1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_kern_tls.h" 41 #include "opt_tcpdebug.h" 42 43 #include <sys/param.h> 44 #include <sys/systm.h> 45 #include <sys/arb.h> 46 #include <sys/callout.h> 47 #include <sys/eventhandler.h> 48 #ifdef TCP_HHOOK 49 #include <sys/hhook.h> 50 #endif 51 #include <sys/kernel.h> 52 #ifdef TCP_HHOOK 53 #include <sys/khelp.h> 54 #endif 55 #ifdef KERN_TLS 56 #include <sys/ktls.h> 57 #endif 58 #include <sys/qmath.h> 59 #include <sys/stats.h> 60 #include <sys/sysctl.h> 61 #include <sys/jail.h> 62 #include <sys/malloc.h> 63 #include <sys/refcount.h> 64 #include <sys/mbuf.h> 65 #ifdef INET6 66 #include <sys/domain.h> 67 #endif 68 #include <sys/priv.h> 69 #include <sys/proc.h> 70 #include <sys/sdt.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/protosw.h> 74 #include <sys/random.h> 75 76 #include <vm/uma.h> 77 78 #include <net/route.h> 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/vnet.h> 82 83 #include <netinet/in.h> 84 #include <netinet/in_fib.h> 85 #include <netinet/in_kdtrace.h> 86 #include <netinet/in_pcb.h> 87 #include <netinet/in_systm.h> 88 #include <netinet/in_var.h> 89 #include <netinet/ip.h> 90 #include <netinet/ip_icmp.h> 91 #include <netinet/ip_var.h> 92 #ifdef INET6 93 #include <netinet/icmp6.h> 94 #include <netinet/ip6.h> 95 #include <netinet6/in6_fib.h> 96 #include <netinet6/in6_pcb.h> 97 #include <netinet6/ip6_var.h> 98 #include <netinet6/scope6_var.h> 99 #include <netinet6/nd6.h> 100 #endif 101 102 #include <netinet/tcp.h> 103 #include <netinet/tcp_fsm.h> 104 #include <netinet/tcp_seq.h> 105 #include <netinet/tcp_timer.h> 106 #include <netinet/tcp_var.h> 107 #include <netinet/tcp_log_buf.h> 108 #include <netinet/tcp_syncache.h> 109 #include <netinet/tcp_hpts.h> 110 #include <netinet/cc/cc.h> 111 #ifdef INET6 112 #include <netinet6/tcp6_var.h> 113 #endif 114 #include <netinet/tcpip.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #ifdef TCPDEBUG 120 #include <netinet/tcp_debug.h> 121 #endif 122 #ifdef INET6 123 #include <netinet6/ip6protosw.h> 124 #endif 125 #ifdef TCP_OFFLOAD 126 #include <netinet/tcp_offload.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 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 137 #ifdef INET6 138 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 139 #endif 140 141 #ifdef NETFLIX_EXP_DETECTION 142 /* Sack attack detection thresholds and such */ 143 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack_attack, CTLFLAG_RW, 0, 144 "Sack Attack detection thresholds"); 145 int32_t tcp_force_detection = 0; 146 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, force_detection, 147 CTLFLAG_RW, 148 &tcp_force_detection, 0, 149 "Do we force detection even if the INP has it off?"); 150 int32_t tcp_sack_to_ack_thresh = 700; /* 70 % */ 151 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sack_to_ack_thresh, 152 CTLFLAG_RW, 153 &tcp_sack_to_ack_thresh, 700, 154 "Percentage of sacks to acks we must see above (10.1 percent is 101)?"); 155 int32_t tcp_sack_to_move_thresh = 600; /* 60 % */ 156 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, move_thresh, 157 CTLFLAG_RW, 158 &tcp_sack_to_move_thresh, 600, 159 "Percentage of sack moves we must see above (10.1 percent is 101)"); 160 int32_t tcp_restoral_thresh = 650; /* 65 % (sack:2:ack -5%) */ 161 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, restore_thresh, 162 CTLFLAG_RW, 163 &tcp_restoral_thresh, 550, 164 "Percentage of sack to ack percentage we must see below to restore(10.1 percent is 101)"); 165 int32_t tcp_sad_decay_val = 800; 166 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, decay_per, 167 CTLFLAG_RW, 168 &tcp_sad_decay_val, 800, 169 "The decay percentage (10.1 percent equals 101 )"); 170 int32_t tcp_map_minimum = 500; 171 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, nummaps, 172 CTLFLAG_RW, 173 &tcp_map_minimum, 500, 174 "Number of Map enteries before we start detection"); 175 int32_t tcp_attack_on_turns_on_logging = 0; 176 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, attacks_logged, 177 CTLFLAG_RW, 178 &tcp_attack_on_turns_on_logging, 0, 179 "When we have a positive hit on attack, do we turn on logging?"); 180 int32_t tcp_sad_pacing_interval = 2000; 181 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_pacing_int, 182 CTLFLAG_RW, 183 &tcp_sad_pacing_interval, 2000, 184 "What is the minimum pacing interval for a classified attacker?"); 185 186 int32_t tcp_sad_low_pps = 100; 187 SYSCTL_INT(_net_inet_tcp_sack_attack, OID_AUTO, sad_low_pps, 188 CTLFLAG_RW, 189 &tcp_sad_low_pps, 100, 190 "What is the input pps that below which we do not decay?"); 191 #endif 192 193 struct rwlock tcp_function_lock; 194 195 static int 196 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 197 { 198 int error, new; 199 200 new = V_tcp_mssdflt; 201 error = sysctl_handle_int(oidp, &new, 0, req); 202 if (error == 0 && req->newptr) { 203 if (new < TCP_MINMSS) 204 error = EINVAL; 205 else 206 V_tcp_mssdflt = new; 207 } 208 return (error); 209 } 210 211 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 212 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 213 &sysctl_net_inet_tcp_mss_check, "I", 214 "Default TCP Maximum Segment Size"); 215 216 #ifdef INET6 217 static int 218 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 219 { 220 int error, new; 221 222 new = V_tcp_v6mssdflt; 223 error = sysctl_handle_int(oidp, &new, 0, req); 224 if (error == 0 && req->newptr) { 225 if (new < TCP_MINMSS) 226 error = EINVAL; 227 else 228 V_tcp_v6mssdflt = new; 229 } 230 return (error); 231 } 232 233 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 234 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 235 &sysctl_net_inet_tcp_mss_v6_check, "I", 236 "Default TCP Maximum Segment Size for IPv6"); 237 #endif /* INET6 */ 238 239 /* 240 * Minimum MSS we accept and use. This prevents DoS attacks where 241 * we are forced to a ridiculous low MSS like 20 and send hundreds 242 * of packets instead of one. The effect scales with the available 243 * bandwidth and quickly saturates the CPU and network interface 244 * with packet generation and sending. Set to zero to disable MINMSS 245 * checking. This setting prevents us from sending too small packets. 246 */ 247 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 248 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 249 &VNET_NAME(tcp_minmss), 0, 250 "Minimum TCP Maximum Segment Size"); 251 252 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 253 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 254 &VNET_NAME(tcp_do_rfc1323), 0, 255 "Enable rfc1323 (high performance TCP) extensions"); 256 257 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 258 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 259 &VNET_NAME(tcp_ts_offset_per_conn), 0, 260 "Initialize TCP timestamps per connection instead of per host pair"); 261 262 static int tcp_log_debug = 0; 263 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 264 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 265 266 static int tcp_tcbhashsize; 267 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 268 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 269 270 static int do_tcpdrain = 1; 271 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 272 "Enable tcp_drain routine for extra help when low on mbufs"); 273 274 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 275 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 276 277 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 278 #define V_icmp_may_rst VNET(icmp_may_rst) 279 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 280 &VNET_NAME(icmp_may_rst), 0, 281 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 282 283 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 284 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 285 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 286 &VNET_NAME(tcp_isn_reseed_interval), 0, 287 "Seconds between reseeding of ISN secret"); 288 289 static int tcp_soreceive_stream; 290 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 291 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 292 293 VNET_DEFINE(uma_zone_t, sack_hole_zone); 294 #define V_sack_hole_zone VNET(sack_hole_zone) 295 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 296 static int 297 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 298 { 299 int error; 300 uint32_t new; 301 302 new = V_tcp_map_entries_limit; 303 error = sysctl_handle_int(oidp, &new, 0, req); 304 if (error == 0 && req->newptr) { 305 /* only allow "0" and value > minimum */ 306 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 307 error = EINVAL; 308 else 309 V_tcp_map_entries_limit = new; 310 } 311 return (error); 312 } 313 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 314 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW, 315 &VNET_NAME(tcp_map_entries_limit), 0, 316 &sysctl_net_inet_tcp_map_limit_check, "IU", 317 "Total sendmap entries limit"); 318 319 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 320 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 321 &VNET_NAME(tcp_map_split_limit), 0, 322 "Total sendmap split entries limit"); 323 324 #ifdef TCP_HHOOK 325 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 326 #endif 327 328 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 329 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 330 #define V_ts_offset_secret VNET(ts_offset_secret) 331 332 static int tcp_default_fb_init(struct tcpcb *tp); 333 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 334 static int tcp_default_handoff_ok(struct tcpcb *tp); 335 static struct inpcb *tcp_notify(struct inpcb *, int); 336 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 337 static void tcp_mtudisc(struct inpcb *, int); 338 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 339 void *ip4hdr, const void *ip6hdr); 340 341 342 static struct tcp_function_block tcp_def_funcblk = { 343 .tfb_tcp_block_name = "freebsd", 344 .tfb_tcp_output = tcp_output, 345 .tfb_tcp_do_segment = tcp_do_segment, 346 .tfb_tcp_ctloutput = tcp_default_ctloutput, 347 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 348 .tfb_tcp_fb_init = tcp_default_fb_init, 349 .tfb_tcp_fb_fini = tcp_default_fb_fini, 350 }; 351 352 static int tcp_fb_cnt = 0; 353 struct tcp_funchead t_functions; 354 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 355 356 static struct tcp_function_block * 357 find_tcp_functions_locked(struct tcp_function_set *fs) 358 { 359 struct tcp_function *f; 360 struct tcp_function_block *blk=NULL; 361 362 TAILQ_FOREACH(f, &t_functions, tf_next) { 363 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 364 blk = f->tf_fb; 365 break; 366 } 367 } 368 return(blk); 369 } 370 371 static struct tcp_function_block * 372 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 373 { 374 struct tcp_function_block *rblk=NULL; 375 struct tcp_function *f; 376 377 TAILQ_FOREACH(f, &t_functions, tf_next) { 378 if (f->tf_fb == blk) { 379 rblk = blk; 380 if (s) { 381 *s = f; 382 } 383 break; 384 } 385 } 386 return (rblk); 387 } 388 389 struct tcp_function_block * 390 find_and_ref_tcp_functions(struct tcp_function_set *fs) 391 { 392 struct tcp_function_block *blk; 393 394 rw_rlock(&tcp_function_lock); 395 blk = find_tcp_functions_locked(fs); 396 if (blk) 397 refcount_acquire(&blk->tfb_refcnt); 398 rw_runlock(&tcp_function_lock); 399 return(blk); 400 } 401 402 struct tcp_function_block * 403 find_and_ref_tcp_fb(struct tcp_function_block *blk) 404 { 405 struct tcp_function_block *rblk; 406 407 rw_rlock(&tcp_function_lock); 408 rblk = find_tcp_fb_locked(blk, NULL); 409 if (rblk) 410 refcount_acquire(&rblk->tfb_refcnt); 411 rw_runlock(&tcp_function_lock); 412 return(rblk); 413 } 414 415 static struct tcp_function_block * 416 find_and_ref_tcp_default_fb(void) 417 { 418 struct tcp_function_block *rblk; 419 420 rw_rlock(&tcp_function_lock); 421 rblk = tcp_func_set_ptr; 422 refcount_acquire(&rblk->tfb_refcnt); 423 rw_runlock(&tcp_function_lock); 424 return (rblk); 425 } 426 427 void 428 tcp_switch_back_to_default(struct tcpcb *tp) 429 { 430 struct tcp_function_block *tfb; 431 432 KASSERT(tp->t_fb != &tcp_def_funcblk, 433 ("%s: called by the built-in default stack", __func__)); 434 435 /* 436 * Release the old stack. This function will either find a new one 437 * or panic. 438 */ 439 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 440 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 441 refcount_release(&tp->t_fb->tfb_refcnt); 442 443 /* 444 * Now, we'll find a new function block to use. 445 * Start by trying the current user-selected 446 * default, unless this stack is the user-selected 447 * default. 