1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
12 *
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
17 *
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21 *
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
24 *
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
27 * are met:
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * SUCH DAMAGE.
48 */
49
50 #include <sys/cdefs.h>
51 #include "opt_inet.h"
52 #include "opt_inet6.h"
53 #include "opt_ipsec.h"
54 #include "opt_rss.h"
55
56 #include <sys/param.h>
57 #include <sys/arb.h>
58 #include <sys/kernel.h>
59 #ifdef TCP_HHOOK
60 #include <sys/hhook.h>
61 #endif
62 #include <sys/malloc.h>
63 #include <sys/mbuf.h>
64 #include <sys/proc.h> /* for proc0 declaration */
65 #include <sys/protosw.h>
66 #include <sys/qmath.h>
67 #include <sys/sdt.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
74 #include <sys/stats.h>
75
76 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
77
78 #include <vm/uma.h>
79
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
84 #include <net/vnet.h>
85
86 #define TCPSTATES /* for logging */
87
88 #include <netinet/in.h>
89 #include <netinet/in_kdtrace.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_rss.h>
92 #include <netinet/in_systm.h>
93 #include <netinet/ip.h>
94 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
95 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
96 #include <netinet/ip_var.h>
97 #include <netinet/ip_options.h>
98 #include <netinet/ip6.h>
99 #include <netinet/icmp6.h>
100 #include <netinet6/in6_pcb.h>
101 #include <netinet6/in6_rss.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #include <netinet6/nd6.h>
105 #include <netinet/tcp.h>
106 #include <netinet/tcp_fsm.h>
107 #include <netinet/tcp_seq.h>
108 #include <netinet/tcp_timer.h>
109 #include <netinet/tcp_var.h>
110 #include <netinet/tcp_log_buf.h>
111 #include <netinet6/tcp6_var.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/cc/cc.h>
114 #include <netinet/tcp_fastopen.h>
115 #ifdef TCPPCAP
116 #include <netinet/tcp_pcap.h>
117 #endif
118 #include <netinet/tcp_syncache.h>
119 #ifdef TCP_OFFLOAD
120 #include <netinet/tcp_offload.h>
121 #endif
122 #include <netinet/tcp_ecn.h>
123 #include <netinet/udp.h>
124
125 #include <netipsec/ipsec_support.h>
126
127 #include <machine/in_cksum.h>
128
129 #include <security/mac/mac_framework.h>
130
131 const int tcprexmtthresh = 3;
132
133 VNET_DEFINE(int, tcp_log_in_vain) = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
135 &VNET_NAME(tcp_log_in_vain), 0,
136 "Log all incoming TCP segments to closed ports");
137
138 VNET_DEFINE(int, blackhole) = 0;
139 #define V_blackhole VNET(blackhole)
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(blackhole), 0,
142 "Do not send RST on segments to closed ports");
143
144 VNET_DEFINE(bool, blackhole_local) = false;
145 #define V_blackhole_local VNET(blackhole_local)
146 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
147 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
148 "Enforce net.inet.tcp.blackhole for locally originated packets");
149
150 VNET_DEFINE(int, tcp_delack_enabled) = 1;
151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
152 &VNET_NAME(tcp_delack_enabled), 0,
153 "Delay ACK to try and piggyback it onto a data packet");
154
155 VNET_DEFINE(int, drop_synfin) = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
157 &VNET_NAME(drop_synfin), 0,
158 "Drop TCP packets with SYN+FIN set");
159
160 VNET_DEFINE(int, tcp_do_prr) = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
162 &VNET_NAME(tcp_do_prr), 1,
163 "Enable Proportional Rate Reduction per RFC 6937");
164
165 VNET_DEFINE(int, tcp_do_newcwv) = 0;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
167 &VNET_NAME(tcp_do_newcwv), 0,
168 "Enable New Congestion Window Validation per RFC7661");
169
170 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
172 &VNET_NAME(tcp_do_rfc3042), 0,
173 "Enable RFC 3042 (Limited Transmit)");
174
175 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
177 &VNET_NAME(tcp_do_rfc3390), 0,
178 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
179
180 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
182 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
183 "Slow-start flight size (initial congestion window) in number of segments");
184
185 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc3465), 0,
188 "Enable RFC 3465 (Appropriate Byte Counting)");
189
190 VNET_DEFINE(int, tcp_abc_l_var) = 2;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_abc_l_var), 2,
193 "Cap the max cwnd increment during slow-start to this number of segments");
194
195 VNET_DEFINE(int, tcp_insecure_syn) = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
199
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_rst), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
204
205 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
206 #define V_tcp_recvspace VNET(tcp_recvspace)
207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
209
210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
212 &VNET_NAME(tcp_do_autorcvbuf), 0,
213 "Enable automatic receive buffer sizing");
214
215 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_max), 0,
218 "Max size of automatic receive buffer");
219
220 VNET_DEFINE(struct inpcbinfo, tcbinfo);
221
222 /*
223 * TCP statistics are stored in an array of counter(9)s, which size matches
224 * size of struct tcpstat. TCP running connection count is a regular array.
225 */
226 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
227 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
228 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
229 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
230 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
231 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
232 "TCP connection counts by TCP state");
233
234 /*
235 * Kernel module interface for updating tcpstat. The first argument is an index
236 * into tcpstat treated as an array.
237 */
238 void
kmod_tcpstat_add(int statnum,int val)239 kmod_tcpstat_add(int statnum, int val)
240 {
241
242 counter_u64_add(VNET(tcpstat)[statnum], val);
243 }
244
245 /*
246 * Make sure that we only start a SACK loss recovery when
247 * receiving a duplicate ACK with a SACK block, and also
248 * complete SACK loss recovery in case the other end
249 * reneges.
250 */
251 static bool inline
tcp_is_sack_recovery(struct tcpcb * tp,struct tcpopt * to)252 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
253 {
254 return ((tp->t_flags & TF_SACK_PERMIT) &&
255 ((to->to_flags & TOF_SACK) ||
256 (!TAILQ_EMPTY(&tp->snd_holes))));
257 }
258
259 #ifdef TCP_HHOOK
260 /*
261 * Wrapper for the TCP established input helper hook.
262 */
263 void
hhook_run_tcp_est_in(struct tcpcb * tp,struct tcphdr * th,struct tcpopt * to)264 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
265 {
266 struct tcp_hhook_data hhook_data;
267
268 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
269 hhook_data.tp = tp;
270 hhook_data.th = th;
271 hhook_data.to = to;
272
273 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
274 &tp->t_osd);
275 }
276 }
277 #endif
278
279 /*
280 * CC wrapper hook functions
281 */
282 void
cc_ack_received(struct tcpcb * tp,struct tcphdr * th,uint16_t nsegs,uint16_t type)283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
284 uint16_t type)
285 {
286 #ifdef STATS
287 int32_t gput;
288 #endif
289
290 INP_WLOCK_ASSERT(tptoinpcb(tp));
291
292 tp->t_ccv.nsegs = nsegs;
293 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th);
294 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
295 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
296 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
297 tp->t_ccv.flags |= CCF_CWND_LIMITED;
298 else
299 tp->t_ccv.flags &= ~CCF_CWND_LIMITED;
300
301 if (type == CC_ACK) {
302 #ifdef STATS
303 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
304 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
305 if (!IN_RECOVERY(tp->t_flags))
306 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
307 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs));
308 if ((tp->t_flags & TF_GPUTINPROG) &&
309 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
310 /*
311 * Compute goodput in bits per millisecond.
312 */
313 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
314 max(1, tcp_ts_getticks() - tp->gput_ts);
315 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
316 gput);
317 /*
318 * XXXLAS: This is a temporary hack, and should be
319 * chained off VOI_TCP_GPUT when stats(9) grows an API
320 * to deal with chained VOIs.
321 */
322 if (tp->t_stats_gput_prev > 0)
323 stats_voi_update_abs_s32(tp->t_stats,
324 VOI_TCP_GPUT_ND,
325 ((gput - tp->t_stats_gput_prev) * 100) /
326 tp->t_stats_gput_prev);
327 tp->t_flags &= ~TF_GPUTINPROG;
328 tp->t_stats_gput_prev = gput;
329 }
330 #endif /* STATS */
331 if (tp->snd_cwnd > tp->snd_ssthresh) {
332 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack;
333 if (tp->t_bytes_acked >= tp->snd_cwnd) {
334 tp->t_bytes_acked -= tp->snd_cwnd;
335 tp->t_ccv.flags |= CCF_ABC_SENTAWND;
336 }
337 } else {
338 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND;
339 tp->t_bytes_acked = 0;
340 }
341 }
342
343 if (CC_ALGO(tp)->ack_received != NULL) {
344 /* XXXLAS: Find a way to live without this */
345 tp->t_ccv.curack = th->th_ack;
346 CC_ALGO(tp)->ack_received(&tp->t_ccv, type);
347 }
348 #ifdef STATS
349 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
350 #endif
351 }
352
353 void
cc_conn_init(struct tcpcb * tp)354 cc_conn_init(struct tcpcb *tp)
355 {
356 struct hc_metrics_lite metrics;
357 struct inpcb *inp = tptoinpcb(tp);
358 u_int maxseg;
359 int rtt;
360
361 INP_WLOCK_ASSERT(inp);
362
363 tcp_hc_get(&inp->inp_inc, &metrics);
364 maxseg = tcp_maxseg(tp);
365
366 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
367 tp->t_srtt = rtt;
368 TCPSTAT_INC(tcps_usedrtt);
369 if (metrics.rmx_rttvar) {
370 tp->t_rttvar = metrics.rmx_rttvar;
371 TCPSTAT_INC(tcps_usedrttvar);
372 } else {
373 /* default variation is +- 1 rtt */
374 tp->t_rttvar =
375 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
376 }
377 TCPT_RANGESET(tp->t_rxtcur,
378 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
379 tp->t_rttmin, TCPTV_REXMTMAX);
380 }
381 if (metrics.rmx_ssthresh) {
382 /*
383 * There's some sort of gateway or interface
384 * buffer limit on the path. Use this to set
385 * the slow start threshold, but set the
386 * threshold to no less than 2*mss.
387 */
388 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
389 TCPSTAT_INC(tcps_usedssthresh);
390 }
391
392 /*
393 * Set the initial slow-start flight size.
394 *
395 * If a SYN or SYN/ACK was lost and retransmitted, we have to
396 * reduce the initial CWND to one segment as congestion is likely
397 * requiring us to be cautious.
398 */
399 if (tp->snd_cwnd == 1)
400 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
401 else
402 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
403
404 if (CC_ALGO(tp)->conn_init != NULL)
405 CC_ALGO(tp)->conn_init(&tp->t_ccv);
406 }
407
408 void inline
cc_cong_signal(struct tcpcb * tp,struct tcphdr * th,uint32_t type)409 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
410 {
411 INP_WLOCK_ASSERT(tptoinpcb(tp));
412
413 #ifdef STATS
414 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
415 #endif
416
417 switch(type) {
418 case CC_NDUPACK:
419 if (!IN_FASTRECOVERY(tp->t_flags)) {
420 tp->snd_recover = tp->snd_max;
421 if (tp->t_flags2 & TF2_ECN_PERMIT)
422 tp->t_flags2 |= TF2_ECN_SND_CWR;
423 }
424 break;
425 case CC_ECN:
426 if (!IN_CONGRECOVERY(tp->t_flags) ||
427 /*
428 * Allow ECN reaction on ACK to CWR, if
429 * that data segment was also CE marked.
430 */
431 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
432 EXIT_CONGRECOVERY(tp->t_flags);
433 TCPSTAT_INC(tcps_ecn_rcwnd);
434 tp->snd_recover = tp->snd_max + 1;
435 if (tp->t_flags2 & TF2_ECN_PERMIT)
436 tp->t_flags2 |= TF2_ECN_SND_CWR;
437 }
438 break;
439 case CC_RTO:
440 tp->t_dupacks = 0;
441 tp->t_bytes_acked = 0;
442 if ((tp->t_rxtshift > 1) ||
443 !((tp->t_flags & TF_SACK_PERMIT) &&
444 (!TAILQ_EMPTY(&tp->snd_holes))))
445 EXIT_RECOVERY(tp->t_flags);
446 if (tp->t_flags2 & TF2_ECN_PERMIT)
447 tp->t_flags2 |= TF2_ECN_SND_CWR;
448 break;
449 case CC_RTO_ERR:
450 TCPSTAT_INC(tcps_sndrexmitbad);
451 /* RTO was unnecessary, so reset everything. */
452 tp->snd_cwnd = tp->snd_cwnd_prev;
453 tp->snd_ssthresh = tp->snd_ssthresh_prev;
454 tp->snd_recover = tp->snd_recover_prev;
455 if (tp->t_flags & TF_WASFRECOVERY)
456 ENTER_FASTRECOVERY(tp->t_flags);
457 if (tp->t_flags & TF_WASCRECOVERY)
458 ENTER_CONGRECOVERY(tp->t_flags);
459 tp->snd_nxt = tp->snd_max;
460 tp->t_flags &= ~TF_PREVVALID;
461 tp->t_badrxtwin = 0;
462 break;
463 }
464 if (SEQ_LT(tp->snd_fack, tp->snd_una) ||
465 SEQ_GT(tp->snd_fack, tp->snd_max)) {
466 tp->snd_fack = tp->snd_una;
467 }
468
469 if (CC_ALGO(tp)->cong_signal != NULL) {
470 if (th != NULL)
471 tp->t_ccv.curack = th->th_ack;
472 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type);
473 }
474 }
475
476 void inline
cc_post_recovery(struct tcpcb * tp,struct tcphdr * th)477 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
478 {
479 INP_WLOCK_ASSERT(tptoinpcb(tp));
480
481 if (CC_ALGO(tp)->post_recovery != NULL) {
482 if (SEQ_LT(tp->snd_fack, th->th_ack) ||
483 SEQ_GT(tp->snd_fack, tp->snd_max)) {
484 tp->snd_fack = th->th_ack;
485 }
486 tp->t_ccv.curack = th->th_ack;
487 CC_ALGO(tp)->post_recovery(&tp->t_ccv);
488 }
489 EXIT_RECOVERY(tp->t_flags);
490
491 tp->t_bytes_acked = 0;
492 tp->sackhint.delivered_data = 0;
493 tp->sackhint.prr_delivered = 0;
494 tp->sackhint.prr_out = 0;
495 }
496
497 /*
498 * Indicate whether this ack should be delayed. We can delay the ack if
499 * following conditions are met:
500 * - There is no delayed ack timer in progress.
501 * - Our last ack wasn't a 0-sized window. We never want to delay
502 * the ack that opens up a 0-sized window.
503 * - LRO wasn't used for this segment. We make sure by checking that the
504 * segment size is not larger than the MSS.
505 */
506 #define DELAY_ACK(tp, tlen) \
507 ((!tcp_timer_active(tp, TT_DELACK) && \
508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
509 (tlen <= tp->t_maxseg) && \
510 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
511
512 void inline
cc_ecnpkt_handler_flags(struct tcpcb * tp,uint16_t flags,uint8_t iptos)513 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
514 {
515 INP_WLOCK_ASSERT(tptoinpcb(tp));
516
517 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
518 switch (iptos & IPTOS_ECN_MASK) {
519 case IPTOS_ECN_CE:
520 tp->t_ccv.flags |= CCF_IPHDR_CE;
521 break;
522 case IPTOS_ECN_ECT0:
523 /* FALLTHROUGH */
524 case IPTOS_ECN_ECT1:
525 /* FALLTHROUGH */
526 case IPTOS_ECN_NOTECT:
527 tp->t_ccv.flags &= ~CCF_IPHDR_CE;
528 break;
529 }
530
531 if (flags & TH_CWR)
532 tp->t_ccv.flags |= CCF_TCPHDR_CWR;
533 else
534 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR;
535
536 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv);
537
538 if (tp->t_ccv.flags & CCF_ACKNOW) {
539 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
540 tp->t_flags |= TF_ACKNOW;
541 }
542 }
543 }
544
545 void inline
cc_ecnpkt_handler(struct tcpcb * tp,struct tcphdr * th,uint8_t iptos)546 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
547 {
548 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
549 }
550
551 /*
552 * TCP input handling is split into multiple parts:
553 * tcp6_input is a thin wrapper around tcp_input for the extended
554 * ip6_protox[] call format in ip6_input
555 * tcp_input handles primary segment validation, inpcb lookup and
556 * SYN processing on listen sockets
557 * tcp_do_segment processes the ACK and text of the segment for
558 * establishing, established and closing connections
559 */
560 #ifdef INET6
561 int
tcp6_input_with_port(struct mbuf ** mp,int * offp,int proto,uint16_t port)562 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
563 {
564 struct mbuf *m;
565 struct in6_ifaddr *ia6;
566 struct ip6_hdr *ip6;
567
568 m = *mp;
569 if (m->m_len < *offp + sizeof(struct tcphdr)) {
570 m = m_pullup(m, *offp + sizeof(struct tcphdr));
571 if (m == NULL) {
572 *mp = m;
573 TCPSTAT_INC(tcps_rcvshort);
574 return (IPPROTO_DONE);
575 }
576 }
577
578 /*
579 * draft-itojun-ipv6-tcp-to-anycast
580 * better place to put this in?
