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


6fafa0f5f853b481cabd900abf4f5ac82c9e3767
[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 /*
1138 * Close a TCP control block:
1139 * discard all space held by the tcp
1140 * discard internet protocol block
1141 * wake up any sleepers
1142 */
1143 struct tcpcb *
1144 tcp_close(tp)
1145 register struct tcpcb *tp;
1146 {
1147 struct inpcb *inp = tp->t_inpcb;
1148 struct socket *so = inp->inp_socket;
1149 #if INET6
1150 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1151 #endif /* INET6 */
1152 struct route *ro;
1153 struct rtentry *rt;
1154 int dosavessthresh;
1155
1156 /* tcp_close was called previously, bail */
1157 if (inp->inp_ppcb == NULL)
1158 return(NULL);
1159
1160 tcp_canceltimers(tp);
1161 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0);
1162
1163 /*
1164 * If another thread for this tcp is currently in ip (indicated by
1165 * the TF_SENDINPROG flag), defer the cleanup until after it returns
1166 * back to tcp. This is done to serialize the close until after all
1167 * pending output is finished, in order to avoid having the PCB be
1168 * detached and the cached route cleaned, only for ip to cache the
1169 * route back into the PCB again. Note that we've cleared all the
1170 * timers at this point. Set TF_CLOSING to indicate to tcp_output()
1171 * that is should call us again once it returns from ip; at that
1172 * point both flags should be cleared and we can proceed further
1173 * with the cleanup.
1174 */
1175 if ((tp->t_flags & TF_CLOSING) ||
1176 inp->inp_sndinprog_cnt > 0) {
1177 tp->t_flags |= TF_CLOSING;
1178 return (NULL);
1179 }
1180
1181 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1182 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1183
1184 #if INET6
1185 ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route);
1186 #else
1187 ro = &inp->inp_route;
1188 #endif
1189 rt = ro->ro_rt;
1190 if (rt != NULL)
1191 RT_LOCK_SPIN(rt);
1192
1193 /*
1194 * If we got enough samples through the srtt filter,
1195 * save the rtt and rttvar in the routing entry.
1196 * 'Enough' is arbitrarily defined as the 16 samples.
1197 * 16 samples is enough for the srtt filter to converge
1198 * to within 5% of the correct value; fewer samples and
1199 * we could save a very bogus rtt.
1200 *
1201 * Don't update the default route's characteristics and don't
1202 * update anything that the user "locked".
1203 */
1204 if (tp->t_rttupdated >= 16) {
1205 register u_int32_t i = 0;
1206
1207 #if INET6
1208 if (isipv6) {
1209 struct sockaddr_in6 *sin6;
1210
1211 if (rt == NULL)
1212 goto no_valid_rt;
1213 sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt);
1214 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1215 goto no_valid_rt;
1216 }
1217 else
1218 #endif /* INET6 */
1219 if (ROUTE_UNUSABLE(ro) ||
1220 SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) {
1221 DTRACE_TCP4(state__change, void, NULL,
1222 struct inpcb *, inp, struct tcpcb *, tp,
1223 int32_t, TCPS_CLOSED);
1224 tp->t_state = TCPS_CLOSED;
1225 goto no_valid_rt;
1226 }
1227
1228 RT_LOCK_ASSERT_HELD(rt);
1229 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1230 i = tp->t_srtt *
1231 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1232 if (rt->rt_rmx.rmx_rtt && i)
1233 /*
1234 * filter this update to half the old & half
1235 * the new values, converting scale.
1236 * See route.h and tcp_var.h for a
1237 * description of the scaling constants.
1238 */
1239 rt->rt_rmx.rmx_rtt =
1240 (rt->rt_rmx.rmx_rtt + i) / 2;
1241 else
1242 rt->rt_rmx.rmx_rtt = i;
1243 tcpstat.tcps_cachedrtt++;
1244 }
1245 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1246 i = tp->t_rttvar *
1247 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1248 if (rt->rt_rmx.rmx_rttvar && i)
1249 rt->rt_rmx.rmx_rttvar =
1250 (rt->rt_rmx.rmx_rttvar + i) / 2;
1251 else
1252 rt->rt_rmx.rmx_rttvar = i;
1253 tcpstat.tcps_cachedrttvar++;
1254 }
1255 /*
1256 * The old comment here said:
1257 * update the pipelimit (ssthresh) if it has been updated
1258 * already or if a pipesize was specified & the threshhold
1259 * got below half the pipesize. I.e., wait for bad news
1260 * before we start updating, then update on both good
1261 * and bad news.
1262 *
1263 * But we want to save the ssthresh even if no pipesize is
1264 * specified explicitly in the route, because such
1265 * connections still have an implicit pipesize specified
1266 * by the global tcp_sendspace. In the absence of a reliable
1267 * way to calculate the pipesize, it will have to do.
1268 */
1269 i = tp->snd_ssthresh;
1270 if (rt->rt_rmx.rmx_sendpipe != 0)
1271 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
1272 else
1273 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
1274 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1275 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
1276 || dosavessthresh) {
1277 /*
1278 * convert the limit from user data bytes to
1279 * packets then to packet data bytes.
1280 */
1281 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
1282 if (i < 2)
1283 i = 2;
1284 i *= (u_int32_t)(tp->t_maxseg +
1285 #if INET6
1286 (isipv6 ? sizeof (struct ip6_hdr) +
1287 sizeof (struct tcphdr) :
1288 #endif
1289 sizeof (struct tcpiphdr)
1290 #if INET6
1291 )
1292 #endif
1293 );
1294 if (rt->rt_rmx.rmx_ssthresh)
1295 rt->rt_rmx.rmx_ssthresh =
1296 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1297 else
1298 rt->rt_rmx.rmx_ssthresh = i;
1299 tcpstat.tcps_cachedssthresh++;
1300 }
1301 }
1302
1303 /*
1304 * Mark route for deletion if no information is cached.
1305 */
1306 if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) {
1307 if (!(rt->rt_rmx.rmx_locks & RTV_RTT) &&
1308 rt->rt_rmx.rmx_rtt == 0) {
1309 rt->rt_flags |= RTF_DELCLONE;
1310 }
1311 }
1312
1313 no_valid_rt:
1314 if (rt != NULL)
1315 RT_UNLOCK(rt);
1316
1317 /* free the reassembly queue, if any */
1318 (void) tcp_freeq(tp);
1319 if (TCP_ECN_ENABLED(tp)) {
1320 if (tp->ecn_flags & TE_RECV_ECN_CE)
1321 tcpstat.tcps_ecn_conn_recv_ce++;
1322 if (tp->ecn_flags & TE_RECV_ECN_ECE)
1323 tcpstat.tcps_ecn_conn_recv_ece++;
1324 if (tp->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) {
1325 if (tp->t_stat.txretransmitbytes > 0 ||
1326 tp->t_stat.rxoutoforderbytes > 0)
1327 tcpstat.tcps_ecn_conn_pl_ce++;
1328 else
1329 tcpstat.tcps_ecn_conn_nopl_ce++;
1330 } else {
1331 if (tp->t_stat.txretransmitbytes > 0 ||
1332 tp->t_stat.rxoutoforderbytes > 0)
1333 tcpstat.tcps_ecn_conn_plnoce++;
1334 }
1335 }
1336
1337 tcp_free_sackholes(tp);
1338 if (tp->t_bwmeas != NULL) {
1339 tcp_bwmeas_free(tp);
1340 }
1341 tcp_rxtseg_clean(tp);
1342 /* Free the packet list */
1343 if (tp->t_pktlist_head != NULL)
1344 m_freem_list(tp->t_pktlist_head);
1345 TCP_PKTLIST_CLEAR(tp);
1346
1347 #if MPTCP
1348 /* Clear MPTCP state */
1349 if ((so->so_flags & SOF_MPTCP_TRUE) ||
1350 (so->so_flags & SOF_MP_SUBFLOW)) {
1351 soevent(so, (SO_FILT_HINT_LOCKED | SO_FILT_HINT_DELETEOK));
1352 }
1353 tp->t_mpflags = 0;
1354 tp->t_mptcb = NULL;
1355 #endif /* MPTCP */
1356
1357 if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER)
1358 inp->inp_saved_ppcb = (caddr_t) tp;
1359
1360 tp->t_state = TCPS_CLOSED;
1361
1362 /* Issue a wakeup before detach so that we don't miss
1363 * a wakeup
1364 */
1365 sodisconnectwakeup(so);
1366
1367 /*
1368 * Clean up any LRO state
1369 */
1370 if (tp->t_flagsext & TF_LRO_OFFLOADED) {
1371 tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr,
1372 inp->inp_lport, inp->inp_fport);
1373 tp->t_flagsext &= ~TF_LRO_OFFLOADED;
1374 }
1375
1376 /*
1377 * If this is a socket that does not want to wakeup the device
1378 * for it's traffic, the application might need to know that the
1379 * socket is closed, send a notification.
