tcp_subr.c revision 302233
1232809Sjmallett/*- 2232809Sjmallett * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 3232809Sjmallett * The Regents of the University of California. All rights reserved. 4232809Sjmallett * 5232809Sjmallett * Redistribution and use in source and binary forms, with or without 6232809Sjmallett * modification, are permitted provided that the following conditions 7232809Sjmallett * are met: 8232809Sjmallett * 1. Redistributions of source code must retain the above copyright 9232809Sjmallett * notice, this list of conditions and the following disclaimer. 10232809Sjmallett * 2. Redistributions in binary form must reproduce the above copyright 11232809Sjmallett * notice, this list of conditions and the following disclaimer in the 12232809Sjmallett * documentation and/or other materials provided with the distribution. 13232809Sjmallett * 4. Neither the name of the University nor the names of its contributors 14232809Sjmallett * may be used to endorse or promote products derived from this software 15232809Sjmallett * without specific prior written permission. 16232809Sjmallett * 17232809Sjmallett * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18232809Sjmallett * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19232809Sjmallett * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20232809Sjmallett * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21232809Sjmallett * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22232809Sjmallett * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23232809Sjmallett * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24232809Sjmallett * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25232809Sjmallett * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26232809Sjmallett * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27232809Sjmallett * SUCH DAMAGE. 28232809Sjmallett * 29232809Sjmallett * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 30232809Sjmallett */ 31232809Sjmallett 32232809Sjmallett#include <sys/cdefs.h> 33232809Sjmallett__FBSDID("$FreeBSD: stable/10/sys/netinet/tcp_subr.c 302233 2016-06-27 21:44:27Z bdrewery $"); 34232809Sjmallett 35232809Sjmallett#include "opt_compat.h" 36232809Sjmallett#include "opt_inet.h" 37232809Sjmallett#include "opt_inet6.h" 38232809Sjmallett#include "opt_ipsec.h" 39232809Sjmallett#include "opt_kdtrace.h" 40232809Sjmallett#include "opt_tcpdebug.h" 41232809Sjmallett 42232809Sjmallett#include <sys/param.h> 43232809Sjmallett#include <sys/systm.h> 44232809Sjmallett#include <sys/callout.h> 45232809Sjmallett#include <sys/hhook.h> 46232809Sjmallett#include <sys/kernel.h> 47232809Sjmallett#include <sys/khelp.h> 48232809Sjmallett#include <sys/sysctl.h> 49232809Sjmallett#include <sys/jail.h> 50232809Sjmallett#include <sys/malloc.h> 51232809Sjmallett#include <sys/mbuf.h> 52232809Sjmallett#ifdef INET6 53232809Sjmallett#include <sys/domain.h> 54232809Sjmallett#endif 55232809Sjmallett#include <sys/priv.h> 56232809Sjmallett#include <sys/proc.h> 57232809Sjmallett#include <sys/sdt.h> 58232809Sjmallett#include <sys/socket.h> 59232809Sjmallett#include <sys/socketvar.h> 60232809Sjmallett#include <sys/protosw.h> 61232809Sjmallett#include <sys/random.h> 62232812Sjmallett 63232809Sjmallett#include <vm/uma.h> 64232809Sjmallett 65232812Sjmallett#include <net/route.h> 66232809Sjmallett#include <net/if.h> 67232809Sjmallett#include <net/vnet.h> 68232809Sjmallett 69232809Sjmallett#include <netinet/cc.h> 70232809Sjmallett#include <netinet/in.h> 71232809Sjmallett#include <netinet/in_kdtrace.h> 72232809Sjmallett#include <netinet/in_pcb.h> 73232809Sjmallett#include <netinet/in_systm.h> 74232809Sjmallett#include <netinet/in_var.h> 75232809Sjmallett#include <netinet/ip.h> 76232809Sjmallett#include <netinet/ip_icmp.h> 77232809Sjmallett#include <netinet/ip_var.h> 78232809Sjmallett#ifdef INET6 79232809Sjmallett#include <netinet/ip6.h> 80232809Sjmallett#include <netinet6/in6_pcb.h> 81232809Sjmallett#include <netinet6/ip6_var.h> 82232809Sjmallett#include <netinet6/scope6_var.h> 83232809Sjmallett#include <netinet6/nd6.h> 84232809Sjmallett#endif 85232809Sjmallett 86232809Sjmallett#ifdef TCP_RFC7413 87232809Sjmallett#include <netinet/tcp_fastopen.h> 88232809Sjmallett#endif 89232809Sjmallett#include <netinet/tcp_fsm.h> 90232809Sjmallett#include <netinet/tcp_seq.h> 91232809Sjmallett#include <netinet/tcp_timer.h> 92232809Sjmallett#include <netinet/tcp_var.h> 93232809Sjmallett#include <netinet/tcp_syncache.h> 94232809Sjmallett#ifdef INET6 95232809Sjmallett#include <netinet6/tcp6_var.h> 96232809Sjmallett#endif 97232809Sjmallett#include <netinet/tcpip.h> 98232809Sjmallett#ifdef TCPDEBUG 99232809Sjmallett#include <netinet/tcp_debug.h> 100232809Sjmallett#endif 101232809Sjmallett#ifdef INET6 102232809Sjmallett#include <netinet6/ip6protosw.h> 103232809Sjmallett#endif 104232809Sjmallett#ifdef TCP_OFFLOAD 105232809Sjmallett#include <netinet/tcp_offload.h> 106232809Sjmallett#endif 107232809Sjmallett 108232809Sjmallett#ifdef IPSEC 109232809Sjmallett#include <netipsec/ipsec.h> 110232809Sjmallett#include <netipsec/xform.h> 111232809Sjmallett#ifdef INET6 112232809Sjmallett#include <netipsec/ipsec6.h> 113232809Sjmallett#endif 114232809Sjmallett#include <netipsec/key.h> 115232809Sjmallett#include <sys/syslog.h> 116232809Sjmallett#endif /*IPSEC*/ 117232809Sjmallett 118232809Sjmallett#include <machine/in_cksum.h> 119232809Sjmallett#include <sys/md5.h> 120232809Sjmallett 121232809Sjmallett#include <security/mac/mac_framework.h> 122232809Sjmallett 123232809SjmallettVNET_DEFINE(int, tcp_mssdflt) = TCP_MSS; 124232809Sjmallett#ifdef INET6 125232809SjmallettVNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS; 126232809Sjmallett#endif 127232809Sjmallett 128232809Sjmallettstatic int 129232809Sjmallettsysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS) 130232809Sjmallett{ 131232809Sjmallett int error, new; 132232809Sjmallett 133232809Sjmallett new = V_tcp_mssdflt; 134232809Sjmallett error = sysctl_handle_int(oidp, &new, 0, req); 135232809Sjmallett if (error == 0 && req->newptr) { 136232809Sjmallett if (new < TCP_MINMSS) 137232809Sjmallett error = EINVAL; 138232809Sjmallett else 139232809Sjmallett V_tcp_mssdflt = new; 140232809Sjmallett } 141232809Sjmallett return (error); 142232809Sjmallett} 143232809Sjmallett 144232809SjmallettSYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, 145232809Sjmallett CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0, 146232809Sjmallett &sysctl_net_inet_tcp_mss_check, "I", 147232809Sjmallett "Default TCP Maximum Segment Size"); 148232809Sjmallett 149232809Sjmallett#ifdef INET6 150232809Sjmallettstatic int 151232809Sjmallettsysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS) 152232809Sjmallett{ 153232809Sjmallett int error, new; 154232809Sjmallett 155232809Sjmallett new = V_tcp_v6mssdflt; 156232809Sjmallett error = sysctl_handle_int(oidp, &new, 0, req); 157232809Sjmallett if (error == 0 && req->newptr) { 158232809Sjmallett if (new < TCP_MINMSS) 159232809Sjmallett error = EINVAL; 160232809Sjmallett else 161232809Sjmallett V_tcp_v6mssdflt = new; 162232809Sjmallett } 163232809Sjmallett return (error); 164232809Sjmallett} 165232809Sjmallett 166232809SjmallettSYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, 167232809Sjmallett CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0, 168232809Sjmallett &sysctl_net_inet_tcp_mss_v6_check, "I", 169232809Sjmallett "Default TCP Maximum Segment Size for IPv6"); 170232809Sjmallett#endif /* INET6 */ 171232809Sjmallett 172232809Sjmallett/* 173232809Sjmallett * Minimum MSS we accept and use. This prevents DoS attacks where 174232809Sjmallett * we are forced to a ridiculous low MSS like 20 and send hundreds 175232809Sjmallett * of packets instead of one. The effect scales with the available 176232809Sjmallett * bandwidth and quickly saturates the CPU and network interface 177232809Sjmallett * with packet generation and sending. Set to zero to disable MINMSS 178232809Sjmallett * checking. This setting prevents us from sending too small packets. 179232809Sjmallett */ 180232809SjmallettVNET_DEFINE(int, tcp_minmss) = TCP_MINMSS; 181232809SjmallettSYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW, 182232809Sjmallett &VNET_NAME(tcp_minmss), 0, 183232809Sjmallett "Minimum TCP Maximum Segment Size"); 184232809Sjmallett 185232809SjmallettVNET_DEFINE(int, tcp_do_rfc1323) = 1; 186232809SjmallettSYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW, 187232809Sjmallett &VNET_NAME(tcp_do_rfc1323), 0, 188232809Sjmallett "Enable rfc1323 (high performance TCP) extensions"); 189232809Sjmallett 190232809Sjmallettstatic int tcp_log_debug = 0; 191232809SjmallettSYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW, 192232809Sjmallett &tcp_log_debug, 0, "Log errors caused by incoming TCP segments"); 193232809Sjmallett 194232809Sjmallettstatic int tcp_tcbhashsize = 0; 195232809SjmallettSYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN, 196232809Sjmallett &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable"); 197232809Sjmallett 198232809Sjmallettstatic int do_tcpdrain = 1; 199232809SjmallettSYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0, 200232809Sjmallett "Enable tcp_drain routine for extra help when low on mbufs"); 201232809Sjmallett 202232809SjmallettSYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD, 203232809Sjmallett &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs"); 204232809Sjmallett 205232809Sjmallettstatic VNET_DEFINE(int, icmp_may_rst) = 1; 206232809Sjmallett#define V_icmp_may_rst VNET(icmp_may_rst) 207232809SjmallettSYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, 208232809Sjmallett &VNET_NAME(icmp_may_rst), 0, 209232809Sjmallett "Certain ICMP unreachable messages may abort connections in SYN_SENT"); 210232809Sjmallett 211232809Sjmallettstatic VNET_DEFINE(int, tcp_isn_reseed_interval) = 0; 212232809Sjmallett#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval) 213232809SjmallettSYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW, 214232809Sjmallett &VNET_NAME(tcp_isn_reseed_interval), 0, 215232809Sjmallett "Seconds between