581 */
582 ip6 = mtod(m, struct ip6_hdr *);
583 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
584 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
585 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
586 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
587 *mp = NULL;
588 return (IPPROTO_DONE);
589 }
590
591 *mp = m;
592 return (tcp_input_with_port(mp, offp, proto, port));
593 }
594
595 int
tcp6_input(struct mbuf ** mp,int * offp,int proto)596 tcp6_input(struct mbuf **mp, int *offp, int proto)
597 {
598
599 return(tcp6_input_with_port(mp, offp, proto, 0));
600 }
601 #endif /* INET6 */
602
603 int
tcp_input_with_port(struct mbuf ** mp,int * offp,int proto,uint16_t port)604 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
605 {
606 struct mbuf *m = *mp;
607 struct tcphdr *th = NULL;
608 struct ip *ip = NULL;
609 struct inpcb *inp = NULL;
610 struct tcpcb *tp = NULL;
611 struct socket *so = NULL;
612 u_char *optp = NULL;
613 int off0;
614 int optlen = 0;
615 #ifdef INET
616 int len;
617 uint8_t ipttl;
618 #endif
619 int tlen = 0, off;
620 int drop_hdrlen;
621 int thflags;
622 int rstreason = 0; /* For badport_bandlim accounting purposes */
623 int lookupflag;
624 uint8_t iptos;
625 struct m_tag *fwd_tag = NULL;
626 #ifdef INET6
627 struct ip6_hdr *ip6 = NULL;
628 int isipv6;
629 #else
630 const void *ip6 = NULL;
631 #endif /* INET6 */
632 struct tcpopt to; /* options in this segment */
633 char *s = NULL; /* address and port logging */
634
635 NET_EPOCH_ASSERT();
636
637 #ifdef INET6
638 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
639 #endif
640
641 off0 = *offp;
642 m = *mp;
643 *mp = NULL;
644 to.to_flags = 0;
645 TCPSTAT_INC(tcps_rcvtotal);
646
647 m->m_pkthdr.tcp_tun_port = port;
648 #ifdef INET6
649 if (isipv6) {
650 ip6 = mtod(m, struct ip6_hdr *);
651 th = (struct tcphdr *)((caddr_t)ip6 + off0);
652 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
653 if (port)
654 goto skip6_csum;
655 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
656 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
657 th->th_sum = m->m_pkthdr.csum_data;
658 else
659 th->th_sum = in6_cksum_pseudo(ip6, tlen,
660 IPPROTO_TCP, m->m_pkthdr.csum_data);
661 th->th_sum ^= 0xffff;
662 } else
663 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
664 if (th->th_sum) {
665 TCPSTAT_INC(tcps_rcvbadsum);
666 goto drop;
667 }
668 skip6_csum:
669 /*
670 * Be proactive about unspecified IPv6 address in source.
671 * As we use all-zero to indicate unbounded/unconnected pcb,
672 * unspecified IPv6 address can be used to confuse us.
673 *
674 * Note that packets with unspecified IPv6 destination is
675 * already dropped in ip6_input.
676 */
677 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst),
678 ("%s: unspecified destination v6 address", __func__));
679 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
680 IP6STAT_INC(ip6s_badscope); /* XXX */
681 goto drop;
682 }
683 iptos = IPV6_TRAFFIC_CLASS(ip6);
684 }
685 #endif
686 #if defined(INET) && defined(INET6)
687 else
688 #endif
689 #ifdef INET
690 {
691 /*
692 * Get IP and TCP header together in first mbuf.
693 * Note: IP leaves IP header in first mbuf.
694 */
695 if (off0 > sizeof (struct ip)) {
696 ip_stripoptions(m);
697 off0 = sizeof(struct ip);
698 }
699 if (m->m_len < sizeof (struct tcpiphdr)) {
700 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
701 == NULL) {
702 TCPSTAT_INC(tcps_rcvshort);
703 return (IPPROTO_DONE);
704 }
705 }
706 ip = mtod(m, struct ip *);
707 th = (struct tcphdr *)((caddr_t)ip + off0);
708 tlen = ntohs(ip->ip_len) - off0;
709
710 iptos = ip->ip_tos;
711 if (port)
712 goto skip_csum;
713 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
714 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
715 th->th_sum = m->m_pkthdr.csum_data;
716 else
717 th->th_sum = in_pseudo(ip->ip_src.s_addr,
718 ip->ip_dst.s_addr,
719 htonl(m->m_pkthdr.csum_data + tlen +
720 IPPROTO_TCP));
721 th->th_sum ^= 0xffff;
722 } else {
723 struct ipovly *ipov = (struct ipovly *)ip;
724
725 /*
726 * Checksum extended TCP header and data.
727 */
728 len = off0 + tlen;
729 ipttl = ip->ip_ttl;
730 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
731 ipov->ih_len = htons(tlen);
732 th->th_sum = in_cksum(m, len);
733 /* Reset length for SDT probes. */
734 ip->ip_len = htons(len);
735 /* Reset TOS bits */
736 ip->ip_tos = iptos;
737 /* Re-initialization for later version check */
738 ip->ip_ttl = ipttl;
739 ip->ip_v = IPVERSION;
740 ip->ip_hl = off0 >> 2;
741 }
742 skip_csum:
743 if (th->th_sum && (port == 0)) {
744 TCPSTAT_INC(tcps_rcvbadsum);
745 goto drop;
746 }
747 KASSERT(ip->ip_dst.s_addr != INADDR_ANY,
748 ("%s: unspecified destination v4 address", __func__));
749 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) {
750 IPSTAT_INC(ips_badaddr);
751 goto drop;
752 }
753 }
754 #endif /* INET */
755
756 /*
757 * Check that TCP offset makes sense,
758 * pull out TCP options and adjust length. XXX
759 */
760 off = th->th_off << 2;
761 if (off < sizeof (struct tcphdr) || off > tlen) {
762 TCPSTAT_INC(tcps_rcvbadoff);
763 goto drop;
764 }
765 tlen -= off; /* tlen is used instead of ti->ti_len */
766 if (off > sizeof (struct tcphdr)) {
767 #ifdef INET6
768 if (isipv6) {
769 if (m->m_len < off0 + off) {
770 m = m_pullup(m, off0 + off);
771 if (m == NULL) {
772 TCPSTAT_INC(tcps_rcvshort);
773 return (IPPROTO_DONE);
774 }
775 }
776 ip6 = mtod(m, struct ip6_hdr *);
777 th = (struct tcphdr *)((caddr_t)ip6 + off0);
778 }
779 #endif
780 #if defined(INET) && defined(INET6)
781 else
782 #endif
783 #ifdef INET
784 {
785 if (m->m_len < sizeof(struct ip) + off) {
786 if ((m = m_pullup(m, sizeof (struct ip) + off))
787 == NULL) {
788 TCPSTAT_INC(tcps_rcvshort);
789 return (IPPROTO_DONE);
790 }
791 ip = mtod(m, struct ip *);
792 th = (struct tcphdr *)((caddr_t)ip + off0);
793 }
794 }
795 #endif
796 optlen = off - sizeof (struct tcphdr);
797 optp = (u_char *)(th + 1);
798 }
799 thflags = tcp_get_flags(th);
800
801 /*
802 * Convert TCP protocol specific fields to host format.
803 */
804 tcp_fields_to_host(th);
805
806 /*
807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
808 */
809 drop_hdrlen = off0 + off;
810
811 /*
812 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
813 */
814 if (
815 #ifdef INET6
816 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
817 #ifdef INET
818 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
819 #endif
820 #endif
821 #if defined(INET) && !defined(INET6)
822 (m->m_flags & M_IP_NEXTHOP)
823 #endif
824 )
825 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
826
827 /*
828 * For initial SYN packets we don't need write lock on matching
829 * PCB, be it a listening one or a synchronized one. The packet
830 * shall not modify its state.
831 */
832 lookupflag = INPLOOKUP_WILDCARD |
833 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
834 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
835 findpcb:
836 tp = NULL;
837 #ifdef INET6
838 if (isipv6 && fwd_tag != NULL) {
839 struct sockaddr_in6 *next_hop6;
840
841 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
842 /*
843 * Transparently forwarded. Pretend to be the destination.
844 * Already got one like this?
845 */
846 inp = in6_pcblookup_mbuf(&V_tcbinfo,
847 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
848 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m);
849 if (!inp) {
850 /*
851 * It's new. Try to find the ambushing socket.
852 * Because we've rewritten the destination address,
853 * any hardware-generated hash is ignored.
854 */
855 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
856 th->th_sport, &next_hop6->sin6_addr,
857 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
858 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
859 }
860 } else if (isipv6) {
861 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
862 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
863 m->m_pkthdr.rcvif, m);
864 }
865 #endif /* INET6 */
866 #if defined(INET6) && defined(INET)
867 else
868 #endif
869 #ifdef INET
870 if (fwd_tag != NULL) {
871 struct sockaddr_in *next_hop;
872
873 next_hop = (struct sockaddr_in *)(fwd_tag+1);
874 /*
875 * Transparently forwarded. Pretend to be the destination.
876 * already got one like this?
877 */
878 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
879 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD,
880 m->m_pkthdr.rcvif, m);
881 if (!inp) {
882 /*
883 * It's new. Try to find the ambushing socket.
884 * Because we've rewritten the destination address,
885 * any hardware-generated hash is ignored.
886 */
887 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
888 th->th_sport, next_hop->sin_addr,
889 next_hop->sin_port ? ntohs(next_hop->sin_port) :
890 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
891 }
892 } else
893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
894 th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
895 m->m_pkthdr.rcvif, m);
896 #endif /* INET */
897
898 /*
899 * If the INPCB does not exist then all data in the incoming
900 * segment is discarded and an appropriate RST is sent back.
901 * XXX MRT Send RST using which routing table?
902 */
903 if (inp == NULL) {
904 if (rstreason != 0) {
905 /* We came here after second (safety) lookup. */
906 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
907 goto dropwithreset;
908 }
909 /*
910 * Log communication attempts to ports that are not
911 * in use.
912 */
913 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
914 V_tcp_log_in_vain == 2) {
915 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
916 log(LOG_INFO, "%s; %s: Connection attempt "
917 "to closed port\n", s, __func__);
918 }
919 rstreason = BANDLIM_RST_CLOSEDPORT;
920 goto dropwithreset;
921 }
922 INP_LOCK_ASSERT(inp);
923
924 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
925 !SOLISTENING(inp->inp_socket)) {
926 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
927 inp->inp_flowid = m->m_pkthdr.flowid;
928 inp->inp_flowtype = M_HASHTYPE_GET(m);
929 #ifdef RSS
930 } else {
931 /* assign flowid by software RSS hash */
932 #ifdef INET6
933 if (isipv6) {
934 rss_proto_software_hash_v6(&inp->in6p_faddr,
935 &inp->in6p_laddr,
936 inp->inp_fport,
937 inp->inp_lport,
938 IPPROTO_TCP,
939 &inp->inp_flowid,
940 &inp->inp_flowtype);
941 } else
942 #endif /* INET6 */
943 {
944 rss_proto_software_hash_v4(inp->inp_faddr,
945 inp->inp_laddr,
946 inp->inp_fport,
947 inp->inp_lport,
948 IPPROTO_TCP,
949 &inp->inp_flowid,
950 &inp->inp_flowtype);
951 }
952 #endif /* RSS */
953 }
954 }
955 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
956 #ifdef INET6
957 if (isipv6 && IPSEC_ENABLED(ipv6) &&
958 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
959 goto dropunlock;
960 }
961 #ifdef INET
962 else
963 #endif
964 #endif /* INET6 */
965 #ifdef INET
966 if (IPSEC_ENABLED(ipv4) &&
967 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
968 goto dropunlock;
969 }
970 #endif /* INET */
971 #endif /* IPSEC */
972
973 /*
974 * Check the minimum TTL for socket.
975 */
976 if (inp->inp_ip_minttl != 0) {
977 #ifdef INET6
978 if (isipv6) {
979 if (inp->inp_ip_minttl > ip6->ip6_hlim)
980 goto dropunlock;
981 } else
982 #endif
983 if (inp->inp_ip_minttl > ip->ip_ttl)
984 goto dropunlock;
985 }
986
987 tp = intotcpcb(inp);
988 switch (tp->t_state) {
989 case TCPS_TIME_WAIT:
990 /*
991 * A previous connection in TIMEWAIT state is supposed to catch
992 * stray or duplicate segments arriving late. If this segment
993 * was a legitimate new connection attempt, the old INPCB gets
994 * removed and we can try again to find a listening socket.
995 */
996 tcp_dooptions(&to, optp, optlen,
997 (thflags & TH_SYN) ? TO_SYN : 0);
998 /*
999 * tcp_twcheck unlocks the inp always, and frees the m if fails.
1000 */
1001 if (tcp_twcheck(inp, &to, th, m, tlen))
1002 goto findpcb;
1003 return (IPPROTO_DONE);
1004 case TCPS_CLOSED:
1005 /*
1006 * The TCPCB may no longer exist if the connection is winding
1007 * down or it is in the CLOSED state. Either way we drop the
1008 * segment and send an appropriate response.
1009 */
1010 rstreason = BANDLIM_RST_CLOSEDPORT;
1011 goto dropwithreset;
1012 }
1013
1014 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1015 rstreason = BANDLIM_RST_CLOSEDPORT;
1016 goto dropwithreset;
1017 }
1018
1019 #ifdef TCP_OFFLOAD
1020 if (tp->t_flags & TF_TOE) {
1021 tcp_offload_input(tp, m);
1022 m = NULL; /* consumed by the TOE driver */
1023 goto dropunlock;
1024 }
1025 #endif
1026
1027 #ifdef MAC
1028 if (mac_inpcb_check_deliver(inp, m))
1029 goto dropunlock;
1030 #endif
1031 so = inp->inp_socket;
1032 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1033 /*
1034 * When the socket is accepting connections (the INPCB is in LISTEN
1035 * state) we look into the SYN cache if this is a new connection
1036 * attempt or the completion of a previous one.
1037 */
1038 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1039 ("%s: so accepting but tp %p not listening", __func__, tp));
1040 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1041 struct in_conninfo inc;
1042
1043 bzero(&inc, sizeof(inc));
1044 #ifdef INET6
1045 if (isipv6) {
1046 inc.inc_flags |= INC_ISIPV6;
1047 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1048 inc.inc_flags |= INC_IPV6MINMTU;
1049 inc.inc6_faddr = ip6->ip6_src;
1050 inc.inc6_laddr = ip6->ip6_dst;
1051 } else
1052 #endif
1053 {
1054 inc.inc_faddr = ip->ip_src;
1055 inc.inc_laddr = ip->ip_dst;
1056 }
1057 inc.inc_fport = th->th_sport;
1058 inc.inc_lport = th->th_dport;
1059 inc.inc_fibnum = so->so_fibnum;
1060
1061 /*
1062 * Check for an existing connection attempt in syncache if
1063 * the flag is only ACK. A successful lookup creates a new
1064 * socket appended to the listen queue in SYN_RECEIVED state.
1065 */
1066 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1067 /*
1068 * Parse the TCP options here because
1069 * syncookies need access to the reflected
1070 * timestamp.
1071 */
1072 tcp_dooptions(&to, optp, optlen, 0);
1073 /*
1074 * NB: syncache_expand() doesn't unlock inp.
1075 */
1076 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1077 if (rstreason < 0) {
1078 /*
1079 * A failing TCP MD5 signature comparison
1080 * must result in the segment being dropped
1081 * and must not produce any response back
1082 * to the sender.
1083 */
1084 goto dropunlock;
1085 } else if (rstreason == 0) {
1086 /*
1087 * No syncache entry, or ACK was not for our
1088 * SYN/ACK. Do our protection against double
1089 * ACK. If peer sent us 2 ACKs, then for the
1090 * first one syncache_expand() successfully
1091 * converted syncache entry into a socket,
1092 * while we were waiting on the inpcb lock. We
1093 * don't want to sent RST for the second ACK,
1094 * so we perform second lookup without wildcard
1095 * match, hoping to find the new socket. If
1096 * the ACK is stray indeed, rstreason would
1097 * hint the above code that the lookup was a
1098 * second attempt.
1099 *
1100 * NB: syncache did its own logging
1101 * of the failure cause.
1102 */
1103 INP_WUNLOCK(inp);
1104 rstreason = BANDLIM_RST_OPENPORT;
1105 lookupflag &= ~INPLOOKUP_WILDCARD;
1106 goto findpcb;
1107 }
1108 tfo_socket_result:
1109 if (so == NULL) {
1110 /*
1111 * We completed the 3-way handshake
1112 * but could not allocate a socket
1113 * either due to memory shortage,
1114 * listen queue length limits or
1115 * global socket limits. Send RST
1116 * or wait and have the remote end
1117 * retransmit the ACK for another
1118 * try.
1119 */
1120 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1121 log(LOG_DEBUG, "%s; %s: Listen socket: "
1122 "Socket allocation failed due to "
1123 "limits or memory shortage, %s\n",
1124 s, __func__,
1125 V_tcp_sc_rst_sock_fail ?
1126 "sending RST" : "try again");
1127 if (V_tcp_sc_rst_sock_fail) {
1128 rstreason = BANDLIM_UNLIMITED;
1129 goto dropwithreset;
1130 } else
1131 goto dropunlock;
1132 }
1133 /*
1134 * Socket is created in state SYN_RECEIVED.
1135 * Unlock the listen socket, lock the newly
1136 * created socket and update the tp variable.
1137 * If we came here via jump to tfo_socket_result,
1138 * then listening socket is read-locked.
1139 */
1140 INP_UNLOCK(inp); /* listen socket */
1141 inp = sotoinpcb(so);
1142 /*
1143 * New connection inpcb is already locked by
1144 * syncache_expand().
1145 */
1146 INP_WLOCK_ASSERT(inp);
1147 tp = intotcpcb(inp);
1148 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1149 ("%s: ", __func__));
1150 /*
1151 * Process the segment and the data it
1152 * contains. tcp_do_segment() consumes
1153 * the mbuf chain and unlocks the inpcb.
