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


65a171fed3df0d3e01c65efda38d85265d40217a
[apple/xnu.git] / bsd / netinet / tcp_subr.c
1 /*
2 * Copyright (c) 2000-2015 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
30 * The Regents of the University of California. All rights reserved.
31 *
32 * Redistribution and use in source and binary forms, with or without
33 * modification, are permitted provided that the following conditions
34 * are met:
35 * 1. Redistributions of source code must retain the above copyright
36 * notice, this list of conditions and the following disclaimer.
37 * 2. Redistributions in binary form must reproduce the above copyright
38 * notice, this list of conditions and the following disclaimer in the
39 * documentation and/or other materials provided with the distribution.
40 * 3. All advertising materials mentioning features or use of this software
41 * must display the following acknowledgement:
42 * This product includes software developed by the University of
43 * California, Berkeley and its contributors.
44 * 4. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
61 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.22 2001/08/22 00:59:12 silby Exp $
62 */
63 /*
64 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
65 * support for mandatory and extensible security protections. This notice
66 * is included in support of clause 2.2 (b) of the Apple Public License,
67 * Version 2.0.
68 */
69
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/callout.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/domain.h>
78 #include <sys/proc.h>
79 #include <sys/kauth.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/protosw.h>
83 #include <sys/random.h>
84 #include <sys/syslog.h>
85 #include <sys/mcache.h>
86 #include <kern/locks.h>
87 #include <kern/zalloc.h>
88
89 #include <dev/random/randomdev.h>
90
91 #include <net/route.h>
92 #include <net/if.h>
93 #include <net/content_filter.h>
94
95 #define tcp_minmssoverload fring
96 #define _IP_VHL
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_icmp.h>
101 #if INET6
102 #include <netinet/ip6.h>
103 #endif
104 #include <netinet/in_pcb.h>
105 #if INET6
106 #include <netinet6/in6_pcb.h>
107 #endif
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
110 #include <netinet/icmp_var.h>
111 #if INET6
112 #include <netinet6/ip6_var.h>
113 #endif
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_var.h>
119 #include <netinet/tcp_cc.h>
120 #include <netinet/tcp_cache.h>
121 #include <kern/thread_call.h>
122
123 #if INET6
124 #include <netinet6/tcp6_var.h>
125 #endif
126 #include <netinet/tcpip.h>
127 #if TCPDEBUG
128 #include <netinet/tcp_debug.h>
129 #endif
130 #include <netinet6/ip6protosw.h>
131
132 #if IPSEC
133 #include <netinet6/ipsec.h>
134 #if INET6
135 #include <netinet6/ipsec6.h>
136 #endif
137 #endif /*IPSEC*/
138
139 #if NECP
140 #include <net/necp.h>
141 #endif /* NECP */
142
143 #undef tcp_minmssoverload
144
145 #if CONFIG_MACF_NET
146 #include <security/mac_framework.h>
147 #endif /* MAC_NET */
148
149 #include <corecrypto/ccaes.h>
150 #include <libkern/crypto/aes.h>
151 #include <libkern/crypto/md5.h>
152 #include <sys/kdebug.h>
153 #include <mach/sdt.h>
154
155 #include <netinet/lro_ext.h>
156
157 #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2))
158
159 extern int tcp_lq_overflow;
160
161 extern struct tcptimerlist tcp_timer_list;
162 extern struct tcptailq tcp_tw_tailq;
163
164 int tcp_mssdflt = TCP_MSS;
165 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED,
166 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
167
168 #if INET6
169 int tcp_v6mssdflt = TCP6_MSS;
170 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
171 CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_v6mssdflt , 0,
172 "Default TCP Maximum Segment Size for IPv6");
173 #endif
174
175 extern int tcp_do_autorcvbuf;
176
177 int tcp_sysctl_fastopenkey(struct sysctl_oid *, void *, int ,
178 struct sysctl_req *);
179 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, fastopen_key,
180 CTLTYPE_STRING | CTLFLAG_WR,
181 0 , 0, tcp_sysctl_fastopenkey, "S", "TCP Fastopen key");
182
183 /* Current count of half-open TFO connections */
184 int tcp_tfo_halfcnt = 0;
185
186 /* Maximum of half-open TFO connection backlog */
187 int tcp_tfo_backlog = 10;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_backlog, CTLFLAG_RW | CTLFLAG_LOCKED,
189 &tcp_tfo_backlog, 0, "Backlog queue for half-open TFO connections");
190
191 int tcp_fastopen = TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED,
193 &tcp_fastopen, 0, "Enable TCP Fastopen (RFC 7413)");
194
195 int tcp_tfo_fallback_min = 10;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, fastopen_fallback_min, CTLFLAG_RW | CTLFLAG_LOCKED,
197 &tcp_tfo_fallback_min, 0, "Mininum number of trials without TFO when in fallback mode");
198
199 /*
200 * Minimum MSS we accept and use. This prevents DoS attacks where
201 * we are forced to a ridiculous low MSS like 20 and send hundreds
202 * of packets instead of one. The effect scales with the available
203 * bandwidth and quickly saturates the CPU and network interface
204 * with packet generation and sending. Set to zero to disable MINMSS
205 * checking. This setting prevents us from sending too small packets.
206 */
207 int tcp_minmss = TCP_MINMSS;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED,
209 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
210 int tcp_do_rfc1323 = 1;
211 #if (DEVELOPMENT || DEBUG)
212 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323,
213 CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_do_rfc1323 , 0,
214 "Enable rfc1323 (high performance TCP) extensions");
215 #endif /* (DEVELOPMENT || DEBUG) */
216
217 // Not used
218 static int tcp_do_rfc1644 = 0;
219 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW | CTLFLAG_LOCKED,
220 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
221
222 static int do_tcpdrain = 0;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW | CTLFLAG_LOCKED, &do_tcpdrain, 0,
224 "Enable tcp_drain routine for extra help when low on mbufs");
225
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
227 &tcbinfo.ipi_count, 0, "Number of active PCBs");
228
229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount,
230 CTLFLAG_RD | CTLFLAG_LOCKED,
231 &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state");
232
233 static int icmp_may_rst = 1;
234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED, &icmp_may_rst, 0,
235 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
236
237 static int tcp_strict_rfc1948 = 0;
238 static int tcp_isn_reseed_interval = 0;
239 #if (DEVELOPMENT || DEBUG)
240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948,
241 CTLFLAG_RW | CTLFLAG_LOCKED,
242 &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly");
243
244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval,
245 CTLFLAG_RW | CTLFLAG_LOCKED,
246 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
247 #endif /* (DEVELOPMENT || DEBUG) */
248
249 int tcp_TCPTV_MIN = 100; /* 100ms minimum RTT */
250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED,
251 &tcp_TCPTV_MIN, 0, "min rtt value allowed");
252
253 int tcp_rexmt_slop = TCPTV_REXMTSLOP;
254 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmt_slop, CTLFLAG_RW,
255 &tcp_rexmt_slop, 0, "Slop added to retransmit timeout");
256
257 __private_extern__ int tcp_use_randomport = 0;
258 SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED,
259 &tcp_use_randomport, 0, "Randomize TCP port numbers");
260
261 __private_extern__ int tcp_win_scale = 3;
262 SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor,
263 CTLFLAG_RW | CTLFLAG_LOCKED,
264 &tcp_win_scale, 0, "Window scaling factor");
265
266 static void tcp_cleartaocache(void);
267 static void tcp_notify(struct inpcb *, int);
268
269 struct zone *sack_hole_zone;
270 struct zone *tcp_reass_zone;
271 struct zone *tcp_bwmeas_zone;
272 struct zone *tcp_rxt_seg_zone;
273
274 extern int slowlink_wsize; /* window correction for slow links */
275 extern int path_mtu_discovery;
276
277 extern u_int32_t tcp_autorcvbuf_max;
278 extern u_int32_t tcp_autorcvbuf_inc_shift;
279 static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb);
280
281 #define TCP_BWMEAS_BURST_MINSIZE 6
282 #define TCP_BWMEAS_BURST_MAXSIZE 25
283
284 static uint32_t bwmeas_elm_size;
285
286 /*
287 * Target size of TCP PCB hash tables. Must be a power of two.
288 *
289 * Note that this can be overridden by the kernel environment
290 * variable net.inet.tcp.tcbhashsize
291 */
292 #ifndef TCBHASHSIZE
293 #define TCBHASHSIZE CONFIG_TCBHASHSIZE
294 #endif
295
296 __private_extern__ int tcp_tcbhashsize = TCBHASHSIZE;
297 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED,
298 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
299
300 /*
301 * This is the actual shape of what we allocate using the zone
302 * allocator. Doing it this way allows us to protect both structures
303 * using the same generation count, and also eliminates the overhead
304 * of allocating tcpcbs separately. By hiding the structure here,
305 * we avoid changing most of the rest of the code (although it needs
306 * to be changed, eventually, for greater efficiency).
307 */
308 #define ALIGNMENT 32
309 struct inp_tp {
310 struct inpcb inp;
311 struct tcpcb tcb __attribute__((aligned(ALIGNMENT)));
312 };
313 #undef ALIGNMENT
314
315 int get_inpcb_str_size(void);
316 int get_tcp_str_size(void);
317
318 static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *);
319
320 static lck_attr_t *tcp_uptime_mtx_attr = NULL; /* mutex attributes */
321 static lck_grp_t *tcp_uptime_mtx_grp = NULL; /* mutex group definition */
322 static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL; /* mutex group attributes */
323 int tcp_notsent_lowat_check(struct socket *so);
324
325 static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */
326
327 void
328 tcp_tfo_gen_cookie(struct inpcb *inp, u_char *out, size_t blk_size)
329 {
330 u_char in[CCAES_BLOCK_SIZE];
331 #if INET6
332 int isipv6 = inp->inp_vflag & INP_IPV6;
333 #endif
334
335 VERIFY(blk_size == CCAES_BLOCK_SIZE);
336
337 bzero(&in[0], CCAES_BLOCK_SIZE);
338 bzero(&out[0], CCAES_BLOCK_SIZE);
339
340 #if INET6
341 if (isipv6)
342 memcpy(in, &inp->in6p_faddr, sizeof(struct in6_addr));
343 else
344 #endif /* INET6 */
345 memcpy(in, &inp->inp_faddr, sizeof(struct in_addr));
346
347 aes_encrypt_cbc(in, NULL, 1, out, &tfo_ctx);
348 }
349
350 __private_extern__ int
351 tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1,
352 __unused int arg2, struct sysctl_req *req)
353 {
354 int error = 0;
355 /* TFO-key is expressed as a string in hex format (+1 to account for \0 char) */
356 char keystring[TCP_FASTOPEN_KEYLEN * 2 + 1];
357 u_int32_t key[TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)];
358 int i;
359
360 /* -1, because newlen is len without the terminating \0 character */
361 if (req->newlen != (sizeof(keystring) - 1)) {
362 error = EINVAL;
363 goto exit;
364 }
365
366 /* sysctl_io_string copies keystring into the oldptr of the sysctl_req.
367 * Make sure everything is zero, to avoid putting garbage in there or
368 * leaking the stack.
369 */
370 bzero(keystring, sizeof(keystring));
371
372 error = sysctl_io_string(req, keystring, sizeof(keystring), 0, NULL);
373 if (error)
374 goto exit;
375
376 for (i = 0; i < (TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)); i++) {
377 /* We jump over the keystring in 8-character (4 byte in hex) steps */
378 if (sscanf(&keystring[i * 8], "%8x", &key[i]) != 1) {
379 error = EINVAL;
380 goto exit;
381 }
382 }
383
384 aes_encrypt_key128((u_char *)key, &tfo_ctx);
385
386 exit:
387 return (error);
388 }
389
390 int get_inpcb_str_size(void)
391 {
392 return sizeof(struct inpcb);
393 }
394
395 int get_tcp_str_size(void)
396 {
397 return sizeof(struct tcpcb);
398 }
399
400 int tcp_freeq(struct tcpcb *tp);
401
402 static int scale_to_powerof2(int size);
403
404 /*
405 * This helper routine returns one of the following scaled value of size:
406 * 1. Rounded down power of two value of size if the size value passed as
407 * argument is not a power of two and the rounded up value overflows.
408 * OR
409 * 2. Rounded up power of two value of size if the size value passed as
410 * argument is not a power of two and the rounded up value does not overflow
411 * OR
412 * 3. Same value as argument size if it is already a power of two.
413 */
414 static int scale_to_powerof2(int size) {
415 /* Handle special case of size = 0 */
416 int ret = size ? size : 1;
417
418 if (!powerof2(ret)) {
419 while(!powerof2(size)) {
420 /*
421 * Clear out least significant
422 * set bit till size is left with
423 * its highest set bit at which point
424 * it is rounded down power of two.
