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 VNET_DEFINE(uma_zone_t, sack_hole_zone); 296 #define V_sack_hole_zone VNET(sack_hole_zone) 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 1137 static struct mtx isn_mtx; 1138 1139 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 1140 #define ISN_LOCK() mtx_lock(&isn_mtx) 1141 #define ISN_UNLOCK() mtx_unlock(&isn_mtx) 1142 1143 INPCBSTORAGE_DEFINE(tcpcbstor, tcpcb, "tcpinp", "tcp_inpcb", "tcphash"); 1144 1145 /* 1146 * Take a value and get the next power of 2 that doesn't overflow. 1147 * Used to size the tcp_inpcb hash buckets. 1148 */ 1149 static int 1150 maketcp_hashsize(int size) 1151 { 1152 int hashsize; 1153 1154 /* 1155 * auto tune. 1156 * get the next power of 2 higher than maxsockets. 1157 */ 1158 hashsize = 1 << fls(size); 1159 /* catch overflow, and just go one power of 2 smaller */ 1160 if (hashsize < size) { 1161 hashsize = 1 << (fls(size) - 1); 1162 } 1163 return (hashsize); 1164 } 1165 1166 static volatile int next_tcp_stack_id = 1; 1167 1168 /* 1169 * Register a TCP function block with the name provided in the names 1170 * array. (Note that this function does NOT automatically register 1171 * blk->tfb_tcp_block_name as a stack name. Therefore, you should 1172 * explicitly include blk->tfb_tcp_block_name in the list of names if 1173 * you wish to register the stack with that name.) 1174 * 1175 * Either all name registrations will succeed or all will fail. If 1176 * a name registration fails, the function will update the num_names 1177 * argument to point to the array index of the name that encountered 1178 * the failure. 1179 * 1180 * Returns 0 on success, or an error code on failure. 1181 */ 1182 int 1183 register_tcp_functions_as_names(struct tcp_function_block *blk, int wait, 1184 const char *names[], int *num_names) 1185 { 1186 struct tcp_function *f[TCP_FUNCTION_NAME_NUM_MAX]; 1187 struct tcp_function_set fs; 1188 int error, i, num_registered; 1189 1190 KASSERT(names != NULL, ("%s: Called with NULL name list", __func__)); 1191 KASSERT(*num_names > 0, 1192 ("%s: Called with non-positive length of name list", __func__)); 1193 KASSERT(rw_initialized(&tcp_function_lock), 1194 ("%s: called too early", __func__)); 1195 1196 if (*num_names > TCP_FUNCTION_NAME_NUM_MAX) { 1197 /* Too many names. */ 1198 *num_names = 0; 1199 return (E2BIG); 1200 } 1201 if ((blk->tfb_tcp_output == NULL) || 1202 (blk->tfb_tcp_do_segment == NULL) || 1203 (blk->tfb_tcp_ctloutput == NULL) || 1204 (blk->tfb_tcp_handoff_ok == NULL) || 1205 (strlen(blk->tfb_tcp_block_name) == 0)) { 1206 /* These functions are required and a name is needed. */ 1207 *num_names = 0; 1208 return (EINVAL); 1209 } 1210 1211 for (i = 0; i < *num_names; i++) { 1212 f[i] = malloc(sizeof(struct tcp_function), M_TCPFUNCTIONS, wait); 1213 if (f[i] == NULL) { 1214 while (--i >= 0) 1215 free(f[i], M_TCPFUNCTIONS); 1216 *num_names = 0; 1217 return (ENOMEM); 1218 } 1219 } 1220 1221 num_registered = 0; 1222 rw_wlock(&tcp_function_lock); 1223 if (find_tcp_fb_locked(blk, NULL) != NULL) { 1224 /* A TCP function block can only be registered once. */ 1225 error = EALREADY; 1226 goto cleanup; 1227 } 1228 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) { 1229 error = EINVAL; 1230 goto cleanup; 1231 } 1232 refcount_init(&blk->tfb_refcnt, 0); 1233 blk->tfb_id = atomic_fetchadd_int(&next_tcp_stack_id, 1); 1234 for (i = 0; i < *num_names; i++) { 1235 (void)strlcpy(fs.function_set_name, names[i], 1236 sizeof(fs.function_set_name)); 1237 if (find_tcp_functions_locked(&fs) != NULL) { 1238 /* Duplicate name space not allowed */ 1239 error = EALREADY; 1240 goto cleanup; 1241 } 1242 f[i]->tf_fb = blk; 1243 (void)strlcpy(f[i]->tf_name, names[i], sizeof(f[i]->tf_name)); 1244 TAILQ_INSERT_TAIL(&t_functions, f[i], tf_next); 1245 tcp_fb_cnt++; 1246 num_registered++; 1247 } 1248 rw_wunlock(&tcp_function_lock); 1249 return (0); 1250 1251 cleanup: 1252 /* Remove the entries just added. */ 1253 for (i = 0; i < *num_names; i++) { 1254 if (i < num_registered) { 1255 TAILQ_REMOVE(&t_functions, f[i], tf_next); 1256 tcp_fb_cnt--; 1257 } 1258 f[i]->tf_fb = NULL; 1259 free(f[i], M_TCPFUNCTIONS); 1260 } 1261 rw_wunlock(&tcp_function_lock); 1262 *num_names = num_registered; 1263 return (error); 1264 } 1265 1266 /* 1267 * Register a TCP function block using the name provided in the name 1268 * argument. 1269 * 1270 * Returns 0 on success, or an error code on failure. 1271 */ 1272 int 1273 register_tcp_functions_as_name(struct tcp_function_block *blk, const char *name, 1274 int wait) 1275 { 1276 const char *name_list[1]; 1277 int num_names, rv; 1278 1279 num_names = 1; 1280 if (name != NULL) 1281 name_list[0] = name; 1282 else 1283 name_list[0] = blk->tfb_tcp_block_name; 1284 rv = register_tcp_functions_as_names(blk, wait, name_list, &num_names); 1285 return (rv); 1286 } 1287 1288 /* 1289 * Register a TCP function block using the name defined in 1290 * blk->tfb_tcp_block_name. 1291 * 1292 * Returns 0 on success, or an error code on failure. 1293 */ 1294 int 1295 register_tcp_functions(struct tcp_function_block *blk, int wait) 1296 { 1297 1298 return (register_tcp_functions_as_name(blk, NULL, wait)); 1299 } 1300 1301 /* 1302 * Deregister all names associated with a function block. This 1303 * functionally removes the function block from use within the system. 1304 * 1305 * When called with a true quiesce argument, mark the function block 1306 * as being removed so no more stacks will use it and determine 1307 * whether the removal would succeed. 1308 * 1309 * When called with a false quiesce argument, actually attempt the 1310 * removal. 1311 * 1312 * When called with a force argument, attempt to switch all TCBs to 1313 * use the default stack instead of returning EBUSY. 1314 * 1315 * Returns 0 on success (or if the removal would succeed), or an error 1316 * code on failure. 1317 */ 1318 int 1319 deregister_tcp_functions(struct tcp_function_block *blk, bool quiesce, 1320 bool force) 1321 { 1322 struct tcp_function *f; 1323 VNET_ITERATOR_DECL(vnet_iter); 1324 1325 if (blk == &tcp_def_funcblk) { 1326 /* You can't un-register the default */ 1327 return (EPERM); 1328 } 1329 rw_wlock(&tcp_function_lock); 1330 VNET_LIST_RLOCK_NOSLEEP(); 1331 VNET_FOREACH(vnet_iter) { 1332 CURVNET_SET(vnet_iter); 1333 if (blk == V_tcp_func_set_ptr) { 1334 /* You can't free the current default in some vnet. */ 1335 CURVNET_RESTORE(); 1336 VNET_LIST_RUNLOCK_NOSLEEP(); 1337 rw_wunlock(&tcp_function_lock); 1338 return (EBUSY); 1339 } 1340 CURVNET_RESTORE(); 1341 } 1342 VNET_LIST_RUNLOCK_NOSLEEP(); 1343 /* Mark the block so no more stacks can use it. */ 1344 blk->tfb_flags |= TCP_FUNC_BEING_REMOVED; 1345 /* 1346 * If TCBs are still attached to the stack, attempt to switch them 1347 * to the default stack. 1348 */ 1349 if (force && blk->tfb_refcnt) { 1350 struct inpcb *inp; 1351 struct tcpcb *tp; 1352 VNET_ITERATOR_DECL(vnet_iter); 1353 1354 rw_wunlock(&tcp_function_lock); 1355 1356 VNET_LIST_RLOCK(); 1357 VNET_FOREACH(vnet_iter) { 1358 CURVNET_SET(vnet_iter); 1359 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1360 INPLOOKUP_WLOCKPCB); 1361 1362 while ((inp = inp_next(&inpi)) != NULL) { 1363 tp = intotcpcb(inp); 1364 if (tp == NULL || tp->t_fb != blk) 1365 continue; 1366 tcp_switch_back_to_default(tp); 1367 } 1368 CURVNET_RESTORE(); 1369 } 1370 VNET_LIST_RUNLOCK(); 1371 1372 rw_wlock(&tcp_function_lock); 1373 } 1374 if (blk->tfb_refcnt) { 1375 /* TCBs still attached. */ 1376 rw_wunlock(&tcp_function_lock); 1377 return (EBUSY); 1378 } 1379 if (quiesce) { 1380 /* Skip removal. */ 1381 rw_wunlock(&tcp_function_lock); 1382 return (0); 1383 } 1384 /* Remove any function names that map to this function block. */ 1385 while (find_tcp_fb_locked(blk, &f) != NULL) { 1386 TAILQ_REMOVE(&t_functions, f, tf_next); 1387 tcp_fb_cnt--; 1388 f->tf_fb = NULL; 1389 free(f, M_TCPFUNCTIONS); 1390 } 1391 rw_wunlock(&tcp_function_lock); 1392 return (0); 1393 } 1394 1395 static void 1396 tcp_drain(void *ctx __unused, int flags __unused) 1397 { 1398 struct epoch_tracker et; 1399 VNET_ITERATOR_DECL(vnet_iter); 1400 1401 if (!do_tcpdrain) 1402 return; 1403 1404 NET_EPOCH_ENTER(et); 1405 VNET_LIST_RLOCK_NOSLEEP(); 1406 VNET_FOREACH(vnet_iter) { 1407 CURVNET_SET(vnet_iter); 1408 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 1409 INPLOOKUP_WLOCKPCB); 1410 struct inpcb *inpb; 1411 struct tcpcb *tcpb; 1412 1413 /* 1414 * Walk the tcpbs, if existing, and flush the reassembly queue, 1415 * if there is one... 1416 * XXX: The "Net/3" implementation doesn't imply that the TCP 1417 * reassembly queue should be flushed, but in a situation 1418 * where we're really low on mbufs, this is potentially 1419 * useful. 1420 */ 1421 while ((inpb = inp_next(&inpi)) != NULL) { 1422 if ((tcpb = intotcpcb(inpb)) != NULL) { 1423 tcp_reass_flush(tcpb); 1424 tcp_clean_sackreport(tcpb); 1425 #ifdef TCP_BLACKBOX 1426 tcp_log_drain(tcpb); 1427 #endif 1428 } 1429 } 1430 CURVNET_RESTORE(); 1431 } 1432 VNET_LIST_RUNLOCK_NOSLEEP(); 1433 NET_EPOCH_EXIT(et); 1434 } 1435 1436 static void 1437 tcp_vnet_init(void *arg __unused) 1438 { 1439 1440 #ifdef TCP_HHOOK 1441 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 1442 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1443 printf("%s: WARNING: unable to register helper hook\n", __func__); 1444 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 1445 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 1446 printf("%s: WARNING: unable to register helper hook\n", __func__); 1447 #endif 1448 #ifdef STATS 1449 if (tcp_stats_init()) 1450 printf("%s: WARNING: unable to initialise TCP stats\n", 1451 __func__); 1452 #endif 1453 in_pcbinfo_init(&V_tcbinfo, &tcpcbstor, tcp_tcbhashsize, 1454 tcp_tcbhashsize); 1455 1456 syncache_init(); 1457 tcp_hc_init(); 1458 1459 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 1460 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 1461 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 1462 1463 tcp_fastopen_init(); 1464 1465 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 1466 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 1467 1468 V_tcp_msl = TCPTV_MSL; 1469 V_tcp_msl_local = TCPTV_MSL_LOCAL; 1470 arc4rand(&V_ts_offset_secret, sizeof(V_ts_offset_secret), 0); 1471 } 1472 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 1473 tcp_vnet_init, NULL); 1474 1475 static void 1476 tcp_init(void *arg __unused) 1477 { 1478 int hashsize; 1479 1480 tcp_reass_global_init(); 1481 1482 /* XXX virtualize those below? */ 1483 tcp_delacktime = TCPTV_DELACK; 1484 tcp_keepinit = TCPTV_KEEP_INIT; 1485 tcp_keepidle = TCPTV_KEEP_IDLE; 1486 tcp_keepintvl = TCPTV_KEEPINTVL; 1487 tcp_maxpersistidle = TCPTV_KEEP_IDLE; 1488 tcp_rexmit_initial = TCPTV_RTOBASE; 1489 tcp_rexmit_min = TCPTV_MIN; 1490 tcp_rexmit_max = TCPTV_REXMTMAX; 1491 tcp_persmin = TCPTV_PERSMIN; 1492 tcp_persmax = TCPTV_PERSMAX; 1493 tcp_rexmit_slop = TCPTV_CPU_VAR; 1494 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 1495 1496 /* Setup the tcp function block list */ 1497 TAILQ_INIT(&t_functions); 1498 rw_init(&tcp_function_lock, "tcp_func_lock"); 1499 register_tcp_functions(&tcp_def_funcblk, M_WAITOK); 1500 sx_init(&tcpoudp_lock, "TCP over UDP configuration"); 1501 #ifdef TCP_BLACKBOX 1502 /* Initialize the TCP logging data. */ 1503 tcp_log_init(); 1504 #endif 1505 1506 if (tcp_soreceive_stream) { 1507 #ifdef INET 1508 tcp_protosw.pr_soreceive = soreceive_stream; 1509 #endif 1510 #ifdef INET6 1511 tcp6_protosw.pr_soreceive = soreceive_stream; 1512 #endif /* INET6 */ 1513 } 1514 1515 #ifdef INET6 1516 max_protohdr_grow(sizeof(struct ip6_hdr) + sizeof(struct tcphdr)); 1517 #else /* INET6 */ 1518 max_protohdr_grow(sizeof(struct tcpiphdr)); 1519 #endif /* INET6 */ 1520 1521 ISN_LOCK_INIT(); 1522 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 1523 SHUTDOWN_PRI_DEFAULT); 1524 EVENTHANDLER_REGISTER(vm_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1525 EVENTHANDLER_REGISTER(mbuf_lowmem, tcp_drain, NULL, LOWMEM_PRI_DEFAULT); 1526 1527 tcp_inp_lro_direct_queue = counter_u64_alloc(M_WAITOK); 1528 tcp_inp_lro_wokeup_queue = counter_u64_alloc(M_WAITOK); 1529 tcp_inp_lro_compressed = counter_u64_alloc(M_WAITOK); 1530 tcp_inp_lro_locks_taken = counter_u64_alloc(M_WAITOK); 1531 tcp_extra_mbuf = counter_u64_alloc(M_WAITOK); 1532 tcp_would_have_but = counter_u64_alloc(M_WAITOK); 1533 tcp_comp_total = counter_u64_alloc(M_WAITOK); 1534 tcp_uncomp_total = counter_u64_alloc(M_WAITOK); 1535 tcp_bad_csums = counter_u64_alloc(M_WAITOK); 1536 tcp_pacing_failures = counter_u64_alloc(M_WAITOK); 1537 tcp_dgp_failures = counter_u64_alloc(M_WAITOK); 1538 1539 hashsize = tcp_tcbhashsize; 1540 if (hashsize == 0) { 1541 /* 1542 * Auto tune the hash size based on maxsockets. 1543 * A perfect hash would have a 1:1 mapping 1544 * (hashsize = maxsockets) however it's been 1545 * suggested that O(2) average is better. 1546 */ 1547 hashsize = maketcp_hashsize(maxsockets / 4); 1548 /* 1549 * Our historical default is 512, 1550 * do not autotune lower than this. 1551 */ 1552 if (hashsize < 512) 1553 hashsize = 512; 1554 if (bootverbose) 1555 printf("%s: %s auto tuned to %d\n", __func__, 1556 "net.inet.tcp.tcbhashsize", hashsize); 1557 } 1558 /* 1559 * We require a hashsize to be a power of two. 1560 * Previously if it was not a power of two we would just reset it 1561 * back to 512, which could be a nasty surprise if you did not notice 1562 * the error message. 1563 * Instead what we do is clip it to the closest power of two lower 1564 * than the specified hash value. 1565 */ 1566 if (!powerof2(hashsize)) { 1567 int oldhashsize = hashsize; 1568 1569 hashsize = maketcp_hashsize(hashsize); 1570 /* prevent absurdly low value */ 1571 if (hashsize < 16) 1572 hashsize = 16; 1573 printf("%s: WARNING: TCB hash size not a power of 2, " 1574 "clipped from %d to %d.