1154 */
1155 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1156 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen,
1157 tlen, iptos);
1158 return (IPPROTO_DONE);
1159 }
1160 /*
1161 * Segment flag validation for new connection attempts:
1162 *
1163 * Our (SYN|ACK) response was rejected.
1164 * Check with syncache and remove entry to prevent
1165 * retransmits.
1166 *
1167 * NB: syncache_chkrst does its own logging of failure
1168 * causes.
1169 */
1170 if (thflags & TH_RST) {
1171 syncache_chkrst(&inc, th, m, port);
1172 goto dropunlock;
1173 }
1174 /*
1175 * We can't do anything without SYN.
1176 */
1177 if ((thflags & TH_SYN) == 0) {
1178 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1179 log(LOG_DEBUG, "%s; %s: Listen socket: "
1180 "SYN is missing, segment ignored\n",
1181 s, __func__);
1182 TCPSTAT_INC(tcps_badsyn);
1183 goto dropunlock;
1184 }
1185 /*
1186 * (SYN|ACK) is bogus on a listen socket.
1187 */
1188 if (thflags & TH_ACK) {
1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1190 log(LOG_DEBUG, "%s; %s: Listen socket: "
1191 "SYN|ACK invalid, segment rejected\n",
1192 s, __func__);
1193 syncache_badack(&inc, port); /* XXX: Not needed! */
1194 TCPSTAT_INC(tcps_badsyn);
1195 rstreason = BANDLIM_RST_OPENPORT;
1196 goto dropwithreset;
1197 }
1198 /*
1199 * If the drop_synfin option is enabled, drop all
1200 * segments with both the SYN and FIN bits set.
1201 * This prevents e.g. nmap from identifying the
1202 * TCP/IP stack.
1203 * XXX: Poor reasoning. nmap has other methods
1204 * and is constantly refining its stack detection
1205 * strategies.
1206 * XXX: This is a violation of the TCP specification
1207 * and was used by RFC1644.
1208 */
1209 if ((thflags & TH_FIN) && V_drop_synfin) {
1210 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1211 log(LOG_DEBUG, "%s; %s: Listen socket: "
1212 "SYN|FIN segment ignored (based on "
1213 "sysctl setting)\n", s, __func__);
1214 TCPSTAT_INC(tcps_badsyn);
1215 goto dropunlock;
1216 }
1217 /*
1218 * Segment's flags are (SYN) or (SYN|FIN).
1219 *
1220 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1221 * as they do not affect the state of the TCP FSM.
1222 * The data pointed to by TH_URG and th_urp is ignored.
1223 */
1224 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1225 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1226 KASSERT(thflags & (TH_SYN),
1227 ("%s: Listen socket: TH_SYN not set", __func__));
1228 INP_RLOCK_ASSERT(inp);
1229 #ifdef INET6
1230 /*
1231 * If deprecated address is forbidden,
1232 * we do not accept SYN to deprecated interface
1233 * address to prevent any new inbound connection from
1234 * getting established.
1235 * When we do not accept SYN, we send a TCP RST,
1236 * with deprecated source address (instead of dropping
1237 * it). We compromise it as it is much better for peer
1238 * to send a RST, and RST will be the final packet
1239 * for the exchange.
1240 *
1241 * If we do not forbid deprecated addresses, we accept
1242 * the SYN packet. RFC2462 does not suggest dropping
1243 * SYN in this case.
1244 * If we decipher RFC2462 5.5.4, it says like this:
1245 * 1. use of deprecated addr with existing
1246 * communication is okay - "SHOULD continue to be
1247 * used"
1248 * 2. use of it with new communication:
1249 * (2a) "SHOULD NOT be used if alternate address
1250 * with sufficient scope is available"
1251 * (2b) nothing mentioned otherwise.
1252 * Here we fall into (2b) case as we have no choice in
1253 * our source address selection - we must obey the peer.
1254 *
1255 * The wording in RFC2462 is confusing, and there are
1256 * multiple description text for deprecated address
1257 * handling - worse, they are not exactly the same.
1258 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1259 */
1260 if (isipv6 && !V_ip6_use_deprecated) {
1261 struct in6_ifaddr *ia6;
1262
1263 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1264 if (ia6 != NULL &&
1265 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1266 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1267 log(LOG_DEBUG, "%s; %s: Listen socket: "
1268 "Connection attempt to deprecated "
1269 "IPv6 address rejected\n",
1270 s, __func__);
1271 rstreason = BANDLIM_RST_OPENPORT;
1272 goto dropwithreset;
1273 }
1274 }
1275 #endif /* INET6 */
1276 /*
1277 * Basic sanity checks on incoming SYN requests:
1278 * Don't respond if the destination is a link layer
1279 * broadcast according to RFC1122 4.2.3.10, p. 104.
1280 * If it is from this socket it must be forged.
1281 * Don't respond if the source or destination is a
1282 * global or subnet broad- or multicast address.
1283 * Note that it is quite possible to receive unicast
1284 * link-layer packets with a broadcast IP address. Use
1285 * in_broadcast() to find them.
1286 */
1287 if (m->m_flags & (M_BCAST|M_MCAST)) {
1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1289 log(LOG_DEBUG, "%s; %s: Listen socket: "
1290 "Connection attempt from broad- or multicast "
1291 "link layer address ignored\n", s, __func__);
1292 goto dropunlock;
1293 }
1294 #ifdef INET6
1295 if (isipv6) {
1296 if (th->th_dport == th->th_sport &&
1297 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 "Connection attempt to/from self "
1301 "ignored\n", s, __func__);
1302 goto dropunlock;
1303 }
1304 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1305 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1307 log(LOG_DEBUG, "%s; %s: Listen socket: "
1308 "Connection attempt from/to multicast "
1309 "address ignored\n", s, __func__);
1310 goto dropunlock;
1311 }
1312 }
1313 #endif
1314 #if defined(INET) && defined(INET6)
1315 else
1316 #endif
1317 #ifdef INET
1318 {
1319 if (th->th_dport == th->th_sport &&
1320 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1322 log(LOG_DEBUG, "%s; %s: Listen socket: "
1323 "Connection attempt from/to self "
1324 "ignored\n", s, __func__);
1325 goto dropunlock;
1326 }
1327 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1328 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1329 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1330 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332 log(LOG_DEBUG, "%s; %s: Listen socket: "
1333 "Connection attempt from/to broad- "
1334 "or multicast address ignored\n",
1335 s, __func__);
1336 goto dropunlock;
1337 }
1338 }
1339 #endif
1340 /*
1341 * SYN appears to be valid. Create compressed TCP state
1342 * for syncache.
1343 */
1344 TCP_PROBE3(debug__input, tp, th, m);
1345 tcp_dooptions(&to, optp, optlen, TO_SYN);
1346 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1347 iptos, port)) != NULL)
1348 goto tfo_socket_result;
1349
1350 /*
1351 * Entry added to syncache and mbuf consumed.
1352 * Only the listen socket is unlocked by syncache_add().
1353 */
1354 return (IPPROTO_DONE);
1355 } else if (tp->t_state == TCPS_LISTEN) {
1356 /*
1357 * When a listen socket is torn down the SO_ACCEPTCONN
1358 * flag is removed first while connections are drained
1359 * from the accept queue in a unlock/lock cycle of the
1360 * ACCEPT_LOCK, opening a race condition allowing a SYN
1361 * attempt go through unhandled.
1362 */
1363 goto dropunlock;
1364 }
1365 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1366 if (tp->t_flags & TF_SIGNATURE) {
1367 tcp_dooptions(&to, optp, optlen, thflags);
1368 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1369 TCPSTAT_INC(tcps_sig_err_nosigopt);
1370 goto dropunlock;
1371 }
1372 if (!TCPMD5_ENABLED() ||
1373 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1374 goto dropunlock;
1375 }
1376 #endif
1377 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1378
1379 /*
1380 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1381 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1382 * the inpcb, and unlocks pcbinfo.
1383 *
1384 * XXXGL: in case of a pure SYN arriving on existing connection
1385 * TCP stacks won't need to modify the PCB, they would either drop
1386 * the segment silently, or send a challenge ACK. However, we try
1387 * to upgrade the lock, because calling convention for stacks is
1388 * write-lock on PCB. If upgrade fails, drop the SYN.
1389 */
1390 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1391 goto dropunlock;
1392
1393 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos);
1394 return (IPPROTO_DONE);
1395
1396 dropwithreset:
1397 /*
1398 * When blackholing do not respond with a RST but
1399 * completely ignore the segment and drop it.
1400 */
1401 if (((rstreason == BANDLIM_RST_OPENPORT && V_blackhole == 3) ||
1402 (rstreason == BANDLIM_RST_CLOSEDPORT &&
1403 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) &&
1404 (V_blackhole_local || (
1405 #ifdef INET6
1406 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
1407 #endif
1408 #ifdef INET
1409 !in_localip(ip->ip_src)
1410 #else
1411 true
1412 #endif
1413 )))
1414 goto dropunlock;
1415 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1416 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1417 m = NULL; /* mbuf chain got consumed. */
1418
1419 dropunlock:
1420 if (m != NULL)
1421 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1422
1423 if (inp != NULL)
1424 INP_UNLOCK(inp);
1425
1426 drop:
1427 if (s != NULL)
1428 free(s, M_TCPLOG);
1429 if (m != NULL)
1430 m_freem(m);
1431 return (IPPROTO_DONE);
1432 }
1433
1434 /*
1435 * Automatic sizing of receive socket buffer. Often the send
1436 * buffer size is not optimally adjusted to the actual network
1437 * conditions at hand (delay bandwidth product). Setting the
1438 * buffer size too small limits throughput on links with high
1439 * bandwidth and high delay (eg. trans-continental/oceanic links).
1440 *
1441 * On the receive side the socket buffer memory is only rarely
1442 * used to any significant extent. This allows us to be much
1443 * more aggressive in scaling the receive socket buffer. For
1444 * the case that the buffer space is actually used to a large
1445 * extent and we run out of kernel memory we can simply drop
1446 * the new segments; TCP on the sender will just retransmit it
1447 * later. Setting the buffer size too big may only consume too
1448 * much kernel memory if the application doesn't read() from
1449 * the socket or packet loss or reordering makes use of the
1450 * reassembly queue.
1451 *
1452 * The criteria to step up the receive buffer one notch are:
1453 * 1. Application has not set receive buffer size with
1454 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1455 * 2. the number of bytes received during 1/2 of an sRTT
1456 * is at least 3/8 of the current socket buffer size.
1457 * 3. receive buffer size has not hit maximal automatic size;
1458 *
1459 * If all of the criteria are met we increaset the socket buffer
1460 * by a 1/2 (bounded by the max). This allows us to keep ahead
1461 * of slow-start but also makes it so our peer never gets limited
1462 * by our rwnd which we then open up causing a burst.
1463 *
1464 * This algorithm does two steps per RTT at most and only if
1465 * we receive a bulk stream w/o packet losses or reorderings.
1466 * Shrinking the buffer during idle times is not necessary as
1467 * it doesn't consume any memory when idle.
1468 *
1469 * TODO: Only step up if the application is actually serving
1470 * the buffer to better manage the socket buffer resources.
1471 */
1472 int
tcp_autorcvbuf(struct mbuf * m,struct tcphdr * th,struct socket * so,struct tcpcb * tp,int tlen)1473 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1474 struct tcpcb *tp, int tlen)
1475 {
1476 int newsize = 0;
1477
1478 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1479 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1480 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1481 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1482 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1483 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1484 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1485 }
1486 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1487
1488 /* Start over with next RTT. */
1489 tp->rfbuf_ts = 0;
1490 tp->rfbuf_cnt = 0;
1491 } else {
1492 tp->rfbuf_cnt += tlen; /* add up */
1493 }
1494 return (newsize);
1495 }
1496
1497 int
tcp_input(struct mbuf ** mp,int * offp,int proto)1498 tcp_input(struct mbuf **mp, int *offp, int proto)
1499 {
1500 return(tcp_input_with_port(mp, offp, proto, 0));
1501 }
1502
1503 static void
tcp_handle_wakeup(struct tcpcb * tp)1504 tcp_handle_wakeup(struct tcpcb *tp)
1505 {
1506
1507 INP_WLOCK_ASSERT(tptoinpcb(tp));
1508
1509 if (tp->t_flags & TF_WAKESOR) {
1510 struct socket *so = tptosocket(tp);
1511
1512 tp->t_flags &= ~TF_WAKESOR;
1513 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1514 sorwakeup_locked(so);
1515 }
1516 }
1517
1518 void
tcp_do_segment(struct tcpcb * tp,struct mbuf * m,struct tcphdr * th,int drop_hdrlen,int tlen,uint8_t iptos)1519 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th,
1520 int drop_hdrlen, int tlen, uint8_t iptos)
1521 {
1522 uint16_t thflags;
1523 int acked, ourfinisacked, needoutput = 0;
1524 sackstatus_t sack_changed;
1525 int rstreason, todrop, win, incforsyn = 0;
1526 uint32_t tiwin;
1527 uint16_t nsegs;
1528 char *s;
1529 struct inpcb *inp = tptoinpcb(tp);
1530 struct socket *so = tptosocket(tp);
1531 struct in_conninfo *inc = &inp->inp_inc;
1532 struct mbuf *mfree;
1533 struct tcpopt to;
1534 int tfo_syn;
1535 u_int maxseg = 0;
1536
1537 thflags = tcp_get_flags(th);
1538 tp->sackhint.last_sack_ack = 0;
1539 sack_changed = SACK_NOCHANGE;
1540 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1541
1542 NET_EPOCH_ASSERT();
1543 INP_WLOCK_ASSERT(inp);
1544 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1545 __func__));
1546 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1547 __func__));
1548
1549 #ifdef TCPPCAP
1550 /* Save segment, if requested. */
1551 tcp_pcap_add(th, m, &(tp->t_inpkts));
1552 #endif
1553 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1554 tlen, NULL, true);
1555
1556 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1557 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1558 log(LOG_DEBUG, "%s; %s: "
1559 "SYN|FIN segment ignored (based on "
1560 "sysctl setting)\n", s, __func__);
1561 free(s, M_TCPLOG);
1562 }
1563 goto drop;
1564 }
1565
1566 /*
1567 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1568 * check SEQ.ACK first.
1569 */
1570 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1571 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1572 rstreason = BANDLIM_UNLIMITED;
1573 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1574 goto dropwithreset;
1575 }
1576
1577 /*
1578 * Segment received on connection.
1579 * Reset idle time and keep-alive timer.
1580 * XXX: This should be done after segment
1581 * validation to ignore broken/spoofed segs.
1582 */
1583 if (tp->t_idle_reduce &&
1584 (tp->snd_max == tp->snd_una) &&
1585 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1586 cc_after_idle(tp);
1587 tp->t_rcvtime = ticks;
1588
1589 if (thflags & TH_FIN)
1590 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1591 /*
1592 * Scale up the window into a 32-bit value.
1593 * For the SYN_SENT state the scale is zero.
1594 */
1595 tiwin = th->th_win << tp->snd_scale;
1596 #ifdef STATS
1597 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1598 #endif
1599
1600 /*
1601 * TCP ECN processing.
1602 */
1603 if (tcp_ecn_input_segment(tp, thflags, tlen,
1604 tcp_packets_this_ack(tp, th->th_ack),
1605 iptos))
1606 cc_cong_signal(tp, th, CC_ECN);
1607
1608 /*
1609 * Parse options on any incoming segment.
1610 */
1611 tcp_dooptions(&to, (u_char *)(th + 1),
1612 (th->th_off << 2) - sizeof(struct tcphdr),
1613 (thflags & TH_SYN) ? TO_SYN : 0);
1614 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) {
1615 /*
1616 * We don't look at sack's from the
1617 * peer because the MSS is too small which
1618 * can subject us to an attack.
1619 */
1620 to.to_flags &= ~TOF_SACK;
1621 }
1622 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1623 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1624 (to.to_flags & TOF_SIGNATURE) == 0) {
1625 TCPSTAT_INC(tcps_sig_err_sigopt);
1626 /* XXX: should drop? */
1627 }
1628 #endif
1629 /*
1630 * If echoed timestamp is later than the current time,
1631 * fall back to non RFC1323 RTT calculation. Normalize
1632 * timestamp if syncookies were used when this connection
1633 * was established.
1634 */
1635 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1636 to.to_tsecr -= tp->ts_offset;
1637 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) {
1638 to.to_tsecr = 0;
1639 } else if (tp->t_rxtshift == 1 &&
1640 tp->t_flags & TF_PREVVALID &&
1641 tp->t_badrxtwin != 0 &&
1642 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) {
1643 cc_cong_signal(tp, th, CC_RTO_ERR);
1644 }
1645 }
1646 /*
1647 * Process options only when we get SYN/ACK back. The SYN case
1648 * for incoming connections is handled in tcp_syncache.
1649 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1650 * or <SYN,ACK>) segment itself is never scaled.
1651 * XXX this is traditional behavior, may need to be cleaned up.
1652 */
1653 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1654 /* Handle parallel SYN for ECN */
1655 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1656 if ((to.to_flags & TOF_SCALE) &&
1657 (tp->t_flags & TF_REQ_SCALE) &&
1658 !(tp->t_flags & TF_NOOPT)) {
1659 tp->t_flags |= TF_RCVD_SCALE;
1660 tp->snd_scale = to.to_wscale;
1661 } else {
1662 tp->t_flags &= ~TF_REQ_SCALE;
1663 }
1664 /*
1665 * Initial send window. It will be updated with
1666 * the next incoming segment to the scaled value.