448 */ 449 tfb = find_and_ref_tcp_default_fb(); 450 if (tfb == tp->t_fb) { 451 refcount_release(&tfb->tfb_refcnt); 452 tfb = NULL; 453 } 454 /* Does the stack accept this connection? */ 455 if (tfb != NULL && tfb->tfb_tcp_handoff_ok != NULL && 456 (*tfb->tfb_tcp_handoff_ok)(tp)) { 457 refcount_release(&tfb->tfb_refcnt); 458 tfb = NULL; 459 } 460 /* Try to use that stack. */ 461 if (tfb != NULL) { 462 /* Initialize the new stack. If it succeeds, we are done. */ 463 tp->t_fb = tfb; 464 if (tp->t_fb->tfb_tcp_fb_init == NULL || 465 (*tp->t_fb->tfb_tcp_fb_init)(tp) == 0) 466 return; 467 468 /* 469 * Initialization failed. Release the reference count on 470 * the stack. 471 */ 472 refcount_release(&tfb->tfb_refcnt); 473 } 474 475 /* 476 * If that wasn't feasible, use the built-in default 477 * stack which is not allowed to reject anyone. 478 */ 479 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 480 if (tfb == NULL) { 481 /* there always should be a default */ 482 panic("Can't refer to tcp_def_funcblk"); 483 } 484 if (tfb->tfb_tcp_handoff_ok != NULL) { 485 if ((*tfb->tfb_tcp_handoff_ok) (tp)) { 486 /* The default stack cannot say no */ 487 panic("Default stack rejects a new session?"); 488 } 489 } 490 tp->t_fb = tfb; 491 if (tp->t_fb->tfb_tcp_fb_init != NULL && 492 (*tp->t_fb->tfb_tcp_fb_init)(tp)) { 493 /* The default stack cannot fail */ 494 panic("Default stack initialization failed"); 495 } 496 } 497 498 static int 499 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 500 { 501 int error=ENOENT; 502 struct tcp_function_set fs; 503 struct tcp_function_block *blk; 504 505 memset(&fs, 0, sizeof(fs)); 506 rw_rlock(&tcp_function_lock); 507 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 508 if (blk) { 509 /* Found him */ 510 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 511 fs.pcbcnt = blk->tfb_refcnt; 512 } 513 rw_runlock(&tcp_function_lock); 514 error = sysctl_handle_string(oidp, fs.function_set_name, 515 sizeof(fs.function_set_name), req); 516 517 /* Check for error or no change */ 518 if (error != 0 || req->newptr == NULL) 519 return(error); 520 521 rw_wlock(&tcp_function_lock); 522 blk = find_tcp_functions_locked(&fs); 523 if ((blk == NULL) || 524 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 525 error = ENOENT; 526 goto done; 527 } 528 tcp_func_set_ptr = blk; 529 done: 530 rw_wunlock(&tcp_function_lock); 531 return (error); 532 } 533 534 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 535 CTLTYPE_STRING | CTLFLAG_RW, 536 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 537 "Set/get the default TCP functions"); 538 539 static int 540 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 541 { 542 int error, cnt, linesz; 543 struct tcp_function *f; 544 char *buffer, *cp; 545 size_t bufsz, outsz; 546 bool alias; 547 548 cnt = 0; 549 rw_rlock(&tcp_function_lock); 550 TAILQ_FOREACH(f, &t_functions, tf_next) { 551 cnt++; 552 } 553 rw_runlock(&tcp_function_lock); 554 555 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 556 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 557 558 error = 0; 559 cp = buffer; 560 561 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 562 "Alias", "PCB count"); 563 cp += linesz; 564 bufsz -= linesz; 565 outsz = linesz; 566 567 rw_rlock(&tcp_function_lock); 568 TAILQ_FOREACH(f, &t_functions, tf_next) { 569 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 570 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 571 f->tf_fb->tfb_tcp_block_name, 572 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 573 alias ? f->tf_name : "-", 574 f->tf_fb->tfb_refcnt); 575 if (linesz >= bufsz) { 576 error = EOVERFLOW; 577 break; 578 } 579 cp += linesz; 580 bufsz -= linesz; 581 outsz += linesz; 582 } 583 rw_runlock(&tcp_function_lock); 584 if (error == 0) 585 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 586 free(buffer, M_TEMP); 587 return (error); 588 } 589 590 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 591 CTLTYPE_STRING|CTLFLAG_RD, 592 NULL, 0, sysctl_net_inet_list_available, "A", 593 "list available TCP Function sets"); 594 595 /* 596 * Exports one (struct tcp_function_info) for each alias/name. 597 */ 598 static int 599 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 600 { 601 int cnt, error; 602 struct tcp_function *f; 603 struct tcp_function_info tfi; 604 605 /* 606 * We don't allow writes. 607 */ 608 if (req->newptr != NULL) 609 return (EINVAL); 610 611 /* 612 * Wire the old buffer so we can directly copy the functions to 613 * user space without dropping the lock. 614 */ 615 if (req->oldptr != NULL) { 616 error = sysctl_wire_old_buffer(req, 0); 617 if (error) 618 return (error); 619 } 620 621 /* 622 * Walk the list and copy out matching entries. If INVARIANTS 623 * is compiled in, also walk the list to verify the length of 624 * the list matches what we have recorded. 625 */ 626 rw_rlock(&tcp_function_lock); 627 628 cnt = 0; 629 #ifndef INVARIANTS 630 if (req->oldptr == NULL) { 631 cnt = tcp_fb_cnt; 632 goto skip_loop; 633 } 634 #endif 635 TAILQ_FOREACH(f, &t_functions, tf_next) { 636 #ifdef INVARIANTS 637 cnt++; 638 #endif 639 if (req->oldptr != NULL) { 640 bzero(&tfi, sizeof(tfi)); 641 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 642 tfi.tfi_id = f->tf_fb->tfb_id; 643 (void)strlcpy(tfi.tfi_alias, f->tf_name, 644 sizeof(tfi.tfi_alias)); 645 (void)strlcpy(tfi.tfi_name, 646 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 647 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 648 /* 649 * Don't stop on error, as that is the 650 * mechanism we use to accumulate length 651 * information if the buffer was too short. 652 */ 653 } 654 } 655 KASSERT(cnt == tcp_fb_cnt, 656 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 657 #ifndef INVARIANTS 658 skip_loop: 659 #endif 660 rw_runlock(&tcp_function_lock); 661 if (req->oldptr == NULL) 662 error = SYSCTL_OUT(req, NULL, 663 (cnt + 1) * sizeof(struct tcp_function_info)); 664 665 return (error); 666 } 667 668 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 669 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 670 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 671 "List TCP function block name-to-ID mappings"); 672 673 /* 674 * tfb_tcp_handoff_ok() function for the default stack. 675 * Note that we'll basically try to take all comers. 676 */ 677 static int 678 tcp_default_handoff_ok(struct tcpcb *tp) 679 { 680 681 return (0); 682 } 683 684 /* 685 * tfb_tcp_fb_init() function for the default stack. 686 * 687 * This handles making sure we have appropriate timers set if you are 688 * transitioning a socket that has some amount of setup done. 689 * 690 * The init() fuction from the default can *never* return non-zero i.e. 691 * it is required to always succeed since it is the stack of last resort! 692 */ 693 static int 694 tcp_default_fb_init(struct tcpcb *tp) 695 { 696 697 struct socket *so; 698 699 INP_WLOCK_ASSERT(tp->t_inpcb); 700 701 KASSERT(tp->t_state >= 0 && tp->t_state < TCPS_TIME_WAIT, 702 ("%s: connection %p in unexpected state %d", __func__, tp, 703 tp->t_state)); 704 705 /* 706 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 707 * know what to do for unexpected states (which includes TIME_WAIT). 708 */ 709 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 710 return (0); 711 712 /* 713 * Make sure some kind of transmission timer is set if there is 714 * outstanding data. 715 */ 716 so = tp->t_inpcb->inp_socket; 717 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 718 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 719 tcp_timer_active(tp, TT_PERSIST))) { 720 /* 721 * If the session has established and it looks like it should 722 * be in the persist state, set the persist timer. Otherwise, 723 * set the retransmit timer. 724 */ 725 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 726 (int32_t)(tp->snd_nxt - tp->snd_una) < 727 (int32_t)sbavail(&so->so_snd)) 728 tcp_setpersist(tp); 729 else 730 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 731 } 732 733 /* All non-embryonic sessions get a keepalive timer. */ 734 if (!tcp_timer_active(tp, TT_KEEP)) 735 tcp_timer_activate(tp, TT_KEEP, 736 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 737 TP_KEEPINIT(tp)); 738 739 return (0); 740 } 741 742 /* 743 * tfb_tcp_fb_fini() function for the default stack. 744 * 745 * This changes state as necessary (or prudent) to prepare for another stack 746 * to assume responsibility for the connection. 747 */ 748 static void 749 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 750 { 751 752 INP_WLOCK_ASSERT(tp->t_inpcb); 753 return; 754 } 755 756 /* 757 * Target size of TCP PCB hash tables. Must be a power of two. 758 * 759 * Note that this can be overridden by the kernel environment 760 * variable net.inet.tcp.tcbhashsize 761 */ 762 #ifndef TCBHASHSIZE 763 #define TCBHASHSIZE 0 764 #endif 765 766 /* 767 * XXX 768 * Callouts should be moved into struct tcp directly. They are currently 769 * separate because the tcpcb structure is exported to userland for sysctl 770 * parsing purposes, which do not know about callouts. 771 */ 772 struct tcpcb_mem { 773 struct tcpcb tcb; 774 struct tcp_timer tt; 775 struct cc_var ccv; 776 #ifdef TCP_HHOOK 777 struct osd osd; 778 #endif 779 }; 780 781 VNET_DEFINE_STATIC(uma_zone_t, tcpcb_zone); 782 #define V_tcpcb_zone VNET(tcpcb_zone) 783 784 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 785 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 786 787 static struct mtx isn_mtx; 788 789 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 790 #define ISN_LOCK() mtx_lock(&isn_mtx) 791 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 792 793 /* 794 * TCP initialization. 795 */ 796 static void 797 tcp_zone_change(void *tag) 798 { 799 800 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 801 uma_zone_set_max(V_tcpcb_zone, maxsockets); 802 tcp_tw_zone_change(); 803 } 804 805 static int 806 tcp_inpcb_init(void *mem, int size, int flags) 807 { 808 struct inpcb *inp = mem; 809 810 INP_LOCK_INIT(inp, "inp", "tcpinp"); 811 return (0); 812 } 813 814 /* 815 * Take a value and get the next power of 2 that doesn't overflow. 816 * Used to size the tcp_inpcb hash buckets. 817 */ 818 static int 819 maketcp_hashsize(int size) 820 { 821 int hashsize; 822 823 /* 824 * auto tune. 825 * get the next power of 2 higher than maxsockets. 826 */ 827 hashsize = 1 << fls(size); 828 /* catch overflow, and just go one power of 2 smaller */ 829 if (hashsize < size) { 830 hashsize = 1 << (fls(size) - 1); 831 } 832 return (hashsize); 833 } 834 835 static volatile int next_tcp_stack_id = 1; 836 837 /* 838 * Register a TCP function block with the name provided in the names 839 * array. (Note that this function does NOT automatically register 840 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 841 * explicitly include blk->tfb_tcp_block_name in the list of names if 842 * you wish to register the stack with that name.) 843 * 844 * Either all name registrations will succeed or all will fail. If 845 * a name registration fails, the function will update the num_names 846 * argument to point to the array index of the name that encountered 847 * the failure. 848 * 849 * Returns 0 on success, or an error code on failure. 