1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2019 Joyent, Inc.
26 * Copyright (c) 2014, 2016 by Delphix. All rights reserved.
27 */
28
29 /* This file contains all TCP input processing functions. */
30
31 #include <sys/types.h>
32 #include <sys/stream.h>
33 #include <sys/strsun.h>
34 #include <sys/strsubr.h>
35 #include <sys/stropts.h>
36 #include <sys/strlog.h>
37 #define _SUN_TPI_VERSION 2
38 #include <sys/tihdr.h>
39 #include <sys/suntpi.h>
40 #include <sys/xti_inet.h>
41 #include <sys/squeue_impl.h>
42 #include <sys/squeue.h>
43 #include <sys/tsol/tnet.h>
44
45 #include <inet/common.h>
46 #include <inet/ip.h>
47 #include <inet/tcp.h>
48 #include <inet/tcp_impl.h>
49 #include <inet/tcp_cluster.h>
50 #include <inet/proto_set.h>
51 #include <inet/ipsec_impl.h>
52
53 /*
54 * RFC7323-recommended phrasing of TSTAMP option, for easier parsing
55 */
56
57 #ifdef _BIG_ENDIAN
58 #define TCPOPT_NOP_NOP_TSTAMP ((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | \
59 (TCPOPT_TSTAMP << 8) | 10)
60 #else
61 #define TCPOPT_NOP_NOP_TSTAMP ((10 << 24) | (TCPOPT_TSTAMP << 16) | \
62 (TCPOPT_NOP << 8) | TCPOPT_NOP)
63 #endif
64
65 /*
66 * PAWS needs a timer for 24 days. This is the number of ticks in 24 days
67 */
68 #define PAWS_TIMEOUT ((clock_t)(24*24*60*60*hz))
69
70 /*
71 * Since tcp_listener is not cleared atomically with tcp_detached
72 * being cleared we need this extra bit to tell a detached connection
73 * apart from one that is in the process of being accepted.
74 */
75 #define TCP_IS_DETACHED_NONEAGER(tcp) \
76 (TCP_IS_DETACHED(tcp) && \
77 (!(tcp)->tcp_hard_binding))
78
79 /*
80 * Steps to do when a tcp_t moves to TIME-WAIT state.
81 *
82 * This connection is done, we don't need to account for it. Decrement
83 * the listener connection counter if needed.
84 *
85 * Decrement the connection counter of the stack. Note that this counter
86 * is per CPU. So the total number of connections in a stack is the sum of all
87 * of them. Since there is no lock for handling all of them exclusively, the
88 * resulting sum is only an approximation.
89 *
90 * Unconditionally clear the exclusive binding bit so this TIME-WAIT
91 * connection won't interfere with new ones.
92 *
93 * Start the TIME-WAIT timer. If upper layer has not closed the connection,
94 * the timer is handled within the context of this tcp_t. When the timer
95 * fires, tcp_clean_death() is called. If upper layer closes the connection
96 * during this period, tcp_time_wait_append() will be called to add this
97 * tcp_t to the global TIME-WAIT list. Note that this means that the
98 * actual wait time in TIME-WAIT state will be longer than the
99 * tcps_time_wait_interval since the period before upper layer closes the
100 * connection is not accounted for when tcp_time_wait_append() is called.
101 *
102 * If upper layer has closed the connection, call tcp_time_wait_append()
103 * directly.
104 *
105 */
106 #define SET_TIME_WAIT(tcps, tcp, connp) \
107 { \
108 (tcp)->tcp_state = TCPS_TIME_WAIT; \
109 if ((tcp)->tcp_listen_cnt != NULL) \
110 TCP_DECR_LISTEN_CNT(tcp); \
111 atomic_dec_64( \
112 (uint64_t *)&(tcps)->tcps_sc[CPU->cpu_seqid]->tcp_sc_conn_cnt); \
113 (connp)->conn_exclbind = 0; \
114 if (!TCP_IS_DETACHED(tcp)) { \
115 TCP_TIMER_RESTART(tcp, (tcps)->tcps_time_wait_interval); \
116 } else { \
117 tcp_time_wait_append(tcp); \
118 TCP_DBGSTAT(tcps, tcp_rput_time_wait); \
119 } \
120 }
121
122 /*
123 * If tcp_drop_ack_unsent_cnt is greater than 0, when TCP receives more
124 * than tcp_drop_ack_unsent_cnt number of ACKs which acknowledge unsent
125 * data, TCP will not respond with an ACK. RFC 793 requires that
126 * TCP responds with an ACK for such a bogus ACK. By not following
127 * the RFC, we prevent TCP from getting into an ACK storm if somehow
128 * an attacker successfully spoofs an acceptable segment to our
129 * peer; or when our peer is "confused."
130 */
131 static uint32_t tcp_drop_ack_unsent_cnt = 10;
132
133 /*
134 * To protect TCP against attacker using a small window and requesting
135 * large amount of data (DoS attack by conuming memory), TCP checks the
136 * window advertised in the last ACK of the 3-way handshake. TCP uses
137 * the tcp_mss (the size of one packet) value for comparion. The window
138 * should be larger than tcp_mss. But while a sane TCP should advertise
139 * a receive window larger than or equal to 4*MSS to avoid stop and go
140 * tarrfic, not all TCP stacks do that. This is especially true when
141 * tcp_mss is a big value.
142 *
143 * To work around this issue, an additional fixed value for comparison
144 * is also used. If the advertised window is smaller than both tcp_mss
145 * and tcp_init_wnd_chk, the ACK is considered as invalid. So for large
146 * tcp_mss value (say, 8K), a window larger than tcp_init_wnd_chk but
147 * smaller than 8K is considered to be OK.
148 */
149 static uint32_t tcp_init_wnd_chk = 4096;
150
151 /* Process ICMP source quench message or not. */
152 static boolean_t tcp_icmp_source_quench = B_FALSE;
153
154 static boolean_t tcp_outbound_squeue_switch = B_FALSE;
155
156 static mblk_t *tcp_conn_create_v4(conn_t *, conn_t *, mblk_t *,
157 ip_recv_attr_t *);
158 static mblk_t *tcp_conn_create_v6(conn_t *, conn_t *, mblk_t *,
159 ip_recv_attr_t *);
160 static boolean_t tcp_drop_q0(tcp_t *);
161 static void tcp_icmp_error_ipv6(tcp_t *, mblk_t *, ip_recv_attr_t *);
162 static mblk_t *tcp_input_add_ancillary(tcp_t *, mblk_t *, ip_pkt_t *,
163 ip_recv_attr_t *);
164 static void tcp_input_listener(void *, mblk_t *, void *, ip_recv_attr_t *);
165 static void tcp_process_options(tcp_t *, tcpha_t *);
166 static mblk_t *tcp_reass(tcp_t *, mblk_t *, uint32_t);
167 static void tcp_reass_elim_overlap(tcp_t *, mblk_t *);
168 static void tcp_rsrv_input(void *, mblk_t *, void *, ip_recv_attr_t *);
169 static void tcp_set_rto(tcp_t *, hrtime_t);
170 static void tcp_setcred_data(mblk_t *, ip_recv_attr_t *);
171
172 /*
173 * Set the MSS associated with a particular tcp based on its current value,
174 * and a new one passed in. Observe minimums and maximums, and reset other
175 * state variables that we want to view as multiples of MSS.
176 *
177 * The value of MSS could be either increased or descreased.
178 */
179 void
180 tcp_mss_set(tcp_t *tcp, uint32_t mss)
181 {
182 uint32_t mss_max;
183 tcp_stack_t *tcps = tcp->tcp_tcps;
184 conn_t *connp = tcp->tcp_connp;
185
186 if (connp->conn_ipversion == IPV4_VERSION)
187 mss_max = tcps->tcps_mss_max_ipv4;
188 else
189 mss_max = tcps->tcps_mss_max_ipv6;
190
191 if (mss < tcps->tcps_mss_min)
192 mss = tcps->tcps_mss_min;
193 if (mss > mss_max)
194 mss = mss_max;
195 /*
196 * Unless naglim has been set by our client to
197 * a non-mss value, force naglim to track mss.
198 * This can help to aggregate small writes.
199 */
200 if (mss < tcp->tcp_naglim || tcp->tcp_mss == tcp->tcp_naglim)
201 tcp->tcp_naglim = mss;
202 /*
203 * TCP should be able to buffer at least 4 MSS data for obvious
204 * performance reason.
205 */
206 if ((mss << 2) > connp->conn_sndbuf)
207 connp->conn_sndbuf = mss << 2;
208
209 /*
210 * Set the send lowater to at least twice of MSS.
211 */
212 if ((mss << 1) > connp->conn_sndlowat)
213 connp->conn_sndlowat = mss << 1;
214
215 /*
216 * Update tcp_cwnd according to the new value of MSS. Keep the
217 * previous ratio to preserve the transmit rate.
218 */
219 tcp->tcp_cwnd = (tcp->tcp_cwnd / tcp->tcp_mss) * mss;
220 tcp->tcp_cwnd_cnt = 0;
221
222 tcp->tcp_mss = mss;
223 (void) tcp_maxpsz_set(tcp, B_TRUE);
224 }
225
226 /*
227 * Extract option values from a tcp header. We put any found values into the
228 * tcpopt struct and return a bitmask saying which options were found.
229 */
230 int
231 tcp_parse_options(tcpha_t *tcpha, tcp_opt_t *tcpopt)
232 {
233 uchar_t *endp;
234 int len;
235 uint32_t mss;
236 uchar_t *up = (uchar_t *)tcpha;
237 int found = 0;
238 int32_t sack_len;
239 tcp_seq sack_begin, sack_end;
240 tcp_t *tcp;
241
242 endp = up + TCP_HDR_LENGTH(tcpha);
243 up += TCP_MIN_HEADER_LENGTH;
244 /*
245 * If timestamp option is aligned as recommended in RFC 7323 Appendix
246 * A, and is the only option, return quickly.
247 */
248 if (TCP_HDR_LENGTH(tcpha) == (uint32_t)TCP_MIN_HEADER_LENGTH +
249 TCPOPT_REAL_TS_LEN &&
250 OK_32PTR(up) &&
251 *(uint32_t *)up == TCPOPT_NOP_NOP_TSTAMP) {
252 tcpopt->tcp_opt_ts_val = ABE32_TO_U32((up+4));
253 tcpopt->tcp_opt_ts_ecr = ABE32_TO_U32((up+8));
254
255 return (TCP_OPT_TSTAMP_PRESENT);
256 }
257 while (up < endp) {
258 len = endp - up;
259 switch (*up) {
260 case TCPOPT_EOL:
261 break;
262
263 case TCPOPT_NOP:
264 up++;
265 continue;
266
267 case TCPOPT_MAXSEG:
268 if (len < TCPOPT_MAXSEG_LEN ||
269 up[1] != TCPOPT_MAXSEG_LEN)
270 break;
271
272 mss = BE16_TO_U16(up+2);
273 /* Caller must handle tcp_mss_min and tcp_mss_max_* */
274 tcpopt->tcp_opt_mss = mss;
275 found |= TCP_OPT_MSS_PRESENT;
276
277 up += TCPOPT_MAXSEG_LEN;
278 continue;
279
280 case TCPOPT_WSCALE:
281 if (len < TCPOPT_WS_LEN || up[1] != TCPOPT_WS_LEN)
282 break;
283
284 if (up[2] > TCP_MAX_WINSHIFT)
285 tcpopt->tcp_opt_wscale = TCP_MAX_WINSHIFT;
286 else
287 tcpopt->tcp_opt_wscale = up[2];
288 found |= TCP_OPT_WSCALE_PRESENT;
289
290 up += TCPOPT_WS_LEN;
291 continue;
292
293 case TCPOPT_SACK_PERMITTED:
294 if (len < TCPOPT_SACK_OK_LEN ||
295 up[1] != TCPOPT_SACK_OK_LEN)
296 break;
297 found |= TCP_OPT_SACK_OK_PRESENT;
298 up += TCPOPT_SACK_OK_LEN;
299 continue;
300
301 case TCPOPT_SACK:
302 if (len <= 2 || up[1] <= 2 || len < up[1])
303 break;
304
305 /* If TCP is not interested in SACK blks... */
306 if ((tcp = tcpopt->tcp) == NULL) {
307 up += up[1];
308 continue;
309 }
310 sack_len = up[1] - TCPOPT_HEADER_LEN;
311 up += TCPOPT_HEADER_LEN;
312
313 /*
314 * If the list is empty, allocate one and assume
315 * nothing is sack'ed.
316 */
317 if (tcp->tcp_notsack_list == NULL) {
318 tcp_notsack_update(&(tcp->tcp_notsack_list),
319 tcp->tcp_suna, tcp->tcp_snxt,
320 &(tcp->tcp_num_notsack_blk),
321 &(tcp->tcp_cnt_notsack_list));
322
323 /*
324 * Make sure tcp_notsack_list is not NULL.
325 * This happens when kmem_alloc(KM_NOSLEEP)
326 * returns NULL.
327 */
328 if (tcp->tcp_notsack_list == NULL) {
329 up += sack_len;
330 continue;
331 }
332 tcp->tcp_fack = tcp->tcp_suna;
333 }
334
335 while (sack_len > 0) {
336 if (up + 8 > endp) {
337 up = endp;
338 break;
339 }
340 sack_begin = BE32_TO_U32(up);
341 up += 4;
342 sack_end = BE32_TO_U32(up);
343 up += 4;
344 sack_len -= 8;
345 /*
346 * Bounds checking. Make sure the SACK
347 * info is within tcp_suna and tcp_snxt.
348 * If this SACK blk is out of bound, ignore
349 * it but continue to parse the following
350 * blks.
351 */
352 if (SEQ_LEQ(sack_end, sack_begin) ||
353 SEQ_LT(sack_begin, tcp->tcp_suna) ||
354 SEQ_GT(sack_end, tcp->tcp_snxt)) {
355 continue;
356 }
357 tcp_notsack_insert(&(tcp->tcp_notsack_list),
358 sack_begin, sack_end,
359 &(tcp->tcp_num_notsack_blk),
360 &(tcp->tcp_cnt_notsack_list));
361 if (SEQ_GT(sack_end, tcp->tcp_fack)) {
362 tcp->tcp_fack = sack_end;
363 }
364 }
365 found |= TCP_OPT_SACK_PRESENT;
366 continue;
367
368 case TCPOPT_TSTAMP:
369 if (len < TCPOPT_TSTAMP_LEN ||
370 up[1] != TCPOPT_TSTAMP_LEN)
371 break;
372
373 tcpopt->tcp_opt_ts_val = BE32_TO_U32(up+2);
374 tcpopt->tcp_opt_ts_ecr = BE32_TO_U32(up+6);
375
376 found |= TCP_OPT_TSTAMP_PRESENT;
377
378 up += TCPOPT_TSTAMP_LEN;
379 continue;
380
381 default:
382 if (len <= 1 || len < (int)up[1] || up[1] == 0)
383 break;
384 up += up[1];
385 continue;
386 }
387 break;
388 }
389 return (found);
390 }
391
392 /*
393 * Process all TCP option in SYN segment. Note that this function should
394 * be called after tcp_set_destination() is called so that the necessary info
395 * from IRE is already set in the tcp structure.
396 *
397 * This function sets up the correct tcp_mss value according to the
398 * MSS option value and our header size. It also sets up the window scale
399 * and timestamp values, and initialize SACK info blocks. But it does not
400 * change receive window size after setting the tcp_mss value. The caller
401 * should do the appropriate change.
402 */
403 static void
404 tcp_process_options(tcp_t *tcp, tcpha_t *tcpha)
405 {
406 int options;
407 tcp_opt_t tcpopt;
408 uint32_t mss_max;
409 char *tmp_tcph;
410 tcp_stack_t *tcps = tcp->tcp_tcps;
411 conn_t *connp = tcp->tcp_connp;
412
413 tcpopt.tcp = NULL;
414 options = tcp_parse_options(tcpha, &tcpopt);
415
416 /*
417 * Process MSS option. Note that MSS option value does not account
418 * for IP or TCP options. This means that it is equal to MTU - minimum
419 * IP+TCP header size, which is 40 bytes for IPv4 and 60 bytes for
420 * IPv6.
421 */
422 if (!(options & TCP_OPT_MSS_PRESENT)) {
423 if (connp->conn_ipversion == IPV4_VERSION)
424 tcpopt.tcp_opt_mss = tcps->tcps_mss_def_ipv4;
425 else
426 tcpopt.tcp_opt_mss = tcps->tcps_mss_def_ipv6;
427 } else {
428 if (connp->conn_ipversion == IPV4_VERSION)
429 mss_max = tcps->tcps_mss_max_ipv4;
430 else
431 mss_max = tcps->tcps_mss_max_ipv6;
432 if (tcpopt.tcp_opt_mss < tcps->tcps_mss_min)
433 tcpopt.tcp_opt_mss = tcps->tcps_mss_min;
434 else if (tcpopt.tcp_opt_mss > mss_max)
435 tcpopt.tcp_opt_mss = mss_max;
436 }
437
438 /* Process Window Scale option. */
439 if (options & TCP_OPT_WSCALE_PRESENT) {
440 tcp->tcp_snd_ws = tcpopt.tcp_opt_wscale;
441 tcp->tcp_snd_ws_ok = B_TRUE;
442 } else {
443 tcp->tcp_snd_ws = B_FALSE;
444 tcp->tcp_snd_ws_ok = B_FALSE;
445 tcp->tcp_rcv_ws = B_FALSE;
446 }
447
448 /* Process Timestamp option. */
449 if ((options & TCP_OPT_TSTAMP_PRESENT) &&
450 (tcp->tcp_snd_ts_ok || TCP_IS_DETACHED(tcp))) {
451 tmp_tcph = (char *)tcp->tcp_tcpha;
452
453 tcp->tcp_snd_ts_ok = B_TRUE;
454 tcp->tcp_ts_recent = tcpopt.tcp_opt_ts_val;
455 tcp->tcp_last_rcv_lbolt = ddi_get_lbolt64();
456 ASSERT(OK_32PTR(tmp_tcph));
457 ASSERT(connp->conn_ht_ulp_len == TCP_MIN_HEADER_LENGTH);
458
459 /* Fill in our template header with basic timestamp option. */
460 tmp_tcph += connp->conn_ht_ulp_len;
461 tmp_tcph[0] = TCPOPT_NOP;
462 tmp_tcph[1] = TCPOPT_NOP;
463 tmp_tcph[2] = TCPOPT_TSTAMP;
464 tmp_tcph[3] = TCPOPT_TSTAMP_LEN;
465 connp->conn_ht_iphc_len += TCPOPT_REAL_TS_LEN;
466 connp->conn_ht_ulp_len += TCPOPT_REAL_TS_LEN;
467 tcp->tcp_tcpha->tha_offset_and_reserved += (3 << 4);
468 } else {
469 tcp->tcp_snd_ts_ok = B_FALSE;
470 }
471
472 /*
473 * Process SACK options. If SACK is enabled for this connection,
474 * then allocate the SACK info structure. Note the following ways
475 * when tcp_snd_sack_ok is set to true.
476 *
477 * For active connection: in tcp_set_destination() called in
478 * tcp_connect().
479 *
480 * For passive connection: in tcp_set_destination() called in
481 * tcp_input_listener().
482 *
483 * That's the reason why the extra TCP_IS_DETACHED() check is there.
484 * That check makes sure that if we did not send a SACK OK option,
485 * we will not enable SACK for this connection even though the other
486 * side sends us SACK OK option. For active connection, the SACK
487 * info structure has already been allocated. So we need to free
488 * it if SACK is disabled.
489 */
490 if ((options & TCP_OPT_SACK_OK_PRESENT) &&
491 (tcp->tcp_snd_sack_ok ||
492 (tcps->tcps_sack_permitted != 0 && TCP_IS_DETACHED(tcp)))) {
493 ASSERT(tcp->tcp_num_sack_blk == 0);
494 ASSERT(tcp->tcp_notsack_list == NULL);
495
496 tcp->tcp_snd_sack_ok = B_TRUE;
497 if (tcp->tcp_snd_ts_ok) {
498 tcp->tcp_max_sack_blk = 3;
499 } else {
500 tcp->tcp_max_sack_blk = 4;
501 }
502 } else if (tcp->tcp_snd_sack_ok) {
503 /*
504 * Resetting tcp_snd_sack_ok to B_FALSE so that
505 * no SACK info will be used for this
506 * connection. This assumes that SACK usage
507 * permission is negotiated. This may need
508 * to be changed once this is clarified.
509 */
510 ASSERT(tcp->tcp_num_sack_blk == 0);
511 ASSERT(tcp->tcp_notsack_list == NULL);
512 tcp->tcp_snd_sack_ok = B_FALSE;
513 }
514
515 /*
516 * Now we know the exact TCP/IP header length, subtract
517 * that from tcp_mss to get our side's MSS.
518 */
519 tcp->tcp_mss -= connp->conn_ht_iphc_len;
520
521 /*
522 * Here we assume that the other side's header size will be equal to
523 * our header size. We calculate the real MSS accordingly. Need to
524 * take into additional stuffs IPsec puts in.
525 *
526 * Real MSS = Opt.MSS - (our TCP/IP header - min TCP/IP header)
527 */
528 tcpopt.tcp_opt_mss -= connp->conn_ht_iphc_len +
529 tcp->tcp_ipsec_overhead -
530 ((connp->conn_ipversion == IPV4_VERSION ?
531 IP_SIMPLE_HDR_LENGTH : IPV6_HDR_LEN) + TCP_MIN_HEADER_LENGTH);
532
533 /*
534 * Set MSS to the smaller one of both ends of the connection.
535 * We should not have called tcp_mss_set() before, but our
536 * side of the MSS should have been set to a proper value
537 * by tcp_set_destination(). tcp_mss_set() will also set up the
538 * STREAM head parameters properly.
539 *
540 * If we have a larger-than-16-bit window but the other side
541 * didn't want to do window scale, tcp_rwnd_set() will take
542 * care of that.
543 */
544 tcp_mss_set(tcp, MIN(tcpopt.tcp_opt_mss, tcp->tcp_mss));
545
546 /*
547 * Initialize tcp_cwnd value. After tcp_mss_set(), tcp_mss has been
548 * updated properly.
549 */
550 TCP_SET_INIT_CWND(tcp, tcp->tcp_mss, tcps->tcps_slow_start_initial);
551 }
552
553 /*
554 * Add a new piece to the tcp reassembly queue. If the gap at the beginning
555 * is filled, return as much as we can. The message passed in may be
556 * multi-part, chained using b_cont. "start" is the starting sequence
557 * number for this piece.
558 */
559 static mblk_t *
560 tcp_reass(tcp_t *tcp, mblk_t *mp, uint32_t start)
561 {
562 uint32_t end, bytes;
563 mblk_t *mp1;
564 mblk_t *mp2;
565 mblk_t *next_mp;
566 uint32_t u1;
567 tcp_stack_t *tcps = tcp->tcp_tcps;
568
569
570 /* Walk through all the new pieces. */
571 do {
572 ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <=
573 (uintptr_t)INT_MAX);
574 end = start + (int)(mp->b_wptr - mp->b_rptr);
575 next_mp = mp->b_cont;
576 if (start == end) {
577 /* Empty. Blast it. */
578 freeb(mp);
579 continue;
580 }
581 bytes = end - start;
582 mp->b_cont = NULL;
583 TCP_REASS_SET_SEQ(mp, start);
584 TCP_REASS_SET_END(mp, end);
585 mp1 = tcp->tcp_reass_tail;
586 if (mp1 == NULL || SEQ_GEQ(start, TCP_REASS_END(mp1))) {
587 if (mp1 != NULL) {
588 /*
589 * New stuff is beyond the tail; link it on the
590 * end.
591 */
592 mp1->b_cont = mp;
593 } else {
594 tcp->tcp_reass_head = mp;
595 }
596 tcp->tcp_reass_tail = mp;
597 TCPS_BUMP_MIB(tcps, tcpInDataUnorderSegs);
598 TCPS_UPDATE_MIB(tcps, tcpInDataUnorderBytes, bytes);
599 tcp->tcp_cs.tcp_in_data_unorder_segs++;
600 tcp->tcp_cs.tcp_in_data_unorder_bytes += bytes;
601 continue;
602 }
603 mp1 = tcp->tcp_reass_head;
604 u1 = TCP_REASS_SEQ(mp1);
605 /* New stuff at the front? */
606 if (SEQ_LT(start, u1)) {
607 /* Yes... Check for overlap. */
608 mp->b_cont = mp1;
609 tcp->tcp_reass_head = mp;
610 tcp_reass_elim_overlap(tcp, mp);
611 continue;
612 }
613 /*
614 * The new piece fits somewhere between the head and tail.
615 * We find our slot, where mp1 precedes us and mp2 trails.
616 */
617 for (; (mp2 = mp1->b_cont) != NULL; mp1 = mp2) {
618 u1 = TCP_REASS_SEQ(mp2);
619 if (SEQ_LEQ(start, u1))
620 break;
621 }
622 /* Link ourselves in */
623 mp->b_cont = mp2;
624 mp1->b_cont = mp;
625
626 /* Trim overlap with following mblk(s) first */
627 tcp_reass_elim_overlap(tcp, mp);
628
629 /* Trim overlap with preceding mblk */
630 tcp_reass_elim_overlap(tcp, mp1);
631
632 } while (start = end, mp = next_mp);
633 mp1 = tcp->tcp_reass_head;
634 /* Anything ready to go? */
635 if (TCP_REASS_SEQ(mp1) != tcp->tcp_rnxt)
636 return (NULL);
637 /* Eat what we can off the queue */
638 for (;;) {
639 mp = mp1->b_cont;
640 end = TCP_REASS_END(mp1);
641 TCP_REASS_SET_SEQ(mp1, 0);
642 TCP_REASS_SET_END(mp1, 0);
643 if (!mp) {
644 tcp->tcp_reass_tail = NULL;
645 break;
646 }
647 if (end != TCP_REASS_SEQ(mp)) {
648 mp1->b_cont = NULL;
649 break;
650 }
651 mp1 = mp;
652 }
653 mp1 = tcp->tcp_reass_head;
654 tcp->tcp_reass_head = mp;
655 return (mp1);
656 }
657
658 /* Eliminate any overlap that mp may have over later mblks */
659 static void
660 tcp_reass_elim_overlap(tcp_t *tcp, mblk_t *mp)
661 {
662 uint32_t end;
663 mblk_t *mp1;
664 uint32_t u1;
665 tcp_stack_t *tcps = tcp->tcp_tcps;
666
667 end = TCP_REASS_END(mp);
668 while ((mp1 = mp->b_cont) != NULL) {
669 u1 = TCP_REASS_SEQ(mp1);
670 if (!SEQ_GT(end, u1))
671 break;
672 if (!SEQ_GEQ(end, TCP_REASS_END(mp1))) {
673 mp->b_wptr -= end - u1;
674 TCP_REASS_SET_END(mp, u1);
675 TCPS_BUMP_MIB(tcps, tcpInDataPartDupSegs);
676 TCPS_UPDATE_MIB(tcps, tcpInDataPartDupBytes,
677 end - u1);
678 break;
679 }
680 mp->b_cont = mp1->b_cont;
681 TCP_REASS_SET_SEQ(mp1, 0);
682 TCP_REASS_SET_END(mp1, 0);
683 freeb(mp1);
684 TCPS_BUMP_MIB(tcps, tcpInDataDupSegs);
685 TCPS_UPDATE_MIB(tcps, tcpInDataDupBytes, end - u1);
686 }
687 if (!mp1)
688 tcp->tcp_reass_tail = mp;
689 }
690
691 /*
692 * This function does PAWS protection check, per RFC 7323 section 5. Requires
693 * that timestamp options are already processed into tcpoptp. Returns B_TRUE if
694 * the segment passes the PAWS test, else returns B_FALSE.
695 */
696 boolean_t
697 tcp_paws_check(tcp_t *tcp, const tcp_opt_t *tcpoptp)
698 {
699 if (TSTMP_LT(tcpoptp->tcp_opt_ts_val,
700 tcp->tcp_ts_recent)) {
701 if (LBOLT_FASTPATH64 <
702 (tcp->tcp_last_rcv_lbolt + PAWS_TIMEOUT)) {
703 /* This segment is not acceptable. */
704 return (B_FALSE);
705 } else {
706 /*
707 * Connection has been idle for
708 * too long. Reset the timestamp
709 */
710 tcp->tcp_ts_recent =
711 tcpoptp->tcp_opt_ts_val;
712 }
713 }
714 return (B_TRUE);
715 }
716
717 /*
718 * Defense for the SYN attack -
719 * 1. When q0 is full, drop from the tail (tcp_eager_prev_drop_q0) the oldest
720 * one from the list of droppable eagers. This list is a subset of q0.
721 * see comments before the definition of MAKE_DROPPABLE().
722 * 2. Don't drop a SYN request before its first timeout. This gives every
723 * request at least til the first timeout to complete its 3-way handshake.
724 * 3. Maintain tcp_syn_rcvd_timeout as an accurate count of how many
725 * requests currently on the queue that has timed out. This will be used
726 * as an indicator of whether an attack is under way, so that appropriate
727 * actions can be taken. (It's incremented in tcp_timer() and decremented
728 * either when eager goes into ESTABLISHED, or gets freed up.)
729 * 4. The current threshold is - # of timeout > q0len/4 => SYN alert on
730 * # of timeout drops back to <= q0len/32 => SYN alert off
731 */
732 static boolean_t
733 tcp_drop_q0(tcp_t *tcp)
734 {
735 tcp_t *eager;
736 mblk_t *mp;
737 tcp_stack_t *tcps = tcp->tcp_tcps;
738
739 ASSERT(MUTEX_HELD(&tcp->tcp_eager_lock));
740 ASSERT(tcp->tcp_eager_next_q0 != tcp->tcp_eager_prev_q0);
741
742 /* Pick oldest eager from the list of droppable eagers */
743 eager = tcp->tcp_eager_prev_drop_q0;
744
745 /* If list is empty. return B_FALSE */
746 if (eager == tcp) {
747 return (B_FALSE);
748 }
749
750 /* If allocated, the mp will be freed in tcp_clean_death_wrapper() */
751 if ((mp = allocb(0, BPRI_HI)) == NULL)
752 return (B_FALSE);
753
754 /*
755 * Take this eager out from the list of droppable eagers since we are
756 * going to drop it.
