1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/arb.h> 45 #include <sys/callout.h> 46 #include <sys/eventhandler.h> 47 #ifdef TCP_HHOOK 48 #include <sys/hhook.h> 49 #endif 50 #include <sys/kernel.h> 51 #ifdef TCP_HHOOK 52 #include <sys/khelp.h> 53 #endif 54 #ifdef KERN_TLS 55 #include <sys/ktls.h> 56 #endif 57 #include <sys/qmath.h> 58 #include <sys/stats.h> 59 #include <sys/sysctl.h> 60 #include <sys/jail.h> 61 #include <sys/malloc.h> 62 #include <sys/refcount.h> 63 #include <sys/mbuf.h> 64 #include <sys/priv.h> 65 #include <sys/proc.h> 66 #include <sys/sdt.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/protosw.h> 70 #include <sys/random.h> 71 72 #include <vm/uma.h> 73 74 #include <net/route.h> 75 #include <net/route/nhop.h> 76 #include <net/if.h> 77 #include <net/if_var.h> 78 #include <net/if_private.h> 79 #include <net/vnet.h> 80 81 #include <netinet/in.h> 82 #include <netinet/in_fib.h> 83 #include <netinet/in_kdtrace.h> 84 #include <netinet/in_pcb.h> 85 #include <netinet/in_systm.h> 86 #include <netinet/in_var.h> 87 #include <netinet/ip.h> 88 #include <netinet/ip_icmp.h> 89 #include <netinet/ip_var.h> 90 #ifdef INET6 91 #include <netinet/icmp6.h> 92 #include <netinet/ip6.h> 93 #include <netinet6/in6_fib.h> 94 #include <netinet6/in6_pcb.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/scope6_var.h> 97 #include <netinet6/nd6.h> 98 #endif 99 100 #include <netinet/tcp.h> 101 #ifdef INVARIANTS 102 #define TCPSTATES 103 #endif 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet/tcp_ecn.h> 109 #include <netinet/tcp_log_buf.h> 110 #include <netinet/tcp_syncache.h> 111 #include <netinet/tcp_hpts.h> 112 #include <netinet/tcp_lro.h> 113 #include <netinet/cc/cc.h> 114 #include <netinet/tcpip.h> 115 #include <netinet/tcp_fastopen.h> 116 #include <netinet/tcp_accounting.h> 117 #ifdef TCPPCAP 118 #include <netinet/tcp_pcap.h> 119 #endif 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 #include <netinet/udp.h> 124 #include <netinet/udp_var.h> 125 #ifdef INET6 126 #include <netinet6/tcp6_var.h> 127 #endif 128 129 #include <netipsec/ipsec_support.h> 130 131 #include <machine/in_cksum.h> 132 #include <crypto/siphash/siphash.h> 133 134 #include <security/mac/mac_framework.h> 135 136 #ifdef INET6 137 static ip6proto_ctlinput_t tcp6_ctlinput; 138 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 139 #endif 140 141 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 142 #ifdef INET6 143 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 144 #endif 145 146 #ifdef NETFLIX_EXP_DETECTION 147 /* Sack attack detection thresholds and such */ 148 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, 149 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 150 "Sack Attack detection thresholds"); 151 int32_t tcp_force_detection = 0; 152 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 153 CTLFLAG_RW, 154 &tcp_force_detection, 0, 155 "Do we force detection even if the INP has it off?"); 156 int32_t tcp_sad_limit = 10000; 157 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, limit, 158 CTLFLAG_RW, 159 &tcp_sad_limit, 10000, 160 "If SaD is enabled, what is the limit to sendmap entries (0 = unlimited)?"); 161 int32_t tcp_sad_limit = 10000; 162 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, limit, 163 CTLFLAG_RW, 164 &tcp_sad_limit, 10000, 165 "If SaD is enabled, what is the limit to sendmap entries (0 = unlimited)?"); 166 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 167 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 168 CTLFLAG_RW, 169 &tcp_sack_to_ack_thresh, 700, 170 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 171 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 172 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 173 CTLFLAG_RW, 174 &tcp_sack_to_move_thresh, 600, 175 "Percentage of sack moves we must see above (10.1 percent is 101)"); 176 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 177 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 178 CTLFLAG_RW, 179 &tcp_restoral_thresh, 550, 180 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 181 int32_t tcp_sad_decay_val = 800; 182 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 183 CTLFLAG_RW, 184 &tcp_sad_decay_val, 800, 185 "The decay percentage (10.1 percent equals 101 )"); 186 int32_t tcp_map_minimum = 500; 187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 188 CTLFLAG_RW, 189 &tcp_map_minimum, 500, 190 "Number of Map enteries before we start detection"); 191 int32_t tcp_sad_pacing_interval = 2000; 192 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 193 CTLFLAG_RW, 194 &tcp_sad_pacing_interval, 2000, 195 "What is the minimum pacing interval for a classified attacker?"); 196 197 int32_t tcp_sad_low_pps = 100; 198 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 199 CTLFLAG_RW, 200 &tcp_sad_low_pps, 100, 201 "What is the input pps that below which we do not decay?"); 202 #endif 203 uint32_t tcp_ack_war_time_window = 1000; 204 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 205 CTLFLAG_RW, 206 &tcp_ack_war_time_window, 1000, 207 "If the tcp_stack does ack-war prevention how many milliseconds are in its time window?"); 208 uint32_t tcp_ack_war_cnt = 5; 209 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, 210 CTLFLAG_RW, 211 &tcp_ack_war_cnt, 5, 212 "If the tcp_stack does ack-war prevention how many acks can be sent in its time window?"); 213 214 struct rwlock tcp_function_lock; 215 216 static int 217 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 218 { 219 int error, new; 220 221 new = V_tcp_mssdflt; 222 error = sysctl_handle_int(oidp, &new, 0, req); 223 if (error == 0 && req->newptr) { 224 if (new < TCP_MINMSS) 225 error = EINVAL; 226 else 227 V_tcp_mssdflt = new; 228 } 229 return (error); 230 } 231 232 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 233 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 234 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 235 "Default TCP Maximum Segment Size"); 236 237 #ifdef INET6 238 static int 239 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 240 { 241 int error, new; 242 243 new = V_tcp_v6mssdflt; 244 error = sysctl_handle_int(oidp, &new, 0, req); 245 if (error == 0 && req->newptr) { 246 if (new < TCP_MINMSS) 247 error = EINVAL; 248 else 249 V_tcp_v6mssdflt = new; 250 } 251 return (error); 252 } 253 254 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 255 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 256 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 257 "Default TCP Maximum Segment Size for IPv6"); 258 #endif /* INET6 */ 259 260 /* 261 * Minimum MSS we accept and use. This prevents DoS attacks where 262 * we are forced to a ridiculous low MSS like 20 and send hundreds 263 * of packets instead of one. The effect scales with the available 264 * bandwidth and quickly saturates the CPU and network interface 265 * with packet generation and sending. Set to zero to disable MINMSS 266 * checking. This setting prevents us from sending too small packets. 267 */ 268 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 269 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 270 &VNET_NAME(tcp_minmss), 0, 271 "Minimum TCP Maximum Segment Size"); 272 273 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 274 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 275 &VNET_NAME(tcp_do_rfc1323), 0, 276 "Enable rfc1323 (high performance TCP) extensions"); 277 278 /* 279 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 280 * Some stacks negotiate TS, but never send them after connection setup. Some 281 * stacks negotiate TS, but don't send them when sending keep-alive segments. 282 * These include modern widely deployed TCP stacks. 283 * Therefore tolerating violations for now... 284 */ 285 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 286 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 287 &VNET_NAME(tcp_tolerate_missing_ts), 0, 288 "Tolerate missing TCP timestamps"); 289 290 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 291 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 292 &VNET_NAME(tcp_ts_offset_per_conn), 0, 293 "Initialize TCP timestamps per connection instead of per host pair"); 294 295 /* How many connections are pacing */ 296 static volatile uint32_t number_of_tcp_connections_pacing = 0; 297 static uint32_t shadow_num_connections = 0; 298 299 static int tcp_pacing_limit = 10000; 300 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 301 &tcp_pacing_limit, 1000, 302 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 303 304 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 305 &shadow_num_connections, 0, "Number of TCP connections being paced"); 306 307 static int tcp_log_debug = 0; 308 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 309 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 310 311 static int tcp_tcbhashsize; 312 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 313 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 314 315 static int do_tcpdrain = 1; 316 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 317 "Enable tcp_drain routine for extra help when low on mbufs"); 318 319 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 320 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 321 322 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 323 #define V_icmp_may_rst VNET(icmp_may_rst) 324 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 325 &VNET_NAME(icmp_may_rst), 0, 326 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 327 328 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 329 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 330 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 331 &VNET_NAME(tcp_isn_reseed_interval), 0, 332 "Seconds between reseeding of ISN secret"); 333 334 static int tcp_soreceive_stream; 335 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 336 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 337 338 VNET_DEFINE(uma_zone_t, sack_hole_zone); 339 #define V_sack_hole_zone VNET(sack_hole_zone) 340 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 341 static int 342 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 343 { 344 int error; 345 uint32_t new; 346 347 new = V_tcp_map_entries_limit; 348 error = sysctl_handle_int(oidp, &new, 0, req); 349 if (error == 0 && req->newptr) { 350 /* only allow "0" and value > minimum */ 351 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 352 error = EINVAL; 353 else 354 V_tcp_map_entries_limit = new; 355 } 356 return (error); 357 } 358 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 359 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 360 &VNET_NAME(tcp_map_entries_limit), 0, 361 &sysctl_net_inet_tcp_map_limit_check, "IU", 362 "Total sendmap entries limit"); 363 364 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 365 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 366 &VNET_NAME(tcp_map_split_limit), 0, 367 "Total sendmap split entries limit"); 368 369 #ifdef TCP_HHOOK 370 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 371 #endif 372 373 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 374 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 375 #define V_ts_offset_secret VNET(ts_offset_secret) 376 377 static int tcp_default_fb_init(struct tcpcb *tp, void **ptr); 378 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 379 static int tcp_default_handoff_ok(struct tcpcb *tp); 380 static struct inpcb *tcp_notify(struct inpcb *, int); 381 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 382 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 383 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 384 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 385 const void *ip4hdr, const void *ip6hdr); 386 static ipproto_ctlinput_t tcp_ctlinput; 387 static udp_tun_icmp_t tcp_ctlinput_viaudp; 388 389 static struct tcp_function_block tcp_def_funcblk = { 390 .tfb_tcp_block_name = "freebsd", 391 .tfb_tcp_output = tcp_default_output, 392 .tfb_tcp_do_segment = tcp_do_segment, 393 .tfb_tcp_ctloutput = tcp_default_ctloutput, 394 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 395 .tfb_tcp_fb_init = tcp_default_fb_init, 396 .tfb_tcp_fb_fini = tcp_default_fb_fini, 397 }; 398 399 static int tcp_fb_cnt = 0; 400 struct tcp_funchead t_functions; 401 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 402 403 void 404 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 405 { 406 TCPSTAT_INC(tcps_dsack_count); 407 tp->t_dsack_pack++; 408 if (tlp == 0) { 409 if (SEQ_GT(end, start)) { 410 tp->t_dsack_bytes += (end - start); 411 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 412 } else { 413 tp->t_dsack_tlp_bytes += (start - end); 414 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 415 } 416 } else { 417 if (SEQ_GT(end, start)) { 418 tp->t_dsack_bytes += (end - start); 419 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 420 } else { 421 tp->t_dsack_tlp_bytes += (start - end); 422 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 423 } 424 } 425 } 426 427 static struct tcp_function_block * 428 find_tcp_functions_locked(struct tcp_function_set *fs) 429 { 430 struct tcp_function *f; 431 struct tcp_function_block *blk=NULL; 432 433 TAILQ_FOREACH(f, &t_functions, tf_next) { 434 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 435 blk = f->tf_fb; 436 break; 437 } 438 } 439 return(blk); 440 } 441 442 static struct tcp_function_block * 443 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 444 { 445 struct tcp_function_block *rblk=NULL; 446 struct tcp_function *f; 447 448 TAILQ_FOREACH(f, &t_functions, tf_next) { 449 if (f->tf_fb == blk) { 450 rblk = blk; 451 if (s) { 452 *s = f; 453 } 454 break; 455 } 456 } 457 return (rblk); 458 } 459 460 struct tcp_function_block * 461 find_and_ref_tcp_functions(struct tcp_function_set *fs) 462 { 463 struct tcp_function_block *blk; 464 465 rw_rlock(&tcp_function_lock); 466 blk = find_tcp_functions_locked(fs); 467 if (blk) 468 refcount_acquire(&blk->tfb_refcnt); 469 rw_runlock(&tcp_function_lock); 470 return(blk); 471 } 472 473 struct tcp_function_block * 474 find_and_ref_tcp_fb(struct tcp_function_block *blk) 475 { 476 struct tcp_function_block *rblk; 477 478 rw_rlock(&tcp_function_lock); 479 rblk = find_tcp_fb_locked(blk, NULL); 480 if (rblk) 481 refcount_acquire(&rblk->tfb_refcnt); 482 rw_runlock(&tcp_function_lock); 483 return(rblk); 484 } 485 486 /* Find a matching alias for the given tcp_function_block. */ 487 int 488 find_tcp_function_alias(struct tcp_function_block *blk, 489 struct tcp_function_set *fs) 490 { 491 struct tcp_function *f; 492 int found; 493 494 found = 0; 495 rw_rlock(&tcp_function_lock); 496 TAILQ_FOREACH(f, &t_functions, tf_next) { 497 if ((f->tf_fb == blk) && 498 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 499 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 500 /* Matching function block with different name. */ 501 strncpy(fs->function_set_name, f->tf_name, 502 TCP_FUNCTION_NAME_LEN_MAX); 503 found = 1; 504 break; 505 } 506 } 507 /* Null terminate the string appropriately. */ 508 if (found) { 509 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 510 } else { 511 fs->function_set_name[0] = '\0'; 512 } 513 rw_runlock(&tcp_function_lock); 514 return (found); 515 } 516 517 static struct tcp_function_block * 518 find_and_ref_tcp_default_fb(void) 519 { 520 struct tcp_function_block *rblk; 521 522 rw_rlock(&tcp_function_lock); 523 rblk = tcp_func_set_ptr; 524 refcount_acquire(&rblk->tfb_refcnt); 525 rw_runlock(&tcp_function_lock); 526 return (rblk); 527 } 528 529 void 530 tcp_switch_back_to_default(struct tcpcb *tp) 531 { 532 struct tcp_function_block *tfb; 533 void *ptr = NULL; 534 535 KASSERT(tp->t_fb != &tcp_def_funcblk, 536 ("%s: called by the built-in default stack", __func__)); 537 538 /* 539 * Now, we'll find a new function block to use. 540 * Start by trying the current user-selected 541 * default, unless this stack is the user-selected 542 * default. 543 */ 544 tfb = find_and_ref_tcp_default_fb(); 545 if (tfb == tp->t_fb) { 546 refcount_release(&tfb->tfb_refcnt); 547 tfb = NULL; 548 } 549 /* Does the stack accept this connection? */ 550 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 551 (*tfb->tfb_tcp_handoff_ok)(tp)) { 552 refcount_release(&tfb->tfb_refcnt); 553 tfb = NULL; 554 } 555 /* Try to use that stack. */ 556 if (tfb != NULL) { 557 /* Initialize the new stack. If it succeeds, we are done. */ 558 if (tfb->tfb_tcp_fb_init == NULL || 559 (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) { 560 /* Release the old stack */ 561 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 562 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 563 refcount_release(&tp->t_fb->tfb_refcnt); 564 /* Now set in all the pointers */ 565 tp->t_fb = tfb; 566 tp->t_fb_ptr = ptr; 567 return; 568 } 569 /* 570 * Initialization failed. Release the reference count on 571 * the looked up default stack. 