1 /*
2 * Copyright (c) 2000-2022 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 "tcp_includes.h"
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/domain.h>
78 #include <sys/proc.h>
79 #include <sys/kauth.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/protosw.h>
83 #include <sys/random.h>
84 #include <sys/syslog.h>
85 #include <sys/mcache.h>
86 #include <kern/locks.h>
87 #include <kern/zalloc.h>
88
89 #include <dev/random/randomdev.h>
90
91 #include <net/route.h>
92 #include <net/if.h>
93 #include <net/content_filter.h>
94 #include <net/ntstat.h>
95 #include <net/multi_layer_pkt_log.h>
96
97 #define tcp_minmssoverload fring
98 #define _IP_VHL
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_icmp.h>
103 #include <netinet/ip6.h>
104 #include <netinet/icmp6.h>
105 #include <netinet/in_pcb.h>
106 #include <netinet6/in6_pcb.h>
107 #include <netinet/in_var.h>
108 #include <netinet/ip_var.h>
109 #include <netinet/icmp_var.h>
110 #include <netinet6/ip6_var.h>
111 #include <netinet/mptcp_var.h>
112 #include <netinet/tcp.h>
113 #include <netinet/tcp_fsm.h>
114 #include <netinet/tcp_seq.h>
115 #include <netinet/tcp_timer.h>
116 #include <netinet/tcp_var.h>
117 #include <netinet/tcp_cc.h>
118 #include <netinet/tcp_cache.h>
119 #include <kern/thread_call.h>
120
121 #include <netinet6/tcp6_var.h>
122 #include <netinet/tcpip.h>
123 #if TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126 #include <netinet/tcp_log.h>
127
128 #include <netinet6/ip6protosw.h>
129
130 #if IPSEC
131 #include <netinet6/ipsec.h>
132 #include <netinet6/ipsec6.h>
133 #endif /* IPSEC */
134
135 #if NECP
136 #include <net/necp.h>
137 #endif /* NECP */
138
139 #undef tcp_minmssoverload
140
141 #include <corecrypto/ccaes.h>
142 #include <libkern/crypto/aes.h>
143 #include <libkern/crypto/md5.h>
144 #include <sys/kdebug.h>
145 #include <mach/sdt.h>
146 #include <atm/atm_internal.h>
147 #include <pexpert/pexpert.h>
148
149 #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2))
150
151 static tcp_cc tcp_ccgen;
152
153 extern struct tcptimerlist tcp_timer_list;
154 extern struct tcptailq tcp_tw_tailq;
155
156 SYSCTL_SKMEM_TCP_INT(TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED,
157 int, tcp_mssdflt, TCP_MSS, "Default TCP Maximum Segment Size");
158
159 SYSCTL_SKMEM_TCP_INT(TCPCTL_V6MSSDFLT, v6mssdflt,
160 CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_v6mssdflt, TCP6_MSS,
161 "Default TCP Maximum Segment Size for IPv6");
162
163 int tcp_sysctl_fastopenkey(struct sysctl_oid *, void *, int,
164 struct sysctl_req *);
165 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, fastopen_key, CTLTYPE_STRING | CTLFLAG_WR,
166 0, 0, tcp_sysctl_fastopenkey, "S", "TCP Fastopen key");
167
168 /* Current count of half-open TFO connections */
169 int tcp_tfo_halfcnt = 0;
170
171 /* Maximum of half-open TFO connection backlog */
172 SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen_backlog,
173 CTLFLAG_RW | CTLFLAG_LOCKED, int, tcp_tfo_backlog, 10,
174 "Backlog queue for half-open TFO connections");
175
176 SYSCTL_SKMEM_TCP_INT(OID_AUTO, fastopen, CTLFLAG_RW | CTLFLAG_LOCKED,
177 int, tcp_fastopen, TCP_FASTOPEN_CLIENT | TCP_FASTOPEN_SERVER,
178 "Enable TCP Fastopen (RFC 7413)");
179
180 SYSCTL_SKMEM_TCP_INT(OID_AUTO, now_init, CTLFLAG_RD | CTLFLAG_LOCKED,
181 uint32_t, tcp_now_init, 0, "Initial tcp now value");
182
183 SYSCTL_SKMEM_TCP_INT(OID_AUTO, microuptime_init, CTLFLAG_RD | CTLFLAG_LOCKED,
184 uint32_t, tcp_microuptime_init, 0, "Initial tcp uptime value in micro seconds");
185
186 /*
187 * Minimum MSS we accept and use. This prevents DoS attacks where
188 * we are forced to a ridiculous low MSS like 20 and send hundreds
189 * of packets instead of one. The effect scales with the available
190 * bandwidth and quickly saturates the CPU and network interface
191 * with packet generation and sending. Set to zero to disable MINMSS
192 * checking. This setting prevents us from sending too small packets.
193 */
194 SYSCTL_SKMEM_TCP_INT(OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED,
195 int, tcp_minmss, TCP_MINMSS, "Minmum TCP Maximum Segment Size");
196
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
198 &tcbinfo.ipi_count, 0, "Number of active PCBs");
199
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tw_pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
201 &tcbinfo.ipi_twcount, 0, "Number of pcbs in time-wait state");
202
203 SYSCTL_SKMEM_TCP_INT(OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED,
204 static int, icmp_may_rst, 1,
205 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
206
207 static int tcp_strict_rfc1948 = 0;
208 static int tcp_isn_reseed_interval = 0;
209 #if (DEVELOPMENT || DEBUG)
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW | CTLFLAG_LOCKED,
211 &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly");
212
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval,
214 CTLFLAG_RW | CTLFLAG_LOCKED,
215 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
216 #endif /* (DEVELOPMENT || DEBUG) */
217
218 SYSCTL_SKMEM_TCP_INT(OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED,
219 int, tcp_TCPTV_MIN, 100, "min rtt value allowed");
220
221 SYSCTL_SKMEM_TCP_INT(OID_AUTO, rexmt_slop, CTLFLAG_RW,
222 int, tcp_rexmt_slop, TCPTV_REXMTSLOP, "Slop added to retransmit timeout");
223
224 SYSCTL_SKMEM_TCP_INT(OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED,
225 __private_extern__ int, tcp_use_randomport, 0,
226 "Randomize TCP port numbers");
227
228 SYSCTL_SKMEM_TCP_INT(OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED,
229 __private_extern__ int, tcp_win_scale, 3, "Window scaling factor");
230
231 #if (DEVELOPMENT || DEBUG)
232 SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache,
233 CTLFLAG_RW | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1,
234 "Initalize RTT from route cache");
235 #else
236 SYSCTL_SKMEM_TCP_INT(OID_AUTO, init_rtt_from_cache,
237 CTLFLAG_RD | CTLFLAG_LOCKED, static int, tcp_init_rtt_from_cache, 1,
238 "Initalize RTT from route cache");
239 #endif /* (DEVELOPMENT || DEBUG) */
240
241 static int tso_debug = 0;
242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso_debug, CTLFLAG_RW | CTLFLAG_LOCKED,
243 &tso_debug, 0, "TSO verbosity");
244
245 static int tcp_rxt_seg_max = 1024;
246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rxt_seg_max, CTLFLAG_RW | CTLFLAG_LOCKED,
247 &tcp_rxt_seg_max, 0, "");
248
249 static unsigned long tcp_rxt_seg_drop = 0;
250 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, rxt_seg_drop, CTLFLAG_RD | CTLFLAG_LOCKED,
251 &tcp_rxt_seg_drop, "");
252
253 static void tcp_notify(struct inpcb *, int);
254
255 static KALLOC_TYPE_DEFINE(tcp_bwmeas_zone, struct bwmeas, NET_KT_DEFAULT);
256 KALLOC_TYPE_DEFINE(tcp_reass_zone, struct tseg_qent, NET_KT_DEFAULT);
257 KALLOC_TYPE_DEFINE(tcp_rxt_seg_zone, struct tcp_rxt_seg, NET_KT_DEFAULT);
258
259 extern int slowlink_wsize; /* window correction for slow links */
260 extern int path_mtu_discovery;
261
262 uint32_t tcp_now_remainder_us = 0; /* remaining micro seconds for tcp_now */
263
264 static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb);
265
266 #define TCP_BWMEAS_BURST_MINSIZE 6
267 #define TCP_BWMEAS_BURST_MAXSIZE 25
268
269 /*
270 * Target size of TCP PCB hash tables. Must be a power of two.
271 *
272 * Note that this can be overridden by the kernel environment
273 * variable net.inet.tcp.tcbhashsize
274 */
275 #ifndef TCBHASHSIZE
276 #define TCBHASHSIZE CONFIG_TCBHASHSIZE
277 #endif
278
279 __private_extern__ int tcp_tcbhashsize = TCBHASHSIZE;
280 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED,
281 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
282
283 /*
284 * This is the actual shape of what we allocate using the zone
285 * allocator. Doing it this way allows us to protect both structures
286 * using the same generation count, and also eliminates the overhead
287 * of allocating tcpcbs separately. By hiding the structure here,
288 * we avoid changing most of the rest of the code (although it needs
289 * to be changed, eventually, for greater efficiency).
290 */
291 #define ALIGNMENT 32
292 struct inp_tp {
293 struct inpcb inp;
294 struct tcpcb tcb __attribute__((aligned(ALIGNMENT)));
295 };
296 #undef ALIGNMENT
297
298 static KALLOC_TYPE_DEFINE(tcpcbzone, struct inp_tp, NET_KT_DEFAULT);
299
300 int get_inpcb_str_size(void);
301 int get_tcp_str_size(void);
302
303 os_log_t tcp_mpkl_log_object = NULL;
304
305 static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *);
306
307 int tcp_notsent_lowat_check(struct socket *so);
308 static void tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs,
309 struct if_lim_perf_stat *stat);
310 static void tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs,
311 struct if_tcp_ecn_perf_stat *stat);
312
313 static aes_encrypt_ctx tfo_ctx; /* Crypto-context for TFO */
314
315 void
tcp_tfo_gen_cookie(struct inpcb * inp,u_char * out,size_t blk_size)316 tcp_tfo_gen_cookie(struct inpcb *inp, u_char *out, size_t blk_size)
317 {
318 u_char in[CCAES_BLOCK_SIZE];
319 int isipv6 = inp->inp_vflag & INP_IPV6;
320
321 VERIFY(blk_size == CCAES_BLOCK_SIZE);
322
323 bzero(&in[0], CCAES_BLOCK_SIZE);
324 bzero(&out[0], CCAES_BLOCK_SIZE);
325
326 if (isipv6) {
327 memcpy(in, &inp->in6p_faddr, sizeof(struct in6_addr));
328 } else {
329 memcpy(in, &inp->inp_faddr, sizeof(struct in_addr));
330 }
331
332 aes_encrypt_cbc(in, NULL, 1, out, &tfo_ctx);
333 }
334
335 __private_extern__ int
tcp_sysctl_fastopenkey(__unused struct sysctl_oid * oidp,__unused void * arg1,__unused int arg2,struct sysctl_req * req)336 tcp_sysctl_fastopenkey(__unused struct sysctl_oid *oidp, __unused void *arg1,
337 __unused int arg2, struct sysctl_req *req)
338 {
339 int error = 0;
340 /*
341 * TFO-key is expressed as a string in hex format
342 * +1 to account for the \0 char
343 * +1 because sysctl_io_string() expects a string length but the sysctl command
344 * now includes the terminating \0 in newlen -- see rdar://77205344
345 */
346 char keystring[TCP_FASTOPEN_KEYLEN * 2 + 2];
347 u_int32_t key[TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)];
348 int i;
349
350 /*
351 * sysctl_io_string copies keystring into the oldptr of the sysctl_req.
