Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_subr.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD: releng/11.1/sys/netinet/tcp_subr.c 319654 2017-06-07 12:50:54Z tuexen $"); 34 35 #include "opt_compat.h" 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_tcpdebug.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/callout.h> 44 #include <sys/eventhandler.h> 45 #include <sys/hhook.h> 46 #include <sys/kernel.h> 47 #include <sys/khelp.h> 48 #include <sys/sysctl.h> 49 #include <sys/jail.h> 50 #include <sys/malloc.h> 51 #include <sys/refcount.h> 52 #include <sys/mbuf.h> 53 #ifdef INET6 54 #include <sys/domain.h> 55 #endif 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/sdt.h> 59 #include <sys/socket.h> 60 #include <sys/socketvar.h> 61 #include <sys/protosw.h> 62 #include <sys/random.h> 63 64 #include <vm/uma.h> 65 66 #include <net/route.h> 67 #include <net/if.h> 68 #include <net/if_var.h> 69 #include <net/vnet.h> 70 71 #include <netinet/in.h> 72 #include <netinet/in_fib.h> 73 #include <netinet/in_kdtrace.h> 74 #include <netinet/in_pcb.h> 75 #include <netinet/in_systm.h> 76 #include <netinet/in_var.h> 77 #include <netinet/ip.h> 78 #include <netinet/ip_icmp.h> 79 #include <netinet/ip_var.h> 80 #ifdef INET6 81 #include <netinet/icmp6.h> 82 #include <netinet/ip6.h> 83 #include <netinet6/in6_fib.h> 84 #include <netinet6/in6_pcb.h> 85 #include <netinet6/ip6_var.h> 86 #include <netinet6/scope6_var.h> 87 #include <netinet6/nd6.h> 88 #endif 89 90 #ifdef TCP_RFC7413 91 #include <netinet/tcp_fastopen.h> 92 #endif 93 #include <netinet/tcp.h> 94 #include <netinet/tcp_fsm.h> 95 #include <netinet/tcp_seq.h> 96 #include <netinet/tcp_timer.h> 97 #include <netinet/tcp_var.h> 98 #include <netinet/tcp_syncache.h> 99 #include <netinet/cc/cc.h> 100 #ifdef INET6 101 #include <netinet6/tcp6_var.h> 102 #endif 103 #include <netinet/tcpip.h> 104 #ifdef TCPPCAP 105 #include <netinet/tcp_pcap.h> 106 #endif 107 #ifdef TCPDEBUG 108 #include <netinet/tcp_debug.h> 109 #endif 110 #ifdef INET6 111 #include <netinet6/ip6protosw.h> 112 #endif 113 #ifdef TCP_OFFLOAD 114 #include <netinet/tcp_offload.h> 115 #endif 116 117 #include <netipsec/ipsec_support.h> 118 119 #include <machine/in_cksum.h> 120 #include <sys/md5.h> 121 122 #include <security/mac/mac_framework.h> 123 124 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 125 #ifdef INET6 126 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 127 #endif 128 129 struct rwlock tcp_function_lock; 130 131 static int 132 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 133 { 134 int error, new; 135 136 new = V_tcp_mssdflt; 137 error = sysctl_handle_int(oidp, &new, 0, req); 138 if (error == 0 && req->newptr) { 139 if (new < TCP_MINMSS) 140 error = EINVAL; 141 else 142 V_tcp_mssdflt = new; 143 } 144 return (error); 145 } 146 147 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 148 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 149 &sysctl_net_inet_tcp_mss_check, "I", 150 "Default TCP Maximum Segment Size"); 151 152 #ifdef INET6 153 static int 154 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 155 { 156 int error, new; 157 158 new = V_tcp_v6mssdflt; 159 error = sysctl_handle_int(oidp, &new, 0, req); 160 if (error == 0 && req->newptr) { 161 if (new < TCP_MINMSS) 162 error = EINVAL; 163 else 164 V_tcp_v6mssdflt = new; 165 } 166 return (error); 167 } 168 169 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 171 &sysctl_net_inet_tcp_mss_v6_check, "I", 172 "Default TCP Maximum Segment Size for IPv6"); 173 #endif /* INET6 */ 174 175 /* 176 * Minimum MSS we accept and use. This prevents DoS attacks where 177 * we are forced to a ridiculous low MSS like 20 and send hundreds 178 * of packets instead of one. The effect scales with the available 179 * bandwidth and quickly saturates the CPU and network interface 180 * with packet generation and sending. Set to zero to disable MINMSS 181 * checking. This setting prevents us from sending too small packets. 182 */ 183 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 185 &VNET_NAME(tcp_minmss), 0, 186 "Minimum TCP Maximum Segment Size"); 187 188 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 189 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 190 &VNET_NAME(tcp_do_rfc1323), 0, 191 "Enable rfc1323 (high performance TCP) extensions"); 192 193 static int tcp_log_debug = 0; 194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 195 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 196 197 static int tcp_tcbhashsize; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, 199 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 200 201 static int do_tcpdrain = 1; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 203 "Enable tcp_drain routine for extra help when low on mbufs"); 204 205 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 206 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 207 208 static VNET_DEFINE(int, icmp_may_rst) = 1; 209 #define V_icmp_may_rst VNET(icmp_may_rst) 210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(icmp_may_rst), 0, 212 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 213 214 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 215 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_isn_reseed_interval), 0, 218 "Seconds between reseeding of ISN secret"); 219 220 static int tcp_soreceive_stream; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 222 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 223 224 VNET_DEFINE(uma_zone_t, sack_hole_zone); 225 #define V_sack_hole_zone VNET(sack_hole_zone) 226 227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 228 229 static struct inpcb *tcp_notify(struct inpcb *, int); 230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 231 static void tcp_mtudisc(struct inpcb *, int); 232 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 233 void *ip4hdr, const void *ip6hdr); 234 235 236 static struct tcp_function_block tcp_def_funcblk = { 237 "default", 238 tcp_output, 239 tcp_do_segment, 240 tcp_default_ctloutput, 241 NULL, 242 NULL, 243 NULL, 244 NULL, 245 NULL, 246 NULL, 247 0, 248 0 249 }; 250 251 int t_functions_inited = 0; 252 struct tcp_funchead t_functions; 253 static struct tcp_function_block *tcp_func_set_ptr = &tcp_def_funcblk; 254 255 static void 256 init_tcp_functions(void) 257 { 258 if (t_functions_inited == 0) { 259 TAILQ_INIT(&t_functions); 260 rw_init_flags(&tcp_function_lock, "tcp_func_lock" , 0); 261 t_functions_inited = 1; 262 } 263 } 264 265 static struct tcp_function_block * 266 find_tcp_functions_locked(struct tcp_function_set *fs) 267 { 268 struct tcp_function *f; 269 struct tcp_function_block *blk=NULL; 270 271 TAILQ_FOREACH(f, &t_functions, tf_next) { 272 if (strcmp(f->tf_fb->tfb_tcp_block_name, fs->function_set_name) == 0) { 273 blk = f->tf_fb; 274 break; 275 } 276 } 277 return(blk); 278 } 279 280 static struct tcp_function_block * 281 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 282 { 283 struct tcp_function_block *rblk=NULL; 284 struct tcp_function *f; 285 286 TAILQ_FOREACH(f, &t_functions, tf_next) { 287 if (f->tf_fb == blk) { 288 rblk = blk; 289 if (s) { 290 *s = f; 291 } 292 break; 293 } 294 } 295 return (rblk); 296 } 297 298 struct tcp_function_block * 299 find_and_ref_tcp_functions(struct tcp_function_set *fs) 300 { 301 struct tcp_function_block *blk; 302 303 rw_rlock(&tcp_function_lock); 304 blk = find_tcp_functions_locked(fs); 305 if (blk) 306 refcount_acquire(&blk->tfb_refcnt); 307 rw_runlock(&tcp_function_lock); 308 return(blk); 309 } 310 311 struct tcp_function_block * 312 find_and_ref_tcp_fb(struct tcp_function_block *blk) 313 { 314 struct tcp_function_block *rblk; 315 316 rw_rlock(&tcp_function_lock); 317 rblk = find_tcp_fb_locked(blk, NULL); 318 if (rblk) 319 refcount_acquire(&rblk->tfb_refcnt); 320 rw_runlock(&tcp_function_lock); 321 return(rblk); 322 } 323 324 325 static int 326 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 327 { 328 int error=ENOENT; 329 struct tcp_function_set fs; 330 struct tcp_function_block *blk; 331 332 memset(&fs, 0, sizeof(fs)); 333 rw_rlock(&tcp_function_lock); 334 blk = find_tcp_fb_locked(tcp_func_set_ptr, NULL); 335 if (blk) { 336 /* Found him */ 337 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 338 fs.pcbcnt = blk->tfb_refcnt; 339 } 340 rw_runlock(&tcp_function_lock); 341 error = sysctl_handle_string(oidp, fs.function_set_name, 342 sizeof(fs.function_set_name), req); 343 344 /* Check for error or no change */ 345 if (error != 0 || req->newptr == NULL) 346 return(error); 347 348 rw_wlock(&tcp_function_lock); 349 blk = find_tcp_functions_locked(&fs); 350 if ((blk == NULL) || 351 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 352 error = ENOENT; 353 goto done; 354 } 355 tcp_func_set_ptr = blk; 356 done: 357 rw_wunlock(&tcp_function_lock); 358 return (error); 359 } 360 361 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 362 CTLTYPE_STRING | CTLFLAG_RW, 363 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 364 "Set/get the default TCP functions"); 365 366 static int 367 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 368 { 369 int error, cnt, linesz; 370 struct tcp_function *f; 371 char *buffer, *cp; 372 size_t bufsz, outsz; 373 374 cnt = 0; 375 rw_rlock(&tcp_function_lock); 376 TAILQ_FOREACH(f, &t_functions, tf_next) { 377 cnt++; 378 } 379 rw_runlock(&tcp_function_lock); 380 381 bufsz = (cnt+2) * (TCP_FUNCTION_NAME_LEN_MAX + 12) + 1; 382 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 383 384 error = 0; 385 cp = buffer; 386 387 linesz = snprintf(cp, bufsz, "\n%-32s%c %s\n", "Stack", 'D', "PCB count"); 388 cp += linesz; 389 bufsz -= linesz; 390 outsz = linesz; 391 392 rw_rlock(&tcp_function_lock); 393 TAILQ_FOREACH(f, &t_functions, tf_next) { 394 linesz = snprintf(cp, bufsz, "%-32s%c %u\n", 395 f->tf_fb->tfb_tcp_block_name, 396 (f->tf_fb == tcp_func_set_ptr) ? '*' : ' ', 397 f->tf_fb->tfb_refcnt); 398 if (linesz >= bufsz) { 399 error = EOVERFLOW; 400 break; 401 } 402 cp += linesz; 403 bufsz -= linesz; 404 outsz += linesz; 405 } 406 rw_runlock(&tcp_function_lock); 407 if (error == 0) 408 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 409 free(buffer, M_TEMP); 410 return (error); 411 } 412 413 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 414 CTLTYPE_STRING|CTLFLAG_RD, 415 NULL, 0, sysctl_net_inet_list_available, "A", 416 "list available TCP Function sets"); 417 418 /* 419 * Target size of TCP PCB hash tables. Must be a power of two. 420 * 421 * Note that this can be overridden by the kernel environment 422 * variable net.inet.tcp.tcbhashsize 423 */ 424 #ifndef TCBHASHSIZE 425 #define TCBHASHSIZE 0 426 #endif 427 428 /* 429 * XXX 430 * Callouts should be moved into struct tcp directly. They are currently 431 * separate because the tcpcb structure is exported to userland for sysctl 432 * parsing purposes, which do not know about callouts. 