1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 #include <sys/cdefs.h> 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include "opt_ipsec.h" 36 #include "opt_kern_tls.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/arb.h> 41 #include <sys/callout.h> 42 #include <sys/eventhandler.h> 43 #ifdef TCP_HHOOK 44 #include <sys/hhook.h> 45 #endif 46 #include <sys/kernel.h> 47 #ifdef TCP_HHOOK 48 #include <sys/khelp.h> 49 #endif 50 #ifdef KERN_TLS 51 #include <sys/ktls.h> 52 #endif 53 #include <sys/qmath.h> 54 #include <sys/stats.h> 55 #include <sys/sysctl.h> 56 #include <sys/jail.h> 57 #include <sys/malloc.h> 58 #include <sys/refcount.h> 59 #include <sys/mbuf.h> 60 #include <sys/priv.h> 61 #include <sys/proc.h> 62 #include <sys/sdt.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/protosw.h> 66 #include <sys/random.h> 67 68 #include <vm/uma.h> 69 70 #include <net/route.h> 71 #include <net/route/nhop.h> 72 #include <net/if.h> 73 #include <net/if_var.h> 74 #include <net/if_private.h> 75 #include <net/vnet.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_fib.h> 79 #include <netinet/in_kdtrace.h> 80 #include <netinet/in_pcb.h> 81 #include <netinet/in_systm.h> 82 #include <netinet/in_var.h> 83 #include <netinet/ip.h> 84 #include <netinet/ip_icmp.h> 85 #include <netinet/ip_var.h> 86 #ifdef INET6 87 #include <netinet/icmp6.h> 88 #include <netinet/ip6.h> 89 #include <netinet6/in6_fib.h> 90 #include <netinet6/in6_pcb.h> 91 #include <netinet6/ip6_var.h> 92 #include <netinet6/scope6_var.h> 93 #include <netinet6/nd6.h> 94 #endif 95 96 #include <netinet/tcp.h> 97 #ifdef INVARIANTS 98 #define TCPSTATES 99 #endif 100 #include <netinet/tcp_fsm.h> 101 #include <netinet/tcp_seq.h> 102 #include <netinet/tcp_timer.h> 103 #include <netinet/tcp_var.h> 104 #include <netinet/tcp_ecn.h> 105 #include <netinet/tcp_log_buf.h> 106 #include <netinet/tcp_syncache.h> 107 #include <netinet/tcp_hpts.h> 108 #include <netinet/tcp_lro.h> 109 #include <netinet/cc/cc.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/tcp_fastopen.h> 112 #include <netinet/tcp_accounting.h> 113 #ifdef TCPPCAP 114 #include <netinet/tcp_pcap.h> 115 #endif 116 #ifdef TCP_OFFLOAD 117 #include <netinet/tcp_offload.h> 118 #endif 119 #include <netinet/udp.h> 120 #include <netinet/udp_var.h> 121 #ifdef INET6 122 #include <netinet6/tcp6_var.h> 123 #endif 124 125 #include <netipsec/ipsec_support.h> 126 127 #include <machine/in_cksum.h> 128 #include <crypto/siphash/siphash.h> 129 130 #include <security/mac/mac_framework.h> 131 132 #ifdef INET6 133 static ip6proto_ctlinput_t tcp6_ctlinput; 134 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 135 #endif 136 137 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 138 #ifdef INET6 139 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 140 #endif 141 142 VNET_DEFINE(uint32_t, tcp_ack_war_time_window) = 1000; 143 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 144 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ack_war_time_window), 0, 145 "Time interval in ms used to limit the number (ack_war_cnt) of challenge ACKs sent per TCP connection"); 146 VNET_DEFINE(uint32_t, tcp_ack_war_cnt) = 5; 147 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, CTLFLAG_VNET | CTLFLAG_RW, 148 &VNET_NAME(tcp_ack_war_cnt), 0, 149 "Maximum number of challenge ACKs sent per TCP connection during the time interval (ack_war_timewindow)"); 150 151 struct rwlock tcp_function_lock; 152 153 static int 154 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 155 { 156 int error, new; 157 158 new = V_tcp_mssdflt; 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_mssdflt = new; 165 } 166 return (error); 167 } 168 169 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 170 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 171 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 172 "Default TCP Maximum Segment Size"); 173 174 #ifdef INET6 175 static int 176 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 177 { 178 int error, new; 179 180 new = V_tcp_v6mssdflt; 181 error = sysctl_handle_int(oidp, &new, 0, req); 182 if (error == 0 && req->newptr) { 183 if (new < TCP_MINMSS) 184 error = EINVAL; 185 else 186 V_tcp_v6mssdflt = new; 187 } 188 return (error); 189 } 190 191 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 192 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 193 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 194 "Default TCP Maximum Segment Size for IPv6"); 195 #endif /* INET6 */ 196 197 /* 198 * Minimum MSS we accept and use. This prevents DoS attacks where 199 * we are forced to a ridiculous low MSS like 20 and send hundreds 200 * of packets instead of one. The effect scales with the available 201 * bandwidth and quickly saturates the CPU and network interface 202 * with packet generation and sending. Set to zero to disable MINMSS 203 * checking. This setting prevents us from sending too small packets. 204 */ 205 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_minmss), 0, 208 "Minimum TCP Maximum Segment Size"); 209 210 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 211 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_do_rfc1323), 0, 213 "Enable rfc1323 (high performance TCP) extensions"); 214 215 /* 216 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 217 * Some stacks negotiate TS, but never send them after connection setup. Some 218 * stacks negotiate TS, but don't send them when sending keep-alive segments. 219 * These include modern widely deployed TCP stacks. 220 * Therefore tolerating violations for now... 221 */ 222 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_tolerate_missing_ts), 0, 225 "Tolerate missing TCP timestamps"); 226 227 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_ts_offset_per_conn), 0, 230 "Initialize TCP timestamps per connection instead of per host pair"); 231 232 /* How many connections are pacing */ 233 static volatile uint32_t number_of_tcp_connections_pacing = 0; 234 static uint32_t shadow_num_connections = 0; 235 static counter_u64_t tcp_pacing_failures; 236 static counter_u64_t tcp_dgp_failures; 237 static uint32_t shadow_tcp_pacing_dgp = 0; 238 static volatile uint32_t number_of_dgp_connections = 0; 239 240 static int tcp_pacing_limit = 10000; 241 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 242 &tcp_pacing_limit, 1000, 243 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 244 245 static int tcp_dgp_limit = -1; 246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, dgp_limit, CTLFLAG_RW, 247 &tcp_dgp_limit, -1, 248 "If the TCP stack does DGP, is there a limit (-1 = no, 0 = no dgp N = number of connections)"); 249 250 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 251 &shadow_num_connections, 0, "Number of TCP connections being paced"); 252 253 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD, 254 &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit"); 255 256 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, dgp_failures, CTLFLAG_RD, 257 &tcp_dgp_failures, "Number of times we failed to enable dgp to avoid exceeding the limit"); 258 259 static int tcp_log_debug = 0; 260 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 261 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 262 263 /* 264 * Target size of TCP PCB hash tables. Must be a power of two. 265 * 266 * Note that this can be overridden by the kernel environment 267 * variable net.inet.tcp.tcbhashsize 268 */ 269 #ifndef TCBHASHSIZE 270 #define TCBHASHSIZE 0 271 #endif 272 static int tcp_tcbhashsize = TCBHASHSIZE; 273 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 274 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 275 276 static int do_tcpdrain = 1; 277 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 278 "Enable tcp_drain routine for extra help when low on mbufs"); 279 280 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 281 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 282 283 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 284 #define V_icmp_may_rst VNET(icmp_may_rst) 285 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 286 &VNET_NAME(icmp_may_rst), 0, 287 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 288 289 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 290 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 291 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 292 &VNET_NAME(tcp_isn_reseed_interval), 0, 293 "Seconds between reseeding of ISN secret"); 294 295 static int tcp_soreceive_stream; 296 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 297 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 298 299 VNET_DEFINE(uma_zone_t, sack_hole_zone); 300 #define V_sack_hole_zone VNET(sack_hole_zone) 301 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 302 static int 303 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 304 { 305 int error; 306 uint32_t new; 307 308 new = V_tcp_map_entries_limit; 309 error = sysctl_handle_int(oidp, &new, 0, req); 310 if (error == 0 && req->newptr) { 311 /* only allow "0" and value > minimum */ 312 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 313 error = EINVAL; 314 else 315 V_tcp_map_entries_limit = new; 316 } 317 return (error); 318 } 319 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 320 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 321 &VNET_NAME(tcp_map_entries_limit), 0, 322 &sysctl_net_inet_tcp_map_limit_check, "IU", 323 "Total sendmap entries limit"); 324 325 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 326 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 327 &VNET_NAME(tcp_map_split_limit), 0, 328 "Total sendmap split entries limit"); 329 330 #ifdef TCP_HHOOK 331 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 332 #endif 333 334 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 335 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 336 #define V_ts_offset_secret VNET(ts_offset_secret) 337 338 static int tcp_default_fb_init(struct tcpcb *tp, void **ptr); 339 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 340 static int tcp_default_handoff_ok(struct tcpcb *tp); 341 static struct inpcb *tcp_notify(struct inpcb *, int); 342 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 343 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 344 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 345 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 346 const void *ip4hdr, const void *ip6hdr); 347 static void tcp_default_switch_failed(struct tcpcb *tp); 348 static ipproto_ctlinput_t tcp_ctlinput; 349 static udp_tun_icmp_t tcp_ctlinput_viaudp; 350 351 static struct tcp_function_block tcp_def_funcblk = { 352 .tfb_tcp_block_name = "freebsd", 353 .tfb_tcp_output = tcp_default_output, 354 .tfb_tcp_do_segment = tcp_do_segment, 355 .tfb_tcp_ctloutput = tcp_default_ctloutput, 356 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 357 .tfb_tcp_fb_init = tcp_default_fb_init, 358 .tfb_tcp_fb_fini = tcp_default_fb_fini, 359 .tfb_switch_failed = tcp_default_switch_failed, 360 .tfb_flags = TCP_FUNC_DEFAULT_OK, 361 }; 362 363 static int tcp_fb_cnt = 0; 364 struct tcp_funchead t_functions; 365 VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk; 366 #define V_tcp_func_set_ptr VNET(tcp_func_set_ptr) 367 368 void 369 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 370 { 371 TCPSTAT_INC(tcps_dsack_count); 372 tp->t_dsack_pack++; 373 if (tlp == 0) { 374 if (SEQ_GT(end, start)) { 375 tp->t_dsack_bytes += (end - start); 376 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 377 } else { 378 tp->t_dsack_tlp_bytes += (start - end); 379 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 380 } 381 } else { 382 if (SEQ_GT(end, start)) { 383 tp->t_dsack_bytes += (end - start); 384 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 385 } else { 386 tp->t_dsack_tlp_bytes += (start - end); 387 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 388 } 389 } 390 } 391 392 static struct tcp_function_block * 393 find_tcp_functions_locked(struct tcp_function_set *fs) 394 { 395 struct tcp_function *f; 396 struct tcp_function_block *blk = NULL; 397 398 rw_assert(&tcp_function_lock, RA_LOCKED); 399 TAILQ_FOREACH(f, &t_functions, tf_next) { 400 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 401 blk = f->tf_fb; 402 break; 403 } 404 } 405 return (blk); 406 } 407 408 static struct tcp_function_block * 409 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 410 { 411 struct tcp_function_block *rblk = NULL; 412 struct tcp_function *f; 413 414 rw_assert(&tcp_function_lock, RA_LOCKED); 415 TAILQ_FOREACH(f, &t_functions, tf_next) { 416 if (f->tf_fb == blk) { 417 rblk = blk; 418 if (s) { 419 *s = f; 420 } 421 break; 422 } 423 } 424 return (rblk); 425 } 426 427 struct tcp_function_block * 428 find_and_ref_tcp_functions(struct tcp_function_set *fs) 429 { 430 struct tcp_function_block *blk; 431 432 rw_rlock(&tcp_function_lock); 433 blk = find_tcp_functions_locked(fs); 434 if (blk) 435 refcount_acquire(&blk->tfb_refcnt); 436 rw_runlock(&tcp_function_lock); 437 return (blk); 438 } 439 440 struct tcp_function_block * 441 find_and_ref_tcp_fb(struct tcp_function_block *blk) 442 { 443 struct tcp_function_block *rblk; 444 445 rw_rlock(&tcp_function_lock); 446 rblk = find_tcp_fb_locked(blk, NULL); 447 if (rblk) 448 refcount_acquire(&rblk->tfb_refcnt); 449 rw_runlock(&tcp_function_lock); 450 return (rblk); 451 } 452 453 /* Find a matching alias for the given tcp_function_block. */ 454 int 455 find_tcp_function_alias(struct tcp_function_block *blk, 456 struct tcp_function_set *fs) 457 { 458 struct tcp_function *f; 459 int found; 460 461 found = 0; 462 rw_rlock(&tcp_function_lock); 463 TAILQ_FOREACH(f, &t_functions, tf_next) { 464 if ((f->tf_fb == blk) && 465 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 466 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 467 /* Matching function block with different name. */ 468 strncpy(fs->function_set_name, f->tf_name, 469 TCP_FUNCTION_NAME_LEN_MAX); 470 found = 1; 471 break; 472 } 473 } 474 /* Null terminate the string appropriately. */ 475 if (found) { 476 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 477 } else { 478 fs->function_set_name[0] = '\0'; 479 } 480 rw_runlock(&tcp_function_lock); 481 return (found); 482 } 483 484 static struct tcp_function_block * 485 find_and_ref_tcp_default_fb(void) 486 { 487 struct tcp_function_block *rblk; 488 489 rw_rlock(&tcp_function_lock); 490 rblk = V_tcp_func_set_ptr; 491 refcount_acquire(&rblk->tfb_refcnt); 492 rw_runlock(&tcp_function_lock); 493 return (rblk); 494 } 495 496 void 497 tcp_switch_back_to_default(struct tcpcb *tp) 498 { 499 struct tcp_function_block *tfb; 500 void *ptr = NULL; 501 502 KASSERT(tp->t_fb != &tcp_def_funcblk, 503 ("%s: called by the built-in default stack", __func__)); 504 505 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 506 tp->t_fb->tfb_tcp_timer_stop_all(tp); 507 508 /* 509 * Now, we'll find a new function block to use. 510 * Start by trying the current user-selected 511 * default, unless this stack is the user-selected 512 * default. 513 */ 514 tfb = find_and_ref_tcp_default_fb(); 515 if (tfb == tp->t_fb) { 516 refcount_release(&tfb->tfb_refcnt); 517 tfb = NULL; 518 } 519 /* Does the stack accept this connection? */ 520 if (tfb != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { 521 refcount_release(&tfb->tfb_refcnt); 522 tfb = NULL; 523 } 524 /* Try to use that stack. */ 525 if (tfb != NULL) { 526 /* Initialize the new stack. If it succeeds, we are done. */ 527 if (tfb->tfb_tcp_fb_init == NULL || 528 (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) { 529 /* Release the old stack */ 530 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 531 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 532 refcount_release(&tp->t_fb->tfb_refcnt); 533 /* Now set in all the pointers */ 534 tp->t_fb = tfb; 535 tp->t_fb_ptr = ptr; 536 return; 537 } 538 /* 539 * Initialization failed. Release the reference count on 540 * the looked up default stack. 541 */ 542 refcount_release(&tfb->tfb_refcnt); 543 } 544 545 /* 546 * If that wasn't feasible, use the built-in default 547 * stack which is not allowed to reject anyone. 548 */ 549 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 550 if (tfb == NULL) { 551 /* there always should be a default */ 552 panic("Can't refer to tcp_def_funcblk"); 553 } 554 if ((*tfb->tfb_tcp_handoff_ok)(tp)) { 555 /* The default stack cannot say no */ 556 panic("Default stack rejects a new session?"); 557 } 558 if (tfb->tfb_tcp_fb_init != NULL && 559 (*tfb->tfb_tcp_fb_init)(tp, &ptr)) { 560 /* The default stack cannot fail */ 561 panic("Default stack initialization failed"); 562 } 563 /* Now release the old stack */ 564 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 565 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 566 refcount_release(&tp->t_fb->tfb_refcnt); 567 /* And set in the pointers to the new */ 568 tp->t_fb = tfb; 569 tp->t_fb_ptr = ptr; 570 } 571 572 static bool 573 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 574 const struct sockaddr *sa, void *ctx) 575 { 576 struct ip *iph; 577 #ifdef INET6 578 struct ip6_hdr *ip6; 579 #endif 580 struct udphdr *uh; 581 struct tcphdr *th; 582 int thlen; 583 uint16_t port; 584 585 TCPSTAT_INC(tcps_tunneled_pkts); 586 if ((m->m_flags & M_PKTHDR) == 0) { 587 /* Can't handle one that is not a pkt hdr */ 588 TCPSTAT_INC(tcps_tunneled_errs); 589 goto out; 590 } 591 thlen = sizeof(struct tcphdr); 592 if (m->m_len < off + sizeof(struct udphdr) + thlen && 593 (m = m_pullup(m, off + sizeof(struct udphdr) + thlen)) == NULL) { 594 TCPSTAT_INC(tcps_tunneled_errs); 595 goto out; 596 } 597 iph = mtod(m, struct ip *); 598 uh = (struct udphdr *)((caddr_t)iph + off); 599 th = (struct tcphdr *)(uh + 1); 600 thlen = th->th_off << 2; 601 if (m->m_len < off + sizeof(struct udphdr) + thlen) { 602 m = m_pullup(m, off + sizeof(struct udphdr) + thlen); 603 if (m == NULL) { 604 TCPSTAT_INC(tcps_tunneled_errs); 605 goto out; 606 } else { 607 iph = mtod(m, struct ip *); 608 uh = (struct udphdr *)((caddr_t)iph + off); 609 th = (struct tcphdr *)(uh + 1); 610 } 611 } 612 m->m_pkthdr.tcp_tun_port = port = uh->uh_sport; 613 bcopy(th, uh, m->m_len - off); 614 m->m_len -= sizeof(struct udphdr); 615 m->m_pkthdr.len -= sizeof(struct udphdr); 616 /* 617 * We use the same algorithm for 618 * both UDP and TCP for c-sum. So 619 * the code in tcp_input will skip 620 * the checksum. So we do nothing 621 * with the flag (m->m_pkthdr.csum_flags). 