352 * Make sure everything is zero, to avoid putting garbage in there or
353 * leaking the stack.
354 */
355 bzero(keystring, sizeof(keystring));
356
357 error = sysctl_io_string(req, keystring, sizeof(keystring), 0, NULL);
358 if (error) {
359 os_log(OS_LOG_DEFAULT,
360 "%s: sysctl_io_string() error %d, req->newlen %lu, sizeof(keystring) %lu",
361 __func__, error, req->newlen, sizeof(keystring));
362 goto exit;
363 }
364 if (req->newptr == USER_ADDR_NULL) {
365 goto exit;
366 }
367
368 if (strlen(keystring) != TCP_FASTOPEN_KEYLEN * 2) {
369 os_log(OS_LOG_DEFAULT,
370 "%s: strlen(keystring) %lu != TCP_FASTOPEN_KEYLEN * 2 %u, newlen %lu",
371 __func__, strlen(keystring), TCP_FASTOPEN_KEYLEN * 2, req->newlen);
372 error = EINVAL;
373 goto exit;
374 }
375
376 for (i = 0; i < (TCP_FASTOPEN_KEYLEN / sizeof(u_int32_t)); i++) {
377 /*
378 * We jump over the keystring in 8-character (4 byte in hex)
379 * steps
380 */
381 if (sscanf(&keystring[i * 8], "%8x", &key[i]) != 1) {
382 error = EINVAL;
383 os_log(OS_LOG_DEFAULT,
384 "%s: sscanf() != 1, error EINVAL", __func__);
385 goto exit;
386 }
387 }
388
389 aes_encrypt_key128((u_char *)key, &tfo_ctx);
390
391 exit:
392 return error;
393 }
394
395 int
get_inpcb_str_size(void)396 get_inpcb_str_size(void)
397 {
398 return sizeof(struct inpcb);
399 }
400
401 int
get_tcp_str_size(void)402 get_tcp_str_size(void)
403 {
404 return sizeof(struct tcpcb);
405 }
406
407 static int scale_to_powerof2(int size);
408
409 /*
410 * This helper routine returns one of the following scaled value of size:
411 * 1. Rounded down power of two value of size if the size value passed as
412 * argument is not a power of two and the rounded up value overflows.
413 * OR
414 * 2. Rounded up power of two value of size if the size value passed as
415 * argument is not a power of two and the rounded up value does not overflow
416 * OR
417 * 3. Same value as argument size if it is already a power of two.
418 */
419 static int
scale_to_powerof2(int size)420 scale_to_powerof2(int size)
421 {
422 /* Handle special case of size = 0 */
423 int ret = size ? size : 1;
424
425 if (!powerof2(ret)) {
426 while (!powerof2(size)) {
427 /*
428 * Clear out least significant
429 * set bit till size is left with
430 * its highest set bit at which point
431 * it is rounded down power of two.
432 */
433 size = size & (size - 1);
434 }
435
436 /* Check for overflow when rounding up */
437 if (0 == (size << 1)) {
438 ret = size;
439 } else {
440 ret = size << 1;
441 }
442 }
443
444 return ret;
445 }
446
447 /*
448 * Round the floating point to the next integer
449 * Eg. 1.3 will round up to 2.
450 */
451 uint32_t
tcp_ceil(double a)452 tcp_ceil(double a)
453 {
454 double res = (uint32_t) a;
455 return (uint32_t)(res + (res < a));
456 }
457
458 uint32_t
tcp_round_to(uint32_t val,uint32_t round)459 tcp_round_to(uint32_t val, uint32_t round)
460 {
461 /*
462 * Round up or down based on the middle. Meaning, if we round upon a
463 * multiple of 10, 16 will round to 20 and 14 will round to 10.
464 */
465 return ((val + (round / 2)) / round) * round;
466 }
467
468 /*
469 * Round up to the next multiple of base.
470 * Eg. for a base of 64, 65 will become 128,
471 * 2896 will become 2944.
472 */
473 uint32_t
tcp_round_up(uint32_t val,uint32_t base)474 tcp_round_up(uint32_t val, uint32_t base)
475 {
476 if (base == 1 || val % base == 0) {
477 return val;
478 }
479
480 return ((val + base) / base) * base;
481 }
482
483 static void
tcp_tfo_init(void)484 tcp_tfo_init(void)
485 {
486 u_char key[TCP_FASTOPEN_KEYLEN];
487
488 read_frandom(key, sizeof(key));
489 aes_encrypt_key128(key, &tfo_ctx);
490 }
491
492 /*
493 * Tcp initialization
494 */
495 void
tcp_init(struct protosw * pp,struct domain * dp)496 tcp_init(struct protosw *pp, struct domain *dp)
497 {
498 #pragma unused(dp)
499 static int tcp_initialized = 0;
500 struct inpcbinfo *pcbinfo;
501 uint32_t logging_config;
502
503 VERIFY((pp->pr_flags & (PR_INITIALIZED | PR_ATTACHED)) == PR_ATTACHED);
504
505 if (tcp_initialized) {
506 return;
507 }
508 tcp_initialized = 1;
509
510 #if DEBUG || DEVELOPMENT
511 (void) PE_parse_boot_argn("tcp_rxt_seg_max", &tcp_rxt_seg_max,
512 sizeof(tcp_rxt_seg_max));
513 #endif /* DEBUG || DEVELOPMENT */
514
515 tcp_ccgen = 1;
516 tcp_keepinit = TCPTV_KEEP_INIT;
517 tcp_keepidle = TCPTV_KEEP_IDLE;
518 tcp_keepintvl = TCPTV_KEEPINTVL;
519 tcp_keepcnt = TCPTV_KEEPCNT;
520 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
521 tcp_msl = TCPTV_MSL;
522
523 microuptime(&tcp_uptime);
524 read_frandom(&tcp_now, sizeof(tcp_now));
525
526 /* Starts tcp internal clock at a random value */
527 tcp_now = tcp_now & 0x3fffffff;
528
529 /* expose initial uptime/now via systcl for utcp to keep time sync */
530 tcp_now_init = tcp_now;
531 tcp_microuptime_init =
532 (uint32_t)(tcp_uptime.tv_usec + (tcp_uptime.tv_sec * USEC_PER_SEC));
533 SYSCTL_SKMEM_UPDATE_FIELD(tcp.microuptime_init, tcp_microuptime_init);
534 SYSCTL_SKMEM_UPDATE_FIELD(tcp.now_init, tcp_now_init);
535
536 tcp_tfo_init();
537
538 LIST_INIT(&tcb);
539 tcbinfo.ipi_listhead = &tcb;
540
541 pcbinfo = &tcbinfo;
542
543 /*
544 * allocate group, lock attributes and lock for tcp pcb mutexes
545 */
546 pcbinfo->ipi_lock_grp = lck_grp_alloc_init("tcppcb",
547 LCK_GRP_ATTR_NULL);
548 lck_attr_setdefault(&pcbinfo->ipi_lock_attr);
549 lck_rw_init(&pcbinfo->ipi_lock, pcbinfo->ipi_lock_grp,
550 &pcbinfo->ipi_lock_attr);
551
552 if (tcp_tcbhashsize == 0) {
553 /* Set to default */
554 tcp_tcbhashsize = 512;
555 }
556
557 if (!powerof2(tcp_tcbhashsize)) {
558 int old_hash_size = tcp_tcbhashsize;
559 tcp_tcbhashsize = scale_to_powerof2(tcp_tcbhashsize);
560 /* Lower limit of 16 */
561 if (tcp_tcbhashsize < 16) {
562 tcp_tcbhashsize = 16;
563 }
564 printf("WARNING: TCB hash size not a power of 2, "
565 "scaled from %d to %d.\n",
566 old_hash_size,
567 tcp_tcbhashsize);
568 }
569
570 tcbinfo.ipi_hashbase = hashinit(tcp_tcbhashsize, M_PCB,
571 &tcbinfo.ipi_hashmask);
572 tcbinfo.ipi_porthashbase = hashinit(tcp_tcbhashsize, M_PCB,
573 &tcbinfo.ipi_porthashmask);
574 tcbinfo.ipi_zone = tcpcbzone;
575
576 tcbinfo.ipi_gc = tcp_gc;
577 tcbinfo.ipi_timer = tcp_itimer;
578 in_pcbinfo_attach(&tcbinfo);
579
580 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
581 if (max_protohdr < TCP_MINPROTOHDR) {
582 max_protohdr = (int)P2ROUNDUP(TCP_MINPROTOHDR, sizeof(uint32_t));
583 }
584 if (max_linkhdr + max_protohdr > MCLBYTES) {
585 panic("tcp_init");
586 }
587 #undef TCP_MINPROTOHDR
588
589 /* Initialize time wait and timer lists */
590 TAILQ_INIT(&tcp_tw_tailq);
591
592 bzero(&tcp_timer_list, sizeof(tcp_timer_list));
593 LIST_INIT(&tcp_timer_list.lhead);
594 /*
595 * allocate group and attribute for the tcp timer list
596 */
597 tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist",
598 LCK_GRP_ATTR_NULL);
599 lck_mtx_init(&tcp_timer_list.mtx, tcp_timer_list.mtx_grp,
600 LCK_ATTR_NULL);
601
602 tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL);
603 if (tcp_timer_list.call == NULL) {
604 panic("failed to allocate call entry 1 in tcp_init");
605 }
606
607 /* Initialize TCP Cache */
608 tcp_cache_init();
609
610 tcp_mpkl_log_object = MPKL_CREATE_LOGOBJECT("com.apple.xnu.tcp");
611 if (tcp_mpkl_log_object == NULL) {
612 panic("MPKL_CREATE_LOGOBJECT failed");
613 }
614
615 logging_config = atm_get_diagnostic_config();
616 if (logging_config & 0x80000000) {
617 tcp_log_privacy = 1;
618 }
619
620 if (PE_parse_boot_argn("tcp_log", &tcp_log_enable_flags, sizeof(tcp_log_enable_flags))) {
621 os_log(OS_LOG_DEFAULT, "tcp_init: set tcp_log_enable_flags to 0x%x", tcp_log_enable_flags);
622 }
623
624 /*
625 * If more than 4GB of actual memory is available, increase the
626 * maximum allowed receive and send socket buffer size.
627 */
628 if (mem_actual >= (1ULL << (GBSHIFT + 2))) {
629 tcp_autorcvbuf_max = 4 * 1024 * 1024;
630 tcp_autosndbuf_max = 4 * 1024 * 1024;
631
632 SYSCTL_SKMEM_UPDATE_FIELD(tcp.autorcvbufmax, tcp_autorcvbuf_max);
633 SYSCTL_SKMEM_UPDATE_FIELD(tcp.autosndbufmax, tcp_autosndbuf_max);
634 }
635 }
636
637 /*
638 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
639 * tcp_template used to store this data in mbufs, but we now recopy it out
640 * of the tcpcb each time to conserve mbufs.
