]> git.saurik.com Git - apple/xnu.git/blob - bsd/netinet/tcp_subr.c
projects / apple / xnu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
xnu-7195.101.1.tar.gz
[apple/xnu.git] / bsd / netinet / tcp_subr.c
1 /*
2 * Copyright (c) 2000-2020 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 */
62 /*
63 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
64 * support for mandatory and extensible security protections. This notice
65 * is included in support of clause 2.2 (b) of the Apple Public License,
66 * Version 2.0.
67 */
68
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/kernel.h>
72 #include <sys/sysctl.h>
73 #include <sys/malloc.h>
74 #include <sys/mbuf.h>
75 #include <sys/domain.h>
76 #include <sys/proc.h>
77 #include <sys/kauth.h>
78 #include <sys/socket.h>
79 #include <sys/socketvar.h>
80 #include <sys/protosw.h>
81 #include <sys/random.h>
82 #include <sys/syslog.h>
83 #include <sys/mcache.h>
84 #include <kern/locks.h>
85 #include <kern/zalloc.h>
86
87 #include <dev/random/randomdev.h>
88
89 #include <net/route.h>
90 #include <net/if.h>
91 #include <net/content_filter.h>
92 #include <net/ntstat.h>
93 #include <net/multi_layer_pkt_log.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 #include <netinet/ip6.h>
102 #include <netinet/icmp6.h>
103 #include <netinet/in_pcb.h>
104 #include <netinet6/in6_pcb.h>
105 #include <netinet/in_var.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/icmp_var.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet/mptcp_var.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_var.h>
115 #include <netinet/tcp_cc.h>
116 #include <netinet/tcp_cache.h>
117 #include <kern/thread_call.h>
118
119 #include <netinet6/tcp6_var.h>
120 #include <netinet/tcpip.h>
121 #if TCPDEBUG
122 #include <netinet/tcp_debug.h>
123 #endif
124 #include <netinet/tcp_log.h>
125
126 #include <netinet6/ip6protosw.h>
127
128 #if IPSEC
129 #include <netinet6/ipsec.h>
130 #include <netinet6/ipsec6.h>
131 #endif /* IPSEC */
132
133 #if NECP
134 #include <net/necp.h>
135 #endif /* NECP */
136
137 #undef tcp_minmssoverload
138
139 #include <corecrypto/ccaes.h>
140 #include <libkern/crypto/aes.h>
141 #include <libkern/crypto/md5.h>
142 #include <sys/kdebug.h>
143 #include <mach/sdt.h>
144 #include <atm/atm_internal.h>
145 #include <pexpert/pexpert.h>
146
147 #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2))
148
149 static tcp_cc tcp_ccgen;
150
151 extern struct tcptimerlist tcp_timer_list;
152 extern struct tcptailq tcp_tw_tailq;
153
154 SYSCTL_SKMEM_TCP_INT(TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED,
155 int, tcp_mssdflt, TCP_MSS, "Default TCP Maximum Segment Size");
156
157 SYSCTL_SKMEM_TCP_INT(TCPCTL_V6MSSDFLT, v6mssdflt,
158 CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_v6mssdflt, TCP6_MSS,
159 "Default TCP Maximum Segment Size for IPv6");
160
161 int tcp_sysctl_fastopenkey(struct sysctl_oid *, void *, int,
162 struct sysctl_req *);
163 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, fastopen_key, CTLTYPE_STRING | CTLFLAG_WR,
164 0, 0, tcp_sysctl_fastopenkey, "S", "TCP Fastopen key");
165
166 /* Current count of half-open TFO connections */
167 int tcp_tfo_halfcnt = 0;
168
169 /* Maximum of half-open TFO connection backlog */
170 SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen_backlog,
171 CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_tfo_backlog, 10,
172 "Backlog queue for half-open TFO connections");
173
174 SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED,
175 int, tcp_fastopen, TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER,
176 "Enable TCP Fastopen (RFC 7413)");
177
178 SYSCTL_SKMEM_TCP_INT(OID_AUTO, now_init, CTLFLAG_RD | CTLFLAG_LOCKED,
179 uint32_t, tcp_now_init, 0, "Initial tcp now value");
180
181 SYSCTL_SKMEM_TCP_INT(OID_AUTO, microuptime_init, CTLFLAG_RD | CTLFLAG_LOCKED,
182 uint32_t, tcp_microuptime_init, 0, "Initial tcp uptime value in micro seconds");
183
184 /*
185 * Minimum MSS we accept and use. This prevents DoS attacks where
186 * we are forced to a ridiculous low MSS like 20 and send hundreds
187 * of packets instead of one. The effect scales with the available
188 * bandwidth and quickly saturates the CPU and network interface
189 * with packet generation and sending. Set to zero to disable MINMSS
190 * checking. This setting prevents us from sending too small packets.
191 */
192 SYSCTL_SKMEM_TCP_INT(OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED,
193 int, tcp_minmss, TCP_MINMSS, "Minmum TCP Maximum Segment Size");
194
195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
196 &tcbinfo.ipi_count, 0, "Number of active PCBs");
197
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
199 &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state");
200
201 SYSCTL_SKMEM_TCP_INT(OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED,
202 static int, icmp_may_rst, 1,
203 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
204
205 static int tcp_strict_rfc1948 = 0;
206 static int tcp_isn_reseed_interval = 0;
207 #if (DEVELOPMENT || DEBUG)
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW | CTLFLAG_LOCKED,
209 &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly");
210
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval,
212 CTLFLAG_RW | CTLFLAG_LOCKED,
213 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
214 #endif /* (DEVELOPMENT || DEBUG) */
215
216 SYSCTL_SKMEM_TCP_INT(OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED,
217 int, tcp_TCPTV_MIN, 100, "min rtt value allowed");
218
219 SYSCTL_SKMEM_TCP_INT(OID_AUTO, rexmt_slop, CTLFLAG_RW,
220 int, tcp_rexmt_slop, TCPTV_REXMTSLOP, "Slop added to retransmit timeout");
221
222 SYSCTL_SKMEM_TCP_INT(OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED,
223 __private_extern__ int, tcp_use_randomport, 0,
224 "Randomize TCP port numbers");
225
226 SYSCTL_SKMEM_TCP_INT(OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED,
227 __private_extern__ int, tcp_win_scale, 3, "Window scaling factor");
228
229 #if (DEVELOPMENT || DEBUG)
230 SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache,
231 CTLFLAG_RW | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1,
232 "Initalize RTT from route cache");
233 #else
234 SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache,
235 CTLFLAG_RD | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1,
236 "Initalize RTT from route cache");
237 #endif /* (DEVELOPMENT || DEBUG) */
238
239 static int tso_debug = 0;
240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso_debug, CTLFLAG_RW | CTLFLAG_LOCKED,
241 &tso_debug, 0, "TSO verbosity");
242
243 static void tcp_notify(struct inpcb *, int);
244
245 struct zone *sack_hole_zone;
246 struct zone *tcp_reass_zone;
247 struct zone *tcp_bwmeas_zone;
248 struct zone *tcp_rxt_seg_zone;
249
250 extern int slowlink_wsize; /* window correction for slow links */
251 extern int path_mtu_discovery;
252
253 static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb);
254
255 #define TCP_BWMEAS_BURST_MINSIZE 6
256 #define TCP_BWMEAS_BURST_MAXSIZE 25
257
258 /*
259 * Target size of TCP PCB hash tables. Must be a power of two.
260 *
261 * Note that this can be overridden by the kernel environment
262 * variable net.inet.tcp.tcbhashsize
263 */
264 #ifndef TCBHASHSIZE
265 #define TCBHASHSIZE CONFIG_TCBHASHSIZE
266 #endif
267
268 __private_extern__ int tcp_tcbhashsize = TCBHASHSIZE;
269 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED,
270 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
271
272 /*
273 * This is the actual shape of what we allocate using the zone
274 * allocator. Doing it this way allows us to protect both structures
275 * using the same generation count, and also eliminates the overhead
276 * of allocating tcpcbs separately. By hiding the structure here,
277 * we avoid changing most of the rest of the code (although it needs
278 * to be changed, eventually, for greater efficiency).
279 */
280 #define ALIGNMENT 32
281 struct inp_tp {
282 struct inpcb inp;
283 struct tcpcb tcb __attribute__((aligned(ALIGNMENT)));
284 };
285 #undef ALIGNMENT
286
287 int get_inpcb_str_size(void);
288 int get_tcp_str_size(void);
289
290 os_log_t tcp_mpkl_log_object = NULL;
291
292 static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *);
293
294 static lck_attr_t *tcp_uptime_mtx_attr = NULL;
295 static lck_grp_t *tcp_uptime_mtx_grp = NULL;
296 static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL;
297 int tcp_notsent_lowat_check(struct socket *so);
298 static void tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs,
299 struct if_lim_perf_stat *stat);
300 static void tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs,
301 struct if_tcp_ecn_perf_stat *stat);
302
303 static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */
304
305 void
306 tcp_tfo_gen_cookie(struct inpcb *inp, u_char *out, size_t blk_size)
307 {
308 u_char in[CCAES_BLOCK_SIZE];
309 int isipv6 = inp->inp_vflag & INP_IPV6;
310
311 VERIFY(blk_size == CCAES_BLOCK_SIZE);
312
313 bzero(&in[0], CCAES_BLOCK_SIZE);
314 bzero(&out[0], CCAES_BLOCK_SIZE);
315
316 if (isipv6) {
317 memcpy(in, &inp->in6p_faddr, sizeof(struct in6_addr));
318 } else {
319 memcpy(in, &inp->inp_faddr, sizeof(struct in_addr));
320 }
321
322 aes_encrypt_cbc(in, NULL, 1, out, &tfo_ctx);
323 }
324
325 __private_extern__ int
326 tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1,
327 __unused int arg2, struct sysctl_req *req)
328 {
329 int error = 0;
330 /*
331 * TFO-key is expressed as a string in hex format
332 * (+1 to account for \0 char)
333 */
334 char keystring[TCP_FASTOPEN_KEYLEN * 2 + 1];
335 u_int32_t key[TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)];
336 int i;
337
338 /* -1, because newlen is len without the terminating \0 character */
339 if (req->newlen != (sizeof(keystring) - 1)) {
340 error = EINVAL;
341 goto exit;
342 }
343
344 /*
345 * sysctl_io_string copies keystring into the oldptr of the sysctl_req.
346 * Make sure everything is zero, to avoid putting garbage in there or
347 * leaking the stack.
348 */
349 bzero(keystring, sizeof(keystring));
350
351 error = sysctl_io_string(req, keystring, sizeof(keystring), 0, NULL);
352 if (error) {
353 goto exit;
354 }
355
356 for (i = 0; i < (TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)); i++) {
357 /*
358 * We jump over the keystring in 8-character (4 byte in hex)
359 * steps
360 */
361 if (sscanf(&keystring[i * 8], "%8x", &key[i]) != 1) {
362 error = EINVAL;
363 goto exit;
364 }
365 }
366
367 aes_encrypt_key128((u_char *)key, &tfo_ctx);
368
369 exit:
370 return error;
371 }
372
373 int
374 get_inpcb_str_size(void)
375 {
376 return sizeof(struct inpcb);
377 }
378
379 int
380 get_tcp_str_size(void)
381 {
382 return sizeof(struct tcpcb);
383 }
384
385 static int scale_to_powerof2(int size);
386
387 /*
388 * This helper routine returns one of the following scaled value of size:
389 * 1. Rounded down power of two value of size if the size value passed as
390 * argument is not a power of two and the rounded up value overflows.
391 * OR
392 * 2. Rounded up power of two value of size if the size value passed as
393 * argument is not a power of two and the rounded up value does not overflow
394 * OR
395 * 3. Same value as argument size if it is already a power of two.
396 */
397 static int
398 scale_to_powerof2(int size)
399 {
400 /* Handle special case of size = 0 */
401 int ret = size ? size : 1;
402
403 if (!powerof2(ret)) {
404 while (!powerof2(size)) {
405 /*
406 * Clear out least significant
407 * set bit till size is left with
408 * its highest set bit at which point
409 * it is rounded down power of two.
410 */
411 size = size & (size - 1);
412 }
413
414 /* Check for overflow when rounding up */
415 if (0 == (size << 1)) {
416 ret = size;
417 } else {
418 ret = size << 1;
419 }
420 }
421
422 return ret;
423 }
424
425 static void
426 tcp_tfo_init(void)
427 {
428 u_char key[TCP_FASTOPEN_KEYLEN];
429
430 read_frandom(key, sizeof(key));
431 aes_encrypt_key128(key, &tfo_ctx);
432 }
433
434 /*
435 * Tcp initialization
436 */
437 void
438 tcp_init(struct protosw *pp, struct domain *dp)
439 {
440 #pragma unused(dp)
441 static int tcp_initialized = 0;
442 vm_size_t str_size;
443 struct inpcbinfo *pcbinfo;
444 uint32_t logging_config;
445
446 VERIFY((pp->pr_flags & (PR_INITIALIZED | PR_ATTACHED)) == PR_ATTACHED);
447
448 if (tcp_initialized) {
449 return;
450 }
451 tcp_initialized = 1;
452
453 tcp_ccgen = 1;
454 tcp_keepinit = TCPTV_KEEP_INIT;
455 tcp_keepidle = TCPTV_KEEP_IDLE;
456 tcp_keepintvl = TCPTV_KEEPINTVL;
457 tcp_keepcnt = TCPTV_KEEPCNT;
458 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
459 tcp_msl = TCPTV_MSL;
460
461 microuptime(&tcp_uptime);
462 read_frandom(&tcp_now, sizeof(tcp_now));
463
464 /* Starts tcp internal clock at a random value */
465 tcp_now = tcp_now & 0x3fffffff;
466
467 /* expose initial uptime/now via systcl for utcp to keep time sync */
468 tcp_now_init = tcp_now;
469 tcp_microuptime_init =
470 (uint32_t)(tcp_uptime.tv_usec + (tcp_uptime.tv_sec * USEC_PER_SEC));
471 SYSCTL_SKMEM_UPDATE_FIELD(tcp.microuptime_init, tcp_microuptime_init);
472 SYSCTL_SKMEM_UPDATE_FIELD(tcp.now_init, tcp_now_init);
473
474 tcp_tfo_init();
475
476 LIST_INIT(&tcb);
477 tcbinfo.ipi_listhead = &tcb;
478
479 pcbinfo = &tcbinfo;
480 /*
481 * allocate lock group attribute and group for tcp pcb mutexes
482 */
483 pcbinfo->ipi_lock_grp_attr = lck_grp_attr_alloc_init();
484 pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb",
485 pcbinfo->ipi_lock_grp_attr);
486
487 /*
488 * allocate the lock attribute for tcp pcb mutexes
489 */
490 pcbinfo->ipi_lock_attr = lck_attr_alloc_init();
491
492 if ((pcbinfo->ipi_lock = lck_rw_alloc_init(pcbinfo->ipi_lock_grp,
493 pcbinfo->ipi_lock_attr)) == NULL) {
494 panic("%s: unable to allocate PCB lock\n", __func__);
495 /* NOTREACHED */
496 }
497
498 if (tcp_tcbhashsize == 0) {
499 /* Set to default */
500 tcp_tcbhashsize = 512;
501 }
502
503 if (!powerof2(tcp_tcbhashsize)) {
504 int old_hash_size = tcp_tcbhashsize;
505 tcp_tcbhashsize = scale_to_powerof2(tcp_tcbhashsize);
506 /* Lower limit of 16 */
507 if (tcp_tcbhashsize < 16) {
508 tcp_tcbhashsize = 16;
509 }
510 printf("WARNING: TCB hash size not a power of 2, "
511 "scaled from %d to %d.\n",
512 old_hash_size,
513 tcp_tcbhashsize);
514 }
515
516 tcbinfo.ipi_hashbase = hashinit(tcp_tcbhashsize, M_PCB,
517 &tcbinfo.ipi_hashmask);
518 tcbinfo.ipi_porthashbase = hashinit(tcp_tcbhashsize, M_PCB,
519 &tcbinfo.ipi_porthashmask);
520 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t));
521 tcbinfo.ipi_zone = zone_create("tcpcb", str_size, ZC_NONE);
522
523 tcbinfo.ipi_gc = tcp_gc;
524 tcbinfo.ipi_timer = tcp_itimer;
525 in_pcbinfo_attach(&tcbinfo);
526
527 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t));
528 sack_hole_zone = zone_create("sack_hole zone", str_size, ZC_NONE);
529
530 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t));
531 tcp_reass_zone = zone_create("tcp_reass_zone", str_size, ZC_NONE);
532
533 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct bwmeas), sizeof(u_int64_t));
534 tcp_bwmeas_zone = zone_create("tcp_bwmeas_zone", str_size, ZC_ZFREE_CLEARMEM);
535
536 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tcp_ccstate), sizeof(u_int64_t));
537 tcp_cc_zone = zone_create("tcp_cc_zone", str_size, ZC_NONE);
538
539 str_size = (vm_size_t)P2ROUNDUP(sizeof(struct tcp_rxt_seg), sizeof(u_int64_t));
540 tcp_rxt_seg_zone = zone_create("tcp_rxt_seg_zone", str_size, ZC_NONE);
541
542 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
543 if (max_protohdr < TCP_MINPROTOHDR) {
544 _max_protohdr = TCP_MINPROTOHDR;
545 _max_protohdr = (int)max_protohdr; /* round it up */
546 }
547 if (max_linkhdr + max_protohdr > MCLBYTES) {
548 panic("tcp_init");
549 }
550 #undef TCP_MINPROTOHDR
551
552 /* Initialize time wait and timer lists */
553 TAILQ_INIT(&tcp_tw_tailq);
554
555 bzero(&tcp_timer_list, sizeof(tcp_timer_list));
556 LIST_INIT(&tcp_timer_list.lhead);
557 /*
558 * allocate lock group attribute, group and attribute for
559 * the tcp timer list
560 */
561 tcp_timer_list.mtx_grp_attr = lck_grp_attr_alloc_init();
562 tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist",
563 tcp_timer_list.mtx_grp_attr);
564 tcp_timer_list.mtx_attr = lck_attr_alloc_init();
565 if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp,
566 tcp_timer_list.mtx_attr)) == NULL) {
567 panic("failed to allocate memory for tcp_timer_list.mtx\n");
568 }
569 ;
570 tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL);
571 if (tcp_timer_list.call == NULL) {
572 panic("failed to allocate call entry 1 in tcp_init\n");
573 }
574
575 /*
576 * allocate lock group attribute, group and attribute for
577 * tcp_uptime_lock
578 */
579 tcp_uptime_mtx_grp_attr = lck_grp_attr_alloc_init();
580 tcp_uptime_mtx_grp = lck_grp_alloc_init("tcpuptime",
581 tcp_uptime_mtx_grp_attr);
582 tcp_uptime_mtx_attr = lck_attr_alloc_init();
583 tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp,
584 tcp_uptime_mtx_attr);
585
586 /* Initialize TCP Cache */
587 tcp_cache_init();
588
589 tcp_mpkl_log_object = MPKL_CREATE_LOGOBJECT("com.apple.xnu.tcp");
590 if (tcp_mpkl_log_object == NULL) {
591 panic("MPKL_CREATE_LOGOBJECT failed");
592 }
593
594 logging_config = atm_get_diagnostic_config();
595 if (logging_config & 0x80000000) {
596 tcp_log_privacy = 1;
597 }
598
599 PE_parse_boot_argn("tcp_log", &tcp_log_enable_flags, sizeof(tcp_log_enable_flags));
600
601 /*
602 * If more than 4GB of actual memory is available, increase the
603 * maximum allowed receive and send socket buffer size.