reseeding of ISN secret"); 216232809Sjmallett 217232809Sjmallettstatic int tcp_soreceive_stream = 0; 218232809SjmallettSYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN, 219232809Sjmallett &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets"); 220232809Sjmallett 221232809Sjmallett#ifdef TCP_SIGNATURE 222232809Sjmallettstatic int tcp_sig_checksigs = 1; 223232809SjmallettSYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW, 224232809Sjmallett &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic"); 225232809Sjmallett#endif 226232809Sjmallett 227232809SjmallettVNET_DEFINE(uma_zone_t, sack_hole_zone); 228232809Sjmallett#define V_sack_hole_zone VNET(sack_hole_zone) 229232809Sjmallett 230232809SjmallettVNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]); 231232809Sjmallett 232232809Sjmallettstatic struct inpcb *tcp_notify(struct inpcb *, int); 233232809Sjmallettstatic struct inpcb *tcp_mtudisc_notify(struct inpcb *, int); 234232809Sjmallettstatic char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, 235232809Sjmallett void *ip4hdr, const void *ip6hdr); 236232809Sjmallettstatic void tcp_timer_discard(struct tcpcb *, uint32_t); 237232809Sjmallett 238232809Sjmallett/* 239232809Sjmallett * Target size of TCP PCB hash tables. Must be a power of two. 240232809Sjmallett * 241232809Sjmallett * Note that this can be overridden by the kernel environment 242232809Sjmallett * variable net.inet.tcp.tcbhashsize 243232809Sjmallett */ 244232809Sjmallett#ifndef TCBHASHSIZE 245232809Sjmallett#define TCBHASHSIZE 0 246232809Sjmallett#endif 247232809Sjmallett 248232809Sjmallett/* 249232809Sjmallett * XXX 250232809Sjmallett * Callouts should be moved into struct tcp directly. They are currently 251232809Sjmallett * separate because the tcpcb structure is exported to userland for sysctl 252232809Sjmallett * parsing purposes, which do not know about callouts. 253232809Sjmallett */ 254232809Sjmallettstruct tcpcb_mem { 255232809Sjmallett struct tcpcb tcb; 256232809Sjmallett struct tcp_timer tt; 257232809Sjmallett struct cc_var ccv; 258232809Sjmallett struct osd osd; 259232809Sjmallett}; 260232809Sjmallett 261232809Sjmallettstatic VNET_DEFINE(uma_zone_t, tcpcb_zone); 262232809Sjmallett#define V_tcpcb_zone VNET(tcpcb_zone) 263232809Sjmallett 264232809SjmallettMALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers"); 265232809Sjmallettstatic struct mtx isn_mtx; 266232809Sjmallett 267232809Sjmallett#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF) 268232809Sjmallett#define ISN_LOCK() mtx_lock(&isn_mtx) 269232809Sjmallett#define ISN_UNLOCK() mtx_unlock(&isn_mtx) 270232809Sjmallett 271232809Sjmallett/* 272232809Sjmallett * TCP initialization. 273232809Sjmallett */ 274232809Sjmallettstatic void 275232809Sjmalletttcp_zone_change(void *tag) 276232809Sjmallett{ 277232809Sjmallett 278232809Sjmallett uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets); 279232809Sjmallett uma_zone_set_max(V_tcpcb_zone, maxsockets); 280232809Sjmallett tcp_tw_zone_change(); 281232809Sjmallett} 282232809Sjmallett 283232809Sjmallettstatic int 284232809Sjmalletttcp_inpcb_init(void *mem, int size, int flags) 285232809Sjmallett{ 286232809Sjmallett struct inpcb *inp = mem; 287232809Sjmallett 288232809Sjmallett INP_LOCK_INIT(inp, "inp", "tcpinp"); 289232809Sjmallett return (0); 290232809Sjmallett} 291232809Sjmallett 292232809Sjmallett/* 293232809Sjmallett * Take a value and get the next power of 2 that doesn't overflow. 294232809Sjmallett * Used to size the tcp_inpcb hash buckets. 295232809Sjmallett */ 296232809Sjmallettstatic int 297232809Sjmallettmaketcp_hashsize(int size) 298232809Sjmallett{ 299232809Sjmallett int hashsize; 300232809Sjmallett 301232809Sjmallett /* 302232809Sjmallett * auto tune. 303232809Sjmallett * get the next power of 2 higher than maxsockets. 304232809Sjmallett */ 305232809Sjmallett hashsize = 1 << fls(size); 306232809Sjmallett /* catch overflow, and just go one power of 2 smaller */ 307232809Sjmallett if (hashsize < size) { 308232809Sjmallett hashsize = 1 << (fls(size) - 1); 309232809Sjmallett } 310232809Sjmallett return (hashsize); 311232809Sjmallett} 312232809Sjmallett 313232809Sjmallettvoid 314232809Sjmalletttcp_init(void) 315232809Sjmallett{ 316232809Sjmallett const char *tcbhash_tuneable; 317232809Sjmallett int hashsize; 318232809Sjmallett 319232809Sjmallett tcbhash_tuneable = "net.inet.tcp.tcbhashsize"; 320232809Sjmallett 321232809Sjmallett if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, 322232809Sjmallett &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 323232809Sjmallett printf("%s: WARNING: unable to register helper hook\n", __func__); 324232809Sjmallett if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, 325232809Sjmallett &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0) 326232809Sjmallett printf("%s: WARNING: unable to register helper hook\n", __func__); 327232809Sjmallett 328232809Sjmallett hashsize = TCBHASHSIZE; 329232809Sjmallett TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize); 330232809Sjmallett if (hashsize == 0) { 331232809Sjmallett /* 332232809Sjmallett * Auto tune the hash size based on maxsockets. 333232809Sjmallett * A perfect hash would have a 1:1 mapping 334232809Sjmallett * (hashsize = maxsockets) however it's been 335232809Sjmallett * suggested that O(2) average is better. 336232809Sjmallett */ 337232809Sjmallett hashsize = maketcp_hashsize(maxsockets / 4); 338232809Sjmallett /* 339232809Sjmallett * Our historical default is 512, 340232809Sjmallett * do not autotune lower than this. 341232809Sjmallett */ 342232809Sjmallett if (hashsize < 512) 343232809Sjmallett hashsize = 512; 344232809Sjmallett if (bootverbose && IS_DEFAULT_VNET(curvnet)) 345232809Sjmallett printf("%s: %s auto tuned to %d\n", __func__, 346232809Sjmallett tcbhash_tuneable, hashsize); 347232809Sjmallett } 348232809Sjmallett /* 349232809Sjmallett * We require a hashsize to be a power of two. 350232809Sjmallett * Previously if it was not a power of two we would just reset it 351232809Sjmallett * back to 512, which could be a nasty surprise if you did not notice 352232809Sjmallett * the error message. 353232809Sjmallett * Instead what we do is clip it to the closest power of two lower 354232809Sjmallett * than the specified hash value. 355232809Sjmallett */ 356232809Sjmallett if (!powerof2(hashsize)) { 357232809Sjmallett int oldhashsize = hashsize; 358232809Sjmallett 359232809Sjmallett hashsize = maketcp_hashsize(hashsize); 360232809Sjmallett /* prevent absurdly low value */ 361232809Sjmallett if (hashsize < 16) 362232809Sjmallett hashsize = 16; 363232809Sjmallett printf("%s: WARNING: TCB hash size not a power of 2, " 364232809Sjmallett "clipped from %d to %d.\n", __func__, oldhashsize, 365232809Sjmallett hashsize); 366232809Sjmallett } 367232809Sjmallett in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize, 368232809Sjmallett "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE, 369232809Sjmallett IPI_HASHFIELDS_4TUPLE); 370232809Sjmallett 371232809Sjmallett /* 372232809Sjmallett * These have to be type stable for the benefit of the timers. 373232809Sjmallett */ 374232809Sjmallett V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem), 375232809Sjmallett NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 376232809Sjmallett uma_zone_set_max(V_tcpcb_zone, maxsockets); 377232809Sjmallett uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached"); 378232809Sjmallett 379232809Sjmallett tcp_tw_init(); 380232809Sjmallett syncache_init(); 381232809Sjmallett tcp_hc_init(); 382232809Sjmallett 383232809Sjmallett TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack); 384232809Sjmallett V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole), 385232809Sjmallett NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 386232809Sjmallett 387232809Sjmallett /* Skip initialization of globals for non-default instances. */ 388232809Sjmallett if (!IS_DEFAULT_VNET(curvnet)) 389232809Sjmallett return; 390232809Sjmallett 391232809Sjmallett tcp_reass_global_init(); 392232809Sjmallett 393232809Sjmallett /* XXX virtualize those bellow? */ 394232809Sjmallett tcp_delacktime = TCPTV_DELACK; 395232809Sjmallett tcp_keepinit = TCPTV_KEEP_INIT; 396232809Sjmallett tcp_keepidle = TCPTV_KEEP_IDLE; 397232809Sjmallett tcp_keepintvl = TCPTV_KEEPINTVL; 398232809Sjmallett tcp_maxpersistidle = TCPTV_KEEP_IDLE; 399232809Sjmallett tcp_msl = TCPTV_MSL; 400232809Sjmallett tcp_rexmit_min = TCPTV_MIN; 401232809Sjmallett if (tcp_rexmit_min < 1) 402232809Sjmallett tcp_rexmit_min = 1; 403232809Sjmallett tcp_persmin = TCPTV_PERSMIN; 404232809Sjmallett tcp_persmax = TCPTV_PERSMAX; 405232809Sjmallett tcp_rexmit_slop = TCPTV_CPU_VAR; 406232809Sjmallett tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT; 407232809Sjmallett tcp_tcbhashsize = hashsize; 408232809Sjmallett 409232809Sjmallett TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream); 410232809Sjmallett if (tcp_soreceive_stream) { 411232809Sjmallett#ifdef INET 412232809Sjmallett tcp_usrreqs.pru_soreceive = soreceive_stream; 413232809Sjmallett#endif 414232809Sjmallett#ifdef INET6 415232809Sjmallett tcp6_usrreqs.pru_soreceive = soreceive_stream; 416232809Sjmallett#endif /* INET6 */ 417232809Sjmallett } 418232809Sjmallett 419232809Sjmallett#ifdef INET6 420232809Sjmallett#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr)) 421232809Sjmallett#else /* INET6 */ 422232809Sjmallett#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr)) 423232809Sjmallett#endif /* INET6 */ 424232809Sjmallett if (max_protohdr < TCP_MINPROTOHDR) 425232809Sjmallett max_protohdr = TCP_MINPROTOHDR; 426232809Sjmallett if (max_linkhdr + TCP_MINPROTOHDR > MHLEN) 427232809Sjmallett panic("tcp_init"); 428232809Sjmallett#undef TCP_MINPROTOHDR 429232809Sjmallett 430232809Sjmallett ISN_LOCK_INIT(); 431232809Sjmallett EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL, 432232809Sjmallett SHUTDOWN_PRI_DEFAULT); 433232809Sjmallett EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL, 434232809Sjmallett EVENTHANDLER_PRI_ANY); 435232809Sjmallett 436232809Sjmallett#ifdef TCP_RFC7413 437232809Sjmallett tcp_fastopen_init(); 438232809Sjmallett#endif 439232809Sjmallett} 440232809Sjmallett 441232809Sjmallett#ifdef VIMAGE 442232809Sjmallettvoid 443tcp_destroy(void) 444{ 445 446#ifdef TCP_RFC7413 447 tcp_fastopen_destroy(); 448#endif 449 tcp_hc_destroy(); 450 syncache_destroy(); 451 tcp_tw_destroy(); 452 in_pcbinfo_destroy(&V_tcbinfo); 453 uma_zdestroy(V_sack_hole_zone); 454 uma_zdestroy(V_tcpcb_zone); 455} 456#endif 457 458void 459tcp_fini(void *xtp) 460{ 461 462} 463 464/* 465 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb. 466 * tcp_template used to store this data in mbufs, but we now recopy it out 467 * of the tcpcb each time to conserve mbufs. 