572 */ 573 refcount_release(&tfb->tfb_refcnt); 574 } 575 576 /* 577 * If that wasn't feasible, use the built-in default 578 * stack which is not allowed to reject anyone. 579 */ 580 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 581 if (tfb == NULL) { 582 /* there always should be a default */ 583 panic("Can't refer to tcp_def_funcblk"); 584 } 585 if (tfb->tfb_tcp_handoff_ok != NULL) { 586 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 587 /* The default stack cannot say no */ 588 panic("Default stack rejects a new session?"); 589 } 590 } 591 if (tfb->tfb_tcp_fb_init != NULL && 592 (*tfb->tfb_tcp_fb_init)(tp, &ptr)) { 593 /* The default stack cannot fail */ 594 panic("Default stack initialization failed"); 595 } 596 /* Now release the old stack */ 597 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 598 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 599 refcount_release(&tp->t_fb->tfb_refcnt); 600 /* And set in the pointers to the new */ 601 tp->t_fb = tfb; 602 tp->t_fb_ptr = ptr; 603 } 604 605 static bool 606 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 607 const struct sockaddr *sa, void *ctx) 608 { 609 struct ip *iph; 610 #ifdef INET6 611 struct ip6_hdr *ip6; 612 #endif 613 struct udphdr *uh; 614 struct tcphdr *th; 615 int thlen; 616 uint16_t port; 617 618 TCPSTAT_INC(tcps_tunneled_pkts); 619 if ((m->m_flags & M_PKTHDR) == 0) { 620 /* Can't handle one that is not a pkt hdr */ 621 TCPSTAT_INC(tcps_tunneled_errs); 622 goto out; 623 } 624 thlen = sizeof(struct tcphdr); 625 if (m->m_len < off + sizeof(struct udphdr) + thlen && 626 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 627 TCPSTAT_INC(tcps_tunneled_errs); 628 goto out; 629 } 630 iph = mtod(m, struct ip *); 631 uh = (struct udphdr *)((caddr_t)iph + off); 632 th = (struct tcphdr *)(uh + 1); 633 thlen = th->th_off << 2; 634 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 635 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 636 if (m == NULL) { 637 TCPSTAT_INC(tcps_tunneled_errs); 638 goto out; 639 } else { 640 iph = mtod(m, struct ip *); 641 uh = (struct udphdr *)((caddr_t)iph + off); 642 th = (struct tcphdr *)(uh + 1); 643 } 644 } 645 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 646 bcopy(th, uh, m->m_len - off); 647 m->m_len -= sizeof(struct udphdr); 648 m->m_pkthdr.len -= sizeof(struct udphdr); 649 /* 650 * We use the same algorithm for 651 * both UDP and TCP for c-sum. So 652 * the code in tcp_input will skip 653 * the checksum. So we do nothing 654 * with the flag (m->m_pkthdr.csum_flags). 655 */ 656 switch (iph->ip_v) { 657 #ifdef INET 658 case IPVERSION: 659 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 660 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 661 break; 662 #endif 663 #ifdef INET6 664 case IPV6_VERSION >> 4: 665 ip6 = mtod(m, struct ip6_hdr *); 666 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 667 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 668 break; 669 #endif 670 default: 671 goto out; 672 break; 673 } 674 return (true); 675 out: 676 m_freem(m); 677 678 return (true); 679 } 680 681 static int 682 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 683 { 684 int error=ENOENT; 685 struct tcp_function_set fs; 686 struct tcp_function_block *blk; 687 688 memset(&fs, 0, sizeof(fs)); 689 rw_rlock(&tcp_function_lock); 690 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 691 if (blk) { 692 /* Found him */ 693 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 694 fs.pcbcnt = blk->tfb_refcnt; 695 } 696 rw_runlock(&tcp_function_lock); 697 error = sysctl_handle_string(oidp, fs.function_set_name, 698 sizeof(fs.function_set_name), req); 699 700 /* Check for error or no change */ 701 if (error != 0 || req->newptr == NULL) 702 return(error); 703 704 rw_wlock(&tcp_function_lock); 705 blk = find_tcp_functions_locked(&fs); 706 if ((blk == NULL) || 707 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 708 error = ENOENT; 709 goto done; 710 } 711 tcp_func_set_ptr = blk; 712 done: 713 rw_wunlock(&tcp_function_lock); 714 return (error); 715 } 716 717 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 718 CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 719 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 720 "Set/get the default TCP functions"); 721 722 static int 723 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 724 { 725 int error, cnt, linesz; 726 struct tcp_function *f; 727 char *buffer, *cp; 728 size_t bufsz, outsz; 729 bool alias; 730 731 cnt = 0; 732 rw_rlock(&tcp_function_lock); 733 TAILQ_FOREACH(f, &t_functions, tf_next) { 734 cnt++; 735 } 736 rw_runlock(&tcp_function_lock); 737 738 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 739 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 740 741 error = 0; 742 cp = buffer; 743 744 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 745 "Alias", "PCB count"); 746 cp += linesz; 747 bufsz -= linesz; 748 outsz = linesz; 749 750 rw_rlock(&tcp_function_lock); 751 TAILQ_FOREACH(f, &t_functions, tf_next) { 752 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 753 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 754 f->tf_fb->tfb_tcp_block_name, 755 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 756 alias ? f->tf_name : "-", 757 f->tf_fb->tfb_refcnt); 758 if (linesz >= bufsz) { 759 error = EOVERFLOW; 760 break; 761 } 762 cp += linesz; 763 bufsz -= linesz; 764 outsz += linesz; 765 } 766 rw_runlock(&tcp_function_lock); 767 if (error == 0) 768 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 769 free(buffer, M_TEMP); 770 return (error); 771 } 772 773 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 774 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 775 NULL, 0, sysctl_net_inet_list_available, "A", 776 "list available TCP Function sets"); 777 778 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 779 780 #ifdef INET 781 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 782 #define V_udp4_tun_socket VNET(udp4_tun_socket) 783 #endif 784 #ifdef INET6 785 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 786 #define V_udp6_tun_socket VNET(udp6_tun_socket) 787 #endif 788 789 static struct sx tcpoudp_lock; 790 791 static void 792 tcp_over_udp_stop(void) 793 { 794 795 sx_assert(&tcpoudp_lock, SA_XLOCKED); 796 797 #ifdef INET 798 if (V_udp4_tun_socket != NULL) { 799 soclose(V_udp4_tun_socket); 800 V_udp4_tun_socket = NULL; 801 } 802 #endif 803 #ifdef INET6 804 if (V_udp6_tun_socket != NULL) { 805 soclose(V_udp6_tun_socket); 806 V_udp6_tun_socket = NULL; 807 } 808 #endif 809 } 810 811 static int 812 tcp_over_udp_start(void) 813 { 814 uint16_t port; 815 int ret; 816 #ifdef INET 817 struct sockaddr_in sin; 818 #endif 819 #ifdef INET6 820 struct sockaddr_in6 sin6; 821 #endif 822 823 sx_assert(&tcpoudp_lock, SA_XLOCKED); 824 825 port = V_tcp_udp_tunneling_port; 826 if (ntohs(port) == 0) { 827 /* Must have a port set */ 828 return (EINVAL); 829 } 830 #ifdef INET 831 if (V_udp4_tun_socket != NULL) { 832 /* Already running -- must stop first */ 833 return (EALREADY); 834 } 835 #endif 836 #ifdef INET6 837 if (V_udp6_tun_socket != NULL) { 838 /* Already running -- must stop first */ 839 return (EALREADY); 840 } 841 #endif 842 #ifdef INET 843 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 844 SOCK_DGRAM, IPPROTO_UDP, 845 curthread->td_ucred, curthread))) { 846 tcp_over_udp_stop(); 847 return (ret); 848 } 849 /* Call the special UDP hook. */ 850 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 851 tcp_recv_udp_tunneled_packet, 852 tcp_ctlinput_viaudp, 853 NULL))) { 854 tcp_over_udp_stop(); 855 return (ret); 856 } 857 /* Ok, we have a socket, bind it to the port. */ 858 memset(&sin, 0, sizeof(struct sockaddr_in)); 859 sin.sin_len = sizeof(struct sockaddr_in); 860 sin.sin_family = AF_INET; 861 sin.sin_port = htons(port); 862 if ((ret = sobind(V_udp4_tun_socket, 863 (struct sockaddr *)&sin, curthread))) { 864 tcp_over_udp_stop(); 865 return (ret); 866 } 867 #endif 868 #ifdef INET6 869 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 870 SOCK_DGRAM, IPPROTO_UDP, 871 curthread->td_ucred, curthread))) { 872 tcp_over_udp_stop(); 873 return (ret); 874 } 875 /* Call the special UDP hook. */ 876 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 877 tcp_recv_udp_tunneled_packet, 878 tcp6_ctlinput_viaudp, 879 NULL))) { 880 tcp_over_udp_stop(); 881 return (ret); 882 } 883 /* Ok, we have a socket, bind it to the port. */ 884 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 885 sin6.sin6_len = sizeof(struct sockaddr_in6); 886 sin6.sin6_family = AF_INET6; 887 sin6.sin6_port = htons(port); 888 if ((ret = sobind(V_udp6_tun_socket, 889 (struct sockaddr *)&sin6, curthread))) { 890 tcp_over_udp_stop(); 891 return (ret); 892 } 893 #endif 894 return (0); 895 } 896 897 static int 898 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 899 { 900 int error; 901 uint32_t old, new; 902 903 old = V_tcp_udp_tunneling_port; 904 new = old; 905 error = sysctl_handle_int(oidp, &new, 0, req); 906 if ((error == 0) && 907 (req->newptr != NULL)) { 908 if ((new < TCP_TUNNELING_PORT_MIN) || 909 (new > TCP_TUNNELING_PORT_MAX)) { 910 error = EINVAL; 911 } else { 912 sx_xlock(&tcpoudp_lock); 913 V_tcp_udp_tunneling_port = new; 914 if (old != 0) { 915 tcp_over_udp_stop(); 916 } 917 if (new != 0) { 918 error = tcp_over_udp_start(); 919 if (error != 0) { 920 V_tcp_udp_tunneling_port = 0; 921 } 922 } 923 sx_xunlock(&tcpoudp_lock); 924 } 925 } 926 return (error); 927 } 928 929 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 930 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 931 &VNET_NAME(tcp_udp_tunneling_port), 932 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 933 "Tunneling port for tcp over udp"); 934 935 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 936 937 static int 938 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 939 { 940 int error, new; 941 942 new = V_tcp_udp_tunneling_overhead; 943 error = sysctl_handle_int(oidp, &new, 0, req); 944 if (error == 0 && req->newptr) { 945 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 946 (new > TCP_TUNNELING_OVERHEAD_MAX)) 947 error = EINVAL; 948 else 949 V_tcp_udp_tunneling_overhead = new; 950 } 951 return (error); 952 } 953 954 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 955 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 956 &VNET_NAME(tcp_udp_tunneling_overhead), 957 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 958 "MSS reduction when using tcp over udp"); 959 960 /* 961 * Exports one (struct tcp_function_info) for each alias/name. 962 */ 963 static int 964 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 965 { 966 int cnt, error; 967 struct tcp_function *f; 968 struct tcp_function_info tfi; 969 970 /* 971 * We don't allow writes. 972 */ 973 if (req->newptr != NULL) 974 return (EINVAL); 975 976 /* 977 * Wire the old buffer so we can directly copy the functions to 978 * user space without dropping the lock. 979 */ 980 if (req->oldptr != NULL) { 981 error = sysctl_wire_old_buffer(req, 0); 982 if (error) 983 return (error); 984 } 985 986 /* 987 * Walk the list and copy out matching entries. If INVARIANTS 988 * is compiled in, also walk the list to verify the length of 989 * the list matches what we have recorded. 990 */ 991 rw_rlock(&tcp_function_lock); 992 993 cnt = 0; 994 #ifndef INVARIANTS 995 if (req->oldptr == NULL) { 996 cnt = tcp_fb_cnt; 997 goto skip_loop; 998 } 999 #endif 1000 TAILQ_FOREACH(f, &t_functions, tf_next) { 1001 #ifdef INVARIANTS 1002 cnt++; 1003 #endif 1004 if (req->oldptr != NULL) { 1005 bzero(&tfi, sizeof(tfi)); 1006 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 1007 tfi.tfi_id = f->tf_fb->tfb_id; 1008 (void)strlcpy(tfi.tfi_alias, f->tf_name, 1009 sizeof(tfi.tfi_alias)); 1010 (void)strlcpy(tfi.tfi_name, 1011 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 1012 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 1013 /* 1014 * Don't stop on error, as that is the 1015 * mechanism we use to accumulate length 1016 * information if the buffer was too short. 1017 */ 1018 } 1019 } 1020 KASSERT(cnt == tcp_fb_cnt, 1021 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1022 #ifndef INVARIANTS 1023 skip_loop: 1024 #endif 1025 rw_runlock(&tcp_function_lock); 1026 if (req->oldptr == NULL) 1027 error = SYSCTL_OUT(req, NULL, 1028 (cnt + 1) * sizeof(struct tcp_function_info)); 1029 1030 return (error); 1031 } 1032 1033 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1034 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1035 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1036 "List TCP function block name-to-ID mappings"); 1037 1038 /* 1039 * tfb_tcp_handoff_ok() function for the default stack. 1040 * Note that we'll basically try to take all comers. 1041 */ 1042 static int 1043 tcp_default_handoff_ok(struct tcpcb *tp) 1044 { 1045 1046 return (0); 1047 } 1048 1049 /* 1050 * tfb_tcp_fb_init() function for the default stack. 1051 * 1052 * This handles making sure we have appropriate timers set if you are 1053 * transitioning a socket that has some amount of setup done. 1054 * 1055 * The init() fuction from the default can *never* return non-zero i.e. 1056 * it is required to always succeed since it is the stack of last resort! 1057 */ 1058 static int 1059 tcp_default_fb_init(struct tcpcb *tp, void **ptr) 1060 { 1061 struct socket *so = tptosocket(tp); 1062 int rexmt; 1063 1064 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1065 /* We don't use the pointer */ 1066 *ptr = NULL; 1067 1068 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 1069 ("%s: connection %p in unexpected state %d", __func__, tp, 1070 tp->t_state)); 1071 1072 /* Make sure we get no interesting mbuf queuing behavior */ 1073 /* All mbuf queue/ack compress flags should be off */ 1074 tcp_lro_features_off(tptoinpcb(tp)); 1075 1076 /* Cancel the GP measurement in progress */ 1077 tp->t_flags &= ~TF_GPUTINPROG; 1078 /* Validate the timers are not in usec, if they are convert */ 1079 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 1080 if ((tp->t_state == TCPS_SYN_SENT) || 1081 (tp->t_state == TCPS_SYN_RECEIVED)) 1082 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift]; 1083 else 1084 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 1085 if (tp->t_rxtshift == 0) 1086 tp->t_rxtcur = rexmt; 1087 else 1088 TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX); 1089 1090 /* 1091 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1092 * know what to do for unexpected states (which includes TIME_WAIT). 1093 */ 1094 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1095 return (0); 1096 1097 /* 1098 * Make sure some kind of transmission timer is set if there is 1099 * outstanding data. 1100 */ 1101 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1102 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1103 tcp_timer_active(tp, TT_PERSIST))) { 1104 /* 1105 * If the session has established and it looks like it should 1106 * be in the persist state, set the persist timer. Otherwise, 1107 * set the retransmit timer. 1108 */ 1109 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1110 (int32_t)(tp->snd_nxt - tp->snd_una) < 1111 (int32_t)sbavail(&so->so_snd)) 1112 tcp_setpersist(tp); 1113 else 1114 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 1115 } 1116 1117 /* All non-embryonic sessions get a keepalive timer. */ 1118 if (!tcp_timer_active(tp, TT_KEEP)) 1119 tcp_timer_activate(tp, TT_KEEP, 1120 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1121 TP_KEEPINIT(tp)); 1122 1123 /* 1124 * Make sure critical variables are initialized 1125 * if transitioning while in Recovery. 1126 */ 1127 if IN_FASTRECOVERY(tp->t_flags) { 1128 if (tp->sackhint.recover_fs == 0) 1129 tp->sackhint.recover_fs = max(1, 1130 tp->snd_nxt - tp->snd_una); 1131 } 1132 1133 return (0); 1134 } 1135 1136 /* 1137 * tfb_tcp_fb_fini() function for the default stack. 1138 * 1139 * This changes state as necessary (or prudent) to prepare for another stack 1140 * to assume responsibility for the connection. 1141 */ 1142 static void 1143 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1144 { 1145 1146 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1147 } 1148 1149 /* 1150 * Target size of TCP PCB hash tables. Must be a power of two. 1151 * 1152 * Note that this can be overridden by the kernel environment 1153 * variable net.inet.tcp.tcbhashsize 1154 */ 1155 #ifndef TCBHASHSIZE 1156 #define TCBHASHSIZE 0 1157 #endif 1158 1159 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1160 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1161 1162 static struct mtx isn_mtx; 1163 1164 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1165 #define ISN_LOCK() mtx_lock(&isn_mtx) 1166 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1167 1168 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1169 1170 /* 1171 * Take a value and get the next power of 2 that doesn't overflow. 