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 "opt_inet.h" 33 #include "opt_inet6.h" 34 #include "opt_ipsec.h" 35 #include "opt_kern_tls.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/arb.h> 40 #include <sys/callout.h> 41 #include <sys/eventhandler.h> 42 #ifdef TCP_HHOOK 43 #include <sys/hhook.h> 44 #endif 45 #include <sys/kernel.h> 46 #ifdef TCP_HHOOK 47 #include <sys/khelp.h> 48 #endif 49 #ifdef KERN_TLS 50 #include <sys/ktls.h> 51 #endif 52 #include <sys/qmath.h> 53 #include <sys/stats.h> 54 #include <sys/sysctl.h> 55 #include <sys/jail.h> 56 #include <sys/malloc.h> 57 #include <sys/refcount.h> 58 #include <sys/mbuf.h> 59 #include <sys/priv.h> 60 #include <sys/sdt.h> 61 #include <sys/socket.h> 62 #include <sys/socketvar.h> 63 #include <sys/protosw.h> 64 #include <sys/random.h> 65 66 #include <vm/uma.h> 67 68 #include <net/route.h> 69 #include <net/route/nhop.h> 70 #include <net/if.h> 71 #include <net/if_var.h> 72 #include <net/if_private.h> 73 #include <net/vnet.h> 74 75 #include <netinet/in.h> 76 #include <netinet/in_fib.h> 77 #include <netinet/in_kdtrace.h> 78 #include <netinet/in_pcb.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/in_var.h> 81 #include <netinet/ip.h> 82 #include <netinet/ip_icmp.h> 83 #include <netinet/ip_var.h> 84 #include <netinet/icmp_var.h> 85 #ifdef INET6 86 #include <netinet/icmp6.h> 87 #include <netinet/ip6.h> 88 #include <netinet6/in6_fib.h> 89 #include <netinet6/in6_pcb.h> 90 #include <netinet6/ip6_var.h> 91 #include <netinet6/scope6_var.h> 92 #include <netinet6/nd6.h> 93 #endif 94 95 #include <netinet/tcp.h> 96 #ifdef INVARIANTS 97 #define TCPSTATES 98 #endif 99 #include <netinet/tcp_fsm.h> 100 #include <netinet/tcp_seq.h> 101 #include <netinet/tcp_timer.h> 102 #include <netinet/tcp_var.h> 103 #include <netinet/tcp_ecn.h> 104 #include <netinet/tcp_log_buf.h> 105 #include <netinet/tcp_syncache.h> 106 #include <netinet/tcp_hpts.h> 107 #include <netinet/tcp_lro.h> 108 #include <netinet/cc/cc.h> 109 #include <netinet/tcpip.h> 110 #include <netinet/tcp_fastopen.h> 111 #include <netinet/tcp_accounting.h> 112 #ifdef TCP_OFFLOAD 113 #include <netinet/tcp_offload.h> 114 #endif 115 #include <netinet/udp.h> 116 #include <netinet/udp_var.h> 117 #ifdef INET6 118 #include <netinet6/tcp6_var.h> 119 #endif 120 121 #include <netipsec/ipsec_support.h> 122 123 #include <machine/in_cksum.h> 124 #include <crypto/siphash/siphash.h> 125 126 #include <security/mac/mac_framework.h> 127 128 #ifdef INET6 129 static ip6proto_ctlinput_t tcp6_ctlinput; 130 static udp_tun_icmp_t tcp6_ctlinput_viaudp; 131 #endif 132 133 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 134 #ifdef INET6 135 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 136 #endif 137 138 VNET_DEFINE(uint32_t, tcp_ack_war_time_window) = 1000; 139 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_timewindow, 140 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ack_war_time_window), 0, 141 "Time interval in ms used to limit the number (ack_war_cnt) of challenge ACKs sent per TCP connection"); 142 VNET_DEFINE(uint32_t, tcp_ack_war_cnt) = 5; 143 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, ack_war_cnt, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_ack_war_cnt), 0, 145 "Maximum number of challenge ACKs sent per TCP connection during the time interval (ack_war_timewindow)"); 146 147 struct rwlock tcp_function_lock; 148 149 static int 150 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 151 { 152 int error, new; 153 154 new = V_tcp_mssdflt; 155 error = sysctl_handle_int(oidp, &new, 0, req); 156 if (error == 0 && req->newptr) { 157 if (new < TCP_MINMSS) 158 error = EINVAL; 159 else 160 V_tcp_mssdflt = new; 161 } 162 return (error); 163 } 164 165 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 166 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 167 &VNET_NAME(tcp_mssdflt), 0, &sysctl_net_inet_tcp_mss_check, "I", 168 "Default TCP Maximum Segment Size"); 169 170 #ifdef INET6 171 static int 172 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 173 { 174 int error, new; 175 176 new = V_tcp_v6mssdflt; 177 error = sysctl_handle_int(oidp, &new, 0, req); 178 if (error == 0 && req->newptr) { 179 if (new < TCP_MINMSS) 180 error = EINVAL; 181 else 182 V_tcp_v6mssdflt = new; 183 } 184 return (error); 185 } 186 187 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 188 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 189 &VNET_NAME(tcp_v6mssdflt), 0, &sysctl_net_inet_tcp_mss_v6_check, "I", 190 "Default TCP Maximum Segment Size for IPv6"); 191 #endif /* INET6 */ 192 193 /* 194 * Minimum MSS we accept and use. This prevents DoS attacks where 195 * we are forced to a ridiculous low MSS like 20 and send hundreds 196 * of packets instead of one. The effect scales with the available 197 * bandwidth and quickly saturates the CPU and network interface 198 * with packet generation and sending. Set to zero to disable MINMSS 199 * checking. This setting prevents us from sending too small packets. 200 */ 201 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(tcp_minmss), 0, 204 "Minimum TCP Maximum Segment Size"); 205 206 VNET_DEFINE(int, tcp_do_rfc1323) = 1; 207 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_do_rfc1323), 0, 209 "Enable rfc1323 (high performance TCP) extensions"); 210 211 /* 212 * As of June 2021, several TCP stacks violate RFC 7323 from September 2014. 213 * Some stacks negotiate TS, but never send them after connection setup. Some 214 * stacks negotiate TS, but don't send them when sending keep-alive segments. 215 * These include modern widely deployed TCP stacks. 216 * Therefore tolerating violations for now... 217 */ 218 VNET_DEFINE(int, tcp_tolerate_missing_ts) = 1; 219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tolerate_missing_ts, CTLFLAG_VNET | CTLFLAG_RW, 220 &VNET_NAME(tcp_tolerate_missing_ts), 0, 221 "Tolerate missing TCP timestamps"); 222 223 VNET_DEFINE(int, tcp_ts_offset_per_conn) = 1; 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ts_offset_per_conn, CTLFLAG_VNET | CTLFLAG_RW, 225 &VNET_NAME(tcp_ts_offset_per_conn), 0, 226 "Initialize TCP timestamps per connection instead of per host pair"); 227 228 /* How many connections are pacing */ 229 static volatile uint32_t number_of_tcp_connections_pacing = 0; 230 static uint32_t shadow_num_connections = 0; 231 static counter_u64_t tcp_pacing_failures; 232 static counter_u64_t tcp_dgp_failures; 233 static uint32_t shadow_tcp_pacing_dgp = 0; 234 static volatile uint32_t number_of_dgp_connections = 0; 235 236 static int tcp_pacing_limit = 10000; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pacing_limit, CTLFLAG_RW, 238 &tcp_pacing_limit, 1000, 239 "If the TCP stack does pacing, is there a limit (-1 = no, 0 = no pacing N = number of connections)"); 240 241 static int tcp_dgp_limit = -1; 242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, dgp_limit, CTLFLAG_RW, 243 &tcp_dgp_limit, -1, 244 "If the TCP stack does DGP, is there a limit (-1 = no, 0 = no dgp N = number of connections)"); 245 246 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pacing_count, CTLFLAG_RD, 247 &shadow_num_connections, 0, "Number of TCP connections being paced"); 248 249 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, pacing_failures, CTLFLAG_RD, 250 &tcp_pacing_failures, "Number of times we failed to enable pacing to avoid exceeding the limit"); 251 252 SYSCTL_COUNTER_U64(_net_inet_tcp, OID_AUTO, dgp_failures, CTLFLAG_RD, 253 &tcp_dgp_failures, "Number of times we failed to enable dgp to avoid exceeding the limit"); 254 255 static int tcp_log_debug = 0; 256 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 257 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 258 259 /* 260 * Target size of TCP PCB hash tables. Must be a power of two. 261 * 262 * Note that this can be overridden by the kernel environment 263 * variable net.inet.tcp.tcbhashsize 264 */ 265 #ifndef TCBHASHSIZE 266 #define TCBHASHSIZE 0 267 #endif 268 static int tcp_tcbhashsize = TCBHASHSIZE; 269 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 270 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 271 272 static int do_tcpdrain = 1; 273 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 274 "Enable tcp_drain routine for extra help when low on mbufs"); 275 276 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD, 277 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 278 279 VNET_DEFINE_STATIC(int, icmp_may_rst) = 1; 280 #define V_icmp_may_rst VNET(icmp_may_rst) 281 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW, 282 &VNET_NAME(icmp_may_rst), 0, 283 "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 284 285 VNET_DEFINE_STATIC(int, tcp_isn_reseed_interval) = 0; 286 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW, 288 &VNET_NAME(tcp_isn_reseed_interval), 0, 289 "Seconds between reseeding of ISN secret"); 290 291 static int tcp_soreceive_stream; 292 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 293 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 294 295 uma_zone_t tcp_sack_hole_zone; 296 297 VNET_DEFINE(uint32_t, tcp_map_entries_limit) = 0; /* unlimited */ 298 static int 299 sysctl_net_inet_tcp_map_limit_check(SYSCTL_HANDLER_ARGS) 300 { 301 int error; 302 uint32_t new; 303 304 new = V_tcp_map_entries_limit; 305 error = sysctl_handle_int(oidp, &new, 0, req); 306 if (error == 0 && req->newptr) { 307 /* only allow "0" and value > minimum */ 308 if (new > 0 && new < TCP_MIN_MAP_ENTRIES_LIMIT) 309 error = EINVAL; 310 else 311 V_tcp_map_entries_limit = new; 312 } 313 return (error); 314 } 315 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, map_limit, 316 CTLFLAG_VNET | CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 317 &VNET_NAME(tcp_map_entries_limit), 0, 318 &sysctl_net_inet_tcp_map_limit_check, "IU", 319 "Total sendmap entries limit"); 320 321 VNET_DEFINE(uint32_t, tcp_map_split_limit) = 0; /* unlimited */ 322 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, split_limit, CTLFLAG_VNET | CTLFLAG_RW, 323 &VNET_NAME(tcp_map_split_limit), 0, 324 "Total sendmap split entries limit"); 325 326 #ifdef TCP_HHOOK 327 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 328 #endif 329 330 #define TS_OFFSET_SECRET_LENGTH SIPHASH_KEY_LENGTH 331 VNET_DEFINE_STATIC(u_char, ts_offset_secret[TS_OFFSET_SECRET_LENGTH]); 332 #define V_ts_offset_secret VNET(ts_offset_secret) 333 334 static int tcp_default_fb_init(struct tcpcb *tp, void **ptr); 335 static void tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged); 336 static int tcp_default_handoff_ok(struct tcpcb *tp); 337 static struct inpcb *tcp_notify(struct inpcb *, int); 338 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 339 static struct inpcb *tcp_mtudisc(struct inpcb *, int); 340 static struct inpcb *tcp_drop_syn_sent(struct inpcb *, int); 341 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 342 const void *ip4hdr, const void *ip6hdr); 343 static void tcp_default_switch_failed(struct tcpcb *tp); 344 345 #ifdef INET 346 static ipproto_ctlinput_t tcp_ctlinput; 347 static udp_tun_icmp_t tcp_ctlinput_viaudp; 348 #endif 349 350 static struct tcp_function_block tcp_def_funcblk = { 351 .tfb_tcp_block_name = "freebsd", 352 .tfb_tcp_output = tcp_default_output, 353 .tfb_tcp_do_segment = tcp_do_segment, 354 .tfb_tcp_ctloutput = tcp_default_ctloutput, 355 .tfb_tcp_handoff_ok = tcp_default_handoff_ok, 356 .tfb_tcp_fb_init = tcp_default_fb_init, 357 .tfb_tcp_fb_fini = tcp_default_fb_fini, 358 .tfb_switch_failed = tcp_default_switch_failed, 359 .tfb_flags = TCP_FUNC_DEFAULT_OK, 360 }; 361 362 static int tcp_fb_cnt = 0; 363 struct tcp_funchead t_functions; 364 VNET_DEFINE_STATIC(struct tcp_function_block *, tcp_func_set_ptr) = &tcp_def_funcblk; 365 #define V_tcp_func_set_ptr VNET(tcp_func_set_ptr) 366 367 void 368 tcp_record_dsack(struct tcpcb *tp, tcp_seq start, tcp_seq end, int tlp) 369 { 370 TCPSTAT_INC(tcps_dsack_count); 371 tp->t_dsack_pack++; 372 if (tlp == 0) { 373 if (SEQ_GT(end, start)) { 374 tp->t_dsack_bytes += (end - start); 375 TCPSTAT_ADD(tcps_dsack_bytes, (end - start)); 376 } else { 377 tp->t_dsack_tlp_bytes += (start - end); 378 TCPSTAT_ADD(tcps_dsack_bytes, (start - end)); 379 } 380 } else { 381 if (SEQ_GT(end, start)) { 382 tp->t_dsack_bytes += (end - start); 383 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (end - start)); 384 } else { 385 tp->t_dsack_tlp_bytes += (start - end); 386 TCPSTAT_ADD(tcps_dsack_tlp_bytes, (start - end)); 387 } 388 } 389 } 390 391 static struct tcp_function_block * 392 find_tcp_functions_locked(struct tcp_function_set *fs) 393 { 394 struct tcp_function *f; 395 struct tcp_function_block *blk = NULL; 396 397 rw_assert(&tcp_function_lock, RA_LOCKED); 398 TAILQ_FOREACH(f, &t_functions, tf_next) { 399 if (strcmp(f->tf_name, fs->function_set_name) == 0) { 400 blk = f->tf_fb; 401 break; 402 } 403 } 404 return (blk); 405 } 406 407 static struct tcp_function_block * 408 find_tcp_fb_locked(struct tcp_function_block *blk, struct tcp_function **s) 409 { 410 struct tcp_function_block *rblk = NULL; 411 struct tcp_function *f; 412 413 rw_assert(&tcp_function_lock, RA_LOCKED); 414 TAILQ_FOREACH(f, &t_functions, tf_next) { 415 if (f->tf_fb == blk) { 416 rblk = blk; 417 if (s) { 418 *s = f; 419 } 420 break; 421 } 422 } 423 return (rblk); 424 } 425 426 struct tcp_function_block * 427 find_and_ref_tcp_functions(struct tcp_function_set *fs) 428 { 429 struct tcp_function_block *blk; 430 431 rw_rlock(&tcp_function_lock); 432 blk = find_tcp_functions_locked(fs); 433 if (blk) 434 refcount_acquire(&blk->tfb_refcnt); 435 rw_runlock(&tcp_function_lock); 436 return (blk); 437 } 438 439 struct tcp_function_block * 440 find_and_ref_tcp_fb(struct tcp_function_block *blk) 441 { 442 struct tcp_function_block *rblk; 443 444 rw_rlock(&tcp_function_lock); 445 rblk = find_tcp_fb_locked(blk, NULL); 446 if (rblk) 447 refcount_acquire(&rblk->tfb_refcnt); 448 rw_runlock(&tcp_function_lock); 449 return (rblk); 450 } 451 452 /* Find a matching alias for the given tcp_function_block. */ 453 int 454 find_tcp_function_alias(struct tcp_function_block *blk, 455 struct tcp_function_set *fs) 456 { 457 struct tcp_function *f; 458 int found; 459 460 found = 0; 461 rw_rlock(&tcp_function_lock); 462 TAILQ_FOREACH(f, &t_functions, tf_next) { 463 if ((f->tf_fb == blk) && 464 (strncmp(f->tf_name, blk->tfb_tcp_block_name, 465 TCP_FUNCTION_NAME_LEN_MAX) != 0)) { 466 /* Matching function block with different name. */ 467 strncpy(fs->function_set_name, f->tf_name, 468 TCP_FUNCTION_NAME_LEN_MAX); 469 found = 1; 470 break; 471 } 472 } 473 /* Null terminate the string appropriately. */ 474 if (found) { 475 fs->function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0'; 476 } else { 477 fs->function_set_name[0] = '\0'; 478 } 479 rw_runlock(&tcp_function_lock); 480 return (found); 481 } 482 483 static struct tcp_function_block * 484 find_and_ref_tcp_default_fb(void) 485 { 486 struct tcp_function_block *rblk; 487 488 rw_rlock(&tcp_function_lock); 489 rblk = V_tcp_func_set_ptr; 490 refcount_acquire(&rblk->tfb_refcnt); 491 rw_runlock(&tcp_function_lock); 492 return (rblk); 493 } 494 495 void 496 tcp_switch_back_to_default(struct tcpcb *tp) 497 { 498 struct tcp_function_block *tfb; 499 void *ptr = NULL; 500 501 KASSERT(tp->t_fb != &tcp_def_funcblk, 502 ("%s: called by the built-in default stack", __func__)); 503 504 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 505 tp->t_fb->tfb_tcp_timer_stop_all(tp); 506 507 /* 508 * Now, we'll find a new function block to use. 509 * Start by trying the current user-selected 510 * default, unless this stack is the user-selected 511 * default. 512 */ 513 tfb = find_and_ref_tcp_default_fb(); 514 if (tfb == tp->t_fb) { 515 refcount_release(&tfb->tfb_refcnt); 516 tfb = NULL; 517 } 518 /* Does the stack accept this connection? */ 519 if (tfb != NULL && (*tfb->tfb_tcp_handoff_ok)(tp)) { 520 refcount_release(&tfb->tfb_refcnt); 521 tfb = NULL; 522 } 523 /* Try to use that stack. */ 524 if (tfb != NULL) { 525 /* Initialize the new stack. If it succeeds, we are done. */ 526 if (tfb->tfb_tcp_fb_init == NULL || 527 (*tfb->tfb_tcp_fb_init)(tp, &ptr) == 0) { 528 /* Release the old stack */ 529 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 530 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 531 refcount_release(&tp->t_fb->tfb_refcnt); 532 /* Now set in all the pointers */ 533 tp->t_fb = tfb; 534 tp->t_fb_ptr = ptr; 535 return; 536 } 537 /* 538 * Initialization failed. Release the reference count on 539 * the looked up default stack. 540 */ 541 refcount_release(&tfb->tfb_refcnt); 542 } 543 544 /* 545 * If that wasn't feasible, use the built-in default 546 * stack which is not allowed to reject anyone. 