468 */ 469void 470tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr) 471{ 472 struct tcphdr *th = (struct tcphdr *)tcp_ptr; 473 474 INP_WLOCK_ASSERT(inp); 475 476#ifdef INET6 477 if ((inp->inp_vflag & INP_IPV6) != 0) { 478 struct ip6_hdr *ip6; 479 480 ip6 = (struct ip6_hdr *)ip_ptr; 481 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) | 482 (inp->inp_flow & IPV6_FLOWINFO_MASK); 483 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) | 484 (IPV6_VERSION & IPV6_VERSION_MASK); 485 ip6->ip6_nxt = IPPROTO_TCP; 486 ip6->ip6_plen = htons(sizeof(struct tcphdr)); 487 ip6->ip6_src = inp->in6p_laddr; 488 ip6->ip6_dst = inp->in6p_faddr; 489 } 490#endif /* INET6 */ 491#if defined(INET6) && defined(INET) 492 else 493#endif 494#ifdef INET 495 { 496 struct ip *ip; 497 498 ip = (struct ip *)ip_ptr; 499 ip->ip_v = IPVERSION; 500 ip->ip_hl = 5; 501 ip->ip_tos = inp->inp_ip_tos; 502 ip->ip_len = 0; 503 ip->ip_id = 0; 504 ip->ip_off = 0; 505 ip->ip_ttl = inp->inp_ip_ttl; 506 ip->ip_sum = 0; 507 ip->ip_p = IPPROTO_TCP; 508 ip->ip_src = inp->inp_laddr; 509 ip->ip_dst = inp->inp_faddr; 510 } 511#endif /* INET */ 512 th->th_sport = inp->inp_lport; 513 th->th_dport = inp->inp_fport; 514 th->th_seq = 0; 515 th->th_ack = 0; 516 th->th_x2 = 0; 517 th->th_off = 5; 518 th->th_flags = 0; 519 th->th_win = 0; 520 th->th_urp = 0; 521 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */ 522} 523 524/* 525 * Create template to be used to send tcp packets on a connection. 526 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only 527 * use for this function is in keepalives, which use tcp_respond. 528 */ 529struct tcptemp * 530tcpip_maketemplate(struct inpcb *inp) 531{ 532 struct tcptemp *t; 533 534 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT); 535 if (t == NULL) 536 return (NULL); 537 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t); 538 return (t); 539} 540 541/* 542 * Send a single message to the TCP at address specified by 543 * the given TCP/IP header. If m == NULL, then we make a copy 544 * of the tcpiphdr at ti and send directly to the addressed host. 545 * This is used to force keep alive messages out using the TCP 546 * template for a connection. If flags are given then we send 547 * a message back to the TCP which originated the * segment ti, 548 * and discard the mbuf containing it and any other attached mbufs. 549 * 550 * In any case the ack and sequence number of the transmitted 551 * segment are as specified by the parameters. 552 * 553 * NOTE: If m != NULL, then ti must point to *inside* the mbuf. 554 */ 555void 556tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m, 557 tcp_seq ack, tcp_seq seq, int flags) 558{ 559 int tlen; 560 int win = 0; 561 struct ip *ip; 562 struct tcphdr *nth; 563#ifdef INET6 564 struct ip6_hdr *ip6; 565 int isipv6; 566#endif /* INET6 */ 567 int ipflags = 0; 568 struct inpcb *inp; 569 570 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL")); 571 572#ifdef INET6 573 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4); 574 ip6 = ipgen; 575#endif /* INET6 */ 576 ip = ipgen; 577 578 if (tp != NULL) { 579 inp = tp->t_inpcb; 580 KASSERT(inp != NULL, ("tcp control block w/o inpcb")); 581 INP_WLOCK_ASSERT(inp); 582 } else 583 inp = NULL; 584 585 if (tp != NULL) { 586 if (!(flags & TH_RST)) { 587 win = sbspace(&inp->inp_socket->so_rcv); 588 if (win > (long)TCP_MAXWIN << tp->rcv_scale) 589 win = (long)TCP_MAXWIN << tp->rcv_scale; 590 } 591 } 592 if (m == NULL) { 593 m = m_gethdr(M_NOWAIT, MT_DATA); 594 if (m == NULL) 595 return; 596 tlen = 0; 597 m->m_data += max_linkhdr; 598#ifdef INET6 599 if (isipv6) { 600 bcopy((caddr_t)ip6, mtod(m, caddr_t), 601 sizeof(struct ip6_hdr)); 602 ip6 = mtod(m, struct ip6_hdr *); 603 nth = (struct tcphdr *)(ip6 + 1); 604 } else 605#endif /* INET6 */ 606 { 607 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip)); 608 ip = mtod(m, struct ip *); 609 nth = (struct tcphdr *)(ip + 1); 610 } 611 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr)); 612 flags = TH_ACK; 613 } else { 614 /* 615 * reuse the mbuf. 616 * XXX MRT We inherrit the FIB, which is lucky. 617 */ 618 m_freem(m->m_next); 619 m->m_next = NULL; 620 m->m_data = (caddr_t)ipgen; 621 /* m_len is set later */ 622 tlen = 0; 623#define xchg(a,b,type) { type t; t=a; a=b; b=t; } 624#ifdef INET6 625 if (isipv6) { 626 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr); 627 nth = (struct tcphdr *)(ip6 + 1); 628 } else 629#endif /* INET6 */ 630 { 631 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t); 632 nth = (struct tcphdr *)(ip + 1); 633 } 634 if (th != nth) { 635 /* 636 * this is usually a case when an extension header 637 * exists between the IPv6 header and the 638 * TCP header. 639 */ 640 nth->th_sport = th->th_sport; 641 nth->th_dport = th->th_dport; 642 } 643 xchg(nth->th_dport, nth->th_sport, uint16_t); 644#undef xchg 645 } 646#ifdef INET6 647 if (isipv6) { 648 ip6->ip6_flow = 0; 649 ip6->ip6_vfc = IPV6_VERSION; 650 ip6->ip6_nxt = IPPROTO_TCP; 651 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr); 652 ip6->ip6_plen = htons(tlen - sizeof(*ip6)); 653 } 654#endif 655#if defined(INET) && defined(INET6) 656 else 657#endif 658#ifdef INET 659 { 660 tlen += sizeof (struct tcpiphdr); 661 ip->ip_len = htons(tlen); 662 ip->ip_ttl = V_ip_defttl; 663 if (V_path_mtu_discovery) 664 ip->ip_off |= htons(IP_DF); 665 } 666#endif 667 m->m_len = tlen; 668 m->m_pkthdr.len = tlen; 669 m->m_pkthdr.rcvif = NULL; 670#ifdef MAC 671 if (inp != NULL) { 672 /* 673 * Packet is associated with a socket, so allow the 674 * label of the response to reflect the socket label. 675 */ 676 INP_WLOCK_ASSERT(inp); 677 mac_inpcb_create_mbuf(inp, m); 678 } else { 679 /* 680 * Packet is not associated with a socket, so possibly 681 * update the label in place. 682 */ 683 mac_netinet_tcp_reply(m); 684 } 685#endif 686 nth->th_seq = htonl(seq); 687 nth->th_ack = htonl(ack); 688 nth->th_x2 = 0; 689 nth->th_off = sizeof (struct tcphdr) >> 2; 690 nth->th_flags = flags; 691 if (tp != NULL) 692 nth->th_win = htons((u_short) (win >> tp->rcv_scale)); 693 else 694 nth->th_win = htons((u_short)win); 695 nth->th_urp = 0; 696 697 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); 698#ifdef INET6 699 if (isipv6) { 700 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; 701 nth->th_sum = in6_cksum_pseudo(ip6, 702 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0); 703 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb : 704 NULL, NULL); 705 } 706#endif /* INET6 */ 707#if defined(INET6) && defined(INET) 708 else 709#endif 710#ifdef INET 711 { 712 m->m_pkthdr.csum_flags = CSUM_TCP; 713 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, 714 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p))); 715 } 716#endif /* INET */ 717#ifdef TCPDEBUG 718 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG)) 719 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0); 720#endif 721 if (flags & TH_RST) 722 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *), 723 tp, nth); 724 725 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth); 726#ifdef INET6 727 if (isipv6) 728 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp); 729#endif /* INET6 */ 730#if defined(INET) && defined(INET6) 731 else 732#endif 733#ifdef INET 734 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp); 735#endif 736} 737 738/* 739 * Create a new TCP control block, making an 740 * empty reassembly queue and hooking it to the argument 741 * protocol control block. The `inp' parameter must have 742 * come from the zone allocator set up in tcp_init(). 743 */ 744struct tcpcb * 745tcp_newtcpcb(struct inpcb *inp) 746{ 747 struct tcpcb_mem *tm; 748 struct tcpcb *tp; 749#ifdef INET6 750 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 751#endif /* INET6 */ 752 753 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO); 754 if (tm == NULL) 755 return (NULL); 756 tp = &tm->tcb; 757 758 /* Initialise cc_var struct for this tcpcb. */ 759 tp->ccv = &tm->ccv; 760 tp->ccv->type = IPPROTO_TCP; 761 tp->ccv->ccvc.tcp = tp; 762 763 /* 764 * Use the current system default CC algorithm. 765 */ 766 CC_LIST_RLOCK(); 767 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!")); 768 CC_ALGO(tp) = CC_DEFAULT(); 769 CC_LIST_RUNLOCK(); 770 771 if (CC_ALGO(tp)->cb_init != NULL) 772 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) { 773 uma_zfree(V_tcpcb_zone, tm); 774 return (NULL); 775 } 776 777 tp->osd = &tm->osd; 778 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) { 779 uma_zfree(V_tcpcb_zone, tm); 780 return (NULL); 781 } 782 783#ifdef VIMAGE 784 tp->t_vnet = inp->inp_vnet; 785#endif 786 tp->t_timers = &tm->tt; 787 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */ 788 tp->t_maxseg = tp->t_maxopd = 789#ifdef INET6 790 isipv6 ? V_tcp_v6mssdflt : 791#endif /* INET6 */ 792 V_tcp_mssdflt; 793 794 /* Set up our timeouts. */ 795 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE); 796 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE); 797 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE); 798 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE); 799 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE); 800 801 if (V_tcp_do_rfc1323) 802 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP); 803 if (V_tcp_do_sack) 804 tp->t_flags |= TF_SACK_PERMIT; 805 TAILQ_INIT(&tp->snd_holes); 806 /* 807 * The tcpcb will hold a reference on its inpcb until tcp_discardcb() 808 * is called. 