1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
| 1/* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
|
30 * $FreeBSD: head/sys/netinet/tcp_reass.c 130398 2004-06-13 02:50:07Z rwatson $
| 30 * $FreeBSD: head/sys/netinet/tcp_reass.c 130480 2004-06-14 18:16:22Z rwatson $
|
31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40 41#include <sys/param.h> 42#include <sys/kernel.h> 43#include <sys/mac.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/signalvar.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/syslog.h> 53#include <sys/systm.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <vm/uma.h> 58 59#include <net/if.h> 60#include <net/route.h> 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip6.h> 71#include <netinet/icmp6.h> 72#include <netinet6/in6_pcb.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/nd6.h> 75#include <netinet/tcp.h> 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#include <netinet6/tcp6_var.h> 81#include <netinet/tcpip.h> 82#ifdef TCPDEBUG 83#include <netinet/tcp_debug.h> 84#endif /* TCPDEBUG */ 85 86#ifdef FAST_IPSEC 87#include <netipsec/ipsec.h> 88#include <netipsec/ipsec6.h> 89#endif /*FAST_IPSEC*/ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#include <netinet6/ipsec6.h> 94#include <netkey/key.h> 95#endif /*IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99static const int tcprexmtthresh = 3; 100tcp_cc tcp_ccgen; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &log_in_vain, 0, "Log all incoming TCP connections"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST when dropping refused connections"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119#ifdef TCP_DROP_SYNFIN 120static int drop_synfin = 0; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 123#endif 124 125static int tcp_do_rfc3042 = 1; 126SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 128 129static int tcp_do_rfc3390 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 131 &tcp_do_rfc3390, 0, 132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 133 134SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 135 "TCP Segment Reassembly Queue"); 136 137static int tcp_reass_maxseg = 0; 138SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 139 &tcp_reass_maxseg, 0, 140 "Global maximum number of TCP Segments in Reassembly Queue"); 141 142int tcp_reass_qsize = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 144 &tcp_reass_qsize, 0, 145 "Global number of TCP Segments currently in Reassembly Queue"); 146 147static int tcp_reass_maxqlen = 48; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 149 &tcp_reass_maxqlen, 0, 150 "Maximum number of TCP Segments per individual Reassembly Queue"); 151 152static int tcp_reass_overflows = 0; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 154 &tcp_reass_overflows, 0, 155 "Global number of TCP Segment Reassembly Queue Overflows"); 156 157struct inpcbhead tcb; 158#define tcb6 tcb /* for KAME src sync over BSD*'s */ 159struct inpcbinfo tcbinfo; 160struct mtx *tcbinfo_mtx; 161 162static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 163static void tcp_pulloutofband(struct socket *, 164 struct tcphdr *, struct mbuf *, int); 165static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 166 struct mbuf *); 167static void tcp_xmit_timer(struct tcpcb *, int); 168static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 169static int tcp_timewait(struct tcptw *, struct tcpopt *, 170 struct tcphdr *, struct mbuf *, int); 171 172/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 173#ifdef INET6 174#define ND6_HINT(tp) \ 175do { \ 176 if ((tp) && (tp)->t_inpcb && \ 177 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 178 nd6_nud_hint(NULL, NULL, 0); \ 179} while (0) 180#else 181#define ND6_HINT(tp) 182#endif 183 184/* 185 * Indicate whether this ack should be delayed. We can delay the ack if 186 * - there is no delayed ack timer in progress and 187 * - our last ack wasn't a 0-sized window. We never want to delay 188 * the ack that opens up a 0-sized window and 189 * - delayed acks are enabled or 190 * - this is a half-synchronized T/TCP connection. 191 */ 192#define DELAY_ACK(tp) \ 193 ((!callout_active(tp->tt_delack) && \ 194 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 195 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 196 197/* Initialize TCP reassembly queue */ 198uma_zone_t tcp_reass_zone; 199void 200tcp_reass_init() 201{ 202 tcp_reass_maxseg = nmbclusters / 16; 203 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 204 &tcp_reass_maxseg); 205 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 207 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 208} 209 210static int 211tcp_reass(tp, th, tlenp, m) 212 register struct tcpcb *tp; 213 register struct tcphdr *th; 214 int *tlenp; 215 struct mbuf *m; 216{ 217 struct tseg_qent *q; 218 struct tseg_qent *p = NULL; 219 struct tseg_qent *nq; 220 struct tseg_qent *te = NULL; 221 struct socket *so = tp->t_inpcb->inp_socket; 222 int flags; 223 224 /* 225 * XXX: tcp_reass() is rather inefficient with its data structures 226 * and should be rewritten (see NetBSD for optimizations). While 227 * doing that it should move to its own file tcp_reass.c. 228 */ 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 /* 238 * Limit the number of segments in the reassembly queue to prevent 239 * holding on to too many segments (and thus running out of mbufs). 240 * Make sure to let the missing segment through which caused this 241 * queue. Always keep one global queue entry spare to be able to 242 * process the missing segment. 243 */ 244 if (th->th_seq != tp->rcv_nxt && 245 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 246 tp->t_segqlen >= tcp_reass_maxqlen)) { 247 tcp_reass_overflows++; 248 tcpstat.tcps_rcvmemdrop++; 249 m_freem(m); 250 return (0); 251 } 252 253 /* 254 * Allocate a new queue entry. If we can't, or hit the zone limit 255 * just drop the pkt. 256 */ 257 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 258 if (te == NULL) { 259 tcpstat.tcps_rcvmemdrop++; 260 m_freem(m); 261 return (0); 262 } 263 tp->t_segqlen++; 264 tcp_reass_qsize++; 265 266 /* 267 * Find a segment which begins after this one does. 268 */ 269 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 270 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 271 break; 272 p = q; 273 } 274 275 /* 276 * If there is a preceding segment, it may provide some of 277 * our data already. If so, drop the data from the incoming 278 * segment. If it provides all of our data, drop us. 279 */ 280 if (p != NULL) { 281 register int i; 282 /* conversion to int (in i) handles seq wraparound */ 283 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 284 if (i > 0) { 285 if (i >= *tlenp) { 286 tcpstat.tcps_rcvduppack++; 287 tcpstat.tcps_rcvdupbyte += *tlenp; 288 m_freem(m); 289 uma_zfree(tcp_reass_zone, te); 290 tp->t_segqlen--; 291 tcp_reass_qsize--; 292 /* 293 * Try to present any queued data 294 * at the left window edge to the user. 295 * This is needed after the 3-WHS 296 * completes. 297 */ 298 goto present; /* ??? */ 299 } 300 m_adj(m, i); 301 *tlenp -= i; 302 th->th_seq += i; 303 } 304 } 305 tcpstat.tcps_rcvoopack++; 306 tcpstat.tcps_rcvoobyte += *tlenp; 307 308 /* 309 * While we overlap succeeding segments trim them or, 310 * if they are completely covered, dequeue them. 311 */ 312 while (q) { 313 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 314 if (i <= 0) 315 break; 316 if (i < q->tqe_len) { 317 q->tqe_th->th_seq += i; 318 q->tqe_len -= i; 319 m_adj(q->tqe_m, i); 320 break; 321 } 322 323 nq = LIST_NEXT(q, tqe_q); 324 LIST_REMOVE(q, tqe_q); 325 m_freem(q->tqe_m); 326 uma_zfree(tcp_reass_zone, q); 327 tp->t_segqlen--; 328 tcp_reass_qsize--; 329 q = nq; 330 } 331 332 /* Insert the new segment queue entry into place. */ 333 te->tqe_m = m; 334 te->tqe_th = th; 335 te->tqe_len = *tlenp; 336 337 if (p == NULL) { 338 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 339 } else { 340 LIST_INSERT_AFTER(p, te, tqe_q); 341 } 342 343present: 344 /* 345 * Present data to user, advancing rcv_nxt through 346 * completed sequence space. 347 */ 348 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 349 return (0); 350 q = LIST_FIRST(&tp->t_segq); 351 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 352 return (0); 353 do { 354 tp->rcv_nxt += q->tqe_len; 355 flags = q->tqe_th->th_flags & TH_FIN; 356 nq = LIST_NEXT(q, tqe_q); 357 LIST_REMOVE(q, tqe_q);
| 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40 41#include <sys/param.h> 42#include <sys/kernel.h> 43#include <sys/mac.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/signalvar.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/syslog.h> 53#include <sys/systm.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <vm/uma.h> 58 59#include <net/if.h> 60#include <net/route.h> 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip6.h> 71#include <netinet/icmp6.h> 72#include <netinet6/in6_pcb.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/nd6.h> 75#include <netinet/tcp.h> 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#include <netinet6/tcp6_var.h> 81#include <netinet/tcpip.h> 82#ifdef TCPDEBUG 83#include <netinet/tcp_debug.h> 84#endif /* TCPDEBUG */ 85 86#ifdef FAST_IPSEC 87#include <netipsec/ipsec.h> 88#include <netipsec/ipsec6.h> 89#endif /*FAST_IPSEC*/ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#include <netinet6/ipsec6.h> 94#include <netkey/key.h> 95#endif /*IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99static const int tcprexmtthresh = 3; 100tcp_cc tcp_ccgen; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &log_in_vain, 0, "Log all incoming TCP connections"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST when dropping refused connections"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119#ifdef TCP_DROP_SYNFIN 120static int drop_synfin = 0; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 123#endif 124 125static int tcp_do_rfc3042 = 1; 126SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 128 129static int tcp_do_rfc3390 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 131 &tcp_do_rfc3390, 0, 132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 133 134SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 135 "TCP Segment Reassembly Queue"); 136 137static int tcp_reass_maxseg = 0; 138SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 139 &tcp_reass_maxseg, 0, 140 "Global maximum number of TCP Segments in Reassembly Queue"); 141 142int tcp_reass_qsize = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 144 &tcp_reass_qsize, 0, 145 "Global number of TCP Segments currently in Reassembly Queue"); 146 147static int tcp_reass_maxqlen = 48; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 149 &tcp_reass_maxqlen, 0, 150 "Maximum number of TCP Segments per individual Reassembly Queue"); 151 152static int tcp_reass_overflows = 0; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 154 &tcp_reass_overflows, 0, 155 "Global number of TCP Segment Reassembly Queue Overflows"); 156 157struct inpcbhead tcb; 158#define tcb6 tcb /* for KAME src sync over BSD*'s */ 159struct inpcbinfo tcbinfo; 160struct mtx *tcbinfo_mtx; 161 162static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 163static void tcp_pulloutofband(struct socket *, 164 struct tcphdr *, struct mbuf *, int); 165static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 166 struct mbuf *); 167static void tcp_xmit_timer(struct tcpcb *, int); 168static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 169static int tcp_timewait(struct tcptw *, struct tcpopt *, 170 struct tcphdr *, struct mbuf *, int); 171 172/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 173#ifdef INET6 174#define ND6_HINT(tp) \ 175do { \ 176 if ((tp) && (tp)->t_inpcb && \ 177 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 178 nd6_nud_hint(NULL, NULL, 0); \ 179} while (0) 180#else 181#define ND6_HINT(tp) 182#endif 183 184/* 185 * Indicate whether this ack should be delayed. We can delay the ack if 186 * - there is no delayed ack timer in progress and 187 * - our last ack wasn't a 0-sized window. We never want to delay 188 * the ack that opens up a 0-sized window and 189 * - delayed acks are enabled or 190 * - this is a half-synchronized T/TCP connection. 191 */ 192#define DELAY_ACK(tp) \ 193 ((!callout_active(tp->tt_delack) && \ 194 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 195 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 196 197/* Initialize TCP reassembly queue */ 198uma_zone_t tcp_reass_zone; 199void 200tcp_reass_init() 201{ 202 tcp_reass_maxseg = nmbclusters / 16; 203 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 204 &tcp_reass_maxseg); 205 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 207 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 208} 209 210static int 211tcp_reass(tp, th, tlenp, m) 212 register struct tcpcb *tp; 213 register struct tcphdr *th; 214 int *tlenp; 215 struct mbuf *m; 216{ 217 struct tseg_qent *q; 218 struct tseg_qent *p = NULL; 219 struct tseg_qent *nq; 220 struct tseg_qent *te = NULL; 221 struct socket *so = tp->t_inpcb->inp_socket; 222 int flags; 223 224 /* 225 * XXX: tcp_reass() is rather inefficient with its data structures 226 * and should be rewritten (see NetBSD for optimizations). While 227 * doing that it should move to its own file tcp_reass.c. 228 */ 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 /* 238 * Limit the number of segments in the reassembly queue to prevent 239 * holding on to too many segments (and thus running out of mbufs). 240 * Make sure to let the missing segment through which caused this 241 * queue. Always keep one global queue entry spare to be able to 242 * process the missing segment. 243 */ 244 if (th->th_seq != tp->rcv_nxt && 245 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 246 tp->t_segqlen >= tcp_reass_maxqlen)) { 247 tcp_reass_overflows++; 248 tcpstat.tcps_rcvmemdrop++; 249 m_freem(m); 250 return (0); 251 } 252 253 /* 254 * Allocate a new queue entry. If we can't, or hit the zone limit 255 * just drop the pkt. 256 */ 257 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 258 if (te == NULL) { 259 tcpstat.tcps_rcvmemdrop++; 260 m_freem(m); 261 return (0); 262 } 263 tp->t_segqlen++; 264 tcp_reass_qsize++; 265 266 /* 267 * Find a segment which begins after this one does. 268 */ 269 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 270 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 271 break; 272 p = q; 273 } 274 275 /* 276 * If there is a preceding segment, it may provide some of 277 * our data already. If so, drop the data from the incoming 278 * segment. If it provides all of our data, drop us. 279 */ 280 if (p != NULL) { 281 register int i; 282 /* conversion to int (in i) handles seq wraparound */ 283 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 284 if (i > 0) { 285 if (i >= *tlenp) { 286 tcpstat.tcps_rcvduppack++; 287 tcpstat.tcps_rcvdupbyte += *tlenp; 288 m_freem(m); 289 uma_zfree(tcp_reass_zone, te); 290 tp->t_segqlen--; 291 tcp_reass_qsize--; 292 /* 293 * Try to present any queued data 294 * at the left window edge to the user. 295 * This is needed after the 3-WHS 296 * completes. 297 */ 298 goto present; /* ??? */ 299 } 300 m_adj(m, i); 301 *tlenp -= i; 302 th->th_seq += i; 303 } 304 } 305 tcpstat.tcps_rcvoopack++; 306 tcpstat.tcps_rcvoobyte += *tlenp; 307 308 /* 309 * While we overlap succeeding segments trim them or, 310 * if they are completely covered, dequeue them. 311 */ 312 while (q) { 313 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 314 if (i <= 0) 315 break; 316 if (i < q->tqe_len) { 317 q->tqe_th->th_seq += i; 318 q->tqe_len -= i; 319 m_adj(q->tqe_m, i); 320 break; 321 } 322 323 nq = LIST_NEXT(q, tqe_q); 324 LIST_REMOVE(q, tqe_q); 325 m_freem(q->tqe_m); 326 uma_zfree(tcp_reass_zone, q); 327 tp->t_segqlen--; 328 tcp_reass_qsize--; 329 q = nq; 330 } 331 332 /* Insert the new segment queue entry into place. */ 333 te->tqe_m = m; 334 te->tqe_th = th; 335 te->tqe_len = *tlenp; 336 337 if (p == NULL) { 338 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 339 } else { 340 LIST_INSERT_AFTER(p, te, tqe_q); 341 } 342 343present: 344 /* 345 * Present data to user, advancing rcv_nxt through 346 * completed sequence space. 347 */ 348 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 349 return (0); 350 q = LIST_FIRST(&tp->t_segq); 351 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 352 return (0); 353 do { 354 tp->rcv_nxt += q->tqe_len; 355 flags = q->tqe_th->th_flags & TH_FIN; 356 nq = LIST_NEXT(q, tqe_q); 357 LIST_REMOVE(q, tqe_q);
|
358 if (so->so_state & SS_CANTRCVMORE)
| 358 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
|
359 m_freem(q->tqe_m); 360 else 361 sbappendstream(&so->so_rcv, q->tqe_m); 362 uma_zfree(tcp_reass_zone, q); 363 tp->t_segqlen--; 364 tcp_reass_qsize--; 365 q = nq; 366 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 367 ND6_HINT(tp); 368 sorwakeup(so); 369 return (flags); 370} 371 372/* 373 * TCP input routine, follows pages 65-76 of the 374 * protocol specification dated September, 1981 very closely. 375 */ 376#ifdef INET6 377int 378tcp6_input(mp, offp, proto) 379 struct mbuf **mp; 380 int *offp, proto; 381{ 382 register struct mbuf *m = *mp; 383 struct in6_ifaddr *ia6; 384 385 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 386 387 /* 388 * draft-itojun-ipv6-tcp-to-anycast 389 * better place to put this in? 390 */ 391 ia6 = ip6_getdstifaddr(m); 392 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 393 struct ip6_hdr *ip6; 394 395 ip6 = mtod(m, struct ip6_hdr *); 396 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 397 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 398 return IPPROTO_DONE; 399 } 400 401 tcp_input(m, *offp); 402 return IPPROTO_DONE; 403} 404#endif 405 406void 407tcp_input(m, off0) 408 register struct mbuf *m; 409 int off0; 410{ 411 register struct tcphdr *th; 412 register struct ip *ip = NULL; 413 register struct ipovly *ipov; 414 register struct inpcb *inp = NULL; 415 u_char *optp = NULL; 416 int optlen = 0; 417 int len, tlen, off; 418 int drop_hdrlen; 419 register struct tcpcb *tp = 0; 420 register int thflags; 421 struct socket *so = 0; 422 int todrop, acked, ourfinisacked, needoutput = 0; 423 u_long tiwin; 424 struct tcpopt to; /* options in this segment */ 425 struct rmxp_tao tao; /* our TAO cache entry */ 426 int headlocked = 0; 427 struct sockaddr_in *next_hop = NULL; 428 int rstreason; /* For badport_bandlim accounting purposes */ 429 430 struct ip6_hdr *ip6 = NULL; 431#ifdef INET6 432 int isipv6; 433#else 434 const int isipv6 = 0; 435#endif 436 437#ifdef TCPDEBUG 438 /* 439 * The size of tcp_saveipgen must be the size of the max ip header, 440 * now IPv6. 441 */ 442 u_char tcp_saveipgen[40]; 443 struct tcphdr tcp_savetcp; 444 short ostate = 0; 445#endif 446 447 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 448 next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD); 449#ifdef INET6 450 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 451#endif 452 bzero(&tao, sizeof(tao)); 453 bzero((char *)&to, sizeof(to)); 454 455 tcpstat.tcps_rcvtotal++; 456 457 if (isipv6) { 458#ifdef INET6 459 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 460 ip6 = mtod(m, struct ip6_hdr *); 461 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 462 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 463 tcpstat.tcps_rcvbadsum++; 464 goto drop; 465 } 466 th = (struct tcphdr *)((caddr_t)ip6 + off0); 467 468 /* 469 * Be proactive about unspecified IPv6 address in source. 470 * As we use all-zero to indicate unbounded/unconnected pcb, 471 * unspecified IPv6 address can be used to confuse us. 472 * 473 * Note that packets with unspecified IPv6 destination is 474 * already dropped in ip6_input. 475 */ 476 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 477 /* XXX stat */ 478 goto drop; 479 } 480#else 481 th = NULL; /* XXX: avoid compiler warning */ 482#endif 483 } else { 484 /* 485 * Get IP and TCP header together in first mbuf. 486 * Note: IP leaves IP header in first mbuf. 487 */ 488 if (off0 > sizeof (struct ip)) { 489 ip_stripoptions(m, (struct mbuf *)0); 490 off0 = sizeof(struct ip); 491 } 492 if (m->m_len < sizeof (struct tcpiphdr)) { 493 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 494 tcpstat.tcps_rcvshort++; 495 return; 496 } 497 } 498 ip = mtod(m, struct ip *); 499 ipov = (struct ipovly *)ip; 500 th = (struct tcphdr *)((caddr_t)ip + off0); 501 tlen = ip->ip_len; 502 503 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 504 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 505 th->th_sum = m->m_pkthdr.csum_data; 506 else 507 th->th_sum = in_pseudo(ip->ip_src.s_addr, 508 ip->ip_dst.s_addr, 509 htonl(m->m_pkthdr.csum_data + 510 ip->ip_len + 511 IPPROTO_TCP)); 512 th->th_sum ^= 0xffff; 513#ifdef TCPDEBUG 514 ipov->ih_len = (u_short)tlen; 515 ipov->ih_len = htons(ipov->ih_len); 516#endif 517 } else { 518 /* 519 * Checksum extended TCP header and data. 520 */ 521 len = sizeof (struct ip) + tlen; 522 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525 th->th_sum = in_cksum(m, len); 526 } 527 if (th->th_sum) { 528 tcpstat.tcps_rcvbadsum++; 529 goto drop; 530 } 531#ifdef INET6 532 /* Re-initialization for later version check */ 533 ip->ip_v = IPVERSION; 534#endif 535 } 536 537 /* 538 * Check that TCP offset makes sense, 539 * pull out TCP options and adjust length. XXX 540 */ 541 off = th->th_off << 2; 542 if (off < sizeof (struct tcphdr) || off > tlen) { 543 tcpstat.tcps_rcvbadoff++; 544 goto drop; 545 } 546 tlen -= off; /* tlen is used instead of ti->ti_len */ 547 if (off > sizeof (struct tcphdr)) { 548 if (isipv6) { 549#ifdef INET6 550 IP6_EXTHDR_CHECK(m, off0, off, ); 551 ip6 = mtod(m, struct ip6_hdr *); 552 th = (struct tcphdr *)((caddr_t)ip6 + off0); 553#endif 554 } else { 555 if (m->m_len < sizeof(struct ip) + off) { 556 if ((m = m_pullup(m, sizeof (struct ip) + off)) 557 == 0) { 558 tcpstat.tcps_rcvshort++; 559 return; 560 } 561 ip = mtod(m, struct ip *); 562 ipov = (struct ipovly *)ip; 563 th = (struct tcphdr *)((caddr_t)ip + off0); 564 } 565 } 566 optlen = off - sizeof (struct tcphdr); 567 optp = (u_char *)(th + 1); 568 } 569 thflags = th->th_flags; 570 571#ifdef TCP_DROP_SYNFIN 572 /* 573 * If the drop_synfin option is enabled, drop all packets with 574 * both the SYN and FIN bits set. This prevents e.g. nmap from 575 * identifying the TCP/IP stack. 576 * 577 * This is a violation of the TCP specification. 578 */ 579 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 580 goto drop; 581#endif 582 583 /* 584 * Convert TCP protocol specific fields to host format. 585 */ 586 th->th_seq = ntohl(th->th_seq); 587 th->th_ack = ntohl(th->th_ack); 588 th->th_win = ntohs(th->th_win); 589 th->th_urp = ntohs(th->th_urp); 590 591 /* 592 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 593 * until after ip6_savecontrol() is called and before other functions 594 * which don't want those proto headers. 595 * Because ip6_savecontrol() is going to parse the mbuf to 596 * search for data to be passed up to user-land, it wants mbuf 597 * parameters to be unchanged. 598 * XXX: the call of ip6_savecontrol() has been obsoleted based on 599 * latest version of the advanced API (20020110). 600 */ 601 drop_hdrlen = off0 + off; 602 603 /* 604 * Locate pcb for segment. 605 */ 606 INP_INFO_WLOCK(&tcbinfo); 607 headlocked = 1; 608findpcb: 609 /* IPFIREWALL_FORWARD section */ 610 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 611 /* 612 * Transparently forwarded. Pretend to be the destination. 613 * already got one like this? 614 */ 615 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 616 ip->ip_dst, th->th_dport, 617 0, m->m_pkthdr.rcvif); 618 if (!inp) { 619 /* It's new. Try find the ambushing socket. */ 620 inp = in_pcblookup_hash(&tcbinfo, 621 ip->ip_src, th->th_sport, 622 next_hop->sin_addr, 623 next_hop->sin_port ? 