757 */
758 MAKE_UNDROPPABLE(eager);
759
760 if (tcp->tcp_connp->conn_debug) {
761 (void) strlog(TCP_MOD_ID, 0, 3, SL_TRACE,
762 "tcp_drop_q0: listen half-open queue (max=%d) overflow"
763 " (%d pending) on %s, drop one", tcps->tcps_conn_req_max_q0,
764 tcp->tcp_conn_req_cnt_q0,
765 tcp_display(tcp, NULL, DISP_PORT_ONLY));
766 }
767
768 TCPS_BUMP_MIB(tcps, tcpHalfOpenDrop);
769
770 /* Put a reference on the conn as we are enqueueing it in the sqeue */
771 CONN_INC_REF(eager->tcp_connp);
772
773 SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
774 tcp_clean_death_wrapper, eager->tcp_connp, NULL,
775 SQ_FILL, SQTAG_TCP_DROP_Q0);
776
777 return (B_TRUE);
778 }
779
780 /*
781 * Handle a SYN on an AF_INET6 socket; can be either IPv4 or IPv6
782 */
783 static mblk_t *
784 tcp_conn_create_v6(conn_t *lconnp, conn_t *connp, mblk_t *mp,
785 ip_recv_attr_t *ira)
786 {
787 tcp_t *ltcp = lconnp->conn_tcp;
788 tcp_t *tcp = connp->conn_tcp;
789 mblk_t *tpi_mp;
790 ipha_t *ipha;
791 ip6_t *ip6h;
792 sin6_t sin6;
793 uint_t ifindex = ira->ira_ruifindex;
794 tcp_stack_t *tcps = tcp->tcp_tcps;
795
796 if (ira->ira_flags & IRAF_IS_IPV4) {
797 ipha = (ipha_t *)mp->b_rptr;
798
799 connp->conn_ipversion = IPV4_VERSION;
800 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &connp->conn_laddr_v6);
801 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &connp->conn_faddr_v6);
802 connp->conn_saddr_v6 = connp->conn_laddr_v6;
803
804 sin6 = sin6_null;
805 sin6.sin6_addr = connp->conn_faddr_v6;
806 sin6.sin6_port = connp->conn_fport;
807 sin6.sin6_family = AF_INET6;
808 sin6.__sin6_src_id = ip_srcid_find_addr(&connp->conn_laddr_v6,
809 IPCL_ZONEID(lconnp), tcps->tcps_netstack);
810
811 if (connp->conn_recv_ancillary.crb_recvdstaddr) {
812 sin6_t sin6d;
813
814 sin6d = sin6_null;
815 sin6d.sin6_addr = connp->conn_laddr_v6;
816 sin6d.sin6_port = connp->conn_lport;
817 sin6d.sin6_family = AF_INET;
818 tpi_mp = mi_tpi_extconn_ind(NULL,
819 (char *)&sin6d, sizeof (sin6_t),
820 (char *)&tcp,
821 (t_scalar_t)sizeof (intptr_t),
822 (char *)&sin6d, sizeof (sin6_t),
823 (t_scalar_t)ltcp->tcp_conn_req_seqnum);
824 } else {
825 tpi_mp = mi_tpi_conn_ind(NULL,
826 (char *)&sin6, sizeof (sin6_t),
827 (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
828 (t_scalar_t)ltcp->tcp_conn_req_seqnum);
829 }
830 } else {
831 ip6h = (ip6_t *)mp->b_rptr;
832
833 connp->conn_ipversion = IPV6_VERSION;
834 connp->conn_laddr_v6 = ip6h->ip6_dst;
835 connp->conn_faddr_v6 = ip6h->ip6_src;
836 connp->conn_saddr_v6 = connp->conn_laddr_v6;
837
838 sin6 = sin6_null;
839 sin6.sin6_addr = connp->conn_faddr_v6;
840 sin6.sin6_port = connp->conn_fport;
841 sin6.sin6_family = AF_INET6;
842 sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
843 sin6.__sin6_src_id = ip_srcid_find_addr(&connp->conn_laddr_v6,
844 IPCL_ZONEID(lconnp), tcps->tcps_netstack);
845
846 if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src)) {
847 /* Pass up the scope_id of remote addr */
848 sin6.sin6_scope_id = ifindex;
849 } else {
850 sin6.sin6_scope_id = 0;
851 }
852 if (connp->conn_recv_ancillary.crb_recvdstaddr) {
853 sin6_t sin6d;
854
855 sin6d = sin6_null;
856 sin6.sin6_addr = connp->conn_laddr_v6;
857 sin6d.sin6_port = connp->conn_lport;
858 sin6d.sin6_family = AF_INET6;
859 if (IN6_IS_ADDR_LINKSCOPE(&connp->conn_laddr_v6))
860 sin6d.sin6_scope_id = ifindex;
861
862 tpi_mp = mi_tpi_extconn_ind(NULL,
863 (char *)&sin6d, sizeof (sin6_t),
864 (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
865 (char *)&sin6d, sizeof (sin6_t),
866 (t_scalar_t)ltcp->tcp_conn_req_seqnum);
867 } else {
868 tpi_mp = mi_tpi_conn_ind(NULL,
869 (char *)&sin6, sizeof (sin6_t),
870 (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
871 (t_scalar_t)ltcp->tcp_conn_req_seqnum);
872 }
873 }
874
875 tcp->tcp_mss = tcps->tcps_mss_def_ipv6;
876 return (tpi_mp);
877 }
878
879 /* Handle a SYN on an AF_INET socket */
880 static mblk_t *
881 tcp_conn_create_v4(conn_t *lconnp, conn_t *connp, mblk_t *mp,
882 ip_recv_attr_t *ira)
883 {
884 tcp_t *ltcp = lconnp->conn_tcp;
885 tcp_t *tcp = connp->conn_tcp;
886 sin_t sin;
887 mblk_t *tpi_mp = NULL;
888 tcp_stack_t *tcps = tcp->tcp_tcps;
889 ipha_t *ipha;
890
891 ASSERT(ira->ira_flags & IRAF_IS_IPV4);
892 ipha = (ipha_t *)mp->b_rptr;
893
894 connp->conn_ipversion = IPV4_VERSION;
895 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &connp->conn_laddr_v6);
896 IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &connp->conn_faddr_v6);
897 connp->conn_saddr_v6 = connp->conn_laddr_v6;
898
899 sin = sin_null;
900 sin.sin_addr.s_addr = connp->conn_faddr_v4;
901 sin.sin_port = connp->conn_fport;
902 sin.sin_family = AF_INET;
903 if (lconnp->conn_recv_ancillary.crb_recvdstaddr) {
904 sin_t sind;
905
906 sind = sin_null;
907 sind.sin_addr.s_addr = connp->conn_laddr_v4;
908 sind.sin_port = connp->conn_lport;
909 sind.sin_family = AF_INET;
910 tpi_mp = mi_tpi_extconn_ind(NULL,
911 (char *)&sind, sizeof (sin_t), (char *)&tcp,
912 (t_scalar_t)sizeof (intptr_t), (char *)&sind,
913 sizeof (sin_t), (t_scalar_t)ltcp->tcp_conn_req_seqnum);
914 } else {
915 tpi_mp = mi_tpi_conn_ind(NULL,
916 (char *)&sin, sizeof (sin_t),
917 (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
918 (t_scalar_t)ltcp->tcp_conn_req_seqnum);
919 }
920
921 tcp->tcp_mss = tcps->tcps_mss_def_ipv4;
922 return (tpi_mp);
923 }
924
925 /*
926 * Called via squeue to get on to eager's perimeter. It sends a
927 * TH_RST if eager is in the fanout table. The listener wants the
928 * eager to disappear either by means of tcp_eager_blowoff() or
929 * tcp_eager_cleanup() being called. tcp_eager_kill() can also be
930 * called (via squeue) if the eager cannot be inserted in the
931 * fanout table in tcp_input_listener().
932 */
933 /* ARGSUSED */
934 void
935 tcp_eager_kill(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
936 {
937 conn_t *econnp = (conn_t *)arg;
938 tcp_t *eager = econnp->conn_tcp;
939 tcp_t *listener = eager->tcp_listener;
940
941 /*
942 * We could be called because listener is closing. Since
943 * the eager was using listener's queue's, we avoid
944 * using the listeners queues from now on.
945 */
946 ASSERT(eager->tcp_detached);
947 econnp->conn_rq = NULL;
948 econnp->conn_wq = NULL;
949
950 /*
951 * An eager's conn_fanout will be NULL if it's a duplicate
952 * for an existing 4-tuples in the conn fanout table.
953 * We don't want to send an RST out in such case.
954 */
955 if (econnp->conn_fanout != NULL && eager->tcp_state > TCPS_LISTEN) {
956 tcp_xmit_ctl("tcp_eager_kill, can't wait",
957 eager, eager->tcp_snxt, 0, TH_RST);
958 }
959
960 /* We are here because listener wants this eager gone */
961 if (listener != NULL) {
962 mutex_enter(&listener->tcp_eager_lock);
963 tcp_eager_unlink(eager);
964 if (eager->tcp_tconnind_started) {
965 /*
966 * The eager has sent a conn_ind up to the
967 * listener but listener decides to close
968 * instead. We need to drop the extra ref
969 * placed on eager in tcp_input_data() before
970 * sending the conn_ind to listener.
971 */
972 CONN_DEC_REF(econnp);
973 }
974 mutex_exit(&listener->tcp_eager_lock);
975 CONN_DEC_REF(listener->tcp_connp);
976 }
977
978 if (eager->tcp_state != TCPS_CLOSED)
979 tcp_close_detached(eager);
980 }
981
982 /*
983 * Reset any eager connection hanging off this listener marked
984 * with 'seqnum' and then reclaim it's resources.
985 */
986 boolean_t
987 tcp_eager_blowoff(tcp_t *listener, t_scalar_t seqnum)
988 {
989 tcp_t *eager;
990 mblk_t *mp;
991
992 eager = listener;
993 mutex_enter(&listener->tcp_eager_lock);
994 do {
995 eager = eager->tcp_eager_next_q;
996 if (eager == NULL) {
997 mutex_exit(&listener->tcp_eager_lock);
998 return (B_FALSE);
999 }
1000 } while (eager->tcp_conn_req_seqnum != seqnum);
1001
1002 if (eager->tcp_closemp_used) {
1003 mutex_exit(&listener->tcp_eager_lock);
1004 return (B_TRUE);
1005 }
1006 eager->tcp_closemp_used = B_TRUE;
1007 TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
1008 CONN_INC_REF(eager->tcp_connp);
1009 mutex_exit(&listener->tcp_eager_lock);
1010 mp = &eager->tcp_closemp;
1011 SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp, tcp_eager_kill,
1012 eager->tcp_connp, NULL, SQ_FILL, SQTAG_TCP_EAGER_BLOWOFF);
1013 return (B_TRUE);
1014 }
1015
1016 /*
1017 * Reset any eager connection hanging off this listener
1018 * and then reclaim it's resources.
1019 */
1020 void
1021 tcp_eager_cleanup(tcp_t *listener, boolean_t q0_only)
1022 {
1023 tcp_t *eager;
1024 mblk_t *mp;
1025 tcp_stack_t *tcps = listener->tcp_tcps;
1026
1027 ASSERT(MUTEX_HELD(&listener->tcp_eager_lock));
1028
1029 if (!q0_only) {
1030 /* First cleanup q */
1031 TCP_STAT(tcps, tcp_eager_blowoff_q);
1032 eager = listener->tcp_eager_next_q;
1033 while (eager != NULL) {
1034 if (!eager->tcp_closemp_used) {
1035 eager->tcp_closemp_used = B_TRUE;
1036 TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
1037 CONN_INC_REF(eager->tcp_connp);
1038 mp = &eager->tcp_closemp;
1039 SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
1040 tcp_eager_kill, eager->tcp_connp, NULL,
1041 SQ_FILL, SQTAG_TCP_EAGER_CLEANUP);
1042 }
1043 eager = eager->tcp_eager_next_q;
1044 }
1045 }
1046 /* Then cleanup q0 */
1047 TCP_STAT(tcps, tcp_eager_blowoff_q0);
1048 eager = listener->tcp_eager_next_q0;
1049 while (eager != listener) {
1050 if (!eager->tcp_closemp_used) {
1051 eager->tcp_closemp_used = B_TRUE;
1052 TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
1053 CONN_INC_REF(eager->tcp_connp);
1054 mp = &eager->tcp_closemp;
1055 SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
1056 tcp_eager_kill, eager->tcp_connp, NULL, SQ_FILL,
1057 SQTAG_TCP_EAGER_CLEANUP_Q0);
1058 }
1059 eager = eager->tcp_eager_next_q0;
1060 }
1061 }
1062
1063 /*
1064 * If we are an eager connection hanging off a listener that hasn't
1065 * formally accepted the connection yet, get off its list and blow off
1066 * any data that we have accumulated.
1067 */
1068 void
1069 tcp_eager_unlink(tcp_t *tcp)
1070 {
1071 tcp_t *listener = tcp->tcp_listener;
1072
1073 ASSERT(listener != NULL);
1074 ASSERT(MUTEX_HELD(&listener->tcp_eager_lock));
1075 if (tcp->tcp_eager_next_q0 != NULL) {
1076 ASSERT(tcp->tcp_eager_prev_q0 != NULL);
1077
1078 /* Remove the eager tcp from q0 */
1079 tcp->tcp_eager_next_q0->tcp_eager_prev_q0 =
1080 tcp->tcp_eager_prev_q0;
1081 tcp->tcp_eager_prev_q0->tcp_eager_next_q0 =
1082 tcp->tcp_eager_next_q0;
1083 ASSERT(listener->tcp_conn_req_cnt_q0 > 0);
1084 listener->tcp_conn_req_cnt_q0--;
1085
1086 tcp->tcp_eager_next_q0 = NULL;
1087 tcp->tcp_eager_prev_q0 = NULL;
1088
1089 /*
1090 * Take the eager out, if it is in the list of droppable
1091 * eagers.
1092 */
1093 MAKE_UNDROPPABLE(tcp);
1094
1095 if (tcp->tcp_syn_rcvd_timeout != 0) {
1096 /* we have timed out before */
1097 ASSERT(listener->tcp_syn_rcvd_timeout > 0);
1098 listener->tcp_syn_rcvd_timeout--;
1099 }
1100 } else {
1101 tcp_t **tcpp = &listener->tcp_eager_next_q;
1102 tcp_t *prev = NULL;
1103
1104 for (; tcpp[0]; tcpp = &tcpp[0]->tcp_eager_next_q) {
1105 if (tcpp[0] == tcp) {
1106 if (listener->tcp_eager_last_q == tcp) {
1107 /*
1108 * If we are unlinking the last
1109 * element on the list, adjust
1110 * tail pointer. Set tail pointer
1111 * to nil when list is empty.
1112 */
1113 ASSERT(tcp->tcp_eager_next_q == NULL);
1114 if (listener->tcp_eager_last_q ==
1115 listener->tcp_eager_next_q) {
1116 listener->tcp_eager_last_q =
1117 NULL;
1118 } else {
1119 /*
1120 * We won't get here if there
1121 * is only one eager in the
1122 * list.
1123 */
1124 ASSERT(prev != NULL);
1125 listener->tcp_eager_last_q =
1126 prev;
1127 }
1128 }
1129 tcpp[0] = tcp->tcp_eager_next_q;
1130 tcp->tcp_eager_next_q = NULL;
1131 tcp->tcp_eager_last_q = NULL;
1132 ASSERT(listener->tcp_conn_req_cnt_q > 0);
1133 listener->tcp_conn_req_cnt_q--;
1134 break;
1135 }
1136 prev = tcpp[0];
1137 }
1138 }
1139 tcp->tcp_listener = NULL;
1140 }
1141
1142 /* BEGIN CSTYLED */
1143 /*
1144 *
1145 * The sockfs ACCEPT path:
1146 * =======================
1147 *
1148 * The eager is now established in its own perimeter as soon as SYN is
1149 * received in tcp_input_listener(). When sockfs receives conn_ind, it
1150 * completes the accept processing on the acceptor STREAM. The sending
1151 * of conn_ind part is common for both sockfs listener and a TLI/XTI
1152 * listener but a TLI/XTI listener completes the accept processing
1153 * on the listener perimeter.
1154 *
1155 * Common control flow for 3 way handshake:
1156 * ----------------------------------------
1157 *
1158 * incoming SYN (listener perimeter) -> tcp_input_listener()
1159 *
1160 * incoming SYN-ACK-ACK (eager perim) -> tcp_input_data()
1161 * send T_CONN_IND (listener perim) -> tcp_send_conn_ind()
1162 *
1163 * Sockfs ACCEPT Path:
1164 * -------------------
1165 *
1166 * open acceptor stream (tcp_open allocates tcp_tli_accept()
1167 * as STREAM entry point)
1168 *
1169 * soaccept() sends T_CONN_RES on the acceptor STREAM to tcp_tli_accept()
1170 *
1171 * tcp_tli_accept() extracts the eager and makes the q->q_ptr <-> eager
1172 * association (we are not behind eager's squeue but sockfs is protecting us
1173 * and no one knows about this stream yet. The STREAMS entry point q->q_info
1174 * is changed to point at tcp_wput().
1175 *
1176 * tcp_accept_common() sends any deferred eagers via tcp_send_pending() to
1177 * listener (done on listener's perimeter).
1178 *
1179 * tcp_tli_accept() calls tcp_accept_finish() on eagers perimeter to finish
1180 * accept.
1181 *
1182 * TLI/XTI client ACCEPT path:
1183 * ---------------------------
1184 *
1185 * soaccept() sends T_CONN_RES on the listener STREAM.
1186 *
1187 * tcp_tli_accept() -> tcp_accept_swap() complete the processing and send
1188 * a M_SETOPS mblk to eager perimeter to finish accept (tcp_accept_finish()).
1189 *
1190 * Locks:
1191 * ======
1192 *
1193 * listener->tcp_eager_lock protects the listeners->tcp_eager_next_q0 and
1194 * and listeners->tcp_eager_next_q.
1195 *
1196 * Referencing:
1197 * ============
1198 *
1199 * 1) We start out in tcp_input_listener by eager placing a ref on
1200 * listener and listener adding eager to listeners->tcp_eager_next_q0.
1201 *
1202 * 2) When a SYN-ACK-ACK arrives, we send the conn_ind to listener. Before
1203 * doing so we place a ref on the eager. This ref is finally dropped at the
1204 * end of tcp_accept_finish() while unwinding from the squeue, i.e. the
1205 * reference is dropped by the squeue framework.
1206 *
1207 * 3) The ref on listener placed in 1 above is dropped in tcp_accept_finish
1208 *
1209 * The reference must be released by the same entity that added the reference
1210 * In the above scheme, the eager is the entity that adds and releases the
1211 * references. Note that tcp_accept_finish executes in the squeue of the eager
1212 * (albeit after it is attached to the acceptor stream). Though 1. executes
1213 * in the listener's squeue, the eager is nascent at this point and the
1214 * reference can be considered to have been added on behalf of the eager.
1215 *
1216 * Eager getting a Reset or listener closing:
1217 * ==========================================
1218 *
1219 * Once the listener and eager are linked, the listener never does the unlink.
1220 * If the listener needs to close, tcp_eager_cleanup() is called which queues
1221 * a message on all eager perimeter. The eager then does the unlink, clears
1222 * any pointers to the listener's queue and drops the reference to the
1223 * listener. The listener waits in tcp_close outside the squeue until its
1224 * refcount has dropped to 1. This ensures that the listener has waited for
1225 * all eagers to clear their association with the listener.
1226 *
1227 * Similarly, if eager decides to go away, it can unlink itself and close.
1228 * When the T_CONN_RES comes down, we check if eager has closed. Note that
1229 * the reference to eager is still valid because of the extra ref we put
1230 * in tcp_send_conn_ind.
1231 *
1232 * Listener can always locate the eager under the protection
1233 * of the listener->tcp_eager_lock, and then do a refhold
1234 * on the eager during the accept processing.
1235 *
1236 * The acceptor stream accesses the eager in the accept processing
1237 * based on the ref placed on eager before sending T_conn_ind.
1238 * The only entity that can negate this refhold is a listener close
1239 * which is mutually exclusive with an active acceptor stream.
1240 *
1241 * Eager's reference on the listener
1242 * ===================================
1243 *
1244 * If the accept happens (even on a closed eager) the eager drops its
1245 * reference on the listener at the start of tcp_accept_finish. If the
1246 * eager is killed due to an incoming RST before the T_conn_ind is sent up,
1247 * the reference is dropped in tcp_closei_local. If the listener closes,
1248 * the reference is dropped in tcp_eager_kill. In all cases the reference
1249 * is dropped while executing in the eager's context (squeue).
1250 */
1251 /* END CSTYLED */
1252
1253 /* Process the SYN packet, mp, directed at the listener 'tcp' */
1254
1255 /*
1256 * THIS FUNCTION IS DIRECTLY CALLED BY IP VIA SQUEUE FOR SYN.
1257 * tcp_input_data will not see any packets for listeners since the listener
1258 * has conn_recv set to tcp_input_listener.
1259 */
1260 /* ARGSUSED */
1261 static void
1262 tcp_input_listener(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
1263 {
1264 tcpha_t *tcpha;
1265 uint32_t seg_seq;
1266 tcp_t *eager;
1267 int err;
1268 conn_t *econnp = NULL;
1269 squeue_t *new_sqp;
1270 mblk_t *mp1;
1271 uint_t ip_hdr_len;
1272 conn_t *lconnp = (conn_t *)arg;
1273 tcp_t *listener = lconnp->conn_tcp;
1274 tcp_stack_t *tcps = listener->tcp_tcps;
1275 ip_stack_t *ipst = tcps->tcps_netstack->netstack_ip;
1276 uint_t flags;
1277 mblk_t *tpi_mp;
1278 uint_t ifindex = ira->ira_ruifindex;
1279 boolean_t tlc_set = B_FALSE;
1280
1281 ip_hdr_len = ira->ira_ip_hdr_length;
1282 tcpha = (tcpha_t *)&mp->b_rptr[ip_hdr_len];
1283 flags = (unsigned int)tcpha->tha_flags & 0xFF;
1284
1285 DTRACE_TCP5(receive, mblk_t *, NULL, ip_xmit_attr_t *, lconnp->conn_ixa,
1286 __dtrace_tcp_void_ip_t *, mp->b_rptr, tcp_t *, listener,
1287 __dtrace_tcp_tcph_t *, tcpha);
1288
1289 if (!(flags & TH_SYN)) {
1290 if ((flags & TH_RST) || (flags & TH_URG)) {
1291 freemsg(mp);
1292 return;
1293 }
1294 if (flags & TH_ACK) {
1295 /* Note this executes in listener's squeue */
1296 tcp_xmit_listeners_reset(mp, ira, ipst, lconnp);
1297 return;
1298 }
1299
1300 freemsg(mp);
1301 return;
1302 }
1303
1304 if (listener->tcp_state != TCPS_LISTEN)
1305 goto error2;
1306
1307 ASSERT(IPCL_IS_BOUND(lconnp));
1308
1309 mutex_enter(&listener->tcp_eager_lock);
1310
1311 /*
1312 * The system is under memory pressure, so we need to do our part
1313 * to relieve the pressure. So we only accept new request if there
1314 * is nothing waiting to be accepted or waiting to complete the 3-way
1315 * handshake. This means that busy listener will not get too many
1316 * new requests which they cannot handle in time while non-busy
1317 * listener is still functioning properly.
1318 */
1319 if (tcps->tcps_reclaim && (listener->tcp_conn_req_cnt_q > 0 ||
1320 listener->tcp_conn_req_cnt_q0 > 0)) {
1321 mutex_exit(&listener->tcp_eager_lock);
1322 TCP_STAT(tcps, tcp_listen_mem_drop);
1323 goto error2;
1324 }
1325
1326 if (listener->tcp_conn_req_cnt_q >= listener->tcp_conn_req_max) {
1327 mutex_exit(&listener->tcp_eager_lock);
1328 TCP_STAT(tcps, tcp_listendrop);
1329 TCPS_BUMP_MIB(tcps, tcpListenDrop);
1330 if (lconnp->conn_debug) {
1331 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE|SL_ERROR,
1332 "tcp_input_listener: listen backlog (max=%d) "
1333 "overflow (%d pending) on %s",
1334 listener->tcp_conn_req_max,
1335 listener->tcp_conn_req_cnt_q,
1336 tcp_display(listener, NULL, DISP_PORT_ONLY));
1337 }
1338 goto error2;
1339 }
1340
1341 if (listener->tcp_conn_req_cnt_q0 >=
1342 listener->tcp_conn_req_max + tcps->tcps_conn_req_max_q0) {
1343 /*
1344 * Q0 is full. Drop a pending half-open req from the queue
1345 * to make room for the new SYN req. Also mark the time we
1346 * drop a SYN.
1347 *
1348 * A more aggressive defense against SYN attack will
1349 * be to set the "tcp_syn_defense" flag now.
1350 */
1351 TCP_STAT(tcps, tcp_listendropq0);
1352 listener->tcp_last_rcv_lbolt = ddi_get_lbolt64();
1353 if (!tcp_drop_q0(listener)) {
1354 mutex_exit(&listener->tcp_eager_lock);
1355 TCPS_BUMP_MIB(tcps, tcpListenDropQ0);
1356 if (lconnp->conn_debug) {
1357 (void) strlog(TCP_MOD_ID, 0, 3, SL_TRACE,
1358 "tcp_input_listener: listen half-open "
1359 "queue (max=%d) full (%d pending) on %s",
1360 tcps->tcps_conn_req_max_q0,
1361 listener->tcp_conn_req_cnt_q0,
1362 tcp_display(listener, NULL,
1363 DISP_PORT_ONLY));
1364 }
1365 goto error2;
1366 }
1367 }
1368
1369 /*
1370 * Enforce the limit set on the number of connections per listener.
1371 * Note that tlc_cnt starts with 1. So need to add 1 to tlc_max
1372 * for comparison.
1373 */
1374 if (listener->tcp_listen_cnt != NULL) {
1375 tcp_listen_cnt_t *tlc = listener->tcp_listen_cnt;
1376 int64_t now;
1377
1378 if (atomic_inc_32_nv(&tlc->tlc_cnt) > tlc->tlc_max + 1) {
1379 mutex_exit(&listener->tcp_eager_lock);
1380 now = ddi_get_lbolt64();
1381 atomic_dec_32(&tlc->tlc_cnt);
1382 TCP_STAT(tcps, tcp_listen_cnt_drop);
1383 tlc->tlc_drop++;
1384 if (now - tlc->tlc_report_time >
1385 MSEC_TO_TICK(TCP_TLC_REPORT_INTERVAL)) {
1386 zcmn_err(lconnp->conn_zoneid, CE_WARN,
1387 "Listener (port %d) connection max (%u) "
1388 "reached: %u attempts dropped total\n",
1389 ntohs(listener->tcp_connp->conn_lport),
1390 tlc->tlc_max, tlc->tlc_drop);
1391 tlc->tlc_report_time = now;
1392 }
1393 goto error2;
1394 }
1395 tlc_set = B_TRUE;
1396 }
1397
1398 mutex_exit(&listener->tcp_eager_lock);
1399
1400 /*
1401 * IP sets ira_sqp to either the senders conn_sqp (for loopback)
1402 * or based on the ring (for packets from GLD). Otherwise it is
1403 * set based on lbolt i.e., a somewhat random number.
1404 */
1405 ASSERT(ira->ira_sqp != NULL);
1406 new_sqp = ira->ira_sqp;
1407
1408 econnp = (conn_t *)tcp_get_conn(arg2, tcps);
1409 if (econnp == NULL)
1410 goto error2;
1411
1412 ASSERT(econnp->conn_netstack == lconnp->conn_netstack);
1413 econnp->conn_sqp = new_sqp;
1414 econnp->conn_initial_sqp = new_sqp;
1415 econnp->conn_ixa->ixa_sqp = new_sqp;
1416
1417 econnp->conn_fport = tcpha->tha_lport;
1418 econnp->conn_lport = tcpha->tha_fport;
1419
1420 err = conn_inherit_parent(lconnp, econnp);
1421 if (err != 0)
1422 goto error3;
1423
1424 /* We already know the laddr of the new connection is ours */
1425 econnp->conn_ixa->ixa_src_generation = ipst->ips_src_generation;
1426
1427 ASSERT(OK_32PTR(mp->b_rptr));
1428 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION ||
1429 IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
1430
1431 if (lconnp->conn_family == AF_INET) {
1432 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION);
1433 tpi_mp = tcp_conn_create_v4(lconnp, econnp, mp, ira);
1434 } else {
1435 tpi_mp = tcp_conn_create_v6(lconnp, econnp, mp, ira);
1436 }
1437
1438 if (tpi_mp == NULL)
1439 goto error3;
1440
1441 eager = econnp->conn_tcp;
1442 eager->tcp_detached = B_TRUE;
1443 SOCK_CONNID_INIT(eager->tcp_connid);
1444
1445 /*
1446 * Initialize the eager's tcp_t and inherit some parameters from
1447 * the listener.
1448 */
1449 tcp_init_values(eager, listener);
1450
1451 ASSERT((econnp->conn_ixa->ixa_flags &
1452 (IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
1453 IXAF_VERIFY_PMTU | IXAF_VERIFY_LSO)) ==
1454 (IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
1455 IXAF_VERIFY_PMTU | IXAF_VERIFY_LSO));
1456
1457 if (!tcps->tcps_dev_flow_ctl)
1458 econnp->conn_ixa->ixa_flags |= IXAF_NO_DEV_FLOW_CTL;
1459
1460 /* Prepare for diffing against previous packets */
1461 eager->tcp_recvifindex = 0;
1462 eager->tcp_recvhops = 0xffffffffU;
1463
1464 if (!(ira->ira_flags & IRAF_IS_IPV4) && econnp->conn_bound_if == 0) {
1465 if (IN6_IS_ADDR_LINKSCOPE(&econnp->conn_faddr_v6) ||
1466 IN6_IS_ADDR_LINKSCOPE(&econnp->conn_laddr_v6)) {
1467 econnp->conn_incoming_ifindex = ifindex;
1468 econnp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
1469 econnp->conn_ixa->ixa_scopeid = ifindex;
1470 }
1471 }
1472
1473 if ((ira->ira_flags & (IRAF_IS_IPV4|IRAF_IPV4_OPTIONS)) ==
1474 (IRAF_IS_IPV4|IRAF_IPV4_OPTIONS) &&
1475 tcps->tcps_rev_src_routes) {
1476 ipha_t *ipha = (ipha_t *)mp->b_rptr;
1477 ip_pkt_t *ipp = &econnp->conn_xmit_ipp;
1478
1479 /* Source routing option copyover (reverse it) */
1480 err = ip_find_hdr_v4(ipha, ipp, B_TRUE);
1481 if (err != 0) {
1482 freemsg(tpi_mp);
1483 goto error3;
1484 }
1485 ip_pkt_source_route_reverse_v4(ipp);
1486 }
1487
1488 ASSERT(eager->tcp_conn.tcp_eager_conn_ind == NULL);
1489 ASSERT(!eager->tcp_tconnind_started);
1490 /*
1491 * If the SYN came with a credential, it's a loopback packet or a
1492 * labeled packet; attach the credential to the TPI message.