1172 * Used to size the tcp_inpcb hash buckets. 1173 */ 1174 static int 1175 maketcp_hashsize(int size) 1176 { 1177 int hashsize; 1178 1179 /* 1180 * auto tune. 1181 * get the next power of 2 higher than maxsockets. 1182 */ 1183 hashsize = 1 << fls(size); 1184 /* catch overflow, and just go one power of 2 smaller */ 1185 if (hashsize < size) { 1186 hashsize = 1 << (fls(size) - 1); 1187 } 1188 return (hashsize); 1189 } 1190 1191 static volatile int next_tcp_stack_id = 1; 1192 1193 /* 1194 * Register a TCP function block with the name provided in the names 1195 * array. (Note that this function does NOT automatically register 1196 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1197 * explicitly include blk->tfb_tcp_block_name in the list of names if 1198 * you wish to register the stack with that name.) 1199 * 1200 * Either all name registrations will succeed or all will fail. If 1201 * a name registration fails, the function will update the num_names 1202 * argument to point to the array index of the name that encountered 1203 * the failure. 1204 * 1205 * Returns 0 on success, or an error code on failure. 1206 */ 1207 int 1208 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1209 const char *names[], int *num_names) 1210 { 1211 struct tcp_function *n; 1212 struct tcp_function_set fs; 1213 int error, i; 1214 1215 KASSERT(names != NULL && *num_names > 0, 1216 ("%s: Called with 0-length name list", __func__)); 1217 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1218 KASSERT(rw_initialized(&tcp_function_lock), 1219 ("%s: called too early", __func__)); 1220 1221 if ((blk->tfb_tcp_output == NULL) || 1222 (blk->tfb_tcp_do_segment == NULL) || 1223 (blk->tfb_tcp_ctloutput == NULL) || 1224 (strlen(blk->tfb_tcp_block_name) == 0)) { 1225 /* 1226 * These functions are required and you 1227 * need a name. 1228 */ 1229 *num_names = 0; 1230 return (EINVAL); 1231 } 1232 1233 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1234 *num_names = 0; 1235 return (EINVAL); 1236 } 1237 1238 refcount_init(&blk->tfb_refcnt, 0); 1239 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1240 for (i = 0; i < *num_names; i++) { 1241 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1242 if (n == NULL) { 1243 error = ENOMEM; 1244 goto cleanup; 1245 } 1246 n->tf_fb = blk; 1247 1248 (void)strlcpy(fs.function_set_name, names[i], 1249 sizeof(fs.function_set_name)); 1250 rw_wlock(&tcp_function_lock); 1251 if (find_tcp_functions_locked(&fs) != NULL) { 1252 /* Duplicate name space not allowed */ 1253 rw_wunlock(&tcp_function_lock); 1254 free(n, M_TCPFUNCTIONS); 1255 error = EALREADY; 1256 goto cleanup; 1257 } 1258 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 1259 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 1260 tcp_fb_cnt++; 1261 rw_wunlock(&tcp_function_lock); 1262 } 1263 return(0); 1264 1265 cleanup: 1266 /* 1267 * Deregister the names we just added. Because registration failed 1268 * for names[i], we don't need to deregister that name. 1269 */ 1270 *num_names = i; 1271 rw_wlock(&tcp_function_lock); 1272 while (--i >= 0) { 1273 TAILQ_FOREACH(n, &t_functions, tf_next) { 1274 if (!strncmp(n->tf_name, names[i], 1275 TCP_FUNCTION_NAME_LEN_MAX)) { 1276 TAILQ_REMOVE(&t_functions, n, tf_next); 1277 tcp_fb_cnt--; 1278 n->tf_fb = NULL; 1279 free(n, M_TCPFUNCTIONS); 1280 break; 1281 } 1282 } 1283 } 1284 rw_wunlock(&tcp_function_lock); 1285 return (error); 1286 } 1287 1288 /* 1289 * Register a TCP function block using the name provided in the name 1290 * argument. 1291 * 1292 * Returns 0 on success, or an error code on failure. 1293 */ 1294 int 1295 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1296 int wait) 1297 { 1298 const char *name_list[1]; 1299 int num_names, rv; 1300 1301 num_names = 1; 1302 if (name != NULL) 1303 name_list[0] = name; 1304 else 1305 name_list[0] = blk->tfb_tcp_block_name; 1306 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1307 return (rv); 1308 } 1309 1310 /* 1311 * Register a TCP function block using the name defined in 1312 * blk->tfb_tcp_block_name. 1313 * 1314 * Returns 0 on success, or an error code on failure. 1315 */ 1316 int 1317 register_tcp_functions(struct tcp_function_block *blk, int wait) 1318 { 1319 1320 return (register_tcp_functions_as_name(blk, NULL, wait)); 1321 } 1322 1323 /* 1324 * Deregister all names associated with a function block. This 1325 * functionally removes the function block from use within the system. 1326 * 1327 * When called with a true quiesce argument, mark the function block 1328 * as being removed so no more stacks will use it and determine 1329 * whether the removal would succeed. 1330 * 1331 * When called with a false quiesce argument, actually attempt the 1332 * removal. 1333 * 1334 * When called with a force argument, attempt to switch all TCBs to 1335 * use the default stack instead of returning EBUSY. 1336 * 1337 * Returns 0 on success (or if the removal would succeed, or an error 1338 * code on failure. 1339 */ 1340 int 1341 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1342 bool force) 1343 { 1344 struct tcp_function *f; 1345 1346 if (blk == &tcp_def_funcblk) { 1347 /* You can't un-register the default */ 1348 return (EPERM); 1349 } 1350 rw_wlock(&tcp_function_lock); 1351 if (blk == tcp_func_set_ptr) { 1352 /* You can't free the current default */ 1353 rw_wunlock(&tcp_function_lock); 1354 return (EBUSY); 1355 } 1356 /* Mark the block so no more stacks can use it. */ 1357 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1358 /* 1359 * If TCBs are still attached to the stack, attempt to switch them 1360 * to the default stack. 1361 */ 1362 if (force && blk->tfb_refcnt) { 1363 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1364 INPLOOKUP_WLOCKPCB); 1365 struct inpcb *inp; 1366 struct tcpcb *tp; 1367 VNET_ITERATOR_DECL(vnet_iter); 1368 1369 rw_wunlock(&tcp_function_lock); 1370 1371 VNET_LIST_RLOCK(); 1372 VNET_FOREACH(vnet_iter) { 1373 CURVNET_SET(vnet_iter); 1374 while ((inp = inp_next(&inpi)) != NULL) { 1375 tp = intotcpcb(inp); 1376 if (tp == NULL || tp->t_fb != blk) 1377 continue; 1378 tcp_switch_back_to_default(tp); 1379 } 1380 CURVNET_RESTORE(); 1381 } 1382 VNET_LIST_RUNLOCK(); 1383 1384 rw_wlock(&tcp_function_lock); 1385 } 1386 if (blk->tfb_refcnt) { 1387 /* TCBs still attached. */ 1388 rw_wunlock(&tcp_function_lock); 1389 return (EBUSY); 1390 } 1391 if (quiesce) { 1392 /* Skip removal. */ 1393 rw_wunlock(&tcp_function_lock); 1394 return (0); 1395 } 1396 /* Remove any function names that map to this function block. */ 1397 while (find_tcp_fb_locked(blk, &f) != NULL) { 1398 TAILQ_REMOVE(&t_functions, f, tf_next); 1399 tcp_fb_cnt--; 1400 f->tf_fb = NULL; 1401 free(f, M_TCPFUNCTIONS); 1402 } 1403 rw_wunlock(&tcp_function_lock); 1404 return (0); 1405 } 1406 1407 static void 1408 tcp_drain(void) 1409 { 1410 struct epoch_tracker et; 1411 VNET_ITERATOR_DECL(vnet_iter); 1412 1413 if (!do_tcpdrain) 1414 return; 1415 1416 NET_EPOCH_ENTER(et); 1417 VNET_LIST_RLOCK_NOSLEEP(); 1418 VNET_FOREACH(vnet_iter) { 1419 CURVNET_SET(vnet_iter); 1420 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1421 INPLOOKUP_WLOCKPCB); 1422 struct inpcb *inpb; 1423 struct tcpcb *tcpb; 1424 1425 /* 1426 * Walk the tcpbs, if existing, and flush the reassembly queue, 1427 * if there is one... 1428 * XXX: The "Net/3" implementation doesn't imply that the TCP 1429 * reassembly queue should be flushed, but in a situation 1430 * where we're really low on mbufs, this is potentially 1431 * useful. 1432 */ 1433 while ((inpb = inp_next(&inpi)) != NULL) { 1434 if ((tcpb = intotcpcb(inpb)) != NULL) { 1435 tcp_reass_flush(tcpb); 1436 tcp_clean_sackreport(tcpb); 1437 #ifdef TCP_BLACKBOX 1438 tcp_log_drain(tcpb); 1439 #endif 1440 #ifdef TCPPCAP 1441 if (tcp_pcap_aggressive_free) { 1442 /* Free the TCP PCAP queues. */ 1443 tcp_pcap_drain(&(tcpb->t_inpkts)); 1444 tcp_pcap_drain(&(tcpb->t_outpkts)); 1445 } 1446 #endif 1447 } 1448 } 1449 CURVNET_RESTORE(); 1450 } 1451 VNET_LIST_RUNLOCK_NOSLEEP(); 1452 NET_EPOCH_EXIT(et); 1453 } 1454 1455 static void 1456 tcp_vnet_init(void *arg __unused) 1457 { 1458 1459 #ifdef TCP_HHOOK 1460 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1461 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1462 printf("%s: WARNING: unable to register helper hook\n", __func__); 1463 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1464 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1465 printf("%s: WARNING: unable to register helper hook\n", __func__); 1466 #endif 1467 #ifdef STATS 1468 if (tcp_stats_init()) 1469 printf("%s: WARNING: unable to initialise TCP stats\n", 1470 __func__); 1471 #endif 1472 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1473 tcp_tcbhashsize); 1474 1475 syncache_init(); 1476 tcp_hc_init(); 1477 1478 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1479 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1480 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1481 1482 tcp_fastopen_init(); 1483 1484 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1485 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1486 1487 V_tcp_msl = TCPTV_MSL; 1488 } 1489 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1490 tcp_vnet_init, NULL); 1491 1492 static void 1493 tcp_init(void *arg __unused) 1494 { 1495 const char *tcbhash_tuneable; 1496 int hashsize; 1497 1498 tcp_reass_global_init(); 1499 1500 /* XXX virtualize those below? */ 1501 tcp_delacktime = TCPTV_DELACK; 1502 tcp_keepinit = TCPTV_KEEP_INIT; 1503 tcp_keepidle = TCPTV_KEEP_IDLE; 1504 tcp_keepintvl = TCPTV_KEEPINTVL; 1505 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1506 tcp_rexmit_initial = TCPTV_RTOBASE; 1507 if (tcp_rexmit_initial < 1) 1508 tcp_rexmit_initial = 1; 1509 tcp_rexmit_min = TCPTV_MIN; 1510 if (tcp_rexmit_min < 1) 1511 tcp_rexmit_min = 1; 1512 tcp_persmin = TCPTV_PERSMIN; 1513 tcp_persmax = TCPTV_PERSMAX; 1514 tcp_rexmit_slop = TCPTV_CPU_VAR; 1515 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1516 1517 /* Setup the tcp function block list */ 1518 TAILQ_INIT(&t_functions); 1519 rw_init(&tcp_function_lock, "tcp_func_lock"); 1520 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1521 sx_init(&tcpoudp_lock, "TCP over UDP configuration"); 1522 #ifdef TCP_BLACKBOX 1523 /* Initialize the TCP logging data. */ 1524 tcp_log_init(); 1525 #endif 1526 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1527 1528 if (tcp_soreceive_stream) { 1529 #ifdef INET 1530 tcp_protosw.pr_soreceive = soreceive_stream; 1531 #endif 1532 #ifdef INET6 1533 tcp6_protosw.pr_soreceive = soreceive_stream; 1534 #endif /* INET6 */ 1535 } 1536 1537 #ifdef INET6 1538 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1539 #else /* INET6 */ 1540 max_protohdr_grow(sizeof(struct tcpiphdr)); 1541 #endif /* INET6 */ 1542 1543 ISN_LOCK_INIT(); 1544 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1545 SHUTDOWN_PRI_DEFAULT); 1546 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1547 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1548 1549 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1550 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1551 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1552 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1553 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1554 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1555 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1556 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1557 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1558 #ifdef TCPPCAP 1559 tcp_pcap_init(); 1560 #endif 1561 1562 hashsize = TCBHASHSIZE; 1563 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1564 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1565 if (hashsize == 0) { 1566 /* 1567 * Auto tune the hash size based on maxsockets. 1568 * A perfect hash would have a 1:1 mapping 1569 * (hashsize = maxsockets) however it's been 1570 * suggested that O(2) average is better. 1571 */ 1572 hashsize = maketcp_hashsize(maxsockets / 4); 1573 /* 1574 * Our historical default is 512, 1575 * do not autotune lower than this. 1576 */ 1577 if (hashsize < 512) 1578 hashsize = 512; 1579 if (bootverbose) 1580 printf("%s: %s auto tuned to %d\n", __func__, 1581 tcbhash_tuneable, hashsize); 1582 } 1583 /* 1584 * We require a hashsize to be a power of two. 1585 * Previously if it was not a power of two we would just reset it 1586 * back to 512, which could be a nasty surprise if you did not notice 1587 * the error message. 1588 * Instead what we do is clip it to the closest power of two lower 1589 * than the specified hash value. 1590 */ 1591 if (!powerof2(hashsize)) { 1592 int oldhashsize = hashsize; 1593 1594 hashsize = maketcp_hashsize(hashsize); 1595 /* prevent absurdly low value */ 1596 if (hashsize < 16) 1597 hashsize = 16; 1598 printf("%s: WARNING: TCB hash size not a power of 2, " 1599 "clipped from %d to %d.\n", __func__, oldhashsize, 1600 hashsize); 1601 } 1602 tcp_tcbhashsize = hashsize; 1603 1604 #ifdef INET 1605 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1606 #endif 1607 #ifdef INET6 1608 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1609 #endif 1610 } 1611 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1612 1613 #ifdef VIMAGE 1614 static void 1615 tcp_destroy(void *unused __unused) 1616 { 1617 int n; 1618 #ifdef TCP_HHOOK 1619 int error; 1620 #endif 1621 1622 /* 1623 * All our processes are gone, all our sockets should be cleaned 1624 * up, which means, we should be past the tcp_discardcb() calls. 1625 * Sleep to let all tcpcb timers really disappear and cleanup. 1626 */ 1627 for (;;) { 1628 INP_INFO_WLOCK(&V_tcbinfo); 1629 n = V_tcbinfo.ipi_count; 1630 INP_INFO_WUNLOCK(&V_tcbinfo); 1631 if (n == 0) 1632 break; 1633 pause("tcpdes", hz / 10); 1634 } 1635 tcp_hc_destroy(); 1636 syncache_destroy(); 1637 in_pcbinfo_destroy(&V_tcbinfo); 1638 /* tcp_discardcb() clears the sack_holes up. */ 1639 uma_zdestroy(V_sack_hole_zone); 1640 1641 /* 1642 * Cannot free the zone until all tcpcbs are released as we attach 1643 * the allocations to them. 1644 */ 1645 tcp_fastopen_destroy(); 1646 1647 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1648 VNET_PCPUSTAT_FREE(tcpstat); 1649 1650 #ifdef TCP_HHOOK 1651 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1652 if (error != 0) { 1653 printf("%s: WARNING: unable to deregister helper hook " 1654 "type=%d, id=%d: error %d returned\n", __func__, 1655 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1656 } 1657 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1658 if (error != 0) { 1659 printf("%s: WARNING: unable to deregister helper hook " 1660 "type=%d, id=%d: error %d returned\n", __func__, 1661 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1662 } 1663 #endif 1664 } 1665 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1666 #endif 1667 1668 void 1669 tcp_fini(void *xtp) 1670 { 1671 1672 } 1673 1674 /* 1675 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1676 * tcp_template used to store this data in mbufs, but we now recopy it out 1677 * of the tcpcb each time to conserve mbufs. 1678 */ 1679 void 1680 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1681 { 1682 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1683 1684 INP_WLOCK_ASSERT(inp); 1685 1686 #ifdef INET6 1687 if ((inp->inp_vflag & INP_IPV6) != 0) { 1688 struct ip6_hdr *ip6; 1689 1690 ip6 = (struct ip6_hdr *)ip_ptr; 1691 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1692 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1693 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1694 (IPV6_VERSION & IPV6_VERSION_MASK); 1695 if (port == 0) 1696 ip6->ip6_nxt = IPPROTO_TCP; 1697 else 1698 ip6->ip6_nxt = IPPROTO_UDP; 1699 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1700 ip6->ip6_src = inp->in6p_laddr; 1701 ip6->ip6_dst = inp->in6p_faddr; 1702 } 1703 #endif /* INET6 */ 1704 #if defined(INET6) && defined(INET) 1705 else 1706 #endif 1707 #ifdef INET 1708 { 1709 struct ip *ip; 1710 1711 ip = (struct ip *)ip_ptr; 1712 ip->ip_v = IPVERSION; 1713 ip->ip_hl = 5; 1714 ip->ip_tos = inp->inp_ip_tos; 1715 ip->ip_len = 0; 1716 ip->ip_id = 0; 1717 ip->ip_off = 0; 1718 ip->ip_ttl = inp->inp_ip_ttl; 1719 ip->ip_sum = 0; 1720 if (port == 0) 1721 ip->ip_p = IPPROTO_TCP; 1722 else 1723 ip->ip_p = IPPROTO_UDP; 1724 ip->ip_src = inp->inp_laddr; 1725 ip->ip_dst = inp->inp_faddr; 1726 } 1727 #endif /* INET */ 1728 th->th_sport = inp->inp_lport; 1729 th->th_dport = inp->inp_fport; 1730 th->th_seq = 0; 1731 th->th_ack = 0; 1732 th->th_off = 5; 1733 tcp_set_flags(th, 0); 1734 th->th_win = 0; 1735 th->th_urp = 0; 1736 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1737 } 1738 1739 /* 1740 * Create template to be used to send tcp packets on a connection. 