1667 */
1668 tp->snd_wnd = th->th_win;
1669 if ((to.to_flags & TOF_TS) &&
1670 (tp->t_flags & TF_REQ_TSTMP) &&
1671 !(tp->t_flags & TF_NOOPT)) {
1672 tp->t_flags |= TF_RCVD_TSTMP;
1673 tp->ts_recent = to.to_tsval;
1674 tp->ts_recent_age = tcp_ts_getticks();
1675 } else {
1676 tp->t_flags &= ~TF_REQ_TSTMP;
1677 }
1678 if (to.to_flags & TOF_MSS) {
1679 tcp_mss(tp, to.to_mss);
1680 }
1681 if ((tp->t_flags & TF_SACK_PERMIT) &&
1682 (!(to.to_flags & TOF_SACKPERM) ||
1683 (tp->t_flags & TF_NOOPT))) {
1684 tp->t_flags &= ~TF_SACK_PERMIT;
1685 }
1686 if (tp->t_flags & TF_FASTOPEN) {
1687 if ((to.to_flags & TOF_FASTOPEN) &&
1688 !(tp->t_flags & TF_NOOPT)) {
1689 uint16_t mss;
1690
1691 if (to.to_flags & TOF_MSS) {
1692 mss = to.to_mss;
1693 } else {
1694 if ((inp->inp_vflag & INP_IPV6) != 0) {
1695 mss = TCP6_MSS;
1696 } else {
1697 mss = TCP_MSS;
1698 }
1699 }
1700 tcp_fastopen_update_cache(tp, mss,
1701 to.to_tfo_len, to.to_tfo_cookie);
1702 } else {
1703 tcp_fastopen_disable_path(tp);
1704 }
1705 }
1706 }
1707
1708 /*
1709 * If timestamps were negotiated during SYN/ACK and a
1710 * segment without a timestamp is received, silently drop
1711 * the segment, unless it is a RST segment or missing timestamps are
1712 * tolerated.
1713 * See section 3.2 of RFC 7323.
1714 */
1715 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1716 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1717 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1718 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1719 "segment processed normally\n",
1720 s, __func__);
1721 free(s, M_TCPLOG);
1722 }
1723 } else {
1724 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1725 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1726 "segment silently dropped\n", s, __func__);
1727 free(s, M_TCPLOG);
1728 }
1729 goto drop;
1730 }
1731 }
1732 /*
1733 * If timestamps were not negotiated during SYN/ACK and a
1734 * segment with a timestamp is received, ignore the
1735 * timestamp and process the packet normally.
1736 * See section 3.2 of RFC 7323.
1737 */
1738 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1739 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1740 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1741 "segment processed normally\n", s, __func__);
1742 free(s, M_TCPLOG);
1743 }
1744 }
1745
1746 /*
1747 * Header prediction: check for the two common cases
1748 * of a uni-directional data xfer. If the packet has
1749 * no control flags, is in-sequence, the window didn't
1750 * change and we're not retransmitting, it's a
1751 * candidate. If the length is zero and the ack moved
1752 * forward, we're the sender side of the xfer. Just
1753 * free the data acked & wake any higher level process
1754 * that was blocked waiting for space. If the length
1755 * is non-zero and the ack didn't move, we're the
1756 * receiver side. If we're getting packets in-order
1757 * (the reassembly queue is empty), add the data to
1758 * the socket buffer and note that we need a delayed ack.
1759 * Make sure that the hidden state-flags are also off.
1760 * Since we check for TCPS_ESTABLISHED first, it can only
1761 * be TH_NEEDSYN.
1762 */
1763 if (tp->t_state == TCPS_ESTABLISHED &&
1764 th->th_seq == tp->rcv_nxt &&
1765 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1766 tp->snd_nxt == tp->snd_max &&
1767 tiwin && tiwin == tp->snd_wnd &&
1768 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1769 SEGQ_EMPTY(tp) &&
1770 ((to.to_flags & TOF_TS) == 0 ||
1771 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1772 /*
1773 * If last ACK falls within this segment's sequence numbers,
1774 * record the timestamp.
1775 * NOTE that the test is modified according to the latest
1776 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1777 */
1778 if ((to.to_flags & TOF_TS) != 0 &&
1779 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1780 tp->ts_recent_age = tcp_ts_getticks();
1781 tp->ts_recent = to.to_tsval;
1782 }
1783
1784 if (tlen == 0) {
1785 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1786 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1787 !IN_RECOVERY(tp->t_flags) &&
1788 (to.to_flags & TOF_SACK) == 0 &&
1789 TAILQ_EMPTY(&tp->snd_holes)) {
1790 /*
1791 * This is a pure ack for outstanding data.
1792 */
1793 TCPSTAT_INC(tcps_predack);
1794
1795 /*
1796 * "bad retransmit" recovery without timestamps.
1797 */
1798 if ((to.to_flags & TOF_TS) == 0 &&
1799 tp->t_rxtshift == 1 &&
1800 tp->t_flags & TF_PREVVALID &&
1801 tp->t_badrxtwin != 0 &&
1802 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1803 cc_cong_signal(tp, th, CC_RTO_ERR);
1804 }
1805
1806 /*
1807 * Recalculate the transmit timer / rtt.
1808 *
1809 * Some boxes send broken timestamp replies
1810 * during the SYN+ACK phase, ignore
1811 * timestamps of 0 or we could calculate a
1812 * huge RTT and blow up the retransmit timer.
1813 */
1814 if ((to.to_flags & TOF_TS) != 0 &&
1815 to.to_tsecr) {
1816 uint32_t t;
1817
1818 t = tcp_ts_getticks() - to.to_tsecr;
1819 if (!tp->t_rttlow || tp->t_rttlow > t)
1820 tp->t_rttlow = t;
1821 tcp_xmit_timer(tp,
1822 TCP_TS_TO_TICKS(t) + 1);
1823 } else if (tp->t_rtttime &&
1824 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1825 if (!tp->t_rttlow ||
1826 tp->t_rttlow > ticks - tp->t_rtttime)
1827 tp->t_rttlow = ticks - tp->t_rtttime;
1828 tcp_xmit_timer(tp,
1829 ticks - tp->t_rtttime);
1830 }
1831 acked = BYTES_THIS_ACK(tp, th);
1832
1833 #ifdef TCP_HHOOK
1834 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1835 hhook_run_tcp_est_in(tp, th, &to);
1836 #endif
1837
1838 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1839 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1840 sbdrop(&so->so_snd, acked);
1841 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1842 SEQ_LEQ(th->th_ack, tp->snd_recover))
1843 tp->snd_recover = th->th_ack - 1;
1844
1845 /*
1846 * Let the congestion control algorithm update
1847 * congestion control related information. This
1848 * typically means increasing the congestion
1849 * window.
1850 */
1851 cc_ack_received(tp, th, nsegs, CC_ACK);
1852
1853 tp->snd_una = th->th_ack;
1854 /*
1855 * Pull snd_wl2 up to prevent seq wrap relative
1856 * to th_ack.
1857 */
1858 tp->snd_wl2 = th->th_ack;
1859 tp->t_dupacks = 0;
1860 m_freem(m);
1861
1862 /*
1863 * If all outstanding data are acked, stop
1864 * retransmit timer, otherwise restart timer
1865 * using current (possibly backed-off) value.
1866 * If process is waiting for space,
1867 * wakeup/selwakeup/signal. If data
1868 * are ready to send, let tcp_output
1869 * decide between more output or persist.
1870 */
1871 TCP_PROBE3(debug__input, tp, th, m);
1872 /*
1873 * Clear t_acktime if remote side has ACKd
1874 * all data in the socket buffer.
1875 * Otherwise, update t_acktime if we received
1876 * a sufficiently large ACK.
1877 */
1878 if (sbavail(&so->so_snd) == 0)
1879 tp->t_acktime = 0;
1880 else if (acked > 1)
1881 tp->t_acktime = ticks;
1882 if (tp->snd_una == tp->snd_max)
1883 tcp_timer_activate(tp, TT_REXMT, 0);
1884 else if (!tcp_timer_active(tp, TT_PERSIST))
1885 tcp_timer_activate(tp, TT_REXMT,
1886 TP_RXTCUR(tp));
1887 sowwakeup(so);
1888 /*
1889 * Only call tcp_output when there
1890 * is new data available to be sent
1891 * or we need to send an ACK.
1892 */
1893 if ((tp->t_flags & TF_ACKNOW) ||
1894 (sbavail(&so->so_snd) >=
1895 SEQ_SUB(tp->snd_max, tp->snd_una))) {
1896 (void) tcp_output(tp);
1897 }
1898 goto check_delack;
1899 }
1900 } else if (th->th_ack == tp->snd_una &&
1901 tlen <= sbspace(&so->so_rcv)) {
1902 int newsize = 0; /* automatic sockbuf scaling */
1903
1904 /*
1905 * This is a pure, in-sequence data packet with
1906 * nothing on the reassembly queue and we have enough
1907 * buffer space to take it.
1908 */
1909 /* Clean receiver SACK report if present */
1910 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1911 tcp_clean_sackreport(tp);
1912 TCPSTAT_INC(tcps_preddat);
1913 tp->rcv_nxt += tlen;
1914 if (tlen &&
1915 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1916 (tp->t_fbyte_in == 0)) {
1917 tp->t_fbyte_in = ticks;
1918 if (tp->t_fbyte_in == 0)
1919 tp->t_fbyte_in = 1;
1920 if (tp->t_fbyte_out && tp->t_fbyte_in)
1921 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1922 }
1923 /*
1924 * Pull snd_wl1 up to prevent seq wrap relative to
1925 * th_seq.
1926 */
1927 tp->snd_wl1 = th->th_seq;
1928 /*
1929 * Pull rcv_up up to prevent seq wrap relative to
1930 * rcv_nxt.
1931 */
1932 tp->rcv_up = tp->rcv_nxt;
1933 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1934 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1935 TCP_PROBE3(debug__input, tp, th, m);
1936
1937 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1938
1939 /* Add data to socket buffer. */
1940 SOCKBUF_LOCK(&so->so_rcv);
1941 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1942 m_freem(m);
1943 } else {
1944 /*
1945 * Set new socket buffer size.
1946 * Give up when limit is reached.
1947 */
1948 if (newsize)
1949 if (!sbreserve_locked(so, SO_RCV,
1950 newsize, NULL))
1951 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1952 m_adj(m, drop_hdrlen); /* delayed header drop */
1953 sbappendstream_locked(&so->so_rcv, m, 0);
1954 }
1955 /* NB: sorwakeup_locked() does an implicit unlock. */
1956 sorwakeup_locked(so);
1957 if (DELAY_ACK(tp, tlen)) {
1958 tp->t_flags |= TF_DELACK;
1959 } else {
1960 tp->t_flags |= TF_ACKNOW;
1961 (void) tcp_output(tp);
1962 }
1963 goto check_delack;
1964 }
1965 }
1966
1967 /*
1968 * Calculate amount of space in receive window,
1969 * and then do TCP input processing.
1970 * Receive window is amount of space in rcv queue,
1971 * but not less than advertised window.
1972 */
1973 win = sbspace(&so->so_rcv);
1974 if (win < 0)
1975 win = 0;
1976 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1977
1978 switch (tp->t_state) {
1979 /*
1980 * If the state is SYN_RECEIVED:
1981 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1982 */
1983 case TCPS_SYN_RECEIVED:
1984 if (thflags & TH_RST) {
1985 /* Handle RST segments later. */
1986 break;
1987 }
1988 if ((thflags & TH_ACK) &&
1989 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1990 SEQ_GT(th->th_ack, tp->snd_max))) {
1991 rstreason = BANDLIM_RST_OPENPORT;
1992 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1993 goto dropwithreset;
1994 }
1995 if (tp->t_flags & TF_FASTOPEN) {
1996 /*
1997 * When a TFO connection is in SYN_RECEIVED, the
1998 * only valid packets are the initial SYN, a
1999 * retransmit/copy of the initial SYN (possibly with
2000 * a subset of the original data), a valid ACK, a
2001 * FIN, or a RST.
2002 */
2003 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2004 rstreason = BANDLIM_RST_OPENPORT;
2005 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2006 goto dropwithreset;
2007 } else if (thflags & TH_SYN) {
2008 /* non-initial SYN is ignored */
2009 if ((tcp_timer_active(tp, TT_DELACK) ||
2010 tcp_timer_active(tp, TT_REXMT)))
2011 goto drop;
2012 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2013 goto drop;
2014 }
2015 }
2016 break;
2017
2018 /*
2019 * If the state is SYN_SENT:
2020 * if seg contains a RST with valid ACK (SEQ.ACK has already
2021 * been verified), then drop the connection.
2022 * if seg contains a RST without an ACK, drop the seg.
2023 * if seg does not contain SYN, then drop the seg.
2024 * Otherwise this is an acceptable SYN segment
2025 * initialize tp->rcv_nxt and tp->irs
2026 * if seg contains ack then advance tp->snd_una
2027 * if seg contains an ECE and ECN support is enabled, the stream
2028 * is ECN capable.
2029 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2030 * arrange for segment to be acked (eventually)
2031 * continue processing rest of data/controls, beginning with URG
2032 */
2033 case TCPS_SYN_SENT:
2034 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2035 TCP_PROBE5(connect__refused, NULL, tp,
2036 m, tp, th);
2037 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2038 tp = tcp_drop(tp, ECONNREFUSED);
2039 }
2040 if (thflags & TH_RST)
2041 goto drop;
2042 if (!(thflags & TH_SYN))
2043 goto drop;
2044
2045 tp->irs = th->th_seq;
2046 tcp_rcvseqinit(tp);
2047 if (thflags & TH_ACK) {
2048 int tfo_partial_ack = 0;
2049
2050 TCPSTAT_INC(tcps_connects);
2051 soisconnected(so);
2052 #ifdef MAC
2053 mac_socketpeer_set_from_mbuf(m, so);
2054 #endif
2055 /* Do window scaling on this connection? */
2056 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2057 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2058 tp->rcv_scale = tp->request_r_scale;
2059 }
2060 tp->rcv_adv += min(tp->rcv_wnd,
2061 TCP_MAXWIN << tp->rcv_scale);
2062 tp->snd_una++; /* SYN is acked */
2063 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2064 tp->snd_nxt = tp->snd_una;
2065 /*
2066 * If not all the data that was sent in the TFO SYN
2067 * has been acked, resend the remainder right away.
2068 */
2069 if ((tp->t_flags & TF_FASTOPEN) &&
2070 (tp->snd_una != tp->snd_max)) {
2071 tp->snd_nxt = th->th_ack;
2072 tfo_partial_ack = 1;
2073 }
2074 /*
2075 * If there's data, delay ACK; if there's also a FIN
2076 * ACKNOW will be turned on later.
2077 */
2078 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2079 tcp_timer_activate(tp, TT_DELACK,
2080 tcp_delacktime);
2081 else
2082 tp->t_flags |= TF_ACKNOW;
2083
2084 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2085
2086 /*
2087 * Received <SYN,ACK> in SYN_SENT[*] state.
2088 * Transitions:
2089 * SYN_SENT --> ESTABLISHED
2090 * SYN_SENT* --> FIN_WAIT_1
2091 */
2092 tp->t_starttime = ticks;
2093 if (tp->t_flags & TF_NEEDFIN) {
2094 tp->t_acktime = ticks;
2095 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2096 tp->t_flags &= ~TF_NEEDFIN;
2097 thflags &= ~TH_SYN;
2098 } else {
2099 tcp_state_change(tp, TCPS_ESTABLISHED);
2100 TCP_PROBE5(connect__established, NULL, tp,
2101 m, tp, th);
2102 cc_conn_init(tp);
2103 tcp_timer_activate(tp, TT_KEEP,
2104 TP_KEEPIDLE(tp));
2105 }
2106 } else {
2107 /*
2108 * Received initial SYN in SYN-SENT[*] state =>
2109 * simultaneous open.
2110 * If it succeeds, connection is * half-synchronized.
2111 * Otherwise, do 3-way handshake:
2112 * SYN-SENT -> SYN-RECEIVED
2113 * SYN-SENT* -> SYN-RECEIVED*
2114 */
2115 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2116 tcp_timer_activate(tp, TT_REXMT, 0);
2117 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2118 }
2119
2120 /*
2121 * Advance th->th_seq to correspond to first data byte.
2122 * If data, trim to stay within window,
2123 * dropping FIN if necessary.
2124 */
2125 th->th_seq++;
2126 if (tlen > tp->rcv_wnd) {
2127 todrop = tlen - tp->rcv_wnd;
2128 m_adj(m, -todrop);
2129 tlen = tp->rcv_wnd;
2130 thflags &= ~TH_FIN;
2131 TCPSTAT_INC(tcps_rcvpackafterwin);
2132 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2133 }
2134 tp->snd_wl1 = th->th_seq - 1;
2135 tp->rcv_up = th->th_seq;
2136 /*
2137 * Client side of transaction: already sent SYN and data.
2138 * If the remote host used T/TCP to validate the SYN,
2139 * our data will be ACK'd; if so, enter normal data segment
2140 * processing in the middle of step 5, ack processing.
2141 * Otherwise, goto step 6.
2142 */
2143 if (thflags & TH_ACK)
2144 goto process_ACK;
2145
2146 goto step6;
2147 }
2148
2149 /*
2150 * States other than LISTEN or SYN_SENT.
2151 * First check the RST flag and sequence number since reset segments
2152 * are exempt from the timestamp and connection count tests. This
2153 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2154 * below which allowed reset segments in half the sequence space
2155 * to fall though and be processed (which gives forged reset
2156 * segments with a random sequence number a 50 percent chance of
2157 * killing a connection).
2158 * Then check timestamp, if present.
2159 * Then check the connection count, if present.