547 */ 548 tfb = find_and_ref_tcp_fb(&tcp_def_funcblk); 549 if (tfb == NULL) { 550 /* there always should be a default */ 551 panic("Can't refer to tcp_def_funcblk"); 552 } 553 if ((*tfb->tfb_tcp_handoff_ok)(tp)) { 554 /* The default stack cannot say no */ 555 panic("Default stack rejects a new session?"); 556 } 557 if (tfb->tfb_tcp_fb_init != NULL && 558 (*tfb->tfb_tcp_fb_init)(tp, &ptr)) { 559 /* The default stack cannot fail */ 560 panic("Default stack initialization failed"); 561 } 562 /* Now release the old stack */ 563 if (tp->t_fb->tfb_tcp_fb_fini != NULL) 564 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0); 565 refcount_release(&tp->t_fb->tfb_refcnt); 566 /* And set in the pointers to the new */ 567 tp->t_fb = tfb; 568 tp->t_fb_ptr = ptr; 569 } 570 571 static bool 572 tcp_recv_udp_tunneled_packet(struct mbuf *m, int off, struct inpcb *inp, 573 const struct sockaddr *sa, void *ctx) 574 { 575 struct ip *iph; 576 #ifdef INET6 577 struct ip6_hdr *ip6; 578 #endif 579 struct udphdr *uh; 580 struct tcphdr *th; 581 int len, thlen; 582 uint16_t port; 583 584 TCPSTAT_INC(tcps_tunneled_pkts); 585 if ((m->m_flags & M_PKTHDR) == 0) { 586 /* Can't handle one that is not a pkt hdr */ 587 TCPSTAT_INC(tcps_tunneled_errs); 588 m_freem(m); 589 return (true); 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 return (true); 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 return (true); 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 - sizeof(struct udphdr)); 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 len = ntohs(iph->ip_len) - sizeof(struct udphdr); 627 if (__predict_false(len != m->m_pkthdr.len)) { 628 TCPSTAT_INC(tcps_tunneled_errs); 629 m_freem(m); 630 return (true); 631 } else { 632 iph->ip_len = htons(len); 633 tcp_input_with_port(&m, &off, IPPROTO_TCP, port); 634 } 635 break; 636 #endif 637 #ifdef INET6 638 case IPV6_VERSION >> 4: 639 ip6 = mtod(m, struct ip6_hdr *); 640 len = ntohs(ip6->ip6_plen) - sizeof(struct udphdr); 641 if (__predict_false(len + sizeof(struct ip6_hdr) != 642 m->m_pkthdr.len)) { 643 TCPSTAT_INC(tcps_tunneled_errs); 644 m_freem(m); 645 return (true); 646 } else { 647 ip6->ip6_plen = htons(len); 648 tcp6_input_with_port(&m, &off, IPPROTO_TCP, port); 649 } 650 break; 651 #endif 652 default: 653 m_freem(m); 654 break; 655 } 656 return (true); 657 } 658 659 static int 660 sysctl_net_inet_default_tcp_functions(SYSCTL_HANDLER_ARGS) 661 { 662 struct tcp_function_set fs; 663 struct tcp_function_block *blk; 664 int error; 665 666 memset(&fs, 0, sizeof(struct tcp_function_set)); 667 rw_rlock(&tcp_function_lock); 668 blk = find_tcp_fb_locked(V_tcp_func_set_ptr, NULL); 669 if (blk != NULL) { 670 /* Found him */ 671 strcpy(fs.function_set_name, blk->tfb_tcp_block_name); 672 fs.pcbcnt = blk->tfb_refcnt; 673 } 674 rw_runlock(&tcp_function_lock); 675 error = sysctl_handle_string(oidp, fs.function_set_name, 676 sizeof(fs.function_set_name), req); 677 678 /* Check for error or no change */ 679 if (error != 0 || req->newptr == NULL) 680 return (error); 681 682 rw_wlock(&tcp_function_lock); 683 blk = find_tcp_functions_locked(&fs); 684 if ((blk == NULL) || 685 (blk->tfb_flags & TCP_FUNC_BEING_REMOVED)) { 686 error = ENOENT; 687 goto done; 688 } 689 if ((blk->tfb_flags & TCP_FUNC_DEFAULT_OK) == 0) { 690 error = EINVAL; 691 goto done; 692 } 693 V_tcp_func_set_ptr = blk; 694 done: 695 rw_wunlock(&tcp_function_lock); 696 return (error); 697 } 698 699 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_default, 700 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 701 NULL, 0, sysctl_net_inet_default_tcp_functions, "A", 702 "Set/get the default TCP functions"); 703 704 static int 705 sysctl_net_inet_list_available(SYSCTL_HANDLER_ARGS) 706 { 707 int error, cnt, linesz; 708 struct tcp_function *f; 709 char *buffer, *cp; 710 size_t bufsz, outsz; 711 bool alias; 712 713 cnt = 0; 714 rw_rlock(&tcp_function_lock); 715 TAILQ_FOREACH(f, &t_functions, tf_next) { 716 cnt++; 717 } 718 rw_runlock(&tcp_function_lock); 719 720 bufsz = (cnt+2) * ((TCP_FUNCTION_NAME_LEN_MAX * 2) + 13) + 1; 721 buffer = malloc(bufsz, M_TEMP, M_WAITOK); 722 723 error = 0; 724 cp = buffer; 725 726 linesz = snprintf(cp, bufsz, "\n%-32s%c %-32s %s\n", "Stack", 'D', 727 "Alias", "PCB count"); 728 cp += linesz; 729 bufsz -= linesz; 730 outsz = linesz; 731 732 rw_rlock(&tcp_function_lock); 733 TAILQ_FOREACH(f, &t_functions, tf_next) { 734 alias = (f->tf_name != f->tf_fb->tfb_tcp_block_name); 735 linesz = snprintf(cp, bufsz, "%-32s%c %-32s %u\n", 736 f->tf_fb->tfb_tcp_block_name, 737 (f->tf_fb == V_tcp_func_set_ptr) ? '*' : ' ', 738 alias ? f->tf_name : "-", 739 f->tf_fb->tfb_refcnt); 740 if (linesz >= bufsz) { 741 error = EOVERFLOW; 742 break; 743 } 744 cp += linesz; 745 bufsz -= linesz; 746 outsz += linesz; 747 } 748 rw_runlock(&tcp_function_lock); 749 if (error == 0) 750 error = sysctl_handle_string(oidp, buffer, outsz + 1, req); 751 free(buffer, M_TEMP); 752 return (error); 753 } 754 755 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, functions_available, 756 CTLFLAG_VNET | CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 757 NULL, 0, sysctl_net_inet_list_available, "A", 758 "list available TCP Function sets"); 759 760 VNET_DEFINE(int, tcp_udp_tunneling_port) = TCP_TUNNELING_PORT_DEFAULT; 761 762 #ifdef INET 763 VNET_DEFINE(struct socket *, udp4_tun_socket) = NULL; 764 #define V_udp4_tun_socket VNET(udp4_tun_socket) 765 #endif 766 #ifdef INET6 767 VNET_DEFINE(struct socket *, udp6_tun_socket) = NULL; 768 #define V_udp6_tun_socket VNET(udp6_tun_socket) 769 #endif 770 771 static struct sx tcpoudp_lock; 772 773 static void 774 tcp_over_udp_stop(void) 775 { 776 777 sx_assert(&tcpoudp_lock, SA_XLOCKED); 778 779 #ifdef INET 780 if (V_udp4_tun_socket != NULL) { 781 soclose(V_udp4_tun_socket); 782 V_udp4_tun_socket = NULL; 783 } 784 #endif 785 #ifdef INET6 786 if (V_udp6_tun_socket != NULL) { 787 soclose(V_udp6_tun_socket); 788 V_udp6_tun_socket = NULL; 789 } 790 #endif 791 } 792 793 static int 794 tcp_over_udp_start(void) 795 { 796 uint16_t port; 797 int ret; 798 #ifdef INET 799 struct sockaddr_in sin; 800 #endif 801 #ifdef INET6 802 struct sockaddr_in6 sin6; 803 #endif 804 805 sx_assert(&tcpoudp_lock, SA_XLOCKED); 806 807 port = V_tcp_udp_tunneling_port; 808 if (ntohs(port) == 0) { 809 /* Must have a port set */ 810 return (EINVAL); 811 } 812 #ifdef INET 813 if (V_udp4_tun_socket != NULL) { 814 /* Already running -- must stop first */ 815 return (EALREADY); 816 } 817 #endif 818 #ifdef INET6 819 if (V_udp6_tun_socket != NULL) { 820 /* Already running -- must stop first */ 821 return (EALREADY); 822 } 823 #endif 824 #ifdef INET 825 if ((ret = socreate(PF_INET, &V_udp4_tun_socket, 826 SOCK_DGRAM, IPPROTO_UDP, 827 curthread->td_ucred, curthread))) { 828 tcp_over_udp_stop(); 829 return (ret); 830 } 831 /* Call the special UDP hook. */ 832 if ((ret = udp_set_kernel_tunneling(V_udp4_tun_socket, 833 tcp_recv_udp_tunneled_packet, 834 tcp_ctlinput_viaudp, 835 NULL))) { 836 tcp_over_udp_stop(); 837 return (ret); 838 } 839 /* Ok, we have a socket, bind it to the port. */ 840 memset(&sin, 0, sizeof(struct sockaddr_in)); 841 sin.sin_len = sizeof(struct sockaddr_in); 842 sin.sin_family = AF_INET; 843 sin.sin_port = htons(port); 844 if ((ret = sobind(V_udp4_tun_socket, 845 (struct sockaddr *)&sin, curthread))) { 846 tcp_over_udp_stop(); 847 return (ret); 848 } 849 #endif 850 #ifdef INET6 851 if ((ret = socreate(PF_INET6, &V_udp6_tun_socket, 852 SOCK_DGRAM, IPPROTO_UDP, 853 curthread->td_ucred, curthread))) { 854 tcp_over_udp_stop(); 855 return (ret); 856 } 857 /* Call the special UDP hook. */ 858 if ((ret = udp_set_kernel_tunneling(V_udp6_tun_socket, 859 tcp_recv_udp_tunneled_packet, 860 tcp6_ctlinput_viaudp, 861 NULL))) { 862 tcp_over_udp_stop(); 863 return (ret); 864 } 865 /* Ok, we have a socket, bind it to the port. */ 866 memset(&sin6, 0, sizeof(struct sockaddr_in6)); 867 sin6.sin6_len = sizeof(struct sockaddr_in6); 868 sin6.sin6_family = AF_INET6; 869 sin6.sin6_port = htons(port); 870 if ((ret = sobind(V_udp6_tun_socket, 871 (struct sockaddr *)&sin6, curthread))) { 872 tcp_over_udp_stop(); 873 return (ret); 874 } 875 #endif 876 return (0); 877 } 878 879 static int 880 sysctl_net_inet_tcp_udp_tunneling_port_check(SYSCTL_HANDLER_ARGS) 881 { 882 int error; 883 uint32_t old, new; 884 885 old = V_tcp_udp_tunneling_port; 886 new = old; 887 error = sysctl_handle_int(oidp, &new, 0, req); 888 if ((error == 0) && 889 (req->newptr != NULL)) { 890 if ((new < TCP_TUNNELING_PORT_MIN) || 891 (new > TCP_TUNNELING_PORT_MAX)) { 892 error = EINVAL; 893 } else { 894 sx_xlock(&tcpoudp_lock); 895 V_tcp_udp_tunneling_port = new; 896 if (old != 0) { 897 tcp_over_udp_stop(); 898 } 899 if (new != 0) { 900 error = tcp_over_udp_start(); 901 if (error != 0) { 902 V_tcp_udp_tunneling_port = 0; 903 } 904 } 905 sx_xunlock(&tcpoudp_lock); 906 } 907 } 908 return (error); 909 } 910 911 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_port, 912 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 913 &VNET_NAME(tcp_udp_tunneling_port), 914 0, &sysctl_net_inet_tcp_udp_tunneling_port_check, "IU", 915 "Tunneling port for tcp over udp"); 916 917 VNET_DEFINE(int, tcp_udp_tunneling_overhead) = TCP_TUNNELING_OVERHEAD_DEFAULT; 918 919 static int 920 sysctl_net_inet_tcp_udp_tunneling_overhead_check(SYSCTL_HANDLER_ARGS) 921 { 922 int error, new; 923 924 new = V_tcp_udp_tunneling_overhead; 925 error = sysctl_handle_int(oidp, &new, 0, req); 926 if (error == 0 && req->newptr) { 927 if ((new < TCP_TUNNELING_OVERHEAD_MIN) || 928 (new > TCP_TUNNELING_OVERHEAD_MAX)) 929 error = EINVAL; 930 else 931 V_tcp_udp_tunneling_overhead = new; 932 } 933 return (error); 934 } 935 936 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, udp_tunneling_overhead, 937 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 938 &VNET_NAME(tcp_udp_tunneling_overhead), 939 0, &sysctl_net_inet_tcp_udp_tunneling_overhead_check, "IU", 940 "MSS reduction when using tcp over udp"); 941 942 /* 943 * Exports one (struct tcp_function_info) for each alias/name. 944 */ 945 static int 946 sysctl_net_inet_list_func_info(SYSCTL_HANDLER_ARGS) 947 { 948 int cnt, error; 949 struct tcp_function *f; 950 struct tcp_function_info tfi; 951 952 /* 953 * We don't allow writes. 954 */ 955 if (req->newptr != NULL) 956 return (EINVAL); 957 958 /* 959 * Wire the old buffer so we can directly copy the functions to 960 * user space without dropping the lock. 961 */ 962 if (req->oldptr != NULL) { 963 error = sysctl_wire_old_buffer(req, 0); 964 if (error) 965 return (error); 966 } 967 968 /* 969 * Walk the list and copy out matching entries. If INVARIANTS 970 * is compiled in, also walk the list to verify the length of 971 * the list matches what we have recorded. 972 */ 973 rw_rlock(&tcp_function_lock); 974 975 cnt = 0; 976 #ifndef INVARIANTS 977 if (req->oldptr == NULL) { 978 cnt = tcp_fb_cnt; 979 goto skip_loop; 980 } 981 #endif 982 TAILQ_FOREACH(f, &t_functions, tf_next) { 983 #ifdef INVARIANTS 984 cnt++; 985 #endif 986 if (req->oldptr != NULL) { 987 bzero(&tfi, sizeof(tfi)); 988 tfi.tfi_refcnt = f->tf_fb->tfb_refcnt; 989 tfi.tfi_id = f->tf_fb->tfb_id; 990 (void)strlcpy(tfi.tfi_alias, f->tf_name, 991 sizeof(tfi.tfi_alias)); 992 (void)strlcpy(tfi.tfi_name, 993 f->tf_fb->tfb_tcp_block_name, sizeof(tfi.tfi_name)); 994 error = SYSCTL_OUT(req, &tfi, sizeof(tfi)); 995 /* 996 * Don't stop on error, as that is the 997 * mechanism we use to accumulate length 998 * information if the buffer was too short. 999 */ 1000 } 1001 } 1002 KASSERT(cnt == tcp_fb_cnt, 1003 ("%s: cnt (%d) != tcp_fb_cnt (%d)", __func__, cnt, tcp_fb_cnt)); 1004 #ifndef INVARIANTS 1005 skip_loop: 1006 #endif 1007 rw_runlock(&tcp_function_lock); 1008 if (req->oldptr == NULL) 1009 error = SYSCTL_OUT(req, NULL, 1010 (cnt + 1) * sizeof(struct tcp_function_info)); 1011 1012 return (error); 1013 } 1014 1015 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, function_info, 1016 CTLTYPE_OPAQUE | CTLFLAG_SKIP | CTLFLAG_RD | CTLFLAG_MPSAFE, 1017 NULL, 0, sysctl_net_inet_list_func_info, "S,tcp_function_info", 1018 "List TCP function block name-to-ID mappings"); 1019 1020 /* 1021 * tfb_tcp_handoff_ok() function for the default stack. 1022 * Note that we'll basically try to take all comers. 1023 */ 1024 static int 1025 tcp_default_handoff_ok(struct tcpcb *tp) 1026 { 1027 1028 return (0); 1029 } 1030 1031 /* 1032 * tfb_tcp_fb_init() function for the default stack. 1033 * 1034 * This handles making sure we have appropriate timers set if you are 1035 * transitioning a socket that has some amount of setup done. 1036 * 1037 * The init() fuction from the default can *never* return non-zero i.e. 1038 * it is required to always succeed since it is the stack of last resort! 1039 */ 1040 static int 1041 tcp_default_fb_init(struct tcpcb *tp, void **ptr) 1042 { 1043 struct socket *so = tptosocket(tp); 1044 int rexmt; 1045 1046 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1047 /* We don't use the pointer */ 1048 *ptr = NULL; 1049 1050 /* Make sure we get no interesting mbuf queuing behavior */ 1051 /* All mbuf queue/ack compress flags should be off */ 1052 tcp_lro_features_off(tp); 1053 1054 /* Cancel the GP measurement in progress */ 1055 tp->t_flags &= ~TF_GPUTINPROG; 1056 /* Validate the timers are not in usec, if they are convert */ 1057 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 1058 if ((tp->t_state == TCPS_SYN_SENT) || 1059 (tp->t_state == TCPS_SYN_RECEIVED)) 1060 rexmt = tcp_rexmit_initial * tcp_backoff[tp->t_rxtshift]; 1061 else 1062 rexmt = TCP_REXMTVAL(tp) * tcp_backoff[tp->t_rxtshift]; 1063 if (tp->t_rxtshift == 0) 1064 tp->t_rxtcur = rexmt; 1065 else 1066 TCPT_RANGESET(tp->t_rxtcur, rexmt, tp->t_rttmin, 1067 tcp_rexmit_max); 1068 1069 /* 1070 * Nothing to do for ESTABLISHED or LISTEN states. And, we don't 1071 * know what to do for unexpected states (which includes TIME_WAIT). 1072 */ 1073 if (tp->t_state <= TCPS_LISTEN || tp->t_state >= TCPS_TIME_WAIT) 1074 return (0); 1075 1076 /* 1077 * Make sure some kind of transmission timer is set if there is 1078 * outstanding data. 1079 */ 1080 if ((!TCPS_HAVEESTABLISHED(tp->t_state) || sbavail(&so->so_snd) || 1081 tp->snd_una != tp->snd_max) && !(tcp_timer_active(tp, TT_REXMT) || 1082 tcp_timer_active(tp, TT_PERSIST))) { 1083 /* 1084 * If the session has established and it looks like it should 1085 * be in the persist state, set the persist timer. Otherwise, 1086 * set the retransmit timer. 1087 */ 1088 if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->snd_wnd == 0 && 1089 (int32_t)(tp->snd_nxt - tp->snd_una) < 1090 (int32_t)sbavail(&so->so_snd)) 1091 tcp_setpersist(tp); 1092 else 1093 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 1094 } 1095 1096 /* All non-embryonic sessions get a keepalive timer. */ 1097 if (!tcp_timer_active(tp, TT_KEEP)) 1098 tcp_timer_activate(tp, TT_KEEP, 1099 TCPS_HAVEESTABLISHED(tp->t_state) ? TP_KEEPIDLE(tp) : 1100 TP_KEEPINIT(tp)); 1101 1102 /* 1103 * Make sure critical variables are initialized 1104 * if transitioning while in Recovery. 1105 */ 1106 if IN_FASTRECOVERY(tp->t_flags) { 1107 if (tp->sackhint.recover_fs == 0) 1108 tp->sackhint.recover_fs = max(1, 1109 tp->snd_nxt - tp->snd_una); 1110 } 1111 1112 return (0); 1113 } 1114 1115 /* 1116 * tfb_tcp_fb_fini() function for the default stack. 1117 * 1118 * This changes state as necessary (or prudent) to prepare for another stack 1119 * to assume responsibility for the connection. 1120 */ 1121 static void 1122 tcp_default_fb_fini(struct tcpcb *tp, int tcb_is_purged) 1123 { 1124 1125 INP_WLOCK_ASSERT(tptoinpcb(tp)); 1126 1127 #ifdef TCP_BLACKBOX 1128 tcp_log_flowend(tp); 1129 #endif 1130 tp->t_acktime = 0; 1131 return; 1132 } 1133 1134 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 1135 MALLOC_DEFINE(M_TCPFUNCTIONS, "tcpfunc", "TCP function set memory"); 1136 #ifdef TCP_REQUEST_TRK 1137 MALLOC_DEFINE(M_TCPREQTRK, "tcpreqtrk", "TCP request tracking"); 1138 #endif 1139 1140 static struct mtx isn_mtx; 1141 1142 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1143 #define ISN_LOCK() mtx_lock(&isn_mtx) 1144 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1145 1146 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcphash"); 1147 1148 /* 1149 * Take a value and get the next power of 2 that doesn't overflow. 1150 * Used to size the tcp_inpcb hash buckets. 1151 */ 1152 static int 1153 maketcp_hashsize(int size) 1154 { 1155 int hashsize; 1156 1157 /* 1158 * auto tune. 1159 * get the next power of 2 higher than maxsockets. 1160 */ 1161 hashsize = 1 << fls(size); 1162 /* catch overflow, and just go one power of 2 smaller */ 1163 if (hashsize < size) { 1164 hashsize = 1 << (fls(size) - 1); 1165 } 1166 return (hashsize); 1167 } 1168 1169 static volatile int next_tcp_stack_id = 1; 1170 1171 /* 1172 * Register a TCP function block with the name provided in the names 1173 * array. (Note that this function does NOT automatically register 1174 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1175 * explicitly include blk->tfb_tcp_block_name in the list of names if 1176 * you wish to register the stack with that name.) 1177 * 1178 * Either all name registrations will succeed or all will fail. If 1179 * a name registration fails, the function will update the num_names 1180 * argument to point to the array index of the name that encountered 1181 * the failure. 1182 * 1183 * Returns 0 on success, or an error code on failure. 1184 */ 1185 int 1186 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1187 const char *names[], int *num_names) 1188 { 1189 struct tcp_function *f[TCP_FUNCTION_NAME_NUM_MAX]; 1190 struct tcp_function_set fs; 1191 int error, i, num_registered; 1192 1193 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1194 KASSERT(*num_names > 0, 1195 ("%s: Called with non-positive length of name list", __func__)); 1196 KASSERT(rw_initialized(&tcp_function_lock), 1197 ("%s: called too early", __func__)); 1198 1199 if (*num_names > TCP_FUNCTION_NAME_NUM_MAX) { 1200 /* Too many names. */ 1201 *num_names = 0; 1202 return (E2BIG); 1203 } 1204 if ((blk->tfb_tcp_output == NULL) || 1205 (blk->tfb_tcp_do_segment == NULL) || 1206 (blk->tfb_tcp_ctloutput == NULL) || 1207 (blk->tfb_tcp_handoff_ok == NULL) || 1208 (strlen(blk->tfb_tcp_block_name) == 0)) { 1209 /* These functions are required and a name is needed. */ 1210 *num_names = 0; 1211 return (EINVAL); 1212 } 1213 1214 for (i = 0; i < *num_names; i++) { 1215 f[i] = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1216 if (f[i] == NULL) { 1217 while (--i >= 0) 1218 free(f[i], M_TCPFUNCTIONS); 1219 *num_names = 0; 1220 return (ENOMEM); 1221 } 1222 } 1223 1224 num_registered = 0; 1225 rw_wlock(&tcp_function_lock); 1226 if (find_tcp_fb_locked(blk, NULL) != NULL) { 1227 /* A TCP function block can only be registered once. */ 1228 error = EALREADY; 1229 goto cleanup; 1230 } 1231 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1232 error = EINVAL; 1233 goto cleanup; 1234 } 1235 refcount_init(&blk->tfb_refcnt, 0); 1236 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1237 for (i = 0; i < *num_names; i++) { 1238 (void)strlcpy(fs.function_set_name, names[i], 1239 sizeof(fs.function_set_name)); 1240 if (find_tcp_functions_locked(&fs) != NULL) { 1241 /* Duplicate name space not allowed */ 1242 error = EALREADY; 1243 goto cleanup; 1244 } 1245 f[i]->tf_fb = blk; 1246 (void)strlcpy(f[i]->tf_name, names[i], sizeof(f[i]->tf_name)); 1247 TAILQ_INSERT_TAIL(&t_functions, f[i], tf_next); 1248 tcp_fb_cnt++; 1249 num_registered++; 1250 } 1251 rw_wunlock(&tcp_function_lock); 1252 return (0); 1253 1254 cleanup: 1255 /* Remove the entries just added. */ 1256 for (i = 0; i < *num_names; i++) { 1257 if (i < num_registered) { 1258 TAILQ_REMOVE(&t_functions, f[i], tf_next); 1259 tcp_fb_cnt--; 1260 } 1261 f[i]->tf_fb = NULL; 1262 free(f[i], M_TCPFUNCTIONS); 1263 } 1264 rw_wunlock(&tcp_function_lock); 1265 *num_names = num_registered; 1266 return (error); 1267 } 1268 1269 /* 1270 * Register a TCP function block using the name provided in the name 1271 * argument. 1272 * 1273 * Returns 0 on success, or an error code on failure. 