425 */
426 size = size & (size -1);
427 }
428
429 /* Check for overflow when rounding up */
430 if (0 == (size << 1)) {
431 ret = size;
432 } else {
433 ret = size << 1;
434 }
435 }
436
437 return ret;
438 }
439
440 static void
441 tcp_tfo_init()
442 {
443 u_char key[TCP_FASTOPEN_KEYLEN];
444
445 read_random(key, sizeof(key));
446 aes_encrypt_key128(key, &tfo_ctx);
447 }
448
449 /*
450 * Tcp initialization
451 */
452 void
453 tcp_init(struct protosw *pp, struct domain *dp)
454 {
455 #pragma unused(dp)
456 static int tcp_initialized = 0;
457 vm_size_t str_size;
458 struct inpcbinfo *pcbinfo;
459
460 VERIFY((pp->pr_flags & (PR_INITIALIZED|PR_ATTACHED)) == PR_ATTACHED);
461
462 if (tcp_initialized)
463 return;
464 tcp_initialized = 1;
465
466 tcp_ccgen = 1;
467 tcp_cleartaocache();
468
469 tcp_keepinit = TCPTV_KEEP_INIT;
470 tcp_keepidle = TCPTV_KEEP_IDLE;
471 tcp_keepintvl = TCPTV_KEEPINTVL;
472 tcp_keepcnt = TCPTV_KEEPCNT;
473 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
474 tcp_msl = TCPTV_MSL;
475
476 microuptime(&tcp_uptime);
477 read_random(&tcp_now, sizeof(tcp_now));
478 tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */
479
480 tcp_tfo_init();
481
482 LIST_INIT(&tcb);
483 tcbinfo.ipi_listhead = &tcb;
484
485 pcbinfo = &tcbinfo;
486 /*
487 * allocate lock group attribute and group for tcp pcb mutexes
488 */
489 pcbinfo->ipi_lock_grp_attr = lck_grp_attr_alloc_init();
490 pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb", pcbinfo->ipi_lock_grp_attr);
491
492 /*
493 * allocate the lock attribute for tcp pcb mutexes
494 */
495 pcbinfo->ipi_lock_attr = lck_attr_alloc_init();
496
497 if ((pcbinfo->ipi_lock = lck_rw_alloc_init(pcbinfo->ipi_lock_grp,
498 pcbinfo->ipi_lock_attr)) == NULL) {
499 panic("%s: unable to allocate PCB lock\n", __func__);
500 /* NOTREACHED */
501 }
502
503 if (tcp_tcbhashsize == 0) {
504 /* Set to default */
505 tcp_tcbhashsize = 512;
506 }
507
508 if (!powerof2(tcp_tcbhashsize)) {
509 int old_hash_size = tcp_tcbhashsize;
510 tcp_tcbhashsize = scale_to_powerof2(tcp_tcbhashsize);
511 /* Lower limit of 16 */
512 if (tcp_tcbhashsize < 16) {
513 tcp_tcbhashsize = 16;
514 }
515 printf("WARNING: TCB hash size not a power of 2, "
516 "scaled from %d to %d.\n",
517 old_hash_size,
518 tcp_tcbhashsize);
519 }
520
521 tcbinfo.ipi_hashbase = hashinit(tcp_tcbhashsize, M_PCB, &tcbinfo.ipi_hashmask);
522 tcbinfo.ipi_porthashbase = hashinit(tcp_tcbhashsize, M_PCB,
523 &tcbinfo.ipi_porthashmask);
524 str_size = P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t));
525 tcbinfo.ipi_zone = zinit(str_size, 120000*str_size, 8192, "tcpcb");
526 zone_change(tcbinfo.ipi_zone, Z_CALLERACCT, FALSE);
527 zone_change(tcbinfo.ipi_zone, Z_EXPAND, TRUE);
528
529 tcbinfo.ipi_gc = tcp_gc;
530 tcbinfo.ipi_timer = tcp_itimer;
531 in_pcbinfo_attach(&tcbinfo);
532
533 str_size = P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t));
534 sack_hole_zone = zinit(str_size, 120000*str_size, 8192, "sack_hole zone");
535 zone_change(sack_hole_zone, Z_CALLERACCT, FALSE);
536 zone_change(sack_hole_zone, Z_EXPAND, TRUE);
537
538 str_size = P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t));
539 tcp_reass_zone = zinit(str_size, (nmbclusters >> 4) * str_size,
540 0, "tcp_reass_zone");
541 if (tcp_reass_zone == NULL) {
542 panic("%s: failed allocating tcp_reass_zone", __func__);
543 /* NOTREACHED */
544 }
545 zone_change(tcp_reass_zone, Z_CALLERACCT, FALSE);
546 zone_change(tcp_reass_zone, Z_EXPAND, TRUE);
547
548 bwmeas_elm_size = P2ROUNDUP(sizeof(struct bwmeas), sizeof(u_int64_t));
549 tcp_bwmeas_zone = zinit(bwmeas_elm_size, (100 * bwmeas_elm_size), 0, "tcp_bwmeas_zone");
550 if (tcp_bwmeas_zone == NULL) {
551 panic("%s: failed allocating tcp_bwmeas_zone", __func__);
552 /* NOTREACHED */
553 }
554 zone_change(tcp_bwmeas_zone, Z_CALLERACCT, FALSE);
555 zone_change(tcp_bwmeas_zone, Z_EXPAND, TRUE);
556
557 str_size = P2ROUNDUP(sizeof(struct tcp_ccstate), sizeof(u_int64_t));
558 tcp_cc_zone = zinit(str_size, 20000 * str_size, 0, "tcp_cc_zone");
559 zone_change(tcp_cc_zone, Z_CALLERACCT, FALSE);
560 zone_change(tcp_cc_zone, Z_EXPAND, TRUE);
561
562 str_size = P2ROUNDUP(sizeof(struct tcp_rxt_seg), sizeof(u_int64_t));
563 tcp_rxt_seg_zone = zinit(str_size, 10000 * str_size, 0,
564 "tcp_rxt_seg_zone");
565 zone_change(tcp_rxt_seg_zone, Z_CALLERACCT, FALSE);
566 zone_change(tcp_rxt_seg_zone, Z_EXPAND, TRUE);
567
568 #if INET6
569 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
570 #else /* INET6 */
571 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
572 #endif /* INET6 */
573 if (max_protohdr < TCP_MINPROTOHDR) {
574 _max_protohdr = TCP_MINPROTOHDR;
575 _max_protohdr = max_protohdr; /* round it up */
576 }
577 if (max_linkhdr + max_protohdr > MCLBYTES)
578 panic("tcp_init");
579 #undef TCP_MINPROTOHDR
580
581 /* Initialize time wait and timer lists */
582 TAILQ_INIT(&tcp_tw_tailq);
583
584 bzero(&tcp_timer_list, sizeof(tcp_timer_list));
585 LIST_INIT(&tcp_timer_list.lhead);
586 /*
587 * allocate lock group attribute, group and attribute for the tcp timer list
588 */
589 tcp_timer_list.mtx_grp_attr = lck_grp_attr_alloc_init();
590 tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist", tcp_timer_list.mtx_grp_attr);
591 tcp_timer_list.mtx_attr = lck_attr_alloc_init();
592 if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp, tcp_timer_list.mtx_attr)) == NULL) {
593 panic("failed to allocate memory for tcp_timer_list.mtx\n");
594 };
595 if ((tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL)) == NULL) {
596 panic("failed to allocate call entry 1 in tcp_init\n");
597 }
598
599 /*
600 * allocate lock group attribute, group and attribute for tcp_uptime_lock
601 */
602 tcp_uptime_mtx_grp_attr = lck_grp_attr_alloc_init();
603 tcp_uptime_mtx_grp = lck_grp_alloc_init("tcpuptime", tcp_uptime_mtx_grp_attr);
604 tcp_uptime_mtx_attr = lck_attr_alloc_init();
605 tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, tcp_uptime_mtx_attr);
606
607 /* Initialize TCP LRO data structures */
608 tcp_lro_init();
609
610 /* Initialize TCP Cache */
611 tcp_cache_init();
612
613 /*
614 * If more than 60 MB of mbuf pool is available, increase the
615 * maximum allowed receive and send socket buffer size.
616 */
617 if (nmbclusters > 30720) {
618 tcp_autorcvbuf_max = 1024 * 1024;
619 tcp_autosndbuf_max = 1024 * 1024;
620 }
621 }
622
623 /*
624 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
625 * tcp_template used to store this data in mbufs, but we now recopy it out
626 * of the tcpcb each time to conserve mbufs.
627 */
628 void
629 tcp_fillheaders(tp, ip_ptr, tcp_ptr)
630 struct tcpcb *tp;
631 void *ip_ptr;
632 void *tcp_ptr;
633 {
634 struct inpcb *inp = tp->t_inpcb;
635 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
636
637 #if INET6
638 if ((inp->inp_vflag & INP_IPV6) != 0) {
639 struct ip6_hdr *ip6;
640
641 ip6 = (struct ip6_hdr *)ip_ptr;
642 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
643 (inp->inp_flow & IPV6_FLOWINFO_MASK);
644 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
645 (IPV6_VERSION & IPV6_VERSION_MASK);
646 ip6->ip6_nxt = IPPROTO_TCP;
647 ip6->ip6_plen = sizeof(struct tcphdr);
648 ip6->ip6_src = inp->in6p_laddr;
649 ip6->ip6_dst = inp->in6p_faddr;
650 tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr,
651 htonl(sizeof (struct tcphdr) + IPPROTO_TCP));
652 } else
653 #endif
654 {
655 struct ip *ip = (struct ip *) ip_ptr;
656
657 ip->ip_vhl = IP_VHL_BORING;
658 ip->ip_tos = 0;
659 ip->ip_len = 0;
660 ip->ip_id = 0;
661 ip->ip_off = 0;
662 ip->ip_ttl = 0;
663 ip->ip_sum = 0;
664 ip->ip_p = IPPROTO_TCP;
665 ip->ip_src = inp->inp_laddr;
666 ip->ip_dst = inp->inp_faddr;
667 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
668 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
669 }
670
671 tcp_hdr->th_sport = inp->inp_lport;
672 tcp_hdr->th_dport = inp->inp_fport;
673 tcp_hdr->th_seq = 0;
674 tcp_hdr->th_ack = 0;
675 tcp_hdr->th_x2 = 0;
676 tcp_hdr->th_off = 5;
677 tcp_hdr->th_flags = 0;
678 tcp_hdr->th_win = 0;
679 tcp_hdr->th_urp = 0;
680 }
681
682 /*
683 * Create template to be used to send tcp packets on a connection.
684 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
685 * use for this function is in keepalives, which use tcp_respond.
686 */
687 struct tcptemp *
688 tcp_maketemplate(tp)
689 struct tcpcb *tp;
690 {
691 struct mbuf *m;
692 struct tcptemp *n;
693
694 m = m_get(M_DONTWAIT, MT_HEADER);
695 if (m == NULL)
696 return (0);
697 m->m_len = sizeof(struct tcptemp);
698 n = mtod(m, struct tcptemp *);
699
700 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
701 return (n);
702 }
703
704 /*
705 * Send a single message to the TCP at address specified by
706 * the given TCP/IP header. If m == 0, then we make a copy
707 * of the tcpiphdr at ti and send directly to the addressed host.
708 * This is used to force keep alive messages out using the TCP
709 * template for a connection. If flags are given then we send
710 * a message back to the TCP which originated the * segment ti,
711 * and discard the mbuf containing it and any other attached mbufs.
712 *
713 * In any case the ack and sequence number of the transmitted
714 * segment are as specified by the parameters.
715 *
716 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
717 */
718 void
719 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
720 tcp_seq ack, tcp_seq seq, int flags, struct tcp_respond_args *tra)
721 {
722 int tlen;
723 int win = 0;
724 struct route *ro = 0;
725 struct route sro;
726 struct ip *ip;
727 struct tcphdr *nth;
728 #if INET6
729 struct route_in6 *ro6 = 0;
730 struct route_in6 sro6;
731 struct ip6_hdr *ip6;
732 int isipv6;
733 #endif /* INET6 */
734 struct ifnet *outif;
735
736 #if INET6
737 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
738 ip6 = ipgen;
739 #endif /* INET6 */
740 ip = ipgen;
741
742 if (tp) {
743 if (!(flags & TH_RST)) {
744 win = tcp_sbspace(tp);
745 if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale)
746 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
747 }
748 #if INET6
749 if (isipv6)
750 ro6 = &tp->t_inpcb->in6p_route;
751 else
752 #endif /* INET6 */
753 ro = &tp->t_inpcb->inp_route;
754 } else {
755 #if INET6
756 if (isipv6) {
757 ro6 = &sro6;
758 bzero(ro6, sizeof *ro6);
759 } else
760 #endif /* INET6 */
761 {
762 ro = &sro;
763 bzero(ro, sizeof *ro);
764 }
765 }
766 if (m == 0) {
767 m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */
768 if (m == NULL)
769 return;
770 tlen = 0;
771 m->m_data += max_linkhdr;
772 #if INET6
773 if (isipv6) {
774 VERIFY((MHLEN - max_linkhdr) >=
775 (sizeof (*ip6) + sizeof (*nth)));
776 bcopy((caddr_t)ip6, mtod(m, caddr_t),
777 sizeof(struct ip6_hdr));
778 ip6 = mtod(m, struct ip6_hdr *);
779 nth = (struct tcphdr *)(void *)(ip6 + 1);
780 } else
781 #endif /* INET6 */
782 {
783 VERIFY((MHLEN - max_linkhdr) >=
784 (sizeof (*ip) + sizeof (*nth)));
785 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
786 ip = mtod(m, struct ip *);
787 nth = (struct tcphdr *)(void *)(ip + 1);
788 }
789 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
790 #if MPTCP
791 if ((tp) && (tp->t_mpflags & TMPF_RESET))
792 flags = (TH_RST | TH_ACK);
793 else
794 #endif
795 flags = TH_ACK;
796 } else {
797 m_freem(m->m_next);
798 m->m_next = 0;
799 m->m_data = (caddr_t)ipgen;
800 /* m_len is set later */
801 tlen = 0;
802 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
803 #if INET6
804 if (isipv6) {
805 /* Expect 32-bit aligned IP on strict-align platforms */
806 IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6);
807 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
808 nth = (struct tcphdr *)(void *)(ip6 + 1);
809 } else
810 #endif /* INET6 */
811 {
812 /* Expect 32-bit aligned IP on strict-align platforms */
813 IP_HDR_STRICT_ALIGNMENT_CHECK(ip);
814 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
815 nth = (struct tcphdr *)(void *)(ip + 1);
816 }
817 if (th != nth) {
818 /*
819 * this is usually a case when an extension header
820 * exists between the IPv6 header and the
821 * TCP header.
822 */
823 nth->th_sport = th->th_sport;
824 nth->th_dport = th->th_dport;
825 }
826 xchg(nth->th_dport, nth->th_sport, n_short);
827 #undef xchg
828 }
829 #if INET6
830 if (isipv6) {
831 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
832 tlen));
833 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
834 } else
835 #endif
836 {
837 tlen += sizeof (struct tcpiphdr);
838 ip->ip_len = tlen;
839 ip->ip_ttl = ip_defttl;
840 }
841 m->m_len = tlen;
842 m->m_pkthdr.len = tlen;
843 m->m_pkthdr.rcvif = 0;
844 #if CONFIG_MACF_NET
845 if (tp != NULL && tp->t_inpcb != NULL) {
846 /*
847 * Packet is associated with a socket, so allow the
848 * label of the response to reflect the socket label.
849 */
850 mac_mbuf_label_associate_inpcb(tp->t_inpcb, m);
851 } else {
852 /*
853 * Packet is not associated with a socket, so possibly
854 * update the label in place.