\n", __func__, oldhashsize, 1575 hashsize); 1576 } 1577 tcp_tcbhashsize = hashsize; 1578 1579 #ifdef INET 1580 IPPROTO_REGISTER(IPPROTO_TCP, tcp_input, tcp_ctlinput); 1581 #endif 1582 #ifdef INET6 1583 IP6PROTO_REGISTER(IPPROTO_TCP, tcp6_input, tcp6_ctlinput); 1584 #endif 1585 } 1586 SYSINIT(tcp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, tcp_init, NULL); 1587 1588 #ifdef VIMAGE 1589 static void 1590 tcp_destroy(void *unused __unused) 1591 { 1592 #ifdef TCP_HHOOK 1593 int error; 1594 #endif 1595 1596 tcp_hc_destroy(); 1597 syncache_destroy(); 1598 in_pcbinfo_destroy(&V_tcbinfo); 1599 /* tcp_discardcb() clears the sack_holes up. */ 1600 uma_zdestroy(V_sack_hole_zone); 1601 1602 /* 1603 * Cannot free the zone until all tcpcbs are released as we attach 1604 * the allocations to them. 1605 */ 1606 tcp_fastopen_destroy(); 1607 1608 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 1609 VNET_PCPUSTAT_FREE(tcpstat); 1610 1611 #ifdef TCP_HHOOK 1612 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]); 1613 if (error != 0) { 1614 printf("%s: WARNING: unable to deregister helper hook " 1615 "type=%d, id=%d: error %d returned\n", __func__, 1616 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error); 1617 } 1618 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]); 1619 if (error != 0) { 1620 printf("%s: WARNING: unable to deregister helper hook " 1621 "type=%d, id=%d: error %d returned\n", __func__, 1622 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error); 1623 } 1624 #endif 1625 } 1626 VNET_SYSUNINIT(tcp, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, tcp_destroy, NULL); 1627 #endif 1628 1629 void 1630 tcp_fini(void *xtp) 1631 { 1632 1633 } 1634 1635 /* 1636 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 1637 * tcp_template used to store this data in mbufs, but we now recopy it out 1638 * of the tcpcb each time to conserve mbufs. 1639 */ 1640 void 1641 tcpip_fillheaders(struct inpcb *inp, uint16_t port, void *ip_ptr, void *tcp_ptr) 1642 { 1643 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 1644 1645 INP_WLOCK_ASSERT(inp); 1646 1647 #ifdef INET6 1648 if ((inp->inp_vflag & INP_IPV6) != 0) { 1649 struct ip6_hdr *ip6; 1650 1651 ip6 = (struct ip6_hdr *)ip_ptr; 1652 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 1653 (inp->inp_flow & IPV6_FLOWINFO_MASK); 1654 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 1655 (IPV6_VERSION & IPV6_VERSION_MASK); 1656 if (port == 0) 1657 ip6->ip6_nxt = IPPROTO_TCP; 1658 else 1659 ip6->ip6_nxt = IPPROTO_UDP; 1660 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 1661 ip6->ip6_src = inp->in6p_laddr; 1662 ip6->ip6_dst = inp->in6p_faddr; 1663 } 1664 #endif /* INET6 */ 1665 #if defined(INET6) && defined(INET) 1666 else 1667 #endif 1668 #ifdef INET 1669 { 1670 struct ip *ip; 1671 1672 ip = (struct ip *)ip_ptr; 1673 ip->ip_v = IPVERSION; 1674 ip->ip_hl = 5; 1675 ip->ip_tos = inp->inp_ip_tos; 1676 ip->ip_len = 0; 1677 ip->ip_id = 0; 1678 ip->ip_off = 0; 1679 ip->ip_ttl = inp->inp_ip_ttl; 1680 ip->ip_sum = 0; 1681 if (port == 0) 1682 ip->ip_p = IPPROTO_TCP; 1683 else 1684 ip->ip_p = IPPROTO_UDP; 1685 ip->ip_src = inp->inp_laddr; 1686 ip->ip_dst = inp->inp_faddr; 1687 } 1688 #endif /* INET */ 1689 th->th_sport = inp->inp_lport; 1690 th->th_dport = inp->inp_fport; 1691 th->th_seq = 0; 1692 th->th_ack = 0; 1693 th->th_off = 5; 1694 tcp_set_flags(th, 0); 1695 th->th_win = 0; 1696 th->th_urp = 0; 1697 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 1698 } 1699 1700 /* 1701 * Create template to be used to send tcp packets on a connection. 1702 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 1703 * use for this function is in keepalives, which use tcp_respond. 1704 */ 1705 struct tcptemp * 1706 tcpip_maketemplate(struct inpcb *inp) 1707 { 1708 struct tcptemp *t; 1709 1710 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 1711 if (t == NULL) 1712 return (NULL); 1713 tcpip_fillheaders(inp, 0, (void *)&t->tt_ipgen, (void *)&t->tt_t); 1714 return (t); 1715 } 1716 1717 /* 1718 * Send a single message to the TCP at address specified by 1719 * the given TCP/IP header. If m == NULL, then we make a copy 1720 * of the tcpiphdr at th and send directly to the addressed host. 1721 * This is used to force keep alive messages out using the TCP 1722 * template for a connection. If flags are given then we send 1723 * a message back to the TCP which originated the segment th, 1724 * and discard the mbuf containing it and any other attached mbufs. 1725 * 1726 * In any case the ack and sequence number of the transmitted 1727 * segment are as specified by the parameters. 1728 * 1729 * NOTE: If m != NULL, then th must point to *inside* the mbuf. 1730 */ 1731 1732 void 1733 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 1734 tcp_seq ack, tcp_seq seq, uint16_t flags) 1735 { 1736 struct tcpopt to; 1737 struct inpcb *inp; 1738 struct ip *ip; 1739 struct mbuf *optm; 1740 struct udphdr *uh = NULL; 1741 struct tcphdr *nth; 1742 struct tcp_log_buffer *lgb; 1743 u_char *optp; 1744 #ifdef INET6 1745 struct ip6_hdr *ip6; 1746 int isipv6; 1747 #endif /* INET6 */ 1748 int optlen, tlen, win, ulen; 1749 int ect = 0; 1750 bool incl_opts; 1751 uint16_t port; 1752 int output_ret; 1753 #ifdef INVARIANTS 1754 int thflags = tcp_get_flags(th); 1755 #endif 1756 1757 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 1758 NET_EPOCH_ASSERT(); 1759 1760 #ifdef INET6 1761 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 1762 ip6 = ipgen; 1763 #endif /* INET6 */ 1764 ip = ipgen; 1765 1766 if (tp != NULL) { 1767 inp = tptoinpcb(tp); 1768 INP_LOCK_ASSERT(inp); 1769 } else 1770 inp = NULL; 1771 1772 if (m != NULL) { 1773 #ifdef INET6 1774 if (isipv6 && ip6 && (ip6->ip6_nxt == IPPROTO_UDP)) 1775 port = m->m_pkthdr.tcp_tun_port; 1776 else 1777 #endif 1778 if (ip && (ip->ip_p == IPPROTO_UDP)) 1779 port = m->m_pkthdr.tcp_tun_port; 1780 else 1781 port = 0; 1782 } else 1783 port = tp->t_port; 1784 1785 incl_opts = false; 1786 win = 0; 1787 if (tp != NULL) { 1788 if (!(flags & TH_RST)) { 1789 win = sbspace(&inp->inp_socket->so_rcv); 1790 if (win > TCP_MAXWIN << tp->rcv_scale) 1791 win = TCP_MAXWIN << tp->rcv_scale; 1792 } 1793 if ((tp->t_flags & TF_NOOPT) == 0) 1794 incl_opts = true; 1795 } 1796 if (m == NULL) { 1797 m = m_gethdr(M_NOWAIT, MT_DATA); 1798 if (m == NULL) 1799 return; 1800 m->m_data += max_linkhdr; 1801 #ifdef INET6 1802 if (isipv6) { 1803 bcopy((caddr_t)ip6, mtod(m, caddr_t), 1804 sizeof(struct ip6_hdr)); 1805 ip6 = mtod(m, struct ip6_hdr *); 1806 nth = (struct tcphdr *)(ip6 + 1); 1807 if (port) { 1808 /* Insert a UDP header */ 1809 uh = (struct udphdr *)nth; 1810 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1811 uh->uh_dport = port; 1812 nth = (struct tcphdr *)(uh + 1); 1813 } 1814 } else 1815 #endif /* INET6 */ 1816 { 1817 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 1818 ip = mtod(m, struct ip *); 1819 nth = (struct tcphdr *)(ip + 1); 1820 if (port) { 1821 /* Insert a UDP header */ 1822 uh = (struct udphdr *)nth; 1823 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1824 uh->uh_dport = port; 1825 nth = (struct tcphdr *)(uh + 1); 1826 } 1827 } 1828 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1829 flags = TH_ACK; 1830 } else if ((!M_WRITABLE(m)) || (port != 0)) { 1831 struct mbuf *n; 1832 1833 /* Can't reuse 'm', allocate a new mbuf. */ 1834 n = m_gethdr(M_NOWAIT, MT_DATA); 1835 if (n == NULL) { 1836 m_freem(m); 1837 return; 1838 } 1839 1840 if (!m_dup_pkthdr(n, m, M_NOWAIT)) { 1841 m_freem(m); 1842 m_freem(n); 1843 return; 1844 } 1845 1846 n->m_data += max_linkhdr; 1847 /* m_len is set later */ 1848 #define xchg(a,b,type) { type t; t=a; a=b; b=t; } 1849 #ifdef INET6 1850 if (isipv6) { 1851 bcopy((caddr_t)ip6, mtod(n, caddr_t), 1852 sizeof(struct ip6_hdr)); 1853 ip6 = mtod(n, struct ip6_hdr *); 1854 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1855 nth = (struct tcphdr *)(ip6 + 1); 1856 if (port) { 1857 /* Insert a UDP header */ 1858 uh = (struct udphdr *)nth; 1859 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1860 uh->uh_dport = port; 1861 nth = (struct tcphdr *)(uh + 1); 1862 } 1863 } else 1864 #endif /* INET6 */ 1865 { 1866 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip)); 1867 ip = mtod(n, struct ip *); 1868 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1869 nth = (struct tcphdr *)(ip + 1); 1870 if (port) { 1871 /* Insert a UDP header */ 1872 uh = (struct udphdr *)nth; 1873 uh->uh_sport = htons(V_tcp_udp_tunneling_port); 1874 uh->uh_dport = port; 1875 nth = (struct tcphdr *)(uh + 1); 1876 } 1877 } 1878 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 1879 xchg(nth->th_dport, nth->th_sport, uint16_t); 1880 th = nth; 1881 m_freem(m); 1882 m = n; 1883 } else { 1884 /* 1885 * reuse the mbuf. 1886 * XXX MRT We inherit the FIB, which is lucky. 1887 */ 1888 m_freem(m->m_next); 1889 m->m_next = NULL; 1890 m->m_data = (caddr_t)ipgen; 1891 /* clear any receive flags for proper bpf timestamping */ 1892 m->m_flags &= ~(M_TSTMP | M_TSTMP_LRO); 1893 /* m_len is set later */ 1894 #ifdef INET6 1895 if (isipv6) { 1896 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 1897 nth = (struct tcphdr *)(ip6 + 1); 1898 } else 1899 #endif /* INET6 */ 1900 { 1901 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 1902 nth = (struct tcphdr *)(ip + 1); 1903 } 1904 if (th != nth) { 1905 /* 1906 * this is usually a case when an extension header 1907 * exists between the IPv6 header and the 1908 * TCP header. 1909 */ 1910 nth->th_sport = th->th_sport; 1911 nth->th_dport = th->th_dport; 1912 } 1913 xchg(nth->th_dport, nth->th_sport, uint16_t); 1914 #undef xchg 1915 } 1916 tlen = 0; 1917 #ifdef INET6 1918 if (isipv6) 1919 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 1920 #endif 1921 #if defined(INET) && defined(INET6) 1922 else 1923 #endif 1924 #ifdef INET 1925 tlen = sizeof (struct tcpiphdr); 1926 #endif 1927 if (port) 1928 tlen += sizeof (struct udphdr); 1929 #ifdef INVARIANTS 1930 m->m_len = 0; 1931 KASSERT(M_TRAILINGSPACE(m) >= tlen, 1932 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)", 1933 m, tlen, (long)M_TRAILINGSPACE(m))); 1934 #endif 1935 m->m_len = tlen; 1936 to.to_flags = 0; 1937 if (incl_opts) { 1938 ect = tcp_ecn_output_established(tp, &flags, 0, false); 1939 /* Make sure we have room. */ 1940 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) { 1941 m->m_next = m_get(M_NOWAIT, MT_DATA); 1942 if (m->m_next) { 1943 optp = mtod(m->m_next, u_char *); 1944 optm = m->m_next; 1945 } else 1946 incl_opts = false; 1947 } else { 1948 optp = (u_char *) (nth + 1); 1949 optm = m; 1950 } 1951 } 1952 if (incl_opts) { 1953 /* Timestamps. */ 1954 if (tp->t_flags & TF_RCVD_TSTMP) { 1955 to.to_tsval = tcp_ts_getticks() + tp->ts_offset; 1956 to.to_tsecr = tp->ts_recent; 1957 to.to_flags |= TOF_TS; 1958 } 1959 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1960 /* TCP-MD5 (RFC2385). */ 1961 if (tp->t_flags & TF_SIGNATURE) 1962 to.to_flags |= TOF_SIGNATURE; 1963 #endif 1964 /* Add the options. */ 1965 tlen += optlen = tcp_addoptions(&to, optp); 1966 1967 /* Update m_len in the correct mbuf. */ 1968 optm->m_len += optlen; 1969 } else 1970 optlen = 0; 1971 #ifdef INET6 1972 if (isipv6) { 1973 if (uh) { 1974 ulen = tlen - sizeof(struct ip6_hdr); 1975 uh->uh_ulen = htons(ulen); 1976 } 1977 ip6->ip6_flow = htonl(ect << IPV6_FLOWLABEL_LEN); 1978 ip6->ip6_vfc = IPV6_VERSION; 1979 if (port) 1980 ip6->ip6_nxt = IPPROTO_UDP; 1981 else 1982 ip6->ip6_nxt = IPPROTO_TCP; 1983 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 1984 } 1985 #endif 1986 #if defined(INET) && defined(INET6) 1987 else 1988 #endif 1989 #ifdef INET 1990 { 1991 if (uh) { 1992 ulen = tlen - sizeof(struct ip); 1993 uh->uh_ulen = htons(ulen); 1994 } 1995 ip->ip_len = htons(tlen); 1996 if (inp != NULL) { 1997 ip->ip_tos = inp->inp_ip_tos & ~IPTOS_ECN_MASK; 1998 ip->ip_ttl = inp->inp_ip_ttl; 1999 } else { 2000 ip->ip_tos = 0; 2001 ip->ip_ttl = V_ip_defttl; 2002 } 2003 ip->ip_tos |= ect; 2004 if (port) { 2005 ip->ip_p = IPPROTO_UDP; 2006 } else { 2007 ip->ip_p = IPPROTO_TCP; 2008 } 2009 if (V_path_mtu_discovery) 2010 ip->ip_off |= htons(IP_DF); 2011 } 2012 #endif 2013 m->m_pkthdr.len = tlen; 2014 m->m_pkthdr.rcvif = NULL; 2015 #ifdef MAC 2016 if (inp != NULL) { 2017 /* 2018 * Packet is associated with a socket, so allow the 2019 * label of the response to reflect the socket label. 2020 */ 2021 INP_LOCK_ASSERT(inp); 2022 mac_inpcb_create_mbuf(inp, m); 2023 } else { 2024 /* 2025 * Packet is not associated with a socket, so possibly 2026 * update the label in place. 2027 */ 2028 mac_netinet_tcp_reply(m); 2029 } 2030 #endif 2031 nth->th_seq = htonl(seq); 2032 nth->th_ack = htonl(ack); 2033 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2; 2034 tcp_set_flags(nth, flags); 2035 if (tp && (flags & TH_RST)) { 2036 /* Log the reset */ 2037 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2038 } 2039 if (tp != NULL) 2040 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 2041 else 2042 nth->th_win = htons((u_short)win); 2043 nth->th_urp = 0; 2044 2045 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 2046 if (to.to_flags & TOF_SIGNATURE) { 2047 if (!TCPMD5_ENABLED() || 2048 TCPMD5_OUTPUT(m, nth, to.to_signature) != 0) { 2049 m_freem(m); 2050 return; 2051 } 2052 } 2053 #endif 2054 2055 #ifdef INET6 2056 if (isipv6) { 2057 if (port) { 2058 m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; 2059 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2060 uh->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); 2061 nth->th_sum = 0; 2062 } else { 2063 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 2064 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2065 nth->th_sum = in6_cksum_pseudo(ip6, 2066 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 2067 } 2068 ip6->ip6_hlim = in6_selecthlim(inp, NULL); 2069 } 2070 #endif /* INET6 */ 2071 #if defined(INET6) && defined(INET) 2072 else 2073 #endif 2074 #ifdef INET 2075 { 2076 if (port) { 2077 uh->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2078 htons(ulen + IPPROTO_UDP)); 2079 m->m_pkthdr.csum_flags = CSUM_UDP; 2080 m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); 2081 nth->th_sum = 0; 2082 } else { 2083 m->m_pkthdr.csum_flags = CSUM_TCP; 2084 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 2085 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 2086 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 2087 } 2088 } 2089 #endif /* INET */ 2090 TCP_PROBE3(debug__output, tp, th, m); 2091 if (flags & TH_RST) 2092 TCP_PROBE5(accept__refused, NULL, NULL, m, tp, nth); 2093 lgb = NULL; 2094 if ((tp != NULL) && tcp_bblogging_on(tp)) { 2095 if (INP_WLOCKED(inp)) { 2096 union tcp_log_stackspecific log; 2097 struct timeval tv; 2098 2099 memset(&log, 0, sizeof(log)); 2100 log.u_bbr.inhpts = tcp_in_hpts(tp); 2101 log.u_bbr.flex8 = 4; 2102 log.u_bbr.pkts_out = tp->t_maxseg; 2103 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 2104 log.u_bbr.delivered = 0; 2105 lgb = tcp_log_event(tp, nth, NULL, NULL, TCP_LOG_OUT, 2106 ERRNO_UNK, 0, &log, false, NULL, NULL, 0, &tv); 2107 } else { 2108 /* 2109 * We can not log the packet, since we only own the 2110 * read lock, but a write lock is needed. The read lock 2111 * is not upgraded to a write lock, since only getting 2112 * the read lock was done intentionally to improve the 2113 * handling of SYN flooding attacks. 2114 * This happens only for pure SYN segments received in 2115 * the initial CLOSED state, or received in a more 2116 * advanced state than listen and the UDP encapsulation 2117 * port is unexpected. 