]> git.saurik.com Git - apple/xnu.git/blob - bsd/netinet/tcp_input.c
projects / apple / xnu.git / blob
? search:


b931b642df36ea5481eab35827bd71d00d9ffd2f
[apple/xnu.git] / bsd / netinet / tcp_input.c
1 /*
2 * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
3 *
4 * @APPLE_LICENSE_HEADER_START@
5 *
6 * The contents of this file constitute Original Code as defined in and
7 * are subject to the Apple Public Source License Version 1.1 (the
8 * "License"). You may not use this file except in compliance with the
9 * License. Please obtain a copy of the License at
10 * http://www.apple.com/publicsource and read it before using this file.
11 *
12 * This Original Code and all software distributed under the License are
13 * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
14 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
15 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the
17 * License for the specific language governing rights and limitations
18 * under the License.
19 *
20 * @APPLE_LICENSE_HEADER_END@
21 */
22 /*
23 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
24 * The Regents of the University of California. All rights reserved.
25 *
26 * Redistribution and use in source and binary forms, with or without
27 * modification, are permitted provided that the following conditions
28 * are met:
29 * 1. Redistributions of source code must retain the above copyright
30 * notice, this list of conditions and the following disclaimer.
31 * 2. Redistributions in binary form must reproduce the above copyright
32 * notice, this list of conditions and the following disclaimer in the
33 * documentation and/or other materials provided with the distribution.
34 * 3. All advertising materials mentioning features or use of this software
35 * must display the following acknowledgement:
36 * This product includes software developed by the University of
37 * California, Berkeley and its contributors.
38 * 4. Neither the name of the University nor the names of its contributors
39 * may be used to endorse or promote products derived from this software
40 * without specific prior written permission.
41 *
42 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
45 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
52 * SUCH DAMAGE.
53 *
54 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
55 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.16 2001/08/22 00:59:12 silby Exp $
56 */
57
58
59 #include <sys/param.h>
60 #include <sys/systm.h>
61 #include <sys/kernel.h>
62 #include <sys/sysctl.h>
63 #include <sys/malloc.h>
64 #include <sys/mbuf.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/syslog.h>
70
71 #include <kern/cpu_number.h> /* before tcp_seq.h, for tcp_random18() */
72
73 #include <net/if.h>
74 #include <net/if_types.h>
75 #include <net/route.h>
76
77 #include <netinet/in.h>
78 #include <netinet/in_systm.h>
79 #include <netinet/ip.h>
80 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
81 #include <netinet/in_var.h>
82 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
83 #include <netinet/in_pcb.h>
84 #include <netinet/ip_var.h>
85 #if INET6
86 #include <netinet/ip6.h>
87 #include <netinet/icmp6.h>
88 #include <netinet6/nd6.h>
89 #include <netinet6/ip6_var.h>
90 #include <netinet6/in6_pcb.h>
91 #endif
92 #include <netinet/tcp.h>
93 #include <netinet/tcp_fsm.h>
94 #include <netinet/tcp_seq.h>
95 #include <netinet/tcp_timer.h>
96 #include <netinet/tcp_var.h>
97 #if INET6
98 #include <netinet6/tcp6_var.h>
99 #endif
100 #include <netinet/tcpip.h>
101 #if TCPDEBUG
102 #include <netinet/tcp_debug.h>
103 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
104 struct tcphdr tcp_savetcp;
105 #endif /* TCPDEBUG */
106
107 #if IPSEC
108 #include <netinet6/ipsec.h>
109 #if INET6
110 #include <netinet6/ipsec6.h>
111 #endif
112 #include <netkey/key.h>
113 #endif /*IPSEC*/
114
115 #include <sys/kdebug.h>
116
117 #ifndef __APPLE__
118 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
119 #endif
120
121 #define DBG_LAYER_BEG NETDBG_CODE(DBG_NETTCP, 0)
122 #define DBG_LAYER_END NETDBG_CODE(DBG_NETTCP, 2)
123 #define DBG_FNC_TCP_INPUT NETDBG_CODE(DBG_NETTCP, (3 << 8))
124 #define DBG_FNC_TCP_NEWCONN NETDBG_CODE(DBG_NETTCP, (7 << 8))
125
126 static int tcprexmtthresh = 3;
127 tcp_cc tcp_ccgen;
128 extern int apple_hwcksum_rx;
129
130 #if IPSEC
131 extern int ipsec_bypass;
132 extern lck_mtx_t *sadb_mutex;
133 #endif
134
135 struct tcpstat tcpstat;
136 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD,
137 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
138
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141 &log_in_vain, 0, "Log all incoming TCP connections");
142
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145 &blackhole, 0, "Do not send RST when dropping refused connections");
146
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149 &tcp_delack_enabled, 0,
150 "Delay ACK to try and piggyback it onto a data packet");
151
152 int tcp_lq_overflow = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW,
154 &tcp_lq_overflow, 0,
155 "Listen Queue Overflow");
156
157 #if TCP_DROP_SYNFIN
158 static int drop_synfin = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
160 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
161 #endif
162
163 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
164 "TCP Segment Reassembly Queue");
165
166 __private_extern__ int tcp_reass_maxseg = 0;
167 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RW,
168 &tcp_reass_maxseg, 0,
169 "Global maximum number of TCP Segments in Reassembly Queue");
170
171 __private_extern__ int tcp_reass_qsize = 0;
172 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
173 &tcp_reass_qsize, 0,
174 "Global number of TCP Segments currently in Reassembly Queue");
175
176 static int tcp_reass_overflows = 0;
177 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
178 &tcp_reass_overflows, 0,
179 "Global number of TCP Segment Reassembly Queue Overflows");
180
181
182 __private_extern__ int slowlink_wsize = 8192;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, slowlink_wsize, CTLFLAG_RW,
184 &slowlink_wsize, 0, "Maximum advertised window size for slowlink");
185
186
187 u_long tcp_now;
188 struct inpcbhead tcb;
189 #define tcb6 tcb /* for KAME src sync over BSD*'s */
190 struct inpcbinfo tcbinfo;
191
192 static void tcp_dooptions(struct tcpcb *,
193 u_char *, int, struct tcphdr *, struct tcpopt *);
194 static void tcp_pulloutofband(struct socket *,
195 struct tcphdr *, struct mbuf *, int);
196 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
197 struct mbuf *);
198 static void tcp_xmit_timer(struct tcpcb *, int);
199 static int tcp_newreno __P((struct tcpcb *, struct tcphdr *));
200
201 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
202 #if INET6
203 #define ND6_HINT(tp) \
204 do { \
205 if ((tp) && (tp)->t_inpcb && \
206 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \
207 (tp)->t_inpcb->in6p_route.ro_rt) \
208 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
209 } while (0)
210 #else
211 #define ND6_HINT(tp)
212 #endif
213
214 extern u_long *delack_bitmask;
215
216 extern void ipfwsyslog( int level, char *format,...);
217 extern int ChkAddressOK( __uint32_t dstaddr, __uint32_t srcaddr );
218 extern int fw_verbose;
219
220 #define log_in_vain_log( a ) { \
221 if ( (log_in_vain == 3 ) && (fw_verbose == 2)) { /* Apple logging, log to ipfw.log */ \
222 ipfwsyslog a ; \
223 } \
224 else log a ; \
225 }
226
227 /*
228 * Indicate whether this ack should be delayed.
229 * We can delay the ack if:
230 * - delayed acks are enabled (set to 1) and
231 * - our last ack wasn't a 0-sized window. We never want to delay
232 * the ack that opens up a 0-sized window.
233 * - delayed acks are enabled (set to 2, "more compatible") and
234 * - our last ack wasn't a 0-sized window.
235 * - if the peer hasn't sent us a TH_PUSH data packet (this solves 3649245)
236 * - the peer hasn't sent us a TH_PUSH data packet, if he did, take this as a clue that we
237 * need to ACK with no delay. This helps higher level protocols who won't send
238 * us more data even if the window is open because their last "segment" hasn't been ACKed
239 *
240 *
241 */
242 #define DELAY_ACK(tp) \
243 (((tcp_delack_enabled == 1) && ((tp->t_flags & TF_RXWIN0SENT) == 0)) || \
244 (((tcp_delack_enabled == 2) && (tp->t_flags & TF_RXWIN0SENT) == 0) && \
245 ((thflags & TH_PUSH) == 0) && ((tp->t_flags & TF_DELACK) == 0)))
246
247
248 static int tcpdropdropablreq(struct socket *head);
249
250
251 static int
252 tcp_reass(tp, th, tlenp, m)
253 register struct tcpcb *tp;
254 register struct tcphdr *th;
255 int *tlenp;
256 struct mbuf *m;
257 {
258 struct tseg_qent *q;
259 struct tseg_qent *p = NULL;
260 struct tseg_qent *nq;
261 struct tseg_qent *te;
262 struct socket *so = tp->t_inpcb->inp_socket;
263 int flags;
264 int dowakeup = 0;
265
266 /*
267 * Call with th==0 after become established to
268 * force pre-ESTABLISHED data up to user socket.
269 */
270 if (th == 0)
271 goto present;
272
273 /*
274 * Limit the number of segments in the reassembly queue to prevent
275 * holding on to too many segments (and thus running out of mbufs).
276 * Make sure to let the missing segment through which caused this
277 * queue. Always keep one global queue entry spare to be able to
278 * process the missing segment.
279 */
280 if (th->th_seq != tp->rcv_nxt &&
281 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
282 tcp_reass_overflows++;
283 tcpstat.tcps_rcvmemdrop++;
284 m_freem(m);
285 return (0);
286 }
287
288 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */
289 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ,
290 M_NOWAIT);
291 if (te == NULL) {
292 tcpstat.tcps_rcvmemdrop++;
293 m_freem(m);
294 return (0);
295 }
296 tcp_reass_qsize++;
297
298 /*
299 * Find a segment which begins after this one does.
300 */
301 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
302 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
303 break;
304 p = q;
305 }
306
307 /*
308 * If there is a preceding segment, it may provide some of
309 * our data already. If so, drop the data from the incoming
310 * segment. If it provides all of our data, drop us.
311 */
312 if (p != NULL) {
313 register int i;
314 /* conversion to int (in i) handles seq wraparound */
315 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
316 if (i > 0) {
317 if (i >= *tlenp) {
318 tcpstat.tcps_rcvduppack++;
319 tcpstat.tcps_rcvdupbyte += *tlenp;
320 m_freem(m);
321 FREE(te, M_TSEGQ);
322 tcp_reass_qsize--;
323 /*
324 * Try to present any queued data
325 * at the left window edge to the user.
326 * This is needed after the 3-WHS
327 * completes.
328 */
329 goto present; /* ??? */
330 }
331 m_adj(m, i);
332 *tlenp -= i;
333 th->th_seq += i;
334 }
335 }
336 tcpstat.tcps_rcvoopack++;
337 tcpstat.tcps_rcvoobyte += *tlenp;
338
339 /*
340 * While we overlap succeeding segments trim them or,
341 * if they are completely covered, dequeue them.
342 */
343 while (q) {
344 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
345 if (i <= 0)
346 break;
347 if (i < q->tqe_len) {
348 q->tqe_th->th_seq += i;
349 q->tqe_len -= i;
350 m_adj(q->tqe_m, i);
351 break;
352 }
353
354 nq = LIST_NEXT(q, tqe_q);
355 LIST_REMOVE(q, tqe_q);
356 m_freem(q->tqe_m);
357 FREE(q, M_TSEGQ);
358 tcp_reass_qsize--;
359 q = nq;
360 }
361
362 /* Insert the new segment queue entry into place. */
363 te->tqe_m = m;
364 te->tqe_th = th;
365 te->tqe_len = *tlenp;
366
367 if (p == NULL) {
368 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
369 } else {
370 LIST_INSERT_AFTER(p, te, tqe_q);
371 }
372
373 present:
374 /*
375 * Present data to user, advancing rcv_nxt through
376 * completed sequence space.
377 */
378 if (!TCPS_HAVEESTABLISHED(tp->t_state))
379 return (0);
380 q = LIST_FIRST(&tp->t_segq);
381 if (!q || q->tqe_th->th_seq != tp->rcv_nxt)
382 return (0);
383 do {
384 tp->rcv_nxt += q->tqe_len;
385 flags = q->tqe_th->th_flags & TH_FIN;
386 nq = LIST_NEXT(q, tqe_q);
387 LIST_REMOVE(q, tqe_q);
388 if (so->so_state & SS_CANTRCVMORE)
389 m_freem(q->tqe_m);
390 else {
391 if (sbappend(&so->so_rcv, q->tqe_m))
392 dowakeup = 1;
393 }
394 FREE(q, M_TSEGQ);
395 tcp_reass_qsize--;
396 q = nq;
397 } while (q && q->tqe_th->th_seq == tp->rcv_nxt);
398 ND6_HINT(tp);
399
400 #if INET6
401 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
402
403 KERNEL_DEBUG(DBG_LAYER_BEG,
404 ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
405 (((tp->t_inpcb->in6p_laddr.s6_addr16[0] & 0xffff) << 16) |
406 (tp->t_inpcb->in6p_faddr.s6_addr16[0] & 0xffff)),
407 0,0,0);
408 }
409 else
410 #endif
411 {
412 KERNEL_DEBUG(DBG_LAYER_BEG,
413 ((tp->t_inpcb->inp_fport << 16) | tp->t_inpcb->inp_lport),
414 (((tp->t_inpcb->inp_laddr.s_addr & 0xffff) << 16) |
415 (tp->t_inpcb->inp_faddr.s_addr & 0xffff)),
416 0,0,0);
417 }
418 if (dowakeup)
419 sorwakeup(so); /* done with socket lock held */
420 return (flags);
421
422 }
423
424
425 /*
426 * TCP input routine, follows pages 65-76 of the
427 * protocol specification dated September, 1981 very closely.
428 */
429 #if INET6
430 int
431 tcp6_input(mp, offp)
432 struct mbuf **mp;
433 int *offp;
434 {
435 register struct mbuf *m = *mp;
436 struct in6_ifaddr *ia6;
437
438 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), return IPPROTO_DONE);
439
440 /*
441 * draft-itojun-ipv6-tcp-to-anycast
442 * better place to put this in?