604 */
605 if (mem_actual >= (1ULL << (GBSHIFT + 2))) {
606 tcp_autorcvbuf_max = 4 * 1024 * 1024;
607 tcp_autosndbuf_max = 4 * 1024 * 1024;
608
609 SYSCTL_SKMEM_UPDATE_FIELD(tcp.autorcvbufmax, tcp_autorcvbuf_max);
610 SYSCTL_SKMEM_UPDATE_FIELD(tcp.autosndbufmax, tcp_autosndbuf_max);
611 }
612 }
613
614 /*
615 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
616 * tcp_template used to store this data in mbufs, but we now recopy it out
617 * of the tcpcb each time to conserve mbufs.
618 */
619 void
620 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr)
621 {
622 struct inpcb *inp = tp->t_inpcb;
623 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
624
625 if ((inp->inp_vflag & INP_IPV6) != 0) {
626 struct ip6_hdr *ip6;
627
628 ip6 = (struct ip6_hdr *)ip_ptr;
629 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
630 (inp->inp_flow & IPV6_FLOWINFO_MASK);
631 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
632 (IPV6_VERSION & IPV6_VERSION_MASK);
633 ip6->ip6_plen = htons(sizeof(struct tcphdr));
634 ip6->ip6_nxt = IPPROTO_TCP;
635 ip6->ip6_hlim = 0;
636 ip6->ip6_src = inp->in6p_laddr;
637 ip6->ip6_dst = inp->in6p_faddr;
638 tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr,
639 htonl(sizeof(struct tcphdr) + IPPROTO_TCP));
640 } else {
641 struct ip *ip = (struct ip *) ip_ptr;
642
643 ip->ip_vhl = IP_VHL_BORING;
644 ip->ip_tos = 0;
645 ip->ip_len = 0;
646 ip->ip_id = 0;
647 ip->ip_off = 0;
648 ip->ip_ttl = 0;
649 ip->ip_sum = 0;
650 ip->ip_p = IPPROTO_TCP;
651 ip->ip_src = inp->inp_laddr;
652 ip->ip_dst = inp->inp_faddr;
653 tcp_hdr->th_sum =
654 in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
655 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
656 }
657
658 tcp_hdr->th_sport = inp->inp_lport;
659 tcp_hdr->th_dport = inp->inp_fport;
660 tcp_hdr->th_seq = 0;
661 tcp_hdr->th_ack = 0;
662 tcp_hdr->th_x2 = 0;
663 tcp_hdr->th_off = 5;
664 tcp_hdr->th_flags = 0;
665 tcp_hdr->th_win = 0;
666 tcp_hdr->th_urp = 0;
667 }
668
669 /*
670 * Create template to be used to send tcp packets on a connection.
671 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
672 * use for this function is in keepalives, which use tcp_respond.
673 */
674 struct tcptemp *
675 tcp_maketemplate(struct tcpcb *tp)
676 {
677 struct mbuf *m;
678 struct tcptemp *n;
679
680 m = m_get(M_DONTWAIT, MT_HEADER);
681 if (m == NULL) {
682 return NULL;
683 }
684 m->m_len = sizeof(struct tcptemp);
685 n = mtod(m, struct tcptemp *);
686
687 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
688 return n;
689 }
690
691 /*
692 * Send a single message to the TCP at address specified by
693 * the given TCP/IP header. If m == 0, then we make a copy
694 * of the tcpiphdr at ti and send directly to the addressed host.
695 * This is used to force keep alive messages out using the TCP
696 * template for a connection. If flags are given then we send
697 * a message back to the TCP which originated the * segment ti,
698 * and discard the mbuf containing it and any other attached mbufs.
699 *
700 * In any case the ack and sequence number of the transmitted
701 * segment are as specified by the parameters.
702 *
703 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
704 */
705 void
706 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
707 tcp_seq ack, tcp_seq seq, uint8_t flags, struct tcp_respond_args *tra)
708 {
709 uint16_t tlen;
710 int win = 0;
711 struct route *ro = 0;
712 struct route sro;
713 struct ip *ip;
714 struct tcphdr *nth;
715 struct route_in6 *ro6 = 0;
716 struct route_in6 sro6;
717 struct ip6_hdr *ip6;
718 int isipv6;
719 struct ifnet *outif;
720 int sotc = SO_TC_UNSPEC;
721 bool check_qos_marking_again = FALSE;
722
723 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
724 ip6 = ipgen;
725 ip = ipgen;
726
727 if (tp) {
728 check_qos_marking_again = tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE ? FALSE : TRUE;
729 if (!(flags & TH_RST)) {
730 win = tcp_sbspace(tp);
731 if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale) {
732 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
733 }
734 }
735 if (isipv6) {
736 ro6 = &tp->t_inpcb->in6p_route;
737 } else {
738 ro = &tp->t_inpcb->inp_route;
739 }
740 } else {
741 if (isipv6) {
742 ro6 = &sro6;
743 bzero(ro6, sizeof(*ro6));
744 } else {
745 ro = &sro;
746 bzero(ro, sizeof(*ro));
747 }
748 }
749 if (m == 0) {
750 m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */
751 if (m == NULL) {
752 return;
753 }
754 tlen = 0;
755 m->m_data += max_linkhdr;
756 if (isipv6) {
757 VERIFY((MHLEN - max_linkhdr) >=
758 (sizeof(*ip6) + sizeof(*nth)));
759 bcopy((caddr_t)ip6, mtod(m, caddr_t),
760 sizeof(struct ip6_hdr));
761 ip6 = mtod(m, struct ip6_hdr *);
762 nth = (struct tcphdr *)(void *)(ip6 + 1);
763 } else {
764 VERIFY((MHLEN - max_linkhdr) >=
765 (sizeof(*ip) + sizeof(*nth)));
766 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
767 ip = mtod(m, struct ip *);
768 nth = (struct tcphdr *)(void *)(ip + 1);
769 }
770 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
771 #if MPTCP
772 if ((tp) && (tp->t_mpflags & TMPF_RESET)) {
773 flags = (TH_RST | TH_ACK);
774 } else
775 #endif
776 flags = TH_ACK;
777 } else {
778 m_freem(m->m_next);
779 m->m_next = 0;
780 m->m_data = (caddr_t)ipgen;
781 /* m_len is set later */
782 tlen = 0;
783 #define xchg(a, b, type) { type t; t = a; a = b; b = t; }
784 if (isipv6) {
785 /* Expect 32-bit aligned IP on strict-align platforms */
786 IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6);
787 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
788 nth = (struct tcphdr *)(void *)(ip6 + 1);
789 } else {
790 /* Expect 32-bit aligned IP on strict-align platforms */
791 IP_HDR_STRICT_ALIGNMENT_CHECK(ip);
792 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
793 nth = (struct tcphdr *)(void *)(ip + 1);
794 }
795 if (th != nth) {
796 /*
797 * this is usually a case when an extension header
798 * exists between the IPv6 header and the
799 * TCP header.
800 */
801 nth->th_sport = th->th_sport;
802 nth->th_dport = th->th_dport;
803 }
804 xchg(nth->th_dport, nth->th_sport, n_short);
805 #undef xchg
806 }
807 if (isipv6) {
808 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) +
809 tlen));
810 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
811 } else {
812 tlen += sizeof(struct tcpiphdr);
813 ip->ip_len = tlen;
814 ip->ip_ttl = (uint8_t)ip_defttl;
815 }
816 m->m_len = tlen;
817 m->m_pkthdr.len = tlen;
818 m->m_pkthdr.rcvif = 0;
819 if (tra->keep_alive) {
820 m->m_pkthdr.pkt_flags |= PKTF_KEEPALIVE;
821 }
822
823 nth->th_seq = htonl(seq);
824 nth->th_ack = htonl(ack);
825 nth->th_x2 = 0;
826 nth->th_off = sizeof(struct tcphdr) >> 2;
827 nth->th_flags = flags;
828 if (tp) {
829 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
830 } else {
831 nth->th_win = htons((u_short)win);
832 }
833 nth->th_urp = 0;
834 if (isipv6) {
835 nth->th_sum = 0;
836 nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst,
837 htonl((tlen - sizeof(struct ip6_hdr)) + IPPROTO_TCP));
838 m->m_pkthdr.csum_flags = CSUM_TCPIPV6;
839 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
840 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
841 ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
842 } else {
843 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
844 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
845 m->m_pkthdr.csum_flags = CSUM_TCP;
846 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
847 }
848 #if TCPDEBUG
849 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
850 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
851 }
852 #endif
853
854 #if NECP
855 necp_mark_packet_from_socket(m, tp ? tp->t_inpcb : NULL, 0, 0, 0, 0);
856 #endif /* NECP */
857
858 #if IPSEC
859 if (tp != NULL && tp->t_inpcb->inp_sp != NULL &&
860 ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
861 m_freem(m);
862 return;
863 }
864 #endif
865
866 if (tp != NULL) {
867 u_int32_t svc_flags = 0;
868 if (isipv6) {
869 svc_flags |= PKT_SCF_IPV6;
870 }
871 sotc = tp->t_inpcb->inp_socket->so_traffic_class;
872 if ((flags & TH_RST) == 0) {
873 set_packet_service_class(m, tp->t_inpcb->inp_socket,
874 sotc, svc_flags);
875 } else {
876 m_set_service_class(m, MBUF_SC_BK_SYS);
877 }
878
879 /* Embed flowhash and flow control flags */
880 m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB;
881 m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash;
882 m->m_pkthdr.pkt_flags |= (PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC | PKTF_FLOW_ADV);
883 m->m_pkthdr.pkt_proto = IPPROTO_TCP;
884 m->m_pkthdr.tx_tcp_pid = tp->t_inpcb->inp_socket->last_pid;
885 m->m_pkthdr.tx_tcp_e_pid = tp->t_inpcb->inp_socket->e_pid;
886
887 if (flags & TH_RST) {
888 m->m_pkthdr.comp_gencnt = tp->t_comp_gencnt;
889 }
890 } else {
891 if (flags & TH_RST) {
892 m->m_pkthdr.comp_gencnt = TCP_ACK_COMPRESSION_DUMMY;
893 m_set_service_class(m, MBUF_SC_BK_SYS);
894 }
895 }
896
897 if (isipv6) {
898 struct ip6_out_args ip6oa;
899 bzero(&ip6oa, sizeof(ip6oa));
900 ip6oa.ip6oa_boundif = tra->ifscope;
901 ip6oa.ip6oa_flags = IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR;
902 ip6oa.ip6oa_sotc = SO_TC_UNSPEC;
903 ip6oa.ip6oa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
904
905 if (tra->ifscope != IFSCOPE_NONE) {
906 ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF;
907 }
908 if (tra->nocell) {
909 ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR;
910 }
911 if (tra->noexpensive) {
912 ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE;
913 }
914 if (tra->noconstrained) {
915 ip6oa.ip6oa_flags |= IP6OAF_NO_CONSTRAINED;
916 }
917 if (tra->awdl_unrestricted) {
918 ip6oa.ip6oa_flags |= IP6OAF_AWDL_UNRESTRICTED;
919 }
920 if (tra->intcoproc_allowed) {
921 ip6oa.ip6oa_flags |= IP6OAF_INTCOPROC_ALLOWED;
922 }
923 ip6oa.ip6oa_sotc = sotc;
924 if (tp != NULL) {
925 if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) {
926 ip6oa.ip6oa_flags |= IP6OAF_QOSMARKING_ALLOWED;
927 }
928 ip6oa.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount;
929 if (check_qos_marking_again) {
930 ip6oa.ip6oa_flags |= IP6OAF_REDO_QOSMARKING_POLICY;
931 }
932 ip6oa.ip6oa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype;
933 }
934 (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL,
935 NULL, &ip6oa);
936
937 if (check_qos_marking_again) {
938 struct inpcb *inp = tp->t_inpcb;
939 inp->inp_policyresult.results.qos_marking_gencount = ip6oa.qos_marking_gencount;
940 if (ip6oa.ip6oa_flags & IP6OAF_QOSMARKING_ALLOWED) {
941 inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED;
942 } else {
943 inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED;
944 }
945 }
946
947 if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL &&
948 (outif = ro6->ro_rt->rt_ifp) !=
949 tp->t_inpcb->in6p_last_outifp) {
950 tp->t_inpcb->in6p_last_outifp = outif;
951 }
952
953 if (ro6 == &sro6) {
954 ROUTE_RELEASE(ro6);
955 }
956 } else {
957 struct ip_out_args ipoa;
958 bzero(&ipoa, sizeof(ipoa));
959 ipoa.ipoa_boundif = tra->ifscope;
960 ipoa.ipoa_flags = IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR;
961 ipoa.ipoa_sotc = SO_TC_UNSPEC;
962 ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
963
964 if (tra->ifscope != IFSCOPE_NONE) {
965 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
966 }
967 if (tra->nocell) {
968 ipoa.ipoa_flags |= IPOAF_NO_CELLULAR;
969 }
970 if (tra->noexpensive) {
971 ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE;
972 }
973 if (tra->noconstrained) {
974 ipoa.ipoa_flags |= IPOAF_NO_CONSTRAINED;
975 }
976 if (tra->awdl_unrestricted) {
977 ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED;
978 }
979 ipoa.ipoa_sotc = sotc;
980 if (tp != NULL) {
981 if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) {
982 ipoa.ipoa_flags |= IPOAF_QOSMARKING_ALLOWED;
983 }
984 if (!(tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE)) {
985 ipoa.ipoa_flags |= IPOAF_REDO_QOSMARKING_POLICY;
986 }
987 ipoa.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount;
988 ipoa.ipoa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype;
989 }
990 if (ro != &sro) {
991 /* Copy the cached route and take an extra reference */
992 inp_route_copyout(tp->t_inpcb, &sro);
993 }
994 /*
995 * For consistency, pass a local route copy.
996 */
997 (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa);
998
999 if (check_qos_marking_again) {
1000 struct inpcb *inp = tp->t_inpcb;
1001 inp->inp_policyresult.results.qos_marking_gencount = ipoa.qos_marking_gencount;
1002 if (ipoa.ipoa_flags & IPOAF_QOSMARKING_ALLOWED) {
1003 inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED;
1004 } else {
1005 inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED;
1006 }
1007 }
1008 if (tp != NULL && sro.ro_rt != NULL &&
1009 (outif = sro.ro_rt->rt_ifp) !=
1010 tp->t_inpcb->inp_last_outifp) {
1011 tp->t_inpcb->inp_last_outifp = outif;
1012 }
1013 if (ro != &sro) {
1014 /* Synchronize cached PCB route */
1015 inp_route_copyin(tp->t_inpcb, &sro);
1016 } else {
1017 ROUTE_RELEASE(&sro);
1018 }
1019 }
1020 }
1021
1022 /*
1023 * Create a new TCP control block, making an
1024 * empty reassembly queue and hooking it to the argument
1025 * protocol control block. The `inp' parameter must have
1026 * come from the zone allocator set up in tcp_init().
1027 */
1028 struct tcpcb *
1029 tcp_newtcpcb(struct inpcb *inp)
1030 {
1031 struct inp_tp *it;
1032 struct tcpcb *tp;
1033 struct socket *so = inp->inp_socket;
1034 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1035 uint32_t random_32;
1036
1037 calculate_tcp_clock();
1038
1039 if ((so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) == 0) {
1040 it = (struct inp_tp *)(void *)inp;
1041 tp = &it->tcb;
1042 } else {
1043 tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb;
1044 }
1045
1046 bzero((char *) tp, sizeof(struct tcpcb));
1047 LIST_INIT(&tp->t_segq);
1048 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
1049
1050 tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP);
1051 tp->t_flagsext |= TF_SACK_ENABLE;
1052
1053 TAILQ_INIT(&tp->snd_holes);
1054 SLIST_INIT(&tp->t_rxt_segments);
1055 SLIST_INIT(&tp->t_notify_ack);
1056 tp->t_inpcb = inp;
1057 /*
1058 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1059 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1060 * reasonable initial retransmit time.
1061 */
1062 tp->t_srtt = TCPTV_SRTTBASE;
1063 tp->t_rttvar =
1064 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1065 tp->t_rttmin = tcp_TCPTV_MIN;
1066 tp->t_rxtcur = TCPTV_RTOBASE;
1067
1068 if (tcp_use_newreno) {
1069 /* use newreno by default */
1070 tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX;
1071 } else {
1072 tp->tcp_cc_index = TCP_CC_ALGO_CUBIC_INDEX;
1073 }
1074
1075 tcp_cc_allocate_state(tp);
1076
1077 if (CC_ALGO(tp)->init != NULL) {
1078 CC_ALGO(tp)->init(tp);
1079 }
1080
1081 tp->snd_cwnd = tcp_initial_cwnd(tp);
1082 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1083 tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1084 tp->t_rcvtime = tcp_now;
1085 tp->tentry.timer_start = tcp_now;
1086 tp->rcv_unackwin = tcp_now;
1087 tp->t_persist_timeout = tcp_max_persist_timeout;
1088 tp->t_persist_stop = 0;
1089 tp->t_flagsext |= TF_RCVUNACK_WAITSS;
1090 tp->t_rexmtthresh = (uint8_t)tcprexmtthresh;
1091 tp->rfbuf_ts = tcp_now;
1092 tp->rfbuf_space = tcp_initial_cwnd(tp);
1093 tp->t_forced_acks = TCP_FORCED_ACKS_COUNT;
1094
1095 /* Enable bandwidth measurement on this connection */
1096 tp->t_flagsext |= TF_MEASURESNDBW;
1097 if (tp->t_bwmeas == NULL) {
1098 tp->t_bwmeas = tcp_bwmeas_alloc(tp);
1099 if (tp->t_bwmeas == NULL) {
1100 tp->t_flagsext &= ~TF_MEASURESNDBW;
1101 }
1102 }
1103
1104 /* Clear time wait tailq entry */
1105 tp->t_twentry.tqe_next = NULL;
1106 tp->t_twentry.tqe_prev = NULL;
1107
1108 read_frandom(&random_32, sizeof(random_32));
1109 if (__probable(tcp_do_ack_compression)) {
1110 tp->t_comp_gencnt = random_32;
1111 if (tp->t_comp_gencnt <= TCP_ACK_COMPRESSION_DUMMY) {
1112 tp->t_comp_gencnt = TCP_ACK_COMPRESSION_DUMMY + 1;
1113 }
1114 tp->t_comp_lastinc = tcp_now;
1115 }
1116
1117 if (__probable(tcp_randomize_timestamps)) {
1118 tp->t_ts_offset = random_32;
1119 }
1120
1121 /*
1122 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1123 * because the socket may be bound to an IPv6 wildcard address,
1124 * which may match an IPv4-mapped IPv6 address.