850 */ 851 int 852 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 853 const char *names[], int *num_names) 854 { 855 struct tcp_function *n; 856 struct tcp_function_set fs; 857 int error, i; 858 859 KASSERT(names != NULL && *num_names > 0, 860 ("%s: Called with 0-length name list", __func__)); 861 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 862 KASSERT(rw_initialized(&tcp_function_lock), 863 ("%s: called too early", __func__)); 864 865 if ((blk->tfb_tcp_output == NULL) || 866 (blk->tfb_tcp_do_segment == NULL) || 867 (blk->tfb_tcp_ctloutput == NULL) || 868 (strlen(blk->tfb_tcp_block_name) == 0)) { 869 /* 870 * These functions are required and you 871 * need a name. 872 */ 873 *num_names = 0; 874 return (EINVAL); 875 } 876 if (blk->tfb_tcp_timer_stop_all || 877 blk->tfb_tcp_timer_activate || 878 blk->tfb_tcp_timer_active || 879 blk->tfb_tcp_timer_stop) { 880 /* 881 * If you define one timer function you 882 * must have them all. 883 */ 884 if ((blk->tfb_tcp_timer_stop_all == NULL) || 885 (blk->tfb_tcp_timer_activate == NULL) || 886 (blk->tfb_tcp_timer_active == NULL) || 887 (blk->tfb_tcp_timer_stop == NULL)) { 888 *num_names = 0; 889 return (EINVAL); 890 } 891 } 892 893 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 894 *num_names = 0; 895 return (EINVAL); 896 } 897 898 refcount_init(&blk->tfb_refcnt, 0); 899 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 900 for (i = 0; i < *num_names; i++) { 901 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 902 if (n == NULL) { 903 error = ENOMEM; 904 goto cleanup; 905 } 906 n->tf_fb = blk; 907 908 (void)strlcpy(fs.function_set_name, names[i], 909 sizeof(fs.function_set_name)); 910 rw_wlock(&tcp_function_lock); 911 if (find_tcp_functions_locked(&fs) != NULL) { 912 /* Duplicate name space not allowed */ 913 rw_wunlock(&tcp_function_lock); 914 free(n, M_TCPFUNCTIONS); 915 error = EALREADY; 916 goto cleanup; 917 } 918 (void)strlcpy(n->tf_name, names[i], sizeof(n->tf_name)); 919 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 920 tcp_fb_cnt++; 921 rw_wunlock(&tcp_function_lock); 922 } 923 return(0); 924 925 cleanup: 926 /* 927 * Deregister the names we just added. Because registration failed 928 * for names[i], we don't need to deregister that name. 929 */ 930 *num_names = i; 931 rw_wlock(&tcp_function_lock); 932 while (--i >= 0) { 933 TAILQ_FOREACH(n, &t_functions, tf_next) { 934 if (!strncmp(n->tf_name, names[i], 935 TCP_FUNCTION_NAME_LEN_MAX)) { 936 TAILQ_REMOVE(&t_functions, n, tf_next); 937 tcp_fb_cnt--; 938 n->tf_fb = NULL; 939 free(n, M_TCPFUNCTIONS); 940 break; 941 } 942 } 943 } 944 rw_wunlock(&tcp_function_lock); 945 return (error); 946 } 947 948 /* 949 * Register a TCP function block using the name provided in the name 950 * argument. 951 * 952 * Returns 0 on success, or an error code on failure. 953 */ 954 int 955 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 956 int wait) 957 { 958 const char *name_list[1]; 959 int num_names, rv; 960 961 num_names = 1; 962 if (name != NULL) 963 name_list[0] = name; 964 else 965 name_list[0] = blk->tfb_tcp_block_name; 966 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 967 return (rv); 968 } 969 970 /* 971 * Register a TCP function block using the name defined in 972 * blk->tfb_tcp_block_name. 973 * 974 * Returns 0 on success, or an error code on failure. 975 */ 976 int 977 register_tcp_functions(struct tcp_function_block *blk, int wait) 978 { 979 980 return (register_tcp_functions_as_name(blk, NULL, wait)); 981 } 982 983 /* 984 * Deregister all names associated with a function block. This 985 * functionally removes the function block from use within the system. 986 * 987 * When called with a true quiesce argument, mark the function block 988 * as being removed so no more stacks will use it and determine 989 * whether the removal would succeed. 990 * 991 * When called with a false quiesce argument, actually attempt the 992 * removal. 993 * 994 * When called with a force argument, attempt to switch all TCBs to 995 * use the default stack instead of returning EBUSY. 996 * 997 * Returns 0 on success (or if the removal would succeed, or an error 998 * code on failure. 999 */ 1000 int 1001 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1002 bool force) 1003 { 1004 struct tcp_function *f; 1005 1006 if (blk == &tcp_def_funcblk) { 1007 /* You can't un-register the default */ 1008 return (EPERM); 1009 } 1010 rw_wlock(&tcp_function_lock); 1011 if (blk == tcp_func_set_ptr) { 1012 /* You can't free the current default */ 1013 rw_wunlock(&tcp_function_lock); 1014 return (EBUSY); 1015 } 1016 /* Mark the block so no more stacks can use it. */ 1017 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1018 /* 1019 * If TCBs are still attached to the stack, attempt to switch them 1020 * to the default stack. 1021 */ 1022 if (force && blk->tfb_refcnt) { 1023 struct inpcb *inp; 1024 struct tcpcb *tp; 1025 VNET_ITERATOR_DECL(vnet_iter); 1026 1027 rw_wunlock(&tcp_function_lock); 1028 1029 VNET_LIST_RLOCK(); 1030 VNET_FOREACH(vnet_iter) { 1031 CURVNET_SET(vnet_iter); 1032 INP_INFO_WLOCK(&V_tcbinfo); 1033 CK_LIST_FOREACH(inp, V_tcbinfo.ipi_listhead, inp_list) { 1034 INP_WLOCK(inp); 1035 if (inp->inp_flags & INP_TIMEWAIT) { 1036 INP_WUNLOCK(inp); 1037 continue; 1038 } 1039 tp = intotcpcb(inp); 1040 if (tp == NULL || tp->t_fb != blk) { 1041 INP_WUNLOCK(inp); 1042 continue; 1043 } 1044 tcp_switch_back_to_default(tp); 1045 INP_WUNLOCK(inp); 1046 } 1047 INP_INFO_WUNLOCK(&V_tcbinfo); 1048 CURVNET_RESTORE(); 1049 } 1050 VNET_LIST_RUNLOCK(); 1051 1052 rw_wlock(&tcp_function_lock); 1053 } 1054 if (blk->tfb_refcnt) { 1055 /* TCBs still attached. */ 1056 rw_wunlock(&tcp_function_lock); 1057 return (EBUSY); 1058 } 1059 if (quiesce) { 1060 /* Skip removal. */ 1061 rw_wunlock(&tcp_function_lock); 1062 return (0); 1063 } 1064 /* Remove any function names that map to this function block. */ 1065 while (find_tcp_fb_locked(blk, &f) != NULL) { 1066 TAILQ_REMOVE(&t_functions, f, tf_next); 1067 tcp_fb_cnt--; 1068 f->tf_fb = NULL; 1069 free(f, M_TCPFUNCTIONS); 1070 } 1071 rw_wunlock(&tcp_function_lock); 1072 return (0); 1073 } 1074 1075 void 1076 tcp_init(void) 1077 { 1078 const char *tcbhash_tuneable; 1079 int hashsize; 1080 1081 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 1082 1083 #ifdef TCP_HHOOK 1084 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1085 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1086 printf("%s: WARNING: unable to register helper hook\n", __func__); 1087 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1088 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1089 printf("%s: WARNING: unable to register helper hook\n", __func__); 1090 #endif 1091 #ifdef STATS 1092 if (tcp_stats_init()) 1093 printf("%s: WARNING: unable to initialise TCP stats\n", 1094 __func__); 1095 #endif 1096 hashsize = TCBHASHSIZE; 1097 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 1098 if (hashsize == 0) { 1099 /* 1100 * Auto tune the hash size based on maxsockets. 1101 * A perfect hash would have a 1:1 mapping 1102 * (hashsize = maxsockets) however it's been 1103 * suggested that O(2) average is better. 1104 */ 1105 hashsize = maketcp_hashsize(maxsockets / 4); 1106 /* 1107 * Our historical default is 512, 1108 * do not autotune lower than this. 1109 */ 1110 if (hashsize < 512) 1111 hashsize = 512; 1112 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 1113 printf("%s: %s auto tuned to %d\n", __func__, 1114 tcbhash_tuneable, hashsize); 1115 } 1116 /* 1117 * We require a hashsize to be a power of two. 1118 * Previously if it was not a power of two we would just reset it 1119 * back to 512, which could be a nasty surprise if you did not notice 1120 * the error message. 1121 * Instead what we do is clip it to the closest power of two lower 1122 * than the specified hash value. 1123 */ 1124 if (!powerof2(hashsize)) { 1125 int oldhashsize = hashsize; 1126 1127 hashsize = maketcp_hashsize(hashsize); 1128 /* prevent absurdly low value */ 1129 if (hashsize < 16) 1130 hashsize = 16; 1131 printf("%s: WARNING: TCB hash size not a power of 2, " 1132 "clipped from %d to %d.\n", __func__, oldhashsize, 1133 hashsize); 1134 } 1135 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 1136 "tcp_inpcb", tcp_inpcb_init, IPI_HASHFIELDS_4TUPLE); 1137 1138 /* 1139 * These have to be type stable for the benefit of the timers. 1140 */ 1141 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 1142 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1143 uma_zone_set_max(V_tcpcb_zone, maxsockets); 1144 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 1145 1146 tcp_tw_init(); 1147 syncache_init(); 1148 tcp_hc_init(); 1149 1150 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1151 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1152 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1153 1154 tcp_fastopen_init(); 1155 1156 /* Skip initialization of globals for non-default instances. */ 1157 if (!IS_DEFAULT_VNET(curvnet)) 1158 return; 1159 1160 tcp_reass_global_init(); 1161 1162 /* XXX virtualize those bellow? */ 1163 tcp_delacktime = TCPTV_DELACK; 1164 tcp_keepinit = TCPTV_KEEP_INIT; 1165 tcp_keepidle = TCPTV_KEEP_IDLE; 1166 tcp_keepintvl = TCPTV_KEEPINTVL; 1167 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1168 tcp_msl = TCPTV_MSL; 1169 tcp_rexmit_initial = TCPTV_RTOBASE; 1170 if (tcp_rexmit_initial < 1) 1171 tcp_rexmit_initial = 1; 1172 tcp_rexmit_min = TCPTV_MIN; 1173 if (tcp_rexmit_min < 1) 1174 tcp_rexmit_min = 1; 1175 tcp_persmin = TCPTV_PERSMIN; 1176 tcp_persmax = TCPTV_PERSMAX; 1177 tcp_rexmit_slop = TCPTV_CPU_VAR; 1178 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1179 tcp_tcbhashsize = hashsize; 1180 1181 /* Setup the tcp function block list */ 1182 TAILQ_INIT(&t_functions); 1183 rw_init(&tcp_function_lock, "tcp_func_lock"); 1184 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1185 #ifdef TCP_BLACKBOX 1186 /* Initialize the TCP logging data. */ 1187 tcp_log_init(); 1188 #endif 1189 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1190 1191 if (tcp_soreceive_stream) { 1192 #ifdef INET 1193 tcp_usrreqs.pru_soreceive = soreceive_stream; 1194 #endif 1195 #ifdef INET6 1196 tcp6_usrreqs.pru_soreceive = soreceive_stream; 1197 #endif /* INET6 */ 1198 } 1199 1200 #ifdef INET6 1201 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 1202 #else /* INET6 */ 1203 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 1204 #endif /* INET6 */ 1205 if (max_protohdr < TCP_MINPROTOHDR) 1206 max_protohdr = TCP_MINPROTOHDR; 1207 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 1208 panic("tcp_init"); 1209 #undef TCP_MINPROTOHDR 1210 1211 ISN_LOCK_INIT(); 1212 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1213 SHUTDOWN_PRI_DEFAULT); 1214 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 1215 EVENTHANDLER_PRI_ANY); 1216 1217 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1218 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1219 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1220 tcp_inp_lro_single_push = counter_u64_alloc(M_WAITOK); 1221 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1222 tcp_inp_lro_sack_wake = counter_u64_alloc(M_WAITOK); 1223 #ifdef TCPPCAP 1224 tcp_pcap_init(); 1225 #endif 1226 } 1227 1228 #ifdef VIMAGE 1229 static void 1230 tcp_destroy(void *unused __unused) 1231 { 1232 int n; 1233 #ifdef TCP_HHOOK 1234 int error; 1235 #endif 1236 1237 /* 1238 * All our processes are gone, all our sockets should be cleaned 1239 * up, which means, we should be past the tcp_discardcb() calls. 1240 * Sleep to let all tcpcb timers really disappear and cleanup. 1241 */ 1242 for (;;) { 1243 INP_LIST_RLOCK(&V_tcbinfo); 1244 n = V_tcbinfo.ipi_count; 1245 INP_LIST_RUNLOCK(&V_tcbinfo); 1246 if (n == 0) 1247 break; 1248 pause("tcpdes", hz / 10); 1249 } 1250 tcp_hc_destroy(); 1251 syncache_destroy(); 1252 tcp_tw_destroy(); 1253 in_pcbinfo_destroy(&V_tcbinfo); 1254 /* tcp_discardcb() clears the sack_holes up. */ 1255 uma_zdestroy(V_sack_hole_zone); 1256 uma_zdestroy(V_tcpcb_zone); 1257 1258 /* 1259 * Cannot free the zone until all tcpcbs are released as we attach 1260 * the allocations to them. 1261 */ 1262 tcp_fastopen_destroy(); 1263 1264 #ifdef TCP_HHOOK 1265 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1266 if (error != 0) { 1267 printf("%s: WARNING: unable to deregister helper hook " 1268 "type=%d, id=%d: error %d returned\n", __func__, 1269 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1270 } 1271 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1272 if (error != 0) { 1273 printf("%s: WARNING: unable to deregister helper hook " 1274 "type=%d, id=%d: error %d returned\n", __func__, 1275 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1276 } 1277 #endif 1278 } 1279 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1280 #endif 1281 1282 void 1283 tcp_fini(void *xtp) 1284 { 1285 1286 } 1287 1288 /* 1289 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1290 * tcp_template used to store this data in mbufs, but we now recopy it out 1291 * of the tcpcb each time to conserve mbufs. 1292 */ 1293 void 1294 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 1295 { 1296 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1297 1298 INP_WLOCK_ASSERT(inp); 1299 1300 #ifdef INET6 1301 if ((inp->inp_vflag & INP_IPV6) != 0) { 1302 struct ip6_hdr *ip6; 1303 1304 ip6 = (struct ip6_hdr *)ip_ptr; 1305 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1306 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1307 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1308 (IPV6_VERSION & IPV6_VERSION_MASK); 1309 ip6->ip6_nxt = IPPROTO_TCP; 1310 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1311 ip6->ip6_src = inp->in6p_laddr; 1312 ip6->ip6_dst = inp->in6p_faddr; 1313 } 1314 #endif /* INET6 */ 1315 #if defined(INET6) && defined(INET) 1316 else 1317 #endif 1318 #ifdef INET 1319 { 1320 struct ip *ip; 1321 1322 ip = (struct ip *)ip_ptr; 1323 ip->ip_v = IPVERSION; 1324 ip->ip_hl = 5; 1325 ip->ip_tos = inp->inp_ip_tos; 1326 ip->ip_len = 0; 1327 ip->ip_id = 0; 1328 ip->ip_off = 0; 1329 ip->ip_ttl = inp->inp_ip_ttl; 1330 ip->ip_sum = 0; 1331 ip->ip_p = IPPROTO_TCP; 1332 ip->ip_src = inp->inp_laddr; 1333 ip->ip_dst = inp->inp_faddr; 1334 } 1335 #endif /* INET */ 1336 th->th_sport = inp->inp_lport; 1337 th->th_dport = inp->inp_fport; 1338 th->th_seq = 0; 1339 th->th_ack = 0; 1340 th->th_x2 = 0; 1341 th->th_off = 5; 1342 th->th_flags = 0; 1343 th->th_win = 0; 1344 th->th_urp = 0; 1345 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1346 } 1347 1348 /* 1349 * Create template to be used to send tcp packets on a connection. 