855 */
856 mac_netinet_tcp_reply(m);
857 }
858 #endif
859
860 nth->th_seq = htonl(seq);
861 nth->th_ack = htonl(ack);
862 nth->th_x2 = 0;
863 nth->th_off = sizeof (struct tcphdr) >> 2;
864 nth->th_flags = flags;
865 if (tp)
866 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
867 else
868 nth->th_win = htons((u_short)win);
869 nth->th_urp = 0;
870 #if INET6
871 if (isipv6) {
872 nth->th_sum = 0;
873 nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst,
874 htonl((tlen - sizeof (struct ip6_hdr)) + IPPROTO_TCP));
875 m->m_pkthdr.csum_flags = CSUM_TCPIPV6;
876 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
877 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
878 ro6 && ro6->ro_rt ?
879 ro6->ro_rt->rt_ifp :
880 NULL);
881 } else
882 #endif /* INET6 */
883 {
884 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
885 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
886 m->m_pkthdr.csum_flags = CSUM_TCP;
887 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
888 }
889 #if TCPDEBUG
890 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
891 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
892 #endif
893
894 #if NECP
895 necp_mark_packet_from_socket(m, tp ? tp->t_inpcb : NULL, 0, 0);
896 #endif /* NECP */
897
898 #if IPSEC
899 if (tp != NULL && tp->t_inpcb->inp_sp != NULL &&
900 ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
901 m_freem(m);
902 return;
903 }
904 #endif
905
906 if (tp != NULL) {
907 u_int32_t svc_flags = 0;
908 if (isipv6) {
909 svc_flags |= PKT_SCF_IPV6;
910 }
911 set_packet_service_class(m, tp->t_inpcb->inp_socket,
912 MBUF_SC_UNSPEC, svc_flags);
913
914 /* Embed flowhash and flow control flags */
915 m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB;
916 m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash;
917 m->m_pkthdr.pkt_flags |= PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC;
918 #if MPTCP
919 /* Disable flow advisory when using MPTCP. */
920 if (!(tp->t_mpflags & TMPF_MPTCP_TRUE))
921 #endif /* MPTCP */
922 m->m_pkthdr.pkt_flags |= PKTF_FLOW_ADV;
923 m->m_pkthdr.pkt_proto = IPPROTO_TCP;
924 }
925
926 #if INET6
927 if (isipv6) {
928 struct ip6_out_args ip6oa = { tra->ifscope, { 0 },
929 IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR, 0 };
930
931 if (tra->ifscope != IFSCOPE_NONE)
932 ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF;
933 if (tra->nocell)
934 ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR;
935 if (tra->noexpensive)
936 ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE;
937 if (tra->awdl_unrestricted)
938 ip6oa.ip6oa_flags |= IP6OAF_AWDL_UNRESTRICTED;
939
940 (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL,
941 NULL, &ip6oa);
942
943 if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL &&
944 (outif = ro6->ro_rt->rt_ifp) !=
945 tp->t_inpcb->in6p_last_outifp)
946 tp->t_inpcb->in6p_last_outifp = outif;
947
948 if (ro6 == &sro6)
949 ROUTE_RELEASE(ro6);
950 } else
951 #endif /* INET6 */
952 {
953 struct ip_out_args ipoa = { tra->ifscope, { 0 },
954 IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR, 0 };
955
956 if (tra->ifscope != IFSCOPE_NONE)
957 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
958 if (tra->nocell)
959 ipoa.ipoa_flags |= IPOAF_NO_CELLULAR;
960 if (tra->noexpensive)
961 ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE;
962 if (tra->awdl_unrestricted)
963 ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED;
964
965 if (ro != &sro) {
966 /* Copy the cached route and take an extra reference */
967 inp_route_copyout(tp->t_inpcb, &sro);
968 }
969 /*
970 * For consistency, pass a local route copy.
971 */
972 (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa);
973
974 if (tp != NULL && sro.ro_rt != NULL &&
975 (outif = sro.ro_rt->rt_ifp) !=
976 tp->t_inpcb->inp_last_outifp)
977 tp->t_inpcb->inp_last_outifp = outif;
978
979 if (ro != &sro) {
980 /* Synchronize cached PCB route */
981 inp_route_copyin(tp->t_inpcb, &sro);
982 } else {
983 ROUTE_RELEASE(&sro);
984 }
985 }
986 }
987
988 /*
989 * Create a new TCP control block, making an
990 * empty reassembly queue and hooking it to the argument
991 * protocol control block. The `inp' parameter must have
992 * come from the zone allocator set up in tcp_init().
993 */
994 struct tcpcb *
995 tcp_newtcpcb(inp)
996 struct inpcb *inp;
997 {
998 struct inp_tp *it;
999 register struct tcpcb *tp;
1000 register struct socket *so = inp->inp_socket;
1001 #if INET6
1002 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1003 #endif /* INET6 */
1004
1005 calculate_tcp_clock();
1006
1007 if ((so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) == 0) {
1008 it = (struct inp_tp *)(void *)inp;
1009 tp = &it->tcb;
1010 } else {
1011 tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb;
1012 }
1013
1014 bzero((char *) tp, sizeof(struct tcpcb));
1015 LIST_INIT(&tp->t_segq);
1016 tp->t_maxseg = tp->t_maxopd =
1017 #if INET6
1018 isipv6 ? tcp_v6mssdflt :
1019 #endif /* INET6 */
1020 tcp_mssdflt;
1021
1022 if (tcp_do_rfc1323)
1023 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
1024 if (tcp_do_sack)
1025 tp->t_flagsext |= TF_SACK_ENABLE;
1026
1027 TAILQ_INIT(&tp->snd_holes);
1028 SLIST_INIT(&tp->t_rxt_segments);
1029 tp->t_inpcb = inp; /* XXX */
1030 /*
1031 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1032 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1033 * reasonable initial retransmit time.
1034 */
1035 tp->t_srtt = TCPTV_SRTTBASE;
1036 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1037 tp->t_rttmin = tcp_TCPTV_MIN;
1038 tp->t_rxtcur = TCPTV_RTOBASE;
1039
1040 if (tcp_use_newreno)
1041 /* use newreno by default */
1042 tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX;
1043 else
1044 tp->tcp_cc_index = TCP_CC_ALGO_CUBIC_INDEX;
1045
1046 tcp_cc_allocate_state(tp);
1047
1048 if (CC_ALGO(tp)->init != NULL)
1049 CC_ALGO(tp)->init(tp);
1050
1051 tp->snd_cwnd = TCP_CC_CWND_INIT_BYTES;
1052 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1053 tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1054 tp->t_rcvtime = tcp_now;
1055 tp->tentry.timer_start = tcp_now;
1056 tp->t_persist_timeout = tcp_max_persist_timeout;
1057 tp->t_persist_stop = 0;
1058 tp->t_flagsext |= TF_RCVUNACK_WAITSS;
1059 tp->t_rexmtthresh = tcprexmtthresh;
1060
1061 /* Clear time wait tailq entry */
1062 tp->t_twentry.tqe_next = NULL;
1063 tp->t_twentry.tqe_prev = NULL;
1064
1065 /*
1066 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1067 * because the socket may be bound to an IPv6 wildcard address,
1068 * which may match an IPv4-mapped IPv6 address.
1069 */
1070 inp->inp_ip_ttl = ip_defttl;
1071 inp->inp_ppcb = (caddr_t)tp;
1072 return (tp); /* XXX */
1073 }
1074
1075 /*
1076 * Drop a TCP connection, reporting
1077 * the specified error. If connection is synchronized,
1078 * then send a RST to peer.
1079 */
1080 struct tcpcb *
1081 tcp_drop(tp, errno)
1082 register struct tcpcb *tp;
1083 int errno;
1084 {
1085 struct socket *so = tp->t_inpcb->inp_socket;
1086 #if CONFIG_DTRACE
1087 struct inpcb *inp = tp->t_inpcb;
1088 #endif
1089
1090 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1091 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1092 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1093 tp->t_state = TCPS_CLOSED;
1094 (void) tcp_output(tp);
1095 tcpstat.tcps_drops++;
1096 } else
1097 tcpstat.tcps_conndrops++;
1098 if (errno == ETIMEDOUT && tp->t_softerror)
1099 errno = tp->t_softerror;
1100 so->so_error = errno;
1101 return (tcp_close(tp));
1102 }
1103
1104 void
1105 tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt)
1106 {
1107 u_int32_t rtt = rt->rt_rmx.rmx_rtt;
1108 int isnetlocal = (tp->t_flags & TF_LOCAL);
1109
1110 if (rtt != 0) {
1111 /*
1112 * XXX the lock bit for RTT indicates that the value
1113 * is also a minimum value; this is subject to time.
1114 */
1115 if (rt->rt_rmx.rmx_locks & RTV_RTT)
1116 tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ);
1117 else
1118 tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN;
1119 tp->t_srtt = rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1120 tcpstat.tcps_usedrtt++;
1121 if (rt->rt_rmx.rmx_rttvar) {
1122 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1123 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1124 tcpstat.tcps_usedrttvar++;
1125 } else {
1126 /* default variation is +- 1 rtt */
1127 tp->t_rttvar =
1128 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1129 }
1130 TCPT_RANGESET(tp->t_rxtcur,
1131 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
1132 tp->t_rttmin, TCPTV_REXMTMAX,
1133 TCP_ADD_REXMTSLOP(tp));
1134 }
1135 }
1136
1137 static inline void
1138 tcp_update_ecn_perf_stats(struct tcpcb *tp,
1139 struct if_tcp_ecn_perf_stat *stat)
1140 {
1141 u_int64_t curval, oldval;
1142 struct inpcb *inp = tp->t_inpcb;
1143
1144 /* Average RTT */
1145 curval = (tp->t_srtt >> TCP_RTT_SHIFT);
1146 if (curval > 0 && tp->t_rttupdated >= 16) {
1147 if (stat->rtt_avg == 0) {
1148 stat->rtt_avg = curval;
1149 } else {
1150 oldval = stat->rtt_avg;
1151 stat->rtt_avg =
1152 ((oldval << 4) - oldval + curval) >> 4;
1153 }
1154 }
1155
1156 /* RTT variance */
1157 curval = tp->t_rttvar >> TCP_RTTVAR_SHIFT;
1158 if (curval > 0 && tp->t_rttupdated >= 16) {
1159 if (stat->rtt_var == 0) {
1160 stat->rtt_var = curval;
1161 } else {
1162 oldval = stat->rtt_var;
1163 stat->rtt_var =
1164 ((oldval << 4) - oldval + curval) >> 4;
1165 }
1166 }
1167
1168 /* Percentage of Out-of-order packets, shift by 10 for precision */
1169 curval = (tp->t_rcvoopack << 10);
1170 if (inp->inp_stat != NULL && inp->inp_stat->rxpackets > 0 &&
1171 curval > 0) {
1172 /* Compute percentage */
1173 curval = (curval * 100)/inp->inp_stat->rxpackets;
1174 if (stat->oo_percent == 0) {
1175 stat->oo_percent = curval;
1176 } else {
1177 oldval = stat->oo_percent;
1178 stat->oo_percent =
1179 ((oldval << 4) - oldval + curval) >> 4;
1180 }
1181 }
1182
1183 /* Total number of SACK recovery episodes */
1184 stat->sack_episodes += tp->t_sack_recovery_episode;
1185
1186 /* Percentage of reordered packets, shift by 10 for precision */
1187 curval = tp->t_reordered_pkts + tp->t_pawsdrop + tp->t_dsack_sent +
1188 tp->t_dsack_recvd;
1189 curval = curval << 10;
1190 if (inp->inp_stat != NULL && (inp->inp_stat->rxpackets > 0 ||
1191 inp->inp_stat->txpackets > 0) && curval > 0) {
1192 /* Compute percentage */
1193 curval = (curval * 100) /
1194 (inp->inp_stat->rxpackets + inp->inp_stat->txpackets);
1195 if (stat->reorder_percent == 0) {
1196 stat->reorder_percent = curval;
1197 } else {
1198 oldval = stat->reorder_percent;
1199 stat->reorder_percent =
1200 ((oldval << 4) - oldval + curval) >> 4;
1201 }
1202 }
1203
1204 /* Percentage of retransmit bytes, shift by 10 for precision */
1205 curval = tp->t_stat.txretransmitbytes << 10;
1206 if (inp->inp_stat != NULL && inp->inp_stat->txbytes > 0
1207 && curval > 0) {
1208 curval = (curval * 100) / inp->inp_stat->txbytes;
1209 if (stat->rxmit_percent == 0) {
1210 stat->rxmit_percent = curval;
1211 } else {
1212 oldval = stat->rxmit_percent;
1213 stat->rxmit_percent =
1214 ((oldval << 4) - oldval + curval) >> 4;
1215 }
1216 }
1217 return;
1218 }
1219
1220 /*
1221 * Close a TCP control block:
1222 * discard all space held by the tcp
1223 * discard internet protocol block
1224 * wake up any sleepers
1225 */
1226 struct tcpcb *
1227 tcp_close(tp)
1228 register struct tcpcb *tp;
1229 {
1230 struct inpcb *inp = tp->t_inpcb;
1231 struct socket *so = inp->inp_socket;
1232 #if INET6
1233 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1234 #endif /* INET6 */
1235 struct route *ro;
1236 struct rtentry *rt;
1237 int dosavessthresh;
1238
1239 /* tcp_close was called previously, bail */
1240 if (inp->inp_ppcb == NULL)
1241 return(NULL);
1242
1243 tcp_canceltimers(tp);
1244 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0);
1245
1246 /*
1247 * If another thread for this tcp is currently in ip (indicated by
1248 * the TF_SENDINPROG flag), defer the cleanup until after it returns
1249 * back to tcp. This is done to serialize the close until after all
1250 * pending output is finished, in order to avoid having the PCB be
1251 * detached and the cached route cleaned, only for ip to cache the
1252 * route back into the PCB again. Note that we've cleared all the
1253 * timers at this point. Set TF_CLOSING to indicate to tcp_output()
1254 * that is should call us again once it returns from ip; at that
1255 * point both flags should be cleared and we can proceed further
1256 * with the cleanup.
1257 */
1258 if ((tp->t_flags & TF_CLOSING) ||
1259 inp->inp_sndinprog_cnt > 0) {
1260 tp->t_flags |= TF_CLOSING;
1261 return (NULL);
1262 }
1263
1264 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1265 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1266
1267 #if INET6
1268 ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route);
1269 #else
1270 ro = &inp->inp_route;
1271 #endif
1272 rt = ro->ro_rt;
1273 if (rt != NULL)
1274 RT_LOCK_SPIN(rt);
1275
1276 /*
1277 * If we got enough samples through the srtt filter,
1278 * save the rtt and rttvar in the routing entry.