1741 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1742 * use for this function is in keepalives, which use tcp_respond. 1743 */ 1744 struct tcptemp * 1745 tcpip_maketemplate(struct inpcb *inp) 1746 { 1747 struct tcptemp *t; 1748 1749 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1750 if (t == NULL) 1751 return (NULL); 1752 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1753 return (t); 1754 } 1755 1756 /* 1757 * Send a single message to the TCP at address specified by 1758 * the given TCP/IP header. If m == NULL, then we make a copy 1759 * of the tcpiphdr at th and send directly to the addressed host. 1760 * This is used to force keep alive messages out using the TCP 1761 * template for a connection. If flags are given then we send 1762 * a message back to the TCP which originated the segment th, 1763 * and discard the mbuf containing it and any other attached mbufs. 1764 * 1765 * In any case the ack and sequence number of the transmitted 1766 * segment are as specified by the parameters. 1767 * 1768 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1769 */ 1770 void 1771 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1772 tcp_seq ack, tcp_seq seq, uint16_t flags) 1773 { 1774 struct tcpopt to; 1775 struct inpcb *inp; 1776 struct ip *ip; 1777 struct mbuf *optm; 1778 struct udphdr *uh = NULL; 1779 struct tcphdr *nth; 1780 struct tcp_log_buffer *lgb; 1781 u_char *optp; 1782 #ifdef INET6 1783 struct ip6_hdr *ip6; 1784 int isipv6; 1785 #endif /* INET6 */ 1786 int optlen, tlen, win, ulen; 1787 int ect = 0; 1788 bool incl_opts; 1789 uint16_t port; 1790 int output_ret; 1791 #ifdef INVARIANTS 1792 int thflags = tcp_get_flags(th); 1793 #endif 1794 1795 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1796 NET_EPOCH_ASSERT(); 1797 1798 #ifdef INET6 1799 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1800 ip6 = ipgen; 1801 #endif /* INET6 */ 1802 ip = ipgen; 1803 1804 if (tp != NULL) { 1805 inp = tptoinpcb(tp); 1806 INP_LOCK_ASSERT(inp); 1807 } else 1808 inp = NULL; 1809 1810 if (m != NULL) { 1811 #ifdef INET6 1812 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1813 port = m->m_pkthdr.tcp_tun_port; 1814 else 1815 #endif 1816 if (ip && (ip->ip_p == IPPROTO_UDP)) 1817 port = m->m_pkthdr.tcp_tun_port; 1818 else 1819 port = 0; 1820 } else 1821 port = tp->t_port; 1822 1823 incl_opts = false; 1824 win = 0; 1825 if (tp != NULL) { 1826 if (!(flags & TH_RST)) { 1827 win = sbspace(&inp->inp_socket->so_rcv); 1828 if (win > TCP_MAXWIN << tp->rcv_scale) 1829 win = TCP_MAXWIN << tp->rcv_scale; 1830 } 1831 if ((tp->t_flags & TF_NOOPT) == 0) 1832 incl_opts = true; 1833 } 1834 if (m == NULL) { 1835 m = m_gethdr(M_NOWAIT, MT_DATA); 1836 if (m == NULL) 1837 return; 1838 m->m_data += max_linkhdr; 1839 #ifdef INET6 1840 if (isipv6) { 1841 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1842 sizeof(struct ip6_hdr)); 1843 ip6 = mtod(m, struct ip6_hdr *); 1844 nth = (struct tcphdr *)(ip6 + 1); 1845 if (port) { 1846 /* Insert a UDP header */ 1847 uh = (struct udphdr *)nth; 1848 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1849 uh->uh_dport = port; 1850 nth = (struct tcphdr *)(uh + 1); 1851 } 1852 } else 1853 #endif /* INET6 */ 1854 { 1855 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1856 ip = mtod(m, struct ip *); 1857 nth = (struct tcphdr *)(ip + 1); 1858 if (port) { 1859 /* Insert a UDP header */ 1860 uh = (struct udphdr *)nth; 1861 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1862 uh->uh_dport = port; 1863 nth = (struct tcphdr *)(uh + 1); 1864 } 1865 } 1866 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1867 flags = TH_ACK; 1868 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1869 struct mbuf *n; 1870 1871 /* Can't reuse 'm', allocate a new mbuf. */ 1872 n = m_gethdr(M_NOWAIT, MT_DATA); 1873 if (n == NULL) { 1874 m_freem(m); 1875 return; 1876 } 1877 1878 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1879 m_freem(m); 1880 m_freem(n); 1881 return; 1882 } 1883 1884 n->m_data += max_linkhdr; 1885 /* m_len is set later */ 1886 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1887 #ifdef INET6 1888 if (isipv6) { 1889 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1890 sizeof(struct ip6_hdr)); 1891 ip6 = mtod(n, struct ip6_hdr *); 1892 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1893 nth = (struct tcphdr *)(ip6 + 1); 1894 if (port) { 1895 /* Insert a UDP header */ 1896 uh = (struct udphdr *)nth; 1897 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1898 uh->uh_dport = port; 1899 nth = (struct tcphdr *)(uh + 1); 1900 } 1901 } else 1902 #endif /* INET6 */ 1903 { 1904 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1905 ip = mtod(n, struct ip *); 1906 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1907 nth = (struct tcphdr *)(ip + 1); 1908 if (port) { 1909 /* Insert a UDP header */ 1910 uh = (struct udphdr *)nth; 1911 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1912 uh->uh_dport = port; 1913 nth = (struct tcphdr *)(uh + 1); 1914 } 1915 } 1916 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1917 xchg(nth->th_dport, nth->th_sport, uint16_t); 1918 th = nth; 1919 m_freem(m); 1920 m = n; 1921 } else { 1922 /* 1923 * reuse the mbuf. 1924 * XXX MRT We inherit the FIB, which is lucky. 1925 */ 1926 m_freem(m->m_next); 1927 m->m_next = NULL; 1928 m->m_data = (caddr_t)ipgen; 1929 /* m_len is set later */ 1930 #ifdef INET6 1931 if (isipv6) { 1932 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1933 nth = (struct tcphdr *)(ip6 + 1); 1934 } else 1935 #endif /* INET6 */ 1936 { 1937 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1938 nth = (struct tcphdr *)(ip + 1); 1939 } 1940 if (th != nth) { 1941 /* 1942 * this is usually a case when an extension header 1943 * exists between the IPv6 header and the 1944 * TCP header. 1945 */ 1946 nth->th_sport = th->th_sport; 1947 nth->th_dport = th->th_dport; 1948 } 1949 xchg(nth->th_dport, nth->th_sport, uint16_t); 1950 #undef xchg 1951 } 1952 tlen = 0; 1953 #ifdef INET6 1954 if (isipv6) 1955 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1956 #endif 1957 #if defined(INET) && defined(INET6) 1958 else 1959 #endif 1960 #ifdef INET 1961 tlen = sizeof (struct tcpiphdr); 1962 #endif 1963 if (port) 1964 tlen += sizeof (struct udphdr); 1965 #ifdef INVARIANTS 1966 m->m_len = 0; 1967 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1968 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1969 m, tlen, (long)M_TRAILINGSPACE(m))); 1970 #endif 1971 m->m_len = tlen; 1972 to.to_flags = 0; 1973 if (incl_opts) { 1974 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1975 /* Make sure we have room. */ 1976 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1977 m->m_next = m_get(M_NOWAIT, MT_DATA); 1978 if (m->m_next) { 1979 optp = mtod(m->m_next, u_char *); 1980 optm = m->m_next; 1981 } else 1982 incl_opts = false; 1983 } else { 1984 optp = (u_char *) (nth + 1); 1985 optm = m; 1986 } 1987 } 1988 if (incl_opts) { 1989 /* Timestamps. */ 1990 if (tp->t_flags & TF_RCVD_TSTMP) { 1991 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1992 to.to_tsecr = tp->ts_recent; 1993 to.to_flags |= TOF_TS; 1994 } 1995 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1996 /* TCP-MD5 (RFC2385). */ 1997 if (tp->t_flags & TF_SIGNATURE) 1998 to.to_flags |= TOF_SIGNATURE; 1999 #endif 2000 /* Add the options. */ 2001 tlen += optlen = tcp_addoptions(&to, optp); 2002 2003 /* Update m_len in the correct mbuf. */ 2004 optm->m_len += optlen; 2005 } else 2006 optlen = 0; 2007 #ifdef INET6 2008 if (isipv6) { 2009 if (uh) { 2010 ulen = tlen - sizeof(struct ip6_hdr); 2011 uh->uh_ulen = htons(ulen); 2012 } 2013 ip6->ip6_flow = htonl(ect << 20); 2014 ip6->ip6_vfc = IPV6_VERSION; 2015 if (port) 2016 ip6->ip6_nxt = IPPROTO_UDP; 2017 else 2018 ip6->ip6_nxt = IPPROTO_TCP; 2019 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 2020 } 2021 #endif 2022 #if defined(INET) && defined(INET6) 2023 else 2024 #endif 2025 #ifdef INET 2026 { 2027 if (uh) { 2028 ulen = tlen - sizeof(struct ip); 2029 uh->uh_ulen = htons(ulen); 2030 } 2031 ip->ip_tos = ect; 2032 ip->ip_len = htons(tlen); 2033 ip->ip_ttl = V_ip_defttl; 2034 if (port) { 2035 ip->ip_p = IPPROTO_UDP; 2036 } else { 2037 ip->ip_p = IPPROTO_TCP; 2038 } 2039 if (V_path_mtu_discovery) 2040 ip->ip_off |= htons(IP_DF); 2041 } 2042 #endif 2043 m->m_pkthdr.len = tlen; 2044 m->m_pkthdr.rcvif = NULL; 2045 #ifdef MAC 2046 if (inp != NULL) { 2047 /* 2048 * Packet is associated with a socket, so allow the 2049 * label of the response to reflect the socket label. 2050 */ 2051 INP_LOCK_ASSERT(inp); 2052 mac_inpcb_create_mbuf(inp, m); 2053 } else { 2054 /* 2055 * Packet is not associated with a socket, so possibly 2056 * update the label in place. 2057 */ 2058 mac_netinet_tcp_reply(m); 2059 } 2060 #endif 2061 nth->th_seq = htonl(seq); 2062 nth->th_ack = htonl(ack); 2063 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2064 tcp_set_flags(nth, flags); 2065 if (tp != NULL) 2066 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2067 else 2068 nth->th_win = htons((u_short)win); 2069 nth->th_urp = 0; 2070 2071 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2072 if (to.to_flags & TOF_SIGNATURE) { 2073 if (!TCPMD5_ENABLED() || 2074 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2075 m_freem(m); 2076 return; 2077 } 2078 } 2079 #endif 2080 2081 #ifdef INET6 2082 if (isipv6) { 2083 if (port) { 2084 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2085 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2086 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2087 nth->th_sum = 0; 2088 } else { 2089 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2090 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2091 nth->th_sum = in6_cksum_pseudo(ip6, 2092 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2093 } 2094 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2095 } 2096 #endif /* INET6 */ 2097 #if defined(INET6) && defined(INET) 2098 else 2099 #endif 2100 #ifdef INET 2101 { 2102 if (port) { 2103 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2104 htons(ulen + IPPROTO_UDP)); 2105 m->m_pkthdr.csum_flags = CSUM_UDP; 2106 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2107 nth->th_sum = 0; 2108 } else { 2109 m->m_pkthdr.csum_flags = CSUM_TCP; 2110 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2111 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2112 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2113 } 2114 } 2115 #endif /* INET */ 2116 TCP_PROBE3(debug__output, tp, th, m); 2117 if (flags & TH_RST) 2118 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2119 lgb = NULL; 2120 if ((tp != NULL) && tcp_bblogging_on(tp)) { 2121 if (INP_WLOCKED(inp)) { 2122 union tcp_log_stackspecific log; 2123 struct timeval tv; 2124 2125 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2126 log.u_bbr.inhpts = inp->inp_in_hpts; 2127 log.u_bbr.flex8 = 4; 2128 log.u_bbr.pkts_out = tp->t_maxseg; 2129 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2130 log.u_bbr.delivered = 0; 2131 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT, 2132 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2133 } else { 2134 /* 2135 * We can not log the packet, since we only own the 2136 * read lock, but a write lock is needed. The read lock 2137 * is not upgraded to a write lock, since only getting 2138 * the read lock was done intentionally to improve the 2139 * handling of SYN flooding attacks. 2140 * This happens only for pure SYN segments received in 2141 * the initial CLOSED state, or received in a more 2142 * advanced state than listen and the UDP encapsulation 2143 * port is unexpected. 2144 * The incoming SYN segments do not really belong to 2145 * the TCP connection and the handling does not change 2146 * the state of the TCP connection. Therefore, the 2147 * sending of the RST segments is not logged. Please 2148 * note that also the incoming SYN segments are not 2149 * logged. 2150 * 2151 * The following code ensures that the above description 2152 * is and stays correct. 2153 */ 2154 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2155 (tp->t_state == TCPS_CLOSED || 2156 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2157 ("%s: Logging of TCP segment with flags 0x%b and " 2158 "UDP encapsulation port %u skipped in state %s", 2159 __func__, thflags, PRINT_TH_FLAGS, 2160 ntohs(port), tcpstates[tp->t_state])); 2161 } 2162 } 2163 2164 if (flags & TH_ACK) 2165 TCPSTAT_INC(tcps_sndacks); 2166 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2167 TCPSTAT_INC(tcps_sndctrl); 2168 TCPSTAT_INC(tcps_sndtotal); 2169 2170 #ifdef INET6 2171 if (isipv6) { 2172 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2173 output_ret = ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 2174 } 2175 #endif /* INET6 */ 2176 #if defined(INET) && defined(INET6) 2177 else 2178 #endif 2179 #ifdef INET 2180 { 2181 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2182 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2183 } 2184 #endif 2185 if (lgb != NULL) 2186 lgb->tlb_errno = output_ret; 2187 } 2188 2189 /* 2190 * Create a new TCP control block, making an empty reassembly queue and hooking 2191 * it to the argument protocol control block. The `inp' parameter must have 2192 * come from the zone allocator set up by tcpcbstor declaration. 2193 */ 2194 struct tcpcb * 2195 tcp_newtcpcb(struct inpcb *inp) 2196 { 2197 struct tcpcb *tp = intotcpcb(inp); 2198 #ifdef INET6 2199 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2200 #endif /* INET6 */ 2201 2202 /* 2203 * Historically allocation was done with M_ZERO. There is a lot of 2204 * code that rely on that. For now take safe approach and zero whole 2205 * tcpcb. This definitely can be optimized. 2206 */ 2207 bzero(&tp->t_start_zero, t_zero_size); 2208 2209 /* Initialise cc_var struct for this tcpcb. */ 2210 tp->t_ccv.type = IPPROTO_TCP; 2211 tp->t_ccv.ccvc.tcp = tp; 2212 rw_rlock(&tcp_function_lock); 2213 tp->t_fb = tcp_func_set_ptr; 2214 refcount_acquire(&tp->t_fb->tfb_refcnt); 2215 rw_runlock(&tcp_function_lock); 2216 /* 2217 * Use the current system default CC algorithm. 2218 */ 2219 cc_attach(tp, CC_DEFAULT_ALGO()); 2220 2221 if (CC_ALGO(tp)->cb_init != NULL) 2222 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2223 cc_detach(tp); 2224 if (tp->t_fb->tfb_tcp_fb_fini) 2225 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2226 refcount_release(&tp->t_fb->tfb_refcnt); 2227 return (NULL); 2228 } 2229 2230 #ifdef TCP_HHOOK 2231 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2232 if (tp->t_fb->tfb_tcp_fb_fini) 2233 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2234 refcount_release(&tp->t_fb->tfb_refcnt); 2235 return (NULL); 2236 } 2237 #endif 2238 2239 TAILQ_INIT(&tp->t_segq); 2240 tp->t_maxseg = 2241 #ifdef INET6 2242 isipv6 ? V_tcp_v6mssdflt : 2243 #endif /* INET6 */ 2244 V_tcp_mssdflt; 2245 2246 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED); 2247 for (int i = 0; i < TT_N; i++) 2248 tp->t_timers[i] = SBT_MAX; 2249 2250 switch (V_tcp_do_rfc1323) { 2251 case 0: 2252 break; 2253 default: 2254 case 1: 2255 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2256 break; 2257 case 2: 2258 tp->t_flags = TF_REQ_SCALE; 2259 break; 2260 case 3: 2261 tp->t_flags = TF_REQ_TSTMP; 2262 break; 2263 } 2264 if (V_tcp_do_sack) 2265 tp->t_flags |= TF_SACK_PERMIT; 2266 TAILQ_INIT(&tp->snd_holes); 2267 2268 /* 2269 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2270 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2271 * reasonable initial retransmit time. 2272 */ 2273 tp->t_srtt = TCPTV_SRTTBASE; 2274 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2275 tp->t_rttmin = tcp_rexmit_min; 2276 tp->t_rxtcur = tcp_rexmit_initial; 2277 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2278 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2279 tp->t_rcvtime = ticks; 2280 /* We always start with ticks granularity */ 2281 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 2282 /* 2283 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2284 * because the socket may be bound to an IPv6 wildcard address, 2285 * which may match an IPv4-mapped IPv6 address. 2286 */ 2287 inp->inp_ip_ttl = V_ip_defttl; 2288 #ifdef TCPHPTS 2289 /* 2290 * If using hpts lets drop a random number in so 2291 * not all new connections fall on the same CPU. 2292 */ 2293 inp->inp_hpts_cpu = hpts_random_cpu(inp); 2294 #endif 2295 #ifdef TCPPCAP 2296 /* 2297 * Init the TCP PCAP queues. 