1125 */
1126 inp->inp_ip_ttl = (uint8_t)ip_defttl;
1127 inp->inp_ppcb = (caddr_t)tp;
1128 return tp; /* XXX */
1129 }
1130
1131 /*
1132 * Drop a TCP connection, reporting
1133 * the specified error. If connection is synchronized,
1134 * then send a RST to peer.
1135 */
1136 struct tcpcb *
1137 tcp_drop(struct tcpcb *tp, int errno)
1138 {
1139 struct socket *so = tp->t_inpcb->inp_socket;
1140 #if CONFIG_DTRACE
1141 struct inpcb *inp = tp->t_inpcb;
1142 #endif
1143
1144 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1145 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1146 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1147 tp->t_state = TCPS_CLOSED;
1148 (void) tcp_output(tp);
1149 tcpstat.tcps_drops++;
1150 } else {
1151 tcpstat.tcps_conndrops++;
1152 }
1153 if (errno == ETIMEDOUT && tp->t_softerror) {
1154 errno = tp->t_softerror;
1155 }
1156 so->so_error = (u_short)errno;
1157
1158 TCP_LOG_CONNECTION_SUMMARY(tp);
1159
1160 return tcp_close(tp);
1161 }
1162
1163 void
1164 tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt)
1165 {
1166 u_int32_t rtt = rt->rt_rmx.rmx_rtt;
1167 int isnetlocal = (tp->t_flags & TF_LOCAL);
1168
1169 TCP_LOG_RTM_RTT(tp, rt);
1170
1171 if (rtt != 0 && tcp_init_rtt_from_cache != 0) {
1172 /*
1173 * XXX the lock bit for RTT indicates that the value
1174 * is also a minimum value; this is subject to time.
1175 */
1176 if (rt->rt_rmx.rmx_locks & RTV_RTT) {
1177 tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ);
1178 } else {
1179 tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN :
1180 TCPTV_REXMTMIN;
1181 }
1182
1183 tp->t_srtt =
1184 rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1185 tcpstat.tcps_usedrtt++;
1186
1187 if (rt->rt_rmx.rmx_rttvar) {
1188 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1189 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1190 tcpstat.tcps_usedrttvar++;
1191 } else {
1192 /* default variation is +- 1 rtt */
1193 tp->t_rttvar =
1194 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1195 }
1196
1197 /*
1198 * The RTO formula in the route metric case is based on:
1199 * 4 * srtt + 8 * rttvar
1200 * modulo the min, max and slop
1201 */
1202 TCPT_RANGESET(tp->t_rxtcur,
1203 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
1204 tp->t_rttmin, TCPTV_REXMTMAX,
1205 TCP_ADD_REXMTSLOP(tp));
1206 }
1207
1208 TCP_LOG_RTT_INFO(tp);
1209 }
1210
1211 static inline void
1212 tcp_create_ifnet_stats_per_flow(struct tcpcb *tp,
1213 struct ifnet_stats_per_flow *ifs)
1214 {
1215 struct inpcb *inp;
1216 struct socket *so;
1217 if (tp == NULL || ifs == NULL) {
1218 return;
1219 }
1220
1221 bzero(ifs, sizeof(*ifs));
1222 inp = tp->t_inpcb;
1223 so = inp->inp_socket;
1224
1225 ifs->ipv4 = (inp->inp_vflag & INP_IPV6) ? 0 : 1;
1226 ifs->local = (tp->t_flags & TF_LOCAL) ? 1 : 0;
1227 ifs->connreset = (so->so_error == ECONNRESET) ? 1 : 0;
1228 ifs->conntimeout = (so->so_error == ETIMEDOUT) ? 1 : 0;
1229 ifs->ecn_flags = tp->ecn_flags;
1230 ifs->txretransmitbytes = tp->t_stat.txretransmitbytes;
1231 ifs->rxoutoforderbytes = tp->t_stat.rxoutoforderbytes;
1232 ifs->rxmitpkts = tp->t_stat.rxmitpkts;
1233 ifs->rcvoopack = tp->t_rcvoopack;
1234 ifs->pawsdrop = tp->t_pawsdrop;
1235 ifs->sack_recovery_episodes = tp->t_sack_recovery_episode;
1236 ifs->reordered_pkts = tp->t_reordered_pkts;
1237 ifs->dsack_sent = tp->t_dsack_sent;
1238 ifs->dsack_recvd = tp->t_dsack_recvd;
1239 ifs->srtt = tp->t_srtt;
1240 ifs->rttupdated = tp->t_rttupdated;
1241 ifs->rttvar = tp->t_rttvar;
1242 ifs->rttmin = get_base_rtt(tp);
1243 if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_sndbw_max > 0) {
1244 ifs->bw_sndbw_max = tp->t_bwmeas->bw_sndbw_max;
1245 } else {
1246 ifs->bw_sndbw_max = 0;
1247 }
1248 if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_rcvbw_max > 0) {
1249 ifs->bw_rcvbw_max = tp->t_bwmeas->bw_rcvbw_max;
1250 } else {
1251 ifs->bw_rcvbw_max = 0;
1252 }
1253 ifs->bk_txpackets = so->so_tc_stats[MBUF_TC_BK].txpackets;
1254 ifs->txpackets = inp->inp_stat->txpackets;
1255 ifs->rxpackets = inp->inp_stat->rxpackets;
1256 }
1257
1258 static inline void
1259 tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs,
1260 struct if_tcp_ecn_perf_stat *stat)
1261 {
1262 u_int64_t curval, oldval;
1263 stat->total_txpkts += ifs->txpackets;
1264 stat->total_rxpkts += ifs->rxpackets;
1265 stat->total_rxmitpkts += ifs->rxmitpkts;
1266 stat->total_oopkts += ifs->rcvoopack;
1267 stat->total_reorderpkts += (ifs->reordered_pkts +
1268 ifs->pawsdrop + ifs->dsack_sent + ifs->dsack_recvd);
1269
1270 /* Average RTT */
1271 curval = ifs->srtt >> TCP_RTT_SHIFT;
1272 if (curval > 0 && ifs->rttupdated >= 16) {
1273 if (stat->rtt_avg == 0) {
1274 stat->rtt_avg = curval;
1275 } else {
1276 oldval = stat->rtt_avg;
1277 stat->rtt_avg = ((oldval << 4) - oldval + curval) >> 4;
1278 }
1279 }
1280
1281 /* RTT variance */
1282 curval = ifs->rttvar >> TCP_RTTVAR_SHIFT;
1283 if (curval > 0 && ifs->rttupdated >= 16) {
1284 if (stat->rtt_var == 0) {
1285 stat->rtt_var = curval;
1286 } else {
1287 oldval = stat->rtt_var;
1288 stat->rtt_var =
1289 ((oldval << 4) - oldval + curval) >> 4;
1290 }
1291 }
1292
1293 /* SACK episodes */
1294 stat->sack_episodes += ifs->sack_recovery_episodes;
1295 if (ifs->connreset) {
1296 stat->rst_drop++;
1297 }
1298 }
1299
1300 static inline void
1301 tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs,
1302 struct if_lim_perf_stat *stat)
1303 {
1304 u_int64_t curval, oldval;
1305
1306 stat->lim_total_txpkts += ifs->txpackets;
1307 stat->lim_total_rxpkts += ifs->rxpackets;
1308 stat->lim_total_retxpkts += ifs->rxmitpkts;
1309 stat->lim_total_oopkts += ifs->rcvoopack;
1310
1311 if (ifs->bw_sndbw_max > 0) {
1312 /* convert from bytes per ms to bits per second */
1313 ifs->bw_sndbw_max *= 8000;
1314 stat->lim_ul_max_bandwidth = MAX(stat->lim_ul_max_bandwidth,
1315 ifs->bw_sndbw_max);
1316 }
1317
1318 if (ifs->bw_rcvbw_max > 0) {
1319 /* convert from bytes per ms to bits per second */
1320 ifs->bw_rcvbw_max *= 8000;
1321 stat->lim_dl_max_bandwidth = MAX(stat->lim_dl_max_bandwidth,
1322 ifs->bw_rcvbw_max);
1323 }
1324
1325 /* Average RTT */
1326 curval = ifs->srtt >> TCP_RTT_SHIFT;
1327 if (curval > 0 && ifs->rttupdated >= 16) {
1328 if (stat->lim_rtt_average == 0) {
1329 stat->lim_rtt_average = curval;
1330 } else {
1331 oldval = stat->lim_rtt_average;
1332 stat->lim_rtt_average =
1333 ((oldval << 4) - oldval + curval) >> 4;
1334 }
1335 }
1336
1337 /* RTT variance */
1338 curval = ifs->rttvar >> TCP_RTTVAR_SHIFT;
1339 if (curval > 0 && ifs->rttupdated >= 16) {
1340 if (stat->lim_rtt_variance == 0) {
1341 stat->lim_rtt_variance = curval;
1342 } else {
1343 oldval = stat->lim_rtt_variance;
1344 stat->lim_rtt_variance =
1345 ((oldval << 4) - oldval + curval) >> 4;
1346 }
1347 }
1348
1349 if (stat->lim_rtt_min == 0) {
1350 stat->lim_rtt_min = ifs->rttmin;
1351 } else {
1352 stat->lim_rtt_min = MIN(stat->lim_rtt_min, ifs->rttmin);
1353 }
1354
1355 /* connection timeouts */
1356 stat->lim_conn_attempts++;
1357 if (ifs->conntimeout) {
1358 stat->lim_conn_timeouts++;
1359 }
1360
1361 /* bytes sent using background delay-based algorithms */
1362 stat->lim_bk_txpkts += ifs->bk_txpackets;
1363 }
1364
1365 /*
1366 * Close a TCP control block:
1367 * discard all space held by the tcp
1368 * discard internet protocol block
1369 * wake up any sleepers
1370 */
1371 struct tcpcb *
1372 tcp_close(struct tcpcb *tp)
1373 {
1374 struct inpcb *inp = tp->t_inpcb;
1375 struct socket *so = inp->inp_socket;
1376 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1377 struct route *ro;
1378 struct rtentry *rt;
1379 int dosavessthresh;
1380 struct ifnet_stats_per_flow ifs;
1381
1382 /* tcp_close was called previously, bail */
1383 if (inp->inp_ppcb == NULL) {
1384 return NULL;
1385 }
1386
1387 tcp_del_fsw_flow(tp);
1388
1389 tcp_canceltimers(tp);
1390 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp, 0, 0, 0, 0);
1391
1392 /*
1393 * If another thread for this tcp is currently in ip (indicated by
1394 * the TF_SENDINPROG flag), defer the cleanup until after it returns
1395 * back to tcp. This is done to serialize the close until after all
1396 * pending output is finished, in order to avoid having the PCB be
1397 * detached and the cached route cleaned, only for ip to cache the
1398 * route back into the PCB again. Note that we've cleared all the
1399 * timers at this point. Set TF_CLOSING to indicate to tcp_output()
1400 * that is should call us again once it returns from ip; at that
1401 * point both flags should be cleared and we can proceed further
1402 * with the cleanup.
1403 */
1404 if ((tp->t_flags & TF_CLOSING) ||
1405 inp->inp_sndinprog_cnt > 0) {
1406 tp->t_flags |= TF_CLOSING;
1407 return NULL;
1408 }
1409
1410 TCP_LOG_CONNECTION_SUMMARY(tp);
1411
1412 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1413 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1414
1415 ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route);
1416 rt = ro->ro_rt;
1417 if (rt != NULL) {
1418 RT_LOCK_SPIN(rt);
1419 }
1420
1421 /*
1422 * If we got enough samples through the srtt filter,
1423 * save the rtt and rttvar in the routing entry.
1424 * 'Enough' is arbitrarily defined as the 16 samples.
1425 * 16 samples is enough for the srtt filter to converge
1426 * to within 5% of the correct value; fewer samples and
1427 * we could save a very bogus rtt.
1428 *
1429 * Don't update the default route's characteristics and don't
1430 * update anything that the user "locked".
1431 */
1432 if (tp->t_rttupdated >= 16) {
1433 u_int32_t i = 0;
1434 bool log_rtt = false;
1435
1436 if (isipv6) {
1437 struct sockaddr_in6 *sin6;
1438
1439 if (rt == NULL) {
1440 goto no_valid_rt;
1441 }
1442 sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt);
1443 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
1444 goto no_valid_rt;
1445 }
1446 } else if (ROUTE_UNUSABLE(ro) ||
1447 SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) {
1448 DTRACE_TCP4(state__change, void, NULL,
1449 struct inpcb *, inp, struct tcpcb *, tp,
1450 int32_t, TCPS_CLOSED);
1451 tp->t_state = TCPS_CLOSED;
1452 goto no_valid_rt;
1453 }
1454
1455 RT_LOCK_ASSERT_HELD(rt);
1456 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1457 i = tp->t_srtt *
1458 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1459 if (rt->rt_rmx.rmx_rtt && i) {
1460 /*
1461 * filter this update to half the old & half
1462 * the new values, converting scale.
1463 * See route.h and tcp_var.h for a
1464 * description of the scaling constants.
1465 */
1466 rt->rt_rmx.rmx_rtt =
1467 (rt->rt_rmx.rmx_rtt + i) / 2;
1468 } else {
1469 rt->rt_rmx.rmx_rtt = i;
1470 }
1471 tcpstat.tcps_cachedrtt++;
1472 log_rtt = true;
1473 }
1474 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1475 i = tp->t_rttvar *
1476 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1477 if (rt->rt_rmx.rmx_rttvar && i) {
1478 rt->rt_rmx.rmx_rttvar =
1479 (rt->rt_rmx.rmx_rttvar + i) / 2;
1480 } else {
1481 rt->rt_rmx.rmx_rttvar = i;
1482 }
1483 tcpstat.tcps_cachedrttvar++;
1484 log_rtt = true;
1485 }
1486 if (log_rtt) {
1487 TCP_LOG_RTM_RTT(tp, rt);
1488 TCP_LOG_RTT_INFO(tp);
1489 }
1490 /*
1491 * The old comment here said:
1492 * update the pipelimit (ssthresh) if it has been updated
1493 * already or if a pipesize was specified & the threshhold
1494 * got below half the pipesize. I.e., wait for bad news
1495 * before we start updating, then update on both good
1496 * and bad news.
1497 *
1498 * But we want to save the ssthresh even if no pipesize is
1499 * specified explicitly in the route, because such
1500 * connections still have an implicit pipesize specified
1501 * by the global tcp_sendspace. In the absence of a reliable
1502 * way to calculate the pipesize, it will have to do.
1503 */
1504 i = tp->snd_ssthresh;
1505 if (rt->rt_rmx.rmx_sendpipe != 0) {
1506 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
1507 } else {
1508 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
1509 }
1510 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1511 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) ||
1512 dosavessthresh) {
1513 /*
1514 * convert the limit from user data bytes to
1515 * packets then to packet data bytes.
1516 */
1517 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
1518 if (i < 2) {
1519 i = 2;
1520 }
1521 i *= (u_int32_t)(tp->t_maxseg +
1522 isipv6 ? sizeof(struct ip6_hdr) +
1523 sizeof(struct tcphdr) :
1524 sizeof(struct tcpiphdr));
1525 if (rt->rt_rmx.rmx_ssthresh) {
1526 rt->rt_rmx.rmx_ssthresh =
1527 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1528 } else {
1529 rt->rt_rmx.rmx_ssthresh = i;
1530 }
1531 tcpstat.tcps_cachedssthresh++;
1532 }
1533 }
1534
1535 /*
1536 * Mark route for deletion if no information is cached.
1537 */
1538 if (rt != NULL && (so->so_flags & SOF_OVERFLOW)) {
1539 if (!(rt->rt_rmx.rmx_locks & RTV_RTT) &&
1540 rt->rt_rmx.rmx_rtt == 0) {
1541 rt->rt_flags |= RTF_DELCLONE;
1542 }
1543 }
1544
1545 no_valid_rt:
1546 if (rt != NULL) {
1547 RT_UNLOCK(rt);
1548 }
1549
1550 /* free the reassembly queue, if any */
1551 (void) tcp_freeq(tp);
1552
1553 /* performance stats per interface */
1554 tcp_create_ifnet_stats_per_flow(tp, &ifs);
1555 tcp_update_stats_per_flow(&ifs, inp->inp_last_outifp);
1556
1557 tcp_free_sackholes(tp);
1558 tcp_notify_ack_free(tp);
1559
1560 inp_decr_sndbytes_allunsent(so, tp->snd_una);
1561
1562 if (tp->t_bwmeas != NULL) {
1563 tcp_bwmeas_free(tp);
1564 }
1565 tcp_rxtseg_clean(tp);
1566 /* Free the packet list */
1567 if (tp->t_pktlist_head != NULL) {
1568 m_freem_list(tp->t_pktlist_head);
1569 }
1570 TCP_PKTLIST_CLEAR(tp);
1571
1572 if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) {
1573 inp->inp_saved_ppcb = (caddr_t) tp;
1574 }
1575
1576 tp->t_state = TCPS_CLOSED;
1577
1578 /*
1579 * Issue a wakeup before detach so that we don't miss
1580 * a wakeup
1581 */
1582 sodisconnectwakeup(so);
1583
1584 /*
1585 * Make sure to clear the TCP Keep Alive Offload as it is
1586 * ref counted on the interface
1587 */
1588 tcp_clear_keep_alive_offload(so);
1589
1590 /*
1591 * If this is a socket that does not want to wakeup the device
1592 * for it's traffic, the application might need to know that the
1593 * socket is closed, send a notification.