2118 * The incoming SYN segments do not really belong to 2119 * the TCP connection and the handling does not change 2120 * the state of the TCP connection. Therefore, the 2121 * sending of the RST segments is not logged. Please 2122 * note that also the incoming SYN segments are not 2123 * logged. 2124 * 2125 * The following code ensures that the above description 2126 * is and stays correct. 2127 */ 2128 KASSERT((thflags & (TH_ACK|TH_SYN)) == TH_SYN && 2129 (tp->t_state == TCPS_CLOSED || 2130 (tp->t_state > TCPS_LISTEN && tp->t_port != port)), 2131 ("%s: Logging of TCP segment with flags 0x%b and " 2132 "UDP encapsulation port %u skipped in state %s", 2133 __func__, thflags, PRINT_TH_FLAGS, 2134 ntohs(port), tcpstates[tp->t_state])); 2135 } 2136 } 2137 2138 if (flags & TH_ACK) 2139 TCPSTAT_INC(tcps_sndacks); 2140 else if (flags & (TH_SYN|TH_FIN|TH_RST)) 2141 TCPSTAT_INC(tcps_sndctrl); 2142 TCPSTAT_INC(tcps_sndtotal); 2143 2144 #ifdef INET6 2145 if (isipv6) { 2146 TCP_PROBE5(send, NULL, tp, ip6, tp, nth); 2147 output_ret = ip6_output(m, inp ? inp->in6p_outputopts : NULL, 2148 NULL, 0, NULL, NULL, inp); 2149 } 2150 #endif /* INET6 */ 2151 #if defined(INET) && defined(INET6) 2152 else 2153 #endif 2154 #ifdef INET 2155 { 2156 TCP_PROBE5(send, NULL, tp, ip, tp, nth); 2157 output_ret = ip_output(m, NULL, NULL, 0, NULL, inp); 2158 } 2159 #endif 2160 if (lgb != NULL) 2161 lgb->tlb_errno = output_ret; 2162 } 2163 2164 /* 2165 * Check that no more than V_tcp_ack_war_cnt per V_tcp_ack_war_time_window 2166 * are sent. *epoch_end is the end of the current epoch and is updated, if the 2167 * current epoch ended in the past. *ack_cnt is the counter used during the 2168 * current epoch. It might be reset and incremented. 2169 * The function returns true if a challenge ACK should be sent. 2170 */ 2171 bool 2172 tcp_challenge_ack_check(sbintime_t *epoch_end, uint32_t *ack_cnt) 2173 { 2174 sbintime_t now; 2175 2176 /* 2177 * The sending of a challenge ACK could be triggered by a blind attacker 2178 * to detect an existing TCP connection. To mitigate that, increment 2179 * also the global counter which would be incremented if the attacker 2180 * would have guessed wrongly. 2181 */ 2182 (void)badport_bandlim(BANDLIM_TCP_RST); 2183 2184 if (V_tcp_ack_war_time_window == 0 || V_tcp_ack_war_cnt == 0) { 2185 /* ACK war protection is disabled. */ 2186 return (true); 2187 } else { 2188 /* Start new epoch, if the previous one is already over. */ 2189 now = getsbinuptime(); 2190 if (*epoch_end < now) { 2191 *ack_cnt = 0; 2192 *epoch_end = now + V_tcp_ack_war_time_window * SBT_1MS; 2193 } 2194 /* 2195 * Send a challenge ACK, if less than tcp_ack_war_cnt have been 2196 * sent in the current epoch. 2197 */ 2198 if (*ack_cnt < V_tcp_ack_war_cnt) { 2199 (*ack_cnt)++; 2200 return (true); 2201 } else { 2202 return (false); 2203 } 2204 } 2205 } 2206 2207 /* 2208 * Send a challenge ack (no data, no SACK option), but not more than 2209 * V_tcp_ack_war_cnt per V_tcp_ack_war_time_window (per TCP connection). 2210 */ 2211 void 2212 tcp_send_challenge_ack(struct tcpcb *tp, struct tcphdr *th, struct mbuf *m) 2213 { 2214 if (tcp_challenge_ack_check(&tp->t_challenge_ack_end, 2215 &tp->t_challenge_ack_cnt)) { 2216 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2217 tp->snd_nxt, TH_ACK); 2218 tp->last_ack_sent = tp->rcv_nxt; 2219 } else { 2220 m_freem(m); 2221 } 2222 } 2223 2224 /* 2225 * Create a new TCP control block, making an empty reassembly queue and hooking 2226 * it to the argument protocol control block. The `inp' parameter must have 2227 * come from the zone allocator set up by tcpcbstor declaration. 2228 * The caller can provide a pointer to a tcpcb of the listener to inherit the 2229 * TCP function block from the listener. 2230 */ 2231 struct tcpcb * 2232 tcp_newtcpcb(struct inpcb *inp, struct tcpcb *listening_tcb) 2233 { 2234 struct tcpcb *tp = intotcpcb(inp); 2235 #ifdef INET6 2236 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2237 #endif /* INET6 */ 2238 2239 /* 2240 * Historically allocation was done with M_ZERO. There is a lot of 2241 * code that rely on that. For now take safe approach and zero whole 2242 * tcpcb. This definitely can be optimized. 2243 */ 2244 bzero(&tp->t_start_zero, t_zero_size); 2245 2246 /* Initialise cc_var struct for this tcpcb. */ 2247 tp->t_ccv.tp = tp; 2248 rw_rlock(&tcp_function_lock); 2249 if (listening_tcb != NULL) { 2250 INP_LOCK_ASSERT(tptoinpcb(listening_tcb)); 2251 KASSERT(listening_tcb->t_fb != NULL, 2252 ("tcp_newtcpcb: listening_tcb->t_fb is NULL")); 2253 if (listening_tcb->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) { 2254 rw_runlock(&tcp_function_lock); 2255 return (NULL); 2256 } 2257 tp->t_fb = listening_tcb->t_fb; 2258 } else { 2259 tp->t_fb = V_tcp_func_set_ptr; 2260 } 2261 refcount_acquire(&tp->t_fb->tfb_refcnt); 2262 KASSERT((tp->t_fb->tfb_flags & TCP_FUNC_BEING_REMOVED) == 0, 2263 ("tcp_newtcpcb: using TFB being removed")); 2264 rw_runlock(&tcp_function_lock); 2265 CC_LIST_RLOCK(); 2266 if (listening_tcb != NULL) { 2267 if (CC_ALGO(listening_tcb)->flags & CC_MODULE_BEING_REMOVED) { 2268 CC_LIST_RUNLOCK(); 2269 if (tp->t_fb->tfb_tcp_fb_fini) 2270 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2271 refcount_release(&tp->t_fb->tfb_refcnt); 2272 return (NULL); 2273 } 2274 CC_ALGO(tp) = CC_ALGO(listening_tcb); 2275 } else 2276 CC_ALGO(tp) = CC_DEFAULT_ALGO(); 2277 cc_refer(CC_ALGO(tp)); 2278 CC_LIST_RUNLOCK(); 2279 if (CC_ALGO(tp)->cb_init != NULL) 2280 if (CC_ALGO(tp)->cb_init(&tp->t_ccv, NULL) > 0) { 2281 cc_detach(tp); 2282 if (tp->t_fb->tfb_tcp_fb_fini) 2283 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2284 refcount_release(&tp->t_fb->tfb_refcnt); 2285 return (NULL); 2286 } 2287 2288 #ifdef TCP_HHOOK 2289 if (khelp_init_osd(HELPER_CLASS_TCP, &tp->t_osd)) { 2290 if (CC_ALGO(tp)->cb_destroy != NULL) 2291 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2292 CC_DATA(tp) = NULL; 2293 cc_detach(tp); 2294 if (tp->t_fb->tfb_tcp_fb_fini) 2295 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2296 refcount_release(&tp->t_fb->tfb_refcnt); 2297 return (NULL); 2298 } 2299 #endif 2300 2301 TAILQ_INIT(&tp->t_segq); 2302 STAILQ_INIT(&tp->t_inqueue); 2303 tp->t_maxseg = 2304 #ifdef INET6 2305 isipv6 ? V_tcp_v6mssdflt : 2306 #endif /* INET6 */ 2307 V_tcp_mssdflt; 2308 2309 /* All mbuf queue/ack compress flags should be off */ 2310 tcp_lro_features_off(tp); 2311 2312 tp->t_hpts_cpu = HPTS_CPU_NONE; 2313 tp->t_lro_cpu = HPTS_CPU_NONE; 2314 2315 callout_init_rw(&tp->t_callout, &inp->inp_lock, 2316 CALLOUT_TRYLOCK | CALLOUT_RETURNUNLOCKED); 2317 for (int i = 0; i < TT_N; i++) 2318 tp->t_timers[i] = SBT_MAX; 2319 2320 switch (V_tcp_do_rfc1323) { 2321 case 0: 2322 break; 2323 default: 2324 case 1: 2325 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 2326 break; 2327 case 2: 2328 tp->t_flags = TF_REQ_SCALE; 2329 break; 2330 case 3: 2331 tp->t_flags = TF_REQ_TSTMP; 2332 break; 2333 } 2334 if (V_tcp_do_sack) 2335 tp->t_flags |= TF_SACK_PERMIT; 2336 TAILQ_INIT(&tp->snd_holes); 2337 2338 /* 2339 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 2340 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 2341 * reasonable initial retransmit time. 2342 */ 2343 tp->t_srtt = TCPTV_SRTTBASE; 2344 tp->t_rttvar = ((tcp_rexmit_initial - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 2345 tp->t_rttmin = tcp_rexmit_min; 2346 tp->t_rxtcur = tcp_rexmit_initial; 2347 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2348 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 2349 tp->t_rcvtime = ticks; 2350 /* We always start with ticks granularity */ 2351 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 2352 /* 2353 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 2354 * because the socket may be bound to an IPv6 wildcard address, 2355 * which may match an IPv4-mapped IPv6 address. 2356 */ 2357 inp->inp_ip_ttl = V_ip_defttl; 2358 #ifdef TCP_BLACKBOX 2359 /* Initialize the per-TCPCB log data. */ 2360 tcp_log_tcpcbinit(tp); 2361 #endif 2362 tp->t_pacing_rate = -1; 2363 if (tp->t_fb->tfb_tcp_fb_init) { 2364 if ((*tp->t_fb->tfb_tcp_fb_init)(tp, &tp->t_fb_ptr)) { 2365 if (CC_ALGO(tp)->cb_destroy != NULL) 2366 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2367 CC_DATA(tp) = NULL; 2368 cc_detach(tp); 2369 #ifdef TCP_HHOOK 2370 khelp_destroy_osd(&tp->t_osd); 2371 #endif 2372 refcount_release(&tp->t_fb->tfb_refcnt); 2373 return (NULL); 2374 } 2375 } 2376 #ifdef STATS 2377 if (V_tcp_perconn_stats_enable == 1) 2378 tp->t_stats = stats_blob_alloc(V_tcp_perconn_stats_dflt_tpl, 0); 2379 #endif 2380 if (V_tcp_do_lrd) 2381 tp->t_flags |= TF_LRD; 2382 2383 return (tp); 2384 } 2385 2386 /* 2387 * Drop a TCP connection, reporting 2388 * the specified error. If connection is synchronized, 2389 * then send a RST to peer. 2390 */ 2391 struct tcpcb * 2392 tcp_drop(struct tcpcb *tp, int errno) 2393 { 2394 struct socket *so = tptosocket(tp); 2395 2396 NET_EPOCH_ASSERT(); 2397 INP_WLOCK_ASSERT(tptoinpcb(tp)); 2398 2399 if (TCPS_HAVERCVDSYN(tp->t_state)) { 2400 tcp_state_change(tp, TCPS_CLOSED); 2401 /* Don't use tcp_output() here due to possible recursion. */ 2402 (void)tcp_output_nodrop(tp); 2403 TCPSTAT_INC(tcps_drops); 2404 } else 2405 TCPSTAT_INC(tcps_conndrops); 2406 if (errno == ETIMEDOUT && tp->t_softerror) 2407 errno = tp->t_softerror; 2408 so->so_error = errno; 2409 return (tcp_close(tp)); 2410 } 2411 2412 void 2413 tcp_discardcb(struct tcpcb *tp) 2414 { 2415 struct inpcb *inp = tptoinpcb(tp); 2416 struct socket *so = tptosocket(tp); 2417 struct mbuf *m; 2418 #ifdef INET6 2419 bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 2420 #endif 2421 2422 INP_WLOCK_ASSERT(inp); 2423 MPASS(!callout_active(&tp->t_callout)); 2424 MPASS(TAILQ_EMPTY(&tp->snd_holes)); 2425 2426 /* free the reassembly queue, if any */ 2427 tcp_reass_flush(tp); 2428 2429 #ifdef TCP_OFFLOAD 2430 /* Disconnect offload device, if any. */ 2431 if (tp->t_flags & TF_TOE) 2432 tcp_offload_detach(tp); 2433 #endif 2434 2435 /* Allow the CC algorithm to clean up after itself. */ 2436 if (CC_ALGO(tp)->cb_destroy != NULL) 2437 CC_ALGO(tp)->cb_destroy(&tp->t_ccv); 2438 CC_DATA(tp) = NULL; 2439 /* Detach from the CC algorithm */ 2440 cc_detach(tp); 2441 2442 #ifdef TCP_HHOOK 2443 khelp_destroy_osd(&tp->t_osd); 2444 #endif 2445 #ifdef STATS 2446 stats_blob_destroy(tp->t_stats); 2447 #endif 2448 2449 CC_ALGO(tp) = NULL; 2450 if ((m = STAILQ_FIRST(&tp->t_inqueue)) != NULL) { 2451 struct mbuf *prev; 2452 2453 STAILQ_INIT(&tp->t_inqueue); 2454 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, prev) 2455 m_freem(m); 2456 } 2457 TCPSTATES_DEC(tp->t_state); 2458 2459 if (tp->t_fb->tfb_tcp_fb_fini) 2460 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 1); 2461 MPASS(!tcp_in_hpts(tp)); 2462 #ifdef TCP_BLACKBOX 2463 tcp_log_tcpcbfini(tp); 2464 #endif 2465 2466 /* 2467 * If we got enough samples through the srtt filter, 2468 * save the rtt and rttvar in the routing entry. 2469 * 'Enough' is arbitrarily defined as 4 rtt samples. 2470 * 4 samples is enough for the srtt filter to converge 2471 * to within enough % of the correct value; fewer samples 2472 * and we could save a bogus rtt. The danger is not high 2473 * as tcp quickly recovers from everything. 2474 * XXX: Works very well but needs some more statistics! 2475 * 2476 * XXXRRS: Updating must be after the stack fini() since 2477 * that may be converting some internal representation of 2478 * say srtt etc into the general one used by other stacks. 2479 */ 2480 if (tp->t_rttupdated >= 4) { 2481 struct hc_metrics_lite metrics; 2482 uint32_t ssthresh; 2483 2484 bzero(&metrics, sizeof(metrics)); 2485 /* 2486 * Update the ssthresh always when the conditions below 2487 * are satisfied. This gives us better new start value 2488 * for the congestion avoidance for new connections. 2489 * ssthresh is only set if packet loss occurred on a session. 2490 */ 2491 ssthresh = tp->snd_ssthresh; 2492 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 2493 /* 2494 * convert the limit from user data bytes to 2495 * packets then to packet data bytes. 2496 */ 2497 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 2498 if (ssthresh < 2) 2499 ssthresh = 2; 2500 ssthresh *= (tp->t_maxseg + 2501 #ifdef INET6 2502 (isipv6 ? sizeof (struct ip6_hdr) + 2503 sizeof (struct tcphdr) : 2504 #endif 2505 sizeof (struct tcpiphdr) 2506 #ifdef INET6 2507 ) 2508 #endif 2509 ); 2510 } else 2511 ssthresh = 0; 2512 metrics.hc_ssthresh = ssthresh; 2513 2514 metrics.hc_rtt = tp->t_srtt; 2515 metrics.hc_rttvar = tp->t_rttvar; 2516 metrics.hc_cwnd = tp->snd_cwnd; 2517 metrics.hc_sendpipe = 0; 2518 metrics.hc_recvpipe = 0; 2519 2520 tcp_hc_update(&inp->inp_inc, &metrics); 2521 } 2522 2523 refcount_release(&tp->t_fb->tfb_refcnt); 2524 } 2525 2526 /* 2527 * Attempt to close a TCP control block, marking it as dropped, and freeing 2528 * the socket if we hold the only reference. 2529 */ 2530 struct tcpcb * 2531 tcp_close(struct tcpcb *tp) 2532 { 2533 struct inpcb *inp = tptoinpcb(tp); 2534 struct socket *so = tptosocket(tp); 2535 2536 INP_WLOCK_ASSERT(inp); 2537 2538 #ifdef TCP_OFFLOAD 2539 if (tp->t_state == TCPS_LISTEN) 2540 tcp_offload_listen_stop(tp); 2541 #endif 2542 /* 2543 * This releases the TFO pending counter resource for TFO listen 2544 * sockets as well as passively-created TFO sockets that transition 2545 * from SYN_RECEIVED to CLOSED. 2546 */ 2547 if (tp->t_tfo_pending) { 2548 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2549 tp->t_tfo_pending = NULL; 2550 } 2551 tcp_timer_stop(tp); 2552 if (tp->t_fb->tfb_tcp_timer_stop_all != NULL) 2553 tp->t_fb->tfb_tcp_timer_stop_all(tp); 2554 #if defined(INET) && defined(INET6) 2555 if ((inp->inp_vflag & INP_IPV6) != 0) 2556 in6_pcbdisconnect(inp); 2557 else 2558 in_pcbdisconnect(inp); 2559 #elif defined(INET6) 2560 in6_pcbdisconnect(inp); 2561 #else 2562 in_pcbdisconnect(inp); 2563 #endif 2564 TCPSTAT_INC(tcps_closed); 2565 if (tp->t_state != TCPS_CLOSED) 2566 tcp_state_change(tp, TCPS_CLOSED); 2567 tp->t_flags |= TF_DISCONNECTED; 2568 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 2569 tcp_free_sackholes(tp); 2570 soisdisconnected(so); 2571 if (inp->inp_flags & INP_SOCKREF) { 2572 inp->inp_flags &= ~INP_SOCKREF; 2573 INP_WUNLOCK(inp); 2574 sorele(so); 2575 return (NULL); 2576 } 2577 return (tp); 2578 } 2579 2580 /* 2581 * Notify a tcp user of an asynchronous error; 2582 * store error as soft error, but wake up user 2583 * (for now, won't do anything until can select for soft error). 