443 */
444 ia6 = ip6_getdstifaddr(m);
445 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
446 struct ip6_hdr *ip6;
447
448 ip6 = mtod(m, struct ip6_hdr *);
449 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
450 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
451 return IPPROTO_DONE;
452 }
453
454 tcp_input(m, *offp);
455 return IPPROTO_DONE;
456 }
457 #endif
458
459 void
460 tcp_input(m, off0)
461 struct mbuf *m;
462 int off0;
463 {
464 register struct tcphdr *th;
465 register struct ip *ip = NULL;
466 register struct ipovly *ipov;
467 register struct inpcb *inp;
468 u_char *optp = NULL;
469 int optlen = 0;
470 int len, tlen, off;
471 int drop_hdrlen;
472 register struct tcpcb *tp = 0;
473 register int thflags;
474 struct socket *so = 0;
475 int todrop, acked, ourfinisacked, needoutput = 0;
476 struct in_addr laddr;
477 #if INET6
478 struct in6_addr laddr6;
479 #endif
480 int dropsocket = 0;
481 int iss = 0;
482 int nosock = 0;
483 u_long tiwin;
484 struct tcpopt to; /* options in this segment */
485 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
486 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
487 struct sockaddr_in *next_hop = NULL;
488 #if TCPDEBUG
489 short ostate = 0;
490 #endif
491 struct m_tag *fwd_tag;
492
493 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */
494 fwd_tag = m_tag_locate(m, KERNEL_MODULE_TAG_ID, KERNEL_TAG_TYPE_IPFORWARD, NULL);
495 if (fwd_tag != NULL) {
496 struct ip_fwd_tag *ipfwd_tag = (struct ip_fwd_tag *)(fwd_tag+1);
497
498 next_hop = ipfwd_tag->next_hop;
499 m_tag_delete(m, fwd_tag);
500 }
501
502 #if INET6
503 struct ip6_hdr *ip6 = NULL;
504 int isipv6;
505 #endif /* INET6 */
506 int rstreason; /* For badport_bandlim accounting purposes */
507 struct proc *proc0=current_proc();
508
509 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_START,0,0,0,0,0);
510
511 #if INET6
512 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
513 #endif
514 bzero((char *)&to, sizeof(to));
515
516 tcpstat.tcps_rcvtotal++;
517
518
519
520 #if INET6
521 if (isipv6) {
522 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
523 ip6 = mtod(m, struct ip6_hdr *);
524 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
525 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
526 tcpstat.tcps_rcvbadsum++;
527 goto dropnosock;
528 }
529 th = (struct tcphdr *)((caddr_t)ip6 + off0);
530
531 KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport),
532 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
533 th->th_seq, th->th_ack, th->th_win);
534 /*
535 * Be proactive about unspecified IPv6 address in source.
536 * As we use all-zero to indicate unbounded/unconnected pcb,
537 * unspecified IPv6 address can be used to confuse us.
538 *
539 * Note that packets with unspecified IPv6 destination is
540 * already dropped in ip6_input.
541 */
542 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
543 /* XXX stat */
544 goto dropnosock;
545 }
546 } else
547 #endif /* INET6 */
548 {
549 /*
550 * Get IP and TCP header together in first mbuf.
551 * Note: IP leaves IP header in first mbuf.
552 */
553 if (off0 > sizeof (struct ip)) {
554 ip_stripoptions(m, (struct mbuf *)0);
555 off0 = sizeof(struct ip);
556 if (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16)
557 m->m_pkthdr.csum_flags = 0; /* invalidate hwcksuming */
558
559 }
560 if (m->m_len < sizeof (struct tcpiphdr)) {
561 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
562 tcpstat.tcps_rcvshort++;
563 return;
564 }
565 }
566 ip = mtod(m, struct ip *);
567 ipov = (struct ipovly *)ip;
568 th = (struct tcphdr *)((caddr_t)ip + off0);
569 tlen = ip->ip_len;
570
571 KERNEL_DEBUG(DBG_LAYER_BEG, ((th->th_dport << 16) | th->th_sport),
572 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
573 th->th_seq, th->th_ack, th->th_win);
574
575 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
576 if (apple_hwcksum_rx && (m->m_pkthdr.csum_flags & CSUM_TCP_SUM16)) {
577 u_short pseudo;
578 char b[9];
579 *(uint32_t*)&b[0] = *(uint32_t*)&ipov->ih_x1[0];
580 *(uint32_t*)&b[4] = *(uint32_t*)&ipov->ih_x1[4];
581 *(uint8_t*)&b[8] = *(uint8_t*)&ipov->ih_x1[8];
582
583 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
584 ipov->ih_len = (u_short)tlen;
585 HTONS(ipov->ih_len);
586 pseudo = in_cksum(m, sizeof (struct ip));
587
588 *(uint32_t*)&ipov->ih_x1[0] = *(uint32_t*)&b[0];
589 *(uint32_t*)&ipov->ih_x1[4] = *(uint32_t*)&b[4];
590 *(uint8_t*)&ipov->ih_x1[8] = *(uint8_t*)&b[8];
591
592 th->th_sum = in_addword(pseudo, (m->m_pkthdr.csum_data & 0xFFFF));
593 } else {
594 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
595 th->th_sum = m->m_pkthdr.csum_data;
596 else
597 th->th_sum = in_pseudo(ip->ip_src.s_addr,
598 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data +
599 ip->ip_len + IPPROTO_TCP));
600 }
601 th->th_sum ^= 0xffff;
602 } else {
603 char b[9];
604 /*
605 * Checksum extended TCP header and data.
606 */
607 *(uint32_t*)&b[0] = *(uint32_t*)&ipov->ih_x1[0];
608 *(uint32_t*)&b[4] = *(uint32_t*)&ipov->ih_x1[4];
609 *(uint8_t*)&b[8] = *(uint8_t*)&ipov->ih_x1[8];
610
611 len = sizeof (struct ip) + tlen;
612 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
613 ipov->ih_len = (u_short)tlen;
614 HTONS(ipov->ih_len);
615 th->th_sum = in_cksum(m, len);
616
617 *(uint32_t*)&ipov->ih_x1[0] = *(uint32_t*)&b[0];
618 *(uint32_t*)&ipov->ih_x1[4] = *(uint32_t*)&b[4];
619 *(uint8_t*)&ipov->ih_x1[8] = *(uint8_t*)&b[8];
620 }
621 if (th->th_sum) {
622 tcpstat.tcps_rcvbadsum++;
623 goto dropnosock;
624 }
625 #if INET6
626 /* Re-initialization for later version check */
627 ip->ip_v = IPVERSION;
628 #endif
629 }
630
631 /*
632 * Check that TCP offset makes sense,
633 * pull out TCP options and adjust length. XXX
634 */
635 off = th->th_off << 2;
636 if (off < sizeof (struct tcphdr) || off > tlen) {
637 tcpstat.tcps_rcvbadoff++;
638 goto dropnosock;
639 }
640 tlen -= off; /* tlen is used instead of ti->ti_len */
641 if (off > sizeof (struct tcphdr)) {
642 #if INET6
643 if (isipv6) {
644 IP6_EXTHDR_CHECK(m, off0, off, return);
645 ip6 = mtod(m, struct ip6_hdr *);
646 th = (struct tcphdr *)((caddr_t)ip6 + off0);
647 } else
648 #endif /* INET6 */
649 {
650 if (m->m_len < sizeof(struct ip) + off) {
651 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
652 tcpstat.tcps_rcvshort++;
653 return;
654 }
655 ip = mtod(m, struct ip *);
656 ipov = (struct ipovly *)ip;
657 th = (struct tcphdr *)((caddr_t)ip + off0);
658 }
659 }
660 optlen = off - sizeof (struct tcphdr);
661 optp = (u_char *)(th + 1);
662 /*
663 * Do quick retrieval of timestamp options ("options
664 * prediction?"). If timestamp is the only option and it's
665 * formatted as recommended in RFC 1323 appendix A, we
666 * quickly get the values now and not bother calling
667 * tcp_dooptions(), etc.
668 */
669 if ((optlen == TCPOLEN_TSTAMP_APPA ||
670 (optlen > TCPOLEN_TSTAMP_APPA &&
671 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) &&
672 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) &&
673 (th->th_flags & TH_SYN) == 0) {
674 to.to_flag |= TOF_TS;
675 to.to_tsval = ntohl(*(u_int32_t *)(optp + 4));
676 to.to_tsecr = ntohl(*(u_int32_t *)(optp + 8));
677 optp = NULL; /* we've parsed the options */
678 }
679 }
680 thflags = th->th_flags;
681
682 #if TCP_DROP_SYNFIN
683 /*
684 * If the drop_synfin option is enabled, drop all packets with
685 * both the SYN and FIN bits set. This prevents e.g. nmap from
686 * identifying the TCP/IP stack.
687 *
688 * This is incompatible with RFC1644 extensions (T/TCP).
689 */
690 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
691 goto dropnosock;
692 #endif
693
694 /*
695 * Convert TCP protocol specific fields to host format.
696 */
697 NTOHL(th->th_seq);
698 NTOHL(th->th_ack);
699 NTOHS(th->th_win);
700 NTOHS(th->th_urp);
701
702 /*
703 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options,
704 * until after ip6_savecontrol() is called and before other functions
705 * which don't want those proto headers.
706 * Because ip6_savecontrol() is going to parse the mbuf to
707 * search for data to be passed up to user-land, it wants mbuf
708 * parameters to be unchanged.
709 */
710 drop_hdrlen = off0 + off;
711
712 /*
713 * Locate pcb for segment.
714 */
715 findpcb:
716 #if IPFIREWALL_FORWARD
717 if (next_hop != NULL
718 #if INET6
719 && isipv6 == NULL /* IPv6 support is not yet */
720 #endif /* INET6 */
721 ) {
722 /*
723 * Diverted. Pretend to be the destination.
724 * already got one like this?
725 */
726 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
727 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif);
728 if (!inp) {
729 /*
730 * No, then it's new. Try find the ambushing socket
731 */
732 if (!next_hop->sin_port) {
733 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src,
734 th->th_sport, next_hop->sin_addr,
735 th->th_dport, 1, m->m_pkthdr.rcvif);
736 } else {
737 inp = in_pcblookup_hash(&tcbinfo,
738 ip->ip_src, th->th_sport,
739 next_hop->sin_addr,
740 ntohs(next_hop->sin_port), 1,
741 m->m_pkthdr.rcvif);
742 }
743 }
744 } else
745 #endif /* IPFIREWALL_FORWARD */
746 {
747 #if INET6
748 if (isipv6)
749 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport,
750 &ip6->ip6_dst, th->th_dport, 1,
751 m->m_pkthdr.rcvif);
752 else
753 #endif /* INET6 */
754 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport,
755 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif);
756 }
757
758 #if IPSEC
759 if (ipsec_bypass == 0) {
760 lck_mtx_lock(sadb_mutex);
761 #if INET6
762 if (isipv6) {
763 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) {
764 ipsec6stat.in_polvio++;
765 lck_mtx_unlock(sadb_mutex);
766 goto dropnosock;
767 }
768 } else
769 #endif /* INET6 */
770 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) {
771 ipsecstat.in_polvio++;
772 lck_mtx_unlock(sadb_mutex);
773 goto dropnosock;
774 }
775 lck_mtx_unlock(sadb_mutex);
776 }
777 #endif /*IPSEC*/
778
779 /*
780 * If the state is CLOSED (i.e., TCB does not exist) then
781 * all data in the incoming segment is discarded.
782 * If the TCB exists but is in CLOSED state, it is embryonic,
783 * but should either do a listen or a connect soon.
784 */
785 if (inp == NULL) {
786 if (log_in_vain) {
787 #if INET6
788 char dbuf[MAX_IPv6_STR_LEN], sbuf[MAX_IPv6_STR_LEN];
789 #else /* INET6 */
790 char dbuf[MAX_IPv4_STR_LEN], sbuf[MAX_IPv4_STR_LEN];
791 #endif /* INET6 */
792
793 #if INET6
794 if (isipv6) {
795 inet_ntop(AF_INET6, &ip6->ip6_dst, dbuf, sizeof(dbuf));
796 inet_ntop(AF_INET6, &ip6->ip6_src, sbuf, sizeof(sbuf));
797 } else
798 #endif
799 {
800 inet_ntop(AF_INET, &ip->ip_dst, dbuf, sizeof(dbuf));
801 inet_ntop(AF_INET, &ip->ip_src, sbuf, sizeof(sbuf));
802 }
803 switch (log_in_vain) {
804 case 1:
805 if(thflags & TH_SYN)
806 log(LOG_INFO,
807 "Connection attempt to TCP %s:%d from %s:%d\n",
808 dbuf, ntohs(th->th_dport),
809 sbuf,
810 ntohs(th->th_sport));
811 break;
812 case 2:
813 log(LOG_INFO,
814 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n",
815 dbuf, ntohs(th->th_dport), sbuf,
816 ntohs(th->th_sport), thflags);
817 break;
818 case 3:
819 if ((thflags & TH_SYN) &&
820 !(m->m_flags & (M_BCAST | M_MCAST)) &&
821 #if INET6
822 ((isipv6 && !IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) ||
823 (!isipv6 && ip->ip_dst.s_addr != ip->ip_src.s_addr))
824 #else
825 ip->ip_dst.s_addr != ip->ip_src.s_addr
826 #endif
827 )
828 log_in_vain_log((LOG_INFO,
829 "Stealth Mode connection attempt to TCP %s:%d from %s:%d\n",
830 dbuf, ntohs(th->th_dport),
831 sbuf,
832 ntohs(th->th_sport)));
833 break;
834 default:
835 break;
836 }
837 }
838 if (blackhole) {
839 if (m->m_pkthdr.rcvif && m->m_pkthdr.rcvif->if_type != IFT_LOOP)
840 switch (blackhole) {
841 case 1:
842 if (thflags & TH_SYN)
843 goto dropnosock;
844 break;
845 case 2:
846 goto dropnosock;
847 default:
848 goto dropnosock;
849 }
850 }
851 rstreason = BANDLIM_RST_CLOSEDPORT;
852 goto dropwithresetnosock;
853 }
854 so = inp->inp_socket;
855 if (so == NULL) {
856 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)
857 inp = NULL; // pretend we didn't find it
858 #if TEMPDEBUG
859 printf("tcp_input: no more socket for inp=%x\n", inp);
860 #endif
861 goto dropnosock;
862 }
863 tcp_lock(so, 1, 2);
864 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
865 tcp_unlock(so, 1, 2);
866 inp = NULL; // pretend we didn't find it
867 goto dropnosock;
868 }
869
870 tp = intotcpcb(inp);
871 if (tp == 0) {
872 rstreason = BANDLIM_RST_CLOSEDPORT;
873 goto dropwithreset;
874 }
875 if (tp->t_state == TCPS_CLOSED)
876 goto drop;
877
878 #ifdef __APPLE__
879 /*
880 * Bogus state when listening port owned by SharedIP with loopback as the
881 * only configured interface: BlueBox does not filters loopback
882 */
883 if (tp->t_state == TCP_NSTATES)
884 goto drop;
885 #endif
886
887 /* Unscale the window into a 32-bit value. */
888 if ((thflags & TH_SYN) == 0)
889 tiwin = th->th_win << tp->snd_scale;
890 else
891 tiwin = th->th_win;
892
893 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
894 #if TCPDEBUG
895 if (so->so_options & SO_DEBUG) {
896 ostate = tp->t_state;
897 #if INET6
898 if (isipv6)
899 bcopy((char *)ip6, (char *)tcp_saveipgen,
900 sizeof(*ip6));
901 else
902 #endif /* INET6 */
903 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
904 tcp_savetcp = *th;
905 }
906 #endif
907 if (so->so_options & SO_ACCEPTCONN) {
908 register struct tcpcb *tp0 = tp;
909 struct socket *so2;
910 struct socket *oso;
911 struct sockaddr_storage from;
912 #if INET6
913 struct inpcb *oinp = sotoinpcb(so);
914 #endif /* INET6 */
915 int ogencnt = so->so_gencnt;
916
917 #if !IPSEC
918 /*
919 * Current IPsec implementation makes incorrect IPsec
920 * cache if this check is done here.