1274 */ 1275 int 1276 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1277 int wait) 1278 { 1279 const char *name_list[1]; 1280 int num_names, rv; 1281 1282 num_names = 1; 1283 if (name != NULL) 1284 name_list[0] = name; 1285 else 1286 name_list[0] = blk->tfb_tcp_block_name; 1287 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1288 return (rv); 1289 } 1290 1291 /* 1292 * Register a TCP function block using the name defined in 1293 * blk->tfb_tcp_block_name. 1294 * 1295 * Returns 0 on success, or an error code on failure. 1296 */ 1297 int 1298 register_tcp_functions(struct tcp_function_block *blk, int wait) 1299 { 1300 1301 return (register_tcp_functions_as_name(blk, NULL, wait)); 1302 } 1303 1304 /* 1305 * Deregister all names associated with a function block. This 1306 * functionally removes the function block from use within the system. 1307 * 1308 * When called with a true quiesce argument, mark the function block 1309 * as being removed so no more stacks will use it and determine 1310 * whether the removal would succeed. 1311 * 1312 * When called with a false quiesce argument, actually attempt the 1313 * removal. 1314 * 1315 * When called with a force argument, attempt to switch all TCBs to 1316 * use the default stack instead of returning EBUSY. 1317 * 1318 * Returns 0 on success (or if the removal would succeed), or an error 1319 * code on failure. 1320 */ 1321 int 1322 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1323 bool force) 1324 { 1325 struct tcp_function *f; 1326 VNET_ITERATOR_DECL(vnet_iter); 1327 1328 if (blk == &tcp_def_funcblk) { 1329 /* You can't un-register the default */ 1330 return (EPERM); 1331 } 1332 rw_wlock(&tcp_function_lock); 1333 VNET_LIST_RLOCK_NOSLEEP(); 1334 VNET_FOREACH(vnet_iter) { 1335 CURVNET_SET(vnet_iter); 1336 if (blk == V_tcp_func_set_ptr) { 1337 /* You can't free the current default in some vnet. */ 1338 CURVNET_RESTORE(); 1339 VNET_LIST_RUNLOCK_NOSLEEP(); 1340 rw_wunlock(&tcp_function_lock); 1341 return (EBUSY); 1342 } 1343 CURVNET_RESTORE(); 1344 } 1345 VNET_LIST_RUNLOCK_NOSLEEP(); 1346 /* Mark the block so no more stacks can use it. */ 1347 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1348 /* 1349 * If TCBs are still attached to the stack, attempt to switch them 1350 * to the default stack. 1351 */ 1352 if (force && blk->tfb_refcnt) { 1353 struct inpcb *inp; 1354 struct tcpcb *tp; 1355 VNET_ITERATOR_DECL(vnet_iter); 1356 1357 rw_wunlock(&tcp_function_lock); 1358 1359 VNET_LIST_RLOCK(); 1360 VNET_FOREACH(vnet_iter) { 1361 CURVNET_SET(vnet_iter); 1362 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1363 INPLOOKUP_WLOCKPCB); 1364 1365 while ((inp = inp_next(&inpi)) != NULL) { 1366 tp = intotcpcb(inp); 1367 if (tp == NULL || tp->t_fb != blk) 1368 continue; 1369 tcp_switch_back_to_default(tp); 1370 } 1371 CURVNET_RESTORE(); 1372 } 1373 VNET_LIST_RUNLOCK(); 1374 1375 rw_wlock(&tcp_function_lock); 1376 } 1377 if (blk->tfb_refcnt) { 1378 /* TCBs still attached. */ 1379 rw_wunlock(&tcp_function_lock); 1380 return (EBUSY); 1381 } 1382 if (quiesce) { 1383 /* Skip removal. */ 1384 rw_wunlock(&tcp_function_lock); 1385 return (0); 1386 } 1387 /* Remove any function names that map to this function block. */ 1388 while (find_tcp_fb_locked(blk, &f) != NULL) { 1389 TAILQ_REMOVE(&t_functions, f, tf_next); 1390 tcp_fb_cnt--; 1391 f->tf_fb = NULL; 1392 free(f, M_TCPFUNCTIONS); 1393 } 1394 rw_wunlock(&tcp_function_lock); 1395 return (0); 1396 } 1397 1398 static void 1399 tcp_drain(void *ctx __unused, int flags __unused) 1400 { 1401 struct epoch_tracker et; 1402 VNET_ITERATOR_DECL(vnet_iter); 1403 1404 if (!do_tcpdrain) 1405 return; 1406 1407 NET_EPOCH_ENTER(et); 1408 VNET_LIST_RLOCK_NOSLEEP(); 1409 VNET_FOREACH(vnet_iter) { 1410 CURVNET_SET(vnet_iter); 1411 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1412 INPLOOKUP_WLOCKPCB); 1413 struct inpcb *inpb; 1414 struct tcpcb *tcpb; 1415 1416 /* 1417 * Walk the tcpbs, if existing, and flush the reassembly queue, 1418 * if there is one... 1419 * XXX: The "Net/3" implementation doesn't imply that the TCP 1420 * reassembly queue should be flushed, but in a situation 1421 * where we're really low on mbufs, this is potentially 1422 * useful. 1423 */ 1424 while ((inpb = inp_next(&inpi)) != NULL) { 1425 if ((tcpb = intotcpcb(inpb)) != NULL) { 1426 tcp_reass_flush(tcpb); 1427 tcp_clean_sackreport(tcpb); 1428 #ifdef TCP_BLACKBOX 1429 tcp_log_drain(tcpb); 1430 #endif 1431 } 1432 } 1433 CURVNET_RESTORE(); 1434 } 1435 VNET_LIST_RUNLOCK_NOSLEEP(); 1436 NET_EPOCH_EXIT(et); 1437 } 1438 1439 static void 1440 tcp_vnet_init(void *arg __unused) 1441 { 1442 1443 #ifdef TCP_HHOOK 1444 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1445 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1446 printf("%s: WARNING: unable to register helper hook\n", __func__); 1447 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1448 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1449 printf("%s: WARNING: unable to register helper hook\n", __func__); 1450 #endif 1451 #ifdef STATS 1452 if (tcp_stats_init()) 1453 printf("%s: WARNING: unable to initialise TCP stats\n", 1454 __func__); 1455 #endif 1456 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1457 tcp_tcbhashsize, tcp_tcbhashsize); 1458 1459 syncache_init(); 1460 tcp_hc_init(); 1461 1462 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1463 1464 tcp_fastopen_vnet_init(); 1465 1466 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1467 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1468 1469 V_tcp_msl = TCPTV_MSL; 1470 V_tcp_msl_local = TCPTV_MSL_LOCAL; 1471 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1472 } 1473 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1474 tcp_vnet_init, NULL); 1475 1476 static void 1477 tcp_init(void *arg __unused) 1478 { 1479 int hashsize; 1480 1481 tcp_reass_global_init(); 1482 1483 /* XXX virtualize those below? */ 1484 tcp_delacktime = TCPTV_DELACK; 1485 tcp_keepinit = TCPTV_KEEP_INIT; 1486 tcp_keepidle = TCPTV_KEEP_IDLE; 1487 tcp_keepintvl = TCPTV_KEEPINTVL; 1488 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1489 tcp_rexmit_initial = TCPTV_RTOBASE; 1490 tcp_rexmit_min = TCPTV_MIN; 1491 tcp_rexmit_max = TCPTV_REXMTMAX; 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(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1524 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1525 1526 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1527 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1528 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1529 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1530 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1531 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1532 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1533 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1534 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1535 tcp_pacing_failures = counter_u64_alloc(M_WAITOK); 1536 tcp_dgp_failures = counter_u64_alloc(M_WAITOK); 1537 1538 hashsize = tcp_tcbhashsize; 1539 if (hashsize == 0) { 1540 /* 1541 * Auto tune the hash size based on maxsockets. 1542 * A perfect hash would have a 1:1 mapping 1543 * (hashsize = maxsockets) however it's been 1544 * suggested that O(2) average is better. 1545 */ 1546 hashsize = maketcp_hashsize(maxsockets / 4); 1547 /* 1548 * Our historical default is 512, 1549 * do not autotune lower than this. 1550 */ 1551 if (hashsize < 512) 1552 hashsize = 512; 1553 if (bootverbose) 1554 printf("%s: %s auto tuned to %d\n", __func__, 1555 "net.inet.tcp.tcbhashsize", hashsize); 1556 } 1557 /* 1558 * We require a hashsize to be a power of two. 1559 * Previously if it was not a power of two we would just reset it 1560 * back to 512, which could be a nasty surprise if you did not notice 1561 * the error message. 1562 * Instead what we do is clip it to the closest power of two lower 1563 * than the specified hash value. 1564 */ 1565 if (!powerof2(hashsize)) { 1566 int oldhashsize = hashsize; 1567 1568 hashsize = maketcp_hashsize(hashsize); 1569 /* prevent absurdly low value */ 1570 if (hashsize < 16) 1571 hashsize = 16; 1572 printf("%s: WARNING: TCB hash size not a power of 2, " 1573 "clipped from %d to %d.\n", __func__, oldhashsize, 1574 hashsize); 1575 } 1576 tcp_tcbhashsize = hashsize; 1577 1578 #ifdef INET 1579 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1580 #endif 1581 #ifdef INET6 1582 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1583 #endif 1584 1585 tcp_fastopen_init(); 1586 tcp_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1587 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1588 } 1589 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1590 1591 #ifdef VIMAGE 1592 static void 1593 tcp_vnet_destroy(void *unused __unused) 1594 { 1595 #ifdef TCP_HHOOK 1596 int error; 1597 #endif 1598 1599 tcp_hc_destroy(); 1600 syncache_destroy(); 1601 in_pcbinfo_destroy(&V_tcbinfo); 1602 1603 /* 1604 * Cannot free the zone until all tcpcbs are released as we attach 1605 * the allocations to them. 1606 */ 1607 tcp_fastopen_vnet_destroy(); 1608 1609 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1610 VNET_PCPUSTAT_FREE(tcpstat); 1611 1612 #ifdef TCP_HHOOK 1613 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1614 if (error != 0) { 1615 printf("%s: WARNING: unable to deregister helper hook " 1616 "type=%d, id=%d: error %d returned\n", __func__, 1617 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1618 } 1619 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1620 if (error != 0) { 1621 printf("%s: WARNING: unable to deregister helper hook " 1622 "type=%d, id=%d: error %d returned\n", __func__, 1623 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1624 } 1625 #endif 1626 } 1627 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_vnet_destroy, 1628 NULL); 1629 #endif 1630 1631 /* 1632 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1633 * tcp_template used to store this data in mbufs, but we now recopy it out 1634 * of the tcpcb each time to conserve mbufs. 1635 */ 1636 void 1637 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1638 { 1639 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1640 1641 INP_WLOCK_ASSERT(inp); 1642 1643 #ifdef INET6 1644 if ((inp->inp_vflag & INP_IPV6) != 0) { 1645 struct ip6_hdr *ip6; 1646 1647 ip6 = (struct ip6_hdr *)ip_ptr; 1648 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1649 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1650 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1651 (IPV6_VERSION & IPV6_VERSION_MASK); 1652 if (port == 0) 1653 ip6->ip6_nxt = IPPROTO_TCP; 1654 else 1655 ip6->ip6_nxt = IPPROTO_UDP; 1656 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1657 ip6->ip6_src = inp->in6p_laddr; 1658 ip6->ip6_dst = inp->in6p_faddr; 1659 } 1660 #endif /* INET6 */ 1661 #if defined(INET6) && defined(INET) 1662 else 1663 #endif 1664 #ifdef INET 1665 { 1666 struct ip *ip; 1667 1668 ip = (struct ip *)ip_ptr; 1669 ip->ip_v = IPVERSION; 1670 ip->ip_hl = 5; 1671 ip->ip_tos = inp->inp_ip_tos; 1672 ip->ip_len = 0; 1673 ip->ip_id = 0; 1674 ip->ip_off = 0; 1675 ip->ip_ttl = inp->inp_ip_ttl; 1676 ip->ip_sum = 0; 1677 if (port == 0) 1678 ip->ip_p = IPPROTO_TCP; 1679 else 1680 ip->ip_p = IPPROTO_UDP; 1681 ip->ip_src = inp->inp_laddr; 1682 ip->ip_dst = inp->inp_faddr; 1683 } 1684 #endif /* INET */ 1685 th->th_sport = inp->inp_lport; 1686 th->th_dport = inp->inp_fport; 1687 th->th_seq = 0; 1688 th->th_ack = 0; 1689 th->th_off = 5; 1690 tcp_set_flags(th, 0); 1691 th->th_win = 0; 1692 th->th_urp = 0; 1693 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1694 } 1695 1696 /* 1697 * Create template to be used to send tcp packets on a connection. 1698 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1699 * use for this function is in keepalives, which use tcp_respond. 1700 */ 1701 struct tcptemp * 1702 tcpip_maketemplate(struct inpcb *inp) 1703 { 1704 struct tcptemp *t; 1705 1706 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1707 if (t == NULL) 1708 return (NULL); 1709 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1710 return (t); 1711 } 1712 1713 /* 1714 * Send a single message to the TCP at address specified by 1715 * the given TCP/IP header. If m == NULL, then we make a copy 1716 * of the tcpiphdr at th and send directly to the addressed host. 1717 * This is used to force keep alive messages out using the TCP 1718 * template for a connection. If flags are given then we send 1719 * a message back to the TCP which originated the segment th, 1720 * and discard the mbuf containing it and any other attached mbufs. 1721 * 1722 * In any case the ack and sequence number of the transmitted 1723 * segment are as specified by the parameters. 1724 * 1725 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1726 */ 1727 1728 void 1729 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1730 tcp_seq ack, tcp_seq seq, uint16_t flags) 1731 { 1732 struct tcpopt to; 1733 struct inpcb *inp; 1734 struct ip *ip; 1735 struct mbuf *optm; 1736 struct udphdr *uh = NULL; 1737 struct tcphdr *nth; 1738 struct tcp_log_buffer *lgb; 1739 u_char *optp; 1740 #ifdef INET6 1741 struct ip6_hdr *ip6; 1742 int isipv6; 1743 #endif /* INET6 */ 1744 int optlen, tlen, win, ulen; 1745 int ect = 0; 1746 bool incl_opts; 1747 uint16_t port; 1748 int output_ret; 1749 #ifdef INVARIANTS 1750 int thflags = tcp_get_flags(th); 1751 #endif 1752 1753 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1754 NET_EPOCH_ASSERT(); 1755 1756 #ifdef INET6 1757 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1758 ip6 = ipgen; 1759 #endif /* INET6 */ 1760 ip = ipgen; 1761 1762 if (tp != NULL) { 1763 inp = tptoinpcb(tp); 1764 INP_LOCK_ASSERT(inp); 1765 } else 1766 inp = NULL; 1767 1768 if (m != NULL) { 1769 #ifdef INET6 1770 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1771 port = m->m_pkthdr.tcp_tun_port; 1772 else 1773 #endif 1774 if (ip && (ip->ip_p == IPPROTO_UDP)) 1775 port = m->m_pkthdr.tcp_tun_port; 1776 else 1777 port = 0; 1778 } else 1779 port = tp->t_port; 1780 1781 incl_opts = false; 1782 win = 0; 1783 if (tp != NULL) { 1784 if (!(flags & TH_RST)) { 1785 win = sbspace(&inp->inp_socket->so_rcv); 1786 if (win > TCP_MAXWIN << tp->rcv_scale) 1787 win = TCP_MAXWIN << tp->rcv_scale; 1788 } 1789 if ((tp->t_flags & TF_NOOPT) == 0) 1790 incl_opts = true; 1791 } 1792 if (m == NULL) { 1793 m = m_gethdr(M_NOWAIT, MT_DATA); 1794 if (m == NULL) 1795 return; 1796 m->m_data += max_linkhdr; 1797 #ifdef INET6 1798 if (isipv6) { 1799 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1800 sizeof(struct ip6_hdr)); 1801 ip6 = mtod(m, struct ip6_hdr *); 1802 nth = (struct tcphdr *)(ip6 + 1); 1803 if (port) { 1804 /* Insert a UDP header */ 1805 uh = (struct udphdr *)nth; 1806 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1807 uh->uh_dport = port; 1808 nth = (struct tcphdr *)(uh + 1); 1809 } 1810 } else 1811 #endif /* INET6 */ 1812 { 1813 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1814 ip = mtod(m, struct ip *); 1815 nth = (struct tcphdr *)(ip + 1); 1816 if (port) { 1817 /* Insert a UDP header */ 1818 uh = (struct udphdr *)nth; 1819 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1820 uh->uh_dport = port; 1821 nth = (struct tcphdr *)(uh + 1); 1822 } 1823 } 1824 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1825 flags = TH_ACK; 1826 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1827 struct mbuf *n; 1828 1829 /* Can't reuse 'm', allocate a new mbuf. */ 1830 n = m_gethdr(M_NOWAIT, MT_DATA); 1831 if (n == NULL) { 1832 m_freem(m); 1833 return; 1834 } 1835 1836 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1837 m_freem(m); 1838 m_freem(n); 1839 return; 1840 } 1841 1842 n->m_data += max_linkhdr; 1843 /* m_len is set later */ 1844 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1845 #ifdef INET6 1846 if (isipv6) { 1847 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1848 sizeof(struct ip6_hdr)); 1849 ip6 = mtod(n, struct ip6_hdr *); 1850 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1851 nth = (struct tcphdr *)(ip6 + 1); 1852 if (port) { 1853 /* Insert a UDP header */ 1854 uh = (struct udphdr *)nth; 1855 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1856 uh->uh_dport = port; 1857 nth = (struct tcphdr *)(uh + 1); 1858 } 1859 } else 1860 #endif /* INET6 */ 1861 { 1862 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1863 ip = mtod(n, struct ip *); 1864 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1865 nth = (struct tcphdr *)(ip + 1); 1866 if (port) { 1867 /* Insert a UDP header */ 1868 uh = (struct udphdr *)nth; 1869 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1870 uh->uh_dport = port; 1871 nth = (struct tcphdr *)(uh + 1); 1872 } 1873 } 1874 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1875 xchg(nth->th_dport, nth->th_sport, uint16_t); 1876 th = nth; 1877 m_freem(m); 1878 m = n; 1879 } else { 1880 /* 1881 * reuse the mbuf. 1882 * XXX MRT We inherit the FIB, which is lucky. 1883 */ 1884 m_freem(m->m_next); 1885 m->m_next = NULL; 1886 m->m_data = (caddr_t)ipgen; 1887 /* clear any receive flags for proper bpf timestamping */ 1888 m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO); 1889 /* m_len is set later */ 1890 #ifdef INET6 1891 if (isipv6) { 1892 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1893 nth = (struct tcphdr *)(ip6 + 1); 1894 } else 1895 #endif /* INET6 */ 1896 { 1897 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1898 nth = (struct tcphdr *)(ip + 1); 1899 } 1900 if (th != nth) { 1901 /* 1902 * this is usually a case when an extension header 1903 * exists between the IPv6 header and the 1904 * TCP header. 