp->icmp_nextmtu); 2741 2742 /* If no alternative MTU was proposed, try the next smaller one. */ 2743 if (!mtu) 2744 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2745 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2746 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2747 2748 return (mtu); 2749} 2750 2751static void 2752tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2753{ 2754 struct ip *ip; 2755 struct tcphdr *th; 2756 struct inpcb *inp; 2757 struct tcpcb *tp; 2758 struct inpcb *(*notify)(struct inpcb *, int); 2759 struct in_conninfo inc; 2760 tcp_seq icmp_tcp_seq; 2761 int errno, mtu; 2762 2763 errno = icmp_errmap(icp); 2764 switch (errno) { 2765 case 0: 2766 return; 2767 case EMSGSIZE: 2768 notify = tcp_mtudisc_notify; 2769 break; 2770 case ECONNREFUSED: 2771 if (V_icmp_may_rst) 2772 notify = tcp_drop_syn_sent; 2773 else 2774 notify = tcp_notify; 2775 break; 2776 case EHOSTUNREACH: 2777 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2778 notify = tcp_drop_syn_sent; 2779 else 2780 notify = tcp_notify; 2781 break; 2782 default: 2783 notify = tcp_notify; 2784 } 2785 2786 ip = &icp->icmp_ip; 2787 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2788 icmp_tcp_seq = th->th_seq; 2789 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2790 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2791 if (inp != NULL) { 2792 tp = intotcpcb(inp); 2793#ifdef TCP_OFFLOAD 2794 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2795 /* 2796 * MTU discovery for offloaded connections. Let 2797 * the TOE driver verify seq# and process it. 2798 */ 2799 mtu = tcp_next_pmtu(icp, ip); 2800 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2801 goto out; 2802 } 2803#endif 2804 if (tp->t_port != port) 2805 goto out; 2806 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2807 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2808 if (errno == EMSGSIZE) { 2809 /* 2810 * MTU discovery: we got a needfrag and 2811 * will potentially try a lower MTU. 2812 */ 2813 mtu = tcp_next_pmtu(icp, ip); 2814 2815 /* 2816 * Only process the offered MTU if it 2817 * is smaller than the current one. 2818 */ 2819 if (mtu < tp->t_maxseg + 2820 sizeof(struct tcpiphdr)) { 2821 bzero(&inc, sizeof(inc)); 2822 inc.inc_faddr = ip->ip_dst; 2823 inc.inc_fibnum = 2824 inp->inp_inc.inc_fibnum; 2825 tcp_hc_updatemtu(&inc, mtu); 2826 inp = tcp_mtudisc(inp, mtu); 2827 } 2828 } else 2829 inp = (*notify)(inp, errno); 2830 } 2831 } else { 2832 bzero(&inc, sizeof(inc)); 2833 inc.inc_fport = th->th_dport; 2834 inc.inc_lport = th->th_sport; 2835 inc.inc_faddr = ip->ip_dst; 2836 inc.inc_laddr = ip->ip_src; 2837 syncache_unreach(&inc, icmp_tcp_seq, port); 2838 } 2839out: 2840 if (inp != NULL) 2841 INP_WUNLOCK(inp); 2842} 2843 2844static void 2845tcp_ctlinput(struct icmp *icmp) 2846{ 2847 tcp_ctlinput_with_port(icmp, htons(0)); 2848} 2849 2850static void 2851tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2852{ 2853 /* Its a tunneled TCP over UDP icmp */ 2854 struct icmp *icmp = param.icmp; 2855 struct ip *outer_ip, *inner_ip; 2856 struct udphdr *udp; 2857 struct tcphdr *th, ttemp; 2858 int i_hlen, o_len; 2859 uint16_t port; 2860 2861 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2862 inner_ip = &icmp->icmp_ip; 2863 i_hlen = inner_ip->ip_hl << 2; 2864 o_len = ntohs(outer_ip->ip_len); 2865 if (o_len < 2866 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2867 /* Not enough data present */ 2868 return; 2869 } 2870 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2871 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2872 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2873 return; 2874 } 2875 port = udp->uh_dport; 2876 th = (struct tcphdr *)(udp + 1); 2877 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2878 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2879 /* Now adjust down the size of the outer IP header */ 2880 o_len -= sizeof(struct udphdr); 2881 outer_ip->ip_len = htons(o_len); 2882 /* Now call in to the normal handling code */ 2883 tcp_ctlinput_with_port(icmp, port); 2884} 2885#endif /* INET */ 2886 2887#ifdef INET6 2888static inline int 2889tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2890{ 2891 int mtu = ntohl(icmp6->icmp6_mtu); 2892 2893 /* 2894 * If no alternative MTU was proposed, or the proposed MTU was too 2895 * small, set to the min. 2896 */ 2897 if (mtu < IPV6_MMTU) 2898 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2899 return (mtu); 2900} 2901 2902static void 2903tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2904{ 2905 struct in6_addr *dst; 2906 struct inpcb *(*notify)(struct inpcb *, int); 2907 struct ip6_hdr *ip6; 2908 struct mbuf *m; 2909 struct inpcb *inp; 2910 struct tcpcb *tp; 2911 struct icmp6_hdr *icmp6; 2912 struct in_conninfo inc; 2913 struct tcp_ports { 2914 uint16_t th_sport; 2915 uint16_t th_dport; 2916 } t_ports; 2917 tcp_seq icmp_tcp_seq; 2918 unsigned int mtu; 2919 unsigned int off; 2920 int errno; 2921 2922 icmp6 = ip6cp->ip6c_icmp6; 2923 m = ip6cp->ip6c_m; 2924 ip6 = ip6cp->ip6c_ip6; 2925 off = ip6cp->ip6c_off; 2926 dst = &ip6cp->ip6c_finaldst->sin6_addr; 2927 2928 errno = icmp6_errmap(icmp6); 2929 switch (errno) { 2930 case 0: 2931 return; 2932 case EMSGSIZE: 2933 notify = tcp_mtudisc_notify; 2934 break; 2935 case ECONNREFUSED: 2936 if (V_icmp_may_rst) 2937 notify = tcp_drop_syn_sent; 2938 else 2939 notify = tcp_notify; 2940 break; 2941 case EHOSTUNREACH: 2942 /* 2943 * There are only four ICMPs that may reset connection: 2944 * - administratively prohibited 2945 * - port unreachable 2946 * - time exceeded in transit 2947 * - unknown next header 2948 */ 2949 if (V_icmp_may_rst && 2950 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 2951 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 2952 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 2953 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 2954 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 2955 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 2956 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 2957 notify = tcp_drop_syn_sent; 2958 else 2959 notify = tcp_notify; 2960 break; 2961 default: 2962 notify = tcp_notify; 2963 } 2964 2965 /* Check if we can safely get the ports from the tcp hdr */ 2966 if (m == NULL || 2967 (m->m_pkthdr.len < 2968 (int32_t) (off + sizeof(struct tcp_ports)))) { 2969 return; 2970 } 2971 bzero(&t_ports, sizeof(struct tcp_ports)); 2972 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2973 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2974 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2975 off += sizeof(struct tcp_ports); 2976 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2977 goto out; 2978 } 2979 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2980 if (inp != NULL) { 2981 tp = intotcpcb(inp); 2982#ifdef TCP_OFFLOAD 2983 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2984 /* MTU discovery for offloaded connections. */ 2985 mtu = tcp6_next_pmtu(icmp6); 2986 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2987 goto out; 2988 } 2989#endif 2990 if (tp->t_port != port) 2991 goto out; 2992 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2993 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2994 if (errno == EMSGSIZE) { 2995 /* 2996 * MTU discovery: 2997 * If we got a needfrag set the MTU 2998 * in the route to the suggested new 2999 * value (if given) and then notify. 