1350 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1351 * use for this function is in keepalives, which use tcp_respond. 1352 */ 1353 struct tcptemp * 1354 tcpip_maketemplate(struct inpcb *inp) 1355 { 1356 struct tcptemp *t; 1357 1358 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1359 if (t == NULL) 1360 return (NULL); 1361 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1362 return (t); 1363 } 1364 1365 /* 1366 * Send a single message to the TCP at address specified by 1367 * the given TCP/IP header. If m == NULL, then we make a copy 1368 * of the tcpiphdr at th and send directly to the addressed host. 1369 * This is used to force keep alive messages out using the TCP 1370 * template for a connection. If flags are given then we send 1371 * a message back to the TCP which originated the segment th, 1372 * and discard the mbuf containing it and any other attached mbufs. 1373 * 1374 * In any case the ack and sequence number of the transmitted 1375 * segment are as specified by the parameters. 1376 * 1377 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1378 */ 1379 void 1380 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1381 tcp_seq ack, tcp_seq seq, int flags) 1382 { 1383 struct tcpopt to; 1384 struct inpcb *inp; 1385 struct ip *ip; 1386 struct mbuf *optm; 1387 struct tcphdr *nth; 1388 u_char *optp; 1389 #ifdef INET6 1390 struct ip6_hdr *ip6; 1391 int isipv6; 1392 #endif /* INET6 */ 1393 int optlen, tlen, win; 1394 bool incl_opts; 1395 1396 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1397 NET_EPOCH_ASSERT(); 1398 1399 #ifdef INET6 1400 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1401 ip6 = ipgen; 1402 #endif /* INET6 */ 1403 ip = ipgen; 1404 1405 if (tp != NULL) { 1406 inp = tp->t_inpcb; 1407 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 1408 INP_WLOCK_ASSERT(inp); 1409 } else 1410 inp = NULL; 1411 1412 incl_opts = false; 1413 win = 0; 1414 if (tp != NULL) { 1415 if (!(flags & TH_RST)) { 1416 win = sbspace(&inp->inp_socket->so_rcv); 1417 if (win > TCP_MAXWIN << tp->rcv_scale) 1418 win = TCP_MAXWIN << tp->rcv_scale; 1419 } 1420 if ((tp->t_flags & TF_NOOPT) == 0) 1421 incl_opts = true; 1422 } 1423 if (m == NULL) { 1424 m = m_gethdr(M_NOWAIT, MT_DATA); 1425 if (m == NULL) 1426 return; 1427 m->m_data += max_linkhdr; 1428 #ifdef INET6 1429 if (isipv6) { 1430 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1431 sizeof(struct ip6_hdr)); 1432 ip6 = mtod(m, struct ip6_hdr *); 1433 nth = (struct tcphdr *)(ip6 + 1); 1434 } else 1435 #endif /* INET6 */ 1436 { 1437 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1438 ip = mtod(m, struct ip *); 1439 nth = (struct tcphdr *)(ip + 1); 1440 } 1441 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1442 flags = TH_ACK; 1443 } else if (!M_WRITABLE(m)) { 1444 struct mbuf *n; 1445 1446 /* Can't reuse 'm', allocate a new mbuf. */ 1447 n = m_gethdr(M_NOWAIT, MT_DATA); 1448 if (n == NULL) { 1449 m_freem(m); 1450 return; 1451 } 1452 1453 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1454 m_freem(m); 1455 m_freem(n); 1456 return; 1457 } 1458 1459 n->m_data += max_linkhdr; 1460 /* m_len is set later */ 1461 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1462 #ifdef INET6 1463 if (isipv6) { 1464 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1465 sizeof(struct ip6_hdr)); 1466 ip6 = mtod(n, struct ip6_hdr *); 1467 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1468 nth = (struct tcphdr *)(ip6 + 1); 1469 } else 1470 #endif /* INET6 */ 1471 { 1472 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1473 ip = mtod(n, struct ip *); 1474 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1475 nth = (struct tcphdr *)(ip + 1); 1476 } 1477 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1478 xchg(nth->th_dport, nth->th_sport, uint16_t); 1479 th = nth; 1480 m_freem(m); 1481 m = n; 1482 } else { 1483 /* 1484 * reuse the mbuf. 1485 * XXX MRT We inherit the FIB, which is lucky. 1486 */ 1487 m_freem(m->m_next); 1488 m->m_next = NULL; 1489 m->m_data = (caddr_t)ipgen; 1490 /* m_len is set later */ 1491 #ifdef INET6 1492 if (isipv6) { 1493 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1494 nth = (struct tcphdr *)(ip6 + 1); 1495 } else 1496 #endif /* INET6 */ 1497 { 1498 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1499 nth = (struct tcphdr *)(ip + 1); 1500 } 1501 if (th != nth) { 1502 /* 1503 * this is usually a case when an extension header 1504 * exists between the IPv6 header and the 1505 * TCP header. 1506 */ 1507 nth->th_sport = th->th_sport; 1508 nth->th_dport = th->th_dport; 1509 } 1510 xchg(nth->th_dport, nth->th_sport, uint16_t); 1511 #undef xchg 1512 } 1513 tlen = 0; 1514 #ifdef INET6 1515 if (isipv6) 1516 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1517 #endif 1518 #if defined(INET) && defined(INET6) 1519 else 1520 #endif 1521 #ifdef INET 1522 tlen = sizeof (struct tcpiphdr); 1523 #endif 1524 #ifdef INVARIANTS 1525 m->m_len = 0; 1526 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1527 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1528 m, tlen, (long)M_TRAILINGSPACE(m))); 1529 #endif 1530 m->m_len = tlen; 1531 to.to_flags = 0; 1532 if (incl_opts) { 1533 /* Make sure we have room. */ 1534 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1535 m->m_next = m_get(M_NOWAIT, MT_DATA); 1536 if (m->m_next) { 1537 optp = mtod(m->m_next, u_char *); 1538 optm = m->m_next; 1539 } else 1540 incl_opts = false; 1541 } else { 1542 optp = (u_char *) (nth + 1); 1543 optm = m; 1544 } 1545 } 1546 if (incl_opts) { 1547 /* Timestamps. */ 1548 if (tp->t_flags & TF_RCVD_TSTMP) { 1549 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1550 to.to_tsecr = tp->ts_recent; 1551 to.to_flags |= TOF_TS; 1552 } 1553 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1554 /* TCP-MD5 (RFC2385). */ 1555 if (tp->t_flags & TF_SIGNATURE) 1556 to.to_flags |= TOF_SIGNATURE; 1557 #endif 1558 /* Add the options. */ 1559 tlen += optlen = tcp_addoptions(&to, optp); 1560 1561 /* Update m_len in the correct mbuf. */ 1562 optm->m_len += optlen; 1563 } else 1564 optlen = 0; 1565 #ifdef INET6 1566 if (isipv6) { 1567 ip6->ip6_flow = 0; 1568 ip6->ip6_vfc = IPV6_VERSION; 1569 ip6->ip6_nxt = IPPROTO_TCP; 1570 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1571 } 1572 #endif 1573 #if defined(INET) && defined(INET6) 1574 else 1575 #endif 1576 #ifdef INET 1577 { 1578 ip->ip_len = htons(tlen); 1579 ip->ip_ttl = V_ip_defttl; 1580 if (V_path_mtu_discovery) 1581 ip->ip_off |= htons(IP_DF); 1582 } 1583 #endif 1584 m->m_pkthdr.len = tlen; 1585 m->m_pkthdr.rcvif = NULL; 1586 #ifdef MAC 1587 if (inp != NULL) { 1588 /* 1589 * Packet is associated with a socket, so allow the 1590 * label of the response to reflect the socket label. 1591 */ 1592 INP_WLOCK_ASSERT(inp); 1593 mac_inpcb_create_mbuf(inp, m); 1594 } else { 1595 /* 1596 * Packet is not associated with a socket, so possibly 1597 * update the label in place. 1598 */ 1599 mac_netinet_tcp_reply(m); 1600 } 1601 #endif 1602 nth->th_seq = htonl(seq); 1603 nth->th_ack = htonl(ack); 1604 nth->th_x2 = 0; 1605 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1606 nth->th_flags = flags; 1607 if (tp != NULL) 1608 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1609 else 1610 nth->th_win = htons((u_short)win); 1611 nth->th_urp = 0; 1612 1613 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1614 if (to.to_flags & TOF_SIGNATURE) { 1615 if (!TCPMD5_ENABLED() || 1616 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1617 m_freem(m); 1618 return; 1619 } 1620 } 1621 #endif 1622 1623 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1624 #ifdef INET6 1625 if (isipv6) { 1626 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1627 nth->th_sum = in6_cksum_pseudo(ip6, 1628 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1629 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1630 NULL, NULL); 1631 } 1632 #endif /* INET6 */ 1633 #if defined(INET6) && defined(INET) 1634 else 1635 #endif 1636 #ifdef INET 1637 { 1638 m->m_pkthdr.csum_flags = CSUM_TCP; 1639 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1640 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1641 } 1642 #endif /* INET */ 1643 #ifdef TCPDEBUG 1644 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1645 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1646 #endif 1647 TCP_PROBE3(debug__output, tp, th, m); 1648 if (flags & TH_RST) 1649 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1650 1651 #ifdef INET6 1652 if (isipv6) { 1653 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 1654 (void)ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1655 } 1656 #endif /* INET6 */ 1657 #if defined(INET) && defined(INET6) 1658 else 1659 #endif 1660 #ifdef INET 1661 { 1662 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 1663 (void)ip_output(m, NULL, NULL, 0, NULL, inp); 1664 } 1665 #endif 1666 } 1667 1668 /* 1669 * Create a new TCP control block, making an 1670 * empty reassembly queue and hooking it to the argument 1671 * protocol control block. The `inp' parameter must have 1672 * come from the zone allocator set up in tcp_init(). 1673 */ 1674 struct tcpcb * 1675 tcp_newtcpcb(struct inpcb *inp) 1676 { 1677 struct tcpcb_mem *tm; 1678 struct tcpcb *tp; 1679 #ifdef INET6 1680 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1681 #endif /* INET6 */ 1682 1683 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1684 if (tm == NULL) 1685 return (NULL); 1686 tp = &tm->tcb; 1687 1688 /* Initialise cc_var struct for this tcpcb. */ 1689 tp->ccv = &tm->ccv; 1690 tp->ccv->type = IPPROTO_TCP; 1691 tp->ccv->ccvc.tcp = tp; 1692 rw_rlock(&tcp_function_lock); 1693 tp->t_fb = tcp_func_set_ptr; 1694 refcount_acquire(&tp->t_fb->tfb_refcnt); 1695 rw_runlock(&tcp_function_lock); 1696 /* 1697 * Use the current system default CC algorithm. 1698 */ 1699 CC_LIST_RLOCK(); 1700 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1701 CC_ALGO(tp) = CC_DEFAULT(); 1702 CC_LIST_RUNLOCK(); 1703 1704 if (CC_ALGO(tp)->cb_init != NULL) 1705 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1706 if (tp->t_fb->tfb_tcp_fb_fini) 1707 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1708 refcount_release(&tp->t_fb->tfb_refcnt); 1709 uma_zfree(V_tcpcb_zone, tm); 1710 return (NULL); 1711 } 1712 1713 #ifdef TCP_HHOOK 1714 tp->osd = &tm->osd; 1715 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1716 if (tp->t_fb->tfb_tcp_fb_fini) 1717 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 1718 refcount_release(&tp->t_fb->tfb_refcnt); 1719 uma_zfree(V_tcpcb_zone, tm); 1720 return (NULL); 1721 } 1722 #endif 1723 1724 #ifdef VIMAGE 1725 tp->t_vnet = inp->inp_vnet; 1726 #endif 1727 tp->t_timers = &tm->tt; 1728 TAILQ_INIT(&tp->t_segq); 1729 tp->t_maxseg = 1730 #ifdef INET6 1731 isipv6 ? V_tcp_v6mssdflt : 1732 #endif /* INET6 */ 1733 V_tcp_mssdflt; 1734 1735 /* Set up our timeouts. */ 1736 callout_init(&tp->t_timers->tt_rexmt, 1); 1737 callout_init(&tp->t_timers->tt_persist, 1); 1738 callout_init(&tp->t_timers->tt_keep, 1); 1739 callout_init(&tp->t_timers->tt_2msl, 1); 1740 callout_init(&tp->t_timers->tt_delack, 1); 1741 1742 if (V_tcp_do_rfc1323) 1743 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1744 if (V_tcp_do_sack) 1745 tp->t_flags |= TF_SACK_PERMIT; 1746 TAILQ_INIT(&tp->snd_holes); 1747 /* 1748 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1749 * is called. 1750 */ 1751 in_pcbref(inp); /* Reference for tcpcb */ 1752 tp->t_inpcb = inp; 1753 1754 /* 1755 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1756 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1757 * reasonable initial retransmit time. 1758 */ 1759 tp->t_srtt = TCPTV_SRTTBASE; 1760 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1761 tp->t_rttmin = tcp_rexmit_min; 1762 tp->t_rxtcur = tcp_rexmit_initial; 1763 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1764 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1765 tp->t_rcvtime = ticks; 1766 /* 1767 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1768 * because the socket may be bound to an IPv6 wildcard address, 1769 * which may match an IPv4-mapped IPv6 address. 