1594 */
1595 if ((so->so_options & SO_NOWAKEFROMSLEEP) &&
1596 inp->inp_state != INPCB_STATE_DEAD &&
1597 !(inp->inp_flags2 & INP2_TIMEWAIT)) {
1598 socket_post_kev_msg_closed(so);
1599 }
1600
1601 if (CC_ALGO(tp)->cleanup != NULL) {
1602 CC_ALGO(tp)->cleanup(tp);
1603 }
1604
1605 if (tp->t_ccstate != NULL) {
1606 zfree(tcp_cc_zone, tp->t_ccstate);
1607 tp->t_ccstate = NULL;
1608 }
1609 tp->tcp_cc_index = TCP_CC_ALGO_NONE;
1610
1611 /* Can happen if we close the socket before receiving the third ACK */
1612 if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) {
1613 OSDecrementAtomic(&tcp_tfo_halfcnt);
1614
1615 /* Panic if something has gone terribly wrong. */
1616 VERIFY(tcp_tfo_halfcnt >= 0);
1617
1618 tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID;
1619 }
1620
1621 if (SOCK_CHECK_DOM(so, PF_INET6)) {
1622 in6_pcbdetach(inp);
1623 } else {
1624 in_pcbdetach(inp);
1625 }
1626
1627 /*
1628 * Call soisdisconnected after detach because it might unlock the socket
1629 */
1630 soisdisconnected(so);
1631 tcpstat.tcps_closed++;
1632 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END,
1633 tcpstat.tcps_closed, 0, 0, 0, 0);
1634 return NULL;
1635 }
1636
1637 int
1638 tcp_freeq(struct tcpcb *tp)
1639 {
1640 struct tseg_qent *q;
1641 int rv = 0;
1642
1643 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
1644 LIST_REMOVE(q, tqe_q);
1645 m_freem(q->tqe_m);
1646 zfree(tcp_reass_zone, q);
1647 rv = 1;
1648 }
1649 tp->t_reassqlen = 0;
1650 return rv;
1651 }
1652
1653
1654 void
1655 tcp_drain(void)
1656 {
1657 struct inpcb *inp;
1658 struct tcpcb *tp;
1659
1660 if (!lck_rw_try_lock_exclusive(tcbinfo.ipi_lock)) {
1661 return;
1662 }
1663
1664 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1665 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) !=
1666 WNT_STOPUSING) {
1667 socket_lock(inp->inp_socket, 1);
1668 if (in_pcb_checkstate(inp, WNT_RELEASE, 1)
1669 == WNT_STOPUSING) {
1670 /* lost a race, try the next one */
1671 socket_unlock(inp->inp_socket, 1);
1672 continue;
1673 }
1674 tp = intotcpcb(inp);
1675
1676 so_drain_extended_bk_idle(inp->inp_socket);
1677
1678 socket_unlock(inp->inp_socket, 1);
1679 }
1680 }
1681 lck_rw_done(tcbinfo.ipi_lock);
1682 }
1683
1684 /*
1685 * Notify a tcp user of an asynchronous error;
1686 * store error as soft error, but wake up user
1687 * (for now, won't do anything until can select for soft error).
1688 *
1689 * Do not wake up user since there currently is no mechanism for
1690 * reporting soft errors (yet - a kqueue filter may be added).
1691 */
1692 static void
1693 tcp_notify(struct inpcb *inp, int error)
1694 {
1695 struct tcpcb *tp;
1696
1697 if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD)) {
1698 return; /* pcb is gone already */
1699 }
1700 tp = (struct tcpcb *)inp->inp_ppcb;
1701
1702 VERIFY(tp != NULL);
1703 /*
1704 * Ignore some errors if we are hooked up.
1705 * If connection hasn't completed, has retransmitted several times,
1706 * and receives a second error, give up now. This is better
1707 * than waiting a long time to establish a connection that
1708 * can never complete.
1709 */
1710 if (tp->t_state == TCPS_ESTABLISHED &&
1711 (error == EHOSTUNREACH || error == ENETUNREACH ||
1712 error == EHOSTDOWN)) {
1713 if (inp->inp_route.ro_rt) {
1714 rtfree(inp->inp_route.ro_rt);
1715 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1716 }
1717 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1718 tp->t_softerror) {
1719 tcp_drop(tp, error);
1720 } else {
1721 tp->t_softerror = error;
1722 }
1723 }
1724
1725 struct bwmeas *
1726 tcp_bwmeas_alloc(struct tcpcb *tp)
1727 {
1728 struct bwmeas *elm;
1729 elm = zalloc_flags(tcp_bwmeas_zone, Z_ZERO | Z_WAITOK);
1730 elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE;
1731 elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg;
1732 return elm;
1733 }
1734
1735 void
1736 tcp_bwmeas_free(struct tcpcb *tp)
1737 {
1738 zfree(tcp_bwmeas_zone, tp->t_bwmeas);
1739 tp->t_bwmeas = NULL;
1740 tp->t_flagsext &= ~(TF_MEASURESNDBW);
1741 }
1742
1743 int
1744 get_tcp_inp_list(struct inpcb **inp_list, int n, inp_gen_t gencnt)
1745 {
1746 struct tcpcb *tp;
1747 struct inpcb *inp;
1748 int i = 0;
1749
1750 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1751 if (inp->inp_gencnt <= gencnt &&
1752 inp->inp_state != INPCB_STATE_DEAD) {
1753 inp_list[i++] = inp;
1754 }
1755 if (i >= n) {
1756 break;
1757 }
1758 }
1759
1760 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
1761 inp = tp->t_inpcb;
1762 if (inp->inp_gencnt <= gencnt &&
1763 inp->inp_state != INPCB_STATE_DEAD) {
1764 inp_list[i++] = inp;
1765 }
1766 if (i >= n) {
1767 break;
1768 }
1769 }
1770 return i;
1771 }
1772
1773 /*
1774 * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format.
1775 * The otcpcb data structure is passed to user space and must not change.
1776 */
1777 static void
1778 tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp)
1779 {
1780 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1781 otp->t_dupacks = tp->t_dupacks;
1782 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1783 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1784 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1785 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1786 otp->t_inpcb =
1787 (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb);
1788 otp->t_state = tp->t_state;
1789 otp->t_flags = tp->t_flags;
1790 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1791 otp->snd_una = tp->snd_una;
1792 otp->snd_max = tp->snd_max;
1793 otp->snd_nxt = tp->snd_nxt;
1794 otp->snd_up = tp->snd_up;
1795 otp->snd_wl1 = tp->snd_wl1;
1796 otp->snd_wl2 = tp->snd_wl2;
1797 otp->iss = tp->iss;
1798 otp->irs = tp->irs;
1799 otp->rcv_nxt = tp->rcv_nxt;
1800 otp->rcv_adv = tp->rcv_adv;
1801 otp->rcv_wnd = tp->rcv_wnd;
1802 otp->rcv_up = tp->rcv_up;
1803 otp->snd_wnd = tp->snd_wnd;
1804 otp->snd_cwnd = tp->snd_cwnd;
1805 otp->snd_ssthresh = tp->snd_ssthresh;
1806 otp->t_maxopd = tp->t_maxopd;
1807 otp->t_rcvtime = tp->t_rcvtime;
1808 otp->t_starttime = tp->t_starttime;
1809 otp->t_rtttime = tp->t_rtttime;
1810 otp->t_rtseq = tp->t_rtseq;
1811 otp->t_rxtcur = tp->t_rxtcur;
1812 otp->t_maxseg = tp->t_maxseg;
1813 otp->t_srtt = tp->t_srtt;
1814 otp->t_rttvar = tp->t_rttvar;
1815 otp->t_rxtshift = tp->t_rxtshift;
1816 otp->t_rttmin = tp->t_rttmin;
1817 otp->t_rttupdated = tp->t_rttupdated;
1818 otp->max_sndwnd = tp->max_sndwnd;
1819 otp->t_softerror = tp->t_softerror;
1820 otp->t_oobflags = tp->t_oobflags;
1821 otp->t_iobc = tp->t_iobc;
1822 otp->snd_scale = tp->snd_scale;
1823 otp->rcv_scale = tp->rcv_scale;
1824 otp->request_r_scale = tp->request_r_scale;
1825 otp->requested_s_scale = tp->requested_s_scale;
1826 otp->ts_recent = tp->ts_recent;
1827 otp->ts_recent_age = tp->ts_recent_age;
1828 otp->last_ack_sent = tp->last_ack_sent;
1829 otp->cc_send = 0;
1830 otp->cc_recv = 0;
1831 otp->snd_recover = tp->snd_recover;
1832 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1833 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1834 otp->t_badrxtwin = 0;
1835 }
1836
1837 static int
1838 tcp_pcblist SYSCTL_HANDLER_ARGS
1839 {
1840 #pragma unused(oidp, arg1, arg2)
1841 int error, i = 0, n;
1842 struct inpcb **inp_list;
1843 inp_gen_t gencnt;
1844 struct xinpgen xig;
1845
1846 /*
1847 * The process of preparing the TCB list is too time-consuming and
1848 * resource-intensive to repeat twice on every request.
1849 */
1850 lck_rw_lock_shared(tcbinfo.ipi_lock);
1851 if (req->oldptr == USER_ADDR_NULL) {
1852 n = tcbinfo.ipi_count;
1853 req->oldidx = 2 * (sizeof(xig))
1854 + (n + n / 8) * sizeof(struct xtcpcb);
1855 lck_rw_done(tcbinfo.ipi_lock);
1856 return 0;
1857 }
1858
1859 if (req->newptr != USER_ADDR_NULL) {
1860 lck_rw_done(tcbinfo.ipi_lock);
1861 return EPERM;
1862 }
1863
1864 /*
1865 * OK, now we're committed to doing something.
1866 */
1867 gencnt = tcbinfo.ipi_gencnt;
1868 n = tcbinfo.ipi_count;
1869
1870 bzero(&xig, sizeof(xig));
1871 xig.xig_len = sizeof(xig);
1872 xig.xig_count = n;
1873 xig.xig_gen = gencnt;
1874 xig.xig_sogen = so_gencnt;
1875 error = SYSCTL_OUT(req, &xig, sizeof(xig));
1876 if (error) {
1877 lck_rw_done(tcbinfo.ipi_lock);
1878 return error;
1879 }
1880 /*
1881 * We are done if there is no pcb
1882 */
1883 if (n == 0) {
1884 lck_rw_done(tcbinfo.ipi_lock);
1885 return 0;
1886 }
1887
1888 inp_list = _MALLOC(n * sizeof(*inp_list), M_TEMP, M_WAITOK);
1889 if (inp_list == 0) {
1890 lck_rw_done(tcbinfo.ipi_lock);
1891 return ENOMEM;
1892 }
1893
1894 n = get_tcp_inp_list(inp_list, n, gencnt);
1895
1896 error = 0;
1897 for (i = 0; i < n; i++) {
1898 struct xtcpcb xt;
1899 caddr_t inp_ppcb;
1900 struct inpcb *inp;
1901
1902 inp = inp_list[i];
1903
1904 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
1905 continue;
1906 }
1907 socket_lock(inp->inp_socket, 1);
1908 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
1909 socket_unlock(inp->inp_socket, 1);
1910 continue;
1911 }
1912 if (inp->inp_gencnt > gencnt) {
1913 socket_unlock(inp->inp_socket, 1);
1914 continue;
1915 }
1916
1917 bzero(&xt, sizeof(xt));
1918 xt.xt_len = sizeof(xt);
1919 /* XXX should avoid extra copy */
1920 inpcb_to_compat(inp, &xt.xt_inp);
1921 inp_ppcb = inp->inp_ppcb;
1922 if (inp_ppcb != NULL) {
1923 tcpcb_to_otcpcb((struct tcpcb *)(void *)inp_ppcb,
1924 &xt.xt_tp);
1925 } else {
1926 bzero((char *) &xt.xt_tp, sizeof(xt.xt_tp));
1927 }
1928 if (inp->inp_socket) {
1929 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1930 }
1931
1932 socket_unlock(inp->inp_socket, 1);
1933
1934 error = SYSCTL_OUT(req, &xt, sizeof(xt));
1935 }
1936 if (!error) {
1937 /*
1938 * Give the user an updated idea of our state.
1939 * If the generation differs from what we told
1940 * her before, she knows that something happened
1941 * while we were processing this request, and it
1942 * might be necessary to retry.
1943 */
1944 bzero(&xig, sizeof(xig));
1945 xig.xig_len = sizeof(xig);
1946 xig.xig_gen = tcbinfo.ipi_gencnt;
1947 xig.xig_sogen = so_gencnt;
1948 xig.xig_count = tcbinfo.ipi_count;
1949 error = SYSCTL_OUT(req, &xig, sizeof(xig));
1950 }
1951 FREE(inp_list, M_TEMP);
1952 lck_rw_done(tcbinfo.ipi_lock);
1953 return error;
1954 }
1955
1956 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1957 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1958 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1959
1960 #if XNU_TARGET_OS_OSX
1961
1962 static void
1963 tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp)
1964 {
1965 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1966 otp->t_dupacks = tp->t_dupacks;
1967 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1968 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1969 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1970 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1971 otp->t_state = tp->t_state;
1972 otp->t_flags = tp->t_flags;
1973 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1974 otp->snd_una = tp->snd_una;
1975 otp->snd_max = tp->snd_max;
1976 otp->snd_nxt = tp->snd_nxt;
1977 otp->snd_up = tp->snd_up;
1978 otp->snd_wl1 = tp->snd_wl1;
1979 otp->snd_wl2 = tp->snd_wl2;
1980 otp->iss = tp->iss;
1981 otp->irs = tp->irs;
1982 otp->rcv_nxt = tp->rcv_nxt;
1983 otp->rcv_adv = tp->rcv_adv;
1984 otp->rcv_wnd = tp->rcv_wnd;
1985 otp->rcv_up = tp->rcv_up;
1986 otp->snd_wnd = tp->snd_wnd;
1987 otp->snd_cwnd = tp->snd_cwnd;
1988 otp->snd_ssthresh = tp->snd_ssthresh;
1989 otp->t_maxopd = tp->t_maxopd;
1990 otp->t_rcvtime = tp->t_rcvtime;
1991 otp->t_starttime = tp->t_starttime;
1992 otp->t_rtttime = tp->t_rtttime;
1993 otp->t_rtseq = tp->t_rtseq;
1994 otp->t_rxtcur = tp->t_rxtcur;
1995 otp->t_maxseg = tp->t_maxseg;
1996 otp->t_srtt = tp->t_srtt;
1997 otp->t_rttvar = tp->t_rttvar;
1998 otp->t_rxtshift = tp->t_rxtshift;
1999 otp->t_rttmin = tp->t_rttmin;
2000 otp->t_rttupdated = tp->t_rttupdated;
2001 otp->max_sndwnd = tp->max_sndwnd;
2002 otp->t_softerror = tp->t_softerror;
2003 otp->t_oobflags = tp->t_oobflags;
2004 otp->t_iobc = tp->t_iobc;
2005 otp->snd_scale = tp->snd_scale;
2006 otp->rcv_scale = tp->rcv_scale;
2007 otp->request_r_scale = tp->request_r_scale;
2008 otp->requested_s_scale = tp->requested_s_scale;
2009 otp->ts_recent = tp->ts_recent;
2010 otp->ts_recent_age = tp->ts_recent_age;
2011 otp->last_ack_sent = tp->last_ack_sent;
2012 otp->cc_send = 0;
2013 otp->cc_recv = 0;
2014 otp->snd_recover = tp->snd_recover;
2015 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
2016 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
2017 otp->t_badrxtwin = 0;
2018 }
2019
2020
2021 static int
2022 tcp_pcblist64 SYSCTL_HANDLER_ARGS
2023 {
2024 #pragma unused(oidp, arg1, arg2)
2025 int error, i = 0, n;
2026 struct inpcb **inp_list;
2027 inp_gen_t gencnt;
2028 struct xinpgen xig;
2029
2030 /*
2031 * The process of preparing the TCB list is too time-consuming and
2032 * resource-intensive to repeat twice on every request.
2033 */
2034 lck_rw_lock_shared(tcbinfo.ipi_lock);
2035 if (req->oldptr == USER_ADDR_NULL) {
2036 n = tcbinfo.ipi_count;
2037 req->oldidx = 2 * (sizeof(xig))
2038 + (n + n / 8) * sizeof(struct xtcpcb64);
2039 lck_rw_done(tcbinfo.ipi_lock);
2040 return 0;
2041 }
2042
2043 if (req->newptr != USER_ADDR_NULL) {
2044 lck_rw_done(tcbinfo.ipi_lock);
2045 return EPERM;
2046 }
2047
2048 /*
2049 * OK, now we're committed to doing something.
2050 */
2051 gencnt = tcbinfo.ipi_gencnt;
2052 n = tcbinfo.ipi_count;
2053
2054 bzero(&xig, sizeof(xig));
2055 xig.xig_len = sizeof(xig);
2056 xig.xig_count = n;
2057 xig.xig_gen = gencnt;
2058 xig.xig_sogen = so_gencnt;
2059 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2060 if (error) {
2061 lck_rw_done(tcbinfo.ipi_lock);
2062 return error;
2063 }
2064 /*
2065 * We are done if there is no pcb
2066 */
2067 if (n == 0) {
2068 lck_rw_done(tcbinfo.ipi_lock);
2069 return 0;
2070 }
2071
2072 inp_list = _MALLOC(n * sizeof(*inp_list), M_TEMP, M_WAITOK);
2073 if (inp_list == 0) {
2074 lck_rw_done(tcbinfo.ipi_lock);
2075 return ENOMEM;
2076 }
2077
2078 n = get_tcp_inp_list(inp_list, n, gencnt);
2079
2080 error = 0;
2081 for (i = 0; i < n; i++) {
2082 struct xtcpcb64 xt;
2083 struct inpcb *inp;
2084
2085 inp = inp_list[i];
2086
2087 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
2088 continue;
2089 }
2090 socket_lock(inp->inp_socket, 1);
2091 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
2092 socket_unlock(inp->inp_socket, 1);
2093 continue;
2094 }
2095 if (inp->inp_gencnt > gencnt) {
2096 socket_unlock(inp->inp_socket, 1);
2097 continue;
2098 }
2099
2100 bzero(&xt, sizeof(xt));
2101 xt.xt_len = sizeof(xt);
2102 inpcb_to_xinpcb64(inp, &xt.xt_inpcb);
2103 xt.xt_inpcb.inp_ppcb =
2104 (uint64_t)VM_KERNEL_ADDRPERM(inp->inp_ppcb);
2105 if (inp->inp_ppcb != NULL) {
2106 tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb,
2107 &xt);
2108 }
2109 if (inp->inp_socket) {
2110 sotoxsocket64(inp->inp_socket,
2111 &xt.xt_inpcb.xi_socket);
2112 }
2113
2114 socket_unlock(inp->inp_socket, 1);
2115
2116 error = SYSCTL_OUT(req, &xt, sizeof(xt));
2117 }
2118 if (!error) {
2119 /*
2120 * Give the user an updated idea of our state.
2121 * If the generation differs from what we told
2122 * her before, she knows that something happened
2123 * while we were processing this request, and it
2124 * might be necessary to retry.