3000 */ 3001 mtu = tcp6_next_pmtu(icmp6); 3002 3003 bzero(&inc, sizeof(inc)); 3004 inc.inc_fibnum = M_GETFIB(m); 3005 inc.inc_flags |= INC_ISIPV6; 3006 inc.inc6_faddr = *dst; 3007 if (in6_setscope(&inc.inc6_faddr, 3008 m->m_pkthdr.rcvif, NULL)) 3009 goto out; 3010 /* 3011 * Only process the offered MTU if it 3012 * is smaller than the current one. 3013 */ 3014 if (mtu < tp->t_maxseg + 3015 sizeof (struct tcphdr) + 3016 sizeof (struct ip6_hdr)) { 3017 tcp_hc_updatemtu(&inc, mtu); 3018 tcp_mtudisc(inp, mtu); 3019 ICMP6STAT_INC(icp6s_pmtuchg); 3020 } 3021 } else 3022 inp = (*notify)(inp, errno); 3023 } 3024 } else { 3025 bzero(&inc, sizeof(inc)); 3026 inc.inc_fibnum = M_GETFIB(m); 3027 inc.inc_flags |= INC_ISIPV6; 3028 inc.inc_fport = t_ports.th_dport; 3029 inc.inc_lport = t_ports.th_sport; 3030 inc.inc6_faddr = *dst; 3031 inc.inc6_laddr = ip6->ip6_src; 3032 syncache_unreach(&inc, icmp_tcp_seq, port); 3033 } 3034out: 3035 if (inp != NULL) 3036 INP_WUNLOCK(inp); 3037} 3038 3039static void 3040tcp6_ctlinput(struct ip6ctlparam *ctl) 3041{ 3042 tcp6_ctlinput_with_port(ctl, htons(0)); 3043} 3044 3045static void 3046tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3047{ 3048 struct ip6ctlparam *ip6cp = param.ip6cp; 3049 struct mbuf *m; 3050 struct udphdr *udp; 3051 uint16_t port; 3052 3053 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3054 if (m == NULL) { 3055 return; 3056 } 3057 udp = mtod(m, struct udphdr *); 3058 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3059 return; 3060 } 3061 port = udp->uh_dport; 3062 m_adj(m, sizeof(struct udphdr)); 3063 if ((m->m_flags & M_PKTHDR) == 0) { 3064 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3065 } 3066 /* Now call in to the normal handling code */ 3067 tcp6_ctlinput_with_port(ip6cp, port); 3068} 3069 3070#endif /* INET6 */ 3071 3072static uint32_t 3073tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3074{ 3075 SIPHASH_CTX ctx; 3076 uint32_t hash[2]; 3077 3078 KASSERT(len >= SIPHASH_KEY_LENGTH, 3079 ("%s: keylen %u too short ", __func__, len)); 3080 SipHash24_Init(&ctx); 3081 SipHash_SetKey(&ctx, (uint8_t *)key); 3082 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3083 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3084 switch (inc->inc_flags & INC_ISIPV6) { 3085#ifdef INET 3086 case 0: 3087 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3088 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3089 break; 3090#endif 3091#ifdef INET6 3092 case INC_ISIPV6: 3093 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3094 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3095 break; 3096#endif 3097 } 3098 SipHash_Final((uint8_t *)hash, &ctx); 3099 3100 return (hash[0] ^ hash[1]); 3101} 3102 3103uint32_t 3104tcp_new_ts_offset(struct in_conninfo *inc) 3105{ 3106 struct in_conninfo inc_store, *local_inc; 3107 3108 if (!V_tcp_ts_offset_per_conn) { 3109 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3110 inc_store.inc_lport = 0; 3111 inc_store.inc_fport = 0; 3112 local_inc = &inc_store; 3113 } else { 3114 local_inc = inc; 3115 } 3116 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3117 sizeof(V_ts_offset_secret))); 3118} 3119 3120/* 3121 * Following is where TCP initial sequence number generation occurs. 3122 * 3123 * There are two places where we must use initial sequence numbers: 3124 * 1. In SYN-ACK packets. 3125 * 2. In SYN packets. 3126 * 3127 * All ISNs for SYN-ACK packets are generated by the syncache. See 3128 * tcp_syncache.c for details. 3129 * 3130 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3131 * depends on this property. In addition, these ISNs should be 3132 * unguessable so as to prevent connection hijacking. To satisfy 3133 * the requirements of this situation, the algorithm outlined in 3134 * RFC 1948 is used, with only small modifications. 3135 * 3136 * Implementation details: 3137 * 3138 * Time is based off the system timer, and is corrected so that it 3139 * increases by one megabyte per second. This allows for proper 3140 * recycling on high speed LANs while still leaving over an hour 3141 * before rollover. 3142 * 3143 * As reading the *exact* system time is too expensive to be done 3144 * whenever setting up a TCP connection, we increment the time 3145 * offset in two ways. First, a small random positive increment 3146 * is added to isn_offset for each connection that is set up. 3147 * Second, the function tcp_isn_tick fires once per clock tick 3148 * and increments isn_offset as necessary so that sequence numbers 3149 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3150 * random positive increments serve only to ensure that the same 3151 * exact sequence number is never sent out twice (as could otherwise 3152 * happen when a port is recycled in less than the system tick 3153 * interval.) 3154 * 3155 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3156 * between seeding of isn_secret. This is normally set to zero, 3157 * as reseeding should not be necessary. 3158 * 3159 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3160 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3161 * general, this means holding an exclusive (write) lock. 3162 */ 3163 3164#define ISN_BYTES_PER_SECOND 1048576 3165#define ISN_STATIC_INCREMENT 4096 3166#define ISN_RANDOM_INCREMENT (4096 - 1) 3167#define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3168 3169VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3170VNET_DEFINE_STATIC(int, isn_last); 3171VNET_DEFINE_STATIC(int, isn_last_reseed); 3172VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3173VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3174 3175#define V_isn_secret VNET(isn_secret) 3176#define V_isn_last VNET(isn_last) 3177#define V_isn_last_reseed VNET(isn_last_reseed) 3178#define V_isn_offset VNET(isn_offset) 3179#define V_isn_offset_old VNET(isn_offset_old) 3180 3181tcp_seq 3182tcp_new_isn(struct in_conninfo *inc) 3183{ 3184 tcp_seq new_isn; 3185 u_int32_t projected_offset; 3186 3187 ISN_LOCK(); 3188 /* Seed if this is the first use, reseed if requested. */ 3189 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3190 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3191 < (u_int)ticks))) { 3192 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3193 V_isn_last_reseed = ticks; 3194 } 3195 3196 /* Compute the hash and return the ISN. */ 3197 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3198 sizeof(V_isn_secret)); 3199 V_isn_offset += ISN_STATIC_INCREMENT + 3200 (arc4random() & ISN_RANDOM_INCREMENT); 3201 if (ticks != V_isn_last) { 3202 projected_offset = V_isn_offset_old + 3203 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3204 if (SEQ_GT(projected_offset, V_isn_offset)) 3205 V_isn_offset = projected_offset; 3206 V_isn_offset_old = V_isn_offset; 3207 V_isn_last = ticks; 3208 } 3209 new_isn += V_isn_offset; 3210 ISN_UNLOCK(); 3211 return (new_isn); 3212} 3213 3214/* 3215 * When a specific ICMP unreachable message is received and the 3216 * connection state is SYN-SENT, drop the connection. This behavior 3217 * is controlled by the icmp_may_rst sysctl. 