1493 */
1494 if (ira->ira_cred != NULL)
1495 mblk_setcred(tpi_mp, ira->ira_cred, ira->ira_cpid);
1496
1497 eager->tcp_conn.tcp_eager_conn_ind = tpi_mp;
1498 ASSERT(eager->tcp_ordrel_mp == NULL);
1499
1500 /* Inherit the listener's non-STREAMS flag */
1501 if (IPCL_IS_NONSTR(lconnp)) {
1502 econnp->conn_flags |= IPCL_NONSTR;
1503 /* All non-STREAMS tcp_ts are sockets */
1504 eager->tcp_issocket = B_TRUE;
1505 } else {
1506 /*
1507 * Pre-allocate the T_ordrel_ind mblk for TPI socket so that
1508 * at close time, we will always have that to send up.
1509 * Otherwise, we need to do special handling in case the
1510 * allocation fails at that time.
1511 */
1512 if ((eager->tcp_ordrel_mp = mi_tpi_ordrel_ind()) == NULL)
1513 goto error3;
1514 }
1515 /*
1516 * Now that the IP addresses and ports are setup in econnp we
1517 * can do the IPsec policy work.
1518 */
1519 if (ira->ira_flags & IRAF_IPSEC_SECURE) {
1520 if (lconnp->conn_policy != NULL) {
1521 /*
1522 * Inherit the policy from the listener; use
1523 * actions from ira
1524 */
1525 if (!ip_ipsec_policy_inherit(econnp, lconnp, ira)) {
1526 CONN_DEC_REF(econnp);
1527 freemsg(mp);
1528 goto error3;
1529 }
1530 }
1531 }
1532
1533 /*
1534 * tcp_set_destination() may set tcp_rwnd according to the route
1535 * metrics. If it does not, the eager's receive window will be set
1536 * to the listener's receive window later in this function.
1537 */
1538 eager->tcp_rwnd = 0;
1539
1540 if (is_system_labeled()) {
1541 ip_xmit_attr_t *ixa = econnp->conn_ixa;
1542
1543 ASSERT(ira->ira_tsl != NULL);
1544 /* Discard any old label */
1545 if (ixa->ixa_free_flags & IXA_FREE_TSL) {
1546 ASSERT(ixa->ixa_tsl != NULL);
1547 label_rele(ixa->ixa_tsl);
1548 ixa->ixa_free_flags &= ~IXA_FREE_TSL;
1549 ixa->ixa_tsl = NULL;
1550 }
1551 if ((lconnp->conn_mlp_type != mlptSingle ||
1552 lconnp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1553 ira->ira_tsl != NULL) {
1554 /*
1555 * If this is an MLP connection or a MAC-Exempt
1556 * connection with an unlabeled node, packets are to be
1557 * exchanged using the security label of the received
1558 * SYN packet instead of the server application's label.
1559 * tsol_check_dest called from ip_set_destination
1560 * might later update TSF_UNLABELED by replacing
1561 * ixa_tsl with a new label.
1562 */
1563 label_hold(ira->ira_tsl);
1564 ip_xmit_attr_replace_tsl(ixa, ira->ira_tsl);
1565 DTRACE_PROBE2(mlp_syn_accept, conn_t *,
1566 econnp, ts_label_t *, ixa->ixa_tsl)
1567 } else {
1568 ixa->ixa_tsl = crgetlabel(econnp->conn_cred);
1569 DTRACE_PROBE2(syn_accept, conn_t *,
1570 econnp, ts_label_t *, ixa->ixa_tsl)
1571 }
1572 /*
1573 * conn_connect() called from tcp_set_destination will verify
1574 * the destination is allowed to receive packets at the
1575 * security label of the SYN-ACK we are generating. As part of
1576 * that, tsol_check_dest() may create a new effective label for
1577 * this connection.
1578 * Finally conn_connect() will call conn_update_label.
1579 * All that remains for TCP to do is to call
1580 * conn_build_hdr_template which is done as part of
1581 * tcp_set_destination.
1582 */
1583 }
1584
1585 /*
1586 * Since we will clear tcp_listener before we clear tcp_detached
1587 * in the accept code we need tcp_hard_binding aka tcp_accept_inprogress
1588 * so we can tell a TCP_IS_DETACHED_NONEAGER apart.
1589 */
1590 eager->tcp_hard_binding = B_TRUE;
1591
1592 tcp_bind_hash_insert(&tcps->tcps_bind_fanout[
1593 TCP_BIND_HASH(econnp->conn_lport)], eager, 0);
1594
1595 CL_INET_CONNECT(econnp, B_FALSE, err);
1596 if (err != 0) {
1597 tcp_bind_hash_remove(eager);
1598 goto error3;
1599 }
1600
1601 SOCK_CONNID_BUMP(eager->tcp_connid);
1602
1603 /*
1604 * Adapt our mss, ttl, ... based on the remote address.
1605 */
1606
1607 if (tcp_set_destination(eager) != 0) {
1608 TCPS_BUMP_MIB(tcps, tcpAttemptFails);
1609 /* Undo the bind_hash_insert */
1610 tcp_bind_hash_remove(eager);
1611 goto error3;
1612 }
1613
1614 /* Process all TCP options. */
1615 tcp_process_options(eager, tcpha);
1616
1617 /* Is the other end ECN capable? */
1618 if (tcps->tcps_ecn_permitted >= 1 &&
1619 (tcpha->tha_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) {
1620 eager->tcp_ecn_ok = B_TRUE;
1621 }
1622
1623 /*
1624 * The listener's conn_rcvbuf should be the default window size or a
1625 * window size changed via SO_RCVBUF option. First round up the
1626 * eager's tcp_rwnd to the nearest MSS. Then find out the window
1627 * scale option value if needed. Call tcp_rwnd_set() to finish the
1628 * setting.
1629 *
1630 * Note if there is a rpipe metric associated with the remote host,
1631 * we should not inherit receive window size from listener.
1632 */
1633 eager->tcp_rwnd = MSS_ROUNDUP(
1634 (eager->tcp_rwnd == 0 ? econnp->conn_rcvbuf :
1635 eager->tcp_rwnd), eager->tcp_mss);
1636 if (eager->tcp_snd_ws_ok)
1637 tcp_set_ws_value(eager);
1638 /*
1639 * Note that this is the only place tcp_rwnd_set() is called for
1640 * accepting a connection. We need to call it here instead of
1641 * after the 3-way handshake because we need to tell the other
1642 * side our rwnd in the SYN-ACK segment.
1643 */
1644 (void) tcp_rwnd_set(eager, eager->tcp_rwnd);
1645
1646 ASSERT(eager->tcp_connp->conn_rcvbuf != 0 &&
1647 eager->tcp_connp->conn_rcvbuf == eager->tcp_rwnd);
1648
1649 ASSERT(econnp->conn_rcvbuf != 0 &&
1650 econnp->conn_rcvbuf == eager->tcp_rwnd);
1651
1652 /* Put a ref on the listener for the eager. */
1653 CONN_INC_REF(lconnp);
1654 mutex_enter(&listener->tcp_eager_lock);
1655 listener->tcp_eager_next_q0->tcp_eager_prev_q0 = eager;
1656 eager->tcp_eager_next_q0 = listener->tcp_eager_next_q0;
1657 listener->tcp_eager_next_q0 = eager;
1658 eager->tcp_eager_prev_q0 = listener;
1659
1660 /* Set tcp_listener before adding it to tcp_conn_fanout */
1661 eager->tcp_listener = listener;
1662 eager->tcp_saved_listener = listener;
1663
1664 /*
1665 * Set tcp_listen_cnt so that when the connection is done, the counter
1666 * is decremented.
1667 */
1668 eager->tcp_listen_cnt = listener->tcp_listen_cnt;
1669
1670 /*
1671 * Tag this detached tcp vector for later retrieval
1672 * by our listener client in tcp_accept().
1673 */
1674 eager->tcp_conn_req_seqnum = listener->tcp_conn_req_seqnum;
1675 listener->tcp_conn_req_cnt_q0++;
1676 if (++listener->tcp_conn_req_seqnum == -1) {
1677 /*
1678 * -1 is "special" and defined in TPI as something
1679 * that should never be used in T_CONN_IND
1680 */
1681 ++listener->tcp_conn_req_seqnum;
1682 }
1683 mutex_exit(&listener->tcp_eager_lock);
1684
1685 if (listener->tcp_syn_defense) {
1686 /* Don't drop the SYN that comes from a good IP source */
1687 ipaddr_t *addr_cache;
1688
1689 addr_cache = (ipaddr_t *)(listener->tcp_ip_addr_cache);
1690 if (addr_cache != NULL && econnp->conn_faddr_v4 ==
1691 addr_cache[IP_ADDR_CACHE_HASH(econnp->conn_faddr_v4)]) {
1692 eager->tcp_dontdrop = B_TRUE;
1693 }
1694 }
1695
1696 /*
1697 * We need to insert the eager in its own perimeter but as soon
1698 * as we do that, we expose the eager to the classifier and
1699 * should not touch any field outside the eager's perimeter.
1700 * So do all the work necessary before inserting the eager
1701 * in its own perimeter. Be optimistic that conn_connect()
1702 * will succeed but undo everything if it fails.
1703 */
1704 seg_seq = ntohl(tcpha->tha_seq);
1705 eager->tcp_irs = seg_seq;
1706 eager->tcp_rack = seg_seq;
1707 eager->tcp_rnxt = seg_seq + 1;
1708 eager->tcp_tcpha->tha_ack = htonl(eager->tcp_rnxt);
1709 TCPS_BUMP_MIB(tcps, tcpPassiveOpens);
1710 eager->tcp_state = TCPS_SYN_RCVD;
1711 DTRACE_TCP6(state__change, void, NULL, ip_xmit_attr_t *,
1712 econnp->conn_ixa, void, NULL, tcp_t *, eager, void, NULL,
1713 int32_t, TCPS_LISTEN);
1714
1715 mp1 = tcp_xmit_mp(eager, eager->tcp_xmit_head, eager->tcp_mss,
1716 NULL, NULL, eager->tcp_iss, B_FALSE, NULL, B_FALSE);
1717 if (mp1 == NULL) {
1718 /*
1719 * Increment the ref count as we are going to
1720 * enqueueing an mp in squeue
1721 */
1722 CONN_INC_REF(econnp);
1723 goto error;
1724 }
1725
1726 /*
1727 * We need to start the rto timer. In normal case, we start
1728 * the timer after sending the packet on the wire (or at
1729 * least believing that packet was sent by waiting for
1730 * conn_ip_output() to return). Since this is the first packet
1731 * being sent on the wire for the eager, our initial tcp_rto
1732 * is at least tcp_rexmit_interval_min which is a fairly
1733 * large value to allow the algorithm to adjust slowly to large
1734 * fluctuations of RTT during first few transmissions.
1735 *
1736 * Starting the timer first and then sending the packet in this
1737 * case shouldn't make much difference since tcp_rexmit_interval_min
1738 * is of the order of several 100ms and starting the timer
1739 * first and then sending the packet will result in difference
1740 * of few micro seconds.
1741 *
1742 * Without this optimization, we are forced to hold the fanout
1743 * lock across the ipcl_bind_insert() and sending the packet
1744 * so that we don't race against an incoming packet (maybe RST)
1745 * for this eager.
1746 *
1747 * It is necessary to acquire an extra reference on the eager
1748 * at this point and hold it until after tcp_send_data() to
1749 * ensure against an eager close race.
1750 */
1751
1752 CONN_INC_REF(econnp);
1753
1754 TCP_TIMER_RESTART(eager, eager->tcp_rto);
1755
1756 /*
1757 * Insert the eager in its own perimeter now. We are ready to deal
1758 * with any packets on eager.
1759 */
1760 if (ipcl_conn_insert(econnp) != 0)
1761 goto error;
1762
1763 ASSERT(econnp->conn_ixa->ixa_notify_cookie == econnp->conn_tcp);
1764 freemsg(mp);
1765 /*
1766 * Send the SYN-ACK. Use the right squeue so that conn_ixa is
1767 * only used by one thread at a time.
1768 */
1769 if (econnp->conn_sqp == lconnp->conn_sqp) {
1770 DTRACE_TCP5(send, mblk_t *, NULL, ip_xmit_attr_t *,
1771 econnp->conn_ixa, __dtrace_tcp_void_ip_t *, mp1->b_rptr,
1772 tcp_t *, eager, __dtrace_tcp_tcph_t *,
1773 &mp1->b_rptr[econnp->conn_ixa->ixa_ip_hdr_length]);
1774 (void) conn_ip_output(mp1, econnp->conn_ixa);
1775 CONN_DEC_REF(econnp);
1776 } else {
1777 SQUEUE_ENTER_ONE(econnp->conn_sqp, mp1, tcp_send_synack,
1778 econnp, NULL, SQ_PROCESS, SQTAG_TCP_SEND_SYNACK);
1779 }
1780 return;
1781 error:
1782 freemsg(mp1);
1783 eager->tcp_closemp_used = B_TRUE;
1784 TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
1785 mp1 = &eager->tcp_closemp;
1786 SQUEUE_ENTER_ONE(econnp->conn_sqp, mp1, tcp_eager_kill,
1787 econnp, NULL, SQ_FILL, SQTAG_TCP_CONN_REQ_2);
1788
1789 /*
1790 * If a connection already exists, send the mp to that connections so
1791 * that it can be appropriately dealt with.
1792 */
1793 ipst = tcps->tcps_netstack->netstack_ip;
1794
1795 if ((econnp = ipcl_classify(mp, ira, ipst)) != NULL) {
1796 if (!IPCL_IS_CONNECTED(econnp)) {
1797 /*
1798 * Something bad happened. ipcl_conn_insert()
1799 * failed because a connection already existed
1800 * in connected hash but we can't find it
1801 * anymore (someone blew it away). Just
1802 * free this message and hopefully remote
1803 * will retransmit at which time the SYN can be
1804 * treated as a new connection or dealth with
1805 * a TH_RST if a connection already exists.
1806 */
1807 CONN_DEC_REF(econnp);
1808 freemsg(mp);
1809 } else {
1810 SQUEUE_ENTER_ONE(econnp->conn_sqp, mp, tcp_input_data,
1811 econnp, ira, SQ_FILL, SQTAG_TCP_CONN_REQ_1);
1812 }
1813 } else {
1814 /* Nobody wants this packet */
1815 freemsg(mp);
1816 }
1817 return;
1818 error3:
1819 CONN_DEC_REF(econnp);
1820 error2:
1821 freemsg(mp);
1822 if (tlc_set)
1823 atomic_dec_32(&listener->tcp_listen_cnt->tlc_cnt);
1824 }
1825
1826 /*
1827 * In an ideal case of vertical partition in NUMA architecture, its
1828 * beneficial to have the listener and all the incoming connections
1829 * tied to the same squeue. The other constraint is that incoming
1830 * connections should be tied to the squeue attached to interrupted
1831 * CPU for obvious locality reason so this leaves the listener to
1832 * be tied to the same squeue. Our only problem is that when listener
1833 * is binding, the CPU that will get interrupted by the NIC whose
1834 * IP address the listener is binding to is not even known. So
1835 * the code below allows us to change that binding at the time the
1836 * CPU is interrupted by virtue of incoming connection's squeue.
1837 *
1838 * This is usefull only in case of a listener bound to a specific IP
1839 * address. For other kind of listeners, they get bound the
1840 * very first time and there is no attempt to rebind them.
1841 */
1842 void
1843 tcp_input_listener_unbound(void *arg, mblk_t *mp, void *arg2,
1844 ip_recv_attr_t *ira)
1845 {
1846 conn_t *connp = (conn_t *)arg;
1847 squeue_t *sqp = (squeue_t *)arg2;
1848 squeue_t *new_sqp;
1849 uint32_t conn_flags;
1850
1851 /*
1852 * IP sets ira_sqp to either the senders conn_sqp (for loopback)
1853 * or based on the ring (for packets from GLD). Otherwise it is
1854 * set based on lbolt i.e., a somewhat random number.
1855 */
1856 ASSERT(ira->ira_sqp != NULL);
1857 new_sqp = ira->ira_sqp;
1858
1859 if (connp->conn_fanout == NULL)
1860 goto done;
1861
1862 if (!(connp->conn_flags & IPCL_FULLY_BOUND)) {
1863 mutex_enter(&connp->conn_fanout->connf_lock);
1864 mutex_enter(&connp->conn_lock);
1865 /*
1866 * No one from read or write side can access us now
1867 * except for already queued packets on this squeue.
1868 * But since we haven't changed the squeue yet, they
1869 * can't execute. If they are processed after we have
1870 * changed the squeue, they are sent back to the
1871 * correct squeue down below.
1872 * But a listner close can race with processing of
1873 * incoming SYN. If incoming SYN processing changes
1874 * the squeue then the listener close which is waiting
1875 * to enter the squeue would operate on the wrong
1876 * squeue. Hence we don't change the squeue here unless
1877 * the refcount is exactly the minimum refcount. The
1878 * minimum refcount of 4 is counted as - 1 each for
1879 * TCP and IP, 1 for being in the classifier hash, and
1880 * 1 for the mblk being processed.
1881 */
1882
1883 if (connp->conn_ref != 4 ||
1884 connp->conn_tcp->tcp_state != TCPS_LISTEN) {
1885 mutex_exit(&connp->conn_lock);
1886 mutex_exit(&connp->conn_fanout->connf_lock);
1887 goto done;
1888 }
1889 if (connp->conn_sqp != new_sqp) {
1890 while (connp->conn_sqp != new_sqp)
1891 (void) atomic_cas_ptr(&connp->conn_sqp, sqp,
1892 new_sqp);
1893 /* No special MT issues for outbound ixa_sqp hint */
1894 connp->conn_ixa->ixa_sqp = new_sqp;
1895 }
1896
1897 do {
1898 conn_flags = connp->conn_flags;
1899 conn_flags |= IPCL_FULLY_BOUND;
1900 (void) atomic_cas_32(&connp->conn_flags,
1901 connp->conn_flags, conn_flags);
1902 } while (!(connp->conn_flags & IPCL_FULLY_BOUND));
1903
1904 mutex_exit(&connp->conn_fanout->connf_lock);
1905 mutex_exit(&connp->conn_lock);
1906
1907 /*
1908 * Assume we have picked a good squeue for the listener. Make
1909 * subsequent SYNs not try to change the squeue.
1910 */
1911 connp->conn_recv = tcp_input_listener;
1912 }
1913
1914 done:
1915 if (connp->conn_sqp != sqp) {
1916 CONN_INC_REF(connp);
1917 SQUEUE_ENTER_ONE(connp->conn_sqp, mp, connp->conn_recv, connp,
1918 ira, SQ_FILL, SQTAG_TCP_CONN_REQ_UNBOUND);
1919 } else {
1920 tcp_input_listener(connp, mp, sqp, ira);
1921 }
1922 }
1923
1924 /*
1925 * Send up all messages queued on tcp_rcv_list.
1926 */
1927 uint_t
1928 tcp_rcv_drain(tcp_t *tcp)
1929 {
1930 mblk_t *mp;
1931 uint_t ret = 0;
1932 #ifdef DEBUG
1933 uint_t cnt = 0;
1934 #endif
1935 queue_t *q = tcp->tcp_connp->conn_rq;
1936
1937 /* Can't drain on an eager connection */
1938 if (tcp->tcp_listener != NULL)
1939 return (ret);
1940
1941 /* Can't be a non-STREAMS connection */
1942 ASSERT(!IPCL_IS_NONSTR(tcp->tcp_connp));
1943
1944 /* No need for the push timer now. */
1945 if (tcp->tcp_push_tid != 0) {
1946 (void) TCP_TIMER_CANCEL(tcp, tcp->tcp_push_tid);
1947 tcp->tcp_push_tid = 0;
1948 }
1949
1950 /*
1951 * Handle two cases here: we are currently fused or we were
1952 * previously fused and have some urgent data to be delivered
1953 * upstream. The latter happens because we either ran out of
1954 * memory or were detached and therefore sending the SIGURG was
1955 * deferred until this point. In either case we pass control
1956 * over to tcp_fuse_rcv_drain() since it may need to complete
1957 * some work.
1958 */
1959 if ((tcp->tcp_fused || tcp->tcp_fused_sigurg)) {
1960 if (tcp_fuse_rcv_drain(q, tcp, tcp->tcp_fused ? NULL :
1961 &tcp->tcp_fused_sigurg_mp))
1962 return (ret);
1963 }
1964
1965 while ((mp = tcp->tcp_rcv_list) != NULL) {
1966 tcp->tcp_rcv_list = mp->b_next;
1967 mp->b_next = NULL;
1968 #ifdef DEBUG
1969 cnt += msgdsize(mp);
1970 #endif
1971 putnext(q, mp);
1972 }
1973 #ifdef DEBUG
1974 ASSERT(cnt == tcp->tcp_rcv_cnt);
1975 #endif
1976 tcp->tcp_rcv_last_head = NULL;
1977 tcp->tcp_rcv_last_tail = NULL;
1978 tcp->tcp_rcv_cnt = 0;
1979
1980 if (canputnext(q))
1981 return (tcp_rwnd_reopen(tcp));
1982
1983 return (ret);
1984 }
1985
1986 /*
1987 * Queue data on tcp_rcv_list which is a b_next chain.
1988 * tcp_rcv_last_head/tail is the last element of this chain.
1989 * Each element of the chain is a b_cont chain.
1990 *
1991 * M_DATA messages are added to the current element.
1992 * Other messages are added as new (b_next) elements.
1993 */
1994 void
1995 tcp_rcv_enqueue(tcp_t *tcp, mblk_t *mp, uint_t seg_len, cred_t *cr)
1996 {
1997 ASSERT(seg_len == msgdsize(mp));
1998 ASSERT(tcp->tcp_rcv_list == NULL || tcp->tcp_rcv_last_head != NULL);
1999
2000 if (is_system_labeled()) {
2001 ASSERT(cr != NULL || msg_getcred(mp, NULL) != NULL);
2002 /*
2003 * Provide for protocols above TCP such as RPC. NOPID leaves
2004 * db_cpid unchanged.
2005 * The cred could have already been set.
2006 */
2007 if (cr != NULL)
2008 mblk_setcred(mp, cr, NOPID);
2009 }
2010
2011 if (tcp->tcp_rcv_list == NULL) {
2012 ASSERT(tcp->tcp_rcv_last_head == NULL);
2013 tcp->tcp_rcv_list = mp;
2014 tcp->tcp_rcv_last_head = mp;
2015 } else if (DB_TYPE(mp) == DB_TYPE(tcp->tcp_rcv_last_head)) {
2016 tcp->tcp_rcv_last_tail->b_cont = mp;
2017 } else {
2018 tcp->tcp_rcv_last_head->b_next = mp;
2019 tcp->tcp_rcv_last_head = mp;
2020 }
2021
2022 while (mp->b_cont)
2023 mp = mp->b_cont;
2024
2025 tcp->tcp_rcv_last_tail = mp;
2026 tcp->tcp_rcv_cnt += seg_len;
2027 tcp->tcp_rwnd -= seg_len;
2028 }
2029
2030 /* Generate an ACK-only (no data) segment for a TCP endpoint */
2031 mblk_t *
2032 tcp_ack_mp(tcp_t *tcp)
2033 {
2034 uint32_t seq_no;
2035 tcp_stack_t *tcps = tcp->tcp_tcps;
2036 conn_t *connp = tcp->tcp_connp;
2037
2038 /*
2039 * There are a few cases to be considered while setting the sequence no.
2040 * Essentially, we can come here while processing an unacceptable pkt
2041 * in the TCPS_SYN_RCVD state, in which case we set the sequence number
2042 * to snxt (per RFC 793), note the swnd wouldn't have been set yet.
2043 * If we are here for a zero window probe, stick with suna. In all
2044 * other cases, we check if suna + swnd encompasses snxt and set
2045 * the sequence number to snxt, if so. If snxt falls outside the
2046 * window (the receiver probably shrunk its window), we will go with
2047 * suna + swnd, otherwise the sequence no will be unacceptable to the
2048 * receiver.
2049 */
2050 if (tcp->tcp_zero_win_probe) {
2051 seq_no = tcp->tcp_suna;
2052 } else if (tcp->tcp_state == TCPS_SYN_RCVD) {
2053 ASSERT(tcp->tcp_swnd == 0);
2054 seq_no = tcp->tcp_snxt;
2055 } else {
2056 seq_no = SEQ_GT(tcp->tcp_snxt,
2057 (tcp->tcp_suna + tcp->tcp_swnd)) ?
2058 (tcp->tcp_suna + tcp->tcp_swnd) : tcp->tcp_snxt;
2059 }
2060
2061 if (tcp->tcp_valid_bits) {
2062 /*
2063 * For the complex case where we have to send some
2064 * controls (FIN or SYN), let tcp_xmit_mp do it.
2065 */
2066 return (tcp_xmit_mp(tcp, NULL, 0, NULL, NULL, seq_no, B_FALSE,
2067 NULL, B_FALSE));
2068 } else {
2069 /* Generate a simple ACK */
2070 int data_length;
2071 uchar_t *rptr;
2072 tcpha_t *tcpha;
2073 mblk_t *mp1;
2074 int32_t total_hdr_len;
2075 int32_t tcp_hdr_len;
2076 int32_t num_sack_blk = 0;
2077 int32_t sack_opt_len;
2078 ip_xmit_attr_t *ixa = connp->conn_ixa;
2079
2080 /*
2081 * Allocate space for TCP + IP headers
2082 * and link-level header
2083 */
2084 if (tcp->tcp_snd_sack_ok && tcp->tcp_num_sack_blk > 0) {
2085 num_sack_blk = MIN(tcp->tcp_max_sack_blk,
2086 tcp->tcp_num_sack_blk);
2087 sack_opt_len = num_sack_blk * sizeof (sack_blk_t) +
2088 TCPOPT_NOP_LEN * 2 + TCPOPT_HEADER_LEN;
2089 total_hdr_len = connp->conn_ht_iphc_len + sack_opt_len;
2090 tcp_hdr_len = connp->conn_ht_ulp_len + sack_opt_len;
2091 } else {
2092 total_hdr_len = connp->conn_ht_iphc_len;
2093 tcp_hdr_len = connp->conn_ht_ulp_len;
2094 }
2095 mp1 = allocb(total_hdr_len + tcps->tcps_wroff_xtra, BPRI_MED);
2096 if (!mp1)
2097 return (NULL);
2098
2099 /* Update the latest receive window size in TCP header. */
2100 tcp->tcp_tcpha->tha_win =
2101 htons(tcp->tcp_rwnd >> tcp->tcp_rcv_ws);
2102 /* copy in prototype TCP + IP header */
2103 rptr = mp1->b_rptr + tcps->tcps_wroff_xtra;
2104 mp1->b_rptr = rptr;
2105 mp1->b_wptr = rptr + total_hdr_len;
2106 bcopy(connp->conn_ht_iphc, rptr, connp->conn_ht_iphc_len);
2107
2108 tcpha = (tcpha_t *)&rptr[ixa->ixa_ip_hdr_length];
2109
2110 /* Set the TCP sequence number. */
2111 tcpha->tha_seq = htonl(seq_no);
2112
2113 /* Set up the TCP flag field. */
2114 tcpha->tha_flags = (uchar_t)TH_ACK;
2115 if (tcp->tcp_ecn_echo_on)
2116 tcpha->tha_flags |= TH_ECE;
2117
2118 tcp->tcp_rack = tcp->tcp_rnxt;
2119 tcp->tcp_rack_cnt = 0;
2120
2121 /* fill in timestamp option if in use */
2122 if (tcp->tcp_snd_ts_ok) {
2123 uint32_t llbolt = (uint32_t)LBOLT_FASTPATH;
2124
2125 U32_TO_BE32(llbolt,
2126 (char *)tcpha + TCP_MIN_HEADER_LENGTH+4);
2127 U32_TO_BE32(tcp->tcp_ts_recent,
2128 (char *)tcpha + TCP_MIN_HEADER_LENGTH+8);
2129 }
2130
2131 /* Fill in SACK options */
2132 if (num_sack_blk > 0) {
2133 uchar_t *wptr = (uchar_t *)tcpha +
2134 connp->conn_ht_ulp_len;
2135 sack_blk_t *tmp;
2136 int32_t i;
2137
2138 wptr[0] = TCPOPT_NOP;
2139 wptr[1] = TCPOPT_NOP;
2140 wptr[2] = TCPOPT_SACK;
2141 wptr[3] = TCPOPT_HEADER_LEN + num_sack_blk *
2142 sizeof (sack_blk_t);
2143 wptr += TCPOPT_REAL_SACK_LEN;
2144
2145 tmp = tcp->tcp_sack_list;
2146 for (i = 0; i < num_sack_blk; i++) {
2147 U32_TO_BE32(tmp[i].begin, wptr);
2148 wptr += sizeof (tcp_seq);
2149 U32_TO_BE32(tmp[i].end, wptr);
2150 wptr += sizeof (tcp_seq);
2151 }
2152 tcpha->tha_offset_and_reserved +=
2153 ((num_sack_blk * 2 + 1) << 4);
2154 }
2155
2156 ixa->ixa_pktlen = total_hdr_len;
2157
2158 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2159 ((ipha_t *)rptr)->ipha_length = htons(total_hdr_len);
2160 } else {
2161 ip6_t *ip6 = (ip6_t *)rptr;
2162
2163 ip6->ip6_plen = htons(total_hdr_len - IPV6_HDR_LEN);
2164 }
2165
2166 /*
2167 * Prime pump for checksum calculation in IP. Include the
2168 * adjustment for a source route if any.
2169 */
2170 data_length = tcp_hdr_len + connp->conn_sum;
2171 data_length = (data_length >> 16) + (data_length & 0xFFFF);
2172 tcpha->tha_sum = htons(data_length);
2173
2174 if (tcp->tcp_ip_forward_progress) {
2175 tcp->tcp_ip_forward_progress = B_FALSE;
2176 connp->conn_ixa->ixa_flags |= IXAF_REACH_CONF;
2177 } else {
2178 connp->conn_ixa->ixa_flags &= ~IXAF_REACH_CONF;
2179 }
2180 return (mp1);
2181 }
2182 }
2183
2184 /*
2185 * Dummy socket upcalls for if/when the conn_t gets detached from a
2186 * direct-callback sonode via a user-driven close(). Easy to catch with
2187 * DTrace FBT, and should be mostly harmless.