1279 * 'Enough' is arbitrarily defined as the 16 samples.
1280 * 16 samples is enough for the srtt filter to converge
1281 * to within 5% of the correct value; fewer samples and
1282 * we could save a very bogus rtt.
1283 *
1284 * Don't update the default route's characteristics and don't
1285 * update anything that the user "locked".
1286 */
1287 if (tp->t_rttupdated >= 16) {
1288 register u_int32_t i = 0;
1289
1290 #if INET6
1291 if (isipv6) {
1292 struct sockaddr_in6 *sin6;
1293
1294 if (rt == NULL)
1295 goto no_valid_rt;
1296 sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt);
1297 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1298 goto no_valid_rt;
1299 }
1300 else
1301 #endif /* INET6 */
1302 if (ROUTE_UNUSABLE(ro) ||
1303 SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) {
1304 DTRACE_TCP4(state__change, void, NULL,
1305 struct inpcb *, inp, struct tcpcb *, tp,
1306 int32_t, TCPS_CLOSED);
1307 tp->t_state = TCPS_CLOSED;
1308 goto no_valid_rt;
1309 }
1310
1311 RT_LOCK_ASSERT_HELD(rt);
1312 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1313 i = tp->t_srtt *
1314 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1315 if (rt->rt_rmx.rmx_rtt && i)
1316 /*
1317 * filter this update to half the old & half
1318 * the new values, converting scale.
1319 * See route.h and tcp_var.h for a
1320 * description of the scaling constants.
1321 */
1322 rt->rt_rmx.rmx_rtt =
1323 (rt->rt_rmx.rmx_rtt + i) / 2;
1324 else
1325 rt->rt_rmx.rmx_rtt = i;
1326 tcpstat.tcps_cachedrtt++;
1327 }
1328 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1329 i = tp->t_rttvar *
1330 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1331 if (rt->rt_rmx.rmx_rttvar && i)
1332 rt->rt_rmx.rmx_rttvar =
1333 (rt->rt_rmx.rmx_rttvar + i) / 2;
1334 else
1335 rt->rt_rmx.rmx_rttvar = i;
1336 tcpstat.tcps_cachedrttvar++;
1337 }
1338 /*
1339 * The old comment here said:
1340 * update the pipelimit (ssthresh) if it has been updated
1341 * already or if a pipesize was specified & the threshhold
1342 * got below half the pipesize. I.e., wait for bad news
1343 * before we start updating, then update on both good
1344 * and bad news.
1345 *
1346 * But we want to save the ssthresh even if no pipesize is
1347 * specified explicitly in the route, because such
1348 * connections still have an implicit pipesize specified
1349 * by the global tcp_sendspace. In the absence of a reliable
1350 * way to calculate the pipesize, it will have to do.
1351 */
1352 i = tp->snd_ssthresh;
1353 if (rt->rt_rmx.rmx_sendpipe != 0)
1354 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
1355 else
1356 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
1357 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1358 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
1359 || dosavessthresh) {
1360 /*
1361 * convert the limit from user data bytes to
1362 * packets then to packet data bytes.
1363 */
1364 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
1365 if (i < 2)
1366 i = 2;
1367 i *= (u_int32_t)(tp->t_maxseg +
1368 #if INET6
1369 (isipv6 ? sizeof (struct ip6_hdr) +
1370 sizeof (struct tcphdr) :
1371 #endif
1372 sizeof (struct tcpiphdr)
1373 #if INET6
1374 )
1375 #endif
1376 );
1377 if (rt->rt_rmx.rmx_ssthresh)
1378 rt->rt_rmx.rmx_ssthresh =
1379 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1380 else
1381 rt->rt_rmx.rmx_ssthresh = i;
1382 tcpstat.tcps_cachedssthresh++;
1383 }
1384 }
1385
1386 /*
1387 * Mark route for deletion if no information is cached.
1388 */
1389 if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) {
1390 if (!(rt->rt_rmx.rmx_locks & RTV_RTT) &&
1391 rt->rt_rmx.rmx_rtt == 0) {
1392 rt->rt_flags |= RTF_DELCLONE;
1393 }
1394 }
1395
1396 no_valid_rt:
1397 if (rt != NULL)
1398 RT_UNLOCK(rt);
1399
1400 /* free the reassembly queue, if any */
1401 (void) tcp_freeq(tp);
1402
1403 /* Collect ECN related statistics */
1404 if (tp->ecn_flags & TE_SETUPSENT) {
1405 if (tp->ecn_flags & TE_CLIENT_SETUP) {
1406 INP_INC_IFNET_STAT(inp, ecn_client_setup);
1407 if (TCP_ECN_ENABLED(tp)) {
1408 INP_INC_IFNET_STAT(inp,
1409 ecn_client_success);
1410 } else if (tp->ecn_flags & TE_LOST_SYN) {
1411 INP_INC_IFNET_STAT(inp, ecn_syn_lost);
1412 } else {
1413 INP_INC_IFNET_STAT(inp,
1414 ecn_peer_nosupport);
1415 }
1416 } else {
1417 INP_INC_IFNET_STAT(inp, ecn_server_setup);
1418 if (TCP_ECN_ENABLED(tp)) {
1419 INP_INC_IFNET_STAT(inp,
1420 ecn_server_success);
1421 } else if (tp->ecn_flags & TE_LOST_SYNACK) {
1422 INP_INC_IFNET_STAT(inp,
1423 ecn_synack_lost);
1424 } else {
1425 INP_INC_IFNET_STAT(inp,
1426 ecn_peer_nosupport);
1427 }
1428 }
1429 }
1430 if (TCP_ECN_ENABLED(tp)) {
1431 if (tp->ecn_flags & TE_RECV_ECN_CE) {
1432 tcpstat.tcps_ecn_conn_recv_ce++;
1433 INP_INC_IFNET_STAT(inp, ecn_conn_recv_ce);
1434 }
1435 if (tp->ecn_flags & TE_RECV_ECN_ECE) {
1436 tcpstat.tcps_ecn_conn_recv_ece++;
1437 INP_INC_IFNET_STAT(inp, ecn_conn_recv_ece);
1438 }
1439 if (tp->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) {
1440 if (tp->t_stat.txretransmitbytes > 0 ||
1441 tp->t_stat.rxoutoforderbytes > 0) {
1442 tcpstat.tcps_ecn_conn_pl_ce++;
1443 INP_INC_IFNET_STAT(inp, ecn_conn_plce);
1444 } else {
1445 tcpstat.tcps_ecn_conn_nopl_ce++;
1446 INP_INC_IFNET_STAT(inp, ecn_conn_noplce);
1447 }
1448 } else {
1449 if (tp->t_stat.txretransmitbytes > 0 ||
1450 tp->t_stat.rxoutoforderbytes > 0) {
1451 tcpstat.tcps_ecn_conn_plnoce++;
1452 INP_INC_IFNET_STAT(inp, ecn_conn_plnoce);
1453 }
1454 }
1455
1456 }
1457
1458 /* Aggregate performance stats */
1459 if (inp->inp_last_outifp != NULL) {
1460 struct ifnet *ifp = inp->inp_last_outifp;
1461 ifnet_lock_shared(ifp);
1462 if ((ifp->if_refflags & (IFRF_ATTACHED | IFRF_DETACHING)) ==
1463 IFRF_ATTACHED) {
1464 if (inp->inp_vflag & INP_IPV6) {
1465 if (TCP_ECN_ENABLED(tp)) {
1466 ifp->if_ipv6_stat->timestamp
1467 = net_uptime();
1468 tcp_update_ecn_perf_stats(tp,
1469 &ifp->if_ipv6_stat->ecn_on);
1470 } else {
1471 ifp->if_ipv6_stat->timestamp
1472 = net_uptime();
1473 tcp_update_ecn_perf_stats(tp,
1474 &ifp->if_ipv6_stat->ecn_off);
1475 }
1476 } else {
1477 if (TCP_ECN_ENABLED(tp)) {
1478 ifp->if_ipv4_stat->timestamp
1479 = net_uptime();
1480 tcp_update_ecn_perf_stats(tp,
1481 &ifp->if_ipv4_stat->ecn_on);
1482 } else {
1483 ifp->if_ipv4_stat->timestamp
1484 = net_uptime();
1485 tcp_update_ecn_perf_stats(tp,
1486 &ifp->if_ipv4_stat->ecn_off);
1487 }
1488 }
1489 }
1490 ifnet_lock_done(ifp);
1491 }
1492
1493 tcp_free_sackholes(tp);
1494 if (tp->t_bwmeas != NULL) {
1495 tcp_bwmeas_free(tp);
1496 }
1497 tcp_rxtseg_clean(tp);
1498 /* Free the packet list */
1499 if (tp->t_pktlist_head != NULL)
1500 m_freem_list(tp->t_pktlist_head);
1501 TCP_PKTLIST_CLEAR(tp);
1502
1503 #if MPTCP
1504 /* Clear MPTCP state */
1505 if ((so->so_flags & SOF_MPTCP_TRUE) ||
1506 (so->so_flags & SOF_MP_SUBFLOW)) {
1507 soevent(so, (SO_FILT_HINT_LOCKED | SO_FILT_HINT_DELETEOK));
1508 }
1509 tp->t_mpflags = 0;
1510 tp->t_mptcb = NULL;
1511 #endif /* MPTCP */
1512
1513 if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER)
1514 inp->inp_saved_ppcb = (caddr_t) tp;
1515
1516 tp->t_state = TCPS_CLOSED;
1517
1518 /* Issue a wakeup before detach so that we don't miss
1519 * a wakeup
1520 */
1521 sodisconnectwakeup(so);
1522
1523 /*
1524 * Clean up any LRO state
1525 */
1526 if (tp->t_flagsext & TF_LRO_OFFLOADED) {
1527 tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr,
1528 inp->inp_lport, inp->inp_fport);
1529 tp->t_flagsext &= ~TF_LRO_OFFLOADED;
1530 }
1531
1532 /*
1533 * If this is a socket that does not want to wakeup the device
1534 * for it's traffic, the application might need to know that the
1535 * socket is closed, send a notification.
1536 */
1537 if ((so->so_options & SO_NOWAKEFROMSLEEP) &&
1538 inp->inp_state != INPCB_STATE_DEAD &&
1539 !(inp->inp_flags2 & INP2_TIMEWAIT))
1540 socket_post_kev_msg_closed(so);
1541
1542 if (CC_ALGO(tp)->cleanup != NULL) {
1543 CC_ALGO(tp)->cleanup(tp);
1544 }
1545
1546 if (tp->t_ccstate != NULL) {
1547 zfree(tcp_cc_zone, tp->t_ccstate);
1548 tp->t_ccstate = NULL;
1549 }
1550 tp->tcp_cc_index = TCP_CC_ALGO_NONE;
1551
1552 /* Can happen if we close the socket before receiving the third ACK */
1553 if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) {
1554 OSDecrementAtomic(&tcp_tfo_halfcnt);
1555
1556 /* Panic if something has gone terribly wrong. */
1557 VERIFY(tcp_tfo_halfcnt >= 0);
1558
1559 tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID;
1560 }
1561
1562 #if INET6
1563 if (SOCK_CHECK_DOM(so, PF_INET6))
1564 in6_pcbdetach(inp);
1565 else
1566 #endif /* INET6 */
1567 in_pcbdetach(inp);
1568
1569 /* Call soisdisconnected after detach because it might unlock the socket */
1570 soisdisconnected(so);
1571 tcpstat.tcps_closed++;
1572 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END,
1573 tcpstat.tcps_closed, 0, 0, 0, 0);
1574 return(NULL);
1575 }
1576
1577 int
1578 tcp_freeq(tp)
1579 struct tcpcb *tp;
1580 {
1581
1582 register struct tseg_qent *q;
1583 int rv = 0;
1584
1585 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
1586 LIST_REMOVE(q, tqe_q);
1587 m_freem(q->tqe_m);
1588 zfree(tcp_reass_zone, q);
1589 rv = 1;
1590 }
1591 tp->t_reassqlen = 0;
1592 return (rv);
1593 }
1594
1595
1596 /*
1597 * Walk the tcpbs, if existing, and flush the reassembly queue,
1598 * if there is one when do_tcpdrain is enabled
1599 * Also defunct the extended background idle socket
1600 * Do it next time if the pcbinfo lock is in use
1601 */
1602 void
1603 tcp_drain()
1604 {
1605 struct inpcb *inp;
1606 struct tcpcb *tp;
1607
1608 if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock))
1609 return;
1610
1611 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1612 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) !=
1613 WNT_STOPUSING) {
1614 tcp_lock(inp->inp_socket, 1, 0);
1615 if (in_pcb_checkstate(inp, WNT_RELEASE, 1)
1616 == WNT_STOPUSING) {
1617 /* lost a race, try the next one */
1618 tcp_unlock(inp->inp_socket, 1, 0);
1619 continue;
1620 }
1621 tp = intotcpcb(inp);
1622
1623 if (do_tcpdrain)
1624 tcp_freeq(tp);
1625
1626 so_drain_extended_bk_idle(inp->inp_socket);
1627
1628 tcp_unlock(inp->inp_socket, 1, 0);
1629 }
1630 }
1631 lck_rw_done(tcbinfo.ipi_lock);
1632
1633 }
1634
1635 /*
1636 * Notify a tcp user of an asynchronous error;
1637 * store error as soft error, but wake up user
1638 * (for now, won't do anything until can select for soft error).
1639 *
1640 * Do not wake up user since there currently is no mechanism for
1641 * reporting soft errors (yet - a kqueue filter may be added).
1642 */
1643 static void
1644 tcp_notify(inp, error)
1645 struct inpcb *inp;
1646 int error;
1647 {
1648 struct tcpcb *tp;
1649
1650 if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD))
1651 return; /* pcb is gone already */
1652
1653 tp = (struct tcpcb *)inp->inp_ppcb;
1654
1655 /*
1656 * Ignore some errors if we are hooked up.
1657 * If connection hasn't completed, has retransmitted several times,
1658 * and receives a second error, give up now. This is better
1659 * than waiting a long time to establish a connection that
1660 * can never complete.