921 * So delay this until duplicated socket is created.
922 */
923 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
924 /*
925 * Note: dropwithreset makes sure we don't
926 * send a RST in response to a RST.
927 */
928 if (thflags & TH_ACK) {
929 tcpstat.tcps_badsyn++;
930 rstreason = BANDLIM_RST_OPENPORT;
931 goto dropwithreset;
932 }
933 goto drop;
934 }
935 #endif
936 KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_START,0,0,0,0,0);
937
938 #if INET6
939 /*
940 * If deprecated address is forbidden,
941 * we do not accept SYN to deprecated interface
942 * address to prevent any new inbound connection from
943 * getting established.
944 * When we do not accept SYN, we send a TCP RST,
945 * with deprecated source address (instead of dropping
946 * it). We compromise it as it is much better for peer
947 * to send a RST, and RST will be the final packet
948 * for the exchange.
949 *
950 * If we do not forbid deprecated addresses, we accept
951 * the SYN packet. RFC2462 does not suggest dropping
952 * SYN in this case.
953 * If we decipher RFC2462 5.5.4, it says like this:
954 * 1. use of deprecated addr with existing
955 * communication is okay - "SHOULD continue to be
956 * used"
957 * 2. use of it with new communication:
958 * (2a) "SHOULD NOT be used if alternate address
959 * with sufficient scope is available"
960 * (2b) nothing mentioned otherwise.
961 * Here we fall into (2b) case as we have no choice in
962 * our source address selection - we must obey the peer.
963 *
964 * The wording in RFC2462 is confusing, and there are
965 * multiple description text for deprecated address
966 * handling - worse, they are not exactly the same.
967 * I believe 5.5.4 is the best one, so we follow 5.5.4.
968 */
969 if (isipv6 && !ip6_use_deprecated) {
970 struct in6_ifaddr *ia6;
971
972 if ((ia6 = ip6_getdstifaddr(m)) &&
973 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
974 tp = NULL;
975 rstreason = BANDLIM_RST_OPENPORT;
976 goto dropwithreset;
977 }
978 }
979 #endif
980 if (so->so_filt) {
981 if (isipv6) {
982 struct sockaddr_in6 *sin6 = (struct sockaddr_in6*)&from;
983
984 sin6->sin6_len = sizeof(*sin6);
985 sin6->sin6_family = AF_INET6;
986 sin6->sin6_port = th->th_sport;
987 sin6->sin6_flowinfo = 0;
988 sin6->sin6_addr = ip6->ip6_src;
989 sin6->sin6_scope_id = 0;
990 } else {
991 struct sockaddr_in *sin = (struct sockaddr_in*)&from;
992
993 sin->sin_len = sizeof(*sin);
994 sin->sin_family = AF_INET;
995 sin->sin_port = th->th_sport;
996 sin->sin_addr = ip->ip_src;
997 }
998 so2 = sonewconn(so, 0, (struct sockaddr*)&from);
999 } else {
1000 so2 = sonewconn(so, 0, NULL);
1001 }
1002 if (so2 == 0) {
1003 tcpstat.tcps_listendrop++;
1004 if (tcpdropdropablreq(so)) {
1005 if (so->so_filt)
1006 so2 = sonewconn(so, 0, (struct sockaddr*)&from);
1007 else
1008 so2 = sonewconn(so, 0, NULL);
1009 }
1010 if (!so2)
1011 goto drop;
1012 }
1013 /*
1014 * Make sure listening socket did not get closed during socket allocation,
1015 * not only this is incorrect but it is know to cause panic
1016 */
1017 if (so->so_gencnt != ogencnt)
1018 goto drop;
1019
1020 oso = so;
1021 tcp_unlock(so, 0, 0); /* Unlock but keep a reference on listener for now */
1022
1023 so = so2;
1024 tcp_lock(so, 1, 0);
1025 /*
1026 * This is ugly, but ....
1027 *
1028 * Mark socket as temporary until we're
1029 * committed to keeping it. The code at
1030 * ``drop'' and ``dropwithreset'' check the
1031 * flag dropsocket to see if the temporary
1032 * socket created here should be discarded.
1033 * We mark the socket as discardable until
1034 * we're committed to it below in TCPS_LISTEN.
1035 */
1036 dropsocket++;
1037 inp = (struct inpcb *)so->so_pcb;
1038 #if INET6
1039 if (isipv6)
1040 inp->in6p_laddr = ip6->ip6_dst;
1041 else {
1042 inp->inp_vflag &= ~INP_IPV6;
1043 inp->inp_vflag |= INP_IPV4;
1044 #endif /* INET6 */
1045 inp->inp_laddr = ip->ip_dst;
1046 #if INET6
1047 }
1048 #endif /* INET6 */
1049 inp->inp_lport = th->th_dport;
1050 if (in_pcbinshash(inp, 0) != 0) {
1051 /*
1052 * Undo the assignments above if we failed to
1053 * put the PCB on the hash lists.
1054 */
1055 #if INET6
1056 if (isipv6)
1057 inp->in6p_laddr = in6addr_any;
1058 else
1059 #endif /* INET6 */
1060 inp->inp_laddr.s_addr = INADDR_ANY;
1061 inp->inp_lport = 0;
1062 tcp_lock(oso, 0, 0); /* release ref on parent */
1063 tcp_unlock(oso, 1, 0);
1064 goto drop;
1065 }
1066 #if IPSEC
1067 /*
1068 * To avoid creating incorrectly cached IPsec
1069 * association, this is need to be done here.
1070 *
1071 * Subject: (KAME-snap 748)
1072 * From: Wayne Knowles <w.knowles@niwa.cri.nz>
1073 * ftp://ftp.kame.net/pub/mail-list/snap-users/748
1074 */
1075 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
1076 /*
1077 * Note: dropwithreset makes sure we don't
1078 * send a RST in response to a RST.
1079 */
1080 tcp_lock(oso, 0, 0); /* release ref on parent */
1081 tcp_unlock(oso, 1, 0);
1082 if (thflags & TH_ACK) {
1083 tcpstat.tcps_badsyn++;
1084 rstreason = BANDLIM_RST_OPENPORT;
1085 goto dropwithreset;
1086 }
1087 goto drop;
1088 }
1089 #endif
1090 #if INET6
1091 if (isipv6) {
1092 /*
1093 * Inherit socket options from the listening
1094 * socket.
1095 * Note that in6p_inputopts are not (even
1096 * should not be) copied, since it stores
1097 * previously received options and is used to
1098 * detect if each new option is different than
1099 * the previous one and hence should be passed
1100 * to a user.
1101 * If we copied in6p_inputopts, a user would
1102 * not be able to receive options just after
1103 * calling the accept system call.
1104 */
1105 inp->inp_flags |=
1106 oinp->inp_flags & INP_CONTROLOPTS;
1107 if (oinp->in6p_outputopts)
1108 inp->in6p_outputopts =
1109 ip6_copypktopts(oinp->in6p_outputopts,
1110 M_NOWAIT);
1111 } else
1112 #endif /* INET6 */
1113 inp->inp_options = ip_srcroute();
1114 tcp_lock(oso, 0, 0);
1115 #if IPSEC
1116 /* copy old policy into new socket's */
1117 if (sotoinpcb(oso)->inp_sp)
1118 {
1119 int error = 0;
1120 lck_mtx_lock(sadb_mutex);
1121 /* Is it a security hole here to silently fail to copy the policy? */
1122 if (inp->inp_sp != NULL)
1123 error = ipsec_init_policy(so, &inp->inp_sp);
1124 if (error != 0 || ipsec_copy_policy(sotoinpcb(oso)->inp_sp, inp->inp_sp))
1125 printf("tcp_input: could not copy policy\n");
1126 lck_mtx_unlock(sadb_mutex);
1127 }
1128 #endif
1129 tcp_unlock(oso, 1, 0); /* now drop the reference on the listener */
1130 tp = intotcpcb(inp);
1131 tp->t_state = TCPS_LISTEN;
1132 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT|TF_NODELAY);
1133 tp->t_inpcb->inp_ip_ttl = tp0->t_inpcb->inp_ip_ttl;
1134 /* Compute proper scaling value from buffer space */
1135 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1136 TCP_MAXWIN << tp->request_r_scale <
1137 so->so_rcv.sb_hiwat)
1138 tp->request_r_scale++;
1139
1140 KERNEL_DEBUG(DBG_FNC_TCP_NEWCONN | DBG_FUNC_END,0,0,0,0,0);
1141 }
1142 }
1143
1144 #if 1
1145 lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
1146 #endif
1147 /*
1148 * Radar 3529618
1149 * This is the second part of the MSS DoS prevention code (after
1150 * minmss on the sending side) and it deals with too many too small
1151 * tcp packets in a too short timeframe (1 second).
1152 *
1153 * For every full second we count the number of received packets
1154 * and bytes. If we get a lot of packets per second for this connection
1155 * (tcp_minmssoverload) we take a closer look at it and compute the
1156 * average packet size for the past second. If that is less than
1157 * tcp_minmss we get too many packets with very small payload which
1158 * is not good and burdens our system (and every packet generates
1159 * a wakeup to the process connected to our socket). We can reasonable
1160 * expect this to be small packet DoS attack to exhaust our CPU
1161 * cycles.
1162 *
1163 * Care has to be taken for the minimum packet overload value. This
1164 * value defines the minimum number of packets per second before we
1165 * start to worry. This must not be too low to avoid killing for
1166 * example interactive connections with many small packets like
1167 * telnet or SSH.
1168 *
1169 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables
1170 * this check.
1171 *
1172 * Account for packet if payload packet, skip over ACK, etc.
1173 */
1174 if (tcp_minmss && tcp_minmssoverload &&
1175 tp->t_state == TCPS_ESTABLISHED && tlen > 0) {
1176 if (tp->rcv_reset > tcp_now) {
1177 tp->rcv_pps++;
1178 tp->rcv_byps += tlen + off;
1179 if (tp->rcv_pps > tcp_minmssoverload) {
1180 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) {
1181 char ipstrbuf[MAX_IPv6_STR_LEN];
1182 printf("too many small tcp packets from "
1183 "%s:%u, av. %lubyte/packet, "
1184 "dropping connection\n",
1185 #ifdef INET6
1186 isipv6 ?
1187 inet_ntop(AF_INET6, &inp->in6p_faddr, ipstrbuf,
1188 sizeof(ipstrbuf)) :
1189 #endif
1190 inet_ntop(AF_INET, &inp->inp_faddr, ipstrbuf,
1191 sizeof(ipstrbuf)),
1192 inp->inp_fport,
1193 tp->rcv_byps / tp->rcv_pps);
1194 tp = tcp_drop(tp, ECONNRESET);
1195 /* tcpstat.tcps_minmssdrops++; */
1196 goto drop;
1197 }
1198 }
1199 } else {
1200 tp->rcv_reset = tcp_now + PR_SLOWHZ;
1201 tp->rcv_pps = 1;
1202 tp->rcv_byps = tlen + off;
1203 }
1204 }
1205
1206 /*
1207 * Segment received on connection.
1208 * Reset idle time and keep-alive timer.
1209 */
1210 tp->t_rcvtime = 0;
1211 if (TCPS_HAVEESTABLISHED(tp->t_state))
1212 tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp);
1213
1214 /*
1215 * Process options if not in LISTEN state,
1216 * else do it below (after getting remote address).
1217 */
1218 if (tp->t_state != TCPS_LISTEN && optp)
1219 tcp_dooptions(tp, optp, optlen, th, &to);
1220
1221 /*
1222 * Header prediction: check for the two common cases
1223 * of a uni-directional data xfer. If the packet has
1224 * no control flags, is in-sequence, the window didn't
1225 * change and we're not retransmitting, it's a
1226 * candidate. If the length is zero and the ack moved
1227 * forward, we're the sender side of the xfer. Just
1228 * free the data acked & wake any higher level process
1229 * that was blocked waiting for space. If the length
1230 * is non-zero and the ack didn't move, we're the
1231 * receiver side. If we're getting packets in-order
1232 * (the reassembly queue is empty), add the data to
1233 * the socket buffer and note that we need a delayed ack.
1234 * Make sure that the hidden state-flags are also off.
1235 * Since we check for TCPS_ESTABLISHED above, it can only
1236 * be TH_NEEDSYN.
1237 */
1238 if (tp->t_state == TCPS_ESTABLISHED &&
1239 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1240 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1241 ((to.to_flag & TOF_TS) == 0 ||
1242 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1243 /*
1244 * Using the CC option is compulsory if once started:
1245 * the segment is OK if no T/TCP was negotiated or
1246 * if the segment has a CC option equal to CCrecv
1247 */
1248 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1249 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
1250 th->th_seq == tp->rcv_nxt &&
1251 tiwin && tiwin == tp->snd_wnd &&
1252 tp->snd_nxt == tp->snd_max) {
1253
1254 /*
1255 * If last ACK falls within this segment's sequence numbers,
1256 * record the timestamp.
1257 * NOTE that the test is modified according to the latest
1258 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1259 */
1260 if ((to.to_flag & TOF_TS) != 0 &&
1261 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1262 tp->ts_recent_age = tcp_now;
1263 tp->ts_recent = to.to_tsval;
1264 }
1265
1266 if (tlen == 0) {
1267 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1268 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1269 tp->snd_cwnd >= tp->snd_wnd &&
1270 tp->t_dupacks < tcprexmtthresh) {
1271 /*
1272 * this is a pure ack for outstanding data.
1273 */
1274 ++tcpstat.tcps_predack;
1275 /*
1276 * "bad retransmit" recovery
1277 */
1278 if (tp->t_rxtshift == 1 &&
1279 tcp_now < tp->t_badrxtwin) {
1280 tp->snd_cwnd = tp->snd_cwnd_prev;
1281 tp->snd_ssthresh =
1282 tp->snd_ssthresh_prev;
1283 tp->snd_nxt = tp->snd_max;
1284 tp->t_badrxtwin = 0;
1285 }
1286 if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0)) /* Makes sure we already have a TS */
1287 tcp_xmit_timer(tp,
1288 tcp_now - to.to_tsecr + 1);
1289 else if (tp->t_rtttime &&
1290 SEQ_GT(th->th_ack, tp->t_rtseq))
1291 tcp_xmit_timer(tp, tp->t_rtttime);
1292 acked = th->th_ack - tp->snd_una;
1293 tcpstat.tcps_rcvackpack++;
1294 tcpstat.tcps_rcvackbyte += acked;
1295 sbdrop(&so->so_snd, acked);
1296 tp->snd_una = th->th_ack;
1297 m_freem(m);
1298 ND6_HINT(tp); /* some progress has been done */
1299
1300 /*
1301 * If all outstanding data are acked, stop
1302 * retransmit timer, otherwise restart timer
1303 * using current (possibly backed-off) value.
1304 * If process is waiting for space,
1305 * wakeup/selwakeup/signal. If data
1306 * are ready to send, let tcp_output
1307 * decide between more output or persist.