2188 */
2189
2190 /* ARGSUSED */
2191 static sock_upper_handle_t
2192 tcp_dummy_newconn(sock_upper_handle_t x, sock_lower_handle_t y,
2193 sock_downcalls_t *z, cred_t *cr, pid_t pid, sock_upcalls_t **ignored)
2194 {
2195 ASSERT(0); /* Panic in debug, otherwise ignore. */
2196 return (NULL);
2197 }
2198
2199 /* ARGSUSED */
2200 static void
2201 tcp_dummy_connected(sock_upper_handle_t x, sock_connid_t y, cred_t *cr,
2202 pid_t pid)
2203 {
2204 ASSERT(x == NULL);
2205 /* Normally we'd crhold(cr) and attach it to socket state. */
2206 /* LINTED */
2207 }
2208
2209 /* ARGSUSED */
2210 static int
2211 tcp_dummy_disconnected(sock_upper_handle_t x, sock_connid_t y, int blah)
2212 {
2213 ASSERT(0); /* Panic in debug, otherwise ignore. */
2214 return (-1);
2215 }
2216
2217 /* ARGSUSED */
2218 static void
2219 tcp_dummy_opctl(sock_upper_handle_t x, sock_opctl_action_t y, uintptr_t blah)
2220 {
2221 ASSERT(x == NULL);
2222 /* We really want this one to be a harmless NOP for now. */
2223 /* LINTED */
2224 }
2225
2226 /* ARGSUSED */
2227 static ssize_t
2228 tcp_dummy_recv(sock_upper_handle_t x, mblk_t *mp, size_t len, int flags,
2229 int *error, boolean_t *push)
2230 {
2231 ASSERT(x == NULL);
2232
2233 /*
2234 * Consume the message, set ESHUTDOWN, and return an error.
2235 * Nobody's home!
2236 */
2237 freemsg(mp);
2238 *error = ESHUTDOWN;
2239 return (-1);
2240 }
2241
2242 /* ARGSUSED */
2243 static void
2244 tcp_dummy_set_proto_props(sock_upper_handle_t x, struct sock_proto_props *y)
2245 {
2246 ASSERT(0); /* Panic in debug, otherwise ignore. */
2247 }
2248
2249 /* ARGSUSED */
2250 static void
2251 tcp_dummy_txq_full(sock_upper_handle_t x, boolean_t y)
2252 {
2253 ASSERT(0); /* Panic in debug, otherwise ignore. */
2254 }
2255
2256 /* ARGSUSED */
2257 static void
2258 tcp_dummy_signal_oob(sock_upper_handle_t x, ssize_t len)
2259 {
2260 ASSERT(x == NULL);
2261 /* Otherwise, this would signal socket state about OOB data. */
2262 }
2263
2264 /* ARGSUSED */
2265 static void
2266 tcp_dummy_set_error(sock_upper_handle_t x, int err)
2267 {
2268 ASSERT(0); /* Panic in debug, otherwise ignore. */
2269 }
2270
2271 /* ARGSUSED */
2272 static void
2273 tcp_dummy_onearg(sock_upper_handle_t x)
2274 {
2275 ASSERT(0); /* Panic in debug, otherwise ignore. */
2276 }
2277
2278 static sock_upcalls_t tcp_dummy_upcalls = {
2279 tcp_dummy_newconn,
2280 tcp_dummy_connected,
2281 tcp_dummy_disconnected,
2282 tcp_dummy_opctl,
2283 tcp_dummy_recv,
2284 tcp_dummy_set_proto_props,
2285 tcp_dummy_txq_full,
2286 tcp_dummy_signal_oob,
2287 tcp_dummy_onearg,
2288 tcp_dummy_set_error,
2289 tcp_dummy_onearg
2290 };
2291
2292 /*
2293 * Handle M_DATA messages from IP. Its called directly from IP via
2294 * squeue for received IP packets.
2295 *
2296 * The first argument is always the connp/tcp to which the mp belongs.
2297 * There are no exceptions to this rule. The caller has already put
2298 * a reference on this connp/tcp and once tcp_input_data() returns,
2299 * the squeue will do the refrele.
2300 *
2301 * The TH_SYN for the listener directly go to tcp_input_listener via
2302 * squeue. ICMP errors go directly to tcp_icmp_input().
2303 *
2304 * sqp: NULL = recursive, sqp != NULL means called from squeue
2305 */
2306 void
2307 tcp_input_data(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
2308 {
2309 int32_t bytes_acked;
2310 int32_t gap;
2311 mblk_t *mp1;
2312 uint_t flags;
2313 uint32_t new_swnd = 0;
2314 uchar_t *iphdr;
2315 uchar_t *rptr;
2316 int32_t rgap;
2317 uint32_t seg_ack;
2318 int seg_len;
2319 uint_t ip_hdr_len;
2320 uint32_t seg_seq;
2321 tcpha_t *tcpha;
2322 int urp;
2323 tcp_opt_t tcpopt;
2324 ip_pkt_t ipp;
2325 boolean_t ofo_seg = B_FALSE; /* Out of order segment */
2326 uint32_t cwnd;
2327 uint32_t add;
2328 int npkt;
2329 int mss;
2330 conn_t *connp = (conn_t *)arg;
2331 squeue_t *sqp = (squeue_t *)arg2;
2332 tcp_t *tcp = connp->conn_tcp;
2333 tcp_stack_t *tcps = tcp->tcp_tcps;
2334 sock_upcalls_t *sockupcalls;
2335
2336 /*
2337 * RST from fused tcp loopback peer should trigger an unfuse.
2338 */
2339 if (tcp->tcp_fused) {
2340 TCP_STAT(tcps, tcp_fusion_aborted);
2341 tcp_unfuse(tcp);
2342 }
2343
2344 iphdr = mp->b_rptr;
2345 rptr = mp->b_rptr;
2346 ASSERT(OK_32PTR(rptr));
2347
2348 ip_hdr_len = ira->ira_ip_hdr_length;
2349 if (connp->conn_recv_ancillary.crb_all != 0) {
2350 /*
2351 * Record packet information in the ip_pkt_t
2352 */
2353 ipp.ipp_fields = 0;
2354 if (ira->ira_flags & IRAF_IS_IPV4) {
2355 (void) ip_find_hdr_v4((ipha_t *)rptr, &ipp,
2356 B_FALSE);
2357 } else {
2358 uint8_t nexthdrp;
2359
2360 /*
2361 * IPv6 packets can only be received by applications
2362 * that are prepared to receive IPv6 addresses.
2363 * The IP fanout must ensure this.
2364 */
2365 ASSERT(connp->conn_family == AF_INET6);
2366
2367 (void) ip_find_hdr_v6(mp, (ip6_t *)rptr, B_TRUE, &ipp,
2368 &nexthdrp);
2369 ASSERT(nexthdrp == IPPROTO_TCP);
2370
2371 /* Could have caused a pullup? */
2372 iphdr = mp->b_rptr;
2373 rptr = mp->b_rptr;
2374 }
2375 }
2376 ASSERT(DB_TYPE(mp) == M_DATA);
2377 ASSERT(mp->b_next == NULL);
2378
2379 tcpha = (tcpha_t *)&rptr[ip_hdr_len];
2380 seg_seq = ntohl(tcpha->tha_seq);
2381 seg_ack = ntohl(tcpha->tha_ack);
2382 ASSERT((uintptr_t)(mp->b_wptr - rptr) <= (uintptr_t)INT_MAX);
2383 seg_len = (int)(mp->b_wptr - rptr) -
2384 (ip_hdr_len + TCP_HDR_LENGTH(tcpha));
2385 if ((mp1 = mp->b_cont) != NULL && mp1->b_datap->db_type == M_DATA) {
2386 do {
2387 ASSERT((uintptr_t)(mp1->b_wptr - mp1->b_rptr) <=
2388 (uintptr_t)INT_MAX);
2389 seg_len += (int)(mp1->b_wptr - mp1->b_rptr);
2390 } while ((mp1 = mp1->b_cont) != NULL &&
2391 mp1->b_datap->db_type == M_DATA);
2392 }
2393
2394 DTRACE_TCP5(receive, mblk_t *, NULL, ip_xmit_attr_t *, connp->conn_ixa,
2395 __dtrace_tcp_void_ip_t *, iphdr, tcp_t *, tcp,
2396 __dtrace_tcp_tcph_t *, tcpha);
2397
2398 if (tcp->tcp_state == TCPS_TIME_WAIT) {
2399 tcp_time_wait_processing(tcp, mp, seg_seq, seg_ack,
2400 seg_len, tcpha, ira);
2401 return;
2402 }
2403
2404 if (sqp != NULL) {
2405 /*
2406 * This is the correct place to update tcp_last_recv_time. Note
2407 * that it is also updated for tcp structure that belongs to
2408 * global and listener queues which do not really need updating.
2409 * But that should not cause any harm. And it is updated for
2410 * all kinds of incoming segments, not only for data segments.
2411 */
2412 tcp->tcp_last_recv_time = LBOLT_FASTPATH;
2413 }
2414
2415 flags = (unsigned int)tcpha->tha_flags & 0xFF;
2416
2417 TCPS_BUMP_MIB(tcps, tcpHCInSegs);
2418 DTRACE_PROBE2(tcp__trace__recv, mblk_t *, mp, tcp_t *, tcp);
2419
2420 if ((flags & TH_URG) && sqp != NULL) {
2421 /*
2422 * TCP can't handle urgent pointers that arrive before
2423 * the connection has been accept()ed since it can't
2424 * buffer OOB data. Discard segment if this happens.
2425 *
2426 * We can't just rely on a non-null tcp_listener to indicate
2427 * that the accept() has completed since unlinking of the
2428 * eager and completion of the accept are not atomic.
2429 * tcp_detached, when it is not set (B_FALSE) indicates
2430 * that the accept() has completed.
2431 *
2432 * Nor can it reassemble urgent pointers, so discard
2433 * if it's not the next segment expected.
2434 *
2435 * Otherwise, collapse chain into one mblk (discard if
2436 * that fails). This makes sure the headers, retransmitted
2437 * data, and new data all are in the same mblk.
2438 */
2439 ASSERT(mp != NULL);
2440 if (tcp->tcp_detached || !pullupmsg(mp, -1)) {
2441 freemsg(mp);
2442 return;
2443 }
2444 /* Update pointers into message */
2445 iphdr = rptr = mp->b_rptr;
2446 tcpha = (tcpha_t *)&rptr[ip_hdr_len];
2447 if (SEQ_GT(seg_seq, tcp->tcp_rnxt)) {
2448 /*
2449 * Since we can't handle any data with this urgent
2450 * pointer that is out of sequence, we expunge
2451 * the data. This allows us to still register
2452 * the urgent mark and generate the M_PCSIG,
2453 * which we can do.
2454 */
2455 mp->b_wptr = (uchar_t *)tcpha + TCP_HDR_LENGTH(tcpha);
2456 seg_len = 0;
2457 }
2458 }
2459
2460 sockupcalls = connp->conn_upcalls;
2461 /* A conn_t may have belonged to a now-closed socket. Be careful. */
2462 if (sockupcalls == NULL)
2463 sockupcalls = &tcp_dummy_upcalls;
2464
2465 switch (tcp->tcp_state) {
2466 case TCPS_SYN_SENT:
2467 if (connp->conn_final_sqp == NULL &&
2468 tcp_outbound_squeue_switch && sqp != NULL) {
2469 ASSERT(connp->conn_initial_sqp == connp->conn_sqp);
2470 connp->conn_final_sqp = sqp;
2471 if (connp->conn_final_sqp != connp->conn_sqp) {
2472 DTRACE_PROBE1(conn__final__sqp__switch,
2473 conn_t *, connp);
2474 CONN_INC_REF(connp);
2475 SQUEUE_SWITCH(connp, connp->conn_final_sqp);
2476 SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2477 tcp_input_data, connp, ira, ip_squeue_flag,
2478 SQTAG_CONNECT_FINISH);
2479 return;
2480 }
2481 DTRACE_PROBE1(conn__final__sqp__same, conn_t *, connp);
2482 }
2483 if (flags & TH_ACK) {
2484 /*
2485 * Note that our stack cannot send data before a
2486 * connection is established, therefore the
2487 * following check is valid. Otherwise, it has
2488 * to be changed.
2489 */
2490 if (SEQ_LEQ(seg_ack, tcp->tcp_iss) ||
2491 SEQ_GT(seg_ack, tcp->tcp_snxt)) {
2492 freemsg(mp);
2493 if (flags & TH_RST)
2494 return;
2495 tcp_xmit_ctl("TCPS_SYN_SENT-Bad_seq",
2496 tcp, seg_ack, 0, TH_RST);
2497 return;
2498 }
2499 ASSERT(tcp->tcp_suna + 1 == seg_ack);
2500 }
2501 if (flags & TH_RST) {
2502 if (flags & TH_ACK) {
2503 DTRACE_TCP5(connect__refused, mblk_t *, NULL,
2504 ip_xmit_attr_t *, connp->conn_ixa,
2505 void_ip_t *, iphdr, tcp_t *, tcp,
2506 tcph_t *, tcpha);
2507 (void) tcp_clean_death(tcp, ECONNREFUSED);
2508 }
2509 freemsg(mp);
2510 return;
2511 }
2512 if (!(flags & TH_SYN)) {
2513 freemsg(mp);
2514 return;
2515 }
2516
2517 /* Process all TCP options. */
2518 tcp_process_options(tcp, tcpha);
2519 /*
2520 * The following changes our rwnd to be a multiple of the
2521 * MIN(peer MSS, our MSS) for performance reason.
2522 */
2523 (void) tcp_rwnd_set(tcp, MSS_ROUNDUP(connp->conn_rcvbuf,
2524 tcp->tcp_mss));
2525
2526 /* Is the other end ECN capable? */
2527 if (tcp->tcp_ecn_ok) {
2528 if ((flags & (TH_ECE|TH_CWR)) != TH_ECE) {
2529 tcp->tcp_ecn_ok = B_FALSE;
2530 }
2531 }
2532 /*
2533 * Clear ECN flags because it may interfere with later
2534 * processing.
2535 */
2536 flags &= ~(TH_ECE|TH_CWR);
2537
2538 tcp->tcp_irs = seg_seq;
2539 tcp->tcp_rack = seg_seq;
2540 tcp->tcp_rnxt = seg_seq + 1;
2541 tcp->tcp_tcpha->tha_ack = htonl(tcp->tcp_rnxt);
2542 if (!TCP_IS_DETACHED(tcp)) {
2543 /* Allocate room for SACK options if needed. */
2544 connp->conn_wroff = connp->conn_ht_iphc_len;
2545 if (tcp->tcp_snd_sack_ok)
2546 connp->conn_wroff += TCPOPT_MAX_SACK_LEN;
2547 if (!tcp->tcp_loopback)
2548 connp->conn_wroff += tcps->tcps_wroff_xtra;
2549
2550 (void) proto_set_tx_wroff(connp->conn_rq, connp,
2551 connp->conn_wroff);
2552 }
2553 if (flags & TH_ACK) {
2554 /*
2555 * If we can't get the confirmation upstream, pretend
2556 * we didn't even see this one.
2557 *
2558 * XXX: how can we pretend we didn't see it if we
2559 * have updated rnxt et. al.
2560 *
2561 * For loopback we defer sending up the T_CONN_CON
2562 * until after some checks below.
2563 */
2564 mp1 = NULL;
2565 /*
2566 * tcp_sendmsg() checks tcp_state without entering
2567 * the squeue so tcp_state should be updated before
2568 * sending up connection confirmation. Probe the
2569 * state change below when we are sure the connection
2570 * confirmation has been sent.
2571 */
2572 tcp->tcp_state = TCPS_ESTABLISHED;
2573 if (!tcp_conn_con(tcp, iphdr, mp,
2574 tcp->tcp_loopback ? &mp1 : NULL, ira)) {
2575 tcp->tcp_state = TCPS_SYN_SENT;
2576 freemsg(mp);
2577 return;
2578 }
2579 TCPS_CONN_INC(tcps);
2580 /* SYN was acked - making progress */
2581 tcp->tcp_ip_forward_progress = B_TRUE;
2582
2583 /* One for the SYN */
2584 tcp->tcp_suna = tcp->tcp_iss + 1;
2585 tcp->tcp_valid_bits &= ~TCP_ISS_VALID;
2586
2587 /*
2588 * If SYN was retransmitted, need to reset all
2589 * retransmission info. This is because this
2590 * segment will be treated as a dup ACK.
2591 */
2592 if (tcp->tcp_rexmit) {
2593 tcp->tcp_rexmit = B_FALSE;
2594 tcp->tcp_rexmit_nxt = tcp->tcp_snxt;
2595 tcp->tcp_rexmit_max = tcp->tcp_snxt;
2596 tcp->tcp_ms_we_have_waited = 0;
2597
2598 /*
2599 * Set tcp_cwnd back to 1 MSS, per
2600 * recommendation from
2601 * draft-floyd-incr-init-win-01.txt,
2602 * Increasing TCP's Initial Window.
2603 */
2604 tcp->tcp_cwnd = tcp->tcp_mss;
2605 }
2606
2607 tcp->tcp_swl1 = seg_seq;
2608 tcp->tcp_swl2 = seg_ack;
2609
2610 new_swnd = ntohs(tcpha->tha_win);
2611 tcp->tcp_swnd = new_swnd;
2612 if (new_swnd > tcp->tcp_max_swnd)
2613 tcp->tcp_max_swnd = new_swnd;
2614
2615 /*
2616 * Always send the three-way handshake ack immediately
2617 * in order to make the connection complete as soon as
2618 * possible on the accepting host.
2619 */
2620 flags |= TH_ACK_NEEDED;
2621
2622 /*
2623 * Trace connect-established here.
2624 */
2625 DTRACE_TCP5(connect__established, mblk_t *, NULL,
2626 ip_xmit_attr_t *, tcp->tcp_connp->conn_ixa,
2627 void_ip_t *, iphdr, tcp_t *, tcp, tcph_t *, tcpha);
2628
2629 /* Trace change from SYN_SENT -> ESTABLISHED here */
2630 DTRACE_TCP6(state__change, void, NULL, ip_xmit_attr_t *,
2631 connp->conn_ixa, void, NULL, tcp_t *, tcp,
2632 void, NULL, int32_t, TCPS_SYN_SENT);
2633
2634 /*
2635 * Special case for loopback. At this point we have
2636 * received SYN-ACK from the remote endpoint. In
2637 * order to ensure that both endpoints reach the
2638 * fused state prior to any data exchange, the final
2639 * ACK needs to be sent before we indicate T_CONN_CON
2640 * to the module upstream.
2641 */
2642 if (tcp->tcp_loopback) {
2643 mblk_t *ack_mp;
2644
2645 ASSERT(!tcp->tcp_unfusable);
2646 ASSERT(mp1 != NULL);
2647 /*
2648 * For loopback, we always get a pure SYN-ACK
2649 * and only need to send back the final ACK
2650 * with no data (this is because the other
2651 * tcp is ours and we don't do T/TCP). This
2652 * final ACK triggers the passive side to
2653 * perform fusion in ESTABLISHED state.
2654 */
2655 if ((ack_mp = tcp_ack_mp(tcp)) != NULL) {
2656 if (tcp->tcp_ack_tid != 0) {
2657 (void) TCP_TIMER_CANCEL(tcp,
2658 tcp->tcp_ack_tid);
2659 tcp->tcp_ack_tid = 0;
2660 }
2661 tcp_send_data(tcp, ack_mp);
2662 TCPS_BUMP_MIB(tcps, tcpHCOutSegs);
2663 TCPS_BUMP_MIB(tcps, tcpOutAck);
2664
2665 if (!IPCL_IS_NONSTR(connp)) {
2666 /* Send up T_CONN_CON */
2667 if (ira->ira_cred != NULL) {
2668 mblk_setcred(mp1,
2669 ira->ira_cred,
2670 ira->ira_cpid);
2671 }
2672 putnext(connp->conn_rq, mp1);
2673 } else {
2674 (*sockupcalls->su_connected)
2675 (connp->conn_upper_handle,
2676 tcp->tcp_connid,
2677 ira->ira_cred,
2678 ira->ira_cpid);
2679 freemsg(mp1);
2680 }
2681
2682 freemsg(mp);
2683 return;
2684 }
2685 /*
2686 * Forget fusion; we need to handle more
2687 * complex cases below. Send the deferred
2688 * T_CONN_CON message upstream and proceed
2689 * as usual. Mark this tcp as not capable
2690 * of fusion.
2691 */
2692 TCP_STAT(tcps, tcp_fusion_unfusable);
2693 tcp->tcp_unfusable = B_TRUE;
2694 if (!IPCL_IS_NONSTR(connp)) {
2695 if (ira->ira_cred != NULL) {
2696 mblk_setcred(mp1, ira->ira_cred,
2697 ira->ira_cpid);
2698 }
2699 putnext(connp->conn_rq, mp1);
2700 } else {
2701 (*sockupcalls->su_connected)
2702 (connp->conn_upper_handle,
2703 tcp->tcp_connid, ira->ira_cred,
2704 ira->ira_cpid);
2705 freemsg(mp1);
2706 }
2707 }
2708
2709 /*
2710 * Check to see if there is data to be sent. If
2711 * yes, set the transmit flag. Then check to see
2712 * if received data processing needs to be done.
2713 * If not, go straight to xmit_check. This short
2714 * cut is OK as we don't support T/TCP.
2715 */
2716 if (tcp->tcp_unsent)
2717 flags |= TH_XMIT_NEEDED;
2718
2719 if (seg_len == 0 && !(flags & TH_URG)) {
2720 freemsg(mp);
2721 goto xmit_check;
2722 }
2723
2724 flags &= ~TH_SYN;
2725 seg_seq++;
2726 break;
2727 }
2728 tcp->tcp_state = TCPS_SYN_RCVD;
2729 DTRACE_TCP6(state__change, void, NULL, ip_xmit_attr_t *,
2730 connp->conn_ixa, void_ip_t *, NULL, tcp_t *, tcp,
2731 tcph_t *, NULL, int32_t, TCPS_SYN_SENT);
2732 mp1 = tcp_xmit_mp(tcp, tcp->tcp_xmit_head, tcp->tcp_mss,
2733 NULL, NULL, tcp->tcp_iss, B_FALSE, NULL, B_FALSE);
2734 if (mp1 != NULL) {
2735 tcp_send_data(tcp, mp1);
2736 TCP_TIMER_RESTART(tcp, tcp->tcp_rto);
2737 }
2738 freemsg(mp);
2739 return;
2740 case TCPS_SYN_RCVD:
2741 if (flags & TH_ACK) {
2742 uint32_t pinit_wnd;
2743
2744 /*
2745 * In this state, a SYN|ACK packet is either bogus
2746 * because the other side must be ACKing our SYN which
2747 * indicates it has seen the ACK for their SYN and
2748 * shouldn't retransmit it or we're crossing SYNs
2749 * on active open.
2750 */
2751 if ((flags & TH_SYN) && !tcp->tcp_active_open) {
2752 freemsg(mp);
2753 tcp_xmit_ctl("TCPS_SYN_RCVD-bad_syn",
2754 tcp, seg_ack, 0, TH_RST);
2755 return;
2756 }
2757 /*
2758 * NOTE: RFC 793 pg. 72 says this should be
2759 * tcp->tcp_suna <= seg_ack <= tcp->tcp_snxt
2760 * but that would mean we have an ack that ignored
2761 * our SYN.
2762 */
2763 if (SEQ_LEQ(seg_ack, tcp->tcp_suna) ||
2764 SEQ_GT(seg_ack, tcp->tcp_snxt)) {
2765 freemsg(mp);
2766 tcp_xmit_ctl("TCPS_SYN_RCVD-bad_ack",
2767 tcp, seg_ack, 0, TH_RST);
2768 return;
2769 }
2770 /*
2771 * No sane TCP stack will send such a small window
2772 * without receiving any data. Just drop this invalid
2773 * ACK. We also shorten the abort timeout in case
2774 * this is an attack.
2775 */
2776 pinit_wnd = ntohs(tcpha->tha_win) << tcp->tcp_snd_ws;
2777 if (pinit_wnd < tcp->tcp_mss &&
2778 pinit_wnd < tcp_init_wnd_chk) {
2779 freemsg(mp);
2780 TCP_STAT(tcps, tcp_zwin_ack_syn);
2781 tcp->tcp_second_ctimer_threshold =
2782 tcp_early_abort * SECONDS;
2783 return;
2784 }
2785 }
2786 break;
2787 case TCPS_LISTEN:
2788 /*
2789 * Only a TLI listener can come through this path when a
2790 * acceptor is going back to be a listener and a packet
2791 * for the acceptor hits the classifier. For a socket
2792 * listener, this can never happen because a listener
2793 * can never accept connection on itself and hence a
2794 * socket acceptor can not go back to being a listener.
2795 */
2796 ASSERT(!TCP_IS_SOCKET(tcp));
2797 /*FALLTHRU*/
2798 case TCPS_CLOSED:
2799 case TCPS_BOUND: {
2800 conn_t *new_connp;
2801 ip_stack_t *ipst = tcps->tcps_netstack->netstack_ip;
2802
2803 /*
2804 * Don't accept any input on a closed tcp as this TCP logically
2805 * does not exist on the system. Don't proceed further with
2806 * this TCP. For instance, this packet could trigger another
2807 * close of this tcp which would be disastrous for tcp_refcnt.
2808 * tcp_close_detached / tcp_clean_death / tcp_closei_local must
2809 * be called at most once on a TCP. In this case we need to
2810 * refeed the packet into the classifier and figure out where
2811 * the packet should go.
2812 */
2813 new_connp = ipcl_classify(mp, ira, ipst);
2814 if (new_connp != NULL) {
2815 /* Drops ref on new_connp */
2816 tcp_reinput(new_connp, mp, ira, ipst);
2817 return;
2818 }
2819 /* We failed to classify. For now just drop the packet */
2820 freemsg(mp);
2821 return;
2822 }
2823 case TCPS_IDLE:
2824 /*
2825 * Handle the case where the tcp_clean_death() has happened
2826 * on a connection (application hasn't closed yet) but a packet
2827 * was already queued on squeue before tcp_clean_death()
2828 * was processed. Calling tcp_clean_death() twice on same
2829 * connection can result in weird behaviour.
2830 */
2831 freemsg(mp);
2832 return;
2833 default:
2834 break;
2835 }
2836
2837 /*
2838 * Already on the correct queue/perimeter.
2839 * If this is a detached connection and not an eager
2840 * connection hanging off a listener then new data
2841 * (past the FIN) will cause a reset.
2842 * We do a special check here where it
2843 * is out of the main line, rather than check
2844 * if we are detached every time we see new
2845 * data down below.
2846 */
2847 if (TCP_IS_DETACHED_NONEAGER(tcp) &&
2848 (seg_len > 0 && SEQ_GT(seg_seq + seg_len, tcp->tcp_rnxt))) {
2849 TCPS_BUMP_MIB(tcps, tcpInClosed);
2850 DTRACE_PROBE2(tcp__trace__recv, mblk_t *, mp, tcp_t *, tcp);
2851 freemsg(mp);
2852 tcp_xmit_ctl("new data when detached", tcp,
2853 tcp->tcp_snxt, 0, TH_RST);
2854 (void) tcp_clean_death(tcp, EPROTO);
2855 return;
2856 }
2857
2858 mp->b_rptr = (uchar_t *)tcpha + TCP_HDR_LENGTH(tcpha);
2859 urp = ntohs(tcpha->tha_urp) - TCP_OLD_URP_INTERPRETATION;
2860 new_swnd = ntohs(tcpha->tha_win) <<
2861 ((tcpha->tha_flags & TH_SYN) ? 0 : tcp->tcp_snd_ws);
2862
2863 /*
2864 * We are interested in two TCP options: timestamps (if negotiated) and
2865 * SACK (if negotiated). Skip option parsing if neither is negotiated.
2866 */
2867 if (tcp->tcp_snd_ts_ok || tcp->tcp_snd_sack_ok) {
2868 int options;
2869 if (tcp->tcp_snd_sack_ok)
2870 tcpopt.tcp = tcp;
2871 else
2872 tcpopt.tcp = NULL;
2873 options = tcp_parse_options(tcpha, &tcpopt);
2874 /*
2875 * RST segments must not be subject to PAWS and are not
2876 * required to have timestamps.
2877 * We do not drop keepalive segments without
2878 * timestamps, to maintain compatibility with legacy TCP stacks.
2879 */
2880 boolean_t keepalive = (seg_len == 0 || seg_len == 1) &&
2881 (seg_seq + 1 == tcp->tcp_rnxt);
2882 if (tcp->tcp_snd_ts_ok && !(flags & TH_RST) && !keepalive) {
2883 /*
2884 * Per RFC 7323 section 3.2., silently drop non-RST
2885 * segments without expected TSopt. This is a 'SHOULD'
2886 * requirement.
2887 * We accept keepalives without TSopt to maintain
2888 * interoperability with tcp implementations that omit
2889 * the TSopt on these. Keepalive data is discarded, so
2890 * there is no risk corrupting data by accepting these.
2891 */
2892 if (!(options & TCP_OPT_TSTAMP_PRESENT)) {
2893 /*
2894 * Leave a breadcrumb for people to detect this
2895 * behavior.
2896 */
2897 DTRACE_TCP1(droppedtimestamp, tcp_t *, tcp);
2898 freemsg(mp);
2899 return;
2900 }
2901
2902 if (!tcp_paws_check(tcp, &tcpopt)) {
2903 /*
2904 * This segment is not acceptable.
2905 * Drop it and send back an ACK.
2906 */
2907 freemsg(mp);
2908 flags |= TH_ACK_NEEDED;
2909 goto ack_check;
2910 }
2911 }
2912 }
2913 try_again:;
2914 mss = tcp->tcp_mss;
2915 gap = seg_seq - tcp->tcp_rnxt;
2916 rgap = tcp->tcp_rwnd - (gap + seg_len);
2917 /*
2918 * gap is the amount of sequence space between what we expect to see
2919 * and what we got for seg_seq. A positive value for gap means
2920 * something got lost. A negative value means we got some old stuff.