1661 */
1662 if (tp->t_state == TCPS_ESTABLISHED &&
1663 (error == EHOSTUNREACH || error == ENETUNREACH ||
1664 error == EHOSTDOWN)) {
1665 return;
1666 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1667 tp->t_softerror)
1668 tcp_drop(tp, error);
1669 else
1670 tp->t_softerror = error;
1671 #if 0
1672 wakeup((caddr_t) &so->so_timeo);
1673 sorwakeup(so);
1674 sowwakeup(so);
1675 #endif
1676 }
1677
1678 struct bwmeas*
1679 tcp_bwmeas_alloc(struct tcpcb *tp)
1680 {
1681 struct bwmeas *elm;
1682 elm = zalloc(tcp_bwmeas_zone);
1683 if (elm == NULL)
1684 return(elm);
1685
1686 bzero(elm, bwmeas_elm_size);
1687 elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE;
1688 elm->bw_maxsizepkts = TCP_BWMEAS_BURST_MAXSIZE;
1689 elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg;
1690 elm->bw_maxsize = elm->bw_maxsizepkts * tp->t_maxseg;
1691 return(elm);
1692 }
1693
1694 void
1695 tcp_bwmeas_free(struct tcpcb* tp)
1696 {
1697 zfree(tcp_bwmeas_zone, tp->t_bwmeas);
1698 tp->t_bwmeas = NULL;
1699 tp->t_flagsext &= ~(TF_MEASURESNDBW);
1700 }
1701
1702 /*
1703 * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format.
1704 * The otcpcb data structure is passed to user space and must not change.
1705 */
1706 static void
1707 tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp)
1708 {
1709 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1710 otp->t_dupacks = tp->t_dupacks;
1711 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1712 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1713 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1714 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1715 otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb);
1716 otp->t_state = tp->t_state;
1717 otp->t_flags = tp->t_flags;
1718 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1719 otp->snd_una = tp->snd_una;
1720 otp->snd_max = tp->snd_max;
1721 otp->snd_nxt = tp->snd_nxt;
1722 otp->snd_up = tp->snd_up;
1723 otp->snd_wl1 = tp->snd_wl1;
1724 otp->snd_wl2 = tp->snd_wl2;
1725 otp->iss = tp->iss;
1726 otp->irs = tp->irs;
1727 otp->rcv_nxt = tp->rcv_nxt;
1728 otp->rcv_adv = tp->rcv_adv;
1729 otp->rcv_wnd = tp->rcv_wnd;
1730 otp->rcv_up = tp->rcv_up;
1731 otp->snd_wnd = tp->snd_wnd;
1732 otp->snd_cwnd = tp->snd_cwnd;
1733 otp->snd_ssthresh = tp->snd_ssthresh;
1734 otp->t_maxopd = tp->t_maxopd;
1735 otp->t_rcvtime = tp->t_rcvtime;
1736 otp->t_starttime = tp->t_starttime;
1737 otp->t_rtttime = tp->t_rtttime;
1738 otp->t_rtseq = tp->t_rtseq;
1739 otp->t_rxtcur = tp->t_rxtcur;
1740 otp->t_maxseg = tp->t_maxseg;
1741 otp->t_srtt = tp->t_srtt;
1742 otp->t_rttvar = tp->t_rttvar;
1743 otp->t_rxtshift = tp->t_rxtshift;
1744 otp->t_rttmin = tp->t_rttmin;
1745 otp->t_rttupdated = tp->t_rttupdated;
1746 otp->max_sndwnd = tp->max_sndwnd;
1747 otp->t_softerror = tp->t_softerror;
1748 otp->t_oobflags = tp->t_oobflags;
1749 otp->t_iobc = tp->t_iobc;
1750 otp->snd_scale = tp->snd_scale;
1751 otp->rcv_scale = tp->rcv_scale;
1752 otp->request_r_scale = tp->request_r_scale;
1753 otp->requested_s_scale = tp->requested_s_scale;
1754 otp->ts_recent = tp->ts_recent;
1755 otp->ts_recent_age = tp->ts_recent_age;
1756 otp->last_ack_sent = tp->last_ack_sent;
1757 otp->cc_send = tp->cc_send;
1758 otp->cc_recv = tp->cc_recv;
1759 otp->snd_recover = tp->snd_recover;
1760 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1761 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1762 otp->t_badrxtwin = 0;
1763 }
1764
1765 static int
1766 tcp_pcblist SYSCTL_HANDLER_ARGS
1767 {
1768 #pragma unused(oidp, arg1, arg2)
1769 int error, i = 0, n;
1770 struct inpcb *inp, **inp_list;
1771 struct tcpcb *tp;
1772 inp_gen_t gencnt;
1773 struct xinpgen xig;
1774
1775 /*
1776 * The process of preparing the TCB list is too time-consuming and
1777 * resource-intensive to repeat twice on every request.
1778 */
1779 lck_rw_lock_shared(tcbinfo.ipi_lock);
1780 if (req->oldptr == USER_ADDR_NULL) {
1781 n = tcbinfo.ipi_count;
1782 req->oldidx = 2 * (sizeof xig)
1783 + (n + n/8) * sizeof(struct xtcpcb);
1784 lck_rw_done(tcbinfo.ipi_lock);
1785 return 0;
1786 }
1787
1788 if (req->newptr != USER_ADDR_NULL) {
1789 lck_rw_done(tcbinfo.ipi_lock);
1790 return EPERM;
1791 }
1792
1793 /*
1794 * OK, now we're committed to doing something.
1795 */
1796 gencnt = tcbinfo.ipi_gencnt;
1797 n = tcbinfo.ipi_count;
1798
1799 bzero(&xig, sizeof(xig));
1800 xig.xig_len = sizeof xig;
1801 xig.xig_count = n;
1802 xig.xig_gen = gencnt;
1803 xig.xig_sogen = so_gencnt;
1804 error = SYSCTL_OUT(req, &xig, sizeof xig);
1805 if (error) {
1806 lck_rw_done(tcbinfo.ipi_lock);
1807 return error;
1808 }
1809 /*
1810 * We are done if there is no pcb
1811 */
1812 if (n == 0) {
1813 lck_rw_done(tcbinfo.ipi_lock);
1814 return 0;
1815 }
1816
1817 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1818 if (inp_list == 0) {
1819 lck_rw_done(tcbinfo.ipi_lock);
1820 return ENOMEM;
1821 }
1822
1823 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1824 if (inp->inp_gencnt <= gencnt &&
1825 inp->inp_state != INPCB_STATE_DEAD)
1826 inp_list[i++] = inp;
1827 if (i >= n) break;
1828 }
1829
1830 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
1831 inp = tp->t_inpcb;
1832 if (inp->inp_gencnt <= gencnt &&
1833 inp->inp_state != INPCB_STATE_DEAD)
1834 inp_list[i++] = inp;
1835 if (i >= n) break;
1836 }
1837
1838 n = i;
1839
1840 error = 0;
1841 for (i = 0; i < n; i++) {
1842 inp = inp_list[i];
1843 if (inp->inp_gencnt <= gencnt &&
1844 inp->inp_state != INPCB_STATE_DEAD) {
1845 struct xtcpcb xt;
1846 caddr_t inp_ppcb;
1847
1848 bzero(&xt, sizeof(xt));
1849 xt.xt_len = sizeof xt;
1850 /* XXX should avoid extra copy */
1851 inpcb_to_compat(inp, &xt.xt_inp);
1852 inp_ppcb = inp->inp_ppcb;
1853 if (inp_ppcb != NULL) {
1854 tcpcb_to_otcpcb(
1855 (struct tcpcb *)(void *)inp_ppcb,
1856 &xt.xt_tp);
1857 } else {
1858 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1859 }
1860 if (inp->inp_socket)
1861 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1862 error = SYSCTL_OUT(req, &xt, sizeof xt);
1863 }
1864 }
1865 if (!error) {
1866 /*
1867 * Give the user an updated idea of our state.
1868 * If the generation differs from what we told
1869 * her before, she knows that something happened
1870 * while we were processing this request, and it
1871 * might be necessary to retry.
1872 */
1873 bzero(&xig, sizeof(xig));
1874 xig.xig_len = sizeof xig;
1875 xig.xig_gen = tcbinfo.ipi_gencnt;
1876 xig.xig_sogen = so_gencnt;
1877 xig.xig_count = tcbinfo.ipi_count;
1878 error = SYSCTL_OUT(req, &xig, sizeof xig);
1879 }
1880 FREE(inp_list, M_TEMP);
1881 lck_rw_done(tcbinfo.ipi_lock);
1882 return error;
1883 }
1884
1885 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1886 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1887 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1888
1889
1890 static void
1891 tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp)
1892 {
1893 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1894 otp->t_dupacks = tp->t_dupacks;
1895 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1896 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1897 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1898 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1899 otp->t_state = tp->t_state;
1900 otp->t_flags = tp->t_flags;
1901 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1902 otp->snd_una = tp->snd_una;
1903 otp->snd_max = tp->snd_max;
1904 otp->snd_nxt = tp->snd_nxt;
1905 otp->snd_up = tp->snd_up;
1906 otp->snd_wl1 = tp->snd_wl1;
1907 otp->snd_wl2 = tp->snd_wl2;
1908 otp->iss = tp->iss;
1909 otp->irs = tp->irs;
1910 otp->rcv_nxt = tp->rcv_nxt;
1911 otp->rcv_adv = tp->rcv_adv;
1912 otp->rcv_wnd = tp->rcv_wnd;
1913 otp->rcv_up = tp->rcv_up;
1914 otp->snd_wnd = tp->snd_wnd;
1915 otp->snd_cwnd = tp->snd_cwnd;
1916 otp->snd_ssthresh = tp->snd_ssthresh;
1917 otp->t_maxopd = tp->t_maxopd;
1918 otp->t_rcvtime = tp->t_rcvtime;
1919 otp->t_starttime = tp->t_starttime;
1920 otp->t_rtttime = tp->t_rtttime;
1921 otp->t_rtseq = tp->t_rtseq;
1922 otp->t_rxtcur = tp->t_rxtcur;
1923 otp->t_maxseg = tp->t_maxseg;
1924 otp->t_srtt = tp->t_srtt;
1925 otp->t_rttvar = tp->t_rttvar;
1926 otp->t_rxtshift = tp->t_rxtshift;
1927 otp->t_rttmin = tp->t_rttmin;
1928 otp->t_rttupdated = tp->t_rttupdated;
1929 otp->max_sndwnd = tp->max_sndwnd;
1930 otp->t_softerror = tp->t_softerror;
1931 otp->t_oobflags = tp->t_oobflags;
1932 otp->t_iobc = tp->t_iobc;
1933 otp->snd_scale = tp->snd_scale;
1934 otp->rcv_scale = tp->rcv_scale;
1935 otp->request_r_scale = tp->request_r_scale;
1936 otp->requested_s_scale = tp->requested_s_scale;
1937 otp->ts_recent = tp->ts_recent;
1938 otp->ts_recent_age = tp->ts_recent_age;
1939 otp->last_ack_sent = tp->last_ack_sent;
1940 otp->cc_send = tp->cc_send;
1941 otp->cc_recv = tp->cc_recv;
1942 otp->snd_recover = tp->snd_recover;
1943 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1944 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1945 otp->t_badrxtwin = 0;
1946 }
1947
1948
1949 static int
1950 tcp_pcblist64 SYSCTL_HANDLER_ARGS
1951 {
1952 #pragma unused(oidp, arg1, arg2)
1953 int error, i = 0, n;
1954 struct inpcb *inp, **inp_list;
1955 struct tcpcb *tp;
1956 inp_gen_t gencnt;
1957 struct xinpgen xig;
1958
1959 /*
1960 * The process of preparing the TCB list is too time-consuming and
1961 * resource-intensive to repeat twice on every request.
1962 */
1963 lck_rw_lock_shared(tcbinfo.ipi_lock);
1964 if (req->oldptr == USER_ADDR_NULL) {
1965 n = tcbinfo.ipi_count;
1966 req->oldidx = 2 * (sizeof xig)
1967 + (n + n/8) * sizeof(struct xtcpcb64);
1968 lck_rw_done(tcbinfo.ipi_lock);
1969 return 0;
1970 }
1971
1972 if (req->newptr != USER_ADDR_NULL) {
1973 lck_rw_done(tcbinfo.ipi_lock);
1974 return EPERM;
1975 }
1976
1977 /*
1978 * OK, now we're committed to doing something.
1979 */
1980 gencnt = tcbinfo.ipi_gencnt;
1981 n = tcbinfo.ipi_count;
1982
1983 bzero(&xig, sizeof(xig));
1984 xig.xig_len = sizeof xig;
1985 xig.xig_count = n;
1986 xig.xig_gen = gencnt;
1987 xig.xig_sogen = so_gencnt;
1988 error = SYSCTL_OUT(req, &xig, sizeof xig);
1989 if (error) {
1990 lck_rw_done(tcbinfo.ipi_lock);
1991 return error;
1992 }
1993 /*
1994 * We are done if there is no pcb
1995 */
1996 if (n == 0) {
1997 lck_rw_done(tcbinfo.ipi_lock);
1998 return 0;
1999 }
2000
2001 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
2002 if (inp_list == 0) {
2003 lck_rw_done(tcbinfo.ipi_lock);
2004 return ENOMEM;
2005 }
2006
2007 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
2008 if (inp->inp_gencnt <= gencnt &&
2009 inp->inp_state != INPCB_STATE_DEAD)
2010 inp_list[i++] = inp;
2011 if (i >= n) break;
2012 }
2013
2014 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
2015 inp = tp->t_inpcb;
2016 if (inp->inp_gencnt <= gencnt &&
2017 inp->inp_state != INPCB_STATE_DEAD)
2018 inp_list[i++] = inp;
2019 if (i >= n) break;
2020 }
2021
2022 n = i;
2023
2024 error = 0;
2025 for (i = 0; i < n; i++) {
2026 inp = inp_list[i];
2027 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) {
2028 struct xtcpcb64 xt;
2029
2030 bzero(&xt, sizeof(xt));
2031 xt.xt_len = sizeof xt;
2032 inpcb_to_xinpcb64(inp, &xt.xt_inpcb);
2033 xt.xt_inpcb.inp_ppcb = (uint64_t)VM_KERNEL_ADDRPERM(inp->inp_ppcb);
2034 if (inp->inp_ppcb != NULL)
2035 tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt);
2036 if (inp->inp_socket)
2037 sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket);
2038 error = SYSCTL_OUT(req, &xt, sizeof xt);
2039 }
2040 }
2041 if (!error) {
2042 /*
2043 * Give the user an updated idea of our state.