2584 * 2585 * Do not wake up user since there currently is no mechanism for 2586 * reporting soft errors (yet - a kqueue filter may be added). 2587 */ 2588 static struct inpcb * 2589 tcp_notify(struct inpcb *inp, int error) 2590 { 2591 struct tcpcb *tp; 2592 2593 INP_WLOCK_ASSERT(inp); 2594 2595 tp = intotcpcb(inp); 2596 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 2597 2598 /* 2599 * Ignore some errors if we are hooked up. 2600 * If connection hasn't completed, has retransmitted several times, 2601 * and receives a second error, give up now. This is better 2602 * than waiting a long time to establish a connection that 2603 * can never complete. 2604 */ 2605 if (tp->t_state == TCPS_ESTABLISHED && 2606 (error == EHOSTUNREACH || error == ENETUNREACH || 2607 error == EHOSTDOWN)) { 2608 if (inp->inp_route.ro_nh) { 2609 NH_FREE(inp->inp_route.ro_nh); 2610 inp->inp_route.ro_nh = (struct nhop_object *)NULL; 2611 } 2612 return (inp); 2613 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 2614 tp->t_softerror) { 2615 tp = tcp_drop(tp, error); 2616 if (tp != NULL) 2617 return (inp); 2618 else 2619 return (NULL); 2620 } else { 2621 tp->t_softerror = error; 2622 return (inp); 2623 } 2624 #if 0 2625 wakeup( &so->so_timeo); 2626 sorwakeup(so); 2627 sowwakeup(so); 2628 #endif 2629 } 2630 2631 static int 2632 tcp_pcblist(SYSCTL_HANDLER_ARGS) 2633 { 2634 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2635 INPLOOKUP_RLOCKPCB); 2636 struct xinpgen xig; 2637 struct inpcb *inp; 2638 int error; 2639 2640 if (req->newptr != NULL) 2641 return (EPERM); 2642 2643 if (req->oldptr == NULL) { 2644 int n; 2645 2646 n = V_tcbinfo.ipi_count + 2647 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2648 n += imax(n / 8, 10); 2649 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 2650 return (0); 2651 } 2652 2653 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2654 return (error); 2655 2656 bzero(&xig, sizeof(xig)); 2657 xig.xig_len = sizeof xig; 2658 xig.xig_count = V_tcbinfo.ipi_count + 2659 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2660 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2661 xig.xig_sogen = so_gencnt; 2662 error = SYSCTL_OUT(req, &xig, sizeof xig); 2663 if (error) 2664 return (error); 2665 2666 error = syncache_pcblist(req); 2667 if (error) 2668 return (error); 2669 2670 while ((inp = inp_next(&inpi)) != NULL) { 2671 if (inp->inp_gencnt <= xig.xig_gen && 2672 cr_canseeinpcb(req->td->td_ucred, inp) == 0) { 2673 struct xtcpcb xt; 2674 2675 tcp_inptoxtp(inp, &xt); 2676 error = SYSCTL_OUT(req, &xt, sizeof xt); 2677 if (error) { 2678 INP_RUNLOCK(inp); 2679 break; 2680 } else 2681 continue; 2682 } 2683 } 2684 2685 if (!error) { 2686 /* 2687 * Give the user an updated idea of our state. 2688 * If the generation differs from what we told 2689 * her before, she knows that something happened 2690 * while we were processing this request, and it 2691 * might be necessary to retry. 2692 */ 2693 xig.xig_gen = V_tcbinfo.ipi_gencnt; 2694 xig.xig_sogen = so_gencnt; 2695 xig.xig_count = V_tcbinfo.ipi_count + 2696 counter_u64_fetch(V_tcps_states[TCPS_SYN_RECEIVED]); 2697 error = SYSCTL_OUT(req, &xig, sizeof xig); 2698 } 2699 2700 return (error); 2701 } 2702 2703 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 2704 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_NEEDGIANT, 2705 NULL, 0, tcp_pcblist, "S,xtcpcb", 2706 "List of active TCP connections"); 2707 2708 #define SND_TAG_STATUS_MAXLEN 128 2709 2710 #ifdef KERN_TLS 2711 2712 static struct sx ktlslist_lock; 2713 SX_SYSINIT(ktlslistlock, &ktlslist_lock, "ktlslist"); 2714 static uint64_t ktls_glob_gen = 1; 2715 2716 static int 2717 tcp_ktlslist_locked(SYSCTL_HANDLER_ARGS, bool export_keys) 2718 { 2719 struct xinpgen xig; 2720 struct inpcb *inp; 2721 struct socket *so; 2722 struct ktls_session *ksr, *kss; 2723 char *buf; 2724 struct xktls_session *xktls; 2725 uint64_t ipi_gencnt; 2726 size_t buflen, len, sz; 2727 u_int cnt; 2728 int error; 2729 bool ek, p; 2730 2731 sx_assert(&ktlslist_lock, SA_XLOCKED); 2732 if (req->newptr != NULL) 2733 return (EPERM); 2734 2735 len = 0; 2736 cnt = 0; 2737 ipi_gencnt = V_tcbinfo.ipi_gencnt; 2738 bzero(&xig, sizeof(xig)); 2739 xig.xig_len = sizeof(xig); 2740 xig.xig_gen = ktls_glob_gen++; 2741 xig.xig_sogen = so_gencnt; 2742 2743 struct inpcb_iterator inpi = INP_ALL_ITERATOR(&V_tcbinfo, 2744 INPLOOKUP_RLOCKPCB); 2745 while ((inp = inp_next(&inpi)) != NULL) { 2746 if (inp->inp_gencnt > ipi_gencnt || 2747 cr_canseeinpcb(req->td->td_ucred, inp) != 0) 2748 continue; 2749 2750 so = inp->inp_socket; 2751 if (so != NULL && so->so_gencnt <= xig.xig_sogen) { 2752 p = false; 2753 ek = export_keys && cr_canexport_ktlskeys( 2754 req->td, inp); 2755 ksr = so->so_rcv.sb_tls_info; 2756 if (ksr != NULL) { 2757 ksr->gen = xig.xig_gen; 2758 p = true; 2759 if (ek) { 2760 sz = SIZE_T_MAX; 2761 ktls_session_copy_keys(ksr, 2762 NULL, &sz); 2763 len += sz; 2764 } 2765 if (ksr->snd_tag != NULL && 2766 ksr->snd_tag->sw->snd_tag_status_str != 2767 NULL) { 2768 sz = SND_TAG_STATUS_MAXLEN; 2769 in_pcbref(inp); 2770 INP_RUNLOCK(inp); 2771 error = ksr->snd_tag->sw-> 2772 snd_tag_status_str( 2773 ksr->snd_tag, NULL, &sz); 2774 if (in_pcbrele_rlock(inp)) 2775 return (EDEADLK); 2776 if (error == 0) 2777 len += sz; 2778 } 2779 } 2780 kss = so->so_snd.sb_tls_info; 2781 if (kss != NULL) { 2782 kss->gen = xig.xig_gen; 2783 p = true; 2784 if (ek) { 2785 sz = SIZE_T_MAX; 2786 ktls_session_copy_keys(kss, 2787 NULL, &sz); 2788 len += sz; 2789 } 2790 if (kss->snd_tag != NULL && 2791 kss->snd_tag->sw->snd_tag_status_str != 2792 NULL) { 2793 sz = SND_TAG_STATUS_MAXLEN; 2794 in_pcbref(inp); 2795 INP_RUNLOCK(inp); 2796 error = kss->snd_tag->sw-> 2797 snd_tag_status_str( 2798 kss->snd_tag, NULL, &sz); 2799 if (in_pcbrele_rlock(inp)) 2800 return (EDEADLK); 2801 if (error == 0) 2802 len += sz; 2803 } 2804 } 2805 if (p) { 2806 len += sizeof(*xktls); 2807 len = roundup2(len, __alignof(struct 2808 xktls_session)); 2809 } 2810 } 2811 } 2812 if (req->oldptr == NULL) { 2813 len += 2 * sizeof(xig); 2814 len += 3 * len / 4; 2815 req->oldidx = len; 2816 return (0); 2817 } 2818 2819 if ((error = sysctl_wire_old_buffer(req, 0)) != 0) 2820 return (error); 2821 2822 error = SYSCTL_OUT(req, &xig, sizeof xig); 2823 if (error != 0) 2824 return (error); 2825 2826 buflen = roundup2(sizeof(*xktls) + 2 * TLS_MAX_PARAM_SIZE + 2827 2 * SND_TAG_STATUS_MAXLEN, __alignof(struct xktls_session)); 2828 buf = malloc(buflen, M_TEMP, M_WAITOK | M_ZERO); 2829 struct inpcb_iterator inpi1 = INP_ALL_ITERATOR(&V_tcbinfo, 2830 INPLOOKUP_RLOCKPCB); 2831 while ((inp = inp_next(&inpi1)) != NULL) { 2832 if (inp->inp_gencnt > ipi_gencnt || 2833 cr_canseeinpcb(req->td->td_ucred, inp) != 0) 2834 continue; 2835 2836 so = inp->inp_socket; 2837 if (so == NULL) 2838 continue; 2839 2840 p = false; 2841 ek = export_keys && cr_canexport_ktlskeys(req->td, inp); 2842 ksr = so->so_rcv.sb_tls_info; 2843 kss = so->so_snd.sb_tls_info; 2844 xktls = (struct xktls_session *)buf; 2845 if (ksr != NULL && ksr->gen == xig.xig_gen) { 2846 p = true; 2847 ktls_session_to_xktls_onedir(ksr, ek, &xktls->rcv); 2848 } 2849 if (kss != NULL && kss->gen == xig.xig_gen) { 2850 p = true; 2851 ktls_session_to_xktls_onedir(kss, ek, &xktls->snd); 2852 } 2853 if (!p) 2854 continue; 2855 2856 xktls->inp_gencnt = inp->inp_gencnt; 2857 xktls->so_pcb = (kvaddr_t)inp; 2858 memcpy(&xktls->coninf, &inp->inp_inc, sizeof(xktls->coninf)); 2859 len = sizeof(*xktls); 2860 if (ksr != NULL && ksr->gen == xig.xig_gen) { 2861 if (ek) { 2862 sz = buflen - len; 2863 ktls_session_copy_keys(ksr, buf + len, &sz); 2864 len += sz; 2865 } else { 2866 xktls->rcv.cipher_key_len = 0; 2867 xktls->rcv.auth_key_len = 0; 2868 } 2869 if (ksr->snd_tag != NULL && 2870 ksr->snd_tag->sw->snd_tag_status_str != NULL) { 2871 sz = SND_TAG_STATUS_MAXLEN; 2872 in_pcbref(inp); 2873 INP_RUNLOCK(inp); 2874 error = ksr->snd_tag->sw->snd_tag_status_str( 2875 ksr->snd_tag, buf + len, &sz); 2876 if (in_pcbrele_rlock(inp)) 2877 return (EDEADLK); 2878 if (error == 0) { 2879 xktls->rcv.drv_st_len = sz; 2880 len += sz; 2881 } 2882 } 2883 } 2884 if (kss != NULL && kss->gen == xig.xig_gen) { 2885 if (ek) { 2886 sz = buflen - len; 2887 ktls_session_copy_keys(kss, buf + len, &sz); 2888 len += sz; 2889 } else { 2890 xktls->snd.cipher_key_len = 0; 2891 xktls->snd.auth_key_len = 0; 2892 } 2893 if (kss->snd_tag != NULL && 2894 kss->snd_tag->sw->snd_tag_status_str != NULL) { 2895 sz = SND_TAG_STATUS_MAXLEN; 2896 in_pcbref(inp); 2897 INP_RUNLOCK(inp); 2898 error = kss->snd_tag->sw->snd_tag_status_str( 2899 kss->snd_tag, buf + len, &sz); 2900 if (in_pcbrele_rlock(inp)) 2901 return (EDEADLK); 2902 if (error == 0) { 2903 xktls->snd.drv_st_len = sz; 2904 len += sz; 2905 } 2906 } 2907 } 2908 len = roundup2(len, __alignof(*xktls)); 2909 xktls->tsz = len; 2910 xktls->fsz = sizeof(*xktls); 2911 2912 error = SYSCTL_OUT(req, xktls, len); 2913 if (error != 0) { 2914 INP_RUNLOCK(inp); 2915 break; 2916 } 2917 cnt++; 2918 } 2919 2920 if (error == 0) { 2921 xig.xig_sogen = so_gencnt; 2922 xig.xig_count = cnt; 2923 error = SYSCTL_OUT(req, &xig, sizeof(xig)); 2924 } 2925 2926 zfree(buf, M_TEMP); 2927 return (error); 2928 } 2929 2930 static int 2931 tcp_ktlslist1(SYSCTL_HANDLER_ARGS, bool export_keys) 2932 { 2933 int repeats, error; 2934 2935 for (repeats = 0; repeats < 100; repeats++) { 2936 if (sx_xlock_sig(&ktlslist_lock)) 2937 return (EINTR); 2938 error = tcp_ktlslist_locked(oidp, arg1, arg2, req, 2939 export_keys); 2940 sx_xunlock(&ktlslist_lock); 2941 if (error != EDEADLK) 2942 break; 2943 if (sig_intr() != 0) { 2944 error = EINTR; 2945 break; 2946 } 2947 req->oldidx = 0; 2948 } 2949 return (error); 2950 } 2951 2952 static int 2953 tcp_ktlslist_nokeys(SYSCTL_HANDLER_ARGS) 2954 { 2955 return (tcp_ktlslist1(oidp, arg1, arg2, req, false)); 2956 } 2957 2958 static int 2959 tcp_ktlslist_wkeys(SYSCTL_HANDLER_ARGS) 2960 { 2961 return (tcp_ktlslist1(oidp, arg1, arg2, req, true)); 2962 } 2963 2964 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST, ktlslist, 2965 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2966 NULL, 0, tcp_ktlslist_nokeys, "S,xktls_session", 2967 "List of active kTLS sessions for TCP connections"); 2968 SYSCTL_PROC(_net_inet_tcp, TCPCTL_KTLSLIST_WKEYS, ktlslist_wkeys, 2969 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 2970 NULL, 0, tcp_ktlslist_wkeys, "S,xktls_session", 2971 "List of active kTLS sessions for TCP connections with keys"); 2972 #endif /* KERN_TLS */ 2973 2974 #ifdef INET 2975 static int 2976 tcp_getcred(SYSCTL_HANDLER_ARGS) 2977 { 2978 struct xucred xuc; 2979 struct sockaddr_in addrs[2]; 2980 struct epoch_tracker et; 2981 struct inpcb *inp; 2982 int error; 2983 2984 if (req->newptr == NULL) 2985 return (EINVAL); 2986 error = priv_check(req->td, PRIV_NETINET_GETCRED); 2987 if (error) 2988 return (error); 2989 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 2990 if (error) 2991 return (error); 2992 NET_EPOCH_ENTER(et); 2993 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 2994 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 2995 NET_EPOCH_EXIT(et); 2996 if (inp != NULL) { 2997 if (error == 0) 2998 error = cr_canseeinpcb(req->td->td_ucred, inp); 2999 if (error == 0) 3000 cru2x(inp->inp_cred, &xuc); 3001 INP_RUNLOCK(inp); 3002 } else 3003 error = ENOENT; 3004 if (error == 0) 3005 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 3006 return (error); 3007 } 3008 3009 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 3010 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 3011 0, 0, tcp_getcred, "S,xucred", 3012 "Get the xucred of a TCP connection"); 3013 #endif /* INET */ 3014 3015 #ifdef INET6 3016 static int 3017 tcp6_getcred(SYSCTL_HANDLER_ARGS) 3018 { 3019 struct epoch_tracker et; 3020 struct xucred xuc; 3021 struct sockaddr_in6 addrs[2]; 3022 struct inpcb *inp; 3023 int error; 3024 #ifdef INET 3025 int mapped = 0; 3026 #endif 3027 3028 if (req->newptr == NULL) 3029 return (EINVAL); 3030 error = priv_check(req->td, PRIV_NETINET_GETCRED); 3031 if (error) 3032 return (error); 3033 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 3034 if (error) 3035 return (error); 3036 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 3037 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 3038 return (error); 3039 } 3040 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 3041 #ifdef INET 3042 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 3043 mapped = 1; 3044 else 3045 #endif 3046 return (EINVAL); 3047 } 3048 3049 NET_EPOCH_ENTER(et); 3050 #ifdef INET 3051 if (mapped == 1) 3052 inp = in_pcblookup(&V_tcbinfo, 3053 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 3054 addrs[1].sin6_port, 3055 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 3056 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 3057 else 3058 #endif 3059 inp = in6_pcblookup(&V_tcbinfo, 3060 &addrs[1].sin6_addr, addrs[1].sin6_port, 3061 &addrs[0].sin6_addr, addrs[0].sin6_port, 3062 INPLOOKUP_RLOCKPCB, NULL); 3063 NET_EPOCH_EXIT(et); 3064 if (inp != NULL) { 3065 if (error == 0) 3066 error = cr_canseeinpcb(req->td->td_ucred, inp); 3067 if (error == 0) 3068 cru2x(inp->inp_cred, &xuc); 3069 INP_RUNLOCK(inp); 3070 } else 3071 error = ENOENT; 3072 if (error == 0) 3073 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 3074 return (error); 3075 } 3076 3077 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 3078 CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_PRISON | CTLFLAG_NEEDGIANT, 3079 0, 0, tcp6_getcred, "S,xucred", 3080 "Get the xucred of a TCP6 connection"); 3081 #endif /* INET6 */ 3082 3083 #ifdef INET 3084 /* Path MTU to try next when a fragmentation-needed message is received. */ 3085 static inline int 3086 tcp_next_pmtu(const struct icmp *icp, const struct ip *ip) 3087 { 3088 int mtu = ntohs(icp->icmp_nextmtu); 3089 3090 /* If no alternative MTU was proposed, try the next smaller one. */ 3091 if (!mtu) 3092 mtu = ip_next_mtu(ntohs(ip->ip_len), 1); 3093 if (mtu < V_tcp_minmss + sizeof(struct tcpiphdr)) 3094 mtu = V_tcp_minmss + sizeof(struct tcpiphdr); 3095 3096 return (mtu); 3097 } 3098 3099 static void 3100 tcp_ctlinput_with_port(struct icmp *icp, uint16_t port) 3101 { 3102 struct ip *ip; 3103 struct tcphdr *th; 3104 struct inpcb *inp; 3105 struct tcpcb *tp; 3106 struct inpcb *(*notify)(struct inpcb *, int); 3107 struct in_conninfo inc; 3108 tcp_seq icmp_tcp_seq; 3109 int errno, mtu; 3110 3111 errno = icmp_errmap(icp); 3112 switch (errno) { 3113 case 0: 3114 return; 3115 case EMSGSIZE: 3116 notify = tcp_mtudisc_notify; 3117 break; 3118 case ECONNREFUSED: 3119 if (V_icmp_may_rst) 3120 notify = tcp_drop_syn_sent; 3121 else 3122 notify = tcp_notify; 3123 break; 3124 case EHOSTUNREACH: 3125 