624 ntohs(next_hop->sin_port) : 625 th->th_dport, 626 1, m->m_pkthdr.rcvif); 627 } 628 } else { 629 if (isipv6) { 630#ifdef INET6 631 inp = in6_pcblookup_hash(&tcbinfo, 632 &ip6->ip6_src, th->th_sport, 633 &ip6->ip6_dst, th->th_dport, 634 1, m->m_pkthdr.rcvif); 635#endif 636 } else 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 ip->ip_dst, th->th_dport, 640 1, m->m_pkthdr.rcvif); 641 } 642 643#if defined(IPSEC) || defined(FAST_IPSEC) 644 if (isipv6) { 645 if (inp != NULL && ipsec6_in_reject(m, inp)) { 646#ifdef IPSEC 647 ipsec6stat.in_polvio++; 648#endif /*IPSEC*/ 649 goto drop; 650 } 651 } else if (inp != NULL && ipsec4_in_reject(m, inp)) { 652#ifdef IPSEC 653 ipsecstat.in_polvio++; 654#endif /*IPSEC*/ 655 goto drop; 656 } 657#endif /*IPSEC || FAST_IPSEC*/ 658 659 /* 660 * If the state is CLOSED (i.e., TCB does not exist) then 661 * all data in the incoming segment is discarded. 662 * If the TCB exists but is in CLOSED state, it is embryonic, 663 * but should either do a listen or a connect soon. 664 */ 665 if (inp == NULL) { 666 if (log_in_vain) { 667#ifdef INET6 668 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 669#else 670 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 671#endif 672 673 if (isipv6) { 674#ifdef INET6 675 strcpy(dbuf, "["); 676 strcpy(sbuf, "["); 677 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 678 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 679 strcat(dbuf, "]"); 680 strcat(sbuf, "]"); 681#endif 682 } else { 683 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 684 strcpy(sbuf, inet_ntoa(ip->ip_src)); 685 } 686 switch (log_in_vain) { 687 case 1: 688 if ((thflags & TH_SYN) == 0) 689 break; 690 /* FALLTHROUGH */ 691 case 2: 692 log(LOG_INFO, 693 "Connection attempt to TCP %s:%d " 694 "from %s:%d flags:0x%02x\n", 695 dbuf, ntohs(th->th_dport), sbuf, 696 ntohs(th->th_sport), thflags); 697 break; 698 default: 699 break; 700 } 701 } 702 if (blackhole) { 703 switch (blackhole) { 704 case 1: 705 if (thflags & TH_SYN) 706 goto drop; 707 break; 708 case 2: 709 goto drop; 710 default: 711 goto drop; 712 } 713 } 714 rstreason = BANDLIM_RST_CLOSEDPORT; 715 goto dropwithreset; 716 } 717 INP_LOCK(inp); 718 if (inp->inp_vflag & INP_TIMEWAIT) { 719 /* 720 * The only option of relevance is TOF_CC, and only if 721 * present in a SYN segment. See tcp_timewait(). 722 */ 723 if (thflags & TH_SYN) 724 tcp_dooptions(&to, optp, optlen, 1); 725 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 726 &to, th, m, tlen)) 727 goto findpcb; 728 /* 729 * tcp_timewait unlocks inp. 730 */ 731 INP_INFO_WUNLOCK(&tcbinfo); 732 return; 733 } 734 tp = intotcpcb(inp); 735 if (tp == 0) { 736 INP_UNLOCK(inp); 737 rstreason = BANDLIM_RST_CLOSEDPORT; 738 goto dropwithreset; 739 } 740 if (tp->t_state == TCPS_CLOSED) 741 goto drop; 742 743 /* Unscale the window into a 32-bit value. */ 744 if ((thflags & TH_SYN) == 0) 745 tiwin = th->th_win << tp->snd_scale; 746 else 747 tiwin = th->th_win; 748 749#ifdef MAC 750 if (mac_check_inpcb_deliver(inp, m)) 751 goto drop; 752#endif 753 so = inp->inp_socket; 754#ifdef TCPDEBUG 755 if (so->so_options & SO_DEBUG) { 756 ostate = tp->t_state; 757 if (isipv6) 758 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 759 else 760 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 761 tcp_savetcp = *th; 762 } 763#endif 764 if (so->so_options & SO_ACCEPTCONN) { 765 struct in_conninfo inc; 766 767#ifdef INET6 768 inc.inc_isipv6 = isipv6; 769#endif 770 if (isipv6) { 771 inc.inc6_faddr = ip6->ip6_src; 772 inc.inc6_laddr = ip6->ip6_dst; 773 } else { 774 inc.inc_faddr = ip->ip_src; 775 inc.inc_laddr = ip->ip_dst; 776 } 777 inc.inc_fport = th->th_sport; 778 inc.inc_lport = th->th_dport; 779 780 /* 781 * If the state is LISTEN then ignore segment if it contains 782 * a RST. If the segment contains an ACK then it is bad and 783 * send a RST. If it does not contain a SYN then it is not 784 * interesting; drop it. 785 * 786 * If the state is SYN_RECEIVED (syncache) and seg contains 787 * an ACK, but not for our SYN/ACK, send a RST. If the seg 788 * contains a RST, check the sequence number to see if it 789 * is a valid reset segment. 790 */ 791 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 792 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 793 if (!syncache_expand(&inc, th, &so, m)) { 794 /* 795 * No syncache entry, or ACK was not 796 * for our SYN/ACK. Send a RST. 797 */ 798 tcpstat.tcps_badsyn++; 799 rstreason = BANDLIM_RST_OPENPORT; 800 goto dropwithreset; 801 } 802 if (so == NULL) { 803 /* 804 * Could not complete 3-way handshake, 805 * connection is being closed down, and 806 * syncache will free mbuf. 807 */ 808 INP_UNLOCK(inp); 809 INP_INFO_WUNLOCK(&tcbinfo); 810 return; 811 } 812 /* 813 * Socket is created in state SYN_RECEIVED. 814 * Continue processing segment. 815 */ 816 INP_UNLOCK(inp); 817 inp = sotoinpcb(so); 818 INP_LOCK(inp); 819 tp = intotcpcb(inp); 820 /* 821 * This is what would have happened in 822 * tcp_output() when the SYN,ACK was sent. 823 */ 824 tp->snd_up = tp->snd_una; 825 tp->snd_max = tp->snd_nxt = tp->iss + 1; 826 tp->last_ack_sent = tp->rcv_nxt; 827 /* 828 * RFC1323: The window in SYN & SYN/ACK 829 * segments is never scaled. 830 */ 831 tp->snd_wnd = tiwin; /* unscaled */ 832 goto after_listen; 833 } 834 if (thflags & TH_RST) { 835 syncache_chkrst(&inc, th); 836 goto drop; 837 } 838 if (thflags & TH_ACK) { 839 syncache_badack(&inc); 840 tcpstat.tcps_badsyn++; 841 rstreason = BANDLIM_RST_OPENPORT; 842 goto dropwithreset; 843 } 844 goto drop; 845 } 846 847 /* 848 * Segment's flags are (SYN) or (SYN|FIN). 849 */ 850#ifdef INET6 851 /* 852 * If deprecated address is forbidden, 853 * we do not accept SYN to deprecated interface 854 * address to prevent any new inbound connection from 855 * getting established. 856 * When we do not accept SYN, we send a TCP RST, 857 * with deprecated source address (instead of dropping 858 * it). We compromise it as it is much better for peer 859 * to send a RST, and RST will be the final packet 860 * for the exchange. 861 * 862 * If we do not forbid deprecated addresses, we accept 863 * the SYN packet. RFC2462 does not suggest dropping 864 * SYN in this case. 865 * If we decipher RFC2462 5.5.4, it says like this: 866 * 1. use of deprecated addr with existing 867 * communication is okay - "SHOULD continue to be 868 * used" 869 * 2. use of it with new communication: 870 * (2a) "SHOULD NOT be used if alternate address 871 * with sufficient scope is available" 872 * (2b) nothing mentioned otherwise. 873 * Here we fall into (2b) case as we have no choice in 874 * our source address selection - we must obey the peer. 875 * 876 * The wording in RFC2462 is confusing, and there are 877 * multiple description text for deprecated address 878 * handling - worse, they are not exactly the same. 879 * I believe 5.5.4 is the best one, so we follow 5.5.4. 880 */ 881 if (isipv6 && !ip6_use_deprecated) { 882 struct in6_ifaddr *ia6; 883 884 if ((ia6 = ip6_getdstifaddr(m)) && 885 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 886 INP_UNLOCK(inp); 887 tp = NULL; 888 rstreason = BANDLIM_RST_OPENPORT; 889 goto dropwithreset; 890 } 891 } 892#endif 893 /* 894 * If it is from this socket, drop it, it must be forged. 895 * Don't bother responding if the destination was a broadcast. 896 */ 897 if (th->th_dport == th->th_sport) { 898 if (isipv6) { 899 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 900 &ip6->ip6_src)) 901 goto drop; 902 } else { 903 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 904 goto drop; 905 } 906 } 907 /* 908 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 909 * 910 * Note that it is quite possible to receive unicast 911 * link-layer packets with a broadcast IP address. Use 912 * in_broadcast() to find them. 913 */ 914 if (m->m_flags & (M_BCAST|M_MCAST)) 915 goto drop; 916 if (isipv6) { 917 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 918 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 919 goto drop; 920 } else { 921 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 922 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 923 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 924 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 925 goto drop; 926 } 927 /* 928 * SYN appears to be valid; create compressed TCP state 929 * for syncache, or perform t/tcp connection. 930 */ 931 if (so->so_qlen <= so->so_qlimit) { 932#ifdef TCPDEBUG 933 if (so->so_options & SO_DEBUG) 934 tcp_trace(TA_INPUT, ostate, tp, 935 (void *)tcp_saveipgen, &tcp_savetcp, 0); 936#endif 937 tcp_dooptions(&to, optp, optlen, 1); 938 if (!syncache_add(&inc, &to, th, &so, m)) 939 goto drop; 940 if (so == NULL) { 941 /* 942 * Entry added to syncache, mbuf used to 943 * send SYN,ACK packet. 944 */ 945 KASSERT(headlocked, ("headlocked")); 946 INP_UNLOCK(inp); 947 INP_INFO_WUNLOCK(&tcbinfo); 948 return; 949 } 950 /* 951 * Segment passed TAO tests. 952 */ 953 INP_UNLOCK(inp); 954 inp = sotoinpcb(so); 955 INP_LOCK(inp); 956 tp = intotcpcb(inp); 957 tp->snd_wnd = tiwin; 958 tp->t_starttime = ticks; 959 tp->t_state = TCPS_ESTABLISHED; 960 961 /* 962 * T/TCP logic: 963 * If there is a FIN or if there is data, then 964 * delay SYN,ACK(SYN) in the hope of piggy-backing 965 * it on a response segment. Otherwise must send 966 * ACK now in case the other side is slow starting. 967 */ 968 if (thflags & TH_FIN || tlen != 0) 969 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 970 else 971 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 972 tcpstat.tcps_connects++; 973 soisconnected(so); 974 goto trimthenstep6; 975 } 976 goto drop; 977 } 978after_listen: 979 980 /* XXX temp debugging */ 981 /* should not happen - syncache should pick up these connections */ 982 if (tp->t_state == TCPS_LISTEN) 983 panic("tcp_input: TCPS_LISTEN"); 984 985 /* 986 * This is the second part of the MSS DoS prevention code (after 987 * minmss on the sending side) and it deals with too many too small 988 * tcp packets in a too short timeframe (1 second). 989 * 990 * For every full second we count the number of received packets 991 * and bytes. If we get a lot of packets per second for this connection 992 * (tcp_minmssoverload) we take a closer look at it and compute the 993 * average packet size for the past second. If that is less than 994 * tcp_minmss we get too many packets with very small payload which 995 * is not good and burdens our system (and every packet generates 996 * a wakeup to the process connected to our socket). We can reasonable 997 * expect this to be small packet DoS attack to exhaust our CPU 998 * cycles. 999 * 1000 * Care has to be taken for the minimum packet overload value. This 1001 * value defines the minimum number of packets per second before we 1002 * start to worry. This must not be too low to avoid killing for 1003 * example interactive connections with many small packets like 1004 * telnet or SSH. 1005 * 1006 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1007 * this check. 1008 * 1009 * Account for packet if payload packet, skip over ACK, etc. 1010 */ 1011 if (tcp_minmss && tcp_minmssoverload && 1012 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1013 if (tp->rcv_second > ticks) { 1014 tp->rcv_pps++; 1015 tp->rcv_byps += tlen + off; 1016 if (tp->rcv_pps > tcp_minmssoverload) { 1017 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1018 printf("too many small tcp packets from " 1019 "%s:%u, av. %lubyte/packet, " 1020 "dropping connection\n", 1021#ifdef INET6 1022 isipv6 ? 1023 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1024#endif 1025 inet_ntoa(inp->inp_inc.inc_faddr), 1026 inp->inp_inc.inc_fport, 1027 tp->rcv_byps / tp->rcv_pps); 1028 tp = tcp_drop(tp, ECONNRESET); 1029 tcpstat.tcps_minmssdrops++; 1030 goto drop; 1031 } 1032 } 1033 } else { 1034 tp->rcv_second = ticks + hz; 1035 tp->rcv_pps = 1; 1036 tp->rcv_byps = tlen + off; 1037 } 1038 } 1039 1040 /* 1041 * Segment received on connection. 1042 * Reset idle time and keep-alive timer. 1043 */ 1044 tp->t_rcvtime = ticks; 1045 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1046 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1047 1048 /* 1049 * Process options only when we get SYN/ACK back. The SYN case 1050 * for incoming connections is handled in tcp_syncache. 1051 * XXX this is traditional behavior, may need to be cleaned up. 1052 */ 1053 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1054 if (thflags & TH_SYN) { 1055 if (to.to_flags & TOF_SCALE) { 1056 tp->t_flags |= TF_RCVD_SCALE; 1057 tp->requested_s_scale = to.to_requested_s_scale; 1058 } 1059 if (to.to_flags & TOF_TS) { 1060 tp->t_flags |= TF_RCVD_TSTMP; 1061 tp->ts_recent = to.to_tsval; 1062 tp->ts_recent_age = ticks; 1063 } 1064 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1065 tp->t_flags |= TF_RCVD_CC; 1066 if (to.to_flags & TOF_MSS) 1067 tcp_mss(tp, to.to_mss); 1068 } 1069 1070 /* 1071 * Header prediction: check for the two common cases 1072 * of a uni-directional data xfer. If the packet has 1073 * no control flags, is in-sequence, the window didn't 1074 * change and we're not retransmitting, it's a 1075 * candidate. If the length is zero and the ack moved 1076 * forward, we're the sender side of the xfer. Just 1077 * free the data acked & wake any higher level process 1078 * that was blocked waiting for space. If the length 1079 * is non-zero and the ack didn't move, we're the 1080 * receiver side. If we're getting packets in-order 1081 * (the reassembly queue is empty), add the data to 1082 * the socket buffer and note that we need a delayed ack. 1083 * Make sure that the hidden state-flags are also off. 1084 * Since we check for TCPS_ESTABLISHED above, it can only 1085 * be TH_NEEDSYN. 1086 */ 1087 if (tp->t_state == TCPS_ESTABLISHED && 1088 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1089 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1090 ((to.to_flags & TOF_TS) == 0 || 1091 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1092 /* 1093 * Using the CC option is compulsory if once started: 1094 * the segment is OK if no T/TCP was negotiated or 1095 * if the segment has a CC option equal to CCrecv 1096 */ 1097 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1098 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1099 th->th_seq == tp->rcv_nxt && 1100 tiwin && tiwin == tp->snd_wnd && 1101 tp->snd_nxt == tp->snd_max) { 1102 1103 /* 1104 * If last ACK falls within this segment's sequence numbers, 1105 * record the timestamp. 1106 * NOTE that the test is modified according to the latest 1107 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1108 */ 1109 if ((to.to_flags & TOF_TS) != 0 && 1110 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1111 tp->ts_recent_age = ticks; 1112 tp->ts_recent = to.to_tsval; 1113 } 1114 1115 if (tlen == 0) { 1116 if (SEQ_GT(th->th_ack, tp->snd_una) && 1117 SEQ_LEQ(th->th_ack, tp->snd_max) && 1118 tp->snd_cwnd >= tp->snd_wnd && 1119 ((!tcp_do_newreno && 1120 tp->t_dupacks < tcprexmtthresh) || 1121 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1122 KASSERT(headlocked, ("headlocked")); 1123 INP_INFO_WUNLOCK(&tcbinfo); 1124 /* 1125 * this is a pure ack for outstanding data. 1126 */ 1127 ++tcpstat.tcps_predack; 1128 /* 1129 * "bad retransmit" recovery 1130 */ 1131 if (tp->t_rxtshift == 1 && 1132 ticks < tp->t_badrxtwin) { 1133 ++tcpstat.tcps_sndrexmitbad; 1134 tp->snd_cwnd = tp->snd_cwnd_prev; 1135 tp->snd_ssthresh = 1136 tp->snd_ssthresh_prev; 1137 tp->snd_recover = tp->snd_recover_prev; 1138 if (tp->t_flags & TF_WASFRECOVERY) 1139 ENTER_FASTRECOVERY(tp); 1140 tp->snd_nxt = tp->snd_max; 1141 tp->t_badrxtwin = 0; 1142 } 1143 1144 /* 1145 * Recalculate the transmit timer / rtt. 1146 * 1147 * Some boxes send broken timestamp replies 1148 * during the SYN+ACK phase, ignore 1149 * timestamps of 0 or we could calculate a 1150 * huge RTT and blow up the retransmit timer. 1151 */ 1152 if ((to.to_flags & TOF_TS) != 0 && 1153 to.to_tsecr) { 1154 tcp_xmit_timer(tp, 1155 ticks - to.to_tsecr + 1); 1156 } else if (tp->t_rtttime && 1157 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1158 tcp_xmit_timer(tp, 1159 ticks - tp->t_rtttime); 1160 } 1161 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1162 acked = th->th_ack - tp->snd_una; 1163 tcpstat.tcps_rcvackpack++; 1164 tcpstat.tcps_rcvackbyte += acked; 1165 sbdrop(&so->so_snd, acked); 1166 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1167 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1168 tp->snd_recover = th->th_ack - 1; 1169 tp->snd_una = th->th_ack; 1170 /* 1171 * pull snd_wl2 up to prevent seq wrap relative 1172 * to th_ack. 1173 */ 1174 tp->snd_wl2 = th->th_ack; 1175 tp->t_dupacks = 0; 1176 m_freem(m); 1177 ND6_HINT(tp); /* some progress has been done */ 1178 1179 /* 1180 * If all outstanding data are acked, stop 1181 * retransmit timer, otherwise restart timer 1182 * using current (possibly backed-off) value. 1183 * If process is waiting for space, 1184 * wakeup/selwakeup/signal. If data 1185 * are ready to send, let tcp_output 1186 * decide between more output or persist. 1187 1188#ifdef TCPDEBUG 1189 if (so->so_options & SO_DEBUG) 1190 tcp_trace(TA_INPUT, ostate, tp, 1191 (void *)tcp_saveipgen, 1192 &tcp_savetcp, 0); 1193#endif 1194 */ 1195 if (tp->snd_una == tp->snd_max) 1196 callout_stop(tp->tt_rexmt); 1197 else if (!callout_active(tp->tt_persist)) 1198 callout_reset(tp->tt_rexmt, 1199 tp->t_rxtcur, 1200 tcp_timer_rexmt, tp); 1201 1202 sowwakeup(so); 1203 if (so->so_snd.sb_cc) 1204 (void) tcp_output(tp); 1205 goto check_delack; 1206 } 1207 } else if (th->th_ack == tp->snd_una && 1208 LIST_EMPTY(&tp->t_segq) && 1209 tlen <= sbspace(&so->so_rcv)) { 1210 KASSERT(headlocked, ("headlocked")); 1211 INP_INFO_WUNLOCK(&tcbinfo); 1212 /* 1213 * this is a pure, in-sequence data packet 1214 * with nothing on the reassembly queue and 1215 * we have enough buffer space to take it. 1216 */ 1217 ++tcpstat.tcps_preddat; 1218 tp->rcv_nxt += tlen; 1219 /* 1220 * Pull snd_wl1 up to prevent seq wrap relative to 1221 * th_seq. 1222 */ 1223 tp->snd_wl1 = th->th_seq; 1224 /* 1225 * Pull rcv_up up to prevent seq wrap relative to 1226 * rcv_nxt. 1227 */ 1228 tp->rcv_up = tp->rcv_nxt; 1229 tcpstat.tcps_rcvpack++; 1230 tcpstat.tcps_rcvbyte += tlen; 1231 ND6_HINT(tp); /* some progress has been done */ 1232 /* 1233#ifdef TCPDEBUG 1234 if (so->so_options & SO_DEBUG) 1235 tcp_trace(TA_INPUT, ostate, tp, 1236 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1237#endif 1238 * Add data to socket buffer. 1239 */
| 359 m_freem(q->tqe_m); 360 else 361 sbappendstream(&so->so_rcv, q->tqe_m); 362 uma_zfree(tcp_reass_zone, q); 363 tp->t_segqlen--; 364 tcp_reass_qsize--; 365 q = nq; 366 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 367 ND6_HINT(tp); 368 sorwakeup(so); 369 return (flags); 370} 371 372/* 373 * TCP input routine, follows pages 65-76 of the 374 * protocol specification dated September, 1981 very closely. 375 */ 376#ifdef INET6 377int 378tcp6_input(mp, offp, proto) 379 struct mbuf **mp; 380 int *offp, proto; 381{ 382 register struct mbuf *m = *mp; 383 struct in6_ifaddr *ia6; 384 385 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 386 387 /* 388 * draft-itojun-ipv6-tcp-to-anycast 389 * better place to put this in? 390 */ 391 ia6 = ip6_getdstifaddr(m); 392 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 393 struct ip6_hdr *ip6; 394 395 ip6 = mtod(m, struct ip6_hdr *); 396 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 397 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 398 return IPPROTO_DONE; 399 } 400 401 tcp_input(m, *offp); 402 return IPPROTO_DONE; 403} 404#endif 405 406void 407tcp_input(m, off0) 408 register struct mbuf *m; 409 int off0; 410{ 411 register struct tcphdr *th; 412 register struct ip *ip = NULL; 413 register struct ipovly *ipov; 414 register struct inpcb *inp = NULL; 415 u_char *optp = NULL; 416 int optlen = 0; 417 int len, tlen, off; 418 int drop_hdrlen; 419 register struct tcpcb *tp = 0; 420 register int thflags; 421 struct socket *so = 0; 422 int todrop, acked, ourfinisacked, needoutput = 0; 423 u_long tiwin; 424 struct tcpopt to; /* options in this segment */ 425 struct rmxp_tao tao; /* our TAO cache entry */ 426 int headlocked = 0; 427 struct sockaddr_in *next_hop = NULL; 428 int rstreason; /* For badport_bandlim accounting purposes */ 429 430 struct ip6_hdr *ip6 = NULL; 431#ifdef INET6 432 int isipv6; 433#else 434 const int isipv6 = 0; 435#endif 436 437#ifdef TCPDEBUG 438 /* 439 * The size of tcp_saveipgen must be the size of the max ip header, 440 * now IPv6. 441 */ 442 u_char tcp_saveipgen[40]; 443 struct tcphdr tcp_savetcp; 444 short ostate = 0; 445#endif 446 447 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 448 next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD); 449#ifdef INET6 450 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 451#endif 452 bzero(&tao, sizeof(tao)); 453 bzero((char *)&to, sizeof(to)); 454 455 tcpstat.tcps_rcvtotal++; 456 457 if (isipv6) { 458#ifdef INET6 459 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 460 ip6 = mtod(m, struct ip6_hdr *); 461 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 462 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 463 tcpstat.tcps_rcvbadsum++; 464 goto drop; 465 } 466 th = (struct tcphdr *)((caddr_t)ip6 + off0); 467 468 /* 469 * Be proactive about unspecified IPv6 address in source. 470 * As we use all-zero to indicate unbounded/unconnected pcb, 471 * unspecified IPv6 address can be used to confuse us. 472 * 473 * Note that packets with unspecified IPv6 destination is 474 * already dropped in ip6_input. 475 */ 476 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 477 /* XXX stat */ 478 goto drop; 479 } 480#else 481 th = NULL; /* XXX: avoid compiler warning */ 482#endif 483 } else { 484 /* 485 * Get IP and TCP header together in first mbuf. 486 * Note: IP leaves IP header in first mbuf. 487 */ 488 if (off0 > sizeof (struct ip)) { 489 ip_stripoptions(m, (struct mbuf *)0); 490 off0 = sizeof(struct ip); 491 } 492 if (m->m_len < sizeof (struct tcpiphdr)) { 493 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 494 tcpstat.tcps_rcvshort++; 495 return; 496 } 497 } 498 ip = mtod(m, struct ip *); 499 ipov = (struct ipovly *)ip; 500 th = (struct tcphdr *)((caddr_t)ip + off0); 501 tlen = ip->ip_len; 502 503 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 504 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 505 th->th_sum = m->m_pkthdr.csum_data; 506 else 507 th->th_sum = in_pseudo(ip->ip_src.s_addr, 508 ip->ip_dst.s_addr, 509 htonl(m->m_pkthdr.csum_data + 510 ip->ip_len + 511 IPPROTO_TCP)); 512 th->th_sum ^= 0xffff; 513#ifdef TCPDEBUG 514 ipov->ih_len = (u_short)tlen; 515 ipov->ih_len = htons(ipov->ih_len); 516#endif 517 } else { 518 /* 519 * Checksum extended TCP header and data. 520 */ 521 len = sizeof (struct ip) + tlen; 522 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525 th->th_sum = in_cksum(m, len); 526 } 527 if (th->th_sum) { 528 tcpstat.tcps_rcvbadsum++; 529 goto drop; 530 } 531#ifdef INET6 532 /* Re-initialization for later version check */ 533 ip->ip_v = IPVERSION; 534#endif 535 } 536 537 /* 538 * Check that TCP offset makes sense, 539 * pull out TCP options and adjust length. XXX 540 */ 541 off = th->th_off << 2; 542 if (off < sizeof (struct tcphdr) || off > tlen) { 543 tcpstat.tcps_rcvbadoff++; 544 goto drop; 545 } 546 tlen -= off; /* tlen is used instead of ti->ti_len */ 547 if (off > sizeof (struct tcphdr)) { 548 if (isipv6) { 549#ifdef INET6 550 IP6_EXTHDR_CHECK(m, off0, off, ); 551 ip6 = mtod(m, struct ip6_hdr *); 552 th = (struct tcphdr *)((caddr_t)ip6 + off0); 553#endif 554 } else { 555 if (m->m_len < sizeof(struct ip) + off) { 556 if ((m = m_pullup(m, sizeof (struct ip) + off)) 557 == 0) { 558 tcpstat.tcps_rcvshort++; 559 return; 560 } 561 ip = mtod(m, struct ip *); 562 ipov = (struct ipovly *)ip; 563 th = (struct tcphdr *)((caddr_t)ip + off0); 564 } 565 } 566 optlen = off - sizeof (struct tcphdr); 567 optp = (u_char *)(th + 1); 568 } 569 thflags = th->th_flags; 570 571#ifdef TCP_DROP_SYNFIN 572 /* 573 * If the drop_synfin option is enabled, drop all packets with 574 * both the SYN and FIN bits set. This prevents e.g. nmap from 575 * identifying the TCP/IP stack. 