433 */ 434 struct tcpcb_mem { 435 struct tcpcb tcb; 436 struct tcp_timer tt; 437 struct cc_var ccv; 438 struct osd osd; 439 }; 440 441 static VNET_DEFINE(uma_zone_t, tcpcb_zone); 442 #define V_tcpcb_zone VNET(tcpcb_zone) 443 444 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 445 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 446 447 static struct mtx isn_mtx; 448 449 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 450 #define ISN_LOCK() mtx_lock(&isn_mtx) 451 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 452 453 /* 454 * TCP initialization. 455 */ 456 static void 457 tcp_zone_change(void *tag) 458 { 459 460 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 461 uma_zone_set_max(V_tcpcb_zone, maxsockets); 462 tcp_tw_zone_change(); 463 } 464 465 static int 466 tcp_inpcb_init(void *mem, int size, int flags) 467 { 468 struct inpcb *inp = mem; 469 470 INP_LOCK_INIT(inp, "inp", "tcpinp"); 471 return (0); 472 } 473 474 /* 475 * Take a value and get the next power of 2 that doesn't overflow. 476 * Used to size the tcp_inpcb hash buckets. 477 */ 478 static int 479 maketcp_hashsize(int size) 480 { 481 int hashsize; 482 483 /* 484 * auto tune. 485 * get the next power of 2 higher than maxsockets. 486 */ 487 hashsize = 1 << fls(size); 488 /* catch overflow, and just go one power of 2 smaller */ 489 if (hashsize < size) { 490 hashsize = 1 << (fls(size) - 1); 491 } 492 return (hashsize); 493 } 494 495 int 496 register_tcp_functions(struct tcp_function_block *blk, int wait) 497 { 498 struct tcp_function_block *lblk; 499 struct tcp_function *n; 500 struct tcp_function_set fs; 501 502 if (t_functions_inited == 0) { 503 init_tcp_functions(); 504 } 505 if ((blk->tfb_tcp_output == NULL) || 506 (blk->tfb_tcp_do_segment == NULL) || 507 (blk->tfb_tcp_ctloutput == NULL) || 508 (strlen(blk->tfb_tcp_block_name) == 0)) { 509 /* 510 * These functions are required and you 511 * need a name. 512 */ 513 return (EINVAL); 514 } 515 if (blk->tfb_tcp_timer_stop_all || 516 blk->tfb_tcp_timer_activate || 517 blk->tfb_tcp_timer_active || 518 blk->tfb_tcp_timer_stop) { 519 /* 520 * If you define one timer function you 521 * must have them all. 522 */ 523 if ((blk->tfb_tcp_timer_stop_all == NULL) || 524 (blk->tfb_tcp_timer_activate == NULL) || 525 (blk->tfb_tcp_timer_active == NULL) || 526 (blk->tfb_tcp_timer_stop == NULL)) { 527 return (EINVAL); 528 } 529 } 530 n = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 531 if (n == NULL) { 532 return (ENOMEM); 533 } 534 n->tf_fb = blk; 535 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 536 rw_wlock(&tcp_function_lock); 537 lblk = find_tcp_functions_locked(&fs); 538 if (lblk) { 539 /* Duplicate name space not allowed */ 540 rw_wunlock(&tcp_function_lock); 541 free(n, M_TCPFUNCTIONS); 542 return (EALREADY); 543 } 544 refcount_init(&blk->tfb_refcnt, 0); 545 blk->tfb_flags = 0; 546 TAILQ_INSERT_TAIL(&t_functions, n, tf_next); 547 rw_wunlock(&tcp_function_lock); 548 return(0); 549 } 550 551 int 552 deregister_tcp_functions(struct tcp_function_block *blk) 553 { 554 struct tcp_function_block *lblk; 555 struct tcp_function *f; 556 int error=ENOENT; 557 558 if (strcmp(blk->tfb_tcp_block_name, "default") == 0) { 559 /* You can't un-register the default */ 560 return (EPERM); 561 } 562 rw_wlock(&tcp_function_lock); 563 if (blk == tcp_func_set_ptr) { 564 /* You can't free the current default */ 565 rw_wunlock(&tcp_function_lock); 566 return (EBUSY); 567 } 568 if (blk->tfb_refcnt) { 569 /* Still tcb attached, mark it. */ 570 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 571 rw_wunlock(&tcp_function_lock); 572 return (EBUSY); 573 } 574 lblk = find_tcp_fb_locked(blk, &f); 575 if (lblk) { 576 /* Found */ 577 TAILQ_REMOVE(&t_functions, f, tf_next); 578 f->tf_fb = NULL; 579 free(f, M_TCPFUNCTIONS); 580 error = 0; 581 } 582 rw_wunlock(&tcp_function_lock); 583 return (error); 584 } 585 586 void 587 tcp_init(void) 588 { 589 const char *tcbhash_tuneable; 590 int hashsize; 591 592 tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 593 594 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 595 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 596 printf("%s: WARNING: unable to register helper hook\n", __func__); 597 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 598 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 599 printf("%s: WARNING: unable to register helper hook\n", __func__); 600 hashsize = TCBHASHSIZE; 601 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 602 if (hashsize == 0) { 603 /* 604 * Auto tune the hash size based on maxsockets. 605 * A perfect hash would have a 1:1 mapping 606 * (hashsize = maxsockets) however it's been 607 * suggested that O(2) average is better. 608 */ 609 hashsize = maketcp_hashsize(maxsockets / 4); 610 /* 611 * Our historical default is 512, 612 * do not autotune lower than this. 613 */ 614 if (hashsize < 512) 615 hashsize = 512; 616 if (bootverbose && IS_DEFAULT_VNET(curvnet)) 617 printf("%s: %s auto tuned to %d\n", __func__, 618 tcbhash_tuneable, hashsize); 619 } 620 /* 621 * We require a hashsize to be a power of two. 622 * Previously if it was not a power of two we would just reset it 623 * back to 512, which could be a nasty surprise if you did not notice 624 * the error message. 625 * Instead what we do is clip it to the closest power of two lower 626 * than the specified hash value. 627 */ 628 if (!powerof2(hashsize)) { 629 int oldhashsize = hashsize; 630 631 hashsize = maketcp_hashsize(hashsize); 632 /* prevent absurdly low value */ 633 if (hashsize < 16) 634 hashsize = 16; 635 printf("%s: WARNING: TCB hash size not a power of 2, " 636 "clipped from %d to %d.\n", __func__, oldhashsize, 637 hashsize); 638 } 639 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 640 "tcp_inpcb", tcp_inpcb_init, NULL, 0, IPI_HASHFIELDS_4TUPLE); 641 642 /* 643 * These have to be type stable for the benefit of the timers. 644 */ 645 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 646 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 647 uma_zone_set_max(V_tcpcb_zone, maxsockets); 648 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 649 650 tcp_tw_init(); 651 syncache_init(); 652 tcp_hc_init(); 653 654 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 655 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 656 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 657 658 /* Skip initialization of globals for non-default instances. */ 659 if (!IS_DEFAULT_VNET(curvnet)) 660 return; 661 662 tcp_reass_global_init(); 663 664 /* XXX virtualize those bellow? */ 665 tcp_delacktime = TCPTV_DELACK; 666 tcp_keepinit = TCPTV_KEEP_INIT; 667 tcp_keepidle = TCPTV_KEEP_IDLE; 668 tcp_keepintvl = TCPTV_KEEPINTVL; 669 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 670 tcp_msl = TCPTV_MSL; 671 tcp_rexmit_min = TCPTV_MIN; 672 if (tcp_rexmit_min < 1) 673 tcp_rexmit_min = 1; 674 tcp_persmin = TCPTV_PERSMIN; 675 tcp_persmax = TCPTV_PERSMAX; 676 tcp_rexmit_slop = TCPTV_CPU_VAR; 677 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 678 tcp_tcbhashsize = hashsize; 679 /* Setup the tcp function block list */ 680 init_tcp_functions(); 681 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 682 683 if (tcp_soreceive_stream) { 684 #ifdef INET 685 tcp_usrreqs.pru_soreceive = soreceive_stream; 686 #endif 687 #ifdef INET6 688 tcp6_usrreqs.pru_soreceive = soreceive_stream; 689 #endif /* INET6 */ 690 } 691 692 #ifdef INET6 693 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 694 #else /* INET6 */ 695 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 696 #endif /* INET6 */ 697 if (max_protohdr < TCP_MINPROTOHDR) 698 max_protohdr = TCP_MINPROTOHDR; 699 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 700 panic("tcp_init"); 701 #undef TCP_MINPROTOHDR 702 703 ISN_LOCK_INIT(); 704 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 705 SHUTDOWN_PRI_DEFAULT); 706 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 707 EVENTHANDLER_PRI_ANY); 708 #ifdef TCPPCAP 709 tcp_pcap_init(); 710 #endif 711 712 #ifdef TCP_RFC7413 713 tcp_fastopen_init(); 714 #endif 715 } 716 717 #ifdef VIMAGE 718 static void 719 tcp_destroy(void *unused __unused) 720 { 721 int error, n; 722 723 /* 724 * All our processes are gone, all our sockets should be cleaned 725 * up, which means, we should be past the tcp_discardcb() calls. 726 * Sleep to let all tcpcb timers really disappear and cleanup. 