2125 */
2126 bzero(&xig, sizeof(xig));
2127 xig.xig_len = sizeof(xig);
2128 xig.xig_gen = tcbinfo.ipi_gencnt;
2129 xig.xig_sogen = so_gencnt;
2130 xig.xig_count = tcbinfo.ipi_count;
2131 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2132 }
2133 FREE(inp_list, M_TEMP);
2134 lck_rw_done(tcbinfo.ipi_lock);
2135 return error;
2136 }
2137
2138 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64,
2139 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2140 tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections");
2141
2142 #endif /* XNU_TARGET_OS_OSX */
2143
2144 static int
2145 tcp_pcblist_n SYSCTL_HANDLER_ARGS
2146 {
2147 #pragma unused(oidp, arg1, arg2)
2148 int error = 0;
2149
2150 error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo);
2151
2152 return error;
2153 }
2154
2155
2156 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n,
2157 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2158 tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections");
2159
2160 static int
2161 tcp_progress_indicators SYSCTL_HANDLER_ARGS
2162 {
2163 #pragma unused(oidp, arg1, arg2)
2164
2165 return ntstat_tcp_progress_indicators(req);
2166 }
2167
2168 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, progress,
2169 CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, 0,
2170 tcp_progress_indicators, "S", "Various items that indicate the current state of progress on the link");
2171
2172
2173 __private_extern__ void
2174 tcp_get_ports_used(uint32_t ifindex, int protocol, uint32_t flags,
2175 bitstr_t *bitfield)
2176 {
2177 inpcb_get_ports_used(ifindex, protocol, flags, bitfield,
2178 &tcbinfo);
2179 }
2180
2181 __private_extern__ uint32_t
2182 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags)
2183 {
2184 return inpcb_count_opportunistic(ifindex, &tcbinfo, flags);
2185 }
2186
2187 __private_extern__ uint32_t
2188 tcp_find_anypcb_byaddr(struct ifaddr *ifa)
2189 {
2190 return inpcb_find_anypcb_byaddr(ifa, &tcbinfo);
2191 }
2192
2193 static void
2194 tcp_handle_msgsize(struct ip *ip, struct inpcb *inp)
2195 {
2196 struct rtentry *rt = NULL;
2197 u_short ifscope = IFSCOPE_NONE;
2198 int mtu;
2199 struct sockaddr_in icmpsrc = {
2200 .sin_len = sizeof(struct sockaddr_in),
2201 .sin_family = AF_INET, .sin_port = 0, .sin_addr = { .s_addr = 0 },
2202 .sin_zero = { 0, 0, 0, 0, 0, 0, 0, 0 }
2203 };
2204 struct icmp *icp = NULL;
2205
2206 icp = (struct icmp *)(void *)
2207 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2208
2209 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
2210
2211 /*
2212 * MTU discovery:
2213 * If we got a needfrag and there is a host route to the
2214 * original destination, and the MTU is not locked, then
2215 * set the MTU in the route to the suggested new value
2216 * (if given) and then notify as usual. The ULPs will
2217 * notice that the MTU has changed and adapt accordingly.
2218 * If no new MTU was suggested, then we guess a new one
2219 * less than the current value. If the new MTU is
2220 * unreasonably small (defined by sysctl tcp_minmss), then
2221 * we reset the MTU to the interface value and enable the
2222 * lock bit, indicating that we are no longer doing MTU
2223 * discovery.
2224 */
2225 if (ROUTE_UNUSABLE(&(inp->inp_route)) == false) {
2226 rt = inp->inp_route.ro_rt;
2227 }
2228
2229 /*
2230 * icmp6_mtudisc_update scopes the routing lookup
2231 * to the incoming interface (delivered from mbuf
2232 * packet header.
2233 * That is mostly ok but for asymmetric networks
2234 * that may be an issue.
2235 * Frag needed OR Packet too big really communicates
2236 * MTU for the out data path.
2237 * Take the interface scope from cached route or
2238 * the last outgoing interface from inp
2239 */
2240 if (rt != NULL) {
2241 ifscope = (rt->rt_ifp != NULL) ?
2242 rt->rt_ifp->if_index : IFSCOPE_NONE;
2243 } else {
2244 ifscope = (inp->inp_last_outifp != NULL) ?
2245 inp->inp_last_outifp->if_index : IFSCOPE_NONE;
2246 }
2247
2248 if ((rt == NULL) ||
2249 !(rt->rt_flags & RTF_HOST) ||
2250 (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING))) {
2251 rt = rtalloc1_scoped((struct sockaddr *)&icmpsrc, 0,
2252 RTF_CLONING | RTF_PRCLONING, ifscope);
2253 } else if (rt) {
2254 RT_LOCK(rt);
2255 rtref(rt);
2256 RT_UNLOCK(rt);
2257 }
2258
2259 if (rt != NULL) {
2260 RT_LOCK(rt);
2261 if ((rt->rt_flags & RTF_HOST) &&
2262 !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
2263 mtu = ntohs(icp->icmp_nextmtu);
2264 /*
2265 * XXX Stock BSD has changed the following
2266 * to compare with icp->icmp_ip.ip_len
2267 * to converge faster when sent packet
2268 * < route's MTU. We may want to adopt
2269 * that change.
2270 */
2271 if (mtu == 0) {
2272 mtu = ip_next_mtu(rt->rt_rmx.
2273 rmx_mtu, 1);
2274 }
2275 #if DEBUG_MTUDISC
2276 printf("MTU for %s reduced to %d\n",
2277 inet_ntop(AF_INET,
2278 &icmpsrc.sin_addr, ipv4str,
2279 sizeof(ipv4str)), mtu);
2280 #endif
2281 if (mtu < max(296, (tcp_minmss +
2282 sizeof(struct tcpiphdr)))) {
2283 rt->rt_rmx.rmx_locks |= RTV_MTU;
2284 } else if (rt->rt_rmx.rmx_mtu > mtu) {
2285 rt->rt_rmx.rmx_mtu = mtu;
2286 }
2287 }
2288 RT_UNLOCK(rt);
2289 rtfree(rt);
2290 }
2291 }
2292
2293 void
2294 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp)
2295 {
2296 tcp_seq icmp_tcp_seq;
2297 struct ip *ip = vip;
2298 struct in_addr faddr;
2299 struct inpcb *inp;
2300 struct tcpcb *tp;
2301 struct tcphdr *th;
2302 struct icmp *icp;
2303 void (*notify)(struct inpcb *, int) = tcp_notify;
2304
2305 faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr;
2306 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) {
2307 return;
2308 }
2309
2310 if ((unsigned)cmd >= PRC_NCMDS) {
2311 return;
2312 }
2313
2314 /* Source quench is deprecated */
2315 if (cmd == PRC_QUENCH) {
2316 return;
2317 }
2318
2319 if (cmd == PRC_MSGSIZE) {
2320 notify = tcp_mtudisc;
2321 } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2322 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2323 cmd == PRC_TIMXCEED_INTRANS) && ip) {
2324 notify = tcp_drop_syn_sent;
2325 }
2326 /*
2327 * Hostdead is ugly because it goes linearly through all PCBs.
2328 * XXX: We never get this from ICMP, otherwise it makes an
2329 * excellent DoS attack on machines with many connections.
2330 */
2331 else if (cmd == PRC_HOSTDEAD) {
2332 ip = NULL;
2333 } else if (inetctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) {
2334 return;
2335 }
2336
2337
2338 if (ip == NULL) {
2339 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
2340 return;
2341 }
2342
2343 icp = (struct icmp *)(void *)
2344 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2345 th = (struct tcphdr *)(void *)((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2));
2346 icmp_tcp_seq = ntohl(th->th_seq);
2347
2348 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
2349 ip->ip_src, th->th_sport, 0, NULL);
2350
2351 if (inp == NULL ||
2352 inp->inp_socket == NULL) {
2353 return;
2354 }
2355
2356 socket_lock(inp->inp_socket, 1);
2357 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) ==
2358 WNT_STOPUSING) {
2359 socket_unlock(inp->inp_socket, 1);
2360 return;
2361 }
2362
2363 if (PRC_IS_REDIRECT(cmd)) {
2364 /* signal EHOSTDOWN, as it flushes the cached route */
2365 (*notify)(inp, EHOSTDOWN);
2366 } else {
2367 tp = intotcpcb(inp);
2368 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2369 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2370 if (cmd == PRC_MSGSIZE) {
2371 tcp_handle_msgsize(ip, inp);
2372 }
2373
2374 (*notify)(inp, inetctlerrmap[cmd]);
2375 }
2376 }
2377 socket_unlock(inp->inp_socket, 1);
2378 }
2379
2380 void
2381 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp)
2382 {
2383 tcp_seq icmp_tcp_seq;
2384 struct in6_addr *dst;
2385 void (*notify)(struct inpcb *, int) = tcp_notify;
2386 struct ip6_hdr *ip6;
2387 struct mbuf *m;
2388 struct inpcb *inp;
2389 struct tcpcb *tp;
2390 struct icmp6_hdr *icmp6;
2391 struct ip6ctlparam *ip6cp = NULL;
2392 const struct sockaddr_in6 *sa6_src = NULL;
2393 unsigned int mtu;
2394 unsigned int off;
2395
2396 struct tcp_ports {
2397 uint16_t th_sport;
2398 uint16_t th_dport;
2399 } t_ports;
2400
2401 if (sa->sa_family != AF_INET6 ||
2402 sa->sa_len != sizeof(struct sockaddr_in6)) {
2403 return;
2404 }
2405
2406 /* Source quench is deprecated */
2407 if (cmd == PRC_QUENCH) {
2408 return;
2409 }
2410
2411 if ((unsigned)cmd >= PRC_NCMDS) {
2412 return;
2413 }
2414
2415 /* if the parameter is from icmp6, decode it. */
2416 if (d != NULL) {
2417 ip6cp = (struct ip6ctlparam *)d;
2418 icmp6 = ip6cp->ip6c_icmp6;
2419 m = ip6cp->ip6c_m;
2420 ip6 = ip6cp->ip6c_ip6;
2421 off = ip6cp->ip6c_off;
2422 sa6_src = ip6cp->ip6c_src;
2423 dst = ip6cp->ip6c_finaldst;
2424 } else {
2425 m = NULL;
2426 ip6 = NULL;
2427 off = 0; /* fool gcc */
2428 sa6_src = &sa6_any;
2429 dst = NULL;
2430 }
2431
2432 if (cmd == PRC_MSGSIZE) {
2433 notify = tcp_mtudisc;
2434 } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2435 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) &&
2436 ip6 != NULL) {
2437 notify = tcp_drop_syn_sent;
2438 }
2439 /*
2440 * Hostdead is ugly because it goes linearly through all PCBs.
2441 * XXX: We never get this from ICMP, otherwise it makes an
2442 * excellent DoS attack on machines with many connections.
2443 */
2444 else if (cmd == PRC_HOSTDEAD) {
2445 ip6 = NULL;
2446 } else if (inet6ctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) {
2447 return;
2448 }
2449
2450
2451 if (ip6 == NULL) {
2452 in6_pcbnotify(&tcbinfo, sa, 0, (struct sockaddr *)(size_t)sa6_src,
2453 0, cmd, NULL, notify);
2454 return;
2455 }
2456
2457 /* Check if we can safely get the ports from the tcp hdr */
2458 if (m == NULL ||
2459 (m->m_pkthdr.len <
2460 (int32_t) (off + sizeof(struct tcp_ports)))) {
2461 return;
2462 }
2463 bzero(&t_ports, sizeof(struct tcp_ports));
2464 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2465
2466 off += sizeof(struct tcp_ports);
2467 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2468 return;
2469 }
2470 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2471 icmp_tcp_seq = ntohl(icmp_tcp_seq);
2472
2473 if (cmd == PRC_MSGSIZE) {
2474 mtu = ntohl(icmp6->icmp6_mtu);
2475 /*
2476 * If no alternative MTU was proposed, or the proposed
2477 * MTU was too small, set to the min.
2478 */
2479 if (mtu < IPV6_MMTU) {
2480 mtu = IPV6_MMTU - 8;
2481 }
2482 }
2483
2484 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_dst, t_ports.th_dport,
2485 &ip6->ip6_src, t_ports.th_sport, 0, NULL);
2486
2487 if (inp == NULL ||
2488 inp->inp_socket == NULL) {
2489 return;
2490 }
2491
2492 socket_lock(inp->inp_socket, 1);
2493 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) ==
2494 WNT_STOPUSING) {
2495 socket_unlock(inp->inp_socket, 1);
2496 return;
2497 }
2498
2499 if (PRC_IS_REDIRECT(cmd)) {
2500 /* signal EHOSTDOWN, as it flushes the cached route */
2501 (*notify)(inp, EHOSTDOWN);
2502 } else {
2503 tp = intotcpcb(inp);
2504 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2505 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2506 if (cmd == PRC_MSGSIZE) {
2507 /*
2508 * Only process the offered MTU if it
2509 * is smaller than the current one.
2510 */
2511 if (mtu < tp->t_maxseg +
2512 (sizeof(struct tcphdr) + sizeof(struct ip6_hdr))) {
2513 (*notify)(inp, inetctlerrmap[cmd]);
2514 }
2515 } else {
2516 (*notify)(inp, inetctlerrmap[cmd]);
2517 }
2518 }
2519 }
2520 socket_unlock(inp->inp_socket, 1);
2521 }
2522
2523
2524 /*
2525 * Following is where TCP initial sequence number generation occurs.
2526 *
2527 * There are two places where we must use initial sequence numbers:
2528 * 1. In SYN-ACK packets.
2529 * 2. In SYN packets.
2530 *
2531 * The ISNs in SYN-ACK packets have no monotonicity requirement,
2532 * and should be as unpredictable as possible to avoid the possibility
2533 * of spoofing and/or connection hijacking. To satisfy this
2534 * requirement, SYN-ACK ISNs are generated via the arc4random()
2535 * function. If exact RFC 1948 compliance is requested via sysctl,
2536 * these ISNs will be generated just like those in SYN packets.
2537 *
2538 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2539 * depends on this property. In addition, these ISNs should be
2540 * unguessable so as to prevent connection hijacking. To satisfy
2541 * the requirements of this situation, the algorithm outlined in
2542 * RFC 1948 is used to generate sequence numbers.
2543 *
2544 * For more information on the theory of operation, please see
2545 * RFC 1948.
2546 *
2547 * Implementation details:
2548 *
2549 * Time is based off the system timer, and is corrected so that it
2550 * increases by one megabyte per second. This allows for proper
2551 * recycling on high speed LANs while still leaving over an hour
2552 * before rollover.
2553 *
2554 * Two sysctls control the generation of ISNs:
2555 *
2556 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2557 * between seeding of isn_secret. This is normally set to zero,
2558 * as reseeding should not be necessary.
2559 *
2560 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
2561 * strictly. When strict compliance is requested, reseeding is
2562 * disabled and SYN-ACKs will be generated in the same manner as
2563 * SYNs. Strict mode is disabled by default.
2564 *
2565 */
2566
2567 #define ISN_BYTES_PER_SECOND 1048576
2568
2569 tcp_seq
2570 tcp_new_isn(struct tcpcb *tp)
2571 {
2572 u_int32_t md5_buffer[4];
2573 tcp_seq new_isn;
2574 struct timeval timenow;
2575 u_char isn_secret[32];
2576 long isn_last_reseed = 0;
2577 MD5_CTX isn_ctx;
2578
2579 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
2580 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) &&
2581 tcp_strict_rfc1948 == 0)
2582 #ifdef __APPLE__
2583 { return RandomULong(); }
2584 #else
2585 { return arc4random(); }
2586 #endif
2587 getmicrotime(&timenow);
2588
2589 /* Seed if this is the first use, reseed if requested. */
2590 if ((isn_last_reseed == 0) ||
2591 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
2592 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval * hz)
2593 < (u_int)timenow.tv_sec))) {
2594 #ifdef __APPLE__
2595 read_frandom(&isn_secret, sizeof(isn_secret));
2596 #else
2597 read_random_unlimited(&isn_secret, sizeof(isn_secret));
2598 #endif
2599 isn_last_reseed = timenow.tv_sec;
2600 }
2601
2602 /* Compute the md5 hash and return the ISN. */
2603 MD5Init(&isn_ctx);
2604 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport,
2605 sizeof(u_short));
2606 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport,
2607 sizeof(u_short));
2608 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2609 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2610 sizeof(struct in6_addr));
2611 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2612 sizeof(struct in6_addr));
2613 } else {
2614 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2615 sizeof(struct in_addr));
2616 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2617 sizeof(struct in_addr));
2618 }
2619 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
2620 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2621 new_isn = (tcp_seq) md5_buffer[0];
2622 new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz);
2623 return new_isn;
2624 }
2625
2626
2627 /*
2628 * When a specific ICMP unreachable message is received and the
2629 * connection state is SYN-SENT, drop the connection. This behavior
2630 * is controlled by the icmp_may_rst sysctl.
2631 */
2632 void
2633 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2634 {
2635 struct tcpcb *tp = intotcpcb(inp);
2636
2637 if (tp && tp->t_state == TCPS_SYN_SENT) {
2638 tcp_drop(tp, errno);
2639 }
2640 }
2641
2642 /*
2643 * When `need fragmentation' ICMP is received, update our idea of the MSS
2644 * based on the new value in the route. Also nudge TCP to send something,
2645 * since we know the packet we just sent was dropped.
2646 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2647 */
2648 void
2649 tcp_mtudisc(struct inpcb *inp, __unused int errno)
2650 {
2651 struct tcpcb *tp = intotcpcb(inp);
2652 struct rtentry *rt;
2653 struct socket *so = inp->inp_socket;
2654 int mss;
2655 u_int32_t mtu;
2656 u_int32_t protoHdrOverhead = sizeof(struct tcpiphdr);
2657 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
2658
2659 /*
2660 * Nothing left to send after the socket is defunct or TCP is in the closed state
2661 */
2662 if ((so->so_state & SS_DEFUNCT) || (tp != NULL && tp->t_state == TCPS_CLOSED)) {
2663 return;
2664 }
2665
2666 if (isipv6) {
2667 protoHdrOverhead = sizeof(struct ip6_hdr) +
2668 sizeof(struct tcphdr);
2669 }
2670
2671 if (tp != NULL) {
2672 if (isipv6) {
2673 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2674 } else {
2675 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2676 }
2677 if (!rt || !rt->rt_rmx.rmx_mtu) {
2678 tp->t_maxopd = tp->t_maxseg =
2679 isipv6 ? tcp_v6mssdflt :
2680 tcp_mssdflt;
2681
2682 /* Route locked during lookup above */
2683 if (rt != NULL) {
2684 RT_UNLOCK(rt);
2685 }
2686 return;
2687 }
2688 mtu = rt->rt_rmx.rmx_mtu;
2689
2690 /* Route locked during lookup above */
2691 RT_UNLOCK(rt);
2692
2693 #if NECP
2694 // Adjust MTU if necessary.