2044 * If the generation differs from what we told
2045 * her before, she knows that something happened
2046 * while we were processing this request, and it
2047 * might be necessary to retry.
2048 */
2049 bzero(&xig, sizeof(xig));
2050 xig.xig_len = sizeof xig;
2051 xig.xig_gen = tcbinfo.ipi_gencnt;
2052 xig.xig_sogen = so_gencnt;
2053 xig.xig_count = tcbinfo.ipi_count;
2054 error = SYSCTL_OUT(req, &xig, sizeof xig);
2055 }
2056 FREE(inp_list, M_TEMP);
2057 lck_rw_done(tcbinfo.ipi_lock);
2058 return error;
2059 }
2060
2061 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64,
2062 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2063 tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections");
2064
2065
2066 static int
2067 tcp_pcblist_n SYSCTL_HANDLER_ARGS
2068 {
2069 #pragma unused(oidp, arg1, arg2)
2070 int error = 0;
2071
2072 error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo);
2073
2074 return error;
2075 }
2076
2077
2078 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n,
2079 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2080 tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections");
2081
2082
2083 __private_extern__ void
2084 tcp_get_ports_used(uint32_t ifindex, int protocol, uint32_t flags,
2085 bitstr_t *bitfield)
2086 {
2087 inpcb_get_ports_used(ifindex, protocol, flags,
2088 bitfield, &tcbinfo);
2089 }
2090
2091 __private_extern__ uint32_t
2092 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags)
2093 {
2094 return inpcb_count_opportunistic(ifindex, &tcbinfo, flags);
2095 }
2096
2097 __private_extern__ uint32_t
2098 tcp_find_anypcb_byaddr(struct ifaddr *ifa)
2099 {
2100 return inpcb_find_anypcb_byaddr(ifa, &tcbinfo);
2101 }
2102
2103 void
2104 tcp_ctlinput(cmd, sa, vip)
2105 int cmd;
2106 struct sockaddr *sa;
2107 void *vip;
2108 {
2109 tcp_seq icmp_tcp_seq;
2110 struct ip *ip = vip;
2111 struct in_addr faddr;
2112 struct inpcb *inp;
2113 struct tcpcb *tp;
2114
2115 void (*notify)(struct inpcb *, int) = tcp_notify;
2116
2117 faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr;
2118 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
2119 return;
2120
2121 if ((unsigned)cmd >= PRC_NCMDS)
2122 return;
2123
2124 if (cmd == PRC_MSGSIZE)
2125 notify = tcp_mtudisc;
2126 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2127 cmd == PRC_UNREACH_PORT) && ip)
2128 notify = tcp_drop_syn_sent;
2129 else if (PRC_IS_REDIRECT(cmd)) {
2130 ip = 0;
2131 notify = in_rtchange;
2132 } else if (cmd == PRC_HOSTDEAD)
2133 ip = 0;
2134 /* Source quench is deprecated */
2135 else if (cmd == PRC_QUENCH)
2136 return;
2137 else if (inetctlerrmap[cmd] == 0)
2138 return;
2139 if (ip) {
2140 struct tcphdr th;
2141 struct icmp *icp;
2142
2143 icp = (struct icmp *)(void *)
2144 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2145 bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)),
2146 &th, sizeof (th));
2147 inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport,
2148 ip->ip_src, th.th_sport, 0, NULL);
2149 if (inp != NULL && inp->inp_socket != NULL) {
2150 tcp_lock(inp->inp_socket, 1, 0);
2151 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
2152 tcp_unlock(inp->inp_socket, 1, 0);
2153 return;
2154 }
2155 icmp_tcp_seq = htonl(th.th_seq);
2156 tp = intotcpcb(inp);
2157 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2158 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2159 if (cmd == PRC_MSGSIZE) {
2160
2161 /*
2162 * MTU discovery:
2163 * If we got a needfrag and there is a host route to the
2164 * original destination, and the MTU is not locked, then
2165 * set the MTU in the route to the suggested new value
2166 * (if given) and then notify as usual. The ULPs will
2167 * notice that the MTU has changed and adapt accordingly.
2168 * If no new MTU was suggested, then we guess a new one
2169 * less than the current value. If the new MTU is
2170 * unreasonably small (defined by sysctl tcp_minmss), then
2171 * we reset the MTU to the interface value and enable the
2172 * lock bit, indicating that we are no longer doing MTU
2173 * discovery.
2174 */
2175 struct rtentry *rt;
2176 int mtu;
2177 struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET,
2178 0 , { 0 }, { 0,0,0,0,0,0,0,0 } };
2179 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
2180
2181 rt = rtalloc1((struct sockaddr *)&icmpsrc, 0,
2182 RTF_CLONING | RTF_PRCLONING);
2183 if (rt != NULL) {
2184 RT_LOCK(rt);
2185 if ((rt->rt_flags & RTF_HOST) &&
2186 !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
2187 mtu = ntohs(icp->icmp_nextmtu);
2188 if (!mtu)
2189 mtu = ip_next_mtu(rt->rt_rmx.
2190 rmx_mtu, 1);
2191 #if DEBUG_MTUDISC
2192 printf("MTU for %s reduced to %d\n",
2193 inet_ntop(AF_INET,
2194 &icmpsrc.sin_addr, ipv4str,
2195 sizeof (ipv4str)), mtu);
2196 #endif
2197 if (mtu < max(296, (tcp_minmss +
2198 sizeof (struct tcpiphdr)))) {
2199 /* rt->rt_rmx.rmx_mtu =
2200 rt->rt_ifp->if_mtu; */
2201 rt->rt_rmx.rmx_locks |= RTV_MTU;
2202 } else if (rt->rt_rmx.rmx_mtu > mtu) {
2203 rt->rt_rmx.rmx_mtu = mtu;
2204 }
2205 }
2206 RT_UNLOCK(rt);
2207 rtfree(rt);
2208 }
2209 }
2210
2211 (*notify)(inp, inetctlerrmap[cmd]);
2212 }
2213 tcp_unlock(inp->inp_socket, 1, 0);
2214 }
2215 } else
2216 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
2217 }
2218
2219 #if INET6
2220 void
2221 tcp6_ctlinput(cmd, sa, d)
2222 int cmd;
2223 struct sockaddr *sa;
2224 void *d;
2225 {
2226 struct tcphdr th;
2227 void (*notify)(struct inpcb *, int) = tcp_notify;
2228 struct ip6_hdr *ip6;
2229 struct mbuf *m;
2230 struct ip6ctlparam *ip6cp = NULL;
2231 const struct sockaddr_in6 *sa6_src = NULL;
2232 int off;
2233 struct tcp_portonly {
2234 u_int16_t th_sport;
2235 u_int16_t th_dport;
2236 } *thp;
2237
2238 if (sa->sa_family != AF_INET6 ||
2239 sa->sa_len != sizeof(struct sockaddr_in6))
2240 return;
2241
2242 if ((unsigned)cmd >= PRC_NCMDS)
2243 return;
2244
2245 if (cmd == PRC_MSGSIZE)
2246 notify = tcp_mtudisc;
2247 else if (!PRC_IS_REDIRECT(cmd) && (inet6ctlerrmap[cmd] == 0))
2248 return;
2249 /* Source quench is deprecated */
2250 else if (cmd == PRC_QUENCH)
2251 return;
2252
2253 /* if the parameter is from icmp6, decode it. */
2254 if (d != NULL) {
2255 ip6cp = (struct ip6ctlparam *)d;
2256 m = ip6cp->ip6c_m;
2257 ip6 = ip6cp->ip6c_ip6;
2258 off = ip6cp->ip6c_off;
2259 sa6_src = ip6cp->ip6c_src;
2260 } else {
2261 m = NULL;
2262 ip6 = NULL;
2263 off = 0; /* fool gcc */
2264 sa6_src = &sa6_any;
2265 }
2266
2267 if (ip6) {
2268 /*
2269 * XXX: We assume that when IPV6 is non NULL,
2270 * M and OFF are valid.
2271 */
2272
2273 /* check if we can safely examine src and dst ports */
2274 if (m->m_pkthdr.len < off + sizeof(*thp))
2275 return;
2276
2277 bzero(&th, sizeof(th));
2278 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2279
2280 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
2281 (struct sockaddr *)ip6cp->ip6c_src,
2282 th.th_sport, cmd, NULL, notify);
2283 } else {
2284 in6_pcbnotify(&tcbinfo, sa, 0,
2285 (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify);
2286 }
2287 }
2288 #endif /* INET6 */
2289
2290
2291 /*
2292 * Following is where TCP initial sequence number generation occurs.
2293 *
2294 * There are two places where we must use initial sequence numbers:
2295 * 1. In SYN-ACK packets.
2296 * 2. In SYN packets.
2297 *
2298 * The ISNs in SYN-ACK packets have no monotonicity requirement,
2299 * and should be as unpredictable as possible to avoid the possibility
2300 * of spoofing and/or connection hijacking. To satisfy this
2301 * requirement, SYN-ACK ISNs are generated via the arc4random()
2302 * function. If exact RFC 1948 compliance is requested via sysctl,
2303 * these ISNs will be generated just like those in SYN packets.
2304 *
2305 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2306 * depends on this property. In addition, these ISNs should be
2307 * unguessable so as to prevent connection hijacking. To satisfy
2308 * the requirements of this situation, the algorithm outlined in
2309 * RFC 1948 is used to generate sequence numbers.
2310 *
2311 * For more information on the theory of operation, please see
2312 * RFC 1948.
2313 *
2314 * Implementation details:
2315 *
2316 * Time is based off the system timer, and is corrected so that it
2317 * increases by one megabyte per second. This allows for proper
2318 * recycling on high speed LANs while still leaving over an hour
2319 * before rollover.
2320 *
2321 * Two sysctls control the generation of ISNs:
2322 *
2323 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2324 * between seeding of isn_secret. This is normally set to zero,
2325 * as reseeding should not be necessary.
2326 *
2327 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
2328 * strictly. When strict compliance is requested, reseeding is
2329 * disabled and SYN-ACKs will be generated in the same manner as
2330 * SYNs. Strict mode is disabled by default.
2331 *
2332 */
2333
2334 #define ISN_BYTES_PER_SECOND 1048576
2335
2336 tcp_seq
2337 tcp_new_isn(tp)
2338 struct tcpcb *tp;
2339 {
2340 u_int32_t md5_buffer[4];
2341 tcp_seq new_isn;
2342 struct timeval timenow;
2343 u_char isn_secret[32];
2344 int isn_last_reseed = 0;
2345 MD5_CTX isn_ctx;
2346
2347 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
2348 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT))
2349 && tcp_strict_rfc1948 == 0)
2350 #ifdef __APPLE__
2351 return RandomULong();
2352 #else
2353 return arc4random();
2354 #endif
2355 getmicrotime(&timenow);
2356
2357 /* Seed if this is the first use, reseed if requested. */
2358 if ((isn_last_reseed == 0) ||
2359 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
2360 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
2361 < (u_int)timenow.tv_sec))) {
2362 #ifdef __APPLE__
2363 read_random(&isn_secret, sizeof(isn_secret));
2364 #else
2365 read_random_unlimited(&isn_secret, sizeof(isn_secret));
2366 #endif
2367 isn_last_reseed = timenow.tv_sec;
2368 }
2369
2370 /* Compute the md5 hash and return the ISN. */
2371 MD5Init(&isn_ctx);
2372 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2373 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2374 #if INET6
2375 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2376 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2377 sizeof(struct in6_addr));
2378 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2379 sizeof(struct in6_addr));
2380 } else
2381 #endif
2382 {
2383 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2384 sizeof(struct in_addr));
2385 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2386 sizeof(struct in_addr));
2387 }
2388 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
2389 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2390 new_isn = (tcp_seq) md5_buffer[0];
2391 new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz);
2392 return new_isn;
2393 }
2394
2395
2396 /*
2397 * When a specific ICMP unreachable message is received and the
2398 * connection state is SYN-SENT, drop the connection. This behavior
2399 * is controlled by the icmp_may_rst sysctl.
2400 */
2401 void
2402 tcp_drop_syn_sent(inp, errno)
2403 struct inpcb *inp;
2404 int errno;
2405 {
2406 struct tcpcb *tp = intotcpcb(inp);
2407
2408 if (tp && tp->t_state == TCPS_SYN_SENT)
2409 tcp_drop(tp, errno);
2410 }
2411
2412 /*
2413 * When `need fragmentation' ICMP is received, update our idea of the MSS
2414 * based on the new value in the route. Also nudge TCP to send something,
2415 * since we know the packet we just sent was dropped.
2416 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2417 */
2418 void
2419 tcp_mtudisc(
2420 struct inpcb *inp,
2421 __unused int errno
2422 )
2423 {
2424 struct tcpcb *tp = intotcpcb(inp);
2425 struct rtentry *rt;
2426 struct rmxp_tao *taop;
2427 struct socket *so = inp->inp_socket;
2428 int offered;
2429 int mss;
2430 u_int32_t mtu;
2431 #if INET6
2432 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
2433 #endif /* INET6 */
2434
2435 if (tp) {
2436 #if INET6
2437 if (isipv6)
2438 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2439 else
2440 #endif /* INET6 */
2441 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2442 if (!rt || !rt->rt_rmx.rmx_mtu) {
2443 tp->t_maxopd = tp->t_maxseg =
2444 #if INET6
2445 isipv6 ? tcp_v6mssdflt :
2446 #endif /* INET6 */
2447 tcp_mssdflt;
2448
2449 /* Route locked during lookup above */
2450 if (rt != NULL)
2451 RT_UNLOCK(rt);
2452 return;
2453 }
2454 taop = rmx_taop(rt->rt_rmx);
2455 offered = taop->tao_mssopt;
2456 mtu = rt->rt_rmx.rmx_mtu;
2457
2458 /* Route locked during lookup above */
2459 RT_UNLOCK(rt);
2460
2461 #if NECP
2462 // Adjust MTU if necessary.
2463 mtu = necp_socket_get_effective_mtu(inp, mtu);
2464 #endif /* NECP */
2465
2466 mss = mtu -
2467 #if INET6
2468 (isipv6 ?