727 */ 728 for (;;) { 729 INP_LIST_RLOCK(&V_tcbinfo); 730 n = V_tcbinfo.ipi_count; 731 INP_LIST_RUNLOCK(&V_tcbinfo); 732 if (n == 0) 733 break; 734 pause("tcpdes", hz / 10); 735 } 736 tcp_hc_destroy(); 737 syncache_destroy(); 738 tcp_tw_destroy(); 739 in_pcbinfo_destroy(&V_tcbinfo); 740 /* tcp_discardcb() clears the sack_holes up. */ 741 uma_zdestroy(V_sack_hole_zone); 742 uma_zdestroy(V_tcpcb_zone); 743 744 #ifdef TCP_RFC7413 745 /* 746 * Cannot free the zone until all tcpcbs are released as we attach 747 * the allocations to them. 748 */ 749 tcp_fastopen_destroy(); 750 #endif 751 752 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 753 if (error != 0) { 754 printf("%s: WARNING: unable to deregister helper hook " 755 "type=%d, id=%d: error %d returned\n", __func__, 756 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 757 } 758 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 759 if (error != 0) { 760 printf("%s: WARNING: unable to deregister helper hook " 761 "type=%d, id=%d: error %d returned\n", __func__, 762 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 763 } 764 } 765 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 766 #endif 767 768 void 769 tcp_fini(void *xtp) 770 { 771 772 } 773 774 /* 775 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 776 * tcp_template used to store this data in mbufs, but we now recopy it out 777 * of the tcpcb each time to conserve mbufs. 778 */ 779 void 780 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 781 { 782 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 783 784 INP_WLOCK_ASSERT(inp); 785 786 #ifdef INET6 787 if ((inp->inp_vflag & INP_IPV6) != 0) { 788 struct ip6_hdr *ip6; 789 790 ip6 = (struct ip6_hdr *)ip_ptr; 791 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 792 (inp->inp_flow & IPV6_FLOWINFO_MASK); 793 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 794 (IPV6_VERSION & IPV6_VERSION_MASK); 795 ip6->ip6_nxt = IPPROTO_TCP; 796 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 797 ip6->ip6_src = inp->in6p_laddr; 798 ip6->ip6_dst = inp->in6p_faddr; 799 } 800 #endif /* INET6 */ 801 #if defined(INET6) && defined(INET) 802 else 803 #endif 804 #ifdef INET 805 { 806 struct ip *ip; 807 808 ip = (struct ip *)ip_ptr; 809 ip->ip_v = IPVERSION; 810 ip->ip_hl = 5; 811 ip->ip_tos = inp->inp_ip_tos; 812 ip->ip_len = 0; 813 ip->ip_id = 0; 814 ip->ip_off = 0; 815 ip->ip_ttl = inp->inp_ip_ttl; 816 ip->ip_sum = 0; 817 ip->ip_p = IPPROTO_TCP; 818 ip->ip_src = inp->inp_laddr; 819 ip->ip_dst = inp->inp_faddr; 820 } 821 #endif /* INET */ 822 th->th_sport = inp->inp_lport; 823 th->th_dport = inp->inp_fport; 824 th->th_seq = 0; 825 th->th_ack = 0; 826 th->th_x2 = 0; 827 th->th_off = 5; 828 th->th_flags = 0; 829 th->th_win = 0; 830 th->th_urp = 0; 831 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 832 } 833 834 /* 835 * Create template to be used to send tcp packets on a connection. 836 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 837 * use for this function is in keepalives, which use tcp_respond. 838 */ 839 struct tcptemp * 840 tcpip_maketemplate(struct inpcb *inp) 841 { 842 struct tcptemp *t; 843 844 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 845 if (t == NULL) 846 return (NULL); 847 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 848 return (t); 849 } 850 851 /* 852 * Send a single message to the TCP at address specified by 853 * the given TCP/IP header. If m == NULL, then we make a copy 854 * of the tcpiphdr at th and send directly to the addressed host. 855 * This is used to force keep alive messages out using the TCP 856 * template for a connection. If flags are given then we send 857 * a message back to the TCP which originated the segment th, 858 * and discard the mbuf containing it and any other attached mbufs. 859 * 860 * In any case the ack and sequence number of the transmitted 861 * segment are as specified by the parameters. 862 * 863 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 864 */ 865 void 866 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 867 tcp_seq ack, tcp_seq seq, int flags) 868 { 869 struct tcpopt to; 870 struct inpcb *inp; 871 struct ip *ip; 872 struct mbuf *optm; 873 struct tcphdr *nth; 874 u_char *optp; 875 #ifdef INET6 876 struct ip6_hdr *ip6; 877 int isipv6; 878 #endif /* INET6 */ 879 int optlen, tlen, win; 880 bool incl_opts; 881 882 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 883 884 #ifdef INET6 885 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 886 ip6 = ipgen; 887 #endif /* INET6 */ 888 ip = ipgen; 889 890 if (tp != NULL) { 891 inp = tp->t_inpcb; 892 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 893 INP_WLOCK_ASSERT(inp); 894 } else 895 inp = NULL; 896 897 incl_opts = false; 898 win = 0; 899 if (tp != NULL) { 900 if (!(flags & TH_RST)) { 901 win = sbspace(&inp->inp_socket->so_rcv); 902 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 903 win = (long)TCP_MAXWIN << tp->rcv_scale; 904 } 905 if ((tp->t_flags & TF_NOOPT) == 0) 906 incl_opts = true; 907 } 908 if (m == NULL) { 909 m = m_gethdr(M_NOWAIT, MT_DATA); 910 if (m == NULL) 911 return; 912 m->m_data += max_linkhdr; 913 #ifdef INET6 914 if (isipv6) { 915 bcopy((caddr_t)ip6, mtod(m, caddr_t), 916 sizeof(struct ip6_hdr)); 917 ip6 = mtod(m, struct ip6_hdr *); 918 nth = (struct tcphdr *)(ip6 + 1); 919 } else 920 #endif /* INET6 */ 921 { 922 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 923 ip = mtod(m, struct ip *); 924 nth = (struct tcphdr *)(ip + 1); 925 } 926 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 927 flags = TH_ACK; 928 } else if (!M_WRITABLE(m)) { 929 struct mbuf *n; 930 931 /* Can't reuse 'm', allocate a new mbuf. */ 932 n = m_gethdr(M_NOWAIT, MT_DATA); 933 if (n == NULL) { 934 m_freem(m); 935 return; 936 } 937 938 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 939 m_freem(m); 940 m_freem(n); 941 return; 942 } 943 944 n->m_data += max_linkhdr; 945 /* m_len is set later */ 946 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 947 #ifdef INET6 948 if (isipv6) { 949 bcopy((caddr_t)ip6, mtod(n, caddr_t), 950 sizeof(struct ip6_hdr)); 951 ip6 = mtod(n, struct ip6_hdr *); 952 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 953 nth = (struct tcphdr *)(ip6 + 1); 954 } else 955 #endif /* INET6 */ 956 { 957 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 958 ip = mtod(n, struct ip *); 959 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 960 nth = (struct tcphdr *)(ip + 1); 961 } 962 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 963 xchg(nth->th_dport, nth->th_sport, uint16_t); 964 th = nth; 965 m_freem(m); 966 m = n; 967 } else { 968 /* 969 * reuse the mbuf. 970 * XXX MRT We inherit the FIB, which is lucky. 971 */ 972 m_freem(m->m_next); 973 m->m_next = NULL; 974 m->m_data = (caddr_t)ipgen; 975 /* m_len is set later */ 976 #ifdef INET6 977 if (isipv6) { 978 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 979 nth = (struct tcphdr *)(ip6 + 1); 980 } else 981 #endif /* INET6 */ 982 { 983 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 984 nth = (struct tcphdr *)(ip + 1); 985 } 986 if (th != nth) { 987 /* 988 * this is usually a case when an extension header 989 * exists between the IPv6 header and the 990 * TCP header. 991 */ 992 nth->th_sport = th->th_sport; 993 nth->th_dport = th->th_dport; 994 } 995 xchg(nth->th_dport, nth->th_sport, uint16_t); 996 #undef xchg 997 } 998 tlen = 0; 999 #ifdef INET6 1000 if (isipv6) 1001 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1002 #endif 1003 #if defined(INET) && defined(INET6) 1004 else 1005 #endif 1006 #ifdef INET 1007 tlen = sizeof (struct tcpiphdr); 1008 #endif 1009 #ifdef INVARIANTS 1010 m->m_len = 0; 1011 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1012 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1013 m, tlen, (long)M_TRAILINGSPACE(m))); 1014 #endif 1015 m->m_len = tlen; 1016 to.to_flags = 0; 1017 if (incl_opts) { 1018 /* Make sure we have room. */ 1019 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1020 m->m_next = m_get(M_NOWAIT, MT_DATA); 1021 if (m->m_next) { 1022 optp = mtod(m->m_next, u_char *); 1023 optm = m->m_next; 1024 } else 1025 incl_opts = false; 1026 } else { 1027 optp = (u_char *) (nth + 1); 1028 optm = m; 1029 } 1030 } 1031 if (incl_opts) { 1032 /* Timestamps. */ 1033 if (tp->t_flags & TF_RCVD_TSTMP) { 1034 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1035 to.to_tsecr = tp->ts_recent; 1036 to.to_flags |= TOF_TS; 1037 } 1038 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1039 /* TCP-MD5 (RFC2385). */ 1040 if (tp->t_flags & TF_SIGNATURE) 1041 to.to_flags |= TOF_SIGNATURE; 1042 #endif 1043 /* Add the options. */ 1044 tlen += optlen = tcp_addoptions(&to, optp); 1045 1046 /* Update m_len in the correct mbuf. */ 1047 optm->m_len += optlen; 1048 } else 1049 optlen = 0; 1050 #ifdef INET6 1051 if (isipv6) { 1052 ip6->ip6_flow = 0; 1053 ip6->ip6_vfc = IPV6_VERSION; 1054 ip6->ip6_nxt = IPPROTO_TCP; 1055 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1056 } 1057 #endif 1058 #if defined(INET) && defined(INET6) 1059 else 1060 #endif 1061 #ifdef INET 1062 { 1063 ip->ip_len = htons(tlen); 1064 ip->ip_ttl = V_ip_defttl; 1065 if (V_path_mtu_discovery) 1066 ip->ip_off |= htons(IP_DF); 1067 } 1068 #endif 1069 m->m_pkthdr.len = tlen; 1070 m->m_pkthdr.rcvif = NULL; 1071 #ifdef MAC 1072 if (inp != NULL) { 1073 /* 1074 * Packet is associated with a socket, so allow the 1075 * label of the response to reflect the socket label. 1076 */ 1077 INP_WLOCK_ASSERT(inp); 1078 mac_inpcb_create_mbuf(inp, m); 1079 } else { 1080 /* 1081 * Packet is not associated with a socket, so possibly 1082 * update the label in place. 