1308 */
1309 if (tp->snd_una == tp->snd_max)
1310 tp->t_timer[TCPT_REXMT] = 0;
1311 else if (tp->t_timer[TCPT_PERSIST] == 0)
1312 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1313
1314 sowwakeup(so); /* has to be done with socket lock held */
1315 if ((so->so_snd.sb_cc) || (tp->t_flags & TF_ACKNOW))
1316 (void) tcp_output(tp);
1317 tcp_unlock(so, 1, 0);
1318 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
1319 return;
1320 }
1321 } else if (th->th_ack == tp->snd_una &&
1322 LIST_EMPTY(&tp->t_segq) &&
1323 tlen <= sbspace(&so->so_rcv)) {
1324 /*
1325 * this is a pure, in-sequence data packet
1326 * with nothing on the reassembly queue and
1327 * we have enough buffer space to take it.
1328 */
1329 ++tcpstat.tcps_preddat;
1330 tp->rcv_nxt += tlen;
1331 tcpstat.tcps_rcvpack++;
1332 tcpstat.tcps_rcvbyte += tlen;
1333 ND6_HINT(tp); /* some progress has been done */
1334 /*
1335 * Add data to socket buffer.
1336 */
1337 m_adj(m, drop_hdrlen); /* delayed header drop */
1338 if (sbappend(&so->so_rcv, m))
1339 sorwakeup(so);
1340 #if INET6
1341 if (isipv6) {
1342 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
1343 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
1344 th->th_seq, th->th_ack, th->th_win);
1345 }
1346 else
1347 #endif
1348 {
1349 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
1350 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
1351 th->th_seq, th->th_ack, th->th_win);
1352 }
1353 if (DELAY_ACK(tp)) {
1354 tp->t_flags |= TF_DELACK;
1355 } else {
1356 tp->t_flags |= TF_ACKNOW;
1357 tcp_output(tp);
1358 }
1359 tcp_unlock(so, 1, 0);
1360 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
1361 return;
1362 }
1363 }
1364
1365 /*
1366 * Calculate amount of space in receive window,
1367 * and then do TCP input processing.
1368 * Receive window is amount of space in rcv queue,
1369 * but not less than advertised window.
1370 */
1371 #if 1
1372 lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
1373 #endif
1374 { int win;
1375
1376 win = sbspace(&so->so_rcv);
1377 if (win < 0)
1378 win = 0;
1379 else { /* clip rcv window to 4K for modems */
1380 if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0)
1381 win = min(win, slowlink_wsize);
1382 }
1383 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1384 }
1385
1386 switch (tp->t_state) {
1387
1388 /*
1389 * If the state is LISTEN then ignore segment if it contains an RST.
1390 * If the segment contains an ACK then it is bad and send a RST.
1391 * If it does not contain a SYN then it is not interesting; drop it.
1392 * If it is from this socket, drop it, it must be forged.
1393 * Don't bother responding if the destination was a broadcast.
1394 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
1395 * tp->iss, and send a segment:
1396 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
1397 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
1398 * Fill in remote peer address fields if not previously specified.
1399 * Enter SYN_RECEIVED state, and process any other fields of this
1400 * segment in this state.
1401 */
1402 case TCPS_LISTEN: {
1403 register struct sockaddr_in *sin;
1404 #if INET6
1405 register struct sockaddr_in6 *sin6;
1406 #endif
1407
1408 #if 1
1409 lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
1410 #endif
1411 if (thflags & TH_RST)
1412 goto drop;
1413 if (thflags & TH_ACK) {
1414 rstreason = BANDLIM_RST_OPENPORT;
1415 goto dropwithreset;
1416 }
1417 if ((thflags & TH_SYN) == 0)
1418 goto drop;
1419 if (th->th_dport == th->th_sport) {
1420 #if INET6
1421 if (isipv6) {
1422 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1423 &ip6->ip6_src))
1424 goto drop;
1425 } else
1426 #endif /* INET6 */
1427 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1428 goto drop;
1429 }
1430 /*
1431 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1432 * in_broadcast() should never return true on a received
1433 * packet with M_BCAST not set.
1434 *
1435 * Packets with a multicast source address should also
1436 * be discarded.
1437 */
1438 if (m->m_flags & (M_BCAST|M_MCAST))
1439 goto drop;
1440 #if INET6
1441 if (isipv6) {
1442 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1443 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1444 goto drop;
1445 } else
1446 #endif
1447 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1448 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1449 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1450 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1451 goto drop;
1452 #if INET6
1453 if (isipv6) {
1454 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6,
1455 M_SONAME, M_NOWAIT);
1456 if (sin6 == NULL)
1457 goto drop;
1458 bzero(sin6, sizeof(*sin6));
1459 sin6->sin6_family = AF_INET6;
1460 sin6->sin6_len = sizeof(*sin6);
1461 sin6->sin6_addr = ip6->ip6_src;
1462 sin6->sin6_port = th->th_sport;
1463 laddr6 = inp->in6p_laddr;
1464 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1465 inp->in6p_laddr = ip6->ip6_dst;
1466 if (in6_pcbconnect(inp, (struct sockaddr *)sin6,
1467 proc0)) {
1468 inp->in6p_laddr = laddr6;
1469 FREE(sin6, M_SONAME);
1470 goto drop;
1471 }
1472 FREE(sin6, M_SONAME);
1473 } else
1474 #endif
1475 {
1476 #if 1
1477 lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
1478 #endif
1479 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
1480 M_NOWAIT);
1481 if (sin == NULL)
1482 goto drop;
1483 sin->sin_family = AF_INET;
1484 sin->sin_len = sizeof(*sin);
1485 sin->sin_addr = ip->ip_src;
1486 sin->sin_port = th->th_sport;
1487 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
1488 laddr = inp->inp_laddr;
1489 if (inp->inp_laddr.s_addr == INADDR_ANY)
1490 inp->inp_laddr = ip->ip_dst;
1491 if (in_pcbconnect(inp, (struct sockaddr *)sin, proc0)) {
1492 inp->inp_laddr = laddr;
1493 FREE(sin, M_SONAME);
1494 goto drop;
1495 }
1496 FREE(sin, M_SONAME);
1497 }
1498 if ((taop = tcp_gettaocache(inp)) == NULL) {
1499 taop = &tao_noncached;
1500 bzero(taop, sizeof(*taop));
1501 }
1502 tcp_dooptions(tp, optp, optlen, th, &to);
1503 if (iss)
1504 tp->iss = iss;
1505 else {
1506 tp->iss = tcp_new_isn(tp);
1507 }
1508 tp->irs = th->th_seq;
1509 tcp_sendseqinit(tp);
1510 tcp_rcvseqinit(tp);
1511 tp->snd_recover = tp->snd_una;
1512 /*
1513 * Initialization of the tcpcb for transaction;
1514 * set SND.WND = SEG.WND,
1515 * initialize CCsend and CCrecv.
1516 */
1517 tp->snd_wnd = tiwin; /* initial send-window */
1518 tp->cc_send = CC_INC(tcp_ccgen);
1519 tp->cc_recv = to.to_cc;
1520 /*
1521 * Perform TAO test on incoming CC (SEG.CC) option, if any.
1522 * - compare SEG.CC against cached CC from the same host,
1523 * if any.
1524 * - if SEG.CC > chached value, SYN must be new and is accepted
1525 * immediately: save new CC in the cache, mark the socket
1526 * connected, enter ESTABLISHED state, turn on flag to
1527 * send a SYN in the next segment.
1528 * A virtual advertised window is set in rcv_adv to
1529 * initialize SWS prevention. Then enter normal segment
1530 * processing: drop SYN, process data and FIN.
1531 * - otherwise do a normal 3-way handshake.
1532 */
1533 if ((to.to_flag & TOF_CC) != 0) {
1534 if (((tp->t_flags & TF_NOPUSH) != 0) &&
1535 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
1536
1537 taop->tao_cc = to.to_cc;
1538
1539 tp->t_state = TCPS_ESTABLISHED;
1540
1541 /*
1542 * If there is a FIN, or if there is data and the
1543 * connection is local, then delay SYN,ACK(SYN) in
1544 * the hope of piggy-backing it on a response
1545 * segment. Otherwise must send ACK now in case
1546 * the other side is slow starting.
1547 */
1548 if (DELAY_ACK(tp) && ((thflags & TH_FIN) ||
1549 (tlen != 0 &&
1550 #if INET6
1551 (isipv6 && in6_localaddr(&inp->in6p_faddr))
1552 ||
1553 (!isipv6 &&
1554 #endif /* INET6 */
1555 in_localaddr(inp->inp_faddr)
1556 #if INET6
1557 )
1558 #endif /* INET6 */
1559 ))) {
1560 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
1561 }
1562 else {
1563 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1564 }
1565
1566 /*
1567 * Limit the `virtual advertised window' to TCP_MAXWIN
1568 * here. Even if we requested window scaling, it will
1569 * become effective only later when our SYN is acked.
1570 */
1571 if (tp->t_flags & TF_SLOWLINK && slowlink_wsize > 0) /* clip window size for for slow link */
1572 tp->rcv_adv += min(tp->rcv_wnd, slowlink_wsize);
1573 else
1574 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
1575 tcpstat.tcps_connects++;
1576 soisconnected(so);
1577 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
1578 dropsocket = 0; /* committed to socket */
1579 tcpstat.tcps_accepts++;
1580 goto trimthenstep6;
1581 }
1582 /* else do standard 3-way handshake */
1583 } else {
1584 /*
1585 * No CC option, but maybe CC.NEW:
1586 * invalidate cached value.
1587 */
1588 taop->tao_cc = 0;
1589 }
1590 /*
1591 * TAO test failed or there was no CC option,
1592 * do a standard 3-way handshake.
1593 */
1594 tp->t_flags |= TF_ACKNOW;
1595 tp->t_state = TCPS_SYN_RECEIVED;
1596 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
1597 dropsocket = 0; /* committed to socket */
1598 tcpstat.tcps_accepts++;
1599 goto trimthenstep6;
1600 }
1601
1602 /*
1603 * If the state is SYN_RECEIVED:
1604 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1605 */
1606 case TCPS_SYN_RECEIVED:
1607 if ((thflags & TH_ACK) &&
1608 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1609 SEQ_GT(th->th_ack, tp->snd_max))) {
1610 rstreason = BANDLIM_RST_OPENPORT;
1611 goto dropwithreset;
1612 }
1613 break;
1614
1615 /*
1616 * If the state is SYN_SENT:
1617 * if seg contains an ACK, but not for our SYN, drop the input.
1618 * if seg contains a RST, then drop the connection.
1619 * if seg does not contain SYN, then drop it.
1620 * Otherwise this is an acceptable SYN segment
1621 * initialize tp->rcv_nxt and tp->irs
1622 * if seg contains ack then advance tp->snd_una
1623 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1624 * arrange for segment to be acked (eventually)
1625 * continue processing rest of data/controls, beginning with URG
1626 */
1627 case TCPS_SYN_SENT:
1628 if ((taop = tcp_gettaocache(inp)) == NULL) {
1629 taop = &tao_noncached;
1630 bzero(taop, sizeof(*taop));
1631 }
1632
1633 if ((thflags & TH_ACK) &&
1634 (SEQ_LEQ(th->th_ack, tp->iss) ||
1635 SEQ_GT(th->th_ack, tp->snd_max))) {
1636 /*
1637 * If we have a cached CCsent for the remote host,
1638 * hence we haven't just crashed and restarted,
1639 * do not send a RST. This may be a retransmission
1640 * from the other side after our earlier ACK was lost.
1641 * Our new SYN, when it arrives, will serve as the
1642 * needed ACK.
1643 */
1644 if (taop->tao_ccsent != 0)
1645 goto drop;
1646 else {
1647 rstreason = BANDLIM_UNLIMITED;
1648 goto dropwithreset;
1649 }
1650 }
1651 if (thflags & TH_RST) {
1652 if (thflags & TH_ACK) {
1653 tp = tcp_drop(tp, ECONNREFUSED);
1654 postevent(so, 0, EV_RESET);
1655 }
1656 goto drop;
1657 }
1658 if ((thflags & TH_SYN) == 0)
1659 goto drop;
1660 tp->snd_wnd = th->th_win; /* initial send window */
1661 tp->cc_recv = to.to_cc; /* foreign CC */
1662
1663 tp->irs = th->th_seq;
1664 tcp_rcvseqinit(tp);
1665 if (thflags & TH_ACK) {
1666 /*
1667 * Our SYN was acked. If segment contains CC.ECHO
1668 * option, check it to make sure this segment really
1669 * matches our SYN. If not, just drop it as old
1670 * duplicate, but send an RST if we're still playing
1671 * by the old rules. If no CC.ECHO option, make sure
1672 * we don't get fooled into using T/TCP.
1673 */
1674 if (to.to_flag & TOF_CCECHO) {
1675 if (tp->cc_send != to.to_ccecho) {
1676 if (taop->tao_ccsent != 0)
1677 goto drop;
1678 else {
1679 rstreason = BANDLIM_UNLIMITED;
1680 goto dropwithreset;
1681 }
1682 }
1683 } else
1684 tp->t_flags &= ~TF_RCVD_CC;
1685 tcpstat.tcps_connects++;
1686 soisconnected(so);
1687 /* Do window scaling on this connection? */
1688 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1689 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1690 tp->snd_scale = tp->requested_s_scale;
1691 tp->rcv_scale = tp->request_r_scale;
1692 }
1693 /* Segment is acceptable, update cache if undefined. */
1694 if (taop->tao_ccsent == 0)
1695 taop->tao_ccsent = to.to_ccecho;
1696
1697 tp->rcv_adv += tp->rcv_wnd;
1698 tp->snd_una++; /* SYN is acked */
1699 /*
1700 * If there's data, delay ACK; if there's also a FIN
1701 * ACKNOW will be turned on later.
1702 */
1703 if (DELAY_ACK(tp) && tlen != 0) {
1704 tp->t_flags |= TF_DELACK;
1705 }
1706 else {
1707 tp->t_flags |= TF_ACKNOW;
1708 }
1709 /*
1710 * Received <SYN,ACK> in SYN_SENT[*] state.
1711 * Transitions:
1712 * SYN_SENT --> ESTABLISHED
1713 * SYN_SENT* --> FIN_WAIT_1
1714 */
1715 if (tp->t_flags & TF_NEEDFIN) {
1716 tp->t_state = TCPS_FIN_WAIT_1;
1717 tp->t_flags &= ~TF_NEEDFIN;
1718 thflags &= ~TH_SYN;
1719 } else {
1720 tp->t_state = TCPS_ESTABLISHED;
1721 tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp);
1722 }
1723 } else {
1724 /*
1725 * Received initial SYN in SYN-SENT[*] state => simul-
1726 * taneous open. If segment contains CC option and there is
1727 * a cached CC, apply TAO test; if it succeeds, connection is
1728 * half-synchronized. Otherwise, do 3-way handshake:
1729 * SYN-SENT -> SYN-RECEIVED
1730 * SYN-SENT* -> SYN-RECEIVED*
1731 * If there was no CC option, clear cached CC value.