576 * 577 * This is a violation of the TCP specification. 578 */ 579 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 580 goto drop; 581#endif 582 583 /* 584 * Convert TCP protocol specific fields to host format. 585 */ 586 th->th_seq = ntohl(th->th_seq); 587 th->th_ack = ntohl(th->th_ack); 588 th->th_win = ntohs(th->th_win); 589 th->th_urp = ntohs(th->th_urp); 590 591 /* 592 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 593 * until after ip6_savecontrol() is called and before other functions 594 * which don't want those proto headers. 595 * Because ip6_savecontrol() is going to parse the mbuf to 596 * search for data to be passed up to user-land, it wants mbuf 597 * parameters to be unchanged. 598 * XXX: the call of ip6_savecontrol() has been obsoleted based on 599 * latest version of the advanced API (20020110). 600 */ 601 drop_hdrlen = off0 + off; 602 603 /* 604 * Locate pcb for segment. 605 */ 606 INP_INFO_WLOCK(&tcbinfo); 607 headlocked = 1; 608findpcb: 609 /* IPFIREWALL_FORWARD section */ 610 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 611 /* 612 * Transparently forwarded. Pretend to be the destination. 613 * already got one like this? 614 */ 615 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 616 ip->ip_dst, th->th_dport, 617 0, m->m_pkthdr.rcvif); 618 if (!inp) { 619 /* It's new. Try find the ambushing socket. */ 620 inp = in_pcblookup_hash(&tcbinfo, 621 ip->ip_src, th->th_sport, 622 next_hop->sin_addr, 623 next_hop->sin_port ? 624 ntohs(next_hop->sin_port) : 625 th->th_dport, 626 1, m->m_pkthdr.rcvif); 627 } 628 } else { 629 if (isipv6) { 630#ifdef INET6 631 inp = in6_pcblookup_hash(&tcbinfo, 632 &ip6->ip6_src, th->th_sport, 633 &ip6->ip6_dst, th->th_dport, 634 1, m->m_pkthdr.rcvif); 635#endif 636 } else 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 ip->ip_dst, th->th_dport, 640 1, m->m_pkthdr.rcvif); 641 } 642 643#if defined(IPSEC) || defined(FAST_IPSEC) 644 if (isipv6) { 645 if (inp != NULL && ipsec6_in_reject(m, inp)) { 646#ifdef IPSEC 647 ipsec6stat.in_polvio++; 648#endif /*IPSEC*/ 649 goto drop; 650 } 651 } else if (inp != NULL && ipsec4_in_reject(m, inp)) { 652#ifdef IPSEC 653 ipsecstat.in_polvio++; 654#endif /*IPSEC*/ 655 goto drop; 656 } 657#endif /*IPSEC || FAST_IPSEC*/ 658 659 /* 660 * If the state is CLOSED (i.e., TCB does not exist) then 661 * all data in the incoming segment is discarded. 662 * If the TCB exists but is in CLOSED state, it is embryonic, 663 * but should either do a listen or a connect soon. 664 */ 665 if (inp == NULL) { 666 if (log_in_vain) { 667#ifdef INET6 668 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 669#else 670 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 671#endif 672 673 if (isipv6) { 674#ifdef INET6 675 strcpy(dbuf, "["); 676 strcpy(sbuf, "["); 677 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 678 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 679 strcat(dbuf, "]"); 680 strcat(sbuf, "]"); 681#endif 682 } else { 683 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 684 strcpy(sbuf, inet_ntoa(ip->ip_src)); 685 } 686 switch (log_in_vain) { 687 case 1: 688 if ((thflags & TH_SYN) == 0) 689 break; 690 /* FALLTHROUGH */ 691 case 2: 692 log(LOG_INFO, 693 "Connection attempt to TCP %s:%d " 694 "from %s:%d flags:0x%02x\n", 695 dbuf, ntohs(th->th_dport), sbuf, 696 ntohs(th->th_sport), thflags); 697 break; 698 default: 699 break; 700 } 701 } 702 if (blackhole) { 703 switch (blackhole) { 704 case 1: 705 if (thflags & TH_SYN) 706 goto drop; 707 break; 708 case 2: 709 goto drop; 710 default: 711 goto drop; 712 } 713 } 714 rstreason = BANDLIM_RST_CLOSEDPORT; 715 goto dropwithreset; 716 } 717 INP_LOCK(inp); 718 if (inp->inp_vflag & INP_TIMEWAIT) { 719 /* 720 * The only option of relevance is TOF_CC, and only if 721 * present in a SYN segment. See tcp_timewait(). 722 */ 723 if (thflags & TH_SYN) 724 tcp_dooptions(&to, optp, optlen, 1); 725 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 726 &to, th, m, tlen)) 727 goto findpcb; 728 /* 729 * tcp_timewait unlocks inp. 730 */ 731 INP_INFO_WUNLOCK(&tcbinfo); 732 return; 733 } 734 tp = intotcpcb(inp); 735 if (tp == 0) { 736 INP_UNLOCK(inp); 737 rstreason = BANDLIM_RST_CLOSEDPORT; 738 goto dropwithreset; 739 } 740 if (tp->t_state == TCPS_CLOSED) 741 goto drop; 742 743 /* Unscale the window into a 32-bit value. */ 744 if ((thflags & TH_SYN) == 0) 745 tiwin = th->th_win << tp->snd_scale; 746 else 747 tiwin = th->th_win; 748 749#ifdef MAC 750 if (mac_check_inpcb_deliver(inp, m)) 751 goto drop; 752#endif 753 so = inp->inp_socket; 754#ifdef TCPDEBUG 755 if (so->so_options & SO_DEBUG) { 756 ostate = tp->t_state; 757 if (isipv6) 758 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 759 else 760 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 761 tcp_savetcp = *th; 762 } 763#endif 764 if (so->so_options & SO_ACCEPTCONN) { 765 struct in_conninfo inc; 766 767#ifdef INET6 768 inc.inc_isipv6 = isipv6; 769#endif 770 if (isipv6) { 771 inc.inc6_faddr = ip6->ip6_src; 772 inc.inc6_laddr = ip6->ip6_dst; 773 } else { 774 inc.inc_faddr = ip->ip_src; 775 inc.inc_laddr = ip->ip_dst; 776 } 777 inc.inc_fport = th->th_sport; 778 inc.inc_lport = th->th_dport; 779 780 /* 781 * If the state is LISTEN then ignore segment if it contains 782 * a RST. If the segment contains an ACK then it is bad and 783 * send a RST. If it does not contain a SYN then it is not 784 * interesting; drop it. 785 * 786 * If the state is SYN_RECEIVED (syncache) and seg contains 787 * an ACK, but not for our SYN/ACK, send a RST. If the seg 788 * contains a RST, check the sequence number to see if it 789 * is a valid reset segment. 790 */ 791 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 792 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 793 if (!syncache_expand(&inc, th, &so, m)) { 794 /* 795 * No syncache entry, or ACK was not 796 * for our SYN/ACK. Send a RST. 797 */ 798 tcpstat.tcps_badsyn++; 799 rstreason = BANDLIM_RST_OPENPORT; 800 goto dropwithreset; 801 } 802 if (so == NULL) { 803 /* 804 * Could not complete 3-way handshake, 805 * connection is being closed down, and 806 * syncache will free mbuf. 807 */ 808 INP_UNLOCK(inp); 809 INP_INFO_WUNLOCK(&tcbinfo); 810 return; 811 } 812 /* 813 * Socket is created in state SYN_RECEIVED. 814 * Continue processing segment. 815 */ 816 INP_UNLOCK(inp); 817 inp = sotoinpcb(so); 818 INP_LOCK(inp); 819 tp = intotcpcb(inp); 820 /* 821 * This is what would have happened in 822 * tcp_output() when the SYN,ACK was sent. 823 */ 824 tp->snd_up = tp->snd_una; 825 tp->snd_max = tp->snd_nxt = tp->iss + 1; 826 tp->last_ack_sent = tp->rcv_nxt; 827 /* 828 * RFC1323: The window in SYN & SYN/ACK 829 * segments is never scaled. 830 */ 831 tp->snd_wnd = tiwin; /* unscaled */ 832 goto after_listen; 833 } 834 if (thflags & TH_RST) { 835 syncache_chkrst(&inc, th); 836 goto drop; 837 } 838 if (thflags & TH_ACK) { 839 syncache_badack(&inc); 840 tcpstat.tcps_badsyn++; 841 rstreason = BANDLIM_RST_OPENPORT; 842 goto dropwithreset; 843 } 844 goto drop; 845 } 846 847 /* 848 * Segment's flags are (SYN) or (SYN|FIN). 849 */ 850#ifdef INET6 851 /* 852 * If deprecated address is forbidden, 853 * we do not accept SYN to deprecated interface 854 * address to prevent any new inbound connection from 855 * getting established. 856 * When we do not accept SYN, we send a TCP RST, 857 * with deprecated source address (instead of dropping 858 * it). We compromise it as it is much better for peer 859 * to send a RST, and RST will be the final packet 860 * for the exchange. 861 * 862 * If we do not forbid deprecated addresses, we accept 863 * the SYN packet. RFC2462 does not suggest dropping 864 * SYN in this case. 865 * If we decipher RFC2462 5.5.4, it says like this: 866 * 1. use of deprecated addr with existing 867 * communication is okay - "SHOULD continue to be 868 * used" 869 * 2. use of it with new communication: 870 * (2a) "SHOULD NOT be used if alternate address 871 * with sufficient scope is available" 872 * (2b) nothing mentioned otherwise. 873 * Here we fall into (2b) case as we have no choice in 874 * our source address selection - we must obey the peer. 875 * 876 * The wording in RFC2462 is confusing, and there are 877 * multiple description text for deprecated address 878 * handling - worse, they are not exactly the same. 879 * I believe 5.5.4 is the best one, so we follow 5.5.4. 880 */ 881 if (isipv6 && !ip6_use_deprecated) { 882 struct in6_ifaddr *ia6; 883 884 if ((ia6 = ip6_getdstifaddr(m)) && 885 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 886 INP_UNLOCK(inp); 887 tp = NULL; 888 rstreason = BANDLIM_RST_OPENPORT; 889 goto dropwithreset; 890 } 891 } 892#endif 893 /* 894 * If it is from this socket, drop it, it must be forged. 895 * Don't bother responding if the destination was a broadcast. 896 */ 897 if (th->th_dport == th->th_sport) { 898 if (isipv6) { 899 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 900 &ip6->ip6_src)) 901 goto drop; 902 } else { 903 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 904 goto drop; 905 } 906 } 907 /* 908 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 909 * 910 * Note that it is quite possible to receive unicast 911 * link-layer packets with a broadcast IP address. Use 912 * in_broadcast() to find them. 913 */ 914 if (m->m_flags & (M_BCAST|M_MCAST)) 915 goto drop; 916 if (isipv6) { 917 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 918 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 919 goto drop; 920 } else { 921 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 922 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 923 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 924 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 925 goto drop; 926 } 927 /* 928 * SYN appears to be valid; create compressed TCP state 929 * for syncache, or perform t/tcp connection. 930 */ 931 if (so->so_qlen <= so->so_qlimit) { 932#ifdef TCPDEBUG 933 if (so->so_options & SO_DEBUG) 934 tcp_trace(TA_INPUT, ostate, tp, 935 (void *)tcp_saveipgen, &tcp_savetcp, 0); 936#endif 937 tcp_dooptions(&to, optp, optlen, 1); 938 if (!syncache_add(&inc, &to, th, &so, m)) 939 goto drop; 940 if (so == NULL) { 941 /* 942 * Entry added to syncache, mbuf used to 943 * send SYN,ACK packet. 944 */ 945 KASSERT(headlocked, ("headlocked")); 946 INP_UNLOCK(inp); 947 INP_INFO_WUNLOCK(&tcbinfo); 948 return; 949 } 950 /* 951 * Segment passed TAO tests. 952 */ 953 INP_UNLOCK(inp); 954 inp = sotoinpcb(so); 955 INP_LOCK(inp); 956 tp = intotcpcb(inp); 957 tp->snd_wnd = tiwin; 958 tp->t_starttime = ticks; 959 tp->t_state = TCPS_ESTABLISHED; 960 961 /* 962 * T/TCP logic: 963 * If there is a FIN or if there is data, then 964 * delay SYN,ACK(SYN) in the hope of piggy-backing 965 * it on a response segment. Otherwise must send 966 * ACK now in case the other side is slow starting. 967 */ 968 if (thflags & TH_FIN || tlen != 0) 969 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 970 else 971 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 972 tcpstat.tcps_connects++; 973 soisconnected(so); 974 goto trimthenstep6; 975 } 976 goto drop; 977 } 978after_listen: 979 980 /* XXX temp debugging */ 981 /* should not happen - syncache should pick up these connections */ 982 if (tp->t_state == TCPS_LISTEN) 983 panic("tcp_input: TCPS_LISTEN"); 984 985 /* 986 * This is the second part of the MSS DoS prevention code (after 987 * minmss on the sending side) and it deals with too many too small 988 * tcp packets in a too short timeframe (1 second). 989 * 990 * For every full second we count the number of received packets 991 * and bytes. If we get a lot of packets per second for this connection 992 * (tcp_minmssoverload) we take a closer look at it and compute the 993 * average packet size for the past second. If that is less than 994 * tcp_minmss we get too many packets with very small payload which 995 * is not good and burdens our system (and every packet generates 996 * a wakeup to the process connected to our socket). We can reasonable 997 * expect this to be small packet DoS attack to exhaust our CPU 998 * cycles. 999 * 1000 * Care has to be taken for the minimum packet overload value. This 1001 * value defines the minimum number of packets per second before we 1002 * start to worry. This must not be too low to avoid killing for 1003 * example interactive connections with many small packets like 1004 * telnet or SSH. 1005 * 1006 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1007 * this check. 1008 * 1009 * Account for packet if payload packet, skip over ACK, etc. 1010 */ 1011 if (tcp_minmss && tcp_minmssoverload && 1012 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1013 if (tp->rcv_second > ticks) { 1014 tp->rcv_pps++; 1015 tp->rcv_byps += tlen + off; 1016 if (tp->rcv_pps > tcp_minmssoverload) { 1017 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1018 printf("too many small tcp packets from " 1019 "%s:%u, av. %lubyte/packet, " 1020 "dropping connection\n", 1021#ifdef INET6 1022 isipv6 ? 1023 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1024#endif 1025 inet_ntoa(inp->inp_inc.inc_faddr), 1026 inp->inp_inc.inc_fport, 1027 tp->rcv_byps / tp->rcv_pps); 1028 tp = tcp_drop(tp, ECONNRESET); 1029 tcpstat.tcps_minmssdrops++; 1030 goto drop; 1031 } 1032 } 1033 } else { 1034 tp->rcv_second = ticks + hz; 1035 tp->rcv_pps = 1; 1036 tp->rcv_byps = tlen + off; 1037 } 1038 } 1039 1040 /* 1041 * Segment received on connection. 1042 * Reset idle time and keep-alive timer. 1043 */ 1044 tp->t_rcvtime = ticks; 1045 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1046 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1047 1048 /* 1049 * Process options only when we get SYN/ACK back. The SYN case 1050 * for incoming connections is handled in tcp_syncache. 1051 * XXX this is traditional behavior, may need to be cleaned up. 1052 */ 1053 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1054 if (thflags & TH_SYN) { 1055 if (to.to_flags & TOF_SCALE) { 1056 tp->t_flags |= TF_RCVD_SCALE; 1057 tp->requested_s_scale = to.to_requested_s_scale; 1058 } 1059 if (to.to_flags & TOF_TS) { 1060 tp->t_flags |= TF_RCVD_TSTMP; 1061 tp->ts_recent = to.to_tsval; 1062 tp->ts_recent_age = ticks; 1063 } 1064 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1065 tp->t_flags |= TF_RCVD_CC; 1066 if (to.to_flags & TOF_MSS) 1067 tcp_mss(tp, to.to_mss); 1068 } 1069 1070 /* 1071 * Header prediction: check for the two common cases 1072 * of a uni-directional data xfer. If the packet has 1073 * no control flags, is in-sequence, the window didn't 1074 * change and we're not retransmitting, it's a 1075 * candidate. If the length is zero and the ack moved 1076 * forward, we're the sender side of the xfer. Just 1077 * free the data acked & wake any higher level process 1078 * that was blocked waiting for space. If the length 1079 * is non-zero and the ack didn't move, we're the 1080 * receiver side. If we're getting packets in-order 1081 * (the reassembly queue is empty), add the data to 1082 * the socket buffer and note that we need a delayed ack. 1083 * Make sure that the hidden state-flags are also off. 1084 * Since we check for TCPS_ESTABLISHED above, it can only 1085 * be TH_NEEDSYN. 1086 */ 1087 if (tp->t_state == TCPS_ESTABLISHED && 1088 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1089 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1090 ((to.to_flags & TOF_TS) == 0 || 1091 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1092 /* 1093 * Using the CC option is compulsory if once started: 1094 * the segment is OK if no T/TCP was negotiated or 1095 * if the segment has a CC option equal to CCrecv 1096 */ 1097 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1098 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1099 th->th_seq == tp->rcv_nxt && 1100 tiwin && tiwin == tp->snd_wnd && 1101 tp->snd_nxt == tp->snd_max) { 1102 1103 /* 1104 * If last ACK falls within this segment's sequence numbers, 1105 * record the timestamp. 1106 * NOTE that the test is modified according to the latest 1107 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1108 */ 1109 if ((to.to_flags & TOF_TS) != 0 && 1110 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1111 tp->ts_recent_age = ticks; 1112 tp->ts_recent = to.to_tsval; 1113 } 1114 1115 if (tlen == 0) { 1116 if (SEQ_GT(th->th_ack, tp->snd_una) && 1117 SEQ_LEQ(th->th_ack, tp->snd_max) && 1118 tp->snd_cwnd >= tp->snd_wnd && 1119 ((!tcp_do_newreno && 1120 tp->t_dupacks < tcprexmtthresh) || 1121 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1122 KASSERT(headlocked, ("headlocked")); 1123 INP_INFO_WUNLOCK(&tcbinfo); 1124 /* 1125 * this is a pure ack for outstanding data. 1126 */ 1127 ++tcpstat.tcps_predack; 1128 /* 1129 * "bad retransmit" recovery 1130 */ 1131 if (tp->t_rxtshift == 1 && 1132 ticks < tp->t_badrxtwin) { 1133 ++tcpstat.tcps_sndrexmitbad; 1134 tp->snd_cwnd = tp->snd_cwnd_prev; 1135 tp->snd_ssthresh = 1136 tp->snd_ssthresh_prev; 1137 tp->snd_recover = tp->snd_recover_prev; 1138 if (tp->t_flags & TF_WASFRECOVERY) 1139 ENTER_FASTRECOVERY(tp); 1140 tp->snd_nxt = tp->snd_max; 1141 tp->t_badrxtwin = 0; 1142 } 1143 1144 /* 1145 * Recalculate the transmit timer / rtt. 1146 * 1147 * Some boxes send broken timestamp replies 1148 * during the SYN+ACK phase, ignore 1149 * timestamps of 0 or we could calculate a 1150 * huge RTT and blow up the retransmit timer. 1151 */ 1152 if ((to.to_flags & TOF_TS) != 0 && 1153 to.to_tsecr) { 1154 tcp_xmit_timer(tp, 1155 ticks - to.to_tsecr + 1); 1156 } else if (tp->t_rtttime && 1157 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1158 tcp_xmit_timer(tp, 1159 ticks - tp->t_rtttime); 1160 } 1161 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1162 acked = th->th_ack - tp->snd_una; 1163 tcpstat.tcps_rcvackpack++; 1164 tcpstat.tcps_rcvackbyte += acked; 1165 sbdrop(&so->so_snd, acked); 1166 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1167 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1168 tp->snd_recover = th->th_ack - 1; 1169 tp->snd_una = th->th_ack; 1170 /* 1171 * pull snd_wl2 up to prevent seq wrap relative 1172 * to th_ack. 1173 */ 1174 tp->snd_wl2 = th->th_ack; 1175 tp->t_dupacks = 0; 1176 m_freem(m); 1177 ND6_HINT(tp); /* some progress has been done */ 1178 1179 /* 1180 * If all outstanding data are acked, stop 1181 * retransmit timer, otherwise restart timer 1182 * using current (possibly backed-off) value. 1183 * If process is waiting for space, 1184 * wakeup/selwakeup/signal. If data 1185 * are ready to send, let tcp_output 1186 * decide between more output or persist. 1187 1188#ifdef TCPDEBUG 1189 if (so->so_options & SO_DEBUG) 1190 tcp_trace(TA_INPUT, ostate, tp, 1191 (void *)tcp_saveipgen, 1192 &tcp_savetcp, 0); 1193#endif 1194 */ 1195 if (tp->snd_una == tp->snd_max) 1196 callout_stop(tp->tt_rexmt); 1197 else if (!callout_active(tp->tt_persist)) 1198 callout_reset(tp->tt_rexmt, 1199 tp->t_rxtcur, 1200 tcp_timer_rexmt, tp); 1201 1202 sowwakeup(so); 1203 if (so->so_snd.sb_cc) 1204 (void) tcp_output(tp); 1205 goto check_delack; 1206 } 1207 } else if (th->th_ack == tp->snd_una && 1208 LIST_EMPTY(&tp->t_segq) && 1209 tlen <= sbspace(&so->so_rcv)) { 1210 KASSERT(headlocked, ("headlocked")); 1211 INP_INFO_WUNLOCK(&tcbinfo); 1212 /* 1213 * this is a pure, in-sequence data packet 1214 * with nothing on the reassembly queue and 1215 * we have enough buffer space to take it. 1216 */ 1217 ++tcpstat.tcps_preddat; 1218 tp->rcv_nxt += tlen; 1219 /* 1220 * Pull snd_wl1 up to prevent seq wrap relative to 1221 * th_seq. 1222 */ 1223 tp->snd_wl1 = th->th_seq; 1224 /* 1225 * Pull rcv_up up to prevent seq wrap relative to 1226 * rcv_nxt. 1227 */ 1228 tp->rcv_up = tp->rcv_nxt; 1229 tcpstat.tcps_rcvpack++; 1230 tcpstat.tcps_rcvbyte += tlen; 1231 ND6_HINT(tp); /* some progress has been done */ 1232 /* 1233#ifdef TCPDEBUG 1234 if (so->so_options & SO_DEBUG) 1235 tcp_trace(TA_INPUT, ostate, tp, 1236 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1237#endif 1238 * Add data to socket buffer. 1239 */
|
1240 if (so->so_state & SS_CANTRCVMORE) {
| 1240 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
|
1241 m_freem(m); 1242 } else { 1243 m_adj(m, drop_hdrlen); /* delayed header drop */ 1244 sbappendstream(&so->so_rcv, m); 1245 } 1246 sorwakeup(so); 1247 if (DELAY_ACK(tp)) { 1248 tp->t_flags |= TF_DELACK; 1249 } else { 1250 tp->t_flags |= TF_ACKNOW; 1251 tcp_output(tp); 1252 } 1253 goto check_delack; 1254 } 1255 } 1256 1257 /* 1258 * Calculate amount of space in receive window, 1259 * and then do TCP input processing. 1260 * Receive window is amount of space in rcv queue, 1261 * but not less than advertised window. 1262 */ 1263 { int win; 1264 1265 win = sbspace(&so->so_rcv); 1266 if (win < 0) 1267 win = 0; 1268 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1269 } 1270 1271 switch (tp->t_state) { 1272 1273 /* 1274 * If the state is SYN_RECEIVED: 1275 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1276 */ 1277 case TCPS_SYN_RECEIVED: 1278 if ((thflags & TH_ACK) && 1279 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1280 SEQ_GT(th->th_ack, tp->snd_max))) { 1281 rstreason = BANDLIM_RST_OPENPORT; 1282 goto dropwithreset; 1283 } 1284 break; 1285 1286 /* 1287 * If the state is SYN_SENT: 1288 * if seg contains an ACK, but not for our SYN, drop the input. 1289 * if seg contains a RST, then drop the connection. 1290 * if seg does not contain SYN, then drop it. 1291 * Otherwise this is an acceptable SYN segment 1292 * initialize tp->rcv_nxt and tp->irs 1293 * if seg contains ack then advance tp->snd_una 1294 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1295 * arrange for segment to be acked (eventually) 1296 * continue processing rest of data/controls, beginning with URG 1297 */ 1298 case TCPS_SYN_SENT: 1299 if (tcp_do_rfc1644) 1300 tcp_hc_gettao(&inp->inp_inc, &tao); 1301 1302 if ((thflags & TH_ACK) && 1303 (SEQ_LEQ(th->th_ack, tp->iss) || 1304 SEQ_GT(th->th_ack, tp->snd_max))) { 1305 /* 1306 * If we have a cached CCsent for the remote host, 1307 * hence we haven't just crashed and restarted, 1308 * do not send a RST. This may be a retransmission 1309 * from the other side after our earlier ACK was lost. 1310 * Our new SYN, when it arrives, will serve as the 1311 * needed ACK. 1312 */ 1313 if (tao.tao_ccsent != 0) 1314 goto drop; 1315 else { 1316 rstreason = BANDLIM_UNLIMITED; 1317 goto dropwithreset; 1318 } 1319 } 1320 if (thflags & TH_RST) { 1321 if (thflags & TH_ACK) 1322 tp = tcp_drop(tp, ECONNREFUSED); 1323 goto drop; 1324 } 1325 if ((thflags & TH_SYN) == 0) 1326 goto drop; 1327 tp->snd_wnd = th->th_win; /* initial send window */ 1328 tp->cc_recv = to.to_cc; /* foreign CC */ 1329 1330 tp->irs = th->th_seq; 1331 tcp_rcvseqinit(tp); 1332 if (thflags & TH_ACK) { 1333 /* 1334 * Our SYN was acked. If segment contains CC.ECHO 1335 * option, check it to make sure this segment really 1336 * matches our SYN. If not, just drop it as old 1337 * duplicate, but send an RST if we're still playing 1338 * by the old rules. If no CC.ECHO option, make sure 1339 * we don't get fooled into using T/TCP. 1340 */ 1341 if (to.to_flags & TOF_CCECHO) { 1342 if (tp->cc_send != to.to_ccecho) { 1343 if (tao.tao_ccsent != 0) 1344 goto drop; 1345 else { 1346 rstreason = BANDLIM_UNLIMITED; 1347 goto dropwithreset; 1348 } 1349 } 1350 } else 1351 tp->t_flags &= ~TF_RCVD_CC; 1352 tcpstat.tcps_connects++; 1353 soisconnected(so); 1354#ifdef MAC 1355 SOCK_LOCK(so); 1356 mac_set_socket_peer_from_mbuf(m, so); 1357 SOCK_UNLOCK(so); 1358#endif 1359 /* Do window scaling on this connection? */ 1360 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1361 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1362 tp->snd_scale = tp->requested_s_scale; 1363 tp->rcv_scale = tp->request_r_scale; 1364 } 1365 /* Segment is acceptable, update cache if undefined. */ 1366 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1367 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1368 1369 tp->rcv_adv += tp->rcv_wnd; 1370 tp->snd_una++; /* SYN is acked */ 1371 /* 1372 * If there's data, delay ACK; if there's also a FIN 1373 * ACKNOW will be turned on later. 