if (V_icmp_may_rst && icp->icmp_type == ICMP_TIMXCEED) 3126 notify = tcp_drop_syn_sent; 3127 else 3128 notify = tcp_notify; 3129 break; 3130 default: 3131 notify = tcp_notify; 3132 } 3133 3134 ip = &icp->icmp_ip; 3135 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2)); 3136 icmp_tcp_seq = th->th_seq; 3137 inp = in_pcblookup(&V_tcbinfo, ip->ip_dst, th->th_dport, ip->ip_src, 3138 th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 3139 if (inp != NULL) { 3140 tp = intotcpcb(inp); 3141 #ifdef TCP_OFFLOAD 3142 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3143 /* 3144 * MTU discovery for offloaded connections. Let 3145 * the TOE driver verify seq# and process it. 3146 */ 3147 mtu = tcp_next_pmtu(icp, ip); 3148 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3149 goto out; 3150 } 3151 #endif 3152 if (tp->t_port != port) 3153 goto out; 3154 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3155 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3156 if (errno == EMSGSIZE) { 3157 /* 3158 * MTU discovery: we got a needfrag and 3159 * will potentially try a lower MTU. 3160 */ 3161 mtu = tcp_next_pmtu(icp, ip); 3162 3163 /* 3164 * Only process the offered MTU if it 3165 * is smaller than the current one. 3166 */ 3167 if (mtu < tp->t_maxseg + 3168 sizeof(struct tcpiphdr)) { 3169 bzero(&inc, sizeof(inc)); 3170 inc.inc_faddr = ip->ip_dst; 3171 inc.inc_fibnum = 3172 inp->inp_inc.inc_fibnum; 3173 tcp_hc_updatemtu(&inc, mtu); 3174 inp = tcp_mtudisc(inp, mtu); 3175 } 3176 } else 3177 inp = (*notify)(inp, errno); 3178 } 3179 } else { 3180 bzero(&inc, sizeof(inc)); 3181 inc.inc_fport = th->th_dport; 3182 inc.inc_lport = th->th_sport; 3183 inc.inc_faddr = ip->ip_dst; 3184 inc.inc_laddr = ip->ip_src; 3185 syncache_unreach(&inc, icmp_tcp_seq, port); 3186 } 3187 out: 3188 if (inp != NULL) 3189 INP_WUNLOCK(inp); 3190 } 3191 3192 static void 3193 tcp_ctlinput(struct icmp *icmp) 3194 { 3195 tcp_ctlinput_with_port(icmp, htons(0)); 3196 } 3197 3198 static void 3199 tcp_ctlinput_viaudp(udp_tun_icmp_param_t param) 3200 { 3201 /* Its a tunneled TCP over UDP icmp */ 3202 struct icmp *icmp = param.icmp; 3203 struct ip *outer_ip, *inner_ip; 3204 struct udphdr *udp; 3205 struct tcphdr *th, ttemp; 3206 int i_hlen, o_len; 3207 uint16_t port; 3208 3209 outer_ip = (struct ip *)((caddr_t)icmp - sizeof(struct ip)); 3210 inner_ip = &icmp->icmp_ip; 3211 i_hlen = inner_ip->ip_hl << 2; 3212 o_len = ntohs(outer_ip->ip_len); 3213 if (o_len < 3214 (sizeof(struct ip) + 8 + i_hlen + sizeof(struct udphdr) + offsetof(struct tcphdr, th_ack))) { 3215 /* Not enough data present */ 3216 return; 3217 } 3218 /* Ok lets strip out the inner udphdr header by copying up on top of it the tcp hdr */ 3219 udp = (struct udphdr *)(((caddr_t)inner_ip) + i_hlen); 3220 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3221 return; 3222 } 3223 port = udp->uh_dport; 3224 th = (struct tcphdr *)(udp + 1); 3225 memcpy(&ttemp, th, sizeof(struct tcphdr)); 3226 memcpy(udp, &ttemp, sizeof(struct tcphdr)); 3227 /* Now adjust down the size of the outer IP header */ 3228 o_len -= sizeof(struct udphdr); 3229 outer_ip->ip_len = htons(o_len); 3230 /* Now call in to the normal handling code */ 3231 tcp_ctlinput_with_port(icmp, port); 3232 } 3233 #endif /* INET */ 3234 3235 #ifdef INET6 3236 static inline int 3237 tcp6_next_pmtu(const struct icmp6_hdr *icmp6) 3238 { 3239 int mtu = ntohl(icmp6->icmp6_mtu); 3240 3241 /* 3242 * If no alternative MTU was proposed, or the proposed MTU was too 3243 * small, set to the min. 3244 */ 3245 if (mtu < IPV6_MMTU) 3246 mtu = IPV6_MMTU; 3247 return (mtu); 3248 } 3249 3250 static void 3251 tcp6_ctlinput_with_port(struct ip6ctlparam *ip6cp, uint16_t port) 3252 { 3253 struct in6_addr *dst; 3254 struct inpcb *(*notify)(struct inpcb *, int); 3255 struct ip6_hdr *ip6; 3256 struct mbuf *m; 3257 struct inpcb *inp; 3258 struct tcpcb *tp; 3259 struct icmp6_hdr *icmp6; 3260 struct in_conninfo inc; 3261 struct tcp_ports { 3262 uint16_t th_sport; 3263 uint16_t th_dport; 3264 } t_ports; 3265 tcp_seq icmp_tcp_seq; 3266 unsigned int mtu; 3267 unsigned int off; 3268 int errno; 3269 3270 icmp6 = ip6cp->ip6c_icmp6; 3271 m = ip6cp->ip6c_m; 3272 ip6 = ip6cp->ip6c_ip6; 3273 off = ip6cp->ip6c_off; 3274 dst = &ip6cp->ip6c_finaldst->sin6_addr; 3275 3276 errno = icmp6_errmap(icmp6); 3277 switch (errno) { 3278 case 0: 3279 return; 3280 case EMSGSIZE: 3281 notify = tcp_mtudisc_notify; 3282 break; 3283 case ECONNREFUSED: 3284 if (V_icmp_may_rst) 3285 notify = tcp_drop_syn_sent; 3286 else 3287 notify = tcp_notify; 3288 break; 3289 case EHOSTUNREACH: 3290 /* 3291 * There are only four ICMPs that may reset connection: 3292 * - administratively prohibited 3293 * - port unreachable 3294 * - time exceeded in transit 3295 * - unknown next header 3296 */ 3297 if (V_icmp_may_rst && 3298 ((icmp6->icmp6_type == ICMP6_DST_UNREACH && 3299 (icmp6->icmp6_code == ICMP6_DST_UNREACH_ADMIN || 3300 icmp6->icmp6_code == ICMP6_DST_UNREACH_NOPORT)) || 3301 (icmp6->icmp6_type == ICMP6_TIME_EXCEEDED && 3302 icmp6->icmp6_code == ICMP6_TIME_EXCEED_TRANSIT) || 3303 (icmp6->icmp6_type == ICMP6_PARAM_PROB && 3304 icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER))) 3305 notify = tcp_drop_syn_sent; 3306 else 3307 notify = tcp_notify; 3308 break; 3309 default: 3310 notify = tcp_notify; 3311 } 3312 3313 /* Check if we can safely get the ports from the tcp hdr */ 3314 if (m == NULL || 3315 (m->m_pkthdr.len < 3316 (int32_t) (off + sizeof(struct tcp_ports)))) { 3317 return; 3318 } 3319 bzero(&t_ports, sizeof(struct tcp_ports)); 3320 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports); 3321 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_dst, t_ports.th_dport, 3322 &ip6->ip6_src, t_ports.th_sport, INPLOOKUP_WLOCKPCB, NULL); 3323 off += sizeof(struct tcp_ports); 3324 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) { 3325 goto out; 3326 } 3327 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq); 3328 if (inp != NULL) { 3329 tp = intotcpcb(inp); 3330 #ifdef TCP_OFFLOAD 3331 if (tp->t_flags & TF_TOE && errno == EMSGSIZE) { 3332 /* MTU discovery for offloaded connections. */ 3333 mtu = tcp6_next_pmtu(icmp6); 3334 tcp_offload_pmtu_update(tp, icmp_tcp_seq, mtu); 3335 goto out; 3336 } 3337 #endif 3338 if (tp->t_port != port) 3339 goto out; 3340 if (SEQ_GEQ(ntohl(icmp_tcp_seq), tp->snd_una) && 3341 SEQ_LT(ntohl(icmp_tcp_seq), tp->snd_max)) { 3342 if (errno == EMSGSIZE) { 3343 /* 3344 * MTU discovery: 3345 * If we got a needfrag set the MTU 3346 * in the route to the suggested new 3347 * value (if given) and then notify. 3348 */ 3349 mtu = tcp6_next_pmtu(icmp6); 3350 3351 bzero(&inc, sizeof(inc)); 3352 inc.inc_fibnum = M_GETFIB(m); 3353 inc.inc_flags |= INC_ISIPV6; 3354 inc.inc6_faddr = *dst; 3355 if (in6_setscope(&inc.inc6_faddr, 3356 m->m_pkthdr.rcvif, NULL)) 3357 goto out; 3358 /* 3359 * Only process the offered MTU if it 3360 * is smaller than the current one. 3361 */ 3362 if (mtu < tp->t_maxseg + 3363 sizeof (struct tcphdr) + 3364 sizeof (struct ip6_hdr)) { 3365 tcp_hc_updatemtu(&inc, mtu); 3366 tcp_mtudisc(inp, mtu); 3367 ICMP6STAT_INC(icp6s_pmtuchg); 3368 } 3369 } else 3370 inp = (*notify)(inp, errno); 3371 } 3372 } else { 3373 bzero(&inc, sizeof(inc)); 3374 inc.inc_fibnum = M_GETFIB(m); 3375 inc.inc_flags |= INC_ISIPV6; 3376 inc.inc_fport = t_ports.th_dport; 3377 inc.inc_lport = t_ports.th_sport; 3378 inc.inc6_faddr = *dst; 3379 inc.inc6_laddr = ip6->ip6_src; 3380 syncache_unreach(&inc, icmp_tcp_seq, port); 3381 } 3382 out: 3383 if (inp != NULL) 3384 INP_WUNLOCK(inp); 3385 } 3386 3387 static void 3388 tcp6_ctlinput(struct ip6ctlparam *ctl) 3389 { 3390 tcp6_ctlinput_with_port(ctl, htons(0)); 3391 } 3392 3393 static void 3394 tcp6_ctlinput_viaudp(udp_tun_icmp_param_t param) 3395 { 3396 struct ip6ctlparam *ip6cp = param.ip6cp; 3397 struct mbuf *m; 3398 struct udphdr *udp; 3399 uint16_t port; 3400 3401 m = m_pulldown(ip6cp->ip6c_m, ip6cp->ip6c_off, sizeof(struct udphdr), NULL); 3402 if (m == NULL) { 3403 return; 3404 } 3405 udp = mtod(m, struct udphdr *); 3406 if (ntohs(udp->uh_sport) != V_tcp_udp_tunneling_port) { 3407 return; 3408 } 3409 port = udp->uh_dport; 3410 m_adj(m, sizeof(struct udphdr)); 3411 if ((m->m_flags & M_PKTHDR) == 0) { 3412 ip6cp->ip6c_m->m_pkthdr.len -= sizeof(struct udphdr); 3413 } 3414 /* Now call in to the normal handling code */ 3415 tcp6_ctlinput_with_port(ip6cp, port); 3416 } 3417 3418 #endif /* INET6 */ 3419 3420 static uint32_t 3421 tcp_keyed_hash(struct in_conninfo *inc, u_char *key, u_int len) 3422 { 3423 SIPHASH_CTX ctx; 3424 uint32_t hash[2]; 3425 3426 KASSERT(len >= SIPHASH_KEY_LENGTH, 3427 ("%s: keylen %u too short ", __func__, len)); 3428 SipHash24_Init(&ctx); 3429 SipHash_SetKey(&ctx, (uint8_t *)key); 3430 SipHash_Update(&ctx, &inc->inc_fport, sizeof(uint16_t)); 3431 SipHash_Update(&ctx, &inc->inc_lport, sizeof(uint16_t)); 3432 switch (inc->inc_flags & INC_ISIPV6) { 3433 #ifdef INET 3434 case 0: 3435 SipHash_Update(&ctx, &inc->inc_faddr, sizeof(struct in_addr)); 3436 SipHash_Update(&ctx, &inc->inc_laddr, sizeof(struct in_addr)); 3437 break; 3438 #endif 3439 #ifdef INET6 3440 case INC_ISIPV6: 3441 SipHash_Update(&ctx, &inc->inc6_faddr, sizeof(struct in6_addr)); 3442 SipHash_Update(&ctx, &inc->inc6_laddr, sizeof(struct in6_addr)); 3443 break; 3444 #endif 3445 } 3446 SipHash_Final((uint8_t *)hash, &ctx); 3447 3448 return (hash[0] ^ hash[1]); 3449 } 3450 3451 uint32_t 3452 tcp_new_ts_offset(struct in_conninfo *inc) 3453 { 3454 struct in_conninfo inc_store, *local_inc; 3455 3456 if (!V_tcp_ts_offset_per_conn) { 3457 memcpy(&inc_store, inc, sizeof(struct in_conninfo)); 3458 inc_store.inc_lport = 0; 3459 inc_store.inc_fport = 0; 3460 local_inc = &inc_store; 3461 } else { 3462 local_inc = inc; 3463 } 3464 return (tcp_keyed_hash(local_inc, V_ts_offset_secret, 3465 sizeof(V_ts_offset_secret))); 3466 } 3467 3468 /* 3469 * Following is where TCP initial sequence number generation occurs. 3470 * 3471 * There are two places where we must use initial sequence numbers: 3472 * 1. In SYN-ACK packets. 3473 * 2. In SYN packets. 3474 * 3475 * All ISNs for SYN-ACK packets are generated by the syncache. See 3476 * tcp_syncache.c for details. 3477 * 3478 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 3479 * depends on this property. In addition, these ISNs should be 3480 * unguessable so as to prevent connection hijacking. To satisfy 3481 * the requirements of this situation, the algorithm outlined in 3482 * RFC 1948 is used, with only small modifications. 3483 * 3484 * Implementation details: 3485 * 3486 * Time is based off the system timer, and is corrected so that it 3487 * increases by one megabyte per second. This allows for proper 3488 * recycling on high speed LANs while still leaving over an hour 3489 * before rollover. 3490 * 3491 * As reading the *exact* system time is too expensive to be done 3492 * whenever setting up a TCP connection, we increment the time 3493 * offset in two ways. First, a small random positive increment 3494 * is added to isn_offset for each connection that is set up. 3495 * Second, the function tcp_isn_tick fires once per clock tick 3496 * and increments isn_offset as necessary so that sequence numbers 3497 * are incremented at approximately ISN_BYTES_PER_SECOND. The 3498 * random positive increments serve only to ensure that the same 3499 * exact sequence number is never sent out twice (as could otherwise 3500 * happen when a port is recycled in less than the system tick 3501 * interval.) 3502 * 3503 * net.inet.tcp.isn_reseed_interval controls the number of seconds 3504 * between seeding of isn_secret. This is normally set to zero, 3505 * as reseeding should not be necessary. 3506 * 3507 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 3508 * isn_offset_old, and isn_ctx is performed using the ISN lock. In 3509 * general, this means holding an exclusive (write) lock. 3510 */ 3511 3512 #define ISN_BYTES_PER_SECOND 1048576 3513 #define ISN_STATIC_INCREMENT 4096 3514 #define ISN_RANDOM_INCREMENT (4096 - 1) 3515 #define ISN_SECRET_LENGTH SIPHASH_KEY_LENGTH 3516 3517 VNET_DEFINE_STATIC(u_char, isn_secret[ISN_SECRET_LENGTH]); 3518 VNET_DEFINE_STATIC(int, isn_last); 3519 VNET_DEFINE_STATIC(int, isn_last_reseed); 3520 VNET_DEFINE_STATIC(u_int32_t, isn_offset); 3521 VNET_DEFINE_STATIC(u_int32_t, isn_offset_old); 3522 3523 #define V_isn_secret VNET(isn_secret) 3524 #define V_isn_last VNET(isn_last) 3525 #define V_isn_last_reseed VNET(isn_last_reseed) 3526 #define V_isn_offset VNET(isn_offset) 3527 #define V_isn_offset_old VNET(isn_offset_old) 3528 3529 tcp_seq 3530 tcp_new_isn(struct in_conninfo *inc) 3531 { 3532 tcp_seq new_isn; 3533 u_int32_t projected_offset; 3534 3535 ISN_LOCK(); 3536 /* Seed if this is the first use, reseed if requested. */ 3537 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 3538 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 3539 < (u_int)ticks))) { 3540 arc4rand(&V_isn_secret, sizeof(V_isn_secret), 0); 3541 V_isn_last_reseed = ticks; 3542 } 3543 3544 /* Compute the hash and return the ISN. */ 3545 new_isn = (tcp_seq)tcp_keyed_hash(inc, V_isn_secret, 3546 sizeof(V_isn_secret)); 3547 V_isn_offset += ISN_STATIC_INCREMENT + 3548 (arc4random() & ISN_RANDOM_INCREMENT); 3549 if (ticks != V_isn_last) { 3550 projected_offset = V_isn_offset_old + 3551 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 3552 if (SEQ_GT(projected_offset, V_isn_offset)) 3553 V_isn_offset = projected_offset; 3554 V_isn_offset_old = V_isn_offset; 3555 V_isn_last = ticks; 3556 } 3557 new_isn += V_isn_offset; 3558 ISN_UNLOCK(); 3559 return (new_isn); 3560 } 3561 3562 /* 3563 * When a specific ICMP unreachable message is received and the 3564 * connection state is SYN-SENT, drop the connection. This behavior 3565 * is controlled by the icmp_may_rst sysctl. 3566 */ 3567 static struct inpcb * 3568 tcp_drop_syn_sent(struct inpcb *inp, int errno) 3569 { 3570 struct tcpcb *tp; 3571 3572 NET_EPOCH_ASSERT(); 3573 INP_WLOCK_ASSERT(inp); 3574 3575 tp = intotcpcb(inp); 3576 if (tp->t_state != TCPS_SYN_SENT) 3577 return (inp); 3578 3579 if (tp->t_flags & TF_FASTOPEN) 3580 tcp_fastopen_disable_path(tp); 3581 3582 tp = tcp_drop(tp, errno); 3583 if (tp != NULL) 3584 return (inp); 3585 else 3586 return (NULL); 3587 } 3588 3589 /* 3590 * When `need fragmentation' ICMP is received, update our idea of the MSS 3591 * based on the new value. Also nudge TCP to send something, since we 3592 * know the packet we just sent was dropped. 