1380 */
1381 if ((so->so_options & SO_NOWAKEFROMSLEEP) &&
1382 inp->inp_state != INPCB_STATE_DEAD &&
1383 !(inp->inp_flags2 & INP2_TIMEWAIT))
1384 socket_post_kev_msg_closed(so);
1385
1386 if (CC_ALGO(tp)->cleanup != NULL) {
1387 CC_ALGO(tp)->cleanup(tp);
1388 }
1389
1390 if (tp->t_ccstate != NULL) {
1391 zfree(tcp_cc_zone, tp->t_ccstate);
1392 tp->t_ccstate = NULL;
1393 }
1394 tp->tcp_cc_index = TCP_CC_ALGO_NONE;
1395
1396 /* Can happen if we close the socket before receiving the third ACK */
1397 if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) {
1398 OSDecrementAtomic(&tcp_tfo_halfcnt);
1399
1400 /* Panic if something has gone terribly wrong. */
1401 VERIFY(tcp_tfo_halfcnt >= 0);
1402
1403 tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID;
1404 }
1405
1406 #if INET6
1407 if (SOCK_CHECK_DOM(so, PF_INET6))
1408 in6_pcbdetach(inp);
1409 else
1410 #endif /* INET6 */
1411 in_pcbdetach(inp);
1412
1413 /* Call soisdisconnected after detach because it might unlock the socket */
1414 soisdisconnected(so);
1415 tcpstat.tcps_closed++;
1416 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END,
1417 tcpstat.tcps_closed, 0, 0, 0, 0);
1418 return(NULL);
1419 }
1420
1421 int
1422 tcp_freeq(tp)
1423 struct tcpcb *tp;
1424 {
1425
1426 register struct tseg_qent *q;
1427 int rv = 0;
1428
1429 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
1430 LIST_REMOVE(q, tqe_q);
1431 m_freem(q->tqe_m);
1432 zfree(tcp_reass_zone, q);
1433 rv = 1;
1434 }
1435 tp->t_reassqlen = 0;
1436 return (rv);
1437 }
1438
1439
1440 /*
1441 * Walk the tcpbs, if existing, and flush the reassembly queue,
1442 * if there is one when do_tcpdrain is enabled
1443 * Also defunct the extended background idle socket
1444 * Do it next time if the pcbinfo lock is in use
1445 */
1446 void
1447 tcp_drain()
1448 {
1449 struct inpcb *inp;
1450 struct tcpcb *tp;
1451
1452 if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock))
1453 return;
1454
1455 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1456 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) !=
1457 WNT_STOPUSING) {
1458 tcp_lock(inp->inp_socket, 1, 0);
1459 if (in_pcb_checkstate(inp, WNT_RELEASE, 1)
1460 == WNT_STOPUSING) {
1461 /* lost a race, try the next one */
1462 tcp_unlock(inp->inp_socket, 1, 0);
1463 continue;
1464 }
1465 tp = intotcpcb(inp);
1466
1467 if (do_tcpdrain)
1468 tcp_freeq(tp);
1469
1470 so_drain_extended_bk_idle(inp->inp_socket);
1471
1472 tcp_unlock(inp->inp_socket, 1, 0);
1473 }
1474 }
1475 lck_rw_done(tcbinfo.ipi_lock);
1476
1477 }
1478
1479 /*
1480 * Notify a tcp user of an asynchronous error;
1481 * store error as soft error, but wake up user
1482 * (for now, won't do anything until can select for soft error).
1483 *
1484 * Do not wake up user since there currently is no mechanism for
1485 * reporting soft errors (yet - a kqueue filter may be added).
1486 */
1487 static void
1488 tcp_notify(inp, error)
1489 struct inpcb *inp;
1490 int error;
1491 {
1492 struct tcpcb *tp;
1493
1494 if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD))
1495 return; /* pcb is gone already */
1496
1497 tp = (struct tcpcb *)inp->inp_ppcb;
1498
1499 /*
1500 * Ignore some errors if we are hooked up.
1501 * If connection hasn't completed, has retransmitted several times,
1502 * and receives a second error, give up now. This is better
1503 * than waiting a long time to establish a connection that
1504 * can never complete.
1505 */
1506 if (tp->t_state == TCPS_ESTABLISHED &&
1507 (error == EHOSTUNREACH || error == ENETUNREACH ||
1508 error == EHOSTDOWN)) {
1509 return;
1510 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1511 tp->t_softerror)
1512 tcp_drop(tp, error);
1513 else
1514 tp->t_softerror = error;
1515 #if 0
1516 wakeup((caddr_t) &so->so_timeo);
1517 sorwakeup(so);
1518 sowwakeup(so);
1519 #endif
1520 }
1521
1522 struct bwmeas*
1523 tcp_bwmeas_alloc(struct tcpcb *tp)
1524 {
1525 struct bwmeas *elm;
1526 elm = zalloc(tcp_bwmeas_zone);
1527 if (elm == NULL)
1528 return(elm);
1529
1530 bzero(elm, bwmeas_elm_size);
1531 elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE;
1532 elm->bw_maxsizepkts = TCP_BWMEAS_BURST_MAXSIZE;
1533 elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg;
1534 elm->bw_maxsize = elm->bw_maxsizepkts * tp->t_maxseg;
1535 return(elm);
1536 }
1537
1538 void
1539 tcp_bwmeas_free(struct tcpcb* tp)
1540 {
1541 zfree(tcp_bwmeas_zone, tp->t_bwmeas);
1542 tp->t_bwmeas = NULL;
1543 tp->t_flagsext &= ~(TF_MEASURESNDBW);
1544 }
1545
1546 /*
1547 * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format.
1548 * The otcpcb data structure is passed to user space and must not change.
1549 */
1550 static void
1551 tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp)
1552 {
1553 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1554 otp->t_dupacks = tp->t_dupacks;
1555 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1556 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1557 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1558 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1559 otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb);
1560 otp->t_state = tp->t_state;
1561 otp->t_flags = tp->t_flags;
1562 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1563 otp->snd_una = tp->snd_una;
1564 otp->snd_max = tp->snd_max;
1565 otp->snd_nxt = tp->snd_nxt;
1566 otp->snd_up = tp->snd_up;
1567 otp->snd_wl1 = tp->snd_wl1;
1568 otp->snd_wl2 = tp->snd_wl2;
1569 otp->iss = tp->iss;
1570 otp->irs = tp->irs;
1571 otp->rcv_nxt = tp->rcv_nxt;
1572 otp->rcv_adv = tp->rcv_adv;
1573 otp->rcv_wnd = tp->rcv_wnd;
1574 otp->rcv_up = tp->rcv_up;
1575 otp->snd_wnd = tp->snd_wnd;
1576 otp->snd_cwnd = tp->snd_cwnd;
1577 otp->snd_ssthresh = tp->snd_ssthresh;
1578 otp->t_maxopd = tp->t_maxopd;
1579 otp->t_rcvtime = tp->t_rcvtime;
1580 otp->t_starttime = tp->t_starttime;
1581 otp->t_rtttime = tp->t_rtttime;
1582 otp->t_rtseq = tp->t_rtseq;
1583 otp->t_rxtcur = tp->t_rxtcur;
1584 otp->t_maxseg = tp->t_maxseg;
1585 otp->t_srtt = tp->t_srtt;
1586 otp->t_rttvar = tp->t_rttvar;
1587 otp->t_rxtshift = tp->t_rxtshift;
1588 otp->t_rttmin = tp->t_rttmin;
1589 otp->t_rttupdated = tp->t_rttupdated;
1590 otp->max_sndwnd = tp->max_sndwnd;
1591 otp->t_softerror = tp->t_softerror;
1592 otp->t_oobflags = tp->t_oobflags;
1593 otp->t_iobc = tp->t_iobc;
1594 otp->snd_scale = tp->snd_scale;
1595 otp->rcv_scale = tp->rcv_scale;
1596 otp->request_r_scale = tp->request_r_scale;
1597 otp->requested_s_scale = tp->requested_s_scale;
1598 otp->ts_recent = tp->ts_recent;
1599 otp->ts_recent_age = tp->ts_recent_age;
1600 otp->last_ack_sent = tp->last_ack_sent;
1601 otp->cc_send = tp->cc_send;
1602 otp->cc_recv = tp->cc_recv;
1603 otp->snd_recover = tp->snd_recover;
1604 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1605 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1606 otp->t_badrxtwin = 0;
1607 }
1608
1609 static int
1610 tcp_pcblist SYSCTL_HANDLER_ARGS
1611 {
1612 #pragma unused(oidp, arg1, arg2)
1613 int error, i = 0, n;
1614 struct inpcb *inp, **inp_list;
1615 struct tcpcb *tp;
1616 inp_gen_t gencnt;
1617 struct xinpgen xig;
1618
1619 /*
1620 * The process of preparing the TCB list is too time-consuming and
1621 * resource-intensive to repeat twice on every request.