1374 */ 1375 if (DELAY_ACK(tp) && tlen != 0) 1376 callout_reset(tp->tt_delack, tcp_delacktime, 1377 tcp_timer_delack, tp); 1378 else 1379 tp->t_flags |= TF_ACKNOW; 1380 /* 1381 * Received <SYN,ACK> in SYN_SENT[*] state. 1382 * Transitions: 1383 * SYN_SENT --> ESTABLISHED 1384 * SYN_SENT* --> FIN_WAIT_1 1385 */ 1386 tp->t_starttime = ticks; 1387 if (tp->t_flags & TF_NEEDFIN) { 1388 tp->t_state = TCPS_FIN_WAIT_1; 1389 tp->t_flags &= ~TF_NEEDFIN; 1390 thflags &= ~TH_SYN; 1391 } else { 1392 tp->t_state = TCPS_ESTABLISHED; 1393 callout_reset(tp->tt_keep, tcp_keepidle, 1394 tcp_timer_keep, tp); 1395 } 1396 } else { 1397 /* 1398 * Received initial SYN in SYN-SENT[*] state => 1399 * simultaneous open. If segment contains CC option 1400 * and there is a cached CC, apply TAO test. 1401 * If it succeeds, connection is * half-synchronized. 1402 * Otherwise, do 3-way handshake: 1403 * SYN-SENT -> SYN-RECEIVED 1404 * SYN-SENT* -> SYN-RECEIVED* 1405 * If there was no CC option, clear cached CC value. 1406 */ 1407 tp->t_flags |= TF_ACKNOW; 1408 callout_stop(tp->tt_rexmt); 1409 if (to.to_flags & TOF_CC) { 1410 if (tao.tao_cc != 0 && 1411 CC_GT(to.to_cc, tao.tao_cc)) { 1412 /* 1413 * update cache and make transition: 1414 * SYN-SENT -> ESTABLISHED* 1415 * SYN-SENT* -> FIN-WAIT-1* 1416 */ 1417 tao.tao_cc = to.to_cc; 1418 tcp_hc_updatetao(&inp->inp_inc, 1419 TCP_HC_TAO_CC, to.to_cc, 0); 1420 tp->t_starttime = ticks; 1421 if (tp->t_flags & TF_NEEDFIN) { 1422 tp->t_state = TCPS_FIN_WAIT_1; 1423 tp->t_flags &= ~TF_NEEDFIN; 1424 } else { 1425 tp->t_state = TCPS_ESTABLISHED; 1426 callout_reset(tp->tt_keep, 1427 tcp_keepidle, 1428 tcp_timer_keep, 1429 tp); 1430 } 1431 tp->t_flags |= TF_NEEDSYN; 1432 } else 1433 tp->t_state = TCPS_SYN_RECEIVED; 1434 } else { 1435 if (tcp_do_rfc1644) { 1436 /* CC.NEW or no option => invalidate cache */ 1437 tao.tao_cc = 0; 1438 tcp_hc_updatetao(&inp->inp_inc, 1439 TCP_HC_TAO_CC, to.to_cc, 0); 1440 } 1441 tp->t_state = TCPS_SYN_RECEIVED; 1442 } 1443 } 1444 1445trimthenstep6: 1446 /* 1447 * Advance th->th_seq to correspond to first data byte. 1448 * If data, trim to stay within window, 1449 * dropping FIN if necessary. 1450 */ 1451 th->th_seq++; 1452 if (tlen > tp->rcv_wnd) { 1453 todrop = tlen - tp->rcv_wnd; 1454 m_adj(m, -todrop); 1455 tlen = tp->rcv_wnd; 1456 thflags &= ~TH_FIN; 1457 tcpstat.tcps_rcvpackafterwin++; 1458 tcpstat.tcps_rcvbyteafterwin += todrop; 1459 } 1460 tp->snd_wl1 = th->th_seq - 1; 1461 tp->rcv_up = th->th_seq; 1462 /* 1463 * Client side of transaction: already sent SYN and data. 1464 * If the remote host used T/TCP to validate the SYN, 1465 * our data will be ACK'd; if so, enter normal data segment 1466 * processing in the middle of step 5, ack processing. 1467 * Otherwise, goto step 6. 1468 */ 1469 if (thflags & TH_ACK) 1470 goto process_ACK; 1471 1472 goto step6; 1473 1474 /* 1475 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1476 * if segment contains a SYN and CC [not CC.NEW] option: 1477 * if state == TIME_WAIT and connection duration > MSL, 1478 * drop packet and send RST; 1479 * 1480 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1481 * ack the FIN (and data) in retransmission queue. 1482 * Complete close and delete TCPCB. Then reprocess 1483 * segment, hoping to find new TCPCB in LISTEN state; 1484 * 1485 * else must be old SYN; drop it. 1486 * else do normal processing. 1487 */ 1488 case TCPS_LAST_ACK: 1489 case TCPS_CLOSING: 1490 case TCPS_TIME_WAIT: 1491 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1492 if ((thflags & TH_SYN) && 1493 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1494 if (tp->t_state == TCPS_TIME_WAIT && 1495 (ticks - tp->t_starttime) > tcp_msl) { 1496 rstreason = BANDLIM_UNLIMITED; 1497 goto dropwithreset; 1498 } 1499 if (CC_GT(to.to_cc, tp->cc_recv)) { 1500 tp = tcp_close(tp); 1501 goto findpcb; 1502 } 1503 else 1504 goto drop; 1505 } 1506 break; /* continue normal processing */ 1507 } 1508 1509 /* 1510 * States other than LISTEN or SYN_SENT. 1511 * First check the RST flag and sequence number since reset segments 1512 * are exempt from the timestamp and connection count tests. This 1513 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1514 * below which allowed reset segments in half the sequence space 1515 * to fall though and be processed (which gives forged reset 1516 * segments with a random sequence number a 50 percent chance of 1517 * killing a connection). 1518 * Then check timestamp, if present. 1519 * Then check the connection count, if present. 1520 * Then check that at least some bytes of segment are within 1521 * receive window. If segment begins before rcv_nxt, 1522 * drop leading data (and SYN); if nothing left, just ack. 1523 * 1524 * 1525 * If the RST bit is set, check the sequence number to see 1526 * if this is a valid reset segment. 1527 * RFC 793 page 37: 1528 * In all states except SYN-SENT, all reset (RST) segments 1529 * are validated by checking their SEQ-fields. A reset is 1530 * valid if its sequence number is in the window. 1531 * Note: this does not take into account delayed ACKs, so 1532 * we should test against last_ack_sent instead of rcv_nxt. 1533 * The sequence number in the reset segment is normally an 1534 * echo of our outgoing acknowlegement numbers, but some hosts 1535 * send a reset with the sequence number at the rightmost edge 1536 * of our receive window, and we have to handle this case. 1537 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1538 * that brute force RST attacks are possible. To combat this, 1539 * we use a much stricter check while in the ESTABLISHED state, 1540 * only accepting RSTs where the sequence number is equal to 1541 * last_ack_sent. In all other states (the states in which a 1542 * RST is more likely), the more permissive check is used. 1543 * If we have multiple segments in flight, the intial reset 1544 * segment sequence numbers will be to the left of last_ack_sent, 1545 * but they will eventually catch up. 1546 * In any case, it never made sense to trim reset segments to 1547 * fit the receive window since RFC 1122 says: 1548 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1549 * 1550 * A TCP SHOULD allow a received RST segment to include data. 1551 * 1552 * DISCUSSION 1553 * It has been suggested that a RST segment could contain 1554 * ASCII text that encoded and explained the cause of the 1555 * RST. No standard has yet been established for such 1556 * data. 1557 * 1558 * If the reset segment passes the sequence number test examine 1559 * the state: 1560 * SYN_RECEIVED STATE: 1561 * If passive open, return to LISTEN state. 1562 * If active open, inform user that connection was refused. 1563 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1564 * Inform user that connection was reset, and close tcb. 1565 * CLOSING, LAST_ACK STATES: 1566 * Close the tcb. 1567 * TIME_WAIT STATE: 1568 * Drop the segment - see Stevens, vol. 2, p. 964 and 1569 * RFC 1337. 1570 */ 1571 if (thflags & TH_RST) { 1572 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1573 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1574 switch (tp->t_state) { 1575 1576 case TCPS_SYN_RECEIVED: 1577 so->so_error = ECONNREFUSED; 1578 goto close; 1579 1580 case TCPS_ESTABLISHED: 1581 if (tp->last_ack_sent != th->th_seq) { 1582 tcpstat.tcps_badrst++; 1583 goto drop; 1584 } 1585 case TCPS_FIN_WAIT_1: 1586 case TCPS_FIN_WAIT_2: 1587 case TCPS_CLOSE_WAIT: 1588 so->so_error = ECONNRESET; 1589 close: 1590 tp->t_state = TCPS_CLOSED; 1591 tcpstat.tcps_drops++; 1592 tp = tcp_close(tp); 1593 break; 1594 1595 case TCPS_CLOSING: 1596 case TCPS_LAST_ACK: 1597 tp = tcp_close(tp); 1598 break; 1599 1600 case TCPS_TIME_WAIT: 1601 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1602 ("timewait")); 1603 break; 1604 } 1605 } 1606 goto drop; 1607 } 1608 1609 /* 1610 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1611 * and it's less than ts_recent, drop it. 1612 */ 1613 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1614 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1615 1616 /* Check to see if ts_recent is over 24 days old. */ 1617 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1618 /* 1619 * Invalidate ts_recent. If this segment updates 1620 * ts_recent, the age will be reset later and ts_recent 1621 * will get a valid value. If it does not, setting 1622 * ts_recent to zero will at least satisfy the 1623 * requirement that zero be placed in the timestamp 1624 * echo reply when ts_recent isn't valid. The 1625 * age isn't reset until we get a valid ts_recent 1626 * because we don't want out-of-order segments to be 1627 * dropped when ts_recent is old. 1628 */ 1629 tp->ts_recent = 0; 1630 } else { 1631 tcpstat.tcps_rcvduppack++; 1632 tcpstat.tcps_rcvdupbyte += tlen; 1633 tcpstat.tcps_pawsdrop++; 1634 if (tlen) 1635 goto dropafterack; 1636 goto drop; 1637 } 1638 } 1639 1640 /* 1641 * T/TCP mechanism 1642 * If T/TCP was negotiated and the segment doesn't have CC, 1643 * or if its CC is wrong then drop the segment. 1644 * RST segments do not have to comply with this. 1645 */ 1646 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1647 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1648 goto dropafterack; 1649 1650 /* 1651 * In the SYN-RECEIVED state, validate that the packet belongs to 1652 * this connection before trimming the data to fit the receive 1653 * window. Check the sequence number versus IRS since we know 1654 * the sequence numbers haven't wrapped. This is a partial fix 1655 * for the "LAND" DoS attack. 1656 */ 1657 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1658 rstreason = BANDLIM_RST_OPENPORT; 1659 goto dropwithreset; 1660 } 1661 1662 todrop = tp->rcv_nxt - th->th_seq; 1663 if (todrop > 0) { 1664 if (thflags & TH_SYN) { 1665 thflags &= ~TH_SYN; 1666 th->th_seq++; 1667 if (th->th_urp > 1) 1668 th->th_urp--; 1669 else 1670 thflags &= ~TH_URG; 1671 todrop--; 1672 } 1673 /* 1674 * Following if statement from Stevens, vol. 2, p. 960. 1675 */ 1676 if (todrop > tlen 1677 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1678 /* 1679 * Any valid FIN must be to the left of the window. 1680 * At this point the FIN must be a duplicate or out 1681 * of sequence; drop it. 1682 */ 1683 thflags &= ~TH_FIN; 1684 1685 /* 1686 * Send an ACK to resynchronize and drop any data. 1687 * But keep on processing for RST or ACK. 1688 */ 1689 tp->t_flags |= TF_ACKNOW; 1690 todrop = tlen; 1691 tcpstat.tcps_rcvduppack++; 1692 tcpstat.tcps_rcvdupbyte += todrop; 1693 } else { 1694 tcpstat.tcps_rcvpartduppack++; 1695 tcpstat.tcps_rcvpartdupbyte += todrop; 1696 } 1697 drop_hdrlen += todrop; /* drop from the top afterwards */ 1698 th->th_seq += todrop; 1699 tlen -= todrop; 1700 if (th->th_urp > todrop) 1701 th->th_urp -= todrop; 1702 else { 1703 thflags &= ~TH_URG; 1704 th->th_urp = 0; 1705 } 1706 } 1707 1708 /* 1709 * If new data are received on a connection after the 1710 * user processes are gone, then RST the other end. 1711 */ 1712 if ((so->so_state & SS_NOFDREF) && 1713 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1714 tp = tcp_close(tp); 1715 tcpstat.tcps_rcvafterclose++; 1716 rstreason = BANDLIM_UNLIMITED; 1717 goto dropwithreset; 1718 } 1719 1720 /* 1721 * If segment ends after window, drop trailing data 1722 * (and PUSH and FIN); if nothing left, just ACK. 1723 */ 1724 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1725 if (todrop > 0) { 1726 tcpstat.tcps_rcvpackafterwin++; 1727 if (todrop >= tlen) { 1728 tcpstat.tcps_rcvbyteafterwin += tlen; 1729 /* 1730 * If a new connection request is received 1731 * while in TIME_WAIT, drop the old connection 1732 * and start over if the sequence numbers 1733 * are above the previous ones. 1734 */ 1735 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1736 if (thflags & TH_SYN && 1737 tp->t_state == TCPS_TIME_WAIT && 1738 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1739 tp = tcp_close(tp); 1740 goto findpcb; 1741 } 1742 /* 1743 * If window is closed can only take segments at 1744 * window edge, and have to drop data and PUSH from 1745 * incoming segments. Continue processing, but 1746 * remember to ack. Otherwise, drop segment 1747 * and ack. 1748 */ 1749 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1750 tp->t_flags |= TF_ACKNOW; 1751 tcpstat.tcps_rcvwinprobe++; 1752 } else 1753 goto dropafterack; 1754 } else 1755 tcpstat.tcps_rcvbyteafterwin += todrop; 1756 m_adj(m, -todrop); 1757 tlen -= todrop; 1758 thflags &= ~(TH_PUSH|TH_FIN); 1759 } 1760 1761 /* 1762 * If last ACK falls within this segment's sequence numbers, 1763 * record its timestamp. 1764 * NOTE that the test is modified according to the latest 1765 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1766 */ 1767 if ((to.to_flags & TOF_TS) != 0 && 1768 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1769 tp->ts_recent_age = ticks; 1770 tp->ts_recent = to.to_tsval; 1771 } 1772 1773 /* 1774 * If a SYN is in the window, then this is an 1775 * error and we send an RST and drop the connection. 1776 */ 1777 if (thflags & TH_SYN) { 1778 tp = tcp_drop(tp, ECONNRESET); 1779 rstreason = BANDLIM_UNLIMITED; 1780 goto drop; 1781 } 1782 1783 /* 1784 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1785 * flag is on (half-synchronized state), then queue data for 1786 * later processing; else drop segment and return. 1787 */ 1788 if ((thflags & TH_ACK) == 0) { 1789 if (tp->t_state == TCPS_SYN_RECEIVED || 1790 (tp->t_flags & TF_NEEDSYN)) 1791 goto step6; 1792 else 1793 goto drop; 1794 } 1795 1796 /* 1797 * Ack processing. 1798 */ 1799 switch (tp->t_state) { 1800 1801 /* 1802 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1803 * ESTABLISHED state and continue processing. 1804 * The ACK was checked above. 1805 */ 1806 case TCPS_SYN_RECEIVED: 1807 1808 tcpstat.tcps_connects++; 1809 soisconnected(so); 1810 /* Do window scaling? */ 1811 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1812 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1813 tp->snd_scale = tp->requested_s_scale; 1814 tp->rcv_scale = tp->request_r_scale; 1815 } 1816 /* 1817 * Upon successful completion of 3-way handshake, 1818 * update cache.CC, pass any queued data to the user, 1819 * and advance state appropriately. 1820 */ 1821 if (tcp_do_rfc1644) { 1822 tao.tao_cc = tp->cc_recv; 1823 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1824 tp->cc_recv, 0); 1825 } 1826 /* 1827 * Make transitions: 1828 * SYN-RECEIVED -> ESTABLISHED 1829 * SYN-RECEIVED* -> FIN-WAIT-1 1830 */ 1831 tp->t_starttime = ticks; 1832 if (tp->t_flags & TF_NEEDFIN) { 1833 tp->t_state = TCPS_FIN_WAIT_1; 1834 tp->t_flags &= ~TF_NEEDFIN; 1835 } else { 1836 tp->t_state = TCPS_ESTABLISHED; 1837 callout_reset(tp->tt_keep, tcp_keepidle, 1838 tcp_timer_keep, tp); 1839 } 1840 /* 1841 * If segment contains data or ACK, will call tcp_reass() 1842 * later; if not, do so now to pass queued data to user. 1843 */ 1844 if (tlen == 0 && (thflags & TH_FIN) == 0) 1845 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1846 (struct mbuf *)0); 1847 tp->snd_wl1 = th->th_seq - 1; 1848 /* FALLTHROUGH */ 1849 1850 /* 1851 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1852 * ACKs. If the ack is in the range 1853 * tp->snd_una < th->th_ack <= tp->snd_max 1854 * then advance tp->snd_una to th->th_ack and drop 1855 * data from the retransmission queue. If this ACK reflects 1856 * more up to date window information we update our window information. 1857 */ 1858 case TCPS_ESTABLISHED: 1859 case TCPS_FIN_WAIT_1: 1860 case TCPS_FIN_WAIT_2: 1861 case TCPS_CLOSE_WAIT: 1862 case TCPS_CLOSING: 1863 case TCPS_LAST_ACK: 1864 case TCPS_TIME_WAIT: 1865 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1866 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1867 if (tlen == 0 && tiwin == tp->snd_wnd) { 1868 tcpstat.tcps_rcvdupack++; 1869 /* 1870 * If we have outstanding data (other than 1871 * a window probe), this is a completely 1872 * duplicate ack (ie, window info didn't 1873 * change), the ack is the biggest we've 1874 * seen and we've seen exactly our rexmt 1875 * threshhold of them, assume a packet 1876 * has been dropped and retransmit it. 1877 * Kludge snd_nxt & the congestion 1878 * window so we send only this one 1879 * packet. 1880 * 1881 * We know we're losing at the current 1882 * window size so do congestion avoidance 1883 * (set ssthresh to half the current window 1884 * and pull our congestion window back to 1885 * the new ssthresh). 1886 * 1887 * Dup acks mean that packets have left the 1888 * network (they're now cached at the receiver) 1889 * so bump cwnd by the amount in the receiver 1890 * to keep a constant cwnd packets in the 1891 * network. 1892 */ 1893 if (!callout_active(tp->tt_rexmt) || 1894 th->th_ack != tp->snd_una) 1895 tp->t_dupacks = 0; 1896 else if (++tp->t_dupacks > tcprexmtthresh || 1897 (tcp_do_newreno && 1898 IN_FASTRECOVERY(tp))) { 1899 tp->snd_cwnd += tp->t_maxseg; 1900 (void) tcp_output(tp); 1901 goto drop; 1902 } else if (tp->t_dupacks == tcprexmtthresh) { 1903 tcp_seq onxt = tp->snd_nxt; 1904 u_int win; 1905 if (tcp_do_newreno && 1906 SEQ_LEQ(th->th_ack, 1907 tp->snd_recover)) { 1908 tp->t_dupacks = 0; 1909 break; 1910 } 1911 win = min(tp->snd_wnd, tp->snd_cwnd) / 1912 2 / tp->t_maxseg; 1913 if (win < 2) 1914 win = 2; 1915 tp->snd_ssthresh = win * tp->t_maxseg; 1916 ENTER_FASTRECOVERY(tp); 1917 tp->snd_recover = tp->snd_max; 1918 callout_stop(tp->tt_rexmt); 1919 tp->t_rtttime = 0; 1920 tp->snd_nxt = th->th_ack; 1921 tp->snd_cwnd = tp->t_maxseg; 1922 (void) tcp_output(tp); 1923 KASSERT(tp->snd_limited <= 2, 1924 ("tp->snd_limited too big")); 1925 tp->snd_cwnd = tp->snd_ssthresh + 1926 tp->t_maxseg * 1927 (tp->t_dupacks - tp->snd_limited); 1928 if (SEQ_GT(onxt, tp->snd_nxt)) 1929 tp->snd_nxt = onxt; 1930 goto drop; 1931 } else if (tcp_do_rfc3042) { 1932 u_long oldcwnd = tp->snd_cwnd; 1933 tcp_seq oldsndmax = tp->snd_max; 1934 u_int sent; 1935 1936 KASSERT(tp->t_dupacks == 1 || 1937 tp->t_dupacks == 2, 1938 ("dupacks not 1 or 2")); 1939 if (tp->t_dupacks == 1) 1940 tp->snd_limited = 0; 1941 tp->snd_cwnd = 1942 (tp->snd_nxt - tp->snd_una) + 1943 (tp->t_dupacks - tp->snd_limited) * 1944 tp->t_maxseg; 1945 (void) tcp_output(tp); 1946 sent = tp->snd_max - oldsndmax; 1947 if (sent > tp->t_maxseg) { 1948 KASSERT((tp->t_dupacks == 2 && 1949 tp->snd_limited == 0) || 1950 (sent == tp->t_maxseg + 1 && 1951 tp->t_flags & TF_SENTFIN), 1952 ("sent too much")); 1953 tp->snd_limited = 2; 1954 } else if (sent > 0) 1955 ++tp->snd_limited; 1956 tp->snd_cwnd = oldcwnd; 1957 goto drop; 1958 } 1959 } else 1960 tp->t_dupacks = 0; 1961 break; 1962 } 1963 1964 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1965 1966 /* 1967 * If the congestion window was inflated to account 1968 * for the other side's cached packets, retract it. 1969 */ 1970 if (tcp_do_newreno) { 1971 if (IN_FASTRECOVERY(tp)) { 1972 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1973 tcp_newreno_partial_ack(tp, th); 1974 } else { 1975 /* 1976 * Window inflation should have left us 1977 * with approximately snd_ssthresh 1978 * outstanding data. 1979 * But in case we would be inclined to 1980 * send a burst, better to do it via 1981 * the slow start mechanism. 1982 */ 1983 if (SEQ_GT(th->th_ack + 1984 tp->snd_ssthresh, 1985 tp->snd_max)) 1986 tp->snd_cwnd = tp->snd_max - 1987 th->th_ack + 1988 tp->t_maxseg; 1989 else 1990 tp->snd_cwnd = tp->snd_ssthresh; 1991 } 1992 } 1993 } else { 1994 if (tp->t_dupacks >= tcprexmtthresh && 1995 tp->snd_cwnd > tp->snd_ssthresh) 1996 tp->snd_cwnd = tp->snd_ssthresh; 1997 } 1998 tp->t_dupacks = 0; 1999 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2000 tcpstat.tcps_rcvacktoomuch++; 2001 goto dropafterack; 2002 } 2003 /* 2004 * If we reach this point, ACK is not a duplicate, 2005 * i.e., it ACKs something we sent. 2006 */ 2007 if (tp->t_flags & TF_NEEDSYN) { 2008 /* 2009 * T/TCP: Connection was half-synchronized, and our 2010 * SYN has been ACK'd (so connection is now fully 2011 * synchronized). Go to non-starred state, 2012 * increment snd_una for ACK of SYN, and check if 2013 * we can do window scaling. 2014 */ 2015 tp->t_flags &= ~TF_NEEDSYN; 2016 tp->snd_una++; 2017 /* Do window scaling? */ 2018 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2019 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2020 tp->snd_scale = tp->requested_s_scale; 2021 tp->rcv_scale = tp->request_r_scale; 2022 } 2023 } 2024 2025process_ACK: 2026 acked = th->th_ack - tp->snd_una; 2027 tcpstat.tcps_rcvackpack++; 2028 tcpstat.tcps_rcvackbyte += acked; 2029 2030 /* 2031 * If we just performed our first retransmit, and the ACK 2032 * arrives within our recovery window, then it was a mistake 2033 * to do the retransmit in the first place. Recover our 2034 * original cwnd and ssthresh, and proceed to transmit where 2035 * we left off. 2036 */ 2037 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2038 ++tcpstat.tcps_sndrexmitbad; 2039 tp->snd_cwnd = tp->snd_cwnd_prev; 2040 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2041 tp->snd_recover = tp->snd_recover_prev; 2042 if (tp->t_flags & TF_WASFRECOVERY) 2043 ENTER_FASTRECOVERY(tp); 2044 tp->snd_nxt = tp->snd_max; 2045 tp->t_badrxtwin = 0; /* XXX probably not required */ 2046 } 2047 2048 /* 2049 * If we have a timestamp reply, update smoothed 2050 * round trip time. If no timestamp is present but 2051 * transmit timer is running and timed sequence 2052 * number was acked, update smoothed round trip time. 2053 * Since we now have an rtt measurement, cancel the 2054 * timer backoff (cf., Phil Karn's retransmit alg.). 2055 * Recompute the initial retransmit timer. 2056 * 2057 * Some boxes send broken timestamp replies 2058 * during the SYN+ACK phase, ignore 2059 * timestamps of 0 or we could calculate a 2060 * huge RTT and blow up the retransmit timer. 2061 */ 2062 if ((to.to_flags & TOF_TS) != 0 && 2063 to.to_tsecr) { 2064 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2065 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2066 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2067 } 2068 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2069 2070 /* 2071 * If all outstanding data is acked, stop retransmit 2072 * timer and remember to restart (more output or persist). 2073 * If there is more data to be acked, restart retransmit 2074 * timer, using current (possibly backed-off) value. 2075 */ 2076 if (th->th_ack == tp->snd_max) { 2077 callout_stop(tp->tt_rexmt); 2078 needoutput = 1; 2079 } else if (!callout_active(tp->tt_persist)) 2080 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2081 tcp_timer_rexmt, tp); 2082 2083 /* 2084 * If no data (only SYN) was ACK'd, 2085 * skip rest of ACK processing. 2086 */ 2087 if (acked == 0) 2088 goto step6; 2089 2090 /* 2091 * When new data is acked, open the congestion window. 2092 * If the window gives us less than ssthresh packets 2093 * in flight, open exponentially (maxseg per packet). 2094 * Otherwise open linearly: maxseg per window 2095 * (maxseg^2 / cwnd per packet). 2096 */ 2097 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2098 register u_int cw = tp->snd_cwnd; 2099 register u_int incr = tp->t_maxseg; 2100 if (cw > tp->snd_ssthresh) 2101 incr = incr * incr / cw; 2102 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2103 } 2104 if (acked > so->so_snd.sb_cc) { 2105 tp->snd_wnd -= so->so_snd.sb_cc; 2106 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2107 ourfinisacked = 1; 2108 } else { 2109 sbdrop(&so->so_snd, acked); 2110 tp->snd_wnd -= acked; 2111 ourfinisacked = 0; 2112 } 2113 sowwakeup(so); 2114 /* detect una wraparound */ 2115 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2116 SEQ_GT(tp->snd_una, tp->snd_recover) && 2117 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2118 tp->snd_recover = th->th_ack - 1; 2119 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2120 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2121 EXIT_FASTRECOVERY(tp); 2122 tp->snd_una = th->th_ack; 2123 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2124 tp->snd_nxt = tp->snd_una; 2125 2126 switch (tp->t_state) { 2127 2128 /* 2129 * In FIN_WAIT_1 STATE in addition to the processing 2130 * for the ESTABLISHED state if our FIN is now acknowledged 2131 * then enter FIN_WAIT_2. 2132 */ 2133 case TCPS_FIN_WAIT_1: 2134 if (ourfinisacked) { 2135 /* 2136 * If we can't receive any more 2137 * data, then closing user can proceed. 2138 * Starting the timer is contrary to the 2139 * specification, but if we don't get a FIN 2140 * we'll hang forever. 2141 */ 2142 /* XXXjl 2143 * we should release the tp also, and use a 2144 * compressed state. 2145 */