1905 */ 1906 nth->th_sport = th->th_sport; 1907 nth->th_dport = th->th_dport; 1908 } 1909 xchg(nth->th_dport, nth->th_sport, uint16_t); 1910 #undef xchg 1911 } 1912 tlen = 0; 1913 #ifdef INET6 1914 if (isipv6) 1915 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1916 #endif 1917 #if defined(INET) && defined(INET6) 1918 else 1919 #endif 1920 #ifdef INET 1921 tlen = sizeof (struct tcpiphdr); 1922 #endif 1923 if (port) 1924 tlen += sizeof (struct udphdr); 1925 #ifdef INVARIANTS 1926 m->m_len = 0; 1927 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1928 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1929 m, tlen, (long)M_TRAILINGSPACE(m))); 1930 #endif 1931 m->m_len = tlen; 1932 to.to_flags = 0; 1933 if (incl_opts) { 1934 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1935 /* Make sure we have room. */ 1936 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1937 m->m_next = m_get(M_NOWAIT, MT_DATA); 1938 if (m->m_next) { 1939 optp = mtod(m->m_next, u_char *); 1940 optm = m->m_next; 1941 } else 1942 incl_opts = false; 1943 } else { 1944 optp = (u_char *) (nth + 1); 1945 optm = m; 1946 } 1947 } 1948 if (incl_opts) { 1949 /* Timestamps. */ 1950 if (tp->t_flags & TF_RCVD_TSTMP) { 1951 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1952 to.to_tsecr = tp->ts_recent; 1953 to.to_flags |= TOF_TS; 1954 } 1955 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1956 /* TCP-MD5 (RFC2385). */ 1957 if (tp->t_flags & TF_SIGNATURE) 1958 to.to_flags |= TOF_SIGNATURE; 1959 #endif 1960 /* Add the options. */ 1961 tlen += optlen = tcp_addoptions(&to, optp); 1962 1963 /* Update m_len in the correct mbuf. */ 1964 optm->m_len += optlen; 1965 } else 1966 optlen = 0; 1967 #ifdef INET6 1968 if (isipv6) { 1969 if (uh) { 1970 ulen = tlen - sizeof(struct ip6_hdr); 1971 uh->uh_ulen = htons(ulen); 1972 } 1973 ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN); 1974 ip6->ip6_vfc = IPV6_VERSION; 1975 if (port) 1976 ip6->ip6_nxt = IPPROTO_UDP; 1977 else 1978 ip6->ip6_nxt = IPPROTO_TCP; 1979 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1980 } 1981 #endif 1982 #if defined(INET) && defined(INET6) 1983 else 1984 #endif 1985 #ifdef INET 1986 { 1987 if (uh) { 1988 ulen = tlen - sizeof(struct ip); 1989 uh->uh_ulen = htons(ulen); 1990 } 1991 ip->ip_len = htons(tlen); 1992 if (inp != NULL) { 1993 ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK; 1994 ip->ip_ttl = inp->inp_ip_ttl; 1995 } else { 1996 ip->ip_tos = 0; 1997 ip->ip_ttl = V_ip_defttl; 1998 } 1999 ip->ip_tos |= ect; 2000 if (port) { 2001 ip->ip_p = IPPROTO_UDP; 2002 } else { 2003 ip->ip_p = IPPROTO_TCP; 2004 } 2005 if (V_path_mtu_discovery) 2006 ip->ip_off |= htons(IP_DF); 2007 } 2008 #endif 2009 m->m_pkthdr.len = tlen; 2010 m->m_pkthdr.rcvif = NULL; 2011 #ifdef MAC 2012 if (inp != NULL) { 2013 /* 2014 * Packet is associated with a socket, so allow the 2015 * label of the response to reflect the socket label. 2016 */ 2017 INP_LOCK_ASSERT(inp); 2018 mac_inpcb_create_mbuf(inp, m); 2019 } else { 2020 /* 2021 * Packet is not associated with a socket, so possibly 2022 * update the label in place. 2023 */ 2024 mac_netinet_tcp_reply(m); 2025 } 2026 #endif 2027 nth->th_seq = htonl(seq); 2028 nth->th_ack = htonl(ack); 2029 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2030 tcp_set_flags(nth, flags); 2031 if (tp && (flags & TH_RST)) { 2032 /* Log the reset */ 2033 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2034 } 2035 if (tp != NULL) 2036 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2037 else 2038 nth->th_win = htons((u_short)win); 2039 nth->th_urp = 0; 2040 2041 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2042 if (to.to_flags & TOF_SIGNATURE) { 2043 if (!TCPMD5_ENABLED() || 2044 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2045 m_freem(m); 2046 return; 2047 } 2048 } 2049 #endif 2050 2051 #ifdef INET6 2052 if (isipv6) { 2053 if (port) { 2054 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2055 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2056 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2057 nth->th_sum = 0; 2058 } else { 2059 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2060 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2061 nth->th_sum = in6_cksum_pseudo(ip6, 2062 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2063 } 2064 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2065 } 2066 #endif /* INET6 */ 2067 #if defined(INET6) && defined(INET) 2068 else 2069 #endif 2070 #ifdef INET 2071 { 2072 if (port) { 2073 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2074 htons(ulen + IPPROTO_UDP)); 2075 m->m_pkthdr.csum_flags = CSUM_UDP; 2076 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2077 nth->th_sum = 0; 2078 } else { 2079 m->m_pkthdr.csum_flags = CSUM_TCP; 2080 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2081 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2082 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2083 } 2084 } 2085 #endif /* INET */ 2086 TCP_PROBE3(debug__output, tp, th, m); 2087 if (flags & TH_RST) 2088 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2089 lgb = NULL; 2090 if ((tp != NULL) && tcp_bblogging_on(tp)) { 2091 if (INP_WLOCKED(inp)) { 2092 union tcp_log_stackspecific log; 2093 struct timeval tv; 2094 2095 memset(&log, 0, sizeof(log)); 2096 log.u_bbr.inhpts = tcp_in_hpts(tp); 2097 log.u_bbr.flex8 = 4; 2098 log.u_bbr.pkts_out = tp->t_maxseg; 2099 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2100 log.u_bbr.delivered = 0; 2101 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT, 2102 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2103 } else { 2104 /* 2105 * We can not log the packet, since we only own the 2106 * read lock, but a write lock is needed. The read lock 2107 * is not upgraded to a write lock, since only getting 2108 * the read lock was done intentionally to improve the 2109 * handling of SYN flooding attacks. 2110 * This happens only for pure SYN segments received in 2111 * the initial CLOSED state, or received in a more 2112 * advanced state than listen and the UDP encapsulation 2113 * port is unexpected. 2114 * The incoming SYN segments do not really belong to 2115 * the TCP connection and the handling does not change 2116 * the state of the TCP connection. Therefore, the 2117 * sending of the RST segments is not logged. Please 2118 * note that also the incoming SYN segments are not 2119 * logged. 2120 * 2121 * The following code ensures that the above description 2122 * is and stays correct. 2123 */ 2124 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2125 (tp->t_state == TCPS_CLOSED || 2126 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2127 ("%s: Logging of TCP segment with flags 0x%b and " 2128 "UDP encapsulation port %u skipped in state %s", 2129 __func__, thflags, PRINT_TH_FLAGS, 2130 ntohs(port), tcpstates[tp->t_state])); 2131 } 2132 } 2133 2134 if (flags & TH_ACK) 2135 TCPSTAT_INC(tcps_sndacks); 2136 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2137 TCPSTAT_INC(tcps_sndctrl); 2138 TCPSTAT_INC(tcps_sndtotal); 2139 2140 #ifdef INET6 2141 if (isipv6) { 2142 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2143 output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL, 2144 NULL, 0, NULL, NULL, inp); 2145 } 2146 #endif /* INET6 */ 2147 #if defined(INET) && defined(INET6) 2148 else 2149 #endif 2150 #ifdef INET 2151 { 2152 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2153 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2154 } 2155 #endif 2156 if (lgb != NULL) 2157 lgb->tlb_errno = output_ret; 2158 } 2159 2160 /* 2161 * Check that no more than V_tcp_ack_war_cnt per V_tcp_ack_war_time_window 2162 * are sent. *epoch_end is the end of the current epoch and is updated, if the 2163 * current epoch ended in the past. *ack_cnt is the counter used during the 2164 * current epoch. It might be reset and incremented. 2165 * The function returns true if a challenge ACK should be sent. 2166 */ 2167 bool 2168 tcp_challenge_ack_check(sbintime_t *epoch_end, uint32_t *ack_cnt) 2169 { 2170 sbintime_t now; 2171 2172 /* 2173 * The sending of a challenge ACK could be triggered by a blind attacker 2174 * to detect an existing TCP connection. To mitigate that, increment 2175 * also the global counter which would be incremented if the attacker 2176 * would have guessed wrongly. 2177 */ 2178 (void)badport_bandlim(BANDLIM_TCP_RST); 2179 2180 if (V_tcp_ack_war_time_window == 0 || V_tcp_ack_war_cnt == 0) { 2181 /* ACK war protection is disabled. */ 2182 return (true); 2183 } else { 2184 /* Start new epoch, if the previous one is already over. */ 2185 now = getsbinuptime(); 2186 if (*epoch_end < now) { 2187 *ack_cnt = 0; 2188 *epoch_end = now + V_tcp_ack_war_time_window * SBT_1MS; 2189 } 2190 /* 2191 * Send a challenge ACK, if less than tcp_ack_war_cnt have been 2192 * sent in the current epoch. 2193 */ 2194 if (*ack_cnt < V_tcp_ack_war_cnt) { 2195 (*ack_cnt)++; 2196 return (true); 2197 } else { 2198 return (false); 2199 } 2200 } 2201 } 2202 2203 /* 2204 * Send a challenge ack (no data, no SACK option), but not more than 2205 * V_tcp_ack_war_cnt per V_tcp_ack_war_time_window (per TCP connection). 2206 */ 2207 void 2208 tcp_send_challenge_ack(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m) 2209 { 2210 if (tcp_challenge_ack_check(&tp->t_challenge_ack_end, 2211 &tp->t_challenge_ack_cnt)) { 2212 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2213 tp->snd_nxt, TH_ACK); 2214 tp->last_ack_sent = tp->rcv_nxt; 2215 } else { 2216 m_freem(m); 2217 } 2218 } 2219 2220 /* 2221 * Create a new TCP control block, making an empty reassembly queue and hooking 2222 * it to the argument protocol control block. The `inp' parameter must have 2223 * come from the zone allocator set up by tcpcbstor declaration. 2224 * The caller can provide a pointer to a tcpcb of the listener to inherit the 2225 * TCP function block from the listener. 2226 */ 2227 struct tcpcb * 2228 tcp_newtcpcb(struct inpcb *inp, struct tcpcb *listening_tcb) 2229 { 2230 struct tcpcb *tp = intotcpcb(inp); 2231 #ifdef INET6 2232 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2233 #endif /* INET6 */ 2234 2235 /* 2236 * Historically allocation was done with M_ZERO. There is a lot of 2237 * code that rely on that. For now take safe approach and zero whole 2238 * tcpcb. This definitely can be optimized. 2239 */ 2240 bzero(&tp->t_start_zero, t_zero_size); 2241 2242 /* Initialise cc_var struct for this tcpcb. */ 2243 tp->t_ccv.tp = tp; 2244 rw_rlock(&tcp_function_lock); 2245 if (listening_tcb != NULL) { 2246 INP_LOCK_ASSERT(tptoinpcb(listening_tcb)); 2247 KASSERT(listening_tcb->t_fb != NULL, 2248 ("tcp_newtcpcb: listening_tcb->t_fb is NULL")); 2249 if (listening_tcb->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) { 2250 rw_runlock(&tcp_function_lock); 2251 return (NULL); 2252 } 2253 tp->t_fb = listening_tcb->t_fb; 2254 } else { 2255 tp->t_fb = V_tcp_func_set_ptr; 2256 } 2257 refcount_acquire(&tp->t_fb->tfb_refcnt); 2258 KASSERT((tp->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) == 0, 2259 ("tcp_newtcpcb: using TFB being removed")); 2260 rw_runlock(&tcp_function_lock); 2261 CC_LIST_RLOCK(); 2262 if (listening_tcb != NULL) { 2263 if (CC_ALGO(listening_tcb)->flags & CC_MODULE_BEING_REMOVED) { 2264 CC_LIST_RUNLOCK(); 2265 if (tp->t_fb->tfb_tcp_fb_fini) 2266 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2267 refcount_release(&tp->t_fb->tfb_refcnt); 2268 return (NULL); 2269 } 2270 CC_ALGO(tp) = CC_ALGO(listening_tcb); 2271 } else 2272 CC_ALGO(tp) = CC_DEFAULT_ALGO(); 2273 cc_refer(CC_ALGO(tp)); 2274 CC_LIST_RUNLOCK(); 2275 if (CC_ALGO(tp)->cb_init != NULL) 2276 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2277 cc_detach(tp); 2278 if (tp->t_fb->tfb_tcp_fb_fini) 2279 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2280 refcount_release(&tp->t_fb->tfb_refcnt); 2281 return (NULL); 2282 } 2283 2284 #ifdef TCP_HHOOK 2285 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2286 if (CC_ALGO(tp)->cb_destroy != NULL) 2287 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2288 CC_DATA(tp) = NULL; 2289 cc_detach(tp); 2290 if (tp->t_fb->tfb_tcp_fb_fini) 2291 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2292 refcount_release(&tp->t_fb->tfb_refcnt); 2293 return (NULL); 2294 } 2295 #endif 2296 2297 TAILQ_INIT(&tp->t_segq); 2298 STAILQ_INIT(&tp->t_inqueue); 2299 tp->t_maxseg = 2300 #ifdef INET6 2301 isipv6 ? V_tcp_v6mssdflt : 2302 #endif /* INET6 */ 2303 V_tcp_mssdflt; 2304 2305 /* All mbuf queue/ack compress flags should be off */ 2306 tcp_lro_features_off(tp); 2307 2308 tp->t_hpts_cpu = HPTS_CPU_NONE; 2309 tp->t_lro_cpu = HPTS_CPU_NONE; 2310 2311 callout_init_rw(&tp->t_callout, &inp->inp_lock, 2312 CALLOUT_TRYLOCK | 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 TCP_BLACKBOX 2355 /* Initialize the per-TCPCB log data. */ 2356 tcp_log_tcpcbinit(tp); 2357 #endif 2358 tp->t_pacing_rate = -1; 2359 if (tp->t_fb->tfb_tcp_fb_init) { 2360 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) { 2361 if (CC_ALGO(tp)->cb_destroy != NULL) 2362 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2363 CC_DATA(tp) = NULL; 2364 cc_detach(tp); 2365 #ifdef TCP_HHOOK 2366 khelp_destroy_osd(&tp->t_osd); 2367 #endif 2368 refcount_release(&tp->t_fb->tfb_refcnt); 2369 return (NULL); 2370 } 2371 } 2372 #ifdef STATS 2373 if (V_tcp_perconn_stats_enable == 1) 2374 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2375 #endif 2376 if (V_tcp_do_lrd) 2377 tp->t_flags |= TF_LRD; 2378 2379 return (tp); 2380 } 2381 2382 /* 2383 * Drop a TCP connection, reporting 2384 * the specified error. If connection is synchronized, 2385 * then send a RST to peer. 2386 */ 2387 struct tcpcb * 2388 tcp_drop(struct tcpcb *tp, int errno) 2389 { 2390 struct socket *so = tptosocket(tp); 2391 2392 NET_EPOCH_ASSERT(); 2393 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2394 2395 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2396 tcp_state_change(tp, TCPS_CLOSED); 2397 /* Don't use tcp_output() here due to possible recursion. */ 2398 (void)tcp_output_nodrop(tp); 2399 TCPSTAT_INC(tcps_drops); 2400 } else 2401 TCPSTAT_INC(tcps_conndrops); 2402 if (errno == ETIMEDOUT && tp->t_softerror) 2403 errno = tp->t_softerror; 2404 so->so_error = errno; 2405 return (tcp_close(tp)); 2406 } 2407 2408 void 2409 tcp_discardcb(struct tcpcb *tp) 2410 { 2411 struct inpcb *inp = tptoinpcb(tp); 2412 struct socket *so = tptosocket(tp); 2413 struct mbuf *m; 2414 #ifdef INET6 2415 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2416 #endif 2417 2418 INP_WLOCK_ASSERT(inp); 2419 MPASS(!callout_active(&tp->t_callout)); 2420 MPASS(TAILQ_EMPTY(&tp->snd_holes)); 2421 2422 /* free the reassembly queue, if any */ 2423 tcp_reass_flush(tp); 2424 2425 #ifdef TCP_OFFLOAD 2426 /* Disconnect offload device, if any. */ 2427 if (tp->t_flags & TF_TOE) 2428 tcp_offload_detach(tp); 2429 #endif 2430 2431 /* Allow the CC algorithm to clean up after itself. */ 2432 if (CC_ALGO(tp)->cb_destroy != NULL) 2433 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2434 CC_DATA(tp) = NULL; 2435 /* Detach from the CC algorithm */ 2436 cc_detach(tp); 2437 2438 #ifdef TCP_HHOOK 2439 khelp_destroy_osd(&tp->t_osd); 2440 #endif 2441 #ifdef STATS 2442 stats_blob_destroy(tp->t_stats); 2443 #endif 2444 #ifdef TCP_REQUEST_TRK 2445 if (tp->t_tcpreq_info != NULL) { 2446 free(tp->t_tcpreq_info, M_TCPREQTRK); 2447 tp->t_tcpreq_info = NULL; 2448 } 2449 #endif 2450 2451 CC_ALGO(tp) = NULL; 2452 if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) { 2453 struct mbuf *prev; 2454 2455 STAILQ_INIT(&tp->t_inqueue); 2456 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev) 2457 m_freem(m); 2458 } 2459 2460 if (tp->t_fb->tfb_tcp_fb_fini) 2461 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2462 MPASS(!tcp_in_hpts(tp)); 2463 #ifdef TCP_BLACKBOX 2464 tcp_log_tcpcbfini(tp); 2465 #endif 2466 2467 /* 2468 * If we got enough samples through the srtt filter, 2469 * save the rtt and rttvar in the routing entry. 2470 * 'Enough' is arbitrarily defined as 4 rtt samples. 2471 * 4 samples is enough for the srtt filter to converge 2472 * to within enough % of the correct value; fewer samples 2473 * and we could save a bogus rtt. The danger is not high 2474 * as tcp quickly recovers from everything. 2475 * XXX: Works very well but needs some more statistics! 2476 * 2477 * XXXRRS: Updating must be after the stack fini() since 2478 * that may be converting some internal representation of 2479 * say srtt etc into the general one used by other stacks. 2480 */ 2481 if (tp->t_rttupdated >= 4) { 2482 uint32_t ssthresh; 2483 2484 /* 2485 * Update the ssthresh always when the conditions below 2486 * are satisfied. This gives us better new start value 2487 * for the congestion avoidance for new connections. 2488 * ssthresh is only set if packet loss occurred on a session. 2489 */ 2490 ssthresh = tp->snd_ssthresh; 2491 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2492 /* 2493 * convert the limit from user data bytes to 2494 * packets then to packet data bytes. 2495 */ 2496 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2497 if (ssthresh < 2) 2498 ssthresh = 2; 2499 ssthresh *= (tp->t_maxseg + 2500 #ifdef INET6 2501 (isipv6 ? sizeof (struct ip6_hdr) + 2502 sizeof (struct tcphdr) : 2503 #endif 2504 sizeof (struct tcpiphdr) 2505 #ifdef INET6 2506 ) 2507 #endif 2508 ); 2509 } else 2510 ssthresh = 0; 2511 tcp_hc_update(&inp->inp_inc, &(struct tcp_hc_metrics){ 2512 .hc_ssthresh = ssthresh, 2513 .hc_rtt = tp->t_srtt, 2514 .hc_rttvar = tp->t_rttvar, 2515 .hc_cwnd = tp->snd_cwnd, 2516 .hc_sendpipe = so->so_snd.sb_hiwat, 2517 .hc_recvpipe = so->so_rcv.sb_hiwat, 2518 }); 2519 } 2520 2521 refcount_release(&tp->t_fb->tfb_refcnt); 2522 } 2523 2524 /* 2525 * Attempt to close a TCP control block, marking it as dropped, and freeing 2526 * the socket if we hold the only reference. 2527 */ 2528 struct tcpcb * 2529 tcp_close(struct tcpcb *tp) 2530 { 2531 struct inpcb *inp = tptoinpcb(tp); 2532 struct socket *so = tptosocket(tp); 2533 2534 INP_WLOCK_ASSERT(inp); 2535 2536 #ifdef TCP_OFFLOAD 2537 if (tp->t_state == TCPS_LISTEN) 2538 tcp_offload_listen_stop(tp); 2539 #endif 2540 /* 2541 * This releases the TFO pending counter resource for TFO listen 2542 * sockets as well as passively-created TFO sockets that transition 2543 * from SYN_RECEIVED to CLOSED. 2544 */ 2545 if (tp->t_tfo_pending) { 2546 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2547 tp->t_tfo_pending = NULL; 2548 } 2549 tcp_timer_stop(tp); 2550 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 2551 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2552 #if defined(INET) && defined(INET6) 2553 if ((inp->inp_vflag & INP_IPV6) != 0) 2554 in6_pcbdisconnect(inp); 2555 else 2556 in_pcbdisconnect(inp); 2557 #elif defined(INET6) 2558 in6_pcbdisconnect(inp); 2559 #else 2560 in_pcbdisconnect(inp); 2561 #endif 2562 TCPSTAT_INC(tcps_closed); 2563 if (tp->t_state != TCPS_CLOSED) 2564 tcp_state_change(tp, TCPS_CLOSED); 2565 tp->t_flags |= TF_DISCONNECTED; 2566 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2567 tcp_free_sackholes(tp); 2568 soisdisconnected(so); 2569 if (inp->inp_flags & INP_SOCKREF) { 2570 inp->inp_flags &= ~INP_SOCKREF; 2571 INP_WUNLOCK(inp); 2572 sorele(so); 2573 return (NULL); 2574 } 2575 return (tp); 2576 } 2577 2578 /* 2579 * Notify a tcp user of an asynchronous error; 2580 * store error as soft error, but wake up user 2581 * (for now, won't do anything until can select for soft error). 