2921 */
2922 if (gap < 0) {
2923 /* Old stuff present. Is the SYN in there? */
2924 if (seg_seq == tcp->tcp_irs && (flags & TH_SYN) &&
2925 (seg_len != 0)) {
2926 flags &= ~TH_SYN;
2927 seg_seq++;
2928 urp--;
2929 /* Recompute the gaps after noting the SYN. */
2930 goto try_again;
2931 }
2932 TCPS_BUMP_MIB(tcps, tcpInDataDupSegs);
2933 TCPS_UPDATE_MIB(tcps, tcpInDataDupBytes,
2934 (seg_len > -gap ? -gap : seg_len));
2935 /* Remove the old stuff from seg_len. */
2936 seg_len += gap;
2937 /*
2938 * Anything left?
2939 * Make sure to check for unack'd FIN when rest of data
2940 * has been previously ack'd.
2941 */
2942 if (seg_len < 0 || (seg_len == 0 && !(flags & TH_FIN))) {
2943 /*
2944 * Resets are only valid if they lie within our offered
2945 * window. If the RST bit is set, we just ignore this
2946 * segment.
2947 */
2948 if (flags & TH_RST) {
2949 freemsg(mp);
2950 return;
2951 }
2952
2953 /*
2954 * The arriving of dup data packets indicate that we
2955 * may have postponed an ack for too long, or the other
2956 * side's RTT estimate is out of shape. Start acking
2957 * more often.
2958 */
2959 if (SEQ_GEQ(seg_seq + seg_len - gap, tcp->tcp_rack) &&
2960 tcp->tcp_rack_cnt >= 1 &&
2961 tcp->tcp_rack_abs_max > 2) {
2962 tcp->tcp_rack_abs_max--;
2963 }
2964 tcp->tcp_rack_cur_max = 1;
2965
2966 /*
2967 * This segment is "unacceptable". None of its
2968 * sequence space lies within our advertized window.
2969 *
2970 * Adjust seg_len to the original value for tracing.
2971 */
2972 seg_len -= gap;
2973 if (connp->conn_debug) {
2974 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
2975 "tcp_rput: unacceptable, gap %d, rgap %d, "
2976 "flags 0x%x, seg_seq %u, seg_ack %u, "
2977 "seg_len %d, rnxt %u, snxt %u, %s",
2978 gap, rgap, flags, seg_seq, seg_ack,
2979 seg_len, tcp->tcp_rnxt, tcp->tcp_snxt,
2980 tcp_display(tcp, NULL,
2981 DISP_ADDR_AND_PORT));
2982 }
2983
2984 /*
2985 * Arrange to send an ACK in response to the
2986 * unacceptable segment per RFC 793 page 69. There
2987 * is only one small difference between ours and the
2988 * acceptability test in the RFC - we accept ACK-only
2989 * packet with SEG.SEQ = RCV.NXT+RCV.WND and no ACK
2990 * will be generated.
2991 *
2992 * Note that we have to ACK an ACK-only packet at least
2993 * for stacks that send 0-length keep-alives with
2994 * SEG.SEQ = SND.NXT-1 as recommended by RFC1122,
2995 * section 4.2.3.6. As long as we don't ever generate
2996 * an unacceptable packet in response to an incoming
2997 * packet that is unacceptable, it should not cause
2998 * "ACK wars".
2999 */
3000 flags |= TH_ACK_NEEDED;
3001
3002 /*
3003 * Continue processing this segment in order to use the
3004 * ACK information it contains, but skip all other
3005 * sequence-number processing. Processing the ACK
3006 * information is necessary in order to
3007 * re-synchronize connections that may have lost
3008 * synchronization.
3009 *
3010 * We clear seg_len and flag fields related to
3011 * sequence number processing as they are not
3012 * to be trusted for an unacceptable segment.
3013 */
3014 seg_len = 0;
3015 flags &= ~(TH_SYN | TH_FIN | TH_URG);
3016 goto process_ack;
3017 }
3018
3019 /* Fix seg_seq, and chew the gap off the front. */
3020 seg_seq = tcp->tcp_rnxt;
3021 urp += gap;
3022 do {
3023 mblk_t *mp2;
3024 ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <=
3025 (uintptr_t)UINT_MAX);
3026 gap += (uint_t)(mp->b_wptr - mp->b_rptr);
3027 if (gap > 0) {
3028 mp->b_rptr = mp->b_wptr - gap;
3029 break;
3030 }
3031 mp2 = mp;
3032 mp = mp->b_cont;
3033 freeb(mp2);
3034 } while (gap < 0);
3035 /*
3036 * If the urgent data has already been acknowledged, we
3037 * should ignore TH_URG below
3038 */
3039 if (urp < 0)
3040 flags &= ~TH_URG;
3041 }
3042 /*
3043 * rgap is the amount of stuff received out of window. A negative
3044 * value is the amount out of window.
3045 */
3046 if (rgap < 0) {
3047 mblk_t *mp2;
3048
3049 if (tcp->tcp_rwnd == 0) {
3050 TCPS_BUMP_MIB(tcps, tcpInWinProbe);
3051 tcp->tcp_cs.tcp_in_zwnd_probes++;
3052 } else {
3053 TCPS_BUMP_MIB(tcps, tcpInDataPastWinSegs);
3054 TCPS_UPDATE_MIB(tcps, tcpInDataPastWinBytes, -rgap);
3055 }
3056
3057 /*
3058 * seg_len does not include the FIN, so if more than
3059 * just the FIN is out of window, we act like we don't
3060 * see it. (If just the FIN is out of window, rgap
3061 * will be zero and we will go ahead and acknowledge
3062 * the FIN.)
3063 */
3064 flags &= ~TH_FIN;
3065
3066 /* Fix seg_len and make sure there is something left. */
3067 seg_len += rgap;
3068 if (seg_len <= 0) {
3069 /*
3070 * Resets are only valid if they lie within our offered
3071 * window. If the RST bit is set, we just ignore this
3072 * segment.
3073 */
3074 if (flags & TH_RST) {
3075 freemsg(mp);
3076 return;
3077 }
3078
3079 /* Per RFC 793, we need to send back an ACK. */
3080 flags |= TH_ACK_NEEDED;
3081
3082 /*
3083 * Send SIGURG as soon as possible i.e. even
3084 * if the TH_URG was delivered in a window probe
3085 * packet (which will be unacceptable).
3086 *
3087 * We generate a signal if none has been generated
3088 * for this connection or if this is a new urgent
3089 * byte. Also send a zero-length "unmarked" message
3090 * to inform SIOCATMARK that this is not the mark.
3091 *
3092 * tcp_urp_last_valid is cleared when the T_exdata_ind
3093 * is sent up. This plus the check for old data
3094 * (gap >= 0) handles the wraparound of the sequence
3095 * number space without having to always track the
3096 * correct MAX(tcp_urp_last, tcp_rnxt). (BSD tracks
3097 * this max in its rcv_up variable).
3098 *
3099 * This prevents duplicate SIGURGS due to a "late"
3100 * zero-window probe when the T_EXDATA_IND has already
3101 * been sent up.
3102 */
3103 if ((flags & TH_URG) &&
3104 (!tcp->tcp_urp_last_valid || SEQ_GT(urp + seg_seq,
3105 tcp->tcp_urp_last))) {
3106 if (IPCL_IS_NONSTR(connp)) {
3107 if (!TCP_IS_DETACHED(tcp)) {
3108 (*sockupcalls->su_signal_oob)
3109 (connp->conn_upper_handle,
3110 urp);
3111 }
3112 } else {
3113 mp1 = allocb(0, BPRI_MED);
3114 if (mp1 == NULL) {
3115 freemsg(mp);
3116 return;
3117 }
3118 if (!TCP_IS_DETACHED(tcp) &&
3119 !putnextctl1(connp->conn_rq,
3120 M_PCSIG, SIGURG)) {
3121 /* Try again on the rexmit. */
3122 freemsg(mp1);
3123 freemsg(mp);
3124 return;
3125 }
3126 /*
3127 * If the next byte would be the mark
3128 * then mark with MARKNEXT else mark
3129 * with NOTMARKNEXT.
3130 */
3131 if (gap == 0 && urp == 0)
3132 mp1->b_flag |= MSGMARKNEXT;
3133 else
3134 mp1->b_flag |= MSGNOTMARKNEXT;
3135 freemsg(tcp->tcp_urp_mark_mp);
3136 tcp->tcp_urp_mark_mp = mp1;
3137 flags |= TH_SEND_URP_MARK;
3138 }
3139 tcp->tcp_urp_last_valid = B_TRUE;
3140 tcp->tcp_urp_last = urp + seg_seq;
3141 }
3142 /*
3143 * If this is a zero window probe, continue to
3144 * process the ACK part. But we need to set seg_len
3145 * to 0 to avoid data processing. Otherwise just
3146 * drop the segment and send back an ACK.
3147 */
3148 if (tcp->tcp_rwnd == 0 && seg_seq == tcp->tcp_rnxt) {
3149 flags &= ~(TH_SYN | TH_URG);
3150 seg_len = 0;
3151 goto process_ack;
3152 } else {
3153 freemsg(mp);
3154 goto ack_check;
3155 }
3156 }
3157 /* Pitch out of window stuff off the end. */
3158 rgap = seg_len;
3159 mp2 = mp;
3160 do {
3161 ASSERT((uintptr_t)(mp2->b_wptr - mp2->b_rptr) <=
3162 (uintptr_t)INT_MAX);
3163 rgap -= (int)(mp2->b_wptr - mp2->b_rptr);
3164 if (rgap < 0) {
3165 mp2->b_wptr += rgap;
3166 if ((mp1 = mp2->b_cont) != NULL) {
3167 mp2->b_cont = NULL;
3168 freemsg(mp1);
3169 }
3170 break;
3171 }
3172 } while ((mp2 = mp2->b_cont) != NULL);
3173 }
3174 ok:;
3175 /*
3176 * TCP should check ECN info for segments inside the window only.
3177 * Therefore the check should be done here.
3178 */
3179 if (tcp->tcp_ecn_ok) {
3180 if (flags & TH_CWR) {
3181 tcp->tcp_ecn_echo_on = B_FALSE;
3182 }
3183 /*
3184 * Note that both ECN_CE and CWR can be set in the
3185 * same segment. In this case, we once again turn
3186 * on ECN_ECHO.
3187 */
3188 if (connp->conn_ipversion == IPV4_VERSION) {
3189 uchar_t tos = ((ipha_t *)rptr)->ipha_type_of_service;
3190
3191 if ((tos & IPH_ECN_CE) == IPH_ECN_CE) {
3192 tcp->tcp_ecn_echo_on = B_TRUE;
3193 }
3194 } else {
3195 uint32_t vcf = ((ip6_t *)rptr)->ip6_vcf;
3196
3197 if ((vcf & htonl(IPH_ECN_CE << 20)) ==
3198 htonl(IPH_ECN_CE << 20)) {
3199 tcp->tcp_ecn_echo_on = B_TRUE;
3200 }
3201 }
3202 }
3203
3204 /*
3205 * Check whether we can update tcp_ts_recent. This test is from RFC
3206 * 7323, section 5.3.
3207 */
3208 if (tcp->tcp_snd_ts_ok && !(flags & TH_RST) &&
3209 TSTMP_GEQ(tcpopt.tcp_opt_ts_val, tcp->tcp_ts_recent) &&
3210 SEQ_LEQ(seg_seq, tcp->tcp_rack)) {
3211 tcp->tcp_ts_recent = tcpopt.tcp_opt_ts_val;
3212 tcp->tcp_last_rcv_lbolt = LBOLT_FASTPATH64;
3213 }
3214
3215 if (seg_seq != tcp->tcp_rnxt || tcp->tcp_reass_head) {
3216 /*
3217 * FIN in an out of order segment. We record this in
3218 * tcp_valid_bits and the seq num of FIN in tcp_ofo_fin_seq.
3219 * Clear the FIN so that any check on FIN flag will fail.
3220 * Remember that FIN also counts in the sequence number
3221 * space. So we need to ack out of order FIN only segments.
3222 */
3223 if (flags & TH_FIN) {
3224 tcp->tcp_valid_bits |= TCP_OFO_FIN_VALID;
3225 tcp->tcp_ofo_fin_seq = seg_seq + seg_len;
3226 flags &= ~TH_FIN;
3227 flags |= TH_ACK_NEEDED;
3228 }
3229 if (seg_len > 0) {
3230 /* Fill in the SACK blk list. */
3231 if (tcp->tcp_snd_sack_ok) {
3232 tcp_sack_insert(tcp->tcp_sack_list,
3233 seg_seq, seg_seq + seg_len,
3234 &(tcp->tcp_num_sack_blk));
3235 }
3236
3237 /*
3238 * Attempt reassembly and see if we have something
3239 * ready to go.
3240 */
3241 mp = tcp_reass(tcp, mp, seg_seq);
3242 /* Always ack out of order packets */
3243 flags |= TH_ACK_NEEDED | TH_PUSH;
3244 if (mp) {
3245 ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <=
3246 (uintptr_t)INT_MAX);
3247 seg_len = mp->b_cont ? msgdsize(mp) :
3248 (int)(mp->b_wptr - mp->b_rptr);
3249 seg_seq = tcp->tcp_rnxt;
3250 /*
3251 * A gap is filled and the seq num and len
3252 * of the gap match that of a previously
3253 * received FIN, put the FIN flag back in.
3254 */
3255 if ((tcp->tcp_valid_bits & TCP_OFO_FIN_VALID) &&
3256 seg_seq + seg_len == tcp->tcp_ofo_fin_seq) {
3257 flags |= TH_FIN;
3258 tcp->tcp_valid_bits &=
3259 ~TCP_OFO_FIN_VALID;
3260 }
3261 if (tcp->tcp_reass_tid != 0) {
3262 (void) TCP_TIMER_CANCEL(tcp,
3263 tcp->tcp_reass_tid);
3264 /*
3265 * Restart the timer if there is still
3266 * data in the reassembly queue.
3267 */
3268 if (tcp->tcp_reass_head != NULL) {
3269 tcp->tcp_reass_tid = TCP_TIMER(
3270 tcp, tcp_reass_timer,
3271 tcps->tcps_reass_timeout);
3272 } else {
3273 tcp->tcp_reass_tid = 0;
3274 }
3275 }
3276 } else {
3277 /*
3278 * Keep going even with NULL mp.
3279 * There may be a useful ACK or something else
3280 * we don't want to miss.
3281 *
3282 * But TCP should not perform fast retransmit
3283 * because of the ack number. TCP uses
3284 * seg_len == 0 to determine if it is a pure
3285 * ACK. And this is not a pure ACK.
3286 */
3287 seg_len = 0;
3288 ofo_seg = B_TRUE;
3289
3290 if (tcps->tcps_reass_timeout != 0 &&
3291 tcp->tcp_reass_tid == 0) {
3292 tcp->tcp_reass_tid = TCP_TIMER(tcp,
3293 tcp_reass_timer,
3294 tcps->tcps_reass_timeout);
3295 }
3296 }
3297 }
3298 } else if (seg_len > 0) {
3299 TCPS_BUMP_MIB(tcps, tcpInDataInorderSegs);
3300 TCPS_UPDATE_MIB(tcps, tcpInDataInorderBytes, seg_len);
3301 tcp->tcp_cs.tcp_in_data_inorder_segs++;
3302 tcp->tcp_cs.tcp_in_data_inorder_bytes += seg_len;
3303
3304 /*
3305 * If an out of order FIN was received before, and the seq
3306 * num and len of the new segment match that of the FIN,
3307 * put the FIN flag back in.
3308 */
3309 if ((tcp->tcp_valid_bits & TCP_OFO_FIN_VALID) &&
3310 seg_seq + seg_len == tcp->tcp_ofo_fin_seq) {
3311 flags |= TH_FIN;
3312 tcp->tcp_valid_bits &= ~TCP_OFO_FIN_VALID;
3313 }
3314 }
3315 if ((flags & (TH_RST | TH_SYN | TH_URG | TH_ACK)) != TH_ACK) {
3316 if (flags & TH_RST) {
3317 freemsg(mp);
3318 switch (tcp->tcp_state) {
3319 case TCPS_SYN_RCVD:
3320 (void) tcp_clean_death(tcp, ECONNREFUSED);
3321 break;
3322 case TCPS_ESTABLISHED:
3323 case TCPS_FIN_WAIT_1:
3324 case TCPS_FIN_WAIT_2:
3325 case TCPS_CLOSE_WAIT:
3326 (void) tcp_clean_death(tcp, ECONNRESET);
3327 break;
3328 case TCPS_CLOSING:
3329 case TCPS_LAST_ACK:
3330 (void) tcp_clean_death(tcp, 0);
3331 break;
3332 default:
3333 ASSERT(tcp->tcp_state != TCPS_TIME_WAIT);
3334 (void) tcp_clean_death(tcp, ENXIO);
3335 break;
3336 }
3337 return;
3338 }
3339 if (flags & TH_SYN) {
3340 /*
3341 * See RFC 793, Page 71
3342 *
3343 * The seq number must be in the window as it should
3344 * be "fixed" above. If it is outside window, it should
3345 * be already rejected. Note that we allow seg_seq to be
3346 * rnxt + rwnd because we want to accept 0 window probe.
3347 */
3348 ASSERT(SEQ_GEQ(seg_seq, tcp->tcp_rnxt) &&
3349 SEQ_LEQ(seg_seq, tcp->tcp_rnxt + tcp->tcp_rwnd));
3350 freemsg(mp);
3351 /*
3352 * If the ACK flag is not set, just use our snxt as the
3353 * seq number of the RST segment.
3354 */
3355 if (!(flags & TH_ACK)) {
3356 seg_ack = tcp->tcp_snxt;
3357 }
3358 tcp_xmit_ctl("TH_SYN", tcp, seg_ack, seg_seq + 1,
3359 TH_RST|TH_ACK);
3360 ASSERT(tcp->tcp_state != TCPS_TIME_WAIT);
3361 (void) tcp_clean_death(tcp, ECONNRESET);
3362 return;
3363 }
3364 /*
3365 * urp could be -1 when the urp field in the packet is 0
3366 * and TCP_OLD_URP_INTERPRETATION is set. This implies that the urgent
3367 * byte was at seg_seq - 1, in which case we ignore the urgent flag.
3368 */
3369 if ((flags & TH_URG) && urp >= 0) {
3370 if (!tcp->tcp_urp_last_valid ||
3371 SEQ_GT(urp + seg_seq, tcp->tcp_urp_last)) {
3372 /*
3373 * Non-STREAMS sockets handle the urgent data a litte
3374 * differently from STREAMS based sockets. There is no
3375 * need to mark any mblks with the MSG{NOT,}MARKNEXT
3376 * flags to keep SIOCATMARK happy. Instead a
3377 * su_signal_oob upcall is made to update the mark.
3378 * Neither is a T_EXDATA_IND mblk needed to be
3379 * prepended to the urgent data. The urgent data is
3380 * delivered using the su_recv upcall, where we set
3381 * the MSG_OOB flag to indicate that it is urg data.
3382 *
3383 * Neither TH_SEND_URP_MARK nor TH_MARKNEXT_NEEDED
3384 * are used by non-STREAMS sockets.
3385 */
3386 if (IPCL_IS_NONSTR(connp)) {
3387 if (!TCP_IS_DETACHED(tcp)) {
3388 (*sockupcalls->su_signal_oob)
3389 (connp->conn_upper_handle, urp);
3390 }
3391 } else {
3392 /*
3393 * If we haven't generated the signal yet for
3394 * this urgent pointer value, do it now. Also,
3395 * send up a zero-length M_DATA indicating
3396 * whether or not this is the mark. The latter
3397 * is not needed when a T_EXDATA_IND is sent up.
3398 * However, if there are allocation failures
3399 * this code relies on the sender retransmitting
3400 * and the socket code for determining the mark
3401 * should not block waiting for the peer to
3402 * transmit. Thus, for simplicity we always
3403 * send up the mark indication.
3404 */
3405 mp1 = allocb(0, BPRI_MED);
3406 if (mp1 == NULL) {
3407 freemsg(mp);
3408 return;
3409 }
3410 if (!TCP_IS_DETACHED(tcp) &&
3411 !putnextctl1(connp->conn_rq, M_PCSIG,
3412 SIGURG)) {
3413 /* Try again on the rexmit. */
3414 freemsg(mp1);
3415 freemsg(mp);
3416 return;
3417 }
3418 /*
3419 * Mark with NOTMARKNEXT for now.
3420 * The code below will change this to MARKNEXT
3421 * if we are at the mark.
3422 *
3423 * If there are allocation failures (e.g. in
3424 * dupmsg below) the next time tcp_input_data
3425 * sees the urgent segment it will send up the
3426 * MSGMARKNEXT message.
3427 */
3428 mp1->b_flag |= MSGNOTMARKNEXT;
3429 freemsg(tcp->tcp_urp_mark_mp);
3430 tcp->tcp_urp_mark_mp = mp1;
3431 flags |= TH_SEND_URP_MARK;
3432 #ifdef DEBUG
3433 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
3434 "tcp_rput: sent M_PCSIG 2 seq %x urp %x "
3435 "last %x, %s",
3436 seg_seq, urp, tcp->tcp_urp_last,
3437 tcp_display(tcp, NULL, DISP_PORT_ONLY));
3438 #endif /* DEBUG */
3439 }
3440 tcp->tcp_urp_last_valid = B_TRUE;
3441 tcp->tcp_urp_last = urp + seg_seq;
3442 } else if (tcp->tcp_urp_mark_mp != NULL) {
3443 /*
3444 * An allocation failure prevented the previous
3445 * tcp_input_data from sending up the allocated
3446 * MSG*MARKNEXT message - send it up this time
3447 * around.
3448 */
3449 flags |= TH_SEND_URP_MARK;
3450 }
3451
3452 /*
3453 * If the urgent byte is in this segment, make sure that it is
3454 * all by itself. This makes it much easier to deal with the
3455 * possibility of an allocation failure on the T_exdata_ind.
3456 * Note that seg_len is the number of bytes in the segment, and
3457 * urp is the offset into the segment of the urgent byte.
3458 * urp < seg_len means that the urgent byte is in this segment.
3459 */
3460 if (urp < seg_len) {
3461 if (seg_len != 1) {
3462 uint32_t tmp_rnxt;
3463 /*
3464 * Break it up and feed it back in.
3465 * Re-attach the IP header.
3466 */
3467 mp->b_rptr = iphdr;
3468 if (urp > 0) {
3469 /*
3470 * There is stuff before the urgent
3471 * byte.
3472 */
3473 mp1 = dupmsg(mp);
3474 if (!mp1) {
3475 /*
3476 * Trim from urgent byte on.
3477 * The rest will come back.
3478 */
3479 (void) adjmsg(mp,
3480 urp - seg_len);
3481 tcp_input_data(connp,
3482 mp, NULL, ira);
3483 return;
3484 }
3485 (void) adjmsg(mp1, urp - seg_len);
3486 /* Feed this piece back in. */
3487 tmp_rnxt = tcp->tcp_rnxt;
3488 tcp_input_data(connp, mp1, NULL, ira);
3489 /*
3490 * If the data passed back in was not
3491 * processed (ie: bad ACK) sending
3492 * the remainder back in will cause a
3493 * loop. In this case, drop the
3494 * packet and let the sender try
3495 * sending a good packet.
3496 */
3497 if (tmp_rnxt == tcp->tcp_rnxt) {
3498 freemsg(mp);
3499 return;
3500 }
3501 }
3502 if (urp != seg_len - 1) {
3503 uint32_t tmp_rnxt;
3504 /*
3505 * There is stuff after the urgent
3506 * byte.
3507 */
3508 mp1 = dupmsg(mp);
3509 if (!mp1) {
3510 /*
3511 * Trim everything beyond the
3512 * urgent byte. The rest will
3513 * come back.
3514 */
3515 (void) adjmsg(mp,
3516 urp + 1 - seg_len);
3517 tcp_input_data(connp,
3518 mp, NULL, ira);
3519 return;
3520 }
3521 (void) adjmsg(mp1, urp + 1 - seg_len);
3522 tmp_rnxt = tcp->tcp_rnxt;
3523 tcp_input_data(connp, mp1, NULL, ira);
3524 /*
3525 * If the data passed back in was not
3526 * processed (ie: bad ACK) sending
3527 * the remainder back in will cause a
3528 * loop. In this case, drop the
3529 * packet and let the sender try
3530 * sending a good packet.
3531 */
3532 if (tmp_rnxt == tcp->tcp_rnxt) {
3533 freemsg(mp);
3534 return;
3535 }
3536 }
3537 tcp_input_data(connp, mp, NULL, ira);
3538 return;
3539 }
3540 /*
3541 * This segment contains only the urgent byte. We
3542 * have to allocate the T_exdata_ind, if we can.
3543 */
3544 if (IPCL_IS_NONSTR(connp)) {
3545 int error;
3546
3547 (*sockupcalls->su_recv)
3548 (connp->conn_upper_handle, mp, seg_len,
3549 MSG_OOB, &error, NULL);
3550 /*
3551 * We should never be in middle of a
3552 * fallback, the squeue guarantees that.
3553 */
3554 ASSERT(error != EOPNOTSUPP);
3555 mp = NULL;
3556 goto update_ack;
3557 } else if (!tcp->tcp_urp_mp) {
3558 struct T_exdata_ind *tei;
3559 mp1 = allocb(sizeof (struct T_exdata_ind),
3560 BPRI_MED);
3561 if (!mp1) {
3562 /*
3563 * Sigh... It'll be back.
3564 * Generate any MSG*MARK message now.
3565 */
3566 freemsg(mp);
3567 seg_len = 0;
3568 if (flags & TH_SEND_URP_MARK) {
3569
3570
3571 ASSERT(tcp->tcp_urp_mark_mp);
3572 tcp->tcp_urp_mark_mp->b_flag &=
3573 ~MSGNOTMARKNEXT;
3574 tcp->tcp_urp_mark_mp->b_flag |=
3575 MSGMARKNEXT;
3576 }
3577 goto ack_check;
3578 }
3579 mp1->b_datap->db_type = M_PROTO;
3580 tei = (struct T_exdata_ind *)mp1->b_rptr;
3581 tei->PRIM_type = T_EXDATA_IND;
3582 tei->MORE_flag = 0;
3583 mp1->b_wptr = (uchar_t *)&tei[1];
3584 tcp->tcp_urp_mp = mp1;
3585 #ifdef DEBUG
3586 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
3587 "tcp_rput: allocated exdata_ind %s",
3588 tcp_display(tcp, NULL,
3589 DISP_PORT_ONLY));
3590 #endif /* DEBUG */
3591 /*
3592 * There is no need to send a separate MSG*MARK
3593 * message since the T_EXDATA_IND will be sent
3594 * now.
3595 */
3596 flags &= ~TH_SEND_URP_MARK;
3597 freemsg(tcp->tcp_urp_mark_mp);
3598 tcp->tcp_urp_mark_mp = NULL;
3599 }
3600 /*
3601 * Now we are all set. On the next putnext upstream,
3602 * tcp_urp_mp will be non-NULL and will get prepended
3603 * to what has to be this piece containing the urgent
3604 * byte. If for any reason we abort this segment below,
3605 * if it comes back, we will have this ready, or it
3606 * will get blown off in close.
3607 */
3608 } else if (urp == seg_len) {
3609 /*
3610 * The urgent byte is the next byte after this sequence
3611 * number. If this endpoint is non-STREAMS, then there
3612 * is nothing to do here since the socket has already
3613 * been notified about the urg pointer by the
3614 * su_signal_oob call above.
3615 *
3616 * In case of STREAMS, some more work might be needed.
3617 * If there is data it is marked with MSGMARKNEXT and
3618 * and any tcp_urp_mark_mp is discarded since it is not
3619 * needed. Otherwise, if the code above just allocated
3620 * a zero-length tcp_urp_mark_mp message, that message
3621 * is tagged with MSGMARKNEXT. Sending up these
3622 * MSGMARKNEXT messages makes SIOCATMARK work correctly
3623 * even though the T_EXDATA_IND will not be sent up
3624 * until the urgent byte arrives.
3625 */
3626 if (!IPCL_IS_NONSTR(tcp->tcp_connp)) {
3627 if (seg_len != 0) {
3628 flags |= TH_MARKNEXT_NEEDED;
3629 freemsg(tcp->tcp_urp_mark_mp);
3630 tcp->tcp_urp_mark_mp = NULL;
3631 flags &= ~TH_SEND_URP_MARK;
3632 } else if (tcp->tcp_urp_mark_mp != NULL) {
3633 flags |= TH_SEND_URP_MARK;
3634 tcp->tcp_urp_mark_mp->b_flag &=
3635 ~MSGNOTMARKNEXT;
3636 tcp->tcp_urp_mark_mp->b_flag |=
3637 MSGMARKNEXT;
3638 }
3639 }
3640 #ifdef DEBUG
3641 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
3642 "tcp_rput: AT MARK, len %d, flags 0x%x, %s",
3643 seg_len, flags,
3644 tcp_display(tcp, NULL, DISP_PORT_ONLY));
3645 #endif /* DEBUG */
3646 }
3647 #ifdef DEBUG
3648 else {
3649 /* Data left until we hit mark */
3650 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
3651 "tcp_rput: URP %d bytes left, %s",
3652 urp - seg_len, tcp_display(tcp, NULL,
3653 DISP_PORT_ONLY));
3654 }
3655 #endif /* DEBUG */
3656 }
3657
3658 process_ack:
3659 if (!(flags & TH_ACK)) {
3660 freemsg(mp);
3661 goto xmit_check;
3662 }
3663 }
3664 bytes_acked = (int)(seg_ack - tcp->tcp_suna);
3665
3666 if (bytes_acked > 0)
3667 tcp->tcp_ip_forward_progress = B_TRUE;
3668 if (tcp->tcp_state == TCPS_SYN_RCVD) {
3669 /*
3670 * tcp_sendmsg() checks tcp_state without entering
3671 * the squeue so tcp_state should be updated before
3672 * sending up a connection confirmation or a new
3673 * connection indication.
3674 */
3675 tcp->tcp_state = TCPS_ESTABLISHED;
3676
3677 /*
3678 * We are seeing the final ack in the three way
3679 * hand shake of a active open'ed connection
3680 * so we must send up a T_CONN_CON
3681 */
3682 if (tcp->tcp_active_open) {
3683 if (!tcp_conn_con(tcp, iphdr, mp, NULL, ira)) {
3684 freemsg(mp);
3685 tcp->tcp_state = TCPS_SYN_RCVD;
3686 return;
3687 }
3688 /*
3689 * Don't fuse the loopback endpoints for
3690 * simultaneous active opens.