3593 * This duplicates some code in the tcp_mss() function in tcp_input.c. 3594 */ 3595 static struct inpcb * 3596 tcp_mtudisc_notify(struct inpcb *inp, int error) 3597 { 3598 3599 return (tcp_mtudisc(inp, -1)); 3600 } 3601 3602 static struct inpcb * 3603 tcp_mtudisc(struct inpcb *inp, int mtuoffer) 3604 { 3605 struct tcpcb *tp; 3606 struct socket *so; 3607 3608 INP_WLOCK_ASSERT(inp); 3609 3610 tp = intotcpcb(inp); 3611 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 3612 3613 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 3614 3615 so = inp->inp_socket; 3616 SOCK_SENDBUF_LOCK(so); 3617 /* If the mss is larger than the socket buffer, decrease the mss. */ 3618 if (so->so_snd.sb_hiwat < tp->t_maxseg) { 3619 tp->t_maxseg = so->so_snd.sb_hiwat; 3620 if (tp->t_maxseg < V_tcp_mssdflt) { 3621 /* 3622 * The MSS is so small we should not process incoming 3623 * SACK's since we are subject to attack in such a 3624 * case. 3625 */ 3626 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3627 } else { 3628 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3629 } 3630 } 3631 SOCK_SENDBUF_UNLOCK(so); 3632 3633 TCPSTAT_INC(tcps_mturesent); 3634 tp->t_rtttime = 0; 3635 tp->snd_nxt = tp->snd_una; 3636 tcp_free_sackholes(tp); 3637 tp->snd_recover = tp->snd_max; 3638 if (tp->t_flags & TF_SACK_PERMIT) 3639 EXIT_FASTRECOVERY(tp->t_flags); 3640 if (tp->t_fb->tfb_tcp_mtu_chg != NULL) { 3641 /* 3642 * Conceptually the snd_nxt setting 3643 * and freeing sack holes should 3644 * be done by the default stacks 3645 * own tfb_tcp_mtu_chg(). 3646 */ 3647 tp->t_fb->tfb_tcp_mtu_chg(tp); 3648 } 3649 if (tcp_output(tp) < 0) 3650 return (NULL); 3651 else 3652 return (inp); 3653 } 3654 3655 #ifdef INET 3656 /* 3657 * Look-up the routing entry to the peer of this inpcb. If no route 3658 * is found and it cannot be allocated, then return 0. This routine 3659 * is called by TCP routines that access the rmx structure and by 3660 * tcp_mss_update to get the peer/interface MTU. 3661 */ 3662 uint32_t 3663 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 3664 { 3665 struct nhop_object *nh; 3666 struct ifnet *ifp; 3667 uint32_t maxmtu = 0; 3668 3669 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 3670 3671 if (inc->inc_faddr.s_addr != INADDR_ANY) { 3672 nh = fib4_lookup(inc->inc_fibnum, inc->inc_faddr, 0, NHR_NONE, 0); 3673 if (nh == NULL) 3674 return (0); 3675 3676 ifp = nh->nh_ifp; 3677 maxmtu = nh->nh_mtu; 3678 3679 /* Report additional interface capabilities. */ 3680 if (cap != NULL) { 3681 if (ifp->if_capenable & IFCAP_TSO4 && 3682 ifp->if_hwassist & CSUM_TSO) { 3683 cap->ifcap |= CSUM_TSO; 3684 cap->tsomax = ifp->if_hw_tsomax; 3685 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3686 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3687 /* XXXKIB IFCAP2_IPSEC_OFFLOAD_TSO */ 3688 cap->ipsec_tso = (ifp->if_capenable2 & 3689 IFCAP2_BIT(IFCAP2_IPSEC_OFFLOAD)) != 0; 3690 } 3691 } 3692 } 3693 return (maxmtu); 3694 } 3695 #endif /* INET */ 3696 3697 #ifdef INET6 3698 uint32_t 3699 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 3700 { 3701 struct nhop_object *nh; 3702 struct in6_addr dst6; 3703 uint32_t scopeid; 3704 struct ifnet *ifp; 3705 uint32_t maxmtu = 0; 3706 3707 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 3708 3709 if (inc->inc_flags & INC_IPV6MINMTU) 3710 return (IPV6_MMTU); 3711 3712 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 3713 in6_splitscope(&inc->inc6_faddr, &dst6, &scopeid); 3714 nh = fib6_lookup(inc->inc_fibnum, &dst6, scopeid, NHR_NONE, 0); 3715 if (nh == NULL) 3716 return (0); 3717 3718 ifp = nh->nh_ifp; 3719 maxmtu = nh->nh_mtu; 3720 3721 /* Report additional interface capabilities. */ 3722 if (cap != NULL) { 3723 if (ifp->if_capenable & IFCAP_TSO6 && 3724 ifp->if_hwassist & CSUM_TSO) { 3725 cap->ifcap |= CSUM_TSO; 3726 cap->tsomax = ifp->if_hw_tsomax; 3727 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 3728 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 3729 cap->ipsec_tso = false; /* XXXKIB */ 3730 } 3731 } 3732 } 3733 3734 return (maxmtu); 3735 } 3736 3737 /* 3738 * Handle setsockopt(IPV6_USE_MIN_MTU) by a TCP stack. 3739 * 3740 * XXXGL: we are updating inpcb here with INC_IPV6MINMTU flag. 3741 * The right place to do that is ip6_setpktopt() that has just been 3742 * executed. By the way it just filled ip6po_minmtu for us. 3743 */ 3744 void 3745 tcp6_use_min_mtu(struct tcpcb *tp) 3746 { 3747 struct inpcb *inp = tptoinpcb(tp); 3748 3749 INP_WLOCK_ASSERT(inp); 3750 /* 3751 * In case of the IPV6_USE_MIN_MTU socket 3752 * option, the INC_IPV6MINMTU flag to announce 3753 * a corresponding MSS during the initial 3754 * handshake. If the TCP connection is not in 3755 * the front states, just reduce the MSS being 3756 * used. This avoids the sending of TCP 3757 * segments which will be fragmented at the 3758 * IPv6 layer. 3759 */ 3760 inp->inp_inc.inc_flags |= INC_IPV6MINMTU; 3761 if ((tp->t_state >= TCPS_SYN_SENT) && 3762 (inp->inp_inc.inc_flags & INC_ISIPV6)) { 3763 struct ip6_pktopts *opt; 3764 3765 opt = inp->in6p_outputopts; 3766 if (opt != NULL && opt->ip6po_minmtu == IP6PO_MINMTU_ALL && 3767 tp->t_maxseg > TCP6_MSS) { 3768 tp->t_maxseg = TCP6_MSS; 3769 if (tp->t_maxseg < V_tcp_mssdflt) { 3770 /* 3771 * The MSS is so small we should not process incoming 3772 * SACK's since we are subject to attack in such a 3773 * case. 3774 */ 3775 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT; 3776 } else { 3777 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT; 3778 } 3779 } 3780 } 3781 } 3782 #endif /* INET6 */ 3783 3784 /* 3785 * Calculate effective SMSS per RFC5681 definition for a given TCP 3786 * connection at its current state, taking into account SACK and etc. 3787 */ 3788 u_int 3789 tcp_maxseg(const struct tcpcb *tp) 3790 { 3791 u_int optlen; 3792 3793 if (tp->t_flags & TF_NOOPT) 3794 return (tp->t_maxseg); 3795 3796 /* 3797 * Here we have a simplified code from tcp_addoptions(), 3798 * without a proper loop, and having most of paddings hardcoded. 3799 * We might make mistakes with padding here in some edge cases, 3800 * but this is harmless, since result of tcp_maxseg() is used 3801 * only in cwnd and ssthresh estimations. 3802 */ 3803 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3804 if (tp->t_flags & TF_RCVD_TSTMP) 3805 optlen = TCPOLEN_TSTAMP_APPA; 3806 else 3807 optlen = 0; 3808 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3809 if (tp->t_flags & TF_SIGNATURE) 3810 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3811 #endif 3812 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0) { 3813 optlen += TCPOLEN_SACKHDR; 3814 optlen += tp->rcv_numsacks * TCPOLEN_SACK; 3815 optlen = PADTCPOLEN(optlen); 3816 } 3817 } else { 3818 if (tp->t_flags & TF_REQ_TSTMP) 3819 optlen = TCPOLEN_TSTAMP_APPA; 3820 else 3821 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3822 if (tp->t_flags & TF_REQ_SCALE) 3823 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3824 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3825 if (tp->t_flags & TF_SIGNATURE) 3826 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3827 #endif 3828 if (tp->t_flags & TF_SACK_PERMIT) 3829 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3830 } 3831 optlen = min(optlen, TCP_MAXOLEN); 3832 return (tp->t_maxseg - optlen); 3833 } 3834 3835 3836 u_int 3837 tcp_fixed_maxseg(const struct tcpcb *tp) 3838 { 3839 int optlen; 3840 3841 if (tp->t_flags & TF_NOOPT) 3842 return (tp->t_maxseg); 3843 3844 /* 3845 * Here we have a simplified code from tcp_addoptions(), 3846 * without a proper loop, and having most of paddings hardcoded. 3847 * We only consider fixed options that we would send every 3848 * time I.e. SACK is not considered. This is important 3849 * for cc modules to figure out what the modulo of the 3850 * cwnd should be. 3851 */ 3852 if (TCPS_HAVEESTABLISHED(tp->t_state)) { 3853 if (tp->t_flags & TF_RCVD_TSTMP) 3854 optlen = TCPOLEN_TSTAMP_APPA; 3855 else 3856 optlen = 0; 3857 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3858 if (tp->t_flags & TF_SIGNATURE) 3859 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3860 #endif 3861 } else { 3862 if (tp->t_flags & TF_REQ_TSTMP) 3863 optlen = TCPOLEN_TSTAMP_APPA; 3864 else 3865 optlen = PADTCPOLEN(TCPOLEN_MAXSEG); 3866 if (tp->t_flags & TF_REQ_SCALE) 3867 optlen += PADTCPOLEN(TCPOLEN_WINDOW); 3868 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 3869 if (tp->t_flags & TF_SIGNATURE) 3870 optlen += PADTCPOLEN(TCPOLEN_SIGNATURE); 3871 #endif 3872 if (tp->t_flags & TF_SACK_PERMIT) 3873 optlen += PADTCPOLEN(TCPOLEN_SACK_PERMITTED); 3874 } 3875 optlen = min(optlen, TCP_MAXOLEN); 3876 return (tp->t_maxseg - optlen); 3877 } 3878 3879 3880 3881 static int 3882 sysctl_drop(SYSCTL_HANDLER_ARGS) 3883 { 3884 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 3885 struct sockaddr_storage addrs[2]; 3886 struct inpcb *inp; 3887 struct tcpcb *tp; 3888 #ifdef INET 3889 struct sockaddr_in *fin = NULL, *lin = NULL; 3890 #endif 3891 struct epoch_tracker et; 3892 #ifdef INET6 3893 struct sockaddr_in6 *fin6, *lin6; 3894 #endif 3895 int error; 3896 3897 inp = NULL; 3898 #ifdef INET6 3899 fin6 = lin6 = NULL; 3900 #endif 3901 error = 0; 3902 3903 if (req->oldptr != NULL || req->oldlen != 0) 3904 return (EINVAL); 3905 if (req->newptr == NULL) 3906 return (EPERM); 3907 if (req->newlen < sizeof(addrs)) 3908 return (ENOMEM); 3909 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 3910 if (error) 3911 return (error); 3912 3913 switch (addrs[0].ss_family) { 3914 #ifdef INET6 3915 case AF_INET6: 3916 fin6 = (struct sockaddr_in6 *)&addrs[0]; 3917 lin6 = (struct sockaddr_in6 *)&addrs[1]; 3918 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 3919 lin6->sin6_len != sizeof(struct sockaddr_in6)) 3920 return (EINVAL); 3921 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 3922 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 3923 return (EINVAL); 3924 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 3925 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 3926 #ifdef INET 3927 fin = (struct sockaddr_in *)&addrs[0]; 3928 lin = (struct sockaddr_in *)&addrs[1]; 3929 #endif 3930 break; 3931 } 3932 error = sa6_embedscope(fin6, V_ip6_use_defzone); 3933 if (error) 3934 return (error); 3935 error = sa6_embedscope(lin6, V_ip6_use_defzone); 3936 if (error) 3937 return (error); 3938 break; 3939 #endif 3940 #ifdef INET 3941 case AF_INET: 3942 fin = (struct sockaddr_in *)&addrs[0]; 3943 lin = (struct sockaddr_in *)&addrs[1]; 3944 if (fin->sin_len != sizeof(struct sockaddr_in) || 3945 lin->sin_len != sizeof(struct sockaddr_in)) 3946 return (EINVAL); 3947 break; 3948 #endif 3949 default: 3950 return (EINVAL); 3951 } 3952 NET_EPOCH_ENTER(et); 3953 switch (addrs[0].ss_family) { 3954 #ifdef INET6 3955 case AF_INET6: 3956 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 3957 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 3958 INPLOOKUP_WLOCKPCB, NULL); 3959 break; 3960 #endif 3961 #ifdef INET 3962 case AF_INET: 3963 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 3964 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 3965 break; 3966 #endif 3967 } 3968 if (inp != NULL) { 3969 if (!SOLISTENING(inp->inp_socket)) { 3970 tp = intotcpcb(inp); 3971 tp = tcp_drop(tp, ECONNABORTED); 3972 if (tp != NULL) 3973 INP_WUNLOCK(inp); 3974 } else 3975 INP_WUNLOCK(inp); 3976 } else 3977 error = ESRCH; 3978 NET_EPOCH_EXIT(et); 3979 return (error); 3980 } 3981 3982 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 3983 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3984 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_drop, "", 3985 "Drop TCP connection"); 3986 3987 static int 3988 tcp_sysctl_setsockopt(SYSCTL_HANDLER_ARGS) 3989 { 3990 return (sysctl_setsockopt(oidp, arg1, arg2, req, &V_tcbinfo, 3991 &tcp_ctloutput_set)); 3992 } 3993 3994 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, setsockopt, 3995 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 3996 CTLFLAG_MPSAFE, NULL, 0, tcp_sysctl_setsockopt, "", 3997 "Set socket option for TCP endpoint"); 3998 3999 #ifdef KERN_TLS 4000 static int 4001 sysctl_switch_tls(SYSCTL_HANDLER_ARGS) 4002 { 4003 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 4004 struct sockaddr_storage addrs[2]; 4005 struct inpcb *inp; 4006 #ifdef INET 4007 struct sockaddr_in *fin = NULL, *lin = NULL; 4008 #endif 4009 struct epoch_tracker et; 4010 #ifdef INET6 4011 struct sockaddr_in6 *fin6, *lin6; 4012 #endif 4013 int error; 4014 4015 inp = NULL; 4016 #ifdef INET6 4017 fin6 = lin6 = NULL; 4018 #endif 4019 error = 0; 4020 4021 if (req->oldptr != NULL || req->oldlen != 0) 4022 return (EINVAL); 4023 if (req->newptr == NULL) 4024 return (EPERM); 4025 if (req->newlen < sizeof(addrs)) 4026 return (ENOMEM); 4027 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 4028 if (error) 4029 return (error); 4030 4031 switch (addrs[0].ss_family) { 4032 #ifdef INET6 4033 case AF_INET6: 4034 fin6 = (struct sockaddr_in6 *)&addrs[0]; 4035 lin6 = (struct sockaddr_in6 *)&addrs[1]; 4036 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 4037 lin6->sin6_len != sizeof(struct sockaddr_in6)) 4038 return (EINVAL); 4039 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 4040 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 4041 return (EINVAL); 4042 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 4043 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 4044 #ifdef INET 4045 fin = (struct sockaddr_in *)&addrs[0]; 4046 lin = (struct sockaddr_in *)&addrs[1]; 4047 #endif 4048 break; 4049 } 4050 error = sa6_embedscope(fin6, V_ip6_use_defzone); 4051 if (error) 4052 return (error); 4053 error = sa6_embedscope(lin6, V_ip6_use_defzone); 4054 if (error) 4055 return (error); 4056 break; 4057 #endif 4058 #ifdef INET 4059 case AF_INET: 4060 fin = (struct sockaddr_in *)&addrs[0]; 4061 lin = (struct sockaddr_in *)&addrs[1]; 4062 if (fin->sin_len != sizeof(struct sockaddr_in) || 4063 lin->sin_len != sizeof(struct sockaddr_in)) 4064 return (EINVAL); 4065 break; 4066 #endif 4067 default: 4068 return (EINVAL); 4069 } 4070 NET_EPOCH_ENTER(et); 4071 switch (addrs[0].ss_family) { 4072 #ifdef INET6 4073 case AF_INET6: 4074 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 4075 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 4076 INPLOOKUP_WLOCKPCB, NULL); 4077 break; 4078 #endif 4079 #ifdef INET 4080 case AF_INET: 4081 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 4082 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 4083 break; 4084 #endif 4085 } 4086 NET_EPOCH_EXIT(et); 4087 if (inp != NULL) { 4088 struct socket *so; 4089 4090 so = inp->inp_socket; 4091 soref(so); 4092 error = ktls_set_tx_mode(so, 4093 arg2 == 0 ? TCP_TLS_MODE_SW : TCP_TLS_MODE_IFNET); 4094 INP_WUNLOCK(inp); 4095 sorele(so); 4096 } else 4097 error = ESRCH; 4098 return (error); 4099 } 4100 4101 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_sw_tls, 4102 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 4103 CTLFLAG_NEEDGIANT, NULL, 0, sysctl_switch_tls, "", 4104 "Switch TCP connection to SW TLS"); 4105 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, switch_to_ifnet_tls, 4106 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP | 4107 CTLFLAG_NEEDGIANT, NULL, 1, sysctl_switch_tls, "", 4108 "Switch TCP connection to ifnet TLS"); 4109 #endif 4110 4111 /* 4112 * Generate a standardized TCP log line for use throughout the 4113 * tcp subsystem. Memory allocation is done with M_NOWAIT to 4114 * allow use in the interrupt context. 