641 */
642 void
tcp_fillheaders(struct mbuf * m,struct tcpcb * tp,void * ip_ptr,void * tcp_ptr)643 tcp_fillheaders(struct mbuf *m, struct tcpcb *tp, void *ip_ptr, void *tcp_ptr)
644 {
645 struct inpcb *inp = tp->t_inpcb;
646 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
647
648 if ((inp->inp_vflag & INP_IPV6) != 0) {
649 struct ip6_hdr *ip6;
650
651 ip6 = (struct ip6_hdr *)ip_ptr;
652 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
653 (inp->inp_flow & IPV6_FLOWINFO_MASK);
654 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
655 (IPV6_VERSION & IPV6_VERSION_MASK);
656 ip6->ip6_plen = htons(sizeof(struct tcphdr));
657 ip6->ip6_nxt = IPPROTO_TCP;
658 ip6->ip6_hlim = 0;
659 ip6->ip6_src = inp->in6p_laddr;
660 ip6->ip6_dst = inp->in6p_faddr;
661 if (m->m_flags & M_PKTHDR) {
662 uint32_t lifscope = inp->inp_lifscope != 0 ? inp->inp_lifscope : inp->inp_fifscope;
663 uint32_t fifscope = inp->inp_fifscope != 0 ? inp->inp_fifscope : inp->inp_lifscope;
664 ip6_output_setsrcifscope(m, lifscope, NULL);
665 ip6_output_setdstifscope(m, fifscope, NULL);
666 }
667 tcp_hdr->th_sum = in6_pseudo(&inp->in6p_laddr, &inp->in6p_faddr,
668 htonl(sizeof(struct tcphdr) + IPPROTO_TCP));
669 } else {
670 struct ip *ip = (struct ip *) ip_ptr;
671
672 ip->ip_vhl = IP_VHL_BORING;
673 ip->ip_tos = 0;
674 ip->ip_len = 0;
675 ip->ip_id = 0;
676 ip->ip_off = 0;
677 ip->ip_ttl = 0;
678 ip->ip_sum = 0;
679 ip->ip_p = IPPROTO_TCP;
680 ip->ip_src = inp->inp_laddr;
681 ip->ip_dst = inp->inp_faddr;
682 tcp_hdr->th_sum =
683 in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
684 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
685 }
686
687 tcp_hdr->th_sport = inp->inp_lport;
688 tcp_hdr->th_dport = inp->inp_fport;
689 tcp_hdr->th_seq = 0;
690 tcp_hdr->th_ack = 0;
691 tcp_hdr->th_x2 = 0;
692 tcp_hdr->th_off = 5;
693 tcp_hdr->th_flags = 0;
694 tcp_hdr->th_win = 0;
695 tcp_hdr->th_urp = 0;
696 }
697
698 /*
699 * Create template to be used to send tcp packets on a connection.
700 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
701 * use for this function is in keepalives, which use tcp_respond.
702 */
703 struct tcptemp *
tcp_maketemplate(struct tcpcb * tp)704 tcp_maketemplate(struct tcpcb *tp)
705 {
706 struct mbuf *m;
707 struct tcptemp *n;
708
709 m = m_get(M_DONTWAIT, MT_HEADER);
710 if (m == NULL) {
711 return NULL;
712 }
713 m->m_len = sizeof(struct tcptemp);
714 n = mtod(m, struct tcptemp *);
715
716 tcp_fillheaders(m, tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
717 return n;
718 }
719
720 /*
721 * Send a single message to the TCP at address specified by
722 * the given TCP/IP header. If m == 0, then we make a copy
723 * of the tcpiphdr at ti and send directly to the addressed host.
724 * This is used to force keep alive messages out using the TCP
725 * template for a connection. If flags are given then we send
726 * a message back to the TCP which originated the * segment ti,
727 * and discard the mbuf containing it and any other attached mbufs.
728 *
729 * In any case the ack and sequence number of the transmitted
730 * segment are as specified by the parameters.
731 *
732 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
733 */
734 void
tcp_respond(struct tcpcb * tp,void * ipgen,struct tcphdr * th,struct mbuf * m,tcp_seq ack,tcp_seq seq,uint8_t flags,struct tcp_respond_args * tra)735 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
736 tcp_seq ack, tcp_seq seq, uint8_t flags, struct tcp_respond_args *tra)
737 {
738 uint16_t tlen;
739 int win = 0;
740 struct route *ro = 0;
741 struct route sro;
742 struct ip *ip;
743 struct tcphdr *nth;
744 struct route_in6 *ro6 = 0;
745 struct route_in6 sro6;
746 struct ip6_hdr *ip6;
747 int isipv6;
748 struct ifnet *outif;
749 int sotc = SO_TC_UNSPEC;
750 bool check_qos_marking_again = FALSE;
751 uint32_t sifscope = IFSCOPE_NONE, fifscope = IFSCOPE_NONE;
752
753 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
754 ip6 = ipgen;
755 ip = ipgen;
756
757 if (tp) {
758 check_qos_marking_again = tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE ? FALSE : TRUE;
759 sifscope = tp->t_inpcb->inp_lifscope;
760 fifscope = tp->t_inpcb->inp_fifscope;
761 if (!(flags & TH_RST)) {
762 win = tcp_sbspace(tp);
763 if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale) {
764 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
765 }
766 }
767 if (isipv6) {
768 ro6 = &tp->t_inpcb->in6p_route;
769 } else {
770 ro = &tp->t_inpcb->inp_route;
771 }
772 } else {
773 if (isipv6) {
774 ro6 = &sro6;
775 bzero(ro6, sizeof(*ro6));
776 } else {
777 ro = &sro;
778 bzero(ro, sizeof(*ro));
779 }
780 }
781 if (m == 0) {
782 m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */
783 if (m == NULL) {
784 return;
785 }
786 tlen = 0;
787 m->m_data += max_linkhdr;
788 if (isipv6) {
789 VERIFY((MHLEN - max_linkhdr) >=
790 (sizeof(*ip6) + sizeof(*nth)));
791 bcopy((caddr_t)ip6, mtod(m, caddr_t),
792 sizeof(struct ip6_hdr));
793 ip6 = mtod(m, struct ip6_hdr *);
794 nth = (struct tcphdr *)(void *)(ip6 + 1);
795 } else {
796 VERIFY((MHLEN - max_linkhdr) >=
797 (sizeof(*ip) + sizeof(*nth)));
798 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
799 ip = mtod(m, struct ip *);
800 nth = (struct tcphdr *)(void *)(ip + 1);
801 }
802 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
803 #if MPTCP
804 if ((tp) && (tp->t_mpflags & TMPF_RESET)) {
805 flags = (TH_RST | TH_ACK);
806 } else
807 #endif
808 flags = TH_ACK;
809 } else {
810 m_freem(m->m_next);
811 m->m_next = 0;
812 m->m_data = (caddr_t)ipgen;
813 /* m_len is set later */
814 tlen = 0;
815 #define xchg(a, b, type) { type t; t = a; a = b; b = t; }
816 if (isipv6) {
817 ip6_getsrcifaddr_info(m, &sifscope, NULL);
818 ip6_getdstifaddr_info(m, &fifscope, NULL);
819 if (!in6_embedded_scope) {
820 m->m_pkthdr.pkt_flags &= ~PKTF_IFAINFO;
821 }
822 /* Expect 32-bit aligned IP on strict-align platforms */
823 IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6);
824 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
825 nth = (struct tcphdr *)(void *)(ip6 + 1);
826 } else {
827 /* Expect 32-bit aligned IP on strict-align platforms */
828 IP_HDR_STRICT_ALIGNMENT_CHECK(ip);
829 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
830 nth = (struct tcphdr *)(void *)(ip + 1);
831 }
832 if (th != nth) {
833 /*
834 * this is usually a case when an extension header
835 * exists between the IPv6 header and the
836 * TCP header.
837 */
838 nth->th_sport = th->th_sport;
839 nth->th_dport = th->th_dport;
840 }
841 xchg(nth->th_dport, nth->th_sport, n_short);
842 #undef xchg
843 }
844 if (isipv6) {
845 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) +
846 tlen));
847 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
848 ip6_output_setsrcifscope(m, sifscope, NULL);
849 ip6_output_setdstifscope(m, fifscope, NULL);
850 } else {
851 tlen += sizeof(struct tcpiphdr);
852 ip->ip_len = tlen;
853 ip->ip_ttl = (uint8_t)ip_defttl;
854 }
855 m->m_len = tlen;
856 m->m_pkthdr.len = tlen;
857 m->m_pkthdr.rcvif = 0;
858 if (tra->keep_alive) {
859 m->m_pkthdr.pkt_flags |= PKTF_KEEPALIVE;
860 }
861
862 nth->th_seq = htonl(seq);
863 nth->th_ack = htonl(ack);
864 nth->th_x2 = 0;
865 nth->th_off = sizeof(struct tcphdr) >> 2;
866 nth->th_flags = flags;
867 if (tp) {
868 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
869 } else {
870 nth->th_win = htons((u_short)win);
871 }
872 nth->th_urp = 0;
873 if (isipv6) {
874 nth->th_sum = 0;
875 nth->th_sum = in6_pseudo(&ip6->ip6_src, &ip6->ip6_dst,
876 htonl((tlen - sizeof(struct ip6_hdr)) + IPPROTO_TCP));
877 m->m_pkthdr.csum_flags = CSUM_TCPIPV6;
878 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
879 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
880 ro6 && ro6->ro_rt ? ro6->ro_rt->rt_ifp : NULL);
881 } else {
882 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
883 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
884 m->m_pkthdr.csum_flags = CSUM_TCP;
885 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
886 }
887 #if TCPDEBUG
888 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) {
889 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
890 }
891 #endif
892
893 #if NECP
894 necp_mark_packet_from_socket(m, tp ? tp->t_inpcb : NULL, 0, 0, 0, 0);
895 #endif /* NECP */
896
897 #if IPSEC
898 if (tp != NULL && tp->t_inpcb->inp_sp != NULL &&
899 ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
900 m_freem(m);
901 return;
902 }
903 #endif
904
905 if (tp != NULL) {
906 u_int32_t svc_flags = 0;
907 if (isipv6) {
908 svc_flags |= PKT_SCF_IPV6;
909 }
910 sotc = tp->t_inpcb->inp_socket->so_traffic_class;
911 if ((flags & TH_RST) == 0) {
912 set_packet_service_class(m, tp->t_inpcb->inp_socket,
913 sotc, svc_flags);
914 } else {
915 m_set_service_class(m, MBUF_SC_BK_SYS);
916 }
917
918 /* Embed flowhash and flow control flags */
919 m->m_pkthdr.pkt_flowsrc = FLOWSRC_INPCB;
920 m->m_pkthdr.pkt_flowid = tp->t_inpcb->inp_flowhash;
921 m->m_pkthdr.pkt_flags |= (PKTF_FLOW_ID | PKTF_FLOW_LOCALSRC | PKTF_FLOW_ADV);
922 m->m_pkthdr.pkt_proto = IPPROTO_TCP;
923 m->m_pkthdr.tx_tcp_pid = tp->t_inpcb->inp_socket->last_pid;
924 m->m_pkthdr.tx_tcp_e_pid = tp->t_inpcb->inp_socket->e_pid;
925
926 if (flags & TH_RST) {
927 m->m_pkthdr.comp_gencnt = tp->t_comp_gencnt;
928 }
929 } else {
930 if (flags & TH_RST) {
931 m->m_pkthdr.comp_gencnt = TCP_ACK_COMPRESSION_DUMMY;
932 m_set_service_class(m, MBUF_SC_BK_SYS);
933 }
934 }
935
936 if (isipv6) {
937 struct ip6_out_args ip6oa;
938 bzero(&ip6oa, sizeof(ip6oa));
939 ip6oa.ip6oa_boundif = tra->ifscope;
940 ip6oa.ip6oa_flags = IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR;
941 ip6oa.ip6oa_sotc = SO_TC_UNSPEC;