3691 */
3692 if (tcp->tcp_loopback) {
3693 TCP_STAT(tcps, tcp_fusion_unfusable);
3694 tcp->tcp_unfusable = B_TRUE;
3695 }
3696 /*
3697 * For simultaneous active open, trace receipt of final
3698 * ACK as tcp:::connect-established.
3699 */
3700 DTRACE_TCP5(connect__established, mblk_t *, NULL,
3701 ip_xmit_attr_t *, connp->conn_ixa, void_ip_t *,
3702 iphdr, tcp_t *, tcp, tcph_t *, tcpha);
3703 } else if (IPCL_IS_NONSTR(connp)) {
3704 /*
3705 * 3-way handshake has completed, so notify socket
3706 * of the new connection.
3707 *
3708 * We are here means eager is fine but it can
3709 * get a TH_RST at any point between now and till
3710 * accept completes and disappear. We need to
3711 * ensure that reference to eager is valid after
3712 * we get out of eager's perimeter. So we do
3713 * an extra refhold.
3714 */
3715 CONN_INC_REF(connp);
3716
3717 if (!tcp_newconn_notify(tcp, ira)) {
3718 /*
3719 * The state-change probe for SYN_RCVD ->
3720 * ESTABLISHED has not fired yet. We reset
3721 * the state to SYN_RCVD so that future
3722 * state-change probes report correct state
3723 * transistions.
3724 */
3725 tcp->tcp_state = TCPS_SYN_RCVD;
3726 freemsg(mp);
3727 /* notification did not go up, so drop ref */
3728 CONN_DEC_REF(connp);
3729 /* ... and close the eager */
3730 ASSERT(TCP_IS_DETACHED(tcp));
3731 (void) tcp_close_detached(tcp);
3732 return;
3733 }
3734 /*
3735 * tcp_newconn_notify() changes conn_upcalls and
3736 * connp->conn_upper_handle. Fix things now, in case
3737 * there's data attached to this ack.
3738 */
3739 if (connp->conn_upcalls != NULL)
3740 sockupcalls = connp->conn_upcalls;
3741 /*
3742 * For passive open, trace receipt of final ACK as
3743 * tcp:::accept-established.
3744 */
3745 DTRACE_TCP5(accept__established, mlbk_t *, NULL,
3746 ip_xmit_attr_t *, connp->conn_ixa, void_ip_t *,
3747 iphdr, tcp_t *, tcp, tcph_t *, tcpha);
3748 } else {
3749 /*
3750 * 3-way handshake complete - this is a STREAMS based
3751 * socket, so pass up the T_CONN_IND.
3752 */
3753 tcp_t *listener = tcp->tcp_listener;
3754 mblk_t *mp = tcp->tcp_conn.tcp_eager_conn_ind;
3755
3756 tcp->tcp_tconnind_started = B_TRUE;
3757 tcp->tcp_conn.tcp_eager_conn_ind = NULL;
3758 ASSERT(mp != NULL);
3759 /*
3760 * We are here means eager is fine but it can
3761 * get a TH_RST at any point between now and till
3762 * accept completes and disappear. We need to
3763 * ensure that reference to eager is valid after
3764 * we get out of eager's perimeter. So we do
3765 * an extra refhold.
3766 */
3767 CONN_INC_REF(connp);
3768
3769 /*
3770 * The listener also exists because of the refhold
3771 * done in tcp_input_listener. Its possible that it
3772 * might have closed. We will check that once we
3773 * get inside listeners context.
3774 */
3775 CONN_INC_REF(listener->tcp_connp);
3776 if (listener->tcp_connp->conn_sqp ==
3777 connp->conn_sqp) {
3778 /*
3779 * We optimize by not calling an SQUEUE_ENTER
3780 * on the listener since we know that the
3781 * listener and eager squeues are the same.
3782 * We are able to make this check safely only
3783 * because neither the eager nor the listener
3784 * can change its squeue. Only an active connect
3785 * can change its squeue
3786 */
3787 tcp_send_conn_ind(listener->tcp_connp, mp,
3788 listener->tcp_connp->conn_sqp);
3789 CONN_DEC_REF(listener->tcp_connp);
3790 } else if (!tcp->tcp_loopback) {
3791 SQUEUE_ENTER_ONE(listener->tcp_connp->conn_sqp,
3792 mp, tcp_send_conn_ind,
3793 listener->tcp_connp, NULL, SQ_FILL,
3794 SQTAG_TCP_CONN_IND);
3795 } else {
3796 SQUEUE_ENTER_ONE(listener->tcp_connp->conn_sqp,
3797 mp, tcp_send_conn_ind,
3798 listener->tcp_connp, NULL, SQ_NODRAIN,
3799 SQTAG_TCP_CONN_IND);
3800 }
3801 /*
3802 * For passive open, trace receipt of final ACK as
3803 * tcp:::accept-established.
3804 */
3805 DTRACE_TCP5(accept__established, mlbk_t *, NULL,
3806 ip_xmit_attr_t *, connp->conn_ixa, void_ip_t *,
3807 iphdr, tcp_t *, tcp, tcph_t *, tcpha);
3808 }
3809 TCPS_CONN_INC(tcps);
3810
3811 tcp->tcp_suna = tcp->tcp_iss + 1; /* One for the SYN */
3812 bytes_acked--;
3813 /* SYN was acked - making progress */
3814 tcp->tcp_ip_forward_progress = B_TRUE;
3815
3816 /*
3817 * If SYN was retransmitted, need to reset all
3818 * retransmission info as this segment will be
3819 * treated as a dup ACK.
3820 */
3821 if (tcp->tcp_rexmit) {
3822 tcp->tcp_rexmit = B_FALSE;
3823 tcp->tcp_rexmit_nxt = tcp->tcp_snxt;
3824 tcp->tcp_rexmit_max = tcp->tcp_snxt;
3825 tcp->tcp_ms_we_have_waited = 0;
3826 tcp->tcp_cwnd = mss;
3827 }
3828
3829 /*
3830 * We set the send window to zero here.
3831 * This is needed if there is data to be
3832 * processed already on the queue.
3833 * Later (at swnd_update label), the
3834 * "new_swnd > tcp_swnd" condition is satisfied
3835 * the XMIT_NEEDED flag is set in the current
3836 * (SYN_RCVD) state. This ensures tcp_wput_data() is
3837 * called if there is already data on queue in
3838 * this state.
3839 */
3840 tcp->tcp_swnd = 0;
3841
3842 if (new_swnd > tcp->tcp_max_swnd)
3843 tcp->tcp_max_swnd = new_swnd;
3844 tcp->tcp_swl1 = seg_seq;
3845 tcp->tcp_swl2 = seg_ack;
3846 tcp->tcp_valid_bits &= ~TCP_ISS_VALID;
3847
3848 /* Trace change from SYN_RCVD -> ESTABLISHED here */
3849 DTRACE_TCP6(state__change, void, NULL, ip_xmit_attr_t *,
3850 connp->conn_ixa, void, NULL, tcp_t *, tcp, void, NULL,
3851 int32_t, TCPS_SYN_RCVD);
3852
3853 /* Fuse when both sides are in ESTABLISHED state */
3854 if (tcp->tcp_loopback && do_tcp_fusion)
3855 tcp_fuse(tcp, iphdr, tcpha);
3856
3857 }
3858 /* This code follows 4.4BSD-Lite2 mostly. */
3859 if (bytes_acked < 0)
3860 goto est;
3861
3862 /*
3863 * If TCP is ECN capable and the congestion experience bit is
3864 * set, reduce tcp_cwnd and tcp_ssthresh. But this should only be
3865 * done once per window (or more loosely, per RTT).
3866 */
3867 if (tcp->tcp_cwr && SEQ_GT(seg_ack, tcp->tcp_cwr_snd_max))
3868 tcp->tcp_cwr = B_FALSE;
3869 if (tcp->tcp_ecn_ok && (flags & TH_ECE)) {
3870 if (!tcp->tcp_cwr) {
3871 npkt = ((tcp->tcp_snxt - tcp->tcp_suna) >> 1) / mss;
3872 tcp->tcp_cwnd_ssthresh = MAX(npkt, 2) * mss;
3873 tcp->tcp_cwnd = npkt * mss;
3874 /*
3875 * If the cwnd is 0, use the timer to clock out
3876 * new segments. This is required by the ECN spec.
3877 */
3878 if (npkt == 0) {
3879 TCP_TIMER_RESTART(tcp, tcp->tcp_rto);
3880 /*
3881 * This makes sure that when the ACK comes
3882 * back, we will increase tcp_cwnd by 1 MSS.
3883 */
3884 tcp->tcp_cwnd_cnt = 0;
3885 }
3886 tcp->tcp_cwr = B_TRUE;
3887 /*
3888 * This marks the end of the current window of in
3889 * flight data. That is why we don't use
3890 * tcp_suna + tcp_swnd. Only data in flight can
3891 * provide ECN info.
3892 */
3893 tcp->tcp_cwr_snd_max = tcp->tcp_snxt;
3894 tcp->tcp_ecn_cwr_sent = B_FALSE;
3895 }
3896 }
3897
3898 mp1 = tcp->tcp_xmit_head;
3899 if (bytes_acked == 0) {
3900 if (!ofo_seg && seg_len == 0 && new_swnd == tcp->tcp_swnd) {
3901 int dupack_cnt;
3902
3903 TCPS_BUMP_MIB(tcps, tcpInDupAck);
3904 /*
3905 * Fast retransmit. When we have seen exactly three
3906 * identical ACKs while we have unacked data
3907 * outstanding we take it as a hint that our peer
3908 * dropped something.
3909 *
3910 * If TCP is retransmitting, don't do fast retransmit.
3911 */
3912 if (mp1 && tcp->tcp_suna != tcp->tcp_snxt &&
3913 ! tcp->tcp_rexmit) {
3914 /* Do Limited Transmit */
3915 if ((dupack_cnt = ++tcp->tcp_dupack_cnt) <
3916 tcps->tcps_dupack_fast_retransmit) {
3917 /*
3918 * RFC 3042
3919 *
3920 * What we need to do is temporarily
3921 * increase tcp_cwnd so that new
3922 * data can be sent if it is allowed
3923 * by the receive window (tcp_rwnd).
3924 * tcp_wput_data() will take care of
3925 * the rest.
3926 *
3927 * If the connection is SACK capable,
3928 * only do limited xmit when there
3929 * is SACK info.
3930 *
3931 * Note how tcp_cwnd is incremented.
3932 * The first dup ACK will increase
3933 * it by 1 MSS. The second dup ACK
3934 * will increase it by 2 MSS. This
3935 * means that only 1 new segment will
3936 * be sent for each dup ACK.
3937 */
3938 if (tcp->tcp_unsent > 0 &&
3939 (!tcp->tcp_snd_sack_ok ||
3940 (tcp->tcp_snd_sack_ok &&
3941 tcp->tcp_notsack_list != NULL))) {
3942 tcp->tcp_cwnd += mss <<
3943 (tcp->tcp_dupack_cnt - 1);
3944 flags |= TH_LIMIT_XMIT;
3945 }
3946 } else if (dupack_cnt ==
3947 tcps->tcps_dupack_fast_retransmit) {
3948
3949 /*
3950 * If we have reduced tcp_ssthresh
3951 * because of ECN, do not reduce it again
3952 * unless it is already one window of data
3953 * away. After one window of data, tcp_cwr
3954 * should then be cleared. Note that
3955 * for non ECN capable connection, tcp_cwr
3956 * should always be false.
3957 *
3958 * Adjust cwnd since the duplicate
3959 * ack indicates that a packet was
3960 * dropped (due to congestion.)
3961 */
3962 if (!tcp->tcp_cwr) {
3963 npkt = ((tcp->tcp_snxt -
3964 tcp->tcp_suna) >> 1) / mss;
3965 tcp->tcp_cwnd_ssthresh = MAX(npkt, 2) *
3966 mss;
3967 tcp->tcp_cwnd = (npkt +
3968 tcp->tcp_dupack_cnt) * mss;
3969 }
3970 if (tcp->tcp_ecn_ok) {
3971 tcp->tcp_cwr = B_TRUE;
3972 tcp->tcp_cwr_snd_max = tcp->tcp_snxt;
3973 tcp->tcp_ecn_cwr_sent = B_FALSE;
3974 }
3975
3976 /*
3977 * We do Hoe's algorithm. Refer to her
3978 * paper "Improving the Start-up Behavior
3979 * of a Congestion Control Scheme for TCP,"
3980 * appeared in SIGCOMM'96.
3981 *
3982 * Save highest seq no we have sent so far.
3983 * Be careful about the invisible FIN byte.
3984 */
3985 if ((tcp->tcp_valid_bits & TCP_FSS_VALID) &&
3986 (tcp->tcp_unsent == 0)) {
3987 tcp->tcp_rexmit_max = tcp->tcp_fss;
3988 } else {
3989 tcp->tcp_rexmit_max = tcp->tcp_snxt;
3990 }
3991
3992 /*
3993 * For SACK:
3994 * Calculate tcp_pipe, which is the
3995 * estimated number of bytes in
3996 * network.
3997 *
3998 * tcp_fack is the highest sack'ed seq num
3999 * TCP has received.
4000 *
4001 * tcp_pipe is explained in the above quoted
4002 * Fall and Floyd's paper. tcp_fack is
4003 * explained in Mathis and Mahdavi's
4004 * "Forward Acknowledgment: Refining TCP
4005 * Congestion Control" in SIGCOMM '96.
4006 */
4007 if (tcp->tcp_snd_sack_ok) {
4008 if (tcp->tcp_notsack_list != NULL) {
4009 tcp->tcp_pipe = tcp->tcp_snxt -
4010 tcp->tcp_fack;
4011 tcp->tcp_sack_snxt = seg_ack;
4012 flags |= TH_NEED_SACK_REXMIT;
4013 } else {
4014 /*
4015 * Always initialize tcp_pipe
4016 * even though we don't have
4017 * any SACK info. If later
4018 * we get SACK info and
4019 * tcp_pipe is not initialized,
4020 * funny things will happen.
4021 */
4022 tcp->tcp_pipe =
4023 tcp->tcp_cwnd_ssthresh;
4024 }
4025 } else {
4026 flags |= TH_REXMIT_NEEDED;
4027 } /* tcp_snd_sack_ok */
4028
4029 } else {
4030 /*
4031 * Here we perform congestion
4032 * avoidance, but NOT slow start.
4033 * This is known as the Fast
4034 * Recovery Algorithm.
4035 */
4036 if (tcp->tcp_snd_sack_ok &&
4037 tcp->tcp_notsack_list != NULL) {
4038 flags |= TH_NEED_SACK_REXMIT;
4039 tcp->tcp_pipe -= mss;
4040 if (tcp->tcp_pipe < 0)
4041 tcp->tcp_pipe = 0;
4042 } else {
4043 /*
4044 * We know that one more packet has
4045 * left the pipe thus we can update
4046 * cwnd.
4047 */
4048 cwnd = tcp->tcp_cwnd + mss;
4049 if (cwnd > tcp->tcp_cwnd_max)
4050 cwnd = tcp->tcp_cwnd_max;
4051 tcp->tcp_cwnd = cwnd;
4052 if (tcp->tcp_unsent > 0)
4053 flags |= TH_XMIT_NEEDED;
4054 }
4055 }
4056 }
4057 } else if (tcp->tcp_zero_win_probe) {
4058 /*
4059 * If the window has opened, need to arrange
4060 * to send additional data.
4061 */
4062 if (new_swnd != 0) {
4063 /* tcp_suna != tcp_snxt */
4064 /* Packet contains a window update */
4065 TCPS_BUMP_MIB(tcps, tcpInWinUpdate);
4066 tcp->tcp_zero_win_probe = 0;
4067 tcp->tcp_timer_backoff = 0;
4068 tcp->tcp_ms_we_have_waited = 0;
4069
4070 /*
4071 * Transmit starting with tcp_suna since
4072 * the one byte probe is not ack'ed.
4073 * If TCP has sent more than one identical
4074 * probe, tcp_rexmit will be set. That means
4075 * tcp_ss_rexmit() will send out the one
4076 * byte along with new data. Otherwise,
4077 * fake the retransmission.
4078 */
4079 flags |= TH_XMIT_NEEDED;
4080 if (!tcp->tcp_rexmit) {
4081 tcp->tcp_rexmit = B_TRUE;
4082 tcp->tcp_dupack_cnt = 0;
4083 tcp->tcp_rexmit_nxt = tcp->tcp_suna;
4084 tcp->tcp_rexmit_max = tcp->tcp_suna + 1;
4085 }
4086 }
4087 }
4088 goto swnd_update;
4089 }
4090
4091 /*
4092 * Check for "acceptability" of ACK value per RFC 793, pages 72 - 73.
4093 * If the ACK value acks something that we have not yet sent, it might
4094 * be an old duplicate segment. Send an ACK to re-synchronize the
4095 * other side.
4096 * Note: reset in response to unacceptable ACK in SYN_RECEIVE
4097 * state is handled above, so we can always just drop the segment and
4098 * send an ACK here.
4099 *
4100 * In the case where the peer shrinks the window, we see the new window
4101 * update, but all the data sent previously is queued up by the peer.
4102 * To account for this, in tcp_process_shrunk_swnd(), the sequence
4103 * number, which was already sent, and within window, is recorded.
4104 * tcp_snxt is then updated.
4105 *
4106 * If the window has previously shrunk, and an ACK for data not yet
4107 * sent, according to tcp_snxt is recieved, it may still be valid. If
4108 * the ACK is for data within the window at the time the window was
4109 * shrunk, then the ACK is acceptable. In this case tcp_snxt is set to
4110 * the sequence number ACK'ed.
4111 *
4112 * If the ACK covers all the data sent at the time the window was
4113 * shrunk, we can now set tcp_is_wnd_shrnk to B_FALSE.
4114 *
4115 * Should we send ACKs in response to ACK only segments?
4116 */
4117
4118 if (SEQ_GT(seg_ack, tcp->tcp_snxt)) {
4119 if ((tcp->tcp_is_wnd_shrnk) &&
4120 (SEQ_LEQ(seg_ack, tcp->tcp_snxt_shrunk))) {
4121 uint32_t data_acked_ahead_snxt;
4122
4123 data_acked_ahead_snxt = seg_ack - tcp->tcp_snxt;
4124 tcp_update_xmit_tail(tcp, seg_ack);
4125 tcp->tcp_unsent -= data_acked_ahead_snxt;
4126 } else {
4127 TCPS_BUMP_MIB(tcps, tcpInAckUnsent);
4128 /* drop the received segment */
4129 freemsg(mp);
4130
4131 /*
4132 * Send back an ACK. If tcp_drop_ack_unsent_cnt is
4133 * greater than 0, check if the number of such
4134 * bogus ACks is greater than that count. If yes,
4135 * don't send back any ACK. This prevents TCP from
4136 * getting into an ACK storm if somehow an attacker
4137 * successfully spoofs an acceptable segment to our
4138 * peer. If this continues (count > 2 X threshold),
4139 * we should abort this connection.
4140 */
4141 if (tcp_drop_ack_unsent_cnt > 0 &&
4142 ++tcp->tcp_in_ack_unsent >
4143 tcp_drop_ack_unsent_cnt) {
4144 TCP_STAT(tcps, tcp_in_ack_unsent_drop);
4145 if (tcp->tcp_in_ack_unsent > 2 *
4146 tcp_drop_ack_unsent_cnt) {
4147 (void) tcp_clean_death(tcp, EPROTO);
4148 }
4149 return;
4150 }
4151 mp = tcp_ack_mp(tcp);
4152 if (mp != NULL) {
4153 TCPS_BUMP_MIB(tcps, tcpHCOutSegs);
4154 TCPS_BUMP_MIB(tcps, tcpOutAck);
4155 tcp_send_data(tcp, mp);
4156 }
4157 return;
4158 }
4159 } else if (tcp->tcp_is_wnd_shrnk && SEQ_GEQ(seg_ack,
4160 tcp->tcp_snxt_shrunk)) {
4161 tcp->tcp_is_wnd_shrnk = B_FALSE;
4162 }
4163
4164 /*
4165 * TCP gets a new ACK, update the notsack'ed list to delete those
4166 * blocks that are covered by this ACK.
4167 */
4168 if (tcp->tcp_snd_sack_ok && tcp->tcp_notsack_list != NULL) {
4169 tcp_notsack_remove(&(tcp->tcp_notsack_list), seg_ack,
4170 &(tcp->tcp_num_notsack_blk), &(tcp->tcp_cnt_notsack_list));
4171 }
4172
4173 /*
4174 * If we got an ACK after fast retransmit, check to see
4175 * if it is a partial ACK. If it is not and the congestion
4176 * window was inflated to account for the other side's
4177 * cached packets, retract it. If it is, do Hoe's algorithm.
4178 */
4179 if (tcp->tcp_dupack_cnt >= tcps->tcps_dupack_fast_retransmit) {
4180 ASSERT(tcp->tcp_rexmit == B_FALSE);
4181 if (SEQ_GEQ(seg_ack, tcp->tcp_rexmit_max)) {
4182 tcp->tcp_dupack_cnt = 0;
4183 /*
4184 * Restore the orig tcp_cwnd_ssthresh after
4185 * fast retransmit phase.
4186 */
4187 if (tcp->tcp_cwnd > tcp->tcp_cwnd_ssthresh) {
4188 tcp->tcp_cwnd = tcp->tcp_cwnd_ssthresh;
4189 }
4190 tcp->tcp_rexmit_max = seg_ack;
4191 tcp->tcp_cwnd_cnt = 0;
4192
4193 /*
4194 * Remove all notsack info to avoid confusion with
4195 * the next fast retrasnmit/recovery phase.
4196 */
4197 if (tcp->tcp_snd_sack_ok) {
4198 TCP_NOTSACK_REMOVE_ALL(tcp->tcp_notsack_list,
4199 tcp);
4200 }
4201 } else {
4202 if (tcp->tcp_snd_sack_ok &&
4203 tcp->tcp_notsack_list != NULL) {
4204 flags |= TH_NEED_SACK_REXMIT;
4205 tcp->tcp_pipe -= mss;
4206 if (tcp->tcp_pipe < 0)
4207 tcp->tcp_pipe = 0;
4208 } else {
4209 /*
4210 * Hoe's algorithm:
4211 *
4212 * Retransmit the unack'ed segment and
4213 * restart fast recovery. Note that we
4214 * need to scale back tcp_cwnd to the
4215 * original value when we started fast
4216 * recovery. This is to prevent overly
4217 * aggressive behaviour in sending new
4218 * segments.
4219 */
4220 tcp->tcp_cwnd = tcp->tcp_cwnd_ssthresh +
4221 tcps->tcps_dupack_fast_retransmit * mss;
4222 tcp->tcp_cwnd_cnt = tcp->tcp_cwnd;
4223 flags |= TH_REXMIT_NEEDED;
4224 }
4225 }
4226 } else {
4227 tcp->tcp_dupack_cnt = 0;
4228 if (tcp->tcp_rexmit) {
4229 /*
4230 * TCP is retranmitting. If the ACK ack's all
4231 * outstanding data, update tcp_rexmit_max and
4232 * tcp_rexmit_nxt. Otherwise, update tcp_rexmit_nxt
4233 * to the correct value.
4234 *
4235 * Note that SEQ_LEQ() is used. This is to avoid
4236 * unnecessary fast retransmit caused by dup ACKs
4237 * received when TCP does slow start retransmission
4238 * after a time out. During this phase, TCP may
4239 * send out segments which are already received.
4240 * This causes dup ACKs to be sent back.
4241 */
4242 if (SEQ_LEQ(seg_ack, tcp->tcp_rexmit_max)) {
4243 if (SEQ_GT(seg_ack, tcp->tcp_rexmit_nxt)) {
4244 tcp->tcp_rexmit_nxt = seg_ack;
4245 }
4246 if (seg_ack != tcp->tcp_rexmit_max) {
4247 flags |= TH_XMIT_NEEDED;
4248 }
4249 } else {
4250 tcp->tcp_rexmit = B_FALSE;
4251 tcp->tcp_rexmit_nxt = tcp->tcp_snxt;
4252 }
4253 tcp->tcp_ms_we_have_waited = 0;
4254 }
4255 }
4256
4257 TCPS_BUMP_MIB(tcps, tcpInAckSegs);
4258 TCPS_UPDATE_MIB(tcps, tcpInAckBytes, bytes_acked);
4259 tcp->tcp_suna = seg_ack;
4260 if (tcp->tcp_zero_win_probe != 0) {
4261 tcp->tcp_zero_win_probe = 0;
4262 tcp->tcp_timer_backoff = 0;
4263 }
4264
4265 /*
4266 * If tcp_xmit_head is NULL, then it must be the FIN being ack'ed.
4267 * Note that it cannot be the SYN being ack'ed. The code flow
4268 * will not reach here.
4269 */
4270 if (mp1 == NULL) {
4271 goto fin_acked;
4272 }
4273
4274 /*
4275 * Update the congestion window.
4276 *
4277 * If TCP is not ECN capable or TCP is ECN capable but the
4278 * congestion experience bit is not set, increase the tcp_cwnd as
4279 * usual.
4280 */
4281 if (!tcp->tcp_ecn_ok || !(flags & TH_ECE)) {
4282 cwnd = tcp->tcp_cwnd;
4283 add = mss;
4284
4285 if (cwnd >= tcp->tcp_cwnd_ssthresh) {
4286 /*
4287 * This is to prevent an increase of less than 1 MSS of
4288 * tcp_cwnd. With partial increase, tcp_wput_data()
4289 * may send out tinygrams in order to preserve mblk
4290 * boundaries.
4291 *
4292 * By initializing tcp_cwnd_cnt to new tcp_cwnd and
4293 * decrementing it by 1 MSS for every ACKs, tcp_cwnd is
4294 * increased by 1 MSS for every RTTs.
4295 */
4296 if (tcp->tcp_cwnd_cnt <= 0) {
4297 tcp->tcp_cwnd_cnt = cwnd + add;
4298 } else {
4299 tcp->tcp_cwnd_cnt -= add;
4300 add = 0;
4301 }
4302 }
4303 tcp->tcp_cwnd = MIN(cwnd + add, tcp->tcp_cwnd_max);
4304 }
4305
4306 /* See if the latest urgent data has been acknowledged */
4307 if ((tcp->tcp_valid_bits & TCP_URG_VALID) &&
4308 SEQ_GT(seg_ack, tcp->tcp_urg))
4309 tcp->tcp_valid_bits &= ~TCP_URG_VALID;
4310
4311 /*
4312 * Update the RTT estimates. Note that we don't use the TCP
4313 * timestamp option to calculate RTT even if one is present. This is
4314 * because the timestamp option's resolution (CPU tick) is
4315 * too coarse to measure modern datacenter networks' microsecond
4316 * latencies. The timestamp field's resolution is limited by its
4317 * 4-byte width (see RFC1323), and since we always store a
4318 * high-resolution nanosecond presision timestamp along with the data,
4319 * there is no point to ever using the timestamp option.
4320 */
4321 if (SEQ_GT(seg_ack, tcp->tcp_csuna)) {
4322 /*
4323 * An ACK sequence we haven't seen before, so get the RTT
4324 * and update the RTO. But first check if the timestamp is
4325 * valid to use.
4326 */
4327 if ((mp1->b_next != NULL) &&
4328 SEQ_GT(seg_ack, (uint32_t)(uintptr_t)(mp1->b_next))) {
4329 tcp_set_rto(tcp, gethrtime() -
4330 (hrtime_t)(intptr_t)mp1->b_prev);
4331 } else {
4332 TCPS_BUMP_MIB(tcps, tcpRttNoUpdate);
4333 }
4334
4335 /* Remeber the last sequence to be ACKed */
4336 tcp->tcp_csuna = seg_ack;
4337 if (tcp->tcp_set_timer == 1) {
4338 TCP_TIMER_RESTART(tcp, tcp->tcp_rto);
4339 tcp->tcp_set_timer = 0;
4340 }
4341 } else {
4342 TCPS_BUMP_MIB(tcps, tcpRttNoUpdate);
4343 }
4344
4345 /* Eat acknowledged bytes off the xmit queue. */
4346 for (;;) {
4347 mblk_t *mp2;
4348 uchar_t *wptr;
4349
4350 wptr = mp1->b_wptr;
4351 ASSERT((uintptr_t)(wptr - mp1->b_rptr) <= (uintptr_t)INT_MAX);
4352 bytes_acked -= (int)(wptr - mp1->b_rptr);
4353 if (bytes_acked < 0) {
4354 mp1->b_rptr = wptr + bytes_acked;
4355 /*
4356 * Set a new timestamp if all the bytes timed by the
4357 * old timestamp have been ack'ed.
4358 */
4359 if (SEQ_GT(seg_ack,
4360 (uint32_t)(uintptr_t)(mp1->b_next))) {
4361 mp1->b_prev =
4362 (mblk_t *)(intptr_t)gethrtime();
4363 mp1->b_next = NULL;
4364 }
4365 break;
4366 }
4367 mp1->b_next = NULL;
4368 mp1->b_prev = NULL;
4369 mp2 = mp1;
4370 mp1 = mp1->b_cont;
4371
4372 /*
4373 * This notification is required for some zero-copy
4374 * clients to maintain a copy semantic. After the data
4375 * is ack'ed, client is safe to modify or reuse the buffer.
4376 */
4377 if (tcp->tcp_snd_zcopy_aware &&
4378 (mp2->b_datap->db_struioflag & STRUIO_ZCNOTIFY))
4379 tcp_zcopy_notify(tcp);
4380 freeb(mp2);
4381 if (bytes_acked == 0) {
4382 if (mp1 == NULL) {
4383 /* Everything is ack'ed, clear the tail. */
4384 tcp->tcp_xmit_tail = NULL;
4385 /*
4386 * Cancel the timer unless we are still
4387 * waiting for an ACK for the FIN packet.
4388 */
4389 if (tcp->tcp_timer_tid != 0 &&
4390 tcp->tcp_snxt == tcp->tcp_suna) {
4391 (void) TCP_TIMER_CANCEL(tcp,
4392 tcp->tcp_timer_tid);
4393 tcp->tcp_timer_tid = 0;
4394 }
4395 goto pre_swnd_update;
4396 }
4397 if (mp2 != tcp->tcp_xmit_tail)
4398 break;
4399 tcp->tcp_xmit_tail = mp1;
4400 ASSERT((uintptr_t)(mp1->b_wptr - mp1->b_rptr) <=
4401 (uintptr_t)INT_MAX);
4402 tcp->tcp_xmit_tail_unsent = (int)(mp1->b_wptr -
4403 mp1->b_rptr);
4404 break;
4405 }
4406 if (mp1 == NULL) {
4407 /*
4408 * More was acked but there is nothing more
4409 * outstanding. This means that the FIN was
4410 * just acked or that we're talking to a clown.