2469 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2470 #endif /* INET6 */
2471 sizeof(struct tcpiphdr)
2472 #if INET6
2473 )
2474 #endif /* INET6 */
2475 ;
2476
2477 if (offered)
2478 mss = min(mss, offered);
2479 /*
2480 * XXX - The above conditional probably violates the TCP
2481 * spec. The problem is that, since we don't know the
2482 * other end's MSS, we are supposed to use a conservative
2483 * default. But, if we do that, then MTU discovery will
2484 * never actually take place, because the conservative
2485 * default is much less than the MTUs typically seen
2486 * on the Internet today. For the moment, we'll sweep
2487 * this under the carpet.
2488 *
2489 * The conservative default might not actually be a problem
2490 * if the only case this occurs is when sending an initial
2491 * SYN with options and data to a host we've never talked
2492 * to before. Then, they will reply with an MSS value which
2493 * will get recorded and the new parameters should get
2494 * recomputed. For Further Study.
2495 */
2496 if (tp->t_maxopd <= mss)
2497 return;
2498 tp->t_maxopd = mss;
2499
2500 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2501 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2502 mss -= TCPOLEN_TSTAMP_APPA;
2503
2504 #if MPTCP
2505 mss -= mptcp_adj_mss(tp, TRUE);
2506 #endif
2507 if (so->so_snd.sb_hiwat < mss)
2508 mss = so->so_snd.sb_hiwat;
2509
2510 tp->t_maxseg = mss;
2511
2512 /*
2513 * Reset the slow-start flight size as it may depends on the new MSS
2514 */
2515 if (CC_ALGO(tp)->cwnd_init != NULL)
2516 CC_ALGO(tp)->cwnd_init(tp);
2517 tcpstat.tcps_mturesent++;
2518 tp->t_rtttime = 0;
2519 tp->snd_nxt = tp->snd_una;
2520 tcp_output(tp);
2521 }
2522 }
2523
2524 /*
2525 * Look-up the routing entry to the peer of this inpcb. If no route
2526 * is found and it cannot be allocated the return NULL. This routine
2527 * is called by TCP routines that access the rmx structure and by tcp_mss
2528 * to get the interface MTU. If a route is found, this routine will
2529 * hold the rtentry lock; the caller is responsible for unlocking.
2530 */
2531 struct rtentry *
2532 tcp_rtlookup(inp, input_ifscope)
2533 struct inpcb *inp;
2534 unsigned int input_ifscope;
2535 {
2536 struct route *ro;
2537 struct rtentry *rt;
2538 struct tcpcb *tp;
2539
2540 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2541
2542 ro = &inp->inp_route;
2543 if ((rt = ro->ro_rt) != NULL)
2544 RT_LOCK(rt);
2545
2546 if (ROUTE_UNUSABLE(ro)) {
2547 if (rt != NULL) {
2548 RT_UNLOCK(rt);
2549 rt = NULL;
2550 }
2551 ROUTE_RELEASE(ro);
2552 /* No route yet, so try to acquire one */
2553 if (inp->inp_faddr.s_addr != INADDR_ANY) {
2554 unsigned int ifscope;
2555
2556 ro->ro_dst.sa_family = AF_INET;
2557 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
2558 ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr =
2559 inp->inp_faddr;
2560
2561 /*
2562 * If the socket was bound to an interface, then
2563 * the bound-to-interface takes precedence over
2564 * the inbound interface passed in by the caller
2565 * (if we get here as part of the output path then
2566 * input_ifscope is IFSCOPE_NONE).
2567 */
2568 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2569 inp->inp_boundifp->if_index : input_ifscope;
2570
2571 rtalloc_scoped(ro, ifscope);
2572 if ((rt = ro->ro_rt) != NULL)
2573 RT_LOCK(rt);
2574 }
2575 }
2576 if (rt != NULL)
2577 RT_LOCK_ASSERT_HELD(rt);
2578
2579 /*
2580 * Update MTU discovery determination. Don't do it if:
2581 * 1) it is disabled via the sysctl
2582 * 2) the route isn't up
2583 * 3) the MTU is locked (if it is, then discovery has been
2584 * disabled)
2585 */
2586
2587 tp = intotcpcb(inp);
2588
2589 if (!path_mtu_discovery || ((rt != NULL) &&
2590 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2591 tp->t_flags &= ~TF_PMTUD;
2592 else
2593 tp->t_flags |= TF_PMTUD;
2594
2595 #if CONFIG_IFEF_NOWINDOWSCALE
2596 if (tcp_obey_ifef_nowindowscale &&
2597 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2598 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2599 /* Window scaling is enabled on this interface */
2600 tp->t_flags &= ~TF_REQ_SCALE;
2601 }
2602 #endif
2603
2604 if (rt != NULL && rt->rt_ifp != NULL) {
2605 somultipages(inp->inp_socket,
2606 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2607 tcp_set_tso(tp, rt->rt_ifp);
2608 soif2kcl(inp->inp_socket,
2609 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2610 tcp_set_ecn(tp, rt->rt_ifp);
2611 }
2612
2613 /* Note if the peer is local */
2614 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2615 (rt->rt_gateway->sa_family == AF_LINK ||
2616 rt->rt_ifp->if_flags & IFF_LOOPBACK ||
2617 in_localaddr(inp->inp_faddr))) {
2618 tp->t_flags |= TF_LOCAL;
2619 }
2620
2621 /*
2622 * Caller needs to call RT_UNLOCK(rt).
2623 */
2624 return rt;
2625 }
2626
2627 #if INET6
2628 struct rtentry *
2629 tcp_rtlookup6(inp, input_ifscope)
2630 struct inpcb *inp;
2631 unsigned int input_ifscope;
2632 {
2633 struct route_in6 *ro6;
2634 struct rtentry *rt;
2635 struct tcpcb *tp;
2636
2637 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2638
2639 ro6 = &inp->in6p_route;
2640 if ((rt = ro6->ro_rt) != NULL)
2641 RT_LOCK(rt);
2642
2643 if (ROUTE_UNUSABLE(ro6)) {
2644 if (rt != NULL) {
2645 RT_UNLOCK(rt);
2646 rt = NULL;
2647 }
2648 ROUTE_RELEASE(ro6);
2649 /* No route yet, so try to acquire one */
2650 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
2651 struct sockaddr_in6 *dst6;
2652 unsigned int ifscope;
2653
2654 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst;
2655 dst6->sin6_family = AF_INET6;
2656 dst6->sin6_len = sizeof(*dst6);
2657 dst6->sin6_addr = inp->in6p_faddr;
2658
2659 /*
2660 * If the socket was bound to an interface, then
2661 * the bound-to-interface takes precedence over
2662 * the inbound interface passed in by the caller
2663 * (if we get here as part of the output path then
2664 * input_ifscope is IFSCOPE_NONE).
2665 */
2666 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2667 inp->inp_boundifp->if_index : input_ifscope;
2668
2669 rtalloc_scoped((struct route *)ro6, ifscope);
2670 if ((rt = ro6->ro_rt) != NULL)
2671 RT_LOCK(rt);
2672 }
2673 }
2674 if (rt != NULL)
2675 RT_LOCK_ASSERT_HELD(rt);
2676
2677 /*
2678 * Update path MTU Discovery determination
2679 * while looking up the route:
2680 * 1) we have a valid route to the destination
2681 * 2) the MTU is not locked (if it is, then discovery has been
2682 * disabled)
2683 */
2684
2685
2686 tp = intotcpcb(inp);
2687
2688 /*
2689 * Update MTU discovery determination. Don't do it if:
2690 * 1) it is disabled via the sysctl
2691 * 2) the route isn't up
2692 * 3) the MTU is locked (if it is, then discovery has been
2693 * disabled)
2694 */
2695
2696 if (!path_mtu_discovery || ((rt != NULL) &&
2697 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2698 tp->t_flags &= ~TF_PMTUD;
2699 else
2700 tp->t_flags |= TF_PMTUD;
2701
2702 #if CONFIG_IFEF_NOWINDOWSCALE
2703 if (tcp_obey_ifef_nowindowscale &&
2704 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2705 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2706 /* Window scaling is not enabled on this interface */
2707 tp->t_flags &= ~TF_REQ_SCALE;
2708 }
2709 #endif
2710
2711 if (rt != NULL && rt->rt_ifp != NULL) {
2712 somultipages(inp->inp_socket,
2713 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2714 tcp_set_tso(tp, rt->rt_ifp);
2715 soif2kcl(inp->inp_socket,
2716 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2717 tcp_set_ecn(tp, rt->rt_ifp);
2718 }
2719
2720 /* Note if the peer is local */
2721 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2722 (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) ||
2723 IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) ||
2724 rt->rt_gateway->sa_family == AF_LINK ||
2725 in6_localaddr(&inp->in6p_faddr))) {
2726 tp->t_flags |= TF_LOCAL;
2727 }
2728
2729 /*
2730 * Caller needs to call RT_UNLOCK(rt).
2731 */
2732 return rt;
2733 }
2734 #endif /* INET6 */
2735
2736 #if IPSEC
2737 /* compute ESP/AH header size for TCP, including outer IP header. */
2738 size_t
2739 ipsec_hdrsiz_tcp(tp)
2740 struct tcpcb *tp;
2741 {
2742 struct inpcb *inp;
2743 struct mbuf *m;
2744 size_t hdrsiz;
2745 struct ip *ip;
2746 #if INET6
2747 struct ip6_hdr *ip6 = NULL;
2748 #endif /* INET6 */
2749 struct tcphdr *th;
2750
2751 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
2752 return 0;
2753 MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */
2754 if (!m)
2755 return 0;
2756
2757 #if INET6
2758 if ((inp->inp_vflag & INP_IPV6) != 0) {
2759 ip6 = mtod(m, struct ip6_hdr *);
2760 th = (struct tcphdr *)(void *)(ip6 + 1);
2761 m->m_pkthdr.len = m->m_len =
2762 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2763 tcp_fillheaders(tp, ip6, th);
2764 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2765 } else
2766 #endif /* INET6 */
2767 {
2768 ip = mtod(m, struct ip *);
2769 th = (struct tcphdr *)(ip + 1);
2770 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2771 tcp_fillheaders(tp, ip, th);
2772 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2773 }
2774 m_free(m);
2775 return hdrsiz;
2776 }
2777 #endif /*IPSEC*/
2778
2779 /*
2780 * Return a pointer to the cached information about the remote host.
2781 * The cached information is stored in the protocol specific part of
2782 * the route metrics.
2783 */
2784 struct rmxp_tao *
2785 tcp_gettaocache(inp)
2786 struct inpcb *inp;
2787 {
2788 struct rtentry *rt;
2789 struct rmxp_tao *taop;
2790
2791 #if INET6
2792 if ((inp->inp_vflag & INP_IPV6) != 0)
2793 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2794 else
2795 #endif /* INET6 */
2796 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2797
2798 /* Make sure this is a host route and is up. */
2799 if (rt == NULL ||
2800 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) {
2801 /* Route locked during lookup above */
2802 if (rt != NULL)
2803 RT_UNLOCK(rt);
2804 return NULL;
2805 }
2806
2807 taop = rmx_taop(rt->rt_rmx);
2808 /* Route locked during lookup above */
2809 RT_UNLOCK(rt);
2810 return (taop);
2811 }
2812
2813 /*
2814 * Clear all the TAO cache entries, called from tcp_init.
2815 *
2816 * XXX
2817 * This routine is just an empty one, because we assume that the routing
2818 * routing tables are initialized at the same time when TCP, so there is
2819 * nothing in the cache left over.
2820 */
2821 static void
2822 tcp_cleartaocache()
2823 {
2824 }
2825
2826 int
2827 tcp_lock(struct socket *so, int refcount, void *lr)
2828 {
2829 void *lr_saved;
2830
2831 if (lr == NULL)
2832 lr_saved = __builtin_return_address(0);
2833 else
2834 lr_saved = lr;
2835
2836 if (so->so_pcb != NULL) {
2837 lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2838 } else {
2839 panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n",
2840 so, lr_saved, solockhistory_nr(so));
2841 /* NOTREACHED */
2842 }
2843
2844 if (so->so_usecount < 0) {
2845 panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n",
2846 so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so));
2847 /* NOTREACHED */
2848 }
2849 if (refcount)
2850 so->so_usecount++;
2851 so->lock_lr[so->next_lock_lr] = lr_saved;
2852 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX;
2853 return (0);
2854 }
2855
2856 int
2857 tcp_unlock(struct socket *so, int refcount, void *lr)
2858 {
2859 void *lr_saved;
2860
2861 if (lr == NULL)
2862 lr_saved = __builtin_return_address(0);
2863 else
2864 lr_saved = lr;
2865
2866 #ifdef MORE_TCPLOCK_DEBUG
2867 printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x "
2868 "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so),
2869 (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb),
2870 (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)),
2871 so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved));
2872 #endif
2873 if (refcount)
2874 so->so_usecount--;
2875
2876 if (so->so_usecount < 0) {
2877 panic("tcp_unlock: so=%p usecount=%x lrh= %s\n",
2878 so, so->so_usecount, solockhistory_nr(so));
2879 /* NOTREACHED */
2880 }
2881 if (so->so_pcb == NULL) {
2882 panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n",
2883 so, so->so_usecount, lr_saved, solockhistory_nr(so));
2884 /* NOTREACHED */
2885 } else {
2886 lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx,
2887 LCK_MTX_ASSERT_OWNED);
2888 so->unlock_lr[so->next_unlock_lr] = lr_saved;
2889 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX;
2890 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2891 }
2892 return (0);
2893 }
2894
2895 lck_mtx_t *
2896 tcp_getlock(
2897 struct socket *so,
2898 __unused int locktype)
2899 {
2900 struct inpcb *inp = sotoinpcb(so);
2901
2902 if (so->so_pcb) {
2903 if (so->so_usecount < 0)
2904 panic("tcp_getlock: so=%p usecount=%x lrh= %s\n",
2905 so, so->so_usecount, solockhistory_nr(so));
2906 return(&inp->inpcb_mtx);
2907 }
2908 else {
2909 panic("tcp_getlock: so=%p NULL so_pcb %s\n",
2910 so, solockhistory_nr(so));
2911 return (so->so_proto->pr_domain->dom_mtx);
2912 }
2913 }
2914
2915 /*
2916 * Determine if we can grow the recieve socket buffer to avoid sending
2917 * a zero window update to the peer. We allow even socket buffers that
2918 * have fixed size (set by the application) to grow if the resource
2919 * constraints are met. They will also be trimmed after the application
2920 * reads data.