2298 */ 2299 tcp_pcap_tcpcb_init(tp); 2300 #endif 2301 #ifdef TCP_BLACKBOX 2302 /* Initialize the per-TCPCB log data. */ 2303 tcp_log_tcpcbinit(tp); 2304 #endif 2305 tp->t_pacing_rate = -1; 2306 if (tp->t_fb->tfb_tcp_fb_init) { 2307 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) { 2308 refcount_release(&tp->t_fb->tfb_refcnt); 2309 return (NULL); 2310 } 2311 } 2312 #ifdef STATS 2313 if (V_tcp_perconn_stats_enable == 1) 2314 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2315 #endif 2316 if (V_tcp_do_lrd) 2317 tp->t_flags |= TF_LRD; 2318 2319 return (tp); 2320 } 2321 2322 /* 2323 * Drop a TCP connection, reporting 2324 * the specified error. If connection is synchronized, 2325 * then send a RST to peer. 2326 */ 2327 struct tcpcb * 2328 tcp_drop(struct tcpcb *tp, int errno) 2329 { 2330 struct socket *so = tptosocket(tp); 2331 2332 NET_EPOCH_ASSERT(); 2333 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2334 2335 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2336 tcp_state_change(tp, TCPS_CLOSED); 2337 /* Don't use tcp_output() here due to possible recursion. */ 2338 (void)tcp_output_nodrop(tp); 2339 TCPSTAT_INC(tcps_drops); 2340 } else 2341 TCPSTAT_INC(tcps_conndrops); 2342 if (errno == ETIMEDOUT && tp->t_softerror) 2343 errno = tp->t_softerror; 2344 so->so_error = errno; 2345 return (tcp_close(tp)); 2346 } 2347 2348 void 2349 tcp_discardcb(struct tcpcb *tp) 2350 { 2351 struct inpcb *inp = tptoinpcb(tp); 2352 struct socket *so = tptosocket(tp); 2353 #ifdef INET6 2354 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2355 #endif 2356 2357 INP_WLOCK_ASSERT(inp); 2358 2359 tcp_timer_stop(tp); 2360 if (tp->t_fb->tfb_tcp_timer_stop_all) { 2361 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2362 } 2363 2364 /* free the reassembly queue, if any */ 2365 tcp_reass_flush(tp); 2366 2367 #ifdef TCP_OFFLOAD 2368 /* Disconnect offload device, if any. */ 2369 if (tp->t_flags & TF_TOE) 2370 tcp_offload_detach(tp); 2371 #endif 2372 2373 tcp_free_sackholes(tp); 2374 2375 #ifdef TCPPCAP 2376 /* Free the TCP PCAP queues. */ 2377 tcp_pcap_drain(&(tp->t_inpkts)); 2378 tcp_pcap_drain(&(tp->t_outpkts)); 2379 #endif 2380 2381 /* Allow the CC algorithm to clean up after itself. */ 2382 if (CC_ALGO(tp)->cb_destroy != NULL) 2383 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2384 CC_DATA(tp) = NULL; 2385 /* Detach from the CC algorithm */ 2386 cc_detach(tp); 2387 2388 #ifdef TCP_HHOOK 2389 khelp_destroy_osd(&tp->t_osd); 2390 #endif 2391 #ifdef STATS 2392 stats_blob_destroy(tp->t_stats); 2393 #endif 2394 2395 CC_ALGO(tp) = NULL; 2396 2397 #ifdef TCP_BLACKBOX 2398 tcp_log_tcpcbfini(tp); 2399 #endif 2400 TCPSTATES_DEC(tp->t_state); 2401 if (tp->t_fb->tfb_tcp_fb_fini) 2402 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2403 2404 /* 2405 * If we got enough samples through the srtt filter, 2406 * save the rtt and rttvar in the routing entry. 2407 * 'Enough' is arbitrarily defined as 4 rtt samples. 2408 * 4 samples is enough for the srtt filter to converge 2409 * to within enough % of the correct value; fewer samples 2410 * and we could save a bogus rtt. The danger is not high 2411 * as tcp quickly recovers from everything. 2412 * XXX: Works very well but needs some more statistics! 2413 * 2414 * XXXRRS: Updating must be after the stack fini() since 2415 * that may be converting some internal representation of 2416 * say srtt etc into the general one used by other stacks. 2417 * Lets also at least protect against the so being NULL 2418 * as RW stated below. 2419 */ 2420 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2421 struct hc_metrics_lite metrics; 2422 uint32_t ssthresh; 2423 2424 bzero(&metrics, sizeof(metrics)); 2425 /* 2426 * Update the ssthresh always when the conditions below 2427 * are satisfied. This gives us better new start value 2428 * for the congestion avoidance for new connections. 2429 * ssthresh is only set if packet loss occurred on a session. 2430 * 2431 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2432 * being torn down. Ideally this code would not use 'so'. 2433 */ 2434 ssthresh = tp->snd_ssthresh; 2435 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2436 /* 2437 * convert the limit from user data bytes to 2438 * packets then to packet data bytes. 2439 */ 2440 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2441 if (ssthresh < 2) 2442 ssthresh = 2; 2443 ssthresh *= (tp->t_maxseg + 2444 #ifdef INET6 2445 (isipv6 ? sizeof (struct ip6_hdr) + 2446 sizeof (struct tcphdr) : 2447 #endif 2448 sizeof (struct tcpiphdr) 2449 #ifdef INET6 2450 ) 2451 #endif 2452 ); 2453 } else 2454 ssthresh = 0; 2455 metrics.rmx_ssthresh = ssthresh; 2456 2457 metrics.rmx_rtt = tp->t_srtt; 2458 metrics.rmx_rttvar = tp->t_rttvar; 2459 metrics.rmx_cwnd = tp->snd_cwnd; 2460 metrics.rmx_sendpipe = 0; 2461 metrics.rmx_recvpipe = 0; 2462 2463 tcp_hc_update(&inp->inp_inc, &metrics); 2464 } 2465 2466 refcount_release(&tp->t_fb->tfb_refcnt); 2467 } 2468 2469 /* 2470 * Attempt to close a TCP control block, marking it as dropped, and freeing 2471 * the socket if we hold the only reference. 2472 */ 2473 struct tcpcb * 2474 tcp_close(struct tcpcb *tp) 2475 { 2476 struct inpcb *inp = tptoinpcb(tp); 2477 struct socket *so = tptosocket(tp); 2478 2479 INP_WLOCK_ASSERT(inp); 2480 2481 #ifdef TCP_OFFLOAD 2482 if (tp->t_state == TCPS_LISTEN) 2483 tcp_offload_listen_stop(tp); 2484 #endif 2485 /* 2486 * This releases the TFO pending counter resource for TFO listen 2487 * sockets as well as passively-created TFO sockets that transition 2488 * from SYN_RECEIVED to CLOSED. 2489 */ 2490 if (tp->t_tfo_pending) { 2491 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2492 tp->t_tfo_pending = NULL; 2493 } 2494 #ifdef TCPHPTS 2495 tcp_hpts_remove(inp); 2496 #endif 2497 in_pcbdrop(inp); 2498 TCPSTAT_INC(tcps_closed); 2499 if (tp->t_state != TCPS_CLOSED) 2500 tcp_state_change(tp, TCPS_CLOSED); 2501 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2502 soisdisconnected(so); 2503 if (inp->inp_flags & INP_SOCKREF) { 2504 inp->inp_flags &= ~INP_SOCKREF; 2505 INP_WUNLOCK(inp); 2506 sorele(so); 2507 return (NULL); 2508 } 2509 return (tp); 2510 } 2511 2512 /* 2513 * Notify a tcp user of an asynchronous error; 2514 * store error as soft error, but wake up user 2515 * (for now, won't do anything until can select for soft error). 2516 * 2517 * Do not wake up user since there currently is no mechanism for 2518 * reporting soft errors (yet - a kqueue filter may be added). 2519 */ 2520 static struct inpcb * 2521 tcp_notify(struct inpcb *inp, int error) 2522 { 2523 struct tcpcb *tp; 2524 2525 INP_WLOCK_ASSERT(inp); 2526 2527 tp = intotcpcb(inp); 2528 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2529 2530 /* 2531 * Ignore some errors if we are hooked up. 2532 * If connection hasn't completed, has retransmitted several times, 2533 * and receives a second error, give up now. This is better 2534 * than waiting a long time to establish a connection that 2535 * can never complete. 2536 */ 2537 if (tp->t_state == TCPS_ESTABLISHED && 2538 (error == EHOSTUNREACH || error == ENETUNREACH || 2539 error == EHOSTDOWN)) { 2540 if (inp->inp_route.ro_nh) { 2541 NH_FREE(inp->inp_route.ro_nh); 2542 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2543 } 2544 return (inp); 2545 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2546 tp->t_softerror) { 2547 tp = tcp_drop(tp, error); 2548 if (tp != NULL) 2549 return (inp); 2550 else 2551 return (NULL); 2552 } else { 2553 tp->t_softerror = error; 2554 return (inp); 2555 } 2556 #if 0 2557 wakeup( &so->so_timeo); 2558 sorwakeup(so); 2559 sowwakeup(so); 2560 #endif 2561 } 2562 2563 static int 2564 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2565 { 2566 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2567 INPLOOKUP_RLOCKPCB); 2568 struct xinpgen xig; 2569 struct inpcb *inp; 2570 int error; 2571 2572 if (req->newptr != NULL) 2573 return (EPERM); 2574 2575 if (req->oldptr == NULL) { 2576 int n; 2577 2578 n = V_tcbinfo.ipi_count + 2579 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2580 n += imax(n / 8, 10); 2581 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2582 return (0); 2583 } 2584 2585 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2586 return (error); 2587 2588 bzero(&xig, sizeof(xig)); 2589 xig.xig_len = sizeof xig; 2590 xig.xig_count = V_tcbinfo.ipi_count + 2591 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2592 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2593 xig.xig_sogen = so_gencnt; 2594 error = SYSCTL_OUT(req, &xig, sizeof xig); 2595 if (error) 2596 return (error); 2597 2598 error = syncache_pcblist(req); 2599 if (error) 2600 return (error); 2601 2602 while ((inp = inp_next(&inpi)) != NULL) { 2603 if (inp->inp_gencnt <= xig.xig_gen && 2604 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2605 struct xtcpcb xt; 2606 2607 tcp_inptoxtp(inp, &xt); 2608 error = SYSCTL_OUT(req, &xt, sizeof xt); 2609 if (error) { 2610 INP_RUNLOCK(inp); 2611 break; 2612 } else 2613 continue; 2614 } 2615 } 2616 2617 if (!error) { 2618 /* 2619 * Give the user an updated idea of our state. 2620 * If the generation differs from what we told 2621 * her before, she knows that something happened 2622 * while we were processing this request, and it 2623 * might be necessary to retry. 2624 */ 2625 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2626 xig.xig_sogen = so_gencnt; 2627 xig.xig_count = V_tcbinfo.ipi_count + 2628 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2629 error = SYSCTL_OUT(req, &xig, sizeof xig); 2630 } 2631 2632 return (error); 2633 } 2634 2635 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2636 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2637 NULL, 0, tcp_pcblist, "S,xtcpcb", 2638 "List of active TCP connections"); 2639 2640 #ifdef INET 2641 static int 2642 tcp_getcred(SYSCTL_HANDLER_ARGS) 2643 { 2644 struct xucred xuc; 2645 struct sockaddr_in addrs[2]; 2646 struct epoch_tracker et; 2647 struct inpcb *inp; 2648 int error; 2649 2650 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2651 if (error) 2652 return (error); 2653 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2654 if (error) 2655 return (error); 2656 NET_EPOCH_ENTER(et); 2657 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2658 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2659 NET_EPOCH_EXIT(et); 2660 if (inp != NULL) { 2661 if (error == 0) 2662 error = cr_canseeinpcb(req->td->td_ucred, inp); 2663 if (error == 0) 2664 cru2x(inp->inp_cred, &xuc); 2665 INP_RUNLOCK(inp); 2666 } else 2667 error = ENOENT; 2668 if (error == 0) 2669 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2670 return (error); 2671 } 2672 2673 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2674 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2675 0, 0, tcp_getcred, "S,xucred", 2676 "Get the xucred of a TCP connection"); 2677 #endif /* INET */ 2678 2679 #ifdef INET6 2680 static int 2681 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2682 { 2683 struct epoch_tracker et; 2684 struct xucred xuc; 2685 struct sockaddr_in6 addrs[2]; 2686 struct inpcb *inp; 2687 int error; 2688 #ifdef INET 2689 int mapped = 0; 2690 #endif 2691 2692 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2693 if (error) 2694 return (error); 2695 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2696 if (error) 2697 return (error); 2698 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2699 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2700 return (error); 2701 } 2702 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2703 #ifdef INET 2704 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2705 mapped = 1; 2706 else 2707 #endif 2708 return (EINVAL); 2709 } 2710 2711 NET_EPOCH_ENTER(et); 2712 #ifdef INET 2713 if (mapped == 1) 2714 inp = in_pcblookup(&V_tcbinfo, 2715 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2716 addrs[1].sin6_port, 2717 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2718 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2719 else 2720 #endif 2721 inp = in6_pcblookup(&V_tcbinfo, 2722 &addrs[1].sin6_addr, addrs[1].sin6_port, 2723 &addrs[0].sin6_addr, addrs[0].sin6_port, 2724 INPLOOKUP_RLOCKPCB, NULL); 2725 NET_EPOCH_EXIT(et); 2726 if (inp != NULL) { 2727 if (error == 0) 2728 error = cr_canseeinpcb(req->td->td_ucred, inp); 2729 if (error == 0) 2730 cru2x(inp->inp_cred, &xuc); 2731 INP_RUNLOCK(inp); 2732 } else 2733 error = ENOENT; 2734 if (error == 0) 2735 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2736 return (error); 2737 } 2738 2739 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2740 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2741 0, 0, tcp6_getcred, "S,xucred", 2742 "Get the xucred of a TCP6 connection"); 2743 #endif /* INET6 */ 2744 2745 #ifdef INET 2746 /* Path MTU to try next when a fragmentation-needed message is received. */ 2747 static inline int 2748 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2749 { 2750 int mtu = ntohs(icp->icmp_nextmtu); 2751 2752 /* If no alternative MTU was proposed, try the next smaller one. */ 2753 if (!mtu) 2754 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2755 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2756 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2757 2758 return (mtu); 2759 } 2760 2761 static void 2762 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2763 { 2764 struct ip *ip; 2765 struct tcphdr *th; 2766 struct inpcb *inp; 2767 struct tcpcb *tp; 2768 struct inpcb *(*notify)(struct inpcb *, int); 2769 struct in_conninfo inc; 2770 tcp_seq icmp_tcp_seq; 2771 int errno, mtu; 2772 2773 errno = icmp_errmap(icp); 2774 switch (errno) { 2775 case 0: 2776 return; 2777 case EMSGSIZE: 2778 notify = tcp_mtudisc_notify; 2779 break; 2780 case ECONNREFUSED: 2781 if (V_icmp_may_rst) 2782 notify = tcp_drop_syn_sent; 2783 else 2784 notify = tcp_notify; 2785 break; 2786 case EHOSTUNREACH: 2787 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2788 notify = tcp_drop_syn_sent; 2789 else 2790 notify = tcp_notify; 2791 break; 2792 default: 2793 notify = tcp_notify; 2794 } 2795 2796 ip = &icp->icmp_ip; 2797 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2798 icmp_tcp_seq = th->th_seq; 2799 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2800 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2801 if (inp != NULL) { 2802 tp = intotcpcb(inp); 2803 #ifdef TCP_OFFLOAD 2804 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2805 /* 2806 * MTU discovery for offloaded connections. Let 2807 * the TOE driver verify seq# and process it. 2808 */ 2809 mtu = tcp_next_pmtu(icp, ip); 2810 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2811 goto out; 2812 } 2813 #endif 2814 if (tp->t_port != port) 2815 goto out; 2816 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2817 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2818 if (errno == EMSGSIZE) { 2819 /* 2820 * MTU discovery: we got a needfrag and 2821 * will potentially try a lower MTU. 2822 */ 2823 mtu = tcp_next_pmtu(icp, ip); 2824 2825 /* 2826 * Only process the offered MTU if it 2827 * is smaller than the current one. 2828 */ 2829 if (mtu < tp->t_maxseg + 2830 sizeof(struct tcpiphdr)) { 2831 bzero(&inc, sizeof(inc)); 2832 inc.inc_faddr = ip->ip_dst; 2833 inc.inc_fibnum = 2834 inp->inp_inc.inc_fibnum; 2835 tcp_hc_updatemtu(&inc, mtu); 2836 inp = tcp_mtudisc(inp, mtu); 2837 } 2838 } else 2839 inp = (*notify)(inp, errno); 2840 } 2841 } else { 2842 bzero(&inc, sizeof(inc)); 2843 inc.inc_fport = th->th_dport; 2844 inc.inc_lport = th->th_sport; 2845 inc.inc_faddr = ip->ip_dst; 2846 inc.inc_laddr = ip->ip_src; 2847 syncache_unreach(&inc, icmp_tcp_seq, port); 2848 } 2849 out: 2850 if (inp != NULL) 2851 INP_WUNLOCK(inp); 2852 } 2853 2854 static void 2855 tcp_ctlinput(struct icmp *icmp) 2856 { 2857 tcp_ctlinput_with_port(icmp, htons(0)); 2858 } 2859 2860 static void 2861 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2862 { 2863 /* Its a tunneled TCP over UDP icmp */ 2864 struct icmp *icmp = param.icmp; 2865 struct ip *outer_ip, *inner_ip; 2866 struct udphdr *udp; 2867 struct tcphdr *th, ttemp; 2868 int i_hlen, o_len; 2869 uint16_t port; 2870 2871 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2872 inner_ip = &icmp->icmp_ip; 2873 i_hlen = inner_ip->ip_hl << 2; 2874 o_len = ntohs(outer_ip->ip_len); 2875 if (o_len < 2876 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2877 /* Not enough data present */ 2878 return; 2879 } 2880 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2881 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2882 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2883 return; 2884 } 2885 port = udp->uh_dport; 2886 th = (struct tcphdr *)(udp + 1); 2887 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2888 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2889 /* Now adjust down the size of the outer IP header */ 2890 o_len -= sizeof(struct udphdr); 2891 outer_ip->ip_len = htons(o_len); 2892 /* Now call in to the normal handling code */ 2893 tcp_ctlinput_with_port(icmp, port); 