1622 */
1623 lck_rw_lock_shared(tcbinfo.ipi_lock);
1624 if (req->oldptr == USER_ADDR_NULL) {
1625 n = tcbinfo.ipi_count;
1626 req->oldidx = 2 * (sizeof xig)
1627 + (n + n/8) * sizeof(struct xtcpcb);
1628 lck_rw_done(tcbinfo.ipi_lock);
1629 return 0;
1630 }
1631
1632 if (req->newptr != USER_ADDR_NULL) {
1633 lck_rw_done(tcbinfo.ipi_lock);
1634 return EPERM;
1635 }
1636
1637 /*
1638 * OK, now we're committed to doing something.
1639 */
1640 gencnt = tcbinfo.ipi_gencnt;
1641 n = tcbinfo.ipi_count;
1642
1643 bzero(&xig, sizeof(xig));
1644 xig.xig_len = sizeof xig;
1645 xig.xig_count = n;
1646 xig.xig_gen = gencnt;
1647 xig.xig_sogen = so_gencnt;
1648 error = SYSCTL_OUT(req, &xig, sizeof xig);
1649 if (error) {
1650 lck_rw_done(tcbinfo.ipi_lock);
1651 return error;
1652 }
1653 /*
1654 * We are done if there is no pcb
1655 */
1656 if (n == 0) {
1657 lck_rw_done(tcbinfo.ipi_lock);
1658 return 0;
1659 }
1660
1661 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1662 if (inp_list == 0) {
1663 lck_rw_done(tcbinfo.ipi_lock);
1664 return ENOMEM;
1665 }
1666
1667 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1668 if (inp->inp_gencnt <= gencnt &&
1669 inp->inp_state != INPCB_STATE_DEAD)
1670 inp_list[i++] = inp;
1671 if (i >= n) break;
1672 }
1673
1674 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
1675 inp = tp->t_inpcb;
1676 if (inp->inp_gencnt <= gencnt &&
1677 inp->inp_state != INPCB_STATE_DEAD)
1678 inp_list[i++] = inp;
1679 if (i >= n) break;
1680 }
1681
1682 n = i;
1683
1684 error = 0;
1685 for (i = 0; i < n; i++) {
1686 inp = inp_list[i];
1687 if (inp->inp_gencnt <= gencnt &&
1688 inp->inp_state != INPCB_STATE_DEAD) {
1689 struct xtcpcb xt;
1690 caddr_t inp_ppcb;
1691
1692 bzero(&xt, sizeof(xt));
1693 xt.xt_len = sizeof xt;
1694 /* XXX should avoid extra copy */
1695 inpcb_to_compat(inp, &xt.xt_inp);
1696 inp_ppcb = inp->inp_ppcb;
1697 if (inp_ppcb != NULL) {
1698 tcpcb_to_otcpcb(
1699 (struct tcpcb *)(void *)inp_ppcb,
1700 &xt.xt_tp);
1701 } else {
1702 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1703 }
1704 if (inp->inp_socket)
1705 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1706 error = SYSCTL_OUT(req, &xt, sizeof xt);
1707 }
1708 }
1709 if (!error) {
1710 /*
1711 * Give the user an updated idea of our state.
1712 * If the generation differs from what we told
1713 * her before, she knows that something happened
1714 * while we were processing this request, and it
1715 * might be necessary to retry.
1716 */
1717 bzero(&xig, sizeof(xig));
1718 xig.xig_len = sizeof xig;
1719 xig.xig_gen = tcbinfo.ipi_gencnt;
1720 xig.xig_sogen = so_gencnt;
1721 xig.xig_count = tcbinfo.ipi_count;
1722 error = SYSCTL_OUT(req, &xig, sizeof xig);
1723 }
1724 FREE(inp_list, M_TEMP);
1725 lck_rw_done(tcbinfo.ipi_lock);
1726 return error;
1727 }
1728
1729 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1730 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1731 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1732
1733
1734 static void
1735 tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp)
1736 {
1737 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1738 otp->t_dupacks = tp->t_dupacks;
1739 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1740 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1741 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1742 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1743 otp->t_state = tp->t_state;
1744 otp->t_flags = tp->t_flags;
1745 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1746 otp->snd_una = tp->snd_una;
1747 otp->snd_max = tp->snd_max;
1748 otp->snd_nxt = tp->snd_nxt;
1749 otp->snd_up = tp->snd_up;
1750 otp->snd_wl1 = tp->snd_wl1;
1751 otp->snd_wl2 = tp->snd_wl2;
1752 otp->iss = tp->iss;
1753 otp->irs = tp->irs;
1754 otp->rcv_nxt = tp->rcv_nxt;
1755 otp->rcv_adv = tp->rcv_adv;
1756 otp->rcv_wnd = tp->rcv_wnd;
1757 otp->rcv_up = tp->rcv_up;
1758 otp->snd_wnd = tp->snd_wnd;
1759 otp->snd_cwnd = tp->snd_cwnd;
1760 otp->snd_ssthresh = tp->snd_ssthresh;
1761 otp->t_maxopd = tp->t_maxopd;
1762 otp->t_rcvtime = tp->t_rcvtime;
1763 otp->t_starttime = tp->t_starttime;
1764 otp->t_rtttime = tp->t_rtttime;
1765 otp->t_rtseq = tp->t_rtseq;
1766 otp->t_rxtcur = tp->t_rxtcur;
1767 otp->t_maxseg = tp->t_maxseg;
1768 otp->t_srtt = tp->t_srtt;
1769 otp->t_rttvar = tp->t_rttvar;
1770 otp->t_rxtshift = tp->t_rxtshift;
1771 otp->t_rttmin = tp->t_rttmin;
1772 otp->t_rttupdated = tp->t_rttupdated;
1773 otp->max_sndwnd = tp->max_sndwnd;
1774 otp->t_softerror = tp->t_softerror;
1775 otp->t_oobflags = tp->t_oobflags;
1776 otp->t_iobc = tp->t_iobc;
1777 otp->snd_scale = tp->snd_scale;
1778 otp->rcv_scale = tp->rcv_scale;
1779 otp->request_r_scale = tp->request_r_scale;
1780 otp->requested_s_scale = tp->requested_s_scale;
1781 otp->ts_recent = tp->ts_recent;
1782 otp->ts_recent_age = tp->ts_recent_age;
1783 otp->last_ack_sent = tp->last_ack_sent;
1784 otp->cc_send = tp->cc_send;
1785 otp->cc_recv = tp->cc_recv;
1786 otp->snd_recover = tp->snd_recover;
1787 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1788 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1789 otp->t_badrxtwin = 0;
1790 }
1791
1792
1793 static int
1794 tcp_pcblist64 SYSCTL_HANDLER_ARGS
1795 {
1796 #pragma unused(oidp, arg1, arg2)
1797 int error, i = 0, n;
1798 struct inpcb *inp, **inp_list;
1799 struct tcpcb *tp;
1800 inp_gen_t gencnt;
1801 struct xinpgen xig;
1802
1803 /*
1804 * The process of preparing the TCB list is too time-consuming and
1805 * resource-intensive to repeat twice on every request.
1806 */
1807 lck_rw_lock_shared(tcbinfo.ipi_lock);
1808 if (req->oldptr == USER_ADDR_NULL) {
1809 n = tcbinfo.ipi_count;
1810 req->oldidx = 2 * (sizeof xig)
1811 + (n + n/8) * sizeof(struct xtcpcb64);
1812 lck_rw_done(tcbinfo.ipi_lock);
1813 return 0;
1814 }
1815
1816 if (req->newptr != USER_ADDR_NULL) {
1817 lck_rw_done(tcbinfo.ipi_lock);
1818 return EPERM;
1819 }
1820
1821 /*
1822 * OK, now we're committed to doing something.
1823 */
1824 gencnt = tcbinfo.ipi_gencnt;
1825 n = tcbinfo.ipi_count;
1826
1827 bzero(&xig, sizeof(xig));
1828 xig.xig_len = sizeof xig;
1829 xig.xig_count = n;
1830 xig.xig_gen = gencnt;
1831 xig.xig_sogen = so_gencnt;
1832 error = SYSCTL_OUT(req, &xig, sizeof xig);
1833 if (error) {
1834 lck_rw_done(tcbinfo.ipi_lock);
1835 return error;
1836 }
1837 /*
1838 * We are done if there is no pcb
1839 */
1840 if (n == 0) {
1841 lck_rw_done(tcbinfo.ipi_lock);
1842 return 0;
1843 }
1844
1845 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1846 if (inp_list == 0) {
1847 lck_rw_done(tcbinfo.ipi_lock);
1848 return ENOMEM;
1849 }
1850
1851 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1852 if (inp->inp_gencnt <= gencnt &&
1853 inp->inp_state != INPCB_STATE_DEAD)
1854 inp_list[i++] = inp;
1855 if (i >= n) break;
1856 }
1857
1858 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
1859 inp = tp->t_inpcb;
1860 if (inp->inp_gencnt <= gencnt &&
1861 inp->inp_state != INPCB_STATE_DEAD)
1862 inp_list[i++] = inp;
1863 if (i >= n) break;
1864 }
1865
1866 n = i;
1867
1868 error = 0;
1869 for (i = 0; i < n; i++) {
1870 inp = inp_list[i];
1871 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) {
1872 struct xtcpcb64 xt;
1873
1874 bzero(&xt, sizeof(xt));
1875 xt.xt_len = sizeof xt;
1876 inpcb_to_xinpcb64(inp, &xt.xt_inpcb);
1877 xt.xt_inpcb.inp_ppcb = (uint64_t)VM_KERNEL_ADDRPERM(inp->inp_ppcb);
1878 if (inp->inp_ppcb != NULL)
1879 tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt);
1880 if (inp->inp_socket)
1881 sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket);
1882 error = SYSCTL_OUT(req, &xt, sizeof xt);
1883 }
1884 }
1885 if (!error) {
1886 /*
1887 * Give the user an updated idea of our state.