| 1241 m_freem(m); 1242 } else { 1243 m_adj(m, drop_hdrlen); /* delayed header drop */ 1244 sbappendstream(&so->so_rcv, m); 1245 } 1246 sorwakeup(so); 1247 if (DELAY_ACK(tp)) { 1248 tp->t_flags |= TF_DELACK; 1249 } else { 1250 tp->t_flags |= TF_ACKNOW; 1251 tcp_output(tp); 1252 } 1253 goto check_delack; 1254 } 1255 } 1256 1257 /* 1258 * Calculate amount of space in receive window, 1259 * and then do TCP input processing. 1260 * Receive window is amount of space in rcv queue, 1261 * but not less than advertised window. 1262 */ 1263 { int win; 1264 1265 win = sbspace(&so->so_rcv); 1266 if (win < 0) 1267 win = 0; 1268 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1269 } 1270 1271 switch (tp->t_state) { 1272 1273 /* 1274 * If the state is SYN_RECEIVED: 1275 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1276 */ 1277 case TCPS_SYN_RECEIVED: 1278 if ((thflags & TH_ACK) && 1279 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1280 SEQ_GT(th->th_ack, tp->snd_max))) { 1281 rstreason = BANDLIM_RST_OPENPORT; 1282 goto dropwithreset; 1283 } 1284 break; 1285 1286 /* 1287 * If the state is SYN_SENT: 1288 * if seg contains an ACK, but not for our SYN, drop the input. 1289 * if seg contains a RST, then drop the connection. 1290 * if seg does not contain SYN, then drop it. 1291 * Otherwise this is an acceptable SYN segment 1292 * initialize tp->rcv_nxt and tp->irs 1293 * if seg contains ack then advance tp->snd_una 1294 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1295 * arrange for segment to be acked (eventually) 1296 * continue processing rest of data/controls, beginning with URG 1297 */ 1298 case TCPS_SYN_SENT: 1299 if (tcp_do_rfc1644) 1300 tcp_hc_gettao(&inp->inp_inc, &tao); 1301 1302 if ((thflags & TH_ACK) && 1303 (SEQ_LEQ(th->th_ack, tp->iss) || 1304 SEQ_GT(th->th_ack, tp->snd_max))) { 1305 /* 1306 * If we have a cached CCsent for the remote host, 1307 * hence we haven't just crashed and restarted, 1308 * do not send a RST. This may be a retransmission 1309 * from the other side after our earlier ACK was lost. 1310 * Our new SYN, when it arrives, will serve as the 1311 * needed ACK. 1312 */ 1313 if (tao.tao_ccsent != 0) 1314 goto drop; 1315 else { 1316 rstreason = BANDLIM_UNLIMITED; 1317 goto dropwithreset; 1318 } 1319 } 1320 if (thflags & TH_RST) { 1321 if (thflags & TH_ACK) 1322 tp = tcp_drop(tp, ECONNREFUSED); 1323 goto drop; 1324 } 1325 if ((thflags & TH_SYN) == 0) 1326 goto drop; 1327 tp->snd_wnd = th->th_win; /* initial send window */ 1328 tp->cc_recv = to.to_cc; /* foreign CC */ 1329 1330 tp->irs = th->th_seq; 1331 tcp_rcvseqinit(tp); 1332 if (thflags & TH_ACK) { 1333 /* 1334 * Our SYN was acked. If segment contains CC.ECHO 1335 * option, check it to make sure this segment really 1336 * matches our SYN. If not, just drop it as old 1337 * duplicate, but send an RST if we're still playing 1338 * by the old rules. If no CC.ECHO option, make sure 1339 * we don't get fooled into using T/TCP. 1340 */ 1341 if (to.to_flags & TOF_CCECHO) { 1342 if (tp->cc_send != to.to_ccecho) { 1343 if (tao.tao_ccsent != 0) 1344 goto drop; 1345 else { 1346 rstreason = BANDLIM_UNLIMITED; 1347 goto dropwithreset; 1348 } 1349 } 1350 } else 1351 tp->t_flags &= ~TF_RCVD_CC; 1352 tcpstat.tcps_connects++; 1353 soisconnected(so); 1354#ifdef MAC 1355 SOCK_LOCK(so); 1356 mac_set_socket_peer_from_mbuf(m, so); 1357 SOCK_UNLOCK(so); 1358#endif 1359 /* Do window scaling on this connection? */ 1360 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1361 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1362 tp->snd_scale = tp->requested_s_scale; 1363 tp->rcv_scale = tp->request_r_scale; 1364 } 1365 /* Segment is acceptable, update cache if undefined. */ 1366 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1367 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1368 1369 tp->rcv_adv += tp->rcv_wnd; 1370 tp->snd_una++; /* SYN is acked */ 1371 /* 1372 * If there's data, delay ACK; if there's also a FIN 1373 * ACKNOW will be turned on later. 1374 */ 1375 if (DELAY_ACK(tp) && tlen != 0) 1376 callout_reset(tp->tt_delack, tcp_delacktime, 1377 tcp_timer_delack, tp); 1378 else 1379 tp->t_flags |= TF_ACKNOW; 1380 /* 1381 * Received <SYN,ACK> in SYN_SENT[*] state. 1382 * Transitions: 1383 * SYN_SENT --> ESTABLISHED 1384 * SYN_SENT* --> FIN_WAIT_1 1385 */ 1386 tp->t_starttime = ticks; 1387 if (tp->t_flags & TF_NEEDFIN) { 1388 tp->t_state = TCPS_FIN_WAIT_1; 1389 tp->t_flags &= ~TF_NEEDFIN; 1390 thflags &= ~TH_SYN; 1391 } else { 1392 tp->t_state = TCPS_ESTABLISHED; 1393 callout_reset(tp->tt_keep, tcp_keepidle, 1394 tcp_timer_keep, tp); 1395 } 1396 } else { 1397 /* 1398 * Received initial SYN in SYN-SENT[*] state => 1399 * simultaneous open. If segment contains CC option 1400 * and there is a cached CC, apply TAO test. 1401 * If it succeeds, connection is * half-synchronized. 1402 * Otherwise, do 3-way handshake: 1403 * SYN-SENT -> SYN-RECEIVED 1404 * SYN-SENT* -> SYN-RECEIVED* 1405 * If there was no CC option, clear cached CC value. 1406 */ 1407 tp->t_flags |= TF_ACKNOW; 1408 callout_stop(tp->tt_rexmt); 1409 if (to.to_flags & TOF_CC) { 1410 if (tao.tao_cc != 0 && 1411 CC_GT(to.to_cc, tao.tao_cc)) { 1412 /* 1413 * update cache and make transition: 1414 * SYN-SENT -> ESTABLISHED* 1415 * SYN-SENT* -> FIN-WAIT-1* 1416 */ 1417 tao.tao_cc = to.to_cc; 1418 tcp_hc_updatetao(&inp->inp_inc, 1419 TCP_HC_TAO_CC, to.to_cc, 0); 1420 tp->t_starttime = ticks; 1421 if (tp->t_flags & TF_NEEDFIN) { 1422 tp->t_state = TCPS_FIN_WAIT_1; 1423 tp->t_flags &= ~TF_NEEDFIN; 1424 } else { 1425 tp->t_state = TCPS_ESTABLISHED; 1426 callout_reset(tp->tt_keep, 1427 tcp_keepidle, 1428 tcp_timer_keep, 1429 tp); 1430 } 1431 tp->t_flags |= TF_NEEDSYN; 1432 } else 1433 tp->t_state = TCPS_SYN_RECEIVED; 1434 } else { 1435 if (tcp_do_rfc1644) { 1436 /* CC.NEW or no option => invalidate cache */ 1437 tao.tao_cc = 0; 1438 tcp_hc_updatetao(&inp->inp_inc, 1439 TCP_HC_TAO_CC, to.to_cc, 0); 1440 } 1441 tp->t_state = TCPS_SYN_RECEIVED; 1442 } 1443 } 1444 1445trimthenstep6: 1446 /* 1447 * Advance th->th_seq to correspond to first data byte. 1448 * If data, trim to stay within window, 1449 * dropping FIN if necessary. 1450 */ 1451 th->th_seq++; 1452 if (tlen > tp->rcv_wnd) { 1453 todrop = tlen - tp->rcv_wnd; 1454 m_adj(m, -todrop); 1455 tlen = tp->rcv_wnd; 1456 thflags &= ~TH_FIN; 1457 tcpstat.tcps_rcvpackafterwin++; 1458 tcpstat.tcps_rcvbyteafterwin += todrop; 1459 } 1460 tp->snd_wl1 = th->th_seq - 1; 1461 tp->rcv_up = th->th_seq; 1462 /* 1463 * Client side of transaction: already sent SYN and data. 1464 * If the remote host used T/TCP to validate the SYN, 1465 * our data will be ACK'd; if so, enter normal data segment 1466 * processing in the middle of step 5, ack processing. 1467 * Otherwise, goto step 6. 1468 */ 1469 if (thflags & TH_ACK) 1470 goto process_ACK; 1471 1472 goto step6; 1473 1474 /* 1475 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1476 * if segment contains a SYN and CC [not CC.NEW] option: 1477 * if state == TIME_WAIT and connection duration > MSL, 1478 * drop packet and send RST; 1479 * 1480 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1481 * ack the FIN (and data) in retransmission queue. 1482 * Complete close and delete TCPCB. Then reprocess 1483 * segment, hoping to find new TCPCB in LISTEN state; 1484 * 1485 * else must be old SYN; drop it. 1486 * else do normal processing. 1487 */ 1488 case TCPS_LAST_ACK: 1489 case TCPS_CLOSING: 1490 case TCPS_TIME_WAIT: 1491 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1492 if ((thflags & TH_SYN) && 1493 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1494 if (tp->t_state == TCPS_TIME_WAIT && 1495 (ticks - tp->t_starttime) > tcp_msl) { 1496 rstreason = BANDLIM_UNLIMITED; 1497 goto dropwithreset; 1498 } 1499 if (CC_GT(to.to_cc, tp->cc_recv)) { 1500 tp = tcp_close(tp); 1501 goto findpcb; 1502 } 1503 else 1504 goto drop; 1505 } 1506 break; /* continue normal processing */ 1507 } 1508 1509 /* 1510 * States other than LISTEN or SYN_SENT. 1511 * First check the RST flag and sequence number since reset segments 1512 * are exempt from the timestamp and connection count tests. This 1513 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1514 * below which allowed reset segments in half the sequence space 1515 * to fall though and be processed (which gives forged reset 1516 * segments with a random sequence number a 50 percent chance of 1517 * killing a connection). 1518 * Then check timestamp, if present. 1519 * Then check the connection count, if present. 1520 * Then check that at least some bytes of segment are within 1521 * receive window. If segment begins before rcv_nxt, 1522 * drop leading data (and SYN); if nothing left, just ack. 1523 * 1524 * 1525 * If the RST bit is set, check the sequence number to see 1526 * if this is a valid reset segment. 1527 * RFC 793 page 37: 1528 * In all states except SYN-SENT, all reset (RST) segments 1529 * are validated by checking their SEQ-fields. A reset is 1530 * valid if its sequence number is in the window. 1531 * Note: this does not take into account delayed ACKs, so 1532 * we should test against last_ack_sent instead of rcv_nxt. 1533 * The sequence number in the reset segment is normally an 1534 * echo of our outgoing acknowlegement numbers, but some hosts 1535 * send a reset with the sequence number at the rightmost edge 1536 * of our receive window, and we have to handle this case. 1537 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1538 * that brute force RST attacks are possible. To combat this, 1539 * we use a much stricter check while in the ESTABLISHED state, 1540 * only accepting RSTs where the sequence number is equal to 1541 * last_ack_sent. In all other states (the states in which a 1542 * RST is more likely), the more permissive check is used. 1543 * If we have multiple segments in flight, the intial reset 1544 * segment sequence numbers will be to the left of last_ack_sent, 1545 * but they will eventually catch up. 1546 * In any case, it never made sense to trim reset segments to 1547 * fit the receive window since RFC 1122 says: 1548 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1549 * 1550 * A TCP SHOULD allow a received RST segment to include data. 1551 * 1552 * DISCUSSION 1553 * It has been suggested that a RST segment could contain 1554 * ASCII text that encoded and explained the cause of the 1555 * RST. No standard has yet been established for such 1556 * data. 1557 * 1558 * If the reset segment passes the sequence number test examine 1559 * the state: 1560 * SYN_RECEIVED STATE: 1561 * If passive open, return to LISTEN state. 1562 * If active open, inform user that connection was refused. 1563 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1564 * Inform user that connection was reset, and close tcb. 1565 * CLOSING, LAST_ACK STATES: 1566 * Close the tcb. 1567 * TIME_WAIT STATE: 1568 * Drop the segment - see Stevens, vol. 2, p. 964 and 1569 * RFC 1337. 1570 */ 1571 if (thflags & TH_RST) { 1572 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1573 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1574 switch (tp->t_state) { 1575 1576 case TCPS_SYN_RECEIVED: 1577 so->so_error = ECONNREFUSED; 1578 goto close; 1579 1580 case TCPS_ESTABLISHED: 1581 if (tp->last_ack_sent != th->th_seq) { 1582 tcpstat.tcps_badrst++; 1583 goto drop; 1584 } 1585 case TCPS_FIN_WAIT_1: 1586 case TCPS_FIN_WAIT_2: 1587 case TCPS_CLOSE_WAIT: 1588 so->so_error = ECONNRESET; 1589 close: 1590 tp->t_state = TCPS_CLOSED; 1591 tcpstat.tcps_drops++; 1592 tp = tcp_close(tp); 1593 break; 1594 1595 case TCPS_CLOSING: 1596 case TCPS_LAST_ACK: 1597 tp = tcp_close(tp); 1598 break; 1599 1600 case TCPS_TIME_WAIT: 1601 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1602 ("timewait")); 1603 break; 1604 } 1605 } 1606 goto drop; 1607 } 1608 1609 /* 1610 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1611 * and it's less than ts_recent, drop it. 1612 */ 1613 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1614 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1615 1616 /* Check to see if ts_recent is over 24 days old. */ 1617 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1618 /* 1619 * Invalidate ts_recent. If this segment updates 1620 * ts_recent, the age will be reset later and ts_recent 1621 * will get a valid value. If it does not, setting 1622 * ts_recent to zero will at least satisfy the 1623 * requirement that zero be placed in the timestamp 1624 * echo reply when ts_recent isn't valid. The 1625 * age isn't reset until we get a valid ts_recent 1626 * because we don't want out-of-order segments to be 1627 * dropped when ts_recent is old. 1628 */ 1629 tp->ts_recent = 0; 1630 } else { 1631 tcpstat.tcps_rcvduppack++; 1632 tcpstat.tcps_rcvdupbyte += tlen; 1633 tcpstat.tcps_pawsdrop++; 1634 if (tlen) 1635 goto dropafterack; 1636 goto drop; 1637 } 1638 } 1639 1640 /* 1641 * T/TCP mechanism 1642 * If T/TCP was negotiated and the segment doesn't have CC, 1643 * or if its CC is wrong then drop the segment. 1644 * RST segments do not have to comply with this. 1645 */ 1646 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1647 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1648 goto dropafterack; 1649 1650 /* 1651 * In the SYN-RECEIVED state, validate that the packet belongs to 1652 * this connection before trimming the data to fit the receive 1653 * window. Check the sequence number versus IRS since we know 1654 * the sequence numbers haven't wrapped. This is a partial fix 1655 * for the "LAND" DoS attack. 1656 */ 1657 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1658 rstreason = BANDLIM_RST_OPENPORT; 1659 goto dropwithreset; 1660 } 1661 1662 todrop = tp->rcv_nxt - th->th_seq; 1663 if (todrop > 0) { 1664 if (thflags & TH_SYN) { 1665 thflags &= ~TH_SYN; 1666 th->th_seq++; 1667 if (th->th_urp > 1) 1668 th->th_urp--; 1669 else 1670 thflags &= ~TH_URG; 1671 todrop--; 1672 } 1673 /* 1674 * Following if statement from Stevens, vol. 2, p. 960. 1675 */ 1676 if (todrop > tlen 1677 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1678 /* 1679 * Any valid FIN must be to the left of the window. 1680 * At this point the FIN must be a duplicate or out 1681 * of sequence; drop it. 1682 */ 1683 thflags &= ~TH_FIN; 1684 1685 /* 1686 * Send an ACK to resynchronize and drop any data. 1687 * But keep on processing for RST or ACK. 1688 */ 1689 tp->t_flags |= TF_ACKNOW; 1690 todrop = tlen; 1691 tcpstat.tcps_rcvduppack++; 1692 tcpstat.tcps_rcvdupbyte += todrop; 1693 } else { 1694 tcpstat.tcps_rcvpartduppack++; 1695 tcpstat.tcps_rcvpartdupbyte += todrop; 1696 } 1697 drop_hdrlen += todrop; /* drop from the top afterwards */ 1698 th->th_seq += todrop; 1699 tlen -= todrop; 1700 if (th->th_urp > todrop) 1701 th->th_urp -= todrop; 1702 else { 1703 thflags &= ~TH_URG; 1704 th->th_urp = 0; 1705 } 1706 } 1707 1708 /* 1709 * If new data are received on a connection after the 1710 * user processes are gone, then RST the other end. 1711 */ 1712 if ((so->so_state & SS_NOFDREF) && 1713 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1714 tp = tcp_close(tp); 1715 tcpstat.tcps_rcvafterclose++; 1716 rstreason = BANDLIM_UNLIMITED; 1717 goto dropwithreset; 1718 } 1719 1720 /* 1721 * If segment ends after window, drop trailing data 1722 * (and PUSH and FIN); if nothing left, just ACK. 1723 */ 1724 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1725 if (todrop > 0) { 1726 tcpstat.tcps_rcvpackafterwin++; 1727 if (todrop >= tlen) { 1728 tcpstat.tcps_rcvbyteafterwin += tlen; 1729 /* 1730 * If a new connection request is received 1731 * while in TIME_WAIT, drop the old connection 1732 * and start over if the sequence numbers 1733 * are above the previous ones. 1734 */ 1735 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1736 if (thflags & TH_SYN && 1737 tp->t_state == TCPS_TIME_WAIT && 1738 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1739 tp = tcp_close(tp); 1740 goto findpcb; 1741 } 1742 /* 1743 * If window is closed can only take segments at 1744 * window edge, and have to drop data and PUSH from 1745 * incoming segments. Continue processing, but 1746 * remember to ack. Otherwise, drop segment 1747 * and ack. 1748 */ 1749 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1750 tp->t_flags |= TF_ACKNOW; 1751 tcpstat.tcps_rcvwinprobe++; 1752 } else 1753 goto dropafterack; 1754 } else 1755 tcpstat.tcps_rcvbyteafterwin += todrop; 1756 m_adj(m, -todrop); 1757 tlen -= todrop; 1758 thflags &= ~(TH_PUSH|TH_FIN); 1759 } 1760 1761 /* 1762 * If last ACK falls within this segment's sequence numbers, 1763 * record its timestamp. 1764 * NOTE that the test is modified according to the latest 1765 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1766 */ 1767 if ((to.to_flags & TOF_TS) != 0 && 1768 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1769 tp->ts_recent_age = ticks; 1770 tp->ts_recent = to.to_tsval; 1771 } 1772 1773 /* 1774 * If a SYN is in the window, then this is an 1775 * error and we send an RST and drop the connection. 1776 */ 1777 if (thflags & TH_SYN) { 1778 tp = tcp_drop(tp, ECONNRESET); 1779 rstreason = BANDLIM_UNLIMITED; 1780 goto drop; 1781 } 1782 1783 /* 1784 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1785 * flag is on (half-synchronized state), then queue data for 1786 * later processing; else drop segment and return. 1787 */ 1788 if ((thflags & TH_ACK) == 0) { 1789 if (tp->t_state == TCPS_SYN_RECEIVED || 1790 (tp->t_flags & TF_NEEDSYN)) 1791 goto step6; 1792 else 1793 goto drop; 1794 } 1795 1796 /* 1797 * Ack processing. 1798 */ 1799 switch (tp->t_state) { 1800 1801 /* 1802 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1803 * ESTABLISHED state and continue processing. 1804 * The ACK was checked above. 1805 */ 1806 case TCPS_SYN_RECEIVED: 1807 1808 tcpstat.tcps_connects++; 1809 soisconnected(so); 1810 /* Do window scaling? */ 1811 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1812 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1813 tp->snd_scale = tp->requested_s_scale; 1814 tp->rcv_scale = tp->request_r_scale; 1815 } 1816 /* 1817 * Upon successful completion of 3-way handshake, 1818 * update cache.CC, pass any queued data to the user, 1819 * and advance state appropriately. 1820 */ 1821 if (tcp_do_rfc1644) { 1822 tao.tao_cc = tp->cc_recv; 1823 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1824 tp->cc_recv, 0); 1825 } 1826 /* 1827 * Make transitions: 1828 * SYN-RECEIVED -> ESTABLISHED 1829 * SYN-RECEIVED* -> FIN-WAIT-1 1830 */ 1831 tp->t_starttime = ticks; 1832 if (tp->t_flags & TF_NEEDFIN) { 1833 tp->t_state = TCPS_FIN_WAIT_1; 1834 tp->t_flags &= ~TF_NEEDFIN; 1835 } else { 1836 tp->t_state = TCPS_ESTABLISHED; 1837 callout_reset(tp->tt_keep, tcp_keepidle, 1838 tcp_timer_keep, tp); 1839 } 1840 /* 1841 * If segment contains data or ACK, will call tcp_reass() 1842 * later; if not, do so now to pass queued data to user. 1843 */ 1844 if (tlen == 0 && (thflags & TH_FIN) == 0) 1845 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1846 (struct mbuf *)0); 1847 tp->snd_wl1 = th->th_seq - 1; 1848 /* FALLTHROUGH */ 1849 1850 /* 1851 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1852 * ACKs. If the ack is in the range 1853 * tp->snd_una < th->th_ack <= tp->snd_max 1854 * then advance tp->snd_una to th->th_ack and drop 1855 * data from the retransmission queue. If this ACK reflects 1856 * more up to date window information we update our window information. 1857 */ 1858 case TCPS_ESTABLISHED: 1859 case TCPS_FIN_WAIT_1: 1860 case TCPS_FIN_WAIT_2: 1861 case TCPS_CLOSE_WAIT: 1862 case TCPS_CLOSING: 1863 case TCPS_LAST_ACK: 1864 case TCPS_TIME_WAIT: 1865 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1866 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1867 if (tlen == 0 && tiwin == tp->snd_wnd) { 1868 tcpstat.tcps_rcvdupack++; 1869 /* 1870 * If we have outstanding data (other than 1871 * a window probe), this is a completely 1872 * duplicate ack (ie, window info didn't 1873 * change), the ack is the biggest we've 1874 * seen and we've seen exactly our rexmt 1875 * threshhold of them, assume a packet 1876 * has been dropped and retransmit it. 1877 * Kludge snd_nxt & the congestion 1878 * window so we send only this one 1879 * packet. 1880 * 1881 * We know we're losing at the current 1882 * window size so do congestion avoidance 1883 * (set ssthresh to half the current window 1884 * and pull our congestion window back to 1885 * the new ssthresh). 1886 * 1887 * Dup acks mean that packets have left the 1888 * network (they're now cached at the receiver) 1889 * so bump cwnd by the amount in the receiver 1890 * to keep a constant cwnd packets in the 1891 * network. 1892 */ 1893 if (!callout_active(tp->tt_rexmt) || 1894 th->th_ack != tp->snd_una) 1895 tp->t_dupacks = 0; 1896 else if (++tp->t_dupacks > tcprexmtthresh || 1897 (tcp_do_newreno && 1898 IN_FASTRECOVERY(tp))) { 1899 tp->snd_cwnd += tp->t_maxseg; 1900 (void) tcp_output(tp); 1901 goto drop; 1902 } else if (tp->t_dupacks == tcprexmtthresh) { 1903 tcp_seq onxt = tp->snd_nxt; 1904 u_int win; 1905 if (tcp_do_newreno && 1906 SEQ_LEQ(th->th_ack, 1907 tp->snd_recover)) { 1908 tp->t_dupacks = 0; 1909 break; 1910 } 1911 win = min(tp->snd_wnd, tp->snd_cwnd) / 1912 2 / tp->t_maxseg; 1913 if (win < 2) 1914 win = 2; 1915 tp->snd_ssthresh = win * tp->t_maxseg; 1916 ENTER_FASTRECOVERY(tp); 1917 tp->snd_recover = tp->snd_max; 1918 callout_stop(tp->tt_rexmt); 1919 tp->t_rtttime = 0; 1920 tp->snd_nxt = th->th_ack; 1921 tp->snd_cwnd = tp->t_maxseg; 1922 (void) tcp_output(tp); 1923 KASSERT(tp->snd_limited <= 2, 1924 ("tp->snd_limited too big")); 1925 tp->snd_cwnd = tp->snd_ssthresh + 1926 tp->t_maxseg * 1927 (tp->t_dupacks - tp->snd_limited); 1928 if (SEQ_GT(onxt, tp->snd_nxt)) 1929 tp->snd_nxt = onxt; 1930 goto drop; 1931 } else if (tcp_do_rfc3042) { 1932 u_long oldcwnd = tp->snd_cwnd; 1933 tcp_seq oldsndmax = tp->snd_max; 1934 u_int sent; 1935 1936 KASSERT(tp->t_dupacks == 1 || 1937 tp->t_dupacks == 2, 1938 ("dupacks not 1 or 2")); 1939 if (tp->t_dupacks == 1) 1940 tp->snd_limited = 0; 1941 tp->snd_cwnd = 1942 (tp->snd_nxt - tp->snd_una) + 1943 (tp->t_dupacks - tp->snd_limited) * 1944 tp->t_maxseg; 1945 (void) tcp_output(tp); 1946 sent = tp->snd_max - oldsndmax; 1947 if (sent > tp->t_maxseg) { 1948 KASSERT((tp->t_dupacks == 2 && 1949 tp->snd_limited == 0) || 1950 (sent == tp->t_maxseg + 1 && 1951 tp->t_flags & TF_SENTFIN), 1952 ("sent too much")); 1953 tp->snd_limited = 2; 1954 } else if (sent > 0) 1955 ++tp->snd_limited; 1956 tp->snd_cwnd = oldcwnd; 1957 goto drop; 1958 } 1959 } else 1960 tp->t_dupacks = 0; 1961 break; 1962 } 1963 1964 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1965 1966 /* 1967 * If the congestion window was inflated to account 1968 * for the other side's cached packets, retract it. 1969 */ 1970 if (tcp_do_newreno) { 1971 if (IN_FASTRECOVERY(tp)) { 1972 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1973 tcp_newreno_partial_ack(tp, th); 1974 } else { 1975 /* 1976 * Window inflation should have left us 1977 * with approximately snd_ssthresh 1978 * outstanding data. 1979 * But in case we would be inclined to 1980 * send a burst, better to do it via 1981 * the slow start mechanism. 