809 */ 810 in_pcbref(inp); /* Reference for tcpcb */ 811 tp->t_inpcb = inp; 812 813 /* 814 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no 815 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives 816 * reasonable initial retransmit time. 817 */ 818 tp->t_srtt = TCPTV_SRTTBASE; 819 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4; 820 tp->t_rttmin = tcp_rexmit_min; 821 tp->t_rxtcur = TCPTV_RTOBASE; 822 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT; 823 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT; 824 tp->t_rcvtime = ticks; 825 /* 826 * IPv4 TTL initialization is necessary for an IPv6 socket as well, 827 * because the socket may be bound to an IPv6 wildcard address, 828 * which may match an IPv4-mapped IPv6 address. 829 */ 830 inp->inp_ip_ttl = V_ip_defttl; 831 inp->inp_ppcb = tp; 832 return (tp); /* XXX */ 833} 834 835/* 836 * Switch the congestion control algorithm back to NewReno for any active 837 * control blocks using an algorithm which is about to go away. 838 * This ensures the CC framework can allow the unload to proceed without leaving 839 * any dangling pointers which would trigger a panic. 840 * Returning non-zero would inform the CC framework that something went wrong 841 * and it would be unsafe to allow the unload to proceed. However, there is no 842 * way for this to occur with this implementation so we always return zero. 843 */ 844int 845tcp_ccalgounload(struct cc_algo *unload_algo) 846{ 847 struct cc_algo *tmpalgo; 848 struct inpcb *inp; 849 struct tcpcb *tp; 850 VNET_ITERATOR_DECL(vnet_iter); 851 852 /* 853 * Check all active control blocks across all network stacks and change 854 * any that are using "unload_algo" back to NewReno. If "unload_algo" 855 * requires cleanup code to be run, call it. 856 */ 857 VNET_LIST_RLOCK(); 858 VNET_FOREACH(vnet_iter) { 859 CURVNET_SET(vnet_iter); 860 INP_INFO_RLOCK(&V_tcbinfo); 861 /* 862 * New connections already part way through being initialised 863 * with the CC algo we're removing will not race with this code 864 * because the INP_INFO_WLOCK is held during initialisation. We 865 * therefore don't enter the loop below until the connection 866 * list has stabilised. 867 */ 868 LIST_FOREACH(inp, &V_tcb, inp_list) { 869 INP_WLOCK(inp); 870 /* Important to skip tcptw structs. */ 871 if (!(inp->inp_flags & INP_TIMEWAIT) && 872 (tp = intotcpcb(inp)) != NULL) { 873 /* 874 * By holding INP_WLOCK here, we are assured 875 * that the connection is not currently 876 * executing inside the CC module's functions 877 * i.e. it is safe to make the switch back to 878 * NewReno. 879 */ 880 if (CC_ALGO(tp) == unload_algo) { 881 tmpalgo = CC_ALGO(tp); 882 /* NewReno does not require any init. */ 883 CC_ALGO(tp) = &newreno_cc_algo; 884 if (tmpalgo->cb_destroy != NULL) 885 tmpalgo->cb_destroy(tp->ccv); 886 } 887 } 888 INP_WUNLOCK(inp); 889 } 890 INP_INFO_RUNLOCK(&V_tcbinfo); 891 CURVNET_RESTORE(); 892 } 893 VNET_LIST_RUNLOCK(); 894 895 return (0); 896} 897 898/* 899 * Drop a TCP connection, reporting 900 * the specified error. If connection is synchronized, 901 * then send a RST to peer. 902 */ 903struct tcpcb * 904tcp_drop(struct tcpcb *tp, int errno) 905{ 906 struct socket *so = tp->t_inpcb->inp_socket; 907 908 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 909 INP_WLOCK_ASSERT(tp->t_inpcb); 910 911 if (TCPS_HAVERCVDSYN(tp->t_state)) { 912 tcp_state_change(tp, TCPS_CLOSED); 913 (void) tcp_output(tp); 914 TCPSTAT_INC(tcps_drops); 915 } else 916 TCPSTAT_INC(tcps_conndrops); 917 if (errno == ETIMEDOUT && tp->t_softerror) 918 errno = tp->t_softerror; 919 so->so_error = errno; 920 return (tcp_close(tp)); 921} 922 923void 924tcp_discardcb(struct tcpcb *tp) 925{ 926 struct inpcb *inp = tp->t_inpcb; 927 struct socket *so = inp->inp_socket; 928#ifdef INET6 929 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0; 930#endif /* INET6 */ 931 int released; 932 933 INP_WLOCK_ASSERT(inp); 934 935 /* 936 * Make sure that all of our timers are stopped before we delete the 937 * PCB. 938 * 939 * If stopping a timer fails, we schedule a discard function in same 940 * callout, and the last discard function called will take care of 941 * deleting the tcpcb. 942 */ 943 tcp_timer_stop(tp, TT_REXMT); 944 tcp_timer_stop(tp, TT_PERSIST); 945 tcp_timer_stop(tp, TT_KEEP); 946 tcp_timer_stop(tp, TT_2MSL); 947 tcp_timer_stop(tp, TT_DELACK); 948 949 /* 950 * If we got enough samples through the srtt filter, 951 * save the rtt and rttvar in the routing entry. 952 * 'Enough' is arbitrarily defined as 4 rtt samples. 953 * 4 samples is enough for the srtt filter to converge 954 * to within enough % of the correct value; fewer samples 955 * and we could save a bogus rtt. The danger is not high 956 * as tcp quickly recovers from everything. 957 * XXX: Works very well but needs some more statistics! 958 */ 959 if (tp->t_rttupdated >= 4) { 960 struct hc_metrics_lite metrics; 961 u_long ssthresh; 962 963 bzero(&metrics, sizeof(metrics)); 964 /* 965 * Update the ssthresh always when the conditions below 966 * are satisfied. This gives us better new start value 967 * for the congestion avoidance for new connections. 968 * ssthresh is only set if packet loss occured on a session. 969 * 970 * XXXRW: 'so' may be NULL here, and/or socket buffer may be 971 * being torn down. Ideally this code would not use 'so'. 972 */ 973 ssthresh = tp->snd_ssthresh; 974 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) { 975 /* 976 * convert the limit from user data bytes to 977 * packets then to packet data bytes. 978 */ 979 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg; 980 if (ssthresh < 2) 981 ssthresh = 2; 982 ssthresh *= (u_long)(tp->t_maxseg + 983#ifdef INET6 984 (isipv6 ? sizeof (struct ip6_hdr) + 985 sizeof (struct tcphdr) : 986#endif 987 sizeof (struct tcpiphdr) 988#ifdef INET6 989 ) 990#endif 991 ); 992 } else 993 ssthresh = 0; 994 metrics.rmx_ssthresh = ssthresh; 995 996 metrics.rmx_rtt = tp->t_srtt; 997 metrics.rmx_rttvar = tp->t_rttvar; 998 metrics.rmx_cwnd = tp->snd_cwnd; 999 metrics.rmx_sendpipe = 0; 1000 metrics.rmx_recvpipe = 0; 1001 1002 tcp_hc_update(&inp->inp_inc, &metrics); 1003 } 1004 1005 /* free the reassembly queue, if any */ 1006 tcp_reass_flush(tp); 1007 1008#ifdef TCP_OFFLOAD 1009 /* Disconnect offload device, if any. */ 1010 if (tp->t_flags & TF_TOE) 1011 tcp_offload_detach(tp); 1012#endif 1013 1014 tcp_free_sackholes(tp); 1015 1016 /* Allow the CC algorithm to clean up after itself. */ 1017 if (CC_ALGO(tp)->cb_destroy != NULL) 1018 CC_ALGO(tp)->cb_destroy(tp->ccv); 1019 1020 khelp_destroy_osd(tp->osd); 1021 1022 CC_ALGO(tp) = NULL; 1023 inp->inp_ppcb = NULL; 1024 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1025 /* We own the last reference on tcpcb, let's free it. */ 1026 tp->t_inpcb = NULL; 1027 uma_zfree(V_tcpcb_zone, tp); 1028 released = in_pcbrele_wlocked(inp); 1029 KASSERT(!released, ("%s: inp %p should not have been released " 1030 "here", __func__, inp)); 1031 } 1032} 1033 1034void 1035tcp_timer_2msl_discard(void *xtp) 1036{ 1037 1038 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL); 1039} 1040 1041void 1042tcp_timer_keep_discard(void *xtp) 1043{ 1044 1045 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP); 1046} 1047 1048void 1049tcp_timer_persist_discard(void *xtp) 1050{ 1051 1052 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST); 1053} 1054 1055void 1056tcp_timer_rexmt_discard(void *xtp) 1057{ 1058 1059 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT); 1060} 1061 1062void 1063tcp_timer_delack_discard(void *xtp) 1064{ 1065 1066 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK); 1067} 1068 1069void 1070tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type) 1071{ 1072 struct inpcb *inp; 1073 1074 CURVNET_SET(tp->t_vnet); 1075 INP_INFO_WLOCK(&V_tcbinfo); 1076 inp = tp->t_inpcb; 1077 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL", 1078 __func__, tp)); 1079 INP_WLOCK(inp); 1080 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0, 1081 ("%s: tcpcb has to be stopped here", __func__)); 1082 KASSERT((tp->t_timers->tt_flags & timer_type) != 0, 1083 ("%s: discard callout should be running", __func__)); 1084 tp->t_timers->tt_flags &= ~timer_type; 1085 if ((tp->t_timers->tt_flags & TT_MASK) == 0) { 1086 /* We own the last reference on this tcpcb, let's free it. */ 1087 tp->t_inpcb = NULL; 1088 uma_zfree(V_tcpcb_zone, tp); 1089 if (in_pcbrele_wlocked(inp)) { 1090 INP_INFO_WUNLOCK(&V_tcbinfo); 1091 CURVNET_RESTORE(); 1092 return; 1093 } 1094 } 1095 INP_WUNLOCK(inp); 1096 INP_INFO_WUNLOCK(&V_tcbinfo); 1097 CURVNET_RESTORE(); 1098} 1099 1100/* 1101 * Attempt to close a TCP control block, marking it as dropped, and freeing 1102 * the socket if we hold the only reference. 1103 */ 1104struct tcpcb * 1105tcp_close(struct tcpcb *tp) 1106{ 1107 struct inpcb *inp = tp->t_inpcb; 1108 struct socket *so; 1109 1110 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1111 INP_WLOCK_ASSERT(inp); 1112 1113#ifdef TCP_OFFLOAD 1114 if (tp->t_state == TCPS_LISTEN) 1115 tcp_offload_listen_stop(tp); 1116#endif 1117#ifdef TCP_RFC7413 1118 /* 1119 * This releases the TFO pending counter resource for TFO listen 1120 * sockets as well as passively-created TFO sockets that transition 1121 * from SYN_RECEIVED to CLOSED. 1122 */ 1123 if (tp->t_tfo_pending) { 1124 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 1125 tp->t_tfo_pending = NULL; 1126 } 1127#endif 1128 in_pcbdrop(inp); 1129 TCPSTAT_INC(tcps_closed); 1130 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL")); 1131 so = inp->inp_socket; 1132 soisdisconnected(so); 1133 if (inp->inp_flags & INP_SOCKREF) { 1134 KASSERT(so->so_state & SS_PROTOREF, 1135 ("tcp_close: !SS_PROTOREF")); 1136 inp->inp_flags &= ~INP_SOCKREF; 1137 INP_WUNLOCK(inp); 1138 ACCEPT_LOCK(); 1139 SOCK_LOCK(so); 1140 so->so_state &= ~SS_PROTOREF; 1141 sofree(so); 1142 return (NULL); 1143 } 1144 return (tp); 1145} 1146 1147void 1148tcp_drain(void) 1149{ 1150 VNET_ITERATOR_DECL(vnet_iter); 1151 1152 if (!do_tcpdrain) 1153 return; 1154 1155 VNET_LIST_RLOCK_NOSLEEP(); 1156 VNET_FOREACH(vnet_iter) { 1157 CURVNET_SET(vnet_iter); 1158 struct inpcb *inpb; 1159 struct tcpcb *tcpb; 1160 1161 /* 1162 * Walk the tcpbs, if existing, and flush the reassembly queue, 1163 * if there is one... 1164 * XXX: The "Net/3" implementation doesn't imply that the TCP 1165 * reassembly queue should be flushed, but in a situation 1166 * where we're really low on mbufs, this is potentially 1167 * useful. 