4411 */
4412 fin_acked:
4413 ASSERT(tcp->tcp_fin_sent);
4414 tcp->tcp_xmit_tail = NULL;
4415 if (tcp->tcp_fin_sent) {
4416 /* FIN was acked - making progress */
4417 if (!tcp->tcp_fin_acked)
4418 tcp->tcp_ip_forward_progress = B_TRUE;
4419 tcp->tcp_fin_acked = B_TRUE;
4420 if (tcp->tcp_linger_tid != 0 &&
4421 TCP_TIMER_CANCEL(tcp,
4422 tcp->tcp_linger_tid) >= 0) {
4423 tcp_stop_lingering(tcp);
4424 freemsg(mp);
4425 mp = NULL;
4426 }
4427 } else {
4428 /*
4429 * We should never get here because
4430 * we have already checked that the
4431 * number of bytes ack'ed should be
4432 * smaller than or equal to what we
4433 * have sent so far (it is the
4434 * acceptability check of the ACK).
4435 * We can only get here if the send
4436 * queue is corrupted.
4437 *
4438 * Terminate the connection and
4439 * panic the system. It is better
4440 * for us to panic instead of
4441 * continuing to avoid other disaster.
4442 */
4443 tcp_xmit_ctl(NULL, tcp, tcp->tcp_snxt,
4444 tcp->tcp_rnxt, TH_RST|TH_ACK);
4445 panic("Memory corruption "
4446 "detected for connection %s.",
4447 tcp_display(tcp, NULL,
4448 DISP_ADDR_AND_PORT));
4449 /*NOTREACHED*/
4450 }
4451 goto pre_swnd_update;
4452 }
4453 ASSERT(mp2 != tcp->tcp_xmit_tail);
4454 }
4455 if (tcp->tcp_unsent) {
4456 flags |= TH_XMIT_NEEDED;
4457 }
4458 pre_swnd_update:
4459 tcp->tcp_xmit_head = mp1;
4460 swnd_update:
4461 /*
4462 * The following check is different from most other implementations.
4463 * For bi-directional transfer, when segments are dropped, the
4464 * "normal" check will not accept a window update in those
4465 * retransmitted segemnts. Failing to do that, TCP may send out
4466 * segments which are outside receiver's window. As TCP accepts
4467 * the ack in those retransmitted segments, if the window update in
4468 * the same segment is not accepted, TCP will incorrectly calculates
4469 * that it can send more segments. This can create a deadlock
4470 * with the receiver if its window becomes zero.
4471 */
4472 if (SEQ_LT(tcp->tcp_swl2, seg_ack) ||
4473 SEQ_LT(tcp->tcp_swl1, seg_seq) ||
4474 (tcp->tcp_swl1 == seg_seq && new_swnd > tcp->tcp_swnd)) {
4475 /*
4476 * The criteria for update is:
4477 *
4478 * 1. the segment acknowledges some data. Or
4479 * 2. the segment is new, i.e. it has a higher seq num. Or
4480 * 3. the segment is not old and the advertised window is
4481 * larger than the previous advertised window.
4482 */
4483 if (tcp->tcp_unsent && new_swnd > tcp->tcp_swnd)
4484 flags |= TH_XMIT_NEEDED;
4485 tcp->tcp_swnd = new_swnd;
4486 if (new_swnd > tcp->tcp_max_swnd)
4487 tcp->tcp_max_swnd = new_swnd;
4488 tcp->tcp_swl1 = seg_seq;
4489 tcp->tcp_swl2 = seg_ack;
4490 }
4491 est:
4492 if (tcp->tcp_state > TCPS_ESTABLISHED) {
4493
4494 switch (tcp->tcp_state) {
4495 case TCPS_FIN_WAIT_1:
4496 if (tcp->tcp_fin_acked) {
4497 tcp->tcp_state = TCPS_FIN_WAIT_2;
4498 DTRACE_TCP6(state__change, void, NULL,
4499 ip_xmit_attr_t *, connp->conn_ixa,
4500 void, NULL, tcp_t *, tcp, void, NULL,
4501 int32_t, TCPS_FIN_WAIT_1);
4502 /*
4503 * We implement the non-standard BSD/SunOS
4504 * FIN_WAIT_2 flushing algorithm.
4505 * If there is no user attached to this
4506 * TCP endpoint, then this TCP struct
4507 * could hang around forever in FIN_WAIT_2
4508 * state if the peer forgets to send us
4509 * a FIN. To prevent this, we wait only
4510 * 2*MSL (a convenient time value) for
4511 * the FIN to arrive. If it doesn't show up,
4512 * we flush the TCP endpoint. This algorithm,
4513 * though a violation of RFC-793, has worked
4514 * for over 10 years in BSD systems.
4515 * Note: SunOS 4.x waits 675 seconds before
4516 * flushing the FIN_WAIT_2 connection.
4517 */
4518 TCP_TIMER_RESTART(tcp,
4519 tcp->tcp_fin_wait_2_flush_interval);
4520 }
4521 break;
4522 case TCPS_FIN_WAIT_2:
4523 break; /* Shutdown hook? */
4524 case TCPS_LAST_ACK:
4525 freemsg(mp);
4526 if (tcp->tcp_fin_acked) {
4527 (void) tcp_clean_death(tcp, 0);
4528 return;
4529 }
4530 goto xmit_check;
4531 case TCPS_CLOSING:
4532 if (tcp->tcp_fin_acked) {
4533 SET_TIME_WAIT(tcps, tcp, connp);
4534 DTRACE_TCP6(state__change, void, NULL,
4535 ip_xmit_attr_t *, connp->conn_ixa, void,
4536 NULL, tcp_t *, tcp, void, NULL, int32_t,
4537 TCPS_CLOSING);
4538 }
4539 /*FALLTHRU*/
4540 case TCPS_CLOSE_WAIT:
4541 freemsg(mp);
4542 goto xmit_check;
4543 default:
4544 ASSERT(tcp->tcp_state != TCPS_TIME_WAIT);
4545 break;
4546 }
4547 }
4548 if (flags & TH_FIN) {
4549 /* Make sure we ack the fin */
4550 flags |= TH_ACK_NEEDED;
4551 if (!tcp->tcp_fin_rcvd) {
4552 tcp->tcp_fin_rcvd = B_TRUE;
4553 tcp->tcp_rnxt++;
4554 tcpha = tcp->tcp_tcpha;
4555 tcpha->tha_ack = htonl(tcp->tcp_rnxt);
4556
4557 /*
4558 * Generate the ordrel_ind at the end unless the
4559 * conn is detached or it is a STREAMS based eager.
4560 * In the eager case we defer the notification until
4561 * tcp_accept_finish has run.
4562 */
4563 if (!TCP_IS_DETACHED(tcp) && (IPCL_IS_NONSTR(connp) ||
4564 (tcp->tcp_listener == NULL &&
4565 !tcp->tcp_hard_binding)))
4566 flags |= TH_ORDREL_NEEDED;
4567 switch (tcp->tcp_state) {
4568 case TCPS_SYN_RCVD:
4569 tcp->tcp_state = TCPS_CLOSE_WAIT;
4570 DTRACE_TCP6(state__change, void, NULL,
4571 ip_xmit_attr_t *, connp->conn_ixa,
4572 void, NULL, tcp_t *, tcp, void, NULL,
4573 int32_t, TCPS_SYN_RCVD);
4574 /* Keepalive? */
4575 break;
4576 case TCPS_ESTABLISHED:
4577 tcp->tcp_state = TCPS_CLOSE_WAIT;
4578 DTRACE_TCP6(state__change, void, NULL,
4579 ip_xmit_attr_t *, connp->conn_ixa,
4580 void, NULL, tcp_t *, tcp, void, NULL,
4581 int32_t, TCPS_ESTABLISHED);
4582 /* Keepalive? */
4583 break;
4584 case TCPS_FIN_WAIT_1:
4585 if (!tcp->tcp_fin_acked) {
4586 tcp->tcp_state = TCPS_CLOSING;
4587 DTRACE_TCP6(state__change, void, NULL,
4588 ip_xmit_attr_t *, connp->conn_ixa,
4589 void, NULL, tcp_t *, tcp, void,
4590 NULL, int32_t, TCPS_FIN_WAIT_1);
4591 break;
4592 }
4593 /* FALLTHRU */
4594 case TCPS_FIN_WAIT_2:
4595 SET_TIME_WAIT(tcps, tcp, connp);
4596 DTRACE_TCP6(state__change, void, NULL,
4597 ip_xmit_attr_t *, connp->conn_ixa, void,
4598 NULL, tcp_t *, tcp, void, NULL, int32_t,
4599 TCPS_FIN_WAIT_2);
4600 if (seg_len) {
4601 /*
4602 * implies data piggybacked on FIN.
4603 * break to handle data.
4604 */
4605 break;
4606 }
4607 freemsg(mp);
4608 goto ack_check;
4609 }
4610 }
4611 }
4612 if (mp == NULL)
4613 goto xmit_check;
4614 if (seg_len == 0) {
4615 freemsg(mp);
4616 goto xmit_check;
4617 }
4618 if (mp->b_rptr == mp->b_wptr) {
4619 /*
4620 * The header has been consumed, so we remove the
4621 * zero-length mblk here.
4622 */
4623 mp1 = mp;
4624 mp = mp->b_cont;
4625 freeb(mp1);
4626 }
4627 update_ack:
4628 tcpha = tcp->tcp_tcpha;
4629 tcp->tcp_rack_cnt++;
4630 {
4631 uint32_t cur_max;
4632
4633 cur_max = tcp->tcp_rack_cur_max;
4634 if (tcp->tcp_rack_cnt >= cur_max) {
4635 /*
4636 * We have more unacked data than we should - send
4637 * an ACK now.
4638 */
4639 flags |= TH_ACK_NEEDED;
4640 cur_max++;
4641 if (cur_max > tcp->tcp_rack_abs_max)
4642 tcp->tcp_rack_cur_max = tcp->tcp_rack_abs_max;
4643 else
4644 tcp->tcp_rack_cur_max = cur_max;
4645 } else if (TCP_IS_DETACHED(tcp)) {
4646 /* We don't have an ACK timer for detached TCP. */
4647 flags |= TH_ACK_NEEDED;
4648 } else if (seg_len < mss) {
4649 /*
4650 * If we get a segment that is less than an mss, and we
4651 * already have unacknowledged data, and the amount
4652 * unacknowledged is not a multiple of mss, then we
4653 * better generate an ACK now. Otherwise, this may be
4654 * the tail piece of a transaction, and we would rather
4655 * wait for the response.
4656 */
4657 uint32_t udif;
4658 ASSERT((uintptr_t)(tcp->tcp_rnxt - tcp->tcp_rack) <=
4659 (uintptr_t)INT_MAX);
4660 udif = (int)(tcp->tcp_rnxt - tcp->tcp_rack);
4661 if (udif && (udif % mss))
4662 flags |= TH_ACK_NEEDED;
4663 else
4664 flags |= TH_ACK_TIMER_NEEDED;
4665 } else {
4666 /* Start delayed ack timer */
4667 flags |= TH_ACK_TIMER_NEEDED;
4668 }
4669 }
4670 tcp->tcp_rnxt += seg_len;
4671 tcpha->tha_ack = htonl(tcp->tcp_rnxt);
4672
4673 if (mp == NULL)
4674 goto xmit_check;
4675
4676 /* Update SACK list */
4677 if (tcp->tcp_snd_sack_ok && tcp->tcp_num_sack_blk > 0) {
4678 tcp_sack_remove(tcp->tcp_sack_list, tcp->tcp_rnxt,
4679 &(tcp->tcp_num_sack_blk));
4680 }
4681
4682 if (tcp->tcp_urp_mp) {
4683 tcp->tcp_urp_mp->b_cont = mp;
4684 mp = tcp->tcp_urp_mp;
4685 tcp->tcp_urp_mp = NULL;
4686 /* Ready for a new signal. */
4687 tcp->tcp_urp_last_valid = B_FALSE;
4688 #ifdef DEBUG
4689 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
4690 "tcp_rput: sending exdata_ind %s",
4691 tcp_display(tcp, NULL, DISP_PORT_ONLY));
4692 #endif /* DEBUG */
4693 }
4694
4695 /*
4696 * Check for ancillary data changes compared to last segment.
4697 */
4698 if (connp->conn_recv_ancillary.crb_all != 0) {
4699 mp = tcp_input_add_ancillary(tcp, mp, &ipp, ira);
4700 if (mp == NULL)
4701 return;
4702 }
4703
4704 if (IPCL_IS_NONSTR(connp)) {
4705 /*
4706 * Non-STREAMS socket
4707 */
4708 boolean_t push = flags & (TH_PUSH|TH_FIN);
4709 int error;
4710
4711 if ((*sockupcalls->su_recv)(connp->conn_upper_handle,
4712 mp, seg_len, 0, &error, &push) <= 0) {
4713 /*
4714 * We should never be in middle of a
4715 * fallback, the squeue guarantees that.
4716 */
4717 ASSERT(error != EOPNOTSUPP);
4718 if (error == ENOSPC)
4719 tcp->tcp_rwnd -= seg_len;
4720 } else if (push) {
4721 /* PUSH bit set and sockfs is not flow controlled */
4722 flags |= tcp_rwnd_reopen(tcp);
4723 }
4724 } else if (tcp->tcp_listener != NULL || tcp->tcp_hard_binding) {
4725 /*
4726 * Side queue inbound data until the accept happens.
4727 * tcp_accept/tcp_rput drains this when the accept happens.
4728 * M_DATA is queued on b_cont. Otherwise (T_OPTDATA_IND or
4729 * T_EXDATA_IND) it is queued on b_next.
4730 * XXX Make urgent data use this. Requires:
4731 * Removing tcp_listener check for TH_URG
4732 * Making M_PCPROTO and MARK messages skip the eager case
4733 */
4734
4735 tcp_rcv_enqueue(tcp, mp, seg_len, ira->ira_cred);
4736 } else {
4737 /* Active STREAMS socket */
4738 if (mp->b_datap->db_type != M_DATA ||
4739 (flags & TH_MARKNEXT_NEEDED)) {
4740 if (tcp->tcp_rcv_list != NULL) {
4741 flags |= tcp_rcv_drain(tcp);
4742 }
4743 ASSERT(tcp->tcp_rcv_list == NULL ||
4744 tcp->tcp_fused_sigurg);
4745
4746 if (flags & TH_MARKNEXT_NEEDED) {
4747 #ifdef DEBUG
4748 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
4749 "tcp_rput: sending MSGMARKNEXT %s",
4750 tcp_display(tcp, NULL,
4751 DISP_PORT_ONLY));
4752 #endif /* DEBUG */
4753 mp->b_flag |= MSGMARKNEXT;
4754 flags &= ~TH_MARKNEXT_NEEDED;
4755 }
4756
4757 if (is_system_labeled())
4758 tcp_setcred_data(mp, ira);
4759
4760 putnext(connp->conn_rq, mp);
4761 if (!canputnext(connp->conn_rq))
4762 tcp->tcp_rwnd -= seg_len;
4763 } else if ((flags & (TH_PUSH|TH_FIN)) ||
4764 tcp->tcp_rcv_cnt + seg_len >= connp->conn_rcvbuf >> 3) {
4765 if (tcp->tcp_rcv_list != NULL) {
4766 /*
4767 * Enqueue the new segment first and then
4768 * call tcp_rcv_drain() to send all data
4769 * up. The other way to do this is to
4770 * send all queued data up and then call
4771 * putnext() to send the new segment up.
4772 * This way can remove the else part later
4773 * on.
4774 *
4775 * We don't do this to avoid one more call to
4776 * canputnext() as tcp_rcv_drain() needs to
4777 * call canputnext().
4778 */
4779 tcp_rcv_enqueue(tcp, mp, seg_len,
4780 ira->ira_cred);
4781 flags |= tcp_rcv_drain(tcp);
4782 } else {
4783 if (is_system_labeled())
4784 tcp_setcred_data(mp, ira);
4785
4786 putnext(connp->conn_rq, mp);
4787 if (!canputnext(connp->conn_rq))
4788 tcp->tcp_rwnd -= seg_len;
4789 }
4790 } else {
4791 /*
4792 * Enqueue all packets when processing an mblk
4793 * from the co queue and also enqueue normal packets.
4794 */
4795 tcp_rcv_enqueue(tcp, mp, seg_len, ira->ira_cred);
4796 }
4797 /*
4798 * Make sure the timer is running if we have data waiting
4799 * for a push bit. This provides resiliency against
4800 * implementations that do not correctly generate push bits.
4801 */
4802 if (tcp->tcp_rcv_list != NULL && tcp->tcp_push_tid == 0) {
4803 /*
4804 * The connection may be closed at this point, so don't
4805 * do anything for a detached tcp.
4806 */
4807 if (!TCP_IS_DETACHED(tcp))
4808 tcp->tcp_push_tid = TCP_TIMER(tcp,
4809 tcp_push_timer,
4810 tcps->tcps_push_timer_interval);
4811 }
4812 }
4813
4814 xmit_check:
4815 /* Is there anything left to do? */
4816 ASSERT(!(flags & TH_MARKNEXT_NEEDED));
4817 if ((flags & (TH_REXMIT_NEEDED|TH_XMIT_NEEDED|TH_ACK_NEEDED|
4818 TH_NEED_SACK_REXMIT|TH_LIMIT_XMIT|TH_ACK_TIMER_NEEDED|
4819 TH_ORDREL_NEEDED|TH_SEND_URP_MARK)) == 0)
4820 goto done;
4821
4822 /* Any transmit work to do and a non-zero window? */
4823 if ((flags & (TH_REXMIT_NEEDED|TH_XMIT_NEEDED|TH_NEED_SACK_REXMIT|
4824 TH_LIMIT_XMIT)) && tcp->tcp_swnd != 0) {
4825 if (flags & TH_REXMIT_NEEDED) {
4826 uint32_t snd_size = tcp->tcp_snxt - tcp->tcp_suna;
4827
4828 TCPS_BUMP_MIB(tcps, tcpOutFastRetrans);
4829 if (snd_size > mss)
4830 snd_size = mss;
4831 if (snd_size > tcp->tcp_swnd)
4832 snd_size = tcp->tcp_swnd;
4833 mp1 = tcp_xmit_mp(tcp, tcp->tcp_xmit_head, snd_size,
4834 NULL, NULL, tcp->tcp_suna, B_TRUE, &snd_size,
4835 B_TRUE);
4836
4837 if (mp1 != NULL) {
4838 tcp->tcp_xmit_head->b_prev =
4839 (mblk_t *)(intptr_t)gethrtime();
4840 tcp->tcp_csuna = tcp->tcp_snxt;
4841 TCPS_BUMP_MIB(tcps, tcpRetransSegs);
4842 TCPS_UPDATE_MIB(tcps, tcpRetransBytes,
4843 snd_size);
4844 tcp->tcp_cs.tcp_out_retrans_segs++;
4845 tcp->tcp_cs.tcp_out_retrans_bytes += snd_size;
4846 tcp_send_data(tcp, mp1);
4847 }
4848 }
4849 if (flags & TH_NEED_SACK_REXMIT) {
4850 tcp_sack_rexmit(tcp, &flags);
4851 }
4852 /*
4853 * For TH_LIMIT_XMIT, tcp_wput_data() is called to send
4854 * out new segment. Note that tcp_rexmit should not be
4855 * set, otherwise TH_LIMIT_XMIT should not be set.
4856 */
4857 if (flags & (TH_XMIT_NEEDED|TH_LIMIT_XMIT)) {
4858 if (!tcp->tcp_rexmit) {
4859 tcp_wput_data(tcp, NULL, B_FALSE);
4860 } else {
4861 tcp_ss_rexmit(tcp);
4862 }
4863 }
4864 /*
4865 * Adjust tcp_cwnd back to normal value after sending
4866 * new data segments.
4867 */
4868 if (flags & TH_LIMIT_XMIT) {
4869 tcp->tcp_cwnd -= mss << (tcp->tcp_dupack_cnt - 1);
4870 /*
4871 * This will restart the timer. Restarting the
4872 * timer is used to avoid a timeout before the
4873 * limited transmitted segment's ACK gets back.
4874 */
4875 if (tcp->tcp_xmit_head != NULL) {
4876 tcp->tcp_xmit_head->b_prev =
4877 (mblk_t *)(intptr_t)gethrtime();
4878 }
4879 }
4880
4881 /* Anything more to do? */
4882 if ((flags & (TH_ACK_NEEDED|TH_ACK_TIMER_NEEDED|
4883 TH_ORDREL_NEEDED|TH_SEND_URP_MARK)) == 0)
4884 goto done;
4885 }
4886 ack_check:
4887 if (flags & TH_SEND_URP_MARK) {
4888 ASSERT(tcp->tcp_urp_mark_mp);
4889 ASSERT(!IPCL_IS_NONSTR(connp));
4890 /*
4891 * Send up any queued data and then send the mark message
4892 */
4893 if (tcp->tcp_rcv_list != NULL) {
4894 flags |= tcp_rcv_drain(tcp);
4895
4896 }
4897 ASSERT(tcp->tcp_rcv_list == NULL || tcp->tcp_fused_sigurg);
4898 mp1 = tcp->tcp_urp_mark_mp;
4899 tcp->tcp_urp_mark_mp = NULL;
4900 if (is_system_labeled())
4901 tcp_setcred_data(mp1, ira);
4902
4903 putnext(connp->conn_rq, mp1);
4904 #ifdef DEBUG
4905 (void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE,
4906 "tcp_rput: sending zero-length %s %s",
4907 ((mp1->b_flag & MSGMARKNEXT) ? "MSGMARKNEXT" :
4908 "MSGNOTMARKNEXT"),
4909 tcp_display(tcp, NULL, DISP_PORT_ONLY));
4910 #endif /* DEBUG */
4911 flags &= ~TH_SEND_URP_MARK;
4912 }
4913 if (flags & TH_ACK_NEEDED) {
4914 /*
4915 * Time to send an ack for some reason.
4916 */
4917 mp1 = tcp_ack_mp(tcp);
4918
4919 if (mp1 != NULL) {
4920 tcp_send_data(tcp, mp1);
4921 TCPS_BUMP_MIB(tcps, tcpHCOutSegs);
4922 TCPS_BUMP_MIB(tcps, tcpOutAck);
4923 }
4924 if (tcp->tcp_ack_tid != 0) {
4925 (void) TCP_TIMER_CANCEL(tcp, tcp->tcp_ack_tid);
4926 tcp->tcp_ack_tid = 0;
4927 }
4928 }
4929 if (flags & TH_ACK_TIMER_NEEDED) {
4930 /*
4931 * Arrange for deferred ACK or push wait timeout.
4932 * Start timer if it is not already running.
4933 */
4934 if (tcp->tcp_ack_tid == 0) {
4935 tcp->tcp_ack_tid = TCP_TIMER(tcp, tcp_ack_timer,
4936 tcp->tcp_localnet ?
4937 tcps->tcps_local_dack_interval :
4938 tcps->tcps_deferred_ack_interval);
4939 }
4940 }
4941 if (flags & TH_ORDREL_NEEDED) {
4942 /*
4943 * Notify upper layer about an orderly release. If this is
4944 * a non-STREAMS socket, then just make an upcall. For STREAMS
4945 * we send up an ordrel_ind, unless this is an eager, in which
4946 * case the ordrel will be sent when tcp_accept_finish runs.
4947 * Note that for non-STREAMS we make an upcall even if it is an
4948 * eager, because we have an upper handle to send it to.
4949 */
4950 ASSERT(IPCL_IS_NONSTR(connp) || tcp->tcp_listener == NULL);
4951 ASSERT(!tcp->tcp_detached);
4952
4953 if (IPCL_IS_NONSTR(connp)) {
4954 ASSERT(tcp->tcp_ordrel_mp == NULL);
4955 tcp->tcp_ordrel_done = B_TRUE;
4956 (*sockupcalls->su_opctl)(connp->conn_upper_handle,
4957 SOCK_OPCTL_SHUT_RECV, 0);
4958 goto done;
4959 }
4960
4961 if (tcp->tcp_rcv_list != NULL) {
4962 /*
4963 * Push any mblk(s) enqueued from co processing.
4964 */
4965 flags |= tcp_rcv_drain(tcp);
4966 }
4967 ASSERT(tcp->tcp_rcv_list == NULL || tcp->tcp_fused_sigurg);
4968
4969 mp1 = tcp->tcp_ordrel_mp;
4970 tcp->tcp_ordrel_mp = NULL;
4971 tcp->tcp_ordrel_done = B_TRUE;
4972 putnext(connp->conn_rq, mp1);
4973 }
4974 done:
4975 ASSERT(!(flags & TH_MARKNEXT_NEEDED));
4976 }
4977
4978 /*
4979 * Attach ancillary data to a received TCP segments for the
4980 * ancillary pieces requested by the application that are
4981 * different than they were in the previous data segment.
4982 *
4983 * Save the "current" values once memory allocation is ok so that
4984 * when memory allocation fails we can just wait for the next data segment.
4985 */
4986 static mblk_t *
4987 tcp_input_add_ancillary(tcp_t *tcp, mblk_t *mp, ip_pkt_t *ipp,
4988 ip_recv_attr_t *ira)
4989 {
4990 struct T_optdata_ind *todi;
4991 int optlen;
4992 uchar_t *optptr;
4993 struct T_opthdr *toh;
4994 crb_t addflag; /* Which pieces to add */
4995 mblk_t *mp1;
4996 conn_t *connp = tcp->tcp_connp;
4997
4998 optlen = 0;
4999 addflag.crb_all = 0;
5000 /* If app asked for pktinfo and the index has changed ... */
5001 if (connp->conn_recv_ancillary.crb_ip_recvpktinfo &&
5002 ira->ira_ruifindex != tcp->tcp_recvifindex) {
5003 optlen += sizeof (struct T_opthdr) +
5004 sizeof (struct in6_pktinfo);
5005 addflag.crb_ip_recvpktinfo = 1;
5006 }
5007 /* If app asked for hoplimit and it has changed ... */
5008 if (connp->conn_recv_ancillary.crb_ipv6_recvhoplimit &&
5009 ipp->ipp_hoplimit != tcp->tcp_recvhops) {
5010 optlen += sizeof (struct T_opthdr) + sizeof (uint_t);
5011 addflag.crb_ipv6_recvhoplimit = 1;
5012 }
5013 /* If app asked for tclass and it has changed ... */
5014 if (connp->conn_recv_ancillary.crb_ipv6_recvtclass &&
5015 ipp->ipp_tclass != tcp->tcp_recvtclass) {
5016 optlen += sizeof (struct T_opthdr) + sizeof (uint_t);
5017 addflag.crb_ipv6_recvtclass = 1;
5018 }
5019 /*
5020 * If app asked for hopbyhop headers and it has changed ...
5021 * For security labels, note that (1) security labels can't change on
5022 * a connected socket at all, (2) we're connected to at most one peer,
5023 * (3) if anything changes, then it must be some other extra option.
5024 */
5025 if (connp->conn_recv_ancillary.crb_ipv6_recvhopopts &&
5026 ip_cmpbuf(tcp->tcp_hopopts, tcp->tcp_hopoptslen,
5027 (ipp->ipp_fields & IPPF_HOPOPTS),
5028 ipp->ipp_hopopts, ipp->ipp_hopoptslen)) {
5029 optlen += sizeof (struct T_opthdr) + ipp->ipp_hopoptslen;
5030 addflag.crb_ipv6_recvhopopts = 1;
5031 if (!ip_allocbuf((void **)&tcp->tcp_hopopts,
5032 &tcp->tcp_hopoptslen, (ipp->ipp_fields & IPPF_HOPOPTS),
5033 ipp->ipp_hopopts, ipp->ipp_hopoptslen))
5034 return (mp);
5035 }
5036 /* If app asked for dst headers before routing headers ... */
5037 if (connp->conn_recv_ancillary.crb_ipv6_recvrthdrdstopts &&
5038 ip_cmpbuf(tcp->tcp_rthdrdstopts, tcp->tcp_rthdrdstoptslen,
5039 (ipp->ipp_fields & IPPF_RTHDRDSTOPTS),
5040 ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen)) {
5041 optlen += sizeof (struct T_opthdr) +
5042 ipp->ipp_rthdrdstoptslen;
5043 addflag.crb_ipv6_recvrthdrdstopts = 1;
5044 if (!ip_allocbuf((void **)&tcp->tcp_rthdrdstopts,
5045 &tcp->tcp_rthdrdstoptslen,
5046 (ipp->ipp_fields & IPPF_RTHDRDSTOPTS),
5047 ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen))
5048 return (mp);
5049 }
5050 /* If app asked for routing headers and it has changed ... */
5051 if (connp->conn_recv_ancillary.crb_ipv6_recvrthdr &&
5052 ip_cmpbuf(tcp->tcp_rthdr, tcp->tcp_rthdrlen,
5053 (ipp->ipp_fields & IPPF_RTHDR),
5054 ipp->ipp_rthdr, ipp->ipp_rthdrlen)) {
5055 optlen += sizeof (struct T_opthdr) + ipp->ipp_rthdrlen;
5056 addflag.crb_ipv6_recvrthdr = 1;
5057 if (!ip_allocbuf((void **)&tcp->tcp_rthdr,
5058 &tcp->tcp_rthdrlen, (ipp->ipp_fields & IPPF_RTHDR),
5059 ipp->ipp_rthdr, ipp->ipp_rthdrlen))
5060 return (mp);
5061 }
5062 /* If app asked for dest headers and it has changed ... */
5063 if ((connp->conn_recv_ancillary.crb_ipv6_recvdstopts ||
5064 connp->conn_recv_ancillary.crb_old_ipv6_recvdstopts) &&
5065 ip_cmpbuf(tcp->tcp_dstopts, tcp->tcp_dstoptslen,
5066 (ipp->ipp_fields & IPPF_DSTOPTS),
5067 ipp->ipp_dstopts, ipp->ipp_dstoptslen)) {
5068 optlen += sizeof (struct T_opthdr) + ipp->ipp_dstoptslen;
5069 addflag.crb_ipv6_recvdstopts = 1;
5070 if (!ip_allocbuf((void **)&tcp->tcp_dstopts,
5071 &tcp->tcp_dstoptslen, (ipp->ipp_fields & IPPF_DSTOPTS),
5072 ipp->ipp_dstopts, ipp->ipp_dstoptslen))
5073 return (mp);
5074 }
5075
5076 if (optlen == 0) {
5077 /* Nothing to add */
5078 return (mp);
5079 }
5080 mp1 = allocb(sizeof (struct T_optdata_ind) + optlen, BPRI_MED);
5081 if (mp1 == NULL) {
5082 /*
5083 * Defer sending ancillary data until the next TCP segment
5084 * arrives.