2160 * Then check that at least some bytes of segment are within
2161 * receive window. If segment begins before rcv_nxt,
2162 * drop leading data (and SYN); if nothing left, just ack.
2163 */
2164 if (thflags & TH_RST) {
2165 /*
2166 * RFC5961 Section 3.2
2167 *
2168 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2169 * - If RST is in window, we send challenge ACK.
2170 *
2171 * Note: to take into account delayed ACKs, we should
2172 * test against last_ack_sent instead of rcv_nxt.
2173 * Note 2: we handle special case of closed window, not
2174 * covered by the RFC.
2175 */
2176 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2177 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2178 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2179 KASSERT(tp->t_state != TCPS_SYN_SENT,
2180 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2181 __func__, th, tp));
2182
2183 if (V_tcp_insecure_rst ||
2184 tp->last_ack_sent == th->th_seq) {
2185 TCPSTAT_INC(tcps_drops);
2186 /* Drop the connection. */
2187 switch (tp->t_state) {
2188 case TCPS_SYN_RECEIVED:
2189 so->so_error = ECONNREFUSED;
2190 goto close;
2191 case TCPS_ESTABLISHED:
2192 case TCPS_FIN_WAIT_1:
2193 case TCPS_FIN_WAIT_2:
2194 case TCPS_CLOSE_WAIT:
2195 case TCPS_CLOSING:
2196 case TCPS_LAST_ACK:
2197 so->so_error = ECONNRESET;
2198 close:
2199 /* FALLTHROUGH */
2200 default:
2201 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2202 tp = tcp_close(tp);
2203 }
2204 } else {
2205 TCPSTAT_INC(tcps_badrst);
2206 /* Send challenge ACK. */
2207 tcp_respond(tp, mtod(m, void *), th, m,
2208 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2209 tp->last_ack_sent = tp->rcv_nxt;
2210 m = NULL;
2211 }
2212 }
2213 goto drop;
2214 }
2215
2216 /*
2217 * RFC5961 Section 4.2
2218 * Send challenge ACK for any SYN in synchronized state.
2219 */
2220 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2221 tp->t_state != TCPS_SYN_RECEIVED) {
2222 TCPSTAT_INC(tcps_badsyn);
2223 if (V_tcp_insecure_syn &&
2224 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2225 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2226 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2227 tp = tcp_drop(tp, ECONNRESET);
2228 rstreason = BANDLIM_UNLIMITED;
2229 } else {
2230 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2231 /* Send challenge ACK. */
2232 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2233 tp->snd_nxt, TH_ACK);
2234 tp->last_ack_sent = tp->rcv_nxt;
2235 m = NULL;
2236 }
2237 goto drop;
2238 }
2239
2240 /*
2241 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2242 * and it's less than ts_recent, drop it.
2243 */
2244 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2245 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2246 /* Check to see if ts_recent is over 24 days old. */
2247 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2248 /*
2249 * Invalidate ts_recent. If this segment updates
2250 * ts_recent, the age will be reset later and ts_recent
2251 * will get a valid value. If it does not, setting
2252 * ts_recent to zero will at least satisfy the
2253 * requirement that zero be placed in the timestamp
2254 * echo reply when ts_recent isn't valid. The
2255 * age isn't reset until we get a valid ts_recent
2256 * because we don't want out-of-order segments to be
2257 * dropped when ts_recent is old.
2258 */
2259 tp->ts_recent = 0;
2260 } else {
2261 TCPSTAT_INC(tcps_rcvduppack);
2262 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2263 TCPSTAT_INC(tcps_pawsdrop);
2264 if (tlen)
2265 goto dropafterack;
2266 goto drop;
2267 }
2268 }
2269
2270 /*
2271 * In the SYN-RECEIVED state, validate that the packet belongs to
2272 * this connection before trimming the data to fit the receive
2273 * window. Check the sequence number versus IRS since we know
2274 * the sequence numbers haven't wrapped. This is a partial fix
2275 * for the "LAND" DoS attack.
2276 */
2277 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2278 rstreason = BANDLIM_RST_OPENPORT;
2279 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2280 goto dropwithreset;
2281 }
2282
2283 todrop = tp->rcv_nxt - th->th_seq;
2284 if (todrop > 0) {
2285 if (thflags & TH_SYN) {
2286 thflags &= ~TH_SYN;
2287 th->th_seq++;
2288 if (th->th_urp > 1)
2289 th->th_urp--;
2290 else
2291 thflags &= ~TH_URG;
2292 todrop--;
2293 }
2294 /*
2295 * Following if statement from Stevens, vol. 2, p. 960.
2296 */
2297 if (todrop > tlen
2298 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2299 /*
2300 * Any valid FIN must be to the left of the window.
2301 * At this point the FIN must be a duplicate or out
2302 * of sequence; drop it.
2303 */
2304 thflags &= ~TH_FIN;
2305
2306 /*
2307 * Send an ACK to resynchronize and drop any data.
2308 * But keep on processing for RST or ACK.
2309 */
2310 tp->t_flags |= TF_ACKNOW;
2311 todrop = tlen;
2312 TCPSTAT_INC(tcps_rcvduppack);
2313 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2314 } else {
2315 TCPSTAT_INC(tcps_rcvpartduppack);
2316 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2317 }
2318 /*
2319 * DSACK - add SACK block for dropped range
2320 */
2321 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2322 tcp_update_sack_list(tp, th->th_seq,
2323 th->th_seq + todrop);
2324 /*
2325 * ACK now, as the next in-sequence segment
2326 * will clear the DSACK block again
2327 */
2328 tp->t_flags |= TF_ACKNOW;
2329 }
2330 drop_hdrlen += todrop; /* drop from the top afterwards */
2331 th->th_seq += todrop;
2332 tlen -= todrop;
2333 if (th->th_urp > todrop)
2334 th->th_urp -= todrop;
2335 else {
2336 thflags &= ~TH_URG;
2337 th->th_urp = 0;
2338 }
2339 }
2340
2341 /*
2342 * If new data are received on a connection after the
2343 * user processes are gone, then RST the other end if
2344 * no FIN has been processed.
2345 */
2346 if ((tp->t_flags & TF_CLOSED) && tlen > 0 &&
2347 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2348 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2349 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2350 "after socket was closed, "
2351 "sending RST and removing tcpcb\n",
2352 s, __func__, tcpstates[tp->t_state], tlen);
2353 free(s, M_TCPLOG);
2354 }
2355 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2356 /* tcp_close will kill the inp pre-log the Reset */
2357 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2358 tp = tcp_close(tp);
2359 TCPSTAT_INC(tcps_rcvafterclose);
2360 rstreason = BANDLIM_UNLIMITED;
2361 goto dropwithreset;
2362 }
2363
2364 /*
2365 * If segment ends after window, drop trailing data
2366 * (and PUSH and FIN); if nothing left, just ACK.
2367 */
2368 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2369 if (todrop > 0) {
2370 TCPSTAT_INC(tcps_rcvpackafterwin);
2371 if (todrop >= tlen) {
2372 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2373 /*
2374 * If window is closed can only take segments at
2375 * window edge, and have to drop data and PUSH from
2376 * incoming segments. Continue processing, but
2377 * remember to ack. Otherwise, drop segment
2378 * and ack.
2379 */
2380 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2381 tp->t_flags |= TF_ACKNOW;
2382 TCPSTAT_INC(tcps_rcvwinprobe);
2383 } else
2384 goto dropafterack;
2385 } else
2386 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2387 m_adj(m, -todrop);
2388 tlen -= todrop;
2389 thflags &= ~(TH_PUSH|TH_FIN);
2390 }
2391
2392 /*
2393 * If last ACK falls within this segment's sequence numbers,
2394 * record its timestamp.
2395 * NOTE:
2396 * 1) That the test incorporates suggestions from the latest
2397 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2398 * 2) That updating only on newer timestamps interferes with
2399 * our earlier PAWS tests, so this check should be solely
2400 * predicated on the sequence space of this segment.
2401 * 3) That we modify the segment boundary check to be
2402 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2403 * instead of RFC1323's
2404 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2405 * This modified check allows us to overcome RFC1323's
2406 * limitations as described in Stevens TCP/IP Illustrated
2407 * Vol. 2 p.869. In such cases, we can still calculate the
2408 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2409 */
2410 if ((to.to_flags & TOF_TS) != 0 &&
2411 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2412 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2413 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2414 tp->ts_recent_age = tcp_ts_getticks();
2415 tp->ts_recent = to.to_tsval;
2416 }
2417
2418 /*
2419 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2420 * flag is on (half-synchronized state), then queue data for
2421 * later processing; else drop segment and return.
2422 */
2423 if ((thflags & TH_ACK) == 0) {
2424 if (tp->t_state == TCPS_SYN_RECEIVED ||
2425 (tp->t_flags & TF_NEEDSYN)) {
2426 if (tp->t_state == TCPS_SYN_RECEIVED &&
2427 (tp->t_flags & TF_FASTOPEN)) {
2428 tp->snd_wnd = tiwin;
2429 cc_conn_init(tp);
2430 }
2431 goto step6;
2432 } else if (tp->t_flags & TF_ACKNOW)
2433 goto dropafterack;
2434 else
2435 goto drop;
2436 }
2437
2438 /*
2439 * Ack processing.
2440 */
2441 switch (tp->t_state) {
2442 /*
2443 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2444 * ESTABLISHED state and continue processing.
2445 * The ACK was checked above.
2446 */
2447 case TCPS_SYN_RECEIVED:
2448
2449 TCPSTAT_INC(tcps_connects);
2450 if (tp->t_flags & TF_SONOTCONN) {
2451 /*
2452 * Usually SYN_RECEIVED had been created from a LISTEN,
2453 * and solisten_enqueue() has already marked the socket
2454 * layer as connected. If it didn't, which can happen
2455 * only with an accept_filter(9), then the tp is marked
2456 * with TF_SONOTCONN. The other reason for this mark
2457 * to be set is a simultaneous open, a SYN_RECEIVED
2458 * that had been created from SYN_SENT.
2459 */
2460 tp->t_flags &= ~TF_SONOTCONN;
2461 soisconnected(so);
2462 }
2463 /* Do window scaling? */
2464 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2465 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2466 tp->rcv_scale = tp->request_r_scale;
2467 }
2468 tp->snd_wnd = tiwin;
2469 /*
2470 * Make transitions:
2471 * SYN-RECEIVED -> ESTABLISHED
2472 * SYN-RECEIVED* -> FIN-WAIT-1
2473 */
2474 tp->t_starttime = ticks;
2475 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) {
2476 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2477 tp->t_tfo_pending = NULL;
2478 }
2479 if (tp->t_flags & TF_NEEDFIN) {
2480 tp->t_acktime = ticks;
2481 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2482 tp->t_flags &= ~TF_NEEDFIN;
2483 } else {
2484 tcp_state_change(tp, TCPS_ESTABLISHED);
2485 TCP_PROBE5(accept__established, NULL, tp,
2486 m, tp, th);
2487 /*
2488 * TFO connections call cc_conn_init() during SYN
2489 * processing. Calling it again here for such
2490 * connections is not harmless as it would undo the
2491 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2492 * is retransmitted.
2493 */
2494 if (!(tp->t_flags & TF_FASTOPEN))
2495 cc_conn_init(tp);
2496 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2497 }
2498 /*
2499 * Account for the ACK of our SYN prior to
2500 * regular ACK processing below, except for
2501 * simultaneous SYN, which is handled later.
2502 */
2503 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2504 incforsyn = 1;
2505 /*
2506 * If segment contains data or ACK, will call tcp_reass()
2507 * later; if not, do so now to pass queued data to user.
2508 */
2509 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2510 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2511 (struct mbuf *)0);
2512 tcp_handle_wakeup(tp);
2513 }
2514 tp->snd_wl1 = th->th_seq - 1;
2515 /* FALLTHROUGH */
2516
2517 /*
2518 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2519 * ACKs. If the ack is in the range
2520 * tp->snd_una < th->th_ack <= tp->snd_max
2521 * then advance tp->snd_una to th->th_ack and drop
2522 * data from the retransmission queue. If this ACK reflects
2523 * more up to date window information we update our window information.
2524 */
2525 case TCPS_ESTABLISHED:
2526 case TCPS_FIN_WAIT_1:
2527 case TCPS_FIN_WAIT_2:
2528 case TCPS_CLOSE_WAIT:
2529 case TCPS_CLOSING:
2530 case TCPS_LAST_ACK:
2531 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2532 TCPSTAT_INC(tcps_rcvacktoomuch);
2533 goto dropafterack;
2534 }
2535 if (tcp_is_sack_recovery(tp, &to)) {
2536 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2537 if ((sack_changed != SACK_NOCHANGE) &&
2538 (tp->t_flags & TF_LRD)) {
2539 tcp_sack_lost_retransmission(tp, th);
2540 }
2541 } else
2542 /*
2543 * Reset the value so that previous (valid) value
2544 * from the last ack with SACK doesn't get used.
2545 */
2546 tp->sackhint.sacked_bytes = 0;
2547
2548 #ifdef TCP_HHOOK
2549 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2550 hhook_run_tcp_est_in(tp, th, &to);
2551 #endif
2552
2553 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2554 maxseg = tcp_maxseg(tp);
2555 if (tlen == 0 &&
2556 (tiwin == tp->snd_wnd ||
2557 (tp->t_flags & TF_SACK_PERMIT))) {
2558 /*
2559 * If this is the first time we've seen a
2560 * FIN from the remote, this is not a
2561 * duplicate and it needs to be processed
2562 * normally. This happens during a
2563 * simultaneous close.
2564 */
2565 if ((thflags & TH_FIN) &&
2566 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2567 tp->t_dupacks = 0;
2568 break;
2569 }
2570 TCPSTAT_INC(tcps_rcvdupack);
2571 /*
2572 * If we have outstanding data (other than
2573 * a window probe), this is a completely
2574 * duplicate ack (ie, window info didn't
2575 * change and FIN isn't set),
2576 * the ack is the biggest we've
2577 * seen and we've seen exactly our rexmt
2578 * threshold of them, assume a packet
2579 * has been dropped and retransmit it.
2580 * Kludge snd_nxt & the congestion
2581 * window so we send only this one
2582 * packet.
2583 *
2584 * We know we're losing at the current
2585 * window size so do congestion avoidance
2586 * (set ssthresh to half the current window
2587 * and pull our congestion window back to
2588 * the new ssthresh).
2589 *
2590 * Dup acks mean that packets have left the
2591 * network (they're now cached at the receiver)
2592 * so bump cwnd by the amount in the receiver
2593 * to keep a constant cwnd packets in the
2594 * network.
2595 *
2596 * When using TCP ECN, notify the peer that
2597 * we reduced the cwnd.
2598 */
2599 /*
2600 * Following 2 kinds of acks should not affect
2601 * dupack counting:
2602 * 1) Old acks
2603 * 2) Acks with SACK but without any new SACK
2604 * information in them. These could result from
2605 * any anomaly in the network like a switch
2606 * duplicating packets or a possible DoS attack.
2607 */
2608 if (th->th_ack != tp->snd_una ||
2609 (tcp_is_sack_recovery(tp, &to) &&
2610 (sack_changed == SACK_NOCHANGE))) {
2611 break;
2612 } else if (!tcp_timer_active(tp, TT_REXMT)) {
2613 tp->t_dupacks = 0;
2614 } else if (++tp->t_dupacks > tcprexmtthresh ||
2615 IN_FASTRECOVERY(tp->t_flags)) {
2616 cc_ack_received(tp, th, nsegs,
2617 CC_DUPACK);
2618 if (V_tcp_do_prr &&
2619 IN_FASTRECOVERY(tp->t_flags) &&
2620 (tp->t_flags & TF_SACK_PERMIT)) {
2621 tcp_do_prr_ack(tp, th, &to,
2622 sack_changed, &maxseg);
2623 } else if (tcp_is_sack_recovery(tp, &to) &&
2624 IN_FASTRECOVERY(tp->t_flags)) {
2625 int awnd;
2626
2627 /*
2628 * Compute the amount of data in flight first.
2629 * We can inject new data into the pipe iff
2630 * we have less than 1/2 the original window's
2631 * worth of data in flight.
2632 */
2633 if (V_tcp_do_newsack) {
2634 awnd = tcp_compute_pipe(tp);
2635 } else {
2636 awnd = (tp->snd_nxt - tp->snd_fack) +
2637 tp->sackhint.sack_bytes_rexmit;
2638 }
2639 if (awnd < tp->snd_ssthresh) {
2640 tp->snd_cwnd += maxseg;
2641 if (tp->snd_cwnd > tp->snd_ssthresh)
2642 tp->snd_cwnd = tp->snd_ssthresh;
2643 }
2644 } else {
2645 tp->snd_cwnd += maxseg;
2646 }
2647 (void) tcp_output(tp);
2648 goto drop;
2649 } else if (tp->t_dupacks == tcprexmtthresh ||
2650 (tp->t_flags & TF_SACK_PERMIT &&
2651 V_tcp_do_newsack &&
2652 tp->sackhint.sacked_bytes >
2653 (tcprexmtthresh - 1) * maxseg)) {
2654 enter_recovery:
2655 /*
2656 * Above is the RFC6675 trigger condition of
2657 * more than (dupthresh-1)*maxseg sacked data.
2658 * If the count of holes in the
2659 * scoreboard is >= dupthresh, we could
2660 * also enter loss recovery, but don't
2661 * have that value readily available.
2662 */
2663 tp->t_dupacks = tcprexmtthresh;
2664 tcp_seq onxt = tp->snd_nxt;
2665
2666 /*
2667 * If we're doing sack, or prr, check
2668 * to see if we're already in sack
2669 * recovery. If we're not doing sack,
2670 * check to see if we're in newreno
2671 * recovery.