1770 */ 1771 inp->inp_ip_ttl = V_ip_defttl; 1772 inp->inp_ppcb = tp; 1773 #ifdef TCPPCAP 1774 /* 1775 * Init the TCP PCAP queues. 1776 */ 1777 tcp_pcap_tcpcb_init(tp); 1778 #endif 1779 #ifdef TCP_BLACKBOX 1780 /* Initialize the per-TCPCB log data. */ 1781 tcp_log_tcpcbinit(tp); 1782 #endif 1783 if (tp->t_fb->tfb_tcp_fb_init) { 1784 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1785 } 1786 #ifdef STATS 1787 if (V_tcp_perconn_stats_enable == 1) 1788 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 1789 #endif 1790 return (tp); /* XXX */ 1791 } 1792 1793 /* 1794 * Switch the congestion control algorithm back to NewReno for any active 1795 * control blocks using an algorithm which is about to go away. 1796 * This ensures the CC framework can allow the unload to proceed without leaving 1797 * any dangling pointers which would trigger a panic. 1798 * Returning non-zero would inform the CC framework that something went wrong 1799 * and it would be unsafe to allow the unload to proceed. However, there is no 1800 * way for this to occur with this implementation so we always return zero. 1801 */ 1802 int 1803 tcp_ccalgounload(struct cc_algo *unload_algo) 1804 { 1805 struct cc_algo *tmpalgo; 1806 struct inpcb *inp; 1807 struct tcpcb *tp; 1808 VNET_ITERATOR_DECL(vnet_iter); 1809 1810 /* 1811 * Check all active control blocks across all network stacks and change 1812 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1813 * requires cleanup code to be run, call it. 1814 */ 1815 VNET_LIST_RLOCK(); 1816 VNET_FOREACH(vnet_iter) { 1817 CURVNET_SET(vnet_iter); 1818 INP_INFO_WLOCK(&V_tcbinfo); 1819 /* 1820 * New connections already part way through being initialised 1821 * with the CC algo we're removing will not race with this code 1822 * because the INP_INFO_WLOCK is held during initialisation. We 1823 * therefore don't enter the loop below until the connection 1824 * list has stabilised. 1825 */ 1826 CK_LIST_FOREACH(inp, &V_tcb, inp_list) { 1827 INP_WLOCK(inp); 1828 /* Important to skip tcptw structs. */ 1829 if (!(inp->inp_flags & INP_TIMEWAIT) && 1830 (tp = intotcpcb(inp)) != NULL) { 1831 /* 1832 * By holding INP_WLOCK here, we are assured 1833 * that the connection is not currently 1834 * executing inside the CC module's functions 1835 * i.e. it is safe to make the switch back to 1836 * NewReno. 1837 */ 1838 if (CC_ALGO(tp) == unload_algo) { 1839 tmpalgo = CC_ALGO(tp); 1840 if (tmpalgo->cb_destroy != NULL) 1841 tmpalgo->cb_destroy(tp->ccv); 1842 CC_DATA(tp) = NULL; 1843 /* 1844 * NewReno may allocate memory on 1845 * demand for certain stateful 1846 * configuration as needed, but is 1847 * coded to never fail on memory 1848 * allocation failure so it is a safe 1849 * fallback. 1850 */ 1851 CC_ALGO(tp) = &newreno_cc_algo; 1852 } 1853 } 1854 INP_WUNLOCK(inp); 1855 } 1856 INP_INFO_WUNLOCK(&V_tcbinfo); 1857 CURVNET_RESTORE(); 1858 } 1859 VNET_LIST_RUNLOCK(); 1860 1861 return (0); 1862 } 1863 1864 /* 1865 * Drop a TCP connection, reporting 1866 * the specified error. If connection is synchronized, 1867 * then send a RST to peer. 1868 */ 1869 struct tcpcb * 1870 tcp_drop(struct tcpcb *tp, int errno) 1871 { 1872 struct socket *so = tp->t_inpcb->inp_socket; 1873 1874 NET_EPOCH_ASSERT(); 1875 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1876 INP_WLOCK_ASSERT(tp->t_inpcb); 1877 1878 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1879 tcp_state_change(tp, TCPS_CLOSED); 1880 (void) tp->t_fb->tfb_tcp_output(tp); 1881 TCPSTAT_INC(tcps_drops); 1882 } else 1883 TCPSTAT_INC(tcps_conndrops); 1884 if (errno == ETIMEDOUT && tp->t_softerror) 1885 errno = tp->t_softerror; 1886 so->so_error = errno; 1887 return (tcp_close(tp)); 1888 } 1889 1890 void 1891 tcp_discardcb(struct tcpcb *tp) 1892 { 1893 struct inpcb *inp = tp->t_inpcb; 1894 struct socket *so = inp->inp_socket; 1895 #ifdef INET6 1896 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1897 #endif /* INET6 */ 1898 int released __unused; 1899 1900 INP_WLOCK_ASSERT(inp); 1901 1902 /* 1903 * Make sure that all of our timers are stopped before we delete the 1904 * PCB. 1905 * 1906 * If stopping a timer fails, we schedule a discard function in same 1907 * callout, and the last discard function called will take care of 1908 * deleting the tcpcb. 1909 */ 1910 tp->t_timers->tt_draincnt = 0; 1911 tcp_timer_stop(tp, TT_REXMT); 1912 tcp_timer_stop(tp, TT_PERSIST); 1913 tcp_timer_stop(tp, TT_KEEP); 1914 tcp_timer_stop(tp, TT_2MSL); 1915 tcp_timer_stop(tp, TT_DELACK); 1916 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1917 /* 1918 * Call the stop-all function of the methods, 1919 * this function should call the tcp_timer_stop() 1920 * method with each of the function specific timeouts. 1921 * That stop will be called via the tfb_tcp_timer_stop() 1922 * which should use the async drain function of the 1923 * callout system (see tcp_var.h). 1924 */ 1925 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1926 } 1927 1928 /* 1929 * If we got enough samples through the srtt filter, 1930 * save the rtt and rttvar in the routing entry. 1931 * 'Enough' is arbitrarily defined as 4 rtt samples. 1932 * 4 samples is enough for the srtt filter to converge 1933 * to within enough % of the correct value; fewer samples 1934 * and we could save a bogus rtt. The danger is not high 1935 * as tcp quickly recovers from everything. 1936 * XXX: Works very well but needs some more statistics! 1937 */ 1938 if (tp->t_rttupdated >= 4) { 1939 struct hc_metrics_lite metrics; 1940 uint32_t ssthresh; 1941 1942 bzero(&metrics, sizeof(metrics)); 1943 /* 1944 * Update the ssthresh always when the conditions below 1945 * are satisfied. This gives us better new start value 1946 * for the congestion avoidance for new connections. 1947 * ssthresh is only set if packet loss occurred on a session. 1948 * 1949 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1950 * being torn down. Ideally this code would not use 'so'. 1951 */ 1952 ssthresh = tp->snd_ssthresh; 1953 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1954 /* 1955 * convert the limit from user data bytes to 1956 * packets then to packet data bytes. 1957 */ 1958 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1959 if (ssthresh < 2) 1960 ssthresh = 2; 1961 ssthresh *= (tp->t_maxseg + 1962 #ifdef INET6 1963 (isipv6 ? sizeof (struct ip6_hdr) + 1964 sizeof (struct tcphdr) : 1965 #endif 1966 sizeof (struct tcpiphdr) 1967 #ifdef INET6 1968 ) 1969 #endif 1970 ); 1971 } else 1972 ssthresh = 0; 1973 metrics.rmx_ssthresh = ssthresh; 1974 1975 metrics.rmx_rtt = tp->t_srtt; 1976 metrics.rmx_rttvar = tp->t_rttvar; 1977 metrics.rmx_cwnd = tp->snd_cwnd; 1978 metrics.rmx_sendpipe = 0; 1979 metrics.rmx_recvpipe = 0; 1980 1981 tcp_hc_update(&inp->inp_inc, &metrics); 1982 } 1983 1984 /* free the reassembly queue, if any */ 1985 tcp_reass_flush(tp); 1986 1987 #ifdef TCP_OFFLOAD 1988 /* Disconnect offload device, if any. */ 1989 if (tp->t_flags & TF_TOE) 1990 tcp_offload_detach(tp); 1991 #endif 1992 1993 tcp_free_sackholes(tp); 1994 1995 #ifdef TCPPCAP 1996 /* Free the TCP PCAP queues. */ 1997 tcp_pcap_drain(&(tp->t_inpkts)); 1998 tcp_pcap_drain(&(tp->t_outpkts)); 1999 #endif 2000 2001 /* Allow the CC algorithm to clean up after itself. */ 2002 if (CC_ALGO(tp)->cb_destroy != NULL) 2003 CC_ALGO(tp)->cb_destroy(tp->ccv); 2004 CC_DATA(tp) = NULL; 2005 2006 #ifdef TCP_HHOOK 2007 khelp_destroy_osd(tp->osd); 2008 #endif 2009 #ifdef STATS 2010 stats_blob_destroy(tp->t_stats); 2011 #endif 2012 2013 CC_ALGO(tp) = NULL; 2014 inp->inp_ppcb = NULL; 2015 if (tp->t_timers->tt_draincnt == 0) { 2016 /* We own the last reference on tcpcb, let's free it. */ 2017 #ifdef TCP_BLACKBOX 2018 tcp_log_tcpcbfini(tp); 2019 #endif 2020 TCPSTATES_DEC(tp->t_state); 2021 if (tp->t_fb->tfb_tcp_fb_fini) 2022 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2023 refcount_release(&tp->t_fb->tfb_refcnt); 2024 tp->t_inpcb = NULL; 2025 uma_zfree(V_tcpcb_zone, tp); 2026 released = in_pcbrele_wlocked(inp); 2027 KASSERT(!released, ("%s: inp %p should not have been released " 2028 "here", __func__, inp)); 2029 } 2030 } 2031 2032 void 2033 tcp_timer_discard(void *ptp) 2034 { 2035 struct inpcb *inp; 2036 struct tcpcb *tp; 2037 struct epoch_tracker et; 2038 2039 tp = (struct tcpcb *)ptp; 2040 CURVNET_SET(tp->t_vnet); 2041 NET_EPOCH_ENTER(et); 2042 inp = tp->t_inpcb; 2043 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 2044 __func__, tp)); 2045 INP_WLOCK(inp); 2046 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 2047 ("%s: tcpcb has to be stopped here", __func__)); 2048 tp->t_timers->tt_draincnt--; 2049 if (tp->t_timers->tt_draincnt == 0) { 2050 /* We own the last reference on this tcpcb, let's free it. */ 2051 #ifdef TCP_BLACKBOX 2052 tcp_log_tcpcbfini(tp); 2053 #endif 2054 TCPSTATES_DEC(tp->t_state); 2055 if (tp->t_fb->tfb_tcp_fb_fini) 2056 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2057 refcount_release(&tp->t_fb->tfb_refcnt); 2058 tp->t_inpcb = NULL; 2059 uma_zfree(V_tcpcb_zone, tp); 2060 if (in_pcbrele_wlocked(inp)) { 2061 NET_EPOCH_EXIT(et); 2062 CURVNET_RESTORE(); 2063 return; 2064 } 2065 } 2066 INP_WUNLOCK(inp); 2067 NET_EPOCH_EXIT(et); 2068 CURVNET_RESTORE(); 2069 } 2070 2071 /* 2072 * Attempt to close a TCP control block, marking it as dropped, and freeing 2073 * the socket if we hold the only reference. 2074 */ 2075 struct tcpcb * 2076 tcp_close(struct tcpcb *tp) 2077 { 2078 struct inpcb *inp = tp->t_inpcb; 2079 struct socket *so; 2080 2081 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2082 INP_WLOCK_ASSERT(inp); 2083 2084 #ifdef TCP_OFFLOAD 2085 if (tp->t_state == TCPS_LISTEN) 2086 tcp_offload_listen_stop(tp); 2087 #endif 2088 /* 2089 * This releases the TFO pending counter resource for TFO listen 2090 * sockets as well as passively-created TFO sockets that transition 2091 * from SYN_RECEIVED to CLOSED. 2092 */ 2093 if (tp->t_tfo_pending) { 2094 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2095 tp->t_tfo_pending = NULL; 2096 } 2097 in_pcbdrop(inp); 2098 TCPSTAT_INC(tcps_closed); 2099 if (tp->t_state != TCPS_CLOSED) 2100 tcp_state_change(tp, TCPS_CLOSED); 2101 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2102 so = inp->inp_socket; 2103 soisdisconnected(so); 2104 if (inp->inp_flags & INP_SOCKREF) { 2105 KASSERT(so->so_state & SS_PROTOREF, 2106 ("tcp_close: !SS_PROTOREF")); 2107 inp->inp_flags &= ~INP_SOCKREF; 2108 INP_WUNLOCK(inp); 2109 SOCK_LOCK(so); 2110 so->so_state &= ~SS_PROTOREF; 2111 sofree(so); 2112 return (NULL); 2113 } 2114 return (tp); 2115 } 2116 2117 void 2118 tcp_drain(void) 2119 { 2120 VNET_ITERATOR_DECL(vnet_iter); 2121 2122 if (!do_tcpdrain) 2123 return; 2124 2125 VNET_LIST_RLOCK_NOSLEEP(); 2126 VNET_FOREACH(vnet_iter) { 2127 CURVNET_SET(vnet_iter); 2128 struct inpcb *inpb; 2129 struct tcpcb *tcpb; 2130 2131 /* 2132 * Walk the tcpbs, if existing, and flush the reassembly queue, 2133 * if there is one... 2134 * XXX: The "Net/3" implementation doesn't imply that the TCP 2135 * reassembly queue should be flushed, but in a situation 2136 * where we're really low on mbufs, this is potentially 2137 * useful. 2138 */ 2139 INP_INFO_WLOCK(&V_tcbinfo); 2140 CK_LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 2141 INP_WLOCK(inpb); 2142 if (inpb->inp_flags & INP_TIMEWAIT) { 2143 INP_WUNLOCK(inpb); 2144 continue; 2145 } 2146 if ((tcpb = intotcpcb(inpb)) != NULL) { 2147 tcp_reass_flush(tcpb); 2148 tcp_clean_sackreport(tcpb); 2149 #ifdef TCP_BLACKBOX 2150 tcp_log_drain(tcpb); 2151 #endif 2152 #ifdef TCPPCAP 2153 if (tcp_pcap_aggressive_free) { 2154 /* Free the TCP PCAP queues. */ 2155 tcp_pcap_drain(&(tcpb->t_inpkts)); 2156 tcp_pcap_drain(&(tcpb->t_outpkts)); 2157 } 2158 #endif 2159 } 2160 INP_WUNLOCK(inpb); 2161 } 2162 INP_INFO_WUNLOCK(&V_tcbinfo); 2163 CURVNET_RESTORE(); 2164 } 2165 VNET_LIST_RUNLOCK_NOSLEEP(); 2166 } 2167 2168 /* 2169 * Notify a tcp user of an asynchronous error; 2170 * store error as soft error, but wake up user 2171 * (for now, won't do anything until can select for soft error). 2172 * 2173 * Do not wake up user since there currently is no mechanism for 2174 * reporting soft errors (yet - a kqueue filter may be added). 