942 ip6oa.ip6oa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
943
944 if (tra->ifscope != IFSCOPE_NONE) {
945 ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF;
946 }
947 if (tra->nocell) {
948 ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR;
949 }
950 if (tra->noexpensive) {
951 ip6oa.ip6oa_flags |= IP6OAF_NO_EXPENSIVE;
952 }
953 if (tra->noconstrained) {
954 ip6oa.ip6oa_flags |= IP6OAF_NO_CONSTRAINED;
955 }
956 if (tra->awdl_unrestricted) {
957 ip6oa.ip6oa_flags |= IP6OAF_AWDL_UNRESTRICTED;
958 }
959 if (tra->intcoproc_allowed) {
960 ip6oa.ip6oa_flags |= IP6OAF_INTCOPROC_ALLOWED;
961 }
962 if (tra->management_allowed) {
963 ip6oa.ip6oa_flags |= IP6OAF_MANAGEMENT_ALLOWED;
964 }
965 ip6oa.ip6oa_sotc = sotc;
966 if (tp != NULL) {
967 if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) {
968 ip6oa.ip6oa_flags |= IP6OAF_QOSMARKING_ALLOWED;
969 }
970 ip6oa.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount;
971 if (check_qos_marking_again) {
972 ip6oa.ip6oa_flags |= IP6OAF_REDO_QOSMARKING_POLICY;
973 }
974 ip6oa.ip6oa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype;
975 }
976 (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL,
977 NULL, &ip6oa);
978
979 if (check_qos_marking_again) {
980 struct inpcb *inp = tp->t_inpcb;
981 inp->inp_policyresult.results.qos_marking_gencount = ip6oa.qos_marking_gencount;
982 if (ip6oa.ip6oa_flags & IP6OAF_QOSMARKING_ALLOWED) {
983 inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED;
984 } else {
985 inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED;
986 }
987 }
988
989 if (tp != NULL && ro6 != NULL && ro6->ro_rt != NULL &&
990 (outif = ro6->ro_rt->rt_ifp) !=
991 tp->t_inpcb->in6p_last_outifp) {
992 tp->t_inpcb->in6p_last_outifp = outif;
993 #if SKYWALK
994 if (NETNS_TOKEN_VALID(&tp->t_inpcb->inp_netns_token)) {
995 netns_set_ifnet(&tp->t_inpcb->inp_netns_token,
996 tp->t_inpcb->in6p_last_outifp);
997 }
998 #endif /* SKYWALK */
999 }
1000
1001 if (ro6 == &sro6) {
1002 ROUTE_RELEASE(ro6);
1003 }
1004 } else {
1005 struct ip_out_args ipoa;
1006 bzero(&ipoa, sizeof(ipoa));
1007 ipoa.ipoa_boundif = tra->ifscope;
1008 ipoa.ipoa_flags = IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR;
1009 ipoa.ipoa_sotc = SO_TC_UNSPEC;
1010 ipoa.ipoa_netsvctype = _NET_SERVICE_TYPE_UNSPEC;
1011
1012 if (tra->ifscope != IFSCOPE_NONE) {
1013 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
1014 }
1015 if (tra->nocell) {
1016 ipoa.ipoa_flags |= IPOAF_NO_CELLULAR;
1017 }
1018 if (tra->noexpensive) {
1019 ipoa.ipoa_flags |= IPOAF_NO_EXPENSIVE;
1020 }
1021 if (tra->noconstrained) {
1022 ipoa.ipoa_flags |= IPOAF_NO_CONSTRAINED;
1023 }
1024 if (tra->awdl_unrestricted) {
1025 ipoa.ipoa_flags |= IPOAF_AWDL_UNRESTRICTED;
1026 }
1027 if (tra->management_allowed) {
1028 ipoa.ipoa_flags |= IPOAF_MANAGEMENT_ALLOWED;
1029 }
1030 ipoa.ipoa_sotc = sotc;
1031 if (tp != NULL) {
1032 if ((tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_ALLOWED)) {
1033 ipoa.ipoa_flags |= IPOAF_QOSMARKING_ALLOWED;
1034 }
1035 if (!(tp->t_inpcb->inp_socket->so_flags1 & SOF1_QOSMARKING_POLICY_OVERRIDE)) {
1036 ipoa.ipoa_flags |= IPOAF_REDO_QOSMARKING_POLICY;
1037 }
1038 ipoa.qos_marking_gencount = tp->t_inpcb->inp_policyresult.results.qos_marking_gencount;
1039 ipoa.ipoa_netsvctype = tp->t_inpcb->inp_socket->so_netsvctype;
1040 }
1041 if (ro != &sro) {
1042 /* Copy the cached route and take an extra reference */
1043 inp_route_copyout(tp->t_inpcb, &sro);
1044 }
1045 /*
1046 * For consistency, pass a local route copy.
1047 */
1048 (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa);
1049
1050 if (check_qos_marking_again) {
1051 struct inpcb *inp = tp->t_inpcb;
1052 inp->inp_policyresult.results.qos_marking_gencount = ipoa.qos_marking_gencount;
1053 if (ipoa.ipoa_flags & IPOAF_QOSMARKING_ALLOWED) {
1054 inp->inp_socket->so_flags1 |= SOF1_QOSMARKING_ALLOWED;
1055 } else {
1056 inp->inp_socket->so_flags1 &= ~SOF1_QOSMARKING_ALLOWED;
1057 }
1058 }
1059 if (tp != NULL && sro.ro_rt != NULL &&
1060 (outif = sro.ro_rt->rt_ifp) !=
1061 tp->t_inpcb->inp_last_outifp) {
1062 tp->t_inpcb->inp_last_outifp = outif;
1063 #if SKYWALK
1064 if (NETNS_TOKEN_VALID(&tp->t_inpcb->inp_netns_token)) {
1065 netns_set_ifnet(&tp->t_inpcb->inp_netns_token, outif);
1066 }
1067 #endif /* SKYWALK */
1068 }
1069 if (ro != &sro) {
1070 /* Synchronize cached PCB route */
1071 inp_route_copyin(tp->t_inpcb, &sro);
1072 } else {
1073 ROUTE_RELEASE(&sro);
1074 }
1075 }
1076 }
1077
1078 /*
1079 * Create a new TCP control block, making an
1080 * empty reassembly queue and hooking it to the argument
1081 * protocol control block. The `inp' parameter must have
1082 * come from the zone allocator set up in tcp_init().
1083 */
1084 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp)1085 tcp_newtcpcb(struct inpcb *inp)
1086 {
1087 struct inp_tp *it;
1088 struct tcpcb *tp;
1089 struct socket *so = inp->inp_socket;
1090 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1091 uint32_t random_32;
1092
1093 calculate_tcp_clock();
1094
1095 if ((so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) == 0) {
1096 it = (struct inp_tp *)(void *)inp;
1097 tp = &it->tcb;
1098 } else {
1099 tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb;
1100 }
1101
1102 bzero((char *) tp, sizeof(struct tcpcb));
1103 LIST_INIT(&tp->t_segq);
1104 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
1105
1106 tp->t_flags = (TF_REQ_SCALE | TF_REQ_TSTMP);
1107 tp->t_flagsext |= TF_SACK_ENABLE;
1108
1109 TAILQ_INIT(&tp->snd_holes);
1110 SLIST_INIT(&tp->t_rxt_segments);
1111 SLIST_INIT(&tp->t_notify_ack);
1112 tp->t_inpcb = inp;
1113 /*
1114 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1115 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
1116 * reasonable initial retransmit time.
1117 */
1118 tp->t_srtt = TCPTV_SRTTBASE;
1119 tp->t_rttvar =
1120 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
1121 tp->t_rttmin = tcp_TCPTV_MIN;
1122 tp->t_rxtcur = TCPTV_RTOBASE;
1123
1124 if (tcp_use_newreno) {
1125 /* use newreno by default */
1126 tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX;
1127 #if (DEVELOPMENT || DEBUG)
1128 } else if (tcp_use_ledbat) {
1129 /* use ledbat for testing */
1130 tp->tcp_cc_index = TCP_CC_ALGO_BACKGROUND_INDEX;
1131 #endif
1132 } else {
1133 tp->tcp_cc_index = TCP_CC_ALGO_CUBIC_INDEX;
1134 }
1135
1136 tcp_cc_allocate_state(tp);
1137
1138 if (CC_ALGO(tp)->init != NULL) {
1139 CC_ALGO(tp)->init(tp);
1140 }
1141
1142 /* Initialize rledbat if we are using recv_bg */
1143 if (tcp_rledbat == 1 && TCP_RECV_BG(inp->inp_socket) &&
1144 tcp_cc_rledbat.init != NULL) {
1145 tcp_cc_rledbat.init(tp);
1146 }
1147
1148 tp->snd_cwnd = tcp_initial_cwnd(tp);
1149 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1150 tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1151 tp->t_rcvtime = tcp_now;
1152 tp->tentry.timer_start = tcp_now;
1153 tp->rcv_unackwin = tcp_now;
1154 tp->t_persist_timeout = tcp_max_persist_timeout;
1155 tp->t_persist_stop = 0;
1156 tp->t_flagsext |= TF_RCVUNACK_WAITSS;
1157 tp->t_rexmtthresh = (uint8_t)tcprexmtthresh;
1158 tp->rfbuf_ts = tcp_now;
1159 tp->rfbuf_space = tcp_initial_cwnd(tp);
1160 tp->t_forced_acks = TCP_FORCED_ACKS_COUNT;
1161
1162 /* Enable bandwidth measurement on this connection */
1163 tp->t_flagsext |= TF_MEASURESNDBW;
1164 if (tp->t_bwmeas == NULL) {
1165 tp->t_bwmeas = tcp_bwmeas_alloc(tp);
1166 if (tp->t_bwmeas == NULL) {
1167 tp->t_flagsext &= ~TF_MEASURESNDBW;
1168 }
1169 }
1170
1171 /* Clear time wait tailq entry */
1172 tp->t_twentry.tqe_next = NULL;
1173 tp->t_twentry.tqe_prev = NULL;
1174
1175 read_frandom(&random_32, sizeof(random_32));
1176 tp->t_comp_gencnt = random_32;
1177 if (tp->t_comp_gencnt <= TCP_ACK_COMPRESSION_DUMMY) {
1178 tp->t_comp_gencnt = TCP_ACK_COMPRESSION_DUMMY + 1;
1179 }
1180 tp->t_comp_lastinc = tcp_now;
1181
1182 if (__probable(tcp_randomize_timestamps)) {
1183 tp->t_ts_offset = random_32;
1184 }
1185
1186 /*
1187 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
1188 * because the socket may be bound to an IPv6 wildcard address,
1189 * which may match an IPv4-mapped IPv6 address.
1190 */
1191 inp->inp_ip_ttl = (uint8_t)ip_defttl;
1192 inp->inp_ppcb = (caddr_t)tp;
1193 return tp; /* XXX */
1194 }
1195
1196 /*
1197 * Drop a TCP connection, reporting
1198 * the specified error. If connection is synchronized,
1199 * then send a RST to peer.
1200 */
1201 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)1202 tcp_drop(struct tcpcb *tp, int errno)
1203 {
1204 struct socket *so = tp->t_inpcb->inp_socket;
1205 #if CONFIG_DTRACE
1206 struct inpcb *inp = tp->t_inpcb;
1207 #endif
1208
1209 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1210 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1211 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1212 TCP_LOG_STATE(tp, TCPS_CLOSED);
1213 tp->t_state = TCPS_CLOSED;
1214 (void) tcp_output(tp);
1215 tcpstat.tcps_drops++;
1216 } else {
1217 tcpstat.tcps_conndrops++;
1218 }
1219 if (errno == ETIMEDOUT && tp->t_softerror) {
1220 errno = tp->t_softerror;
1221 }
1222 so->so_error = (u_short)errno;
1223
1224 TCP_LOG_CONNECTION_SUMMARY(tp);
1225
1226 return tcp_close(tp);
1227 }
1228
1229 void
tcp_getrt_rtt(struct tcpcb * tp,struct rtentry * rt)1230 tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt)
1231 {
1232 u_int32_t rtt = rt->rt_rmx.rmx_rtt;
1233 int isnetlocal = (tp->t_flags & TF_LOCAL);
1234
1235 TCP_LOG_RTM_RTT(tp, rt);
1236
1237 if (rtt != 0 && tcp_init_rtt_from_cache != 0) {
1238 /*
1239 * XXX the lock bit for RTT indicates that the value
1240 * is also a minimum value; this is subject to time.