2672 */
2673 if (V_tcp_do_prr ||
2674 (tp->t_flags & TF_SACK_PERMIT)) {
2675 if (IN_FASTRECOVERY(tp->t_flags)) {
2676 tp->t_dupacks = 0;
2677 break;
2678 }
2679 } else {
2680 if (SEQ_LEQ(th->th_ack,
2681 tp->snd_recover)) {
2682 tp->t_dupacks = 0;
2683 break;
2684 }
2685 }
2686 /* Congestion signal before ack. */
2687 cc_cong_signal(tp, th, CC_NDUPACK);
2688 cc_ack_received(tp, th, nsegs,
2689 CC_DUPACK);
2690 tcp_timer_activate(tp, TT_REXMT, 0);
2691 tp->t_rtttime = 0;
2692 if (V_tcp_do_prr) {
2693 /*
2694 * snd_ssthresh is already updated by
2695 * cc_cong_signal.
2696 */
2697 if (tcp_is_sack_recovery(tp, &to)) {
2698 /*
2699 * Exclude Limited Transmit
2700 * segments here
2701 */
2702 tp->sackhint.prr_delivered =
2703 maxseg;
2704 } else {
2705 tp->sackhint.prr_delivered =
2706 imin(tp->snd_max - tp->snd_una,
2707 imin(INT_MAX / 65536,
2708 tp->t_dupacks) * maxseg);
2709 }
2710 tp->sackhint.recover_fs = max(1,
2711 tp->snd_nxt - tp->snd_una);
2712 }
2713 if (tcp_is_sack_recovery(tp, &to)) {
2714 TCPSTAT_INC(tcps_sack_recovery_episode);
2715 tp->snd_recover = tp->snd_nxt;
2716 tp->snd_cwnd = maxseg;
2717 (void) tcp_output(tp);
2718 if (SEQ_GT(th->th_ack, tp->snd_una)) {
2719 goto resume_partialack;
2720 }
2721 goto drop;
2722 }
2723 tp->snd_nxt = th->th_ack;
2724 tp->snd_cwnd = maxseg;
2725 (void) tcp_output(tp);
2726 KASSERT(tp->snd_limited <= 2,
2727 ("%s: tp->snd_limited too big",
2728 __func__));
2729 tp->snd_cwnd = tp->snd_ssthresh +
2730 maxseg *
2731 (tp->t_dupacks - tp->snd_limited);
2732 if (SEQ_GT(onxt, tp->snd_nxt))
2733 tp->snd_nxt = onxt;
2734 goto drop;
2735 } else if (V_tcp_do_rfc3042) {
2736 /*
2737 * Process first and second duplicate
2738 * ACKs. Each indicates a segment
2739 * leaving the network, creating room
2740 * for more. Make sure we can send a
2741 * packet on reception of each duplicate
2742 * ACK by increasing snd_cwnd by one
2743 * segment. Restore the original
2744 * snd_cwnd after packet transmission.
2745 */
2746 cc_ack_received(tp, th, nsegs, CC_DUPACK);
2747 uint32_t oldcwnd = tp->snd_cwnd;
2748 tcp_seq oldsndmax = tp->snd_max;
2749 u_int sent;
2750 int avail;
2751
2752 KASSERT(tp->t_dupacks == 1 ||
2753 tp->t_dupacks == 2,
2754 ("%s: dupacks not 1 or 2",
2755 __func__));
2756 if (tp->t_dupacks == 1)
2757 tp->snd_limited = 0;
2758 tp->snd_cwnd =
2759 (tp->snd_nxt - tp->snd_una) +
2760 (tp->t_dupacks - tp->snd_limited) *
2761 maxseg;
2762 /*
2763 * Only call tcp_output when there
2764 * is new data available to be sent
2765 * or we need to send an ACK.
2766 */
2767 SOCKBUF_LOCK(&so->so_snd);
2768 avail = sbavail(&so->so_snd);
2769 SOCKBUF_UNLOCK(&so->so_snd);
2770 if (tp->t_flags & TF_ACKNOW ||
2771 (avail >=
2772 SEQ_SUB(tp->snd_nxt, tp->snd_una))) {
2773 (void) tcp_output(tp);
2774 }
2775 sent = SEQ_SUB(tp->snd_max, oldsndmax);
2776 if (sent > maxseg) {
2777 KASSERT((tp->t_dupacks == 2 &&
2778 tp->snd_limited == 0) ||
2779 (sent == maxseg + 1 &&
2780 tp->t_flags & TF_SENTFIN),
2781 ("%s: sent too much",
2782 __func__));
2783 tp->snd_limited = 2;
2784 } else if (sent > 0) {
2785 ++tp->snd_limited;
2786 }
2787 tp->snd_cwnd = oldcwnd;
2788 goto drop;
2789 }
2790 }
2791 break;
2792 } else {
2793 /*
2794 * This ack is advancing the left edge, reset the
2795 * counter.
2796 */
2797 tp->t_dupacks = 0;
2798 /*
2799 * If this ack also has new SACK info, increment the
2800 * counter as per rfc6675. The variable
2801 * sack_changed tracks all changes to the SACK
2802 * scoreboard, including when partial ACKs without
2803 * SACK options are received, and clear the scoreboard
2804 * from the left side. Such partial ACKs should not be
2805 * counted as dupacks here.
2806 */
2807 if (tcp_is_sack_recovery(tp, &to) &&
2808 (sack_changed != SACK_NOCHANGE)) {
2809 tp->t_dupacks++;
2810 /* limit overhead by setting maxseg last */
2811 if (!IN_FASTRECOVERY(tp->t_flags) &&
2812 (tp->sackhint.sacked_bytes >
2813 ((tcprexmtthresh - 1) *
2814 (maxseg = tcp_maxseg(tp))))) {
2815 goto enter_recovery;
2816 }
2817 }
2818 }
2819
2820 resume_partialack:
2821 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2822 ("%s: th_ack <= snd_una", __func__));
2823
2824 /*
2825 * If the congestion window was inflated to account
2826 * for the other side's cached packets, retract it.
2827 */
2828 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2829 if (IN_FASTRECOVERY(tp->t_flags)) {
2830 if (tp->t_flags & TF_SACK_PERMIT) {
2831 if (V_tcp_do_prr &&
2832 (to.to_flags & TOF_SACK)) {
2833 tcp_timer_activate(tp,
2834 TT_REXMT, 0);
2835 tp->t_rtttime = 0;
2836 tcp_do_prr_ack(tp, th, &to,
2837 sack_changed, &maxseg);
2838 tp->t_flags |= TF_ACKNOW;
2839 (void) tcp_output(tp);
2840 } else {
2841 tcp_sack_partialack(tp, th,
2842 &maxseg);
2843 }
2844 } else {
2845 tcp_newreno_partial_ack(tp, th);
2846 }
2847 } else if (IN_CONGRECOVERY(tp->t_flags) &&
2848 (V_tcp_do_prr)) {
2849 tp->sackhint.delivered_data =
2850 BYTES_THIS_ACK(tp, th);
2851 tp->snd_fack = th->th_ack;
2852 /*
2853 * During ECN cwnd reduction
2854 * always use PRR-SSRB
2855 */
2856 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE,
2857 &maxseg);
2858 (void) tcp_output(tp);
2859 }
2860 }
2861 /*
2862 * If we reach this point, ACK is not a duplicate,
2863 * i.e., it ACKs something we sent.
2864 */
2865 if (tp->t_flags & TF_NEEDSYN) {
2866 /*
2867 * T/TCP: Connection was half-synchronized, and our
2868 * SYN has been ACK'd (so connection is now fully
2869 * synchronized). Go to non-starred state,
2870 * increment snd_una for ACK of SYN, and check if
2871 * we can do window scaling.
2872 */
2873 tp->t_flags &= ~TF_NEEDSYN;
2874 tp->snd_una++;
2875 /* Do window scaling? */
2876 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2877 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2878 tp->rcv_scale = tp->request_r_scale;
2879 /* Send window already scaled. */
2880 }
2881 }
2882
2883 process_ACK:
2884 INP_WLOCK_ASSERT(inp);
2885
2886 /*
2887 * Adjust for the SYN bit in sequence space,
2888 * but don't account for it in cwnd calculations.
2889 * This is for the SYN_RECEIVED, non-simultaneous
2890 * SYN case. SYN_SENT and simultaneous SYN are
2891 * treated elsewhere.
2892 */
2893 if (incforsyn)
2894 tp->snd_una++;
2895 acked = BYTES_THIS_ACK(tp, th);
2896 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2897 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2898 tp->snd_una, th->th_ack, tp, m));
2899 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2900 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2901
2902 /*
2903 * If we just performed our first retransmit, and the ACK
2904 * arrives within our recovery window, then it was a mistake
2905 * to do the retransmit in the first place. Recover our
2906 * original cwnd and ssthresh, and proceed to transmit where
2907 * we left off.
2908 */
2909 if (tp->t_rxtshift == 1 &&
2910 tp->t_flags & TF_PREVVALID &&
2911 tp->t_badrxtwin != 0 &&
2912 to.to_flags & TOF_TS &&
2913 to.to_tsecr != 0 &&
2914 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2915 cc_cong_signal(tp, th, CC_RTO_ERR);
2916
2917 /*
2918 * If we have a timestamp reply, update smoothed
2919 * round trip time. If no timestamp is present but
2920 * transmit timer is running and timed sequence
2921 * number was acked, update smoothed round trip time.
2922 * Since we now have an rtt measurement, cancel the
2923 * timer backoff (cf., Phil Karn's retransmit alg.).
2924 * Recompute the initial retransmit timer.
2925 *
2926 * Some boxes send broken timestamp replies
2927 * during the SYN+ACK phase, ignore
2928 * timestamps of 0 or we could calculate a
2929 * huge RTT and blow up the retransmit timer.
2930 */
2931 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2932 uint32_t t;
2933
2934 t = tcp_ts_getticks() - to.to_tsecr;
2935 if (!tp->t_rttlow || tp->t_rttlow > t)
2936 tp->t_rttlow = t;
2937 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2938 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2939 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2940 tp->t_rttlow = ticks - tp->t_rtttime;
2941 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2942 }
2943
2944 SOCKBUF_LOCK(&so->so_snd);
2945 /*
2946 * Clear t_acktime if remote side has ACKd all data in the
2947 * socket buffer and FIN (if applicable).
2948 * Otherwise, update t_acktime if we received a sufficiently
2949 * large ACK.
2950 */
2951 if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2952 acked == sbavail(&so->so_snd)) ||
2953 acked > sbavail(&so->so_snd))
2954 tp->t_acktime = 0;
2955 else if (acked > 1)
2956 tp->t_acktime = ticks;
2957
2958 /*
2959 * If all outstanding data is acked, stop retransmit
2960 * timer and remember to restart (more output or persist).
2961 * If there is more data to be acked, restart retransmit
2962 * timer, using current (possibly backed-off) value.
2963 */
2964 if (th->th_ack == tp->snd_max) {
2965 tcp_timer_activate(tp, TT_REXMT, 0);
2966 needoutput = 1;
2967 } else if (!tcp_timer_active(tp, TT_PERSIST))
2968 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
2969
2970 /*
2971 * If no data (only SYN) was ACK'd,
2972 * skip rest of ACK processing.
2973 */
2974 if (acked == 0) {
2975 SOCKBUF_UNLOCK(&so->so_snd);
2976 goto step6;
2977 }
2978
2979 /*
2980 * Let the congestion control algorithm update congestion
2981 * control related information. This typically means increasing
2982 * the congestion window.
2983 */
2984 cc_ack_received(tp, th, nsegs, CC_ACK);
2985
2986 if (acked > sbavail(&so->so_snd)) {
2987 if (tp->snd_wnd >= sbavail(&so->so_snd))
2988 tp->snd_wnd -= sbavail(&so->so_snd);
2989 else
2990 tp->snd_wnd = 0;
2991 mfree = sbcut_locked(&so->so_snd,
2992 (int)sbavail(&so->so_snd));
2993 ourfinisacked = 1;
2994 } else {
2995 mfree = sbcut_locked(&so->so_snd, acked);
2996 if (tp->snd_wnd >= (uint32_t) acked)
2997 tp->snd_wnd -= acked;
2998 else
2999 tp->snd_wnd = 0;
3000 ourfinisacked = 0;
3001 }
3002 /* NB: sowwakeup_locked() does an implicit unlock. */
3003 sowwakeup_locked(so);
3004 m_freem(mfree);
3005 /* Detect una wraparound. */
3006 if (!IN_RECOVERY(tp->t_flags) &&
3007 SEQ_GT(tp->snd_una, tp->snd_recover) &&
3008 SEQ_LEQ(th->th_ack, tp->snd_recover))
3009 tp->snd_recover = th->th_ack - 1;
3010 tp->snd_una = th->th_ack;
3011 if (IN_RECOVERY(tp->t_flags) &&
3012 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
3013 cc_post_recovery(tp, th);
3014 }
3015 if (tp->t_flags & TF_SACK_PERMIT) {
3016 if (SEQ_GT(tp->snd_una, tp->snd_recover))
3017 tp->snd_recover = tp->snd_una;
3018 }
3019 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
3020 tp->snd_nxt = tp->snd_una;
3021
3022 switch (tp->t_state) {
3023 /*
3024 * In FIN_WAIT_1 STATE in addition to the processing
3025 * for the ESTABLISHED state if our FIN is now acknowledged
3026 * then enter FIN_WAIT_2.
3027 */
3028 case TCPS_FIN_WAIT_1:
3029 if (ourfinisacked) {
3030 /*
3031 * If we can't receive any more
3032 * data, then closing user can proceed.
3033 * Starting the timer is contrary to the
3034 * specification, but if we don't get a FIN
3035 * we'll hang forever.
3036 */
3037 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3038 tcp_free_sackholes(tp);
3039 soisdisconnected(so);
3040 tcp_timer_activate(tp, TT_2MSL,
3041 (tcp_fast_finwait2_recycle ?
3042 tcp_finwait2_timeout :
3043 TP_MAXIDLE(tp)));
3044 }
3045 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3046 }
3047 break;
3048
3049 /*
3050 * In CLOSING STATE in addition to the processing for
3051 * the ESTABLISHED state if the ACK acknowledges our FIN
3052 * then enter the TIME-WAIT state, otherwise ignore
3053 * the segment.
3054 */
3055 case TCPS_CLOSING:
3056 if (ourfinisacked) {
3057 tcp_twstart(tp);
3058 m_freem(m);
3059 return;
3060 }
3061 break;
3062
3063 /*
3064 * In LAST_ACK, we may still be waiting for data to drain
3065 * and/or to be acked, as well as for the ack of our FIN.
3066 * If our FIN is now acknowledged, delete the TCB,
3067 * enter the closed state and return.
3068 */
3069 case TCPS_LAST_ACK:
3070 if (ourfinisacked) {
3071 tp = tcp_close(tp);
3072 goto drop;
3073 }
3074 break;
3075 }
3076 }
3077
3078 step6:
3079 INP_WLOCK_ASSERT(inp);
3080
3081 /*
3082 * Update window information.
3083 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3084 */
3085 if ((thflags & TH_ACK) &&
3086 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3087 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3088 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3089 /* keep track of pure window updates */
3090 if (tlen == 0 &&
3091 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3092 TCPSTAT_INC(tcps_rcvwinupd);
3093 tp->snd_wnd = tiwin;
3094 tp->snd_wl1 = th->th_seq;
3095 tp->snd_wl2 = th->th_ack;
3096 if (tp->snd_wnd > tp->max_sndwnd)
3097 tp->max_sndwnd = tp->snd_wnd;
3098 needoutput = 1;
3099 }
3100
3101 /*
3102 * Process segments with URG.
3103 */
3104 if ((thflags & TH_URG) && th->th_urp &&
3105 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3106 /*
3107 * This is a kludge, but if we receive and accept
3108 * random urgent pointers, we'll crash in
3109 * soreceive. It's hard to imagine someone
3110 * actually wanting to send this much urgent data.
3111 */
3112 SOCKBUF_LOCK(&so->so_rcv);
3113 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3114 th->th_urp = 0; /* XXX */
3115 thflags &= ~TH_URG; /* XXX */
3116 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3117 goto dodata; /* XXX */
3118 }
3119 /*
3120 * If this segment advances the known urgent pointer,
3121 * then mark the data stream. This should not happen
3122 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3123 * a FIN has been received from the remote side.
3124 * In these states we ignore the URG.
3125 *
3126 * According to RFC961 (Assigned Protocols),
3127 * the urgent pointer points to the last octet
3128 * of urgent data. We continue, however,
3129 * to consider it to indicate the first octet
3130 * of data past the urgent section as the original
3131 * spec states (in one of two places).
3132 */
3133 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3134 tp->rcv_up = th->th_seq + th->th_urp;
3135 so->so_oobmark = sbavail(&so->so_rcv) +
3136 (tp->rcv_up - tp->rcv_nxt) - 1;
3137 if (so->so_oobmark == 0)
3138 so->so_rcv.sb_state |= SBS_RCVATMARK;
3139 sohasoutofband(so);
3140 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3141 }
3142 SOCKBUF_UNLOCK(&so->so_rcv);
3143 /*
3144 * Remove out of band data so doesn't get presented to user.
3145 * This can happen independent of advancing the URG pointer,
3146 * but if two URG's are pending at once, some out-of-band
3147 * data may creep in... ick.