2175 */ 2176 static struct inpcb * 2177 tcp_notify(struct inpcb *inp, int error) 2178 { 2179 struct tcpcb *tp; 2180 2181 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2182 INP_WLOCK_ASSERT(inp); 2183 2184 if ((inp->inp_flags & INP_TIMEWAIT) || 2185 (inp->inp_flags & INP_DROPPED)) 2186 return (inp); 2187 2188 tp = intotcpcb(inp); 2189 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2190 2191 /* 2192 * Ignore some errors if we are hooked up. 2193 * If connection hasn't completed, has retransmitted several times, 2194 * and receives a second error, give up now. This is better 2195 * than waiting a long time to establish a connection that 2196 * can never complete. 2197 */ 2198 if (tp->t_state == TCPS_ESTABLISHED && 2199 (error == EHOSTUNREACH || error == ENETUNREACH || 2200 error == EHOSTDOWN)) { 2201 if (inp->inp_route.ro_rt) { 2202 RTFREE(inp->inp_route.ro_rt); 2203 inp->inp_route.ro_rt = (struct rtentry *)NULL; 2204 } 2205 return (inp); 2206 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2207 tp->t_softerror) { 2208 tp = tcp_drop(tp, error); 2209 if (tp != NULL) 2210 return (inp); 2211 else 2212 return (NULL); 2213 } else { 2214 tp->t_softerror = error; 2215 return (inp); 2216 } 2217 #if 0 2218 wakeup( &so->so_timeo); 2219 sorwakeup(so); 2220 sowwakeup(so); 2221 #endif 2222 } 2223 2224 static int 2225 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2226 { 2227 struct epoch_tracker et; 2228 struct inpcb *inp; 2229 struct xinpgen xig; 2230 int error; 2231 2232 if (req->newptr != NULL) 2233 return (EPERM); 2234 2235 if (req->oldptr == NULL) { 2236 int n; 2237 2238 n = V_tcbinfo.ipi_count + 2239 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2240 n += imax(n / 8, 10); 2241 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2242 return (0); 2243 } 2244 2245 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2246 return (error); 2247 2248 bzero(&xig, sizeof(xig)); 2249 xig.xig_len = sizeof xig; 2250 xig.xig_count = V_tcbinfo.ipi_count + 2251 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2252 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2253 xig.xig_sogen = so_gencnt; 2254 error = SYSCTL_OUT(req, &xig, sizeof xig); 2255 if (error) 2256 return (error); 2257 2258 error = syncache_pcblist(req); 2259 if (error) 2260 return (error); 2261 2262 NET_EPOCH_ENTER(et); 2263 for (inp = CK_LIST_FIRST(V_tcbinfo.ipi_listhead); 2264 inp != NULL; 2265 inp = CK_LIST_NEXT(inp, inp_list)) { 2266 INP_RLOCK(inp); 2267 if (inp->inp_gencnt <= xig.xig_gen) { 2268 int crerr; 2269 2270 /* 2271 * XXX: This use of cr_cansee(), introduced with 2272 * TCP state changes, is not quite right, but for 2273 * now, better than nothing. 2274 */ 2275 if (inp->inp_flags & INP_TIMEWAIT) { 2276 if (intotw(inp) != NULL) 2277 crerr = cr_cansee(req->td->td_ucred, 2278 intotw(inp)->tw_cred); 2279 else 2280 crerr = EINVAL; /* Skip this inp. */ 2281 } else 2282 crerr = cr_canseeinpcb(req->td->td_ucred, inp); 2283 if (crerr == 0) { 2284 struct xtcpcb xt; 2285 2286 tcp_inptoxtp(inp, &xt); 2287 INP_RUNLOCK(inp); 2288 error = SYSCTL_OUT(req, &xt, sizeof xt); 2289 if (error) 2290 break; 2291 else 2292 continue; 2293 } 2294 } 2295 INP_RUNLOCK(inp); 2296 } 2297 NET_EPOCH_EXIT(et); 2298 2299 if (!error) { 2300 /* 2301 * Give the user an updated idea of our state. 2302 * If the generation differs from what we told 2303 * her before, she knows that something happened 2304 * while we were processing this request, and it 2305 * might be necessary to retry. 2306 */ 2307 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2308 xig.xig_sogen = so_gencnt; 2309 xig.xig_count = V_tcbinfo.ipi_count + 2310 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2311 error = SYSCTL_OUT(req, &xig, sizeof xig); 2312 } 2313 2314 return (error); 2315 } 2316 2317 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2318 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 2319 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 2320 2321 #ifdef INET 2322 static int 2323 tcp_getcred(SYSCTL_HANDLER_ARGS) 2324 { 2325 struct xucred xuc; 2326 struct sockaddr_in addrs[2]; 2327 struct epoch_tracker et; 2328 struct inpcb *inp; 2329 int error; 2330 2331 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2332 if (error) 2333 return (error); 2334 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2335 if (error) 2336 return (error); 2337 NET_EPOCH_ENTER(et); 2338 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2339 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2340 NET_EPOCH_EXIT(et); 2341 if (inp != NULL) { 2342 if (inp->inp_socket == NULL) 2343 error = ENOENT; 2344 if (error == 0) 2345 error = cr_canseeinpcb(req->td->td_ucred, inp); 2346 if (error == 0) 2347 cru2x(inp->inp_cred, &xuc); 2348 INP_RUNLOCK(inp); 2349 } else 2350 error = ENOENT; 2351 if (error == 0) 2352 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2353 return (error); 2354 } 2355 2356 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2357 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2358 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 2359 #endif /* INET */ 2360 2361 #ifdef INET6 2362 static int 2363 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2364 { 2365 struct epoch_tracker et; 2366 struct xucred xuc; 2367 struct sockaddr_in6 addrs[2]; 2368 struct inpcb *inp; 2369 int error; 2370 #ifdef INET 2371 int mapped = 0; 2372 #endif 2373 2374 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2375 if (error) 2376 return (error); 2377 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2378 if (error) 2379 return (error); 2380 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2381 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2382 return (error); 2383 } 2384 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2385 #ifdef INET 2386 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2387 mapped = 1; 2388 else 2389 #endif 2390 return (EINVAL); 2391 } 2392 2393 NET_EPOCH_ENTER(et); 2394 #ifdef INET 2395 if (mapped == 1) 2396 inp = in_pcblookup(&V_tcbinfo, 2397 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2398 addrs[1].sin6_port, 2399 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2400 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2401 else 2402 #endif 2403 inp = in6_pcblookup(&V_tcbinfo, 2404 &addrs[1].sin6_addr, addrs[1].sin6_port, 2405 &addrs[0].sin6_addr, addrs[0].sin6_port, 2406 INPLOOKUP_RLOCKPCB, NULL); 2407 NET_EPOCH_EXIT(et); 2408 if (inp != NULL) { 2409 if (inp->inp_socket == NULL) 2410 error = ENOENT; 2411 if (error == 0) 2412 error = cr_canseeinpcb(req->td->td_ucred, inp); 2413 if (error == 0) 2414 cru2x(inp->inp_cred, &xuc); 2415 INP_RUNLOCK(inp); 2416 } else 2417 error = ENOENT; 2418 if (error == 0) 2419 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2420 return (error); 2421 } 2422 2423 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2424 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 2425 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 2426 #endif /* INET6 */ 2427 2428 2429 #ifdef INET 2430 void 2431 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 2432 { 2433 struct ip *ip = vip; 2434 struct tcphdr *th; 2435 struct in_addr faddr; 2436 struct inpcb *inp; 2437 struct tcpcb *tp; 2438 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2439 struct icmp *icp; 2440 struct in_conninfo inc; 2441 tcp_seq icmp_tcp_seq; 2442 int mtu; 2443 2444 faddr = ((struct sockaddr_in *)sa)->sin_addr; 2445 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 2446 return; 2447 2448 if (cmd == PRC_MSGSIZE) 2449 notify = tcp_mtudisc_notify; 2450 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2451 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2452 cmd == PRC_TIMXCEED_INTRANS) && ip) 2453 notify = tcp_drop_syn_sent; 2454 2455 /* 2456 * Hostdead is ugly because it goes linearly through all PCBs. 2457 * XXX: We never get this from ICMP, otherwise it makes an 2458 * excellent DoS attack on machines with many connections. 2459 */ 2460 else if (cmd == PRC_HOSTDEAD) 2461 ip = NULL; 2462 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 2463 return; 2464 2465 if (ip == NULL) { 2466 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 2467 return; 2468 } 2469 2470 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 2471 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2472 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 2473 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2474 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2475 /* signal EHOSTDOWN, as it flushes the cached route */ 2476 inp = (*notify)(inp, EHOSTDOWN); 2477 goto out; 2478 } 2479 icmp_tcp_seq = th->th_seq; 2480 if (inp != NULL) { 2481 if (!(inp->inp_flags & INP_TIMEWAIT) && 2482 !(inp->inp_flags & INP_DROPPED) && 2483 !(inp->inp_socket == NULL)) { 2484 tp = intotcpcb(inp); 2485 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2486 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2487 if (cmd == PRC_MSGSIZE) { 2488 /* 2489 * MTU discovery: 2490 * If we got a needfrag set the MTU 2491 * in the route to the suggested new 2492 * value (if given) and then notify. 2493 */ 2494 mtu = ntohs(icp->icmp_nextmtu); 2495 /* 2496 * If no alternative MTU was 2497 * proposed, try the next smaller 2498 * one. 2499 */ 2500 if (!mtu) 2501 mtu = ip_next_mtu( 2502 ntohs(ip->ip_len), 1); 2503 if (mtu < V_tcp_minmss + 2504 sizeof(struct tcpiphdr)) 2505 mtu = V_tcp_minmss + 2506 sizeof(struct tcpiphdr); 2507 /* 2508 * Only process the offered MTU if it 2509 * is smaller than the current one. 2510 */ 2511 if (mtu < tp->t_maxseg + 2512 sizeof(struct tcpiphdr)) { 2513 bzero(&inc, sizeof(inc)); 2514 inc.inc_faddr = faddr; 2515 inc.inc_fibnum = 2516 inp->inp_inc.inc_fibnum; 2517 tcp_hc_updatemtu(&inc, mtu); 2518 tcp_mtudisc(inp, mtu); 2519 } 2520 } else 2521 inp = (*notify)(inp, 2522 inetctlerrmap[cmd]); 2523 } 2524 } 2525 } else { 2526 bzero(&inc, sizeof(inc)); 2527 inc.inc_fport = th->th_dport; 2528 inc.inc_lport = th->th_sport; 2529 inc.inc_faddr = faddr; 2530 inc.inc_laddr = ip->ip_src; 2531 syncache_unreach(&inc, icmp_tcp_seq); 2532 } 2533 out: 2534 if (inp != NULL) 2535 INP_WUNLOCK(inp); 2536 } 2537 #endif /* INET */ 2538 2539 #ifdef INET6 2540 void 2541 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2542 { 2543 struct in6_addr *dst; 2544 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2545 struct ip6_hdr *ip6; 2546 struct mbuf *m; 2547 struct inpcb *inp; 2548 struct tcpcb *tp; 2549 struct icmp6_hdr *icmp6; 2550 struct ip6ctlparam *ip6cp = NULL; 2551 const struct sockaddr_in6 *sa6_src = NULL; 2552 struct in_conninfo inc; 2553 struct tcp_ports { 2554 uint16_t th_sport; 2555 uint16_t th_dport; 2556 } t_ports; 2557 tcp_seq icmp_tcp_seq; 2558 unsigned int mtu; 2559 unsigned int off; 2560 2561 if (sa->sa_family != AF_INET6 || 2562 sa->sa_len != sizeof(struct sockaddr_in6)) 2563 return; 2564 2565 /* if the parameter is from icmp6, decode it. */ 2566 if (d != NULL) { 2567 ip6cp = (struct ip6ctlparam *)d; 2568 icmp6 = ip6cp->ip6c_icmp6; 2569 m = ip6cp->ip6c_m; 2570 ip6 = ip6cp->ip6c_ip6; 2571 off = ip6cp->ip6c_off; 2572 sa6_src = ip6cp->ip6c_src; 2573 dst = ip6cp->ip6c_finaldst; 2574 } else { 2575 m = NULL; 2576 ip6 = NULL; 2577 off = 0; /* fool gcc */ 2578 sa6_src = &sa6_any; 2579 dst = NULL; 2580 } 2581 2582 if (cmd == PRC_MSGSIZE) 2583 notify = tcp_mtudisc_notify; 2584 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2585 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2586 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2587 notify = tcp_drop_syn_sent; 2588 2589 /* 2590 * Hostdead is ugly because it goes linearly through all PCBs. 2591 * XXX: We never get this from ICMP, otherwise it makes an 2592 * excellent DoS attack on machines with many connections. 2593 */ 2594 else if (cmd == PRC_HOSTDEAD) 2595 ip6 = NULL; 2596 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2597 return; 2598 2599 if (ip6 == NULL) { 2600 in6_pcbnotify(&V_tcbinfo, sa, 0, 2601 (const struct sockaddr *)sa6_src, 2602 0, cmd, NULL, notify); 2603 return; 2604 } 2605 2606 /* Check if we can safely get the ports from the tcp hdr */ 2607 if (m == NULL || 2608 (m->m_pkthdr.len < 2609 (int32_t) (off + sizeof(struct tcp_ports)))) { 2610 return; 2611 } 2612 bzero(&t_ports, sizeof(struct tcp_ports)); 2613 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2614 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2615 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2616 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2617 /* signal EHOSTDOWN, as it flushes the cached route */ 2618 inp = (*notify)(inp, EHOSTDOWN); 2619 goto out; 2620 } 2621 off += sizeof(struct tcp_ports); 2622 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2623 goto out; 2624 } 2625 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2626 if (inp != NULL) { 2627 if (!(inp->inp_flags & INP_TIMEWAIT) && 2628 !