1888 * If the generation differs from what we told
1889 * her before, she knows that something happened
1890 * while we were processing this request, and it
1891 * might be necessary to retry.
1892 */
1893 bzero(&xig, sizeof(xig));
1894 xig.xig_len = sizeof xig;
1895 xig.xig_gen = tcbinfo.ipi_gencnt;
1896 xig.xig_sogen = so_gencnt;
1897 xig.xig_count = tcbinfo.ipi_count;
1898 error = SYSCTL_OUT(req, &xig, sizeof xig);
1899 }
1900 FREE(inp_list, M_TEMP);
1901 lck_rw_done(tcbinfo.ipi_lock);
1902 return error;
1903 }
1904
1905 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64,
1906 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1907 tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections");
1908
1909
1910 static int
1911 tcp_pcblist_n SYSCTL_HANDLER_ARGS
1912 {
1913 #pragma unused(oidp, arg1, arg2)
1914 int error = 0;
1915
1916 error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo);
1917
1918 return error;
1919 }
1920
1921
1922 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n,
1923 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1924 tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections");
1925
1926
1927 __private_extern__ void
1928 tcp_get_ports_used(uint32_t ifindex, int protocol, uint32_t flags,
1929 bitstr_t *bitfield)
1930 {
1931 inpcb_get_ports_used(ifindex, protocol, flags,
1932 bitfield, &tcbinfo);
1933 }
1934
1935 __private_extern__ uint32_t
1936 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags)
1937 {
1938 return inpcb_count_opportunistic(ifindex, &tcbinfo, flags);
1939 }
1940
1941 __private_extern__ uint32_t
1942 tcp_find_anypcb_byaddr(struct ifaddr *ifa)
1943 {
1944 return inpcb_find_anypcb_byaddr(ifa, &tcbinfo);
1945 }
1946
1947 void
1948 tcp_ctlinput(cmd, sa, vip)
1949 int cmd;
1950 struct sockaddr *sa;
1951 void *vip;
1952 {
1953 tcp_seq icmp_tcp_seq;
1954 struct ip *ip = vip;
1955 struct in_addr faddr;
1956 struct inpcb *inp;
1957 struct tcpcb *tp;
1958
1959 void (*notify)(struct inpcb *, int) = tcp_notify;
1960
1961 faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr;
1962 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1963 return;
1964
1965 if ((unsigned)cmd >= PRC_NCMDS)
1966 return;
1967
1968 if (cmd == PRC_MSGSIZE)
1969 notify = tcp_mtudisc;
1970 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1971 cmd == PRC_UNREACH_PORT) && ip)
1972 notify = tcp_drop_syn_sent;
1973 else if (PRC_IS_REDIRECT(cmd)) {
1974 ip = 0;
1975 notify = in_rtchange;
1976 } else if (cmd == PRC_HOSTDEAD)
1977 ip = 0;
1978 /* Source quench is deprecated */
1979 else if (cmd == PRC_QUENCH)
1980 return;
1981 else if (inetctlerrmap[cmd] == 0)
1982 return;
1983 if (ip) {
1984 struct tcphdr th;
1985 struct icmp *icp;
1986
1987 icp = (struct icmp *)(void *)
1988 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1989 bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)),
1990 &th, sizeof (th));
1991 inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport,
1992 ip->ip_src, th.th_sport, 0, NULL);
1993 if (inp != NULL && inp->inp_socket != NULL) {
1994 tcp_lock(inp->inp_socket, 1, 0);
1995 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
1996 tcp_unlock(inp->inp_socket, 1, 0);
1997 return;
1998 }
1999 icmp_tcp_seq = htonl(th.th_seq);
2000 tp = intotcpcb(inp);
2001 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2002 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2003 if (cmd == PRC_MSGSIZE) {
2004
2005 /*
2006 * MTU discovery:
2007 * If we got a needfrag and there is a host route to the
2008 * original destination, and the MTU is not locked, then
2009 * set the MTU in the route to the suggested new value
2010 * (if given) and then notify as usual. The ULPs will
2011 * notice that the MTU has changed and adapt accordingly.
2012 * If no new MTU was suggested, then we guess a new one
2013 * less than the current value. If the new MTU is
2014 * unreasonably small (defined by sysctl tcp_minmss), then
2015 * we reset the MTU to the interface value and enable the
2016 * lock bit, indicating that we are no longer doing MTU
2017 * discovery.
2018 */
2019 struct rtentry *rt;
2020 int mtu;
2021 struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET,
2022 0 , { 0 }, { 0,0,0,0,0,0,0,0 } };
2023 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
2024
2025 rt = rtalloc1((struct sockaddr *)&icmpsrc, 0,
2026 RTF_CLONING | RTF_PRCLONING);
2027 if (rt != NULL) {
2028 RT_LOCK(rt);
2029 if ((rt->rt_flags & RTF_HOST) &&
2030 !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
2031 mtu = ntohs(icp->icmp_nextmtu);
2032 if (!mtu)
2033 mtu = ip_next_mtu(rt->rt_rmx.
2034 rmx_mtu, 1);
2035 #if DEBUG_MTUDISC
2036 printf("MTU for %s reduced to %d\n",
2037 inet_ntop(AF_INET,
2038 &icmpsrc.sin_addr, ipv4str,
2039 sizeof (ipv4str)), mtu);
2040 #endif
2041 if (mtu < max(296, (tcp_minmss +
2042 sizeof (struct tcpiphdr)))) {
2043 /* rt->rt_rmx.rmx_mtu =
2044 rt->rt_ifp->if_mtu; */
2045 rt->rt_rmx.rmx_locks |= RTV_MTU;
2046 } else if (rt->rt_rmx.rmx_mtu > mtu) {
2047 rt->rt_rmx.rmx_mtu = mtu;
2048 }
2049 }
2050 RT_UNLOCK(rt);
2051 rtfree(rt);
2052 }
2053 }
2054
2055 (*notify)(inp, inetctlerrmap[cmd]);
2056 }
2057 tcp_unlock(inp->inp_socket, 1, 0);
2058 }
2059 } else
2060 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
2061 }
2062
2063 #if INET6
2064 void
2065 tcp6_ctlinput(cmd, sa, d)
2066 int cmd;
2067 struct sockaddr *sa;
2068 void *d;
2069 {
2070 struct tcphdr th;
2071 void (*notify)(struct inpcb *, int) = tcp_notify;
2072 struct ip6_hdr *ip6;
2073 struct mbuf *m;
2074 struct ip6ctlparam *ip6cp = NULL;
2075 const struct sockaddr_in6 *sa6_src = NULL;
2076 int off;
2077 struct tcp_portonly {
2078 u_int16_t th_sport;
2079 u_int16_t th_dport;
2080 } *thp;
2081
2082 if (sa->sa_family != AF_INET6 ||
2083 sa->sa_len != sizeof(struct sockaddr_in6))
2084 return;
2085
2086 if ((unsigned)cmd >= PRC_NCMDS)
2087 return;
2088
2089 if (cmd == PRC_MSGSIZE)
2090 notify = tcp_mtudisc;
2091 else if (!PRC_IS_REDIRECT(cmd) && (inet6ctlerrmap[cmd] == 0))
2092 return;
2093 /* Source quench is deprecated */
2094 else if (cmd == PRC_QUENCH)
2095 return;
2096
2097 /* if the parameter is from icmp6, decode it. */
2098 if (d != NULL) {
2099 ip6cp = (struct ip6ctlparam *)d;
2100 m = ip6cp->ip6c_m;
2101 ip6 = ip6cp->ip6c_ip6;
2102 off = ip6cp->ip6c_off;
2103 sa6_src = ip6cp->ip6c_src;
2104 } else {
2105 m = NULL;
2106 ip6 = NULL;
2107 off = 0; /* fool gcc */
2108 sa6_src = &sa6_any;
2109 }
2110
2111 if (ip6) {
2112 /*
2113 * XXX: We assume that when IPV6 is non NULL,
2114 * M and OFF are valid.