2894 } 2895 #endif /* INET */ 2896 2897 #ifdef INET6 2898 static inline int 2899 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2900 { 2901 int mtu = ntohl(icmp6->icmp6_mtu); 2902 2903 /* 2904 * If no alternative MTU was proposed, or the proposed MTU was too 2905 * small, set to the min. 2906 */ 2907 if (mtu < IPV6_MMTU) 2908 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2909 return (mtu); 2910 } 2911 2912 static void 2913 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2914 { 2915 struct in6_addr *dst; 2916 struct inpcb *(*notify)(struct inpcb *, int); 2917 struct ip6_hdr *ip6; 2918 struct mbuf *m; 2919 struct inpcb *inp; 2920 struct tcpcb *tp; 2921 struct icmp6_hdr *icmp6; 2922 struct in_conninfo inc; 2923 struct tcp_ports { 2924 uint16_t th_sport; 2925 uint16_t th_dport; 2926 } t_ports; 2927 tcp_seq icmp_tcp_seq; 2928 unsigned int mtu; 2929 unsigned int off; 2930 int errno; 2931 2932 icmp6 = ip6cp->ip6c_icmp6; 2933 m = ip6cp->ip6c_m; 2934 ip6 = ip6cp->ip6c_ip6; 2935 off = ip6cp->ip6c_off; 2936 dst = &ip6cp->ip6c_finaldst->sin6_addr; 2937 2938 errno = icmp6_errmap(icmp6); 2939 switch (errno) { 2940 case 0: 2941 return; 2942 case EMSGSIZE: 2943 notify = tcp_mtudisc_notify; 2944 break; 2945 case ECONNREFUSED: 2946 if (V_icmp_may_rst) 2947 notify = tcp_drop_syn_sent; 2948 else 2949 notify = tcp_notify; 2950 break; 2951 case EHOSTUNREACH: 2952 /* 2953 * There are only four ICMPs that may reset connection: 2954 * - administratively prohibited 2955 * - port unreachable 2956 * - time exceeded in transit 2957 * - unknown next header 2958 */ 2959 if (V_icmp_may_rst && 2960 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 2961 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 2962 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 2963 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 2964 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 2965 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 2966 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 2967 notify = tcp_drop_syn_sent; 2968 else 2969 notify = tcp_notify; 2970 break; 2971 default: 2972 notify = tcp_notify; 2973 } 2974 2975 /* Check if we can safely get the ports from the tcp hdr */ 2976 if (m == NULL || 2977 (m->m_pkthdr.len < 2978 (int32_t) (off + sizeof(struct tcp_ports)))) { 2979 return; 2980 } 2981 bzero(&t_ports, sizeof(struct tcp_ports)); 2982 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2983 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2984 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2985 off += sizeof(struct tcp_ports); 2986 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2987 goto out; 2988 } 2989 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2990 if (inp != NULL) { 2991 tp = intotcpcb(inp); 2992 #ifdef TCP_OFFLOAD 2993 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2994 /* MTU discovery for offloaded connections. */ 2995 mtu = tcp6_next_pmtu(icmp6); 2996 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2997 goto out; 2998 } 2999 #endif 3000 if (tp->t_port != port) 3001 goto out; 3002 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3003 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3004 if (errno == EMSGSIZE) { 3005 /* 3006 * MTU discovery: 3007 * If we got a needfrag set the MTU 3008 * in the route to the suggested new 3009 * value (if given) and then notify. 3010 */ 3011 mtu = tcp6_next_pmtu(icmp6); 3012 3013 bzero(&inc, sizeof(inc)); 3014 inc.inc_fibnum = M_GETFIB(m); 3015 inc.inc_flags |= INC_ISIPV6; 3016 inc.inc6_faddr = *dst; 3017 if (in6_setscope(&inc.inc6_faddr, 3018 m->m_pkthdr.rcvif, NULL)) 3019 goto out; 3020 /* 3021 * Only process the offered MTU if it 3022 * is smaller than the current one. 3023 */ 3024 if (mtu < tp->t_maxseg + 3025 sizeof (struct tcphdr) + 3026 sizeof (struct ip6_hdr)) { 3027 tcp_hc_updatemtu(&inc, mtu); 3028 tcp_mtudisc(inp, mtu); 3029 ICMP6STAT_INC(icp6s_pmtuchg); 3030 } 3031 } else 3032 inp = (*notify)(inp, errno); 3033 } 3034 } else { 3035 bzero(&inc, sizeof(inc)); 3036 inc.inc_fibnum = M_GETFIB(m); 3037 inc.inc_flags |= INC_ISIPV6; 3038 inc.inc_fport = t_ports.th_dport; 3039 inc.inc_lport = t_ports.th_sport; 3040 inc.inc6_faddr = *dst; 3041 inc.inc6_laddr = ip6->ip6_src; 3042 syncache_unreach(&inc, icmp_tcp_seq, port); 3043 } 3044 out: 3045 if (inp != NULL) 3046 INP_WUNLOCK(inp); 3047 } 3048 3049 static void 3050 tcp6_ctlinput(struct ip6ctlparam *ctl) 3051 { 3052 tcp6_ctlinput_with_port(ctl, htons(0)); 3053 } 3054 3055 static void 3056 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3057 { 3058 struct ip6ctlparam *ip6cp = param.ip6cp; 3059 struct mbuf *m; 3060 struct udphdr *udp; 3061 uint16_t port; 3062 3063 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3064 if (m == NULL) { 3065 return; 3066 } 3067 udp = mtod(m, struct udphdr *); 3068 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3069 return; 3070 } 3071 port = udp->uh_dport; 3072 m_adj(m, sizeof(struct udphdr)); 3073 if ((m->m_flags & M_PKTHDR) == 0) { 3074 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3075 } 3076 /* Now call in to the normal handling code */ 3077 tcp6_ctlinput_with_port(ip6cp, port); 3078 } 3079 3080 #endif /* INET6 */ 3081 3082 static uint32_t 3083 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3084 { 3085 SIPHASH_CTX ctx; 3086 uint32_t hash[2]; 3087 3088 KASSERT(len >= SIPHASH_KEY_LENGTH, 3089 ("%s: keylen %u too short ", __func__, len)); 3090 SipHash24_Init(&ctx); 3091 SipHash_SetKey(&ctx, (uint8_t *)key); 3092 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3093 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3094 switch (inc->inc_flags & INC_ISIPV6) { 3095 #ifdef INET 3096 case 0: 3097 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3098 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3099 break; 3100 #endif 3101 #ifdef INET6 3102 case INC_ISIPV6: 3103 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3104 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3105 break; 3106 #endif 3107 } 3108 SipHash_Final((uint8_t *)hash, &ctx); 3109 3110 return (hash[0] ^ hash[1]); 3111 } 3112 3113 uint32_t 3114 tcp_new_ts_offset(struct in_conninfo *inc) 3115 { 3116 struct in_conninfo inc_store, *local_inc; 3117 3118 if (!V_tcp_ts_offset_per_conn) { 3119 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3120 inc_store.inc_lport = 0; 3121 inc_store.inc_fport = 0; 3122 local_inc = &inc_store; 3123 } else { 3124 local_inc = inc; 3125 } 3126 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3127 sizeof(V_ts_offset_secret))); 3128 } 3129 3130 /* 3131 * Following is where TCP initial sequence number generation occurs. 3132 * 3133 * There are two places where we must use initial sequence numbers: 3134 * 1. In SYN-ACK packets. 3135 * 2. In SYN packets. 3136 * 3137 * All ISNs for SYN-ACK packets are generated by the syncache. See 3138 * tcp_syncache.c for details. 3139 * 3140 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3141 * depends on this property. In addition, these ISNs should be 3142 * unguessable so as to prevent connection hijacking. To satisfy 3143 * the requirements of this situation, the algorithm outlined in 3144 * RFC 1948 is used, with only small modifications. 3145 * 3146 * Implementation details: 3147 * 3148 * Time is based off the system timer, and is corrected so that it 3149 * increases by one megabyte per second. This allows for proper 3150 * recycling on high speed LANs while still leaving over an hour 3151 * before rollover. 3152 * 3153 * As reading the *exact* system time is too expensive to be done 3154 * whenever setting up a TCP connection, we increment the time 3155 * offset in two ways. First, a small random positive increment 3156 * is added to isn_offset for each connection that is set up. 3157 * Second, the function tcp_isn_tick fires once per clock tick 3158 * and increments isn_offset as necessary so that sequence numbers 3159 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3160 * random positive increments serve only to ensure that the same 3161 * exact sequence number is never sent out twice (as could otherwise 3162 * happen when a port is recycled in less than the system tick 3163 * interval.) 3164 * 3165 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3166 * between seeding of isn_secret. This is normally set to zero, 3167 * as reseeding should not be necessary. 3168 * 3169 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3170 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3171 * general, this means holding an exclusive (write) lock. 3172 */ 3173 3174 #define ISN_BYTES_PER_SECOND 1048576 3175 #define ISN_STATIC_INCREMENT 4096 3176 #define ISN_RANDOM_INCREMENT (4096 - 1) 3177 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3178 3179 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3180 VNET_DEFINE_STATIC(int, isn_last); 3181 VNET_DEFINE_STATIC(int, isn_last_reseed); 3182 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3183 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3184 3185 #define V_isn_secret VNET(isn_secret) 3186 #define V_isn_last VNET(isn_last) 3187 #define V_isn_last_reseed VNET(isn_last_reseed) 3188 #define V_isn_offset VNET(isn_offset) 3189 #define V_isn_offset_old VNET(isn_offset_old) 3190 3191 tcp_seq 3192 tcp_new_isn(struct in_conninfo *inc) 3193 { 3194 tcp_seq new_isn; 3195 u_int32_t projected_offset; 3196 3197 ISN_LOCK(); 3198 /* Seed if this is the first use, reseed if requested. */ 3199 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3200 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3201 < (u_int)ticks))) { 3202 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3203 V_isn_last_reseed = ticks; 3204 } 3205 3206 /* Compute the hash and return the ISN. */ 3207 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3208 sizeof(V_isn_secret)); 3209 V_isn_offset += ISN_STATIC_INCREMENT + 3210 (arc4random() & ISN_RANDOM_INCREMENT); 3211 if (ticks != V_isn_last) { 3212 projected_offset = V_isn_offset_old + 3213 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3214 if (SEQ_GT(projected_offset, V_isn_offset)) 3215 V_isn_offset = projected_offset; 3216 V_isn_offset_old = V_isn_offset; 3217 V_isn_last = ticks; 3218 } 3219 new_isn += V_isn_offset; 3220 ISN_UNLOCK(); 3221 return (new_isn); 3222 } 3223 3224 /* 3225 * When a specific ICMP unreachable message is received and the 3226 * connection state is SYN-SENT, drop the connection. This behavior 3227 * is controlled by the icmp_may_rst sysctl. 3228 */ 3229 static struct inpcb * 3230 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3231 { 3232 struct tcpcb *tp; 3233 3234 NET_EPOCH_ASSERT(); 3235 INP_WLOCK_ASSERT(inp); 3236 3237 tp = intotcpcb(inp); 3238 if (tp->t_state != TCPS_SYN_SENT) 3239 return (inp); 3240 3241 if (IS_FASTOPEN(tp->t_flags)) 3242 tcp_fastopen_disable_path(tp); 3243 3244 tp = tcp_drop(tp, errno); 3245 if (tp != NULL) 3246 return (inp); 3247 else 3248 return (NULL); 3249 } 3250 3251 /* 3252 * When `need fragmentation' ICMP is received, update our idea of the MSS 3253 * based on the new value. Also nudge TCP to send something, since we 3254 * know the packet we just sent was dropped. 3255 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3256 */ 3257 static struct inpcb * 3258 tcp_mtudisc_notify(struct inpcb *inp, int error) 3259 { 3260 3261 return (tcp_mtudisc(inp, -1)); 3262 } 3263 3264 static struct inpcb * 3265 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3266 { 3267 struct tcpcb *tp; 3268 struct socket *so; 3269 3270 INP_WLOCK_ASSERT(inp); 3271 3272 tp = intotcpcb(inp); 3273 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3274 3275 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3276 3277 so = inp->inp_socket; 3278 SOCKBUF_LOCK(&so->so_snd); 3279 /* If the mss is larger than the socket buffer, decrease the mss. */ 3280 if (so->so_snd.sb_hiwat < tp->t_maxseg) 3281 tp->t_maxseg = so->so_snd.sb_hiwat; 3282 SOCKBUF_UNLOCK(&so->so_snd); 3283 3284 TCPSTAT_INC(tcps_mturesent); 3285 tp->t_rtttime = 0; 3286 tp->snd_nxt = tp->snd_una; 3287 tcp_free_sackholes(tp); 3288 tp->snd_recover = tp->snd_max; 3289 if (tp->t_flags & TF_SACK_PERMIT) 3290 EXIT_FASTRECOVERY(tp->t_flags); 3291 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3292 /* 3293 * Conceptually the snd_nxt setting 3294 * and freeing sack holes should 3295 * be done by the default stacks 3296 * own tfb_tcp_mtu_chg(). 3297 */ 3298 tp->t_fb->tfb_tcp_mtu_chg(tp); 3299 } 3300 if (tcp_output(tp) < 0) 3301 return (NULL); 3302 else 3303 return (inp); 3304 } 3305 3306 #ifdef INET 3307 /* 3308 * Look-up the routing entry to the peer of this inpcb. If no route 3309 * is found and it cannot be allocated, then return 0. This routine 3310 * is called by TCP routines that access the rmx structure and by 3311 * tcp_mss_update to get the peer/interface MTU. 3312 */ 3313 uint32_t 3314 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3315 { 3316 struct nhop_object *nh; 3317 struct ifnet *ifp; 3318 uint32_t maxmtu = 0; 3319 3320 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3321 3322 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3323 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3324 if (nh == NULL) 3325 return (0); 3326 3327 ifp = nh->nh_ifp; 3328 maxmtu = nh->nh_mtu; 3329 3330 /* Report additional interface capabilities. */ 3331 if (cap != NULL) { 3332 if (ifp->if_capenable & IFCAP_TSO4 && 3333 ifp->if_hwassist & CSUM_TSO) { 3334 cap->ifcap |= CSUM_TSO; 3335 cap->tsomax = ifp->if_hw_tsomax; 3336 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3337 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3338 } 3339 } 3340 } 3341 return (maxmtu); 3342 } 3343 #endif /* INET */ 3344 3345 #ifdef INET6 3346 uint32_t 3347 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3348 { 3349 struct nhop_object *nh; 3350 struct in6_addr dst6; 3351 uint32_t scopeid; 3352 struct ifnet *ifp; 3353 uint32_t maxmtu = 0; 3354 3355 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3356 3357 if (inc->inc_flags & INC_IPV6MINMTU) 3358 return (IPV6_MMTU); 3359 3360 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3361 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3362 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3363 if (nh == NULL) 3364 return (0); 3365 3366 ifp = nh->nh_ifp; 3367 maxmtu = nh->nh_mtu; 3368 3369 /* Report additional interface capabilities. */ 3370 if (cap != NULL) { 3371 if (ifp->if_capenable & IFCAP_TSO6 && 3372 ifp->if_hwassist & CSUM_TSO) { 3373 cap->ifcap |= CSUM_TSO; 3374 cap->tsomax = ifp->if_hw_tsomax; 3375 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3376 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3377 } 3378 } 3379 } 3380 3381 return (maxmtu); 3382 } 3383 3384 /* 3385 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3386 * 3387 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3388 * The right place to do that is ip6_setpktopt() that has just been 3389 * executed. By the way it just filled ip6po_minmtu for us. 3390 */ 3391 void 3392 tcp6_use_min_mtu(struct tcpcb *tp) 3393 { 3394 struct inpcb *inp = tptoinpcb(tp); 3395 3396 INP_WLOCK_ASSERT(inp); 3397 /* 3398 * In case of the IPV6_USE_MIN_MTU socket 3399 * option, the INC_IPV6MINMTU flag to announce 3400 * a corresponding MSS during the initial 3401 * handshake. If the TCP connection is not in 3402 * the front states, just reduce the MSS being 3403 * used. This avoids the sending of TCP 3404 * segments which will be fragmented at the 3405 * IPv6 layer. 3406 */ 3407 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3408 if ((tp->t_state >= TCPS_SYN_SENT) && 3409 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3410 struct ip6_pktopts *opt; 3411 3412 opt = inp->in6p_outputopts; 3413 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3414 tp->t_maxseg > TCP6_MSS) 3415 tp->t_maxseg = TCP6_MSS; 3416 } 3417 } 3418 #endif /* INET6 */ 3419 3420 /* 3421 * Calculate effective SMSS per RFC5681 definition for a given TCP 3422 * connection at its current state, taking into account SACK and etc. 3423 */ 3424 u_int 3425 tcp_maxseg(const struct tcpcb *tp) 3426 { 3427 u_int optlen; 3428 3429 if (tp->t_flags & TF_NOOPT) 3430 return (tp->t_maxseg); 3431 3432 /* 3433 * Here we have a simplified code from tcp_addoptions(), 3434 * without a proper loop, and having most of paddings hardcoded. 3435 * We might make mistakes with padding here in some edge cases, 3436 * but this is harmless, since result of tcp_maxseg() is used 3437 * only in cwnd and ssthresh estimations. 