1083 */ 1084 mac_netinet_tcp_reply(m); 1085 } 1086 #endif 1087 nth->th_seq = htonl(seq); 1088 nth->th_ack = htonl(ack); 1089 nth->th_x2 = 0; 1090 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 1091 nth->th_flags = flags; 1092 if (tp != NULL) 1093 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 1094 else 1095 nth->th_win = htons((u_short)win); 1096 nth->th_urp = 0; 1097 1098 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1099 if (to.to_flags & TOF_SIGNATURE) { 1100 if (!TCPMD5_ENABLED() || 1101 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 1102 m_freem(m); 1103 return; 1104 } 1105 } 1106 #endif 1107 1108 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 1109 #ifdef INET6 1110 if (isipv6) { 1111 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 1112 nth->th_sum = in6_cksum_pseudo(ip6, 1113 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 1114 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 1115 NULL, NULL); 1116 } 1117 #endif /* INET6 */ 1118 #if defined(INET6) && defined(INET) 1119 else 1120 #endif 1121 #ifdef INET 1122 { 1123 m->m_pkthdr.csum_flags = CSUM_TCP; 1124 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 1125 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 1126 } 1127 #endif /* INET */ 1128 #ifdef TCPDEBUG 1129 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 1130 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 1131 #endif 1132 TCP_PROBE3(debug__output, tp, th, m); 1133 if (flags & TH_RST) 1134 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 1135 1136 TCP_PROBE5(send, NULL, tp, m, tp, nth); 1137 #ifdef INET6 1138 if (isipv6) 1139 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp); 1140 #endif /* INET6 */ 1141 #if defined(INET) && defined(INET6) 1142 else 1143 #endif 1144 #ifdef INET 1145 (void) ip_output(m, NULL, NULL, 0, NULL, inp); 1146 #endif 1147 } 1148 1149 /* 1150 * Create a new TCP control block, making an 1151 * empty reassembly queue and hooking it to the argument 1152 * protocol control block. The `inp' parameter must have 1153 * come from the zone allocator set up in tcp_init(). 1154 */ 1155 struct tcpcb * 1156 tcp_newtcpcb(struct inpcb *inp) 1157 { 1158 struct tcpcb_mem *tm; 1159 struct tcpcb *tp; 1160 #ifdef INET6 1161 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1162 #endif /* INET6 */ 1163 1164 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 1165 if (tm == NULL) 1166 return (NULL); 1167 tp = &tm->tcb; 1168 1169 /* Initialise cc_var struct for this tcpcb. */ 1170 tp->ccv = &tm->ccv; 1171 tp->ccv->type = IPPROTO_TCP; 1172 tp->ccv->ccvc.tcp = tp; 1173 rw_rlock(&tcp_function_lock); 1174 tp->t_fb = tcp_func_set_ptr; 1175 refcount_acquire(&tp->t_fb->tfb_refcnt); 1176 rw_runlock(&tcp_function_lock); 1177 if (tp->t_fb->tfb_tcp_fb_init) { 1178 (*tp->t_fb->tfb_tcp_fb_init)(tp); 1179 } 1180 /* 1181 * Use the current system default CC algorithm. 1182 */ 1183 CC_LIST_RLOCK(); 1184 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 1185 CC_ALGO(tp) = CC_DEFAULT(); 1186 CC_LIST_RUNLOCK(); 1187 1188 if (CC_ALGO(tp)->cb_init != NULL) 1189 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 1190 if (tp->t_fb->tfb_tcp_fb_fini) 1191 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1192 refcount_release(&tp->t_fb->tfb_refcnt); 1193 uma_zfree(V_tcpcb_zone, tm); 1194 return (NULL); 1195 } 1196 1197 tp->osd = &tm->osd; 1198 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 1199 if (tp->t_fb->tfb_tcp_fb_fini) 1200 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1201 refcount_release(&tp->t_fb->tfb_refcnt); 1202 uma_zfree(V_tcpcb_zone, tm); 1203 return (NULL); 1204 } 1205 1206 #ifdef VIMAGE 1207 tp->t_vnet = inp->inp_vnet; 1208 #endif 1209 tp->t_timers = &tm->tt; 1210 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 1211 tp->t_maxseg = 1212 #ifdef INET6 1213 isipv6 ? V_tcp_v6mssdflt : 1214 #endif /* INET6 */ 1215 V_tcp_mssdflt; 1216 1217 /* Set up our timeouts. */ 1218 callout_init(&tp->t_timers->tt_rexmt, 1); 1219 callout_init(&tp->t_timers->tt_persist, 1); 1220 callout_init(&tp->t_timers->tt_keep, 1); 1221 callout_init(&tp->t_timers->tt_2msl, 1); 1222 callout_init(&tp->t_timers->tt_delack, 1); 1223 1224 if (V_tcp_do_rfc1323) 1225 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 1226 if (V_tcp_do_sack) 1227 tp->t_flags |= TF_SACK_PERMIT; 1228 TAILQ_INIT(&tp->snd_holes); 1229 /* 1230 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 1231 * is called. 1232 */ 1233 in_pcbref(inp); /* Reference for tcpcb */ 1234 tp->t_inpcb = inp; 1235 1236 /* 1237 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 1238 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 1239 * reasonable initial retransmit time. 1240 */ 1241 tp->t_srtt = TCPTV_SRTTBASE; 1242 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 1243 tp->t_rttmin = tcp_rexmit_min; 1244 tp->t_rxtcur = TCPTV_RTOBASE; 1245 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1246 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 1247 tp->t_rcvtime = ticks; 1248 /* 1249 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 1250 * because the socket may be bound to an IPv6 wildcard address, 1251 * which may match an IPv4-mapped IPv6 address. 1252 */ 1253 inp->inp_ip_ttl = V_ip_defttl; 1254 inp->inp_ppcb = tp; 1255 #ifdef TCPPCAP 1256 /* 1257 * Init the TCP PCAP queues. 1258 */ 1259 tcp_pcap_tcpcb_init(tp); 1260 #endif 1261 return (tp); /* XXX */ 1262 } 1263 1264 /* 1265 * Switch the congestion control algorithm back to NewReno for any active 1266 * control blocks using an algorithm which is about to go away. 1267 * This ensures the CC framework can allow the unload to proceed without leaving 1268 * any dangling pointers which would trigger a panic. 1269 * Returning non-zero would inform the CC framework that something went wrong 1270 * and it would be unsafe to allow the unload to proceed. However, there is no 1271 * way for this to occur with this implementation so we always return zero. 1272 */ 1273 int 1274 tcp_ccalgounload(struct cc_algo *unload_algo) 1275 { 1276 struct cc_algo *tmpalgo; 1277 struct inpcb *inp; 1278 struct tcpcb *tp; 1279 VNET_ITERATOR_DECL(vnet_iter); 1280 1281 /* 1282 * Check all active control blocks across all network stacks and change 1283 * any that are using "unload_algo" back to NewReno. If "unload_algo" 1284 * requires cleanup code to be run, call it. 1285 */ 1286 VNET_LIST_RLOCK(); 1287 VNET_FOREACH(vnet_iter) { 1288 CURVNET_SET(vnet_iter); 1289 INP_INFO_WLOCK(&V_tcbinfo); 1290 /* 1291 * New connections already part way through being initialised 1292 * with the CC algo we're removing will not race with this code 1293 * because the INP_INFO_WLOCK is held during initialisation. We 1294 * therefore don't enter the loop below until the connection 1295 * list has stabilised. 1296 */ 1297 LIST_FOREACH(inp, &V_tcb, inp_list) { 1298 INP_WLOCK(inp); 1299 /* Important to skip tcptw structs. */ 1300 if (!(inp->inp_flags & INP_TIMEWAIT) && 1301 (tp = intotcpcb(inp)) != NULL) { 1302 /* 1303 * By holding INP_WLOCK here, we are assured 1304 * that the connection is not currently 1305 * executing inside the CC module's functions 1306 * i.e. it is safe to make the switch back to 1307 * NewReno. 1308 */ 1309 if (CC_ALGO(tp) == unload_algo) { 1310 tmpalgo = CC_ALGO(tp); 1311 /* NewReno does not require any init. */ 1312 CC_ALGO(tp) = &newreno_cc_algo; 1313 if (tmpalgo->cb_destroy != NULL) 1314 tmpalgo->cb_destroy(tp->ccv); 1315 } 1316 } 1317 INP_WUNLOCK(inp); 1318 } 1319 INP_INFO_WUNLOCK(&V_tcbinfo); 1320 CURVNET_RESTORE(); 1321 } 1322 VNET_LIST_RUNLOCK(); 1323 1324 return (0); 1325 } 1326 1327 /* 1328 * Drop a TCP connection, reporting 1329 * the specified error. If connection is synchronized, 1330 * then send a RST to peer. 1331 */ 1332 struct tcpcb * 1333 tcp_drop(struct tcpcb *tp, int errno) 1334 { 1335 struct socket *so = tp->t_inpcb->inp_socket; 1336 1337 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1338 INP_WLOCK_ASSERT(tp->t_inpcb); 1339 1340 if (TCPS_HAVERCVDSYN(tp->t_state)) { 1341 tcp_state_change(tp, TCPS_CLOSED); 1342 (void) tp->t_fb->tfb_tcp_output(tp); 1343 TCPSTAT_INC(tcps_drops); 1344 } else 1345 TCPSTAT_INC(tcps_conndrops); 1346 if (errno == ETIMEDOUT && tp->t_softerror) 1347 errno = tp->t_softerror; 1348 so->so_error = errno; 1349 return (tcp_close(tp)); 1350 } 1351 1352 void 1353 tcp_discardcb(struct tcpcb *tp) 1354 { 1355 struct inpcb *inp = tp->t_inpcb; 1356 struct socket *so = inp->inp_socket; 1357 #ifdef INET6 1358 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 1359 #endif /* INET6 */ 1360 int released; 1361 1362 INP_WLOCK_ASSERT(inp); 1363 1364 /* 1365 * Make sure that all of our timers are stopped before we delete the 1366 * PCB. 1367 * 1368 * If stopping a timer fails, we schedule a discard function in same 1369 * callout, and the last discard function called will take care of 1370 * deleting the tcpcb. 1371 */ 1372 tp->t_timers->tt_draincnt = 0; 1373 tcp_timer_stop(tp, TT_REXMT); 1374 tcp_timer_stop(tp, TT_PERSIST); 1375 tcp_timer_stop(tp, TT_KEEP); 1376 tcp_timer_stop(tp, TT_2MSL); 1377 tcp_timer_stop(tp, TT_DELACK); 1378 if (tp->t_fb->tfb_tcp_timer_stop_all) { 1379 /* 1380 * Call the stop-all function of the methods, 1381 * this function should call the tcp_timer_stop() 1382 * method with each of the function specific timeouts. 1383 * That stop will be called via the tfb_tcp_timer_stop() 1384 * which should use the async drain function of the 1385 * callout system (see tcp_var.h). 1386 */ 1387 tp->t_fb->tfb_tcp_timer_stop_all(tp); 1388 } 1389 1390 /* 1391 * If we got enough samples through the srtt filter, 1392 * save the rtt and rttvar in the routing entry. 1393 * 'Enough' is arbitrarily defined as 4 rtt samples. 1394 * 4 samples is enough for the srtt filter to converge 1395 * to within enough % of the correct value; fewer samples 1396 * and we could save a bogus rtt. The danger is not high 1397 * as tcp quickly recovers from everything. 1398 * XXX: Works very well but needs some more statistics! 1399 */ 1400 if (tp->t_rttupdated >= 4) { 1401 struct hc_metrics_lite metrics; 1402 u_long ssthresh; 1403 1404 bzero(&metrics, sizeof(metrics)); 1405 /* 1406 * Update the ssthresh always when the conditions below 1407 * are satisfied. This gives us better new start value 1408 * for the congestion avoidance for new connections. 1409 * ssthresh is only set if packet loss occurred on a session. 1410 * 1411 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 1412 * being torn down. Ideally this code would not use 'so'. 1413 */ 1414 ssthresh = tp->snd_ssthresh; 1415 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 1416 /* 1417 * convert the limit from user data bytes to 1418 * packets then to packet data bytes. 