622 */ 623 switch (iph->ip_v) { 624 #ifdef INET 625 case IPVERSION: 626 iph->ip_len = htons(ntohs(iph->ip_len) - sizeof(struct udphdr)); 627 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 628 break; 629 #endif 630 #ifdef INET6 631 case IPV6_VERSION >> 4: 632 ip6 = mtod(m, struct ip6_hdr *); 633 ip6->ip6_plen = htons(ntohs(ip6->ip6_plen) - sizeof(struct udphdr)); 634 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 635 break; 636 #endif 637 default: 638 goto out; 639 break; 640 } 641 return (true); 642 out: 643 m_freem(m); 644 645 return (true); 646 } 647 648 static int 649 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 650 { 651 int error = ENOENT; 652 struct tcp_function_set fs; 653 struct tcp_function_block *blk; 654 655 memset(&fs, 0, sizeof(fs)); 656 rw_rlock(&tcp_function_lock); 657 blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL); 658 if (blk) { 659 /* Found him */ 660 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 661 fs.pcbcnt = blk->tfb_refcnt; 662 } 663 rw_runlock(&tcp_function_lock); 664 error = sysctl_handle_string(oidp, fs.function_set_name, 665 sizeof(fs.function_set_name), req); 666 667 /* Check for error or no change */ 668 if (error != 0 || req->newptr == NULL) 669 return (error); 670 671 rw_wlock(&tcp_function_lock); 672 blk = find_tcp_functions_locked(&fs); 673 if ((blk == NULL) || 674 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 675 error = ENOENT; 676 goto done; 677 } 678 if ((blk->tfb_flags & TCP_FUNC_DEFAULT_OK) == 0) { 679 error = EINVAL; 680 goto done; 681 } 682 V_tcp_func_set_ptr = blk; 683 done: 684 rw_wunlock(&tcp_function_lock); 685 return (error); 686 } 687 688 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 689 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 690 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 691 "Set/get the default TCP functions"); 692 693 static int 694 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 695 { 696 int error, cnt, linesz; 697 struct tcp_function *f; 698 char *buffer, *cp; 699 size_t bufsz, outsz; 700 bool alias; 701 702 cnt = 0; 703 rw_rlock(&tcp_function_lock); 704 TAILQ_FOREACH(f, &t_functions, tf_next) { 705 cnt++; 706 } 707 rw_runlock(&tcp_function_lock); 708 709 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 710 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 711 712 error = 0; 713 cp = buffer; 714 715 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 716 "Alias", "PCB count"); 717 cp += linesz; 718 bufsz -= linesz; 719 outsz = linesz; 720 721 rw_rlock(&tcp_function_lock); 722 TAILQ_FOREACH(f, &t_functions, tf_next) { 723 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 724 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 725 f->tf_fb->tfb_tcp_block_name, 726 (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ', 727 alias ? f->tf_name : "-", 728 f->tf_fb->tfb_refcnt); 729 if (linesz >= bufsz) { 730 error = EOVERFLOW; 731 break; 732 } 733 cp += linesz; 734 bufsz -= linesz; 735 outsz += linesz; 736 } 737 rw_runlock(&tcp_function_lock); 738 if (error == 0) 739 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 740 free(buffer, M_TEMP); 741 return (error); 742 } 743 744 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 745 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 746 NULL, 0, sysctl_net_inet_list_available, "A", 747 "list available TCP Function sets"); 748 749 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 750 751 #ifdef INET 752 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 753 #define V_udp4_tun_socket VNET(udp4_tun_socket) 754 #endif 755 #ifdef INET6 756 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 757 #define V_udp6_tun_socket VNET(udp6_tun_socket) 758 #endif 759 760 static struct sx tcpoudp_lock; 761 762 static void 763 tcp_over_udp_stop(void) 764 { 765 766 sx_assert(&tcpoudp_lock, SA_XLOCKED); 767 768 #ifdef INET 769 if (V_udp4_tun_socket != NULL) { 770 soclose(V_udp4_tun_socket); 771 V_udp4_tun_socket = NULL; 772 } 773 #endif 774 #ifdef INET6 775 if (V_udp6_tun_socket != NULL) { 776 soclose(V_udp6_tun_socket); 777 V_udp6_tun_socket = NULL; 778 } 779 #endif 780 } 781 782 static int 783 tcp_over_udp_start(void) 784 { 785 uint16_t port; 786 int ret; 787 #ifdef INET 788 struct sockaddr_in sin; 789 #endif 790 #ifdef INET6 791 struct sockaddr_in6 sin6; 792 #endif 793 794 sx_assert(&tcpoudp_lock, SA_XLOCKED); 795 796 port = V_tcp_udp_tunneling_port; 797 if (ntohs(port) == 0) { 798 /* Must have a port set */ 799 return (EINVAL); 800 } 801 #ifdef INET 802 if (V_udp4_tun_socket != NULL) { 803 /* Already running -- must stop first */ 804 return (EALREADY); 805 } 806 #endif 807 #ifdef INET6 808 if (V_udp6_tun_socket != NULL) { 809 /* Already running -- must stop first */ 810 return (EALREADY); 811 } 812 #endif 813 #ifdef INET 814 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 815 SOCK_DGRAM, IPPROTO_UDP, 816 curthread->td_ucred, curthread))) { 817 tcp_over_udp_stop(); 818 return (ret); 819 } 820 /* Call the special UDP hook. */ 821 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 822 tcp_recv_udp_tunneled_packet, 823 tcp_ctlinput_viaudp, 824 NULL))) { 825 tcp_over_udp_stop(); 826 return (ret); 827 } 828 /* Ok, we have a socket, bind it to the port. */ 829 memset(&sin, 0, sizeof(struct sockaddr_in)); 830 sin.sin_len = sizeof(struct sockaddr_in); 831 sin.sin_family = AF_INET; 832 sin.sin_port = htons(port); 833 if ((ret = sobind(V_udp4_tun_socket, 834 (struct sockaddr *)&sin, curthread))) { 835 tcp_over_udp_stop(); 836 return (ret); 837 } 838 #endif 839 #ifdef INET6 840 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 841 SOCK_DGRAM, IPPROTO_UDP, 842 curthread->td_ucred, curthread))) { 843 tcp_over_udp_stop(); 844 return (ret); 845 } 846 /* Call the special UDP hook. */ 847 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 848 tcp_recv_udp_tunneled_packet, 849 tcp6_ctlinput_viaudp, 850 NULL))) { 851 tcp_over_udp_stop(); 852 return (ret); 853 } 854 /* Ok, we have a socket, bind it to the port. */ 855 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 856 sin6.sin6_len = sizeof(struct sockaddr_in6); 857 sin6.sin6_family = AF_INET6; 858 sin6.sin6_port = htons(port); 859 if ((ret = sobind(V_udp6_tun_socket, 860 (struct sockaddr *)&sin6, curthread))) { 861 tcp_over_udp_stop(); 862 return (ret); 863 } 864 #endif 865 return (0); 866 } 867 868 static int 869 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 870 { 871 int error; 872 uint32_t old, new; 873 874 old = V_tcp_udp_tunneling_port; 875 new = old; 876 error = sysctl_handle_int(oidp, &new, 0, req); 877 if ((error == 0) && 878 (req->newptr != NULL)) { 879 if ((new < TCP_TUNNELING_PORT_MIN) || 880 (new > TCP_TUNNELING_PORT_MAX)) { 881 error = EINVAL; 882 } else { 883 sx_xlock(&tcpoudp_lock); 884 V_tcp_udp_tunneling_port = new; 885 if (old != 0) { 886 tcp_over_udp_stop(); 887 } 888 if (new != 0) { 889 error = tcp_over_udp_start(); 890 if (error != 0) { 891 V_tcp_udp_tunneling_port = 0; 892 } 893 } 894 sx_xunlock(&tcpoudp_lock); 895 } 896 } 897 return (error); 898 } 899 900 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 901 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 902 &VNET_NAME(tcp_udp_tunneling_port), 903 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 904 "Tunneling port for tcp over udp"); 905 906 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 907 908 static int 909 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 910 { 911 int error, new; 912 913 new = V_tcp_udp_tunneling_overhead; 914 error = sysctl_handle_int(oidp, &new, 0, req); 915 if (error == 0 && req->newptr) { 916 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 917 (new > TCP_TUNNELING_OVERHEAD_MAX)) 918 error = EINVAL; 919 else 920 V_tcp_udp_tunneling_overhead = new; 921 } 922 return (error); 923 } 924 925 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 926 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 927 &VNET_NAME(tcp_udp_tunneling_overhead), 928 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 929 "MSS reduction when using tcp over udp"); 930 931 /* 932 * Exports one (struct tcp_function_info) for each alias/name. 933 */ 934 static int 935 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 936 { 937 int cnt, error; 938 struct tcp_function *f; 939 struct tcp_function_info tfi; 940 941 /* 942 * We don't allow writes. 943 */ 944 if (req->newptr != NULL) 945 return (EINVAL); 946 947 /* 948 * Wire the old buffer so we can directly copy the functions to 949 * user space without dropping the lock. 950 */ 951 if (req->oldptr != NULL) { 952 error = sysctl_wire_old_buffer(req, 0); 953 if (error) 954 return (error); 955 } 956 957 /* 958 * Walk the list and copy out matching entries. If INVARIANTS 959 * is compiled in, also walk the list to verify the length of 960 * the list matches what we have recorded. 961 */ 962 rw_rlock(&tcp_function_lock); 963 964 cnt = 0; 965 #ifndef INVARIANTS 966 if (req->oldptr == NULL) { 967 cnt = tcp_fb_cnt; 968 goto skip_loop; 969 } 970 #endif 971 TAILQ_FOREACH(f, &t_functions, tf_next) { 972 #ifdef INVARIANTS 973 cnt++; 974 #endif 975 if (req->oldptr != NULL) { 976 bzero(&tfi, sizeof(tfi)); 977 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 978 tfi.tfi_id = f->tf_fb->tfb_id; 979 (void)strlcpy(tfi.tfi_alias, f->tf_name, 980 sizeof(tfi.tfi_alias)); 981 (void)strlcpy(tfi.tfi_name, 982 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 983 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 984 /* 985 * Don't stop on error, as that is the 986 * mechanism we use to accumulate length 987 * information if the buffer was too short. 988 */ 989 } 990 } 991 KASSERT(cnt == tcp_fb_cnt, 992 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 993 #ifndef INVARIANTS 994 skip_loop: 995 #endif 996 rw_runlock(&tcp_function_lock); 997 if (req->oldptr == NULL) 998 error = SYSCTL_OUT(req, NULL, 999 (cnt + 1) * sizeof(struct tcp_function_info)); 1000 1001 return (error); 1002 } 1003 1004 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1005 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1006 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1007 "List TCP function block name-to-ID mappings"); 1008 1009 /* 1010 * tfb_tcp_handoff_ok() function for the default stack. 1011 * Note that we'll basically try to take all comers. 1012 */ 1013 static int 1014 tcp_default_handoff_ok(struct tcpcb *tp) 1015 { 1016 1017 return (0); 1018 } 1019 1020 /* 1021 * tfb_tcp_fb_init() function for the default stack. 1022 * 1023 * This handles making sure we have appropriate timers set if you are 1024 * transitioning a socket that has some amount of setup done. 1025 * 1026 * The init() fuction from the default can *never* return non-zero i.e. 1027 * it is required to always succeed since it is the stack of last resort! 1028 */ 1029 static int 1030 tcp_default_fb_init(struct tcpcb *tp, void **ptr) 1031 { 1032 struct socket *so = tptosocket(tp); 1033 int rexmt; 1034 1035 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1036 /* We don't use the pointer */ 1037 *ptr = NULL; 1038 1039 KASSERT(tp->t_state < TCPS_TIME_WAIT, 1040 ("%s: connection %p in unexpected state %d", __func__, tp, 1041 tp->t_state)); 1042 1043 /* Make sure we get no interesting mbuf queuing behavior */ 1044 /* All mbuf queue/ack compress flags should be off */ 1045 tcp_lro_features_off(tp); 1046 1047 /* Cancel the GP measurement in progress */ 1048 tp->t_flags &= ~TF_GPUTINPROG; 1049 /* Validate the timers are not in usec, if they are convert */ 1050 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 1051 if ((tp->t_state == TCPS_SYN_SENT) || 1052 (tp->t_state == TCPS_SYN_RECEIVED)) 1053 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift]; 1054 else 1055 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 1056 if (tp->t_rxtshift == 0) 1057 tp->t_rxtcur = rexmt; 1058 else 1059 TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, TCPTV_REXMTMAX); 1060 1061 /* 1062 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1063 * know what to do for unexpected states (which includes TIME_WAIT). 1064 */ 1065 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1066 return (0); 1067 1068 /* 1069 * Make sure some kind of transmission timer is set if there is 1070 * outstanding data. 1071 */ 1072 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1073 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1074 tcp_timer_active(tp, TT_PERSIST))) { 1075 /* 1076 * If the session has established and it looks like it should 1077 * be in the persist state, set the persist timer. Otherwise, 1078 * set the retransmit timer. 1079 */ 1080 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1081 (int32_t)(tp->snd_nxt - tp->snd_una) < 1082 (int32_t)sbavail(&so->so_snd)) 1083 tcp_setpersist(tp); 1084 else 1085 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 1086 } 1087 1088 /* All non-embryonic sessions get a keepalive timer. */ 1089 if (!tcp_timer_active(tp, TT_KEEP)) 1090 tcp_timer_activate(tp, TT_KEEP, 1091 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1092 TP_KEEPINIT(tp)); 1093 1094 /* 1095 * Make sure critical variables are initialized 1096 * if transitioning while in Recovery. 1097 */ 1098 if IN_FASTRECOVERY(tp->t_flags) { 1099 if (tp->sackhint.recover_fs == 0) 1100 tp->sackhint.recover_fs = max(1, 1101 tp->snd_nxt - tp->snd_una); 1102 } 1103 1104 return (0); 1105 } 1106 1107 /* 1108 * tfb_tcp_fb_fini() function for the default stack. 1109 * 1110 * This changes state as necessary (or prudent) to prepare for another stack 1111 * to assume responsibility for the connection. 1112 */ 1113 static void 1114 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1115 { 1116 1117 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1118 1119 #ifdef TCP_BLACKBOX 1120 tcp_log_flowend(tp); 1121 #endif 1122 tp->t_acktime = 0; 1123 return; 1124 } 1125 1126 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1127 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1128 1129 static struct mtx isn_mtx; 1130 1131 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1132 #define ISN_LOCK() mtx_lock(&isn_mtx) 1133 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1134 1135 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcp", "tcphash"); 1136 1137 /* 1138 * Take a value and get the next power of 2 that doesn't overflow. 1139 * Used to size the tcp_inpcb hash buckets. 1140 */ 1141 static int 1142 maketcp_hashsize(int size) 1143 { 1144 int hashsize; 1145 1146 /* 1147 * auto tune. 1148 * get the next power of 2 higher than maxsockets. 1149 */ 1150 hashsize = 1 << fls(size); 1151 /* catch overflow, and just go one power of 2 smaller */ 1152 if (hashsize < size) { 1153 hashsize = 1 << (fls(size) - 1); 1154 } 1155 return (hashsize); 1156 } 1157 1158 static volatile int next_tcp_stack_id = 1; 1159 1160 /* 1161 * Register a TCP function block with the name provided in the names 1162 * array. (Note that this function does NOT automatically register 1163 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1164 * explicitly include blk->tfb_tcp_block_name in the list of names if 1165 * you wish to register the stack with that name.) 1166 * 1167 * Either all name registrations will succeed or all will fail. If 1168 * a name registration fails, the function will update the num_names 1169 * argument to point to the array index of the name that encountered 1170 * the failure. 1171 * 1172 * Returns 0 on success, or an error code on failure. 1173 */ 1174 int 1175 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1176 const char *names[], int *num_names) 1177 { 1178 struct tcp_function *f[TCP_FUNCTION_NAME_NUM_MAX]; 1179 struct tcp_function_set fs; 1180 int error, i, num_registered; 1181 1182 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1183 KASSERT(*num_names > 0, 1184 ("%s: Called with non-positive length of name list", __func__)); 1185 KASSERT(rw_initialized(&tcp_function_lock), 1186 ("%s: called too early", __func__)); 1187 1188 if (*num_names > TCP_FUNCTION_NAME_NUM_MAX) { 1189 /* Too many names. */ 1190 *num_names = 0; 1191 return (E2BIG); 1192 } 1193 if ((blk->tfb_tcp_output == NULL) || 1194 (blk->tfb_tcp_do_segment == NULL) || 1195 (blk->tfb_tcp_ctloutput == NULL) || 1196 (blk->tfb_tcp_handoff_ok == NULL) || 1197 (strlen(blk->tfb_tcp_block_name) == 0)) { 1198 /* These functions are required and a name is needed. */ 1199 *num_names = 0; 1200 return (EINVAL); 1201 } 1202 1203 for (i = 0; i < *num_names; i++) { 1204 f[i] = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1205 if (f[i] == NULL) { 1206 while (--i >= 0) 1207 free(f[i], M_TCPFUNCTIONS); 1208 *num_names = 0; 1209 return (ENOMEM); 1210 } 1211 } 1212 1213 num_registered = 0; 1214 rw_wlock(&tcp_function_lock); 1215 if (find_tcp_fb_locked(blk, NULL) != NULL) { 1216 /* A TCP function block can only be registered once. */ 1217 error = EALREADY; 1218 goto cleanup; 1219 } 1220 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1221 error = EINVAL; 1222 goto cleanup; 1223 } 1224 refcount_init(&blk->tfb_refcnt, 0); 1225 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1226 for (i = 0; i < *num_names; i++) { 1227 (void)strlcpy(fs.function_set_name, names[i], 1228 sizeof(fs.function_set_name)); 1229 if (find_tcp_functions_locked(&fs) != NULL) { 1230 /* Duplicate name space not allowed */ 1231 error = EALREADY; 1232 goto cleanup; 1233 } 1234 f[i]->tf_fb = blk; 1235 (void)strlcpy(f[i]->tf_name, names[i], sizeof(f[i]->tf_name)); 1236 TAILQ_INSERT_TAIL(&t_functions, f[i], tf_next); 1237 tcp_fb_cnt++; 1238 num_registered++; 1239 } 1240 rw_wunlock(&tcp_function_lock); 1241 return (0); 1242 1243 cleanup: 1244 /* Remove the entries just added. */ 1245 for (i = 0; i < *num_names; i++) { 1246 if (i < num_registered) { 1247 TAILQ_REMOVE(&t_functions, f[i], tf_next); 1248 tcp_fb_cnt--; 1249 } 1250 f[i]->tf_fb = NULL; 1251 free(f[i], M_TCPFUNCTIONS); 1252 } 1253 rw_wunlock(&tcp_function_lock); 1254 *num_names = num_registered; 1255 return (error); 1256 } 1257 1258 /* 1259 * Register a TCP function block using the name provided in the name 1260 * argument. 