2695 mtu = necp_socket_get_effective_mtu(inp, mtu);
2696 #endif /* NECP */
2697 mss = mtu - protoHdrOverhead;
2698
2699 if (tp->t_maxopd) {
2700 mss = min(mss, tp->t_maxopd);
2701 }
2702 /*
2703 * XXX - The above conditional probably violates the TCP
2704 * spec. The problem is that, since we don't know the
2705 * other end's MSS, we are supposed to use a conservative
2706 * default. But, if we do that, then MTU discovery will
2707 * never actually take place, because the conservative
2708 * default is much less than the MTUs typically seen
2709 * on the Internet today. For the moment, we'll sweep
2710 * this under the carpet.
2711 *
2712 * The conservative default might not actually be a problem
2713 * if the only case this occurs is when sending an initial
2714 * SYN with options and data to a host we've never talked
2715 * to before. Then, they will reply with an MSS value which
2716 * will get recorded and the new parameters should get
2717 * recomputed. For Further Study.
2718 */
2719 if (tp->t_maxopd <= mss) {
2720 return;
2721 }
2722 tp->t_maxopd = mss;
2723
2724 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2725 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) {
2726 mss -= TCPOLEN_TSTAMP_APPA;
2727 }
2728
2729 #if MPTCP
2730 mss -= mptcp_adj_mss(tp, TRUE);
2731 #endif
2732 if (so->so_snd.sb_hiwat < mss) {
2733 mss = so->so_snd.sb_hiwat;
2734 }
2735
2736 tp->t_maxseg = mss;
2737
2738 ASSERT(tp->t_maxseg);
2739
2740 /*
2741 * Reset the slow-start flight size as it may depends on the
2742 * new MSS
2743 */
2744 if (CC_ALGO(tp)->cwnd_init != NULL) {
2745 CC_ALGO(tp)->cwnd_init(tp);
2746 }
2747 tcpstat.tcps_mturesent++;
2748 tp->t_rtttime = 0;
2749 tp->snd_nxt = tp->snd_una;
2750 tcp_output(tp);
2751 }
2752 }
2753
2754 /*
2755 * Look-up the routing entry to the peer of this inpcb. If no route
2756 * is found and it cannot be allocated the return NULL. This routine
2757 * is called by TCP routines that access the rmx structure and by tcp_mss
2758 * to get the interface MTU. If a route is found, this routine will
2759 * hold the rtentry lock; the caller is responsible for unlocking.
2760 */
2761 struct rtentry *
2762 tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope)
2763 {
2764 struct route *ro;
2765 struct rtentry *rt;
2766 struct tcpcb *tp;
2767
2768 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2769
2770 ro = &inp->inp_route;
2771 if ((rt = ro->ro_rt) != NULL) {
2772 RT_LOCK(rt);
2773 }
2774
2775 if (ROUTE_UNUSABLE(ro)) {
2776 if (rt != NULL) {
2777 RT_UNLOCK(rt);
2778 rt = NULL;
2779 }
2780 ROUTE_RELEASE(ro);
2781 /* No route yet, so try to acquire one */
2782 if (inp->inp_faddr.s_addr != INADDR_ANY) {
2783 unsigned int ifscope;
2784
2785 ro->ro_dst.sa_family = AF_INET;
2786 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
2787 ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr =
2788 inp->inp_faddr;
2789
2790 /*
2791 * If the socket was bound to an interface, then
2792 * the bound-to-interface takes precedence over
2793 * the inbound interface passed in by the caller
2794 * (if we get here as part of the output path then
2795 * input_ifscope is IFSCOPE_NONE).
2796 */
2797 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2798 inp->inp_boundifp->if_index : input_ifscope;
2799
2800 rtalloc_scoped(ro, ifscope);
2801 if ((rt = ro->ro_rt) != NULL) {
2802 RT_LOCK(rt);
2803 }
2804 }
2805 }
2806 if (rt != NULL) {
2807 RT_LOCK_ASSERT_HELD(rt);
2808 }
2809
2810 /*
2811 * Update MTU discovery determination. Don't do it if:
2812 * 1) it is disabled via the sysctl
2813 * 2) the route isn't up
2814 * 3) the MTU is locked (if it is, then discovery has been
2815 * disabled)
2816 */
2817
2818 tp = intotcpcb(inp);
2819
2820 if (!path_mtu_discovery || ((rt != NULL) &&
2821 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) {
2822 tp->t_flags &= ~TF_PMTUD;
2823 } else {
2824 tp->t_flags |= TF_PMTUD;
2825 }
2826
2827 if (rt != NULL && rt->rt_ifp != NULL) {
2828 somultipages(inp->inp_socket,
2829 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2830 tcp_set_tso(tp, rt->rt_ifp);
2831 soif2kcl(inp->inp_socket,
2832 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2833 tcp_set_ecn(tp, rt->rt_ifp);
2834 if (inp->inp_last_outifp == NULL) {
2835 inp->inp_last_outifp = rt->rt_ifp;
2836 }
2837 }
2838
2839 /* Note if the peer is local */
2840 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2841 (rt->rt_gateway->sa_family == AF_LINK ||
2842 rt->rt_ifp->if_flags & IFF_LOOPBACK ||
2843 in_localaddr(inp->inp_faddr))) {
2844 tp->t_flags |= TF_LOCAL;
2845 }
2846
2847 /*
2848 * Caller needs to call RT_UNLOCK(rt).
2849 */
2850 return rt;
2851 }
2852
2853 struct rtentry *
2854 tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope)
2855 {
2856 struct route_in6 *ro6;
2857 struct rtentry *rt;
2858 struct tcpcb *tp;
2859
2860 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2861
2862 ro6 = &inp->in6p_route;
2863 if ((rt = ro6->ro_rt) != NULL) {
2864 RT_LOCK(rt);
2865 }
2866
2867 if (ROUTE_UNUSABLE(ro6)) {
2868 if (rt != NULL) {
2869 RT_UNLOCK(rt);
2870 rt = NULL;
2871 }
2872 ROUTE_RELEASE(ro6);
2873 /* No route yet, so try to acquire one */
2874 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
2875 struct sockaddr_in6 *dst6;
2876 unsigned int ifscope;
2877
2878 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst;
2879 dst6->sin6_family = AF_INET6;
2880 dst6->sin6_len = sizeof(*dst6);
2881 dst6->sin6_addr = inp->in6p_faddr;
2882
2883 /*
2884 * If the socket was bound to an interface, then
2885 * the bound-to-interface takes precedence over
2886 * the inbound interface passed in by the caller
2887 * (if we get here as part of the output path then
2888 * input_ifscope is IFSCOPE_NONE).
2889 */
2890 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2891 inp->inp_boundifp->if_index : input_ifscope;
2892
2893 rtalloc_scoped((struct route *)ro6, ifscope);
2894 if ((rt = ro6->ro_rt) != NULL) {
2895 RT_LOCK(rt);
2896 }
2897 }
2898 }
2899 if (rt != NULL) {
2900 RT_LOCK_ASSERT_HELD(rt);
2901 }
2902
2903 /*
2904 * Update path MTU Discovery determination
2905 * while looking up the route:
2906 * 1) we have a valid route to the destination
2907 * 2) the MTU is not locked (if it is, then discovery has been
2908 * disabled)
2909 */
2910
2911
2912 tp = intotcpcb(inp);
2913
2914 /*
2915 * Update MTU discovery determination. Don't do it if:
2916 * 1) it is disabled via the sysctl
2917 * 2) the route isn't up
2918 * 3) the MTU is locked (if it is, then discovery has been
2919 * disabled)
2920 */
2921
2922 if (!path_mtu_discovery || ((rt != NULL) &&
2923 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) {
2924 tp->t_flags &= ~TF_PMTUD;
2925 } else {
2926 tp->t_flags |= TF_PMTUD;
2927 }
2928
2929 if (rt != NULL && rt->rt_ifp != NULL) {
2930 somultipages(inp->inp_socket,
2931 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2932 tcp_set_tso(tp, rt->rt_ifp);
2933 soif2kcl(inp->inp_socket,
2934 (rt->rt_ifp->if_eflags & IFEF_2KCL));
2935 tcp_set_ecn(tp, rt->rt_ifp);
2936 if (inp->inp_last_outifp == NULL) {
2937 inp->inp_last_outifp = rt->rt_ifp;
2938 }
2939
2940 /* Note if the peer is local */
2941 if (!(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
2942 (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) ||
2943 IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) ||
2944 rt->rt_gateway->sa_family == AF_LINK ||
2945 in6_localaddr(&inp->in6p_faddr))) {
2946 tp->t_flags |= TF_LOCAL;
2947 }
2948 }
2949
2950 /*
2951 * Caller needs to call RT_UNLOCK(rt).
2952 */
2953 return rt;
2954 }
2955
2956 #if IPSEC
2957 /* compute ESP/AH header size for TCP, including outer IP header. */
2958 size_t
2959 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2960 {
2961 struct inpcb *inp;
2962 struct mbuf *m;
2963 size_t hdrsiz;
2964 struct ip *ip;
2965 struct ip6_hdr *ip6 = NULL;
2966 struct tcphdr *th;
2967
2968 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) {
2969 return 0;
2970 }
2971 MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */
2972 if (!m) {
2973 return 0;
2974 }
2975
2976 if ((inp->inp_vflag & INP_IPV6) != 0) {
2977 ip6 = mtod(m, struct ip6_hdr *);
2978 th = (struct tcphdr *)(void *)(ip6 + 1);
2979 m->m_pkthdr.len = m->m_len =
2980 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2981 tcp_fillheaders(tp, ip6, th);
2982 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2983 } else {
2984 ip = mtod(m, struct ip *);
2985 th = (struct tcphdr *)(ip + 1);
2986 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2987 tcp_fillheaders(tp, ip, th);
2988 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2989 }
2990 m_free(m);
2991 return hdrsiz;
2992 }
2993 #endif /* IPSEC */
2994
2995 int
2996 tcp_lock(struct socket *so, int refcount, void *lr)
2997 {
2998 void *lr_saved;
2999
3000 if (lr == NULL) {
3001 lr_saved = __builtin_return_address(0);
3002 } else {
3003 lr_saved = lr;
3004 }
3005
3006 retry:
3007 if (so->so_pcb != NULL) {
3008 if (so->so_flags & SOF_MP_SUBFLOW) {
3009 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3010 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3011
3012 socket_lock(mp_so, refcount);
3013
3014 /*
3015 * Check if we became non-MPTCP while waiting for the lock.
3016 * If yes, we have to retry to grab the right lock.
3017 */
3018 if (!(so->so_flags & SOF_MP_SUBFLOW)) {
3019 socket_unlock(mp_so, refcount);
3020 goto retry;
3021 }
3022 } else {
3023 lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3024
3025 if (so->so_flags & SOF_MP_SUBFLOW) {
3026 /*
3027 * While waiting for the lock, we might have
3028 * become MPTCP-enabled (see mptcp_subflow_socreate).
3029 */
3030 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3031 goto retry;
3032 }
3033 }
3034 } else {
3035 panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n",
3036 so, lr_saved, solockhistory_nr(so));
3037 /* NOTREACHED */
3038 }
3039
3040 if (so->so_usecount < 0) {
3041 panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n",
3042 so, so->so_pcb, lr_saved, so->so_usecount,
3043 solockhistory_nr(so));
3044 /* NOTREACHED */
3045 }
3046 if (refcount) {
3047 so->so_usecount++;
3048 }
3049 so->lock_lr[so->next_lock_lr] = lr_saved;
3050 so->next_lock_lr = (so->next_lock_lr + 1) % SO_LCKDBG_MAX;
3051 return 0;
3052 }
3053
3054 int
3055 tcp_unlock(struct socket *so, int refcount, void *lr)
3056 {
3057 void *lr_saved;
3058
3059 if (lr == NULL) {
3060 lr_saved = __builtin_return_address(0);
3061 } else {
3062 lr_saved = lr;
3063 }
3064
3065 #ifdef MORE_TCPLOCK_DEBUG
3066 printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x "
3067 "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so),
3068 (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb),
3069 (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)),
3070 so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved));
3071 #endif
3072 if (refcount) {
3073 so->so_usecount--;
3074 }
3075
3076 if (so->so_usecount < 0) {
3077 panic("tcp_unlock: so=%p usecount=%x lrh= %s\n",
3078 so, so->so_usecount, solockhistory_nr(so));
3079 /* NOTREACHED */
3080 }
3081 if (so->so_pcb == NULL) {
3082 panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n",
3083 so, so->so_usecount, lr_saved, solockhistory_nr(so));
3084 /* NOTREACHED */
3085 } else {
3086 so->unlock_lr[so->next_unlock_lr] = lr_saved;
3087 so->next_unlock_lr = (so->next_unlock_lr + 1) % SO_LCKDBG_MAX;
3088
3089 if (so->so_flags & SOF_MP_SUBFLOW) {
3090 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3091 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3092
3093 socket_lock_assert_owned(mp_so);
3094
3095 socket_unlock(mp_so, refcount);
3096 } else {
3097 LCK_MTX_ASSERT(&((struct inpcb *)so->so_pcb)->inpcb_mtx,
3098 LCK_MTX_ASSERT_OWNED);
3099 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3100 }
3101 }
3102 return 0;
3103 }
3104
3105 lck_mtx_t *
3106 tcp_getlock(struct socket *so, int flags)
3107 {
3108 struct inpcb *inp = sotoinpcb(so);
3109
3110 if (so->so_pcb) {
3111 if (so->so_usecount < 0) {
3112 panic("tcp_getlock: so=%p usecount=%x lrh= %s\n",
3113 so, so->so_usecount, solockhistory_nr(so));
3114 }
3115
3116 if (so->so_flags & SOF_MP_SUBFLOW) {
3117 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3118 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3119
3120 return mp_so->so_proto->pr_getlock(mp_so, flags);
3121 } else {
3122 return &inp->inpcb_mtx;
3123 }
3124 } else {
3125 panic("tcp_getlock: so=%p NULL so_pcb %s\n",
3126 so, solockhistory_nr(so));
3127 return so->so_proto->pr_domain->dom_mtx;
3128 }
3129 }
3130
3131 /*
3132 * Determine if we can grow the recieve socket buffer to avoid sending
3133 * a zero window update to the peer. We allow even socket buffers that
3134 * have fixed size (set by the application) to grow if the resource
3135 * constraints are met. They will also be trimmed after the application
3136 * reads data.
3137 */
3138 static void
3139 tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb)
3140 {
3141 u_int32_t rcvbufinc = tp->t_maxseg << 4;
3142 u_int32_t rcvbuf = sb->sb_hiwat;
3143 struct socket *so = tp->t_inpcb->inp_socket;
3144
3145 if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) {
3146 return;
3147 }
3148
3149 if (tcp_do_autorcvbuf == 1 &&
3150 tcp_cansbgrow(sb) &&
3151 (tp->t_flags & TF_SLOWLINK) == 0 &&
3152 (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 &&
3153 (rcvbuf - sb->sb_cc) < rcvbufinc &&
3154 rcvbuf < tcp_autorcvbuf_max &&
3155 (sb->sb_idealsize > 0 &&
3156 sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) {
3157 sbreserve(sb,
3158 min((sb->sb_hiwat + rcvbufinc), tcp_autorcvbuf_max));
3159 }
3160 }
3161
3162 int32_t
3163 tcp_sbspace(struct tcpcb *tp)
3164 {
3165 struct socket *so = tp->t_inpcb->inp_socket;
3166 struct sockbuf *sb = &so->so_rcv;
3167 u_int32_t rcvbuf;
3168 int32_t space;
3169 int32_t pending = 0;
3170
3171 if (so->so_flags & SOF_MP_SUBFLOW) {
3172 /* We still need to grow TCP's buffer to have a BDP-estimate */
3173 tcp_sbrcv_grow_rwin(tp, sb);
3174
3175 return mptcp_sbspace(tptomptp(tp));
3176 }
3177
3178 tcp_sbrcv_grow_rwin(tp, sb);
3179
3180 /* hiwat might have changed */
3181 rcvbuf = sb->sb_hiwat;
3182
3183 space = ((int32_t) imin((rcvbuf - sb->sb_cc),
3184 (sb->sb_mbmax - sb->sb_mbcnt)));
3185 if (space < 0) {
3186 space = 0;
3187 }
3188
3189 #if CONTENT_FILTER
3190 /* Compensate for data being processed by content filters */
3191 pending = cfil_sock_data_space(sb);
3192 #endif /* CONTENT_FILTER */
3193 if (pending > space) {
3194 space = 0;
3195 } else {
3196 space -= pending;
3197 }
3198
3199 /*
3200 * Avoid increasing window size if the current window
3201 * is already very low, we could be in "persist" mode and
3202 * we could break some apps (see rdar://5409343)
3203 */
3204
3205 if (space < tp->t_maxseg) {
3206 return space;
3207 }
3208
3209 /* Clip window size for slower link */
3210
3211 if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) {
3212 return imin(space, slowlink_wsize);
3213 }
3214
3215 return space;
3216 }
3217 /*
3218 * Checks TCP Segment Offloading capability for a given connection
3219 * and interface pair.
3220 */
3221 void
3222 tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp)
3223 {
3224 struct inpcb *inp;
3225 int isipv6;
3226 struct ifnet *tunnel_ifp = NULL;
3227 #define IFNET_TSO_MASK (IFNET_TSO_IPV6 | IFNET_TSO_IPV4)
3228
3229 tp->t_flags &= ~TF_TSO;
3230
3231 if (ifp == NULL) {
3232 return;
3233 }
3234
3235 #if MPTCP
3236 /*
3237 * We can't use TSO if this tcpcb belongs to an MPTCP session.