1419 */ 1420 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 1421 if (ssthresh < 2) 1422 ssthresh = 2; 1423 ssthresh *= (u_long)(tp->t_maxseg + 1424 #ifdef INET6 1425 (isipv6 ? sizeof (struct ip6_hdr) + 1426 sizeof (struct tcphdr) : 1427 #endif 1428 sizeof (struct tcpiphdr) 1429 #ifdef INET6 1430 ) 1431 #endif 1432 ); 1433 } else 1434 ssthresh = 0; 1435 metrics.rmx_ssthresh = ssthresh; 1436 1437 metrics.rmx_rtt = tp->t_srtt; 1438 metrics.rmx_rttvar = tp->t_rttvar; 1439 metrics.rmx_cwnd = tp->snd_cwnd; 1440 metrics.rmx_sendpipe = 0; 1441 metrics.rmx_recvpipe = 0; 1442 1443 tcp_hc_update(&inp->inp_inc, &metrics); 1444 } 1445 1446 /* free the reassembly queue, if any */ 1447 tcp_reass_flush(tp); 1448 1449 #ifdef TCP_OFFLOAD 1450 /* Disconnect offload device, if any. */ 1451 if (tp->t_flags & TF_TOE) 1452 tcp_offload_detach(tp); 1453 #endif 1454 1455 tcp_free_sackholes(tp); 1456 1457 #ifdef TCPPCAP 1458 /* Free the TCP PCAP queues. */ 1459 tcp_pcap_drain(&(tp->t_inpkts)); 1460 tcp_pcap_drain(&(tp->t_outpkts)); 1461 #endif 1462 1463 /* Allow the CC algorithm to clean up after itself. */ 1464 if (CC_ALGO(tp)->cb_destroy != NULL) 1465 CC_ALGO(tp)->cb_destroy(tp->ccv); 1466 1467 khelp_destroy_osd(tp->osd); 1468 1469 CC_ALGO(tp) = NULL; 1470 inp->inp_ppcb = NULL; 1471 if (tp->t_timers->tt_draincnt == 0) { 1472 /* We own the last reference on tcpcb, let's free it. */ 1473 TCPSTATES_DEC(tp->t_state); 1474 if (tp->t_fb->tfb_tcp_fb_fini) 1475 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1476 refcount_release(&tp->t_fb->tfb_refcnt); 1477 tp->t_inpcb = NULL; 1478 uma_zfree(V_tcpcb_zone, tp); 1479 released = in_pcbrele_wlocked(inp); 1480 KASSERT(!released, ("%s: inp %p should not have been released " 1481 "here", __func__, inp)); 1482 } 1483 } 1484 1485 void 1486 tcp_timer_discard(void *ptp) 1487 { 1488 struct inpcb *inp; 1489 struct tcpcb *tp; 1490 1491 tp = (struct tcpcb *)ptp; 1492 CURVNET_SET(tp->t_vnet); 1493 INP_INFO_RLOCK(&V_tcbinfo); 1494 inp = tp->t_inpcb; 1495 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1496 __func__, tp)); 1497 INP_WLOCK(inp); 1498 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1499 ("%s: tcpcb has to be stopped here", __func__)); 1500 tp->t_timers->tt_draincnt--; 1501 if (tp->t_timers->tt_draincnt == 0) { 1502 /* We own the last reference on this tcpcb, let's free it. */ 1503 TCPSTATES_DEC(tp->t_state); 1504 if (tp->t_fb->tfb_tcp_fb_fini) 1505 (*tp->t_fb->tfb_tcp_fb_fini)(tp); 1506 refcount_release(&tp->t_fb->tfb_refcnt); 1507 tp->t_inpcb = NULL; 1508 uma_zfree(V_tcpcb_zone, tp); 1509 if (in_pcbrele_wlocked(inp)) { 1510 INP_INFO_RUNLOCK(&V_tcbinfo); 1511 CURVNET_RESTORE(); 1512 return; 1513 } 1514 } 1515 INP_WUNLOCK(inp); 1516 INP_INFO_RUNLOCK(&V_tcbinfo); 1517 CURVNET_RESTORE(); 1518 } 1519 1520 /* 1521 * Attempt to close a TCP control block, marking it as dropped, and freeing 1522 * the socket if we hold the only reference. 1523 */ 1524 struct tcpcb * 1525 tcp_close(struct tcpcb *tp) 1526 { 1527 struct inpcb *inp = tp->t_inpcb; 1528 struct socket *so; 1529 1530 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1531 INP_WLOCK_ASSERT(inp); 1532 1533 #ifdef TCP_OFFLOAD 1534 if (tp->t_state == TCPS_LISTEN) 1535 tcp_offload_listen_stop(tp); 1536 #endif 1537 #ifdef TCP_RFC7413 1538 /* 1539 * This releases the TFO pending counter resource for TFO listen 1540 * sockets as well as passively-created TFO sockets that transition 1541 * from SYN_RECEIVED to CLOSED. 1542 */ 1543 if (tp->t_tfo_pending) { 1544 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1545 tp->t_tfo_pending = NULL; 1546 } 1547 #endif 1548 in_pcbdrop(inp); 1549 TCPSTAT_INC(tcps_closed); 1550 if (tp->t_state != TCPS_CLOSED) 1551 tcp_state_change(tp, TCPS_CLOSED); 1552 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1553 so = inp->inp_socket; 1554 soisdisconnected(so); 1555 if (inp->inp_flags & INP_SOCKREF) { 1556 KASSERT(so->so_state & SS_PROTOREF, 1557 ("tcp_close: !SS_PROTOREF")); 1558 inp->inp_flags &= ~INP_SOCKREF; 1559 INP_WUNLOCK(inp); 1560 ACCEPT_LOCK(); 1561 SOCK_LOCK(so); 1562 so->so_state &= ~SS_PROTOREF; 1563 sofree(so); 1564 return (NULL); 1565 } 1566 return (tp); 1567 } 1568 1569 void 1570 tcp_drain(void) 1571 { 1572 VNET_ITERATOR_DECL(vnet_iter); 1573 1574 if (!do_tcpdrain) 1575 return; 1576 1577 VNET_LIST_RLOCK_NOSLEEP(); 1578 VNET_FOREACH(vnet_iter) { 1579 CURVNET_SET(vnet_iter); 1580 struct inpcb *inpb; 1581 struct tcpcb *tcpb; 1582 1583 /* 1584 * Walk the tcpbs, if existing, and flush the reassembly queue, 1585 * if there is one... 1586 * XXX: The "Net/3" implementation doesn't imply that the TCP 1587 * reassembly queue should be flushed, but in a situation 1588 * where we're really low on mbufs, this is potentially 1589 * useful. 1590 */ 1591 INP_INFO_WLOCK(&V_tcbinfo); 1592 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1593 if (inpb->inp_flags & INP_TIMEWAIT) 1594 continue; 1595 INP_WLOCK(inpb); 1596 if ((tcpb = intotcpcb(inpb)) != NULL) { 1597 tcp_reass_flush(tcpb); 1598 tcp_clean_sackreport(tcpb); 1599 #ifdef TCPPCAP 1600 if (tcp_pcap_aggressive_free) { 1601 /* Free the TCP PCAP queues. */ 1602 tcp_pcap_drain(&(tcpb->t_inpkts)); 1603 tcp_pcap_drain(&(tcpb->t_outpkts)); 1604 } 1605 #endif 1606 } 1607 INP_WUNLOCK(inpb); 1608 } 1609 INP_INFO_WUNLOCK(&V_tcbinfo); 1610 CURVNET_RESTORE(); 1611 } 1612 VNET_LIST_RUNLOCK_NOSLEEP(); 1613 } 1614 1615 /* 1616 * Notify a tcp user of an asynchronous error; 1617 * store error as soft error, but wake up user 1618 * (for now, won't do anything until can select for soft error). 1619 * 1620 * Do not wake up user since there currently is no mechanism for 1621 * reporting soft errors (yet - a kqueue filter may be added). 1622 */ 1623 static struct inpcb * 1624 tcp_notify(struct inpcb *inp, int error) 1625 { 1626 struct tcpcb *tp; 1627 1628 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 1629 INP_WLOCK_ASSERT(inp); 1630 1631 if ((inp->inp_flags & INP_TIMEWAIT) || 1632 (inp->inp_flags & INP_DROPPED)) 1633 return (inp); 1634 1635 tp = intotcpcb(inp); 1636 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1637 1638 /* 1639 * Ignore some errors if we are hooked up. 1640 * If connection hasn't completed, has retransmitted several times, 1641 * and receives a second error, give up now. This is better 1642 * than waiting a long time to establish a connection that 1643 * can never complete. 1644 */ 1645 if (tp->t_state == TCPS_ESTABLISHED && 1646 (error == EHOSTUNREACH || error == ENETUNREACH || 1647 error == EHOSTDOWN)) { 1648 if (inp->inp_route.ro_rt) { 1649 RTFREE(inp->inp_route.ro_rt); 1650 inp->inp_route.ro_rt = (struct rtentry *)NULL; 1651 } 1652 return (inp); 1653 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1654 tp->t_softerror) { 1655 tp = tcp_drop(tp, error); 1656 if (tp != NULL) 1657 return (inp); 1658 else 1659 return (NULL); 1660 } else { 1661 tp->t_softerror = error; 1662 return (inp); 1663 } 1664 #if 0 1665 wakeup( &so->so_timeo); 1666 sorwakeup(so); 1667 sowwakeup(so); 1668 #endif 1669 } 1670 1671 static int 1672 tcp_pcblist(SYSCTL_HANDLER_ARGS) 1673 { 1674 int error, i, m, n, pcb_count; 1675 struct inpcb *inp, **inp_list; 1676 inp_gen_t gencnt; 1677 struct xinpgen xig; 1678 1679 /* 1680 * The process of preparing the TCB list is too time-consuming and 1681 * resource-intensive to repeat twice on every request. 1682 */ 1683 if (req->oldptr == NULL) { 1684 n = V_tcbinfo.ipi_count + 1685 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1686 n += imax(n / 8, 10); 1687 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1688 return (0); 1689 } 1690 1691 if (req->newptr != NULL) 1692 return (EPERM); 1693 1694 /* 1695 * OK, now we're committed to doing something. 1696 */ 1697 INP_LIST_RLOCK(&V_tcbinfo); 1698 gencnt = V_tcbinfo.ipi_gencnt; 1699 n = V_tcbinfo.ipi_count; 1700 INP_LIST_RUNLOCK(&V_tcbinfo); 1701 1702 m = counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 1703 1704 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1705 + (n + m) * sizeof(struct xtcpcb)); 1706 if (error != 0) 1707 return (error); 1708 1709 xig.xig_len = sizeof xig; 1710 xig.xig_count = n + m; 1711 xig.xig_gen = gencnt; 1712 xig.xig_sogen = so_gencnt; 1713 error = SYSCTL_OUT(req, &xig, sizeof xig); 1714 if (error) 1715 return (error); 1716 1717 error = syncache_pcblist(req, m, &pcb_count); 1718 if (error) 1719 return (error); 1720 1721 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1722 1723 INP_INFO_WLOCK(&V_tcbinfo); 1724 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1725 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1726 INP_WLOCK(inp); 1727 if (inp->inp_gencnt <= gencnt) { 1728 /* 1729 * XXX: This use of cr_cansee(), introduced with 1730 * TCP state changes, is not quite right, but for 1731 * now, better than nothing. 1732 */ 1733 if (inp->inp_flags & INP_TIMEWAIT) { 1734 if (intotw(inp) != NULL) 1735 error = cr_cansee(req->td->td_ucred, 1736 intotw(inp)->tw_cred); 1737 else 1738 error = EINVAL; /* Skip this inp. */ 1739 } else 1740 error = cr_canseeinpcb(req->td->td_ucred, inp); 1741 if (error == 0) { 1742 in_pcbref(inp); 1743 inp_list[i++] = inp; 1744 } 1745 } 1746 INP_WUNLOCK(inp); 1747 } 1748 INP_INFO_WUNLOCK(&V_tcbinfo); 1749 n = i; 1750 1751 error = 0; 1752 for (i = 0; i < n; i++) { 1753 inp = inp_list[i]; 1754 INP_RLOCK(inp); 1755 if (inp->inp_gencnt <= gencnt) { 1756 struct xtcpcb xt; 1757 void *inp_ppcb; 1758 1759 bzero(&xt, sizeof(xt)); 1760 xt.xt_len = sizeof xt; 1761 /* XXX should avoid extra copy */ 1762 bcopy(inp, &xt.xt_inp, sizeof *inp); 1763 inp_ppcb = inp->inp_ppcb; 1764 if (inp_ppcb == NULL) 1765 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1766 else if (inp->inp_flags & INP_TIMEWAIT) { 1767 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1768 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1769 } else { 1770 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1771 if (xt.xt_tp.t_timers) 1772 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1773 } 1774 if (inp->inp_socket != NULL) 1775 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1776 else { 1777 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1778 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1779 } 1780 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1781 INP_RUNLOCK(inp); 1782 error = SYSCTL_OUT(req, &xt, sizeof xt); 1783 } else 1784 INP_RUNLOCK(inp); 1785 } 1786 INP_INFO_RLOCK(&V_tcbinfo); 1787 for (i = 0; i < n; i++) { 1788 inp = inp_list[i]; 1789 INP_RLOCK(inp); 1790 if (!in_pcbrele_rlocked(inp)) 1791 INP_RUNLOCK(inp); 1792 } 1793 INP_INFO_RUNLOCK(&V_tcbinfo); 1794 1795 if (!error) { 1796 /* 1797 * Give the user an updated idea of our state. 