2921 */
2922 static void
2923 tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb)
2924 {
2925 u_int32_t rcvbufinc = tp->t_maxseg << 4;
2926 u_int32_t rcvbuf = sb->sb_hiwat;
2927 struct socket *so = tp->t_inpcb->inp_socket;
2928
2929 /*
2930 * If message delivery is enabled, do not count
2931 * unordered bytes in receive buffer towards hiwat
2932 */
2933 if (so->so_flags & SOF_ENABLE_MSGS)
2934 rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes;
2935
2936 if (tcp_do_autorcvbuf == 1 &&
2937 tcp_cansbgrow(sb) &&
2938 (tp->t_flags & TF_SLOWLINK) == 0 &&
2939 (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 &&
2940 (rcvbuf - sb->sb_cc) < rcvbufinc &&
2941 rcvbuf < tcp_autorcvbuf_max &&
2942 (sb->sb_idealsize > 0 &&
2943 sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) {
2944 sbreserve(sb,
2945 min((sb->sb_hiwat + rcvbufinc), tcp_autorcvbuf_max));
2946 }
2947 }
2948
2949 int32_t
2950 tcp_sbspace(struct tcpcb *tp)
2951 {
2952 struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv;
2953 u_int32_t rcvbuf = sb->sb_hiwat;
2954 int32_t space;
2955 struct socket *so = tp->t_inpcb->inp_socket;
2956 int32_t pending = 0;
2957
2958 /*
2959 * If message delivery is enabled, do not count
2960 * unordered bytes in receive buffer towards hiwat mark.
2961 * This value is used to return correct rwnd that does
2962 * not reflect the extra unordered bytes added to the
2963 * receive socket buffer.
2964 */
2965 if (so->so_flags & SOF_ENABLE_MSGS)
2966 rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes;
2967
2968 tcp_sbrcv_grow_rwin(tp, sb);
2969
2970 space = ((int32_t) imin((rcvbuf - sb->sb_cc),
2971 (sb->sb_mbmax - sb->sb_mbcnt)));
2972 if (space < 0)
2973 space = 0;
2974
2975 #if CONTENT_FILTER
2976 /* Compensate for data being processed by content filters */
2977 pending = cfil_sock_data_space(sb);
2978 #endif /* CONTENT_FILTER */
2979 if (pending > space)
2980 space = 0;
2981 else
2982 space -= pending;
2983
2984 /* Avoid increasing window size if the current window
2985 * is already very low, we could be in "persist" mode and
2986 * we could break some apps (see rdar://5409343)
2987 */
2988
2989 if (space < tp->t_maxseg)
2990 return space;
2991
2992 /* Clip window size for slower link */
2993
2994 if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 )
2995 return imin(space, slowlink_wsize);
2996
2997 return space;
2998 }
2999 /*
3000 * Checks TCP Segment Offloading capability for a given connection
3001 * and interface pair.
3002 */
3003 void
3004 tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp)
3005 {
3006 #if INET6
3007 struct inpcb *inp;
3008 int isipv6;
3009 #endif /* INET6 */
3010 #if MPTCP
3011 /*
3012 * We can't use TSO if this tcpcb belongs to an MPTCP session.
3013 */
3014 if (tp->t_mpflags & TMPF_MPTCP_TRUE) {
3015 tp->t_flags &= ~TF_TSO;
3016 return;
3017 }
3018 #endif
3019 #if INET6
3020 inp = tp->t_inpcb;
3021 isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
3022
3023 if (isipv6) {
3024 if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV6)) {
3025 tp->t_flags |= TF_TSO;
3026 if (ifp->if_tso_v6_mtu != 0)
3027 tp->tso_max_segment_size = ifp->if_tso_v6_mtu;
3028 else
3029 tp->tso_max_segment_size = TCP_MAXWIN;
3030 } else
3031 tp->t_flags &= ~TF_TSO;
3032
3033 } else
3034 #endif /* INET6 */
3035
3036 {
3037 if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) {
3038 tp->t_flags |= TF_TSO;
3039 if (ifp->if_tso_v4_mtu != 0)
3040 tp->tso_max_segment_size = ifp->if_tso_v4_mtu;
3041 else
3042 tp->tso_max_segment_size = TCP_MAXWIN;
3043 } else
3044 tp->t_flags &= ~TF_TSO;
3045 }
3046 }
3047
3048 #define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC)
3049
3050 /* Function to calculate the tcp clock. The tcp clock will get updated
3051 * at the boundaries of the tcp layer. This is done at 3 places:
3052 * 1. Right before processing an input tcp packet
3053 * 2. Whenever a connection wants to access the network using tcp_usrreqs
3054 * 3. When a tcp timer fires or before tcp slow timeout
3055 *
3056 */
3057
3058 void
3059 calculate_tcp_clock()
3060 {
3061 struct timeval tv = tcp_uptime;
3062 struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC};
3063 struct timeval now, hold_now;
3064 uint32_t incr = 0;
3065
3066 microuptime(&now);
3067
3068 /*
3069 * Update coarse-grained networking timestamp (in sec.); the idea
3070 * is to update the counter returnable via net_uptime() when
3071 * we read time.
3072 */
3073 net_update_uptime_secs(now.tv_sec);
3074
3075 timevaladd(&tv, &interval);
3076 if (timevalcmp(&now, &tv, >)) {
3077 /* time to update the clock */
3078 lck_spin_lock(tcp_uptime_lock);
3079 if (timevalcmp(&tcp_uptime, &now, >=)) {
3080 /* clock got updated while waiting for the lock */
3081 lck_spin_unlock(tcp_uptime_lock);
3082 return;
3083 }
3084
3085 microuptime(&now);
3086 hold_now = now;
3087 tv = tcp_uptime;
3088 timevalsub(&now, &tv);
3089
3090 incr = TIMEVAL_TO_TCPHZ(now);
3091 if (incr > 0) {
3092 tcp_uptime = hold_now;
3093 tcp_now += incr;
3094 }
3095
3096 lck_spin_unlock(tcp_uptime_lock);
3097 }
3098 return;
3099 }
3100
3101 /* Compute receive window scaling that we are going to request
3102 * for this connection based on sb_hiwat. Try to leave some
3103 * room to potentially increase the window size upto a maximum
3104 * defined by the constant tcp_autorcvbuf_max.
3105 */
3106 void
3107 tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) {
3108 u_int32_t maxsockbufsize;
3109 if (!tcp_do_rfc1323) {
3110 tp->request_r_scale = 0;
3111 return;
3112 }
3113
3114 tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale);
3115 maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ?
3116 so->so_rcv.sb_hiwat : tcp_autorcvbuf_max;
3117
3118 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
3119 (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize)
3120 tp->request_r_scale++;
3121 tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT);
3122
3123 }
3124
3125 int
3126 tcp_notsent_lowat_check(struct socket *so) {
3127 struct inpcb *inp = sotoinpcb(so);
3128 struct tcpcb *tp = NULL;
3129 int notsent = 0;
3130 if (inp != NULL) {
3131 tp = intotcpcb(inp);
3132 }
3133
3134 notsent = so->so_snd.sb_cc -
3135 (tp->snd_nxt - tp->snd_una);
3136
3137 /* When we send a FIN or SYN, not_sent can be negative.
3138 * In that case also we need to send a write event to the
3139 * process if it is waiting. In the FIN case, it will
3140 * get an error from send because cantsendmore will be set.
3141 */
3142 if (notsent <= tp->t_notsent_lowat) {
3143 return(1);
3144 }
3145
3146 /* When Nagle's algorithm is not disabled, it is better
3147 * to wakeup the client until there is atleast one
3148 * maxseg of data to write.
3149 */
3150 if ((tp->t_flags & TF_NODELAY) == 0 &&
3151 notsent > 0 && notsent < tp->t_maxseg) {
3152 return(1);
3153 }
3154 return(0);
3155 }
3156
3157 void
3158 tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) {
3159 struct tcp_rxt_seg *rxseg = NULL, *prev = NULL, *next = NULL;
3160 u_int32_t rxcount = 0;
3161
3162 if (SLIST_EMPTY(&tp->t_rxt_segments))
3163 tp->t_dsack_lastuna = tp->snd_una;
3164 /*
3165 * First check if there is a segment already existing for this
3166 * sequence space.
3167 */
3168
3169 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3170 if (SEQ_GT(rxseg->rx_start, start))
3171 break;
3172 prev = rxseg;
3173 }
3174 next = rxseg;
3175
3176 /* check if prev seg is for this sequence */
3177 if (prev != NULL && SEQ_LEQ(prev->rx_start, start) &&
3178 SEQ_GEQ(prev->rx_end, end)) {
3179 prev->rx_count++;
3180 return;
3181 }
3182
3183 /*
3184 * There are a couple of possibilities at this point.
3185 * 1. prev overlaps with the beginning of this sequence
3186 * 2. next overlaps with the end of this sequence
3187 * 3. there is no overlap.
3188 */
3189
3190 if (prev != NULL && SEQ_GT(prev->rx_end, start)) {
3191 if (prev->rx_start == start && SEQ_GT(end, prev->rx_end)) {
3192 start = prev->rx_end + 1;
3193 prev->rx_count++;
3194 } else {
3195 prev->rx_end = (start - 1);
3196 rxcount = prev->rx_count;
3197 }
3198 }
3199
3200 if (next != NULL && SEQ_LT(next->rx_start, end)) {
3201 if (SEQ_LEQ(next->rx_end, end)) {
3202 end = next->rx_start - 1;
3203 next->rx_count++;
3204 } else {
3205 next->rx_start = end + 1;
3206 rxcount = next->rx_count;
3207 }
3208 }
3209 if (!SEQ_LT(start, end))
3210 return;
3211
3212 rxseg = (struct tcp_rxt_seg *) zalloc(tcp_rxt_seg_zone);
3213 if (rxseg == NULL) {
3214 return;
3215 }
3216 bzero(rxseg, sizeof(*rxseg));
3217 rxseg->rx_start = start;
3218 rxseg->rx_end = end;
3219 rxseg->rx_count = rxcount + 1;
3220
3221 if (prev != NULL) {
3222 SLIST_INSERT_AFTER(prev, rxseg, rx_link);
3223 } else {
3224 SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link);
3225 }
3226 return;
3227 }
3228
3229 struct tcp_rxt_seg *
3230 tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3231 {
3232 struct tcp_rxt_seg *rxseg;
3233 if (SLIST_EMPTY(&tp->t_rxt_segments))
3234 return (NULL);
3235
3236 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3237 if (SEQ_LEQ(rxseg->rx_start, start) &&
3238 SEQ_GEQ(rxseg->rx_end, end))
3239 return (rxseg);
3240 if (SEQ_GT(rxseg->rx_start, start))
3241 break;
3242 }
3243 return (NULL);
3244 }
3245
3246 void
3247 tcp_rxtseg_clean(struct tcpcb *tp)
3248 {
3249 struct tcp_rxt_seg *rxseg, *next;
3250
3251 SLIST_FOREACH_SAFE(rxseg, &tp->t_rxt_segments, rx_link, next) {
3252 SLIST_REMOVE(&tp->t_rxt_segments, rxseg,
3253 tcp_rxt_seg, rx_link);
3254 zfree(tcp_rxt_seg_zone, rxseg);
3255 }
3256 tp->t_dsack_lastuna = tp->snd_max;
3257 }
3258
3259 boolean_t
3260 tcp_rxtseg_detect_bad_rexmt(struct tcpcb *tp, tcp_seq th_ack)
3261 {
3262 boolean_t bad_rexmt;
3263 struct tcp_rxt_seg *rxseg;
3264
3265 if (SLIST_EMPTY(&tp->t_rxt_segments))
3266 return (FALSE);
3267
3268 /*
3269 * If all of the segments in this window are not cumulatively
3270 * acknowledged, then there can still be undetected packet loss.
3271 * Do not restore congestion window in that case.
3272 */
3273 if (SEQ_LT(th_ack, tp->snd_recover))
3274 return (FALSE);
3275
3276 bad_rexmt = TRUE;
3277 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3278 if (rxseg->rx_count > 1 ||
3279 !(rxseg->rx_flags & TCP_RXT_SPURIOUS)) {
3280 bad_rexmt = FALSE;
3281 break;
3282 }
3283 }
3284 return (bad_rexmt);
3285 }
3286
3287 boolean_t
3288 tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp)
3289 {
3290 boolean_t dsack_for_tlp = FALSE;
3291 struct tcp_rxt_seg *rxseg;
3292 if (SLIST_EMPTY(&tp->t_rxt_segments))
3293 return (FALSE);
3294
3295 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3296 if (rxseg->rx_count == 1 &&
3297 SLIST_NEXT(rxseg,rx_link) == NULL &&
3298 (rxseg->rx_flags & TCP_RXT_DSACK_FOR_TLP)) {
3299 dsack_for_tlp = TRUE;
3300 break;
3301 }
3302 }
3303 return (dsack_for_tlp);
3304 }
3305
3306 u_int32_t
3307 tcp_rxtseg_total_size(struct tcpcb *tp) {
3308 struct tcp_rxt_seg *rxseg;
3309 u_int32_t total_size = 0;
3310
3311 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3312 total_size += (rxseg->rx_end - rxseg->rx_start) + 1;
3313 }
3314 return (total_size);
3315 }
3316
3317 void
3318 tcp_get_connectivity_status(struct tcpcb *tp,
3319 struct tcp_conn_status *connstatus)
3320 {
3321 if (tp == NULL || connstatus == NULL)
3322 return;
3323 bzero(connstatus, sizeof(*connstatus));
3324 if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) {
3325 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3326 connstatus->write_probe_failed = 1;
3327 } else {
3328 connstatus->conn_probe_failed = 1;
3329 }
3330 }
3331 if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX)
3332 connstatus->read_probe_failed = 1;
3333 if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL
3334 && (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY))
3335 connstatus->probe_activated = 1;
3336 return;
3337 }
3338
3339 boolean_t
3340 tfo_enabled(const struct tcpcb *tp)
3341 {
3342 return !!(tp->t_flagsext & TF_FASTOPEN);
3343 }
3344
3345 void
3346 tcp_disable_tfo(struct tcpcb *tp)
3347 {
3348 tp->t_flagsext &= ~TF_FASTOPEN;
3349 }
3350
mirror of XNU