1241 */
1242 if (rt->rt_rmx.rmx_locks & RTV_RTT) {
1243 tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ);
1244 } else {
1245 tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN :
1246 TCPTV_REXMTMIN;
1247 }
1248
1249 tp->t_srtt =
1250 rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1251 tcpstat.tcps_usedrtt++;
1252
1253 if (rt->rt_rmx.rmx_rttvar) {
1254 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1255 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1256 tcpstat.tcps_usedrttvar++;
1257 } else {
1258 /* default variation is +- 1 rtt */
1259 tp->t_rttvar =
1260 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
1261 }
1262
1263 /*
1264 * The RTO formula in the route metric case is based on:
1265 * srtt + 4 * rttvar
1266 * modulo the min, max and slop
1267 */
1268 TCPT_RANGESET(tp->t_rxtcur,
1269 TCP_REXMTVAL(tp),
1270 tp->t_rttmin, TCPTV_REXMTMAX,
1271 TCP_ADD_REXMTSLOP(tp));
1272 }
1273
1274 TCP_LOG_RTT_INFO(tp);
1275 }
1276
1277 static inline void
tcp_create_ifnet_stats_per_flow(struct tcpcb * tp,struct ifnet_stats_per_flow * ifs)1278 tcp_create_ifnet_stats_per_flow(struct tcpcb *tp,
1279 struct ifnet_stats_per_flow *ifs)
1280 {
1281 struct inpcb *inp;
1282 struct socket *so;
1283 if (tp == NULL || ifs == NULL) {
1284 return;
1285 }
1286
1287 bzero(ifs, sizeof(*ifs));
1288 inp = tp->t_inpcb;
1289 so = inp->inp_socket;
1290
1291 ifs->ipv4 = (inp->inp_vflag & INP_IPV6) ? 0 : 1;
1292 ifs->local = (tp->t_flags & TF_LOCAL) ? 1 : 0;
1293 ifs->connreset = (so->so_error == ECONNRESET) ? 1 : 0;
1294 ifs->conntimeout = (so->so_error == ETIMEDOUT) ? 1 : 0;
1295 ifs->ecn_flags = tp->ecn_flags;
1296 ifs->txretransmitbytes = tp->t_stat.txretransmitbytes;
1297 ifs->rxoutoforderbytes = tp->t_stat.rxoutoforderbytes;
1298 ifs->rxmitpkts = tp->t_stat.rxmitpkts;
1299 ifs->rcvoopack = tp->t_rcvoopack;
1300 ifs->pawsdrop = tp->t_pawsdrop;
1301 ifs->sack_recovery_episodes = tp->t_sack_recovery_episode;
1302 ifs->reordered_pkts = tp->t_reordered_pkts;
1303 ifs->dsack_sent = tp->t_dsack_sent;
1304 ifs->dsack_recvd = tp->t_dsack_recvd;
1305 ifs->srtt = tp->t_srtt;
1306 ifs->rttupdated = tp->t_rttupdated;
1307 ifs->rttvar = tp->t_rttvar;
1308 ifs->rttmin = get_base_rtt(tp);
1309 if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_sndbw_max > 0) {
1310 ifs->bw_sndbw_max = tp->t_bwmeas->bw_sndbw_max;
1311 } else {
1312 ifs->bw_sndbw_max = 0;
1313 }
1314 if (tp->t_bwmeas != NULL && tp->t_bwmeas->bw_rcvbw_max > 0) {
1315 ifs->bw_rcvbw_max = tp->t_bwmeas->bw_rcvbw_max;
1316 } else {
1317 ifs->bw_rcvbw_max = 0;
1318 }
1319 ifs->bk_txpackets = so->so_tc_stats[MBUF_TC_BK].txpackets;
1320 ifs->txpackets = inp->inp_stat->txpackets;
1321 ifs->rxpackets = inp->inp_stat->rxpackets;
1322 }
1323
1324 static inline void
tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow * ifs,struct if_tcp_ecn_perf_stat * stat)1325 tcp_flow_ecn_perf_stats(struct ifnet_stats_per_flow *ifs,
1326 struct if_tcp_ecn_perf_stat *stat)
1327 {
1328 u_int64_t curval, oldval;
1329 stat->total_txpkts += ifs->txpackets;
1330 stat->total_rxpkts += ifs->rxpackets;
1331 stat->total_rxmitpkts += ifs->rxmitpkts;
1332 stat->total_oopkts += ifs->rcvoopack;
1333 stat->total_reorderpkts += (ifs->reordered_pkts +
1334 ifs->pawsdrop + ifs->dsack_sent + ifs->dsack_recvd);
1335
1336 /* Average RTT */
1337 curval = ifs->srtt >> TCP_RTT_SHIFT;
1338 if (curval > 0 && ifs->rttupdated >= 16) {
1339 if (stat->rtt_avg == 0) {
1340 stat->rtt_avg = curval;
1341 } else {
1342 oldval = stat->rtt_avg;
1343 stat->rtt_avg = ((oldval << 4) - oldval + curval) >> 4;
1344 }
1345 }
1346
1347 /* RTT variance */
1348 curval = ifs->rttvar >> TCP_RTTVAR_SHIFT;
1349 if (curval > 0 && ifs->rttupdated >= 16) {
1350 if (stat->rtt_var == 0) {
1351 stat->rtt_var = curval;
1352 } else {
1353 oldval = stat->rtt_var;
1354 stat->rtt_var =
1355 ((oldval << 4) - oldval + curval) >> 4;
1356 }
1357 }
1358
1359 /* SACK episodes */
1360 stat->sack_episodes += ifs->sack_recovery_episodes;
1361 if (ifs->connreset) {
1362 stat->rst_drop++;
1363 }
1364 }
1365
1366 static inline void
tcp_flow_lim_stats(struct ifnet_stats_per_flow * ifs,struct if_lim_perf_stat * stat)1367 tcp_flow_lim_stats(struct ifnet_stats_per_flow *ifs,
1368 struct if_lim_perf_stat *stat)
1369 {
1370 u_int64_t curval, oldval;
1371
1372 stat->lim_total_txpkts += ifs->txpackets;
1373 stat->lim_total_rxpkts += ifs->rxpackets;
1374 stat->lim_total_retxpkts += ifs->rxmitpkts;
1375 stat->lim_total_oopkts += ifs->rcvoopack;
1376
1377 if (ifs->bw_sndbw_max > 0) {
1378 /* convert from bytes per ms to bits per second */
1379 ifs->bw_sndbw_max *= 8000;
1380 stat->lim_ul_max_bandwidth = MAX(stat->lim_ul_max_bandwidth,
1381 ifs->bw_sndbw_max);
1382 }
1383
1384 if (ifs->bw_rcvbw_max > 0) {
1385 /* convert from bytes per ms to bits per second */
1386 ifs->bw_rcvbw_max *= 8000;
1387 stat->lim_dl_max_bandwidth = MAX(stat->lim_dl_max_bandwidth,
1388 ifs->bw_rcvbw_max);
1389 }
1390
1391 /* Average RTT */
1392 curval = ifs->srtt >> TCP_RTT_SHIFT;
1393 if (curval > 0 && ifs->rttupdated >= 16) {
1394 if (stat->lim_rtt_average == 0) {
1395 stat->lim_rtt_average = curval;
1396 } else {
1397 oldval = stat->lim_rtt_average;
1398 stat->lim_rtt_average =
1399 ((oldval << 4) - oldval + curval) >> 4;
1400 }
1401 }
1402
1403 /* RTT variance */
1404 curval = ifs->rttvar >> TCP_RTTVAR_SHIFT;
1405 if (curval > 0 && ifs->rttupdated >= 16) {
1406 if (stat->lim_rtt_variance == 0) {
1407 stat->lim_rtt_variance = curval;
1408 } else {
1409 oldval = stat->lim_rtt_variance;
1410 stat->lim_rtt_variance =
1411 ((oldval << 4) - oldval + curval) >> 4;
1412 }
1413 }
1414
1415 if (stat->lim_rtt_min == 0) {
1416 stat->lim_rtt_min = ifs->rttmin;
1417 } else {
1418 stat->lim_rtt_min = MIN(stat->lim_rtt_min, ifs->rttmin);
1419 }
1420
1421 /* connection timeouts */
1422 stat->lim_conn_attempts++;
1423 if (ifs->conntimeout) {
1424 stat->lim_conn_timeouts++;
1425 }
1426
1427 /* bytes sent using background delay-based algorithms */
1428 stat->lim_bk_txpkts += ifs->bk_txpackets;
1429 }
1430
1431 /*
1432 * Close a TCP control block:
1433 * discard all space held by the tcp
1434 * discard internet protocol block
1435 * wake up any sleepers
1436 */
1437 struct tcpcb *
tcp_close(struct tcpcb * tp)1438 tcp_close(struct tcpcb *tp)
1439 {
1440 struct inpcb *inp = tp->t_inpcb;
1441 struct socket *so = inp->inp_socket;
1442 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1443 struct route *ro;
1444 struct rtentry *rt;
1445 int dosavessthresh;
1446 struct ifnet_stats_per_flow ifs;
1447
1448 /* tcp_close was called previously, bail */
1449 if (inp->inp_ppcb == NULL) {
1450 return NULL;
1451 }
1452
1453 tcp_del_fsw_flow(tp);
1454
1455 tcp_canceltimers(tp);
1456 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp, 0, 0, 0, 0);
1457
1458 /*
1459 * If another thread for this tcp is currently in ip (indicated by
1460 * the TF_SENDINPROG flag), defer the cleanup until after it returns
1461 * back to tcp. This is done to serialize the close until after all
1462 * pending output is finished, in order to avoid having the PCB be
1463 * detached and the cached route cleaned, only for ip to cache the
1464 * route back into the PCB again. Note that we've cleared all the
1465 * timers at this point. Set TF_CLOSING to indicate to tcp_output()
1466 * that is should call us again once it returns from ip; at that
1467 * point both flags should be cleared and we can proceed further
1468 * with the cleanup.
1469 */
1470 if ((tp->t_flags & TF_CLOSING) ||
1471 inp->inp_sndinprog_cnt > 0) {
1472 tp->t_flags |= TF_CLOSING;
1473 return NULL;
1474 }
1475
1476 TCP_LOG_CONNECTION_SUMMARY(tp);
1477
1478 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1479 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
1480
1481 ro = (isipv6 ? (struct route *)&inp->in6p_route : &inp->inp_route);
1482 rt = ro->ro_rt;
1483 if (rt != NULL) {
1484 RT_LOCK_SPIN(rt);
1485 }
1486
1487 /*
1488 * If we got enough samples through the srtt filter,
1489 * save the rtt and rttvar in the routing entry.
1490 * 'Enough' is arbitrarily defined as the 16 samples.
1491 * 16 samples is enough for the srtt filter to converge
1492 * to within 5% of the correct value; fewer samples and
1493 * we could save a very bogus rtt.
1494 *
1495 * Don't update the default route's characteristics and don't
1496 * update anything that the user "locked".