1261 * 1262 * Returns 0 on success, or an error code on failure. 1263 */ 1264 int 1265 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1266 int wait) 1267 { 1268 const char *name_list[1]; 1269 int num_names, rv; 1270 1271 num_names = 1; 1272 if (name != NULL) 1273 name_list[0] = name; 1274 else 1275 name_list[0] = blk->tfb_tcp_block_name; 1276 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1277 return (rv); 1278 } 1279 1280 /* 1281 * Register a TCP function block using the name defined in 1282 * blk->tfb_tcp_block_name. 1283 * 1284 * Returns 0 on success, or an error code on failure. 1285 */ 1286 int 1287 register_tcp_functions(struct tcp_function_block *blk, int wait) 1288 { 1289 1290 return (register_tcp_functions_as_name(blk, NULL, wait)); 1291 } 1292 1293 /* 1294 * Deregister all names associated with a function block. This 1295 * functionally removes the function block from use within the system. 1296 * 1297 * When called with a true quiesce argument, mark the function block 1298 * as being removed so no more stacks will use it and determine 1299 * whether the removal would succeed. 1300 * 1301 * When called with a false quiesce argument, actually attempt the 1302 * removal. 1303 * 1304 * When called with a force argument, attempt to switch all TCBs to 1305 * use the default stack instead of returning EBUSY. 1306 * 1307 * Returns 0 on success (or if the removal would succeed), or an error 1308 * code on failure. 1309 */ 1310 int 1311 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1312 bool force) 1313 { 1314 struct tcp_function *f; 1315 VNET_ITERATOR_DECL(vnet_iter); 1316 1317 if (blk == &tcp_def_funcblk) { 1318 /* You can't un-register the default */ 1319 return (EPERM); 1320 } 1321 rw_wlock(&tcp_function_lock); 1322 VNET_LIST_RLOCK_NOSLEEP(); 1323 VNET_FOREACH(vnet_iter) { 1324 CURVNET_SET(vnet_iter); 1325 if (blk == V_tcp_func_set_ptr) { 1326 /* You can't free the current default in some vnet. */ 1327 CURVNET_RESTORE(); 1328 VNET_LIST_RUNLOCK_NOSLEEP(); 1329 rw_wunlock(&tcp_function_lock); 1330 return (EBUSY); 1331 } 1332 CURVNET_RESTORE(); 1333 } 1334 VNET_LIST_RUNLOCK_NOSLEEP(); 1335 /* Mark the block so no more stacks can use it. */ 1336 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1337 /* 1338 * If TCBs are still attached to the stack, attempt to switch them 1339 * to the default stack. 1340 */ 1341 if (force && blk->tfb_refcnt) { 1342 struct inpcb *inp; 1343 struct tcpcb *tp; 1344 VNET_ITERATOR_DECL(vnet_iter); 1345 1346 rw_wunlock(&tcp_function_lock); 1347 1348 VNET_LIST_RLOCK(); 1349 VNET_FOREACH(vnet_iter) { 1350 CURVNET_SET(vnet_iter); 1351 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1352 INPLOOKUP_WLOCKPCB); 1353 1354 while ((inp = inp_next(&inpi)) != NULL) { 1355 tp = intotcpcb(inp); 1356 if (tp == NULL || tp->t_fb != blk) 1357 continue; 1358 tcp_switch_back_to_default(tp); 1359 } 1360 CURVNET_RESTORE(); 1361 } 1362 VNET_LIST_RUNLOCK(); 1363 1364 rw_wlock(&tcp_function_lock); 1365 } 1366 if (blk->tfb_refcnt) { 1367 /* TCBs still attached. */ 1368 rw_wunlock(&tcp_function_lock); 1369 return (EBUSY); 1370 } 1371 if (quiesce) { 1372 /* Skip removal. */ 1373 rw_wunlock(&tcp_function_lock); 1374 return (0); 1375 } 1376 /* Remove any function names that map to this function block. */ 1377 while (find_tcp_fb_locked(blk, &f) != NULL) { 1378 TAILQ_REMOVE(&t_functions, f, tf_next); 1379 tcp_fb_cnt--; 1380 f->tf_fb = NULL; 1381 free(f, M_TCPFUNCTIONS); 1382 } 1383 rw_wunlock(&tcp_function_lock); 1384 return (0); 1385 } 1386 1387 static void 1388 tcp_drain(void) 1389 { 1390 struct epoch_tracker et; 1391 VNET_ITERATOR_DECL(vnet_iter); 1392 1393 if (!do_tcpdrain) 1394 return; 1395 1396 NET_EPOCH_ENTER(et); 1397 VNET_LIST_RLOCK_NOSLEEP(); 1398 VNET_FOREACH(vnet_iter) { 1399 CURVNET_SET(vnet_iter); 1400 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1401 INPLOOKUP_WLOCKPCB); 1402 struct inpcb *inpb; 1403 struct tcpcb *tcpb; 1404 1405 /* 1406 * Walk the tcpbs, if existing, and flush the reassembly queue, 1407 * if there is one... 1408 * XXX: The "Net/3" implementation doesn't imply that the TCP 1409 * reassembly queue should be flushed, but in a situation 1410 * where we're really low on mbufs, this is potentially 1411 * useful. 1412 */ 1413 while ((inpb = inp_next(&inpi)) != NULL) { 1414 if ((tcpb = intotcpcb(inpb)) != NULL) { 1415 tcp_reass_flush(tcpb); 1416 tcp_clean_sackreport(tcpb); 1417 #ifdef TCP_BLACKBOX 1418 tcp_log_drain(tcpb); 1419 #endif 1420 #ifdef TCPPCAP 1421 if (tcp_pcap_aggressive_free) { 1422 /* Free the TCP PCAP queues. */ 1423 tcp_pcap_drain(&(tcpb->t_inpkts)); 1424 tcp_pcap_drain(&(tcpb->t_outpkts)); 1425 } 1426 #endif 1427 } 1428 } 1429 CURVNET_RESTORE(); 1430 } 1431 VNET_LIST_RUNLOCK_NOSLEEP(); 1432 NET_EPOCH_EXIT(et); 1433 } 1434 1435 static void 1436 tcp_vnet_init(void *arg __unused) 1437 { 1438 1439 #ifdef TCP_HHOOK 1440 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1441 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1442 printf("%s: WARNING: unable to register helper hook\n", __func__); 1443 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1444 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1445 printf("%s: WARNING: unable to register helper hook\n", __func__); 1446 #endif 1447 #ifdef STATS 1448 if (tcp_stats_init()) 1449 printf("%s: WARNING: unable to initialise TCP stats\n", 1450 __func__); 1451 #endif 1452 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1453 tcp_tcbhashsize); 1454 1455 syncache_init(); 1456 tcp_hc_init(); 1457 1458 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1459 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1460 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1461 1462 tcp_fastopen_init(); 1463 1464 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1465 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1466 1467 V_tcp_msl = TCPTV_MSL; 1468 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1469 } 1470 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1471 tcp_vnet_init, NULL); 1472 1473 static void 1474 tcp_init(void *arg __unused) 1475 { 1476 int hashsize; 1477 1478 tcp_reass_global_init(); 1479 1480 /* XXX virtualize those below? */ 1481 tcp_delacktime = TCPTV_DELACK; 1482 tcp_keepinit = TCPTV_KEEP_INIT; 1483 tcp_keepidle = TCPTV_KEEP_IDLE; 1484 tcp_keepintvl = TCPTV_KEEPINTVL; 1485 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1486 tcp_rexmit_initial = TCPTV_RTOBASE; 1487 if (tcp_rexmit_initial < 1) 1488 tcp_rexmit_initial = 1; 1489 tcp_rexmit_min = TCPTV_MIN; 1490 if (tcp_rexmit_min < 1) 1491 tcp_rexmit_min = 1; 1492 tcp_persmin = TCPTV_PERSMIN; 1493 tcp_persmax = TCPTV_PERSMAX; 1494 tcp_rexmit_slop = TCPTV_CPU_VAR; 1495 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1496 1497 /* Setup the tcp function block list */ 1498 TAILQ_INIT(&t_functions); 1499 rw_init(&tcp_function_lock, "tcp_func_lock"); 1500 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1501 sx_init(&tcpoudp_lock, "TCP over UDP configuration"); 1502 #ifdef TCP_BLACKBOX 1503 /* Initialize the TCP logging data. */ 1504 tcp_log_init(); 1505 #endif 1506 1507 if (tcp_soreceive_stream) { 1508 #ifdef INET 1509 tcp_protosw.pr_soreceive = soreceive_stream; 1510 #endif 1511 #ifdef INET6 1512 tcp6_protosw.pr_soreceive = soreceive_stream; 1513 #endif /* INET6 */ 1514 } 1515 1516 #ifdef INET6 1517 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1518 #else /* INET6 */ 1519 max_protohdr_grow(sizeof(struct tcpiphdr)); 1520 #endif /* INET6 */ 1521 1522 ISN_LOCK_INIT(); 1523 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1524 SHUTDOWN_PRI_DEFAULT); 1525 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1526 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1527 1528 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1529 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1530 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1531 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1532 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1533 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1534 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1535 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1536 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1537 tcp_pacing_failures = counter_u64_alloc(M_WAITOK); 1538 tcp_dgp_failures = counter_u64_alloc(M_WAITOK); 1539 #ifdef TCPPCAP 1540 tcp_pcap_init(); 1541 #endif 1542 1543 hashsize = tcp_tcbhashsize; 1544 if (hashsize == 0) { 1545 /* 1546 * Auto tune the hash size based on maxsockets. 1547 * A perfect hash would have a 1:1 mapping 1548 * (hashsize = maxsockets) however it's been 1549 * suggested that O(2) average is better. 1550 */ 1551 hashsize = maketcp_hashsize(maxsockets / 4); 1552 /* 1553 * Our historical default is 512, 1554 * do not autotune lower than this. 1555 */ 1556 if (hashsize < 512) 1557 hashsize = 512; 1558 if (bootverbose) 1559 printf("%s: %s auto tuned to %d\n", __func__, 1560 "net.inet.tcp.tcbhashsize", hashsize); 1561 } 1562 /* 1563 * We require a hashsize to be a power of two. 1564 * Previously if it was not a power of two we would just reset it 1565 * back to 512, which could be a nasty surprise if you did not notice 1566 * the error message. 1567 * Instead what we do is clip it to the closest power of two lower 1568 * than the specified hash value. 1569 */ 1570 if (!powerof2(hashsize)) { 1571 int oldhashsize = hashsize; 1572 1573 hashsize = maketcp_hashsize(hashsize); 1574 /* prevent absurdly low value */ 1575 if (hashsize < 16) 1576 hashsize = 16; 1577 printf("%s: WARNING: TCB hash size not a power of 2, " 1578 "clipped from %d to %d.\n", __func__, oldhashsize, 1579 hashsize); 1580 } 1581 tcp_tcbhashsize = hashsize; 1582 1583 #ifdef INET 1584 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1585 #endif 1586 #ifdef INET6 1587 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1588 #endif 1589 } 1590 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1591 1592 #ifdef VIMAGE 1593 static void 1594 tcp_destroy(void *unused __unused) 1595 { 1596 int n; 1597 #ifdef TCP_HHOOK 1598 int error; 1599 #endif 1600 1601 /* 1602 * All our processes are gone, all our sockets should be cleaned 1603 * up, which means, we should be past the tcp_discardcb() calls. 1604 * Sleep to let all tcpcb timers really disappear and cleanup. 1605 */ 1606 for (;;) { 1607 INP_INFO_WLOCK(&V_tcbinfo); 1608 n = V_tcbinfo.ipi_count; 1609 INP_INFO_WUNLOCK(&V_tcbinfo); 1610 if (n == 0) 1611 break; 1612 pause("tcpdes", hz / 10); 1613 } 1614 tcp_hc_destroy(); 1615 syncache_destroy(); 1616 in_pcbinfo_destroy(&V_tcbinfo); 1617 /* tcp_discardcb() clears the sack_holes up. */ 1618 uma_zdestroy(V_sack_hole_zone); 1619 1620 /* 1621 * Cannot free the zone until all tcpcbs are released as we attach 1622 * the allocations to them. 1623 */ 1624 tcp_fastopen_destroy(); 1625 1626 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1627 VNET_PCPUSTAT_FREE(tcpstat); 1628 1629 #ifdef TCP_HHOOK 1630 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1631 if (error != 0) { 1632 printf("%s: WARNING: unable to deregister helper hook " 1633 "type=%d, id=%d: error %d returned\n", __func__, 1634 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1635 } 1636 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1637 if (error != 0) { 1638 printf("%s: WARNING: unable to deregister helper hook " 1639 "type=%d, id=%d: error %d returned\n", __func__, 1640 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1641 } 1642 #endif 1643 } 1644 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1645 #endif 1646 1647 void 1648 tcp_fini(void *xtp) 1649 { 1650 1651 } 1652 1653 /* 1654 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1655 * tcp_template used to store this data in mbufs, but we now recopy it out 1656 * of the tcpcb each time to conserve mbufs. 1657 */ 1658 void 1659 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1660 { 1661 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1662 1663 INP_WLOCK_ASSERT(inp); 1664 1665 #ifdef INET6 1666 if ((inp->inp_vflag & INP_IPV6) != 0) { 1667 struct ip6_hdr *ip6; 1668 1669 ip6 = (struct ip6_hdr *)ip_ptr; 1670 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1671 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1672 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1673 (IPV6_VERSION & IPV6_VERSION_MASK); 1674 if (port == 0) 1675 ip6->ip6_nxt = IPPROTO_TCP; 1676 else 1677 ip6->ip6_nxt = IPPROTO_UDP; 1678 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1679 ip6->ip6_src = inp->in6p_laddr; 1680 ip6->ip6_dst = inp->in6p_faddr; 1681 } 1682 #endif /* INET6 */ 1683 #if defined(INET6) && defined(INET) 1684 else 1685 #endif 1686 #ifdef INET 1687 { 1688 struct ip *ip; 1689 1690 ip = (struct ip *)ip_ptr; 1691 ip->ip_v = IPVERSION; 1692 ip->ip_hl = 5; 1693 ip->ip_tos = inp->inp_ip_tos; 1694 ip->ip_len = 0; 1695 ip->ip_id = 0; 1696 ip->ip_off = 0; 1697 ip->ip_ttl = inp->inp_ip_ttl; 1698 ip->ip_sum = 0; 1699 if (port == 0) 1700 ip->ip_p = IPPROTO_TCP; 1701 else 1702 ip->ip_p = IPPROTO_UDP; 1703 ip->ip_src = inp->inp_laddr; 1704 ip->ip_dst = inp->inp_faddr; 1705 } 1706 #endif /* INET */ 1707 th->th_sport = inp->inp_lport; 1708 th->th_dport = inp->inp_fport; 1709 th->th_seq = 0; 1710 th->th_ack = 0; 1711 th->th_off = 5; 1712 tcp_set_flags(th, 0); 1713 th->th_win = 0; 1714 th->th_urp = 0; 1715 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1716 } 1717 1718 /* 1719 * Create template to be used to send tcp packets on a connection. 1720 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1721 * use for this function is in keepalives, which use tcp_respond. 1722 */ 1723 struct tcptemp * 1724 tcpip_maketemplate(struct inpcb *inp) 1725 { 1726 struct tcptemp *t; 1727 1728 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1729 if (t == NULL) 1730 return (NULL); 1731 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1732 return (t); 1733 } 1734 1735 /* 1736 * Send a single message to the TCP at address specified by 1737 * the given TCP/IP header. If m == NULL, then we make a copy 1738 * of the tcpiphdr at th and send directly to the addressed host. 1739 * This is used to force keep alive messages out using the TCP 1740 * template for a connection. If flags are given then we send 1741 * a message back to the TCP which originated the segment th, 1742 * and discard the mbuf containing it and any other attached mbufs. 1743 * 1744 * In any case the ack and sequence number of the transmitted 1745 * segment are as specified by the parameters. 1746 * 1747 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1748 */ 1749 1750 void 1751 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1752 tcp_seq ack, tcp_seq seq, uint16_t flags) 1753 { 1754 struct tcpopt to; 1755 struct inpcb *inp; 1756 struct ip *ip; 1757 struct mbuf *optm; 1758 struct udphdr *uh = NULL; 1759 struct tcphdr *nth; 1760 struct tcp_log_buffer *lgb; 1761 u_char *optp; 1762 #ifdef INET6 1763 struct ip6_hdr *ip6; 1764 int isipv6; 1765 #endif /* INET6 */ 1766 int optlen, tlen, win, ulen; 1767 int ect = 0; 1768 bool incl_opts; 1769 uint16_t port; 1770 int output_ret; 1771 #ifdef INVARIANTS 1772 int thflags = tcp_get_flags(th); 1773 #endif 1774 1775 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1776 NET_EPOCH_ASSERT(); 1777 1778 #ifdef INET6 1779 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1780 ip6 = ipgen; 1781 #endif /* INET6 */ 1782 ip = ipgen; 1783 1784 if (tp != NULL) { 1785 inp = tptoinpcb(tp); 1786 INP_LOCK_ASSERT(inp); 1787 } else 1788 inp = NULL; 1789 1790 if (m != NULL) { 1791 #ifdef INET6 1792 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1793 port = m->m_pkthdr.tcp_tun_port; 1794 else 1795 #endif 1796 if (ip && (ip->ip_p == IPPROTO_UDP)) 1797 port = m->m_pkthdr.tcp_tun_port; 1798 else 1799 port = 0; 1800 } else 1801 port = tp->t_port; 1802 1803 incl_opts = false; 1804 win = 0; 1805 if (tp != NULL) { 1806 if (!