(inp->inp_flags & INP_DROPPED) && 2629 !(inp->inp_socket == NULL)) { 2630 tp = intotcpcb(inp); 2631 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2632 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2633 if (cmd == PRC_MSGSIZE) { 2634 /* 2635 * MTU discovery: 2636 * If we got a needfrag set the MTU 2637 * in the route to the suggested new 2638 * value (if given) and then notify. 2639 */ 2640 mtu = ntohl(icmp6->icmp6_mtu); 2641 /* 2642 * If no alternative MTU was 2643 * proposed, or the proposed 2644 * MTU was too small, set to 2645 * the min. 2646 */ 2647 if (mtu < IPV6_MMTU) 2648 mtu = IPV6_MMTU - 8; 2649 bzero(&inc, sizeof(inc)); 2650 inc.inc_fibnum = M_GETFIB(m); 2651 inc.inc_flags |= INC_ISIPV6; 2652 inc.inc6_faddr = *dst; 2653 if (in6_setscope(&inc.inc6_faddr, 2654 m->m_pkthdr.rcvif, NULL)) 2655 goto out; 2656 /* 2657 * Only process the offered MTU if it 2658 * is smaller than the current one. 2659 */ 2660 if (mtu < tp->t_maxseg + 2661 sizeof (struct tcphdr) + 2662 sizeof (struct ip6_hdr)) { 2663 tcp_hc_updatemtu(&inc, mtu); 2664 tcp_mtudisc(inp, mtu); 2665 ICMP6STAT_INC(icp6s_pmtuchg); 2666 } 2667 } else 2668 inp = (*notify)(inp, 2669 inet6ctlerrmap[cmd]); 2670 } 2671 } 2672 } else { 2673 bzero(&inc, sizeof(inc)); 2674 inc.inc_fibnum = M_GETFIB(m); 2675 inc.inc_flags |= INC_ISIPV6; 2676 inc.inc_fport = t_ports.th_dport; 2677 inc.inc_lport = t_ports.th_sport; 2678 inc.inc6_faddr = *dst; 2679 inc.inc6_laddr = ip6->ip6_src; 2680 syncache_unreach(&inc, icmp_tcp_seq); 2681 } 2682 out: 2683 if (inp != NULL) 2684 INP_WUNLOCK(inp); 2685 } 2686 #endif /* INET6 */ 2687 2688 static uint32_t 2689 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 2690 { 2691 SIPHASH_CTX ctx; 2692 uint32_t hash[2]; 2693 2694 KASSERT(len >= SIPHASH_KEY_LENGTH, 2695 ("%s: keylen %u too short ", __func__, len)); 2696 SipHash24_Init(&ctx); 2697 SipHash_SetKey(&ctx, (uint8_t *)key); 2698 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 2699 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 2700 switch (inc->inc_flags & INC_ISIPV6) { 2701 #ifdef INET 2702 case 0: 2703 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 2704 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 2705 break; 2706 #endif 2707 #ifdef INET6 2708 case INC_ISIPV6: 2709 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 2710 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 2711 break; 2712 #endif 2713 } 2714 SipHash_Final((uint8_t *)hash, &ctx); 2715 2716 return (hash[0] ^ hash[1]); 2717 } 2718 2719 uint32_t 2720 tcp_new_ts_offset(struct in_conninfo *inc) 2721 { 2722 struct in_conninfo inc_store, *local_inc; 2723 2724 if (!V_tcp_ts_offset_per_conn) { 2725 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 2726 inc_store.inc_lport = 0; 2727 inc_store.inc_fport = 0; 2728 local_inc = &inc_store; 2729 } else { 2730 local_inc = inc; 2731 } 2732 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 2733 sizeof(V_ts_offset_secret))); 2734 } 2735 2736 /* 2737 * Following is where TCP initial sequence number generation occurs. 2738 * 2739 * There are two places where we must use initial sequence numbers: 2740 * 1. In SYN-ACK packets. 2741 * 2. In SYN packets. 2742 * 2743 * All ISNs for SYN-ACK packets are generated by the syncache. See 2744 * tcp_syncache.c for details. 2745 * 2746 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2747 * depends on this property. In addition, these ISNs should be 2748 * unguessable so as to prevent connection hijacking. To satisfy 2749 * the requirements of this situation, the algorithm outlined in 2750 * RFC 1948 is used, with only small modifications. 2751 * 2752 * Implementation details: 2753 * 2754 * Time is based off the system timer, and is corrected so that it 2755 * increases by one megabyte per second. This allows for proper 2756 * recycling on high speed LANs while still leaving over an hour 2757 * before rollover. 2758 * 2759 * As reading the *exact* system time is too expensive to be done 2760 * whenever setting up a TCP connection, we increment the time 2761 * offset in two ways. First, a small random positive increment 2762 * is added to isn_offset for each connection that is set up. 2763 * Second, the function tcp_isn_tick fires once per clock tick 2764 * and increments isn_offset as necessary so that sequence numbers 2765 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2766 * random positive increments serve only to ensure that the same 2767 * exact sequence number is never sent out twice (as could otherwise 2768 * happen when a port is recycled in less than the system tick 2769 * interval.) 2770 * 2771 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2772 * between seeding of isn_secret. This is normally set to zero, 2773 * as reseeding should not be necessary. 2774 * 2775 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2776 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 2777 * general, this means holding an exclusive (write) lock. 2778 */ 2779 2780 #define ISN_BYTES_PER_SECOND 1048576 2781 #define ISN_STATIC_INCREMENT 4096 2782 #define ISN_RANDOM_INCREMENT (4096 - 1) 2783 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 2784 2785 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 2786 VNET_DEFINE_STATIC(int, isn_last); 2787 VNET_DEFINE_STATIC(int, isn_last_reseed); 2788 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 2789 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 2790 2791 #define V_isn_secret VNET(isn_secret) 2792 #define V_isn_last VNET(isn_last) 2793 #define V_isn_last_reseed VNET(isn_last_reseed) 2794 #define V_isn_offset VNET(isn_offset) 2795 #define V_isn_offset_old VNET(isn_offset_old) 2796 2797 tcp_seq 2798 tcp_new_isn(struct in_conninfo *inc) 2799 { 2800 tcp_seq new_isn; 2801 u_int32_t projected_offset; 2802 2803 ISN_LOCK(); 2804 /* Seed if this is the first use, reseed if requested. */ 2805 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2806 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2807 < (u_int)ticks))) { 2808 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 2809 V_isn_last_reseed = ticks; 2810 } 2811 2812 /* Compute the hash and return the ISN. */ 2813 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 2814 sizeof(V_isn_secret)); 2815 V_isn_offset += ISN_STATIC_INCREMENT + 2816 (arc4random() & ISN_RANDOM_INCREMENT); 2817 if (ticks != V_isn_last) { 2818 projected_offset = V_isn_offset_old + 2819 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2820 if (SEQ_GT(projected_offset, V_isn_offset)) 2821 V_isn_offset = projected_offset; 2822 V_isn_offset_old = V_isn_offset; 2823 V_isn_last = ticks; 2824 } 2825 new_isn += V_isn_offset; 2826 ISN_UNLOCK(); 2827 return (new_isn); 2828 } 2829 2830 /* 2831 * When a specific ICMP unreachable message is received and the 2832 * connection state is SYN-SENT, drop the connection. This behavior 2833 * is controlled by the icmp_may_rst sysctl. 2834 */ 2835 struct inpcb * 2836 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2837 { 2838 struct tcpcb *tp; 2839 2840 NET_EPOCH_ASSERT(); 2841 INP_WLOCK_ASSERT(inp); 2842 2843 if ((inp->inp_flags & INP_TIMEWAIT) || 2844 (inp->inp_flags & INP_DROPPED)) 2845 return (inp); 2846 2847 tp = intotcpcb(inp); 2848 if (tp->t_state != TCPS_SYN_SENT) 2849 return (inp); 2850 2851 if (IS_FASTOPEN(tp->t_flags)) 2852 tcp_fastopen_disable_path(tp); 2853 2854 tp = tcp_drop(tp, errno); 2855 if (tp != NULL) 2856 return (inp); 2857 else 2858 return (NULL); 2859 } 2860 2861 /* 2862 * When `need fragmentation' ICMP is received, update our idea of the MSS 2863 * based on the new value. Also nudge TCP to send something, since we 2864 * know the packet we just sent was dropped. 2865 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2866 */ 2867 static struct inpcb * 2868 tcp_mtudisc_notify(struct inpcb *inp, int error) 2869 { 2870 2871 tcp_mtudisc(inp, -1); 2872 return (inp); 2873 } 2874 2875 static void 2876 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2877 { 2878 struct tcpcb *tp; 2879 struct socket *so; 2880 2881 INP_WLOCK_ASSERT(inp); 2882 if ((inp->inp_flags & INP_TIMEWAIT) || 2883 (inp->inp_flags & INP_DROPPED)) 2884 return; 2885 2886 tp = intotcpcb(inp); 2887 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2888 2889 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2890 2891 so = inp->inp_socket; 2892 SOCKBUF_LOCK(&so->so_snd); 2893 /* If the mss is larger than the socket buffer, decrease the mss. */ 2894 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2895 tp->t_maxseg = so->so_snd.sb_hiwat; 2896 SOCKBUF_UNLOCK(&so->so_snd); 2897 2898 TCPSTAT_INC(tcps_mturesent); 2899 tp->t_rtttime = 0; 2900 tp->snd_nxt = tp->snd_una; 2901 tcp_free_sackholes(tp); 2902 tp->snd_recover = tp->snd_max; 2903 if (tp->t_flags & TF_SACK_PERMIT) 2904 EXIT_FASTRECOVERY(tp->t_flags); 2905 tp->t_fb->tfb_tcp_output(tp); 2906 } 2907 2908 #ifdef INET 2909 /* 2910 * Look-up the routing entry to the peer of this inpcb. If no route 2911 * is found and it cannot be allocated, then return 0. This routine 2912 * is called by TCP routines that access the rmx structure and by 2913 * tcp_mss_update to get the peer/interface MTU. 2914 */ 2915 uint32_t 2916 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2917 { 2918 struct nhop4_extended nh4; 2919 struct ifnet *ifp; 2920 uint32_t maxmtu = 0; 2921 2922 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2923 2924 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2925 2926 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2927 NHR_REF, 0, &nh4) != 0) 2928 return (0); 2929 2930 ifp = nh4.nh_ifp; 2931 maxmtu = nh4.nh_mtu; 2932 2933 /* Report additional interface capabilities. */ 2934 if (cap != NULL) { 2935 if (ifp->if_capenable & IFCAP_TSO4 && 2936 ifp->if_hwassist & CSUM_TSO) { 2937 cap->ifcap |= CSUM_TSO; 2938 cap->tsomax = ifp->if_hw_tsomax; 2939 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2940 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2941 } 2942 } 2943 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2944 } 2945 return (maxmtu); 2946 } 2947 #endif /* INET */ 2948 2949 #ifdef INET6 2950 uint32_t 2951 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2952 { 2953 struct nhop6_extended nh6; 2954 struct in6_addr dst6; 2955 uint32_t scopeid; 2956 struct ifnet *ifp; 2957 uint32_t maxmtu = 0; 2958 2959 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2960 2961 if (inc->inc_flags & INC_IPV6MINMTU) 2962 return (IPV6_MMTU); 2963 2964 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2965 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2966 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2967 0, &nh6) != 0) 2968 return (0); 2969 2970 ifp = nh6.nh_ifp; 2971 maxmtu = nh6.nh_mtu; 2972 2973 /* Report additional interface capabilities. */ 2974 if (cap != NULL) { 2975 if (ifp->if_capenable & IFCAP_TSO6 && 2976 ifp->if_hwassist & CSUM_TSO) { 2977 cap->ifcap |= CSUM_TSO; 2978 cap->tsomax = ifp->if_hw_tsomax; 2979 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2980 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2981 } 2982 } 2983 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2984 } 2985 2986 return (maxmtu); 2987 } 2988 #endif /* INET6 */ 2989 2990 /* 2991 * Calculate effective SMSS per RFC5681 definition for a given TCP 2992 * connection at its current state, taking into account SACK and etc. 2993 */ 2994 u_int 2995 tcp_maxseg(const struct tcpcb *tp) 2996 { 2997 u_int optlen; 2998 2999 if (tp->t_flags & TF_NOOPT) 3000 return (tp->t_maxseg); 3001 3002 /* 3003 * Here we have a simplified code from tcp_addoptions(), 3004 * without a proper loop, and having most of paddings hardcoded. 3005 * We might make mistakes with padding here in some edge cases, 3006 * but this is harmless, since result of tcp_maxseg() is used 3007 * only in cwnd and ssthresh estimations. 