2115 */
2116
2117 /* check if we can safely examine src and dst ports */
2118 if (m->m_pkthdr.len < off + sizeof(*thp))
2119 return;
2120
2121 bzero(&th, sizeof(th));
2122 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
2123
2124 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
2125 (struct sockaddr *)ip6cp->ip6c_src,
2126 th.th_sport, cmd, NULL, notify);
2127 } else {
2128 in6_pcbnotify(&tcbinfo, sa, 0,
2129 (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify);
2130 }
2131 }
2132 #endif /* INET6 */
2133
2134
2135 /*
2136 * Following is where TCP initial sequence number generation occurs.
2137 *
2138 * There are two places where we must use initial sequence numbers:
2139 * 1. In SYN-ACK packets.
2140 * 2. In SYN packets.
2141 *
2142 * The ISNs in SYN-ACK packets have no monotonicity requirement,
2143 * and should be as unpredictable as possible to avoid the possibility
2144 * of spoofing and/or connection hijacking. To satisfy this
2145 * requirement, SYN-ACK ISNs are generated via the arc4random()
2146 * function. If exact RFC 1948 compliance is requested via sysctl,
2147 * these ISNs will be generated just like those in SYN packets.
2148 *
2149 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2150 * depends on this property. In addition, these ISNs should be
2151 * unguessable so as to prevent connection hijacking. To satisfy
2152 * the requirements of this situation, the algorithm outlined in
2153 * RFC 1948 is used to generate sequence numbers.
2154 *
2155 * For more information on the theory of operation, please see
2156 * RFC 1948.
2157 *
2158 * Implementation details:
2159 *
2160 * Time is based off the system timer, and is corrected so that it
2161 * increases by one megabyte per second. This allows for proper
2162 * recycling on high speed LANs while still leaving over an hour
2163 * before rollover.
2164 *
2165 * Two sysctls control the generation of ISNs:
2166 *
2167 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2168 * between seeding of isn_secret. This is normally set to zero,
2169 * as reseeding should not be necessary.
2170 *
2171 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
2172 * strictly. When strict compliance is requested, reseeding is
2173 * disabled and SYN-ACKs will be generated in the same manner as
2174 * SYNs. Strict mode is disabled by default.
2175 *
2176 */
2177
2178 #define ISN_BYTES_PER_SECOND 1048576
2179
2180 tcp_seq
2181 tcp_new_isn(tp)
2182 struct tcpcb *tp;
2183 {
2184 u_int32_t md5_buffer[4];
2185 tcp_seq new_isn;
2186 struct timeval timenow;
2187 u_char isn_secret[32];
2188 int isn_last_reseed = 0;
2189 MD5_CTX isn_ctx;
2190
2191 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
2192 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT))
2193 && tcp_strict_rfc1948 == 0)
2194 #ifdef __APPLE__
2195 return RandomULong();
2196 #else
2197 return arc4random();
2198 #endif
2199 getmicrotime(&timenow);
2200
2201 /* Seed if this is the first use, reseed if requested. */
2202 if ((isn_last_reseed == 0) ||
2203 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
2204 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
2205 < (u_int)timenow.tv_sec))) {
2206 #ifdef __APPLE__
2207 read_random(&isn_secret, sizeof(isn_secret));
2208 #else
2209 read_random_unlimited(&isn_secret, sizeof(isn_secret));
2210 #endif
2211 isn_last_reseed = timenow.tv_sec;
2212 }
2213
2214 /* Compute the md5 hash and return the ISN. */
2215 MD5Init(&isn_ctx);
2216 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
2217 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
2218 #if INET6
2219 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2220 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2221 sizeof(struct in6_addr));
2222 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2223 sizeof(struct in6_addr));
2224 } else
2225 #endif
2226 {
2227 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2228 sizeof(struct in_addr));
2229 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2230 sizeof(struct in_addr));
2231 }
2232 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
2233 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2234 new_isn = (tcp_seq) md5_buffer[0];
2235 new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz);
2236 return new_isn;
2237 }
2238
2239
2240 /*
2241 * When a specific ICMP unreachable message is received and the
2242 * connection state is SYN-SENT, drop the connection. This behavior
2243 * is controlled by the icmp_may_rst sysctl.
2244 */
2245 void
2246 tcp_drop_syn_sent(inp, errno)
2247 struct inpcb *inp;
2248 int errno;
2249 {
2250 struct tcpcb *tp = intotcpcb(inp);
2251
2252 if (tp && tp->t_state == TCPS_SYN_SENT)
2253 tcp_drop(tp, errno);
2254 }
2255
2256 /*
2257 * When `need fragmentation' ICMP is received, update our idea of the MSS
2258 * based on the new value in the route. Also nudge TCP to send something,
2259 * since we know the packet we just sent was dropped.
2260 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2261 */
2262 void
2263 tcp_mtudisc(
2264 struct inpcb *inp,
2265 __unused int errno
2266 )
2267 {
2268 struct tcpcb *tp = intotcpcb(inp);
2269 struct rtentry *rt;
2270 struct rmxp_tao *taop;
2271 struct socket *so = inp->inp_socket;
2272 int offered;
2273 int mss;
2274 u_int32_t mtu;
2275 #if INET6
2276 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
2277 #endif /* INET6 */
2278
2279 if (tp) {
2280 #if INET6
2281 if (isipv6)
2282 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2283 else
2284 #endif /* INET6 */
2285 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2286 if (!rt || !rt->rt_rmx.rmx_mtu) {
2287 tp->t_maxopd = tp->t_maxseg =
2288 #if INET6
2289 isipv6 ? tcp_v6mssdflt :
2290 #endif /* INET6 */
2291 tcp_mssdflt;
2292
2293 /* Route locked during lookup above */
2294 if (rt != NULL)
2295 RT_UNLOCK(rt);
2296 return;
2297 }
2298 taop = rmx_taop(rt->rt_rmx);
2299 offered = taop->tao_mssopt;
2300 mtu = rt->rt_rmx.rmx_mtu;
2301
2302 /* Route locked during lookup above */
2303 RT_UNLOCK(rt);
2304
2305 #if NECP
2306 // Adjust MTU if necessary.
2307 mtu = necp_socket_get_effective_mtu(inp, mtu);
2308 #endif /* NECP */
2309
2310 mss = mtu -
2311 #if INET6
2312 (isipv6 ?
2313 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2314 #endif /* INET6 */
2315 sizeof(struct tcpiphdr)
2316 #if INET6
2317 )
2318 #endif /* INET6 */
2319 ;
2320
2321 if (offered)
2322 mss = min(mss, offered);
2323 /*
2324 * XXX - The above conditional probably violates the TCP
2325 * spec. The problem is that, since we don't know the
2326 * other end's MSS, we are supposed to use a conservative
2327 * default. But, if we do that, then MTU discovery will
2328 * never actually take place, because the conservative
2329 * default is much less than the MTUs typically seen
2330 * on the Internet today. For the moment, we'll sweep
2331 * this under the carpet.
2332 *
2333 * The conservative default might not actually be a problem
2334 * if the only case this occurs is when sending an initial
2335 * SYN with options and data to a host we've never talked
2336 * to before. Then, they will reply with an MSS value which
2337 * will get recorded and the new parameters should get
2338 * recomputed. For Further Study.
2339 */
2340 if (tp->t_maxopd <= mss)
2341 return;
2342 tp->t_maxopd = mss;
2343
2344 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2345 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2346 mss -= TCPOLEN_TSTAMP_APPA;
2347
2348 #if MPTCP
2349 mss -= mptcp_adj_mss(tp, TRUE);
2350 #endif
2351 if (so->so_snd.sb_hiwat < mss)
2352 mss = so->so_snd.sb_hiwat;
2353
2354 tp->t_maxseg = mss;
2355
2356 /*
2357 * Reset the slow-start flight size as it may depends on the new MSS
2358 */
2359 if (CC_ALGO(tp)->cwnd_init != NULL)
2360 CC_ALGO(tp)->cwnd_init(tp);
2361 tcpstat.tcps_mturesent++;
2362 tp->t_rtttime = 0;
2363 tp->snd_nxt = tp->snd_una;
2364 tcp_output(tp);
2365 }
2366 }
2367
2368 /*
2369 * Look-up the routing entry to the peer of this inpcb. If no route
2370 * is found and it cannot be allocated the return NULL. This routine
2371 * is called by TCP routines that access the rmx structure and by tcp_mss
2372 * to get the interface MTU. If a route is found, this routine will
2373 * hold the rtentry lock; the caller is responsible for unlocking.