1168 */ 1169 INP_INFO_RLOCK(&V_tcbinfo); 1170 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) { 1171 if (inpb->inp_flags & INP_TIMEWAIT) 1172 continue; 1173 INP_WLOCK(inpb); 1174 if ((tcpb = intotcpcb(inpb)) != NULL) { 1175 tcp_reass_flush(tcpb); 1176 tcp_clean_sackreport(tcpb); 1177 } 1178 INP_WUNLOCK(inpb); 1179 } 1180 INP_INFO_RUNLOCK(&V_tcbinfo); 1181 CURVNET_RESTORE(); 1182 } 1183 VNET_LIST_RUNLOCK_NOSLEEP(); 1184} 1185 1186/* 1187 * Notify a tcp user of an asynchronous error; 1188 * store error as soft error, but wake up user 1189 * (for now, won't do anything until can select for soft error). 1190 * 1191 * Do not wake up user since there currently is no mechanism for 1192 * reporting soft errors (yet - a kqueue filter may be added). 1193 */ 1194static struct inpcb * 1195tcp_notify(struct inpcb *inp, int error) 1196{ 1197 struct tcpcb *tp; 1198 1199 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1200 INP_WLOCK_ASSERT(inp); 1201 1202 if ((inp->inp_flags & INP_TIMEWAIT) || 1203 (inp->inp_flags & INP_DROPPED)) 1204 return (inp); 1205 1206 tp = intotcpcb(inp); 1207 KASSERT(tp != NULL, ("tcp_notify: tp == NULL")); 1208 1209 /* 1210 * Ignore some errors if we are hooked up. 1211 * If connection hasn't completed, has retransmitted several times, 1212 * and receives a second error, give up now. This is better 1213 * than waiting a long time to establish a connection that 1214 * can never complete. 1215 */ 1216 if (tp->t_state == TCPS_ESTABLISHED && 1217 (error == EHOSTUNREACH || error == ENETUNREACH || 1218 error == EHOSTDOWN)) { 1219 return (inp); 1220 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 && 1221 tp->t_softerror) { 1222 tp = tcp_drop(tp, error); 1223 if (tp != NULL) 1224 return (inp); 1225 else 1226 return (NULL); 1227 } else { 1228 tp->t_softerror = error; 1229 return (inp); 1230 } 1231#if 0 1232 wakeup( &so->so_timeo); 1233 sorwakeup(so); 1234 sowwakeup(so); 1235#endif 1236} 1237 1238static int 1239tcp_pcblist(SYSCTL_HANDLER_ARGS) 1240{ 1241 int error, i, m, n, pcb_count; 1242 struct inpcb *inp, **inp_list; 1243 inp_gen_t gencnt; 1244 struct xinpgen xig; 1245 1246 /* 1247 * The process of preparing the TCB list is too time-consuming and 1248 * resource-intensive to repeat twice on every request. 1249 */ 1250 if (req->oldptr == NULL) { 1251 n = V_tcbinfo.ipi_count + syncache_pcbcount(); 1252 n += imax(n / 8, 10); 1253 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb); 1254 return (0); 1255 } 1256 1257 if (req->newptr != NULL) 1258 return (EPERM); 1259 1260 /* 1261 * OK, now we're committed to doing something. 1262 */ 1263 INP_INFO_RLOCK(&V_tcbinfo); 1264 gencnt = V_tcbinfo.ipi_gencnt; 1265 n = V_tcbinfo.ipi_count; 1266 INP_INFO_RUNLOCK(&V_tcbinfo); 1267 1268 m = syncache_pcbcount(); 1269 1270 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig) 1271 + (n + m) * sizeof(struct xtcpcb)); 1272 if (error != 0) 1273 return (error); 1274 1275 xig.xig_len = sizeof xig; 1276 xig.xig_count = n + m; 1277 xig.xig_gen = gencnt; 1278 xig.xig_sogen = so_gencnt; 1279 error = SYSCTL_OUT(req, &xig, sizeof xig); 1280 if (error) 1281 return (error); 1282 1283 error = syncache_pcblist(req, m, &pcb_count); 1284 if (error) 1285 return (error); 1286 1287 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK); 1288 1289 INP_INFO_RLOCK(&V_tcbinfo); 1290 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0; 1291 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) { 1292 INP_WLOCK(inp); 1293 if (inp->inp_gencnt <= gencnt) { 1294 /* 1295 * XXX: This use of cr_cansee(), introduced with 1296 * TCP state changes, is not quite right, but for 1297 * now, better than nothing. 1298 */ 1299 if (inp->inp_flags & INP_TIMEWAIT) { 1300 if (intotw(inp) != NULL) 1301 error = cr_cansee(req->td->td_ucred, 1302 intotw(inp)->tw_cred); 1303 else 1304 error = EINVAL; /* Skip this inp. */ 1305 } else 1306 error = cr_canseeinpcb(req->td->td_ucred, inp); 1307 if (error == 0) { 1308 in_pcbref(inp); 1309 inp_list[i++] = inp; 1310 } 1311 } 1312 INP_WUNLOCK(inp); 1313 } 1314 INP_INFO_RUNLOCK(&V_tcbinfo); 1315 n = i; 1316 1317 error = 0; 1318 for (i = 0; i < n; i++) { 1319 inp = inp_list[i]; 1320 INP_RLOCK(inp); 1321 if (inp->inp_gencnt <= gencnt) { 1322 struct xtcpcb xt; 1323 void *inp_ppcb; 1324 1325 bzero(&xt, sizeof(xt)); 1326 xt.xt_len = sizeof xt; 1327 /* XXX should avoid extra copy */ 1328 bcopy(inp, &xt.xt_inp, sizeof *inp); 1329 inp_ppcb = inp->inp_ppcb; 1330 if (inp_ppcb == NULL) 1331 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1332 else if (inp->inp_flags & INP_TIMEWAIT) { 1333 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp); 1334 xt.xt_tp.t_state = TCPS_TIME_WAIT; 1335 } else { 1336 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp); 1337 if (xt.xt_tp.t_timers) 1338 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer); 1339 } 1340 if (inp->inp_socket != NULL) 1341 sotoxsocket(inp->inp_socket, &xt.xt_socket); 1342 else { 1343 bzero(&xt.xt_socket, sizeof xt.xt_socket); 1344 xt.xt_socket.xso_protocol = IPPROTO_TCP; 1345 } 1346 xt.xt_inp.inp_gencnt = inp->inp_gencnt; 1347 INP_RUNLOCK(inp); 1348 error = SYSCTL_OUT(req, &xt, sizeof xt); 1349 } else 1350 INP_RUNLOCK(inp); 1351 } 1352 INP_INFO_WLOCK(&V_tcbinfo); 1353 for (i = 0; i < n; i++) { 1354 inp = inp_list[i]; 1355 INP_RLOCK(inp); 1356 if (!in_pcbrele_rlocked(inp)) 1357 INP_RUNLOCK(inp); 1358 } 1359 INP_INFO_WUNLOCK(&V_tcbinfo); 1360 1361 if (!error) { 1362 /* 1363 * Give the user an updated idea of our state. 1364 * If the generation differs from what we told 1365 * her before, she knows that something happened 1366 * while we were processing this request, and it 1367 * might be necessary to retry. 1368 */ 1369 INP_INFO_RLOCK(&V_tcbinfo); 1370 xig.xig_gen = V_tcbinfo.ipi_gencnt; 1371 xig.xig_sogen = so_gencnt; 1372 xig.xig_count = V_tcbinfo.ipi_count + pcb_count; 1373 INP_INFO_RUNLOCK(&V_tcbinfo); 1374 error = SYSCTL_OUT(req, &xig, sizeof xig); 1375 } 1376 free(inp_list, M_TEMP); 1377 return (error); 1378} 1379 1380SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, 1381 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0, 1382 tcp_pcblist, "S,xtcpcb", "List of active TCP connections"); 1383 1384#ifdef INET 1385static int 1386tcp_getcred(SYSCTL_HANDLER_ARGS) 1387{ 1388 struct xucred xuc; 1389 struct sockaddr_in addrs[2]; 1390 struct inpcb *inp; 1391 int error; 1392 1393 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1394 if (error) 1395 return (error); 1396 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1397 if (error) 1398 return (error); 1399 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port, 1400 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL); 1401 if (inp != NULL) { 1402 if (inp->inp_socket == NULL) 1403 error = ENOENT; 1404 if (error == 0) 1405 error = cr_canseeinpcb(req->td->td_ucred, inp); 1406 if (error == 0) 1407 cru2x(inp->inp_cred, &xuc); 1408 INP_RUNLOCK(inp); 1409 } else 1410 error = ENOENT; 1411 if (error == 0) 1412 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1413 return (error); 1414} 1415 1416SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, 1417 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1418 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection"); 1419#endif /* INET */ 1420 1421#ifdef INET6 1422static int 1423tcp6_getcred(SYSCTL_HANDLER_ARGS) 1424{ 1425 struct xucred xuc; 1426 struct sockaddr_in6 addrs[2]; 1427 struct inpcb *inp; 1428 int error; 1429#ifdef INET 1430 int mapped = 0; 1431#endif 1432 1433 error = priv_check(req->td, PRIV_NETINET_GETCRED); 1434 if (error) 1435 return (error); 1436 error = SYSCTL_IN(req, addrs, sizeof(addrs)); 1437 if (error) 1438 return (error); 1439 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 || 1440 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) { 1441 return (error); 1442 } 1443 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) { 1444#ifdef INET 1445 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr)) 1446 mapped = 1; 1447 else 1448#endif 1449 return (EINVAL); 1450 } 1451 1452#ifdef INET 1453 if (mapped == 1) 1454 inp = in_pcblookup(&V_tcbinfo, 1455 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12], 1456 addrs[1].sin6_port, 1457 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12], 1458 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL); 1459 else 1460#endif 1461 inp = in6_pcblookup(&V_tcbinfo, 1462 &addrs[1].sin6_addr, addrs[1].sin6_port, 1463 &addrs[0].sin6_addr, addrs[0].sin6_port, 1464 INPLOOKUP_RLOCKPCB, NULL); 1465 if (inp != NULL) { 1466 if (inp->inp_socket == NULL) 1467 error = ENOENT; 1468 if (error == 0) 1469 error = cr_canseeinpcb(req->td->td_ucred, inp); 1470 if (error == 0) 1471 cru2x(inp->inp_cred, &xuc); 1472 INP_RUNLOCK(inp); 1473 } else 1474 error = ENOENT; 1475 if (error == 0) 1476 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred)); 1477 return (error); 1478} 1479 1480SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, 1481 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0, 1482 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection"); 1483#endif /* INET6 */ 1484 1485 1486#ifdef INET 1487void 1488tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip) 1489{ 1490 struct ip *ip = vip; 1491 struct tcphdr *th; 1492 struct in_addr faddr; 1493 struct inpcb *inp; 1494 struct tcpcb *tp; 1495 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1496 struct icmp *icp; 1497 struct in_conninfo inc; 1498 tcp_seq icmp_tcp_seq; 1499 int mtu; 1500 1501 faddr = ((struct sockaddr_in *)sa)->sin_addr; 1502 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) 1503 return; 1504 1505 if (cmd == PRC_MSGSIZE) 1506 notify = tcp_mtudisc_notify; 1507 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB || 1508 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip) 1509 notify = tcp_drop_syn_sent; 1510 /* 1511 * Redirects don't need to be handled up here. 1512 */ 1513 else if (PRC_IS_REDIRECT(cmd)) 1514 return; 1515 /* 1516 * Source quench is depreciated. 