5085 */
5086 return (mp);
5087 }
5088 mp1->b_cont = mp;
5089 mp = mp1;
5090 mp->b_wptr += sizeof (*todi) + optlen;
5091 mp->b_datap->db_type = M_PROTO;
5092 todi = (struct T_optdata_ind *)mp->b_rptr;
5093 todi->PRIM_type = T_OPTDATA_IND;
5094 todi->DATA_flag = 1; /* MORE data */
5095 todi->OPT_length = optlen;
5096 todi->OPT_offset = sizeof (*todi);
5097 optptr = (uchar_t *)&todi[1];
5098 /*
5099 * If app asked for pktinfo and the index has changed ...
5100 * Note that the local address never changes for the connection.
5101 */
5102 if (addflag.crb_ip_recvpktinfo) {
5103 struct in6_pktinfo *pkti;
5104 uint_t ifindex;
5105
5106 ifindex = ira->ira_ruifindex;
5107 toh = (struct T_opthdr *)optptr;
5108 toh->level = IPPROTO_IPV6;
5109 toh->name = IPV6_PKTINFO;
5110 toh->len = sizeof (*toh) + sizeof (*pkti);
5111 toh->status = 0;
5112 optptr += sizeof (*toh);
5113 pkti = (struct in6_pktinfo *)optptr;
5114 pkti->ipi6_addr = connp->conn_laddr_v6;
5115 pkti->ipi6_ifindex = ifindex;
5116 optptr += sizeof (*pkti);
5117 ASSERT(OK_32PTR(optptr));
5118 /* Save as "last" value */
5119 tcp->tcp_recvifindex = ifindex;
5120 }
5121 /* If app asked for hoplimit and it has changed ... */
5122 if (addflag.crb_ipv6_recvhoplimit) {
5123 toh = (struct T_opthdr *)optptr;
5124 toh->level = IPPROTO_IPV6;
5125 toh->name = IPV6_HOPLIMIT;
5126 toh->len = sizeof (*toh) + sizeof (uint_t);
5127 toh->status = 0;
5128 optptr += sizeof (*toh);
5129 *(uint_t *)optptr = ipp->ipp_hoplimit;
5130 optptr += sizeof (uint_t);
5131 ASSERT(OK_32PTR(optptr));
5132 /* Save as "last" value */
5133 tcp->tcp_recvhops = ipp->ipp_hoplimit;
5134 }
5135 /* If app asked for tclass and it has changed ... */
5136 if (addflag.crb_ipv6_recvtclass) {
5137 toh = (struct T_opthdr *)optptr;
5138 toh->level = IPPROTO_IPV6;
5139 toh->name = IPV6_TCLASS;
5140 toh->len = sizeof (*toh) + sizeof (uint_t);
5141 toh->status = 0;
5142 optptr += sizeof (*toh);
5143 *(uint_t *)optptr = ipp->ipp_tclass;
5144 optptr += sizeof (uint_t);
5145 ASSERT(OK_32PTR(optptr));
5146 /* Save as "last" value */
5147 tcp->tcp_recvtclass = ipp->ipp_tclass;
5148 }
5149 if (addflag.crb_ipv6_recvhopopts) {
5150 toh = (struct T_opthdr *)optptr;
5151 toh->level = IPPROTO_IPV6;
5152 toh->name = IPV6_HOPOPTS;
5153 toh->len = sizeof (*toh) + ipp->ipp_hopoptslen;
5154 toh->status = 0;
5155 optptr += sizeof (*toh);
5156 bcopy((uchar_t *)ipp->ipp_hopopts, optptr, ipp->ipp_hopoptslen);
5157 optptr += ipp->ipp_hopoptslen;
5158 ASSERT(OK_32PTR(optptr));
5159 /* Save as last value */
5160 ip_savebuf((void **)&tcp->tcp_hopopts, &tcp->tcp_hopoptslen,
5161 (ipp->ipp_fields & IPPF_HOPOPTS),
5162 ipp->ipp_hopopts, ipp->ipp_hopoptslen);
5163 }
5164 if (addflag.crb_ipv6_recvrthdrdstopts) {
5165 toh = (struct T_opthdr *)optptr;
5166 toh->level = IPPROTO_IPV6;
5167 toh->name = IPV6_RTHDRDSTOPTS;
5168 toh->len = sizeof (*toh) + ipp->ipp_rthdrdstoptslen;
5169 toh->status = 0;
5170 optptr += sizeof (*toh);
5171 bcopy(ipp->ipp_rthdrdstopts, optptr, ipp->ipp_rthdrdstoptslen);
5172 optptr += ipp->ipp_rthdrdstoptslen;
5173 ASSERT(OK_32PTR(optptr));
5174 /* Save as last value */
5175 ip_savebuf((void **)&tcp->tcp_rthdrdstopts,
5176 &tcp->tcp_rthdrdstoptslen,
5177 (ipp->ipp_fields & IPPF_RTHDRDSTOPTS),
5178 ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
5179 }
5180 if (addflag.crb_ipv6_recvrthdr) {
5181 toh = (struct T_opthdr *)optptr;
5182 toh->level = IPPROTO_IPV6;
5183 toh->name = IPV6_RTHDR;
5184 toh->len = sizeof (*toh) + ipp->ipp_rthdrlen;
5185 toh->status = 0;
5186 optptr += sizeof (*toh);
5187 bcopy(ipp->ipp_rthdr, optptr, ipp->ipp_rthdrlen);
5188 optptr += ipp->ipp_rthdrlen;
5189 ASSERT(OK_32PTR(optptr));
5190 /* Save as last value */
5191 ip_savebuf((void **)&tcp->tcp_rthdr, &tcp->tcp_rthdrlen,
5192 (ipp->ipp_fields & IPPF_RTHDR),
5193 ipp->ipp_rthdr, ipp->ipp_rthdrlen);
5194 }
5195 if (addflag.crb_ipv6_recvdstopts) {
5196 toh = (struct T_opthdr *)optptr;
5197 toh->level = IPPROTO_IPV6;
5198 toh->name = IPV6_DSTOPTS;
5199 toh->len = sizeof (*toh) + ipp->ipp_dstoptslen;
5200 toh->status = 0;
5201 optptr += sizeof (*toh);
5202 bcopy(ipp->ipp_dstopts, optptr, ipp->ipp_dstoptslen);
5203 optptr += ipp->ipp_dstoptslen;
5204 ASSERT(OK_32PTR(optptr));
5205 /* Save as last value */
5206 ip_savebuf((void **)&tcp->tcp_dstopts, &tcp->tcp_dstoptslen,
5207 (ipp->ipp_fields & IPPF_DSTOPTS),
5208 ipp->ipp_dstopts, ipp->ipp_dstoptslen);
5209 }
5210 ASSERT(optptr == mp->b_wptr);
5211 return (mp);
5212 }
5213
5214 /* The minimum of smoothed mean deviation in RTO calculation (nsec). */
5215 #define TCP_SD_MIN 400000000
5216
5217 /*
5218 * Set RTO for this connection based on a new round-trip time measurement.
5219 * The formula is from Jacobson and Karels' "Congestion Avoidance and Control"
5220 * in SIGCOMM '88. The variable names are the same as those in Appendix A.2
5221 * of that paper.
5222 *
5223 * m = new measurement
5224 * sa = smoothed RTT average (8 * average estimates).
5225 * sv = smoothed mean deviation (mdev) of RTT (4 * deviation estimates).
5226 */
5227 static void
5228 tcp_set_rto(tcp_t *tcp, hrtime_t rtt)
5229 {
5230 hrtime_t m = rtt;
5231 hrtime_t sa = tcp->tcp_rtt_sa;
5232 hrtime_t sv = tcp->tcp_rtt_sd;
5233 tcp_stack_t *tcps = tcp->tcp_tcps;
5234
5235 TCPS_BUMP_MIB(tcps, tcpRttUpdate);
5236 tcp->tcp_rtt_update++;
5237 tcp->tcp_rtt_sum += m;
5238 tcp->tcp_rtt_cnt++;
5239
5240 /* tcp_rtt_sa is not 0 means this is a new sample. */
5241 if (sa != 0) {
5242 /*
5243 * Update average estimator (see section 2.3 of RFC6298):
5244 * SRTT = 7/8 SRTT + 1/8 rtt
5245 *
5246 * We maintain tcp_rtt_sa as 8 * SRTT, so this reduces to:
5247 * tcp_rtt_sa = 7 * SRTT + rtt
5248 * tcp_rtt_sa = 7 * (tcp_rtt_sa / 8) + rtt
5249 * tcp_rtt_sa = tcp_rtt_sa - (tcp_rtt_sa / 8) + rtt
5250 * tcp_rtt_sa = tcp_rtt_sa + (rtt - (tcp_rtt_sa / 8))
5251 * tcp_rtt_sa = tcp_rtt_sa + (rtt - (tcp_rtt_sa / 2^3))
5252 * tcp_rtt_sa = tcp_rtt_sa + (rtt - (tcp_rtt_sa >> 3))
5253 *
5254 * (rtt - tcp_rtt_sa / 8) is simply the difference
5255 * between the new rtt measurement and the existing smoothed
5256 * RTT average. This is referred to as "Error" in subsequent
5257 * calculations.
5258 */
5259
5260 /* m is now Error. */
5261 m -= sa >> 3;
5262 if ((sa += m) <= 0) {
5263 /*
5264 * Don't allow the smoothed average to be negative.
5265 * We use 0 to denote reinitialization of the
5266 * variables.
5267 */
5268 sa = 1;
5269 }
5270
5271 /*
5272 * Update deviation estimator:
5273 * mdev = 3/4 mdev + 1/4 abs(Error)
5274 *
5275 * We maintain tcp_rtt_sd as 4 * mdev, so this reduces to:
5276 * tcp_rtt_sd = 3 * mdev + abs(Error)
5277 * tcp_rtt_sd = tcp_rtt_sd - (tcp_rtt_sd / 4) + abs(Error)
5278 * tcp_rtt_sd = tcp_rtt_sd - (tcp_rtt_sd / 2^2) + abs(Error)
5279 * tcp_rtt_sd = tcp_rtt_sd - (tcp_rtt_sd >> 2) + abs(Error)
5280 */
5281 if (m < 0)
5282 m = -m;
5283 m -= sv >> 2;
5284 sv += m;
5285 } else {
5286 /*
5287 * This follows BSD's implementation. So the reinitialized
5288 * RTO is 3 * m. We cannot go less than 2 because if the
5289 * link is bandwidth dominated, doubling the window size
5290 * during slow start means doubling the RTT. We want to be
5291 * more conservative when we reinitialize our estimates. 3
5292 * is just a convenient number.
5293 */
5294 sa = m << 3;
5295 sv = m << 1;
5296 }
5297 if (sv < TCP_SD_MIN) {
5298 /*
5299 * Since a receiver doesn't delay its ACKs during a long run of
5300 * segments, sa may not have captured the effect of delayed ACK
5301 * timeouts on the RTT. To make sure we always account for the
5302 * possible delay (and avoid the unnecessary retransmission),
5303 * TCP_SD_MIN is set to 400ms, twice the delayed ACK timeout of
5304 * 200ms on older SunOS/BSD systems and modern Windows systems
5305 * (as of 2019). This means that the minimum possible mean
5306 * deviation is 100 ms.
5307 */
5308 sv = TCP_SD_MIN;
5309 }
5310 tcp->tcp_rtt_sa = sa;
5311 tcp->tcp_rtt_sd = sv;
5312
5313 tcp->tcp_rto = tcp_calculate_rto(tcp, tcps, 0);
5314
5315 /* Now, we can reset tcp_timer_backoff to use the new RTO... */
5316 tcp->tcp_timer_backoff = 0;
5317 }
5318
5319 /*
5320 * On a labeled system we have some protocols above TCP, such as RPC, which
5321 * appear to assume that every mblk in a chain has a db_credp.
5322 */
5323 static void
5324 tcp_setcred_data(mblk_t *mp, ip_recv_attr_t *ira)
5325 {
5326 ASSERT(is_system_labeled());
5327 ASSERT(ira->ira_cred != NULL);
5328
5329 while (mp != NULL) {
5330 mblk_setcred(mp, ira->ira_cred, NOPID);
5331 mp = mp->b_cont;
5332 }
5333 }
5334
5335 uint_t
5336 tcp_rwnd_reopen(tcp_t *tcp)
5337 {
5338 uint_t ret = 0;
5339 uint_t thwin;
5340 conn_t *connp = tcp->tcp_connp;
5341
5342 /* Learn the latest rwnd information that we sent to the other side. */
5343 thwin = ((uint_t)ntohs(tcp->tcp_tcpha->tha_win))
5344 << tcp->tcp_rcv_ws;
5345 /* This is peer's calculated send window (our receive window). */
5346 thwin -= tcp->tcp_rnxt - tcp->tcp_rack;
5347 /*
5348 * Increase the receive window to max. But we need to do receiver
5349 * SWS avoidance. This means that we need to check the increase of
5350 * of receive window is at least 1 MSS.
5351 */
5352 if (connp->conn_rcvbuf - thwin >= tcp->tcp_mss) {
5353 /*
5354 * If the window that the other side knows is less than max
5355 * deferred acks segments, send an update immediately.
5356 */
5357 if (thwin < tcp->tcp_rack_cur_max * tcp->tcp_mss) {
5358 TCPS_BUMP_MIB(tcp->tcp_tcps, tcpOutWinUpdate);
5359 ret = TH_ACK_NEEDED;
5360 }
5361 tcp->tcp_rwnd = connp->conn_rcvbuf;
5362 }
5363 return (ret);
5364 }
5365
5366 /*
5367 * Handle a packet that has been reclassified by TCP.
5368 * This function drops the ref on connp that the caller had.
5369 */
5370 void
5371 tcp_reinput(conn_t *connp, mblk_t *mp, ip_recv_attr_t *ira, ip_stack_t *ipst)
5372 {
5373 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
5374
5375 if (connp->conn_incoming_ifindex != 0 &&
5376 connp->conn_incoming_ifindex != ira->ira_ruifindex) {
5377 freemsg(mp);
5378 CONN_DEC_REF(connp);
5379 return;
5380 }
5381
5382 if (CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss) ||
5383 (ira->ira_flags & IRAF_IPSEC_SECURE)) {
5384 ip6_t *ip6h;
5385 ipha_t *ipha;
5386
5387 if (ira->ira_flags & IRAF_IS_IPV4) {
5388 ipha = (ipha_t *)mp->b_rptr;
5389 ip6h = NULL;
5390 } else {
5391 ipha = NULL;
5392 ip6h = (ip6_t *)mp->b_rptr;
5393 }
5394 mp = ipsec_check_inbound_policy(mp, connp, ipha, ip6h, ira);
5395 if (mp == NULL) {
5396 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards);
5397 /* Note that mp is NULL */
5398 ip_drop_input("ipIfStatsInDiscards", mp, NULL);
5399 CONN_DEC_REF(connp);
5400 return;
5401 }
5402 }
5403
5404 if (IPCL_IS_TCP(connp)) {
5405 /*
5406 * do not drain, certain use cases can blow
5407 * the stack
5408 */
5409 SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
5410 connp->conn_recv, connp, ira,
5411 SQ_NODRAIN, SQTAG_IP_TCP_INPUT);
5412 } else {
5413 /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
5414 (connp->conn_recv)(connp, mp, NULL,
5415 ira);
5416 CONN_DEC_REF(connp);
5417 }
5418
5419 }
5420
5421 /* ARGSUSED */
5422 static void
5423 tcp_rsrv_input(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
5424 {
5425 conn_t *connp = (conn_t *)arg;
5426 tcp_t *tcp = connp->conn_tcp;
5427 queue_t *q = connp->conn_rq;
5428
5429 ASSERT(!IPCL_IS_NONSTR(connp));
5430 mutex_enter(&tcp->tcp_rsrv_mp_lock);
5431 tcp->tcp_rsrv_mp = mp;
5432 mutex_exit(&tcp->tcp_rsrv_mp_lock);
5433
5434 if (TCP_IS_DETACHED(tcp) || q == NULL) {
5435 return;
5436 }
5437
5438 if (tcp->tcp_fused) {
5439 tcp_fuse_backenable(tcp);
5440 return;
5441 }
5442
5443 if (canputnext(q)) {
5444 /* Not flow-controlled, open rwnd */
5445 tcp->tcp_rwnd = connp->conn_rcvbuf;
5446
5447 /*
5448 * Send back a window update immediately if TCP is above
5449 * ESTABLISHED state and the increase of the rcv window
5450 * that the other side knows is at least 1 MSS after flow
5451 * control is lifted.
5452 */
5453 if (tcp->tcp_state >= TCPS_ESTABLISHED &&
5454 tcp_rwnd_reopen(tcp) == TH_ACK_NEEDED) {
5455 tcp_xmit_ctl(NULL, tcp,
5456 (tcp->tcp_swnd == 0) ? tcp->tcp_suna :
5457 tcp->tcp_snxt, tcp->tcp_rnxt, TH_ACK);
5458 }
5459 }
5460 }
5461
5462 /*
5463 * The read side service routine is called mostly when we get back-enabled as a
5464 * result of flow control relief. Since we don't actually queue anything in
5465 * TCP, we have no data to send out of here. What we do is clear the receive
5466 * window, and send out a window update.
5467 */
5468 int
5469 tcp_rsrv(queue_t *q)
5470 {
5471 conn_t *connp = Q_TO_CONN(q);
5472 tcp_t *tcp = connp->conn_tcp;
5473 mblk_t *mp;
5474
5475 /* No code does a putq on the read side */
5476 ASSERT(q->q_first == NULL);
5477
5478 /*
5479 * If tcp->tcp_rsrv_mp == NULL, it means that tcp_rsrv() has already
5480 * been run. So just return.
5481 */
5482 mutex_enter(&tcp->tcp_rsrv_mp_lock);
5483 if ((mp = tcp->tcp_rsrv_mp) == NULL) {
5484 mutex_exit(&tcp->tcp_rsrv_mp_lock);
5485 return (0);
5486 }
5487 tcp->tcp_rsrv_mp = NULL;
5488 mutex_exit(&tcp->tcp_rsrv_mp_lock);
5489
5490 CONN_INC_REF(connp);
5491 SQUEUE_ENTER_ONE(connp->conn_sqp, mp, tcp_rsrv_input, connp,
5492 NULL, SQ_PROCESS, SQTAG_TCP_RSRV);
5493 return (0);
5494 }
5495
5496 /* At minimum we need 8 bytes in the TCP header for the lookup */
5497 #define ICMP_MIN_TCP_HDR 8
5498
5499 /*
5500 * tcp_icmp_input is called as conn_recvicmp to process ICMP error messages
5501 * passed up by IP. The message is always received on the correct tcp_t.
5502 * Assumes that IP has pulled up everything up to and including the ICMP header.
5503 */
5504 /* ARGSUSED2 */
5505 void
5506 tcp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
5507 {
5508 conn_t *connp = (conn_t *)arg1;
5509 icmph_t *icmph;
5510 ipha_t *ipha;
5511 int iph_hdr_length;
5512 tcpha_t *tcpha;
5513 uint32_t seg_seq;
5514 tcp_t *tcp = connp->conn_tcp;
5515
5516 /* Assume IP provides aligned packets */
5517 ASSERT(OK_32PTR(mp->b_rptr));
5518 ASSERT((MBLKL(mp) >= sizeof (ipha_t)));
5519
5520 /*
5521 * It's possible we have a closed, but not yet destroyed, TCP
5522 * connection. Several fields (e.g. conn_ixa->ixa_ire) are invalid
5523 * in the closed state, so don't take any chances and drop the packet.
5524 */
5525 if (tcp->tcp_state == TCPS_CLOSED) {
5526 freemsg(mp);
5527 return;
5528 }
5529
5530 /*
5531 * Verify IP version. Anything other than IPv4 or IPv6 packet is sent
5532 * upstream. ICMPv6 is handled in tcp_icmp_error_ipv6.
5533 */
5534 if (!(ira->ira_flags & IRAF_IS_IPV4)) {
5535 tcp_icmp_error_ipv6(tcp, mp, ira);
5536 return;
5537 }
5538
5539 /* Skip past the outer IP and ICMP headers */
5540 iph_hdr_length = ira->ira_ip_hdr_length;
5541 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
5542 /*
5543 * If we don't have the correct outer IP header length
5544 * or if we don't have a complete inner IP header
5545 * drop it.
5546 */
5547 if (iph_hdr_length < sizeof (ipha_t) ||
5548 (ipha_t *)&icmph[1] + 1 > (ipha_t *)mp->b_wptr) {
5549 noticmpv4:
5550 freemsg(mp);
5551 return;
5552 }
5553 ipha = (ipha_t *)&icmph[1];
5554
5555 /* Skip past the inner IP and find the ULP header */
5556 iph_hdr_length = IPH_HDR_LENGTH(ipha);
5557 tcpha = (tcpha_t *)((char *)ipha + iph_hdr_length);
5558 /*
5559 * If we don't have the correct inner IP header length or if the ULP
5560 * is not IPPROTO_TCP or if we don't have at least ICMP_MIN_TCP_HDR
5561 * bytes of TCP header, drop it.
5562 */
5563 if (iph_hdr_length < sizeof (ipha_t) ||
5564 ipha->ipha_protocol != IPPROTO_TCP ||
5565 (uchar_t *)tcpha + ICMP_MIN_TCP_HDR > mp->b_wptr) {
5566 goto noticmpv4;
5567 }
5568
5569 seg_seq = ntohl(tcpha->tha_seq);
5570 switch (icmph->icmph_type) {
5571 case ICMP_DEST_UNREACHABLE:
5572 switch (icmph->icmph_code) {
5573 case ICMP_FRAGMENTATION_NEEDED:
5574 /*
5575 * Update Path MTU, then try to send something out.
5576 */
5577 tcp_update_pmtu(tcp, B_TRUE);
5578 tcp_rexmit_after_error(tcp);
5579 break;
5580 case ICMP_PORT_UNREACHABLE:
5581 case ICMP_PROTOCOL_UNREACHABLE:
5582 switch (tcp->tcp_state) {
5583 case TCPS_SYN_SENT:
5584 case TCPS_SYN_RCVD:
5585 /*
5586 * ICMP can snipe away incipient
5587 * TCP connections as long as
5588 * seq number is same as initial
5589 * send seq number.
5590 */
5591 if (seg_seq == tcp->tcp_iss) {
5592 (void) tcp_clean_death(tcp,
5593 ECONNREFUSED);
5594 }
5595 break;
5596 }
5597 break;
5598 case ICMP_HOST_UNREACHABLE:
5599 case ICMP_NET_UNREACHABLE:
5600 /* Record the error in case we finally time out. */
5601 if (icmph->icmph_code == ICMP_HOST_UNREACHABLE)
5602 tcp->tcp_client_errno = EHOSTUNREACH;
5603 else
5604 tcp->tcp_client_errno = ENETUNREACH;
5605 if (tcp->tcp_state == TCPS_SYN_RCVD) {
5606 if (tcp->tcp_listener != NULL &&
5607 tcp->tcp_listener->tcp_syn_defense) {
5608 /*
5609 * Ditch the half-open connection if we
5610 * suspect a SYN attack is under way.
5611 */
5612 (void) tcp_clean_death(tcp,
5613 tcp->tcp_client_errno);
5614 }
5615 }
5616 break;
5617 default:
5618 break;
5619 }
5620 break;
5621 case ICMP_SOURCE_QUENCH: {
5622 /*
5623 * use a global boolean to control
5624 * whether TCP should respond to ICMP_SOURCE_QUENCH.
5625 * The default is false.
5626 */
5627 if (tcp_icmp_source_quench) {
5628 /*
5629 * Reduce the sending rate as if we got a
5630 * retransmit timeout
5631 */
5632 uint32_t npkt;
5633
5634 npkt = ((tcp->tcp_snxt - tcp->tcp_suna) >> 1) /
5635 tcp->tcp_mss;
5636 tcp->tcp_cwnd_ssthresh = MAX(npkt, 2) * tcp->tcp_mss;
5637 tcp->tcp_cwnd = tcp->tcp_mss;
5638 tcp->tcp_cwnd_cnt = 0;
5639 }
5640 break;
5641 }
5642 }
5643 freemsg(mp);
5644 }
5645
5646 /*
5647 * tcp_icmp_error_ipv6 is called from tcp_icmp_input to process ICMPv6
5648 * error messages passed up by IP.
5649 * Assumes that IP has pulled up all the extension headers as well
5650 * as the ICMPv6 header.
5651 */
5652 static void
5653 tcp_icmp_error_ipv6(tcp_t *tcp, mblk_t *mp, ip_recv_attr_t *ira)
5654 {
5655 icmp6_t *icmp6;
5656 ip6_t *ip6h;
5657 uint16_t iph_hdr_length = ira->ira_ip_hdr_length;
5658 tcpha_t *tcpha;
5659 uint8_t *nexthdrp;
5660 uint32_t seg_seq;
5661
5662 /*
5663 * Verify that we have a complete IP header.
5664 */
5665 ASSERT((MBLKL(mp) >= sizeof (ip6_t)));
5666
5667 icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
5668 ip6h = (ip6_t *)&icmp6[1];
5669 /*
5670 * Verify if we have a complete ICMP and inner IP header.
5671 */
5672 if ((uchar_t *)&ip6h[1] > mp->b_wptr) {
5673 noticmpv6:
5674 freemsg(mp);
5675 return;
5676 }
5677
5678 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp))
5679 goto noticmpv6;
5680 tcpha = (tcpha_t *)((char *)ip6h + iph_hdr_length);
5681 /*
5682 * Validate inner header. If the ULP is not IPPROTO_TCP or if we don't
5683 * have at least ICMP_MIN_TCP_HDR bytes of TCP header drop the
5684 * packet.
5685 */
5686 if ((*nexthdrp != IPPROTO_TCP) ||
5687 ((uchar_t *)tcpha + ICMP_MIN_TCP_HDR) > mp->b_wptr) {
5688 goto noticmpv6;
5689 }
5690
5691 seg_seq = ntohl(tcpha->tha_seq);
5692 switch (icmp6->icmp6_type) {
5693 case ICMP6_PACKET_TOO_BIG:
5694 /*
5695 * Update Path MTU, then try to send something out.
5696 */
5697 tcp_update_pmtu(tcp, B_TRUE);
5698 tcp_rexmit_after_error(tcp);
5699 break;
5700 case ICMP6_DST_UNREACH:
5701 switch (icmp6->icmp6_code) {
5702 case ICMP6_DST_UNREACH_NOPORT:
5703 if (((tcp->tcp_state == TCPS_SYN_SENT) ||
5704 (tcp->tcp_state == TCPS_SYN_RCVD)) &&
5705 (seg_seq == tcp->tcp_iss)) {
5706 (void) tcp_clean_death(tcp, ECONNREFUSED);
5707 }
5708 break;
5709 case ICMP6_DST_UNREACH_ADMIN:
5710 case ICMP6_DST_UNREACH_NOROUTE:
5711 case ICMP6_DST_UNREACH_BEYONDSCOPE:
5712 case ICMP6_DST_UNREACH_ADDR:
5713 /* Record the error in case we finally time out. */
5714 tcp->tcp_client_errno = EHOSTUNREACH;
5715 if (((tcp->tcp_state == TCPS_SYN_SENT) ||
5716 (tcp->tcp_state == TCPS_SYN_RCVD)) &&
5717 (seg_seq == tcp->tcp_iss)) {
5718 if (tcp->tcp_listener != NULL &&
5719 tcp->tcp_listener->tcp_syn_defense) {
5720 /*
5721 * Ditch the half-open connection if we
5722 * suspect a SYN attack is under way.
5723 */
5724 (void) tcp_clean_death(tcp,
5725 tcp->tcp_client_errno);
5726 }
5727 }
5728
5729
5730 break;
5731 default:
5732 break;
5733 }
5734 break;
5735 case ICMP6_PARAM_PROB:
5736 /* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
5737 if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
5738 (uchar_t *)ip6h + icmp6->icmp6_pptr ==
5739 (uchar_t *)nexthdrp) {
5740 if (tcp->tcp_state == TCPS_SYN_SENT ||
5741 tcp->tcp_state == TCPS_SYN_RCVD) {
5742 (void) tcp_clean_death(tcp, ECONNREFUSED);
5743 }
5744 break;
5745 }
5746 break;
5747
5748 case ICMP6_TIME_EXCEEDED:
5749 default:
5750 break;
5751 }
5752 freemsg(mp);
5753 }
5754
5755 /*
5756 * CALLED OUTSIDE OF SQUEUE! It can not follow any pointers that tcp might
5757 * change. But it can refer to fields like tcp_suna and tcp_snxt.
5758 *
5759 * Function tcp_verifyicmp is called as conn_verifyicmp to verify the ICMP
5760 * error messages received by IP. The message is always received on the correct
5761 * tcp_t.
5762 */
5763 /* ARGSUSED */
5764 boolean_t
5765 tcp_verifyicmp(conn_t *connp, void *arg2, icmph_t *icmph, icmp6_t *icmp6,
5766 ip_recv_attr_t *ira)
5767 {
5768 tcpha_t *tcpha = (tcpha_t *)arg2;
5769 uint32_t seq = ntohl(tcpha->tha_seq);
5770 tcp_t *tcp = connp->conn_tcp;
5771
5772 /*
5773 * TCP sequence number contained in payload of the ICMP error message
5774 * should be within the range SND.UNA <= SEG.SEQ < SND.NXT. Otherwise,
5775 * the message is either a stale ICMP error, or an attack from the
5776 * network. Fail the verification.
5777 */
5778 if (SEQ_LT(seq, tcp->tcp_suna) || SEQ_GEQ(seq, tcp->tcp_snxt))
5779 return (B_FALSE);
5780
5781 /* For "too big" we also check the ignore flag */
5782 if (ira->ira_flags & IRAF_IS_IPV4) {
5783 ASSERT(icmph != NULL);
5784 if (icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
5785 icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED &&
5786 tcp->tcp_tcps->tcps_ignore_path_mtu)
5787 return (B_FALSE);
5788 } else {
5789 ASSERT(icmp6 != NULL);
5790 if (icmp6->icmp6_type == ICMP6_PACKET_TOO_BIG &&
5791 tcp->tcp_tcps->tcps_ignore_path_mtu)
5792 return (B_FALSE);
5793 }
5794 return (B_TRUE);
5795 }