(flags & TH_RST)) { 1807 win = sbspace(&inp->inp_socket->so_rcv); 1808 if (win > TCP_MAXWIN << tp->rcv_scale) 1809 win = TCP_MAXWIN << tp->rcv_scale; 1810 } 1811 if ((tp->t_flags & TF_NOOPT) == 0) 1812 incl_opts = true; 1813 } 1814 if (m == NULL) { 1815 m = m_gethdr(M_NOWAIT, MT_DATA); 1816 if (m == NULL) 1817 return; 1818 m->m_data += max_linkhdr; 1819 #ifdef INET6 1820 if (isipv6) { 1821 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1822 sizeof(struct ip6_hdr)); 1823 ip6 = mtod(m, struct ip6_hdr *); 1824 nth = (struct tcphdr *)(ip6 + 1); 1825 if (port) { 1826 /* Insert a UDP header */ 1827 uh = (struct udphdr *)nth; 1828 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1829 uh->uh_dport = port; 1830 nth = (struct tcphdr *)(uh + 1); 1831 } 1832 } else 1833 #endif /* INET6 */ 1834 { 1835 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1836 ip = mtod(m, struct ip *); 1837 nth = (struct tcphdr *)(ip + 1); 1838 if (port) { 1839 /* Insert a UDP header */ 1840 uh = (struct udphdr *)nth; 1841 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1842 uh->uh_dport = port; 1843 nth = (struct tcphdr *)(uh + 1); 1844 } 1845 } 1846 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1847 flags = TH_ACK; 1848 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1849 struct mbuf *n; 1850 1851 /* Can't reuse 'm', allocate a new mbuf. */ 1852 n = m_gethdr(M_NOWAIT, MT_DATA); 1853 if (n == NULL) { 1854 m_freem(m); 1855 return; 1856 } 1857 1858 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1859 m_freem(m); 1860 m_freem(n); 1861 return; 1862 } 1863 1864 n->m_data += max_linkhdr; 1865 /* m_len is set later */ 1866 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1867 #ifdef INET6 1868 if (isipv6) { 1869 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1870 sizeof(struct ip6_hdr)); 1871 ip6 = mtod(n, struct ip6_hdr *); 1872 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1873 nth = (struct tcphdr *)(ip6 + 1); 1874 if (port) { 1875 /* Insert a UDP header */ 1876 uh = (struct udphdr *)nth; 1877 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1878 uh->uh_dport = port; 1879 nth = (struct tcphdr *)(uh + 1); 1880 } 1881 } else 1882 #endif /* INET6 */ 1883 { 1884 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1885 ip = mtod(n, struct ip *); 1886 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1887 nth = (struct tcphdr *)(ip + 1); 1888 if (port) { 1889 /* Insert a UDP header */ 1890 uh = (struct udphdr *)nth; 1891 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1892 uh->uh_dport = port; 1893 nth = (struct tcphdr *)(uh + 1); 1894 } 1895 } 1896 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1897 xchg(nth->th_dport, nth->th_sport, uint16_t); 1898 th = nth; 1899 m_freem(m); 1900 m = n; 1901 } else { 1902 /* 1903 * reuse the mbuf. 1904 * XXX MRT We inherit the FIB, which is lucky. 1905 */ 1906 m_freem(m->m_next); 1907 m->m_next = NULL; 1908 m->m_data = (caddr_t)ipgen; 1909 /* clear any receive flags for proper bpf timestamping */ 1910 m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO); 1911 /* m_len is set later */ 1912 #ifdef INET6 1913 if (isipv6) { 1914 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1915 nth = (struct tcphdr *)(ip6 + 1); 1916 } else 1917 #endif /* INET6 */ 1918 { 1919 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1920 nth = (struct tcphdr *)(ip + 1); 1921 } 1922 if (th != nth) { 1923 /* 1924 * this is usually a case when an extension header 1925 * exists between the IPv6 header and the 1926 * TCP header. 1927 */ 1928 nth->th_sport = th->th_sport; 1929 nth->th_dport = th->th_dport; 1930 } 1931 xchg(nth->th_dport, nth->th_sport, uint16_t); 1932 #undef xchg 1933 } 1934 tlen = 0; 1935 #ifdef INET6 1936 if (isipv6) 1937 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1938 #endif 1939 #if defined(INET) && defined(INET6) 1940 else 1941 #endif 1942 #ifdef INET 1943 tlen = sizeof (struct tcpiphdr); 1944 #endif 1945 if (port) 1946 tlen += sizeof (struct udphdr); 1947 #ifdef INVARIANTS 1948 m->m_len = 0; 1949 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1950 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1951 m, tlen, (long)M_TRAILINGSPACE(m))); 1952 #endif 1953 m->m_len = tlen; 1954 to.to_flags = 0; 1955 if (incl_opts) { 1956 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1957 /* Make sure we have room. */ 1958 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1959 m->m_next = m_get(M_NOWAIT, MT_DATA); 1960 if (m->m_next) { 1961 optp = mtod(m->m_next, u_char *); 1962 optm = m->m_next; 1963 } else 1964 incl_opts = false; 1965 } else { 1966 optp = (u_char *) (nth + 1); 1967 optm = m; 1968 } 1969 } 1970 if (incl_opts) { 1971 /* Timestamps. */ 1972 if (tp->t_flags & TF_RCVD_TSTMP) { 1973 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1974 to.to_tsecr = tp->ts_recent; 1975 to.to_flags |= TOF_TS; 1976 } 1977 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1978 /* TCP-MD5 (RFC2385). */ 1979 if (tp->t_flags & TF_SIGNATURE) 1980 to.to_flags |= TOF_SIGNATURE; 1981 #endif 1982 /* Add the options. */ 1983 tlen += optlen = tcp_addoptions(&to, optp); 1984 1985 /* Update m_len in the correct mbuf. */ 1986 optm->m_len += optlen; 1987 } else 1988 optlen = 0; 1989 #ifdef INET6 1990 if (isipv6) { 1991 if (uh) { 1992 ulen = tlen - sizeof(struct ip6_hdr); 1993 uh->uh_ulen = htons(ulen); 1994 } 1995 ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN); 1996 ip6->ip6_vfc = IPV6_VERSION; 1997 if (port) 1998 ip6->ip6_nxt = IPPROTO_UDP; 1999 else 2000 ip6->ip6_nxt = IPPROTO_TCP; 2001 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 2002 } 2003 #endif 2004 #if defined(INET) && defined(INET6) 2005 else 2006 #endif 2007 #ifdef INET 2008 { 2009 if (uh) { 2010 ulen = tlen - sizeof(struct ip); 2011 uh->uh_ulen = htons(ulen); 2012 } 2013 ip->ip_len = htons(tlen); 2014 if (inp != NULL) { 2015 ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK; 2016 ip->ip_ttl = inp->inp_ip_ttl; 2017 } else { 2018 ip->ip_tos = 0; 2019 ip->ip_ttl = V_ip_defttl; 2020 } 2021 ip->ip_tos |= ect; 2022 if (port) { 2023 ip->ip_p = IPPROTO_UDP; 2024 } else { 2025 ip->ip_p = IPPROTO_TCP; 2026 } 2027 if (V_path_mtu_discovery) 2028 ip->ip_off |= htons(IP_DF); 2029 } 2030 #endif 2031 m->m_pkthdr.len = tlen; 2032 m->m_pkthdr.rcvif = NULL; 2033 #ifdef MAC 2034 if (inp != NULL) { 2035 /* 2036 * Packet is associated with a socket, so allow the 2037 * label of the response to reflect the socket label. 2038 */ 2039 INP_LOCK_ASSERT(inp); 2040 mac_inpcb_create_mbuf(inp, m); 2041 } else { 2042 /* 2043 * Packet is not associated with a socket, so possibly 2044 * update the label in place. 2045 */ 2046 mac_netinet_tcp_reply(m); 2047 } 2048 #endif 2049 nth->th_seq = htonl(seq); 2050 nth->th_ack = htonl(ack); 2051 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2052 tcp_set_flags(nth, flags); 2053 if (tp && (flags & TH_RST)) { 2054 /* Log the reset */ 2055 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2056 } 2057 if (tp != NULL) 2058 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2059 else 2060 nth->th_win = htons((u_short)win); 2061 nth->th_urp = 0; 2062 2063 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2064 if (to.to_flags & TOF_SIGNATURE) { 2065 if (!TCPMD5_ENABLED() || 2066 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2067 m_freem(m); 2068 return; 2069 } 2070 } 2071 #endif 2072 2073 #ifdef INET6 2074 if (isipv6) { 2075 if (port) { 2076 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2077 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2078 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2079 nth->th_sum = 0; 2080 } else { 2081 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2082 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2083 nth->th_sum = in6_cksum_pseudo(ip6, 2084 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2085 } 2086 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2087 } 2088 #endif /* INET6 */ 2089 #if defined(INET6) && defined(INET) 2090 else 2091 #endif 2092 #ifdef INET 2093 { 2094 if (port) { 2095 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2096 htons(ulen + IPPROTO_UDP)); 2097 m->m_pkthdr.csum_flags = CSUM_UDP; 2098 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2099 nth->th_sum = 0; 2100 } else { 2101 m->m_pkthdr.csum_flags = CSUM_TCP; 2102 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2103 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2104 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2105 } 2106 } 2107 #endif /* INET */ 2108 TCP_PROBE3(debug__output, tp, th, m); 2109 if (flags & TH_RST) 2110 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2111 lgb = NULL; 2112 if ((tp != NULL) && tcp_bblogging_on(tp)) { 2113 if (INP_WLOCKED(inp)) { 2114 union tcp_log_stackspecific log; 2115 struct timeval tv; 2116 2117 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 2118 log.u_bbr.inhpts = tcp_in_hpts(tp); 2119 log.u_bbr.flex8 = 4; 2120 log.u_bbr.pkts_out = tp->t_maxseg; 2121 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2122 log.u_bbr.delivered = 0; 2123 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT, 2124 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2125 } else { 2126 /* 2127 * We can not log the packet, since we only own the 2128 * read lock, but a write lock is needed. The read lock 2129 * is not upgraded to a write lock, since only getting 2130 * the read lock was done intentionally to improve the 2131 * handling of SYN flooding attacks. 2132 * This happens only for pure SYN segments received in 2133 * the initial CLOSED state, or received in a more 2134 * advanced state than listen and the UDP encapsulation 2135 * port is unexpected. 2136 * The incoming SYN segments do not really belong to 2137 * the TCP connection and the handling does not change 2138 * the state of the TCP connection. Therefore, the 2139 * sending of the RST segments is not logged. Please 2140 * note that also the incoming SYN segments are not 2141 * logged. 2142 * 2143 * The following code ensures that the above description 2144 * is and stays correct. 2145 */ 2146 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2147 (tp->t_state == TCPS_CLOSED || 2148 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2149 ("%s: Logging of TCP segment with flags 0x%b and " 2150 "UDP encapsulation port %u skipped in state %s", 2151 __func__, thflags, PRINT_TH_FLAGS, 2152 ntohs(port), tcpstates[tp->t_state])); 2153 } 2154 } 2155 2156 if (flags & TH_ACK) 2157 TCPSTAT_INC(tcps_sndacks); 2158 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2159 TCPSTAT_INC(tcps_sndctrl); 2160 TCPSTAT_INC(tcps_sndtotal); 2161 2162 #ifdef INET6 2163 if (isipv6) { 2164 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2165 output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL, 2166 NULL, 0, NULL, NULL, inp); 2167 } 2168 #endif /* INET6 */ 2169 #if defined(INET) && defined(INET6) 2170 else 2171 #endif 2172 #ifdef INET 2173 { 2174 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2175 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2176 } 2177 #endif 2178 if (lgb != NULL) 2179 lgb->tlb_errno = output_ret; 2180 } 2181 2182 /* 2183 * Send a challenge ack (no data, no SACK option), but not more than 2184 * V_tcp_ack_war_cnt per V_tcp_ack_war_time_window (per TCP connection). 2185 */ 2186 void 2187 tcp_send_challenge_ack(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m) 2188 { 2189 sbintime_t now; 2190 bool send_challenge_ack; 2191 2192 if (V_tcp_ack_war_time_window == 0 || V_tcp_ack_war_cnt == 0) { 2193 /* ACK war protection is disabled. */ 2194 send_challenge_ack = true; 2195 } else { 2196 /* Start new epoch, if the previous one is already over. */ 2197 now = getsbinuptime(); 2198 if (tp->t_challenge_ack_end < now) { 2199 tp->t_challenge_ack_cnt = 0; 2200 tp->t_challenge_ack_end = now + 2201 V_tcp_ack_war_time_window * SBT_1MS; 2202 } 2203 /* 2204 * Send a challenge ACK, if less than tcp_ack_war_cnt have been 2205 * sent in the current epoch. 2206 */ 2207 if (tp->t_challenge_ack_cnt < V_tcp_ack_war_cnt) { 2208 send_challenge_ack = true; 2209 tp->t_challenge_ack_cnt++; 2210 } else { 2211 send_challenge_ack = false; 2212 } 2213 } 2214 if (send_challenge_ack) { 2215 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2216 tp->snd_nxt, TH_ACK); 2217 tp->last_ack_sent = tp->rcv_nxt; 2218 } 2219 } 2220 2221 /* 2222 * Create a new TCP control block, making an empty reassembly queue and hooking 2223 * it to the argument protocol control block. The `inp' parameter must have 2224 * come from the zone allocator set up by tcpcbstor declaration. 2225 * The caller can provide a pointer to a tcpcb of the listener to inherit the 2226 * TCP function block from the listener. 2227 */ 2228 struct tcpcb * 2229 tcp_newtcpcb(struct inpcb *inp, struct tcpcb *listening_tcb) 2230 { 2231 struct tcpcb *tp = intotcpcb(inp); 2232 #ifdef INET6 2233 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2234 #endif /* INET6 */ 2235 2236 /* 2237 * Historically allocation was done with M_ZERO. There is a lot of 2238 * code that rely on that. For now take safe approach and zero whole 2239 * tcpcb. This definitely can be optimized. 2240 */ 2241 bzero(&tp->t_start_zero, t_zero_size); 2242 2243 /* Initialise cc_var struct for this tcpcb. */ 2244 tp->t_ccv.tp = tp; 2245 rw_rlock(&tcp_function_lock); 2246 if (listening_tcb != NULL) { 2247 INP_LOCK_ASSERT(tptoinpcb(listening_tcb)); 2248 KASSERT(listening_tcb->t_fb != NULL, 2249 ("tcp_newtcpcb: listening_tcb->t_fb is NULL")); 2250 if (listening_tcb->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) { 2251 rw_runlock(&tcp_function_lock); 2252 return (NULL); 2253 } 2254 tp->t_fb = listening_tcb->t_fb; 2255 } else { 2256 tp->t_fb = V_tcp_func_set_ptr; 2257 } 2258 refcount_acquire(&tp->t_fb->tfb_refcnt); 2259 KASSERT((tp->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) == 0, 2260 ("tcp_newtcpcb: using TFB being removed")); 2261 rw_runlock(&tcp_function_lock); 2262 CC_LIST_RLOCK(); 2263 if (listening_tcb != NULL) { 2264 if (CC_ALGO(listening_tcb)->flags & CC_MODULE_BEING_REMOVED) { 2265 CC_LIST_RUNLOCK(); 2266 if (tp->t_fb->tfb_tcp_fb_fini) 2267 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2268 refcount_release(&tp->t_fb->tfb_refcnt); 2269 return (NULL); 2270 } 2271 CC_ALGO(tp) = CC_ALGO(listening_tcb); 2272 } else 2273 CC_ALGO(tp) = CC_DEFAULT_ALGO(); 2274 cc_refer(CC_ALGO(tp)); 2275 CC_LIST_RUNLOCK(); 2276 if (CC_ALGO(tp)->cb_init != NULL) 2277 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2278 cc_detach(tp); 2279 if (tp->t_fb->tfb_tcp_fb_fini) 2280 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2281 refcount_release(&tp->t_fb->tfb_refcnt); 2282 return (NULL); 2283 } 2284 2285 #ifdef TCP_HHOOK 2286 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2287 if (CC_ALGO(tp)->cb_destroy != NULL) 2288 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2289 CC_DATA(tp) = NULL; 2290 cc_detach(tp); 2291 if (tp->t_fb->tfb_tcp_fb_fini) 2292 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2293 refcount_release(&tp->t_fb->tfb_refcnt); 2294 return (NULL); 2295 } 2296 #endif 2297 2298 TAILQ_INIT(&tp->t_segq); 2299 STAILQ_INIT(&tp->t_inqueue); 2300 tp->t_maxseg = 2301 #ifdef INET6 2302 isipv6 ? V_tcp_v6mssdflt : 2303 #endif /* INET6 */ 2304 V_tcp_mssdflt; 2305 2306 /* All mbuf queue/ack compress flags should be off */ 2307 tcp_lro_features_off(tp); 2308 2309 tp->t_hpts_cpu = HPTS_CPU_NONE; 2310 tp->t_lro_cpu = HPTS_CPU_NONE; 2311 2312 callout_init_rw(&tp->t_callout, &inp->inp_lock, CALLOUT_RETURNUNLOCKED); 2313 for (int i = 0; i < TT_N; i++) 2314 tp->t_timers[i] = SBT_MAX; 2315 2316 switch (V_tcp_do_rfc1323) { 2317 case 0: 2318 break; 2319 default: 2320 case 1: 2321 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2322 break; 2323 case 2: 2324 tp->t_flags = TF_REQ_SCALE; 2325 break; 2326 case 3: 2327 tp->t_flags = TF_REQ_TSTMP; 2328 break; 2329 } 2330 if (V_tcp_do_sack) 2331 tp->t_flags |= TF_SACK_PERMIT; 2332 TAILQ_INIT(&tp->snd_holes); 2333 2334 /* 2335 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2336 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2337 * reasonable initial retransmit time. 2338 */ 2339 tp->t_srtt = TCPTV_SRTTBASE; 2340 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2341 tp->t_rttmin = tcp_rexmit_min; 2342 tp->t_rxtcur = tcp_rexmit_initial; 2343 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2344 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2345 tp->t_rcvtime = ticks; 2346 /* We always start with ticks granularity */ 2347 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 2348 /* 2349 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2350 * because the socket may be bound to an IPv6 wildcard address, 2351 * which may match an IPv4-mapped IPv6 address. 2352 */ 2353 inp->inp_ip_ttl = V_ip_defttl; 2354 #ifdef TCPPCAP 2355 /* 2356 * Init the TCP PCAP queues. 