1732 */
1733 tp->t_flags |= TF_ACKNOW;
1734 tp->t_timer[TCPT_REXMT] = 0;
1735 if (to.to_flag & TOF_CC) {
1736 if (taop->tao_cc != 0 &&
1737 CC_GT(to.to_cc, taop->tao_cc)) {
1738 /*
1739 * update cache and make transition:
1740 * SYN-SENT -> ESTABLISHED*
1741 * SYN-SENT* -> FIN-WAIT-1*
1742 */
1743 taop->tao_cc = to.to_cc;
1744 if (tp->t_flags & TF_NEEDFIN) {
1745 tp->t_state = TCPS_FIN_WAIT_1;
1746 tp->t_flags &= ~TF_NEEDFIN;
1747 } else {
1748 tp->t_state = TCPS_ESTABLISHED;
1749 tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp);
1750 }
1751 tp->t_flags |= TF_NEEDSYN;
1752 } else
1753 tp->t_state = TCPS_SYN_RECEIVED;
1754 } else {
1755 /* CC.NEW or no option => invalidate cache */
1756 taop->tao_cc = 0;
1757 tp->t_state = TCPS_SYN_RECEIVED;
1758 }
1759 }
1760
1761 trimthenstep6:
1762 /*
1763 * Advance th->th_seq to correspond to first data byte.
1764 * If data, trim to stay within window,
1765 * dropping FIN if necessary.
1766 */
1767 th->th_seq++;
1768 if (tlen > tp->rcv_wnd) {
1769 todrop = tlen - tp->rcv_wnd;
1770 m_adj(m, -todrop);
1771 tlen = tp->rcv_wnd;
1772 thflags &= ~TH_FIN;
1773 tcpstat.tcps_rcvpackafterwin++;
1774 tcpstat.tcps_rcvbyteafterwin += todrop;
1775 }
1776 tp->snd_wl1 = th->th_seq - 1;
1777 tp->rcv_up = th->th_seq;
1778 /*
1779 * Client side of transaction: already sent SYN and data.
1780 * If the remote host used T/TCP to validate the SYN,
1781 * our data will be ACK'd; if so, enter normal data segment
1782 * processing in the middle of step 5, ack processing.
1783 * Otherwise, goto step 6.
1784 */
1785 if (thflags & TH_ACK)
1786 goto process_ACK;
1787 goto step6;
1788 /*
1789 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1790 * if segment contains a SYN and CC [not CC.NEW] option:
1791 * if state == TIME_WAIT and connection duration > MSL,
1792 * drop packet and send RST;
1793 *
1794 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1795 * ack the FIN (and data) in retransmission queue.
1796 * Complete close and delete TCPCB. Then reprocess
1797 * segment, hoping to find new TCPCB in LISTEN state;
1798 *
1799 * else must be old SYN; drop it.
1800 * else do normal processing.
1801 */
1802 case TCPS_LAST_ACK:
1803 case TCPS_CLOSING:
1804 case TCPS_TIME_WAIT:
1805 if ((thflags & TH_SYN) &&
1806 (to.to_flag & TOF_CC) && tp->cc_recv != 0) {
1807 if (tp->t_state == TCPS_TIME_WAIT &&
1808 tp->t_starttime > tcp_msl) {
1809 rstreason = BANDLIM_UNLIMITED;
1810 goto dropwithreset;
1811 }
1812 if (CC_GT(to.to_cc, tp->cc_recv)) {
1813 tp = tcp_close(tp);
1814 tcp_unlock(so, 1, 50);
1815 goto findpcb;
1816 }
1817 else
1818 goto drop;
1819 }
1820 break; /* continue normal processing */
1821
1822 /* Received a SYN while connection is already established.
1823 * This is a "half open connection and other anomalies" described
1824 * in RFC793 page 34, send an ACK so the remote reset the connection
1825 * or recovers by adjusting its sequence numberering
1826 */
1827 case TCPS_ESTABLISHED:
1828 if (thflags & TH_SYN)
1829 goto dropafterack;
1830 break;
1831 }
1832
1833 /*
1834 * States other than LISTEN or SYN_SENT.
1835 * First check the RST flag and sequence number since reset segments
1836 * are exempt from the timestamp and connection count tests. This
1837 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1838 * below which allowed reset segments in half the sequence space
1839 * to fall though and be processed (which gives forged reset
1840 * segments with a random sequence number a 50 percent chance of
1841 * killing a connection).
1842 * Then check timestamp, if present.
1843 * Then check the connection count, if present.
1844 * Then check that at least some bytes of segment are within
1845 * receive window. If segment begins before rcv_nxt,
1846 * drop leading data (and SYN); if nothing left, just ack.
1847 *
1848 *
1849 * If the RST bit is set, check the sequence number to see
1850 * if this is a valid reset segment.
1851 * RFC 793 page 37:
1852 * In all states except SYN-SENT, all reset (RST) segments
1853 * are validated by checking their SEQ-fields. A reset is
1854 * valid if its sequence number is in the window.
1855 * Note: this does not take into account delayed ACKs, so
1856 * we should test against last_ack_sent instead of rcv_nxt.
1857 * The sequence number in the reset segment is normally an
1858 * echo of our outgoing acknowlegement numbers, but some hosts
1859 * send a reset with the sequence number at the rightmost edge
1860 * of our receive window, and we have to handle this case.
1861 * If we have multiple segments in flight, the intial reset
1862 * segment sequence numbers will be to the left of last_ack_sent,
1863 * but they will eventually catch up.
1864 * In any case, it never made sense to trim reset segments to
1865 * fit the receive window since RFC 1122 says:
1866 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1867 *
1868 * A TCP SHOULD allow a received RST segment to include data.
1869 *
1870 * DISCUSSION
1871 * It has been suggested that a RST segment could contain
1872 * ASCII text that encoded and explained the cause of the
1873 * RST. No standard has yet been established for such
1874 * data.
1875 *
1876 * If the reset segment passes the sequence number test examine
1877 * the state:
1878 * SYN_RECEIVED STATE:
1879 * If passive open, return to LISTEN state.
1880 * If active open, inform user that connection was refused.
1881 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1882 * Inform user that connection was reset, and close tcb.
1883 * CLOSING, LAST_ACK STATES:
1884 * Close the tcb.
1885 * TIME_WAIT STATE:
1886 * Drop the segment - see Stevens, vol. 2, p. 964 and
1887 * RFC 1337.
1888 */
1889 if (thflags & TH_RST) {
1890 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1891 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1892 switch (tp->t_state) {
1893
1894 case TCPS_SYN_RECEIVED:
1895 so->so_error = ECONNREFUSED;
1896 goto close;
1897
1898 case TCPS_ESTABLISHED:
1899 case TCPS_FIN_WAIT_1:
1900 case TCPS_CLOSE_WAIT:
1901 /*
1902 Drop through ...
1903 */
1904 case TCPS_FIN_WAIT_2:
1905 so->so_error = ECONNRESET;
1906 close:
1907 postevent(so, 0, EV_RESET);
1908 tp->t_state = TCPS_CLOSED;
1909 tcpstat.tcps_drops++;
1910 tp = tcp_close(tp);
1911 break;
1912
1913 case TCPS_CLOSING:
1914 case TCPS_LAST_ACK:
1915 tp = tcp_close(tp);
1916 break;
1917
1918 case TCPS_TIME_WAIT:
1919 break;
1920 }
1921 }
1922 goto drop;
1923 }
1924
1925 #if 1
1926 lck_mtx_assert(((struct inpcb *)so->so_pcb)->inpcb_mtx, LCK_MTX_ASSERT_OWNED);
1927 #endif
1928 /*
1929 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1930 * and it's less than ts_recent, drop it.
1931 */
1932 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent &&
1933 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1934
1935 /* Check to see if ts_recent is over 24 days old. */
1936 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1937 /*
1938 * Invalidate ts_recent. If this segment updates
1939 * ts_recent, the age will be reset later and ts_recent
1940 * will get a valid value. If it does not, setting
1941 * ts_recent to zero will at least satisfy the
1942 * requirement that zero be placed in the timestamp
1943 * echo reply when ts_recent isn't valid. The
1944 * age isn't reset until we get a valid ts_recent
1945 * because we don't want out-of-order segments to be
1946 * dropped when ts_recent is old.
1947 */
1948 tp->ts_recent = 0;
1949 } else {
1950 tcpstat.tcps_rcvduppack++;
1951 tcpstat.tcps_rcvdupbyte += tlen;
1952 tcpstat.tcps_pawsdrop++;
1953 goto dropafterack;
1954 }
1955 }
1956
1957 /*
1958 * T/TCP mechanism
1959 * If T/TCP was negotiated and the segment doesn't have CC,
1960 * or if its CC is wrong then drop the segment.
1961 * RST segments do not have to comply with this.
1962 */
1963 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1964 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1965 goto dropafterack;
1966
1967 /*
1968 * In the SYN-RECEIVED state, validate that the packet belongs to
1969 * this connection before trimming the data to fit the receive
1970 * window. Check the sequence number versus IRS since we know
1971 * the sequence numbers haven't wrapped. This is a partial fix
1972 * for the "LAND" DoS attack.
1973 */
1974 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1975 rstreason = BANDLIM_RST_OPENPORT;
1976 goto dropwithreset;
1977 }
1978
1979 todrop = tp->rcv_nxt - th->th_seq;
1980 if (todrop > 0) {
1981 if (thflags & TH_SYN) {
1982 thflags &= ~TH_SYN;
1983 th->th_seq++;
1984 if (th->th_urp > 1)
1985 th->th_urp--;
1986 else
1987 thflags &= ~TH_URG;
1988 todrop--;
1989 }
1990 /*
1991 * Following if statement from Stevens, vol. 2, p. 960.
1992 */
1993 if (todrop > tlen
1994 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1995 /*
1996 * Any valid FIN must be to the left of the window.
1997 * At this point the FIN must be a duplicate or out
1998 * of sequence; drop it.
1999 */
2000 thflags &= ~TH_FIN;
2001
2002 /*
2003 * Send an ACK to resynchronize and drop any data.
2004 * But keep on processing for RST or ACK.
2005 */
2006 tp->t_flags |= TF_ACKNOW;
2007 todrop = tlen;
2008 tcpstat.tcps_rcvduppack++;
2009 tcpstat.tcps_rcvdupbyte += todrop;
2010 } else {
2011 tcpstat.tcps_rcvpartduppack++;
2012 tcpstat.tcps_rcvpartdupbyte += todrop;
2013 }
2014 drop_hdrlen += todrop; /* drop from the top afterwards */
2015 th->th_seq += todrop;
2016 tlen -= todrop;
2017 if (th->th_urp > todrop)
2018 th->th_urp -= todrop;
2019 else {
2020 thflags &= ~TH_URG;
2021 th->th_urp = 0;
2022 }
2023 }
2024
2025 /*
2026 * If new data are received on a connection after the
2027 * user processes are gone, then RST the other end.
2028 */
2029 if ((so->so_state & SS_NOFDREF) &&
2030 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2031 tp = tcp_close(tp);
2032 tcpstat.tcps_rcvafterclose++;
2033 rstreason = BANDLIM_UNLIMITED;
2034 goto dropwithreset;
2035 }
2036
2037 /*
2038 * If segment ends after window, drop trailing data
2039 * (and PUSH and FIN); if nothing left, just ACK.
2040 */
2041 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd);
2042 if (todrop > 0) {
2043 tcpstat.tcps_rcvpackafterwin++;
2044 if (todrop >= tlen) {
2045 tcpstat.tcps_rcvbyteafterwin += tlen;
2046 /*
2047 * If a new connection request is received
2048 * while in TIME_WAIT, drop the old connection
2049 * and start over if the sequence numbers
2050 * are above the previous ones.
2051 */
2052 if (thflags & TH_SYN &&
2053 tp->t_state == TCPS_TIME_WAIT &&
2054 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
2055 iss = tcp_new_isn(tp);
2056 tp = tcp_close(tp);
2057 tcp_unlock(so, 1, 0);
2058 goto findpcb;
2059 }
2060 /*
2061 * If window is closed can only take segments at
2062 * window edge, and have to drop data and PUSH from
2063 * incoming segments. Continue processing, but
2064 * remember to ack. Otherwise, drop segment
2065 * and ack.
2066 */
2067 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2068 tp->t_flags |= TF_ACKNOW;
2069 tcpstat.tcps_rcvwinprobe++;
2070 } else
2071 goto dropafterack;
2072 } else
2073 tcpstat.tcps_rcvbyteafterwin += todrop;
2074 m_adj(m, -todrop);
2075 tlen -= todrop;
2076 thflags &= ~(TH_PUSH|TH_FIN);
2077 }
2078
2079 /*
2080 * If last ACK falls within this segment's sequence numbers,
2081 * record its timestamp.
2082 * NOTE that the test is modified according to the latest
2083 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2084 */
2085 if ((to.to_flag & TOF_TS) != 0 &&
2086 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
2087 tp->ts_recent_age = tcp_now;
2088 tp->ts_recent = to.to_tsval;
2089 }
2090
2091 /*
2092 * If a SYN is in the window, then this is an
2093 * error and we send an RST and drop the connection.
2094 */
2095 if (thflags & TH_SYN) {
2096 tp = tcp_drop(tp, ECONNRESET);
2097 rstreason = BANDLIM_UNLIMITED;
2098 postevent(so, 0, EV_RESET);
2099 goto dropwithreset;
2100 }
2101
2102 /*
2103 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2104 * flag is on (half-synchronized state), then queue data for
2105 * later processing; else drop segment and return.
2106 */
2107 if ((thflags & TH_ACK) == 0) {
2108 if (tp->t_state == TCPS_SYN_RECEIVED ||
2109 (tp->t_flags & TF_NEEDSYN))
2110 goto step6;
2111 else
2112 goto drop;
2113 }
2114
2115 /*
2116 * Ack processing.
2117 */
2118 switch (tp->t_state) {
2119
2120 /*
2121 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2122 * ESTABLISHED state and continue processing.
2123 * The ACK was checked above.
2124 */
2125 case TCPS_SYN_RECEIVED:
2126
2127 tcpstat.tcps_connects++;
2128 soisconnected(so);
2129
2130 /* Do window scaling? */
2131 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2132 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2133 tp->snd_scale = tp->requested_s_scale;
2134 tp->rcv_scale = tp->request_r_scale;
2135 }
2136 /*
2137 * Upon successful completion of 3-way handshake,
2138 * update cache.CC if it was undefined, pass any queued
2139 * data to the user, and advance state appropriately.
2140 */
2141 if ((taop = tcp_gettaocache(inp)) != NULL &&
2142 taop->tao_cc == 0)
2143 taop->tao_cc = tp->cc_recv;
2144
2145 /*
2146 * Make transitions:
2147 * SYN-RECEIVED -> ESTABLISHED
2148 * SYN-RECEIVED* -> FIN-WAIT-1
2149 */
2150 if (tp->t_flags & TF_NEEDFIN) {
2151 tp->t_state = TCPS_FIN_WAIT_1;
2152 tp->t_flags &= ~TF_NEEDFIN;
2153 } else {
2154 tp->t_state = TCPS_ESTABLISHED;
2155 tp->t_timer[TCPT_KEEP] = TCP_KEEPIDLE(tp);
2156 }
2157 /*
2158 * If segment contains data or ACK, will call tcp_reass()
2159 * later; if not, do so now to pass queued data to user.