1798 * If the generation differs from what we told 1799 * her before, she knows that something happened 1800 * while we were processing this request, and it 1801 * might be necessary to retry. 1802 */ 1803 INP_LIST_RLOCK(&V_tcbinfo); 1804 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1805 xig.xig_sogen = so_gencnt; 1806 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1807 INP_LIST_RUNLOCK(&V_tcbinfo); 1808 error = SYSCTL_OUT(req, &xig, sizeof xig); 1809 } 1810 free(inp_list, M_TEMP); 1811 return (error); 1812 } 1813 1814 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1815 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1816 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1817 1818 #ifdef INET 1819 static int 1820 tcp_getcred(SYSCTL_HANDLER_ARGS) 1821 { 1822 struct xucred xuc; 1823 struct sockaddr_in addrs[2]; 1824 struct inpcb *inp; 1825 int error; 1826 1827 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1828 if (error) 1829 return (error); 1830 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1831 if (error) 1832 return (error); 1833 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1834 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1835 if (inp != NULL) { 1836 if (inp->inp_socket == NULL) 1837 error = ENOENT; 1838 if (error == 0) 1839 error = cr_canseeinpcb(req->td->td_ucred, inp); 1840 if (error == 0) 1841 cru2x(inp->inp_cred, &xuc); 1842 INP_RUNLOCK(inp); 1843 } else 1844 error = ENOENT; 1845 if (error == 0) 1846 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1847 return (error); 1848 } 1849 1850 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1851 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1852 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1853 #endif /* INET */ 1854 1855 #ifdef INET6 1856 static int 1857 tcp6_getcred(SYSCTL_HANDLER_ARGS) 1858 { 1859 struct xucred xuc; 1860 struct sockaddr_in6 addrs[2]; 1861 struct inpcb *inp; 1862 int error; 1863 #ifdef INET 1864 int mapped = 0; 1865 #endif 1866 1867 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1868 if (error) 1869 return (error); 1870 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1871 if (error) 1872 return (error); 1873 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1874 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1875 return (error); 1876 } 1877 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1878 #ifdef INET 1879 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1880 mapped = 1; 1881 else 1882 #endif 1883 return (EINVAL); 1884 } 1885 1886 #ifdef INET 1887 if (mapped == 1) 1888 inp = in_pcblookup(&V_tcbinfo, 1889 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1890 addrs[1].sin6_port, 1891 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1892 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1893 else 1894 #endif 1895 inp = in6_pcblookup(&V_tcbinfo, 1896 &addrs[1].sin6_addr, addrs[1].sin6_port, 1897 &addrs[0].sin6_addr, addrs[0].sin6_port, 1898 INPLOOKUP_RLOCKPCB, NULL); 1899 if (inp != NULL) { 1900 if (inp->inp_socket == NULL) 1901 error = ENOENT; 1902 if (error == 0) 1903 error = cr_canseeinpcb(req->td->td_ucred, inp); 1904 if (error == 0) 1905 cru2x(inp->inp_cred, &xuc); 1906 INP_RUNLOCK(inp); 1907 } else 1908 error = ENOENT; 1909 if (error == 0) 1910 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1911 return (error); 1912 } 1913 1914 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1915 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1916 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1917 #endif /* INET6 */ 1918 1919 1920 #ifdef INET 1921 void 1922 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1923 { 1924 struct ip *ip = vip; 1925 struct tcphdr *th; 1926 struct in_addr faddr; 1927 struct inpcb *inp; 1928 struct tcpcb *tp; 1929 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1930 struct icmp *icp; 1931 struct in_conninfo inc; 1932 tcp_seq icmp_tcp_seq; 1933 int mtu; 1934 1935 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1936 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1937 return; 1938 1939 if (cmd == PRC_MSGSIZE) 1940 notify = tcp_mtudisc_notify; 1941 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1942 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 1943 cmd == PRC_TIMXCEED_INTRANS) && ip) 1944 notify = tcp_drop_syn_sent; 1945 1946 /* 1947 * Hostdead is ugly because it goes linearly through all PCBs. 1948 * XXX: We never get this from ICMP, otherwise it makes an 1949 * excellent DoS attack on machines with many connections. 1950 */ 1951 else if (cmd == PRC_HOSTDEAD) 1952 ip = NULL; 1953 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1954 return; 1955 1956 if (ip == NULL) { 1957 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1958 return; 1959 } 1960 1961 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip)); 1962 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 1963 INP_INFO_RLOCK(&V_tcbinfo); 1964 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src, 1965 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1966 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 1967 /* signal EHOSTDOWN, as it flushes the cached route */ 1968 inp = (*notify)(inp, EHOSTDOWN); 1969 goto out; 1970 } 1971 icmp_tcp_seq = th->th_seq; 1972 if (inp != NULL) { 1973 if (!(inp->inp_flags & INP_TIMEWAIT) && 1974 !(inp->inp_flags & INP_DROPPED) && 1975 !(inp->inp_socket == NULL)) { 1976 tp = intotcpcb(inp); 1977 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 1978 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 1979 if (cmd == PRC_MSGSIZE) { 1980 /* 1981 * MTU discovery: 1982 * If we got a needfrag set the MTU 1983 * in the route to the suggested new 1984 * value (if given) and then notify. 1985 */ 1986 mtu = ntohs(icp->icmp_nextmtu); 1987 /* 1988 * If no alternative MTU was 1989 * proposed, try the next smaller 1990 * one. 1991 */ 1992 if (!mtu) 1993 mtu = ip_next_mtu( 1994 ntohs(ip->ip_len), 1); 1995 if (mtu < V_tcp_minmss + 1996 sizeof(struct tcpiphdr)) 1997 mtu = V_tcp_minmss + 1998 sizeof(struct tcpiphdr); 1999 /* 2000 * Only process the offered MTU if it 2001 * is smaller than the current one. 2002 */ 2003 if (mtu < tp->t_maxseg + 2004 sizeof(struct tcpiphdr)) { 2005 bzero(&inc, sizeof(inc)); 2006 inc.inc_faddr = faddr; 2007 inc.inc_fibnum = 2008 inp->inp_inc.inc_fibnum; 2009 tcp_hc_updatemtu(&inc, mtu); 2010 tcp_mtudisc(inp, mtu); 2011 } 2012 } else 2013 inp = (*notify)(inp, 2014 inetctlerrmap[cmd]); 2015 } 2016 } 2017 } else { 2018 bzero(&inc, sizeof(inc)); 2019 inc.inc_fport = th->th_dport; 2020 inc.inc_lport = th->th_sport; 2021 inc.inc_faddr = faddr; 2022 inc.inc_laddr = ip->ip_src; 2023 syncache_unreach(&inc, icmp_tcp_seq); 2024 } 2025 out: 2026 if (inp != NULL) 2027 INP_WUNLOCK(inp); 2028 INP_INFO_RUNLOCK(&V_tcbinfo); 2029 } 2030 #endif /* INET */ 2031 2032 #ifdef INET6 2033 void 2034 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 2035 { 2036 struct in6_addr *dst; 2037 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 2038 struct ip6_hdr *ip6; 2039 struct mbuf *m; 2040 struct inpcb *inp; 2041 struct tcpcb *tp; 2042 struct icmp6_hdr *icmp6; 2043 struct ip6ctlparam *ip6cp = NULL; 2044 const struct sockaddr_in6 *sa6_src = NULL; 2045 struct in_conninfo inc; 2046 struct tcp_ports { 2047 uint16_t th_sport; 2048 uint16_t th_dport; 2049 } t_ports; 2050 tcp_seq icmp_tcp_seq; 2051 unsigned int mtu; 2052 unsigned int off; 2053 2054 if (sa->sa_family != AF_INET6 || 2055 sa->sa_len != sizeof(struct sockaddr_in6)) 2056 return; 2057 2058 /* if the parameter is from icmp6, decode it. */ 2059 if (d != NULL) { 2060 ip6cp = (struct ip6ctlparam *)d; 2061 icmp6 = ip6cp->ip6c_icmp6; 2062 m = ip6cp->ip6c_m; 2063 ip6 = ip6cp->ip6c_ip6; 2064 off = ip6cp->ip6c_off; 2065 sa6_src = ip6cp->ip6c_src; 2066 dst = ip6cp->ip6c_finaldst; 2067 } else { 2068 m = NULL; 2069 ip6 = NULL; 2070 off = 0; /* fool gcc */ 2071 sa6_src = &sa6_any; 2072 dst = NULL; 2073 } 2074 2075 if (cmd == PRC_MSGSIZE) 2076 notify = tcp_mtudisc_notify; 2077 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 2078 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL || 2079 cmd == PRC_TIMXCEED_INTRANS) && ip6 != NULL) 2080 notify = tcp_drop_syn_sent; 2081 2082 /* 2083 * Hostdead is ugly because it goes linearly through all PCBs. 2084 * XXX: We never get this from ICMP, otherwise it makes an 2085 * excellent DoS attack on machines with many connections. 2086 */ 2087 else if (cmd == PRC_HOSTDEAD) 2088 ip6 = NULL; 2089 else if ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0) 2090 return; 2091 2092 if (ip6 == NULL) { 2093 in6_pcbnotify(&V_tcbinfo, sa, 0, 2094 (const struct sockaddr *)sa6_src, 2095 0, cmd, NULL, notify); 2096 return; 2097 } 2098 2099 /* Check if we can safely get the ports from the tcp hdr */ 2100 if (m == NULL || 2101 (m->m_pkthdr.len < 2102 (int32_t) (off + sizeof(struct tcp_ports)))) { 2103 return; 2104 } 2105 bzero(&t_ports, sizeof(struct tcp_ports)); 2106 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 2107 INP_INFO_RLOCK(&V_tcbinfo); 2108 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 2109 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 2110 if (inp != NULL && PRC_IS_REDIRECT(cmd)) { 2111 /* signal EHOSTDOWN, as it flushes the cached route */ 2112 inp = (*notify)(inp, EHOSTDOWN); 2113 goto out; 2114 } 2115 off += sizeof(struct tcp_ports); 2116 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 2117 goto out; 2118 } 2119 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 2120 if (inp != NULL) { 2121 if (!