1982 */ 1983 if (SEQ_GT(th->th_ack + 1984 tp->snd_ssthresh, 1985 tp->snd_max)) 1986 tp->snd_cwnd = tp->snd_max - 1987 th->th_ack + 1988 tp->t_maxseg; 1989 else 1990 tp->snd_cwnd = tp->snd_ssthresh; 1991 } 1992 } 1993 } else { 1994 if (tp->t_dupacks >= tcprexmtthresh && 1995 tp->snd_cwnd > tp->snd_ssthresh) 1996 tp->snd_cwnd = tp->snd_ssthresh; 1997 } 1998 tp->t_dupacks = 0; 1999 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2000 tcpstat.tcps_rcvacktoomuch++; 2001 goto dropafterack; 2002 } 2003 /* 2004 * If we reach this point, ACK is not a duplicate, 2005 * i.e., it ACKs something we sent. 2006 */ 2007 if (tp->t_flags & TF_NEEDSYN) { 2008 /* 2009 * T/TCP: Connection was half-synchronized, and our 2010 * SYN has been ACK'd (so connection is now fully 2011 * synchronized). Go to non-starred state, 2012 * increment snd_una for ACK of SYN, and check if 2013 * we can do window scaling. 2014 */ 2015 tp->t_flags &= ~TF_NEEDSYN; 2016 tp->snd_una++; 2017 /* Do window scaling? */ 2018 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2019 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2020 tp->snd_scale = tp->requested_s_scale; 2021 tp->rcv_scale = tp->request_r_scale; 2022 } 2023 } 2024 2025process_ACK: 2026 acked = th->th_ack - tp->snd_una; 2027 tcpstat.tcps_rcvackpack++; 2028 tcpstat.tcps_rcvackbyte += acked; 2029 2030 /* 2031 * If we just performed our first retransmit, and the ACK 2032 * arrives within our recovery window, then it was a mistake 2033 * to do the retransmit in the first place. Recover our 2034 * original cwnd and ssthresh, and proceed to transmit where 2035 * we left off. 2036 */ 2037 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2038 ++tcpstat.tcps_sndrexmitbad; 2039 tp->snd_cwnd = tp->snd_cwnd_prev; 2040 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2041 tp->snd_recover = tp->snd_recover_prev; 2042 if (tp->t_flags & TF_WASFRECOVERY) 2043 ENTER_FASTRECOVERY(tp); 2044 tp->snd_nxt = tp->snd_max; 2045 tp->t_badrxtwin = 0; /* XXX probably not required */ 2046 } 2047 2048 /* 2049 * If we have a timestamp reply, update smoothed 2050 * round trip time. If no timestamp is present but 2051 * transmit timer is running and timed sequence 2052 * number was acked, update smoothed round trip time. 2053 * Since we now have an rtt measurement, cancel the 2054 * timer backoff (cf., Phil Karn's retransmit alg.). 2055 * Recompute the initial retransmit timer. 2056 * 2057 * Some boxes send broken timestamp replies 2058 * during the SYN+ACK phase, ignore 2059 * timestamps of 0 or we could calculate a 2060 * huge RTT and blow up the retransmit timer. 2061 */ 2062 if ((to.to_flags & TOF_TS) != 0 && 2063 to.to_tsecr) { 2064 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2065 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2066 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2067 } 2068 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2069 2070 /* 2071 * If all outstanding data is acked, stop retransmit 2072 * timer and remember to restart (more output or persist). 2073 * If there is more data to be acked, restart retransmit 2074 * timer, using current (possibly backed-off) value. 2075 */ 2076 if (th->th_ack == tp->snd_max) { 2077 callout_stop(tp->tt_rexmt); 2078 needoutput = 1; 2079 } else if (!callout_active(tp->tt_persist)) 2080 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2081 tcp_timer_rexmt, tp); 2082 2083 /* 2084 * If no data (only SYN) was ACK'd, 2085 * skip rest of ACK processing. 2086 */ 2087 if (acked == 0) 2088 goto step6; 2089 2090 /* 2091 * When new data is acked, open the congestion window. 2092 * If the window gives us less than ssthresh packets 2093 * in flight, open exponentially (maxseg per packet). 2094 * Otherwise open linearly: maxseg per window 2095 * (maxseg^2 / cwnd per packet). 2096 */ 2097 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2098 register u_int cw = tp->snd_cwnd; 2099 register u_int incr = tp->t_maxseg; 2100 if (cw > tp->snd_ssthresh) 2101 incr = incr * incr / cw; 2102 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2103 } 2104 if (acked > so->so_snd.sb_cc) { 2105 tp->snd_wnd -= so->so_snd.sb_cc; 2106 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2107 ourfinisacked = 1; 2108 } else { 2109 sbdrop(&so->so_snd, acked); 2110 tp->snd_wnd -= acked; 2111 ourfinisacked = 0; 2112 } 2113 sowwakeup(so); 2114 /* detect una wraparound */ 2115 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2116 SEQ_GT(tp->snd_una, tp->snd_recover) && 2117 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2118 tp->snd_recover = th->th_ack - 1; 2119 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2120 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2121 EXIT_FASTRECOVERY(tp); 2122 tp->snd_una = th->th_ack; 2123 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2124 tp->snd_nxt = tp->snd_una; 2125 2126 switch (tp->t_state) { 2127 2128 /* 2129 * In FIN_WAIT_1 STATE in addition to the processing 2130 * for the ESTABLISHED state if our FIN is now acknowledged 2131 * then enter FIN_WAIT_2. 2132 */ 2133 case TCPS_FIN_WAIT_1: 2134 if (ourfinisacked) { 2135 /* 2136 * If we can't receive any more 2137 * data, then closing user can proceed. 2138 * Starting the timer is contrary to the 2139 * specification, but if we don't get a FIN 2140 * we'll hang forever. 2141 */ 2142 /* XXXjl 2143 * we should release the tp also, and use a 2144 * compressed state. 2145 */
|
2146 if (so->so_state & SS_CANTRCVMORE) {
| 2146 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
|
2147 soisdisconnected(so); 2148 callout_reset(tp->tt_2msl, tcp_maxidle, 2149 tcp_timer_2msl, tp); 2150 } 2151 tp->t_state = TCPS_FIN_WAIT_2; 2152 } 2153 break; 2154 2155 /* 2156 * In CLOSING STATE in addition to the processing for 2157 * the ESTABLISHED state if the ACK acknowledges our FIN 2158 * then enter the TIME-WAIT state, otherwise ignore 2159 * the segment. 2160 */ 2161 case TCPS_CLOSING: 2162 if (ourfinisacked) { 2163 KASSERT(headlocked, ("headlocked")); 2164 tcp_twstart(tp); 2165 INP_INFO_WUNLOCK(&tcbinfo); 2166 m_freem(m); 2167 return; 2168 } 2169 break; 2170 2171 /* 2172 * In LAST_ACK, we may still be waiting for data to drain 2173 * and/or to be acked, as well as for the ack of our FIN. 2174 * If our FIN is now acknowledged, delete the TCB, 2175 * enter the closed state and return. 2176 */ 2177 case TCPS_LAST_ACK: 2178 if (ourfinisacked) { 2179 tp = tcp_close(tp); 2180 goto drop; 2181 } 2182 break; 2183 2184 /* 2185 * In TIME_WAIT state the only thing that should arrive 2186 * is a retransmission of the remote FIN. Acknowledge 2187 * it and restart the finack timer. 2188 */ 2189 case TCPS_TIME_WAIT: 2190 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2191 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2192 tcp_timer_2msl, tp); 2193 goto dropafterack; 2194 } 2195 } 2196 2197step6: 2198 /* 2199 * Update window information. 2200 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2201 */ 2202 if ((thflags & TH_ACK) && 2203 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2204 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2205 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2206 /* keep track of pure window updates */ 2207 if (tlen == 0 && 2208 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2209 tcpstat.tcps_rcvwinupd++; 2210 tp->snd_wnd = tiwin; 2211 tp->snd_wl1 = th->th_seq; 2212 tp->snd_wl2 = th->th_ack; 2213 if (tp->snd_wnd > tp->max_sndwnd) 2214 tp->max_sndwnd = tp->snd_wnd; 2215 needoutput = 1; 2216 } 2217 2218 /* 2219 * Process segments with URG. 2220 */ 2221 if ((thflags & TH_URG) && th->th_urp && 2222 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2223 /* 2224 * This is a kludge, but if we receive and accept 2225 * random urgent pointers, we'll crash in 2226 * soreceive. It's hard to imagine someone 2227 * actually wanting to send this much urgent data. 2228 */ 2229 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2230 th->th_urp = 0; /* XXX */ 2231 thflags &= ~TH_URG; /* XXX */ 2232 goto dodata; /* XXX */ 2233 } 2234 /* 2235 * If this segment advances the known urgent pointer, 2236 * then mark the data stream. This should not happen 2237 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2238 * a FIN has been received from the remote side. 2239 * In these states we ignore the URG. 2240 * 2241 * According to RFC961 (Assigned Protocols), 2242 * the urgent pointer points to the last octet 2243 * of urgent data. We continue, however, 2244 * to consider it to indicate the first octet 2245 * of data past the urgent section as the original 2246 * spec states (in one of two places). 2247 */ 2248 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2249 tp->rcv_up = th->th_seq + th->th_urp; 2250 so->so_oobmark = so->so_rcv.sb_cc + 2251 (tp->rcv_up - tp->rcv_nxt) - 1; 2252 if (so->so_oobmark == 0)
| 2147 soisdisconnected(so); 2148 callout_reset(tp->tt_2msl, tcp_maxidle, 2149 tcp_timer_2msl, tp); 2150 } 2151 tp->t_state = TCPS_FIN_WAIT_2; 2152 } 2153 break; 2154 2155 /* 2156 * In CLOSING STATE in addition to the processing for 2157 * the ESTABLISHED state if the ACK acknowledges our FIN 2158 * then enter the TIME-WAIT state, otherwise ignore 2159 * the segment. 2160 */ 2161 case TCPS_CLOSING: 2162 if (ourfinisacked) { 2163 KASSERT(headlocked, ("headlocked")); 2164 tcp_twstart(tp); 2165 INP_INFO_WUNLOCK(&tcbinfo); 2166 m_freem(m); 2167 return; 2168 } 2169 break; 2170 2171 /* 2172 * In LAST_ACK, we may still be waiting for data to drain 2173 * and/or to be acked, as well as for the ack of our FIN. 2174 * If our FIN is now acknowledged, delete the TCB, 2175 * enter the closed state and return. 2176 */ 2177 case TCPS_LAST_ACK: 2178 if (ourfinisacked) { 2179 tp = tcp_close(tp); 2180 goto drop; 2181 } 2182 break; 2183 2184 /* 2185 * In TIME_WAIT state the only thing that should arrive 2186 * is a retransmission of the remote FIN. Acknowledge 2187 * it and restart the finack timer. 2188 */ 2189 case TCPS_TIME_WAIT: 2190 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2191 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2192 tcp_timer_2msl, tp); 2193 goto dropafterack; 2194 } 2195 } 2196 2197step6: 2198 /* 2199 * Update window information. 2200 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2201 */ 2202 if ((thflags & TH_ACK) && 2203 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2204 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2205 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2206 /* keep track of pure window updates */ 2207 if (tlen == 0 && 2208 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2209 tcpstat.tcps_rcvwinupd++; 2210 tp->snd_wnd = tiwin; 2211 tp->snd_wl1 = th->th_seq; 2212 tp->snd_wl2 = th->th_ack; 2213 if (tp->snd_wnd > tp->max_sndwnd) 2214 tp->max_sndwnd = tp->snd_wnd; 2215 needoutput = 1; 2216 } 2217 2218 /* 2219 * Process segments with URG. 2220 */ 2221 if ((thflags & TH_URG) && th->th_urp && 2222 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2223 /* 2224 * This is a kludge, but if we receive and accept 2225 * random urgent pointers, we'll crash in 2226 * soreceive. It's hard to imagine someone 2227 * actually wanting to send this much urgent data. 2228 */ 2229 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2230 th->th_urp = 0; /* XXX */ 2231 thflags &= ~TH_URG; /* XXX */ 2232 goto dodata; /* XXX */ 2233 } 2234 /* 2235 * If this segment advances the known urgent pointer, 2236 * then mark the data stream. This should not happen 2237 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2238 * a FIN has been received from the remote side. 2239 * In these states we ignore the URG. 2240 * 2241 * According to RFC961 (Assigned Protocols), 2242 * the urgent pointer points to the last octet 2243 * of urgent data. We continue, however, 2244 * to consider it to indicate the first octet 2245 * of data past the urgent section as the original 2246 * spec states (in one of two places). 2247 */ 2248 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2249 tp->rcv_up = th->th_seq + th->th_urp; 2250 so->so_oobmark = so->so_rcv.sb_cc + 2251 (tp->rcv_up - tp->rcv_nxt) - 1; 2252 if (so->so_oobmark == 0)
|
2253 so->so_state |= SS_RCVATMARK;
| 2253 so->so_rcv.sb_state |= SBS_RCVATMARK;
|
2254 sohasoutofband(so); 2255 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2256 } 2257 /* 2258 * Remove out of band data so doesn't get presented to user. 2259 * This can happen independent of advancing the URG pointer, 2260 * but if two URG's are pending at once, some out-of-band 2261 * data may creep in... ick. 2262 */ 2263 if (th->th_urp <= (u_long)tlen && 2264 !(so->so_options & SO_OOBINLINE)) { 2265 /* hdr drop is delayed */ 2266 tcp_pulloutofband(so, th, m, drop_hdrlen); 2267 } 2268 } else { 2269 /* 2270 * If no out of band data is expected, 2271 * pull receive urgent pointer along 2272 * with the receive window. 2273 */ 2274 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2275 tp->rcv_up = tp->rcv_nxt; 2276 } 2277dodata: /* XXX */ 2278 KASSERT(headlocked, ("headlocked")); 2279 /* 2280 * Process the segment text, merging it into the TCP sequencing queue, 2281 * and arranging for acknowledgment of receipt if necessary. 2282 * This process logically involves adjusting tp->rcv_wnd as data 2283 * is presented to the user (this happens in tcp_usrreq.c, 2284 * case PRU_RCVD). If a FIN has already been received on this 2285 * connection then we just ignore the text. 2286 */ 2287 if ((tlen || (thflags & TH_FIN)) && 2288 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2289 m_adj(m, drop_hdrlen); /* delayed header drop */ 2290 /* 2291 * Insert segment which includes th into TCP reassembly queue 2292 * with control block tp. Set thflags to whether reassembly now 2293 * includes a segment with FIN. This handles the common case 2294 * inline (segment is the next to be received on an established 2295 * connection, and the queue is empty), avoiding linkage into 2296 * and removal from the queue and repetition of various 2297 * conversions. 2298 * Set DELACK for segments received in order, but ack 2299 * immediately when segments are out of order (so 2300 * fast retransmit can work). 2301 */ 2302 if (th->th_seq == tp->rcv_nxt && 2303 LIST_EMPTY(&tp->t_segq) && 2304 TCPS_HAVEESTABLISHED(tp->t_state)) { 2305 if (DELAY_ACK(tp)) 2306 tp->t_flags |= TF_DELACK; 2307 else 2308 tp->t_flags |= TF_ACKNOW; 2309 tp->rcv_nxt += tlen; 2310 thflags = th->th_flags & TH_FIN; 2311 tcpstat.tcps_rcvpack++; 2312 tcpstat.tcps_rcvbyte += tlen; 2313 ND6_HINT(tp);
| 2254 sohasoutofband(so); 2255 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2256 } 2257 /* 2258 * Remove out of band data so doesn't get presented to user. 2259 * This can happen independent of advancing the URG pointer, 2260 * but if two URG's are pending at once, some out-of-band 2261 * data may creep in... ick. 2262 */ 2263 if (th->th_urp <= (u_long)tlen && 2264 !(so->so_options & SO_OOBINLINE)) { 2265 /* hdr drop is delayed */ 2266 tcp_pulloutofband(so, th, m, drop_hdrlen); 2267 } 2268 } else { 2269 /* 2270 * If no out of band data is expected, 2271 * pull receive urgent pointer along 2272 * with the receive window. 2273 */ 2274 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2275 tp->rcv_up = tp->rcv_nxt; 2276 } 2277dodata: /* XXX */ 2278 KASSERT(headlocked, ("headlocked")); 2279 /* 2280 * Process the segment text, merging it into the TCP sequencing queue, 2281 * and arranging for acknowledgment of receipt if necessary. 2282 * This process logically involves adjusting tp->rcv_wnd as data 2283 * is presented to the user (this happens in tcp_usrreq.c, 2284 * case PRU_RCVD). If a FIN has already been received on this 2285 * connection then we just ignore the text. 2286 */ 2287 if ((tlen || (thflags & TH_FIN)) && 2288 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2289 m_adj(m, drop_hdrlen); /* delayed header drop */ 2290 /* 2291 * Insert segment which includes th into TCP reassembly queue 2292 * with control block tp. Set thflags to whether reassembly now 2293 * includes a segment with FIN. This handles the common case 2294 * inline (segment is the next to be received on an established 2295 * connection, and the queue is empty), avoiding linkage into 2296 * and removal from the queue and repetition of various 2297 * conversions. 2298 * Set DELACK for segments received in order, but ack 2299 * immediately when segments are out of order (so 2300 * fast retransmit can work). 2301 */ 2302 if (th->th_seq == tp->rcv_nxt && 2303 LIST_EMPTY(&tp->t_segq) && 2304 TCPS_HAVEESTABLISHED(tp->t_state)) { 2305 if (DELAY_ACK(tp)) 2306 tp->t_flags |= TF_DELACK; 2307 else 2308 tp->t_flags |= TF_ACKNOW; 2309 tp->rcv_nxt += tlen; 2310 thflags = th->th_flags & TH_FIN; 2311 tcpstat.tcps_rcvpack++; 2312 tcpstat.tcps_rcvbyte += tlen; 2313 ND6_HINT(tp);
|
2314 if (so->so_state & SS_CANTRCVMORE)
| 2314 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
|
2315 m_freem(m); 2316 else 2317 sbappendstream(&so->so_rcv, m); 2318 sorwakeup(so); 2319 } else { 2320 thflags = tcp_reass(tp, th, &tlen, m); 2321 tp->t_flags |= TF_ACKNOW; 2322 } 2323 2324 /* 2325 * Note the amount of data that peer has sent into 2326 * our window, in order to estimate the sender's 2327 * buffer size. 2328 */ 2329 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2330 } else { 2331 m_freem(m); 2332 thflags &= ~TH_FIN; 2333 } 2334 2335 /* 2336 * If FIN is received ACK the FIN and let the user know 2337 * that the connection is closing. 2338 */ 2339 if (thflags & TH_FIN) { 2340 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2341 socantrcvmore(so); 2342 /* 2343 * If connection is half-synchronized 2344 * (ie NEEDSYN flag on) then delay ACK, 2345 * so it may be piggybacked when SYN is sent. 2346 * Otherwise, since we received a FIN then no 2347 * more input can be expected, send ACK now. 2348 */ 2349 if (tp->t_flags & TF_NEEDSYN) 2350 tp->t_flags |= TF_DELACK; 2351 else 2352 tp->t_flags |= TF_ACKNOW; 2353 tp->rcv_nxt++; 2354 } 2355 switch (tp->t_state) { 2356 2357 /* 2358 * In SYN_RECEIVED and ESTABLISHED STATES 2359 * enter the CLOSE_WAIT state. 2360 */ 2361 case TCPS_SYN_RECEIVED: 2362 tp->t_starttime = ticks; 2363 /*FALLTHROUGH*/ 2364 case TCPS_ESTABLISHED: 2365 tp->t_state = TCPS_CLOSE_WAIT; 2366 break; 2367 2368 /* 2369 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2370 * enter the CLOSING state. 2371 */ 2372 case TCPS_FIN_WAIT_1: 2373 tp->t_state = TCPS_CLOSING; 2374 break; 2375 2376 /* 2377 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2378 * starting the time-wait timer, turning off the other 2379 * standard timers. 2380 */ 2381 case TCPS_FIN_WAIT_2: 2382 KASSERT(headlocked == 1, ("headlocked should be 1")); 2383 tcp_twstart(tp); 2384 INP_INFO_WUNLOCK(&tcbinfo); 2385 return; 2386 2387 /* 2388 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2389 */ 2390 case TCPS_TIME_WAIT: 2391 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2392 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2393 tcp_timer_2msl, tp); 2394 break; 2395 } 2396 } 2397 INP_INFO_WUNLOCK(&tcbinfo); 2398#ifdef TCPDEBUG 2399 if (so->so_options & SO_DEBUG) 2400 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2401 &tcp_savetcp, 0); 2402#endif 2403 2404 /* 2405 * Return any desired output. 2406 */ 2407 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2408 (void) tcp_output(tp); 2409 2410check_delack: 2411 if (tp->t_flags & TF_DELACK) { 2412 tp->t_flags &= ~TF_DELACK; 2413 callout_reset(tp->tt_delack, tcp_delacktime, 2414 tcp_timer_delack, tp); 2415 } 2416 INP_UNLOCK(inp); 2417 return; 2418 2419dropafterack: 2420 /* 2421 * Generate an ACK dropping incoming segment if it occupies 2422 * sequence space, where the ACK reflects our state. 2423 * 2424 * We can now skip the test for the RST flag since all 2425 * paths to this code happen after packets containing 2426 * RST have been dropped. 2427 * 2428 * In the SYN-RECEIVED state, don't send an ACK unless the 2429 * segment we received passes the SYN-RECEIVED ACK test. 2430 * If it fails send a RST. This breaks the loop in the 2431 * "LAND" DoS attack, and also prevents an ACK storm 2432 * between two listening ports that have been sent forged 2433 * SYN segments, each with the source address of the other. 2434 */ 2435 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2436 (SEQ_GT(tp->snd_una, th->th_ack) || 2437 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2438 rstreason = BANDLIM_RST_OPENPORT; 2439 goto dropwithreset; 2440 } 2441#ifdef TCPDEBUG 2442 if (so->so_options & SO_DEBUG) 2443 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2444 &tcp_savetcp, 0); 2445#endif 2446 KASSERT(headlocked, ("headlocked should be 1")); 2447 INP_INFO_WUNLOCK(&tcbinfo); 2448 m_freem(m); 2449 tp->t_flags |= TF_ACKNOW; 2450 (void) tcp_output(tp); 2451 INP_UNLOCK(inp); 2452 return; 2453 2454dropwithreset: 2455 /* 2456 * Generate a RST, dropping incoming segment. 2457 * Make ACK acceptable to originator of segment. 2458 * Don't bother to respond if destination was broadcast/multicast. 2459 */ 2460 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2461 goto drop; 2462 if (isipv6) { 2463 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2464 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2465 goto drop; 2466 } else { 2467 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2468 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2469 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2470 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2471 goto drop; 2472 } 2473 /* IPv6 anycast check is done at tcp6_input() */ 2474 2475 /* 2476 * Perform bandwidth limiting. 2477 */ 2478 if (badport_bandlim(rstreason) < 0) 2479 goto drop; 2480 2481#ifdef TCPDEBUG 2482 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2483 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2484 &tcp_savetcp, 0); 2485#endif 2486 2487 if (thflags & TH_ACK) 2488 /* mtod() below is safe as long as hdr dropping is delayed */ 2489 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2490 TH_RST); 2491 else { 2492 if (thflags & TH_SYN) 2493 tlen++; 2494 /* mtod() below is safe as long as hdr dropping is delayed */ 2495 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2496 (tcp_seq)0, TH_RST|TH_ACK); 2497 } 2498 2499 if (tp) 2500 INP_UNLOCK(inp); 2501 if (headlocked) 2502 INP_INFO_WUNLOCK(&tcbinfo); 2503 return; 2504 2505drop: 2506 /* 2507 * Drop space held by incoming segment and return. 2508 */ 2509#ifdef TCPDEBUG 2510 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2511 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2512 &tcp_savetcp, 0); 2513#endif 2514 if (tp) 2515 INP_UNLOCK(inp); 2516 m_freem(m); 2517 if (headlocked) 2518 INP_INFO_WUNLOCK(&tcbinfo); 2519 return; 2520} 2521 2522/* 2523 * Parse TCP options and place in tcpopt. 