2160 */
2161 if (tlen == 0 && (thflags & TH_FIN) == 0)
2162 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2163 (struct mbuf *)0);
2164 tp->snd_wl1 = th->th_seq - 1;
2165 /* fall into ... */
2166
2167 /*
2168 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2169 * ACKs. If the ack is in the range
2170 * tp->snd_una < th->th_ack <= tp->snd_max
2171 * then advance tp->snd_una to th->th_ack and drop
2172 * data from the retransmission queue. If this ACK reflects
2173 * more up to date window information we update our window information.
2174 */
2175 case TCPS_ESTABLISHED:
2176 case TCPS_FIN_WAIT_1:
2177 case TCPS_FIN_WAIT_2:
2178 case TCPS_CLOSE_WAIT:
2179 case TCPS_CLOSING:
2180 case TCPS_LAST_ACK:
2181 case TCPS_TIME_WAIT:
2182
2183 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2184 if (tlen == 0 && tiwin == tp->snd_wnd) {
2185 tcpstat.tcps_rcvdupack++;
2186 /*
2187 * If we have outstanding data (other than
2188 * a window probe), this is a completely
2189 * duplicate ack (ie, window info didn't
2190 * change), the ack is the biggest we've
2191 * seen and we've seen exactly our rexmt
2192 * threshhold of them, assume a packet
2193 * has been dropped and retransmit it.
2194 * Kludge snd_nxt & the congestion
2195 * window so we send only this one
2196 * packet.
2197 *
2198 * We know we're losing at the current
2199 * window size so do congestion avoidance
2200 * (set ssthresh to half the current window
2201 * and pull our congestion window back to
2202 * the new ssthresh).
2203 *
2204 * Dup acks mean that packets have left the
2205 * network (they're now cached at the receiver)
2206 * so bump cwnd by the amount in the receiver
2207 * to keep a constant cwnd packets in the
2208 * network.
2209 */
2210 if (tp->t_timer[TCPT_REXMT] == 0 ||
2211 th->th_ack != tp->snd_una)
2212 tp->t_dupacks = 0;
2213 else if (++tp->t_dupacks == tcprexmtthresh) {
2214 tcp_seq onxt = tp->snd_nxt;
2215 u_int win =
2216 min(tp->snd_wnd, tp->snd_cwnd) / 2 /
2217 tp->t_maxseg;
2218 if (tcp_do_newreno && SEQ_LT(th->th_ack,
2219 tp->snd_recover)) {
2220 /* False retransmit, should not
2221 * cut window
2222 */
2223 tp->snd_cwnd += tp->t_maxseg;
2224 tp->t_dupacks = 0;
2225 (void) tcp_output(tp);
2226 goto drop;
2227 }
2228 if (win < 2)
2229 win = 2;
2230 tp->snd_ssthresh = win * tp->t_maxseg;
2231 tp->snd_recover = tp->snd_max;
2232 tp->t_timer[TCPT_REXMT] = 0;
2233 tp->t_rtttime = 0;
2234 tp->snd_nxt = th->th_ack;
2235 tp->snd_cwnd = tp->t_maxseg;
2236 (void) tcp_output(tp);
2237 tp->snd_cwnd = tp->snd_ssthresh +
2238 tp->t_maxseg * tp->t_dupacks;
2239 if (SEQ_GT(onxt, tp->snd_nxt))
2240 tp->snd_nxt = onxt;
2241 goto drop;
2242 } else if (tp->t_dupacks > tcprexmtthresh) {
2243 tp->snd_cwnd += tp->t_maxseg;
2244 (void) tcp_output(tp);
2245 goto drop;
2246 }
2247 } else
2248 tp->t_dupacks = 0;
2249 break;
2250 }
2251 /*
2252 * If the congestion window was inflated to account
2253 * for the other side's cached packets, retract it.
2254 */
2255 if (tcp_do_newreno == 0) {
2256 if (tp->t_dupacks >= tcprexmtthresh &&
2257 tp->snd_cwnd > tp->snd_ssthresh)
2258 tp->snd_cwnd = tp->snd_ssthresh;
2259 tp->t_dupacks = 0;
2260 } else if (tp->t_dupacks >= tcprexmtthresh &&
2261 !tcp_newreno(tp, th)) {
2262 /*
2263 * Window inflation should have left us with approx.
2264 * snd_ssthresh outstanding data. But in case we
2265 * would be inclined to send a burst, better to do
2266 * it via the slow start mechanism.
2267 */
2268 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max))
2269 tp->snd_cwnd =
2270 tp->snd_max - th->th_ack + tp->t_maxseg;
2271 else
2272 tp->snd_cwnd = tp->snd_ssthresh;
2273 tp->t_dupacks = 0;
2274 }
2275
2276 if (tp->t_dupacks < tcprexmtthresh)
2277 tp->t_dupacks = 0;
2278
2279 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2280 tcpstat.tcps_rcvacktoomuch++;
2281 goto dropafterack;
2282 }
2283 /*
2284 * If we reach this point, ACK is not a duplicate,
2285 * i.e., it ACKs something we sent.
2286 */
2287 if (tp->t_flags & TF_NEEDSYN) {
2288 /*
2289 * T/TCP: Connection was half-synchronized, and our
2290 * SYN has been ACK'd (so connection is now fully
2291 * synchronized). Go to non-starred state,
2292 * increment snd_una for ACK of SYN, and check if
2293 * we can do window scaling.
2294 */
2295 tp->t_flags &= ~TF_NEEDSYN;
2296 tp->snd_una++;
2297 /* Do window scaling? */
2298 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2299 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2300 tp->snd_scale = tp->requested_s_scale;
2301 tp->rcv_scale = tp->request_r_scale;
2302 }
2303 }
2304
2305 process_ACK:
2306 acked = th->th_ack - tp->snd_una;
2307 tcpstat.tcps_rcvackpack++;
2308 tcpstat.tcps_rcvackbyte += acked;
2309
2310 /*
2311 * If we just performed our first retransmit, and the ACK
2312 * arrives within our recovery window, then it was a mistake
2313 * to do the retransmit in the first place. Recover our
2314 * original cwnd and ssthresh, and proceed to transmit where
2315 * we left off.
2316 */
2317 if (tp->t_rxtshift == 1 && tcp_now < tp->t_badrxtwin) {
2318 tp->snd_cwnd = tp->snd_cwnd_prev;
2319 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2320 tp->snd_nxt = tp->snd_max;
2321 tp->t_badrxtwin = 0; /* XXX probably not required */
2322 }
2323
2324 /*
2325 * If we have a timestamp reply, update smoothed
2326 * round trip time. If no timestamp is present but
2327 * transmit timer is running and timed sequence
2328 * number was acked, update smoothed round trip time.
2329 * Since we now have an rtt measurement, cancel the
2330 * timer backoff (cf., Phil Karn's retransmit alg.).
2331 * Recompute the initial retransmit timer.
2332 * Also makes sure we have a valid time stamp in hand
2333 */
2334 if (((to.to_flag & TOF_TS) != 0) && (to.to_tsecr != 0))
2335 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
2336 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2337 tcp_xmit_timer(tp, tp->t_rtttime);
2338
2339 /*
2340 * If all outstanding data is acked, stop retransmit
2341 * timer and remember to restart (more output or persist).
2342 * If there is more data to be acked, restart retransmit
2343 * timer, using current (possibly backed-off) value.
2344 */
2345 if (th->th_ack == tp->snd_max) {
2346 tp->t_timer[TCPT_REXMT] = 0;
2347 needoutput = 1;
2348 } else if (tp->t_timer[TCPT_PERSIST] == 0)
2349 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
2350
2351 /*
2352 * If no data (only SYN) was ACK'd,
2353 * skip rest of ACK processing.
2354 */
2355 if (acked == 0)
2356 goto step6;
2357
2358 /*
2359 * When new data is acked, open the congestion window.
2360 * If the window gives us less than ssthresh packets
2361 * in flight, open exponentially (maxseg per packet).
2362 * Otherwise open linearly: maxseg per window
2363 * (maxseg^2 / cwnd per packet).
2364 */
2365 {
2366 register u_int cw = tp->snd_cwnd;
2367 register u_int incr = tp->t_maxseg;
2368
2369 if (cw > tp->snd_ssthresh)
2370 incr = incr * incr / cw;
2371 /*
2372 * If t_dupacks != 0 here, it indicates that we are still
2373 * in NewReno fast recovery mode, so we leave the congestion
2374 * window alone.
2375 */
2376 if (tcp_do_newreno == 0 || tp->t_dupacks == 0)
2377 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale);
2378 }
2379 if (acked > so->so_snd.sb_cc) {
2380 tp->snd_wnd -= so->so_snd.sb_cc;
2381 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2382 ourfinisacked = 1;
2383 } else {
2384 sbdrop(&so->so_snd, acked);
2385 tp->snd_wnd -= acked;
2386 ourfinisacked = 0;
2387 }
2388 sowwakeup(so);
2389 /* detect una wraparound */
2390 if (SEQ_GEQ(tp->snd_una, tp->snd_recover) &&
2391 SEQ_LT(th->th_ack, tp->snd_recover))
2392 tp->snd_recover = th->th_ack;
2393 if (SEQ_GT(tp->snd_una, tp->snd_high) &&
2394 SEQ_LEQ(th->th_ack, tp->snd_high))
2395 tp->snd_high = th->th_ack - 1;
2396 tp->snd_una = th->th_ack;
2397 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2398 tp->snd_nxt = tp->snd_una;
2399
2400 switch (tp->t_state) {
2401
2402 /*
2403 * In FIN_WAIT_1 STATE in addition to the processing
2404 * for the ESTABLISHED state if our FIN is now acknowledged
2405 * then enter FIN_WAIT_2.
2406 */
2407 case TCPS_FIN_WAIT_1:
2408 if (ourfinisacked) {
2409 /*
2410 * If we can't receive any more
2411 * data, then closing user can proceed.
2412 * Starting the timer is contrary to the
2413 * specification, but if we don't get a FIN
2414 * we'll hang forever.
2415 */
2416 if (so->so_state & SS_CANTRCVMORE) {
2417 soisdisconnected(so);
2418 tp->t_timer[TCPT_2MSL] = tcp_maxidle;
2419 }
2420 add_to_time_wait(tp);
2421 tp->t_state = TCPS_FIN_WAIT_2;
2422 goto drop;
2423 }
2424 break;
2425
2426 /*
2427 * In CLOSING STATE in addition to the processing for
2428 * the ESTABLISHED state if the ACK acknowledges our FIN
2429 * then enter the TIME-WAIT state, otherwise ignore
2430 * the segment.
2431 */
2432 case TCPS_CLOSING:
2433 if (ourfinisacked) {
2434 tp->t_state = TCPS_TIME_WAIT;
2435 tcp_canceltimers(tp);
2436 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2437 if (tp->cc_recv != 0 &&
2438 tp->t_starttime < tcp_msl)
2439 tp->t_timer[TCPT_2MSL] =
2440 tp->t_rxtcur * TCPTV_TWTRUNC;
2441 else
2442 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2443 add_to_time_wait(tp);
2444 soisdisconnected(so);
2445 }
2446 break;
2447
2448 /*
2449 * In LAST_ACK, we may still be waiting for data to drain
2450 * and/or to be acked, as well as for the ack of our FIN.
2451 * If our FIN is now acknowledged, delete the TCB,
2452 * enter the closed state and return.
2453 */
2454 case TCPS_LAST_ACK:
2455 if (ourfinisacked) {
2456 tp = tcp_close(tp);
2457 goto drop;
2458 }
2459 break;
2460
2461 /*
2462 * In TIME_WAIT state the only thing that should arrive
2463 * is a retransmission of the remote FIN. Acknowledge
2464 * it and restart the finack timer.
2465 */
2466 case TCPS_TIME_WAIT:
2467 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2468 add_to_time_wait(tp);
2469 goto dropafterack;
2470 }
2471 }
2472
2473 step6:
2474 /*
2475 * Update window information.
2476 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2477 */
2478 if ((thflags & TH_ACK) &&
2479 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2480 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2481 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2482 /* keep track of pure window updates */
2483 if (tlen == 0 &&
2484 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2485 tcpstat.tcps_rcvwinupd++;
2486 tp->snd_wnd = tiwin;
2487 tp->snd_wl1 = th->th_seq;
2488 tp->snd_wl2 = th->th_ack;
2489 if (tp->snd_wnd > tp->max_sndwnd)
2490 tp->max_sndwnd = tp->snd_wnd;
2491 needoutput = 1;
2492 }
2493
2494 /*
2495 * Process segments with URG.
2496 */
2497 if ((thflags & TH_URG) && th->th_urp &&
2498 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2499 /*
2500 * This is a kludge, but if we receive and accept
2501 * random urgent pointers, we'll crash in
2502 * soreceive. It's hard to imagine someone
2503 * actually wanting to send this much urgent data.
2504 */
2505 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2506 th->th_urp = 0; /* XXX */
2507 thflags &= ~TH_URG; /* XXX */
2508 goto dodata; /* XXX */
2509 }
2510 /*
2511 * If this segment advances the known urgent pointer,
2512 * then mark the data stream. This should not happen
2513 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2514 * a FIN has been received from the remote side.
2515 * In these states we ignore the URG.
2516 *
2517 * According to RFC961 (Assigned Protocols),
2518 * the urgent pointer points to the last octet
2519 * of urgent data. We continue, however,
2520 * to consider it to indicate the first octet
2521 * of data past the urgent section as the original
2522 * spec states (in one of two places).
2523 */
2524 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2525 tp->rcv_up = th->th_seq + th->th_urp;
2526 so->so_oobmark = so->so_rcv.sb_cc +
2527 (tp->rcv_up - tp->rcv_nxt) - 1;
2528 if (so->so_oobmark == 0) {
2529 so->so_state |= SS_RCVATMARK;
2530 postevent(so, 0, EV_OOB);
2531 }
2532 sohasoutofband(so);
2533 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2534 }
2535 /*
2536 * Remove out of band data so doesn't get presented to user.
2537 * This can happen independent of advancing the URG pointer,
2538 * but if two URG's are pending at once, some out-of-band
2539 * data may creep in... ick.
2540 */
2541 if (th->th_urp <= (u_long)tlen
2542 #if SO_OOBINLINE
2543 && (so->so_options & SO_OOBINLINE) == 0
2544 #endif
2545 )
2546 tcp_pulloutofband(so, th, m,
2547 drop_hdrlen); /* hdr drop is delayed */
2548 } else
2549 /*
2550 * If no out of band data is expected,
2551 * pull receive urgent pointer along
2552 * with the receive window.
2553 */
2554 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2555 tp->rcv_up = tp->rcv_nxt;
2556 dodata: /* XXX */
2557
2558 /*
2559 * Process the segment text, merging it into the TCP sequencing queue,
2560 * and arranging for acknowledgment of receipt if necessary.
2561 * This process logically involves adjusting tp->rcv_wnd as data
2562 * is presented to the user (this happens in tcp_usrreq.c,
2563 * case PRU_RCVD). If a FIN has already been received on this
2564 * connection then we just ignore the text.
2565 */
2566 if ((tlen || (thflags&TH_FIN)) &&
2567 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2568 m_adj(m, drop_hdrlen); /* delayed header drop */
2569 /*
2570 * Insert segment which inludes th into reassembly queue of tcp with
2571 * control block tp. Return TH_FIN if reassembly now includes
2572 * a segment with FIN. This handle the common case inline (segment
2573 * is the next to be received on an established connection, and the
2574 * queue is empty), avoiding linkage into and removal from the queue
2575 * and repetition of various conversions.