1497 */
1498 if (tp->t_rttupdated >= 16) {
1499 u_int32_t i = 0;
1500 bool log_rtt = false;
1501
1502 if (isipv6) {
1503 struct sockaddr_in6 *sin6;
1504
1505 if (rt == NULL) {
1506 goto no_valid_rt;
1507 }
1508 sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt);
1509 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
1510 goto no_valid_rt;
1511 }
1512 } else if (ROUTE_UNUSABLE(ro) ||
1513 SIN(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) {
1514 DTRACE_TCP4(state__change, void, NULL,
1515 struct inpcb *, inp, struct tcpcb *, tp,
1516 int32_t, TCPS_CLOSED);
1517 TCP_LOG_STATE(tp, TCPS_CLOSED);
1518 tp->t_state = TCPS_CLOSED;
1519 goto no_valid_rt;
1520 }
1521
1522 RT_LOCK_ASSERT_HELD(rt);
1523 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1524 i = tp->t_srtt *
1525 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1526 if (rt->rt_rmx.rmx_rtt && i) {
1527 /*
1528 * filter this update to half the old & half
1529 * the new values, converting scale.
1530 * See route.h and tcp_var.h for a
1531 * description of the scaling constants.
1532 */
1533 rt->rt_rmx.rmx_rtt =
1534 (rt->rt_rmx.rmx_rtt + i) / 2;
1535 } else {
1536 rt->rt_rmx.rmx_rtt = i;
1537 }
1538 tcpstat.tcps_cachedrtt++;
1539 log_rtt = true;
1540 }
1541 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1542 i = tp->t_rttvar *
1543 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1544 if (rt->rt_rmx.rmx_rttvar && i) {
1545 rt->rt_rmx.rmx_rttvar =
1546 (rt->rt_rmx.rmx_rttvar + i) / 2;
1547 } else {
1548 rt->rt_rmx.rmx_rttvar = i;
1549 }
1550 tcpstat.tcps_cachedrttvar++;
1551 log_rtt = true;
1552 }
1553 if (log_rtt) {
1554 TCP_LOG_RTM_RTT(tp, rt);
1555 TCP_LOG_RTT_INFO(tp);
1556 }
1557 /*
1558 * The old comment here said:
1559 * update the pipelimit (ssthresh) if it has been updated
1560 * already or if a pipesize was specified & the threshhold
1561 * got below half the pipesize. I.e., wait for bad news
1562 * before we start updating, then update on both good
1563 * and bad news.
1564 *
1565 * But we want to save the ssthresh even if no pipesize is
1566 * specified explicitly in the route, because such
1567 * connections still have an implicit pipesize specified
1568 * by the global tcp_sendspace. In the absence of a reliable
1569 * way to calculate the pipesize, it will have to do.
1570 */
1571 i = tp->snd_ssthresh;
1572 if (rt->rt_rmx.rmx_sendpipe != 0) {
1573 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
1574 } else {
1575 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
1576 }
1577 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1578 i != 0 && rt->rt_rmx.rmx_ssthresh != 0) ||
1579 dosavessthresh) {
1580 /*
1581 * convert the limit from user data bytes to
1582 * packets then to packet data bytes.
1583 */
1584 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
1585 if (i < 2) {
1586 i = 2;
1587 }
1588 i *= (u_int32_t)(tp->t_maxseg +
1589 isipv6 ? sizeof(struct ip6_hdr) +
1590 sizeof(struct tcphdr) :
1591 sizeof(struct tcpiphdr));
1592 if (rt->rt_rmx.rmx_ssthresh) {
1593 rt->rt_rmx.rmx_ssthresh =
1594 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1595 } else {
1596 rt->rt_rmx.rmx_ssthresh = i;
1597 }
1598 tcpstat.tcps_cachedssthresh++;
1599 }
1600 }
1601
1602 /*
1603 * Mark route for deletion if no information is cached.
1604 */
1605 if (rt != NULL && (so->so_flags & SOF_OVERFLOW)) {
1606 if (!(rt->rt_rmx.rmx_locks & RTV_RTT) &&
1607 rt->rt_rmx.rmx_rtt == 0) {
1608 rt->rt_flags |= RTF_DELCLONE;
1609 }
1610 }
1611
1612 no_valid_rt:
1613 if (rt != NULL) {
1614 RT_UNLOCK(rt);
1615 }
1616
1617 /* free the reassembly queue, if any */
1618 (void) tcp_freeq(tp);
1619
1620 /* performance stats per interface */
1621 tcp_create_ifnet_stats_per_flow(tp, &ifs);
1622 tcp_update_stats_per_flow(&ifs, inp->inp_last_outifp);
1623
1624 tcp_free_sackholes(tp);
1625 tcp_notify_ack_free(tp);
1626
1627 inp_decr_sndbytes_allunsent(so, tp->snd_una);
1628
1629 if (tp->t_bwmeas != NULL) {
1630 tcp_bwmeas_free(tp);
1631 }
1632 tcp_rxtseg_clean(tp);
1633 /* Free the packet list */
1634 if (tp->t_pktlist_head != NULL) {
1635 m_freem_list(tp->t_pktlist_head);
1636 }
1637 TCP_PKTLIST_CLEAR(tp);
1638
1639 if (so->so_flags1 & SOF1_CACHED_IN_SOCK_LAYER) {
1640 inp->inp_saved_ppcb = (caddr_t) tp;
1641 }
1642
1643 TCP_LOG_STATE(tp, TCPS_CLOSED);
1644 tp->t_state = TCPS_CLOSED;
1645
1646 /*
1647 * Issue a wakeup before detach so that we don't miss
1648 * a wakeup
1649 */
1650 sodisconnectwakeup(so);
1651
1652 /*
1653 * Make sure to clear the TCP Keep Alive Offload as it is
1654 * ref counted on the interface
1655 */
1656 tcp_clear_keep_alive_offload(so);
1657
1658 /*
1659 * If this is a socket that does not want to wakeup the device
1660 * for it's traffic, the application might need to know that the
1661 * socket is closed, send a notification.
1662 */
1663 if ((so->so_options & SO_NOWAKEFROMSLEEP) &&
1664 inp->inp_state != INPCB_STATE_DEAD &&
1665 !(inp->inp_flags2 & INP2_TIMEWAIT)) {
1666 socket_post_kev_msg_closed(so);
1667 }
1668
1669 if (CC_ALGO(tp)->cleanup != NULL) {
1670 CC_ALGO(tp)->cleanup(tp);
1671 }
1672
1673 tp->tcp_cc_index = TCP_CC_ALGO_NONE;
1674
1675 if (TCP_USE_RLEDBAT(tp, so) && tcp_cc_rledbat.cleanup != NULL) {
1676 tcp_cc_rledbat.cleanup(tp);
1677 }
1678
1679 /* Can happen if we close the socket before receiving the third ACK */
1680 if ((tp->t_tfo_flags & TFO_F_COOKIE_VALID)) {
1681 OSDecrementAtomic(&tcp_tfo_halfcnt);
1682
1683 /* Panic if something has gone terribly wrong. */
1684 VERIFY(tcp_tfo_halfcnt >= 0);
1685
1686 tp->t_tfo_flags &= ~TFO_F_COOKIE_VALID;
1687 }
1688
1689 if (SOCK_CHECK_DOM(so, PF_INET6)) {
1690 in6_pcbdetach(inp);
1691 } else {
1692 in_pcbdetach(inp);
1693 }
1694
1695 /*
1696 * Call soisdisconnected after detach because it might unlock the socket
1697 */
1698 soisdisconnected(so);
1699 tcpstat.tcps_closed++;
1700 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END,
1701 tcpstat.tcps_closed, 0, 0, 0, 0);
1702 return NULL;
1703 }
1704
1705 int
tcp_freeq(struct tcpcb * tp)1706 tcp_freeq(struct tcpcb *tp)
1707 {
1708 struct tseg_qent *q;
1709 int rv = 0;
1710 int count = 0;
1711
1712 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
1713 LIST_REMOVE(q, tqe_q);
1714 tp->t_reassq_mbcnt -= MSIZE + (q->tqe_m->m_flags & M_EXT) ?
1715 q->tqe_m->m_ext.ext_size : 0;
1716 m_freem(q->tqe_m);
1717 zfree(tcp_reass_zone, q);
1718 rv = 1;
1719 count++;
1720 }
1721 tp->t_reassqlen = 0;
1722 if (count > 0) {
1723 OSAddAtomic(-count, &tcp_reass_total_qlen);
1724 }
1725 return rv;
1726 }
1727
1728
1729 void
tcp_drain(void)1730 tcp_drain(void)
1731 {
1732 struct inpcb *inp;
1733 struct tcpcb *tp;
1734
1735 if (!lck_rw_try_lock_exclusive(&tcbinfo.ipi_lock)) {
1736 return;
1737 }
1738
1739 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1740 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) !=
1741 WNT_STOPUSING) {
1742 socket_lock(inp->inp_socket, 1);
1743 if (in_pcb_checkstate(inp, WNT_RELEASE, 1)
1744 == WNT_STOPUSING) {
1745 /* lost a race, try the next one */
1746 socket_unlock(inp->inp_socket, 1);
1747 continue;
1748 }
1749 tp = intotcpcb(inp);
1750
1751 so_drain_extended_bk_idle(inp->inp_socket);
1752
1753 socket_unlock(inp->inp_socket, 1);
1754 }
1755 }
1756 lck_rw_done(&tcbinfo.ipi_lock);
1757 }
1758
1759 /*
1760 * Notify a tcp user of an asynchronous error;
1761 * store error as soft error, but wake up user
1762 * (for now, won't do anything until can select for soft error).
1763 *
1764 * Do not wake up user since there currently is no mechanism for
1765 * reporting soft errors (yet - a kqueue filter may be added).
1766 */
1767 static void
tcp_notify(struct inpcb * inp,int error)1768 tcp_notify(struct inpcb *inp, int error)
1769 {
1770 struct tcpcb *tp;
1771
1772 if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD)) {
1773 return; /* pcb is gone already */
1774 }
1775 tp = (struct tcpcb *)inp->inp_ppcb;
1776
1777 VERIFY(tp != NULL);
1778 /*
1779 * Ignore some errors if we are hooked up.
1780 * If connection hasn't completed, has retransmitted several times,
1781 * and receives a second error, give up now. This is better
1782 * than waiting a long time to establish a connection that
1783 * can never complete.
1784 */
1785 if (tp->t_state == TCPS_ESTABLISHED &&
1786 (error == EHOSTUNREACH || error == ENETUNREACH ||
1787 error == EHOSTDOWN)) {
1788 if (inp->inp_route.ro_rt) {
1789 rtfree(inp->inp_route.ro_rt);
1790 inp->inp_route.ro_rt = (struct rtentry *)NULL;
1791 }
1792 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1793 tp->t_softerror) {
1794 tcp_drop(tp, error);
1795 } else {
1796 tp->t_softerror = error;
1797 }
1798 }
1799
1800 struct bwmeas *
tcp_bwmeas_alloc(struct tcpcb * tp)1801 tcp_bwmeas_alloc(struct tcpcb *tp)
1802 {
1803 struct bwmeas *elm;
1804 elm = zalloc_flags(tcp_bwmeas_zone, Z_ZERO | Z_WAITOK);
1805 elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE;
1806 elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg;
1807 return elm;
1808 }
1809
1810 void
tcp_bwmeas_free(struct tcpcb * tp)1811 tcp_bwmeas_free(struct tcpcb *tp)
1812 {
1813 zfree(tcp_bwmeas_zone, tp->t_bwmeas);
1814 tp->t_bwmeas = NULL;
1815 tp->t_flagsext &= ~(TF_MEASURESNDBW);
1816 }
1817
1818 int
get_tcp_inp_list(struct inpcb ** inp_list,int n,inp_gen_t gencnt)1819 get_tcp_inp_list(struct inpcb **inp_list, int n, inp_gen_t gencnt)
1820 {
1821 struct tcpcb *tp;
1822 struct inpcb *inp;
1823 int i = 0;
1824
1825 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
1826 if (inp->inp_gencnt <= gencnt &&
1827 inp->inp_state != INPCB_STATE_DEAD) {
1828 inp_list[i++] = inp;
1829 }
1830 if (i >= n) {
1831 break;
1832 }
1833 }
1834
1835 TAILQ_FOREACH(tp, &tcp_tw_tailq, t_twentry) {
1836 inp = tp->t_inpcb;
1837 if (inp->inp_gencnt <= gencnt &&
1838 inp->inp_state != INPCB_STATE_DEAD) {
1839 inp_list[i++] = inp;
1840 }
1841 if (i >= n) {
1842 break;
1843 }
1844 }
1845 return i;
1846 }
1847
1848 /*
1849 * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format.