3218 */ 3219static struct inpcb * 3220tcp_drop_syn_sent(struct inpcb *inp, int errno) 3221{ 3222 struct tcpcb *tp; 3223 3224 NET_EPOCH_ASSERT(); 3225 INP_WLOCK_ASSERT(inp); 3226 3227 tp = intotcpcb(inp); 3228 if (tp->t_state != TCPS_SYN_SENT) 3229 return (inp); 3230 3231 if (tp->t_flags & TF_FASTOPEN) 3232 tcp_fastopen_disable_path(tp); 3233 3234 tp = tcp_drop(tp, errno); 3235 if (tp != NULL) 3236 return (inp); 3237 else 3238 return (NULL); 3239} 3240 3241/* 3242 * When `need fragmentation' ICMP is received, update our idea of the MSS 3243 * based on the new value. Also nudge TCP to send something, since we 3244 * know the packet we just sent was dropped. 3245 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3246 */ 3247static struct inpcb * 3248tcp_mtudisc_notify(struct inpcb *inp, int error) 3249{ 3250 3251 return (tcp_mtudisc(inp, -1)); 3252} 3253 3254static struct inpcb * 3255tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3256{ 3257 struct tcpcb *tp; 3258 struct socket *so; 3259 3260 INP_WLOCK_ASSERT(inp); 3261 3262 tp = intotcpcb(inp); 3263 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3264 3265 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3266 3267 so = inp->inp_socket; 3268 SOCKBUF_LOCK(&so->so_snd); 3269 /* If the mss is larger than the socket buffer, decrease the mss. */ 3270 if (so->so_snd.sb_hiwat < tp->t_maxseg) { 3271 tp->t_maxseg = so->so_snd.sb_hiwat; 3272 if (tp->t_maxseg < V_tcp_mssdflt) { 3273 /* 3274 * The MSS is so small we should not process incoming 3275 * SACK's since we are subject to attack in such a 3276 * case. 3277 */ 3278 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3279 } else { 3280 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3281 } 3282 } 3283 SOCKBUF_UNLOCK(&so->so_snd); 3284 3285 TCPSTAT_INC(tcps_mturesent); 3286 tp->t_rtttime = 0; 3287 tp->snd_nxt = tp->snd_una; 3288 tcp_free_sackholes(tp); 3289 tp->snd_recover = tp->snd_max; 3290 if (tp->t_flags & TF_SACK_PERMIT) 3291 EXIT_FASTRECOVERY(tp->t_flags); 3292 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3293 /* 3294 * Conceptually the snd_nxt setting 3295 * and freeing sack holes should 3296 * be done by the default stacks 3297 * own tfb_tcp_mtu_chg(). 3298 */ 3299 tp->t_fb->tfb_tcp_mtu_chg(tp); 3300 } 3301 if (tcp_output(tp) < 0) 3302 return (NULL); 3303 else 3304 return (inp); 3305} 3306 3307#ifdef INET 3308/* 3309 * Look-up the routing entry to the peer of this inpcb. If no route 3310 * is found and it cannot be allocated, then return 0. This routine 3311 * is called by TCP routines that access the rmx structure and by 3312 * tcp_mss_update to get the peer/interface MTU. 3313 */ 3314uint32_t 3315tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3316{ 3317 struct nhop_object *nh; 3318 struct ifnet *ifp; 3319 uint32_t maxmtu = 0; 3320 3321 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3322 3323 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3324 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3325 if (nh == NULL) 3326 return (0); 3327 3328 ifp = nh->nh_ifp; 3329 maxmtu = nh->nh_mtu; 3330 3331 /* Report additional interface capabilities. */ 3332 if (cap != NULL) { 3333 if (ifp->if_capenable & IFCAP_TSO4 && 3334 ifp->if_hwassist & CSUM_TSO) { 3335 cap->ifcap |= CSUM_TSO; 3336 cap->tsomax = ifp->if_hw_tsomax; 3337 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3338 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3339 } 3340 } 3341 } 3342 return (maxmtu); 3343} 3344#endif /* INET */ 3345 3346#ifdef INET6 3347uint32_t 3348tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3349{ 3350 struct nhop_object *nh; 3351 struct in6_addr dst6; 3352 uint32_t scopeid; 3353 struct ifnet *ifp; 3354 uint32_t maxmtu = 0; 3355 3356 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3357 3358 if (inc->inc_flags & INC_IPV6MINMTU) 3359 return (IPV6_MMTU); 3360 3361 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3362 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3363 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3364 if (nh == NULL) 3365 return (0); 3366 3367 ifp = nh->nh_ifp; 3368 maxmtu = nh->nh_mtu; 3369 3370 /* Report additional interface capabilities. */ 3371 if (cap != NULL) { 3372 if (ifp->if_capenable & IFCAP_TSO6 && 3373 ifp->if_hwassist & CSUM_TSO) { 3374 cap->ifcap |= CSUM_TSO; 3375 cap->tsomax = ifp->if_hw_tsomax; 3376 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3377 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3378 } 3379 } 3380 } 3381 3382 return (maxmtu); 3383} 3384 3385/* 3386 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3387 * 3388 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3389 * The right place to do that is ip6_setpktopt() that has just been 3390 * executed. By the way it just filled ip6po_minmtu for us. 3391 */ 3392void 3393tcp6_use_min_mtu(struct tcpcb *tp) 3394{ 3395 struct inpcb *inp = tptoinpcb(tp); 3396 3397 INP_WLOCK_ASSERT(inp); 3398 /* 3399 * In case of the IPV6_USE_MIN_MTU socket 3400 * option, the INC_IPV6MINMTU flag to announce 3401 * a corresponding MSS during the initial 3402 * handshake. If the TCP connection is not in 3403 * the front states, just reduce the MSS being 3404 * used. This avoids the sending of TCP 3405 * segments which will be fragmented at the 3406 * IPv6 layer. 3407 */ 3408 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3409 if ((tp->t_state >= TCPS_SYN_SENT) && 3410 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3411 struct ip6_pktopts *opt; 3412 3413 opt = inp->in6p_outputopts; 3414 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3415 tp->t_maxseg > TCP6_MSS) { 3416 tp->t_maxseg = TCP6_MSS; 3417 if (tp->t_maxseg < V_tcp_mssdflt) { 3418 /* 3419 * The MSS is so small we should not process incoming 3420 * SACK's since we are subject to attack in such a 3421 * case. 3422 */ 3423 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3424 } else { 3425 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3426 } 3427 } 3428 } 3429} 3430#endif /* INET6 */ 3431 3432/* 3433 * Calculate effective SMSS per RFC5681 definition for a given TCP 3434 * connection at its current state, taking into account SACK and etc. 3435 */ 3436u_int 3437tcp_maxseg(const struct tcpcb *tp) 3438{ 3439 u_int optlen; 3440 3441 if (tp->t_flags & TF_NOOPT) 3442 return (tp->t_maxseg); 3443 3444 /* 3445 * Here we have a simplified code from tcp_addoptions(), 3446 * without a proper loop, and having most of paddings hardcoded. 3447 * We might make mistakes with padding here in some edge cases, 3448 * but this is harmless, since result of tcp_maxseg() is used 3449 * only in cwnd and ssthresh estimations. 3450 */ 3451 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3452 if (tp->t_flags & TF_RCVD_TSTMP) 3453 optlen = TCPOLEN_TSTAMP_APPA; 3454 else 3455 optlen = 0; 3456#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3457 if (tp->t_flags & TF_SIGNATURE) 3458 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3459#endif 3460 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3461 optlen += TCPOLEN_SACKHDR; 3462 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3463 optlen = PADTCPOLEN(optlen); 3464 } 3465 } else { 3466 if (tp->t_flags & TF_REQ_TSTMP) 3467 optlen = TCPOLEN_TSTAMP_APPA; 3468 else 3469 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3470 if (tp->t_flags & TF_REQ_SCALE) 3471 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3472#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3473 if (tp->t_flags & TF_SIGNATURE) 3474 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3475#endif 3476 if (tp->t_flags & TF_SACK_PERMIT) 3477 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3478 } 3479 optlen = min(optlen, TCP_MAXOLEN); 3480 return (tp->t_maxseg - optlen); 3481} 3482 3483 3484u_int 3485tcp_fixed_maxseg(const struct tcpcb *tp) 3486{ 3487 int optlen; 3488 3489 if (tp->t_flags & TF_NOOPT) 3490 return (tp->t_maxseg); 3491 3492 /* 3493 * Here we have a simplified code from tcp_addoptions(), 3494 * without a proper loop, and having most of paddings hardcoded. 