2576 * Set DELACK for segments received in order, but ack immediately
2577 * when segments are out of order (so fast retransmit can work).
2578 */
2579 if (th->th_seq == tp->rcv_nxt &&
2580 LIST_EMPTY(&tp->t_segq) &&
2581 TCPS_HAVEESTABLISHED(tp->t_state)) {
2582 if (DELAY_ACK(tp) && ((tp->t_flags & TF_ACKNOW) == 0)) {
2583 tp->t_flags |= TF_DELACK;
2584 }
2585 else {
2586 tp->t_flags |= TF_ACKNOW;
2587 }
2588 tp->rcv_nxt += tlen;
2589 thflags = th->th_flags & TH_FIN;
2590 tcpstat.tcps_rcvpack++;
2591 tcpstat.tcps_rcvbyte += tlen;
2592 ND6_HINT(tp);
2593 if (sbappend(&so->so_rcv, m))
2594 sorwakeup(so);
2595 } else {
2596 thflags = tcp_reass(tp, th, &tlen, m);
2597 tp->t_flags |= TF_ACKNOW;
2598 }
2599
2600 if (tp->t_flags & TF_DELACK)
2601 {
2602 #if INET6
2603 if (isipv6) {
2604 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
2605 (((ip6->ip6_src.s6_addr16[0]) << 16) | (ip6->ip6_dst.s6_addr16[0])),
2606 th->th_seq, th->th_ack, th->th_win);
2607 }
2608 else
2609 #endif
2610 {
2611 KERNEL_DEBUG(DBG_LAYER_END, ((th->th_dport << 16) | th->th_sport),
2612 (((ip->ip_src.s_addr & 0xffff) << 16) | (ip->ip_dst.s_addr & 0xffff)),
2613 th->th_seq, th->th_ack, th->th_win);
2614 }
2615
2616 }
2617 /*
2618 * Note the amount of data that peer has sent into
2619 * our window, in order to estimate the sender's
2620 * buffer size.
2621 */
2622 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2623 } else {
2624 m_freem(m);
2625 thflags &= ~TH_FIN;
2626 }
2627
2628 /*
2629 * If FIN is received ACK the FIN and let the user know
2630 * that the connection is closing.
2631 */
2632 if (thflags & TH_FIN) {
2633 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2634 socantrcvmore(so);
2635 postevent(so, 0, EV_FIN);
2636 /*
2637 * If connection is half-synchronized
2638 * (ie NEEDSYN flag on) then delay ACK,
2639 * so it may be piggybacked when SYN is sent.
2640 * Otherwise, since we received a FIN then no
2641 * more input can be expected, send ACK now.
2642 */
2643 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2644 tp->t_flags |= TF_DELACK;
2645 }
2646 else {
2647 tp->t_flags |= TF_ACKNOW;
2648 }
2649 tp->rcv_nxt++;
2650 }
2651 switch (tp->t_state) {
2652
2653 /*
2654 * In SYN_RECEIVED and ESTABLISHED STATES
2655 * enter the CLOSE_WAIT state.
2656 */
2657 case TCPS_SYN_RECEIVED:
2658 /*FALLTHROUGH*/
2659 case TCPS_ESTABLISHED:
2660 tp->t_state = TCPS_CLOSE_WAIT;
2661 break;
2662
2663 /*
2664 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2665 * enter the CLOSING state.
2666 */
2667 case TCPS_FIN_WAIT_1:
2668 tp->t_state = TCPS_CLOSING;
2669 break;
2670
2671 /*
2672 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2673 * starting the time-wait timer, turning off the other
2674 * standard timers.
2675 */
2676 case TCPS_FIN_WAIT_2:
2677 tp->t_state = TCPS_TIME_WAIT;
2678 tcp_canceltimers(tp);
2679 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2680 if (tp->cc_recv != 0 &&
2681 tp->t_starttime < tcp_msl) {
2682 tp->t_timer[TCPT_2MSL] =
2683 tp->t_rxtcur * TCPTV_TWTRUNC;
2684 /* For transaction client, force ACK now. */
2685 tp->t_flags |= TF_ACKNOW;
2686 }
2687 else
2688 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2689
2690 add_to_time_wait(tp);
2691 soisdisconnected(so);
2692 break;
2693
2694 /*
2695 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2696 */
2697 case TCPS_TIME_WAIT:
2698 tp->t_timer[TCPT_2MSL] = 2 * tcp_msl;
2699 add_to_time_wait(tp);
2700 break;
2701 }
2702 }
2703 #if TCPDEBUG
2704 if (so->so_options & SO_DEBUG)
2705 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2706 &tcp_savetcp, 0);
2707 #endif
2708
2709 /*
2710 * Return any desired output.
2711 */
2712 if (needoutput || (tp->t_flags & TF_ACKNOW))
2713 (void) tcp_output(tp);
2714 tcp_unlock(so, 1, 0);
2715 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2716 return;
2717
2718 dropafterack:
2719 /*
2720 * Generate an ACK dropping incoming segment if it occupies
2721 * sequence space, where the ACK reflects our state.
2722 *
2723 * We can now skip the test for the RST flag since all
2724 * paths to this code happen after packets containing
2725 * RST have been dropped.
2726 *
2727 * In the SYN-RECEIVED state, don't send an ACK unless the
2728 * segment we received passes the SYN-RECEIVED ACK test.
2729 * If it fails send a RST. This breaks the loop in the
2730 * "LAND" DoS attack, and also prevents an ACK storm
2731 * between two listening ports that have been sent forged
2732 * SYN segments, each with the source address of the other.
2733 */
2734 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2735 (SEQ_GT(tp->snd_una, th->th_ack) ||
2736 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2737 rstreason = BANDLIM_RST_OPENPORT;
2738 goto dropwithreset;
2739 }
2740 #if TCPDEBUG
2741 if (so->so_options & SO_DEBUG)
2742 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2743 &tcp_savetcp, 0);
2744 #endif
2745 m_freem(m);
2746 tp->t_flags |= TF_ACKNOW;
2747 (void) tcp_output(tp);
2748 tcp_unlock(so, 1, 0);
2749 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2750 return;
2751 dropwithresetnosock:
2752 nosock = 1;
2753 dropwithreset:
2754 /*
2755 * Generate a RST, dropping incoming segment.
2756 * Make ACK acceptable to originator of segment.
2757 * Don't bother to respond if destination was broadcast/multicast.
2758 */
2759 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2760 goto drop;
2761 #if INET6
2762 if (isipv6) {
2763 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2764 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2765 goto drop;
2766 } else
2767 #endif /* INET6 */
2768 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2769 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2770 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2771 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2772 goto drop;
2773 /* IPv6 anycast check is done at tcp6_input() */
2774
2775 /*
2776 * Perform bandwidth limiting.
2777 */
2778 #if ICMP_BANDLIM
2779 if (badport_bandlim(rstreason) < 0)
2780 goto drop;
2781 #endif
2782
2783 #if TCPDEBUG
2784 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2785 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2786 &tcp_savetcp, 0);
2787 #endif
2788 if (thflags & TH_ACK)
2789 /* mtod() below is safe as long as hdr dropping is delayed */
2790 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2791 TH_RST);
2792 else {
2793 if (thflags & TH_SYN)
2794 tlen++;
2795 /* mtod() below is safe as long as hdr dropping is delayed */
2796 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2797 (tcp_seq)0, TH_RST|TH_ACK);
2798 }
2799 /* destroy temporarily created socket */
2800 if (dropsocket) {
2801 (void) soabort(so);
2802 tcp_unlock(so, 1, 0);
2803 }
2804 else
2805 if ((inp != NULL) && (nosock == 0))
2806 tcp_unlock(so, 1, 0);
2807 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2808 return;
2809 dropnosock:
2810 nosock = 1;
2811 drop:
2812 /*
2813 * Drop space held by incoming segment and return.
2814 */
2815 #if TCPDEBUG
2816 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2817 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2818 &tcp_savetcp, 0);
2819 #endif
2820 m_freem(m);
2821 /* destroy temporarily created socket */
2822 if (dropsocket) {
2823 (void) soabort(so);
2824 tcp_unlock(so, 1, 0);
2825 }
2826 else
2827 if (nosock == 0)
2828 tcp_unlock(so, 1, 0);
2829 KERNEL_DEBUG(DBG_FNC_TCP_INPUT | DBG_FUNC_END,0,0,0,0,0);
2830 return;
2831 }
2832
2833 static void
2834 tcp_dooptions(tp, cp, cnt, th, to)
2835 struct tcpcb *tp;
2836 u_char *cp;
2837 int cnt;
2838 struct tcphdr *th;
2839 struct tcpopt *to;
2840 {
2841 u_short mss = 0;
2842 int opt, optlen;
2843
2844 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2845 opt = cp[0];
2846 if (opt == TCPOPT_EOL)
2847 break;
2848 if (opt == TCPOPT_NOP)
2849 optlen = 1;
2850 else {
2851 if (cnt < 2)
2852 break;
2853 optlen = cp[1];
2854 if (optlen < 2 || optlen > cnt)
2855 break;
2856 }
2857 switch (opt) {
2858
2859 default:
2860 continue;
2861
2862 case TCPOPT_MAXSEG:
2863 if (optlen != TCPOLEN_MAXSEG)
2864 continue;
2865 if (!(th->th_flags & TH_SYN))
2866 continue;
2867 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
2868 NTOHS(mss);
2869 break;
2870
2871 case TCPOPT_WINDOW:
2872 if (optlen != TCPOLEN_WINDOW)
2873 continue;
2874 if (!(th->th_flags & TH_SYN))
2875 continue;
2876 tp->t_flags |= TF_RCVD_SCALE;
2877 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2878 break;
2879
2880 case TCPOPT_TIMESTAMP:
2881 if (optlen != TCPOLEN_TIMESTAMP)
2882 continue;
2883 to->to_flag |= TOF_TS;
2884 bcopy((char *)cp + 2,
2885 (char *)&to->to_tsval, sizeof(to->to_tsval));
2886 NTOHL(to->to_tsval);
2887 bcopy((char *)cp + 6,
2888 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2889 NTOHL(to->to_tsecr);
2890
2891 /*
2892 * A timestamp received in a SYN makes
2893 * it ok to send timestamp requests and replies.
2894 */
2895 if (th->th_flags & TH_SYN) {
2896 tp->t_flags |= TF_RCVD_TSTMP;
2897 tp->ts_recent = to->to_tsval;
2898 tp->ts_recent_age = tcp_now;
2899 }
2900 break;
2901 case TCPOPT_CC:
2902 if (optlen != TCPOLEN_CC)
2903 continue;
2904 to->to_flag |= TOF_CC;
2905 bcopy((char *)cp + 2,
2906 (char *)&to->to_cc, sizeof(to->to_cc));
2907 NTOHL(to->to_cc);
2908 /*
2909 * A CC or CC.new option received in a SYN makes
2910 * it ok to send CC in subsequent segments.
2911 */
2912 if (th->th_flags & TH_SYN)
2913 tp->t_flags |= TF_RCVD_CC;
2914 break;
2915 case TCPOPT_CCNEW:
2916 if (optlen != TCPOLEN_CC)
2917 continue;
2918 if (!(th->th_flags & TH_SYN))
2919 continue;
2920 to->to_flag |= TOF_CCNEW;
2921 bcopy((char *)cp + 2,
2922 (char *)&to->to_cc, sizeof(to->to_cc));
2923 NTOHL(to->to_cc);
2924 /*
2925 * A CC or CC.new option received in a SYN makes
2926 * it ok to send CC in subsequent segments.
2927 */
2928 tp->t_flags |= TF_RCVD_CC;
2929 break;
2930 case TCPOPT_CCECHO:
2931 if (optlen != TCPOLEN_CC)
2932 continue;
2933 if (!(th->th_flags & TH_SYN))
2934 continue;
2935 to->to_flag |= TOF_CCECHO;
2936 bcopy((char *)cp + 2,
2937 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
2938 NTOHL(to->to_ccecho);
2939 break;
2940 }
2941 }
2942 if (th->th_flags & TH_SYN)
2943 tcp_mss(tp, mss); /* sets t_maxseg */
2944 }
2945
2946 /*
2947 * Pull out of band byte out of a segment so
2948 * it doesn't appear in the user's data queue.
2949 * It is still reflected in the segment length for
2950 * sequencing purposes.
2951 */
2952 static void
2953 tcp_pulloutofband(so, th, m, off)
2954 struct socket *so;
2955 struct tcphdr *th;
2956 register struct mbuf *m;
2957 int off; /* delayed to be droped hdrlen */
2958 {
2959 int cnt = off + th->th_urp - 1;
2960
2961 while (cnt >= 0) {
2962 if (m->m_len > cnt) {
2963 char *cp = mtod(m, caddr_t) + cnt;
2964 struct tcpcb *tp = sototcpcb(so);
2965
2966 tp->t_iobc = *cp;
2967 tp->t_oobflags |= TCPOOB_HAVEDATA;
2968 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2969 m->m_len--;
2970 if (m->m_flags & M_PKTHDR)
2971 m->m_pkthdr.len--;
2972 return;
2973 }
2974 cnt -= m->m_len;
2975 m = m->m_next;
2976 if (m == 0)
2977 break;
2978 }
2979 panic("tcp_pulloutofband");
2980 }
2981
2982 /*
2983 * Collect new round-trip time estimate
2984 * and update averages and current timeout.
2985 */
2986 static void
2987 tcp_xmit_timer(tp, rtt)
2988 register struct tcpcb *tp;
2989 int rtt;
2990 {
2991 register int delta;
2992
2993 tcpstat.tcps_rttupdated++;
2994 tp->t_rttupdated++;
2995 if (tp->t_srtt != 0) {
2996 /*
2997 * srtt is stored as fixed point with 5 bits after the
2998 * binary point (i.e., scaled by 8). The following magic
2999 * is equivalent to the smoothing algorithm in rfc793 with
3000 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3001 * point). Adjust rtt to origin 0.
3002 */
3003 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3004 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3005
3006 if ((tp->t_srtt += delta) <= 0)
3007 tp->t_srtt = 1;
3008
3009 /*
3010 * We accumulate a smoothed rtt variance (actually, a
3011 * smoothed mean difference), then set the retransmit
3012 * timer to smoothed rtt + 4 times the smoothed variance.
3013 * rttvar is stored as fixed point with 4 bits after the
3014 * binary point (scaled by 16). The following is
3015 * equivalent to rfc793 smoothing with an alpha of .75
3016 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3017 * rfc793's wired-in beta.
3018 */
3019 if (delta < 0)
3020 delta = -delta;
3021 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3022 if ((tp->t_rttvar += delta) <= 0)
3023 tp->t_rttvar = 1;
3024 } else {
3025 /*
3026 * No rtt measurement yet - use the unsmoothed rtt.
3027 * Set the variance to half the rtt (so our first
3028 * retransmit happens at 3*rtt).