3438 */ 3439 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3440 if (tp->t_flags & TF_RCVD_TSTMP) 3441 optlen = TCPOLEN_TSTAMP_APPA; 3442 else 3443 optlen = 0; 3444 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3445 if (tp->t_flags & TF_SIGNATURE) 3446 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3447 #endif 3448 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3449 optlen += TCPOLEN_SACKHDR; 3450 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3451 optlen = PADTCPOLEN(optlen); 3452 } 3453 } else { 3454 if (tp->t_flags & TF_REQ_TSTMP) 3455 optlen = TCPOLEN_TSTAMP_APPA; 3456 else 3457 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3458 if (tp->t_flags & TF_REQ_SCALE) 3459 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3460 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3461 if (tp->t_flags & TF_SIGNATURE) 3462 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3463 #endif 3464 if (tp->t_flags & TF_SACK_PERMIT) 3465 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3466 } 3467 #undef PAD 3468 optlen = min(optlen, TCP_MAXOLEN); 3469 return (tp->t_maxseg - optlen); 3470 } 3471 3472 3473 u_int 3474 tcp_fixed_maxseg(const struct tcpcb *tp) 3475 { 3476 int optlen; 3477 3478 if (tp->t_flags & TF_NOOPT) 3479 return (tp->t_maxseg); 3480 3481 /* 3482 * Here we have a simplified code from tcp_addoptions(), 3483 * without a proper loop, and having most of paddings hardcoded. 3484 * We only consider fixed options that we would send every 3485 * time I.e. SACK is not considered. This is important 3486 * for cc modules to figure out what the modulo of the 3487 * cwnd should be. 3488 */ 3489 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3490 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3491 if (tp->t_flags & TF_RCVD_TSTMP) 3492 optlen = TCPOLEN_TSTAMP_APPA; 3493 else 3494 optlen = 0; 3495 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3496 if (tp->t_flags & TF_SIGNATURE) 3497 optlen += PAD(TCPOLEN_SIGNATURE); 3498 #endif 3499 } else { 3500 if (tp->t_flags & TF_REQ_TSTMP) 3501 optlen = TCPOLEN_TSTAMP_APPA; 3502 else 3503 optlen = PAD(TCPOLEN_MAXSEG); 3504 if (tp->t_flags & TF_REQ_SCALE) 3505 optlen += PAD(TCPOLEN_WINDOW); 3506 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3507 if (tp->t_flags & TF_SIGNATURE) 3508 optlen += PAD(TCPOLEN_SIGNATURE); 3509 #endif 3510 if (tp->t_flags & TF_SACK_PERMIT) 3511 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3512 } 3513 #undef PAD 3514 optlen = min(optlen, TCP_MAXOLEN); 3515 return (tp->t_maxseg - optlen); 3516 } 3517 3518 3519 3520 static int 3521 sysctl_drop(SYSCTL_HANDLER_ARGS) 3522 { 3523 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3524 struct sockaddr_storage addrs[2]; 3525 struct inpcb *inp; 3526 struct tcpcb *tp; 3527 #ifdef INET 3528 struct sockaddr_in *fin = NULL, *lin = NULL; 3529 #endif 3530 struct epoch_tracker et; 3531 #ifdef INET6 3532 struct sockaddr_in6 *fin6, *lin6; 3533 #endif 3534 int error; 3535 3536 inp = NULL; 3537 #ifdef INET6 3538 fin6 = lin6 = NULL; 3539 #endif 3540 error = 0; 3541 3542 if (req->oldptr != NULL || req->oldlen != 0) 3543 return (EINVAL); 3544 if (req->newptr == NULL) 3545 return (EPERM); 3546 if (req->newlen < sizeof(addrs)) 3547 return (ENOMEM); 3548 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3549 if (error) 3550 return (error); 3551 3552 switch (addrs[0].ss_family) { 3553 #ifdef INET6 3554 case AF_INET6: 3555 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3556 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3557 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3558 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3559 return (EINVAL); 3560 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3561 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3562 return (EINVAL); 3563 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3564 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3565 #ifdef INET 3566 fin = (struct sockaddr_in *)&addrs[0]; 3567 lin = (struct sockaddr_in *)&addrs[1]; 3568 #endif 3569 break; 3570 } 3571 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3572 if (error) 3573 return (error); 3574 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3575 if (error) 3576 return (error); 3577 break; 3578 #endif 3579 #ifdef INET 3580 case AF_INET: 3581 fin = (struct sockaddr_in *)&addrs[0]; 3582 lin = (struct sockaddr_in *)&addrs[1]; 3583 if (fin->sin_len != sizeof(struct sockaddr_in) || 3584 lin->sin_len != sizeof(struct sockaddr_in)) 3585 return (EINVAL); 3586 break; 3587 #endif 3588 default: 3589 return (EINVAL); 3590 } 3591 NET_EPOCH_ENTER(et); 3592 switch (addrs[0].ss_family) { 3593 #ifdef INET6 3594 case AF_INET6: 3595 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3596 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3597 INPLOOKUP_WLOCKPCB, NULL); 3598 break; 3599 #endif 3600 #ifdef INET 3601 case AF_INET: 3602 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3603 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3604 break; 3605 #endif 3606 } 3607 if (inp != NULL) { 3608 if (!SOLISTENING(inp->inp_socket)) { 3609 tp = intotcpcb(inp); 3610 tp = tcp_drop(tp, ECONNABORTED); 3611 if (tp != NULL) 3612 INP_WUNLOCK(inp); 3613 } else 3614 INP_WUNLOCK(inp); 3615 } else 3616 error = ESRCH; 3617 NET_EPOCH_EXIT(et); 3618 return (error); 3619 } 3620 3621 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3622 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3623 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3624 "Drop TCP connection"); 3625 3626 static int 3627 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3628 { 3629 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3630 &tcp_ctloutput_set)); 3631 } 3632 3633 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3634 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3635 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3636 "Set socket option for TCP endpoint"); 3637 3638 #ifdef KERN_TLS 3639 static int 3640 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3641 { 3642 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3643 struct sockaddr_storage addrs[2]; 3644 struct inpcb *inp; 3645 #ifdef INET 3646 struct sockaddr_in *fin = NULL, *lin = NULL; 3647 #endif 3648 struct epoch_tracker et; 3649 #ifdef INET6 3650 struct sockaddr_in6 *fin6, *lin6; 3651 #endif 3652 int error; 3653 3654 inp = NULL; 3655 #ifdef INET6 3656 fin6 = lin6 = NULL; 3657 #endif 3658 error = 0; 3659 3660 if (req->oldptr != NULL || req->oldlen != 0) 3661 return (EINVAL); 3662 if (req->newptr == NULL) 3663 return (EPERM); 3664 if (req->newlen < sizeof(addrs)) 3665 return (ENOMEM); 3666 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3667 if (error) 3668 return (error); 3669 3670 switch (addrs[0].ss_family) { 3671 #ifdef INET6 3672 case AF_INET6: 3673 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3674 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3675 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3676 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3677 return (EINVAL); 3678 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3679 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3680 return (EINVAL); 3681 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3682 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3683 #ifdef INET 3684 fin = (struct sockaddr_in *)&addrs[0]; 3685 lin = (struct sockaddr_in *)&addrs[1]; 3686 #endif 3687 break; 3688 } 3689 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3690 if (error) 3691 return (error); 3692 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3693 if (error) 3694 return (error); 3695 break; 3696 #endif 3697 #ifdef INET 3698 case AF_INET: 3699 fin = (struct sockaddr_in *)&addrs[0]; 3700 lin = (struct sockaddr_in *)&addrs[1]; 3701 if (fin->sin_len != sizeof(struct sockaddr_in) || 3702 lin->sin_len != sizeof(struct sockaddr_in)) 3703 return (EINVAL); 3704 break; 3705 #endif 3706 default: 3707 return (EINVAL); 3708 } 3709 NET_EPOCH_ENTER(et); 3710 switch (addrs[0].ss_family) { 3711 #ifdef INET6 3712 case AF_INET6: 3713 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3714 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3715 INPLOOKUP_WLOCKPCB, NULL); 3716 break; 3717 #endif 3718 #ifdef INET 3719 case AF_INET: 3720 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3721 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3722 break; 3723 #endif 3724 } 3725 NET_EPOCH_EXIT(et); 3726 if (inp != NULL) { 3727 struct socket *so; 3728 3729 so = inp->inp_socket; 3730 soref(so); 3731 error = ktls_set_tx_mode(so, 3732 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3733 INP_WUNLOCK(inp); 3734 sorele(so); 3735 } else 3736 error = ESRCH; 3737 return (error); 3738 } 3739 3740 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3741 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3742 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3743 "Switch TCP connection to SW TLS"); 3744 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3745 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3746 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3747 "Switch TCP connection to ifnet TLS"); 3748 #endif 3749 3750 /* 3751 * Generate a standardized TCP log line for use throughout the 3752 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3753 * allow use in the interrupt context. 3754 * 3755 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3756 * NB: The function may return NULL if memory allocation failed. 3757 * 3758 * Due to header inclusion and ordering limitations the struct ip 3759 * and ip6_hdr pointers have to be passed as void pointers. 3760 */ 3761 char * 3762 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3763 const void *ip6hdr) 3764 { 3765 3766 /* Is logging enabled? */ 3767 if (V_tcp_log_in_vain == 0) 3768 return (NULL); 3769 3770 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3771 } 3772 3773 char * 3774 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3775 const void *ip6hdr) 3776 { 3777 3778 /* Is logging enabled? */ 3779 if (tcp_log_debug == 0) 3780 return (NULL); 3781 3782 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3783 } 3784 3785 static char * 3786 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3787 const void *ip6hdr) 3788 { 3789 char *s, *sp; 3790 size_t size; 3791 #ifdef INET 3792 const struct ip *ip = (const struct ip *)ip4hdr; 3793 #endif 3794 #ifdef INET6 3795 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3796 #endif /* INET6 */ 3797 3798 /* 3799 * The log line looks like this: 3800 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3801 */ 3802 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3803 sizeof(PRINT_TH_FLAGS) + 1 + 3804 #ifdef INET6 3805 2 * INET6_ADDRSTRLEN; 3806 #else 3807 2 * INET_ADDRSTRLEN; 3808 #endif /* INET6 */ 3809 3810 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3811 if (s == NULL) 3812 return (NULL); 3813 3814 strcat(s, "TCP: ["); 3815 sp = s + strlen(s); 3816 3817 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3818 inet_ntoa_r(inc->inc_faddr, sp); 3819 sp = s + strlen(s); 3820 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3821 sp = s + strlen(s); 3822 inet_ntoa_r(inc->inc_laddr, sp); 3823 sp = s + strlen(s); 3824 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3825 #ifdef INET6 3826 } else if (inc) { 3827 ip6_sprintf(sp, &inc->inc6_faddr); 3828 sp = s + strlen(s); 3829 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3830 sp = s + strlen(s); 3831 ip6_sprintf(sp, &inc->inc6_laddr); 3832 sp = s + strlen(s); 3833 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3834 } else if (ip6 && th) { 3835 ip6_sprintf(sp, &ip6->ip6_src); 3836 sp = s + strlen(s); 3837 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3838 sp = s + strlen(s); 3839 ip6_sprintf(sp, &ip6->ip6_dst); 3840 sp = s + strlen(s); 3841 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3842 #endif /* INET6 */ 3843 #ifdef INET 3844 } else if (ip && th) { 3845 inet_ntoa_r(ip->ip_src, sp); 3846 sp = s + strlen(s); 3847 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3848 sp = s + strlen(s); 3849 inet_ntoa_r(ip->ip_dst, sp); 3850 sp = s + strlen(s); 3851 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3852 #endif /* INET */ 3853 } else { 3854 free(s, M_TCPLOG); 3855 return (NULL); 3856 } 3857 sp = s + strlen(s); 3858 if (th) 3859 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3860 if (*(s + size - 1) != '\0') 3861 panic("%s: string too long", __func__); 3862 return (s); 3863 } 3864 3865 /* 3866 * A subroutine which makes it easy to track TCP state changes with DTrace. 3867 * This function shouldn't be called for t_state initializations that don't 3868 * correspond to actual TCP state transitions. 3869 */ 3870 void 3871 tcp_state_change(struct tcpcb *tp, int newstate) 3872 { 3873 #if defined(KDTRACE_HOOKS) 3874 int pstate = tp->t_state; 3875 #endif 3876 3877 TCPSTATES_DEC(tp->t_state); 3878 TCPSTATES_INC(newstate); 3879 tp->t_state = newstate; 3880 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3881 } 3882 3883 /* 3884 * Create an external-format (``xtcpcb'') structure using the information in 3885 * the kernel-format tcpcb structure pointed to by tp. This is done to 3886 * reduce the spew of irrelevant information over this interface, to isolate 3887 * user code from changes in the kernel structure, and potentially to provide 3888 * information-hiding if we decide that some of this information should be 3889 * hidden from users. 3890 */ 3891 void 3892 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3893 { 3894 struct tcpcb *tp = intotcpcb(inp); 3895 sbintime_t now; 3896 3897 bzero(xt, sizeof(*xt)); 3898 xt->t_state = tp->t_state; 3899 xt->t_logstate = tcp_get_bblog_state(tp); 3900 xt->t_flags = tp->t_flags; 3901 xt->t_sndzerowin = tp->t_sndzerowin; 3902 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3903 xt->t_rcvoopack = tp->t_rcvoopack; 3904 xt->t_rcv_wnd = tp->rcv_wnd; 3905 xt->t_snd_wnd = tp->snd_wnd; 3906 xt->t_snd_cwnd = tp->snd_cwnd; 3907 xt->t_snd_ssthresh = tp->snd_ssthresh; 3908 xt->t_dsack_bytes = tp->t_dsack_bytes; 3909 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 3910 xt->t_dsack_pack = tp->t_dsack_pack; 3911 xt->t_maxseg = tp->t_maxseg; 3912 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 3913 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 3914 3915 now = getsbinuptime(); 3916 #define COPYTIMER(which,where) do { \ 3917 if (tp->t_timers[which] != SBT_MAX) \ 3918 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 3919 else \ 3920 xt->where = 0; \ 3921 } while (0) 3922 COPYTIMER(TT_DELACK, tt_delack); 3923 COPYTIMER(TT_REXMT, tt_rexmt); 3924 COPYTIMER(TT_PERSIST, tt_persist); 3925 COPYTIMER(TT_KEEP, tt_keep); 3926 COPYTIMER(TT_2MSL, tt_2msl); 3927 #undef COPYTIMER 3928 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3929 3930 xt->xt_encaps_port = tp->t_port; 3931 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3932 TCP_FUNCTION_NAME_LEN_MAX); 3933 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 3934 #ifdef TCP_BLACKBOX 3935 (void)tcp_log_get_id(tp, xt->xt_logid); 3936 #endif 3937 3938 xt->xt_len = sizeof(struct xtcpcb); 3939 in_pcbtoxinpcb(inp, &xt->xt_inp); 3940 } 3941 3942 void 3943 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 3944 { 3945 uint32_t bit, i; 3946 3947 if ((tp == NULL) || 3948 (status > TCP_EI_STATUS_MAX_VALUE) || 3949 (status == 0)) { 3950 /* Invalid */ 3951 return; 3952 } 3953 if (status > (sizeof(uint32_t) * 8)) { 3954 /* Should this be a KASSERT? */ 3955 return; 3956 } 3957 bit = 1U << (status - 1); 3958 if (bit & tp->t_end_info_status) { 3959 /* already logged */ 3960 return; 3961 } 3962 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 3963 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 3964 tp->t_end_info_bytes[i] = status; 3965 tp->t_end_info_status |= bit; 3966 break; 3967 } 3968 } 3969 } 3970 3971 int 3972 tcp_can_enable_pacing(void) 3973 { 3974 3975 if ((tcp_pacing_limit == -1) || 3976 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 3977 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 3978 shadow_num_connections = number_of_tcp_connections_pacing; 3979 return (1); 3980 } else { 3981 return (0); 3982 } 3983 } 3984 3985 static uint8_t tcp_pacing_warning = 0; 3986 3987 void 3988 tcp_decrement_paced_conn(void) 3989 { 3990 uint32_t ret; 3991 3992 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 3993 shadow_num_connections = number_of_tcp_connections_pacing; 3994 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 3995 if (ret == 0) { 3996 if (tcp_pacing_limit != -1) { 3997 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 3998 tcp_pacing_limit = 0; 3999 } else if (tcp_pacing_warning == 0) { 4000 printf("Warning pacing count is invalid, invalid decrement\n"); 4001 tcp_pacing_warning = 1; 4002 } 4003 } 4004 } 4005 4006 #ifdef TCP_ACCOUNTING 4007 int 4008 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss) 4009 { 4010 if (SEQ_LT(th->th_ack, tp->snd_una)) { 4011 /* Do we have a SACK? */ 4012 if (to->to_flags & TOF_SACK) { 4013 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4014 