2357 */ 2358 tcp_pcap_tcpcb_init(tp); 2359 #endif 2360 #ifdef TCP_BLACKBOX 2361 /* Initialize the per-TCPCB log data. */ 2362 tcp_log_tcpcbinit(tp); 2363 #endif 2364 tp->t_pacing_rate = -1; 2365 if (tp->t_fb->tfb_tcp_fb_init) { 2366 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) { 2367 if (CC_ALGO(tp)->cb_destroy != NULL) 2368 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2369 CC_DATA(tp) = NULL; 2370 cc_detach(tp); 2371 #ifdef TCP_HHOOK 2372 khelp_destroy_osd(&tp->t_osd); 2373 #endif 2374 refcount_release(&tp->t_fb->tfb_refcnt); 2375 return (NULL); 2376 } 2377 } 2378 #ifdef STATS 2379 if (V_tcp_perconn_stats_enable == 1) 2380 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2381 #endif 2382 if (V_tcp_do_lrd) 2383 tp->t_flags |= TF_LRD; 2384 2385 return (tp); 2386 } 2387 2388 /* 2389 * Drop a TCP connection, reporting 2390 * the specified error. If connection is synchronized, 2391 * then send a RST to peer. 2392 */ 2393 struct tcpcb * 2394 tcp_drop(struct tcpcb *tp, int errno) 2395 { 2396 struct socket *so = tptosocket(tp); 2397 2398 NET_EPOCH_ASSERT(); 2399 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2400 2401 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2402 tcp_state_change(tp, TCPS_CLOSED); 2403 /* Don't use tcp_output() here due to possible recursion. */ 2404 (void)tcp_output_nodrop(tp); 2405 TCPSTAT_INC(tcps_drops); 2406 } else 2407 TCPSTAT_INC(tcps_conndrops); 2408 if (errno == ETIMEDOUT && tp->t_softerror) 2409 errno = tp->t_softerror; 2410 so->so_error = errno; 2411 return (tcp_close(tp)); 2412 } 2413 2414 void 2415 tcp_discardcb(struct tcpcb *tp) 2416 { 2417 struct inpcb *inp = tptoinpcb(tp); 2418 struct socket *so = tptosocket(tp); 2419 struct mbuf *m; 2420 #ifdef INET6 2421 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2422 #endif 2423 2424 INP_WLOCK_ASSERT(inp); 2425 MPASS(!callout_active(&tp->t_callout)); 2426 MPASS(TAILQ_EMPTY(&tp->snd_holes)); 2427 2428 /* free the reassembly queue, if any */ 2429 tcp_reass_flush(tp); 2430 2431 #ifdef TCP_OFFLOAD 2432 /* Disconnect offload device, if any. */ 2433 if (tp->t_flags & TF_TOE) 2434 tcp_offload_detach(tp); 2435 #endif 2436 #ifdef TCPPCAP 2437 /* Free the TCP PCAP queues. */ 2438 tcp_pcap_drain(&(tp->t_inpkts)); 2439 tcp_pcap_drain(&(tp->t_outpkts)); 2440 #endif 2441 2442 /* Allow the CC algorithm to clean up after itself. */ 2443 if (CC_ALGO(tp)->cb_destroy != NULL) 2444 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2445 CC_DATA(tp) = NULL; 2446 /* Detach from the CC algorithm */ 2447 cc_detach(tp); 2448 2449 #ifdef TCP_HHOOK 2450 khelp_destroy_osd(&tp->t_osd); 2451 #endif 2452 #ifdef STATS 2453 stats_blob_destroy(tp->t_stats); 2454 #endif 2455 2456 CC_ALGO(tp) = NULL; 2457 if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) { 2458 struct mbuf *prev; 2459 2460 STAILQ_INIT(&tp->t_inqueue); 2461 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev) 2462 m_freem(m); 2463 } 2464 TCPSTATES_DEC(tp->t_state); 2465 2466 if (tp->t_fb->tfb_tcp_fb_fini) 2467 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2468 MPASS(!tcp_in_hpts(tp)); 2469 #ifdef TCP_BLACKBOX 2470 tcp_log_tcpcbfini(tp); 2471 #endif 2472 2473 /* 2474 * If we got enough samples through the srtt filter, 2475 * save the rtt and rttvar in the routing entry. 2476 * 'Enough' is arbitrarily defined as 4 rtt samples. 2477 * 4 samples is enough for the srtt filter to converge 2478 * to within enough % of the correct value; fewer samples 2479 * and we could save a bogus rtt. The danger is not high 2480 * as tcp quickly recovers from everything. 2481 * XXX: Works very well but needs some more statistics! 2482 * 2483 * XXXRRS: Updating must be after the stack fini() since 2484 * that may be converting some internal representation of 2485 * say srtt etc into the general one used by other stacks. 2486 * Lets also at least protect against the so being NULL 2487 * as RW stated below. 2488 */ 2489 if ((tp->t_rttupdated >= 4) && (so != NULL)) { 2490 struct hc_metrics_lite metrics; 2491 uint32_t ssthresh; 2492 2493 bzero(&metrics, sizeof(metrics)); 2494 /* 2495 * Update the ssthresh always when the conditions below 2496 * are satisfied. This gives us better new start value 2497 * for the congestion avoidance for new connections. 2498 * ssthresh is only set if packet loss occurred on a session. 2499 * 2500 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 2501 * being torn down. Ideally this code would not use 'so'. 2502 */ 2503 ssthresh = tp->snd_ssthresh; 2504 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2505 /* 2506 * convert the limit from user data bytes to 2507 * packets then to packet data bytes. 2508 */ 2509 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2510 if (ssthresh < 2) 2511 ssthresh = 2; 2512 ssthresh *= (tp->t_maxseg + 2513 #ifdef INET6 2514 (isipv6 ? sizeof (struct ip6_hdr) + 2515 sizeof (struct tcphdr) : 2516 #endif 2517 sizeof (struct tcpiphdr) 2518 #ifdef INET6 2519 ) 2520 #endif 2521 ); 2522 } else 2523 ssthresh = 0; 2524 metrics.rmx_ssthresh = ssthresh; 2525 2526 metrics.rmx_rtt = tp->t_srtt; 2527 metrics.rmx_rttvar = tp->t_rttvar; 2528 metrics.rmx_cwnd = tp->snd_cwnd; 2529 metrics.rmx_sendpipe = 0; 2530 metrics.rmx_recvpipe = 0; 2531 2532 tcp_hc_update(&inp->inp_inc, &metrics); 2533 } 2534 2535 refcount_release(&tp->t_fb->tfb_refcnt); 2536 } 2537 2538 /* 2539 * Attempt to close a TCP control block, marking it as dropped, and freeing 2540 * the socket if we hold the only reference. 2541 */ 2542 struct tcpcb * 2543 tcp_close(struct tcpcb *tp) 2544 { 2545 struct inpcb *inp = tptoinpcb(tp); 2546 struct socket *so = tptosocket(tp); 2547 2548 INP_WLOCK_ASSERT(inp); 2549 2550 #ifdef TCP_OFFLOAD 2551 if (tp->t_state == TCPS_LISTEN) 2552 tcp_offload_listen_stop(tp); 2553 #endif 2554 /* 2555 * This releases the TFO pending counter resource for TFO listen 2556 * sockets as well as passively-created TFO sockets that transition 2557 * from SYN_RECEIVED to CLOSED. 2558 */ 2559 if (tp->t_tfo_pending) { 2560 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2561 tp->t_tfo_pending = NULL; 2562 } 2563 tcp_timer_stop(tp); 2564 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 2565 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2566 in_pcbdrop(inp); 2567 TCPSTAT_INC(tcps_closed); 2568 if (tp->t_state != TCPS_CLOSED) 2569 tcp_state_change(tp, TCPS_CLOSED); 2570 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2571 tcp_free_sackholes(tp); 2572 soisdisconnected(so); 2573 if (inp->inp_flags & INP_SOCKREF) { 2574 inp->inp_flags &= ~INP_SOCKREF; 2575 INP_WUNLOCK(inp); 2576 sorele(so); 2577 return (NULL); 2578 } 2579 return (tp); 2580 } 2581 2582 /* 2583 * Notify a tcp user of an asynchronous error; 2584 * store error as soft error, but wake up user 2585 * (for now, won't do anything until can select for soft error). 2586 * 2587 * Do not wake up user since there currently is no mechanism for 2588 * reporting soft errors (yet - a kqueue filter may be added). 2589 */ 2590 static struct inpcb * 2591 tcp_notify(struct inpcb *inp, int error) 2592 { 2593 struct tcpcb *tp; 2594 2595 INP_WLOCK_ASSERT(inp); 2596 2597 tp = intotcpcb(inp); 2598 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2599 2600 /* 2601 * Ignore some errors if we are hooked up. 2602 * If connection hasn't completed, has retransmitted several times, 2603 * and receives a second error, give up now. This is better 2604 * than waiting a long time to establish a connection that 2605 * can never complete. 2606 */ 2607 if (tp->t_state == TCPS_ESTABLISHED && 2608 (error == EHOSTUNREACH || error == ENETUNREACH || 2609 error == EHOSTDOWN)) { 2610 if (inp->inp_route.ro_nh) { 2611 NH_FREE(inp->inp_route.ro_nh); 2612 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2613 } 2614 return (inp); 2615 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2616 tp->t_softerror) { 2617 tp = tcp_drop(tp, error); 2618 if (tp != NULL) 2619 return (inp); 2620 else 2621 return (NULL); 2622 } else { 2623 tp->t_softerror = error; 2624 return (inp); 2625 } 2626 #if 0 2627 wakeup( &so->so_timeo); 2628 sorwakeup(so); 2629 sowwakeup(so); 2630 #endif 2631 } 2632 2633 static int 2634 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2635 { 2636 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2637 INPLOOKUP_RLOCKPCB); 2638 struct xinpgen xig; 2639 struct inpcb *inp; 2640 int error; 2641 2642 if (req->newptr != NULL) 2643 return (EPERM); 2644 2645 if (req->oldptr == NULL) { 2646 int n; 2647 2648 n = V_tcbinfo.ipi_count + 2649 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2650 n += imax(n / 8, 10); 2651 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2652 return (0); 2653 } 2654 2655 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2656 return (error); 2657 2658 bzero(&xig, sizeof(xig)); 2659 xig.xig_len = sizeof xig; 2660 xig.xig_count = V_tcbinfo.ipi_count + 2661 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2662 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2663 xig.xig_sogen = so_gencnt; 2664 error = SYSCTL_OUT(req, &xig, sizeof xig); 2665 if (error) 2666 return (error); 2667 2668 error = syncache_pcblist(req); 2669 if (error) 2670 return (error); 2671 2672 while ((inp = inp_next(&inpi)) != NULL) { 2673 if (inp->inp_gencnt <= xig.xig_gen && 2674 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2675 struct xtcpcb xt; 2676 2677 tcp_inptoxtp(inp, &xt); 2678 error = SYSCTL_OUT(req, &xt, sizeof xt); 2679 if (error) { 2680 INP_RUNLOCK(inp); 2681 break; 2682 } else 2683 continue; 2684 } 2685 } 2686 2687 if (!error) { 2688 /* 2689 * Give the user an updated idea of our state. 2690 * If the generation differs from what we told 2691 * her before, she knows that something happened 2692 * while we were processing this request, and it 2693 * might be necessary to retry. 2694 */ 2695 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2696 xig.xig_sogen = so_gencnt; 2697 xig.xig_count = V_tcbinfo.ipi_count + 2698 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2699 error = SYSCTL_OUT(req, &xig, sizeof xig); 2700 } 2701 2702 return (error); 2703 } 2704 2705 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2706 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2707 NULL, 0, tcp_pcblist, "S,xtcpcb", 2708 "List of active TCP connections"); 2709 2710 #ifdef INET 2711 static int 2712 tcp_getcred(SYSCTL_HANDLER_ARGS) 2713 { 2714 struct xucred xuc; 2715 struct sockaddr_in addrs[2]; 2716 struct epoch_tracker et; 2717 struct inpcb *inp; 2718 int error; 2719 2720 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2721 if (error) 2722 return (error); 2723 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2724 if (error) 2725 return (error); 2726 NET_EPOCH_ENTER(et); 2727 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2728 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2729 NET_EPOCH_EXIT(et); 2730 if (inp != NULL) { 2731 if (error == 0) 2732 error = cr_canseeinpcb(req->td->td_ucred, inp); 2733 if (error == 0) 2734 cru2x(inp->inp_cred, &xuc); 2735 INP_RUNLOCK(inp); 2736 } else 2737 error = ENOENT; 2738 if (error == 0) 2739 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2740 return (error); 2741 } 2742 2743 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 2744 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2745 0, 0, tcp_getcred, "S,xucred", 2746 "Get the xucred of a TCP connection"); 2747 #endif /* INET */ 2748 2749 #ifdef INET6 2750 static int 2751 tcp6_getcred(SYSCTL_HANDLER_ARGS) 2752 { 2753 struct epoch_tracker et; 2754 struct xucred xuc; 2755 struct sockaddr_in6 addrs[2]; 2756 struct inpcb *inp; 2757 int error; 2758 #ifdef INET 2759 int mapped = 0; 2760 #endif 2761 2762 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2763 if (error) 2764 return (error); 2765 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2766 if (error) 2767 return (error); 2768 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 2769 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 2770 return (error); 2771 } 2772 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 2773 #ifdef INET 2774 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 2775 mapped = 1; 2776 else 2777 #endif 2778 return (EINVAL); 2779 } 2780 2781 NET_EPOCH_ENTER(et); 2782 #ifdef INET 2783 if (mapped == 1) 2784 inp = in_pcblookup(&V_tcbinfo, 2785 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 2786 addrs[1].sin6_port, 2787 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 2788 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 2789 else 2790 #endif 2791 inp = in6_pcblookup(&V_tcbinfo, 2792 &addrs[1].sin6_addr, addrs[1].sin6_port, 2793 &addrs[0].sin6_addr, addrs[0].sin6_port, 2794 INPLOOKUP_RLOCKPCB, NULL); 2795 NET_EPOCH_EXIT(et); 2796 if (inp != NULL) { 2797 if (error == 0) 2798 error = cr_canseeinpcb(req->td->td_ucred, inp); 2799 if (error == 0) 2800 cru2x(inp->inp_cred, &xuc); 2801 INP_RUNLOCK(inp); 2802 } else 2803 error = ENOENT; 2804 if (error == 0) 2805 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 2806 return (error); 2807 } 2808 2809 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 2810 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 2811 0, 0, tcp6_getcred, "S,xucred", 2812 "Get the xucred of a TCP6 connection"); 2813 #endif /* INET6 */ 2814 2815 #ifdef INET 2816 /* Path MTU to try next when a fragmentation-needed message is received. */ 2817 static inline int 2818 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 2819 { 2820 int mtu = ntohs(icp->icmp_nextmtu); 2821 2822 /* If no alternative MTU was proposed, try the next smaller one. */ 2823 if (!mtu) 2824 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 2825 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 2826 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 2827 2828 return (mtu); 2829 } 2830 2831 static void 2832 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 2833 { 2834 struct ip *ip; 2835 struct tcphdr *th; 2836 struct inpcb *inp; 2837 struct tcpcb *tp; 2838 struct inpcb *(*notify)(struct inpcb *, int); 2839 struct in_conninfo inc; 2840 tcp_seq icmp_tcp_seq; 2841 int errno, mtu; 2842 2843 errno = icmp_errmap(icp); 2844 switch (errno) { 2845 case 0: 2846 return; 2847 case EMSGSIZE: 2848 notify = tcp_mtudisc_notify; 2849 break; 2850 case ECONNREFUSED: 2851 if (V_icmp_may_rst) 2852 notify = tcp_drop_syn_sent; 2853 else 2854 notify = tcp_notify; 2855 break; 2856 case EHOSTUNREACH: 2857 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 2858 notify = tcp_drop_syn_sent; 2859 else 2860 notify = tcp_notify; 2861 break; 2862 default: 2863 notify = tcp_notify; 2864 } 2865 2866 ip = &icp->icmp_ip; 2867 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 2868 icmp_tcp_seq = th->th_seq; 2869 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 2870 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 2871 if (inp != NULL) { 2872 tp = intotcpcb(inp); 2873 #ifdef TCP_OFFLOAD 2874 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 2875 /* 2876 * MTU discovery for offloaded connections. Let 2877 * the TOE driver verify seq# and process it. 2878 */ 2879 mtu = tcp_next_pmtu(icp, ip); 2880 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 2881 goto out; 2882 } 2883 #endif 2884 if (tp->t_port != port) 2885 goto out; 2886 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 2887 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 2888 if (errno == EMSGSIZE) { 2889 /* 2890 * MTU discovery: we got a needfrag and 2891 * will potentially try a lower MTU. 2892 */ 2893 mtu = tcp_next_pmtu(icp, ip); 2894 2895 /* 2896 * Only process the offered MTU if it 2897 * is smaller than the current one. 2898 */ 2899 if (mtu < tp->t_maxseg + 2900 sizeof(struct tcpiphdr)) { 2901 bzero(&inc, sizeof(inc)); 2902 inc.inc_faddr = ip->ip_dst; 2903 inc.inc_fibnum = 2904 inp->inp_inc.inc_fibnum; 2905 tcp_hc_updatemtu(&inc, mtu); 2906 inp = tcp_mtudisc(inp, mtu); 2907 } 2908 } else 2909 inp = (*notify)(inp, errno); 2910 } 2911 } else { 2912 bzero(&inc, sizeof(inc)); 2913 inc.inc_fport = th->th_dport; 2914 inc.inc_lport = th->th_sport; 2915 inc.inc_faddr = ip->ip_dst; 2916 inc.inc_laddr = ip->ip_src; 2917 syncache_unreach(&inc, icmp_tcp_seq, port); 2918 } 2919 out: 2920 if (inp != NULL) 2921 INP_WUNLOCK(inp); 2922 } 2923 2924 static void 2925 tcp_ctlinput(struct icmp *icmp) 2926 { 2927 tcp_ctlinput_with_port(icmp, htons(0)); 2928 } 2929 2930 static void 2931 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 2932 { 2933 /* Its a tunneled TCP over UDP icmp */ 2934 struct icmp *icmp = param.icmp; 2935 struct ip *outer_ip, *inner_ip; 2936 struct udphdr *udp; 2937 struct tcphdr *th, ttemp; 2938 int i_hlen, o_len; 2939 uint16_t port; 2940 2941 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 2942 inner_ip = &icmp->icmp_ip; 2943 i_hlen = inner_ip->ip_hl << 2; 2944 o_len = ntohs(outer_ip->ip_len); 2945 if (o_len < 2946 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 2947 /* Not enough data present */ 2948 return; 2949 } 2950 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 2951 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 2952 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 2953 return; 2954 } 2955 port = udp->uh_dport; 2956 th = (struct tcphdr *)(udp + 1); 2957 memcpy(&ttemp, th, sizeof(struct tcphdr)); 2958 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 2959 /* Now adjust down the size of the outer IP header */ 2960 o_len -= sizeof(struct udphdr); 2961 outer_ip->ip_len = htons(o_len); 2962 /* Now call in to the normal handling code */ 2963 tcp_ctlinput_with_port(icmp, port); 2964 } 2965 #endif /* INET */ 2966 2967 #ifdef INET6 2968 static inline int 2969 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 2970 { 2971 int mtu = ntohl(icmp6->icmp6_mtu); 2972 2973 /* 2974 * If no alternative MTU was proposed, or the proposed MTU was too 2975 * small, set to the min. 2976 */ 2977 if (mtu < IPV6_MMTU) 2978 mtu = IPV6_MMTU - 8; /* XXXNP: what is the adjustment for? */ 2979 return (mtu); 2980 } 2981 2982 static void 2983 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 2984 { 2985 struct in6_addr *dst; 2986 struct inpcb *(*notify)(struct inpcb *, int); 2987 struct ip6_hdr *ip6; 2988 struct mbuf *m; 2989 struct inpcb *inp; 2990 struct tcpcb *tp; 2991 struct icmp6_hdr *icmp6; 2992 struct in_conninfo inc; 2993 struct tcp_ports { 2994 uint16_t th_sport; 2995 uint16_t th_dport; 2996 } t_ports; 2997 tcp_seq icmp_tcp_seq; 2998 unsigned int mtu; 2999 unsigned int off; 3000 int errno; 3001 3002 icmp6 = ip6cp->ip6c_icmp6; 3003 m = ip6cp->ip6c_m; 3004 ip6 = ip6cp->ip6c_ip6; 3005 off = ip6cp->ip6c_off; 3006 dst = &ip6cp->ip6c_finaldst->sin6_addr; 3007 3008 errno = icmp6_errmap(icmp6); 3009 switch (errno) { 3010 case 0: 3011 return; 3012 case EMSGSIZE: 3013 notify = tcp_mtudisc_notify; 3014 break; 3015 case ECONNREFUSED: 3016 if (V_icmp_may_rst) 3017 notify = tcp_drop_syn_sent; 3018 else 3019 notify = tcp_notify; 3020 break; 3021 case EHOSTUNREACH: 3022 /* 3023 * There are only four ICMPs that may reset connection: 3024 * - administratively prohibited 3025 * - port unreachable 3026 * - time exceeded in transit 3027 * - unknown next header 3028 */ 3029 if (V_icmp_may_rst && 3030 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 3031 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 3032 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 3033 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 3034 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 3035 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 3036 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 3037 notify = tcp_drop_syn_sent; 3038 else 3039 notify = tcp_notify; 3040 break; 3041 default: 3042 notify = tcp_notify; 3043 } 3044 3045 /* Check if we can safely get the ports from the tcp hdr */ 3046 if (m == NULL || 3047 (m->m_pkthdr.len < 3048 (int32_t) (off + sizeof(struct tcp_ports)))) { 3049 return; 3050 } 3051 bzero(&t_ports, sizeof(struct tcp_ports)); 3052 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3053 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3054 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3055 off += sizeof(struct tcp_ports); 3056 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3057 goto out; 3058 } 3059 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3060 if (inp != NULL) { 3061 tp = intotcpcb(inp); 3062 #ifdef TCP_OFFLOAD 3063 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3064 /* MTU discovery for offloaded connections. */ 3065 mtu = tcp6_next_pmtu(icmp6); 3066 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3067 goto out; 3068 } 3069 #endif 3070 if (tp->t_port != port) 3071 goto out; 3072 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3073 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3074 if (errno == EMSGSIZE) { 3075 /* 3076 * MTU discovery: 3077 * If we got a needfrag set the MTU 3078 * in the route to the suggested new 3079 * value (if given) and then notify. 