1850 * The otcpcb data structure is passed to user space and must not change.
1851 */
1852 static void
tcpcb_to_otcpcb(struct tcpcb * tp,struct otcpcb * otp)1853 tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp)
1854 {
1855 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
1856 otp->t_dupacks = tp->t_dupacks;
1857 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
1858 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
1859 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
1860 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
1861 otp->t_inpcb =
1862 (_TCPCB_PTR(struct inpcb *))VM_KERNEL_ADDRPERM(tp->t_inpcb);
1863 otp->t_state = tp->t_state;
1864 otp->t_flags = tp->t_flags;
1865 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
1866 otp->snd_una = tp->snd_una;
1867 otp->snd_max = tp->snd_max;
1868 otp->snd_nxt = tp->snd_nxt;
1869 otp->snd_up = tp->snd_up;
1870 otp->snd_wl1 = tp->snd_wl1;
1871 otp->snd_wl2 = tp->snd_wl2;
1872 otp->iss = tp->iss;
1873 otp->irs = tp->irs;
1874 otp->rcv_nxt = tp->rcv_nxt;
1875 otp->rcv_adv = tp->rcv_adv;
1876 otp->rcv_wnd = tp->rcv_wnd;
1877 otp->rcv_up = tp->rcv_up;
1878 otp->snd_wnd = tp->snd_wnd;
1879 otp->snd_cwnd = tp->snd_cwnd;
1880 otp->snd_ssthresh = tp->snd_ssthresh;
1881 otp->t_maxopd = tp->t_maxopd;
1882 otp->t_rcvtime = tp->t_rcvtime;
1883 otp->t_starttime = tp->t_starttime;
1884 otp->t_rtttime = tp->t_rtttime;
1885 otp->t_rtseq = tp->t_rtseq;
1886 otp->t_rxtcur = tp->t_rxtcur;
1887 otp->t_maxseg = tp->t_maxseg;
1888 otp->t_srtt = tp->t_srtt;
1889 otp->t_rttvar = tp->t_rttvar;
1890 otp->t_rxtshift = tp->t_rxtshift;
1891 otp->t_rttmin = tp->t_rttmin;
1892 otp->t_rttupdated = tp->t_rttupdated;
1893 otp->max_sndwnd = tp->max_sndwnd;
1894 otp->t_softerror = tp->t_softerror;
1895 otp->t_oobflags = tp->t_oobflags;
1896 otp->t_iobc = tp->t_iobc;
1897 otp->snd_scale = tp->snd_scale;
1898 otp->rcv_scale = tp->rcv_scale;
1899 otp->request_r_scale = tp->request_r_scale;
1900 otp->requested_s_scale = tp->requested_s_scale;
1901 otp->ts_recent = tp->ts_recent;
1902 otp->ts_recent_age = tp->ts_recent_age;
1903 otp->last_ack_sent = tp->last_ack_sent;
1904 otp->cc_send = 0;
1905 otp->cc_recv = 0;
1906 otp->snd_recover = tp->snd_recover;
1907 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1908 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1909 otp->t_badrxtwin = 0;
1910 }
1911
1912 static int
1913 tcp_pcblist SYSCTL_HANDLER_ARGS
1914 {
1915 #pragma unused(oidp, arg1, arg2)
1916 int error, i = 0, n, sz;
1917 struct inpcb **inp_list;
1918 inp_gen_t gencnt;
1919 struct xinpgen xig;
1920
1921 /*
1922 * The process of preparing the TCB list is too time-consuming and
1923 * resource-intensive to repeat twice on every request.
1924 */
1925 lck_rw_lock_shared(&tcbinfo.ipi_lock);
1926 if (req->oldptr == USER_ADDR_NULL) {
1927 n = tcbinfo.ipi_count;
1928 req->oldidx = 2 * (sizeof(xig))
1929 + (n + n / 8) * sizeof(struct xtcpcb);
1930 lck_rw_done(&tcbinfo.ipi_lock);
1931 return 0;
1932 }
1933
1934 if (req->newptr != USER_ADDR_NULL) {
1935 lck_rw_done(&tcbinfo.ipi_lock);
1936 return EPERM;
1937 }
1938
1939 /*
1940 * OK, now we're committed to doing something.
1941 */
1942 gencnt = tcbinfo.ipi_gencnt;
1943 sz = n = tcbinfo.ipi_count;
1944
1945 bzero(&xig, sizeof(xig));
1946 xig.xig_len = sizeof(xig);
1947 xig.xig_count = n;
1948 xig.xig_gen = gencnt;
1949 xig.xig_sogen = so_gencnt;
1950 error = SYSCTL_OUT(req, &xig, sizeof(xig));
1951 if (error) {
1952 lck_rw_done(&tcbinfo.ipi_lock);
1953 return error;
1954 }
1955 /*
1956 * We are done if there is no pcb
1957 */
1958 if (n == 0) {
1959 lck_rw_done(&tcbinfo.ipi_lock);
1960 return 0;
1961 }
1962
1963 inp_list = kalloc_type(struct inpcb *, n, Z_WAITOK);
1964 if (inp_list == NULL) {
1965 lck_rw_done(&tcbinfo.ipi_lock);
1966 return ENOMEM;
1967 }
1968
1969 n = get_tcp_inp_list(inp_list, n, gencnt);
1970
1971 error = 0;
1972 for (i = 0; i < n; i++) {
1973 struct xtcpcb xt;
1974 caddr_t inp_ppcb;
1975 struct inpcb *inp;
1976
1977 inp = inp_list[i];
1978
1979 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
1980 continue;
1981 }
1982 socket_lock(inp->inp_socket, 1);
1983 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
1984 socket_unlock(inp->inp_socket, 1);
1985 continue;
1986 }
1987 if (inp->inp_gencnt > gencnt) {
1988 socket_unlock(inp->inp_socket, 1);
1989 continue;
1990 }
1991
1992 bzero(&xt, sizeof(xt));
1993 xt.xt_len = sizeof(xt);
1994 /* XXX should avoid extra copy */
1995 inpcb_to_compat(inp, &xt.xt_inp);
1996 inp_ppcb = inp->inp_ppcb;
1997 if (inp_ppcb != NULL) {
1998 tcpcb_to_otcpcb((struct tcpcb *)(void *)inp_ppcb,
1999 &xt.xt_tp);
2000 } else {
2001 bzero((char *) &xt.xt_tp, sizeof(xt.xt_tp));
2002 }
2003 if (inp->inp_socket) {
2004 sotoxsocket(inp->inp_socket, &xt.xt_socket);
2005 }
2006
2007 socket_unlock(inp->inp_socket, 1);
2008
2009 error = SYSCTL_OUT(req, &xt, sizeof(xt));
2010 }
2011 if (!error) {
2012 /*
2013 * Give the user an updated idea of our state.
2014 * If the generation differs from what we told
2015 * her before, she knows that something happened
2016 * while we were processing this request, and it
2017 * might be necessary to retry.
2018 */
2019 bzero(&xig, sizeof(xig));
2020 xig.xig_len = sizeof(xig);
2021 xig.xig_gen = tcbinfo.ipi_gencnt;
2022 xig.xig_sogen = so_gencnt;
2023 xig.xig_count = tcbinfo.ipi_count;
2024 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2025 }
2026
2027 lck_rw_done(&tcbinfo.ipi_lock);
2028 kfree_type(struct inpcb *, sz, inp_list);
2029 return error;
2030 }
2031
2032 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
2033 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2034 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
2035
2036 #if XNU_TARGET_OS_OSX
2037
2038 static void
tcpcb_to_xtcpcb64(struct tcpcb * tp,struct xtcpcb64 * otp)2039 tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp)
2040 {
2041 otp->t_segq = (uint32_t)VM_KERNEL_ADDRPERM(tp->t_segq.lh_first);
2042 otp->t_dupacks = tp->t_dupacks;
2043 otp->t_timer[TCPT_REXMT_EXT] = tp->t_timer[TCPT_REXMT];
2044 otp->t_timer[TCPT_PERSIST_EXT] = tp->t_timer[TCPT_PERSIST];
2045 otp->t_timer[TCPT_KEEP_EXT] = tp->t_timer[TCPT_KEEP];
2046 otp->t_timer[TCPT_2MSL_EXT] = tp->t_timer[TCPT_2MSL];
2047 otp->t_state = tp->t_state;
2048 otp->t_flags = tp->t_flags;
2049 otp->t_force = (tp->t_flagsext & TF_FORCE) ? 1 : 0;
2050 otp->snd_una = tp->snd_una;
2051 otp->snd_max = tp->snd_max;
2052 otp->snd_nxt = tp->snd_nxt;
2053 otp->snd_up = tp->snd_up;
2054 otp->snd_wl1 = tp->snd_wl1;
2055 otp->snd_wl2 = tp->snd_wl2;
2056 otp->iss = tp->iss;
2057 otp->irs = tp->irs;
2058 otp->rcv_nxt = tp->rcv_nxt;
2059 otp->rcv_adv = tp->rcv_adv;
2060 otp->rcv_wnd = tp->rcv_wnd;
2061 otp->rcv_up = tp->rcv_up;
2062 otp->snd_wnd = tp->snd_wnd;
2063 otp->snd_cwnd = tp->snd_cwnd;
2064 otp->snd_ssthresh = tp->snd_ssthresh;
2065 otp->t_maxopd = tp->t_maxopd;
2066 otp->t_rcvtime = tp->t_rcvtime;
2067 otp->t_starttime = tp->t_starttime;
2068 otp->t_rtttime = tp->t_rtttime;
2069 otp->t_rtseq = tp->t_rtseq;
2070 otp->t_rxtcur = tp->t_rxtcur;
2071 otp->t_maxseg = tp->t_maxseg;
2072 otp->t_srtt = tp->t_srtt;
2073 otp->t_rttvar = tp->t_rttvar;
2074 otp->t_rxtshift = tp->t_rxtshift;
2075 otp->t_rttmin = tp->t_rttmin;
2076 otp->t_rttupdated = tp->t_rttupdated;
2077 otp->max_sndwnd = tp->max_sndwnd;
2078 otp->t_softerror = tp->t_softerror;
2079 otp->t_oobflags = tp->t_oobflags;
2080 otp->t_iobc = tp->t_iobc;
2081 otp->snd_scale = tp->snd_scale;
2082 otp->rcv_scale = tp->rcv_scale;
2083 otp->request_r_scale = tp->request_r_scale;
2084 otp->requested_s_scale = tp->requested_s_scale;
2085 otp->ts_recent = tp->ts_recent;
2086 otp->ts_recent_age = tp->ts_recent_age;
2087 otp->last_ack_sent = tp->last_ack_sent;
2088 otp->cc_send = 0;
2089 otp->cc_recv = 0;
2090 otp->snd_recover = tp->snd_recover;
2091 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
2092 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
2093 otp->t_badrxtwin = 0;
2094 }
2095
2096
2097 static int
2098 tcp_pcblist64 SYSCTL_HANDLER_ARGS
2099 {
2100 #pragma unused(oidp, arg1, arg2)
2101 int error, i = 0, n, sz;
2102 struct inpcb **inp_list;
2103 inp_gen_t gencnt;
2104 struct xinpgen xig;
2105
2106 /*
2107 * The process of preparing the TCB list is too time-consuming and
2108 * resource-intensive to repeat twice on every request.