tp->tcp_cnt_counters[ACK_SACK]++; 4015 } 4016 return (ACK_SACK); 4017 } else { 4018 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4019 tp->tcp_cnt_counters[ACK_BEHIND]++; 4020 } 4021 return (ACK_BEHIND); 4022 } 4023 } else if (th->th_ack == tp->snd_una) { 4024 /* Do we have a SACK? */ 4025 if (to->to_flags & TOF_SACK) { 4026 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4027 tp->tcp_cnt_counters[ACK_SACK]++; 4028 } 4029 return (ACK_SACK); 4030 } else if (tiwin != tp->snd_wnd) { 4031 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4032 tp->tcp_cnt_counters[ACK_RWND]++; 4033 } 4034 return (ACK_RWND); 4035 } else { 4036 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4037 tp->tcp_cnt_counters[ACK_DUPACK]++; 4038 } 4039 return (ACK_DUPACK); 4040 } 4041 } else { 4042 if (!SEQ_GT(th->th_ack, tp->snd_max)) { 4043 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4044 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss); 4045 } 4046 } 4047 if (to->to_flags & TOF_SACK) { 4048 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4049 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++; 4050 } 4051 return (ACK_CUMACK_SACK); 4052 } else { 4053 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4054 tp->tcp_cnt_counters[ACK_CUMACK]++; 4055 } 4056 return (ACK_CUMACK); 4057 } 4058 } 4059 } 4060 #endif 4061 4062 void 4063 tcp_change_time_units(struct tcpcb *tp, int granularity) 4064 { 4065 if (tp->t_tmr_granularity == granularity) { 4066 /* We are there */ 4067 return; 4068 } 4069 if (granularity == TCP_TMR_GRANULARITY_USEC) { 4070 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS), 4071 ("Granularity is not TICKS its %u in tp:%p", 4072 tp->t_tmr_granularity, tp)); 4073 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow); 4074 if (tp->t_srtt > 1) { 4075 uint32_t val, frac; 4076 4077 val = tp->t_srtt >> TCP_RTT_SHIFT; 4078 frac = tp->t_srtt & 0x1f; 4079 tp->t_srtt = TICKS_2_USEC(val); 4080 /* 4081 * frac is the fractional part of the srtt (if any) 4082 * but its in ticks and every bit represents 4083 * 1/32nd of a hz. 4084 */ 4085 if (frac) { 4086 if (hz == 1000) { 4087 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4088 } else { 4089 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4090 } 4091 tp->t_srtt += frac; 4092 } 4093 } 4094 if (tp->t_rttvar) { 4095 uint32_t val, frac; 4096 4097 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT; 4098 frac = tp->t_rttvar & 0x1f; 4099 tp->t_rttvar = TICKS_2_USEC(val); 4100 /* 4101 * frac is the fractional part of the srtt (if any) 4102 * but its in ticks and every bit represents 4103 * 1/32nd of a hz. 4104 */ 4105 if (frac) { 4106 if (hz == 1000) { 4107 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4108 } else { 4109 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4110 } 4111 tp->t_rttvar += frac; 4112 } 4113 } 4114 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC; 4115 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) { 4116 /* Convert back to ticks, with */ 4117 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC), 4118 ("Granularity is not USEC its %u in tp:%p", 4119 tp->t_tmr_granularity, tp)); 4120 if (tp->t_srtt > 1) { 4121 uint32_t val, frac; 4122 4123 val = USEC_2_TICKS(tp->t_srtt); 4124 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4125 tp->t_srtt = val << TCP_RTT_SHIFT; 4126 /* 4127 * frac is the fractional part here is left 4128 * over from converting to hz and shifting. 4129 * We need to convert this to the 5 bit 4130 * remainder. 4131 */ 4132 if (frac) { 4133 if (hz == 1000) { 4134 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4135 } else { 4136 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4137 } 4138 tp->t_srtt += frac; 4139 } 4140 } 4141 if (tp->t_rttvar) { 4142 uint32_t val, frac; 4143 4144 val = USEC_2_TICKS(tp->t_rttvar); 4145 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4146 tp->t_rttvar = val << TCP_RTTVAR_SHIFT; 4147 /* 4148 * frac is the fractional part here is left 4149 * over from converting to hz and shifting. 4150 * We need to convert this to the 5 bit 4151 * remainder. 4152 */ 4153 if (frac) { 4154 if (hz == 1000) { 4155 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4156 } else { 4157 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4158 } 4159 tp->t_rttvar += frac; 4160 } 4161 } 4162 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow); 4163 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 4164 } 4165 #ifdef INVARIANTS 4166 else { 4167 panic("Unknown granularity:%d tp:%p", 4168 granularity, tp); 4169 } 4170 #endif 4171 } 4172 4173 void 4174 tcp_handle_orphaned_packets(struct tcpcb *tp) 4175 { 4176 struct mbuf *save, *m, *prev; 4177 /* 4178 * Called when a stack switch is occuring from the fini() 4179 * of the old stack. We assue the init() as already been 4180 * run of the new stack and it has set the inp_flags2 to 4181 * what it supports. This function will then deal with any 4182 * differences i.e. cleanup packets that maybe queued that 4183 * the newstack does not support. 4184 */ 4185 4186 if (tptoinpcb(tp)->inp_flags2 & INP_MBUF_L_ACKS) 4187 return; 4188 if ((tptoinpcb(tp)->inp_flags2 & INP_SUPPORTS_MBUFQ) == 0) { 4189 /* 4190 * It is unsafe to process the packets since a 4191 * reset may be lurking in them (its rare but it 4192 * can occur). If we were to find a RST, then we 4193 * would end up dropping the connection and the 4194 * INP lock, so when we return the caller (tcp_usrreq) 4195 * will blow up when it trys to unlock the inp. 4196 * This new stack does not do any fancy LRO features 4197 * so all we can do is toss the packets. 4198 */ 4199 m = tp->t_in_pkt; 4200 tp->t_in_pkt = NULL; 4201 tp->t_tail_pkt = NULL; 4202 while (m) { 4203 save = m->m_nextpkt; 4204 m->m_nextpkt = NULL; 4205 m_freem(m); 4206 m = save; 4207 } 4208 } else { 4209 /* 4210 * Here we have a stack that does mbuf queuing but 4211 * does not support compressed ack's. We must 4212 * walk all the mbufs and discard any compressed acks. 4213 */ 4214 m = tp->t_in_pkt; 4215 prev = NULL; 4216 while (m) { 4217 if (m->m_flags & M_ACKCMP) { 4218 /* We must toss this packet */ 4219 if (tp->t_tail_pkt == m) 4220 tp->t_tail_pkt = prev; 4221 if (prev) 4222 prev->m_nextpkt = m->m_nextpkt; 4223 else 4224 tp->t_in_pkt = m->m_nextpkt; 4225 m->m_nextpkt = NULL; 4226 m_freem(m); 4227 /* move forward */ 4228 if (prev) 4229 m = prev->m_nextpkt; 4230 else 4231 m = tp->t_in_pkt; 4232 } else { 4233 /* this one is ok */ 4234 prev = m; 4235 m = m->m_nextpkt; 4236 } 4237 } 4238 } 4239 } 4240 4241 #ifdef TCP_REQUEST_TRK 4242 uint32_t 4243 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes) 4244 { 4245 #ifdef KERN_TLS 4246 struct ktls_session *tls; 4247 uint32_t rec_oh, records; 4248 4249 tls = so->so_snd.sb_tls_info; 4250 if (tls == NULL) 4251 return (0); 4252 4253 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen; 4254 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len); 4255 return (records * rec_oh); 4256 #else 4257 return (0); 4258 #endif 4259 } 4260 4261 extern uint32_t tcp_stale_entry_time; 4262 uint32_t tcp_stale_entry_time = 250000; 4263 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW, 4264 &tcp_stale_entry_time, 250000, "Time that a http entry without a sendfile ages out"); 4265 4266 void 4267 tcp_http_log_req_info(struct tcpcb *tp, struct http_sendfile_track *http, 4268 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes) 4269 { 4270 if (tcp_bblogging_on(tp)) { 4271 union tcp_log_stackspecific log; 4272 struct timeval tv; 4273 4274 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 4275 #ifdef TCPHPTS 4276 log.u_bbr.inhpts = tcp_in_hpts(tptoinpcb(tp)); 4277 #endif 4278 log.u_bbr.flex8 = val; 4279 log.u_bbr.rttProp = http->timestamp; 4280 log.u_bbr.delRate = http->start; 4281 log.u_bbr.cur_del_rate = http->end; 4282 log.u_bbr.flex1 = http->start_seq; 4283 log.u_bbr.flex2 = http->end_seq; 4284 log.u_bbr.flex3 = http->flags; 4285 log.u_bbr.flex4 = ((http->localtime >> 32) & 0x00000000ffffffff); 4286 log.u_bbr.flex5 = (http->localtime & 0x00000000ffffffff); 4287 log.u_bbr.flex7 = slot; 4288 log.u_bbr.bw_inuse = offset; 4289 /* nbytes = flex6 | epoch */ 4290 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff); 4291 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff); 4292 /* cspr = lt_epoch | pkts_out */ 4293 log.u_bbr.lt_epoch = ((http->cspr >> 32) & 0x00000000ffffffff); 4294 log.u_bbr.pkts_out |= (http->cspr & 0x00000000ffffffff); 4295 log.u_bbr.applimited = tp->t_http_closed; 4296 log.u_bbr.applimited <<= 8; 4297 log.u_bbr.applimited |= tp->t_http_open; 4298 log.u_bbr.applimited <<= 8; 4299 log.u_bbr.applimited |= tp->t_http_req; 4300 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 4301 TCP_LOG_EVENTP(tp, NULL, 4302 &tptosocket(tp)->so_rcv, 4303 &tptosocket(tp)->so_snd, 4304 TCP_LOG_HTTP_T, 0, 4305 0, &log, false, &tv); 4306 } 4307 } 4308 4309 void 4310 tcp_http_free_a_slot(struct tcpcb *tp, struct http_sendfile_track *ent) 4311 { 4312 if (tp->t_http_req > 0) 4313 tp->t_http_req--; 4314 if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN) { 4315 if (tp->t_http_open > 0) 4316 tp->t_http_open--; 4317 } else { 4318 if (tp->t_http_closed > 0) 4319 tp->t_http_closed--; 4320 } 4321 ent->flags = TCP_HTTP_TRACK_FLG_EMPTY; 4322 } 4323 4324 static void 4325 tcp_http_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest) 4326 { 4327 struct http_sendfile_track *ent; 4328 uint64_t time_delta, oldest_delta; 4329 int i, oldest, oldest_set = 0, cnt_rm = 0; 4330 4331 for(i = 0; i < MAX_TCP_HTTP_REQ; i++) { 4332 ent = &tp->t_http_info[i]; 4333 if (ent->flags != TCP_HTTP_TRACK_FLG_USED) { 4334 /* 4335 * We only care about closed end ranges 4336 * that are allocated and have no sendfile 4337 * ever touching them. They would be in 4338 * state USED. 4339 */ 4340 continue; 4341 } 4342 if (ts >= ent->localtime) 4343 time_delta = ts - ent->localtime; 4344 else 4345 time_delta = 0; 4346 if (time_delta && 4347 ((oldest_delta < time_delta) || (oldest_set == 0))) { 4348 oldest_set = 1; 4349 oldest = i; 4350 oldest_delta = time_delta; 4351 } 4352 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) { 4353 /* 4354 * No sendfile in a our time-limit 4355 * time to purge it. 4356 */ 4357 cnt_rm++; 4358 tcp_http_log_req_info(tp, &tp->t_http_info[i], i, TCP_HTTP_REQ_LOG_STALE, 4359 time_delta, 0); 4360 tcp_http_free_a_slot(tp, ent); 4361 } 4362 } 4363 if ((cnt_rm == 0) && rm_oldest && oldest_set) { 4364 ent = &tp->t_http_info[oldest]; 4365 tcp_http_log_req_info(tp, &tp->t_http_info[i], i, TCP_HTTP_REQ_LOG_STALE, 4366 oldest_delta, 1); 4367 tcp_http_free_a_slot(tp, ent); 4368 } 4369 } 4370 4371 int 4372 tcp_http_check_for_comp(struct tcpcb *tp, tcp_seq ack_point) 4373 { 4374 int i, ret=0; 4375 struct http_sendfile_track *ent; 4376 4377 /* Clean up any old closed end requests that are now completed */ 4378 if (tp->t_http_req == 0) 4379 return(0); 4380 if (tp->t_http_closed == 0) 4381 return(0); 4382 for(i = 0; i < MAX_TCP_HTTP_REQ; i++) { 4383 ent = &tp->t_http_info[i]; 4384 /* Skip empty ones */ 4385 if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY) 4386 continue; 4387 /* Skip open ones */ 4388 if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN) 4389 continue; 4390 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4391 /* We are past it -- free it */ 4392 tcp_http_log_req_info(tp, ent, 4393 i, TCP_HTTP_REQ_LOG_FREED, 0, 0); 4394 tcp_http_free_a_slot(tp, ent); 4395 ret++; 4396 } 4397 } 4398 return (ret); 4399 } 4400 4401 int 4402 tcp_http_is_entry_comp(struct tcpcb *tp, struct http_sendfile_track *ent, tcp_seq ack_point) 4403 { 4404 if (tp->t_http_req == 0) 4405 return(-1); 4406 if (tp->t_http_closed == 0) 4407 return(-1); 4408 if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY) 4409 return(-1); 4410 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4411 return (1); 4412 } 4413 return (0); 4414 } 4415 4416 struct http_sendfile_track * 4417 tcp_http_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip) 4418 { 4419 /* 4420 * Given an ack point (th_ack) walk through our entries and 4421 * return the first one found that th_ack goes past the 4422 * end_seq. 4423 */ 4424 struct http_sendfile_track *ent; 4425 int i; 4426 4427 if (tp->t_http_req == 0) { 4428 /* none open */ 4429 return (NULL); 4430 } 4431 for(i = 0; i < MAX_TCP_HTTP_REQ; i++) { 4432 ent = &tp->t_http_info[i]; 4433 if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY) 4434 continue; 4435 if ((ent->flags & TCP_HTTP_TRACK_FLG_OPEN) == 0) { 4436 if (SEQ_GEQ(th_ack, ent->end_seq)) { 4437 *ip = i; 4438 return (ent); 4439 } 4440 } 4441 } 4442 return (NULL); 4443 } 4444 4445 struct http_sendfile_track * 4446 tcp_http_find_req_for_seq(struct tcpcb *tp, tcp_seq seq) 4447 { 4448 struct http_sendfile_track *ent; 4449 int i; 4450 4451 if (tp->t_http_req == 0) { 4452 /* none open */ 4453 return (NULL); 4454 } 4455 for(i = 0; i < MAX_TCP_HTTP_REQ; i++) { 4456 ent = &tp->t_http_info[i]; 4457 tcp_http_log_req_info(tp, ent, i, TCP_HTTP_REQ_LOG_SEARCH, 4458 (uint64_t)seq, 0); 4459 if (ent->flags == TCP_HTTP_TRACK_FLG_EMPTY) { 4460 continue; 4461 } 4462 if (ent->flags & TCP_HTTP_TRACK_FLG_OPEN) { 4463 /* 4464 * An open end request only needs to 4465 * match the beginning seq or be 4466 * all we have (once we keep going on 4467 * a open end request we may have a seq 4468 * wrap). 4469 */ 4470 if ((SEQ_GEQ(seq, ent->start_seq)) || 4471 (tp->t_http_closed == 0)) 4472 return (ent); 4473 } else { 4474 /* 4475 * For this one we need to 4476 * be a bit more careful if its 4477 * completed at least. 4478 */ 4479 if ((SEQ_GEQ(seq, ent->start_seq)) && 4480 (SEQ_LT(seq, ent->end_seq))) { 4481 return (ent); 4482 } 4483 } 4484 } 4485 return (NULL); 4486 } 4487 4488 /* Should this be in its own file tcp_http.c ? */ 4489 struct http_sendfile_track * 4490 tcp_http_alloc_req_full(struct tcpcb *tp, struct http_req *req, uint64_t ts, int rec_dups) 4491 { 4492 struct http_sendfile_track *fil; 4493 int i, allocated; 4494 4495 /* In case the stack does not check for completions do so now */ 4496 tcp_http_check_for_comp(tp, tp->snd_una); 4497 /* Check for stale entries */ 4498 if (tp->t_http_req) 4499 tcp_http_check_for_stale_entries(tp, ts, 4500 (tp->t_http_req >= MAX_TCP_HTTP_REQ)); 4501 /* Check to see if this is a duplicate of one not started */ 4502 if (tp->t_http_req) { 4503 for(i = 0, allocated = 0; i < MAX_TCP_HTTP_REQ; i++) { 4504 fil = &tp->t_http_info[i]; 4505 if (fil->flags != TCP_HTTP_TRACK_FLG_USED) 4506 continue; 4507 if ((fil->timestamp == req->timestamp) && 4508 (fil->start == req->start) && 4509 ((fil->flags & TCP_HTTP_TRACK_FLG_OPEN) || 4510 (fil->end == req->end))) { 4511 /* 4512 * We already have this request 4513 * and it has not been started with sendfile. 4514 * This probably means the user was returned 4515 * a 4xx of some sort and its going to age 4516 * out, lets not duplicate it. 4517 */ 4518 return(fil); 4519 } 4520 } 4521 } 4522 /* Ok if there is no room at the inn we are in trouble */ 4523 if (tp->t_http_req >= MAX_TCP_HTTP_REQ) { 4524 tcp_trace_point(tp, TCP_TP_HTTP_LOG_FAIL); 4525 for(i = 0; i < MAX_TCP_HTTP_REQ; i++) { 4526 tcp_http_log_req_info(tp, &tp->t_http_info[i], 4527 i, TCP_HTTP_REQ_LOG_ALLOCFAIL, 0, 0); 4528 } 4529 return (NULL); 4530 } 4531 for(i = 0, allocated = 0; i < MAX_TCP_HTTP_REQ; i++) { 4532 fil = &tp->t_http_info[i]; 4533 if (fil->flags == TCP_HTTP_TRACK_FLG_EMPTY) { 4534 allocated = 1; 4535 fil->flags = TCP_HTTP_TRACK_FLG_USED; 4536 fil->timestamp = req->timestamp; 4537 fil->localtime = ts; 4538 fil->start = req->start; 4539 if (req->flags & TCP_LOG_HTTPD_RANGE_END) { 4540 fil->end = req->end; 4541 } else { 4542 fil->end = 0; 4543 fil->flags |= TCP_HTTP_TRACK_FLG_OPEN; 4544 } 4545 /* 4546 * We can set the min boundaries to the TCP Sequence space, 4547 * but it might be found to be further up when sendfile 4548 * actually runs on this range (if it ever does). 4549 */ 4550 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc; 4551 fil->start_seq = tp->snd_una + 4552 tptosocket(tp)->so_snd.sb_ccc; 4553 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start))); 4554 if (tptosocket(tp)->so_snd.sb_tls_info) { 4555 /* 4556 * This session is doing TLS. Take a swag guess 4557 * at the overhead. 4558 */ 4559 fil->end_seq += tcp_estimate_tls_overhead( 4560 tptosocket(tp), (fil->end - fil->start)); 4561 } 4562 tp->t_http_req++; 4563 if (fil->flags & TCP_HTTP_TRACK_FLG_OPEN) 4564 tp->t_http_open++; 4565 else 4566 tp->t_http_closed++; 4567 tcp_http_log_req_info(tp, fil, i, 4568 TCP_HTTP_REQ_LOG_NEW, 0, 0); 4569 break; 4570 } else 4571 fil = NULL; 4572 } 4573 return (fil); 4574 } 4575 4576 void 4577 tcp_http_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts) 4578 { 4579 (void)tcp_http_alloc_req_full(tp, &user->http_req, ts, 1); 4580 } 4581 #endif 4582