4115 * 4116 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 4117 * NB: The function may return NULL if memory allocation failed. 4118 * 4119 * Due to header inclusion and ordering limitations the struct ip 4120 * and ip6_hdr pointers have to be passed as void pointers. 4121 */ 4122 char * 4123 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4124 const void *ip6hdr) 4125 { 4126 4127 /* Is logging enabled? */ 4128 if (V_tcp_log_in_vain == 0) 4129 return (NULL); 4130 4131 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 4132 } 4133 4134 char * 4135 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4136 const void *ip6hdr) 4137 { 4138 4139 /* Is logging enabled? */ 4140 if (tcp_log_debug == 0) 4141 return (NULL); 4142 4143 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 4144 } 4145 4146 static char * 4147 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, const void *ip4hdr, 4148 const void *ip6hdr) 4149 { 4150 char *s, *sp; 4151 size_t size; 4152 #ifdef INET 4153 const struct ip *ip = (const struct ip *)ip4hdr; 4154 #endif 4155 #ifdef INET6 4156 const struct ip6_hdr *ip6 = (const struct ip6_hdr *)ip6hdr; 4157 #endif /* INET6 */ 4158 4159 /* 4160 * The log line looks like this: 4161 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 4162 */ 4163 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 4164 sizeof(PRINT_TH_FLAGS) + 1 + 4165 #ifdef INET6 4166 2 * INET6_ADDRSTRLEN; 4167 #else 4168 2 * INET_ADDRSTRLEN; 4169 #endif /* INET6 */ 4170 4171 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 4172 if (s == NULL) 4173 return (NULL); 4174 4175 strcat(s, "TCP: ["); 4176 sp = s + strlen(s); 4177 4178 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 4179 inet_ntoa_r(inc->inc_faddr, sp); 4180 sp = s + strlen(s); 4181 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4182 sp = s + strlen(s); 4183 inet_ntoa_r(inc->inc_laddr, sp); 4184 sp = s + strlen(s); 4185 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4186 #ifdef INET6 4187 } else if (inc) { 4188 ip6_sprintf(sp, &inc->inc6_faddr); 4189 sp = s + strlen(s); 4190 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 4191 sp = s + strlen(s); 4192 ip6_sprintf(sp, &inc->inc6_laddr); 4193 sp = s + strlen(s); 4194 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 4195 } else if (ip6 && th) { 4196 ip6_sprintf(sp, &ip6->ip6_src); 4197 sp = s + strlen(s); 4198 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4199 sp = s + strlen(s); 4200 ip6_sprintf(sp, &ip6->ip6_dst); 4201 sp = s + strlen(s); 4202 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4203 #endif /* INET6 */ 4204 #ifdef INET 4205 } else if (ip && th) { 4206 inet_ntoa_r(ip->ip_src, sp); 4207 sp = s + strlen(s); 4208 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 4209 sp = s + strlen(s); 4210 inet_ntoa_r(ip->ip_dst, sp); 4211 sp = s + strlen(s); 4212 sprintf(sp, "]:%i", ntohs(th->th_dport)); 4213 #endif /* INET */ 4214 } else { 4215 free(s, M_TCPLOG); 4216 return (NULL); 4217 } 4218 sp = s + strlen(s); 4219 if (th) 4220 sprintf(sp, " tcpflags 0x%b", tcp_get_flags(th), PRINT_TH_FLAGS); 4221 if (*(s + size - 1) != '\0') 4222 panic("%s: string too long", __func__); 4223 return (s); 4224 } 4225 4226 /* 4227 * A subroutine which makes it easy to track TCP state changes with DTrace. 4228 * This function shouldn't be called for t_state initializations that don't 4229 * correspond to actual TCP state transitions. 4230 */ 4231 void 4232 tcp_state_change(struct tcpcb *tp, int newstate) 4233 { 4234 #if defined(KDTRACE_HOOKS) 4235 int pstate = tp->t_state; 4236 #endif 4237 4238 TCPSTATES_DEC(tp->t_state); 4239 TCPSTATES_INC(newstate); 4240 tp->t_state = newstate; 4241 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 4242 } 4243 4244 /* 4245 * Create an external-format (``xtcpcb'') structure using the information in 4246 * the kernel-format tcpcb structure pointed to by tp. This is done to 4247 * reduce the spew of irrelevant information over this interface, to isolate 4248 * user code from changes in the kernel structure, and potentially to provide 4249 * information-hiding if we decide that some of this information should be 4250 * hidden from users. 4251 */ 4252 void 4253 tcp_inptoxtp(const struct inpcb *inp, struct xtcpcb *xt) 4254 { 4255 struct tcpcb *tp = intotcpcb(inp); 4256 sbintime_t now; 4257 4258 bzero(xt, sizeof(*xt)); 4259 xt->t_state = tp->t_state; 4260 xt->t_logstate = tcp_get_bblog_state(tp); 4261 xt->t_flags = tp->t_flags; 4262 xt->t_sndzerowin = tp->t_sndzerowin; 4263 xt->t_sndrexmitpack = tp->t_sndrexmitpack; 4264 xt->t_rcvoopack = tp->t_rcvoopack; 4265 xt->t_rcv_wnd = tp->rcv_wnd; 4266 xt->t_snd_wnd = tp->snd_wnd; 4267 xt->t_snd_cwnd = tp->snd_cwnd; 4268 xt->t_snd_ssthresh = tp->snd_ssthresh; 4269 xt->t_dsack_bytes = tp->t_dsack_bytes; 4270 xt->t_dsack_tlp_bytes = tp->t_dsack_tlp_bytes; 4271 xt->t_dsack_pack = tp->t_dsack_pack; 4272 xt->t_maxseg = tp->t_maxseg; 4273 xt->xt_ecn = (tp->t_flags2 & TF2_ECN_PERMIT) ? 1 : 0 + 4274 (tp->t_flags2 & TF2_ACE_PERMIT) ? 2 : 0; 4275 4276 now = getsbinuptime(); 4277 #define COPYTIMER(which,where) do { \ 4278 if (tp->t_timers[which] != SBT_MAX) \ 4279 xt->where = (tp->t_timers[which] - now) / SBT_1MS; \ 4280 else \ 4281 xt->where = 0; \ 4282 } while (0) 4283 COPYTIMER(TT_DELACK, tt_delack); 4284 COPYTIMER(TT_REXMT, tt_rexmt); 4285 COPYTIMER(TT_PERSIST, tt_persist); 4286 COPYTIMER(TT_KEEP, tt_keep); 4287 COPYTIMER(TT_2MSL, tt_2msl); 4288 #undef COPYTIMER 4289 xt->t_rcvtime = 1000 * (ticks - tp->t_rcvtime) / hz; 4290 4291 xt->xt_encaps_port = tp->t_port; 4292 bcopy(tp->t_fb->tfb_tcp_block_name, xt->xt_stack, 4293 TCP_FUNCTION_NAME_LEN_MAX); 4294 bcopy(CC_ALGO(tp)->name, xt->xt_cc, TCP_CA_NAME_MAX); 4295 #ifdef TCP_BLACKBOX 4296 (void)tcp_log_get_id(tp, xt->xt_logid); 4297 #endif 4298 4299 xt->xt_len = sizeof(struct xtcpcb); 4300 in_pcbtoxinpcb(inp, &xt->xt_inp); 4301 } 4302 4303 void 4304 tcp_log_end_status(struct tcpcb *tp, uint8_t status) 4305 { 4306 uint32_t bit, i; 4307 4308 if ((tp == NULL) || 4309 (status > TCP_EI_STATUS_MAX_VALUE) || 4310 (status == 0)) { 4311 /* Invalid */ 4312 return; 4313 } 4314 if (status > (sizeof(uint32_t) * 8)) { 4315 /* Should this be a KASSERT? */ 4316 return; 4317 } 4318 bit = 1U << (status - 1); 4319 if (bit & tp->t_end_info_status) { 4320 /* already logged */ 4321 return; 4322 } 4323 for (i = 0; i < TCP_END_BYTE_INFO; i++) { 4324 if (tp->t_end_info_bytes[i] == TCP_EI_EMPTY_SLOT) { 4325 tp->t_end_info_bytes[i] = status; 4326 tp->t_end_info_status |= bit; 4327 break; 4328 } 4329 } 4330 } 4331 4332 int 4333 tcp_can_enable_pacing(void) 4334 { 4335 4336 if ((tcp_pacing_limit == -1) || 4337 (tcp_pacing_limit > number_of_tcp_connections_pacing)) { 4338 atomic_fetchadd_int(&number_of_tcp_connections_pacing, 1); 4339 shadow_num_connections = number_of_tcp_connections_pacing; 4340 return (1); 4341 } else { 4342 counter_u64_add(tcp_pacing_failures, 1); 4343 return (0); 4344 } 4345 } 4346 4347 int 4348 tcp_incr_dgp_pacing_cnt(void) 4349 { 4350 if ((tcp_dgp_limit == -1) || 4351 (tcp_dgp_limit > number_of_dgp_connections)) { 4352 atomic_fetchadd_int(&number_of_dgp_connections, 1); 4353 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4354 return (1); 4355 } else { 4356 counter_u64_add(tcp_dgp_failures, 1); 4357 return (0); 4358 } 4359 } 4360 4361 static uint8_t tcp_dgp_warning = 0; 4362 4363 void 4364 tcp_dec_dgp_pacing_cnt(void) 4365 { 4366 uint32_t ret; 4367 4368 ret = atomic_fetchadd_int(&number_of_dgp_connections, -1); 4369 shadow_tcp_pacing_dgp = number_of_dgp_connections; 4370 KASSERT(ret != 0, ("number_of_dgp_connections -1 would cause wrap?")); 4371 if (ret == 0) { 4372 if (tcp_dgp_limit != -1) { 4373 printf("Warning all DGP is now disabled, count decrements invalidly!\n"); 4374 tcp_dgp_limit = 0; 4375 tcp_dgp_warning = 1; 4376 } else if (tcp_dgp_warning == 0) { 4377 printf("Warning DGP pacing is invalid, invalid decrement\n"); 4378 tcp_dgp_warning = 1; 4379 } 4380 } 4381 4382 } 4383 4384 static uint8_t tcp_pacing_warning = 0; 4385 4386 void 4387 tcp_decrement_paced_conn(void) 4388 { 4389 uint32_t ret; 4390 4391 ret = atomic_fetchadd_int(&number_of_tcp_connections_pacing, -1); 4392 shadow_num_connections = number_of_tcp_connections_pacing; 4393 KASSERT(ret != 0, ("tcp_paced_connection_exits -1 would cause wrap?")); 4394 if (ret == 0) { 4395 if (tcp_pacing_limit != -1) { 4396 printf("Warning all pacing is now disabled, count decrements invalidly!\n"); 4397 tcp_pacing_limit = 0; 4398 } else if (tcp_pacing_warning == 0) { 4399 printf("Warning pacing count is invalid, invalid decrement\n"); 4400 tcp_pacing_warning = 1; 4401 } 4402 } 4403 } 4404 4405 static void 4406 tcp_default_switch_failed(struct tcpcb *tp) 4407 { 4408 /* 4409 * If a switch fails we only need to 4410 * care about two things: 4411 * a) The t_flags2 4412 * and 4413 * b) The timer granularity. 4414 * Timeouts, at least for now, don't use the 4415 * old callout system in the other stacks so 4416 * those are hopefully safe. 4417 */ 4418 tcp_lro_features_off(tp); 4419 tcp_change_time_units(tp, TCP_TMR_GRANULARITY_TICKS); 4420 } 4421 4422 #ifdef TCP_ACCOUNTING 4423 int 4424 tcp_do_ack_accounting(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t tiwin, int mss) 4425 { 4426 if (SEQ_LT(th->th_ack, tp->snd_una)) { 4427 /* Do we have a SACK? */ 4428 if (to->to_flags & TOF_SACK) { 4429 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4430 tp->tcp_cnt_counters[ACK_SACK]++; 4431 } 4432 return (ACK_SACK); 4433 } else { 4434 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4435 tp->tcp_cnt_counters[ACK_BEHIND]++; 4436 } 4437 return (ACK_BEHIND); 4438 } 4439 } else if (th->th_ack == tp->snd_una) { 4440 /* Do we have a SACK? */ 4441 if (to->to_flags & TOF_SACK) { 4442 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4443 tp->tcp_cnt_counters[ACK_SACK]++; 4444 } 4445 return (ACK_SACK); 4446 } else if (tiwin != tp->snd_wnd) { 4447 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4448 tp->tcp_cnt_counters[ACK_RWND]++; 4449 } 4450 return (ACK_RWND); 4451 } else { 4452 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4453 tp->tcp_cnt_counters[ACK_DUPACK]++; 4454 } 4455 return (ACK_DUPACK); 4456 } 4457 } else { 4458 if (!SEQ_GT(th->th_ack, tp->snd_max)) { 4459 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4460 tp->tcp_cnt_counters[CNT_OF_ACKS_IN] += (((th->th_ack - tp->snd_una) + mss - 1)/mss); 4461 } 4462 } 4463 if (to->to_flags & TOF_SACK) { 4464 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4465 tp->tcp_cnt_counters[ACK_CUMACK_SACK]++; 4466 } 4467 return (ACK_CUMACK_SACK); 4468 } else { 4469 if (tp->t_flags2 & TF2_TCP_ACCOUNTING) { 4470 tp->tcp_cnt_counters[ACK_CUMACK]++; 4471 } 4472 return (ACK_CUMACK); 4473 } 4474 } 4475 } 4476 #endif 4477 4478 void 4479 tcp_change_time_units(struct tcpcb *tp, int granularity) 4480 { 4481 if (tp->t_tmr_granularity == granularity) { 4482 /* We are there */ 4483 return; 4484 } 4485 if (granularity == TCP_TMR_GRANULARITY_USEC) { 4486 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS), 4487 ("Granularity is not TICKS its %u in tp:%p", 4488 tp->t_tmr_granularity, tp)); 4489 tp->t_rttlow = TICKS_2_USEC(tp->t_rttlow); 4490 if (tp->t_srtt > 1) { 4491 uint32_t val, frac; 4492 4493 val = tp->t_srtt >> TCP_RTT_SHIFT; 4494 frac = tp->t_srtt & 0x1f; 4495 tp->t_srtt = TICKS_2_USEC(val); 4496 /* 4497 * frac is the fractional part of the srtt (if any) 4498 * but its in ticks and every bit represents 4499 * 1/32nd of a hz. 4500 */ 4501 if (frac) { 4502 if (hz == 1000) { 4503 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4504 } else { 4505 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4506 } 4507 tp->t_srtt += frac; 4508 } 4509 } 4510 if (tp->t_rttvar) { 4511 uint32_t val, frac; 4512 4513 val = tp->t_rttvar >> TCP_RTTVAR_SHIFT; 4514 frac = tp->t_rttvar & 0x1f; 4515 tp->t_rttvar = TICKS_2_USEC(val); 4516 /* 4517 * frac is the fractional part of the srtt (if any) 4518 * but its in ticks and every bit represents 4519 * 1/32nd of a hz. 4520 */ 4521 if (frac) { 4522 if (hz == 1000) { 4523 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_MSEC) / (uint64_t)TCP_RTT_SCALE); 4524 } else { 4525 frac = (((uint64_t)frac * (uint64_t)HPTS_USEC_IN_SEC) / ((uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE)); 4526 } 4527 tp->t_rttvar += frac; 4528 } 4529 } 4530 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_USEC; 4531 } else if (granularity == TCP_TMR_GRANULARITY_TICKS) { 4532 /* Convert back to ticks, with */ 4533 KASSERT((tp->t_tmr_granularity == TCP_TMR_GRANULARITY_USEC), 4534 ("Granularity is not USEC its %u in tp:%p", 4535 tp->t_tmr_granularity, tp)); 4536 if (tp->t_srtt > 1) { 4537 uint32_t val, frac; 4538 4539 val = USEC_2_TICKS(tp->t_srtt); 4540 frac = tp->t_srtt % (HPTS_USEC_IN_SEC / hz); 4541 tp->t_srtt = val << TCP_RTT_SHIFT; 4542 /* 4543 * frac is the fractional part here is left 4544 * over from converting to hz and shifting. 4545 * We need to convert this to the 5 bit 4546 * remainder. 4547 */ 4548 if (frac) { 4549 if (hz == 1000) { 4550 frac = (((uint64_t)frac * (uint64_t)TCP_RTT_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4551 } else { 4552 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTT_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4553 } 4554 tp->t_srtt += frac; 4555 } 4556 } 4557 if (tp->t_rttvar) { 4558 uint32_t val, frac; 4559 4560 val = USEC_2_TICKS(tp->t_rttvar); 4561 frac = tp->t_rttvar % (HPTS_USEC_IN_SEC / hz); 4562 tp->t_rttvar = val << TCP_RTTVAR_SHIFT; 4563 /* 4564 * frac is the fractional part here is left 4565 * over from converting to hz and shifting. 4566 * We need to convert this to the 4 bit 4567 * remainder. 