3008 */ 3009 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 3010 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3011 if (tp->t_flags & TF_RCVD_TSTMP) 3012 optlen = TCPOLEN_TSTAMP_APPA; 3013 else 3014 optlen = 0; 3015 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3016 if (tp->t_flags & TF_SIGNATURE) 3017 optlen += PAD(TCPOLEN_SIGNATURE); 3018 #endif 3019 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3020 optlen += TCPOLEN_SACKHDR; 3021 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3022 optlen = PAD(optlen); 3023 } 3024 } else { 3025 if (tp->t_flags & TF_REQ_TSTMP) 3026 optlen = TCPOLEN_TSTAMP_APPA; 3027 else 3028 optlen = PAD(TCPOLEN_MAXSEG); 3029 if (tp->t_flags & TF_REQ_SCALE) 3030 optlen += PAD(TCPOLEN_WINDOW); 3031 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3032 if (tp->t_flags & TF_SIGNATURE) 3033 optlen += PAD(TCPOLEN_SIGNATURE); 3034 #endif 3035 if (tp->t_flags & TF_SACK_PERMIT) 3036 optlen += PAD(TCPOLEN_SACK_PERMITTED); 3037 } 3038 #undef PAD 3039 optlen = min(optlen, TCP_MAXOLEN); 3040 return (tp->t_maxseg - optlen); 3041 } 3042 3043 static int 3044 sysctl_drop(SYSCTL_HANDLER_ARGS) 3045 { 3046 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3047 struct sockaddr_storage addrs[2]; 3048 struct inpcb *inp; 3049 struct tcpcb *tp; 3050 struct tcptw *tw; 3051 struct sockaddr_in *fin, *lin; 3052 struct epoch_tracker et; 3053 #ifdef INET6 3054 struct sockaddr_in6 *fin6, *lin6; 3055 #endif 3056 int error; 3057 3058 inp = NULL; 3059 fin = lin = NULL; 3060 #ifdef INET6 3061 fin6 = lin6 = NULL; 3062 #endif 3063 error = 0; 3064 3065 if (req->oldptr != NULL || req->oldlen != 0) 3066 return (EINVAL); 3067 if (req->newptr == NULL) 3068 return (EPERM); 3069 if (req->newlen < sizeof(addrs)) 3070 return (ENOMEM); 3071 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3072 if (error) 3073 return (error); 3074 3075 switch (addrs[0].ss_family) { 3076 #ifdef INET6 3077 case AF_INET6: 3078 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3079 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3080 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3081 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3082 return (EINVAL); 3083 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3084 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3085 return (EINVAL); 3086 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3087 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3088 fin = (struct sockaddr_in *)&addrs[0]; 3089 lin = (struct sockaddr_in *)&addrs[1]; 3090 break; 3091 } 3092 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3093 if (error) 3094 return (error); 3095 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3096 if (error) 3097 return (error); 3098 break; 3099 #endif 3100 #ifdef INET 3101 case AF_INET: 3102 fin = (struct sockaddr_in *)&addrs[0]; 3103 lin = (struct sockaddr_in *)&addrs[1]; 3104 if (fin->sin_len != sizeof(struct sockaddr_in) || 3105 lin->sin_len != sizeof(struct sockaddr_in)) 3106 return (EINVAL); 3107 break; 3108 #endif 3109 default: 3110 return (EINVAL); 3111 } 3112 NET_EPOCH_ENTER(et); 3113 switch (addrs[0].ss_family) { 3114 #ifdef INET6 3115 case AF_INET6: 3116 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3117 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3118 INPLOOKUP_WLOCKPCB, NULL); 3119 break; 3120 #endif 3121 #ifdef INET 3122 case AF_INET: 3123 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3124 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3125 break; 3126 #endif 3127 } 3128 if (inp != NULL) { 3129 if (inp->inp_flags & INP_TIMEWAIT) { 3130 /* 3131 * XXXRW: There currently exists a state where an 3132 * inpcb is present, but its timewait state has been 3133 * discarded. For now, don't allow dropping of this 3134 * type of inpcb. 3135 */ 3136 tw = intotw(inp); 3137 if (tw != NULL) 3138 tcp_twclose(tw, 0); 3139 else 3140 INP_WUNLOCK(inp); 3141 } else if (!(inp->inp_flags & INP_DROPPED) && 3142 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 3143 tp = intotcpcb(inp); 3144 tp = tcp_drop(tp, ECONNABORTED); 3145 if (tp != NULL) 3146 INP_WUNLOCK(inp); 3147 } else 3148 INP_WUNLOCK(inp); 3149 } else 3150 error = ESRCH; 3151 NET_EPOCH_EXIT(et); 3152 return (error); 3153 } 3154 3155 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3156 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3157 0, sysctl_drop, "", "Drop TCP connection"); 3158 3159 #ifdef KERN_TLS 3160 static int 3161 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3162 { 3163 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3164 struct sockaddr_storage addrs[2]; 3165 struct inpcb *inp; 3166 struct sockaddr_in *fin, *lin; 3167 struct epoch_tracker et; 3168 #ifdef INET6 3169 struct sockaddr_in6 *fin6, *lin6; 3170 #endif 3171 int error; 3172 3173 inp = NULL; 3174 fin = lin = NULL; 3175 #ifdef INET6 3176 fin6 = lin6 = NULL; 3177 #endif 3178 error = 0; 3179 3180 if (req->oldptr != NULL || req->oldlen != 0) 3181 return (EINVAL); 3182 if (req->newptr == NULL) 3183 return (EPERM); 3184 if (req->newlen < sizeof(addrs)) 3185 return (ENOMEM); 3186 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3187 if (error) 3188 return (error); 3189 3190 switch (addrs[0].ss_family) { 3191 #ifdef INET6 3192 case AF_INET6: 3193 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3194 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3195 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3196 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3197 return (EINVAL); 3198 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3199 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3200 return (EINVAL); 3201 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3202 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3203 fin = (struct sockaddr_in *)&addrs[0]; 3204 lin = (struct sockaddr_in *)&addrs[1]; 3205 break; 3206 } 3207 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3208 if (error) 3209 return (error); 3210 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3211 if (error) 3212 return (error); 3213 break; 3214 #endif 3215 #ifdef INET 3216 case AF_INET: 3217 fin = (struct sockaddr_in *)&addrs[0]; 3218 lin = (struct sockaddr_in *)&addrs[1]; 3219 if (fin->sin_len != sizeof(struct sockaddr_in) || 3220 lin->sin_len != sizeof(struct sockaddr_in)) 3221 return (EINVAL); 3222 break; 3223 #endif 3224 default: 3225 return (EINVAL); 3226 } 3227 NET_EPOCH_ENTER(et); 3228 switch (addrs[0].ss_family) { 3229 #ifdef INET6 3230 case AF_INET6: 3231 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3232 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3233 INPLOOKUP_WLOCKPCB, NULL); 3234 break; 3235 #endif 3236 #ifdef INET 3237 case AF_INET: 3238 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3239 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3240 break; 3241 #endif 3242 } 3243 NET_EPOCH_EXIT(et); 3244 if (inp != NULL) { 3245 if ((inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) != 0 || 3246 inp->inp_socket == NULL) { 3247 error = ECONNRESET; 3248 INP_WUNLOCK(inp); 3249 } else { 3250 struct socket *so; 3251 3252 so = inp->inp_socket; 3253 soref(so); 3254 error = ktls_set_tx_mode(so, 3255 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3256 INP_WUNLOCK(inp); 3257 SOCK_LOCK(so); 3258 sorele(so); 3259 } 3260 } else 3261 error = ESRCH; 3262 return (error); 3263 } 3264 3265 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3266 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3267 0, sysctl_switch_tls, "", "Switch TCP connection to SW TLS"); 3268 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3269 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 3270 1, sysctl_switch_tls, "", "Switch TCP connection to ifnet TLS"); 3271 #endif 3272 3273 /* 3274 * Generate a standardized TCP log line for use throughout the 3275 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3276 * allow use in the interrupt context. 3277 * 3278 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3279 * NB: The function may return NULL if memory allocation failed. 3280 * 3281 * Due to header inclusion and ordering limitations the struct ip 3282 * and ip6_hdr pointers have to be passed as void pointers. 3283 */ 3284 char * 3285 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3286 const void *ip6hdr) 3287 { 3288 3289 /* Is logging enabled? */ 3290 if (V_tcp_log_in_vain == 0) 3291 return (NULL); 3292 3293 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3294 } 3295 3296 char * 3297 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3298 const void *ip6hdr) 3299 { 3300 3301 /* Is logging enabled? */ 3302 if (tcp_log_debug == 0) 3303 return (NULL); 3304 3305 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3306 } 3307 3308 static char * 3309 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 3310 const void *ip6hdr) 3311 { 3312 char *s, *sp; 3313 size_t size; 3314 struct ip *ip; 3315 #ifdef INET6 3316 const struct ip6_hdr *ip6; 3317 3318 ip6 = (const struct ip6_hdr *)ip6hdr; 3319 #endif /* INET6 */ 3320 ip = (struct ip *)ip4hdr; 3321 3322 /* 3323 * The log line looks like this: 3324 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3325 */ 3326 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3327 sizeof(PRINT_TH_FLAGS) + 1 + 3328 #ifdef INET6 3329 2 * INET6_ADDRSTRLEN; 3330 #else 3331 2 * INET_ADDRSTRLEN; 3332 #endif /* INET6 */ 3333 3334 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3335 if (s == NULL) 3336 return (NULL); 3337 3338 strcat(s, "TCP: ["); 3339 sp = s + strlen(s); 3340 3341 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3342 inet_ntoa_r(inc->inc_faddr, sp); 3343 sp = s + strlen(s); 3344 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3345 sp = s + strlen(s); 3346 inet_ntoa_r(inc->inc_laddr, sp); 3347 sp = s + strlen(s); 3348 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3349 #ifdef INET6 3350 } else if (inc) { 3351 ip6_sprintf(sp, &inc->inc6_faddr); 3352 sp = s + strlen(s); 3353 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3354 sp = s + strlen(s); 3355 ip6_sprintf(sp, &inc->inc6_laddr); 3356 sp = s + strlen(s); 3357 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3358 } else if (ip6 && th) { 3359 ip6_sprintf(sp, &ip6->ip6_src); 3360 sp = s + strlen(s); 3361 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3362 sp = s + strlen(s); 3363 ip6_sprintf(sp, &ip6->ip6_dst); 3364 sp = s + strlen(s); 3365 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3366 #endif /* INET6 */ 3367 #ifdef INET 3368 } else if (ip && th) { 3369 inet_ntoa_r(ip->ip_src, sp); 3370 sp = s + strlen(s); 3371 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3372 sp = s + strlen(s); 3373 inet_ntoa_r(ip->ip_dst, sp); 3374 sp = s + strlen(s); 3375 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3376 #endif /* INET */ 3377 } else { 3378 free(s, M_TCPLOG); 3379 return (NULL); 3380 } 3381 sp = s + strlen(s); 3382 if (th) 3383 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 3384 if (*(s + size - 1) != '\0') 3385 panic("%s: string too long", __func__); 3386 return (s); 3387 } 3388 3389 /* 3390 * A subroutine which makes it easy to track TCP state changes with DTrace. 3391 * This function shouldn't be called for t_state initializations that don't 3392 * correspond to actual TCP state transitions. 3393 */ 3394 void 3395 tcp_state_change(struct tcpcb *tp, int newstate) 3396 { 3397 #if defined(KDTRACE_HOOKS) 3398 int pstate = tp->t_state; 3399 #endif 3400 3401 TCPSTATES_DEC(tp->t_state); 3402 TCPSTATES_INC(newstate); 3403 tp->t_state = newstate; 3404 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3405 } 3406 3407 /* 3408 * Create an external-format (``xtcpcb'') structure using the information in 3409 * the kernel-format tcpcb structure pointed to by tp. This is done to 3410 * reduce the spew of irrelevant information over this interface, to isolate 3411 * user code from changes in the kernel structure, and potentially to provide 3412 * information-hiding if we decide that some of this information should be 3413 * hidden from users. 3414 */ 3415 void 3416 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3417 { 3418 struct tcpcb *tp = intotcpcb(inp); 3419 sbintime_t now; 3420 3421 bzero(xt, sizeof(*xt)); 3422 if (inp->inp_flags & INP_TIMEWAIT) { 3423 xt->t_state = TCPS_TIME_WAIT; 3424 } else { 3425 xt->t_state = tp->t_state; 3426 xt->t_logstate = tp->t_logstate; 3427 xt->t_flags = tp->t_flags; 3428 xt->t_sndzerowin = tp->t_sndzerowin; 3429 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3430 xt->t_rcvoopack = tp->t_rcvoopack; 3431 3432 now = getsbinuptime(); 3433 #define COPYTIMER(ttt) do { \ 3434 if (callout_active(&tp->t_timers->ttt)) \ 3435 xt->ttt = (tp->t_timers->ttt.c_time - now) / \ 3436 SBT_1MS; \ 3437 else \ 3438 xt->ttt = 0; \ 3439 } while (0) 3440 COPYTIMER(tt_delack); 3441 COPYTIMER(tt_rexmt); 3442 COPYTIMER(tt_persist); 3443 COPYTIMER(tt_keep); 3444 COPYTIMER(tt_2msl); 3445 #undef COPYTIMER 3446 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 3447 3448 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 3449 TCP_FUNCTION_NAME_LEN_MAX); 3450 #ifdef TCP_BLACKBOX 3451 (void)tcp_log_get_id(tp, xt->xt_logid); 3452 #endif 3453 } 3454 3455 xt->xt_len = sizeof(struct xtcpcb); 3456 in_pcbtoxinpcb(inp, &xt->xt_inp); 3457 if (inp->inp_socket == NULL) 3458 xt->xt_inp.xi_socket.xso_protocol = IPPROTO_TCP; 3459 } 3460