(inp->inp_flags & INP_TIMEWAIT) && 2122 !(inp->inp_flags & INP_DROPPED) && 2123 !(inp->inp_socket == NULL)) { 2124 tp = intotcpcb(inp); 2125 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2126 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2127 if (cmd == PRC_MSGSIZE) { 2128 /* 2129 * MTU discovery: 2130 * If we got a needfrag set the MTU 2131 * in the route to the suggested new 2132 * value (if given) and then notify. 2133 */ 2134 mtu = ntohl(icmp6->icmp6_mtu); 2135 /* 2136 * If no alternative MTU was 2137 * proposed, or the proposed 2138 * MTU was too small, set to 2139 * the min. 2140 */ 2141 if (mtu < IPV6_MMTU) 2142 mtu = IPV6_MMTU - 8; 2143 bzero(&inc, sizeof(inc)); 2144 inc.inc_fibnum = M_GETFIB(m); 2145 inc.inc_flags |= INC_ISIPV6; 2146 inc.inc6_faddr = *dst; 2147 if (in6_setscope(&inc.inc6_faddr, 2148 m->m_pkthdr.rcvif, NULL)) 2149 goto out; 2150 /* 2151 * Only process the offered MTU if it 2152 * is smaller than the current one. 2153 */ 2154 if (mtu < tp->t_maxseg + 2155 sizeof (struct tcphdr) + 2156 sizeof (struct ip6_hdr)) { 2157 tcp_hc_updatemtu(&inc, mtu); 2158 tcp_mtudisc(inp, mtu); 2159 ICMP6STAT_INC(icp6s_pmtuchg); 2160 } 2161 } else 2162 inp = (*notify)(inp, 2163 inet6ctlerrmap[cmd]); 2164 } 2165 } 2166 } else { 2167 bzero(&inc, sizeof(inc)); 2168 inc.inc_fibnum = M_GETFIB(m); 2169 inc.inc_flags |= INC_ISIPV6; 2170 inc.inc_fport = t_ports.th_dport; 2171 inc.inc_lport = t_ports.th_sport; 2172 inc.inc6_faddr = *dst; 2173 inc.inc6_laddr = ip6->ip6_src; 2174 syncache_unreach(&inc, icmp_tcp_seq); 2175 } 2176 out: 2177 if (inp != NULL) 2178 INP_WUNLOCK(inp); 2179 INP_INFO_RUNLOCK(&V_tcbinfo); 2180 } 2181 #endif /* INET6 */ 2182 2183 2184 /* 2185 * Following is where TCP initial sequence number generation occurs. 2186 * 2187 * There are two places where we must use initial sequence numbers: 2188 * 1. In SYN-ACK packets. 2189 * 2. In SYN packets. 2190 * 2191 * All ISNs for SYN-ACK packets are generated by the syncache. See 2192 * tcp_syncache.c for details. 2193 * 2194 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 2195 * depends on this property. In addition, these ISNs should be 2196 * unguessable so as to prevent connection hijacking. To satisfy 2197 * the requirements of this situation, the algorithm outlined in 2198 * RFC 1948 is used, with only small modifications. 2199 * 2200 * Implementation details: 2201 * 2202 * Time is based off the system timer, and is corrected so that it 2203 * increases by one megabyte per second. This allows for proper 2204 * recycling on high speed LANs while still leaving over an hour 2205 * before rollover. 2206 * 2207 * As reading the *exact* system time is too expensive to be done 2208 * whenever setting up a TCP connection, we increment the time 2209 * offset in two ways. First, a small random positive increment 2210 * is added to isn_offset for each connection that is set up. 2211 * Second, the function tcp_isn_tick fires once per clock tick 2212 * and increments isn_offset as necessary so that sequence numbers 2213 * are incremented at approximately ISN_BYTES_PER_SECOND. The 2214 * random positive increments serve only to ensure that the same 2215 * exact sequence number is never sent out twice (as could otherwise 2216 * happen when a port is recycled in less than the system tick 2217 * interval.) 2218 * 2219 * net.inet.tcp.isn_reseed_interval controls the number of seconds 2220 * between seeding of isn_secret. This is normally set to zero, 2221 * as reseeding should not be necessary. 2222 * 2223 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 2224 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 2225 * general, this means holding an exclusive (write) lock. 2226 */ 2227 2228 #define ISN_BYTES_PER_SECOND 1048576 2229 #define ISN_STATIC_INCREMENT 4096 2230 #define ISN_RANDOM_INCREMENT (4096 - 1) 2231 2232 static VNET_DEFINE(u_char, isn_secret[32]); 2233 static VNET_DEFINE(int, isn_last); 2234 static VNET_DEFINE(int, isn_last_reseed); 2235 static VNET_DEFINE(u_int32_t, isn_offset); 2236 static VNET_DEFINE(u_int32_t, isn_offset_old); 2237 2238 #define V_isn_secret VNET(isn_secret) 2239 #define V_isn_last VNET(isn_last) 2240 #define V_isn_last_reseed VNET(isn_last_reseed) 2241 #define V_isn_offset VNET(isn_offset) 2242 #define V_isn_offset_old VNET(isn_offset_old) 2243 2244 tcp_seq 2245 tcp_new_isn(struct tcpcb *tp) 2246 { 2247 MD5_CTX isn_ctx; 2248 u_int32_t md5_buffer[4]; 2249 tcp_seq new_isn; 2250 u_int32_t projected_offset; 2251 2252 INP_WLOCK_ASSERT(tp->t_inpcb); 2253 2254 ISN_LOCK(); 2255 /* Seed if this is the first use, reseed if requested. */ 2256 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 2257 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 2258 < (u_int)ticks))) { 2259 read_random(&V_isn_secret, sizeof(V_isn_secret)); 2260 V_isn_last_reseed = ticks; 2261 } 2262 2263 /* Compute the md5 hash and return the ISN. */ 2264 MD5Init(&isn_ctx); 2265 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 2266 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 2267 #ifdef INET6 2268 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 2269 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 2270 sizeof(struct in6_addr)); 2271 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 2272 sizeof(struct in6_addr)); 2273 } else 2274 #endif 2275 { 2276 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 2277 sizeof(struct in_addr)); 2278 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 2279 sizeof(struct in_addr)); 2280 } 2281 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 2282 MD5Final((u_char *) &md5_buffer, &isn_ctx); 2283 new_isn = (tcp_seq) md5_buffer[0]; 2284 V_isn_offset += ISN_STATIC_INCREMENT + 2285 (arc4random() & ISN_RANDOM_INCREMENT); 2286 if (ticks != V_isn_last) { 2287 projected_offset = V_isn_offset_old + 2288 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 2289 if (SEQ_GT(projected_offset, V_isn_offset)) 2290 V_isn_offset = projected_offset; 2291 V_isn_offset_old = V_isn_offset; 2292 V_isn_last = ticks; 2293 } 2294 new_isn += V_isn_offset; 2295 ISN_UNLOCK(); 2296 return (new_isn); 2297 } 2298 2299 /* 2300 * When a specific ICMP unreachable message is received and the 2301 * connection state is SYN-SENT, drop the connection. This behavior 2302 * is controlled by the icmp_may_rst sysctl. 2303 */ 2304 struct inpcb * 2305 tcp_drop_syn_sent(struct inpcb *inp, int errno) 2306 { 2307 struct tcpcb *tp; 2308 2309 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2310 INP_WLOCK_ASSERT(inp); 2311 2312 if ((inp->inp_flags & INP_TIMEWAIT) || 2313 (inp->inp_flags & INP_DROPPED)) 2314 return (inp); 2315 2316 tp = intotcpcb(inp); 2317 if (tp->t_state != TCPS_SYN_SENT) 2318 return (inp); 2319 2320 tp = tcp_drop(tp, errno); 2321 if (tp != NULL) 2322 return (inp); 2323 else 2324 return (NULL); 2325 } 2326 2327 /* 2328 * When `need fragmentation' ICMP is received, update our idea of the MSS 2329 * based on the new value. Also nudge TCP to send something, since we 2330 * know the packet we just sent was dropped. 2331 * This duplicates some code in the tcp_mss() function in tcp_input.c. 2332 */ 2333 static struct inpcb * 2334 tcp_mtudisc_notify(struct inpcb *inp, int error) 2335 { 2336 2337 tcp_mtudisc(inp, -1); 2338 return (inp); 2339 } 2340 2341 static void 2342 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 2343 { 2344 struct tcpcb *tp; 2345 struct socket *so; 2346 2347 INP_WLOCK_ASSERT(inp); 2348 if ((inp->inp_flags & INP_TIMEWAIT) || 2349 (inp->inp_flags & INP_DROPPED)) 2350 return; 2351 2352 tp = intotcpcb(inp); 2353 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 2354 2355 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 2356 2357 so = inp->inp_socket; 2358 SOCKBUF_LOCK(&so->so_snd); 2359 /* If the mss is larger than the socket buffer, decrease the mss. */ 2360 if (so->so_snd.sb_hiwat < tp->t_maxseg) 2361 tp->t_maxseg = so->so_snd.sb_hiwat; 2362 SOCKBUF_UNLOCK(&so->so_snd); 2363 2364 TCPSTAT_INC(tcps_mturesent); 2365 tp->t_rtttime = 0; 2366 tp->snd_nxt = tp->snd_una; 2367 tcp_free_sackholes(tp); 2368 tp->snd_recover = tp->snd_max; 2369 if (tp->t_flags & TF_SACK_PERMIT) 2370 EXIT_FASTRECOVERY(tp->t_flags); 2371 tp->t_fb->tfb_tcp_output(tp); 2372 } 2373 2374 #ifdef INET 2375 /* 2376 * Look-up the routing entry to the peer of this inpcb. If no route 2377 * is found and it cannot be allocated, then return 0. This routine 2378 * is called by TCP routines that access the rmx structure and by 2379 * tcp_mss_update to get the peer/interface MTU. 2380 */ 2381 u_long 2382 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 2383 { 2384 struct nhop4_extended nh4; 2385 struct ifnet *ifp; 2386 u_long maxmtu = 0; 2387 2388 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 2389 2390 if (inc->inc_faddr.s_addr != INADDR_ANY) { 2391 2392 if (fib4_lookup_nh_ext(inc->inc_fibnum, inc->inc_faddr, 2393 NHR_REF, 0, &nh4) != 0) 2394 return (0); 2395 2396 ifp = nh4.nh_ifp; 2397 maxmtu = nh4.nh_mtu; 2398 2399 /* Report additional interface capabilities. */ 2400 if (cap != NULL) { 2401 if (ifp->if_capenable & IFCAP_TSO4 && 2402 ifp->if_hwassist & CSUM_TSO) { 2403 cap->ifcap |= CSUM_TSO; 2404 cap->tsomax = ifp->if_hw_tsomax; 2405 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2406 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2407 } 2408 } 2409 fib4_free_nh_ext(inc->inc_fibnum, &nh4); 2410 } 2411 return (maxmtu); 2412 } 2413 #endif /* INET */ 2414 2415 #ifdef INET6 2416 u_long 2417 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 2418 { 2419 struct nhop6_extended nh6; 2420 struct in6_addr dst6; 2421 uint32_t scopeid; 2422 struct ifnet *ifp; 2423 u_long maxmtu = 0; 2424 2425 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 2426 2427 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 2428 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 2429 if (fib6_lookup_nh_ext(inc->inc_fibnum, &dst6, scopeid, 0, 2430 0, &nh6) != 0) 2431 return (0); 2432 2433 ifp = nh6.nh_ifp; 2434 maxmtu = nh6.nh_mtu; 2435 2436 /* Report additional interface capabilities. */ 2437 if (cap != NULL) { 2438 if (ifp->if_capenable & IFCAP_TSO6 && 2439 ifp->if_hwassist & CSUM_TSO) { 2440 cap->ifcap |= CSUM_TSO; 2441 cap->tsomax = ifp->if_hw_tsomax; 2442 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 2443 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 2444 } 2445 } 2446 fib6_free_nh_ext(inc->inc_fibnum, &nh6); 2447 } 2448 2449 return (maxmtu); 2450 } 2451 #endif /* INET6 */ 2452 2453 /* 2454 * Calculate effective SMSS per RFC5681 definition for a given TCP 2455 * connection at its current state, taking into account SACK and etc. 2456 */ 2457 u_int 2458 tcp_maxseg(const struct tcpcb *tp) 2459 { 2460 u_int optlen; 2461 2462 if (tp->t_flags & TF_NOOPT) 2463 return (tp->t_maxseg); 2464 2465 /* 2466 * Here we have a simplified code from tcp_addoptions(), 2467 * without a proper loop, and having most of paddings hardcoded. 