2524 */ 2525static void 2526tcp_dooptions(to, cp, cnt, is_syn) 2527 struct tcpopt *to; 2528 u_char *cp; 2529 int cnt; 2530 int is_syn; 2531{ 2532 int opt, optlen; 2533 2534 to->to_flags = 0; 2535 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2536 opt = cp[0]; 2537 if (opt == TCPOPT_EOL) 2538 break; 2539 if (opt == TCPOPT_NOP) 2540 optlen = 1; 2541 else { 2542 if (cnt < 2) 2543 break; 2544 optlen = cp[1]; 2545 if (optlen < 2 || optlen > cnt) 2546 break; 2547 } 2548 switch (opt) { 2549 case TCPOPT_MAXSEG: 2550 if (optlen != TCPOLEN_MAXSEG) 2551 continue; 2552 if (!is_syn) 2553 continue; 2554 to->to_flags |= TOF_MSS; 2555 bcopy((char *)cp + 2, 2556 (char *)&to->to_mss, sizeof(to->to_mss)); 2557 to->to_mss = ntohs(to->to_mss); 2558 break; 2559 case TCPOPT_WINDOW: 2560 if (optlen != TCPOLEN_WINDOW) 2561 continue; 2562 if (! is_syn) 2563 continue; 2564 to->to_flags |= TOF_SCALE; 2565 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2566 break; 2567 case TCPOPT_TIMESTAMP: 2568 if (optlen != TCPOLEN_TIMESTAMP) 2569 continue; 2570 to->to_flags |= TOF_TS; 2571 bcopy((char *)cp + 2, 2572 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2573 to->to_tsval = ntohl(to->to_tsval); 2574 bcopy((char *)cp + 6, 2575 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2576 to->to_tsecr = ntohl(to->to_tsecr); 2577 break; 2578 case TCPOPT_CC: 2579 if (optlen != TCPOLEN_CC) 2580 continue; 2581 to->to_flags |= TOF_CC; 2582 bcopy((char *)cp + 2, 2583 (char *)&to->to_cc, sizeof(to->to_cc)); 2584 to->to_cc = ntohl(to->to_cc); 2585 break; 2586 case TCPOPT_CCNEW: 2587 if (optlen != TCPOLEN_CC) 2588 continue; 2589 if (!is_syn) 2590 continue; 2591 to->to_flags |= TOF_CCNEW; 2592 bcopy((char *)cp + 2, 2593 (char *)&to->to_cc, sizeof(to->to_cc)); 2594 to->to_cc = ntohl(to->to_cc); 2595 break; 2596 case TCPOPT_CCECHO: 2597 if (optlen != TCPOLEN_CC) 2598 continue; 2599 if (!is_syn) 2600 continue; 2601 to->to_flags |= TOF_CCECHO; 2602 bcopy((char *)cp + 2, 2603 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2604 to->to_ccecho = ntohl(to->to_ccecho); 2605 break; 2606#ifdef TCP_SIGNATURE 2607 /* 2608 * XXX In order to reply to a host which has set the 2609 * TCP_SIGNATURE option in its initial SYN, we have to 2610 * record the fact that the option was observed here 2611 * for the syncache code to perform the correct response. 2612 */ 2613 case TCPOPT_SIGNATURE: 2614 if (optlen != TCPOLEN_SIGNATURE) 2615 continue; 2616 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2617 break; 2618#endif 2619 default: 2620 continue; 2621 } 2622 } 2623} 2624 2625/* 2626 * Pull out of band byte out of a segment so 2627 * it doesn't appear in the user's data queue. 2628 * It is still reflected in the segment length for 2629 * sequencing purposes. 2630 */ 2631static void 2632tcp_pulloutofband(so, th, m, off) 2633 struct socket *so; 2634 struct tcphdr *th; 2635 register struct mbuf *m; 2636 int off; /* delayed to be droped hdrlen */ 2637{ 2638 int cnt = off + th->th_urp - 1; 2639 2640 while (cnt >= 0) { 2641 if (m->m_len > cnt) { 2642 char *cp = mtod(m, caddr_t) + cnt; 2643 struct tcpcb *tp = sototcpcb(so); 2644 2645 tp->t_iobc = *cp; 2646 tp->t_oobflags |= TCPOOB_HAVEDATA; 2647 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2648 m->m_len--; 2649 if (m->m_flags & M_PKTHDR) 2650 m->m_pkthdr.len--; 2651 return; 2652 } 2653 cnt -= m->m_len; 2654 m = m->m_next; 2655 if (m == 0) 2656 break; 2657 } 2658 panic("tcp_pulloutofband"); 2659} 2660 2661/* 2662 * Collect new round-trip time estimate 2663 * and update averages and current timeout. 2664 */ 2665static void 2666tcp_xmit_timer(tp, rtt) 2667 register struct tcpcb *tp; 2668 int rtt; 2669{ 2670 register int delta; 2671 2672 tcpstat.tcps_rttupdated++; 2673 tp->t_rttupdated++; 2674 if (tp->t_srtt != 0) { 2675 /* 2676 * srtt is stored as fixed point with 5 bits after the 2677 * binary point (i.e., scaled by 8). The following magic 2678 * is equivalent to the smoothing algorithm in rfc793 with 2679 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2680 * point). Adjust rtt to origin 0. 2681 */ 2682 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2683 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2684 2685 if ((tp->t_srtt += delta) <= 0) 2686 tp->t_srtt = 1; 2687 2688 /* 2689 * We accumulate a smoothed rtt variance (actually, a 2690 * smoothed mean difference), then set the retransmit 2691 * timer to smoothed rtt + 4 times the smoothed variance. 2692 * rttvar is stored as fixed point with 4 bits after the 2693 * binary point (scaled by 16). The following is 2694 * equivalent to rfc793 smoothing with an alpha of .75 2695 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2696 * rfc793's wired-in beta. 2697 */ 2698 if (delta < 0) 2699 delta = -delta; 2700 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2701 if ((tp->t_rttvar += delta) <= 0) 2702 tp->t_rttvar = 1; 2703 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2704 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2705 } else { 2706 /* 2707 * No rtt measurement yet - use the unsmoothed rtt. 2708 * Set the variance to half the rtt (so our first 2709 * retransmit happens at 3*rtt). 2710 */ 2711 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2712 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2713 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2714 } 2715 tp->t_rtttime = 0; 2716 tp->t_rxtshift = 0; 2717 2718 /* 2719 * the retransmit should happen at rtt + 4 * rttvar. 2720 * Because of the way we do the smoothing, srtt and rttvar 2721 * will each average +1/2 tick of bias. When we compute 2722 * the retransmit timer, we want 1/2 tick of rounding and 2723 * 1 extra tick because of +-1/2 tick uncertainty in the 2724 * firing of the timer. The bias will give us exactly the 2725 * 1.5 tick we need. But, because the bias is 2726 * statistical, we have to test that we don't drop below 2727 * the minimum feasible timer (which is 2 ticks). 2728 */ 2729 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2730 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2731 2732 /* 2733 * We received an ack for a packet that wasn't retransmitted; 2734 * it is probably safe to discard any error indications we've 2735 * received recently. This isn't quite right, but close enough 2736 * for now (a route might have failed after we sent a segment, 2737 * and the return path might not be symmetrical). 2738 */ 2739 tp->t_softerror = 0; 2740} 2741 2742/* 2743 * Determine a reasonable value for maxseg size. 2744 * If the route is known, check route for mtu. 2745 * If none, use an mss that can be handled on the outgoing 2746 * interface without forcing IP to fragment; if bigger than 2747 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2748 * to utilize large mbufs. If no route is found, route has no mtu, 2749 * or the destination isn't local, use a default, hopefully conservative 2750 * size (usually 512 or the default IP max size, but no more than the mtu 2751 * of the interface), as we can't discover anything about intervening 2752 * gateways or networks. We also initialize the congestion/slow start 2753 * window to be a single segment if the destination isn't local. 2754 * While looking at the routing entry, we also initialize other path-dependent 2755 * parameters from pre-set or cached values in the routing entry. 2756 * 2757 * Also take into account the space needed for options that we 2758 * send regularly. Make maxseg shorter by that amount to assure 2759 * that we can send maxseg amount of data even when the options 2760 * are present. Store the upper limit of the length of options plus 2761 * data in maxopd. 2762 * 2763 * 2764 * In case of T/TCP, we call this routine during implicit connection 2765 * setup as well (offer = -1), to initialize maxseg from the cached 2766 * MSS of our peer. 2767 * 2768 * NOTE that this routine is only called when we process an incoming 2769 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2770 */ 2771void 2772tcp_mss(tp, offer) 2773 struct tcpcb *tp; 2774 int offer; 2775{ 2776 int rtt, mss; 2777 u_long bufsize; 2778 u_long maxmtu; 2779 struct inpcb *inp = tp->t_inpcb; 2780 struct socket *so; 2781 struct hc_metrics_lite metrics; 2782 struct rmxp_tao tao; 2783 int origoffer = offer; 2784#ifdef INET6 2785 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2786 size_t min_protoh = isipv6 ? 2787 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2788 sizeof (struct tcpiphdr); 2789#else 2790 const size_t min_protoh = sizeof(struct tcpiphdr); 2791#endif 2792 bzero(&tao, sizeof(tao)); 2793 2794 /* initialize */ 2795#ifdef INET6 2796 if (isipv6) { 2797 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2798 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2799 } else 2800#endif 2801 { 2802 maxmtu = tcp_maxmtu(&inp->inp_inc); 2803 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2804 } 2805 so = inp->inp_socket; 2806 2807 /* 2808 * no route to sender, stay with default mss and return 2809 */ 2810 if (maxmtu == 0) 2811 return; 2812 2813 /* what have we got? */ 2814 switch (offer) { 2815 case 0: 2816 /* 2817 * Offer == 0 means that there was no MSS on the SYN 2818 * segment, in this case we use tcp_mssdflt. 2819 */ 2820 offer = 2821#ifdef INET6 2822 isipv6 ? tcp_v6mssdflt : 2823#endif 2824 tcp_mssdflt; 2825 break; 2826 2827 case -1: 2828 /* 2829 * Offer == -1 means that we didn't receive SYN yet, 2830 * use cached value in that case; 2831 */ 2832 if (tcp_do_rfc1644) 2833 tcp_hc_gettao(&inp->inp_inc, &tao); 2834 if (tao.tao_mssopt != 0) 2835 offer = tao.tao_mssopt; 2836 /* FALLTHROUGH */ 2837 2838 default: 2839 /* 2840 * Prevent DoS attack with too small MSS. Round up 2841 * to at least minmss. 2842 */ 2843 offer = max(offer, tcp_minmss); 2844 /* 2845 * Sanity check: make sure that maxopd will be large 2846 * enough to allow some data on segments even if the 2847 * all the option space is used (40bytes). Otherwise 2848 * funny things may happen in tcp_output. 2849 */ 2850 offer = max(offer, 64); 2851 if (tcp_do_rfc1644) 2852 tcp_hc_updatetao(&inp->inp_inc, 2853 TCP_HC_TAO_MSSOPT, 0, offer); 2854 } 2855 2856 /* 2857 * rmx information is now retrieved from tcp_hostcache 2858 */ 2859 tcp_hc_get(&inp->inp_inc, &metrics); 2860 2861 /* 2862 * if there's a discovered mtu int tcp hostcache, use it 2863 * else, use the link mtu. 2864 */ 2865 if (metrics.rmx_mtu) 2866 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2867 else { 2868#ifdef INET6 2869 if (isipv6) { 2870 mss = maxmtu - min_protoh; 2871 if (!path_mtu_discovery && 2872 !in6_localaddr(&inp->in6p_faddr)) 2873 mss = min(mss, tcp_v6mssdflt); 2874 } else 2875#endif 2876 { 2877 mss = maxmtu - min_protoh; 2878 if (!path_mtu_discovery && 2879 !in_localaddr(inp->inp_faddr)) 2880 mss = min(mss, tcp_mssdflt); 2881 } 2882 } 2883 mss = min(mss, offer); 2884 2885 /* 2886 * maxopd stores the maximum length of data AND options 2887 * in a segment; maxseg is the amount of data in a normal 2888 * segment. We need to store this value (maxopd) apart 2889 * from maxseg, because now every segment carries options 2890 * and thus we normally have somewhat less data in segments. 2891 */ 2892 tp->t_maxopd = mss; 2893 2894 /* 2895 * In case of T/TCP, origoffer==-1 indicates, that no segments 2896 * were received yet. In this case we just guess, otherwise 2897 * we do the same as before T/TCP. 2898 */ 2899 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2900 (origoffer == -1 || 2901 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2902 mss -= TCPOLEN_TSTAMP_APPA; 2903 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2904 (origoffer == -1 || 2905 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2906 mss -= TCPOLEN_CC_APPA; 2907 tp->t_maxseg = mss; 2908 2909#if (MCLBYTES & (MCLBYTES - 1)) == 0 2910 if (mss > MCLBYTES) 2911 mss &= ~(MCLBYTES-1); 2912#else 2913 if (mss > MCLBYTES) 2914 mss = mss / MCLBYTES * MCLBYTES; 2915#endif 2916 tp->t_maxseg = mss; 2917 2918 /* 2919 * If there's a pipesize, change the socket buffer to that size, 2920 * don't change if sb_hiwat is different than default (then it 2921 * has been changed on purpose with setsockopt). 2922 * Make the socket buffers an integral number of mss units; 2923 * if the mss is larger than the socket buffer, decrease the mss. 2924 */ 2925 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2926 bufsize = metrics.rmx_sendpipe; 2927 else 2928 bufsize = so->so_snd.sb_hiwat; 2929 if (bufsize < mss) 2930 mss = bufsize; 2931 else { 2932 bufsize = roundup(bufsize, mss); 2933 if (bufsize > sb_max) 2934 bufsize = sb_max; 2935 if (bufsize > so->so_snd.sb_hiwat) 2936 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2937 } 2938 tp->t_maxseg = mss; 2939 2940 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2941 bufsize = metrics.rmx_recvpipe; 2942 else 2943 bufsize = so->so_rcv.sb_hiwat; 2944 if (bufsize > mss) { 2945 bufsize = roundup(bufsize, mss); 2946 if (bufsize > sb_max) 2947 bufsize = sb_max; 2948 if (bufsize > so->so_rcv.sb_hiwat) 2949 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2950 } 2951 /* 2952 * While we're here, check the others too 2953 */ 2954 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2955 tp->t_srtt = rtt; 2956 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2957 tcpstat.tcps_usedrtt++; 2958 if (metrics.rmx_rttvar) { 2959 tp->t_rttvar = metrics.rmx_rttvar; 2960 tcpstat.tcps_usedrttvar++; 2961 } else { 2962 /* default variation is +- 1 rtt */ 2963 tp->t_rttvar = 2964 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2965 } 2966 TCPT_RANGESET(tp->t_rxtcur, 2967 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2968 tp->t_rttmin, TCPTV_REXMTMAX); 2969 } 2970 if (metrics.rmx_ssthresh) { 2971 /* 2972 * There's some sort of gateway or interface 2973 * buffer limit on the path. Use this to set 2974 * the slow start threshhold, but set the 2975 * threshold to no less than 2*mss. 2976 */ 2977 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2978 tcpstat.tcps_usedssthresh++; 2979 } 2980 if (metrics.rmx_bandwidth) 2981 tp->snd_bandwidth = metrics.rmx_bandwidth; 2982 2983 /* 2984 * Set the slow-start flight size depending on whether this 2985 * is a local network or not. 2986 * 2987 * Extend this so we cache the cwnd too and retrieve it here. 2988 * Make cwnd even bigger than RFC3390 suggests but only if we 2989 * have previous experience with the remote host. Be careful 2990 * not make cwnd bigger than remote receive window or our own 2991 * send socket buffer. Maybe put some additional upper bound 2992 * on the retrieved cwnd. Should do incremental updates to 2993 * hostcache when cwnd collapses so next connection doesn't 2994 * overloads the path again. 2995 * 2996 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2997 * We currently check only in syncache_socket for that. 2998 */ 2999#define TCP_METRICS_CWND 3000#ifdef TCP_METRICS_CWND 3001 if (metrics.rmx_cwnd) 3002 tp->snd_cwnd = max(mss, 3003 min(metrics.rmx_cwnd / 2, 3004 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3005 else 3006#endif 3007 if (tcp_do_rfc3390) 3008 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3009#ifdef INET6 3010 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3011 (!isipv6 && in_localaddr(inp->inp_faddr))) 3012#else 3013 else if (in_localaddr(inp->inp_faddr)) 3014#endif 3015 tp->snd_cwnd = mss * ss_fltsz_local; 3016 else 3017 tp->snd_cwnd = mss * ss_fltsz; 3018} 3019 3020/* 3021 * Determine the MSS option to send on an outgoing SYN. 3022 */ 3023int 3024tcp_mssopt(inc) 3025 struct in_conninfo *inc; 3026{ 3027 int mss = 0; 3028 u_long maxmtu = 0; 3029 u_long thcmtu = 0; 3030 size_t min_protoh; 3031#ifdef INET6 3032 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3033#endif 3034 3035 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3036 3037#ifdef INET6 3038 if (isipv6) { 3039 mss = tcp_v6mssdflt; 3040 maxmtu = tcp_maxmtu6(inc); 3041 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3042 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3043 } else 3044#endif 3045 { 3046 mss = tcp_mssdflt; 3047 maxmtu = tcp_maxmtu(inc); 3048 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3049 min_protoh = sizeof(struct tcpiphdr); 3050 } 3051 if (maxmtu && thcmtu) 3052 mss = min(maxmtu, thcmtu) - min_protoh; 3053 else if (maxmtu || thcmtu) 3054 mss = max(maxmtu, thcmtu) - min_protoh; 3055 3056 return (mss); 3057} 3058 3059 3060/* 3061 * On a partial ack arrives, force the retransmission of the 3062 * next unacknowledged segment. Do not clear tp->t_dupacks. 3063 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3064 * be started again. 3065 */ 3066static void 3067tcp_newreno_partial_ack(tp, th) 3068 struct tcpcb *tp; 3069 struct tcphdr *th; 3070{ 3071 tcp_seq onxt = tp->snd_nxt; 3072 u_long ocwnd = tp->snd_cwnd; 3073 3074 callout_stop(tp->tt_rexmt); 3075 tp->t_rtttime = 0; 3076 tp->snd_nxt = th->th_ack; 3077 /* 3078 * Set snd_cwnd to one segment beyond acknowledged offset. 3079 * (tp->snd_una has not yet been updated when this function is called.) 3080 */ 3081 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3082 tp->t_flags |= TF_ACKNOW; 3083 (void) tcp_output(tp); 3084 tp->snd_cwnd = ocwnd; 3085 if (SEQ_GT(onxt, tp->snd_nxt)) 3086 tp->snd_nxt = onxt; 3087 /* 3088 * Partial window deflation. Relies on fact that tp->snd_una 3089 * not updated yet. 3090 */ 3091 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3092} 3093 3094/* 3095 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3096 * looking for a pcb in the listen state. Returns 0 otherwise. 3097 */ 3098static int 3099tcp_timewait(tw, to, th, m, tlen) 3100 struct tcptw *tw; 3101 struct tcpopt *to; 3102 struct tcphdr *th; 3103 struct mbuf *m; 3104 int tlen; 3105{ 3106 int thflags; 3107 tcp_seq seq; 3108#ifdef INET6 3109 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3110#else 3111 const int isipv6 = 0; 3112#endif 3113 3114 thflags = th->th_flags; 3115 3116 /* 3117 * NOTE: for FIN_WAIT_2 (to be added later), 3118 * must validate sequence number before accepting RST 3119 */ 3120 3121 /* 3122 * If the segment contains RST: 3123 * Drop the segment - see Stevens, vol. 2, p. 964 and 3124 * RFC 1337. 3125 */ 3126 if (thflags & TH_RST) 3127 goto drop; 3128 3129 /* 3130 * If segment contains a SYN and CC [not CC.NEW] option: 3131 * if connection duration > MSL, drop packet and send RST; 3132 * 3133 * if SEG.CC > CCrecv then is new SYN. 3134 * Complete close and delete TCPCB. Then reprocess 3135 * segment, hoping to find new TCPCB in LISTEN state; 3136 * 3137 * else must be old SYN; drop it. 3138 * else do normal processing. 3139 */ 3140 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3141 if ((ticks - tw->t_starttime) > tcp_msl) 3142 goto reset; 3143 if (CC_GT(to->to_cc, tw->cc_recv)) { 3144 (void) tcp_twclose(tw, 0); 3145 return (1); 3146 } 3147 goto drop; 3148 } 3149 3150#if 0 3151/* PAWS not needed at the moment */ 3152 /* 3153 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3154 * and it's less than ts_recent, drop it. 3155 */ 3156 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3157 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3158 if ((thflags & TH_ACK) == 0) 3159 goto drop; 3160 goto ack; 3161 } 3162 /* 3163 * ts_recent is never updated because we never accept new segments. 3164 */ 3165#endif 3166 3167 /* 3168 * If a new connection request is received 3169 * while in TIME_WAIT, drop the old connection 3170 * and start over if the sequence numbers 3171 * are above the previous ones. 3172 */ 3173 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3174 (void) tcp_twclose(tw, 0); 3175 return (1); 3176 } 3177 3178 /* 3179 * Drop the the segment if it does not contain an ACK. 3180 */ 3181 if ((thflags & TH_ACK) == 0) 3182 goto drop; 3183 3184 /* 3185 * Reset the 2MSL timer if this is a duplicate FIN. 3186 */ 3187 if (thflags & TH_FIN) { 3188 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3189 if (seq + 1 == tw->rcv_nxt) 3190 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3191 } 3192 3193 /* 3194 * Acknowledge the segment if it has data or is not a duplicate ACK. 3195 */ 3196 if (thflags != TH_ACK || tlen != 0 || 3197 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3198 tcp_twrespond(tw, TH_ACK); 3199 goto drop; 3200 3201reset: 3202 /* 3203 * Generate a RST, dropping incoming segment. 3204 * Make ACK acceptable to originator of segment. 3205 * Don't bother to respond if destination was broadcast/multicast. 3206 */ 3207 if (m->m_flags & (M_BCAST|M_MCAST)) 3208 goto drop; 3209 if (isipv6) { 3210 struct ip6_hdr *ip6; 3211 3212 /* IPv6 anycast check is done at tcp6_input() */ 3213 ip6 = mtod(m, struct ip6_hdr *); 3214 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3215 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3216 goto drop; 3217 } else { 3218 struct ip *ip; 3219 3220 ip = mtod(m, struct ip *); 3221 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3222 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3223 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3224 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3225 goto drop; 3226 } 3227 if (thflags & TH_ACK) { 3228 tcp_respond(NULL, 3229 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3230 } else { 3231 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3232 tcp_respond(NULL, 3233 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3234 } 3235 INP_UNLOCK(tw->tw_inpcb); 3236 return (0); 3237 3238drop: 3239 INP_UNLOCK(tw->tw_inpcb); 3240 m_freem(m); 3241 return (0); 3242}