4568 */ 4569 if (frac) { 4570 if (hz == 1000) { 4571 frac = (((uint64_t)frac * (uint64_t)TCP_RTTVAR_SCALE) / (uint64_t)HPTS_USEC_IN_MSEC); 4572 } else { 4573 frac = (((uint64_t)frac * (uint64_t)(hz) * (uint64_t)TCP_RTTVAR_SCALE) /(uint64_t)HPTS_USEC_IN_SEC); 4574 } 4575 tp->t_rttvar += frac; 4576 } 4577 } 4578 tp->t_rttlow = USEC_2_TICKS(tp->t_rttlow); 4579 tp->t_tmr_granularity = TCP_TMR_GRANULARITY_TICKS; 4580 } 4581 #ifdef INVARIANTS 4582 else { 4583 panic("Unknown granularity:%d tp:%p", 4584 granularity, tp); 4585 } 4586 #endif 4587 } 4588 4589 void 4590 tcp_handle_orphaned_packets(struct tcpcb *tp) 4591 { 4592 struct mbuf *save, *m, *prev; 4593 /* 4594 * Called when a stack switch is occuring from the fini() 4595 * of the old stack. We assue the init() as already been 4596 * run of the new stack and it has set the t_flags2 to 4597 * what it supports. This function will then deal with any 4598 * differences i.e. cleanup packets that maybe queued that 4599 * the newstack does not support. 4600 */ 4601 4602 if (tp->t_flags2 & TF2_MBUF_L_ACKS) 4603 return; 4604 if ((tp->t_flags2 & TF2_SUPPORTS_MBUFQ) == 0 && 4605 !STAILQ_EMPTY(&tp->t_inqueue)) { 4606 /* 4607 * It is unsafe to process the packets since a 4608 * reset may be lurking in them (its rare but it 4609 * can occur). If we were to find a RST, then we 4610 * would end up dropping the connection and the 4611 * INP lock, so when we return the caller (tcp_usrreq) 4612 * will blow up when it trys to unlock the inp. 4613 * This new stack does not do any fancy LRO features 4614 * so all we can do is toss the packets. 4615 */ 4616 m = STAILQ_FIRST(&tp->t_inqueue); 4617 STAILQ_INIT(&tp->t_inqueue); 4618 STAILQ_FOREACH_FROM_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) 4619 m_freem(m); 4620 } else { 4621 /* 4622 * Here we have a stack that does mbuf queuing but 4623 * does not support compressed ack's. We must 4624 * walk all the mbufs and discard any compressed acks. 4625 */ 4626 STAILQ_FOREACH_SAFE(m, &tp->t_inqueue, m_stailqpkt, save) { 4627 if (m->m_flags & M_ACKCMP) { 4628 if (m == STAILQ_FIRST(&tp->t_inqueue)) 4629 STAILQ_REMOVE_HEAD(&tp->t_inqueue, 4630 m_stailqpkt); 4631 else 4632 STAILQ_REMOVE_AFTER(&tp->t_inqueue, 4633 prev, m_stailqpkt); 4634 m_freem(m); 4635 } else 4636 prev = m; 4637 } 4638 } 4639 } 4640 4641 #ifdef TCP_REQUEST_TRK 4642 uint32_t 4643 tcp_estimate_tls_overhead(struct socket *so, uint64_t tls_usr_bytes) 4644 { 4645 #ifdef KERN_TLS 4646 struct ktls_session *tls; 4647 uint32_t rec_oh, records; 4648 4649 tls = so->so_snd.sb_tls_info; 4650 if (tls == NULL) 4651 return (0); 4652 4653 rec_oh = tls->params.tls_hlen + tls->params.tls_tlen; 4654 records = ((tls_usr_bytes + tls->params.max_frame_len - 1)/tls->params.max_frame_len); 4655 return (records * rec_oh); 4656 #else 4657 return (0); 4658 #endif 4659 } 4660 4661 extern uint32_t tcp_stale_entry_time; 4662 uint32_t tcp_stale_entry_time = 250000; 4663 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, usrlog_stale, CTLFLAG_RW, 4664 &tcp_stale_entry_time, 250000, "Time that a tcpreq entry without a sendfile ages out"); 4665 4666 void 4667 tcp_req_log_req_info(struct tcpcb *tp, struct tcp_sendfile_track *req, 4668 uint16_t slot, uint8_t val, uint64_t offset, uint64_t nbytes) 4669 { 4670 if (tcp_bblogging_on(tp)) { 4671 union tcp_log_stackspecific log; 4672 struct timeval tv; 4673 4674 memset(&log, 0, sizeof(log)); 4675 log.u_bbr.inhpts = tcp_in_hpts(tp); 4676 log.u_bbr.flex8 = val; 4677 log.u_bbr.rttProp = req->timestamp; 4678 log.u_bbr.delRate = req->start; 4679 log.u_bbr.cur_del_rate = req->end; 4680 log.u_bbr.flex1 = req->start_seq; 4681 log.u_bbr.flex2 = req->end_seq; 4682 log.u_bbr.flex3 = req->flags; 4683 log.u_bbr.flex4 = ((req->localtime >> 32) & 0x00000000ffffffff); 4684 log.u_bbr.flex5 = (req->localtime & 0x00000000ffffffff); 4685 log.u_bbr.flex7 = slot; 4686 log.u_bbr.bw_inuse = offset; 4687 /* nbytes = flex6 | epoch */ 4688 log.u_bbr.flex6 = ((nbytes >> 32) & 0x00000000ffffffff); 4689 log.u_bbr.epoch = (nbytes & 0x00000000ffffffff); 4690 /* cspr = lt_epoch | pkts_out */ 4691 log.u_bbr.lt_epoch = ((req->cspr >> 32) & 0x00000000ffffffff); 4692 log.u_bbr.pkts_out |= (req->cspr & 0x00000000ffffffff); 4693 log.u_bbr.applimited = tp->t_tcpreq_closed; 4694 log.u_bbr.applimited <<= 8; 4695 log.u_bbr.applimited |= tp->t_tcpreq_open; 4696 log.u_bbr.applimited <<= 8; 4697 log.u_bbr.applimited |= tp->t_tcpreq_req; 4698 log.u_bbr.timeStamp = tcp_get_usecs(&tv); 4699 TCP_LOG_EVENTP(tp, NULL, 4700 &tptosocket(tp)->so_rcv, 4701 &tptosocket(tp)->so_snd, 4702 TCP_LOG_REQ_T, 0, 4703 0, &log, false, &tv); 4704 } 4705 } 4706 4707 void 4708 tcp_req_free_a_slot(struct tcpcb *tp, struct tcp_sendfile_track *ent) 4709 { 4710 if (tp->t_tcpreq_req > 0) 4711 tp->t_tcpreq_req--; 4712 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4713 if (tp->t_tcpreq_open > 0) 4714 tp->t_tcpreq_open--; 4715 } else { 4716 if (tp->t_tcpreq_closed > 0) 4717 tp->t_tcpreq_closed--; 4718 } 4719 ent->flags = TCP_TRK_TRACK_FLG_EMPTY; 4720 } 4721 4722 static void 4723 tcp_req_check_for_stale_entries(struct tcpcb *tp, uint64_t ts, int rm_oldest) 4724 { 4725 struct tcp_sendfile_track *ent; 4726 uint64_t time_delta, oldest_delta; 4727 int i, oldest, oldest_set = 0, cnt_rm = 0; 4728 4729 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4730 ent = &tp->t_tcpreq_info[i]; 4731 if (ent->flags != TCP_TRK_TRACK_FLG_USED) { 4732 /* 4733 * We only care about closed end ranges 4734 * that are allocated and have no sendfile 4735 * ever touching them. They would be in 4736 * state USED. 4737 */ 4738 continue; 4739 } 4740 if (ts >= ent->localtime) 4741 time_delta = ts - ent->localtime; 4742 else 4743 time_delta = 0; 4744 if (time_delta && 4745 ((oldest_delta < time_delta) || (oldest_set == 0))) { 4746 oldest_set = 1; 4747 oldest = i; 4748 oldest_delta = time_delta; 4749 } 4750 if (tcp_stale_entry_time && (time_delta >= tcp_stale_entry_time)) { 4751 /* 4752 * No sendfile in a our time-limit 4753 * time to purge it. 4754 */ 4755 cnt_rm++; 4756 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4757 time_delta, 0); 4758 tcp_req_free_a_slot(tp, ent); 4759 } 4760 } 4761 if ((cnt_rm == 0) && rm_oldest && oldest_set) { 4762 ent = &tp->t_tcpreq_info[oldest]; 4763 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], i, TCP_TRK_REQ_LOG_STALE, 4764 oldest_delta, 1); 4765 tcp_req_free_a_slot(tp, ent); 4766 } 4767 } 4768 4769 int 4770 tcp_req_check_for_comp(struct tcpcb *tp, tcp_seq ack_point) 4771 { 4772 int i, ret = 0; 4773 struct tcp_sendfile_track *ent; 4774 4775 /* Clean up any old closed end requests that are now completed */ 4776 if (tp->t_tcpreq_req == 0) 4777 return (0); 4778 if (tp->t_tcpreq_closed == 0) 4779 return (0); 4780 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4781 ent = &tp->t_tcpreq_info[i]; 4782 /* Skip empty ones */ 4783 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4784 continue; 4785 /* Skip open ones */ 4786 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) 4787 continue; 4788 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4789 /* We are past it -- free it */ 4790 tcp_req_log_req_info(tp, ent, 4791 i, TCP_TRK_REQ_LOG_FREED, 0, 0); 4792 tcp_req_free_a_slot(tp, ent); 4793 ret++; 4794 } 4795 } 4796 return (ret); 4797 } 4798 4799 int 4800 tcp_req_is_entry_comp(struct tcpcb *tp, struct tcp_sendfile_track *ent, tcp_seq ack_point) 4801 { 4802 if (tp->t_tcpreq_req == 0) 4803 return (-1); 4804 if (tp->t_tcpreq_closed == 0) 4805 return (-1); 4806 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4807 return (-1); 4808 if (SEQ_GEQ(ack_point, ent->end_seq)) { 4809 return (1); 4810 } 4811 return (0); 4812 } 4813 4814 struct tcp_sendfile_track * 4815 tcp_req_find_a_req_that_is_completed_by(struct tcpcb *tp, tcp_seq th_ack, int *ip) 4816 { 4817 /* 4818 * Given an ack point (th_ack) walk through our entries and 4819 * return the first one found that th_ack goes past the 4820 * end_seq. 4821 */ 4822 struct tcp_sendfile_track *ent; 4823 int i; 4824 4825 if (tp->t_tcpreq_req == 0) { 4826 /* none open */ 4827 return (NULL); 4828 } 4829 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4830 ent = &tp->t_tcpreq_info[i]; 4831 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) 4832 continue; 4833 if ((ent->flags & TCP_TRK_TRACK_FLG_OPEN) == 0) { 4834 if (SEQ_GEQ(th_ack, ent->end_seq)) { 4835 *ip = i; 4836 return (ent); 4837 } 4838 } 4839 } 4840 return (NULL); 4841 } 4842 4843 struct tcp_sendfile_track * 4844 tcp_req_find_req_for_seq(struct tcpcb *tp, tcp_seq seq) 4845 { 4846 struct tcp_sendfile_track *ent; 4847 int i; 4848 4849 if (tp->t_tcpreq_req == 0) { 4850 /* none open */ 4851 return (NULL); 4852 } 4853 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4854 ent = &tp->t_tcpreq_info[i]; 4855 tcp_req_log_req_info(tp, ent, i, TCP_TRK_REQ_LOG_SEARCH, 4856 (uint64_t)seq, 0); 4857 if (ent->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4858 continue; 4859 } 4860 if (ent->flags & TCP_TRK_TRACK_FLG_OPEN) { 4861 /* 4862 * An open end request only needs to 4863 * match the beginning seq or be 4864 * all we have (once we keep going on 4865 * a open end request we may have a seq 4866 * wrap). 4867 */ 4868 if ((SEQ_GEQ(seq, ent->start_seq)) || 4869 (tp->t_tcpreq_closed == 0)) 4870 return (ent); 4871 } else { 4872 /* 4873 * For this one we need to 4874 * be a bit more careful if its 4875 * completed at least. 4876 */ 4877 if ((SEQ_GEQ(seq, ent->start_seq)) && 4878 (SEQ_LT(seq, ent->end_seq))) { 4879 return (ent); 4880 } 4881 } 4882 } 4883 return (NULL); 4884 } 4885 4886 /* Should this be in its own file tcp_req.c ? */ 4887 struct tcp_sendfile_track * 4888 tcp_req_alloc_req_full(struct tcpcb *tp, struct tcp_snd_req *req, uint64_t ts, int rec_dups) 4889 { 4890 struct tcp_sendfile_track *fil; 4891 int i, allocated; 4892 4893 /* In case the stack does not check for completions do so now */ 4894 tcp_req_check_for_comp(tp, tp->snd_una); 4895 /* Check for stale entries */ 4896 if (tp->t_tcpreq_req) 4897 tcp_req_check_for_stale_entries(tp, ts, 4898 (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ)); 4899 /* Check to see if this is a duplicate of one not started */ 4900 if (tp->t_tcpreq_req) { 4901 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4902 fil = &tp->t_tcpreq_info[i]; 4903 if ((fil->flags & TCP_TRK_TRACK_FLG_USED) == 0) 4904 continue; 4905 if ((fil->timestamp == req->timestamp) && 4906 (fil->start == req->start) && 4907 ((fil->flags & TCP_TRK_TRACK_FLG_OPEN) || 4908 (fil->end == req->end))) { 4909 /* 4910 * We already have this request 4911 * and it has not been started with sendfile. 4912 * This probably means the user was returned 4913 * a 4xx of some sort and its going to age 4914 * out, lets not duplicate it. 4915 */ 4916 return (fil); 4917 } 4918 } 4919 } 4920 /* Ok if there is no room at the inn we are in trouble */ 4921 if (tp->t_tcpreq_req >= MAX_TCP_TRK_REQ) { 4922 tcp_trace_point(tp, TCP_TP_REQ_LOG_FAIL); 4923 for (i = 0; i < MAX_TCP_TRK_REQ; i++) { 4924 tcp_req_log_req_info(tp, &tp->t_tcpreq_info[i], 4925 i, TCP_TRK_REQ_LOG_ALLOCFAIL, 0, 0); 4926 } 4927 return (NULL); 4928 } 4929 for (i = 0, allocated = 0; i < MAX_TCP_TRK_REQ; i++) { 4930 fil = &tp->t_tcpreq_info[i]; 4931 if (fil->flags == TCP_TRK_TRACK_FLG_EMPTY) { 4932 allocated = 1; 4933 fil->flags = TCP_TRK_TRACK_FLG_USED; 4934 fil->timestamp = req->timestamp; 4935 fil->playout_ms = req->playout_ms; 4936 fil->localtime = ts; 4937 fil->start = req->start; 4938 if (req->flags & TCP_LOG_HTTPD_RANGE_END) { 4939 fil->end = req->end; 4940 } else { 4941 fil->end = 0; 4942 fil->flags |= TCP_TRK_TRACK_FLG_OPEN; 4943 } 4944 /* 4945 * We can set the min boundaries to the TCP Sequence space, 4946 * but it might be found to be further up when sendfile 4947 * actually runs on this range (if it ever does). 4948 */ 4949 fil->sbcc_at_s = tptosocket(tp)->so_snd.sb_ccc; 4950 fil->start_seq = tp->snd_una + 4951 tptosocket(tp)->so_snd.sb_ccc; 4952 if (req->flags & TCP_LOG_HTTPD_RANGE_END) 4953 fil->end_seq = (fil->start_seq + ((uint32_t)(fil->end - fil->start))); 4954 else 4955 fil->end_seq = 0; 4956 if (tptosocket(tp)->so_snd.sb_tls_info) { 4957 /* 4958 * This session is doing TLS. Take a swag guess 4959 * at the overhead. 4960 */ 4961 fil->end_seq += tcp_estimate_tls_overhead( 4962 tptosocket(tp), (fil->end - fil->start)); 4963 } 4964 tp->t_tcpreq_req++; 4965 if (fil->flags & TCP_TRK_TRACK_FLG_OPEN) 4966 tp->t_tcpreq_open++; 4967 else 4968 tp->t_tcpreq_closed++; 4969 tcp_req_log_req_info(tp, fil, i, 4970 TCP_TRK_REQ_LOG_NEW, 0, 0); 4971 break; 4972 } else 4973 fil = NULL; 4974 } 4975 return (fil); 4976 } 4977 4978 void 4979 tcp_req_alloc_req(struct tcpcb *tp, union tcp_log_userdata *user, uint64_t ts) 4980 { 4981 (void)tcp_req_alloc_req_full(tp, &user->tcp_req, ts, 1); 4982 } 4983 #endif 4984 4985 void 4986 tcp_log_socket_option(struct tcpcb *tp, uint32_t option_num, uint32_t option_val, int err) 4987 { 4988 if (tcp_bblogging_on(tp)) { 4989 struct tcp_log_buffer *l; 4990 4991 l = tcp_log_event(tp, NULL, 4992 &tptosocket(tp)->so_rcv, 4993 &tptosocket(tp)->so_snd, 4994 TCP_LOG_SOCKET_OPT, 4995 err, 0, NULL, 1, 4996 NULL, NULL, 0, NULL); 4997 if (l) { 4998 l->tlb_flex1 = option_num; 4999 l->tlb_flex2 = option_val; 5000 } 5001 } 5002 } 5003 5004 uint32_t 5005 tcp_get_srtt(struct tcpcb *tp, int granularity) 5006 { 5007 uint32_t srtt; 5008 5009 KASSERT(granularity == TCP_TMR_GRANULARITY_USEC || 5010 granularity == TCP_TMR_GRANULARITY_TICKS, 5011 ("%s: called with unexpected granularity %d", __func__, 5012 granularity)); 5013 5014 srtt = tp->t_srtt; 5015 5016 /* 5017 * We only support two granularities. If the stored granularity 5018 * does not match the granularity requested by the caller, 5019 * convert the stored value to the requested unit of granularity. 5020 */ 5021 if (tp->t_tmr_granularity != granularity) { 5022 if (granularity == TCP_TMR_GRANULARITY_USEC) 5023 srtt = TICKS_2_USEC(srtt); 5024 else 5025 srtt = USEC_2_TICKS(srtt); 5026 } 5027 5028 /* 5029 * If the srtt is stored with ticks granularity, we need to 5030 * unshift to get the actual value. We do this after the 5031 * conversion above (if one was necessary) in order to maximize 5032 * precision. 5033 */ 5034 if (tp->t_tmr_granularity == TCP_TMR_GRANULARITY_TICKS) 5035 srtt = srtt >> TCP_RTT_SHIFT; 5036 5037 return (srtt); 5038 } 5039 5040 void 5041 tcp_account_for_send(struct tcpcb *tp, uint32_t len, uint8_t is_rxt, 5042 uint8_t is_tlp, bool hw_tls) 5043 { 5044 5045 if (is_tlp) { 5046 tp->t_sndtlppack++; 5047 tp->t_sndtlpbyte += len; 5048 } 5049 /* To get total bytes sent you must add t_snd_rxt_bytes to t_sndbytes */ 5050 if (is_rxt) 5051 tp->t_snd_rxt_bytes += len; 5052 else 5053 tp->t_sndbytes += len; 5054 5055 #ifdef KERN_TLS 5056 if (hw_tls && is_rxt && len != 0) { 5057 uint64_t rexmit_percent; 5058 5059 rexmit_percent = (1000ULL * tp->t_snd_rxt_bytes) / 5060 (10ULL * (tp->t_snd_rxt_bytes + tp->t_sndbytes)); 5061 if (rexmit_percent > ktls_ifnet_max_rexmit_pct) 5062 ktls_disable_ifnet(tp); 5063 } 5064 #endif 5065 } 5066