2374 */
2375 struct rtentry *
2376 tcp_rtlookup(inp, input_ifscope)
2377 struct inpcb *inp;
2378 unsigned int input_ifscope;
2379 {
2380 struct route *ro;
2381 struct rtentry *rt;
2382 struct tcpcb *tp;
2383
2384 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2385
2386 ro = &inp->inp_route;
2387 if ((rt = ro->ro_rt) != NULL)
2388 RT_LOCK(rt);
2389
2390 if (ROUTE_UNUSABLE(ro)) {
2391 if (rt != NULL) {
2392 RT_UNLOCK(rt);
2393 rt = NULL;
2394 }
2395 ROUTE_RELEASE(ro);
2396 /* No route yet, so try to acquire one */
2397 if (inp->inp_faddr.s_addr != INADDR_ANY) {
2398 unsigned int ifscope;
2399
2400 ro->ro_dst.sa_family = AF_INET;
2401 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
2402 ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr =
2403 inp->inp_faddr;
2404
2405 /*
2406 * If the socket was bound to an interface, then
2407 * the bound-to-interface takes precedence over
2408 * the inbound interface passed in by the caller
2409 * (if we get here as part of the output path then
2410 * input_ifscope is IFSCOPE_NONE).
2411 */
2412 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2413 inp->inp_boundifp->if_index : input_ifscope;
2414
2415 rtalloc_scoped(ro, ifscope);
2416 if ((rt = ro->ro_rt) != NULL)
2417 RT_LOCK(rt);
2418 }
2419 }
2420 if (rt != NULL)
2421 RT_LOCK_ASSERT_HELD(rt);
2422
2423 /*
2424 * Update MTU discovery determination. Don't do it if:
2425 * 1) it is disabled via the sysctl
2426 * 2) the route isn't up
2427 * 3) the MTU is locked (if it is, then discovery has been
2428 * disabled)
2429 */
2430
2431 tp = intotcpcb(inp);
2432
2433 if (!path_mtu_discovery || ((rt != NULL) &&
2434 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2435 tp->t_flags &= ~TF_PMTUD;
2436 else
2437 tp->t_flags |= TF_PMTUD;
2438
2439 #if CONFIG_IFEF_NOWINDOWSCALE
2440 if (tcp_obey_ifef_nowindowscale &&
2441 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2442 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2443 /* Window scaling is enabled on this interface */
2444 tp->t_flags &= ~TF_REQ_SCALE;
2445 }
2446 #endif
2447
2448 if (rt != NULL && rt->rt_ifp != NULL) {
2449 somultipages(inp->inp_socket,
2450 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2451 tcp_set_tso(tp, rt->rt_ifp);
2452 soif2kcl(inp->inp_socket,
2453 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2454 }
2455
2456 /* Note if the peer is local */
2457 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2458 (rt->rt_gateway->sa_family == AF_LINK ||
2459 rt->rt_ifp->if_flags & IFF_LOOPBACK ||
2460 in_localaddr(inp->inp_faddr))) {
2461 tp->t_flags |= TF_LOCAL;
2462 }
2463
2464 /*
2465 * Caller needs to call RT_UNLOCK(rt).
2466 */
2467 return rt;
2468 }
2469
2470 #if INET6
2471 struct rtentry *
2472 tcp_rtlookup6(inp, input_ifscope)
2473 struct inpcb *inp;
2474 unsigned int input_ifscope;
2475 {
2476 struct route_in6 *ro6;
2477 struct rtentry *rt;
2478 struct tcpcb *tp;
2479
2480 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2481
2482 ro6 = &inp->in6p_route;
2483 if ((rt = ro6->ro_rt) != NULL)
2484 RT_LOCK(rt);
2485
2486 if (ROUTE_UNUSABLE(ro6)) {
2487 if (rt != NULL) {
2488 RT_UNLOCK(rt);
2489 rt = NULL;
2490 }
2491 ROUTE_RELEASE(ro6);
2492 /* No route yet, so try to acquire one */
2493 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
2494 struct sockaddr_in6 *dst6;
2495 unsigned int ifscope;
2496
2497 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst;
2498 dst6->sin6_family = AF_INET6;
2499 dst6->sin6_len = sizeof(*dst6);
2500 dst6->sin6_addr = inp->in6p_faddr;
2501
2502 /*
2503 * If the socket was bound to an interface, then
2504 * the bound-to-interface takes precedence over
2505 * the inbound interface passed in by the caller
2506 * (if we get here as part of the output path then
2507 * input_ifscope is IFSCOPE_NONE).
2508 */
2509 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2510 inp->inp_boundifp->if_index : input_ifscope;
2511
2512 rtalloc_scoped((struct route *)ro6, ifscope);
2513 if ((rt = ro6->ro_rt) != NULL)
2514 RT_LOCK(rt);
2515 }
2516 }
2517 if (rt != NULL)
2518 RT_LOCK_ASSERT_HELD(rt);
2519
2520 /*
2521 * Update path MTU Discovery determination
2522 * while looking up the route:
2523 * 1) we have a valid route to the destination
2524 * 2) the MTU is not locked (if it is, then discovery has been
2525 * disabled)
2526 */
2527
2528
2529 tp = intotcpcb(inp);
2530
2531 /*
2532 * Update MTU discovery determination. Don't do it if:
2533 * 1) it is disabled via the sysctl
2534 * 2) the route isn't up
2535 * 3) the MTU is locked (if it is, then discovery has been
2536 * disabled)
2537 */
2538
2539 if (!path_mtu_discovery || ((rt != NULL) &&
2540 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2541 tp->t_flags &= ~TF_PMTUD;
2542 else
2543 tp->t_flags |= TF_PMTUD;
2544
2545 #if CONFIG_IFEF_NOWINDOWSCALE
2546 if (tcp_obey_ifef_nowindowscale &&
2547 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2548 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2549 /* Window scaling is not enabled on this interface */
2550 tp->t_flags &= ~TF_REQ_SCALE;
2551 }
2552 #endif
2553
2554 if (rt != NULL && rt->rt_ifp != NULL) {
2555 somultipages(inp->inp_socket,
2556 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2557 tcp_set_tso(tp, rt->rt_ifp);
2558 soif2kcl(inp->inp_socket,
2559 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2560 }
2561
2562 /* Note if the peer is local */
2563 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2564 (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) ||
2565 IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) ||
2566 rt->rt_gateway->sa_family == AF_LINK ||
2567 in6_localaddr(&inp->in6p_faddr))) {
2568 tp->t_flags |= TF_LOCAL;
2569 }
2570
2571 /*
2572 * Caller needs to call RT_UNLOCK(rt).
2573 */
2574 return rt;
2575 }
2576 #endif /* INET6 */
2577
2578 #if IPSEC
2579 /* compute ESP/AH header size for TCP, including outer IP header. */
2580 size_t
2581 ipsec_hdrsiz_tcp(tp)
2582 struct tcpcb *tp;
2583 {
2584 struct inpcb *inp;
2585 struct mbuf *m;
2586 size_t hdrsiz;
2587 struct ip *ip;
2588 #if INET6
2589 struct ip6_hdr *ip6 = NULL;
2590 #endif /* INET6 */
2591 struct tcphdr *th;
2592
2593 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
2594 return 0;
2595 MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */
2596 if (!m)
2597 return 0;
2598
2599 #if INET6
2600 if ((inp->inp_vflag & INP_IPV6) != 0) {
2601 ip6 = mtod(m, struct ip6_hdr *);
2602 th = (struct tcphdr *)(void *)(ip6 + 1);
2603 m->m_pkthdr.len = m->m_len =
2604 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2605 tcp_fillheaders(tp, ip6, th);
2606 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2607 } else
2608 #endif /* INET6 */
2609 {
2610 ip = mtod(m, struct ip *);
2611 th = (struct tcphdr *)(ip + 1);
2612 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2613 tcp_fillheaders(tp, ip, th);
2614 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2615 }
2616 m_free(m);
2617 return hdrsiz;
2618 }
2619 #endif /*IPSEC*/
2620
2621 /*
2622 * Return a pointer to the cached information about the remote host.
2623 * The cached information is stored in the protocol specific part of
2624 * the route metrics.
2625 */
2626 struct rmxp_tao *
2627 tcp_gettaocache(inp)
2628 struct inpcb *inp;
2629 {
2630 struct rtentry *rt;
2631 struct rmxp_tao *taop;
2632
2633 #if INET6
2634 if ((inp->inp_vflag & INP_IPV6) != 0)
2635 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2636 else
2637 #endif /* INET6 */
2638 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2639
2640 /* Make sure this is a host route and is up. */
2641 if (rt == NULL ||
2642 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) {
2643 /* Route locked during lookup above */
2644 if (rt != NULL)
2645 RT_UNLOCK(rt);
2646 return NULL;
2647 }
2648
2649 taop = rmx_taop(rt->rt_rmx);
2650 /* Route locked during lookup above */
2651 RT_UNLOCK(rt);
2652 return (taop);
2653 }
2654
2655 /*
2656 * Clear all the TAO cache entries, called from tcp_init.