2582 * 2583 * Do not wake up user since there currently is no mechanism for 2584 * reporting soft errors (yet - a kqueue filter may be added). 2585 */ 2586 static struct inpcb * 2587 tcp_notify(struct inpcb *inp, int error) 2588 { 2589 struct tcpcb *tp; 2590 2591 INP_WLOCK_ASSERT(inp); 2592 2593 tp = intotcpcb(inp); 2594 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2595 2596 /* 2597 * Ignore some errors if we are hooked up. 2598 * If connection hasn't completed, has retransmitted several times, 2599 * and receives a second error, give up now. This is better 2600 * than waiting a long time to establish a connection that 2601 * can never complete. 2602 */ 2603 if (tp->t_state == TCPS_ESTABLISHED && 2604 (error == EHOSTUNREACH || error == ENETUNREACH || 2605 error == EHOSTDOWN)) { 2606 if (inp->inp_route.ro_nh) { 2607 NH_FREE(inp->inp_route.ro_nh); 2608 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2609 } 2610 return (inp); 2611 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2612 tp->t_softerror) { 2613 tp = tcp_drop(tp, error); 2614 if (tp != NULL) 2615 return (inp); 2616 else 2617 return (NULL); 2618 } else { 2619 tp->t_softerror = error; 2620 return (inp); 2621 } 2622 #if 0 2623 wakeup( &so->so_timeo); 2624 sorwakeup(so); 2625 sowwakeup(so); 2626 #endif 2627 } 2628 2629 static int 2630 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2631 { 2632 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2633 INPLOOKUP_RLOCKPCB); 2634 struct xinpgen xig; 2635 struct inpcb *inp; 2636 int error; 2637 2638 if (req->newptr != NULL) 2639 return (EPERM); 2640 2641 if (req->oldptr == NULL) { 2642 int n; 2643 2644 n = V_tcbinfo.ipi_count + 2645 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2646 n += imax(n / 8, 10); 2647 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2648 return (0); 2649 } 2650 2651 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2652 return (error); 2653 2654 bzero(&xig, sizeof(xig)); 2655 xig.xig_len = sizeof xig; 2656 xig.xig_count = V_tcbinfo.ipi_count + 2657 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2658 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2659 xig.xig_sogen = so_gencnt; 2660 error = SYSCTL_OUT(req, &xig, sizeof xig); 2661 if (error) 2662 return (error); 2663 2664 error = syncache_pcblist(req); 2665 if (error) 2666 return (error); 2667 2668 while ((inp = inp_next(&inpi)) != NULL) { 2669 if (inp->inp_gencnt <= xig.xig_gen && 2670 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2671 struct xtcpcb xt; 2672 2673 tcp_inptoxtp(inp, &xt); 2674 error = SYSCTL_OUT(req, &xt, sizeof xt); 2675 if (error) { 2676 INP_RUNLOCK(inp); 2677 break; 2678 } else 2679 continue; 2680 } 2681 } 2682 2683 if (!error) { 2684 /* 2685 * Give the user an updated idea of our state. 2686 * If the generation differs from what we told 2687 * her before, she knows that something happened 2688 * while we were processing this request, and it 2689 * might be necessary to retry. 2690 */ 2691 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2692 xig.xig_sogen = so_gencnt; 2693 xig.xig_count = V_tcbinfo.ipi_count + 2694 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2695 error = SYSCTL_OUT(req, &xig, sizeof xig); 2696 } 2697 2698 return (error); 2699 } 2700 2701 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2702 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2703 NULL, 0, tcp_pcblist, "S,xtcpcb", 2704 "List of active TCP connections"); 2705 2706 #define SND_TAG_STATUS_MAXLEN 128 2707 2708 #ifdef KERN_TLS 2709 2710 static struct sx ktlslist_lock; 2711 SX_SYSINIT(ktlslistlock, &ktlslist_lock, "ktlslist"); 2712 static uint64_t ktls_glob_gen = 1; 2713 2714 static int 2715 tcp_ktlslist_locked(SYSCTL_HANDLER_ARGS, bool export_keys) 2716 { 2717 struct xinpgen xig; 2718 struct inpcb *inp; 2719 struct socket *so; 2720 struct ktls_session *ksr, *kss; 2721 char *buf; 2722 struct xktls_session *xktls; 2723 uint64_t ipi_gencnt; 2724 size_t buflen, len, sz; 2725 u_int cnt; 2726 int error; 2727 bool ek, p; 2728 2729 sx_assert(&ktlslist_lock, SA_XLOCKED); 2730 if (req->newptr != NULL) 2731 return (EPERM); 2732 2733 len = 0; 2734 cnt = 0; 2735 ipi_gencnt = V_tcbinfo.ipi_gencnt; 2736 bzero(&xig, sizeof(xig)); 2737 xig.xig_len = sizeof(xig); 2738 xig.xig_gen = ktls_glob_gen++; 2739 xig.xig_sogen = so_gencnt; 2740 2741 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2742 INPLOOKUP_RLOCKPCB); 2743 while ((inp = inp_next(&inpi)) != NULL) { 2744 if (inp->inp_gencnt > ipi_gencnt || 2745 cr_canseeinpcb(req->td->td_ucred, inp) != 0) 2746 continue; 2747 2748 so = inp->inp_socket; 2749 if (so != NULL && so->so_gencnt <= xig.xig_sogen) { 2750 p = false; 2751 ek = export_keys && cr_canexport_ktlskeys( 2752 req->td, inp); 2753 ksr = so->so_rcv.sb_tls_info; 2754 if (ksr != NULL) { 2755 ksr->gen = xig.xig_gen; 2756 p = true; 2757 if (ek) { 2758 sz = SIZE_T_MAX; 2759 ktls_session_copy_keys(ksr, 2760 NULL, &sz); 2761 len += sz; 2762 } 2763 if (ksr->snd_tag != NULL && 2764 ksr->snd_tag->sw->snd_tag_status_str != 2765 NULL) { 2766 sz = SND_TAG_STATUS_MAXLEN; 2767 in_pcbref(inp); 2768 INP_RUNLOCK(inp); 2769 error = ksr->snd_tag->sw-> 2770 snd_tag_status_str( 2771 ksr->snd_tag, NULL, &sz); 2772 if (in_pcbrele_rlock(inp)) 2773 return (EDEADLK); 2774 if (error == 0) 2775 len += sz; 2776 } 2777 } 2778 kss = so->so_snd.sb_tls_info; 2779 if (kss != NULL) { 2780 kss->gen = xig.xig_gen; 2781 p = true; 2782 if (ek) { 2783 sz = SIZE_T_MAX; 2784 ktls_session_copy_keys(kss, 2785 NULL, &sz); 2786 len += sz; 2787 } 2788 if (kss->snd_tag != NULL && 2789 kss->snd_tag->sw->snd_tag_status_str != 2790 NULL) { 2791 sz = SND_TAG_STATUS_MAXLEN; 2792 in_pcbref(inp); 2793 INP_RUNLOCK(inp); 2794 error = kss->snd_tag->sw-> 2795 snd_tag_status_str( 2796 kss->snd_tag, NULL, &sz); 2797 if (in_pcbrele_rlock(inp)) 2798 return (EDEADLK); 2799 if (error == 0) 2800 len += sz; 2801 } 2802 } 2803 if (p) { 2804 len += sizeof(*xktls); 2805 len = roundup2(len, __alignof(struct 2806 xktls_session)); 2807 } 2808 } 2809 } 2810 if (req->oldptr == NULL) { 2811 len += 2 * sizeof(xig); 2812 len += 3 * len / 4; 2813 req->oldidx = len; 2814 return (0); 2815 } 2816 2817 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2818 return (error); 2819 2820 error = SYSCTL_OUT(req, &xig, sizeof xig); 2821 if (error != 0) 2822 return (error); 2823 2824 buflen = roundup2(sizeof(*xktls) + 2 * TLS_MAX_PARAM_SIZE + 2825 2 * SND_TAG_STATUS_MAXLEN, __alignof(struct xktls_session)); 2826 buf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 2827 struct inpcb_iterator inpi1 = INP_ALL_ITERATOR(&V_tcbinfo, 2828 INPLOOKUP_RLOCKPCB); 2829 while ((inp = inp_next(&inpi1)) != NULL) { 2830 if (inp->inp_gencnt > ipi_gencnt || 2831 cr_canseeinpcb(req->td->td_ucred, inp) != 0) 2832 continue; 2833 2834 so = inp->inp_socket; 2835 if (so == NULL) 2836 continue; 2837 2838 p = false; 2839 ek = export_keys && cr_canexport_ktlskeys(req->td, inp); 2840 ksr = so->so_rcv.sb_tls_info; 2841 kss = so->so_snd.sb_tls_info; 2842 xktls = (struct xktls_session *)buf; 2843 if (ksr != NULL && ksr->gen == xig.xig_gen) { 2844 p = true; 2845 ktls_session_to_xktls_onedir(ksr, ek, &xktls->rcv); 2846 } 2847 if (kss != NULL && kss->gen == xig.xig_gen) { 2848 p = true; 2849 ktls_session_to_xktls_onedir(kss, ek, &xktls->snd); 2850 } 2851 if (!p) 2852 continue; 2853 2854 xktls->inp_gencnt = inp->inp_gencnt; 2855 xktls->so_pcb = (kvaddr_t)inp; 2856 memcpy(&xktls->coninf, &inp->inp_inc, sizeof(xktls->coninf)); 2857 len = sizeof(*xktls); 2858 if (ksr != NULL && ksr->gen == xig.xig_gen) { 2859 if (ek) { 2860 sz = buflen - len; 2861 ktls_session_copy_keys(ksr, buf + len, &sz); 2862 len += sz; 2863 } else { 2864 xktls->rcv.cipher_key_len = 0; 2865 xktls->rcv.auth_key_len = 0; 2866 } 2867 if (ksr->snd_tag != NULL && 2868 ksr->snd_tag->sw->snd_tag_status_str != NULL) { 2869 sz = SND_TAG_STATUS_MAXLEN; 2870 in_pcbref(inp); 2871 INP_RUNLOCK(inp); 2872 error = ksr->snd_tag->sw->snd_tag_status_str( 2873 ksr->snd_tag, buf + len, &sz); 2874 if (in_pcbrele_rlock(inp)) 2875 return (EDEADLK); 2876 if (error == 0) { 2877 xktls->rcv.drv_st_len = sz; 2878 len += sz; 2879 } 2880 } 2881 } 2882 if (kss != NULL && kss->gen == xig.xig_gen) { 2883 if (ek) { 2884 sz = buflen - len; 2885 ktls_session_copy_keys(kss, buf + len, &sz); 2886 len += sz; 2887 } else { 2888 xktls->snd.cipher_key_len = 0; 2889 xktls->snd.auth_key_len = 0; 2890 } 2891 if (kss->snd_tag != NULL && 2892 kss->snd_tag->sw->snd_tag_status_str != NULL) { 2893 sz = SND_TAG_STATUS_MAXLEN; 2894 in_pcbref(inp); 2895 INP_RUNLOCK(inp); 2896 error = kss->snd_tag->sw->snd_tag_status_str( 2897 kss->snd_tag, buf + len, &sz); 2898 if (in_pcbrele_rlock(inp)) 2899 return (EDEADLK); 2900 if (error == 0) { 2901 xktls->snd.drv_st_len = sz; 2902 len += sz; 2903 } 2904 } 2905 } 2906 len = roundup2(len, __alignof(*xktls)); 2907 xktls->tsz = len; 2908 xktls->fsz = sizeof(*xktls); 2909 2910 error = SYSCTL_OUT(req, xktls, len); 2911 if (error != 0) { 2912 INP_RUNLOCK(inp); 2913 break; 2914 } 2915 cnt++; 2916 } 2917 2918 if (error == 0) { 2919 xig.xig_sogen = so_gencnt; 2920 xig.xig_count = cnt; 2921 error = SYSCTL_OUT(req, &xig, sizeof(xig)); 2922 } 2923 2924 zfree(buf, M_TEMP); 2925 return (error); 2926 } 2927 2928 static int 2929 tcp_ktlslist1(SYSCTL_HANDLER_ARGS, bool export_keys) 2930 { 2931 int repeats, error; 2932 2933 for (repeats = 0; repeats < 100; repeats++) { 2934 if (sx_xlock_sig(&ktlslist_lock)) 2935 return (EINTR); 2936 error = tcp_ktlslist_locked(oidp, arg1, arg2, req, 2937 export_keys); 2938 sx_xunlock(&ktlslist_lock); 2939 if (error != EDEADLK) 2940 break; 2941 if (sig_intr() != 0) { 2942 error = EINTR; 2943 break; 2944 } 2945 req->oldidx = 0; 2946 } 2947 return (error); 2948 } 2949 2950 static int 2951 tcp_ktlslist_nokeys(SYSCTL_HANDLER_ARGS) 2952 { 2953 return (tcp_ktlslist1(oidp, arg1, arg2, req, false)); 2954 } 2955 2956 static int 2957 tcp_ktlslist_wkeys(SYSCTL_HANDLER_ARGS) 2958 { 2959 return (tcp_ktlslist1(oidp, arg1, arg2, req, true)); 2960 } 2961 2962 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST, ktlslist, 2963 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2964 NULL, 0, tcp_ktlslist_nokeys, "S,xktls_session", 2965 "List of active kTLS sessions for TCP connections"); 2966 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST_WKEYS, ktlslist_wkeys, 2967 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2968 NULL, 0, tcp_ktlslist_wkeys, "S,xktls_session", 2969 "List of active kTLS sessions for TCP connections with keys"); 2970 #endif /* KERN_TLS */ 2971 2972 #ifdef INET 2973 static int 2974 tcp_getcred(SYSCTL_HANDLER_ARGS) 2975 { 2976 struct xucred xuc; 2977 struct sockaddr_in addrs[2]; 2978 struct epoch_tracker et; 2979 struct inpcb *inp; 2980 int error; 2981 2982 if (req->newptr == NULL) 2983 return (EINVAL); 2984 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2985 if (error) 2986 return (error); 2987 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2988 if (error) 2989 return (error); 2990 NET_EPOCH_ENTER(et); 2991 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2992 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2993 NET_EPOCH_EXIT(et); 2994 if (inp != NULL) { 2995 if (error == 0) 2996 error = cr_canseeinpcb(req->td->td_ucred, inp); 2997 if (error == 0) 2998 cru2x(inp->inp_cred, &xuc); 2999 INP_RUNLOCK(inp); 3000 } else 3001 error = ENOENT; 3002 if (error == 0) 3003 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 3004 return (error); 3005 } 3006 3007 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 3008 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 3009 0, 0, tcp_getcred, "S,xucred", 3010 "Get the xucred of a TCP connection"); 3011 #endif /* INET */ 3012 3013 #ifdef INET6 3014 static int 3015 tcp6_getcred(SYSCTL_HANDLER_ARGS) 3016 { 3017 struct epoch_tracker et; 3018 struct xucred xuc; 3019 struct sockaddr_in6 addrs[2]; 3020 struct inpcb *inp; 3021 int error; 3022 #ifdef INET 3023 int mapped = 0; 3024 #endif 3025 3026 if (req->newptr == NULL) 3027 return (EINVAL); 3028 error = priv_check(req->td, PRIV_NETINET_GETCRED); 3029 if (error) 3030 return (error); 3031 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 3032 if (error) 3033 return (error); 3034 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 3035 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 3036 return (error); 3037 } 3038 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 3039 #ifdef INET 3040 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 3041 mapped = 1; 3042 else 3043 #endif 3044 return (EINVAL); 3045 } 3046 3047 NET_EPOCH_ENTER(et); 3048 #ifdef INET 3049 if (mapped == 1) 3050 inp = in_pcblookup(&V_tcbinfo, 3051 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 3052 addrs[1].sin6_port, 3053 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 3054 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 3055 else 3056 #endif 3057 inp = in6_pcblookup(&V_tcbinfo, 3058 &addrs[1].sin6_addr, addrs[1].sin6_port, 3059 &addrs[0].sin6_addr, addrs[0].sin6_port, 3060 INPLOOKUP_RLOCKPCB, NULL); 3061 NET_EPOCH_EXIT(et); 3062 if (inp != NULL) { 3063 if (error == 0) 3064 error = cr_canseeinpcb(req->td->td_ucred, inp); 3065 if (error == 0) 3066 cru2x(inp->inp_cred, &xuc); 3067 INP_RUNLOCK(inp); 3068 } else 3069 error = ENOENT; 3070 if (error == 0) 3071 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 3072 return (error); 3073 } 3074 3075 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 3076 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 3077 0, 0, tcp6_getcred, "S,xucred", 3078 "Get the xucred of a TCP6 connection"); 3079 #endif /* INET6 */ 3080 3081 #ifdef INET 3082 /* Path MTU to try next when a fragmentation-needed message is received. */ 3083 static inline int 3084 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 3085 { 3086 int mtu = ntohs(icp->icmp_nextmtu); 3087 3088 /* If no alternative MTU was proposed, try the next smaller one. */ 3089 if (!mtu) 3090 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 3091 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 3092 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 3093 3094 return (mtu); 3095 } 3096 3097 static void 3098 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 3099 { 3100 struct ip *ip; 3101 struct tcphdr *th; 3102 struct inpcb *inp; 3103 struct tcpcb *tp; 3104 struct inpcb *(*notify)(struct inpcb *, int); 3105 struct in_conninfo inc; 3106 tcp_seq icmp_tcp_seq; 3107 int errno, mtu; 3108 3109 errno = icmp_errmap(icp); 3110 switch (errno) { 3111 case 0: 3112 return; 3113 case EMSGSIZE: 3114 notify = tcp_mtudisc_notify; 3115 break; 3116 case ECONNREFUSED: 3117 if (V_icmp_may_rst) 3118 notify = tcp_drop_syn_sent; 3119 else 3120 notify = tcp_notify; 3121 break; 3122 case EHOSTUNREACH: 3123 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 3124 notify = tcp_drop_syn_sent; 3125 else 3126 notify = tcp_notify; 3127 break; 3128 default: 3129 notify = tcp_notify; 3130 } 3131 3132 ip = &icp->icmp_ip; 3133 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 3134 icmp_tcp_seq = th->th_seq; 3135 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 3136 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 3137 if (inp != NULL) { 3138 tp = intotcpcb(inp); 3139 #ifdef TCP_OFFLOAD 3140 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3141 /* 3142 * MTU discovery for offloaded connections. Let 3143 * the TOE driver verify seq# and process it. 3144 */ 3145 mtu = tcp_next_pmtu(icp, ip); 3146 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3147 goto out; 3148 } 3149 #endif 3150 if (tp->t_port != port) 3151 goto out; 3152 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3153 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3154 if (errno == EMSGSIZE) { 3155 /* 3156 * MTU discovery: we got a needfrag and 3157 * will potentially try a lower MTU. 3158 */ 3159 mtu = tcp_next_pmtu(icp, ip); 3160 3161 /* 3162 * Only process the offered MTU if it 3163 * is smaller than the current one. 3164 */ 3165 if (mtu < tp->t_maxseg + 3166 sizeof(struct tcpiphdr)) { 3167 bzero(&inc, sizeof(inc)); 3168 inc.inc_faddr = ip->ip_dst; 3169 inc.inc_fibnum = 3170 inp->inp_inc.inc_fibnum; 3171 tcp_hc_updatemtu(&inc, mtu); 3172 inp = tcp_mtudisc(inp, mtu); 3173 } 3174 } else 3175 inp = (*notify)(inp, errno); 3176 } 3177 } else { 3178 bzero(&inc, sizeof(inc)); 3179 inc.inc_fport = th->th_dport; 3180 inc.inc_lport = th->th_sport; 3181 inc.inc_faddr = ip->ip_dst; 3182 inc.inc_laddr = ip->ip_src; 3183 syncache_unreach(&inc, icmp_tcp_seq, port); 3184 } 3185 out: 3186 if (inp != NULL) 3187 INP_WUNLOCK(inp); 3188 } 3189 3190 static void 3191 tcp_ctlinput(struct icmp *icmp) 3192 { 3193 tcp_ctlinput_with_port(icmp, htons(0)); 3194 } 3195 3196 static void 3197 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 3198 { 3199 /* Its a tunneled TCP over UDP icmp */ 3200 struct icmp *icmp = param.icmp; 3201 struct ip *outer_ip, *inner_ip; 3202 struct udphdr *udp; 3203 struct tcphdr *th, ttemp; 3204 int i_hlen, o_len; 3205 uint16_t port; 3206 3207 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 3208 inner_ip = &icmp->icmp_ip; 3209 i_hlen = inner_ip->ip_hl << 2; 3210 o_len = ntohs(outer_ip->ip_len); 3211 if (o_len < 3212 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 3213 /* Not enough data present */ 3214 return; 3215 } 3216 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 3217 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 3218 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3219 return; 3220 } 3221 port = udp->uh_dport; 3222 th = (struct tcphdr *)(udp + 1); 3223 memcpy(&ttemp, th, sizeof(struct tcphdr)); 3224 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 3225 /* Now adjust down the size of the outer IP header */ 3226 o_len -= sizeof(struct udphdr); 3227 outer_ip->ip_len = htons(o_len); 3228 /* Now call in to the normal handling code */ 3229 tcp_ctlinput_with_port(icmp, port); 3230 } 3231 #endif /* INET */ 3232 3233 #ifdef INET6 3234 static inline int 3235 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3236 { 3237 int mtu = ntohl(icmp6->icmp6_mtu); 3238 3239 /* 3240 * If no alternative MTU was proposed, or the proposed MTU was too 3241 * small, set to the min. 3242 */ 3243 if (mtu < IPV6_MMTU) 3244 mtu = IPV6_MMTU; 3245 return (mtu); 3246 } 3247 3248 static void 3249 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 3250 { 3251 struct in6_addr *dst; 3252 struct inpcb *(*notify)(struct inpcb *, int); 3253 struct ip6_hdr *ip6; 3254 struct mbuf *m; 3255 struct inpcb *inp; 3256 struct tcpcb *tp; 3257 struct icmp6_hdr *icmp6; 3258 struct in_conninfo inc; 3259 struct tcp_ports { 3260 uint16_t th_sport; 3261 uint16_t th_dport; 3262 } t_ports; 3263 tcp_seq icmp_tcp_seq; 3264 unsigned int mtu; 3265 unsigned int off; 3266 int errno; 3267 3268 icmp6 = ip6cp->ip6c_icmp6; 3269 m = ip6cp->ip6c_m; 3270 ip6 = ip6cp->ip6c_ip6; 3271 off = ip6cp->ip6c_off; 3272 dst = &ip6cp->ip6c_finaldst->sin6_addr; 3273 3274 errno = icmp6_errmap(icmp6); 3275 switch (errno) { 3276 case 0: 3277 return; 3278 case EMSGSIZE: 3279 notify = tcp_mtudisc_notify; 3280 break; 3281 case ECONNREFUSED: 3282 if (V_icmp_may_rst) 3283 notify = tcp_drop_syn_sent; 3284 else 3285 notify = tcp_notify; 3286 break; 3287 case EHOSTUNREACH: 3288 /* 3289 * There are only four ICMPs that may reset connection: 3290 * - administratively prohibited 3291 * - port unreachable 3292 * - time exceeded in transit 3293 * - unknown next header 3294 */ 3295 if (V_icmp_may_rst && 3296 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 3297 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 3298 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 3299 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 3300 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 3301 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 3302 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 3303 notify = tcp_drop_syn_sent; 3304 else 3305 notify = tcp_notify; 3306 break; 3307 default: 3308 notify = tcp_notify; 3309 } 3310 3311 /* Check if we can safely get the ports from the tcp hdr */ 3312 if (m == NULL || 3313 (m->m_pkthdr.len < 3314 (int32_t) (off + sizeof(struct tcp_ports)))) { 3315 return; 3316 } 3317 bzero(&t_ports, sizeof(struct tcp_ports)); 3318 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3319 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3320 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3321 off += sizeof(struct tcp_ports); 3322 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3323 goto out; 3324 } 3325 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3326 if (inp != NULL) { 3327 tp = intotcpcb(inp); 3328 #ifdef TCP_OFFLOAD 3329 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3330 /* MTU discovery for offloaded connections. */ 3331 mtu = tcp6_next_pmtu(icmp6); 3332 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3333 goto out; 3334 } 3335 #endif 3336 if (tp->t_port != port) 3337 goto out; 3338 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3339 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3340 if (errno == EMSGSIZE) { 3341 /* 3342 * MTU discovery: 3343 * If we got a needfrag set the MTU 3344 * in the route to the suggested new 3345 * value (if given) and then notify. 3346 */ 3347 mtu = tcp6_next_pmtu(icmp6); 3348 3349 bzero(&inc, sizeof(inc)); 3350 inc.inc_fibnum = M_GETFIB(m); 3351 inc.inc_flags |= INC_ISIPV6; 3352 inc.inc6_faddr = *dst; 3353 if (in6_setscope(&inc.inc6_faddr, 3354 m->m_pkthdr.rcvif, NULL)) 3355 goto out; 3356 /* 3357 * Only process the offered MTU if it 3358 * is smaller than the current one. 3359 */ 3360 if (mtu < tp->t_maxseg + 3361 sizeof (struct tcphdr) + 3362 sizeof (struct ip6_hdr)) { 3363 tcp_hc_updatemtu(&inc, mtu); 3364 tcp_mtudisc(inp, mtu); 3365 ICMP6STAT_INC(icp6s_pmtuchg); 3366 } 3367 } else 3368 inp = (*notify)(inp, errno); 3369 } 3370 } else { 3371 bzero(&inc, sizeof(inc)); 3372 inc.inc_fibnum = M_GETFIB(m); 3373 inc.inc_flags |= INC_ISIPV6; 3374 inc.inc_fport = t_ports.th_dport; 3375 inc.inc_lport = t_ports.th_sport; 3376 inc.inc6_faddr = *dst; 3377 inc.inc6_laddr = ip6->ip6_src; 3378 syncache_unreach(&inc, icmp_tcp_seq, port); 3379 } 3380 out: 3381 if (inp != NULL) 3382 INP_WUNLOCK(inp); 3383 } 3384 3385 static void 3386 tcp6_ctlinput(struct ip6ctlparam *ctl) 3387 { 3388 tcp6_ctlinput_with_port(ctl, htons(0)); 3389 } 3390 3391 static void 3392 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3393 { 3394 struct ip6ctlparam *ip6cp = param.ip6cp; 3395 struct mbuf *m; 3396 struct udphdr *udp; 3397 uint16_t port; 3398 3399 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3400 if (m == NULL) { 3401 return; 3402 } 3403 udp = mtod(m, struct udphdr *); 3404 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3405 return; 3406 } 3407 port = udp->uh_dport; 3408 m_adj(m, sizeof(struct udphdr)); 3409 if ((m->m_flags & M_PKTHDR) == 0) { 3410 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3411 } 3412 /* Now call in to the normal handling code */ 3413 tcp6_ctlinput_with_port(ip6cp, port); 3414 } 3415 3416 #endif /* INET6 */ 3417 3418 static uint32_t 3419 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3420 { 3421 SIPHASH_CTX ctx; 3422 uint32_t hash[2]; 3423 3424 KASSERT(len >= SIPHASH_KEY_LENGTH, 3425 ("%s: keylen %u too short ", __func__, len)); 3426 SipHash24_Init(&ctx); 3427 SipHash_SetKey(&ctx, (uint8_t *)key); 3428 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3429 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3430 switch (inc->inc_flags & INC_ISIPV6) { 3431 #ifdef INET 3432 case 0: 3433 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3434 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3435 break; 3436 #endif 3437 #ifdef INET6 3438 case INC_ISIPV6: 3439 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3440 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3441 break; 3442 #endif 3443 } 3444 SipHash_Final((uint8_t *)hash, &ctx); 3445 3446 return (hash[0] ^ hash[1]); 3447 } 3448 3449 uint32_t 3450 tcp_new_ts_offset(struct in_conninfo *inc) 3451 { 3452 struct in_conninfo inc_store, *local_inc; 3453 3454 if (!V_tcp_ts_offset_per_conn) { 3455 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3456 inc_store.inc_lport = 0; 3457 inc_store.inc_fport = 0; 3458 local_inc = &inc_store; 3459 } else { 3460 local_inc = inc; 3461 } 3462 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3463 sizeof(V_ts_offset_secret))); 3464 } 3465 3466 /* 3467 * Following is where TCP initial sequence number generation occurs. 3468 * 3469 * There are two places where we must use initial sequence numbers: 3470 * 1. In SYN-ACK packets. 3471 * 2. In SYN packets. 3472 * 3473 * All ISNs for SYN-ACK packets are generated by the syncache. See 3474 * tcp_syncache.c for details. 3475 * 3476 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3477 * depends on this property. In addition, these ISNs should be 3478 * unguessable so as to prevent connection hijacking. To satisfy 3479 * the requirements of this situation, the algorithm outlined in 3480 * RFC 1948 is used, with only small modifications. 3481 * 3482 * Implementation details: 3483 * 3484 * Time is based off the system timer, and is corrected so that it 3485 * increases by one megabyte per second. This allows for proper 3486 * recycling on high speed LANs while still leaving over an hour 3487 * before rollover. 3488 * 3489 * As reading the *exact* system time is too expensive to be done 3490 * whenever setting up a TCP connection, we increment the time 3491 * offset in two ways. First, a small random positive increment 3492 * is added to isn_offset for each connection that is set up. 3493 * Second, the function tcp_isn_tick fires once per clock tick 3494 * and increments isn_offset as necessary so that sequence numbers 3495 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3496 * random positive increments serve only to ensure that the same 3497 * exact sequence number is never sent out twice (as could otherwise 3498 * happen when a port is recycled in less than the system tick 3499 * interval.) 3500 * 3501 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3502 * between seeding of isn_secret. This is normally set to zero, 3503 * as reseeding should not be necessary. 3504 * 3505 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3506 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3507 * general, this means holding an exclusive (write) lock. 3508 */ 3509 3510 #define ISN_BYTES_PER_SECOND 1048576 3511 #define ISN_STATIC_INCREMENT 4096 3512 #define ISN_RANDOM_INCREMENT (4096 - 1) 3513 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3514 3515 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3516 VNET_DEFINE_STATIC(int, isn_last); 3517 VNET_DEFINE_STATIC(int, isn_last_reseed); 3518 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3519 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3520 3521 #define V_isn_secret VNET(isn_secret) 3522 #define V_isn_last VNET(isn_last) 3523 #define V_isn_last_reseed VNET(isn_last_reseed) 3524 #define V_isn_offset VNET(isn_offset) 3525 #define V_isn_offset_old VNET(isn_offset_old) 3526 3527 tcp_seq 3528 tcp_new_isn(struct in_conninfo *inc) 3529 { 3530 tcp_seq new_isn; 3531 u_int32_t projected_offset; 3532 3533 ISN_LOCK(); 3534 /* Seed if this is the first use, reseed if requested. */ 3535 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3536 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3537 < (u_int)ticks))) { 3538 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3539 V_isn_last_reseed = ticks; 3540 } 3541 3542 /* Compute the hash and return the ISN. */ 3543 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3544 sizeof(V_isn_secret)); 3545 V_isn_offset += ISN_STATIC_INCREMENT + 3546 (arc4random() & ISN_RANDOM_INCREMENT); 3547 if (ticks != V_isn_last) { 3548 projected_offset = V_isn_offset_old + 3549 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3550 if (SEQ_GT(projected_offset, V_isn_offset)) 3551 V_isn_offset = projected_offset; 3552 V_isn_offset_old = V_isn_offset; 3553 V_isn_last = ticks; 3554 } 3555 new_isn += V_isn_offset; 3556 ISN_UNLOCK(); 3557 return (new_isn); 3558 } 3559 3560 /* 3561 * When a specific ICMP unreachable message is received and the 3562 * connection state is SYN-SENT, drop the connection. This behavior 3563 * is controlled by the icmp_may_rst sysctl. 3564 */ 3565 static struct inpcb * 3566 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3567 { 3568 struct tcpcb *tp; 3569 3570 NET_EPOCH_ASSERT(); 3571 INP_WLOCK_ASSERT(inp); 3572 3573 tp = intotcpcb(inp); 3574 if (tp->t_state != TCPS_SYN_SENT) 3575 return (inp); 3576 3577 if (tp->t_flags & TF_FASTOPEN) 3578 tcp_fastopen_disable_path(tp); 3579 3580 tp = tcp_drop(tp, errno); 3581 if (tp != NULL) 3582 return (inp); 3583 else 3584 return (NULL); 3585 } 3586 3587 /* 3588 * When `need fragmentation' ICMP is received, update our idea of the MSS 3589 * based on the new value. Also nudge TCP to send something, since we 3590 * know the packet we just sent was dropped. 3591 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3592 */ 3593 static struct inpcb * 3594 tcp_mtudisc_notify(struct inpcb *inp, int error) 3595 { 3596 3597 return (tcp_mtudisc(inp, -1)); 3598 } 3599 3600 static struct inpcb * 3601 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3602 { 3603 struct tcpcb *tp; 3604 struct socket *so; 3605 3606 INP_WLOCK_ASSERT(inp); 3607 3608 tp = intotcpcb(inp); 3609 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3610 3611 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3612 3613 so = inp->inp_socket; 3614 SOCK_SENDBUF_LOCK(so); 3615 /* If the mss is larger than the socket buffer, decrease the mss. */ 3616 if (so->so_snd.sb_hiwat < tp->t_maxseg) { 3617 tp->t_maxseg = so->so_snd.sb_hiwat; 3618 if (tp->t_maxseg < V_tcp_mssdflt) { 3619 /* 3620 * The MSS is so small we should not process incoming 3621 * SACK's since we are subject to attack in such a 3622 * case. 3623 */ 3624 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3625 } else { 3626 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3627 } 3628 } 3629 SOCK_SENDBUF_UNLOCK(so); 3630 3631 TCPSTAT_INC(tcps_mturesent); 3632 tp->t_rtttime = 0; 3633 tp->snd_nxt = tp->snd_una; 3634 tcp_free_sackholes(tp); 3635 tp->snd_recover = tp->snd_max; 3636 if (tp->t_flags & TF_SACK_PERMIT) 3637 EXIT_FASTRECOVERY(tp->t_flags); 3638 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3639 /* 3640 * Conceptually the snd_nxt setting 3641 * and freeing sack holes should 3642 * be done by the default stacks 3643 * own tfb_tcp_mtu_chg(). 3644 */ 3645 tp->t_fb->tfb_tcp_mtu_chg(tp); 3646 } 3647 if (tcp_output(tp) < 0) 3648 return (NULL); 3649 else 3650 return (inp); 3651 } 3652 3653 #ifdef INET 3654 /* 3655 * Look-up the routing entry to the peer of this inpcb. If no route 3656 * is found and it cannot be allocated, then return 0. This routine 3657 * is called by TCP routines that access the rmx structure and by 3658 * tcp_mss_update to get the peer/interface MTU. 3659 */ 3660 uint32_t 3661 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3662 { 3663 struct nhop_object *nh; 3664 struct ifnet *ifp; 3665 uint32_t maxmtu = 0; 3666 3667 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3668 3669 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3670 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3671 if (nh == NULL) 3672 return (0); 3673 3674 ifp = nh->nh_ifp; 3675 maxmtu = nh->nh_mtu; 3676 3677 /* Report additional interface capabilities. */ 3678 if (cap != NULL) { 3679 if (ifp->if_capenable & IFCAP_TSO4 && 3680 ifp->if_hwassist & CSUM_TSO) { 3681 cap->ifcap |= CSUM_TSO; 3682 cap->tsomax = ifp->if_hw_tsomax; 3683 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3684 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3685 /* XXXKIB IFCAP2_IPSEC_OFFLOAD_TSO */ 3686 cap->ipsec_tso = (ifp->if_capenable2 & 3687 IFCAP2_BIT(IFCAP2_IPSEC_OFFLOAD)) != 0; 3688 } 3689 } 3690 } 3691 return (maxmtu); 3692 } 3693 #endif /* INET */ 3694 3695 #ifdef INET6 3696 uint32_t 3697 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3698 { 3699 struct nhop_object *nh; 3700 struct in6_addr dst6; 3701 uint32_t scopeid; 3702 struct ifnet *ifp; 3703 uint32_t maxmtu = 0; 3704 3705 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3706 3707 if (inc->inc_flags & INC_IPV6MINMTU) 3708 return (IPV6_MMTU); 3709 3710 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3711 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3712 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3713 if (nh == NULL) 3714 return (0); 3715 3716 ifp = nh->nh_ifp; 3717 maxmtu = nh->nh_mtu; 3718 3719 /* Report additional interface capabilities. */ 3720 if (cap != NULL) { 3721 if (ifp->if_capenable & IFCAP_TSO6 && 3722 ifp->if_hwassist & CSUM_TSO) { 3723 cap->ifcap |= CSUM_TSO; 3724 cap->tsomax = ifp->if_hw_tsomax; 3725 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3726 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3727 cap->ipsec_tso = false; /* XXXKIB */ 3728 } 3729 } 3730 } 3731 3732 return (maxmtu); 3733 } 3734 3735 /* 3736 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3737 * 3738 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3739 * The right place to do that is ip6_setpktopt() that has just been 3740 * executed. By the way it just filled ip6po_minmtu for us. 3741 */ 3742 void 3743 tcp6_use_min_mtu(struct tcpcb *tp) 3744 { 3745 struct inpcb *inp = tptoinpcb(tp); 3746 3747 INP_WLOCK_ASSERT(inp); 3748 /* 3749 * In case of the IPV6_USE_MIN_MTU socket 3750 * option, the INC_IPV6MINMTU flag to announce 3751 * a corresponding MSS during the initial 3752 * handshake. If the TCP connection is not in 3753 * the front states, just reduce the MSS being 3754 * used. This avoids the sending of TCP 3755 * segments which will be fragmented at the 3756 * IPv6 layer. 3757 */ 3758 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3759 if ((tp->t_state >= TCPS_SYN_SENT) && 3760 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3761 struct ip6_pktopts *opt; 3762 3763 opt = inp->in6p_outputopts; 3764 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3765 tp->t_maxseg > TCP6_MSS) { 3766 tp->t_maxseg = TCP6_MSS; 3767 if (tp->t_maxseg < V_tcp_mssdflt) { 3768 /* 3769 * The MSS is so small we should not process incoming 3770 * SACK's since we are subject to attack in such a 3771 * case. 3772 */ 3773 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3774 } else { 3775 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3776 } 3777 } 3778 } 3779 } 3780 #endif /* INET6 */ 3781 3782 /* 3783 * Calculate effective SMSS per RFC5681 definition for a given TCP 3784 * connection at its current state, taking into account SACK and etc. 3785 */ 3786 u_int 3787 tcp_maxseg(const struct tcpcb *tp) 3788 { 3789 u_int optlen; 3790 3791 if (tp->t_flags & TF_NOOPT) 3792 return (tp->t_maxseg); 3793 3794 /* 3795 * Here we have a simplified code from tcp_addoptions(), 3796 * without a proper loop, and having most of paddings hardcoded. 3797 * We might make mistakes with padding here in some edge cases, 3798 * but this is harmless, since result of tcp_maxseg() is used 3799 * only in cwnd and ssthresh estimations. 3800 */ 3801 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3802 if (tp->t_flags & TF_RCVD_TSTMP) 3803 optlen = TCPOLEN_TSTAMP_APPA; 3804 else 3805 optlen = 0; 3806 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3807 if (tp->t_flags & TF_SIGNATURE) 3808 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3809 #endif 3810 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3811 optlen += TCPOLEN_SACKHDR; 3812 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3813 optlen = PADTCPOLEN(optlen); 3814 } 3815 } else { 3816 if (tp->t_flags & TF_REQ_TSTMP) 3817 optlen = TCPOLEN_TSTAMP_APPA; 3818 else 3819 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3820 if (tp->t_flags & TF_REQ_SCALE) 3821 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3822 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3823 if (tp->t_flags & TF_SIGNATURE) 3824 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3825 #endif 3826 if (tp->t_flags & TF_SACK_PERMIT) 3827 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3828 } 3829 optlen = min(optlen, TCP_MAXOLEN); 3830 return (tp->t_maxseg - optlen); 3831 } 3832 3833 3834 u_int 3835 tcp_fixed_maxseg(const struct tcpcb *tp) 3836 { 3837 int optlen; 3838 3839 if (tp->t_flags & TF_NOOPT) 3840 return (tp->t_maxseg); 3841 3842 /* 3843 * Here we have a simplified code from tcp_addoptions(), 3844 * without a proper loop, and having most of paddings hardcoded. 