3495 * We only consider fixed options that we would send every 3496 * time I.e. SACK is not considered. This is important 3497 * for cc modules to figure out what the modulo of the 3498 * cwnd should be. 3499 */ 3500 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3501 if (tp->t_flags & TF_RCVD_TSTMP) 3502 optlen = TCPOLEN_TSTAMP_APPA; 3503 else 3504 optlen = 0; 3505#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3506 if (tp->t_flags & TF_SIGNATURE) 3507 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3508#endif 3509 } else { 3510 if (tp->t_flags & TF_REQ_TSTMP) 3511 optlen = TCPOLEN_TSTAMP_APPA; 3512 else 3513 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3514 if (tp->t_flags & TF_REQ_SCALE) 3515 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3516#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3517 if (tp->t_flags & TF_SIGNATURE) 3518 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3519#endif 3520 if (tp->t_flags & TF_SACK_PERMIT) 3521 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3522 } 3523 optlen = min(optlen, TCP_MAXOLEN); 3524 return (tp->t_maxseg - optlen); 3525} 3526 3527 3528 3529static int 3530sysctl_drop(SYSCTL_HANDLER_ARGS) 3531{ 3532 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3533 struct sockaddr_storage addrs[2]; 3534 struct inpcb *inp; 3535 struct tcpcb *tp; 3536#ifdef INET 3537 struct sockaddr_in *fin = NULL, *lin = NULL; 3538#endif 3539 struct epoch_tracker et; 3540#ifdef INET6 3541 struct sockaddr_in6 *fin6, *lin6; 3542#endif 3543 int error; 3544 3545 inp = NULL; 3546#ifdef INET6 3547 fin6 = lin6 = NULL; 3548#endif 3549 error = 0; 3550 3551 if (req->oldptr != NULL || req->oldlen != 0) 3552 return (EINVAL); 3553 if (req->newptr == NULL) 3554 return (EPERM); 3555 if (req->newlen < sizeof(addrs)) 3556 return (ENOMEM); 3557 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3558 if (error) 3559 return (error); 3560 3561 switch (addrs[0].ss_family) { 3562#ifdef INET6 3563 case AF_INET6: 3564 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3565 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3566 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3567 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3568 return (EINVAL); 3569 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3570 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3571 return (EINVAL); 3572 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3573 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3574#ifdef INET 3575 fin = (struct sockaddr_in *)&addrs[0]; 3576 lin = (struct sockaddr_in *)&addrs[1]; 3577#endif 3578 break; 3579 } 3580 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3581 if (error) 3582 return (error); 3583 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3584 if (error) 3585 return (error); 3586 break; 3587#endif 3588#ifdef INET 3589 case AF_INET: 3590 fin = (struct sockaddr_in *)&addrs[0]; 3591 lin = (struct sockaddr_in *)&addrs[1]; 3592 if (fin->sin_len != sizeof(struct sockaddr_in) || 3593 lin->sin_len != sizeof(struct sockaddr_in)) 3594 return (EINVAL); 3595 break; 3596#endif 3597 default: 3598 return (EINVAL); 3599 } 3600 NET_EPOCH_ENTER(et); 3601 switch (addrs[0].ss_family) { 3602#ifdef INET6 3603 case AF_INET6: 3604 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3605 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3606 INPLOOKUP_WLOCKPCB, NULL); 3607 break; 3608#endif 3609#ifdef INET 3610 case AF_INET: 3611 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3612 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3613 break; 3614#endif 3615 } 3616 if (inp != NULL) { 3617 if (!SOLISTENING(inp->inp_socket)) { 3618 tp = intotcpcb(inp); 3619 tp = tcp_drop(tp, ECONNABORTED); 3620 if (tp != NULL) 3621 INP_WUNLOCK(inp); 3622 } else 3623 INP_WUNLOCK(inp); 3624 } else 3625 error = ESRCH; 3626 NET_EPOCH_EXIT(et); 3627 return (error); 3628} 3629 3630SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3631 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3632 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3633 "Drop TCP connection"); 3634 3635static int 3636tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3637{ 3638 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3639 &tcp_ctloutput_set)); 3640} 3641 3642SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3643 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3644 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3645 "Set socket option for TCP endpoint"); 3646 3647#ifdef KERN_TLS 3648static int 3649sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3650{ 3651 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3652 struct sockaddr_storage addrs[2]; 3653 struct inpcb *inp; 3654#ifdef INET 3655 struct sockaddr_in *fin = NULL, *lin = NULL; 3656#endif 3657 struct epoch_tracker et; 3658#ifdef INET6 3659 struct sockaddr_in6 *fin6, *lin6; 3660#endif 3661 int error; 3662 3663 inp = NULL; 3664#ifdef INET6 3665 fin6 = lin6 = NULL; 3666#endif 3667 error = 0; 3668 3669 if (req->oldptr != NULL || req->oldlen != 0) 3670 return (EINVAL); 3671 if (req->newptr == NULL) 3672 return (EPERM); 3673 if (req->newlen < sizeof(addrs)) 3674 return (ENOMEM); 3675 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3676 if (error) 3677 return (error); 3678 3679 switch (addrs[0].ss_family) { 3680#ifdef INET6 3681 case AF_INET6: 3682 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3683 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3684 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3685 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3686 return (EINVAL); 3687 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3688 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3689 return (EINVAL); 3690 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3691 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3692#ifdef INET 3693 fin = (struct sockaddr_in *)&addrs[0]; 3694 lin = (struct sockaddr_in *)&addrs[1]; 3695#endif 3696 break; 3697 } 3698 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3699 if (error) 3700 return (error); 3701 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3702 if (error) 3703 return (error); 3704 break; 3705#endif 3706#ifdef INET 3707 case AF_INET: 3708 fin = (struct sockaddr_in *)&addrs[0]; 3709 lin = (struct sockaddr_in *)&addrs[1]; 3710 if (fin->sin_len != sizeof(struct sockaddr_in) || 3711 lin->sin_len != sizeof(struct sockaddr_in)) 3712 return (EINVAL); 3713 break; 3714#endif 3715 default: 3716 return (EINVAL); 3717 } 3718 NET_EPOCH_ENTER(et); 3719 switch (addrs[0].ss_family) { 3720#ifdef INET6 3721 case AF_INET6: 3722 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3723 