2657 *
2658 * XXX
2659 * This routine is just an empty one, because we assume that the routing
2660 * routing tables are initialized at the same time when TCP, so there is
2661 * nothing in the cache left over.
2662 */
2663 static void
2664 tcp_cleartaocache()
2665 {
2666 }
2667
2668 int
2669 tcp_lock(struct socket *so, int refcount, void *lr)
2670 {
2671 void *lr_saved;
2672
2673 if (lr == NULL)
2674 lr_saved = __builtin_return_address(0);
2675 else
2676 lr_saved = lr;
2677
2678 if (so->so_pcb != NULL) {
2679 lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2680 } else {
2681 panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n",
2682 so, lr_saved, solockhistory_nr(so));
2683 /* NOTREACHED */
2684 }
2685
2686 if (so->so_usecount < 0) {
2687 panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n",
2688 so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so));
2689 /* NOTREACHED */
2690 }
2691 if (refcount)
2692 so->so_usecount++;
2693 so->lock_lr[so->next_lock_lr] = lr_saved;
2694 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX;
2695 return (0);
2696 }
2697
2698 int
2699 tcp_unlock(struct socket *so, int refcount, void *lr)
2700 {
2701 void *lr_saved;
2702
2703 if (lr == NULL)
2704 lr_saved = __builtin_return_address(0);
2705 else
2706 lr_saved = lr;
2707
2708 #ifdef MORE_TCPLOCK_DEBUG
2709 printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x "
2710 "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so),
2711 (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb),
2712 (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)),
2713 so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved));
2714 #endif
2715 if (refcount)
2716 so->so_usecount--;
2717
2718 if (so->so_usecount < 0) {
2719 panic("tcp_unlock: so=%p usecount=%x lrh= %s\n",
2720 so, so->so_usecount, solockhistory_nr(so));
2721 /* NOTREACHED */
2722 }
2723 if (so->so_pcb == NULL) {
2724 panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n",
2725 so, so->so_usecount, lr_saved, solockhistory_nr(so));
2726 /* NOTREACHED */
2727 } else {
2728 lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx,
2729 LCK_MTX_ASSERT_OWNED);
2730 so->unlock_lr[so->next_unlock_lr] = lr_saved;
2731 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX;
2732 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2733 }
2734 return (0);
2735 }
2736
2737 lck_mtx_t *
2738 tcp_getlock(
2739 struct socket *so,
2740 __unused int locktype)
2741 {
2742 struct inpcb *inp = sotoinpcb(so);
2743
2744 if (so->so_pcb) {
2745 if (so->so_usecount < 0)
2746 panic("tcp_getlock: so=%p usecount=%x lrh= %s\n",
2747 so, so->so_usecount, solockhistory_nr(so));
2748 return(&inp->inpcb_mtx);
2749 }
2750 else {
2751 panic("tcp_getlock: so=%p NULL so_pcb %s\n",
2752 so, solockhistory_nr(so));
2753 return (so->so_proto->pr_domain->dom_mtx);
2754 }
2755 }
2756
2757 /*
2758 * Determine if we can grow the recieve socket buffer to avoid sending
2759 * a zero window update to the peer. We allow even socket buffers that
2760 * have fixed size (set by the application) to grow if the resource
2761 * constraints are met. They will also be trimmed after the application
2762 * reads data.
2763 */
2764 static void
2765 tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb)
2766 {
2767 u_int32_t rcvbufinc = tp->t_maxseg << 4;
2768 u_int32_t rcvbuf = sb->sb_hiwat;
2769 struct socket *so = tp->t_inpcb->inp_socket;
2770
2771 /*
2772 * If message delivery is enabled, do not count
2773 * unordered bytes in receive buffer towards hiwat
2774 */
2775 if (so->so_flags & SOF_ENABLE_MSGS)
2776 rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes;
2777
2778 if (tcp_do_autorcvbuf == 1 &&
2779 tcp_cansbgrow(sb) &&
2780 (tp->t_flags & TF_SLOWLINK) == 0 &&
2781 (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 &&
2782 (rcvbuf - sb->sb_cc) < rcvbufinc &&
2783 rcvbuf < tcp_autorcvbuf_max &&
2784 (sb->sb_idealsize > 0 &&
2785 sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) {
2786 sbreserve(sb,
2787 min((sb->sb_hiwat + rcvbufinc), tcp_autorcvbuf_max));
2788 }
2789 }
2790
2791 int32_t
2792 tcp_sbspace(struct tcpcb *tp)
2793 {
2794 struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv;
2795 u_int32_t rcvbuf = sb->sb_hiwat;
2796 int32_t space;
2797 struct socket *so = tp->t_inpcb->inp_socket;
2798 int32_t pending = 0;
2799
2800 /*
2801 * If message delivery is enabled, do not count
2802 * unordered bytes in receive buffer towards hiwat mark.
2803 * This value is used to return correct rwnd that does
2804 * not reflect the extra unordered bytes added to the
2805 * receive socket buffer.
2806 */
2807 if (so->so_flags & SOF_ENABLE_MSGS)
2808 rcvbuf = rcvbuf - so->so_msg_state->msg_uno_bytes;
2809
2810 tcp_sbrcv_grow_rwin(tp, sb);
2811
2812 space = ((int32_t) imin((rcvbuf - sb->sb_cc),
2813 (sb->sb_mbmax - sb->sb_mbcnt)));
2814 if (space < 0)
2815 space = 0;
2816
2817 #if CONTENT_FILTER
2818 /* Compensate for data being processed by content filters */
2819 pending = cfil_sock_data_space(sb);
2820 #endif /* CONTENT_FILTER */
2821 if (pending > space)
2822 space = 0;
2823 else
2824 space -= pending;
2825
2826 /* Avoid increasing window size if the current window
2827 * is already very low, we could be in "persist" mode and
2828 * we could break some apps (see rdar://5409343)
2829 */
2830
2831 if (space < tp->t_maxseg)
2832 return space;
2833
2834 /* Clip window size for slower link */
2835
2836 if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 )
2837 return imin(space, slowlink_wsize);
2838
2839 return space;
2840 }
2841 /*
2842 * Checks TCP Segment Offloading capability for a given connection
2843 * and interface pair.
2844 */
2845 void
2846 tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp)
2847 {
2848 #if INET6
2849 struct inpcb *inp;
2850 int isipv6;
2851 #endif /* INET6 */
2852 #if MPTCP
2853 /*
2854 * We can't use TSO if this tcpcb belongs to an MPTCP session.
2855 */
2856 if (tp->t_mpflags & TMPF_MPTCP_TRUE) {
2857 tp->t_flags &= ~TF_TSO;
2858 return;
2859 }
2860 #endif
2861 #if INET6
2862 inp = tp->t_inpcb;
2863 isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2864
2865 if (isipv6) {
2866 if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV6)) {
2867 tp->t_flags |= TF_TSO;
2868 if (ifp->if_tso_v6_mtu != 0)
2869 tp->tso_max_segment_size = ifp->if_tso_v6_mtu;
2870 else
2871 tp->tso_max_segment_size = TCP_MAXWIN;
2872 } else
2873 tp->t_flags &= ~TF_TSO;
2874
2875 } else
2876 #endif /* INET6 */
2877
2878 {
2879 if (ifp && (ifp->if_hwassist & IFNET_TSO_IPV4)) {
2880 tp->t_flags |= TF_TSO;
2881 if (ifp->if_tso_v4_mtu != 0)
2882 tp->tso_max_segment_size = ifp->if_tso_v4_mtu;
2883 else
2884 tp->tso_max_segment_size = TCP_MAXWIN;
2885 } else
2886 tp->t_flags &= ~TF_TSO;
2887 }
2888 }
2889
2890 #define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC)
2891
2892 /* Function to calculate the tcp clock. The tcp clock will get updated
2893 * at the boundaries of the tcp layer. This is done at 3 places:
2894 * 1. Right before processing an input tcp packet
2895 * 2. Whenever a connection wants to access the network using tcp_usrreqs
2896 * 3. When a tcp timer fires or before tcp slow timeout
2897 *
2898 */
2899
2900 void
2901 calculate_tcp_clock()
2902 {
2903 struct timeval tv = tcp_uptime;
2904 struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC};
2905 struct timeval now, hold_now;
2906 uint32_t incr = 0;
2907
2908 microuptime(&now);
2909
2910 /*
2911 * Update coarse-grained networking timestamp (in sec.); the idea
2912 * is to update the counter returnable via net_uptime() when
2913 * we read time.