2468 * We might make mistakes with padding here in some edge cases, 2469 * but this is harmless, since result of tcp_maxseg() is used 2470 * only in cwnd and ssthresh estimations. 2471 */ 2472 #define PAD(len) ((((len) / 4) + !!((len) % 4)) * 4) 2473 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 2474 if (tp->t_flags & TF_RCVD_TSTMP) 2475 optlen = TCPOLEN_TSTAMP_APPA; 2476 else 2477 optlen = 0; 2478 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2479 if (tp->t_flags & TF_SIGNATURE) 2480 optlen += PAD(TCPOLEN_SIGNATURE); 2481 #endif 2482 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 2483 optlen += TCPOLEN_SACKHDR; 2484 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 2485 optlen = PAD(optlen); 2486 } 2487 } else { 2488 if (tp->t_flags & TF_REQ_TSTMP) 2489 optlen = TCPOLEN_TSTAMP_APPA; 2490 else 2491 optlen = PAD(TCPOLEN_MAXSEG); 2492 if (tp->t_flags & TF_REQ_SCALE) 2493 optlen += PAD(TCPOLEN_WINDOW); 2494 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2495 if (tp->t_flags & TF_SIGNATURE) 2496 optlen += PAD(TCPOLEN_SIGNATURE); 2497 #endif 2498 if (tp->t_flags & TF_SACK_PERMIT) 2499 optlen += PAD(TCPOLEN_SACK_PERMITTED); 2500 } 2501 #undef PAD 2502 optlen = min(optlen, TCP_MAXOLEN); 2503 return (tp->t_maxseg - optlen); 2504 } 2505 2506 static int 2507 sysctl_drop(SYSCTL_HANDLER_ARGS) 2508 { 2509 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2510 struct sockaddr_storage addrs[2]; 2511 struct inpcb *inp; 2512 struct tcpcb *tp; 2513 struct tcptw *tw; 2514 struct sockaddr_in *fin, *lin; 2515 #ifdef INET6 2516 struct sockaddr_in6 *fin6, *lin6; 2517 #endif 2518 int error; 2519 2520 inp = NULL; 2521 fin = lin = NULL; 2522 #ifdef INET6 2523 fin6 = lin6 = NULL; 2524 #endif 2525 error = 0; 2526 2527 if (req->oldptr != NULL || req->oldlen != 0) 2528 return (EINVAL); 2529 if (req->newptr == NULL) 2530 return (EPERM); 2531 if (req->newlen < sizeof(addrs)) 2532 return (ENOMEM); 2533 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2534 if (error) 2535 return (error); 2536 2537 switch (addrs[0].ss_family) { 2538 #ifdef INET6 2539 case AF_INET6: 2540 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2541 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2542 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2543 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2544 return (EINVAL); 2545 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2546 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2547 return (EINVAL); 2548 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2549 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2550 fin = (struct sockaddr_in *)&addrs[0]; 2551 lin = (struct sockaddr_in *)&addrs[1]; 2552 break; 2553 } 2554 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2555 if (error) 2556 return (error); 2557 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2558 if (error) 2559 return (error); 2560 break; 2561 #endif 2562 #ifdef INET 2563 case AF_INET: 2564 fin = (struct sockaddr_in *)&addrs[0]; 2565 lin = (struct sockaddr_in *)&addrs[1]; 2566 if (fin->sin_len != sizeof(struct sockaddr_in) || 2567 lin->sin_len != sizeof(struct sockaddr_in)) 2568 return (EINVAL); 2569 break; 2570 #endif 2571 default: 2572 return (EINVAL); 2573 } 2574 INP_INFO_RLOCK(&V_tcbinfo); 2575 switch (addrs[0].ss_family) { 2576 #ifdef INET6 2577 case AF_INET6: 2578 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2579 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2580 INPLOOKUP_WLOCKPCB, NULL); 2581 break; 2582 #endif 2583 #ifdef INET 2584 case AF_INET: 2585 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2586 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2587 break; 2588 #endif 2589 } 2590 if (inp != NULL) { 2591 if (inp->inp_flags & INP_TIMEWAIT) { 2592 /* 2593 * XXXRW: There currently exists a state where an 2594 * inpcb is present, but its timewait state has been 2595 * discarded. For now, don't allow dropping of this 2596 * type of inpcb. 2597 */ 2598 tw = intotw(inp); 2599 if (tw != NULL) 2600 tcp_twclose(tw, 0); 2601 else 2602 INP_WUNLOCK(inp); 2603 } else if (!(inp->inp_flags & INP_DROPPED) && 2604 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2605 tp = intotcpcb(inp); 2606 tp = tcp_drop(tp, ECONNABORTED); 2607 if (tp != NULL) 2608 INP_WUNLOCK(inp); 2609 } else 2610 INP_WUNLOCK(inp); 2611 } else 2612 error = ESRCH; 2613 INP_INFO_RUNLOCK(&V_tcbinfo); 2614 return (error); 2615 } 2616 2617 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2618 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL, 2619 0, sysctl_drop, "", "Drop TCP connection"); 2620 2621 /* 2622 * Generate a standardized TCP log line for use throughout the 2623 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2624 * allow use in the interrupt context. 2625 * 2626 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2627 * NB: The function may return NULL if memory allocation failed. 2628 * 2629 * Due to header inclusion and ordering limitations the struct ip 2630 * and ip6_hdr pointers have to be passed as void pointers. 2631 */ 2632 char * 2633 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2634 const void *ip6hdr) 2635 { 2636 2637 /* Is logging enabled? */ 2638 if (tcp_log_in_vain == 0) 2639 return (NULL); 2640 2641 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2642 } 2643 2644 char * 2645 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2646 const void *ip6hdr) 2647 { 2648 2649 /* Is logging enabled? */ 2650 if (tcp_log_debug == 0) 2651 return (NULL); 2652 2653 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2654 } 2655 2656 static char * 2657 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2658 const void *ip6hdr) 2659 { 2660 char *s, *sp; 2661 size_t size; 2662 struct ip *ip; 2663 #ifdef INET6 2664 const struct ip6_hdr *ip6; 2665 2666 ip6 = (const struct ip6_hdr *)ip6hdr; 2667 #endif /* INET6 */ 2668 ip = (struct ip *)ip4hdr; 2669 2670 /* 2671 * The log line looks like this: 2672 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2673 */ 2674 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2675 sizeof(PRINT_TH_FLAGS) + 1 + 2676 #ifdef INET6 2677 2 * INET6_ADDRSTRLEN; 2678 #else 2679 2 * INET_ADDRSTRLEN; 2680 #endif /* INET6 */ 2681 2682 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2683 if (s == NULL) 2684 return (NULL); 2685 2686 strcat(s, "TCP: ["); 2687 sp = s + strlen(s); 2688 2689 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2690 inet_ntoa_r(inc->inc_faddr, sp); 2691 sp = s + strlen(s); 2692 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2693 sp = s + strlen(s); 2694 inet_ntoa_r(inc->inc_laddr, sp); 2695 sp = s + strlen(s); 2696 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2697 #ifdef INET6 2698 } else if (inc) { 2699 ip6_sprintf(sp, &inc->inc6_faddr); 2700 sp = s + strlen(s); 2701 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2702 sp = s + strlen(s); 2703 ip6_sprintf(sp, &inc->inc6_laddr); 2704 sp = s + strlen(s); 2705 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2706 } else if (ip6 && th) { 2707 ip6_sprintf(sp, &ip6->ip6_src); 2708 sp = s + strlen(s); 2709 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2710 sp = s + strlen(s); 2711 ip6_sprintf(sp, &ip6->ip6_dst); 2712 sp = s + strlen(s); 2713 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2714 #endif /* INET6 */ 2715 #ifdef INET 2716 } else if (ip && th) { 2717 inet_ntoa_r(ip->ip_src, sp); 2718 sp = s + strlen(s); 2719 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2720 sp = s + strlen(s); 2721 inet_ntoa_r(ip->ip_dst, sp); 2722 sp = s + strlen(s); 2723 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2724 #endif /* INET */ 2725 } else { 2726 free(s, M_TCPLOG); 2727 return (NULL); 2728 } 2729 sp = s + strlen(s); 2730 if (th) 2731 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2732 if (*(s + size - 1) != '\0') 2733 panic("%s: string too long", __func__); 2734 return (s); 2735 } 2736 2737 /* 2738 * A subroutine which makes it easy to track TCP state changes with DTrace. 2739 * This function shouldn't be called for t_state initializations that don't 2740 * correspond to actual TCP state transitions. 2741 */ 2742 void 2743 tcp_state_change(struct tcpcb *tp, int newstate) 2744 { 2745 #if defined(KDTRACE_HOOKS) 2746 int pstate = tp->t_state; 2747 #endif 2748 2749 TCPSTATES_DEC(tp->t_state); 2750 TCPSTATES_INC(newstate); 2751 tp->t_state = newstate; 2752 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2753 }
Cache object: 284669cb8cb8e5b6471d60a5b5c3fdc0
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]