3029 */
3030 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3031 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3032 }
3033 tp->t_rtttime = 0;
3034 tp->t_rxtshift = 0;
3035
3036 /*
3037 * the retransmit should happen at rtt + 4 * rttvar.
3038 * Because of the way we do the smoothing, srtt and rttvar
3039 * will each average +1/2 tick of bias. When we compute
3040 * the retransmit timer, we want 1/2 tick of rounding and
3041 * 1 extra tick because of +-1/2 tick uncertainty in the
3042 * firing of the timer. The bias will give us exactly the
3043 * 1.5 tick we need. But, because the bias is
3044 * statistical, we have to test that we don't drop below
3045 * the minimum feasible timer (which is 2 ticks).
3046 */
3047 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3048 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3049
3050 /*
3051 * We received an ack for a packet that wasn't retransmitted;
3052 * it is probably safe to discard any error indications we've
3053 * received recently. This isn't quite right, but close enough
3054 * for now (a route might have failed after we sent a segment,
3055 * and the return path might not be symmetrical).
3056 */
3057 tp->t_softerror = 0;
3058 }
3059
3060 /*
3061 * Determine a reasonable value for maxseg size.
3062 * If the route is known, check route for mtu.
3063 * If none, use an mss that can be handled on the outgoing
3064 * interface without forcing IP to fragment; if bigger than
3065 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3066 * to utilize large mbufs. If no route is found, route has no mtu,
3067 * or the destination isn't local, use a default, hopefully conservative
3068 * size (usually 512 or the default IP max size, but no more than the mtu
3069 * of the interface), as we can't discover anything about intervening
3070 * gateways or networks. We also initialize the congestion/slow start
3071 * window to be a single segment if the destination isn't local.
3072 * While looking at the routing entry, we also initialize other path-dependent
3073 * parameters from pre-set or cached values in the routing entry.
3074 *
3075 * Also take into account the space needed for options that we
3076 * send regularly. Make maxseg shorter by that amount to assure
3077 * that we can send maxseg amount of data even when the options
3078 * are present. Store the upper limit of the length of options plus
3079 * data in maxopd.
3080 *
3081 * NOTE that this routine is only called when we process an incoming
3082 * segment, for outgoing segments only tcp_mssopt is called.
3083 *
3084 * In case of T/TCP, we call this routine during implicit connection
3085 * setup as well (offer = -1), to initialize maxseg from the cached
3086 * MSS of our peer.
3087 */
3088 void
3089 tcp_mss(tp, offer)
3090 struct tcpcb *tp;
3091 int offer;
3092 {
3093 register struct rtentry *rt;
3094 struct ifnet *ifp;
3095 register int rtt, mss;
3096 u_long bufsize;
3097 struct inpcb *inp;
3098 struct socket *so;
3099 struct rmxp_tao *taop;
3100 int origoffer = offer;
3101 #if INET6
3102 int isipv6;
3103 int min_protoh;
3104 #endif
3105
3106 inp = tp->t_inpcb;
3107 #if INET6
3108 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3109 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr)
3110 : sizeof (struct tcpiphdr);
3111 #else
3112 #define min_protoh (sizeof (struct tcpiphdr))
3113 #endif
3114 #if INET6
3115 if (isipv6)
3116 rt = tcp_rtlookup6(inp);
3117 else
3118 #endif /* INET6 */
3119 rt = tcp_rtlookup(inp);
3120 if (rt == NULL) {
3121 tp->t_maxopd = tp->t_maxseg =
3122 #if INET6
3123 isipv6 ? tcp_v6mssdflt :
3124 #endif /* INET6 */
3125 tcp_mssdflt;
3126 return;
3127 }
3128 ifp = rt->rt_ifp;
3129 /*
3130 * Slower link window correction:
3131 * If a value is specificied for slowlink_wsize use it for PPP links
3132 * believed to be on a serial modem (speed <128Kbps). Excludes 9600bps as
3133 * it is the default value adversized by pseudo-devices over ppp.
3134 */
3135 if (ifp->if_type == IFT_PPP && slowlink_wsize > 0 &&
3136 ifp->if_baudrate > 9600 && ifp->if_baudrate <= 128000) {
3137 tp->t_flags |= TF_SLOWLINK;
3138 }
3139 so = inp->inp_socket;
3140
3141 taop = rmx_taop(rt->rt_rmx);
3142 /*
3143 * Offer == -1 means that we didn't receive SYN yet,
3144 * use cached value in that case;
3145 */
3146 if (offer == -1)
3147 offer = taop->tao_mssopt;
3148 /*
3149 * Offer == 0 means that there was no MSS on the SYN segment,
3150 * in this case we use tcp_mssdflt.
3151 */
3152 if (offer == 0)
3153 offer =
3154 #if INET6
3155 isipv6 ? tcp_v6mssdflt :
3156 #endif /* INET6 */
3157 tcp_mssdflt;
3158 else {
3159 /*
3160 * Prevent DoS attack with too small MSS. Round up
3161 * to at least minmss.
3162 */
3163 offer = max(offer, tcp_minmss);
3164 /*
3165 * Sanity check: make sure that maxopd will be large
3166 * enough to allow some data on segments even is the
3167 * all the option space is used (40bytes). Otherwise
3168 * funny things may happen in tcp_output.
3169 */
3170 offer = max(offer, 64);
3171 }
3172 taop->tao_mssopt = offer;
3173
3174 /*
3175 * While we're here, check if there's an initial rtt
3176 * or rttvar. Convert from the route-table units
3177 * to scaled multiples of the slow timeout timer.
3178 */
3179 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3180 /*
3181 * XXX the lock bit for RTT indicates that the value
3182 * is also a minimum value; this is subject to time.
3183 */
3184 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3185 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
3186 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
3187 tcpstat.tcps_usedrtt++;
3188 if (rt->rt_rmx.rmx_rttvar) {
3189 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3190 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
3191 tcpstat.tcps_usedrttvar++;
3192 } else {
3193 /* default variation is +- 1 rtt */
3194 tp->t_rttvar =
3195 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3196 }
3197 TCPT_RANGESET(tp->t_rxtcur,
3198 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3199 tp->t_rttmin, TCPTV_REXMTMAX);
3200 }
3201 /*
3202 * if there's an mtu associated with the route, use it
3203 * else, use the link mtu.
3204 */
3205 if (rt->rt_rmx.rmx_mtu)
3206 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3207 else
3208 {
3209 mss =
3210 #if INET6
3211 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu :
3212 #endif
3213 ifp->if_mtu
3214 #if INET6
3215 )
3216 #endif
3217 - min_protoh;
3218 #if INET6
3219 if (isipv6) {
3220 if (!in6_localaddr(&inp->in6p_faddr))
3221 mss = min(mss, tcp_v6mssdflt);
3222 } else
3223 #endif /* INET6 */
3224 if (!in_localaddr(inp->inp_faddr))
3225 mss = min(mss, tcp_mssdflt);
3226 }
3227 mss = min(mss, offer);
3228 /*
3229 * maxopd stores the maximum length of data AND options
3230 * in a segment; maxseg is the amount of data in a normal
3231 * segment. We need to store this value (maxopd) apart
3232 * from maxseg, because now every segment carries options
3233 * and thus we normally have somewhat less data in segments.
3234 */
3235 tp->t_maxopd = mss;
3236
3237 /*
3238 * In case of T/TCP, origoffer==-1 indicates, that no segments
3239 * were received yet. In this case we just guess, otherwise
3240 * we do the same as before T/TCP.
3241 */
3242 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3243 (origoffer == -1 ||
3244 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3245 mss -= TCPOLEN_TSTAMP_APPA;
3246 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
3247 (origoffer == -1 ||
3248 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3249 mss -= TCPOLEN_CC_APPA;
3250
3251 /*
3252 * If there's a pipesize (ie loopback), change the socket
3253 * buffer to that size only if it's bigger than the current
3254 * sockbuf size. Make the socket buffers an integral
3255 * number of mss units; if the mss is larger than
3256 * the socket buffer, decrease the mss.
3257 */
3258 #if RTV_SPIPE
3259 bufsize = rt->rt_rmx.rmx_sendpipe;
3260 if (bufsize < so->so_snd.sb_hiwat)
3261 #endif
3262 bufsize = so->so_snd.sb_hiwat;
3263 if (bufsize < mss)
3264 mss = bufsize;
3265 else {
3266 bufsize = roundup(bufsize, mss);
3267 if (bufsize > sb_max)
3268 bufsize = sb_max;
3269 (void)sbreserve(&so->so_snd, bufsize);
3270 }
3271 tp->t_maxseg = mss;
3272
3273 #if RTV_RPIPE
3274 bufsize = rt->rt_rmx.rmx_recvpipe;
3275 if (bufsize < so->so_rcv.sb_hiwat)
3276 #endif
3277 bufsize = so->so_rcv.sb_hiwat;
3278 if (bufsize > mss) {
3279 bufsize = roundup(bufsize, mss);
3280 if (bufsize > sb_max)
3281 bufsize = sb_max;
3282 (void)sbreserve(&so->so_rcv, bufsize);
3283 }
3284
3285 /*
3286 * Set the slow-start flight size depending on whether this
3287 * is a local network or not.
3288 */
3289 if (
3290 #if INET6
3291 (isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3292 (!isipv6 &&
3293 #endif
3294 in_localaddr(inp->inp_faddr)
3295 #if INET6
3296 )
3297 #endif
3298 )
3299 tp->snd_cwnd = mss * ss_fltsz_local;
3300 else
3301 tp->snd_cwnd = mss * ss_fltsz;
3302
3303 if (rt->rt_rmx.rmx_ssthresh) {
3304 /*
3305 * There's some sort of gateway or interface
3306 * buffer limit on the path. Use this to set
3307 * the slow start threshhold, but set the
3308 * threshold to no less than 2*mss.
3309 */
3310 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3311 tcpstat.tcps_usedssthresh++;
3312 }
3313 }
3314
3315 /*
3316 * Determine the MSS option to send on an outgoing SYN.
3317 */
3318 int
3319 tcp_mssopt(tp)
3320 struct tcpcb *tp;
3321 {
3322 struct rtentry *rt;
3323 #if INET6
3324 int isipv6;
3325 int min_protoh;
3326 #endif
3327
3328 #if INET6
3329 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3330 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr)
3331 : sizeof (struct tcpiphdr);
3332 #else
3333 #define min_protoh (sizeof (struct tcpiphdr))
3334 #endif
3335 #if INET6
3336 if (isipv6)
3337 rt = tcp_rtlookup6(tp->t_inpcb);
3338 else
3339 #endif /* INET6 */
3340 rt = tcp_rtlookup(tp->t_inpcb);
3341 if (rt == NULL)
3342 return
3343 #if INET6
3344 isipv6 ? tcp_v6mssdflt :
3345 #endif /* INET6 */
3346 tcp_mssdflt;
3347 /*
3348 * Slower link window correction:
3349 * If a value is specificied for slowlink_wsize use it for PPP links
3350 * believed to be on a serial modem (speed <128Kbps). Excludes 9600bps as
3351 * it is the default value adversized by pseudo-devices over ppp.
3352 */
3353 if (rt->rt_ifp->if_type == IFT_PPP && slowlink_wsize > 0 &&
3354 rt->rt_ifp->if_baudrate > 9600 && rt->rt_ifp->if_baudrate <= 128000) {
3355 tp->t_flags |= TF_SLOWLINK;
3356 }
3357
3358 return rt->rt_ifp->if_mtu - min_protoh;
3359 }
3360
3361
3362 /*
3363 * Checks for partial ack. If partial ack arrives, force the retransmission
3364 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return
3365 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to
3366 * be started again. If the ack advances at least to tp->snd_recover, return 0.
3367 */
3368 static int
3369 tcp_newreno(tp, th)
3370 struct tcpcb *tp;
3371 struct tcphdr *th;
3372 {
3373 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
3374 tcp_seq onxt = tp->snd_nxt;
3375 u_long ocwnd = tp->snd_cwnd;
3376 #ifdef __APPLE__
3377 tp->t_timer[TCPT_REXMT] = 0;
3378 #else
3379 callout_stop(tp->tt_rexmt);
3380 #endif
3381 tp->t_rtttime = 0;
3382 tp->snd_nxt = th->th_ack;
3383 /*
3384 * Set snd_cwnd to one segment beyond acknowledged offset
3385 * (tp->snd_una has not yet been updated when this function
3386 * is called)
3387 */
3388 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3389 tp->t_flags |= TF_ACKNOW;
3390 (void) tcp_output(tp);
3391 tp->snd_cwnd = ocwnd;
3392 if (SEQ_GT(onxt, tp->snd_nxt))
3393 tp->snd_nxt = onxt;
3394 /*
3395 * Partial window deflation. Relies on fact that tp->snd_una
3396 * not updated yet.
3397 */
3398 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg);
3399 return (1);
3400 }
3401 return (0);
3402 }
3403
3404 /*
3405 * Drop a random TCP connection that hasn't been serviced yet and
3406 * is eligible for discard. There is a one in qlen chance that
3407 * we will return a null, saying that there are no dropable
3408 * requests. In this case, the protocol specific code should drop
3409 * the new request. This insures fairness.
3410 *
3411 * The listening TCP socket "head" must be locked
3412 */
3413 static int
3414 tcpdropdropablreq(struct socket *head)
3415 {
3416 struct socket *so;
3417 unsigned int i, j, qlen;
3418 static int rnd;
3419 static struct timeval old_runtime;
3420 static unsigned int cur_cnt, old_cnt;
3421 struct timeval tv;
3422 struct inpcb *inp = NULL;
3423
3424 microtime(&tv);
3425 if ((i = (tv.tv_sec - old_runtime.tv_sec)) != 0) {
3426 old_runtime = tv;
3427 old_cnt = cur_cnt / i;
3428 cur_cnt = 0;
3429 }
3430
3431 so = TAILQ_FIRST(&head->so_incomp);
3432 if (!so)
3433 return 0;
3434
3435 qlen = head->so_incqlen;
3436 if (++cur_cnt > qlen || old_cnt > qlen) {
3437 rnd = (314159 * rnd + 66329) & 0xffff;
3438 j = ((qlen + 1) * rnd) >> 16;
3439
3440 while (j-- && so)
3441 so = TAILQ_NEXT(so, so_list);
3442 }
3443 /* Find a connection that is not already closing */
3444 while (so) {
3445 inp = (struct inpcb *)so->so_pcb;
3446
3447 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) != WNT_STOPUSING)
3448 break;
3449
3450 so = TAILQ_NEXT(so, so_list);
3451 }
3452 if (!so)
3453 return 0;
3454
3455 /* Let's remove this connection from the incomplete list */
3456 tcp_lock(so, 1, 0);
3457
3458 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
3459 tcp_unlock(so, 1, 0);
3460 return 0;
3461 }
3462 sototcpcb(so)->t_flags |= TF_LQ_OVERFLOW;
3463 head->so_incqlen--;
3464 head->so_qlen--;
3465 so->so_head = NULL;
3466 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3467 so->so_usecount--; /* No more held by so_head */
3468
3469 tcp_drop(sototcpcb(so), ETIMEDOUT);
3470
3471 tcp_unlock(so, 1, 0);
3472
3473 return 1;
3474
3475 }
3476
3477
mirror of XNU