1517 */ 1518 else if (cmd == PRC_QUENCH) 1519 return; 1520 /* 1521 * Hostdead is ugly because it goes linearly through all PCBs. 1522 * XXX: We never get this from ICMP, otherwise it makes an 1523 * excellent DoS attack on machines with many connections. 1524 */ 1525 else if (cmd == PRC_HOSTDEAD) 1526 ip = NULL; 1527 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) 1528 return; 1529 if (ip != NULL) { 1530 icp = (struct icmp *)((caddr_t)ip 1531 - offsetof(struct icmp, icmp_ip)); 1532 th = (struct tcphdr *)((caddr_t)ip 1533 + (ip->ip_hl << 2)); 1534 INP_INFO_WLOCK(&V_tcbinfo); 1535 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, 1536 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL); 1537 if (inp != NULL) { 1538 if (!(inp->inp_flags & INP_TIMEWAIT) && 1539 !(inp->inp_flags & INP_DROPPED) && 1540 !(inp->inp_socket == NULL)) { 1541 icmp_tcp_seq = htonl(th->th_seq); 1542 tp = intotcpcb(inp); 1543 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) && 1544 SEQ_LT(icmp_tcp_seq, tp->snd_max)) { 1545 if (cmd == PRC_MSGSIZE) { 1546 /* 1547 * MTU discovery: 1548 * If we got a needfrag set the MTU 1549 * in the route to the suggested new 1550 * value (if given) and then notify. 1551 */ 1552 bzero(&inc, sizeof(inc)); 1553 inc.inc_faddr = faddr; 1554 inc.inc_fibnum = 1555 inp->inp_inc.inc_fibnum; 1556 1557 mtu = ntohs(icp->icmp_nextmtu); 1558 /* 1559 * If no alternative MTU was 1560 * proposed, try the next smaller 1561 * one. 1562 */ 1563 if (!mtu) 1564 mtu = ip_next_mtu( 1565 ntohs(ip->ip_len), 1); 1566 if (mtu < V_tcp_minmss 1567 + sizeof(struct tcpiphdr)) 1568 mtu = V_tcp_minmss 1569 + sizeof(struct tcpiphdr); 1570 /* 1571 * Only cache the MTU if it 1572 * is smaller than the interface 1573 * or route MTU. tcp_mtudisc() 1574 * will do right thing by itself. 1575 */ 1576 if (mtu <= tcp_maxmtu(&inc, NULL)) 1577 tcp_hc_updatemtu(&inc, mtu); 1578 tcp_mtudisc(inp, mtu); 1579 } else 1580 inp = (*notify)(inp, 1581 inetctlerrmap[cmd]); 1582 } 1583 } 1584 if (inp != NULL) 1585 INP_WUNLOCK(inp); 1586 } else { 1587 bzero(&inc, sizeof(inc)); 1588 inc.inc_fport = th->th_dport; 1589 inc.inc_lport = th->th_sport; 1590 inc.inc_faddr = faddr; 1591 inc.inc_laddr = ip->ip_src; 1592 syncache_unreach(&inc, th); 1593 } 1594 INP_INFO_WUNLOCK(&V_tcbinfo); 1595 } else 1596 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify); 1597} 1598#endif /* INET */ 1599 1600#ifdef INET6 1601void 1602tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d) 1603{ 1604 struct tcphdr th; 1605 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify; 1606 struct ip6_hdr *ip6; 1607 struct mbuf *m; 1608 struct ip6ctlparam *ip6cp = NULL; 1609 const struct sockaddr_in6 *sa6_src = NULL; 1610 int off; 1611 struct tcp_portonly { 1612 u_int16_t th_sport; 1613 u_int16_t th_dport; 1614 } *thp; 1615 1616 if (sa->sa_family != AF_INET6 || 1617 sa->sa_len != sizeof(struct sockaddr_in6)) 1618 return; 1619 1620 if (cmd == PRC_MSGSIZE) 1621 notify = tcp_mtudisc_notify; 1622 else if (!PRC_IS_REDIRECT(cmd) && 1623 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) 1624 return; 1625 /* Source quench is depreciated. */ 1626 else if (cmd == PRC_QUENCH) 1627 return; 1628 1629 /* if the parameter is from icmp6, decode it. */ 1630 if (d != NULL) { 1631 ip6cp = (struct ip6ctlparam *)d; 1632 m = ip6cp->ip6c_m; 1633 ip6 = ip6cp->ip6c_ip6; 1634 off = ip6cp->ip6c_off; 1635 sa6_src = ip6cp->ip6c_src; 1636 } else { 1637 m = NULL; 1638 ip6 = NULL; 1639 off = 0; /* fool gcc */ 1640 sa6_src = &sa6_any; 1641 } 1642 1643 if (ip6 != NULL) { 1644 struct in_conninfo inc; 1645 /* 1646 * XXX: We assume that when IPV6 is non NULL, 1647 * M and OFF are valid. 1648 */ 1649 1650 /* check if we can safely examine src and dst ports */ 1651 if (m->m_pkthdr.len < off + sizeof(*thp)) 1652 return; 1653 1654 bzero(&th, sizeof(th)); 1655 m_copydata(m, off, sizeof(*thp), (caddr_t)&th); 1656 1657 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport, 1658 (struct sockaddr *)ip6cp->ip6c_src, 1659 th.th_sport, cmd, NULL, notify); 1660 1661 bzero(&inc, sizeof(inc)); 1662 inc.inc_fport = th.th_dport; 1663 inc.inc_lport = th.th_sport; 1664 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr; 1665 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr; 1666 inc.inc_flags |= INC_ISIPV6; 1667 INP_INFO_WLOCK(&V_tcbinfo); 1668 syncache_unreach(&inc, &th); 1669 INP_INFO_WUNLOCK(&V_tcbinfo); 1670 } else 1671 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src, 1672 0, cmd, NULL, notify); 1673} 1674#endif /* INET6 */ 1675 1676 1677/* 1678 * Following is where TCP initial sequence number generation occurs. 1679 * 1680 * There are two places where we must use initial sequence numbers: 1681 * 1. In SYN-ACK packets. 1682 * 2. In SYN packets. 1683 * 1684 * All ISNs for SYN-ACK packets are generated by the syncache. See 1685 * tcp_syncache.c for details. 1686 * 1687 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling 1688 * depends on this property. In addition, these ISNs should be 1689 * unguessable so as to prevent connection hijacking. To satisfy 1690 * the requirements of this situation, the algorithm outlined in 1691 * RFC 1948 is used, with only small modifications. 1692 * 1693 * Implementation details: 1694 * 1695 * Time is based off the system timer, and is corrected so that it 1696 * increases by one megabyte per second. This allows for proper 1697 * recycling on high speed LANs while still leaving over an hour 1698 * before rollover. 1699 * 1700 * As reading the *exact* system time is too expensive to be done 1701 * whenever setting up a TCP connection, we increment the time 1702 * offset in two ways. First, a small random positive increment 1703 * is added to isn_offset for each connection that is set up. 1704 * Second, the function tcp_isn_tick fires once per clock tick 1705 * and increments isn_offset as necessary so that sequence numbers 1706 * are incremented at approximately ISN_BYTES_PER_SECOND. The 1707 * random positive increments serve only to ensure that the same 1708 * exact sequence number is never sent out twice (as could otherwise 1709 * happen when a port is recycled in less than the system tick 1710 * interval.) 1711 * 1712 * net.inet.tcp.isn_reseed_interval controls the number of seconds 1713 * between seeding of isn_secret. This is normally set to zero, 1714 * as reseeding should not be necessary. 1715 * 1716 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset, 1717 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In 1718 * general, this means holding an exclusive (write) lock. 1719 */ 1720 1721#define ISN_BYTES_PER_SECOND 1048576 1722#define ISN_STATIC_INCREMENT 4096 1723#define ISN_RANDOM_INCREMENT (4096 - 1) 1724 1725static VNET_DEFINE(u_char, isn_secret[32]); 1726static VNET_DEFINE(int, isn_last); 1727static VNET_DEFINE(int, isn_last_reseed); 1728static VNET_DEFINE(u_int32_t, isn_offset); 1729static VNET_DEFINE(u_int32_t, isn_offset_old); 1730 1731#define V_isn_secret VNET(isn_secret) 1732#define V_isn_last VNET(isn_last) 1733#define V_isn_last_reseed VNET(isn_last_reseed) 1734#define V_isn_offset VNET(isn_offset) 1735#define V_isn_offset_old VNET(isn_offset_old) 1736 1737tcp_seq 1738tcp_new_isn(struct tcpcb *tp) 1739{ 1740 MD5_CTX isn_ctx; 1741 u_int32_t md5_buffer[4]; 1742 tcp_seq new_isn; 1743 u_int32_t projected_offset; 1744 1745 INP_WLOCK_ASSERT(tp->t_inpcb); 1746 1747 ISN_LOCK(); 1748 /* Seed if this is the first use, reseed if requested. */ 1749 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) && 1750 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz) 1751 < (u_int)ticks))) { 1752 read_random(&V_isn_secret, sizeof(V_isn_secret)); 1753 V_isn_last_reseed = ticks; 1754 } 1755 1756 /* Compute the md5 hash and return the ISN. */ 1757 MD5Init(&isn_ctx); 1758 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short)); 1759 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short)); 1760#ifdef INET6 1761 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) { 1762 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr, 1763 sizeof(struct in6_addr)); 1764 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr, 1765 sizeof(struct in6_addr)); 1766 } else 1767#endif 1768 { 1769 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr, 1770 sizeof(struct in_addr)); 1771 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr, 1772 sizeof(struct in_addr)); 1773 } 1774 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret)); 1775 MD5Final((u_char *) &md5_buffer, &isn_ctx); 1776 new_isn = (tcp_seq) md5_buffer[0]; 1777 V_isn_offset += ISN_STATIC_INCREMENT + 1778 (arc4random() & ISN_RANDOM_INCREMENT); 1779 if (ticks != V_isn_last) { 1780 projected_offset = V_isn_offset_old + 1781 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last); 1782 if (SEQ_GT(projected_offset, V_isn_offset)) 1783 V_isn_offset = projected_offset; 1784 V_isn_offset_old = V_isn_offset; 1785 V_isn_last = ticks; 1786 } 1787 new_isn += V_isn_offset; 1788 ISN_UNLOCK(); 1789 return (new_isn); 1790} 1791 1792/* 1793 * When a specific ICMP unreachable message is received and the 1794 * connection state is SYN-SENT, drop the connection. This behavior 1795 * is controlled by the icmp_may_rst sysctl. 1796 */ 1797struct inpcb * 1798tcp_drop_syn_sent(struct inpcb *inp, int errno) 1799{ 1800 struct tcpcb *tp; 1801 1802 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1803 INP_WLOCK_ASSERT(inp); 1804 1805 if ((inp->inp_flags & INP_TIMEWAIT) || 1806 (inp->inp_flags & INP_DROPPED)) 1807 return (inp); 1808 1809 tp = intotcpcb(inp); 1810 if (tp->t_state != TCPS_SYN_SENT) 1811 return (inp); 1812 1813 tp = tcp_drop(tp, errno); 1814 if (tp != NULL) 1815 return (inp); 1816 else 1817 return (NULL); 1818} 1819 1820/* 1821 * When `need fragmentation' ICMP is received, update our idea of the MSS 1822 * based on the new value. Also nudge TCP to send something, since we 1823 * know the packet we just sent was dropped. 