3238 */
3239 if (tp->t_mpflags & TMPF_MPTCP_TRUE) {
3240 return;
3241 }
3242 #endif
3243 inp = tp->t_inpcb;
3244 isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
3245
3246 /*
3247 * We can't use TSO if the TSO capability of the tunnel interface does
3248 * not match the capability of another interface known by TCP
3249 */
3250 if (inp->inp_policyresult.results.result == NECP_KERNEL_POLICY_RESULT_IP_TUNNEL) {
3251 u_int tunnel_if_index = inp->inp_policyresult.results.result_parameter.tunnel_interface_index;
3252
3253 if (tunnel_if_index != 0) {
3254 ifnet_head_lock_shared();
3255 tunnel_ifp = ifindex2ifnet[tunnel_if_index];
3256 ifnet_head_done();
3257 }
3258
3259 if (tunnel_ifp == NULL) {
3260 return;
3261 }
3262
3263 if ((ifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3264 if (tso_debug > 0) {
3265 os_log(OS_LOG_DEFAULT,
3266 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with ifp %s",
3267 __func__,
3268 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3269 tunnel_ifp->if_xname, ifp->if_xname);
3270 }
3271 return;
3272 }
3273 if (inp->inp_last_outifp != NULL &&
3274 (inp->inp_last_outifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3275 if (tso_debug > 0) {
3276 os_log(OS_LOG_DEFAULT,
3277 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_last_outifp %s",
3278 __func__,
3279 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3280 tunnel_ifp->if_xname, inp->inp_last_outifp->if_xname);
3281 }
3282 return;
3283 }
3284 if ((inp->inp_flags & INP_BOUND_IF) && inp->inp_boundifp != NULL &&
3285 (inp->inp_boundifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3286 if (tso_debug > 0) {
3287 os_log(OS_LOG_DEFAULT,
3288 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_boundifp %s",
3289 __func__,
3290 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3291 tunnel_ifp->if_xname, inp->inp_boundifp->if_xname);
3292 }
3293 return;
3294 }
3295 }
3296
3297 if (isipv6) {
3298 if (ifp->if_hwassist & IFNET_TSO_IPV6) {
3299 tp->t_flags |= TF_TSO;
3300 if (ifp->if_tso_v6_mtu != 0) {
3301 tp->tso_max_segment_size = ifp->if_tso_v6_mtu;
3302 } else {
3303 tp->tso_max_segment_size = TCP_MAXWIN;
3304 }
3305 }
3306 } else {
3307 if (ifp->if_hwassist & IFNET_TSO_IPV4) {
3308 tp->t_flags |= TF_TSO;
3309 if (ifp->if_tso_v4_mtu != 0) {
3310 tp->tso_max_segment_size = ifp->if_tso_v4_mtu;
3311 } else {
3312 tp->tso_max_segment_size = TCP_MAXWIN;
3313 }
3314 if (INTF_ADJUST_MTU_FOR_CLAT46(ifp)) {
3315 tp->tso_max_segment_size -=
3316 CLAT46_HDR_EXPANSION_OVERHD;
3317 }
3318 }
3319 }
3320
3321 if (tso_debug > 1) {
3322 os_log(OS_LOG_DEFAULT, "%s: %u > %u TSO %d ifp %s",
3323 __func__,
3324 ntohs(tp->t_inpcb->inp_lport),
3325 ntohs(tp->t_inpcb->inp_fport),
3326 (tp->t_flags & TF_TSO) != 0,
3327 ifp != NULL ? ifp->if_xname : "<NULL>");
3328 }
3329 }
3330
3331 #define TIMEVAL_TO_TCPHZ(_tv_) ((uint32_t)((_tv_).tv_sec * TCP_RETRANSHZ + \
3332 (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC))
3333
3334 /*
3335 * Function to calculate the tcp clock. The tcp clock will get updated
3336 * at the boundaries of the tcp layer. This is done at 3 places:
3337 * 1. Right before processing an input tcp packet
3338 * 2. Whenever a connection wants to access the network using tcp_usrreqs
3339 * 3. When a tcp timer fires or before tcp slow timeout
3340 *
3341 */
3342
3343 void
3344 calculate_tcp_clock(void)
3345 {
3346 struct timeval tv = tcp_uptime;
3347 struct timeval interval = {.tv_sec = 0, .tv_usec = TCP_RETRANSHZ_TO_USEC};
3348 struct timeval now, hold_now;
3349 uint32_t incr = 0;
3350
3351 microuptime(&now);
3352
3353 /*
3354 * Update coarse-grained networking timestamp (in sec.); the idea
3355 * is to update the counter returnable via net_uptime() when
3356 * we read time.
3357 */
3358 net_update_uptime_with_time(&now);
3359
3360 timevaladd(&tv, &interval);
3361 if (timevalcmp(&now, &tv, >)) {
3362 /* time to update the clock */
3363 lck_spin_lock(tcp_uptime_lock);
3364 if (timevalcmp(&tcp_uptime, &now, >=)) {
3365 /* clock got updated while waiting for the lock */
3366 lck_spin_unlock(tcp_uptime_lock);
3367 return;
3368 }
3369
3370 microuptime(&now);
3371 hold_now = now;
3372 tv = tcp_uptime;
3373 timevalsub(&now, &tv);
3374
3375 incr = TIMEVAL_TO_TCPHZ(now);
3376 if (incr > 0) {
3377 tcp_uptime = hold_now;
3378 tcp_now += incr;
3379 }
3380
3381 lck_spin_unlock(tcp_uptime_lock);
3382 }
3383 }
3384
3385 /*
3386 * Compute receive window scaling that we are going to request
3387 * for this connection based on sb_hiwat. Try to leave some
3388 * room to potentially increase the window size upto a maximum
3389 * defined by the constant tcp_autorcvbuf_max.
3390 */
3391 void
3392 tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so)
3393 {
3394 uint32_t maxsockbufsize;
3395
3396 tp->request_r_scale = MAX((uint8_t)tcp_win_scale, tp->request_r_scale);
3397 maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ?
3398 so->so_rcv.sb_hiwat : tcp_autorcvbuf_max;
3399
3400 /*
3401 * Window scale should not exceed what is needed
3402 * to send the max receive window size; adding 1 to TCP_MAXWIN
3403 * ensures that.
3404 */
3405 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
3406 ((TCP_MAXWIN + 1) << tp->request_r_scale) < maxsockbufsize) {
3407 tp->request_r_scale++;
3408 }
3409 tp->request_r_scale = MIN(tp->request_r_scale, TCP_MAX_WINSHIFT);
3410 }
3411
3412 int
3413 tcp_notsent_lowat_check(struct socket *so)
3414 {
3415 struct inpcb *inp = sotoinpcb(so);
3416 struct tcpcb *tp = NULL;
3417 int notsent = 0;
3418
3419 if (inp != NULL) {
3420 tp = intotcpcb(inp);
3421 }
3422
3423 if (tp == NULL) {
3424 return 0;
3425 }
3426
3427 notsent = so->so_snd.sb_cc -
3428 (tp->snd_nxt - tp->snd_una);
3429
3430 /*
3431 * When we send a FIN or SYN, not_sent can be negative.
3432 * In that case also we need to send a write event to the
3433 * process if it is waiting. In the FIN case, it will
3434 * get an error from send because cantsendmore will be set.
3435 */
3436 if (notsent <= tp->t_notsent_lowat) {
3437 return 1;
3438 }
3439
3440 /*
3441 * When Nagle's algorithm is not disabled, it is better
3442 * to wakeup the client until there is atleast one
3443 * maxseg of data to write.
3444 */
3445 if ((tp->t_flags & TF_NODELAY) == 0 &&
3446 notsent > 0 && notsent < tp->t_maxseg) {
3447 return 1;
3448 }
3449 return 0;
3450 }
3451
3452 void
3453 tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3454 {
3455 struct tcp_rxt_seg *rxseg = NULL, *prev = NULL, *next = NULL;
3456 uint16_t rxcount = 0;
3457
3458 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3459 tp->t_dsack_lastuna = tp->snd_una;
3460 }
3461 /*
3462 * First check if there is a segment already existing for this
3463 * sequence space.
3464 */
3465
3466 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3467 if (SEQ_GT(rxseg->rx_start, start)) {
3468 break;
3469 }
3470 prev = rxseg;
3471 }
3472 next = rxseg;
3473
3474 /* check if prev seg is for this sequence */
3475 if (prev != NULL && SEQ_LEQ(prev->rx_start, start) &&
3476 SEQ_GEQ(prev->rx_end, end)) {
3477 prev->rx_count++;
3478 return;
3479 }
3480
3481 /*
3482 * There are a couple of possibilities at this point.
3483 * 1. prev overlaps with the beginning of this sequence
3484 * 2. next overlaps with the end of this sequence
3485 * 3. there is no overlap.
3486 */
3487
3488 if (prev != NULL && SEQ_GT(prev->rx_end, start)) {
3489 if (prev->rx_start == start && SEQ_GT(end, prev->rx_end)) {
3490 start = prev->rx_end + 1;
3491 prev->rx_count++;
3492 } else {
3493 prev->rx_end = (start - 1);
3494 rxcount = prev->rx_count;
3495 }
3496 }
3497
3498 if (next != NULL && SEQ_LT(next->rx_start, end)) {
3499 if (SEQ_LEQ(next->rx_end, end)) {
3500 end = next->rx_start - 1;
3501 next->rx_count++;
3502 } else {
3503 next->rx_start = end + 1;
3504 rxcount = next->rx_count;
3505 }
3506 }
3507 if (!SEQ_LT(start, end)) {
3508 return;
3509 }
3510
3511 rxseg = (struct tcp_rxt_seg *) zalloc(tcp_rxt_seg_zone);
3512 if (rxseg == NULL) {
3513 return;
3514 }
3515 bzero(rxseg, sizeof(*rxseg));
3516 rxseg->rx_start = start;
3517 rxseg->rx_end = end;
3518 rxseg->rx_count = rxcount + 1;
3519
3520 if (prev != NULL) {
3521 SLIST_INSERT_AFTER(prev, rxseg, rx_link);
3522 } else {
3523 SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link);
3524 }
3525 }
3526
3527 struct tcp_rxt_seg *
3528 tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3529 {
3530 struct tcp_rxt_seg *rxseg;
3531 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3532 return NULL;
3533 }
3534
3535 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3536 if (SEQ_LEQ(rxseg->rx_start, start) &&
3537 SEQ_GEQ(rxseg->rx_end, end)) {
3538 return rxseg;
3539 }
3540 if (SEQ_GT(rxseg->rx_start, start)) {
3541 break;
3542 }
3543 }
3544 return NULL;
3545 }
3546
3547 void
3548 tcp_rxtseg_set_spurious(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3549 {
3550 struct tcp_rxt_seg *rxseg;
3551 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3552 return;
3553 }
3554
3555 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3556 if (SEQ_GEQ(rxseg->rx_start, start) &&
3557 SEQ_LEQ(rxseg->rx_end, end)) {
3558 /*
3559 * If the segment was retransmitted only once, mark it as
3560 * spurious.
3561 */
3562 if (rxseg->rx_count == 1) {
3563 rxseg->rx_flags |= TCP_RXT_SPURIOUS;
3564 }
3565 }
3566
3567 if (SEQ_GEQ(rxseg->rx_start, end)) {
3568 break;
3569 }
3570 }
3571 return;
3572 }
3573
3574 void
3575 tcp_rxtseg_clean(struct tcpcb *tp)
3576 {
3577 struct tcp_rxt_seg *rxseg, *next;
3578
3579 SLIST_FOREACH_SAFE(rxseg, &tp->t_rxt_segments, rx_link, next) {
3580 SLIST_REMOVE(&tp->t_rxt_segments, rxseg,
3581 tcp_rxt_seg, rx_link);
3582 zfree(tcp_rxt_seg_zone, rxseg);
3583 }
3584 tp->t_dsack_lastuna = tp->snd_max;
3585 }
3586
3587 boolean_t
3588 tcp_rxtseg_detect_bad_rexmt(struct tcpcb *tp, tcp_seq th_ack)
3589 {
3590 boolean_t bad_rexmt;
3591 struct tcp_rxt_seg *rxseg;
3592
3593 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3594 return FALSE;
3595 }
3596
3597 /*
3598 * If all of the segments in this window are not cumulatively
3599 * acknowledged, then there can still be undetected packet loss.
3600 * Do not restore congestion window in that case.
3601 */
3602 if (SEQ_LT(th_ack, tp->snd_recover)) {
3603 return FALSE;
3604 }
3605
3606 bad_rexmt = TRUE;
3607 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3608 if (!(rxseg->rx_flags & TCP_RXT_SPURIOUS)) {
3609 bad_rexmt = FALSE;
3610 break;
3611 }
3612 }
3613 return bad_rexmt;
3614 }
3615
3616 boolean_t
3617 tcp_rxtseg_dsack_for_tlp(struct tcpcb *tp)
3618 {
3619 boolean_t dsack_for_tlp = FALSE;
3620 struct tcp_rxt_seg *rxseg;
3621 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3622 return FALSE;
3623 }
3624
3625 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3626 if (rxseg->rx_count == 1 &&
3627 SLIST_NEXT(rxseg, rx_link) == NULL &&
3628 (rxseg->rx_flags & TCP_RXT_DSACK_FOR_TLP)) {
3629 dsack_for_tlp = TRUE;
3630 break;
3631 }
3632 }
3633 return dsack_for_tlp;
3634 }
3635
3636 u_int32_t
3637 tcp_rxtseg_total_size(struct tcpcb *tp)
3638 {
3639 struct tcp_rxt_seg *rxseg;
3640 u_int32_t total_size = 0;
3641
3642 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3643 total_size += (rxseg->rx_end - rxseg->rx_start) + 1;
3644 }
3645 return total_size;
3646 }
3647
3648 void
3649 tcp_get_connectivity_status(struct tcpcb *tp,
3650 struct tcp_conn_status *connstatus)
3651 {
3652 if (tp == NULL || connstatus == NULL) {
3653 return;
3654 }
3655 bzero(connstatus, sizeof(*connstatus));
3656 if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) {
3657 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3658 connstatus->write_probe_failed = 1;
3659 } else {
3660 connstatus->conn_probe_failed = 1;
3661 }
3662 }
3663 if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX) {
3664 connstatus->read_probe_failed = 1;
3665 }
3666 if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL &&
3667 (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY)) {
3668 connstatus->probe_activated = 1;
3669 }
3670 }
3671
3672 boolean_t
3673 tfo_enabled(const struct tcpcb *tp)
3674 {
3675 return (tp->t_flagsext & TF_FASTOPEN)? TRUE : FALSE;
3676 }
3677
3678 void
3679 tcp_disable_tfo(struct tcpcb *tp)
3680 {
3681 tp->t_flagsext &= ~TF_FASTOPEN;
3682 }
3683
3684 static struct mbuf *
3685 tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp,
3686 boolean_t is_probe)
3687 {
3688 struct inpcb *inp = tp->t_inpcb;
3689 struct tcphdr *th;
3690 u_int8_t *data;
3691 int win = 0;
3692 struct mbuf *m;
3693
3694 /*
3695 * The code assumes the IP + TCP headers fit in an mbuf packet header
3696 */
3697 _CASSERT(sizeof(struct ip) + sizeof(struct tcphdr) <= _MHLEN);
3698 _CASSERT(sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= _MHLEN);
3699
3700 MGETHDR(m, M_WAIT, MT_HEADER);
3701 if (m == NULL) {
3702 return NULL;
3703 }
3704 m->m_pkthdr.pkt_proto = IPPROTO_TCP;
3705
3706 data = mbuf_datastart(m);
3707
3708 if (inp->inp_vflag & INP_IPV4) {
3709 bzero(data, sizeof(struct ip) + sizeof(struct tcphdr));
3710 th = (struct tcphdr *)(void *) (data + sizeof(struct ip));
3711 m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
3712 m->m_pkthdr.len = m->m_len;
3713 } else {
3714 VERIFY(inp->inp_vflag & INP_IPV6);
3715
3716 bzero(data, sizeof(struct ip6_hdr)
3717 + sizeof(struct tcphdr));
3718 th = (struct tcphdr *)(void *)(data + sizeof(struct ip6_hdr));
3719 m->m_len = sizeof(struct ip6_hdr) +
3720 sizeof(struct tcphdr);
3721 m->m_pkthdr.len = m->m_len;
3722 }
3723
3724 tcp_fillheaders(tp, data, th);
3725
3726 if (inp->inp_vflag & INP_IPV4) {
3727 struct ip *ip;
3728
3729 ip = (__typeof__(ip))(void *)data;
3730
3731 ip->ip_id = rfc6864 ? 0 : ip_randomid();
3732 ip->ip_off = htons(IP_DF);
3733 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct tcphdr));
3734 ip->ip_ttl = inp->inp_ip_ttl;
3735 ip->ip_tos |= (inp->inp_ip_tos & ~IPTOS_ECN_MASK);
3736 ip->ip_sum = in_cksum_hdr(ip);
3737 } else {
3738 struct ip6_hdr *ip6;
3739
3740 ip6 = (__typeof__(ip6))(void *)data;
3741
3742 ip6->ip6_plen = htons(sizeof(struct tcphdr));
3743 ip6->ip6_hlim = in6_selecthlim(inp, ifp);
3744 ip6->ip6_flow = ip6->ip6_flow & ~IPV6_FLOW_ECN_MASK;
3745
3746 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
3747 ip6->ip6_src.s6_addr16[1] = 0;
3748 }
3749 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
3750 ip6->ip6_dst.s6_addr16[1] = 0;
3751 }
3752 }
3753 th->th_flags = TH_ACK;
3754
3755 win = tcp_sbspace(tp);
3756 if (win > ((int32_t)TCP_MAXWIN << tp->rcv_scale)) {
3757 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
3758 }
3759 th->th_win = htons((u_short) (win >> tp->rcv_scale));
3760
3761 if (is_probe) {
3762 th->th_seq = htonl(tp->snd_una - 1);
3763 } else {
3764 th->th_seq = htonl(tp->snd_una);
3765 }
3766 th->th_ack = htonl(tp->rcv_nxt);
3767
3768 /* Force recompute TCP checksum to be the final value */
3769 th->th_sum = 0;
3770 if (inp->inp_vflag & INP_IPV4) {
3771 th->th_sum = inet_cksum(m, IPPROTO_TCP,
3772 sizeof(struct ip), sizeof(struct tcphdr));
3773 } else {
3774 th->th_sum = inet6_cksum(m, IPPROTO_TCP,
3775 sizeof(struct ip6_hdr), sizeof(struct tcphdr));
3776 }
3777
3778 return m;
3779 }
3780
3781 void
3782 tcp_fill_keepalive_offload_frames(ifnet_t ifp,
3783 struct ifnet_keepalive_offload_frame *frames_array,
3784 u_int32_t frames_array_count, size_t frame_data_offset,
3785 u_int32_t *used_frames_count)
3786 {
3787 struct inpcb *inp;
3788 inp_gen_t gencnt;
3789 u_int32_t frame_index = *used_frames_count;
3790
3791 if (ifp == NULL || frames_array == NULL ||
3792 frames_array_count == 0 ||
3793 frame_index >= frames_array_count ||
3794 frame_data_offset >= IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
3795 return;
3796 }
3797
3798 /*
3799 * This function is called outside the regular TCP processing
3800 * so we need to update the TCP clock.