3148 */
3149 if (th->th_urp <= (uint32_t)tlen &&
3150 !(so->so_options & SO_OOBINLINE)) {
3151 /* hdr drop is delayed */
3152 tcp_pulloutofband(so, th, m, drop_hdrlen);
3153 }
3154 } else {
3155 /*
3156 * If no out of band data is expected,
3157 * pull receive urgent pointer along
3158 * with the receive window.
3159 */
3160 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3161 tp->rcv_up = tp->rcv_nxt;
3162 }
3163 dodata: /* XXX */
3164 INP_WLOCK_ASSERT(inp);
3165
3166 /*
3167 * Process the segment text, merging it into the TCP sequencing queue,
3168 * and arranging for acknowledgment of receipt if necessary.
3169 * This process logically involves adjusting tp->rcv_wnd as data
3170 * is presented to the user (this happens in tcp_usrreq.c,
3171 * case PRU_RCVD). If a FIN has already been received on this
3172 * connection then we just ignore the text.
3173 */
3174 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3175 (tp->t_flags & TF_FASTOPEN));
3176 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3177 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3178 tcp_seq save_start = th->th_seq;
3179 tcp_seq save_rnxt = tp->rcv_nxt;
3180 int save_tlen = tlen;
3181 m_adj(m, drop_hdrlen); /* delayed header drop */
3182 /*
3183 * Insert segment which includes th into TCP reassembly queue
3184 * with control block tp. Set thflags to whether reassembly now
3185 * includes a segment with FIN. This handles the common case
3186 * inline (segment is the next to be received on an established
3187 * connection, and the queue is empty), avoiding linkage into
3188 * and removal from the queue and repetition of various
3189 * conversions.
3190 * Set DELACK for segments received in order, but ack
3191 * immediately when segments are out of order (so
3192 * fast retransmit can work).
3193 */
3194 if (th->th_seq == tp->rcv_nxt &&
3195 SEGQ_EMPTY(tp) &&
3196 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3197 tfo_syn)) {
3198 if (DELAY_ACK(tp, tlen) || tfo_syn)
3199 tp->t_flags |= TF_DELACK;
3200 else
3201 tp->t_flags |= TF_ACKNOW;
3202 tp->rcv_nxt += tlen;
3203 if (tlen &&
3204 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3205 (tp->t_fbyte_in == 0)) {
3206 tp->t_fbyte_in = ticks;
3207 if (tp->t_fbyte_in == 0)
3208 tp->t_fbyte_in = 1;
3209 if (tp->t_fbyte_out && tp->t_fbyte_in)
3210 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3211 }
3212 thflags = tcp_get_flags(th) & TH_FIN;
3213 TCPSTAT_INC(tcps_rcvpack);
3214 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3215 SOCKBUF_LOCK(&so->so_rcv);
3216 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3217 m_freem(m);
3218 else
3219 sbappendstream_locked(&so->so_rcv, m, 0);
3220 tp->t_flags |= TF_WAKESOR;
3221 } else {
3222 /*
3223 * XXX: Due to the header drop above "th" is
3224 * theoretically invalid by now. Fortunately
3225 * m_adj() doesn't actually frees any mbufs
3226 * when trimming from the head.
3227 */
3228 tcp_seq temp = save_start;
3229
3230 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3231 tp->t_flags |= TF_ACKNOW;
3232 }
3233 if ((tp->t_flags & TF_SACK_PERMIT) &&
3234 (save_tlen > 0) &&
3235 TCPS_HAVEESTABLISHED(tp->t_state)) {
3236 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3237 /*
3238 * DSACK actually handled in the fastpath
3239 * above.
3240 */
3241 tcp_update_sack_list(tp, save_start,
3242 save_start + save_tlen);
3243 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3244 if ((tp->rcv_numsacks >= 1) &&
3245 (tp->sackblks[0].end == save_start)) {
3246 /*
3247 * Partial overlap, recorded at todrop
3248 * above.
3249 */
3250 tcp_update_sack_list(tp,
3251 tp->sackblks[0].start,
3252 tp->sackblks[0].end);
3253 } else {
3254 tcp_update_dsack_list(tp, save_start,
3255 save_start + save_tlen);
3256 }
3257 } else if (tlen >= save_tlen) {
3258 /* Update of sackblks. */
3259 tcp_update_dsack_list(tp, save_start,
3260 save_start + save_tlen);
3261 } else if (tlen > 0) {
3262 tcp_update_dsack_list(tp, save_start,
3263 save_start + tlen);
3264 }
3265 }
3266 tcp_handle_wakeup(tp);
3267 #if 0
3268 /*
3269 * Note the amount of data that peer has sent into
3270 * our window, in order to estimate the sender's
3271 * buffer size.
3272 * XXX: Unused.
3273 */
3274 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3275 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3276 else
3277 len = so->so_rcv.sb_hiwat;
3278 #endif
3279 } else {
3280 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
3281 if (tlen > 0) {
3282 if ((thflags & TH_FIN) != 0) {
3283 log(LOG_DEBUG, "%s; %s: %s: "
3284 "Received %d bytes of data and FIN "
3285 "after having received a FIN, "
3286 "just dropping both\n",
3287 s, __func__,
3288 tcpstates[tp->t_state], tlen);
3289 } else {
3290 log(LOG_DEBUG, "%s; %s: %s: "
3291 "Received %d bytes of data "
3292 "after having received a FIN, "
3293 "just dropping it\n",
3294 s, __func__,
3295 tcpstates[tp->t_state], tlen);
3296 }
3297 } else {
3298 if ((thflags & TH_FIN) != 0) {
3299 log(LOG_DEBUG, "%s; %s: %s: "
3300 "Received FIN "
3301 "after having received a FIN, "
3302 "just dropping it\n",
3303 s, __func__,
3304 tcpstates[tp->t_state]);
3305 }
3306 }
3307 free(s, M_TCPLOG);
3308 }
3309 m_freem(m);
3310 thflags &= ~TH_FIN;
3311 }
3312
3313 /*
3314 * If FIN is received ACK the FIN and let the user know
3315 * that the connection is closing.
3316 */
3317 if (thflags & TH_FIN) {
3318 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3319 /* The socket upcall is handled by socantrcvmore. */
3320 socantrcvmore(so);
3321 /*
3322 * If connection is half-synchronized
3323 * (ie NEEDSYN flag on) then delay ACK,
3324 * so it may be piggybacked when SYN is sent.
3325 * Otherwise, since we received a FIN then no
3326 * more input can be expected, send ACK now.
3327 */
3328 if (tp->t_flags & TF_NEEDSYN)
3329 tp->t_flags |= TF_DELACK;
3330 else
3331 tp->t_flags |= TF_ACKNOW;
3332 tp->rcv_nxt++;
3333 }
3334 switch (tp->t_state) {
3335 /*
3336 * In SYN_RECEIVED and ESTABLISHED STATES
3337 * enter the CLOSE_WAIT state.
3338 */
3339 case TCPS_SYN_RECEIVED:
3340 tp->t_starttime = ticks;
3341 /* FALLTHROUGH */
3342 case TCPS_ESTABLISHED:
3343 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3344 break;
3345
3346 /*
3347 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3348 * enter the CLOSING state.
3349 */
3350 case TCPS_FIN_WAIT_1:
3351 tcp_state_change(tp, TCPS_CLOSING);
3352 break;
3353
3354 /*
3355 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3356 * starting the time-wait timer, turning off the other
3357 * standard timers.
3358 */
3359 case TCPS_FIN_WAIT_2:
3360 tcp_twstart(tp);
3361 return;
3362 }
3363 }
3364 TCP_PROBE3(debug__input, tp, th, m);
3365
3366 /*
3367 * Return any desired output.
3368 */
3369 if (needoutput || (tp->t_flags & TF_ACKNOW)) {
3370 (void) tcp_output(tp);
3371 }
3372 check_delack:
3373 INP_WLOCK_ASSERT(inp);
3374
3375 if (tp->t_flags & TF_DELACK) {
3376 tp->t_flags &= ~TF_DELACK;
3377 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3378 }
3379 INP_WUNLOCK(inp);
3380 return;
3381
3382 dropafterack:
3383 /*
3384 * Generate an ACK dropping incoming segment if it occupies
3385 * sequence space, where the ACK reflects our state.
3386 *
3387 * We can now skip the test for the RST flag since all
3388 * paths to this code happen after packets containing
3389 * RST have been dropped.
3390 *
3391 * In the SYN-RECEIVED state, don't send an ACK unless the
3392 * segment we received passes the SYN-RECEIVED ACK test.
3393 * If it fails send a RST. This breaks the loop in the
3394 * "LAND" DoS attack, and also prevents an ACK storm
3395 * between two listening ports that have been sent forged
3396 * SYN segments, each with the source address of the other.
3397 */
3398 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3399 (SEQ_GT(tp->snd_una, th->th_ack) ||
3400 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3401 rstreason = BANDLIM_RST_OPENPORT;
3402 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3403 goto dropwithreset;
3404 }
3405 TCP_PROBE3(debug__input, tp, th, m);
3406 tp->t_flags |= TF_ACKNOW;
3407 (void) tcp_output(tp);
3408 INP_WUNLOCK(inp);
3409 m_freem(m);
3410 return;
3411
3412 dropwithreset:
3413 if (tp != NULL) {
3414 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3415 INP_WUNLOCK(inp);
3416 } else
3417 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3418 return;
3419
3420 drop:
3421 /*
3422 * Drop space held by incoming segment and return.
3423 */
3424 TCP_PROBE3(debug__input, tp, th, m);
3425 if (tp != NULL) {
3426 INP_WUNLOCK(inp);
3427 }
3428 m_freem(m);
3429 }
3430
3431 /*
3432 * Issue RST and make ACK acceptable to originator of segment.
3433 * The mbuf must still include the original packet header.
3434 * tp may be NULL.
3435 */
3436 void
tcp_dropwithreset(struct mbuf * m,struct tcphdr * th,struct tcpcb * tp,int tlen,int rstreason)3437 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3438 int tlen, int rstreason)
3439 {
3440 #ifdef INET
3441 struct ip *ip;
3442 #endif
3443 #ifdef INET6
3444 struct ip6_hdr *ip6;
3445 #endif
3446
3447 if (tp != NULL) {
3448 INP_LOCK_ASSERT(tptoinpcb(tp));
3449 }
3450
3451 /* Don't bother if destination was broadcast/multicast. */
3452 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3453 goto drop;
3454 #ifdef INET6
3455 if (mtod(m, struct ip *)->ip_v == 6) {
3456 ip6 = mtod(m, struct ip6_hdr *);
3457 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3458 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3459 goto drop;
3460 /* IPv6 anycast check is done at tcp6_input() */
3461 }
3462 #endif
3463 #if defined(INET) && defined(INET6)
3464 else
3465 #endif
3466 #ifdef INET
3467 {
3468 ip = mtod(m, struct ip *);
3469 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3470 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3471 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3472 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3473 goto drop;
3474 }
3475 #endif
3476
3477 /* Perform bandwidth limiting. */
3478 if (badport_bandlim(rstreason) < 0)
3479 goto drop;
3480
3481 /* tcp_respond consumes the mbuf chain. */
3482 if (tcp_get_flags(th) & TH_ACK) {
3483 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3484 th->th_ack, TH_RST);
3485 } else {
3486 if (tcp_get_flags(th) & TH_SYN)
3487 tlen++;
3488 if (tcp_get_flags(th) & TH_FIN)
3489 tlen++;
3490 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3491 (tcp_seq)0, TH_RST|TH_ACK);
3492 }
3493 return;
3494 drop:
3495 m_freem(m);
3496 }
3497
3498 /*
3499 * Parse TCP options and place in tcpopt.
3500 */
3501 void
tcp_dooptions(struct tcpopt * to,u_char * cp,int cnt,int flags)3502 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3503 {
3504 int opt, optlen;
3505
3506 to->to_flags = 0;
3507 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3508 opt = cp[0];
3509 if (opt == TCPOPT_EOL)
3510 break;
3511 if (opt == TCPOPT_NOP)
3512 optlen = 1;
3513 else {
3514 if (cnt < 2)
3515 break;
3516 optlen = cp[1];
3517 if (optlen < 2 || optlen > cnt)
3518 break;
3519 }
3520 switch (opt) {
3521 case TCPOPT_MAXSEG:
3522 if (optlen != TCPOLEN_MAXSEG)
3523 continue;
3524 if (!(flags & TO_SYN))
3525 continue;
3526 to->to_flags |= TOF_MSS;
3527 bcopy((char *)cp + 2,
3528 (char *)&to->to_mss, sizeof(to->to_mss));
3529 to->to_mss = ntohs(to->to_mss);
3530 break;
3531 case TCPOPT_WINDOW:
3532 if (optlen != TCPOLEN_WINDOW)
3533 continue;
3534 if (!(flags & TO_SYN))
3535 continue;
3536 to->to_flags |= TOF_SCALE;
3537 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3538 break;
3539 case TCPOPT_TIMESTAMP:
3540 if (optlen != TCPOLEN_TIMESTAMP)
3541 continue;
3542 to->to_flags |= TOF_TS;
3543 bcopy((char *)cp + 2,
3544 (char *)&to->to_tsval, sizeof(to->to_tsval));
3545 to->to_tsval = ntohl(to->to_tsval);
3546 bcopy((char *)cp + 6,
3547 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3548 to->to_tsecr = ntohl(to->to_tsecr);
3549 break;
3550 case TCPOPT_SIGNATURE:
3551 /*
3552 * In order to reply to a host which has set the
3553 * TCP_SIGNATURE option in its initial SYN, we have
3554 * to record the fact that the option was observed
3555 * here for the syncache code to perform the correct
3556 * response.
3557 */
3558 if (optlen != TCPOLEN_SIGNATURE)
3559 continue;
3560 to->to_flags |= TOF_SIGNATURE;
3561 to->to_signature = cp + 2;
3562 break;
3563 case TCPOPT_SACK_PERMITTED:
3564 if (optlen != TCPOLEN_SACK_PERMITTED)
3565 continue;
3566 if (!(flags & TO_SYN))
3567 continue;
3568 if (!V_tcp_do_sack)
3569 continue;
3570 to->to_flags |= TOF_SACKPERM;
3571 break;
3572 case TCPOPT_SACK:
3573 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3574 continue;
3575 if (flags & TO_SYN)
3576 continue;
3577 to->to_flags |= TOF_SACK;
3578 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3579 to->to_sacks = cp + 2;
3580 TCPSTAT_INC(tcps_sack_rcv_blocks);
3581 break;
3582 case TCPOPT_FAST_OPEN:
3583 /*
3584 * Cookie length validation is performed by the
3585 * server side cookie checking code or the client
3586 * side cookie cache update code.
3587 */
3588 if (!(flags & TO_SYN))
3589 continue;
3590 if (!V_tcp_fastopen_client_enable &&
3591 !V_tcp_fastopen_server_enable)
3592 continue;
3593 to->to_flags |= TOF_FASTOPEN;
3594 to->to_tfo_len = optlen - 2;
3595 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3596 break;
3597 default:
3598 continue;
3599 }
3600 }
3601 }
3602
3603 /*
3604 * Pull out of band byte out of a segment so
3605 * it doesn't appear in the user's data queue.
3606 * It is still reflected in the segment length for
3607 * sequencing purposes.
3608 */
3609 void
tcp_pulloutofband(struct socket * so,struct tcphdr * th,struct mbuf * m,int off)3610 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3611 int off)
3612 {
3613 int cnt = off + th->th_urp - 1;
3614
3615 while (cnt >= 0) {
3616 if (m->m_len > cnt) {
3617 char *cp = mtod(m, caddr_t) + cnt;
3618 struct tcpcb *tp = sototcpcb(so);
3619
3620 INP_WLOCK_ASSERT(tptoinpcb(tp));
3621
3622 tp->t_iobc = *cp;
3623 tp->t_oobflags |= TCPOOB_HAVEDATA;
3624 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3625 m->m_len--;
3626 if (m->m_flags & M_PKTHDR)
3627 m->m_pkthdr.len--;
3628 return;
3629 }
3630 cnt -= m->m_len;
3631 m = m->m_next;
3632 if (m == NULL)
3633 break;
3634 }
3635 panic("tcp_pulloutofband");
3636 }
3637
3638 /*
3639 * Collect new round-trip time estimate
3640 * and update averages and current timeout.
3641 */
3642 void
tcp_xmit_timer(struct tcpcb * tp,int rtt)3643 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3644 {
3645 int delta;
3646
3647 INP_WLOCK_ASSERT(tptoinpcb(tp));
3648
3649 TCPSTAT_INC(tcps_rttupdated);
3650 if (tp->t_rttupdated < UCHAR_MAX)
3651 tp->t_rttupdated++;
3652 #ifdef STATS
3653 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3654 imax(0, rtt * 1000 / hz));
3655 #endif
3656 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3657 /*
3658 * srtt is stored as fixed point with 5 bits after the
3659 * binary point (i.e., scaled by 8). The following magic
3660 * is equivalent to the smoothing algorithm in rfc793 with
3661 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3662 * point). Adjust rtt to origin 0.
3663 */
3664 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3665 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3666
3667 if ((tp->t_srtt += delta) <= 0)
3668 tp->t_srtt = 1;
3669
3670 /*
3671 * We accumulate a smoothed rtt variance (actually, a
3672 * smoothed mean difference), then set the retransmit
3673 * timer to smoothed rtt + 4 times the smoothed variance.
3674 * rttvar is stored as fixed point with 4 bits after the
3675 * binary point (scaled by 16). The following is
3676 * equivalent to rfc793 smoothing with an alpha of .75
3677 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3678 * rfc793's wired-in beta.
3679 */
3680 if (delta < 0)
3681 delta = -delta;
3682 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3683 if ((tp->t_rttvar += delta) <= 0)
3684 tp->t_rttvar = 1;
3685 } else {
3686 /*
3687 * No rtt measurement yet - use the unsmoothed rtt.