1824 * This duplicates some code in the tcp_mss() function in tcp_input.c. 1825 */ 1826static struct inpcb * 1827tcp_mtudisc_notify(struct inpcb *inp, int error) 1828{ 1829 1830 return (tcp_mtudisc(inp, -1)); 1831} 1832 1833struct inpcb * 1834tcp_mtudisc(struct inpcb *inp, int mtuoffer) 1835{ 1836 struct tcpcb *tp; 1837 struct socket *so; 1838 1839 INP_WLOCK_ASSERT(inp); 1840 if ((inp->inp_flags & INP_TIMEWAIT) || 1841 (inp->inp_flags & INP_DROPPED)) 1842 return (inp); 1843 1844 tp = intotcpcb(inp); 1845 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL")); 1846 1847 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL); 1848 1849 so = inp->inp_socket; 1850 SOCKBUF_LOCK(&so->so_snd); 1851 /* If the mss is larger than the socket buffer, decrease the mss. */ 1852 if (so->so_snd.sb_hiwat < tp->t_maxseg) 1853 tp->t_maxseg = so->so_snd.sb_hiwat; 1854 SOCKBUF_UNLOCK(&so->so_snd); 1855 1856 TCPSTAT_INC(tcps_mturesent); 1857 tp->t_rtttime = 0; 1858 tp->snd_nxt = tp->snd_una; 1859 tcp_free_sackholes(tp); 1860 tp->snd_recover = tp->snd_max; 1861 if (tp->t_flags & TF_SACK_PERMIT) 1862 EXIT_FASTRECOVERY(tp->t_flags); 1863 tcp_output(tp); 1864 return (inp); 1865} 1866 1867#ifdef INET 1868/* 1869 * Look-up the routing entry to the peer of this inpcb. If no route 1870 * is found and it cannot be allocated, then return 0. This routine 1871 * is called by TCP routines that access the rmx structure and by 1872 * tcp_mss_update to get the peer/interface MTU. 1873 */ 1874u_long 1875tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap) 1876{ 1877 struct route sro; 1878 struct sockaddr_in *dst; 1879 struct ifnet *ifp; 1880 u_long maxmtu = 0; 1881 1882 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer")); 1883 1884 bzero(&sro, sizeof(sro)); 1885 if (inc->inc_faddr.s_addr != INADDR_ANY) { 1886 dst = (struct sockaddr_in *)&sro.ro_dst; 1887 dst->sin_family = AF_INET; 1888 dst->sin_len = sizeof(*dst); 1889 dst->sin_addr = inc->inc_faddr; 1890 in_rtalloc_ign(&sro, 0, inc->inc_fibnum); 1891 } 1892 if (sro.ro_rt != NULL) { 1893 ifp = sro.ro_rt->rt_ifp; 1894 if (sro.ro_rt->rt_mtu == 0) 1895 maxmtu = ifp->if_mtu; 1896 else 1897 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu); 1898 1899 /* Report additional interface capabilities. */ 1900 if (cap != NULL) { 1901 if (ifp->if_capenable & IFCAP_TSO4 && 1902 ifp->if_hwassist & CSUM_TSO) { 1903 cap->ifcap |= CSUM_TSO; 1904 cap->tsomax = ifp->if_hw_tsomax; 1905 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 1906 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 1907 } 1908 } 1909 RTFREE(sro.ro_rt); 1910 } 1911 return (maxmtu); 1912} 1913#endif /* INET */ 1914 1915#ifdef INET6 1916u_long 1917tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap) 1918{ 1919 struct route_in6 sro6; 1920 struct ifnet *ifp; 1921 u_long maxmtu = 0; 1922 1923 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer")); 1924 1925 bzero(&sro6, sizeof(sro6)); 1926 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) { 1927 sro6.ro_dst.sin6_family = AF_INET6; 1928 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6); 1929 sro6.ro_dst.sin6_addr = inc->inc6_faddr; 1930 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum); 1931 } 1932 if (sro6.ro_rt != NULL) { 1933 ifp = sro6.ro_rt->rt_ifp; 1934 if (sro6.ro_rt->rt_mtu == 0) 1935 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp); 1936 else 1937 maxmtu = min(sro6.ro_rt->rt_mtu, 1938 IN6_LINKMTU(sro6.ro_rt->rt_ifp)); 1939 1940 /* Report additional interface capabilities. */ 1941 if (cap != NULL) { 1942 if (ifp->if_capenable & IFCAP_TSO6 && 1943 ifp->if_hwassist & CSUM_TSO) { 1944 cap->ifcap |= CSUM_TSO; 1945 cap->tsomax = ifp->if_hw_tsomax; 1946 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount; 1947 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize; 1948 } 1949 } 1950 RTFREE(sro6.ro_rt); 1951 } 1952 1953 return (maxmtu); 1954} 1955#endif /* INET6 */ 1956 1957#ifdef IPSEC 1958/* compute ESP/AH header size for TCP, including outer IP header. */ 1959size_t 1960ipsec_hdrsiz_tcp(struct tcpcb *tp) 1961{ 1962 struct inpcb *inp; 1963 struct mbuf *m; 1964 size_t hdrsiz; 1965 struct ip *ip; 1966#ifdef INET6 1967 struct ip6_hdr *ip6; 1968#endif 1969 struct tcphdr *th; 1970 1971 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) || 1972 (!key_havesp(IPSEC_DIR_OUTBOUND))) 1973 return (0); 1974 m = m_gethdr(M_NOWAIT, MT_DATA); 1975 if (!m) 1976 return (0); 1977 1978#ifdef INET6 1979 if ((inp->inp_vflag & INP_IPV6) != 0) { 1980 ip6 = mtod(m, struct ip6_hdr *); 1981 th = (struct tcphdr *)(ip6 + 1); 1982 m->m_pkthdr.len = m->m_len = 1983 sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 1984 tcpip_fillheaders(inp, ip6, th); 1985 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1986 } else 1987#endif /* INET6 */ 1988 { 1989 ip = mtod(m, struct ip *); 1990 th = (struct tcphdr *)(ip + 1); 1991 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr); 1992 tcpip_fillheaders(inp, ip, th); 1993 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp); 1994 } 1995 1996 m_free(m); 1997 return (hdrsiz); 1998} 1999#endif /* IPSEC */ 2000 2001#ifdef TCP_SIGNATURE 2002/* 2003 * Callback function invoked by m_apply() to digest TCP segment data 2004 * contained within an mbuf chain. 2005 */ 2006static int 2007tcp_signature_apply(void *fstate, void *data, u_int len) 2008{ 2009 2010 MD5Update(fstate, (u_char *)data, len); 2011 return (0); 2012} 2013 2014/* 2015 * Compute TCP-MD5 hash of a TCP segment. (RFC2385) 2016 * 2017 * Parameters: 2018 * m pointer to head of mbuf chain 2019 * _unused 2020 * len length of TCP segment data, excluding options 2021 * optlen length of TCP segment options 2022 * buf pointer to storage for computed MD5 digest 2023 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2024 * 2025 * We do this over ip, tcphdr, segment data, and the key in the SADB. 2026 * When called from tcp_input(), we can be sure that th_sum has been 2027 * zeroed out and verified already. 2028 * 2029 * Return 0 if successful, otherwise return -1. 2030 * 2031 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a 2032 * search with the destination IP address, and a 'magic SPI' to be 2033 * determined by the application. This is hardcoded elsewhere to 1179 2034 * right now. Another branch of this code exists which uses the SPD to 2035 * specify per-application flows but it is unstable. 2036 */ 2037int 2038tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen, 2039 u_char *buf, u_int direction) 2040{ 2041 union sockaddr_union dst; 2042#ifdef INET 2043 struct ippseudo ippseudo; 2044#endif 2045 MD5_CTX ctx; 2046 int doff; 2047 struct ip *ip; 2048#ifdef INET 2049 struct ipovly *ipovly; 2050#endif 2051 struct secasvar *sav; 2052 struct tcphdr *th; 2053#ifdef INET6 2054 struct ip6_hdr *ip6; 2055 struct in6_addr in6; 2056 char ip6buf[INET6_ADDRSTRLEN]; 2057 uint32_t plen; 2058 uint16_t nhdr; 2059#endif 2060 u_short savecsum; 2061 2062 KASSERT(m != NULL, ("NULL mbuf chain")); 2063 KASSERT(buf != NULL, ("NULL signature pointer")); 2064 2065 /* Extract the destination from the IP header in the mbuf. */ 2066 bzero(&dst, sizeof(union sockaddr_union)); 2067 ip = mtod(m, struct ip *); 2068#ifdef INET6 2069 ip6 = NULL; /* Make the compiler happy. */ 2070#endif 2071 switch (ip->ip_v) { 2072#ifdef INET 2073 case IPVERSION: 2074 dst.sa.sa_len = sizeof(struct sockaddr_in); 2075 dst.sa.sa_family = AF_INET; 2076 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ? 2077 ip->ip_src : ip->ip_dst; 2078 break; 2079#endif 2080#ifdef INET6 2081 case (IPV6_VERSION >> 4): 2082 ip6 = mtod(m, struct ip6_hdr *); 2083 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2084 dst.sa.sa_family = AF_INET6; 2085 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ? 2086 ip6->ip6_src : ip6->ip6_dst; 2087 break; 2088#endif 2089 default: 2090 return (EINVAL); 2091 /* NOTREACHED */ 2092 break; 2093 } 2094 2095 /* Look up an SADB entry which matches the address of the peer. */ 2096 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI)); 2097 if (sav == NULL) { 2098 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__, 2099 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) : 2100#ifdef INET6 2101 (ip->ip_v == (IPV6_VERSION >> 4)) ? 2102 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) : 2103#endif 2104 "(unsupported)")); 2105 return (EINVAL); 2106 } 2107 2108 MD5Init(&ctx); 2109 /* 2110 * Step 1: Update MD5 hash with IP(v6) pseudo-header. 2111 * 2112 * XXX The ippseudo header MUST be digested in network byte order, 2113 * or else we'll fail the regression test. Assume all fields we've 2114 * been doing arithmetic on have been in host byte order. 2115 * XXX One cannot depend on ipovly->ih_len here. When called from 2116 * tcp_output(), the underlying ip_len member has not yet been set. 2117 */ 2118 switch (ip->ip_v) { 2119#ifdef INET 2120 case IPVERSION: 2121 ipovly = (struct ipovly *)ip; 2122 ippseudo.ippseudo_src = ipovly->ih_src; 2123 ippseudo.ippseudo_dst = ipovly->ih_dst; 2124 ippseudo.ippseudo_pad = 0; 2125 ippseudo.ippseudo_p = IPPROTO_TCP; 2126 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + 2127 optlen); 2128 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo)); 2129 2130 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip)); 2131 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen; 2132 break; 2133#endif 2134#ifdef INET6 2135 /* 2136 * RFC 2385, 2.0 Proposal 2137 * For IPv6, the pseudo-header is as described in RFC 2460, namely the 2138 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero- 2139 * extended next header value (to form 32 bits), and 32-bit segment 2140 * length. 2141 * Note: Upper-Layer Packet Length comes before Next Header. 