| 2315 m_freem(m); 2316 else 2317 sbappendstream(&so->so_rcv, m); 2318 sorwakeup(so); 2319 } else { 2320 thflags = tcp_reass(tp, th, &tlen, m); 2321 tp->t_flags |= TF_ACKNOW; 2322 } 2323 2324 /* 2325 * Note the amount of data that peer has sent into 2326 * our window, in order to estimate the sender's 2327 * buffer size. 2328 */ 2329 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2330 } else { 2331 m_freem(m); 2332 thflags &= ~TH_FIN; 2333 } 2334 2335 /* 2336 * If FIN is received ACK the FIN and let the user know 2337 * that the connection is closing. 2338 */ 2339 if (thflags & TH_FIN) { 2340 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2341 socantrcvmore(so); 2342 /* 2343 * If connection is half-synchronized 2344 * (ie NEEDSYN flag on) then delay ACK, 2345 * so it may be piggybacked when SYN is sent. 2346 * Otherwise, since we received a FIN then no 2347 * more input can be expected, send ACK now. 2348 */ 2349 if (tp->t_flags & TF_NEEDSYN) 2350 tp->t_flags |= TF_DELACK; 2351 else 2352 tp->t_flags |= TF_ACKNOW; 2353 tp->rcv_nxt++; 2354 } 2355 switch (tp->t_state) { 2356 2357 /* 2358 * In SYN_RECEIVED and ESTABLISHED STATES 2359 * enter the CLOSE_WAIT state. 2360 */ 2361 case TCPS_SYN_RECEIVED: 2362 tp->t_starttime = ticks; 2363 /*FALLTHROUGH*/ 2364 case TCPS_ESTABLISHED: 2365 tp->t_state = TCPS_CLOSE_WAIT; 2366 break; 2367 2368 /* 2369 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2370 * enter the CLOSING state. 2371 */ 2372 case TCPS_FIN_WAIT_1: 2373 tp->t_state = TCPS_CLOSING; 2374 break; 2375 2376 /* 2377 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2378 * starting the time-wait timer, turning off the other 2379 * standard timers. 2380 */ 2381 case TCPS_FIN_WAIT_2: 2382 KASSERT(headlocked == 1, ("headlocked should be 1")); 2383 tcp_twstart(tp); 2384 INP_INFO_WUNLOCK(&tcbinfo); 2385 return; 2386 2387 /* 2388 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2389 */ 2390 case TCPS_TIME_WAIT: 2391 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2392 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2393 tcp_timer_2msl, tp); 2394 break; 2395 } 2396 } 2397 INP_INFO_WUNLOCK(&tcbinfo); 2398#ifdef TCPDEBUG 2399 if (so->so_options & SO_DEBUG) 2400 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2401 &tcp_savetcp, 0); 2402#endif 2403 2404 /* 2405 * Return any desired output. 2406 */ 2407 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2408 (void) tcp_output(tp); 2409 2410check_delack: 2411 if (tp->t_flags & TF_DELACK) { 2412 tp->t_flags &= ~TF_DELACK; 2413 callout_reset(tp->tt_delack, tcp_delacktime, 2414 tcp_timer_delack, tp); 2415 } 2416 INP_UNLOCK(inp); 2417 return; 2418 2419dropafterack: 2420 /* 2421 * Generate an ACK dropping incoming segment if it occupies 2422 * sequence space, where the ACK reflects our state. 2423 * 2424 * We can now skip the test for the RST flag since all 2425 * paths to this code happen after packets containing 2426 * RST have been dropped. 2427 * 2428 * In the SYN-RECEIVED state, don't send an ACK unless the 2429 * segment we received passes the SYN-RECEIVED ACK test. 2430 * If it fails send a RST. This breaks the loop in the 2431 * "LAND" DoS attack, and also prevents an ACK storm 2432 * between two listening ports that have been sent forged 2433 * SYN segments, each with the source address of the other. 2434 */ 2435 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2436 (SEQ_GT(tp->snd_una, th->th_ack) || 2437 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2438 rstreason = BANDLIM_RST_OPENPORT; 2439 goto dropwithreset; 2440 } 2441#ifdef TCPDEBUG 2442 if (so->so_options & SO_DEBUG) 2443 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2444 &tcp_savetcp, 0); 2445#endif 2446 KASSERT(headlocked, ("headlocked should be 1")); 2447 INP_INFO_WUNLOCK(&tcbinfo); 2448 m_freem(m); 2449 tp->t_flags |= TF_ACKNOW; 2450 (void) tcp_output(tp); 2451 INP_UNLOCK(inp); 2452 return; 2453 2454dropwithreset: 2455 /* 2456 * Generate a RST, dropping incoming segment. 2457 * Make ACK acceptable to originator of segment. 2458 * Don't bother to respond if destination was broadcast/multicast. 2459 */ 2460 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2461 goto drop; 2462 if (isipv6) { 2463 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2464 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2465 goto drop; 2466 } else { 2467 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2468 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2469 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2470 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2471 goto drop; 2472 } 2473 /* IPv6 anycast check is done at tcp6_input() */ 2474 2475 /* 2476 * Perform bandwidth limiting. 2477 */ 2478 if (badport_bandlim(rstreason) < 0) 2479 goto drop; 2480 2481#ifdef TCPDEBUG 2482 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2483 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2484 &tcp_savetcp, 0); 2485#endif 2486 2487 if (thflags & TH_ACK) 2488 /* mtod() below is safe as long as hdr dropping is delayed */ 2489 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2490 TH_RST); 2491 else { 2492 if (thflags & TH_SYN) 2493 tlen++; 2494 /* mtod() below is safe as long as hdr dropping is delayed */ 2495 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2496 (tcp_seq)0, TH_RST|TH_ACK); 2497 } 2498 2499 if (tp) 2500 INP_UNLOCK(inp); 2501 if (headlocked) 2502 INP_INFO_WUNLOCK(&tcbinfo); 2503 return; 2504 2505drop: 2506 /* 2507 * Drop space held by incoming segment and return. 2508 */ 2509#ifdef TCPDEBUG 2510 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2511 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2512 &tcp_savetcp, 0); 2513#endif 2514 if (tp) 2515 INP_UNLOCK(inp); 2516 m_freem(m); 2517 if (headlocked) 2518 INP_INFO_WUNLOCK(&tcbinfo); 2519 return; 2520} 2521 2522/* 2523 * Parse TCP options and place in tcpopt. 2524 */ 2525static void 2526tcp_dooptions(to, cp, cnt, is_syn) 2527 struct tcpopt *to; 2528 u_char *cp; 2529 int cnt; 2530 int is_syn; 2531{ 2532 int opt, optlen; 2533 2534 to->to_flags = 0; 2535 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2536 opt = cp[0]; 2537 if (opt == TCPOPT_EOL) 2538 break; 2539 if (opt == TCPOPT_NOP) 2540 optlen = 1; 2541 else { 2542 if (cnt < 2) 2543 break; 2544 optlen = cp[1]; 2545 if (optlen < 2 || optlen > cnt) 2546 break; 2547 } 2548 switch (opt) { 2549 case TCPOPT_MAXSEG: 2550 if (optlen != TCPOLEN_MAXSEG) 2551 continue; 2552 if (!is_syn) 2553 continue; 2554 to->to_flags |= TOF_MSS; 2555 bcopy((char *)cp + 2, 2556 (char *)&to->to_mss, sizeof(to->to_mss)); 2557 to->to_mss = ntohs(to->to_mss); 2558 break; 2559 case TCPOPT_WINDOW: 2560 if (optlen != TCPOLEN_WINDOW) 2561 continue; 2562 if (! is_syn) 2563 continue; 2564 to->to_flags |= TOF_SCALE; 2565 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2566 break; 2567 case TCPOPT_TIMESTAMP: 2568 if (optlen != TCPOLEN_TIMESTAMP) 2569 continue; 2570 to->to_flags |= TOF_TS; 2571 bcopy((char *)cp + 2, 2572 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2573 to->to_tsval = ntohl(to->to_tsval); 2574 bcopy((char *)cp + 6, 2575 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2576 to->to_tsecr = ntohl(to->to_tsecr); 2577 break; 2578 case TCPOPT_CC: 2579 if (optlen != TCPOLEN_CC) 2580 continue; 2581 to->to_flags |= TOF_CC; 2582 bcopy((char *)cp + 2, 2583 (char *)&to->to_cc, sizeof(to->to_cc)); 2584 to->to_cc = ntohl(to->to_cc); 2585 break; 2586 case TCPOPT_CCNEW: 2587 if (optlen != TCPOLEN_CC) 2588 continue; 2589 if (!is_syn) 2590 continue; 2591 to->to_flags |= TOF_CCNEW; 2592 bcopy((char *)cp + 2, 2593 (char *)&to->to_cc, sizeof(to->to_cc)); 2594 to->to_cc = ntohl(to->to_cc); 2595 break; 2596 case TCPOPT_CCECHO: 2597 if (optlen != TCPOLEN_CC) 2598 continue; 2599 if (!is_syn) 2600 continue; 2601 to->to_flags |= TOF_CCECHO; 2602 bcopy((char *)cp + 2, 2603 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2604 to->to_ccecho = ntohl(to->to_ccecho); 2605 break; 2606#ifdef TCP_SIGNATURE 2607 /* 2608 * XXX In order to reply to a host which has set the 2609 * TCP_SIGNATURE option in its initial SYN, we have to 2610 * record the fact that the option was observed here 2611 * for the syncache code to perform the correct response. 2612 */ 2613 case TCPOPT_SIGNATURE: 2614 if (optlen != TCPOLEN_SIGNATURE) 2615 continue; 2616 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2617 break; 2618#endif 2619 default: 2620 continue; 2621 } 2622 } 2623} 2624 2625/* 2626 * Pull out of band byte out of a segment so 2627 * it doesn't appear in the user's data queue. 2628 * It is still reflected in the segment length for 2629 * sequencing purposes. 2630 */ 2631static void 2632tcp_pulloutofband(so, th, m, off) 2633 struct socket *so; 2634 struct tcphdr *th; 2635 register struct mbuf *m; 2636 int off; /* delayed to be droped hdrlen */ 2637{ 2638 int cnt = off + th->th_urp - 1; 2639 2640 while (cnt >= 0) { 2641 if (m->m_len > cnt) { 2642 char *cp = mtod(m, caddr_t) + cnt; 2643 struct tcpcb *tp = sototcpcb(so); 2644 2645 tp->t_iobc = *cp; 2646 tp->t_oobflags |= TCPOOB_HAVEDATA; 2647 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2648 m->m_len--; 2649 if (m->m_flags & M_PKTHDR) 2650 m->m_pkthdr.len--; 2651 return; 2652 } 2653 cnt -= m->m_len; 2654 m = m->m_next; 2655 if (m == 0) 2656 break; 2657 } 2658 panic("tcp_pulloutofband"); 2659} 2660 2661/* 2662 * Collect new round-trip time estimate 2663 * and update averages and current timeout. 2664 */ 2665static void 2666tcp_xmit_timer(tp, rtt) 2667 register struct tcpcb *tp; 2668 int rtt; 2669{ 2670 register int delta; 2671 2672 tcpstat.tcps_rttupdated++; 2673 tp->t_rttupdated++; 2674 if (tp->t_srtt != 0) { 2675 /* 2676 * srtt is stored as fixed point with 5 bits after the 2677 * binary point (i.e., scaled by 8). The following magic 2678 * is equivalent to the smoothing algorithm in rfc793 with 2679 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2680 * point). Adjust rtt to origin 0. 2681 */ 2682 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2683 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2684 2685 if ((tp->t_srtt += delta) <= 0) 2686 tp->t_srtt = 1; 2687 2688 /* 2689 * We accumulate a smoothed rtt variance (actually, a 2690 * smoothed mean difference), then set the retransmit 2691 * timer to smoothed rtt + 4 times the smoothed variance. 2692 * rttvar is stored as fixed point with 4 bits after the 2693 * binary point (scaled by 16). The following is 2694 * equivalent to rfc793 smoothing with an alpha of .75 2695 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2696 * rfc793's wired-in beta. 2697 */ 2698 if (delta < 0) 2699 delta = -delta; 2700 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2701 if ((tp->t_rttvar += delta) <= 0) 2702 tp->t_rttvar = 1; 2703 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2704 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2705 } else { 2706 /* 2707 * No rtt measurement yet - use the unsmoothed rtt. 2708 * Set the variance to half the rtt (so our first 2709 * retransmit happens at 3*rtt). 2710 */ 2711 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2712 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2713 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2714 } 2715 tp->t_rtttime = 0; 2716 tp->t_rxtshift = 0; 2717 2718 /* 2719 * the retransmit should happen at rtt + 4 * rttvar. 2720 * Because of the way we do the smoothing, srtt and rttvar 2721 * will each average +1/2 tick of bias. When we compute 2722 * the retransmit timer, we want 1/2 tick of rounding and 2723 * 1 extra tick because of +-1/2 tick uncertainty in the 2724 * firing of the timer. The bias will give us exactly the 2725 * 1.5 tick we need. But, because the bias is 2726 * statistical, we have to test that we don't drop below 2727 * the minimum feasible timer (which is 2 ticks). 2728 */ 2729 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2730 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2731 2732 /* 2733 * We received an ack for a packet that wasn't retransmitted; 2734 * it is probably safe to discard any error indications we've 2735 * received recently. This isn't quite right, but close enough 2736 * for now (a route might have failed after we sent a segment, 2737 * and the return path might not be symmetrical). 2738 */ 2739 tp->t_softerror = 0; 2740} 2741 2742/* 2743 * Determine a reasonable value for maxseg size. 2744 * If the route is known, check route for mtu. 2745 * If none, use an mss that can be handled on the outgoing 2746 * interface without forcing IP to fragment; if bigger than 2747 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2748 * to utilize large mbufs. If no route is found, route has no mtu, 2749 * or the destination isn't local, use a default, hopefully conservative 2750 * size (usually 512 or the default IP max size, but no more than the mtu 2751 * of the interface), as we can't discover anything about intervening 2752 * gateways or networks. We also initialize the congestion/slow start 2753 * window to be a single segment if the destination isn't local. 2754 * While looking at the routing entry, we also initialize other path-dependent 2755 * parameters from pre-set or cached values in the routing entry. 2756 * 2757 * Also take into account the space needed for options that we 2758 * send regularly. Make maxseg shorter by that amount to assure 2759 * that we can send maxseg amount of data even when the options 2760 * are present. Store the upper limit of the length of options plus 2761 * data in maxopd. 2762 * 2763 * 2764 * In case of T/TCP, we call this routine during implicit connection 2765 * setup as well (offer = -1), to initialize maxseg from the cached 2766 * MSS of our peer. 2767 * 2768 * NOTE that this routine is only called when we process an incoming 2769 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2770 */ 2771void 2772tcp_mss(tp, offer) 2773 struct tcpcb *tp; 2774 int offer; 2775{ 2776 int rtt, mss; 2777 u_long bufsize; 2778 u_long maxmtu; 2779 struct inpcb *inp = tp->t_inpcb; 2780 struct socket *so; 2781 struct hc_metrics_lite metrics; 2782 struct rmxp_tao tao; 2783 int origoffer = offer; 2784#ifdef INET6 2785 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2786 size_t min_protoh = isipv6 ? 2787 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2788 sizeof (struct tcpiphdr); 2789#else 2790 const size_t min_protoh = sizeof(struct tcpiphdr); 2791#endif 2792 bzero(&tao, sizeof(tao)); 2793 2794 /* initialize */ 2795#ifdef INET6 2796 if (isipv6) { 2797 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2798 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2799 } else 2800#endif 2801 { 2802 maxmtu = tcp_maxmtu(&inp->inp_inc); 2803 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2804 } 2805 so = inp->inp_socket; 2806 2807 /* 2808 * no route to sender, stay with default mss and return 2809 */ 2810 if (maxmtu == 0) 2811 return; 2812 2813 /* what have we got? */ 2814 switch (offer) { 2815 case 0: 2816 /* 2817 * Offer == 0 means that there was no MSS on the SYN 2818 * segment, in this case we use tcp_mssdflt. 2819 */ 2820 offer = 2821#ifdef INET6 2822 isipv6 ? tcp_v6mssdflt : 2823#endif 2824 tcp_mssdflt; 2825 break; 2826 2827 case -1: 2828 /* 2829 * Offer == -1 means that we didn't receive SYN yet, 2830 * use cached value in that case; 2831 */ 2832 if (tcp_do_rfc1644) 2833 tcp_hc_gettao(&inp->inp_inc, &tao); 2834 if (tao.tao_mssopt != 0) 2835 offer = tao.tao_mssopt; 2836 /* FALLTHROUGH */ 2837 2838 default: 2839 /* 2840 * Prevent DoS attack with too small MSS. Round up 2841 * to at least minmss. 2842 */ 2843 offer = max(offer, tcp_minmss); 2844 /* 2845 * Sanity check: make sure that maxopd will be large 2846 * enough to allow some data on segments even if the 2847 * all the option space is used (40bytes). Otherwise 2848 * funny things may happen in tcp_output. 2849 */ 2850 offer = max(offer, 64); 2851 if (tcp_do_rfc1644) 2852 tcp_hc_updatetao(&inp->inp_inc, 2853 TCP_HC_TAO_MSSOPT, 0, offer); 2854 } 2855 2856 /* 2857 * rmx information is now retrieved from tcp_hostcache 2858 */ 2859 tcp_hc_get(&inp->inp_inc, &metrics); 2860 2861 /* 2862 * if there's a discovered mtu int tcp hostcache, use it 2863 * else, use the link mtu. 2864 */ 2865 if (metrics.rmx_mtu) 2866 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2867 else { 2868#ifdef INET6 2869 if (isipv6) { 2870 mss = maxmtu - min_protoh; 2871 if (!path_mtu_discovery && 2872 !in6_localaddr(&inp->in6p_faddr)) 2873 mss = min(mss, tcp_v6mssdflt); 2874 } else 2875#endif 2876 { 2877 mss = maxmtu - min_protoh; 2878 if (!path_mtu_discovery && 2879 !in_localaddr(inp->inp_faddr)) 2880 mss = min(mss, tcp_mssdflt); 2881 } 2882 } 2883 mss = min(mss, offer); 2884 2885 /* 2886 * maxopd stores the maximum length of data AND options 2887 * in a segment; maxseg is the amount of data in a normal 2888 * segment. We need to store this value (maxopd) apart 2889 * from maxseg, because now every segment carries options 2890 * and thus we normally have somewhat less data in segments. 2891 */ 2892 tp->t_maxopd = mss; 2893 2894 /* 2895 * In case of T/TCP, origoffer==-1 indicates, that no segments 2896 * were received yet. In this case we just guess, otherwise 2897 * we do the same as before T/TCP. 2898 */ 2899 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2900 (origoffer == -1 || 2901 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2902 mss -= TCPOLEN_TSTAMP_APPA; 2903 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2904 (origoffer == -1 || 2905 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2906 mss -= TCPOLEN_CC_APPA; 2907 tp->t_maxseg = mss; 2908 2909#if (MCLBYTES & (MCLBYTES - 1)) == 0 2910 if (mss > MCLBYTES) 2911 mss &= ~(MCLBYTES-1); 2912#else 2913 if (mss > MCLBYTES) 2914 mss = mss / MCLBYTES * MCLBYTES; 2915#endif 2916 tp->t_maxseg = mss; 2917 2918 /* 2919 * If there's a pipesize, change the socket buffer to that size, 2920 * don't change if sb_hiwat is different than default (then it 2921 * has been changed on purpose with setsockopt). 2922 * Make the socket buffers an integral number of mss units; 2923 * if the mss is larger than the socket buffer, decrease the mss. 2924 */ 2925 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2926 bufsize = metrics.rmx_sendpipe; 2927 else 2928 bufsize = so->so_snd.sb_hiwat; 2929 if (bufsize < mss) 2930 mss = bufsize; 2931 else { 2932 bufsize = roundup(bufsize, mss); 2933 if (bufsize > sb_max) 2934 bufsize = sb_max; 2935 if (bufsize > so->so_snd.sb_hiwat) 2936 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2937 } 2938 tp->t_maxseg = mss; 2939 2940 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2941 bufsize = metrics.rmx_recvpipe; 2942 else 2943 bufsize = so->so_rcv.sb_hiwat; 2944 if (bufsize > mss) { 2945 bufsize = roundup(bufsize, mss); 2946 if (bufsize > sb_max) 2947 bufsize = sb_max; 2948 if (bufsize > so->so_rcv.sb_hiwat) 2949 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2950 } 2951 /* 2952 * While we're here, check the others too 2953 */ 2954 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2955 tp->t_srtt = rtt; 2956 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2957 tcpstat.tcps_usedrtt++; 2958 if (metrics.rmx_rttvar) { 2959 tp->t_rttvar = metrics.rmx_rttvar; 2960 tcpstat.tcps_usedrttvar++; 2961 } else { 2962 /* default variation is +- 1 rtt */ 2963 tp->t_rttvar = 2964 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2965 } 2966 TCPT_RANGESET(tp->t_rxtcur, 2967 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2968 tp->t_rttmin, TCPTV_REXMTMAX); 2969 } 2970 if (metrics.rmx_ssthresh) { 2971 /* 2972 * There's some sort of gateway or interface 2973 * buffer limit on the path. Use this to set 2974 * the slow start threshhold, but set the 2975 * threshold to no less than 2*mss. 2976 */ 2977 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2978 tcpstat.tcps_usedssthresh++; 2979 } 2980 if (metrics.rmx_bandwidth) 2981 tp->snd_bandwidth = metrics.rmx_bandwidth; 2982 2983 /* 2984 * Set the slow-start flight size depending on whether this 2985 * is a local network or not. 2986 * 2987 * Extend this so we cache the cwnd too and retrieve it here. 2988 * Make cwnd even bigger than RFC3390 suggests but only if we 2989 * have previous experience with the remote host. Be careful 2990 * not make cwnd bigger than remote receive window or our own 2991 * send socket buffer. Maybe put some additional upper bound 2992 * on the retrieved cwnd. Should do incremental updates to 2993 * hostcache when cwnd collapses so next connection doesn't 2994 * overloads the path again. 2995 * 2996 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2997 * We currently check only in syncache_socket for that. 2998 */ 2999#define TCP_METRICS_CWND 3000#ifdef TCP_METRICS_CWND 3001 if (metrics.rmx_cwnd) 3002 tp->snd_cwnd = max(mss, 3003 min(metrics.rmx_cwnd / 2, 3004 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3005 else 3006#endif 3007 if (tcp_do_rfc3390) 3008 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3009#ifdef INET6 3010 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3011 (!isipv6 && in_localaddr(inp->inp_faddr))) 3012#else 3013 else if (in_localaddr(inp->inp_faddr)) 3014#endif 3015 tp->snd_cwnd = mss * ss_fltsz_local; 3016 else 3017 tp->snd_cwnd = mss * ss_fltsz; 3018} 3019 3020/* 3021 * Determine the MSS option to send on an outgoing SYN. 3022 */ 3023int 3024tcp_mssopt(inc) 3025 struct in_conninfo *inc; 3026{ 3027 int mss = 0; 3028 u_long maxmtu = 0; 3029 u_long thcmtu = 0; 3030 size_t min_protoh; 3031#ifdef INET6 3032 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3033#endif 3034 3035 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3036 3037#ifdef INET6 3038 if (isipv6) { 3039 mss = tcp_v6mssdflt; 3040 maxmtu = tcp_maxmtu6(inc); 3041 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3042 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3043 } else 3044#endif 3045 { 3046 mss = tcp_mssdflt; 3047 maxmtu = tcp_maxmtu(inc); 3048 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3049 min_protoh = sizeof(struct tcpiphdr); 3050 } 3051 if (maxmtu && thcmtu) 3052 mss = min(maxmtu, thcmtu) - min_protoh; 3053 else if (maxmtu || thcmtu) 3054 mss = max(maxmtu, thcmtu) - min_protoh; 3055 3056 return (mss); 3057} 3058 3059 3060/* 3061 * On a partial ack arrives, force the retransmission of the 3062 * next unacknowledged segment. Do not clear tp->t_dupacks. 3063 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3064 * be started again. 3065 */ 3066static void 3067tcp_newreno_partial_ack(tp, th) 3068 struct tcpcb *tp; 3069 struct tcphdr *th; 3070{ 3071 tcp_seq onxt = tp->snd_nxt; 3072 u_long ocwnd = tp->snd_cwnd; 3073 3074 callout_stop(tp->tt_rexmt); 3075 tp->t_rtttime = 0; 3076 tp->snd_nxt = th->th_ack; 3077 /* 3078 * Set snd_cwnd to one segment beyond acknowledged offset. 3079 * (tp->snd_una has not yet been updated when this function is called.) 3080 */ 3081 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3082 tp->t_flags |= TF_ACKNOW; 3083 (void) tcp_output(tp); 3084 tp->snd_cwnd = ocwnd; 3085 if (SEQ_GT(onxt, tp->snd_nxt)) 3086 tp->snd_nxt = onxt; 3087 /* 3088 * Partial window deflation. Relies on fact that tp->snd_una 3089 * not updated yet. 3090 */ 3091 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3092} 3093 3094/* 3095 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3096 * looking for a pcb in the listen state. Returns 0 otherwise. 3097 */ 3098static int 3099tcp_timewait(tw, to, th, m, tlen) 3100 struct tcptw *tw; 3101 struct tcpopt *to; 3102 struct tcphdr *th; 3103 struct mbuf *m; 3104 int tlen; 3105{ 3106 int thflags; 3107 tcp_seq seq; 3108#ifdef INET6 3109 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3110#else 3111 const int isipv6 = 0; 3112#endif 3113 3114 thflags = th->th_flags; 3115 3116 /* 3117 * NOTE: for FIN_WAIT_2 (to be added later), 3118 * must validate sequence number before accepting RST 3119 */ 3120 3121 /* 3122 * If the segment contains RST: 3123 * Drop the segment - see Stevens, vol. 2, p. 964 and 3124 * RFC 1337. 3125 */ 3126 if (thflags & TH_RST) 3127 goto drop; 3128 3129 /* 3130 * If segment contains a SYN and CC [not CC.NEW] option: 3131 * if connection duration > MSL, drop packet and send RST; 3132 * 3133 * if SEG.CC > CCrecv then is new SYN. 3134 * Complete close and delete TCPCB. Then reprocess 3135 * segment, hoping to find new TCPCB in LISTEN state; 3136 * 3137 * else must be old SYN; drop it. 3138 * else do normal processing. 3139 */ 3140 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3141 if ((ticks - tw->t_starttime) > tcp_msl) 3142 goto reset; 3143 if (CC_GT(to->to_cc, tw->cc_recv)) { 3144 (void) tcp_twclose(tw, 0); 3145 return (1); 3146 } 3147 goto drop; 3148 } 3149 3150#if 0 3151/* PAWS not needed at the moment */ 3152 /* 3153 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3154 * and it's less than ts_recent, drop it. 3155 */ 3156 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3157 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3158 if ((thflags & TH_ACK) == 0) 3159 goto drop; 3160 goto ack; 3161 } 3162 /* 3163 * ts_recent is never updated because we never accept new segments. 3164 */ 3165#endif 3166 3167 /* 3168 * If a new connection request is received 3169 * while in TIME_WAIT, drop the old connection 3170 * and start over if the sequence numbers 3171 * are above the previous ones. 3172 */ 3173 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3174 (void) tcp_twclose(tw, 0); 3175 return (1); 3176 } 3177 3178 /* 3179 * Drop the the segment if it does not contain an ACK. 3180 */ 3181 if ((thflags & TH_ACK) == 0) 3182 goto drop; 3183 3184 /* 3185 * Reset the 2MSL timer if this is a duplicate FIN. 3186 */ 3187 if (thflags & TH_FIN) { 3188 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3189 if (seq + 1 == tw->rcv_nxt) 3190 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3191 } 3192 3193 /* 3194 * Acknowledge the segment if it has data or is not a duplicate ACK. 3195 */ 3196 if (thflags != TH_ACK || tlen != 0 || 3197 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3198 tcp_twrespond(tw, TH_ACK); 3199 goto drop; 3200 3201reset: 3202 /* 3203 * Generate a RST, dropping incoming segment. 3204 * Make ACK acceptable to originator of segment. 3205 * Don't bother to respond if destination was broadcast/multicast. 3206 */ 3207 if (m->m_flags & (M_BCAST|M_MCAST)) 3208 goto drop; 3209 if (isipv6) { 3210 struct ip6_hdr *ip6; 3211 3212 /* IPv6 anycast check is done at tcp6_input() */ 3213 ip6 = mtod(m, struct ip6_hdr *); 3214 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3215 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3216 goto drop; 3217 } else { 3218 struct ip *ip; 3219 3220 ip = mtod(m, struct ip *); 3221 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3222 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3223 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3224 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3225 goto drop; 3226 } 3227 if (thflags & TH_ACK) { 3228 tcp_respond(NULL, 3229 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3230 } else { 3231 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3232 tcp_respond(NULL, 3233 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3234 } 3235 INP_UNLOCK(tw->tw_inpcb); 3236 return (0); 3237 3238drop: 3239 INP_UNLOCK(tw->tw_inpcb); 3240 m_freem(m); 3241 return (0); 3242}
|