3080 */ 3081 mtu = tcp6_next_pmtu(icmp6); 3082 3083 bzero(&inc, sizeof(inc)); 3084 inc.inc_fibnum = M_GETFIB(m); 3085 inc.inc_flags |= INC_ISIPV6; 3086 inc.inc6_faddr = *dst; 3087 if (in6_setscope(&inc.inc6_faddr, 3088 m->m_pkthdr.rcvif, NULL)) 3089 goto out; 3090 /* 3091 * Only process the offered MTU if it 3092 * is smaller than the current one. 3093 */ 3094 if (mtu < tp->t_maxseg + 3095 sizeof (struct tcphdr) + 3096 sizeof (struct ip6_hdr)) { 3097 tcp_hc_updatemtu(&inc, mtu); 3098 tcp_mtudisc(inp, mtu); 3099 ICMP6STAT_INC(icp6s_pmtuchg); 3100 } 3101 } else 3102 inp = (*notify)(inp, errno); 3103 } 3104 } else { 3105 bzero(&inc, sizeof(inc)); 3106 inc.inc_fibnum = M_GETFIB(m); 3107 inc.inc_flags |= INC_ISIPV6; 3108 inc.inc_fport = t_ports.th_dport; 3109 inc.inc_lport = t_ports.th_sport; 3110 inc.inc6_faddr = *dst; 3111 inc.inc6_laddr = ip6->ip6_src; 3112 syncache_unreach(&inc, icmp_tcp_seq, port); 3113 } 3114 out: 3115 if (inp != NULL) 3116 INP_WUNLOCK(inp); 3117 } 3118 3119 static void 3120 tcp6_ctlinput(struct ip6ctlparam *ctl) 3121 { 3122 tcp6_ctlinput_with_port(ctl, htons(0)); 3123 } 3124 3125 static void 3126 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3127 { 3128 struct ip6ctlparam *ip6cp = param.ip6cp; 3129 struct mbuf *m; 3130 struct udphdr *udp; 3131 uint16_t port; 3132 3133 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3134 if (m == NULL) { 3135 return; 3136 } 3137 udp = mtod(m, struct udphdr *); 3138 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3139 return; 3140 } 3141 port = udp->uh_dport; 3142 m_adj(m, sizeof(struct udphdr)); 3143 if ((m->m_flags & M_PKTHDR) == 0) { 3144 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3145 } 3146 /* Now call in to the normal handling code */ 3147 tcp6_ctlinput_with_port(ip6cp, port); 3148 } 3149 3150 #endif /* INET6 */ 3151 3152 static uint32_t 3153 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3154 { 3155 SIPHASH_CTX ctx; 3156 uint32_t hash[2]; 3157 3158 KASSERT(len >= SIPHASH_KEY_LENGTH, 3159 ("%s: keylen %u too short ", __func__, len)); 3160 SipHash24_Init(&ctx); 3161 SipHash_SetKey(&ctx, (uint8_t *)key); 3162 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3163 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3164 switch (inc->inc_flags & INC_ISIPV6) { 3165 #ifdef INET 3166 case 0: 3167 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3168 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3169 break; 3170 #endif 3171 #ifdef INET6 3172 case INC_ISIPV6: 3173 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3174 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3175 break; 3176 #endif 3177 } 3178 SipHash_Final((uint8_t *)hash, &ctx); 3179 3180 return (hash[0] ^ hash[1]); 3181 } 3182 3183 uint32_t 3184 tcp_new_ts_offset(struct in_conninfo *inc) 3185 { 3186 struct in_conninfo inc_store, *local_inc; 3187 3188 if (!V_tcp_ts_offset_per_conn) { 3189 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3190 inc_store.inc_lport = 0; 3191 inc_store.inc_fport = 0; 3192 local_inc = &inc_store; 3193 } else { 3194 local_inc = inc; 3195 } 3196 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3197 sizeof(V_ts_offset_secret))); 3198 } 3199 3200 /* 3201 * Following is where TCP initial sequence number generation occurs. 3202 * 3203 * There are two places where we must use initial sequence numbers: 3204 * 1. In SYN-ACK packets. 3205 * 2. In SYN packets. 3206 * 3207 * All ISNs for SYN-ACK packets are generated by the syncache. See 3208 * tcp_syncache.c for details. 3209 * 3210 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3211 * depends on this property. In addition, these ISNs should be 3212 * unguessable so as to prevent connection hijacking. To satisfy 3213 * the requirements of this situation, the algorithm outlined in 3214 * RFC 1948 is used, with only small modifications. 3215 * 3216 * Implementation details: 3217 * 3218 * Time is based off the system timer, and is corrected so that it 3219 * increases by one megabyte per second. This allows for proper 3220 * recycling on high speed LANs while still leaving over an hour 3221 * before rollover. 3222 * 3223 * As reading the *exact* system time is too expensive to be done 3224 * whenever setting up a TCP connection, we increment the time 3225 * offset in two ways. First, a small random positive increment 3226 * is added to isn_offset for each connection that is set up. 3227 * Second, the function tcp_isn_tick fires once per clock tick 3228 * and increments isn_offset as necessary so that sequence numbers 3229 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3230 * random positive increments serve only to ensure that the same 3231 * exact sequence number is never sent out twice (as could otherwise 3232 * happen when a port is recycled in less than the system tick 3233 * interval.) 3234 * 3235 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3236 * between seeding of isn_secret. This is normally set to zero, 3237 * as reseeding should not be necessary. 3238 * 3239 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3240 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3241 * general, this means holding an exclusive (write) lock. 3242 */ 3243 3244 #define ISN_BYTES_PER_SECOND 1048576 3245 #define ISN_STATIC_INCREMENT 4096 3246 #define ISN_RANDOM_INCREMENT (4096 - 1) 3247 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3248 3249 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3250 VNET_DEFINE_STATIC(int, isn_last); 3251 VNET_DEFINE_STATIC(int, isn_last_reseed); 3252 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3253 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3254 3255 #define V_isn_secret VNET(isn_secret) 3256 #define V_isn_last VNET(isn_last) 3257 #define V_isn_last_reseed VNET(isn_last_reseed) 3258 #define V_isn_offset VNET(isn_offset) 3259 #define V_isn_offset_old VNET(isn_offset_old) 3260 3261 tcp_seq 3262 tcp_new_isn(struct in_conninfo *inc) 3263 { 3264 tcp_seq new_isn; 3265 u_int32_t projected_offset; 3266 3267 ISN_LOCK(); 3268 /* Seed if this is the first use, reseed if requested. */ 3269 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3270 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3271 < (u_int)ticks))) { 3272 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3273 V_isn_last_reseed = ticks; 3274 } 3275 3276 /* Compute the hash and return the ISN. */ 3277 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3278 sizeof(V_isn_secret)); 3279 V_isn_offset += ISN_STATIC_INCREMENT + 3280 (arc4random() & ISN_RANDOM_INCREMENT); 3281 if (ticks != V_isn_last) { 3282 projected_offset = V_isn_offset_old + 3283 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3284 if (SEQ_GT(projected_offset, V_isn_offset)) 3285 V_isn_offset = projected_offset; 3286 V_isn_offset_old = V_isn_offset; 3287 V_isn_last = ticks; 3288 } 3289 new_isn += V_isn_offset; 3290 ISN_UNLOCK(); 3291 return (new_isn); 3292 } 3293 3294 /* 3295 * When a specific ICMP unreachable message is received and the 3296 * connection state is SYN-SENT, drop the connection. This behavior 3297 * is controlled by the icmp_may_rst sysctl. 3298 */ 3299 static struct inpcb * 3300 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3301 { 3302 struct tcpcb *tp; 3303 3304 NET_EPOCH_ASSERT(); 3305 INP_WLOCK_ASSERT(inp); 3306 3307 tp = intotcpcb(inp); 3308 if (tp->t_state != TCPS_SYN_SENT) 3309 return (inp); 3310 3311 if (tp->t_flags & TF_FASTOPEN) 3312 tcp_fastopen_disable_path(tp); 3313 3314 tp = tcp_drop(tp, errno); 3315 if (tp != NULL) 3316 return (inp); 3317 else 3318 return (NULL); 3319 } 3320 3321 /* 3322 * When `need fragmentation' ICMP is received, update our idea of the MSS 3323 * based on the new value. Also nudge TCP to send something, since we 3324 * know the packet we just sent was dropped. 3325 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3326 */ 3327 static struct inpcb * 3328 tcp_mtudisc_notify(struct inpcb *inp, int error) 3329 { 3330 3331 return (tcp_mtudisc(inp, -1)); 3332 } 3333 3334 static struct inpcb * 3335 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3336 { 3337 struct tcpcb *tp; 3338 struct socket *so; 3339 3340 INP_WLOCK_ASSERT(inp); 3341 3342 tp = intotcpcb(inp); 3343 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3344 3345 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3346 3347 so = inp->inp_socket; 3348 SOCKBUF_LOCK(&so->so_snd); 3349 /* If the mss is larger than the socket buffer, decrease the mss. */ 3350 if (so->so_snd.sb_hiwat < tp->t_maxseg) { 3351 tp->t_maxseg = so->so_snd.sb_hiwat; 3352 if (tp->t_maxseg < V_tcp_mssdflt) { 3353 /* 3354 * The MSS is so small we should not process incoming 3355 * SACK's since we are subject to attack in such a 3356 * case. 3357 */ 3358 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3359 } else { 3360 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3361 } 3362 } 3363 SOCKBUF_UNLOCK(&so->so_snd); 3364 3365 TCPSTAT_INC(tcps_mturesent); 3366 tp->t_rtttime = 0; 3367 tp->snd_nxt = tp->snd_una; 3368 tcp_free_sackholes(tp); 3369 tp->snd_recover = tp->snd_max; 3370 if (tp->t_flags & TF_SACK_PERMIT) 3371 EXIT_FASTRECOVERY(tp->t_flags); 3372 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3373 /* 3374 * Conceptually the snd_nxt setting 3375 * and freeing sack holes should 3376 * be done by the default stacks 3377 * own tfb_tcp_mtu_chg(). 3378 */ 3379 tp->t_fb->tfb_tcp_mtu_chg(tp); 3380 } 3381 if (tcp_output(tp) < 0) 3382 return (NULL); 3383 else 3384 return (inp); 3385 } 3386 3387 #ifdef INET 3388 /* 3389 * Look-up the routing entry to the peer of this inpcb. If no route 3390 * is found and it cannot be allocated, then return 0. This routine 3391 * is called by TCP routines that access the rmx structure and by 3392 * tcp_mss_update to get the peer/interface MTU. 3393 */ 3394 uint32_t 3395 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3396 { 3397 struct nhop_object *nh; 3398 struct ifnet *ifp; 3399 uint32_t maxmtu = 0; 3400 3401 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3402 3403 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3404 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3405 if (nh == NULL) 3406 return (0); 3407 3408 ifp = nh->nh_ifp; 3409 maxmtu = nh->nh_mtu; 3410 3411 /* Report additional interface capabilities. */ 3412 if (cap != NULL) { 3413 if (ifp->if_capenable & IFCAP_TSO4 && 3414 ifp->if_hwassist & CSUM_TSO) { 3415 cap->ifcap |= CSUM_TSO; 3416 cap->tsomax = ifp->if_hw_tsomax; 3417 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3418 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3419 /* XXXKIB IFCAP2_IPSEC_OFFLOAD_TSO */ 3420 cap->ipsec_tso = (ifp->if_capenable2 & 3421 IFCAP2_BIT(IFCAP2_IPSEC_OFFLOAD)) != 0; 3422 } 3423 } 3424 } 3425 return (maxmtu); 3426 } 3427 #endif /* INET */ 3428 3429 #ifdef INET6 3430 uint32_t 3431 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3432 { 3433 struct nhop_object *nh; 3434 struct in6_addr dst6; 3435 uint32_t scopeid; 3436 struct ifnet *ifp; 3437 uint32_t maxmtu = 0; 3438 3439 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3440 3441 if (inc->inc_flags & INC_IPV6MINMTU) 3442 return (IPV6_MMTU); 3443 3444 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3445 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3446 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3447 if (nh == NULL) 3448 return (0); 3449 3450 ifp = nh->nh_ifp; 3451 maxmtu = nh->nh_mtu; 3452 3453 /* Report additional interface capabilities. */ 3454 if (cap != NULL) { 3455 if (ifp->if_capenable & IFCAP_TSO6 && 3456 ifp->if_hwassist & CSUM_TSO) { 3457 cap->ifcap |= CSUM_TSO; 3458 cap->tsomax = ifp->if_hw_tsomax; 3459 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3460 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3461 cap->ipsec_tso = false; /* XXXKIB */ 3462 } 3463 } 3464 } 3465 3466 return (maxmtu); 3467 } 3468 3469 /* 3470 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3471 * 3472 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3473 * The right place to do that is ip6_setpktopt() that has just been 3474 * executed. By the way it just filled ip6po_minmtu for us. 3475 */ 3476 void 3477 tcp6_use_min_mtu(struct tcpcb *tp) 3478 { 3479 struct inpcb *inp = tptoinpcb(tp); 3480 3481 INP_WLOCK_ASSERT(inp); 3482 /* 3483 * In case of the IPV6_USE_MIN_MTU socket 3484 * option, the INC_IPV6MINMTU flag to announce 3485 * a corresponding MSS during the initial 3486 * handshake. If the TCP connection is not in 3487 * the front states, just reduce the MSS being 3488 * used. This avoids the sending of TCP 3489 * segments which will be fragmented at the 3490 * IPv6 layer. 3491 */ 3492 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3493 if ((tp->t_state >= TCPS_SYN_SENT) && 3494 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3495 struct ip6_pktopts *opt; 3496 3497 opt = inp->in6p_outputopts; 3498 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3499 tp->t_maxseg > TCP6_MSS) { 3500 tp->t_maxseg = TCP6_MSS; 3501 if (tp->t_maxseg < V_tcp_mssdflt) { 3502 /* 3503 * The MSS is so small we should not process incoming 3504 * SACK's since we are subject to attack in such a 3505 * case. 3506 */ 3507 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3508 } else { 3509 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3510 } 3511 } 3512 } 3513 } 3514 #endif /* INET6 */ 3515 3516 /* 3517 * Calculate effective SMSS per RFC5681 definition for a given TCP 3518 * connection at its current state, taking into account SACK and etc. 3519 */ 3520 u_int 3521 tcp_maxseg(const struct tcpcb *tp) 3522 { 3523 u_int optlen; 3524 3525 if (tp->t_flags & TF_NOOPT) 3526 return (tp->t_maxseg); 3527 3528 /* 3529 * Here we have a simplified code from tcp_addoptions(), 3530 * without a proper loop, and having most of paddings hardcoded. 3531 * We might make mistakes with padding here in some edge cases, 3532 * but this is harmless, since result of tcp_maxseg() is used 3533 * only in cwnd and ssthresh estimations. 3534 */ 3535 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3536 if (tp->t_flags & TF_RCVD_TSTMP) 3537 optlen = TCPOLEN_TSTAMP_APPA; 3538 else 3539 optlen = 0; 3540 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3541 if (tp->t_flags & TF_SIGNATURE) 3542 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3543 #endif 3544 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3545 optlen += TCPOLEN_SACKHDR; 3546 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3547 optlen = PADTCPOLEN(optlen); 3548 } 3549 } else { 3550 if (tp->t_flags & TF_REQ_TSTMP) 3551 optlen = TCPOLEN_TSTAMP_APPA; 3552 else 3553 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3554 if (tp->t_flags & TF_REQ_SCALE) 3555 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3556 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3557 if (tp->t_flags & TF_SIGNATURE) 3558 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3559 #endif 3560 if (tp->t_flags & TF_SACK_PERMIT) 3561 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3562 } 3563 optlen = min(optlen, TCP_MAXOLEN); 3564 return (tp->t_maxseg - optlen); 3565 } 3566 3567 3568 u_int 3569 tcp_fixed_maxseg(const struct tcpcb *tp) 3570 { 3571 int optlen; 3572 3573 if (tp->t_flags & TF_NOOPT) 3574 return (tp->t_maxseg); 3575 3576 /* 3577 * Here we have a simplified code from tcp_addoptions(), 3578 * without a proper loop, and having most of paddings hardcoded. 