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3724 INPLOOKUP_WLOCKPCB, NULL); 3725 break; 3726#endif 3727#ifdef INET 3728 case AF_INET: 3729 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3730 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3731 break; 3732#endif 3733 } 3734 NET_EPOCH_EXIT(et); 3735 if (inp != NULL) { 3736 struct socket *so; 3737 3738 so = inp->inp_socket; 3739 soref(so); 3740 error = ktls_set_tx_mode(so, 3741 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3742 INP_WUNLOCK(inp); 3743 sorele(so); 3744 } else 3745 error = ESRCH; 3746 return (error); 3747} 3748 3749SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3750 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3751 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3752 "Switch TCP connection to SW TLS"); 3753SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3754 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3755 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3756 "Switch TCP connection to ifnet TLS"); 3757#endif 3758 3759/* 3760 * Generate a standardized TCP log line for use throughout the 3761 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3762 * allow use in the interrupt context. 3763 * 3764 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3765 * NB: The function may return NULL if memory allocation failed. 3766 * 3767 * Due to header inclusion and ordering limitations the struct ip 3768 * and ip6_hdr pointers have to be passed as void pointers. 3769 */ 3770char * 3771tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3772 const void *ip6hdr) 3773{ 3774 3775 /* Is logging enabled? */ 3776 if (V_tcp_log_in_vain == 0) 3777 return (NULL); 3778 3779 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3780} 3781 3782char * 3783tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3784 const void *ip6hdr) 3785{ 3786 3787 /* Is logging enabled? */ 3788 if (tcp_log_debug == 0) 3789 return (NULL); 3790 3791 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3792} 3793 3794static char * 3795tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3796 const void *ip6hdr) 3797{ 3798 char *s, *sp; 3799 size_t size; 3800#ifdef INET 3801 const struct ip *ip = (const struct ip *)ip4hdr; 3802#endif 3803#ifdef INET6 3804 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3805#endif /* INET6 */ 3806 3807 /* 3808 * The log line looks like this: 3809 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3810 */ 3811 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3812 sizeof(PRINT_TH_FLAGS) + 1 + 3813#ifdef INET6 3814 2 * INET6_ADDRSTRLEN; 3815#else 3816 2 * INET_ADDRSTRLEN; 3817#endif /* INET6 */ 3818 3819 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3820 if (s == NULL) 3821 return (NULL); 3822 3823 strcat(s, "TCP: ["); 3824 sp = s + strlen(s); 3825 3826 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3827 inet_ntoa_r(inc->inc_faddr, sp); 3828 sp = s + strlen(s); 3829 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3830 sp = s + strlen(s); 3831 inet_ntoa_r(inc->inc_laddr, sp); 3832 sp = s + strlen(s); 3833 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3834#ifdef INET6 3835 } else if (inc) { 3836 ip6_sprintf(sp, &inc->inc6_faddr); 3837 sp = s + strlen(s); 3838 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3839 sp = s + strlen(s); 3840 ip6_sprintf(sp, &inc->inc6_laddr); 3841 sp = s + strlen(s); 3842 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3843 } else if (ip6 && th) { 3844 ip6_sprintf(sp, &ip6->ip6_src); 3845 sp = s + strlen(s); 3846 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3847 sp = s + strlen(s); 3848 ip6_sprintf(sp, &ip6->ip6_dst); 3849 sp = s + strlen(s); 3850 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3851#endif /* INET6 */ 3852#ifdef INET 3853 } else if (ip && th) { 3854 inet_ntoa_r(ip->ip_src, sp); 3855 sp = s + strlen(s); 3856 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3857 sp = s + strlen(s); 3858 inet_ntoa_r(ip->ip_dst, sp); 3859 sp = s + strlen(s); 3860 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3861#endif /* INET */ 3862 } else { 3863 free(s, M_TCPLOG); 3864 return (NULL); 3865 } 3866 sp = s + strlen(s); 3867 if (th) 3868 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3869 if (*(s + size - 1) != '\0') 3870 panic("%s: string too long", __func__); 3871 return (s); 3872} 3873 3874/* 3875 * A subroutine which makes it easy to track TCP state changes with DTrace. 3876 * This function shouldn't be called for t_state initializations that don't 3877 * correspond to actual TCP state transitions. 3878 */ 3879void 3880tcp_state_change(struct tcpcb *tp, int newstate) 3881{ 3882#if defined(KDTRACE_HOOKS) 3883 int pstate = tp->t_state; 3884#endif 3885 3886 TCPSTATES_DEC(tp->t_state); 3887 TCPSTATES_INC(newstate); 3888 tp->t_state = newstate; 3889 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3890} 3891 3892/* 3893 * Create an external-format (``xtcpcb'') structure using the information in 3894 * the kernel-format tcpcb structure pointed to by tp. This is done to 3895 * reduce the spew of irrelevant information over this interface, to isolate 3896 * user code from changes in the kernel structure, and potentially to provide 3897 * information-hiding if we decide that some of this information should be 3898 * hidden from users. 3899 */ 3900void 3901tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3902{ 3903 struct tcpcb *tp = intotcpcb(inp); 3904 sbintime_t now; 3905 3906 bzero(xt, sizeof(*xt)); 3907 xt->t_state = tp->t_state; 3908 xt->t_logstate = tcp_get_bblog_state(tp); 3909 xt->t_flags = tp->t_flags; 3910 xt->t_sndzerowin = tp->t_sndzerowin; 3911 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3912 xt->t_rcvoopack = tp->t_rcvoopack; 3913 xt->t_rcv_wnd = tp->rcv_wnd; 3914 xt->t_snd_wnd = tp->snd_wnd; 3915 xt->t_snd_cwnd = tp->snd_cwnd; 3916 xt->t_snd_ssthresh = tp->snd_ssthresh; 3917 xt->t_dsack_bytes = tp->t_dsack_bytes; 3918 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 3919 xt->t_dsack_pack = tp->t_dsack_pack; 3920 xt->t_maxseg = tp->t_maxseg; 3921 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3922 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3923 3924 now = getsbinuptime(); 3925#define COPYTIMER(which,where) do { \ 3926 if (tp->t_timers[which] != SBT_MAX) \ 3927 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 3928 else \ 3929 xt->where = 0; \ 3930} while (0) 3931 COPYTIMER(TT_DELACK, tt_delack); 3932 COPYTIMER(TT_REXMT, tt_rexmt); 3933 COPYTIMER(TT_PERSIST, tt_persist); 3934 COPYTIMER(TT_KEEP, tt_keep); 3935 COPYTIMER(TT_2MSL, tt_2msl); 3936#undef COPYTIMER 3937 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3938 3939 xt->xt_encaps_port = tp->t_port; 3940 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3941 TCP_FUNCTION_NAME_LEN_MAX); 3942 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 3943#ifdef TCP_BLACKBOX 3944 (void)tcp_log_get_id(tp, xt->xt_logid); 3945#endif 3946 3947 xt->xt_len = sizeof(struct xtcpcb); 3948 in_pcbtoxinpcb(inp, &xt->xt_inp); 3949} 3950 3951void 3952tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3953{ 3954 uint32_t bit, i; 3955 3956 if ((tp == NULL) || 3957 (status > TCP_EI_STATUS_MAX_VALUE) || 3958 (status == 0)) { 3959 /* Invalid */ 3960 return; 3961 } 3962 if (status > (sizeof(uint32_t) * 8)) { 3963 /* Should this