3845 * We only consider fixed options that we would send every 3846 * time I.e. SACK is not considered. This is important 3847 * for cc modules to figure out what the modulo of the 3848 * cwnd should be. 3849 */ 3850 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3851 if (tp->t_flags & TF_RCVD_TSTMP) 3852 optlen = TCPOLEN_TSTAMP_APPA; 3853 else 3854 optlen = 0; 3855 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3856 if (tp->t_flags & TF_SIGNATURE) 3857 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3858 #endif 3859 } else { 3860 if (tp->t_flags & TF_REQ_TSTMP) 3861 optlen = TCPOLEN_TSTAMP_APPA; 3862 else 3863 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3864 if (tp->t_flags & TF_REQ_SCALE) 3865 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3866 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3867 if (tp->t_flags & TF_SIGNATURE) 3868 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3869 #endif 3870 if (tp->t_flags & TF_SACK_PERMIT) 3871 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3872 } 3873 optlen = min(optlen, TCP_MAXOLEN); 3874 return (tp->t_maxseg - optlen); 3875 } 3876 3877 3878 3879 static int 3880 sysctl_drop(SYSCTL_HANDLER_ARGS) 3881 { 3882 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3883 struct sockaddr_storage addrs[2]; 3884 struct inpcb *inp; 3885 struct tcpcb *tp; 3886 #ifdef INET 3887 struct sockaddr_in *fin = NULL, *lin = NULL; 3888 #endif 3889 struct epoch_tracker et; 3890 #ifdef INET6 3891 struct sockaddr_in6 *fin6, *lin6; 3892 #endif 3893 int error; 3894 3895 inp = NULL; 3896 #ifdef INET6 3897 fin6 = lin6 = NULL; 3898 #endif 3899 error = 0; 3900 3901 if (req->oldptr != NULL || req->oldlen != 0) 3902 return (EINVAL); 3903 if (req->newptr == NULL) 3904 return (EPERM); 3905 if (req->newlen < sizeof(addrs)) 3906 return (ENOMEM); 3907 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3908 if (error) 3909 return (error); 3910 3911 switch (addrs[0].ss_family) { 3912 #ifdef INET6 3913 case AF_INET6: 3914 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3915 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3916 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3917 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3918 return (EINVAL); 3919 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3920 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3921 return (EINVAL); 3922 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3923 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3924 #ifdef INET 3925 fin = (struct sockaddr_in *)&addrs[0]; 3926 lin = (struct sockaddr_in *)&addrs[1]; 3927 #endif 3928 break; 3929 } 3930 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3931 if (error) 3932 return (error); 3933 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3934 if (error) 3935 return (error); 3936 break; 3937 #endif 3938 #ifdef INET 3939 case AF_INET: 3940 fin = (struct sockaddr_in *)&addrs[0]; 3941 lin = (struct sockaddr_in *)&addrs[1]; 3942 if (fin->sin_len != sizeof(struct sockaddr_in) || 3943 lin->sin_len != sizeof(struct sockaddr_in)) 3944 return (EINVAL); 3945 break; 3946 #endif 3947 default: 3948 return (EINVAL); 3949 } 3950 NET_EPOCH_ENTER(et); 3951 switch (addrs[0].ss_family) { 3952 #ifdef INET6 3953 case AF_INET6: 3954 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3955 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3956 INPLOOKUP_WLOCKPCB, NULL); 3957 break; 3958 #endif 3959 #ifdef INET 3960 case AF_INET: 3961 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3962 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3963 break; 3964 #endif 3965 } 3966 if (inp != NULL) { 3967 if (!SOLISTENING(inp->inp_socket)) { 3968 tp = intotcpcb(inp); 3969 tp = tcp_drop(tp, ECONNABORTED); 3970 if (tp != NULL) 3971 INP_WUNLOCK(inp); 3972 } else 3973 INP_WUNLOCK(inp); 3974 } else 3975 error = ESRCH; 3976 NET_EPOCH_EXIT(et); 3977 return (error); 3978 } 3979 3980 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3981 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3982 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3983 "Drop TCP connection"); 3984 3985 static int 3986 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3987 { 3988 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3989 &tcp_ctloutput_set)); 3990 } 3991 3992 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3993 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3994 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3995 "Set socket option for TCP endpoint"); 3996 3997 #ifdef KERN_TLS 3998 static int 3999 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 4000 { 4001 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 4002 struct sockaddr_storage addrs[2]; 4003 struct inpcb *inp; 4004 #ifdef INET 4005 struct sockaddr_in *fin = NULL, *lin = NULL; 4006 #endif 4007 struct epoch_tracker et; 4008 #ifdef INET6 4009 struct sockaddr_in6 *fin6, *lin6; 4010 #endif 4011 int error; 4012 4013 inp = NULL; 4014 #ifdef INET6 4015 fin6 = lin6 = NULL; 4016 #endif 4017 error = 0; 4018 4019 if (req->oldptr != NULL || req->oldlen != 0) 4020 return (EINVAL); 4021 if (req->newptr == NULL) 4022 return (EPERM); 4023 if (req->newlen < sizeof(addrs)) 4024 return (ENOMEM); 4025 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 4026 if (error) 4027 return (error); 4028 4029 switch (addrs[0].ss_family) { 4030 #ifdef INET6 4031 case AF_INET6: 4032 fin6 = (struct sockaddr_in6 *)&addrs[0]; 4033 lin6 = (struct sockaddr_in6 *)&addrs[1]; 4034 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 4035 lin6->sin6_len != sizeof(struct sockaddr_in6)) 4036 return (EINVAL); 4037 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 4038 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 4039 return (EINVAL); 4040 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 4041 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 4042 #ifdef INET 4043 fin = (struct sockaddr_in *)&addrs[0]; 4044 lin = (struct sockaddr_in *)&addrs[1]; 4045 #endif 4046 break; 4047 } 4048 error = sa6_embedscope(fin6, V_ip6_use_defzone); 4049 if (error) 4050 return (error); 4051 error = sa6_embedscope(lin6, V_ip6_use_defzone); 4052 if (error) 4053 return (error); 4054 break; 4055 #endif 4056 #ifdef INET 4057 case AF_INET: 4058 fin = (struct sockaddr_in *)&addrs[0]; 4059 lin = (struct sockaddr_in *)&addrs[1]; 4060 if (fin->sin_len != sizeof(struct sockaddr_in) || 4061 lin->sin_len != sizeof(struct sockaddr_in)) 4062 return (EINVAL); 4063 break; 4064 #endif 4065 default: 4066 return (EINVAL); 4067 } 4068 NET_EPOCH_ENTER(et); 4069 switch (addrs[0].ss_family) { 4070 #ifdef INET6 4071 case AF_INET6: 4072 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 4073 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 4074 INPLOOKUP_WLOCKPCB, NULL); 4075 break; 4076 #endif 4077 #ifdef INET 4078 case AF_INET: 4079 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 4080 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 4081 break; 4082 #endif 4083 } 4084 NET_EPOCH_EXIT(et); 4085 if (inp != NULL) { 4086 struct socket *so; 4087 4088 so = inp->inp_socket; 4089 soref(so); 4090 error = ktls_set_tx_mode(so, 4091 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 4092 INP_WUNLOCK(inp); 4093 sorele(so); 4094 } else 4095 error = ESRCH; 4096 return (error); 4097 } 4098 4099 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 4100 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 4101 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 4102 "Switch TCP connection to SW TLS"); 4103 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 4104 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 4105 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 4106 "Switch TCP connection to ifnet TLS"); 4107 #endif 4108 4109 /* 4110 * Generate a standardized TCP log line for use throughout the 4111 * tcp subsystem. Memory allocation is done with M_NOWAIT to 4112 * allow use in the interrupt context. 4113 * 4114 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 4115 * NB: The function may return NULL if memory allocation failed. 4116 * 4117 * Due to header inclusion and ordering limitations the struct ip 4118 * and ip6_hdr pointers have to be passed as void pointers. 4119 */ 4120 char * 4121 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4122 const void *ip6hdr) 4123 { 4124 4125 /* Is logging enabled? */ 4126 if (V_tcp_log_in_vain == 0) 4127 return (NULL); 4128 4129 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 4130 } 4131 4132 char * 4133 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4134 const void *ip6hdr) 4135 { 4136 4137 /* Is logging enabled? */ 4138 if (tcp_log_debug == 0) 4139 return (NULL); 4140 4141 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 4142 } 4143 4144 static char * 4145 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4146 const void *ip6hdr) 4147 { 4148 char *s, *sp; 4149 size_t size; 4150 #ifdef INET 4151 const struct ip *ip = (const struct ip *)ip4hdr; 4152 #endif 4153 #ifdef INET6 4154 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 4155 #endif /* INET6 */ 4156 4157 /* 4158 * The log line looks like this: 4159 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 4160 */ 4161 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 4162 sizeof(PRINT_TH_FLAGS) + 1 + 4163 #ifdef INET6 4164 2 * INET6_ADDRSTRLEN; 4165 #else 4166 2 * INET_ADDRSTRLEN; 4167 #endif /* INET6 */ 4168 4169 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 4170 if (s == NULL) 4171 return (NULL); 4172 4173 strcat(s, "TCP: ["); 4174 sp = s + strlen(s); 4175 4176 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 4177 inet_ntoa_r(inc->inc_faddr, sp); 4178 sp = s + strlen(s); 4179 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4180 sp = s + strlen(s); 4181 inet_ntoa_r(inc->inc_laddr, sp); 4182 sp = s + strlen(s); 4183 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4184 #ifdef INET6 4185 } else if (inc) { 4186 ip6_sprintf(sp, &inc->inc6_faddr); 4187 sp = s + strlen(s); 4188 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4189 sp = s + strlen(s); 4190 ip6_sprintf(sp, &inc->inc6_laddr); 4191 sp = s + strlen(s); 4192 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4193 } else if (ip6 && th) { 4194 ip6_sprintf(sp, &ip6->ip6_src); 4195 sp = s + strlen(s); 4196 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4197 sp = s + strlen(s); 4198 ip6_sprintf(sp, &ip6->ip6_dst); 4199 sp = s + strlen(s); 4200 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4201 #endif /* INET6 */ 4202 #ifdef INET 4203 } else if (ip && th) { 4204 inet_ntoa_r(ip->ip_src, sp); 4205 sp = s + strlen(s); 4206 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4207 sp = s + strlen(s); 4208 inet_ntoa_r(ip->ip_dst, sp); 4209 sp = s + strlen(s); 4210 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4211 #endif /* INET */ 4212 } else { 4213 free(s, M_TCPLOG); 4214 return (NULL); 4215 } 4216 sp = s + strlen(s); 4217 if (th) 4218 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 4219 if (*(s + size - 1) != '\0') 4220 panic("%s: string too long", __func__); 4221 return (s); 4222 } 4223 4224 /* 4225 * A subroutine which makes it easy to track TCP state changes with DTrace. 4226 * This function shouldn't be called for t_state initializations that don't 4227 * correspond to actual TCP state transitions. 4228 */ 4229 void 4230 tcp_state_change(struct tcpcb *tp, int newstate) 4231 { 4232 #if defined(KDTRACE_HOOKS) 4233 int pstate = tp->t_state; 4234 #endif 4235 4236 TCPSTATES_DEC(tp->t_state); 4237 TCPSTATES_INC(newstate); 4238 tp->t_state = newstate; 4239 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4240 } 4241 4242 /* 4243 * Create an external-format (``xtcpcb'') structure using the information in 4244 * the kernel-format tcpcb structure pointed to by tp. This is done to 4245 * reduce the spew of irrelevant information over this interface, to isolate 4246 * user code from changes in the kernel structure, and potentially to provide 4247 * information-hiding if we decide that some of this information should be 4248 * hidden from users. 4249 */ 4250 void 4251 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4252 { 4253 struct tcpcb *tp = intotcpcb(inp); 4254 sbintime_t now; 4255 4256 bzero(xt, sizeof(*xt)); 4257 xt->t_state = tp->t_state; 4258 xt->t_logstate = tcp_get_bblog_state(tp); 4259 xt->t_flags = tp->t_flags; 4260 xt->t_sndzerowin = tp->t_sndzerowin; 4261 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4262 xt->t_rcvoopack = tp->t_rcvoopack; 4263 xt->t_rcv_wnd = tp->rcv_wnd; 4264 xt->t_snd_wnd = tp->snd_wnd; 4265 xt->t_snd_cwnd = tp->snd_cwnd; 4266 xt->t_snd_ssthresh = tp->snd_ssthresh; 4267 xt->t_dsack_bytes = tp->t_dsack_bytes; 4268 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4269 xt->t_dsack_pack = tp->t_dsack_pack; 4270 xt->t_maxseg = tp->t_maxseg; 4271 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4272 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4273 4274 now = getsbinuptime(); 4275 #define COPYTIMER(which,where) do { \ 4276 if (tp->t_timers[which] != SBT_MAX) \ 4277 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 4278 else \ 4279 xt->where = 0; \ 4280 } while (0) 4281 COPYTIMER(TT_DELACK, tt_delack); 4282 COPYTIMER(TT_REXMT, tt_rexmt); 4283 COPYTIMER(TT_PERSIST, tt_persist); 4284 COPYTIMER(TT_KEEP, tt_keep); 4285 COPYTIMER(TT_2MSL, tt_2msl); 4286 #undef COPYTIMER 4287 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4288 4289 xt->xt_encaps_port = tp->t_port; 4290 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4291 TCP_FUNCTION_NAME_LEN_MAX); 4292 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 4293 #ifdef TCP_BLACKBOX 4294 (void)tcp_log_get_id(tp, xt->xt_logid); 4295 xt->t_lognum = tp->t_lognum; 4296 xt->t_loglimit = tp->t_loglimit; 4297 xt->t_logsn = tp->t_logsn; 4298 #endif 4299 4300 xt->xt_len = sizeof(struct xtcpcb); 4301 in_pcbtoxinpcb(inp, &xt->xt_inp); 4302 } 4303 4304 void 4305 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4306 { 4307 uint32_t bit, i; 4308 4309 if ((tp == NULL) || 4310 (status > TCP_EI_STATUS_MAX_VALUE) || 4311 (status == 0)) { 4312 /* Invalid */ 4313 return; 4314 } 4315 if (status > (sizeof(uint32_t) * 8)) { 4316 /* Should this be a KASSERT? */ 4317 return; 4318 } 4319 bit = 1U << (status - 1); 4320 if (bit & tp->t_end_info_status) { 4321 /* already logged */ 4322 return; 4323 } 4324 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4325 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4326 tp->t_end_info_bytes[i] = status; 4327 tp->t_end_info_status |= bit; 4328 break; 4329 } 4330 } 4331 } 4332 4333 int 4334 tcp_can_enable_pacing(void) 4335 { 4336 4337 if ((tcp_pacing_limit == -1) || 4338 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4339 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4340 shadow_num_connections = number_of_tcp_connections_pacing; 4341 return (1); 4342 } else { 4343 counter_u64_add(tcp_pacing_failures, 1); 4344 return (0); 4345 } 4346 } 4347 4348 int 4349 tcp_incr_dgp_pacing_cnt(void) 4350 { 4351 if ((tcp_dgp_limit == -1) || 4352 (tcp_dgp_limit > number_of_dgp_connections)) { 4353 atomic_fetchadd_int(&number_of_dgp_connections, 1); 4354 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4355 return (1); 4356 } else { 4357 counter_u64_add(tcp_dgp_failures, 1); 4358 return (0); 4359 } 4360 } 4361 4362 static uint8_t tcp_dgp_warning = 0; 4363 4364 void 4365 tcp_dec_dgp_pacing_cnt(void) 4366 { 4367 uint32_t ret; 4368 4369 ret = atomic_fetchadd_int(&number_of_dgp_connections, -1); 4370 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4371 KASSERT(ret != 0, ("number_of_dgp_connections -1 would cause wrap?")); 4372 if (ret == 0) { 4373 if (tcp_dgp_limit != -1) { 4374 printf("Warning all DGP is now disabled, count decrements invalidly!\n"); 4375 tcp_dgp_limit = 0; 4376 tcp_dgp_warning = 1; 4377 } else if (tcp_dgp_warning == 0) { 4378 printf("Warning DGP pacing is invalid, invalid decrement\n"); 4379 tcp_dgp_warning = 1; 4380 } 4381 } 4382 4383 } 4384 4385 static uint8_t tcp_pacing_warning = 0; 4386 4387 void 4388 tcp_decrement_paced_conn(void) 4389 { 4390 uint32_t ret; 4391 4392 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4393 shadow_num_connections = number_of_tcp_connections_pacing; 4394 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4395 if (ret == 0) { 4396 if (tcp_pacing_limit != -1) { 4397 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4398 tcp_pacing_limit = 0; 4399 } else if (tcp_pacing_warning == 0) { 4400 printf("Warning pacing count is invalid, invalid decrement\n"); 4401 tcp_pacing_warning = 1; 4402 } 4403 } 4404 } 4405 4406 static void 4407 tcp_default_switch_failed(struct tcpcb *tp) 4408 { 4409 /* 4410 * If a switch fails we only need to 4411 * care about two things: 4412 * a) The t_flags2 4413 * and 4414 * b) The timer granularity. 