3579 * We only consider fixed options that we would send every 3580 * time I.e. SACK is not considered. This is important 3581 * for cc modules to figure out what the modulo of the 3582 * cwnd should be. 3583 */ 3584 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3585 if (tp->t_flags & TF_RCVD_TSTMP) 3586 optlen = TCPOLEN_TSTAMP_APPA; 3587 else 3588 optlen = 0; 3589 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3590 if (tp->t_flags & TF_SIGNATURE) 3591 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3592 #endif 3593 } else { 3594 if (tp->t_flags & TF_REQ_TSTMP) 3595 optlen = TCPOLEN_TSTAMP_APPA; 3596 else 3597 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3598 if (tp->t_flags & TF_REQ_SCALE) 3599 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3600 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3601 if (tp->t_flags & TF_SIGNATURE) 3602 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3603 #endif 3604 if (tp->t_flags & TF_SACK_PERMIT) 3605 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3606 } 3607 optlen = min(optlen, TCP_MAXOLEN); 3608 return (tp->t_maxseg - optlen); 3609 } 3610 3611 3612 3613 static int 3614 sysctl_drop(SYSCTL_HANDLER_ARGS) 3615 { 3616 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3617 struct sockaddr_storage addrs[2]; 3618 struct inpcb *inp; 3619 struct tcpcb *tp; 3620 #ifdef INET 3621 struct sockaddr_in *fin = NULL, *lin = NULL; 3622 #endif 3623 struct epoch_tracker et; 3624 #ifdef INET6 3625 struct sockaddr_in6 *fin6, *lin6; 3626 #endif 3627 int error; 3628 3629 inp = NULL; 3630 #ifdef INET6 3631 fin6 = lin6 = NULL; 3632 #endif 3633 error = 0; 3634 3635 if (req->oldptr != NULL || req->oldlen != 0) 3636 return (EINVAL); 3637 if (req->newptr == NULL) 3638 return (EPERM); 3639 if (req->newlen < sizeof(addrs)) 3640 return (ENOMEM); 3641 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3642 if (error) 3643 return (error); 3644 3645 switch (addrs[0].ss_family) { 3646 #ifdef INET6 3647 case AF_INET6: 3648 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3649 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3650 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3651 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3652 return (EINVAL); 3653 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3654 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3655 return (EINVAL); 3656 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3657 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3658 #ifdef INET 3659 fin = (struct sockaddr_in *)&addrs[0]; 3660 lin = (struct sockaddr_in *)&addrs[1]; 3661 #endif 3662 break; 3663 } 3664 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3665 if (error) 3666 return (error); 3667 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3668 if (error) 3669 return (error); 3670 break; 3671 #endif 3672 #ifdef INET 3673 case AF_INET: 3674 fin = (struct sockaddr_in *)&addrs[0]; 3675 lin = (struct sockaddr_in *)&addrs[1]; 3676 if (fin->sin_len != sizeof(struct sockaddr_in) || 3677 lin->sin_len != sizeof(struct sockaddr_in)) 3678 return (EINVAL); 3679 break; 3680 #endif 3681 default: 3682 return (EINVAL); 3683 } 3684 NET_EPOCH_ENTER(et); 3685 switch (addrs[0].ss_family) { 3686 #ifdef INET6 3687 case AF_INET6: 3688 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3689 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3690 INPLOOKUP_WLOCKPCB, NULL); 3691 break; 3692 #endif 3693 #ifdef INET 3694 case AF_INET: 3695 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3696 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3697 break; 3698 #endif 3699 } 3700 if (inp != NULL) { 3701 if (!SOLISTENING(inp->inp_socket)) { 3702 tp = intotcpcb(inp); 3703 tp = tcp_drop(tp, ECONNABORTED); 3704 if (tp != NULL) 3705 INP_WUNLOCK(inp); 3706 } else 3707 INP_WUNLOCK(inp); 3708 } else 3709 error = ESRCH; 3710 NET_EPOCH_EXIT(et); 3711 return (error); 3712 } 3713 3714 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3715 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3716 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3717 "Drop TCP connection"); 3718 3719 static int 3720 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3721 { 3722 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3723 &tcp_ctloutput_set)); 3724 } 3725 3726 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3727 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3728 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3729 "Set socket option for TCP endpoint"); 3730 3731 #ifdef KERN_TLS 3732 static int 3733 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 3734 { 3735 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3736 struct sockaddr_storage addrs[2]; 3737 struct inpcb *inp; 3738 #ifdef INET 3739 struct sockaddr_in *fin = NULL, *lin = NULL; 3740 #endif 3741 struct epoch_tracker et; 3742 #ifdef INET6 3743 struct sockaddr_in6 *fin6, *lin6; 3744 #endif 3745 int error; 3746 3747 inp = NULL; 3748 #ifdef INET6 3749 fin6 = lin6 = NULL; 3750 #endif 3751 error = 0; 3752 3753 if (req->oldptr != NULL || req->oldlen != 0) 3754 return (EINVAL); 3755 if (req->newptr == NULL) 3756 return (EPERM); 3757 if (req->newlen < sizeof(addrs)) 3758 return (ENOMEM); 3759 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3760 if (error) 3761 return (error); 3762 3763 switch (addrs[0].ss_family) { 3764 #ifdef INET6 3765 case AF_INET6: 3766 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3767 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3768 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3769 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3770 return (EINVAL); 3771 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3772 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3773 return (EINVAL); 3774 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3775 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3776 #ifdef INET 3777 fin = (struct sockaddr_in *)&addrs[0]; 3778 lin = (struct sockaddr_in *)&addrs[1]; 3779 #endif 3780 break; 3781 } 3782 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3783 if (error) 3784 return (error); 3785 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3786 if (error) 3787 return (error); 3788 break; 3789 #endif 3790 #ifdef INET 3791 case AF_INET: 3792 fin = (struct sockaddr_in *)&addrs[0]; 3793 lin = (struct sockaddr_in *)&addrs[1]; 3794 if (fin->sin_len != sizeof(struct sockaddr_in) || 3795 lin->sin_len != sizeof(struct sockaddr_in)) 3796 return (EINVAL); 3797 break; 3798 #endif 3799 default: 3800 return (EINVAL); 3801 } 3802 NET_EPOCH_ENTER(et); 3803 switch (addrs[0].ss_family) { 3804 #ifdef INET6 3805 case AF_INET6: 3806 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3807 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3808 INPLOOKUP_WLOCKPCB, NULL); 3809 break; 3810 #endif 3811 #ifdef INET 3812 case AF_INET: 3813 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3814 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3815 break; 3816 #endif 3817 } 3818 NET_EPOCH_EXIT(et); 3819 if (inp != NULL) { 3820 struct socket *so; 3821 3822 so = inp->inp_socket; 3823 soref(so); 3824 error = ktls_set_tx_mode(so, 3825 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 3826 INP_WUNLOCK(inp); 3827 sorele(so); 3828 } else 3829 error = ESRCH; 3830 return (error); 3831 } 3832 3833 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 3834 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3835 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 3836 "Switch TCP connection to SW TLS"); 3837 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 3838 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3839 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 3840 "Switch TCP connection to ifnet TLS"); 3841 #endif 3842 3843 /* 3844 * Generate a standardized TCP log line for use throughout the 3845 * tcp subsystem. Memory allocation is done with M_NOWAIT to 3846 * allow use in the interrupt context. 3847 * 3848 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 3849 * NB: The function may return NULL if memory allocation failed. 3850 * 3851 * Due to header inclusion and ordering limitations the struct ip 3852 * and ip6_hdr pointers have to be passed as void pointers. 3853 */ 3854 char * 3855 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3856 const void *ip6hdr) 3857 { 3858 3859 /* Is logging enabled? */ 3860 if (V_tcp_log_in_vain == 0) 3861 return (NULL); 3862 3863 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3864 } 3865 3866 char * 3867 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3868 const void *ip6hdr) 3869 { 3870 3871 /* Is logging enabled? */ 3872 if (tcp_log_debug == 0) 3873 return (NULL); 3874 3875 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 3876 } 3877 3878 static char * 3879 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 3880 const void *ip6hdr) 3881 { 3882 char *s, *sp; 3883 size_t size; 3884 #ifdef INET 3885 const struct ip *ip = (const struct ip *)ip4hdr; 3886 #endif 3887 #ifdef INET6 3888 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 3889 #endif /* INET6 */ 3890 3891 /* 3892 * The log line looks like this: 3893 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 3894 */ 3895 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 3896 sizeof(PRINT_TH_FLAGS) + 1 + 3897 #ifdef INET6 3898 2 * INET6_ADDRSTRLEN; 3899 #else 3900 2 * INET_ADDRSTRLEN; 3901 #endif /* INET6 */ 3902 3903 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 3904 if (s == NULL) 3905 return (NULL); 3906 3907 strcat(s, "TCP: ["); 3908 sp = s + strlen(s); 3909 3910 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 3911 inet_ntoa_r(inc->inc_faddr, sp); 3912 sp = s + strlen(s); 3913 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3914 sp = s + strlen(s); 3915 inet_ntoa_r(inc->inc_laddr, sp); 3916 sp = s + strlen(s); 3917 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3918 #ifdef INET6 3919 } else if (inc) { 3920 ip6_sprintf(sp, &inc->inc6_faddr); 3921 sp = s + strlen(s); 3922 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 3923 sp = s + strlen(s); 3924 ip6_sprintf(sp, &inc->inc6_laddr); 3925 sp = s + strlen(s); 3926 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 3927 } else if (ip6 && th) { 3928 ip6_sprintf(sp, &ip6->ip6_src); 3929 sp = s + strlen(s); 3930 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3931 sp = s + strlen(s); 3932 ip6_sprintf(sp, &ip6->ip6_dst); 3933 sp = s + strlen(s); 3934 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3935 #endif /* INET6 */ 3936 #ifdef INET 3937 } else if (ip && th) { 3938 inet_ntoa_r(ip->ip_src, sp); 3939 sp = s + strlen(s); 3940 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 3941 sp = s + strlen(s); 3942 inet_ntoa_r(ip->ip_dst, sp); 3943 sp = s + strlen(s); 3944 sprintf(sp, "]:%i", ntohs(th->th_dport)); 3945 #endif /* INET */ 3946 } else { 3947 free(s, M_TCPLOG); 3948 return (NULL); 3949 } 3950 sp = s + strlen(s); 3951 if (th) 3952 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 3953 if (*(s + size - 1) != '\0') 3954 panic("%s: string too long", __func__); 3955 return (s); 3956 } 3957 3958 /* 3959 * A subroutine which makes it easy to track TCP state changes with DTrace. 3960 * This function shouldn't be called for t_state initializations that don't 3961 * correspond to actual TCP state transitions. 3962 */ 3963 void 3964 tcp_state_change(struct tcpcb *tp, int newstate) 3965 { 3966 #if defined(KDTRACE_HOOKS) 3967 int pstate = tp->t_state; 3968 #endif 3969 3970 TCPSTATES_DEC(tp->t_state); 3971 TCPSTATES_INC(newstate); 3972 tp->t_state = newstate; 3973 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 3974 } 3975 3976 /* 3977 * Create an external-format (``xtcpcb'') structure using the information in 3978 * the kernel-format tcpcb structure pointed to by tp. This is done to 3979 * reduce the spew of irrelevant information over this interface, to isolate 3980 * user code from changes in the kernel structure, and potentially to provide 3981 * information-hiding if we decide that some of this information should be 3982 * hidden from users. 3983 */ 3984 void 3985 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 3986 { 3987 struct tcpcb *tp = intotcpcb(inp); 3988 sbintime_t now; 3989 3990 bzero(xt, sizeof(*xt)); 3991 xt->t_state = tp->t_state; 3992 xt->t_logstate = tcp_get_bblog_state(tp); 3993 xt->t_flags = tp->t_flags; 3994 xt->t_sndzerowin = tp->t_sndzerowin; 3995 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 3996 xt->t_rcvoopack = tp->t_rcvoopack; 3997 xt->t_rcv_wnd = tp->rcv_wnd; 3998 xt->t_snd_wnd = tp->snd_wnd; 3999 xt->t_snd_cwnd = tp->snd_cwnd; 4000 xt->t_snd_ssthresh = tp->snd_ssthresh; 4001 xt->t_dsack_bytes = tp->t_dsack_bytes; 4002 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4003 xt->t_dsack_pack = tp->t_dsack_pack; 4004 xt->t_maxseg = tp->t_maxseg; 4005 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4006 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4007 4008 now = getsbinuptime(); 4009 #define COPYTIMER(which,where) do { \ 4010 if (tp->t_timers[which] != SBT_MAX) \ 4011 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 4012 else \ 4013 xt->where = 0; \ 4014 } while (0) 4015 COPYTIMER(TT_DELACK, tt_delack); 4016 COPYTIMER(TT_REXMT, tt_rexmt); 4017 COPYTIMER(TT_PERSIST, tt_persist); 4018 COPYTIMER(TT_KEEP, tt_keep); 4019 COPYTIMER(TT_2MSL, tt_2msl); 4020 #undef COPYTIMER 4021 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4022 4023 xt->xt_encaps_port = tp->t_port; 4024 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4025 TCP_FUNCTION_NAME_LEN_MAX); 4026 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 4027 #ifdef TCP_BLACKBOX 4028 (void)tcp_log_get_id(tp, xt->xt_logid); 4029 #endif 4030 4031 xt->xt_len = sizeof(struct xtcpcb); 4032 in_pcbtoxinpcb(inp, &xt->xt_inp); 4033 } 4034 4035 void 4036 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4037 { 4038 uint32_t bit, i; 4039 4040 if ((tp == NULL) || 4041 (status > TCP_EI_STATUS_MAX_VALUE) || 4042 (status == 0)) { 4043 /* Invalid */ 4044 return; 4045 } 4046 if (status > (sizeof(uint32_t) * 8)) { 4047 /* Should this be a KASSERT? */ 4048 return; 4049 } 4050 bit = 1U << (status - 1); 4051 if (bit & tp->t_end_info_status) { 4052 /* already logged */ 4053 return; 4054 } 4055 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4056 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4057 tp->t_end_info_bytes[i] = status; 4058 tp->t_end_info_status |= bit; 4059 break; 4060 } 4061 } 4062 } 4063 4064 int 4065 tcp_can_enable_pacing(void) 4066 { 4067 4068 if ((tcp_pacing_limit == -1) || 4069 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4070 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4071 shadow_num_connections = number_of_tcp_connections_pacing; 4072 return (1); 4073 } else { 4074 counter_u64_add(tcp_pacing_failures, 1); 4075 return (0); 4076 } 4077 } 4078 4079 int 4080 tcp_incr_dgp_pacing_cnt(void) 4081 { 4082 if ((tcp_dgp_limit == -1) || 4083 (tcp_dgp_limit > number_of_dgp_connections)) { 4084 atomic_fetchadd_int(&number_of_dgp_connections, 1); 4085 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4086 return (1); 4087 } else { 4088 counter_u64_add(tcp_dgp_failures, 1); 4089 return (0); 4090 } 4091 } 4092 4093 static uint8_t tcp_dgp_warning = 0; 4094 4095 void 4096 tcp_dec_dgp_pacing_cnt(void) 4097 { 4098 uint32_t ret; 4099 4100 ret = atomic_fetchadd_int(&number_of_dgp_connections, -1); 4101 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4102 KASSERT(ret != 0, ("number_of_dgp_connections -1 would cause wrap?")); 4103 if (ret == 0) { 4104 if (tcp_dgp_limit != -1) { 4105 printf("Warning all DGP is now disabled, count decrements invalidly!\n"); 4106 tcp_dgp_limit = 0; 4107 tcp_dgp_warning = 1; 4108 } else if (tcp_dgp_warning == 0) { 4109 printf("Warning DGP pacing is invalid, invalid decrement\n"); 4110 tcp_dgp_warning = 1; 4111 } 4112 } 4113 4114 } 4115 4116 static uint8_t tcp_pacing_warning = 0; 4117 4118 void 4119 tcp_decrement_paced_conn(void) 4120 { 4121 uint32_t ret; 4122 4123 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4124 shadow_num_connections = number_of_tcp_connections_pacing; 4125 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4126 if (ret == 0) { 4127 if (tcp_pacing_limit != -1) { 4128 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4129 tcp_pacing_limit = 0; 4130 } else if (tcp_pacing_warning == 0) { 4131 printf("Warning pacing count is invalid, invalid decrement\n"); 4132 tcp_pacing_warning = 1; 4133 } 4134 } 4135 } 4136 4137 static void 4138 tcp_default_switch_failed(struct tcpcb *tp) 4139 { 4140 /* 4141 * If a switch fails we only need to 4142 * care about two things: 4143 * a) The t_flags2 4144 * and 4145 * b) The timer granularity. 