be a KASSERT? */ 3964 return; 3965 } 3966 bit = 1U << (status - 1); 3967 if (bit & tp->t_end_info_status) { 3968 /* already logged */ 3969 return; 3970 } 3971 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3972 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3973 tp->t_end_info_bytes[i] = status; 3974 tp->t_end_info_status |= bit; 3975 break; 3976 } 3977 } 3978} 3979 3980int 3981tcp_can_enable_pacing(void) 3982{ 3983 3984 if ((tcp_pacing_limit == -1) || 3985 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 3986 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 3987 shadow_num_connections = number_of_tcp_connections_pacing; 3988 return (1); 3989 } else { 3990 counter_u64_add(tcp_pacing_failures, 1); 3991 return (0); 3992 } 3993} 3994 3995int 3996tcp_incr_dgp_pacing_cnt(void) 3997{ 3998 if ((tcp_dgp_limit == -1) || 3999 (tcp_dgp_limit > number_of_dgp_connections)) { 4000 atomic_fetchadd_int(&number_of_dgp_connections, 1); 4001 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4002 return (1); 4003 } else { 4004 counter_u64_add(tcp_dgp_failures, 1); 4005 return (0); 4006 } 4007} 4008 4009static uint8_t tcp_dgp_warning = 0; 4010 4011void 4012tcp_dec_dgp_pacing_cnt(void) 4013{ 4014 uint32_t ret; 4015 4016 ret = atomic_fetchadd_int(&number_of_dgp_connections, -1); 4017 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4018 KASSERT(ret != 0, ("number_of_dgp_connections -1 would cause wrap?")); 4019 if (ret == 0) { 4020 if (tcp_dgp_limit != -1) { 4021 printf("Warning all DGP is now disabled, count decrements invalidly!\n"); 4022 tcp_dgp_limit = 0; 4023 tcp_dgp_warning = 1; 4024 } else if (tcp_dgp_warning == 0) { 4025 printf("Warning DGP pacing is invalid, invalid decrement\n"); 4026 tcp_dgp_warning = 1; 4027 } 4028 } 4029 4030} 4031 4032static uint8_t tcp_pacing_warning = 0; 4033 4034void 4035tcp_decrement_paced_conn(void) 4036{ 4037 uint32_t ret; 4038 4039 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4040 shadow_num_connections = number_of_tcp_connections_pacing; 4041 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4042 if (ret == 0) { 4043 if (tcp_pacing_limit != -1) { 4044 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4045 tcp_pacing_limit = 0; 4046 } else if (tcp_pacing_warning == 0) { 4047 printf("Warning pacing count is invalid, invalid decrement\n"); 4048 tcp_pacing_warning = 1; 4049 } 4050 } 4051} 4052 4053static void 4054tcp_default_switch_failed(struct tcpcb *tp) 4055{ 4056 /* 4057 * If a switch fails we only need to 4058 * care about two things: 4059 * a) The t_flags2 4060 * and 4061 * b) The timer granularity. 4062 * Timeouts, at least for now, don't use the 4063 * old callout system in the other stacks so 4064 * those are hopefully safe. 4065 */ 4066 tcp_lro_features_off(tp); 4067 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 4068} 4069 4070#ifdef TCP_ACCOUNTING 4071int 4072tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss) 4073{ 4074 if (SEQ_LT(th->th_ack, tp->snd_una)) { 4075 /* Do we have a SACK? */ 4076 if (to->to_flags & TOF_SACK) { 4077 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4078 tp->tcp_cnt_counters[ACK_SACK]++; 4079 } 4080 return (ACK_SACK); 4081 } else { 4082 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4083 tp->tcp_cnt_counters[ACK_BEHIND]++; 4084 } 4085 return (ACK_BEHIND); 4086 } 4087 } else if (th->th_ack == tp->snd_una) { 4088 /* Do we have a SACK? */ 4089 if (to->to_flags & TOF_SACK) { 4090 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4091 tp->tcp_cnt_counters[ACK_SACK]++; 4092 } 4093 return (ACK_SACK); 4094 } else if (tiwin != tp->snd_wnd) { 4095 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4096 tp->tcp_cnt_counters[ACK_RWND]++; 4097 } 4098 return (ACK_RWND); 4099 } else { 4100 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4101 tp->tcp_cnt_counters[ACK_DUPACK]++; 4102 } 4103 return (ACK_DUPACK); 4104 } 4105 } else { 4106 if (!SEQ_GT(th->th_ack, tp->snd_max)) { 4107 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4108 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss); 4109 } 4110 } 4111 if (to->to_flags & TOF_SACK) { 4112 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4113 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++; 4114 } 4115 return (ACK_CUMACK_SACK); 4116 } else { 4117 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4118 tp->tcp_cnt_counters[ACK_CUMACK]++; 4119 } 4120 return (ACK_CUMACK); 4121 } 4122 } 4123} 4124#endif 4125 4126void 4127tcp_change_time_units(struct tcpcb *tp, int granularity) 4128{ 4129 if (tp->t_tmr_granularity == granularity) { 4130 /* We are there */ 4131 return; 4132 } 4133 if (granularity == TCP_TMR_GRANULARITY_USEC) { 4134 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS), 4135 ("Granularity is not TICKS its %u in tp:%p", 4136 tp->t_tmr_granularity, tp)); 4137 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow); 4138 if (tp->t_srtt > 1) { 4139 uint32_t val, frac; 4140 4141 val = tp->t_srtt >> TCP_RTT_SHIFT; 4142 frac = tp->t_srtt & 0x1f; 4143 tp->t_srtt = TICKS_2_USEC(val); 4144 /* 4145 * frac is the fractional part of the srtt (if any) 4146 * but its in ticks and every bit represents 4147 * 1/32nd of a hz. 4148 */ 4149 if (frac) { 4150 if (hz == 1000) { 4151 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4152 } else { 4153 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4154 } 4155 tp->t_srtt += frac; 4156 } 4157 } 4158 if (tp->t_rttvar) { 4159 uint32_t val, frac; 4160 4161 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT; 4162 frac = tp->t_rttvar & 0x1f; 4163 tp->t_rttvar = TICKS_2_USEC(val); 4164 /* 4165 * frac is the fractional part of the srtt (if any) 4166 * but its in ticks and every bit represents 4167 * 1/32nd of a hz. 4168 */ 4169 if (frac) { 4170 if (hz == 1000) { 4171 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4172 } else { 4173 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4174 } 4175 tp->t_rttvar += frac; 4176 } 4177 } 4178 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC; 4179 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) { 4180 /* Convert back to ticks, with */ 4181 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC), 4182 ("Granularity is not USEC its %u in tp:%p", 4183 tp->t_tmr_granularity, tp)); 4184 if (tp->t_srtt > 1) { 4185 uint32_t val, frac; 4186 4187 val = USEC_2_TICKS(tp->t_srtt); 4188 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4189 tp->t_srtt = val << TCP_RTT_SHIFT; 4190 /* 4191 * frac is the fractional part here is left 4192 * over from converting to hz and shifting. 4193 * We need to convert this to the 5 bit 4194 * remainder. 4195 */ 4196 if (frac) { 4197 if (hz == 1000) { 4198 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4199 } else { 4200 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4201 } 4202 tp->t_srtt += frac; 4203 } 4204 } 4205 if (tp->t_rttvar) { 4206 uint32_t val, frac; 4207 4208 val = USEC_2_TICKS(tp->t_rttvar); 4209 frac = tp->t_rttvar % (HPTS_USEC_IN_SEC / hz); 4210 tp->t_rttvar = val << TCP_RTTVAR_SHIFT; 4211 /* 4212 * frac is the fractional part here is left 4213 * over from converting to hz and shifting. 4214 * We need to convert this to the 4 bit 4215 * remainder. 