2914 */
2915 net_update_uptime_secs(now.tv_sec);
2916
2917 timevaladd(&tv, &interval);
2918 if (timevalcmp(&now, &tv, >)) {
2919 /* time to update the clock */
2920 lck_spin_lock(tcp_uptime_lock);
2921 if (timevalcmp(&tcp_uptime, &now, >=)) {
2922 /* clock got updated while waiting for the lock */
2923 lck_spin_unlock(tcp_uptime_lock);
2924 return;
2925 }
2926
2927 microuptime(&now);
2928 hold_now = now;
2929 tv = tcp_uptime;
2930 timevalsub(&now, &tv);
2931
2932 incr = TIMEVAL_TO_TCPHZ(now);
2933 if (incr > 0) {
2934 tcp_uptime = hold_now;
2935 tcp_now += incr;
2936 }
2937
2938 lck_spin_unlock(tcp_uptime_lock);
2939 }
2940 return;
2941 }
2942
2943 /* Compute receive window scaling that we are going to request
2944 * for this connection based on sb_hiwat. Try to leave some
2945 * room to potentially increase the window size upto a maximum
2946 * defined by the constant tcp_autorcvbuf_max.
2947 */
2948 void
2949 tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) {
2950 u_int32_t maxsockbufsize;
2951 if (!tcp_do_rfc1323) {
2952 tp->request_r_scale = 0;
2953 return;
2954 }
2955
2956 tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale);
2957 maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ?
2958 so->so_rcv.sb_hiwat : tcp_autorcvbuf_max;
2959
2960 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
2961 (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize)
2962 tp->request_r_scale++;
2963 tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT);
2964
2965 }
2966
2967 int
2968 tcp_notsent_lowat_check(struct socket *so) {
2969 struct inpcb *inp = sotoinpcb(so);
2970 struct tcpcb *tp = NULL;
2971 int notsent = 0;
2972 if (inp != NULL) {
2973 tp = intotcpcb(inp);
2974 }
2975
2976 notsent = so->so_snd.sb_cc -
2977 (tp->snd_nxt - tp->snd_una);
2978
2979 /* When we send a FIN or SYN, not_sent can be negative.
2980 * In that case also we need to send a write event to the
2981 * process if it is waiting. In the FIN case, it will
2982 * get an error from send because cantsendmore will be set.
2983 */
2984 if (notsent <= tp->t_notsent_lowat) {
2985 return(1);
2986 }
2987
2988 /* When Nagle's algorithm is not disabled, it is better
2989 * to wakeup the client until there is atleast one
2990 * maxseg of data to write.
2991 */
2992 if ((tp->t_flags & TF_NODELAY) == 0 &&
2993 notsent > 0 && notsent < tp->t_maxseg) {
2994 return(1);
2995 }
2996 return(0);
2997 }
2998
2999 void
3000 tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end) {
3001 struct tcp_rxt_seg *rxseg = NULL, *prev = NULL, *next = NULL;
3002 u_int32_t rxcount = 0;
3003
3004 if (SLIST_EMPTY(&tp->t_rxt_segments))
3005 tp->t_dsack_lastuna = tp->snd_una;
3006 /*
3007 * First check if there is a segment already existing for this
3008 * sequence space.
3009 */
3010
3011 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3012 if (SEQ_GT(rxseg->rx_start, start))
3013 break;
3014 prev = rxseg;
3015 }
3016 next = rxseg;
3017
3018 /* check if prev seg is for this sequence */
3019 if (prev != NULL && SEQ_LEQ(prev->rx_start, start) &&
3020 SEQ_GEQ(prev->rx_end, end)) {
3021 prev->rx_count++;
3022 return;
3023 }
3024
3025 /*
3026 * There are a couple of possibilities at this point.
3027 * 1. prev overlaps with the beginning of this sequence
3028 * 2. next overlaps with the end of this sequence
3029 * 3. there is no overlap.
3030 */
3031
3032 if (prev != NULL && SEQ_GT(prev->rx_end, start)) {
3033 if (prev->rx_start == start && SEQ_GT(end, prev->rx_end)) {
3034 start = prev->rx_end + 1;
3035 prev->rx_count++;
3036 } else {
3037 prev->rx_end = (start - 1);
3038 rxcount = prev->rx_count;
3039 }
3040 }
3041
3042 if (next != NULL && SEQ_LT(next->rx_start, end)) {
3043 if (SEQ_LEQ(next->rx_end, end)) {
3044 end = next->rx_start - 1;
3045 next->rx_count++;
3046 } else {
3047 next->rx_start = end + 1;
3048 rxcount = next->rx_count;
3049 }
3050 }
3051 if (!SEQ_LT(start, end))
3052 return;
3053
3054 rxseg = (struct tcp_rxt_seg *) zalloc(tcp_rxt_seg_zone);
3055 if (rxseg == NULL) {
3056 return;
3057 }
3058 bzero(rxseg, sizeof(*rxseg));
3059 rxseg->rx_start = start;
3060 rxseg->rx_end = end;
3061 rxseg->rx_count = rxcount + 1;
3062
3063 if (prev != NULL) {
3064 SLIST_INSERT_AFTER(prev, rxseg, rx_link);
3065 } else {
3066 SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link);
3067 }
3068 return;
3069 }
3070
3071 struct tcp_rxt_seg *
3072 tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3073 {
3074 struct tcp_rxt_seg *rxseg;
3075 if (SLIST_EMPTY(&tp->t_rxt_segments))
3076 return (NULL);
3077
3078 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3079 if (SEQ_LEQ(rxseg->rx_start, start) &&
3080 SEQ_GEQ(rxseg->rx_end, end))
3081 return (rxseg);
3082 if (SEQ_GT(rxseg->rx_start, start))
3083 break;
3084 }
3085 return (NULL);
3086 }
3087
3088 void
3089 tcp_rxtseg_clean(struct tcpcb *tp)
3090 {
3091 struct tcp_rxt_seg *rxseg, *next;
3092
3093 SLIST_FOREACH_SAFE(rxseg, &tp->t_rxt_segments, rx_link, next) {
3094 SLIST_REMOVE(&tp->t_rxt_segments, rxseg,
3095 tcp_rxt_seg, rx_link);
3096 zfree(tcp_rxt_seg_zone, rxseg);
3097 }
3098 tp->t_dsack_lastuna = tp->snd_max;
3099 }
3100
3101 boolean_t
3102 tcp_rxtseg_detect_bad_rexmt(struct tcpcb *tp, tcp_seq th_ack)
3103 {
3104 boolean_t bad_rexmt;
3105 struct tcp_rxt_seg *rxseg;
3106
3107 if (SLIST_EMPTY(&tp->t_rxt_segments))
3108 return (FALSE);
3109
3110 /*
3111 * If all of the segments in this window are not cumulatively
3112 * acknowledged, then there can still be undetected packet loss.
3113 * Do not restore congestion window in that case.
3114 */
3115 if (SEQ_LT(th_ack, tp->snd_recover))
3116 return (FALSE);
3117
3118 bad_rexmt = TRUE;
3119 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3120 if (rxseg->rx_count > 1 ||
3121 !(rxseg->rx_flags & TCP_RXT_SPURIOUS)) {
3122 bad_rexmt = FALSE;
3123 break;
3124 }
3125 }
3126 return (bad_rexmt);
3127 }
3128
3129 boolean_t
3130 tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp)
3131 {
3132 boolean_t dsack_for_tlp = FALSE;
3133 struct tcp_rxt_seg *rxseg;
3134 if (SLIST_EMPTY(&tp->t_rxt_segments))
3135 return (FALSE);
3136
3137 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3138 if (rxseg->rx_count == 1 &&
3139 SLIST_NEXT(rxseg,rx_link) == NULL &&
3140 (rxseg->rx_flags & TCP_RXT_DSACK_FOR_TLP)) {
3141 dsack_for_tlp = TRUE;
3142 break;
3143 }
3144 }
3145 return (dsack_for_tlp);
3146 }
3147
3148 u_int32_t
3149 tcp_rxtseg_total_size(struct tcpcb *tp) {
3150 struct tcp_rxt_seg *rxseg;
3151 u_int32_t total_size = 0;
3152
3153 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3154 total_size += (rxseg->rx_end - rxseg->rx_start) + 1;
3155 }
3156 return (total_size);
3157 }
3158
3159 void
3160 tcp_get_connectivity_status(struct tcpcb *tp,
3161 struct tcp_conn_status *connstatus)
3162 {
3163 if (tp == NULL || connstatus == NULL)
3164 return;
3165 bzero(connstatus, sizeof(*connstatus));
3166 if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) {
3167 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3168 connstatus->write_probe_failed = 1;
3169 } else {
3170 connstatus->conn_probe_failed = 1;
3171 }
3172 }
3173 if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX)
3174 connstatus->read_probe_failed = 1;
3175 if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL
3176 && (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY))
3177 connstatus->probe_activated = 1;
3178 return;
3179 }
3180
3181 boolean_t
3182 tfo_enabled(const struct tcpcb *tp)
3183 {
3184 return !!(tp->t_flagsext & TF_FASTOPEN);
3185 }
3186
3187 void
3188 tcp_disable_tfo(struct tcpcb *tp)
3189 {
3190 tp->t_flagsext &= ~TF_FASTOPEN;
3191 }
3192
mirror of XNU