4146 * Timeouts, at least for now, don't use the 4147 * old callout system in the other stacks so 4148 * those are hopefully safe. 4149 */ 4150 tcp_lro_features_off(tp); 4151 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 4152 } 4153 4154 #ifdef TCP_ACCOUNTING 4155 int 4156 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss) 4157 { 4158 if (SEQ_LT(th->th_ack, tp->snd_una)) { 4159 /* Do we have a SACK? */ 4160 if (to->to_flags & TOF_SACK) { 4161 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4162 tp->tcp_cnt_counters[ACK_SACK]++; 4163 } 4164 return (ACK_SACK); 4165 } else { 4166 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4167 tp->tcp_cnt_counters[ACK_BEHIND]++; 4168 } 4169 return (ACK_BEHIND); 4170 } 4171 } else if (th->th_ack == tp->snd_una) { 4172 /* Do we have a SACK? */ 4173 if (to->to_flags & TOF_SACK) { 4174 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4175 tp->tcp_cnt_counters[ACK_SACK]++; 4176 } 4177 return (ACK_SACK); 4178 } else if (tiwin != tp->snd_wnd) { 4179 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4180 tp->tcp_cnt_counters[ACK_RWND]++; 4181 } 4182 return (ACK_RWND); 4183 } else { 4184 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4185 tp->tcp_cnt_counters[ACK_DUPACK]++; 4186 } 4187 return (ACK_DUPACK); 4188 } 4189 } else { 4190 if (!SEQ_GT(th->th_ack, tp->snd_max)) { 4191 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4192 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss); 4193 } 4194 } 4195 if (to->to_flags & TOF_SACK) { 4196 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4197 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++; 4198 } 4199 return (ACK_CUMACK_SACK); 4200 } else { 4201 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4202 tp->tcp_cnt_counters[ACK_CUMACK]++; 4203 } 4204 return (ACK_CUMACK); 4205 } 4206 } 4207 } 4208 #endif 4209 4210 void 4211 tcp_change_time_units(struct tcpcb *tp, int granularity) 4212 { 4213 if (tp->t_tmr_granularity == granularity) { 4214 /* We are there */ 4215 return; 4216 } 4217 if (granularity == TCP_TMR_GRANULARITY_USEC) { 4218 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS), 4219 ("Granularity is not TICKS its %u in tp:%p", 4220 tp->t_tmr_granularity, tp)); 4221 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow); 4222 if (tp->t_srtt > 1) { 4223 uint32_t val, frac; 4224 4225 val = tp->t_srtt >> TCP_RTT_SHIFT; 4226 frac = tp->t_srtt & 0x1f; 4227 tp->t_srtt = TICKS_2_USEC(val); 4228 /* 4229 * frac is the fractional part of the srtt (if any) 4230 * but its in ticks and every bit represents 4231 * 1/32nd of a hz. 4232 */ 4233 if (frac) { 4234 if (hz == 1000) { 4235 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4236 } else { 4237 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4238 } 4239 tp->t_srtt += frac; 4240 } 4241 } 4242 if (tp->t_rttvar) { 4243 uint32_t val, frac; 4244 4245 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT; 4246 frac = tp->t_rttvar & 0x1f; 4247 tp->t_rttvar = TICKS_2_USEC(val); 4248 /* 4249 * frac is the fractional part of the srtt (if any) 4250 * but its in ticks and every bit represents 4251 * 1/32nd of a hz. 4252 */ 4253 if (frac) { 4254 if (hz == 1000) { 4255 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4256 } else { 4257 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4258 } 4259 tp->t_rttvar += frac; 4260 } 4261 } 4262 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC; 4263 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) { 4264 /* Convert back to ticks, with */ 4265 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC), 4266 ("Granularity is not USEC its %u in tp:%p", 4267 tp->t_tmr_granularity, tp)); 4268 if (tp->t_srtt > 1) { 4269 uint32_t val, frac; 4270 4271 val = USEC_2_TICKS(tp->t_srtt); 4272 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4273 tp->t_srtt = val << TCP_RTT_SHIFT; 4274 /* 4275 * frac is the fractional part here is left 4276 * over from converting to hz and shifting. 4277 * We need to convert this to the 5 bit 4278 * remainder. 4279 */ 4280 if (frac) { 4281 if (hz == 1000) { 4282 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4283 } else { 4284 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4285 } 4286 tp->t_srtt += frac; 4287 } 4288 } 4289 if (tp->t_rttvar) { 4290 uint32_t val, frac; 4291 4292 val = USEC_2_TICKS(tp->t_rttvar); 4293 frac = tp->t_rttvar % (HPTS_USEC_IN_SEC / hz); 4294 tp->t_rttvar = val << TCP_RTTVAR_SHIFT; 4295 /* 4296 * frac is the fractional part here is left 4297 * over from converting to hz and shifting. 4298 * We need to convert this to the 4 bit 4299 * remainder. 4300 */ 4301 if (frac) { 4302 if (hz == 1000) { 4303 frac = (((uint64_t)frac * (uint64_t)TCP_RTTVAR_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4304 } else { 4305 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTTVAR_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4306 } 4307 tp->t_rttvar += frac; 4308 } 4309 } 4310 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow); 4311 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 4312 } 4313 #ifdef INVARIANTS 4314 else { 4315 panic("Unknown granularity:%d tp:%p", 4316 granularity, tp); 4317 } 4318 #endif 4319 } 4320 4321 void 4322 tcp_handle_orphaned_packets(struct tcpcb *tp) 4323 { 4324 struct mbuf *save, *m, *prev; 4325 /* 4326 * Called when a stack switch is occuring from the fini() 4327 * of the old stack. We assue the init() as already been 4328 * run of the new stack and it has set the t_flags2 to 4329 * what it supports. This function will then deal with any 4330 * differences i.e. cleanup packets that maybe queued that 4331 * the newstack does not support. 4332 */ 4333 4334 if (tp->t_flags2 & TF2_MBUF_L_ACKS) 4335 return; 4336 if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 && 4337 !STAILQ_EMPTY(&tp->t_inqueue)) { 4338 /* 4339 * It is unsafe to process the packets since a 4340 * reset may be lurking in them (its rare but it 4341 * can occur). If we were to find a RST, then we 4342 * would end up dropping the connection and the 4343 * INP lock, so when we return the caller (tcp_usrreq) 4344 * will blow up when it trys to unlock the inp. 4345 * This new stack does not do any fancy LRO features 4346 * so all we can do is toss the packets. 4347 */ 4348 m = STAILQ_FIRST(&tp->t_inqueue); 4349 STAILQ_INIT(&tp->t_inqueue); 4350 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) 4351 m_freem(m); 4352 } else { 4353 /* 4354 * Here we have a stack that does mbuf queuing but 4355 * does not support compressed ack's. We must 4356 * walk all the mbufs and discard any compressed acks. 4357 */ 4358 STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) { 4359 if (m->m_flags & M_ACKCMP) { 4360 if (m == STAILQ_FIRST(&tp->t_inqueue)) 4361 STAILQ_REMOVE_HEAD(&tp->t_inqueue, 4362 m_stailqpkt); 4363 else 4364 STAILQ_REMOVE_AFTER(&tp->t_inqueue, 4365 prev, m_stailqpkt); 4366 m_freem(m); 4367 } else 4368 prev = m; 4369 } 4370 } 4371 } 4372 4373 #ifdef TCP_REQUEST_TRK 4374 uint32_t 4375 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes) 4376 { 4377 #ifdef KERN_TLS 4378 struct ktls_session *tls; 4379 uint32_t rec_oh, records; 4380 4381 tls = so->so_snd.sb_tls_info; 4382 if (tls == NULL) 4383 return (0); 4384 4385 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen; 4386 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len); 4387 return (records * rec_oh); 4388 #else 4389 return (0); 4390 #endif 4391 } 4392 4393 extern uint32_t tcp_stale_entry_time; 4394 uint32_t tcp_stale_entry_time = 250000; 4395 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW, 4396 &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out"); 4397 4398 void 4399 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req, 4400 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes) 4401 { 4402 if (tcp_bblogging_on(tp)) { 4403 union tcp_log_stackspecific log; 4404 struct timeval tv; 4405 4406 memset(&log.u_bbr, 0, sizeof(log.u_bbr)); 4407 log.u_bbr.inhpts = tcp_in_hpts(tp); 4408 log.u_bbr.flex8 = val; 4409 log.u_bbr.rttProp = req->timestamp; 4410 log.u_bbr.delRate = req->start; 4411 log.u_bbr.cur_del_rate = req->end; 4412 log.u_bbr.flex1 = req->start_seq; 4413 log.u_bbr.flex2 = req->end_seq; 4414 log.u_bbr.flex3 = req->flags; 4415 log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff); 4416 log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff); 4417 log.u_bbr.flex7 = slot; 4418 log.u_bbr.bw_inuse = offset; 4419 /* nbytes = flex6 | epoch */ 4420 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff); 4421 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff); 4422 /* cspr = lt_epoch | pkts_out */ 4423 log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff); 4424 log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff); 4425 log.u_bbr.applimited = tp->t_tcpreq_closed; 4426 log.u_bbr.applimited <<= 8; 4427 log.u_bbr.applimited |= tp->t_tcpreq_open; 4428 log.u_bbr.applimited <<= 8; 4429 log.u_bbr.applimited |= tp->t_tcpreq_req; 4430 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 4431 TCP_LOG_EVENTP(tp, NULL, 4432 &tptosocket(tp)->so_rcv, 4433 &tptosocket(tp)->so_snd, 4434 TCP_LOG_REQ_T, 0, 4435 0, &log, false, &tv); 4436 } 4437 } 4438 4439 void 4440 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent) 4441 { 4442 if (tp->t_tcpreq_req > 0) 4443 tp->t_tcpreq_req--; 4444 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4445 if (tp->t_tcpreq_open > 0) 4446 tp->t_tcpreq_open--; 4447 } else { 4448 if (tp->t_tcpreq_closed > 0) 4449 tp->t_tcpreq_closed--; 4450 } 4451 ent->flags = TCP_TRK_TRACK_FLG_EMPTY; 4452 } 4453 4454 static void 4455 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest) 4456 { 4457 struct tcp_sendfile_track *ent; 4458 uint64_t time_delta, oldest_delta; 4459 int i, oldest, oldest_set = 0, cnt_rm = 0; 4460 4461 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4462 ent = &tp->t_tcpreq_info[i]; 4463 if (ent->flags != TCP_TRK_TRACK_FLG_USED) { 4464 /* 4465 * We only care about closed end ranges 4466 * that are allocated and have no sendfile 4467 * ever touching them. They would be in 4468 * state USED. 4469 */ 4470 continue; 4471 } 4472 if (ts >= ent->localtime) 4473 time_delta = ts - ent->localtime; 4474 else 4475 time_delta = 0; 4476 if (time_delta && 4477 ((oldest_delta < time_delta) || (oldest_set == 0))) { 4478 oldest_set = 1; 4479 oldest = i; 4480 oldest_delta = time_delta; 4481 } 4482 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) { 4483 /* 4484 * No sendfile in a our time-limit 4485 * time to purge it. 4486 */ 4487 cnt_rm++; 4488 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4489 time_delta, 0); 4490 tcp_req_free_a_slot(tp, ent); 4491 } 4492 } 4493 if ((cnt_rm == 0) && rm_oldest && oldest_set) { 4494 ent = &tp->t_tcpreq_info[oldest]; 4495 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4496 oldest_delta, 1); 4497 tcp_req_free_a_slot(tp, ent); 4498 } 4499 } 4500 4501 int 4502 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point) 4503 { 4504 int i, ret = 0; 4505 struct tcp_sendfile_track *ent; 4506 4507 /* Clean up any old closed end requests that are now completed */ 4508 if (tp->t_tcpreq_req == 0) 4509 return (0); 4510 if (tp->t_tcpreq_closed == 0) 4511 return (0); 4512 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4513 ent = &tp->t_tcpreq_info[i]; 4514 /* Skip empty ones */ 4515 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4516 continue; 4517 /* Skip open ones */ 4518 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) 4519 continue; 4520 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4521 /* We are past it -- free it */ 4522 tcp_req_log_req_info(tp, ent, 4523 i, TCP_TRK_REQ_LOG_FREED, 0, 0); 4524 tcp_req_free_a_slot(tp, ent); 4525 ret++; 4526 } 4527 } 4528 return (ret); 4529 } 4530 4531 int 4532 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point) 4533 { 4534 if (tp->t_tcpreq_req == 0) 4535 return (-1); 4536 if (tp->t_tcpreq_closed == 0) 4537 return (-1); 4538 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4539 return (-1); 4540 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4541 return (1); 4542 } 4543 return (0); 4544 } 4545 4546 struct tcp_sendfile_track * 4547 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip) 4548 { 4549 /* 4550 * Given an ack point (th_ack) walk through our entries and 4551 * return the first one found that th_ack goes past the 4552 * end_seq. 4553 */ 4554 struct tcp_sendfile_track *ent; 4555 int i; 4556 4557 if (tp->t_tcpreq_req == 0) { 4558 /* none open */ 4559 return (NULL); 4560 } 4561 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4562 ent = &tp->t_tcpreq_info[i]; 4563 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4564 continue; 4565 if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) { 4566 if (SEQ_GEQ(th_ack, ent->end_seq)) { 4567 *ip = i; 4568 return (ent); 4569 } 4570 } 4571 } 4572 return (NULL); 4573 } 4574 4575 struct tcp_sendfile_track * 4576 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq) 4577 { 4578 struct tcp_sendfile_track *ent; 4579 int i; 4580 4581 if (tp->t_tcpreq_req == 0) { 4582 /* none open */ 4583 return (NULL); 4584 } 4585 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4586 ent = &tp->t_tcpreq_info[i]; 4587 tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH, 4588 (uint64_t)seq, 0); 4589 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4590 continue; 4591 } 4592 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4593 /* 4594 * An open end request only needs to 4595 * match the beginning seq or be 4596 * all we have (once we keep going on 4597 * a open end request we may have a seq 4598 * wrap). 4599 */ 4600 if ((SEQ_GEQ(seq, ent->start_seq)) || 4601 (tp->t_tcpreq_closed == 0)) 4602 return (ent); 4603 } else { 4604 /* 4605 * For this one we need to 4606 * be a bit more careful if its 4607 * completed at least. 4608 */ 4609 if ((SEQ_GEQ(seq, ent->start_seq)) && 4610 (SEQ_LT(seq, ent->end_seq))) { 4611 return (ent); 4612 } 4613 } 4614 } 4615 return (NULL); 4616 } 4617 4618 /* Should this be in its own file tcp_req.c ? */ 4619 struct tcp_sendfile_track * 4620 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups) 4621 { 4622 struct tcp_sendfile_track *fil; 4623 int i, allocated; 4624 4625 /* In case the stack does not check for completions do so now */ 4626 tcp_req_check_for_comp(tp, tp->snd_una); 4627 /* Check for stale entries */ 4628 if (tp->t_tcpreq_req) 4629 tcp_req_check_for_stale_entries(tp, ts, 4630 (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ)); 4631 /* Check to see if this is a duplicate of one not started */ 4632 if (tp->t_tcpreq_req) { 4633 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4634 fil = &tp->t_tcpreq_info[i]; 4635 if ((fil->flags & TCP_TRK_TRACK_FLG_USED) == 0) 4636 continue; 4637 if ((fil->timestamp == req->timestamp) && 4638 (fil->start == req->start) && 4639 ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) || 4640 (fil->end == req->end))) { 4641 /* 4642 * We already have this request 4643 * and it has not been started with sendfile. 4644 * This probably means the user was returned 4645 * a 4xx of some sort and its going to age 4646 * out, lets not duplicate it. 4647 */ 4648 return (fil); 4649 } 4650 } 4651 } 4652 /* Ok if there is no room at the inn we are in trouble */ 4653 if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) { 4654 tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL); 4655 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4656 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], 4657 i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0); 4658 } 4659 return (NULL); 4660 } 4661 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4662 fil = &tp->t_tcpreq_info[i]; 4663 if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4664 allocated = 1; 4665 fil->flags = TCP_TRK_TRACK_FLG_USED; 4666 fil->timestamp = req->timestamp; 4667 fil->playout_ms = req->playout_ms; 4668 fil->localtime = ts; 4669 fil->start = req->start; 4670 if (req->flags & TCP_LOG_HTTPD_RANGE_END) { 4671 fil->end = req->end; 4672 } else { 4673 fil->end = 0; 4674 fil->flags |= TCP_TRK_TRACK_FLG_OPEN; 4675 } 4676 /* 4677 * We can set the min boundaries to the TCP Sequence space, 4678 * but it might be found to be further up when sendfile 4679 * actually runs on this range (if it ever does). 4680 */ 4681 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc; 4682 fil->start_seq = tp->snd_una + 4683 tptosocket(tp)->so_snd.sb_ccc; 4684 if (req->flags & TCP_LOG_HTTPD_RANGE_END) 4685 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start))); 4686 else 4687 fil->end_seq = 0; 4688 if (tptosocket(tp)->so_snd.sb_tls_info) { 4689 /* 4690 * This session is doing TLS. Take a swag guess 4691 * at the overhead. 4692 */ 4693 fil->end_seq += tcp_estimate_tls_overhead( 4694 tptosocket(tp), (fil->end - fil->start)); 4695 } 4696 tp->t_tcpreq_req++; 4697 if (fil->flags & TCP_TRK_TRACK_FLG_OPEN) 4698 tp->t_tcpreq_open++; 4699 else 4700 tp->t_tcpreq_closed++; 4701 tcp_req_log_req_info(tp, fil, i, 4702 TCP_TRK_REQ_LOG_NEW, 0, 0); 4703 break; 4704 } else 4705 fil = NULL; 4706 } 4707 return (fil); 4708 } 4709 4710 void 4711 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts) 4712 { 4713 (void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1); 4714 } 4715 #endif 4716 4717 void 4718 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err) 4719 { 4720 if (tcp_bblogging_on(tp)) { 4721 struct tcp_log_buffer *l; 4722 4723 l = tcp_log_event(tp, NULL, 4724 &tptosocket(tp)->so_rcv, 4725 &tptosocket(tp)->so_snd, 4726 TCP_LOG_SOCKET_OPT, 4727 err, 0, NULL, 1, 4728 NULL, NULL, 0, NULL); 4729 if (l) { 4730 l->tlb_flex1 = option_num; 4731 l->tlb_flex2 = option_val; 4732 } 4733 } 4734 } 4735 4736 uint32_t 4737 tcp_get_srtt(struct tcpcb *tp, int granularity) 4738 { 4739 uint32_t srtt; 4740 4741 KASSERT(granularity == TCP_TMR_GRANULARITY_USEC || 4742 granularity == TCP_TMR_GRANULARITY_TICKS, 4743 ("%s: called with unexpected granularity %d", __func__, 4744 granularity)); 4745 4746 srtt = tp->t_srtt; 4747 4748 /* 4749 * We only support two granularities. If the stored granularity 4750 * does not match the granularity requested by the caller, 4751 * convert the stored value to the requested unit of granularity. 4752 */ 4753 if (tp->t_tmr_granularity != granularity) { 4754 if (granularity == TCP_TMR_GRANULARITY_USEC) 4755 srtt = TICKS_2_USEC(srtt); 4756 else 4757 srtt = USEC_2_TICKS(srtt); 4758 } 4759 4760 /* 4761 * If the srtt is stored with ticks granularity, we need to 4762 * unshift to get the actual value. We do this after the 4763 * conversion above (if one was necessary) in order to maximize 4764 * precision. 4765 */ 4766 if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS) 4767 srtt = srtt >> TCP_RTT_SHIFT; 4768 4769 return (srtt); 4770 } 4771 4772 void 4773 tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt, 4774 uint8_t is_tlp, bool hw_tls) 4775 { 4776 4777 if (is_tlp) { 4778 tp->t_sndtlppack++; 4779 tp->t_sndtlpbyte += len; 4780 } 4781 /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */ 4782 if (is_rxt) 4783 tp->t_snd_rxt_bytes += len; 4784 else 4785 tp->t_sndbytes += len; 4786 4787 #ifdef KERN_TLS 4788 if (hw_tls && is_rxt && len != 0) { 4789 uint64_t rexmit_percent; 4790 4791 rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) / 4792 (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes)); 4793 if (rexmit_percent > ktls_ifnet_max_rexmit_pct) 4794 ktls_disable_ifnet(tp); 4795 } 4796 #endif 4797 } 4798