3688 * Set the variance to half the rtt (so our first
3689 * retransmit happens at 3*rtt).
3690 */
3691 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3692 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3693 }
3694 tp->t_rtttime = 0;
3695 tp->t_rxtshift = 0;
3696
3697 /*
3698 * the retransmit should happen at rtt + 4 * rttvar.
3699 * Because of the way we do the smoothing, srtt and rttvar
3700 * will each average +1/2 tick of bias. When we compute
3701 * the retransmit timer, we want 1/2 tick of rounding and
3702 * 1 extra tick because of +-1/2 tick uncertainty in the
3703 * firing of the timer. The bias will give us exactly the
3704 * 1.5 tick we need. But, because the bias is
3705 * statistical, we have to test that we don't drop below
3706 * the minimum feasible timer (which is 2 ticks).
3707 */
3708 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3709 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3710
3711 /*
3712 * We received an ack for a packet that wasn't retransmitted;
3713 * it is probably safe to discard any error indications we've
3714 * received recently. This isn't quite right, but close enough
3715 * for now (a route might have failed after we sent a segment,
3716 * and the return path might not be symmetrical).
3717 */
3718 tp->t_softerror = 0;
3719 }
3720
3721 /*
3722 * Determine a reasonable value for maxseg size.
3723 * If the route is known, check route for mtu.
3724 * If none, use an mss that can be handled on the outgoing interface
3725 * without forcing IP to fragment. If no route is found, route has no mtu,
3726 * or the destination isn't local, use a default, hopefully conservative
3727 * size (usually 512 or the default IP max size, but no more than the mtu
3728 * of the interface), as we can't discover anything about intervening
3729 * gateways or networks. We also initialize the congestion/slow start
3730 * window to be a single segment if the destination isn't local.
3731 * While looking at the routing entry, we also initialize other path-dependent
3732 * parameters from pre-set or cached values in the routing entry.
3733 *
3734 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3735 * IP options, e.g. IPSEC data, since length of this data may vary, and
3736 * thus it is calculated for every segment separately in tcp_output().
3737 *
3738 * NOTE that this routine is only called when we process an incoming
3739 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3740 * settings are handled in tcp_mssopt().
3741 */
3742 void
tcp_mss_update(struct tcpcb * tp,int offer,int mtuoffer,struct hc_metrics_lite * metricptr,struct tcp_ifcap * cap)3743 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3744 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3745 {
3746 int mss = 0;
3747 uint32_t maxmtu = 0;
3748 struct inpcb *inp = tptoinpcb(tp);
3749 struct hc_metrics_lite metrics;
3750 #ifdef INET6
3751 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3752 size_t min_protoh = isipv6 ?
3753 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3754 sizeof (struct tcpiphdr);
3755 #else
3756 size_t min_protoh = sizeof(struct tcpiphdr);
3757 #endif
3758
3759 INP_WLOCK_ASSERT(inp);
3760
3761 if (tp->t_port)
3762 min_protoh += V_tcp_udp_tunneling_overhead;
3763 if (mtuoffer != -1) {
3764 KASSERT(offer == -1, ("%s: conflict", __func__));
3765 offer = mtuoffer - min_protoh;
3766 }
3767
3768 /* Initialize. */
3769 #ifdef INET6
3770 if (isipv6) {
3771 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3772 tp->t_maxseg = V_tcp_v6mssdflt;
3773 }
3774 #endif
3775 #if defined(INET) && defined(INET6)
3776 else
3777 #endif
3778 #ifdef INET
3779 {
3780 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3781 tp->t_maxseg = V_tcp_mssdflt;
3782 }
3783 #endif
3784
3785 /*
3786 * No route to sender, stay with default mss and return.
3787 */
3788 if (maxmtu == 0) {
3789 /*
3790 * In case we return early we need to initialize metrics
3791 * to a defined state as tcp_hc_get() would do for us
3792 * if there was no cache hit.
3793 */
3794 if (metricptr != NULL)
3795 bzero(metricptr, sizeof(struct hc_metrics_lite));
3796 return;
3797 }
3798
3799 /* What have we got? */
3800 switch (offer) {
3801 case 0:
3802 /*
3803 * Offer == 0 means that there was no MSS on the SYN
3804 * segment, in this case we use tcp_mssdflt as
3805 * already assigned to t_maxseg above.
3806 */
3807 offer = tp->t_maxseg;
3808 break;
3809
3810 case -1:
3811 /*
3812 * Offer == -1 means that we didn't receive SYN yet.
3813 */
3814 /* FALLTHROUGH */
3815
3816 default:
3817 /*
3818 * Prevent DoS attack with too small MSS. Round up
3819 * to at least minmss.
3820 */
3821 offer = max(offer, V_tcp_minmss);
3822 }
3823
3824 /*
3825 * rmx information is now retrieved from tcp_hostcache.
3826 */
3827 tcp_hc_get(&inp->inp_inc, &metrics);
3828 if (metricptr != NULL)
3829 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3830
3831 /*
3832 * If there's a discovered mtu in tcp hostcache, use it.
3833 * Else, use the link mtu.
3834 */
3835 if (metrics.rmx_mtu)
3836 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3837 else {
3838 #ifdef INET6
3839 if (isipv6) {
3840 mss = maxmtu - min_protoh;
3841 if (!V_path_mtu_discovery &&
3842 !in6_localaddr(&inp->in6p_faddr))
3843 mss = min(mss, V_tcp_v6mssdflt);
3844 }
3845 #endif
3846 #if defined(INET) && defined(INET6)
3847 else
3848 #endif
3849 #ifdef INET
3850 {
3851 mss = maxmtu - min_protoh;
3852 if (!V_path_mtu_discovery &&
3853 !in_localaddr(inp->inp_faddr))
3854 mss = min(mss, V_tcp_mssdflt);
3855 }
3856 #endif
3857 /*
3858 * XXX - The above conditional (mss = maxmtu - min_protoh)
3859 * probably violates the TCP spec.
3860 * The problem is that, since we don't know the
3861 * other end's MSS, we are supposed to use a conservative
3862 * default. But, if we do that, then MTU discovery will
3863 * never actually take place, because the conservative
3864 * default is much less than the MTUs typically seen
3865 * on the Internet today. For the moment, we'll sweep
3866 * this under the carpet.
3867 *
3868 * The conservative default might not actually be a problem
3869 * if the only case this occurs is when sending an initial
3870 * SYN with options and data to a host we've never talked
3871 * to before. Then, they will reply with an MSS value which
3872 * will get recorded and the new parameters should get
3873 * recomputed. For Further Study.
3874 */
3875 }
3876 mss = min(mss, offer);
3877
3878 /*
3879 * Sanity check: make sure that maxseg will be large
3880 * enough to allow some data on segments even if the
3881 * all the option space is used (40bytes). Otherwise
3882 * funny things may happen in tcp_output.
3883 *
3884 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3885 */
3886 mss = max(mss, 64);
3887
3888 tp->t_maxseg = mss;
3889 if (tp->t_maxseg < V_tcp_mssdflt) {
3890 /*
3891 * The MSS is so small we should not process incoming
3892 * SACK's since we are subject to attack in such a
3893 * case.
3894 */
3895 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3896 } else {
3897 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3898 }
3899
3900 }
3901
3902 void
tcp_mss(struct tcpcb * tp,int offer)3903 tcp_mss(struct tcpcb *tp, int offer)
3904 {
3905 int mss;
3906 uint32_t bufsize;
3907 struct inpcb *inp = tptoinpcb(tp);
3908 struct socket *so;
3909 struct hc_metrics_lite metrics;
3910 struct tcp_ifcap cap;
3911
3912 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3913
3914 bzero(&cap, sizeof(cap));
3915 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3916
3917 mss = tp->t_maxseg;
3918
3919 /*
3920 * If there's a pipesize, change the socket buffer to that size,
3921 * don't change if sb_hiwat is different than default (then it
3922 * has been changed on purpose with setsockopt).
3923 * Make the socket buffers an integral number of mss units;
3924 * if the mss is larger than the socket buffer, decrease the mss.
3925 */
3926 so = inp->inp_socket;
3927 SOCKBUF_LOCK(&so->so_snd);
3928 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3929 bufsize = metrics.rmx_sendpipe;
3930 else
3931 bufsize = so->so_snd.sb_hiwat;
3932 if (bufsize < mss)
3933 mss = bufsize;
3934 else {
3935 bufsize = roundup(bufsize, mss);
3936 if (bufsize > sb_max)
3937 bufsize = sb_max;
3938 if (bufsize > so->so_snd.sb_hiwat)
3939 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3940 }
3941 SOCKBUF_UNLOCK(&so->so_snd);
3942 /*
3943 * Sanity check: make sure that maxseg will be large
3944 * enough to allow some data on segments even if the
3945 * all the option space is used (40bytes). Otherwise
3946 * funny things may happen in tcp_output.
3947 *
3948 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3949 */
3950 tp->t_maxseg = max(mss, 64);
3951 if (tp->t_maxseg < V_tcp_mssdflt) {
3952 /*
3953 * The MSS is so small we should not process incoming
3954 * SACK's since we are subject to attack in such a
3955 * case.
3956 */
3957 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3958 } else {
3959 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3960 }
3961
3962 SOCKBUF_LOCK(&so->so_rcv);
3963 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3964 bufsize = metrics.rmx_recvpipe;
3965 else
3966 bufsize = so->so_rcv.sb_hiwat;
3967 if (bufsize > mss) {
3968 bufsize = roundup(bufsize, mss);
3969 if (bufsize > sb_max)
3970 bufsize = sb_max;
3971 if (bufsize > so->so_rcv.sb_hiwat)
3972 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
3973 }
3974 SOCKBUF_UNLOCK(&so->so_rcv);
3975
3976 /* Check the interface for TSO capabilities. */
3977 if (cap.ifcap & CSUM_TSO) {
3978 tp->t_flags |= TF_TSO;
3979 tp->t_tsomax = cap.tsomax;
3980 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3981 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3982 }
3983 }
3984
3985 /*
3986 * Determine the MSS option to send on an outgoing SYN.
3987 */
3988 int
tcp_mssopt(struct in_conninfo * inc)3989 tcp_mssopt(struct in_conninfo *inc)
3990 {
3991 int mss = 0;
3992 uint32_t thcmtu = 0;
3993 uint32_t maxmtu = 0;
3994 size_t min_protoh;
3995
3996 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3997
3998 #ifdef INET6
3999 if (inc->inc_flags & INC_ISIPV6) {
4000 mss = V_tcp_v6mssdflt;
4001 maxmtu = tcp_maxmtu6(inc, NULL);
4002 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
4003 }
4004 #endif
4005 #if defined(INET) && defined(INET6)
4006 else
4007 #endif
4008 #ifdef INET
4009 {
4010 mss = V_tcp_mssdflt;
4011 maxmtu = tcp_maxmtu(inc, NULL);
4012 min_protoh = sizeof(struct tcpiphdr);
4013 }
4014 #endif
4015 #if defined(INET6) || defined(INET)
4016 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
4017 #endif
4018
4019 if (maxmtu && thcmtu)
4020 mss = min(maxmtu, thcmtu) - min_protoh;
4021 else if (maxmtu || thcmtu)
4022 mss = max(maxmtu, thcmtu) - min_protoh;
4023
4024 return (mss);
4025 }
4026
4027 void
tcp_do_prr_ack(struct tcpcb * tp,struct tcphdr * th,struct tcpopt * to,sackstatus_t sack_changed,u_int * maxsegp)4028 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to,
4029 sackstatus_t sack_changed, u_int *maxsegp)
4030 {
4031 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
4032 u_int maxseg;
4033
4034 INP_WLOCK_ASSERT(tptoinpcb(tp));
4035
4036 if (*maxsegp == 0) {
4037 *maxsegp = tcp_maxseg(tp);
4038 }
4039 maxseg = *maxsegp;
4040 /*
4041 * Compute the amount of data that this ACK is indicating
4042 * (del_data) and an estimate of how many bytes are in the
4043 * network.
4044 */
4045 if (tcp_is_sack_recovery(tp, to) ||
4046 (IN_CONGRECOVERY(tp->t_flags) &&
4047 !IN_FASTRECOVERY(tp->t_flags))) {
4048 del_data = tp->sackhint.delivered_data;
4049 if (V_tcp_do_newsack)
4050 pipe = tcp_compute_pipe(tp);
4051 else
4052 pipe = (tp->snd_nxt - tp->snd_fack) +
4053 tp->sackhint.sack_bytes_rexmit;
4054 } else {
4055 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
4056 tp->snd_recover - tp->snd_una)) {
4057 del_data = maxseg;
4058 }
4059 pipe = imax(0, tp->snd_max - tp->snd_una -
4060 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
4061 }
4062 tp->sackhint.prr_delivered += del_data;
4063 /*
4064 * Proportional Rate Reduction
4065 */
4066 if (pipe >= tp->snd_ssthresh) {
4067 if (tp->sackhint.recover_fs == 0)
4068 tp->sackhint.recover_fs =
4069 imax(1, tp->snd_nxt - tp->snd_una);
4070 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
4071 tp->snd_ssthresh, tp->sackhint.recover_fs) -
4072 tp->sackhint.prr_out + maxseg - 1;
4073 } else {
4074 /*
4075 * PRR 6937bis heuristic:
4076 * - A partial ack without SACK block beneath snd_recover
4077 * indicates further loss.
4078 * - An SACK scoreboard update adding a new hole indicates
4079 * further loss, so be conservative and send at most one
4080 * segment.
4081 * - Prevent ACK splitting attacks, by being conservative
4082 * when no new data is acked.
4083 */
4084 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) {
4085 limit = tp->sackhint.prr_delivered -
4086 tp->sackhint.prr_out;
4087 } else {
4088 limit = imax(tp->sackhint.prr_delivered -
4089 tp->sackhint.prr_out, del_data) +
4090 maxseg;
4091 }
4092 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4093 }
4094 snd_cnt = imax(snd_cnt, 0) / maxseg;
4095 /*
4096 * Send snd_cnt new data into the network in response to this ack.
4097 * If there is going to be a SACK retransmission, adjust snd_cwnd
4098 * accordingly.
4099 */
4100 if (IN_FASTRECOVERY(tp->t_flags)) {
4101 if (tcp_is_sack_recovery(tp, to)) {
4102 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4103 tp->sackhint.sack_bytes_rexmit +
4104 (snd_cnt * maxseg);
4105 } else {
4106 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4107 (snd_cnt * maxseg);
4108 }
4109 } else if (IN_CONGRECOVERY(tp->t_flags)) {
4110 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4111 }
4112 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4113 }
4114
4115 /*
4116 * On a partial ack arrives, force the retransmission of the
4117 * next unacknowledged segment. Do not clear tp->t_dupacks.
4118 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4119 * be started again.
4120 */
4121 void
tcp_newreno_partial_ack(struct tcpcb * tp,struct tcphdr * th)4122 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4123 {
4124 tcp_seq onxt = tp->snd_nxt;
4125 uint32_t ocwnd = tp->snd_cwnd;
4126 u_int maxseg = tcp_maxseg(tp);
4127
4128 INP_WLOCK_ASSERT(tptoinpcb(tp));
4129
4130 tcp_timer_activate(tp, TT_REXMT, 0);
4131 tp->t_rtttime = 0;
4132 tp->snd_nxt = th->th_ack;
4133 /*
4134 * Set snd_cwnd to one segment beyond acknowledged offset.
4135 * (tp->snd_una has not yet been updated when this function is called.)
4136 */
4137 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4138 tp->t_flags |= TF_ACKNOW;
4139 (void) tcp_output(tp);
4140 tp->snd_cwnd = ocwnd;
4141 if (SEQ_GT(onxt, tp->snd_nxt))
4142 tp->snd_nxt = onxt;
4143 /*
4144 * Partial window deflation. Relies on fact that tp->snd_una
4145 * not updated yet.
4146 */
4147 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4148 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4149 else
4150 tp->snd_cwnd = 0;
4151 tp->snd_cwnd += maxseg;
4152 }
4153
4154 int
tcp_compute_pipe(struct tcpcb * tp)4155 tcp_compute_pipe(struct tcpcb *tp)
4156 {
4157 if (tp->t_fb->tfb_compute_pipe == NULL) {
4158 return (tp->snd_max - tp->snd_una +
4159 tp->sackhint.sack_bytes_rexmit -
4160 tp->sackhint.sacked_bytes -
4161 tp->sackhint.lost_bytes);
4162 } else {
4163 return((*tp->t_fb->tfb_compute_pipe)(tp));
4164 }
4165 }
4166
4167 uint32_t
tcp_compute_initwnd(uint32_t maxseg)4168 tcp_compute_initwnd(uint32_t maxseg)
4169 {
4170 /*
4171 * Calculate the Initial Window, also used as Restart Window
4172 *
4173 * RFC5681 Section 3.1 specifies the default conservative values.
4174 * RFC3390 specifies slightly more aggressive values.
4175 * RFC6928 increases it to ten segments.
4176 * Support for user specified value for initial flight size.
4177 */
4178 if (V_tcp_initcwnd_segments)
4179 return min(V_tcp_initcwnd_segments * maxseg,
4180 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4181 else if (V_tcp_do_rfc3390)
4182 return min(4 * maxseg, max(2 * maxseg, 4380));
4183 else {
4184 /* Per RFC5681 Section 3.1 */
4185 if (maxseg > 2190)
4186 return (2 * maxseg);
4187 else if (maxseg > 1095)
4188 return (3 * maxseg);
4189 else
4190 return (4 * maxseg);
4191 }
4192 }
4193