4216 */ 4217 if (frac) { 4218 if (hz == 1000) { 4219 frac = (((uint64_t)frac * (uint64_t)TCP_RTTVAR_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4220 } else { 4221 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTTVAR_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4222 } 4223 tp->t_rttvar += frac; 4224 } 4225 } 4226 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow); 4227 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 4228 } 4229#ifdef INVARIANTS 4230 else { 4231 panic("Unknown granularity:%d tp:%p", 4232 granularity, tp); 4233 } 4234#endif 4235} 4236 4237void 4238tcp_handle_orphaned_packets(struct tcpcb *tp) 4239{ 4240 struct mbuf *save, *m, *prev; 4241 /* 4242 * Called when a stack switch is occuring from the fini() 4243 * of the old stack. We assue the init() as already been 4244 * run of the new stack and it has set the t_flags2 to 4245 * what it supports. This function will then deal with any 4246 * differences i.e. cleanup packets that maybe queued that 4247 * the newstack does not support. 4248 */ 4249 4250 if (tp->t_flags2 & TF2_MBUF_L_ACKS) 4251 return; 4252 if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 && 4253 !STAILQ_EMPTY(&tp->t_inqueue)) { 4254 /* 4255 * It is unsafe to process the packets since a 4256 * reset may be lurking in them (its rare but it 4257 * can occur). If we were to find a RST, then we 4258 * would end up dropping the connection and the 4259 * INP lock, so when we return the caller (tcp_usrreq) 4260 * will blow up when it trys to unlock the inp. 4261 * This new stack does not do any fancy LRO features 4262 * so all we can do is toss the packets. 4263 */ 4264 m = STAILQ_FIRST(&tp->t_inqueue); 4265 STAILQ_INIT(&tp->t_inqueue); 4266 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) 4267 m_freem(m); 4268 } else { 4269 /* 4270 * Here we have a stack that does mbuf queuing but 4271 * does not support compressed ack's. We must 4272 * walk all the mbufs and discard any compressed acks. 4273 */ 4274 STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) { 4275 if (m->m_flags & M_ACKCMP) { 4276 if (m == STAILQ_FIRST(&tp->t_inqueue)) 4277 STAILQ_REMOVE_HEAD(&tp->t_inqueue, 4278 m_stailqpkt); 4279 else 4280 STAILQ_REMOVE_AFTER(&tp->t_inqueue, 4281 prev, m_stailqpkt); 4282 m_freem(m); 4283 } else 4284 prev = m; 4285 } 4286 } 4287} 4288 4289#ifdef TCP_REQUEST_TRK 4290uint32_t 4291tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes) 4292{ 4293#ifdef KERN_TLS 4294 struct ktls_session *tls; 4295 uint32_t rec_oh, records; 4296 4297 tls = so->so_snd.sb_tls_info; 4298 if (tls == NULL) 4299 return (0); 4300 4301 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen; 4302 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len); 4303 return (records * rec_oh); 4304#else 4305 return (0); 4306#endif 4307} 4308 4309extern uint32_t tcp_stale_entry_time; 4310uint32_t tcp_stale_entry_time = 250000; 4311SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW, 4312 &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out"); 4313 4314void 4315tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req, 4316 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes) 4317{ 4318 if (tcp_bblogging_on(tp)) { 4319 union tcp_log_stackspecific log; 4320 struct timeval tv; 4321 4322 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 4323 log.u_bbr.inhpts = tcp_in_hpts(tp); 4324 log.u_bbr.flex8 = val; 4325 log.u_bbr.rttProp = req->timestamp; 4326 log.u_bbr.delRate = req->start; 4327 log.u_bbr.cur_del_rate = req->end; 4328 log.u_bbr.flex1 = req->start_seq; 4329 log.u_bbr.flex2 = req->end_seq; 4330 log.u_bbr.flex3 = req->flags; 4331 log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff); 4332 log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff); 4333 log.u_bbr.flex7 = slot; 4334 log.u_bbr.bw_inuse = offset; 4335 /* nbytes = flex6 | epoch */ 4336 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff); 4337 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff); 4338 /* cspr = lt_epoch | pkts_out */ 4339 log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff); 4340 log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff); 4341 log.u_bbr.applimited = tp->t_tcpreq_closed; 4342 log.u_bbr.applimited <<= 8; 4343 log.u_bbr.applimited |= tp->t_tcpreq_open; 4344 log.u_bbr.applimited <<= 8; 4345 log.u_bbr.applimited |= tp->t_tcpreq_req; 4346 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 4347 TCP_LOG_EVENTP(tp, NULL, 4348 &tptosocket(tp)->so_rcv, 4349 &tptosocket(tp)->so_snd, 4350 TCP_LOG_REQ_T, 0, 4351 0, &log, false, &tv); 4352 } 4353} 4354 4355void 4356tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent) 4357{ 4358 if (tp->t_tcpreq_req > 0) 4359 tp->t_tcpreq_req--; 4360 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4361 if (tp->t_tcpreq_open > 0) 4362 tp->t_tcpreq_open--; 4363 } else { 4364 if (tp->t_tcpreq_closed > 0) 4365 tp->t_tcpreq_closed--; 4366 } 4367 ent->flags = TCP_TRK_TRACK_FLG_EMPTY; 4368} 4369 4370static void 4371tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest) 4372{ 4373 struct tcp_sendfile_track *ent; 4374 uint64_t time_delta, oldest_delta; 4375 int i, oldest, oldest_set = 0, cnt_rm = 0; 4376 4377 for(i = 0; i < MAX_TCP_TRK_REQ; i++) { 4378 ent = &tp->t_tcpreq_info[i]; 4379 if (ent->flags != TCP_TRK_TRACK_FLG_USED) { 4380 /* 4381 * We only care about closed end ranges 4382 * that are allocated and have no sendfile 4383 * ever touching them. They would be in 4384 * state USED. 4385 */ 4386 continue; 4387 } 4388 if (ts >= ent->localtime) 4389 time_delta = ts - ent->localtime; 4390 else 4391 time_delta = 0; 4392 if (time_delta && 4393 ((oldest_delta < time_delta) || (oldest_set == 0))) { 4394 oldest_set = 1; 4395 oldest = i; 4396 oldest_delta = time_delta; 4397 } 4398 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) { 4399 /* 4400 * No sendfile in a our time-limit 4401 * time to purge it. 4402 */ 4403 cnt_rm++; 4404 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4405 time_delta, 0); 4406 tcp_req_free_a_slot(tp, ent); 4407 } 4408 } 4409 if ((cnt_rm == 0) && rm_oldest && oldest_set) { 4410 ent = &tp->t_tcpreq_info[oldest]; 4411 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4412 oldest_delta, 1); 4413 tcp_req_free_a_slot(tp, ent); 4414 } 4415} 4416 4417int 4418tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point) 4419{ 4420 int i, ret=0; 4421 struct tcp_sendfile_track *ent; 4422 4423 /* Clean up any old closed end requests that are now completed */ 4424 if (tp->t_tcpreq_req == 0) 4425 return(0); 4426 if (tp->t_tcpreq_closed == 0) 4427 return(0); 4428 for(i = 0; i < MAX_TCP_TRK_REQ; i++) { 4429 ent = &tp->t_tcpreq_info[i]; 4430 /* Skip empty ones */ 4431 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4432 continue; 4433 /* Skip open ones */ 4434 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) 4435 continue; 4436 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4437 /* We are past it -- free it */ 4438 tcp_req_log_req_info(tp, ent, 4439 i, TCP_TRK_REQ_LOG_FREED, 0, 0); 4440 tcp_req_free_a_slot(tp, ent); 4441 ret++; 4442 } 4443 } 4444 return (ret); 4445} 4446 4447int 4448tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point) 4449{ 4450 if (tp->t_tcpreq_req == 0) 4451 return(-1); 4452 if (tp->t_tcpreq_closed == 0) 4453 return(-1); 4454 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4455 return(-1); 4456 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4457 return (1); 4458 } 4459 return (0); 4460} 4461 4462struct tcp_sendfile_track * 4463tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip) 4464{ 4465 /* 4466 * Given an ack point (th_ack) walk through our entries and 4467 * return the first one found that th_ack goes past the 4468 * end_seq. 4469 */ 4470 struct tcp_sendfile_track *ent; 4471 int i; 4472 4473 if (tp->t_tcpreq_req == 0) { 4474 /* none open */ 4475 return (NULL); 4476 } 4477 for(i = 0; i < MAX_TCP_TRK_REQ; i++) { 4478 ent = &tp->t_tcpreq_info[i]; 4479 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4480 continue; 4481 if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) { 4482 if (SEQ_GEQ(th_ack, ent->end_seq)) { 4483 *ip = i; 4484 return (ent); 4485 } 4486 } 4487 } 4488 return (NULL); 4489} 4490 4491struct tcp_sendfile_track * 4492tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq) 4493{ 4494 struct tcp_sendfile_track *ent; 4495 int i; 4496 4497 if (tp->t_tcpreq_req == 0) { 4498 /* none open */ 4499 return (NULL); 4500 } 4501 for(i = 0; i < MAX_TCP_TRK_REQ; i++) { 4502 ent = &tp->t_tcpreq_info[i]; 4503 tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH, 4504 (uint64_t)seq, 0); 4505 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4506 continue; 4507 } 4508 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4509 /* 4510 * An open end request only needs to 4511 * match the beginning seq or be 4512 * all we have (once we keep going on 4513 * a open end request we may have a seq 4514 * wrap). 4515 */ 4516 if ((SEQ_GEQ(seq, ent->start_seq)) || 4517 (tp->t_tcpreq_closed == 0)) 4518 return (ent); 4519 } else { 4520 /* 4521 * For this one we need to 4522 * be a bit more careful if its 4523 * completed at least. 4524 */ 4525 if ((SEQ_GEQ(seq, ent->start_seq)) && 4526 (SEQ_LT(seq, ent->end_seq))) { 4527 return (ent); 4528 } 4529 } 4530 } 4531 return (NULL); 4532} 4533 4534/* Should this be in its own file tcp_req.c ? */ 4535struct tcp_sendfile_track * 4536tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups) 4537{ 4538 struct tcp_sendfile_track *fil; 4539 int i, allocated; 4540 4541 /* In case the stack does not check for completions do so now */ 4542 tcp_req_check_for_comp(tp, tp->snd_una); 4543 /* Check for stale entries */ 4544 if (tp->t_tcpreq_req) 4545 tcp_req_check_for_stale_entries(tp, ts, 4546 (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ)); 4547 /* Check to see if this is a duplicate of one not started */ 4548 if (tp->t_tcpreq_req) { 4549 for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4550 fil = &tp->t_tcpreq_info[i]; 4551 if ((fil->flags & TCP_TRK_TRACK_FLG_USED) == 0) 4552 continue; 4553 if ((fil->timestamp == req->timestamp) && 4554 (fil->start == req->start) && 4555 ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) || 4556 (fil->end == req->end))) { 4557 /* 4558 * We already have this request 4559 * and it has not been started with sendfile. 4560 * This probably means the user was returned 4561 * a 4xx of some sort and its going to age 4562 * out, lets not duplicate it. 4563 */ 4564 return(fil); 4565 } 4566 } 4567 } 4568 /* Ok if there is no room at the inn we are in trouble */ 4569 if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) { 4570 tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL); 4571 for(i = 0; i < MAX_TCP_TRK_REQ; i++) { 4572 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], 4573 i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0); 4574 } 4575 return (NULL); 4576 } 4577 for(i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4578 fil = &tp->t_tcpreq_info[i]; 4579 if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4580 allocated = 1; 4581 fil->flags = TCP_TRK_TRACK_FLG_USED; 4582 fil->timestamp = req->timestamp; 4583 fil->playout_ms = req->playout_ms; 4584 fil->localtime = ts; 4585 fil->start = req->start; 4586 if (req->flags & TCP_LOG_HTTPD_RANGE_END) { 4587 fil->end = req->end; 4588 } else { 4589 fil->end = 0; 4590 fil->flags |= TCP_TRK_TRACK_FLG_OPEN; 4591 } 4592 /* 4593 * We can set the min boundaries to the TCP Sequence space, 4594 * but it might be found to be further up when sendfile 4595 * actually runs on this range (if it ever does). 4596 */ 4597 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc; 4598 fil->start_seq = tp->snd_una + 4599 tptosocket(tp)->so_snd.sb_ccc; 4600 if (req->flags & TCP_LOG_HTTPD_RANGE_END) 4601 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start))); 4602 else 4603 fil->end_seq = 0; 4604 if (tptosocket(tp)->so_snd.sb_tls_info) { 4605 /* 4606 * This session is doing TLS. Take a swag guess 4607 * at the overhead. 4608 */ 4609 fil->end_seq += tcp_estimate_tls_overhead( 4610 tptosocket(tp), (fil->end - fil->start)); 4611 } 4612 tp->t_tcpreq_req++; 4613 if (fil->flags & TCP_TRK_TRACK_FLG_OPEN) 4614 tp->t_tcpreq_open++; 4615 else 4616 tp->t_tcpreq_closed++; 4617 tcp_req_log_req_info(tp, fil, i, 4618 TCP_TRK_REQ_LOG_NEW, 0, 0); 4619 break; 4620 } else 4621 fil = NULL; 4622 } 4623 return (fil); 4624} 4625 4626void 4627tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts) 4628{ 4629 (void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1); 4630} 4631#endif 4632 4633void 4634tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err) 4635{ 4636 if (tcp_bblogging_on(tp)) { 4637 struct tcp_log_buffer *l; 4638 4639 l = tcp_log_event(tp, NULL, 4640 &tptosocket(tp)->so_rcv, 4641 &tptosocket(tp)->so_snd, 4642 TCP_LOG_SOCKET_OPT, 4643 err, 0, NULL, 1, 4644 NULL, NULL, 0, NULL); 4645 if (l) { 4646 l->tlb_flex1 = option_num; 4647 l->tlb_flex2 = option_val; 4648 } 4649 } 4650} 4651 4652uint32_t 4653tcp_get_srtt(struct tcpcb *tp, int granularity) 4654{ 4655 uint32_t srtt; 4656 4657 KASSERT(granularity == TCP_TMR_GRANULARITY_USEC || 4658 granularity == TCP_TMR_GRANULARITY_TICKS, 4659 ("%s: called with unexpected granularity %d", __func__, 4660 granularity)); 4661 4662 srtt = tp->t_srtt; 4663 4664 /* 4665 * We only support two granularities. If the stored granularity 4666 * does not match the granularity requested by the caller, 4667 * convert the stored value to the requested unit of granularity. 4668 */ 4669 if (tp->t_tmr_granularity != granularity) { 4670 if (granularity == TCP_TMR_GRANULARITY_USEC) 4671 srtt = TICKS_2_USEC(srtt); 4672 else 4673 srtt = USEC_2_TICKS(srtt); 4674 } 4675 4676 /* 4677 * If the srtt is stored with ticks granularity, we need to 4678 * unshift to get the actual value. We do this after the 4679 * conversion above (if one was necessary) in order to maximize 4680 * precision. 4681 */ 4682 if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS) 4683 srtt = srtt >> TCP_RTT_SHIFT; 4684 4685 return (srtt); 4686} 4687 4688void 4689tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt, 4690 uint8_t is_tlp, bool hw_tls) 4691{ 4692 4693 if (is_tlp) { 4694 tp->t_sndtlppack++; 4695 tp->t_sndtlpbyte += len; 4696 } 4697 /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */ 4698 if (is_rxt) 4699 tp->t_snd_rxt_bytes += len; 4700 else 4701 tp->t_sndbytes += len; 4702 4703#ifdef KERN_TLS 4704 if (hw_tls && is_rxt && len != 0) { 4705 uint64_t rexmit_percent; 4706 4707 rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) / 4708 (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes)); 4709 if (rexmit_percent > ktls_ifnet_max_rexmit_pct) 4710 ktls_disable_ifnet(tp); 4711 } 4712#endif 4713} 4714