3801 */
3802 calculate_tcp_clock();
3803
3804 lck_rw_lock_shared(tcbinfo.ipi_lock);
3805 gencnt = tcbinfo.ipi_gencnt;
3806 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
3807 struct socket *so;
3808 struct ifnet_keepalive_offload_frame *frame;
3809 struct mbuf *m = NULL;
3810 struct tcpcb *tp = intotcpcb(inp);
3811
3812 if (frame_index >= frames_array_count) {
3813 break;
3814 }
3815
3816 if (inp->inp_gencnt > gencnt ||
3817 inp->inp_state == INPCB_STATE_DEAD) {
3818 continue;
3819 }
3820
3821 if ((so = inp->inp_socket) == NULL ||
3822 (so->so_state & SS_DEFUNCT)) {
3823 continue;
3824 }
3825 /*
3826 * check for keepalive offload flag without socket
3827 * lock to avoid a deadlock
3828 */
3829 if (!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD)) {
3830 continue;
3831 }
3832
3833 if (!(inp->inp_vflag & (INP_IPV4 | INP_IPV6))) {
3834 continue;
3835 }
3836 if (inp->inp_ppcb == NULL ||
3837 in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
3838 continue;
3839 }
3840 socket_lock(so, 1);
3841 /* Release the want count */
3842 if (inp->inp_ppcb == NULL ||
3843 (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) {
3844 socket_unlock(so, 1);
3845 continue;
3846 }
3847 if ((inp->inp_vflag & INP_IPV4) &&
3848 (inp->inp_laddr.s_addr == INADDR_ANY ||
3849 inp->inp_faddr.s_addr == INADDR_ANY)) {
3850 socket_unlock(so, 1);
3851 continue;
3852 }
3853 if ((inp->inp_vflag & INP_IPV6) &&
3854 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ||
3855 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))) {
3856 socket_unlock(so, 1);
3857 continue;
3858 }
3859 if (inp->inp_lport == 0 || inp->inp_fport == 0) {
3860 socket_unlock(so, 1);
3861 continue;
3862 }
3863 if (inp->inp_last_outifp == NULL ||
3864 inp->inp_last_outifp->if_index != ifp->if_index) {
3865 socket_unlock(so, 1);
3866 continue;
3867 }
3868 if ((inp->inp_vflag & INP_IPV4) && frame_data_offset +
3869 sizeof(struct ip) + sizeof(struct tcphdr) >
3870 IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
3871 socket_unlock(so, 1);
3872 continue;
3873 } else if (!(inp->inp_vflag & INP_IPV4) && frame_data_offset +
3874 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) >
3875 IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
3876 socket_unlock(so, 1);
3877 continue;
3878 }
3879 /*
3880 * There is no point in waking up the device for connections
3881 * that are not established. Long lived connection are meant
3882 * for processes that will sent and receive data
3883 */
3884 if (tp->t_state != TCPS_ESTABLISHED) {
3885 socket_unlock(so, 1);
3886 continue;
3887 }
3888 /*
3889 * This inp has all the information that is needed to
3890 * generate an offload frame.
3891 */
3892 frame = &frames_array[frame_index];
3893 frame->type = IFNET_KEEPALIVE_OFFLOAD_FRAME_TCP;
3894 frame->ether_type = (inp->inp_vflag & INP_IPV4) ?
3895 IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV4 :
3896 IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV6;
3897 frame->interval = (uint16_t)(tp->t_keepidle > 0 ? tp->t_keepidle :
3898 tcp_keepidle);
3899 frame->keep_cnt = (uint8_t)TCP_CONN_KEEPCNT(tp);
3900 frame->keep_retry = (uint16_t)TCP_CONN_KEEPINTVL(tp);
3901 if (so->so_options & SO_NOWAKEFROMSLEEP) {
3902 frame->flags |=
3903 IFNET_KEEPALIVE_OFFLOAD_FLAG_NOWAKEFROMSLEEP;
3904 }
3905 frame->local_port = ntohs(inp->inp_lport);
3906 frame->remote_port = ntohs(inp->inp_fport);
3907 frame->local_seq = tp->snd_nxt;
3908 frame->remote_seq = tp->rcv_nxt;
3909 if (inp->inp_vflag & INP_IPV4) {
3910 ASSERT(frame_data_offset + sizeof(struct ip) + sizeof(struct tcphdr) <= UINT8_MAX);
3911 frame->length = (uint8_t)(frame_data_offset +
3912 sizeof(struct ip) + sizeof(struct tcphdr));
3913 frame->reply_length = frame->length;
3914
3915 frame->addr_length = sizeof(struct in_addr);
3916 bcopy(&inp->inp_laddr, frame->local_addr,
3917 sizeof(struct in_addr));
3918 bcopy(&inp->inp_faddr, frame->remote_addr,
3919 sizeof(struct in_addr));
3920 } else {
3921 struct in6_addr *ip6;
3922
3923 ASSERT(frame_data_offset + sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= UINT8_MAX);
3924 frame->length = (uint8_t)(frame_data_offset +
3925 sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
3926 frame->reply_length = frame->length;
3927
3928 frame->addr_length = sizeof(struct in6_addr);
3929 ip6 = (struct in6_addr *)(void *)frame->local_addr;
3930 bcopy(&inp->in6p_laddr, ip6, sizeof(struct in6_addr));
3931 if (IN6_IS_SCOPE_EMBED(ip6)) {
3932 ip6->s6_addr16[1] = 0;
3933 }
3934
3935 ip6 = (struct in6_addr *)(void *)frame->remote_addr;
3936 bcopy(&inp->in6p_faddr, ip6, sizeof(struct in6_addr));
3937 if (IN6_IS_SCOPE_EMBED(ip6)) {
3938 ip6->s6_addr16[1] = 0;
3939 }
3940 }
3941
3942 /*
3943 * First the probe
3944 */
3945 m = tcp_make_keepalive_frame(tp, ifp, TRUE);
3946 if (m == NULL) {
3947 socket_unlock(so, 1);
3948 continue;
3949 }
3950 bcopy(m->m_data, frame->data + frame_data_offset,
3951 m->m_len);
3952 m_freem(m);
3953
3954 /*
3955 * Now the response packet to incoming probes
3956 */
3957 m = tcp_make_keepalive_frame(tp, ifp, FALSE);
3958 if (m == NULL) {
3959 socket_unlock(so, 1);
3960 continue;
3961 }
3962 bcopy(m->m_data, frame->reply_data + frame_data_offset,
3963 m->m_len);
3964 m_freem(m);
3965
3966 frame_index++;
3967 socket_unlock(so, 1);
3968 }
3969 lck_rw_done(tcbinfo.ipi_lock);
3970 *used_frames_count = frame_index;
3971 }
3972
3973 static bool
3974 inp_matches_kao_frame(ifnet_t ifp, struct ifnet_keepalive_offload_frame *frame,
3975 struct inpcb *inp)
3976 {
3977 if (inp->inp_ppcb == NULL) {
3978 return false;
3979 }
3980 /* Release the want count */
3981 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
3982 return false;
3983 }
3984 if (inp->inp_last_outifp == NULL ||
3985 inp->inp_last_outifp->if_index != ifp->if_index) {
3986 return false;
3987 }
3988 if (frame->local_port != ntohs(inp->inp_lport) ||
3989 frame->remote_port != ntohs(inp->inp_fport)) {
3990 return false;
3991 }
3992 if (inp->inp_vflag & INP_IPV4) {
3993 if (memcmp(&inp->inp_laddr, frame->local_addr,
3994 sizeof(struct in_addr)) != 0 ||
3995 memcmp(&inp->inp_faddr, frame->remote_addr,
3996 sizeof(struct in_addr)) != 0) {
3997 return false;
3998 }
3999 } else if (inp->inp_vflag & INP_IPV6) {
4000 if (memcmp(&inp->inp_laddr, frame->local_addr,
4001 sizeof(struct in6_addr)) != 0 ||
4002 memcmp(&inp->inp_faddr, frame->remote_addr,
4003 sizeof(struct in6_addr)) != 0) {
4004 return false;
4005 }
4006 } else {
4007 return false;
4008 }
4009 return true;
4010 }
4011
4012 int
4013 tcp_notify_kao_timeout(ifnet_t ifp,
4014 struct ifnet_keepalive_offload_frame *frame)
4015 {
4016 struct inpcb *inp = NULL;
4017 struct socket *so = NULL;
4018 bool found = false;
4019
4020 /*
4021 * Unlock the list before posting event on the matching socket
4022 */
4023 lck_rw_lock_shared(tcbinfo.ipi_lock);
4024
4025 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
4026 if ((so = inp->inp_socket) == NULL ||
4027 (so->so_state & SS_DEFUNCT)) {
4028 continue;
4029 }
4030 if (!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD)) {
4031 continue;
4032 }
4033 if (!(inp->inp_vflag & (INP_IPV4 | INP_IPV6))) {
4034 continue;
4035 }
4036 if (inp->inp_ppcb == NULL ||
4037 in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
4038 continue;
4039 }
4040 socket_lock(so, 1);
4041 if (inp_matches_kao_frame(ifp, frame, inp)) {
4042 /*
4043 * Keep the matching socket locked
4044 */
4045 found = true;
4046 break;
4047 }
4048 socket_unlock(so, 1);
4049 }
4050 lck_rw_done(tcbinfo.ipi_lock);
4051
4052 if (found) {
4053 ASSERT(inp != NULL);
4054 ASSERT(so != NULL);
4055 ASSERT(so == inp->inp_socket);
4056 /*
4057 * Drop the TCP connection like tcptimers() does
4058 */
4059 struct tcpcb *tp = inp->inp_ppcb;
4060
4061 tcpstat.tcps_keepdrops++;
4062 soevent(so,
4063 (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT));
4064 tp = tcp_drop(tp, ETIMEDOUT);
4065
4066 tcpstat.tcps_ka_offload_drops++;
4067 os_log_info(OS_LOG_DEFAULT, "%s: dropped lport %u fport %u\n",
4068 __func__, frame->local_port, frame->remote_port);
4069
4070 socket_unlock(so, 1);
4071 }
4072
4073 return 0;
4074 }
4075
4076 errno_t
4077 tcp_notify_ack_id_valid(struct tcpcb *tp, struct socket *so,
4078 u_int32_t notify_id)
4079 {
4080 struct tcp_notify_ack_marker *elm;
4081
4082 if (so->so_snd.sb_cc == 0) {
4083 return ENOBUFS;
4084 }
4085
4086 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4087 /* Duplicate id is not allowed */
4088 if (elm->notify_id == notify_id) {
4089 return EINVAL;
4090 }
4091 /* Duplicate position is not allowed */
4092 if (elm->notify_snd_una == tp->snd_una + so->so_snd.sb_cc) {
4093 return EINVAL;
4094 }
4095 }
4096 return 0;
4097 }
4098
4099 errno_t
4100 tcp_add_notify_ack_marker(struct tcpcb *tp, u_int32_t notify_id)
4101 {
4102 struct tcp_notify_ack_marker *nm, *elm = NULL;
4103 struct socket *so = tp->t_inpcb->inp_socket;
4104
4105 MALLOC(nm, struct tcp_notify_ack_marker *, sizeof(*nm),
4106 M_TEMP, M_WAIT | M_ZERO);
4107 if (nm == NULL) {
4108 return ENOMEM;
4109 }
4110 nm->notify_id = notify_id;
4111 nm->notify_snd_una = tp->snd_una + so->so_snd.sb_cc;
4112
4113 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4114 if (SEQ_GT(nm->notify_snd_una, elm->notify_snd_una)) {
4115 break;
4116 }
4117 }
4118
4119 if (elm == NULL) {
4120 VERIFY(SLIST_EMPTY(&tp->t_notify_ack));
4121 SLIST_INSERT_HEAD(&tp->t_notify_ack, nm, notify_next);
4122 } else {
4123 SLIST_INSERT_AFTER(elm, nm, notify_next);
4124 }
4125 tp->t_notify_ack_count++;
4126 return 0;
4127 }
4128
4129 void
4130 tcp_notify_ack_free(struct tcpcb *tp)
4131 {
4132 struct tcp_notify_ack_marker *elm, *next;
4133 if (SLIST_EMPTY(&tp->t_notify_ack)) {
4134 return;
4135 }
4136
4137 SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) {
4138 SLIST_REMOVE(&tp->t_notify_ack, elm, tcp_notify_ack_marker,
4139 notify_next);
4140 FREE(elm, M_TEMP);
4141 }
4142 SLIST_INIT(&tp->t_notify_ack);
4143 tp->t_notify_ack_count = 0;
4144 }
4145
4146 inline void
4147 tcp_notify_acknowledgement(struct tcpcb *tp, struct socket *so)
4148 {
4149 struct tcp_notify_ack_marker *elm;
4150
4151 elm = SLIST_FIRST(&tp->t_notify_ack);
4152 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4153 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_NOTIFY_ACK);
4154 }
4155 }
4156
4157 void
4158 tcp_get_notify_ack_count(struct tcpcb *tp,
4159 struct tcp_notify_ack_complete *retid)
4160 {
4161 struct tcp_notify_ack_marker *elm;
4162 uint32_t complete = 0;
4163
4164 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4165 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4166 ASSERT(complete < UINT32_MAX);
4167 complete++;
4168 } else {
4169 break;
4170 }
4171 }
4172 retid->notify_pending = tp->t_notify_ack_count - complete;
4173 retid->notify_complete_count = min(TCP_MAX_NOTIFY_ACK, complete);
4174 }
4175
4176 void
4177 tcp_get_notify_ack_ids(struct tcpcb *tp,
4178 struct tcp_notify_ack_complete *retid)
4179 {
4180 size_t i = 0;
4181 struct tcp_notify_ack_marker *elm, *next;
4182
4183 SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) {
4184 if (i >= retid->notify_complete_count) {
4185 break;
4186 }
4187 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4188 retid->notify_complete_id[i++] = elm->notify_id;
4189 SLIST_REMOVE(&tp->t_notify_ack, elm,
4190 tcp_notify_ack_marker, notify_next);
4191 FREE(elm, M_TEMP);
4192 tp->t_notify_ack_count--;
4193 } else {
4194 break;
4195 }
4196 }
4197 }
4198
4199 bool
4200 tcp_notify_ack_active(struct socket *so)
4201 {
4202 if ((SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) &&
4203 SOCK_TYPE(so) == SOCK_STREAM) {
4204 struct tcpcb *tp = intotcpcb(sotoinpcb(so));
4205
4206 if (!SLIST_EMPTY(&tp->t_notify_ack)) {
4207 struct tcp_notify_ack_marker *elm;
4208 elm = SLIST_FIRST(&tp->t_notify_ack);
4209 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4210 return true;
4211 }
4212 }
4213 }
4214 return false;
4215 }
4216
4217 inline int32_t
4218 inp_get_sndbytes_allunsent(struct socket *so, u_int32_t th_ack)
4219 {
4220 struct inpcb *inp = sotoinpcb(so);
4221 struct tcpcb *tp = intotcpcb(inp);
4222
4223 if ((so->so_snd.sb_flags & SB_SNDBYTE_CNT) &&
4224 so->so_snd.sb_cc > 0) {
4225 int32_t unsent, sent;
4226 sent = tp->snd_max - th_ack;
4227 if (tp->t_flags & TF_SENTFIN) {
4228 sent--;
4229 }
4230 unsent = so->so_snd.sb_cc - sent;
4231 return unsent;
4232 }
4233 return 0;
4234 }
4235
4236 #define IFP_PER_FLOW_STAT(_ipv4_, _stat_) { \
4237 if (_ipv4_) { \
4238 ifp->if_ipv4_stat->_stat_++; \
4239 } else { \
4240 ifp->if_ipv6_stat->_stat_++; \
4241 } \
4242 }
4243
4244 #define FLOW_ECN_ENABLED(_flags_) \
4245 ((_flags_ & (TE_ECN_ON)) == (TE_ECN_ON))
4246
4247 void
4248 tcp_update_stats_per_flow(struct ifnet_stats_per_flow *ifs,
4249 struct ifnet *ifp)
4250 {
4251 if (ifp == NULL || !IF_FULLY_ATTACHED(ifp)) {
4252 return;
4253 }
4254
4255 ifnet_lock_shared(ifp);
4256 if (ifs->ecn_flags & TE_SETUPSENT) {
4257 if (ifs->ecn_flags & TE_CLIENT_SETUP) {
4258 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_client_setup);
4259 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4260 IFP_PER_FLOW_STAT(ifs->ipv4,
4261 ecn_client_success);
4262 } else if (ifs->ecn_flags & TE_LOST_SYN) {
4263 IFP_PER_FLOW_STAT(ifs->ipv4,
4264 ecn_syn_lost);
4265 } else {
4266 IFP_PER_FLOW_STAT(ifs->ipv4,
4267 ecn_peer_nosupport);
4268 }
4269 } else {
4270 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_server_setup);
4271 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4272 IFP_PER_FLOW_STAT(ifs->ipv4,
4273 ecn_server_success);
4274 } else if (ifs->ecn_flags & TE_LOST_SYN) {
4275 IFP_PER_FLOW_STAT(ifs->ipv4,
4276 ecn_synack_lost);
4277 } else {
4278 IFP_PER_FLOW_STAT(ifs->ipv4,
4279 ecn_peer_nosupport);
4280 }
4281 }
4282 } else {
4283 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off_conn);
4284 }
4285 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4286 if (ifs->ecn_flags & TE_RECV_ECN_CE) {
4287 tcpstat.tcps_ecn_conn_recv_ce++;
4288 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ce);
4289 }
4290 if (ifs->ecn_flags & TE_RECV_ECN_ECE) {
4291 tcpstat.tcps_ecn_conn_recv_ece++;
4292 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ece);
4293 }
4294 if (ifs->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) {
4295 if (ifs->txretransmitbytes > 0 ||
4296 ifs->rxoutoforderbytes > 0) {
4297 tcpstat.tcps_ecn_conn_pl_ce++;
4298 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plce);
4299 } else {
4300 tcpstat.tcps_ecn_conn_nopl_ce++;
4301 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_noplce);
4302 }
4303 } else {
4304 if (ifs->txretransmitbytes > 0 ||
4305 ifs->rxoutoforderbytes > 0) {
4306 tcpstat.tcps_ecn_conn_plnoce++;
4307 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plnoce);
4308 }
4309 }
4310 }
4311
4312 /* Other stats are interesting for non-local connections only */
4313 if (ifs->local) {
4314 ifnet_lock_done(ifp);
4315 return;
4316 }
4317
4318 if (ifs->ipv4) {
4319 ifp->if_ipv4_stat->timestamp = net_uptime();
4320 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4321 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_on);
4322 } else {
4323 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_off);
4324 }
4325 } else {
4326 ifp->if_ipv6_stat->timestamp = net_uptime();
4327 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4328 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_on);
4329 } else {
4330 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_off);
4331 }
4332 }
4333
4334 if (ifs->rxmit_drop) {
4335 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4336 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_on.rxmit_drop);
4337 } else {
4338 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off.rxmit_drop);
4339 }
4340 }
4341 if (ifs->ecn_fallback_synloss) {
4342 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_synloss);
4343 }
4344 if (ifs->ecn_fallback_droprst) {
4345 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprst);
4346 }
4347 if (ifs->ecn_fallback_droprxmt) {
4348 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprxmt);
4349 }
4350 if (ifs->ecn_fallback_ce) {
4351 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_ce);
4352 }
4353 if (ifs->ecn_fallback_reorder) {
4354 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_reorder);
4355 }
4356 if (ifs->ecn_recv_ce > 0) {
4357 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ce);
4358 }
4359 if (ifs->ecn_recv_ece > 0) {
4360 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ece);
4361 }
4362
4363 tcp_flow_lim_stats(ifs, &ifp->if_lim_stat);
4364 ifnet_lock_done(ifp);
4365 }
4366
mirror of XNU