4415 * Timeouts, at least for now, don't use the 4416 * old callout system in the other stacks so 4417 * those are hopefully safe. 4418 */ 4419 tcp_lro_features_off(tp); 4420 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 4421 } 4422 4423 #ifdef TCP_ACCOUNTING 4424 int 4425 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss) 4426 { 4427 if (SEQ_LT(th->th_ack, tp->snd_una)) { 4428 /* Do we have a SACK? */ 4429 if (to->to_flags & TOF_SACK) { 4430 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4431 tp->tcp_cnt_counters[ACK_SACK]++; 4432 } 4433 return (ACK_SACK); 4434 } else { 4435 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4436 tp->tcp_cnt_counters[ACK_BEHIND]++; 4437 } 4438 return (ACK_BEHIND); 4439 } 4440 } else if (th->th_ack == tp->snd_una) { 4441 /* Do we have a SACK? */ 4442 if (to->to_flags & TOF_SACK) { 4443 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4444 tp->tcp_cnt_counters[ACK_SACK]++; 4445 } 4446 return (ACK_SACK); 4447 } else if (tiwin != tp->snd_wnd) { 4448 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4449 tp->tcp_cnt_counters[ACK_RWND]++; 4450 } 4451 return (ACK_RWND); 4452 } else { 4453 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4454 tp->tcp_cnt_counters[ACK_DUPACK]++; 4455 } 4456 return (ACK_DUPACK); 4457 } 4458 } else { 4459 if (!SEQ_GT(th->th_ack, tp->snd_max)) { 4460 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4461 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss); 4462 } 4463 } 4464 if (to->to_flags & TOF_SACK) { 4465 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4466 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++; 4467 } 4468 return (ACK_CUMACK_SACK); 4469 } else { 4470 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4471 tp->tcp_cnt_counters[ACK_CUMACK]++; 4472 } 4473 return (ACK_CUMACK); 4474 } 4475 } 4476 } 4477 #endif 4478 4479 void 4480 tcp_change_time_units(struct tcpcb *tp, int granularity) 4481 { 4482 if (tp->t_tmr_granularity == granularity) { 4483 /* We are there */ 4484 return; 4485 } 4486 if (granularity == TCP_TMR_GRANULARITY_USEC) { 4487 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS), 4488 ("Granularity is not TICKS its %u in tp:%p", 4489 tp->t_tmr_granularity, tp)); 4490 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow); 4491 if (tp->t_srtt > 1) { 4492 uint32_t val, frac; 4493 4494 val = tp->t_srtt >> TCP_RTT_SHIFT; 4495 frac = tp->t_srtt & 0x1f; 4496 tp->t_srtt = TICKS_2_USEC(val); 4497 /* 4498 * frac is the fractional part of the srtt (if any) 4499 * but its in ticks and every bit represents 4500 * 1/32nd of a hz. 4501 */ 4502 if (frac) { 4503 if (hz == 1000) { 4504 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4505 } else { 4506 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4507 } 4508 tp->t_srtt += frac; 4509 } 4510 } 4511 if (tp->t_rttvar) { 4512 uint32_t val, frac; 4513 4514 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT; 4515 frac = tp->t_rttvar & 0x1f; 4516 tp->t_rttvar = TICKS_2_USEC(val); 4517 /* 4518 * frac is the fractional part of the srtt (if any) 4519 * but its in ticks and every bit represents 4520 * 1/32nd of a hz. 4521 */ 4522 if (frac) { 4523 if (hz == 1000) { 4524 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4525 } else { 4526 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4527 } 4528 tp->t_rttvar += frac; 4529 } 4530 } 4531 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC; 4532 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) { 4533 /* Convert back to ticks, with */ 4534 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC), 4535 ("Granularity is not USEC its %u in tp:%p", 4536 tp->t_tmr_granularity, tp)); 4537 if (tp->t_srtt > 1) { 4538 uint32_t val, frac; 4539 4540 val = USEC_2_TICKS(tp->t_srtt); 4541 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4542 tp->t_srtt = val << TCP_RTT_SHIFT; 4543 /* 4544 * frac is the fractional part here is left 4545 * over from converting to hz and shifting. 4546 * We need to convert this to the 5 bit 4547 * remainder. 4548 */ 4549 if (frac) { 4550 if (hz == 1000) { 4551 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4552 } else { 4553 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4554 } 4555 tp->t_srtt += frac; 4556 } 4557 } 4558 if (tp->t_rttvar) { 4559 uint32_t val, frac; 4560 4561 val = USEC_2_TICKS(tp->t_rttvar); 4562 frac = tp->t_rttvar % (HPTS_USEC_IN_SEC / hz); 4563 tp->t_rttvar = val << TCP_RTTVAR_SHIFT; 4564 /* 4565 * frac is the fractional part here is left 4566 * over from converting to hz and shifting. 4567 * We need to convert this to the 4 bit 4568 * remainder. 4569 */ 4570 if (frac) { 4571 if (hz == 1000) { 4572 frac = (((uint64_t)frac * (uint64_t)TCP_RTTVAR_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4573 } else { 4574 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTTVAR_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4575 } 4576 tp->t_rttvar += frac; 4577 } 4578 } 4579 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow); 4580 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 4581 } 4582 #ifdef INVARIANTS 4583 else { 4584 panic("Unknown granularity:%d tp:%p", 4585 granularity, tp); 4586 } 4587 #endif 4588 } 4589 4590 void 4591 tcp_handle_orphaned_packets(struct tcpcb *tp) 4592 { 4593 struct mbuf *save, *m, *prev; 4594 /* 4595 * Called when a stack switch is occuring from the fini() 4596 * of the old stack. We assue the init() as already been 4597 * run of the new stack and it has set the t_flags2 to 4598 * what it supports. This function will then deal with any 4599 * differences i.e. cleanup packets that maybe queued that 4600 * the newstack does not support. 4601 */ 4602 4603 if (tp->t_flags2 & TF2_MBUF_L_ACKS) 4604 return; 4605 if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 && 4606 !STAILQ_EMPTY(&tp->t_inqueue)) { 4607 /* 4608 * It is unsafe to process the packets since a 4609 * reset may be lurking in them (its rare but it 4610 * can occur). If we were to find a RST, then we 4611 * would end up dropping the connection and the 4612 * INP lock, so when we return the caller (tcp_usrreq) 4613 * will blow up when it trys to unlock the inp. 4614 * This new stack does not do any fancy LRO features 4615 * so all we can do is toss the packets. 4616 */ 4617 m = STAILQ_FIRST(&tp->t_inqueue); 4618 STAILQ_INIT(&tp->t_inqueue); 4619 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) 4620 m_freem(m); 4621 } else { 4622 /* 4623 * Here we have a stack that does mbuf queuing but 4624 * does not support compressed ack's. We must 4625 * walk all the mbufs and discard any compressed acks. 4626 */ 4627 STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) { 4628 if (m->m_flags & M_ACKCMP) { 4629 if (m == STAILQ_FIRST(&tp->t_inqueue)) 4630 STAILQ_REMOVE_HEAD(&tp->t_inqueue, 4631 m_stailqpkt); 4632 else 4633 STAILQ_REMOVE_AFTER(&tp->t_inqueue, 4634 prev, m_stailqpkt); 4635 m_freem(m); 4636 } else 4637 prev = m; 4638 } 4639 } 4640 } 4641 4642 #ifdef TCP_REQUEST_TRK 4643 uint32_t 4644 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes) 4645 { 4646 #ifdef KERN_TLS 4647 struct ktls_session *tls; 4648 uint32_t rec_oh, records; 4649 4650 tls = so->so_snd.sb_tls_info; 4651 if (tls == NULL) 4652 return (0); 4653 4654 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen; 4655 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len); 4656 return (records * rec_oh); 4657 #else 4658 return (0); 4659 #endif 4660 } 4661 4662 extern uint32_t tcp_stale_entry_time; 4663 uint32_t tcp_stale_entry_time = 250000; 4664 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW, 4665 &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out"); 4666 4667 void 4668 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req, 4669 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes) 4670 { 4671 if (tcp_bblogging_on(tp)) { 4672 union tcp_log_stackspecific log; 4673 struct timeval tv; 4674 4675 memset(&log, 0, sizeof(log)); 4676 log.u_bbr.inhpts = tcp_in_hpts(tp); 4677 log.u_bbr.flex8 = val; 4678 log.u_bbr.rttProp = req->timestamp; 4679 log.u_bbr.delRate = req->start; 4680 log.u_bbr.cur_del_rate = req->end; 4681 log.u_bbr.flex1 = req->start_seq; 4682 log.u_bbr.flex2 = req->end_seq; 4683 log.u_bbr.flex3 = req->flags; 4684 log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff); 4685 log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff); 4686 log.u_bbr.flex7 = slot; 4687 log.u_bbr.bw_inuse = offset; 4688 /* nbytes = flex6 | epoch */ 4689 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff); 4690 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff); 4691 /* cspr = lt_epoch | pkts_out */ 4692 log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff); 4693 log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff); 4694 log.u_bbr.applimited = tp->t_tcpreq_closed; 4695 log.u_bbr.applimited <<= 8; 4696 log.u_bbr.applimited |= tp->t_tcpreq_open; 4697 log.u_bbr.applimited <<= 8; 4698 log.u_bbr.applimited |= tp->t_tcpreq_req; 4699 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 4700 TCP_LOG_EVENTP(tp, NULL, 4701 &tptosocket(tp)->so_rcv, 4702 &tptosocket(tp)->so_snd, 4703 TCP_LOG_REQ_T, 0, 4704 0, &log, false, &tv); 4705 } 4706 } 4707 4708 void 4709 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent) 4710 { 4711 if (tp->t_tcpreq_req > 0) 4712 tp->t_tcpreq_req--; 4713 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4714 if (tp->t_tcpreq_open > 0) 4715 tp->t_tcpreq_open--; 4716 } else { 4717 if (tp->t_tcpreq_closed > 0) 4718 tp->t_tcpreq_closed--; 4719 } 4720 ent->flags = TCP_TRK_TRACK_FLG_EMPTY; 4721 } 4722 4723 static void 4724 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest) 4725 { 4726 struct tcp_sendfile_track *ent; 4727 uint64_t time_delta, oldest_delta; 4728 int i, oldest, oldest_set = 0, cnt_rm = 0; 4729 4730 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4731 ent = &tp->t_tcpreq_info[i]; 4732 if (ent->flags != TCP_TRK_TRACK_FLG_USED) { 4733 /* 4734 * We only care about closed end ranges 4735 * that are allocated and have no sendfile 4736 * ever touching them. They would be in 4737 * state USED. 4738 */ 4739 continue; 4740 } 4741 if (ts >= ent->localtime) 4742 time_delta = ts - ent->localtime; 4743 else 4744 time_delta = 0; 4745 if (time_delta && 4746 ((oldest_delta < time_delta) || (oldest_set == 0))) { 4747 oldest_set = 1; 4748 oldest = i; 4749 oldest_delta = time_delta; 4750 } 4751 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) { 4752 /* 4753 * No sendfile in a our time-limit 4754 * time to purge it. 4755 */ 4756 cnt_rm++; 4757 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4758 time_delta, 0); 4759 tcp_req_free_a_slot(tp, ent); 4760 } 4761 } 4762 if ((cnt_rm == 0) && rm_oldest && oldest_set) { 4763 ent = &tp->t_tcpreq_info[oldest]; 4764 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4765 oldest_delta, 1); 4766 tcp_req_free_a_slot(tp, ent); 4767 } 4768 } 4769 4770 int 4771 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point) 4772 { 4773 int i, ret = 0; 4774 struct tcp_sendfile_track *ent; 4775 4776 /* Clean up any old closed end requests that are now completed */ 4777 if (tp->t_tcpreq_req == 0) 4778 return (0); 4779 if (tp->t_tcpreq_closed == 0) 4780 return (0); 4781 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4782 ent = &tp->t_tcpreq_info[i]; 4783 /* Skip empty ones */ 4784 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4785 continue; 4786 /* Skip open ones */ 4787 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) 4788 continue; 4789 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4790 /* We are past it -- free it */ 4791 tcp_req_log_req_info(tp, ent, 4792 i, TCP_TRK_REQ_LOG_FREED, 0, 0); 4793 tcp_req_free_a_slot(tp, ent); 4794 ret++; 4795 } 4796 } 4797 return (ret); 4798 } 4799 4800 int 4801 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point) 4802 { 4803 if (tp->t_tcpreq_req == 0) 4804 return (-1); 4805 if (tp->t_tcpreq_closed == 0) 4806 return (-1); 4807 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4808 return (-1); 4809 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4810 return (1); 4811 } 4812 return (0); 4813 } 4814 4815 struct tcp_sendfile_track * 4816 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip) 4817 { 4818 /* 4819 * Given an ack point (th_ack) walk through our entries and 4820 * return the first one found that th_ack goes past the 4821 * end_seq. 4822 */ 4823 struct tcp_sendfile_track *ent; 4824 int i; 4825 4826 if (tp->t_tcpreq_req == 0) { 4827 /* none open */ 4828 return (NULL); 4829 } 4830 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4831 ent = &tp->t_tcpreq_info[i]; 4832 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4833 continue; 4834 if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) { 4835 if (SEQ_GEQ(th_ack, ent->end_seq)) { 4836 *ip = i; 4837 return (ent); 4838 } 4839 } 4840 } 4841 return (NULL); 4842 } 4843 4844 struct tcp_sendfile_track * 4845 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq) 4846 { 4847 struct tcp_sendfile_track *ent; 4848 int i; 4849 4850 if (tp->t_tcpreq_req == 0) { 4851 /* none open */ 4852 return (NULL); 4853 } 4854 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4855 ent = &tp->t_tcpreq_info[i]; 4856 tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH, 4857 (uint64_t)seq, 0); 4858 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4859 continue; 4860 } 4861 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4862 /* 4863 * An open end request only needs to 4864 * match the beginning seq or be 4865 * all we have (once we keep going on 4866 * a open end request we may have a seq 4867 * wrap). 4868 */ 4869 if ((SEQ_GEQ(seq, ent->start_seq)) || 4870 (tp->t_tcpreq_closed == 0)) 4871 return (ent); 4872 } else { 4873 /* 4874 * For this one we need to 4875 * be a bit more careful if its 4876 * completed at least. 4877 */ 4878 if ((SEQ_GEQ(seq, ent->start_seq)) && 4879 (SEQ_LT(seq, ent->end_seq))) { 4880 return (ent); 4881 } 4882 } 4883 } 4884 return (NULL); 4885 } 4886 4887 /* Should this be in its own file tcp_req.c ? */ 4888 struct tcp_sendfile_track * 4889 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups) 4890 { 4891 struct tcp_sendfile_track *fil; 4892 int i, allocated; 4893 4894 /* Allocate the request tracking array on demand */ 4895 if (tp->t_tcpreq_info == NULL) { 4896 tp->t_tcpreq_info = malloc( 4897 sizeof(*tp->t_tcpreq_info) * MAX_TCP_TRK_REQ, 4898 M_TCPREQTRK, M_NOWAIT | M_ZERO); 4899 if (tp->t_tcpreq_info == NULL) 4900 return (NULL); 4901 } 4902 /* In case the stack does not check for completions do so now */ 4903 tcp_req_check_for_comp(tp, tp->snd_una); 4904 /* Check for stale entries */ 4905 if (tp->t_tcpreq_req) 4906 tcp_req_check_for_stale_entries(tp, ts, 4907 (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ)); 4908 /* Check to see if this is a duplicate of one not started */ 4909 if (tp->t_tcpreq_req) { 4910 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4911 fil = &tp->t_tcpreq_info[i]; 4912 if ((fil->flags & TCP_TRK_TRACK_FLG_USED) == 0) 4913 continue; 4914 if ((fil->timestamp == req->timestamp) && 4915 (fil->start == req->start) && 4916 ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) || 4917 (fil->end == req->end))) { 4918 /* 4919 * We already have this request 4920 * and it has not been started with sendfile. 4921 * This probably means the user was returned 4922 * a 4xx of some sort and its going to age 4923 * out, lets not duplicate it. 4924 */ 4925 return (fil); 4926 } 4927 } 4928 } 4929 /* Ok if there is no room at the inn we are in trouble */ 4930 if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) { 4931 tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL); 4932 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4933 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], 4934 i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0); 4935 } 4936 return (NULL); 4937 } 4938 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4939 fil = &tp->t_tcpreq_info[i]; 4940 if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4941 allocated = 1; 4942 fil->flags = TCP_TRK_TRACK_FLG_USED; 4943 fil->timestamp = req->timestamp; 4944 fil->playout_ms = req->playout_ms; 4945 fil->localtime = ts; 4946 fil->start = req->start; 4947 if (req->flags & TCP_LOG_HTTPD_RANGE_END) { 4948 fil->end = req->end; 4949 } else { 4950 fil->end = 0; 4951 fil->flags |= TCP_TRK_TRACK_FLG_OPEN; 4952 } 4953 /* 4954 * We can set the min boundaries to the TCP Sequence space, 4955 * but it might be found to be further up when sendfile 4956 * actually runs on this range (if it ever does). 4957 */ 4958 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc; 4959 fil->start_seq = tp->snd_una + 4960 tptosocket(tp)->so_snd.sb_ccc; 4961 if (req->flags & TCP_LOG_HTTPD_RANGE_END) 4962 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start))); 4963 else 4964 fil->end_seq = 0; 4965 if (tptosocket(tp)->so_snd.sb_tls_info) { 4966 /* 4967 * This session is doing TLS. Take a swag guess 4968 * at the overhead. 4969 */ 4970 fil->end_seq += tcp_estimate_tls_overhead( 4971 tptosocket(tp), (fil->end - fil->start)); 4972 } 4973 tp->t_tcpreq_req++; 4974 if (fil->flags & TCP_TRK_TRACK_FLG_OPEN) 4975 tp->t_tcpreq_open++; 4976 else 4977 tp->t_tcpreq_closed++; 4978 tcp_req_log_req_info(tp, fil, i, 4979 TCP_TRK_REQ_LOG_NEW, 0, 0); 4980 break; 4981 } else 4982 fil = NULL; 4983 } 4984 return (fil); 4985 } 4986 4987 void 4988 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts) 4989 { 4990 (void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1); 4991 } 4992 #endif 4993 4994 void 4995 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err) 4996 { 4997 if (tcp_bblogging_on(tp)) { 4998 struct tcp_log_buffer *l; 4999 5000 l = tcp_log_event(tp, NULL, 5001 &tptosocket(tp)->so_rcv, 5002 &tptosocket(tp)->so_snd, 5003 TCP_LOG_SOCKET_OPT, 5004 err, 0, NULL, 1, 5005 NULL, NULL, 0, NULL); 5006 if (l) { 5007 l->tlb_flex1 = option_num; 5008 l->tlb_flex2 = option_val; 5009 } 5010 } 5011 } 5012 5013 uint32_t 5014 tcp_get_srtt(struct tcpcb *tp, int granularity) 5015 { 5016 uint32_t srtt; 5017 5018 KASSERT(granularity == TCP_TMR_GRANULARITY_USEC || 5019 granularity == TCP_TMR_GRANULARITY_TICKS, 5020 ("%s: called with unexpected granularity %d", __func__, 5021 granularity)); 5022 5023 srtt = tp->t_srtt; 5024 5025 /* 5026 * We only support two granularities. If the stored granularity 5027 * does not match the granularity requested by the caller, 5028 * convert the stored value to the requested unit of granularity. 5029 */ 5030 if (tp->t_tmr_granularity != granularity) { 5031 if (granularity == TCP_TMR_GRANULARITY_USEC) 5032 srtt = TICKS_2_USEC(srtt); 5033 else 5034 srtt = USEC_2_TICKS(srtt); 5035 } 5036 5037 /* 5038 * If the srtt is stored with ticks granularity, we need to 5039 * unshift to get the actual value. We do this after the 5040 * conversion above (if one was necessary) in order to maximize 5041 * precision. 5042 */ 5043 if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS) 5044 srtt = srtt >> TCP_RTT_SHIFT; 5045 5046 return (srtt); 5047 } 5048 5049 void 5050 tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt, 5051 uint8_t is_tlp, bool hw_tls) 5052 { 5053 5054 if (is_tlp) { 5055 tp->t_sndtlppack++; 5056 tp->t_sndtlpbyte += len; 5057 } 5058 /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */ 5059 if (is_rxt) 5060 tp->t_snd_rxt_bytes += len; 5061 else 5062 tp->t_sndbytes += len; 5063 5064 #ifdef KERN_TLS 5065 if (hw_tls && is_rxt && len != 0) { 5066 uint64_t rexmit_percent; 5067 5068 rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) / 5069 (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes)); 5070 if (rexmit_percent > ktls_ifnet_max_rexmit_pct) 5071 ktls_disable_ifnet(tp); 5072 } 5073 #endif 5074 } 5075