2142 */ 2143 case (IPV6_VERSION >> 4): 2144 in6 = ip6->ip6_src; 2145 in6_clearscope(&in6); 2146 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2147 in6 = ip6->ip6_dst; 2148 in6_clearscope(&in6); 2149 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr)); 2150 plen = htonl(len + sizeof(struct tcphdr) + optlen); 2151 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t)); 2152 nhdr = 0; 2153 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2154 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2155 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2156 nhdr = IPPROTO_TCP; 2157 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t)); 2158 2159 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr)); 2160 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen; 2161 break; 2162#endif 2163 default: 2164 return (EINVAL); 2165 /* NOTREACHED */ 2166 break; 2167 } 2168 2169 2170 /* 2171 * Step 2: Update MD5 hash with TCP header, excluding options. 2172 * The TCP checksum must be set to zero. 2173 */ 2174 savecsum = th->th_sum; 2175 th->th_sum = 0; 2176 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 2177 th->th_sum = savecsum; 2178 2179 /* 2180 * Step 3: Update MD5 hash with TCP segment data. 2181 * Use m_apply() to avoid an early m_pullup(). 2182 */ 2183 if (len > 0) 2184 m_apply(m, doff, len, tcp_signature_apply, &ctx); 2185 2186 /* 2187 * Step 4: Update MD5 hash with shared secret. 2188 */ 2189 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth)); 2190 MD5Final(buf, &ctx); 2191 2192 key_sa_recordxfer(sav, m); 2193 KEY_FREESAV(&sav); 2194 return (0); 2195} 2196 2197/* 2198 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385) 2199 * 2200 * Parameters: 2201 * m pointer to head of mbuf chain 2202 * len length of TCP segment data, excluding options 2203 * optlen length of TCP segment options 2204 * buf pointer to storage for computed MD5 digest 2205 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND) 2206 * 2207 * Return 1 if successful, otherwise return 0. 2208 */ 2209int 2210tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen, 2211 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 2212{ 2213 char tmpdigest[TCP_SIGLEN]; 2214 2215 if (tcp_sig_checksigs == 0) 2216 return (1); 2217 if ((tcpbflag & TF_SIGNATURE) == 0) { 2218 if ((to->to_flags & TOF_SIGNATURE) != 0) { 2219 2220 /* 2221 * If this socket is not expecting signature but 2222 * the segment contains signature just fail. 2223 */ 2224 TCPSTAT_INC(tcps_sig_err_sigopt); 2225 TCPSTAT_INC(tcps_sig_rcvbadsig); 2226 return (0); 2227 } 2228 2229 /* Signature is not expected, and not present in segment. */ 2230 return (1); 2231 } 2232 2233 /* 2234 * If this socket is expecting signature but the segment does not 2235 * contain any just fail. 2236 */ 2237 if ((to->to_flags & TOF_SIGNATURE) == 0) { 2238 TCPSTAT_INC(tcps_sig_err_nosigopt); 2239 TCPSTAT_INC(tcps_sig_rcvbadsig); 2240 return (0); 2241 } 2242 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0], 2243 IPSEC_DIR_INBOUND) == -1) { 2244 TCPSTAT_INC(tcps_sig_err_buildsig); 2245 TCPSTAT_INC(tcps_sig_rcvbadsig); 2246 return (0); 2247 } 2248 2249 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) { 2250 TCPSTAT_INC(tcps_sig_rcvbadsig); 2251 return (0); 2252 } 2253 TCPSTAT_INC(tcps_sig_rcvgoodsig); 2254 return (1); 2255} 2256#endif /* TCP_SIGNATURE */ 2257 2258static int 2259sysctl_drop(SYSCTL_HANDLER_ARGS) 2260{ 2261 /* addrs[0] is a foreign socket, addrs[1] is a local one. */ 2262 struct sockaddr_storage addrs[2]; 2263 struct inpcb *inp; 2264 struct tcpcb *tp; 2265 struct tcptw *tw; 2266 struct sockaddr_in *fin, *lin; 2267#ifdef INET6 2268 struct sockaddr_in6 *fin6, *lin6; 2269#endif 2270 int error; 2271 2272 inp = NULL; 2273 fin = lin = NULL; 2274#ifdef INET6 2275 fin6 = lin6 = NULL; 2276#endif 2277 error = 0; 2278 2279 if (req->oldptr != NULL || req->oldlen != 0) 2280 return (EINVAL); 2281 if (req->newptr == NULL) 2282 return (EPERM); 2283 if (req->newlen < sizeof(addrs)) 2284 return (ENOMEM); 2285 error = SYSCTL_IN(req, &addrs, sizeof(addrs)); 2286 if (error) 2287 return (error); 2288 2289 switch (addrs[0].ss_family) { 2290#ifdef INET6 2291 case AF_INET6: 2292 fin6 = (struct sockaddr_in6 *)&addrs[0]; 2293 lin6 = (struct sockaddr_in6 *)&addrs[1]; 2294 if (fin6->sin6_len != sizeof(struct sockaddr_in6) || 2295 lin6->sin6_len != sizeof(struct sockaddr_in6)) 2296 return (EINVAL); 2297 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) { 2298 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr)) 2299 return (EINVAL); 2300 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]); 2301 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]); 2302 fin = (struct sockaddr_in *)&addrs[0]; 2303 lin = (struct sockaddr_in *)&addrs[1]; 2304 break; 2305 } 2306 error = sa6_embedscope(fin6, V_ip6_use_defzone); 2307 if (error) 2308 return (error); 2309 error = sa6_embedscope(lin6, V_ip6_use_defzone); 2310 if (error) 2311 return (error); 2312 break; 2313#endif 2314#ifdef INET 2315 case AF_INET: 2316 fin = (struct sockaddr_in *)&addrs[0]; 2317 lin = (struct sockaddr_in *)&addrs[1]; 2318 if (fin->sin_len != sizeof(struct sockaddr_in) || 2319 lin->sin_len != sizeof(struct sockaddr_in)) 2320 return (EINVAL); 2321 break; 2322#endif 2323 default: 2324 return (EINVAL); 2325 } 2326 INP_INFO_WLOCK(&V_tcbinfo); 2327 switch (addrs[0].ss_family) { 2328#ifdef INET6 2329 case AF_INET6: 2330 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr, 2331 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port, 2332 INPLOOKUP_WLOCKPCB, NULL); 2333 break; 2334#endif 2335#ifdef INET 2336 case AF_INET: 2337 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port, 2338 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL); 2339 break; 2340#endif 2341 } 2342 if (inp != NULL) { 2343 if (inp->inp_flags & INP_TIMEWAIT) { 2344 /* 2345 * XXXRW: There currently exists a state where an 2346 * inpcb is present, but its timewait state has been 2347 * discarded. For now, don't allow dropping of this 2348 * type of inpcb. 2349 */ 2350 tw = intotw(inp); 2351 if (tw != NULL) 2352 tcp_twclose(tw, 0); 2353 else 2354 INP_WUNLOCK(inp); 2355 } else if (!(inp->inp_flags & INP_DROPPED) && 2356 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) { 2357 tp = intotcpcb(inp); 2358 tp = tcp_drop(tp, ECONNABORTED); 2359 if (tp != NULL) 2360 INP_WUNLOCK(inp); 2361 } else 2362 INP_WUNLOCK(inp); 2363 } else 2364 error = ESRCH; 2365 INP_INFO_WUNLOCK(&V_tcbinfo); 2366 return (error); 2367} 2368 2369SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop, 2370 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL, 2371 0, sysctl_drop, "", "Drop TCP connection"); 2372 2373/* 2374 * Generate a standardized TCP log line for use throughout the 2375 * tcp subsystem. Memory allocation is done with M_NOWAIT to 2376 * allow use in the interrupt context. 2377 * 2378 * NB: The caller MUST free(s, M_TCPLOG) the returned string. 2379 * NB: The function may return NULL if memory allocation failed. 2380 * 2381 * Due to header inclusion and ordering limitations the struct ip 2382 * and ip6_hdr pointers have to be passed as void pointers. 2383 */ 2384char * 2385tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2386 const void *ip6hdr) 2387{ 2388 2389 /* Is logging enabled? */ 2390 if (tcp_log_in_vain == 0) 2391 return (NULL); 2392 2393 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2394} 2395 2396char * 2397tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2398 const void *ip6hdr) 2399{ 2400 2401 /* Is logging enabled? */ 2402 if (tcp_log_debug == 0) 2403 return (NULL); 2404 2405 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr)); 2406} 2407 2408static char * 2409tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr, 2410 const void *ip6hdr) 2411{ 2412 char *s, *sp; 2413 size_t size; 2414 struct ip *ip; 2415#ifdef INET6 2416 const struct ip6_hdr *ip6; 2417 2418 ip6 = (const struct ip6_hdr *)ip6hdr; 2419#endif /* INET6 */ 2420 ip = (struct ip *)ip4hdr; 2421 2422 /* 2423 * The log line looks like this: 2424 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>" 2425 */ 2426 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") + 2427 sizeof(PRINT_TH_FLAGS) + 1 + 2428#ifdef INET6 2429 2 * INET6_ADDRSTRLEN; 2430#else 2431 2 * INET_ADDRSTRLEN; 2432#endif /* INET6 */ 2433 2434 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT); 2435 if (s == NULL) 2436 return (NULL); 2437 2438 strcat(s, "TCP: ["); 2439 sp = s + strlen(s); 2440 2441 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) { 2442 inet_ntoa_r(inc->inc_faddr, sp); 2443 sp = s + strlen(s); 2444 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2445 sp = s + strlen(s); 2446 inet_ntoa_r(inc->inc_laddr, sp); 2447 sp = s + strlen(s); 2448 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2449#ifdef INET6 2450 } else if (inc) { 2451 ip6_sprintf(sp, &inc->inc6_faddr); 2452 sp = s + strlen(s); 2453 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport)); 2454 sp = s + strlen(s); 2455 ip6_sprintf(sp, &inc->inc6_laddr); 2456 sp = s + strlen(s); 2457 sprintf(sp, "]:%i", ntohs(inc->inc_lport)); 2458 } else if (ip6 && th) { 2459 ip6_sprintf(sp, &ip6->ip6_src); 2460 sp = s + strlen(s); 2461 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2462 sp = s + strlen(s); 2463 ip6_sprintf(sp, &ip6->ip6_dst); 2464 sp = s + strlen(s); 2465 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2466#endif /* INET6 */ 2467#ifdef INET 2468 } else if (ip && th) { 2469 inet_ntoa_r(ip->ip_src, sp); 2470 sp = s + strlen(s); 2471 sprintf(sp, "]:%i to [", ntohs(th->th_sport)); 2472 sp = s + strlen(s); 2473 inet_ntoa_r(ip->ip_dst, sp); 2474 sp = s + strlen(s); 2475 sprintf(sp, "]:%i", ntohs(th->th_dport)); 2476#endif /* INET */ 2477 } else { 2478 free(s, M_TCPLOG); 2479 return (NULL); 2480 } 2481 sp = s + strlen(s); 2482 if (th) 2483 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS); 2484 if (*(s + size - 1) != '\0') 2485 panic("%s: string too long", __func__); 2486 return (s); 2487} 2488 2489/* 2490 * A subroutine which makes it easy to track TCP state changes with DTrace. 2491 * This function shouldn't be called for t_state initializations that don't 2492 * correspond to actual TCP state transitions. 2493 */ 2494void 2495tcp_state_change(struct tcpcb *tp, int newstate) 2496{ 2497#if defined(KDTRACE_HOOKS) 2498 int pstate = tp->t_state; 2499#endif 2500 2501 tp->t_state = newstate; 2502 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate); 2503} 2504