2109 */
2110 lck_rw_lock_shared(&tcbinfo.ipi_lock);
2111 if (req->oldptr == USER_ADDR_NULL) {
2112 n = tcbinfo.ipi_count;
2113 req->oldidx = 2 * (sizeof(xig))
2114 + (n + n / 8) * sizeof(struct xtcpcb64);
2115 lck_rw_done(&tcbinfo.ipi_lock);
2116 return 0;
2117 }
2118
2119 if (req->newptr != USER_ADDR_NULL) {
2120 lck_rw_done(&tcbinfo.ipi_lock);
2121 return EPERM;
2122 }
2123
2124 /*
2125 * OK, now we're committed to doing something.
2126 */
2127 gencnt = tcbinfo.ipi_gencnt;
2128 sz = n = tcbinfo.ipi_count;
2129
2130 bzero(&xig, sizeof(xig));
2131 xig.xig_len = sizeof(xig);
2132 xig.xig_count = n;
2133 xig.xig_gen = gencnt;
2134 xig.xig_sogen = so_gencnt;
2135 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2136 if (error) {
2137 lck_rw_done(&tcbinfo.ipi_lock);
2138 return error;
2139 }
2140 /*
2141 * We are done if there is no pcb
2142 */
2143 if (n == 0) {
2144 lck_rw_done(&tcbinfo.ipi_lock);
2145 return 0;
2146 }
2147
2148 inp_list = kalloc_type(struct inpcb *, n, Z_WAITOK);
2149 if (inp_list == NULL) {
2150 lck_rw_done(&tcbinfo.ipi_lock);
2151 return ENOMEM;
2152 }
2153
2154 n = get_tcp_inp_list(inp_list, n, gencnt);
2155
2156 error = 0;
2157 for (i = 0; i < n; i++) {
2158 struct xtcpcb64 xt;
2159 struct inpcb *inp;
2160
2161 inp = inp_list[i];
2162
2163 if (in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
2164 continue;
2165 }
2166 socket_lock(inp->inp_socket, 1);
2167 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
2168 socket_unlock(inp->inp_socket, 1);
2169 continue;
2170 }
2171 if (inp->inp_gencnt > gencnt) {
2172 socket_unlock(inp->inp_socket, 1);
2173 continue;
2174 }
2175
2176 bzero(&xt, sizeof(xt));
2177 xt.xt_len = sizeof(xt);
2178 inpcb_to_xinpcb64(inp, &xt.xt_inpcb);
2179 xt.xt_inpcb.inp_ppcb =
2180 (uint64_t)VM_KERNEL_ADDRPERM(inp->inp_ppcb);
2181 if (inp->inp_ppcb != NULL) {
2182 tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb,
2183 &xt);
2184 }
2185 if (inp->inp_socket) {
2186 sotoxsocket64(inp->inp_socket,
2187 &xt.xt_inpcb.xi_socket);
2188 }
2189
2190 socket_unlock(inp->inp_socket, 1);
2191
2192 error = SYSCTL_OUT(req, &xt, sizeof(xt));
2193 }
2194 if (!error) {
2195 /*
2196 * Give the user an updated idea of our state.
2197 * If the generation differs from what we told
2198 * her before, she knows that something happened
2199 * while we were processing this request, and it
2200 * might be necessary to retry.
2201 */
2202 bzero(&xig, sizeof(xig));
2203 xig.xig_len = sizeof(xig);
2204 xig.xig_gen = tcbinfo.ipi_gencnt;
2205 xig.xig_sogen = so_gencnt;
2206 xig.xig_count = tcbinfo.ipi_count;
2207 error = SYSCTL_OUT(req, &xig, sizeof(xig));
2208 }
2209
2210 lck_rw_done(&tcbinfo.ipi_lock);
2211 kfree_type(struct inpcb *, sz, inp_list);
2212 return error;
2213 }
2214
2215 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64,
2216 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2217 tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections");
2218
2219 #endif /* XNU_TARGET_OS_OSX */
2220
2221 static int
2222 tcp_pcblist_n SYSCTL_HANDLER_ARGS
2223 {
2224 #pragma unused(oidp, arg1, arg2)
2225 int error = 0;
2226
2227 error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo);
2228
2229 return error;
2230 }
2231
2232
2233 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n,
2234 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
2235 tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections");
2236
2237 static int
2238 tcp_progress_indicators SYSCTL_HANDLER_ARGS
2239 {
2240 #pragma unused(oidp, arg1, arg2)
2241
2242 return ntstat_tcp_progress_indicators(req);
2243 }
2244
2245 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, progress,
2246 CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, 0,
2247 tcp_progress_indicators, "S", "Various items that indicate the current state of progress on the link");
2248
2249
2250 static int
2251 tcp_progress_probe_enable SYSCTL_HANDLER_ARGS
2252 {
2253 #pragma unused(oidp, arg1, arg2)
2254
2255 return ntstat_tcp_progress_enable(req);
2256 }
2257
2258 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, progress_enable,
2259 CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_LOCKED | CTLFLAG_ANYBODY, 0, 0,
2260 tcp_progress_probe_enable, "S", "Enable/disable TCP keepalive probing on the specified link(s)");
2261
2262
2263 __private_extern__ void
tcp_get_ports_used(ifnet_t ifp,int protocol,uint32_t flags,bitstr_t * bitfield)2264 tcp_get_ports_used(ifnet_t ifp, int protocol, uint32_t flags,
2265 bitstr_t *bitfield)
2266 {
2267 inpcb_get_ports_used(ifp, protocol, flags, bitfield,
2268 &tcbinfo);
2269 }
2270
2271 __private_extern__ uint32_t
tcp_count_opportunistic(unsigned int ifindex,u_int32_t flags)2272 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags)
2273 {
2274 return inpcb_count_opportunistic(ifindex, &tcbinfo, flags);
2275 }
2276
2277 __private_extern__ uint32_t
tcp_find_anypcb_byaddr(struct ifaddr * ifa)2278 tcp_find_anypcb_byaddr(struct ifaddr *ifa)
2279 {
2280 #if SKYWALK
2281 if (netns_is_enabled()) {
2282 return netns_find_anyres_byaddr(ifa, IPPROTO_TCP);
2283 } else
2284 #endif /* SKYWALK */
2285 return inpcb_find_anypcb_byaddr(ifa, &tcbinfo);
2286 }
2287
2288 static void
tcp_handle_msgsize(struct ip * ip,struct inpcb * inp)2289 tcp_handle_msgsize(struct ip *ip, struct inpcb *inp)
2290 {
2291 struct rtentry *rt = NULL;
2292 u_short ifscope = IFSCOPE_NONE;
2293 int mtu;
2294 struct sockaddr_in icmpsrc = {
2295 .sin_len = sizeof(struct sockaddr_in),
2296 .sin_family = AF_INET, .sin_port = 0, .sin_addr = { .s_addr = 0 },
2297 .sin_zero = { 0, 0, 0, 0, 0, 0, 0, 0 }
2298 };
2299 struct icmp *icp = NULL;
2300
2301 icp = (struct icmp *)(void *)
2302 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
2303
2304 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
2305
2306 /*
2307 * MTU discovery:
2308 * If we got a needfrag and there is a host route to the
2309 * original destination, and the MTU is not locked, then
2310 * set the MTU in the route to the suggested new value
2311 * (if given) and then notify as usual. The ULPs will
2312 * notice that the MTU has changed and adapt accordingly.
2313 * If no new MTU was suggested, then we guess a new one
2314 * less than the current value. If the new MTU is
2315 * unreasonably small (defined by sysctl tcp_minmss), then
2316 * we reset the MTU to the interface value and enable the
2317 * lock bit, indicating that we are no longer doing MTU
2318 * discovery.
2319 */
2320 if (ROUTE_UNUSABLE(&(inp->inp_route)) == false) {
2321 rt = inp->inp_route.ro_rt;
2322 }
2323
2324 /*
2325 * icmp6_mtudisc_update scopes the routing lookup
2326 * to the incoming interface (delivered from mbuf
2327 * packet header.
2328 * That is mostly ok but for asymmetric networks
2329 * that may be an issue.
2330 * Frag needed OR Packet too big really communicates
2331 * MTU for the out data path.
2332 * Take the interface scope from cached route or
2333 * the last outgoing interface from inp
2334 */
2335 if (rt != NULL) {
2336 ifscope = (rt->rt_ifp != NULL) ?
2337 rt->rt_ifp->if_index : IFSCOPE_NONE;
2338 } else {
2339 ifscope = (inp->inp_last_outifp != NULL) ?
2340 inp->inp_last_outifp->if_index : IFSCOPE_NONE;
2341 }
2342
2343 if ((rt == NULL) ||
2344 !(rt->rt_flags & RTF_HOST) ||
2345 (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING))) {
2346 rt = rtalloc1_scoped((struct sockaddr *)&icmpsrc, 0,
2347 RTF_CLONING | RTF_PRCLONING, ifscope);
2348 } else if (rt) {
2349 RT_LOCK(rt);
2350 rtref(rt);
2351 RT_UNLOCK(rt);
2352 }
2353
2354 if (rt != NULL) {
2355 RT_LOCK(rt);
2356 if ((rt->rt_flags & RTF_HOST) &&
2357 !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
2358 mtu = ntohs(icp->icmp_nextmtu);
2359 /*
2360 * XXX Stock BSD has changed the following
2361 * to compare with icp->icmp_ip.ip_len
2362 * to converge faster when sent packet
2363 * < route's MTU. We may want to adopt
2364 * that change.
2365 */
2366 if (mtu == 0) {
2367 mtu = ip_next_mtu(rt->rt_rmx.
2368 rmx_mtu, 1);
2369 }
2370 #if DEBUG_MTUDISC
2371 printf("MTU for %s reduced to %d\n",
2372 inet_ntop(AF_INET,
2373 &icmpsrc.sin_addr, ipv4str,
2374 sizeof(ipv4str)), mtu);
2375 #endif
2376 if (mtu < max(296, (tcp_minmss +
2377 sizeof(struct tcpiphdr)))) {
2378 rt->rt_rmx.rmx_locks |= RTV_MTU;
2379 } else if (rt->rt_rmx.rmx_mtu > mtu) {
2380 rt->rt_rmx.rmx_mtu = mtu;
2381 }
2382 }
2383 RT_UNLOCK(rt);
2384 rtfree(rt);
2385 }
2386 }
2387
2388 void
tcp_ctlinput(int cmd,struct sockaddr * sa,void * vip,__unused struct ifnet * ifp)2389 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip, __unused struct ifnet *ifp)
2390 {
2391 tcp_seq icmp_tcp_seq;
2392 struct ipctlparam *ctl_param = vip;
2393 struct ip *ip = NULL;
2394 struct mbuf *m = NULL;
2395 struct in_addr faddr;
2396 struct inpcb *inp;
2397 struct tcpcb *tp;
2398 struct tcphdr *th;
2399 struct icmp *icp;
2400 size_t off;
2401 #if SKYWALK
2402 union sockaddr_in_4_6 sock_laddr;
2403 struct protoctl_ev_val prctl_ev_val;
2404 #endif /* SKYWALK */
2405 void (*notify)(struct inpcb *, int) = tcp_notify;
2406
2407 if (ctl_param != NULL) {
2408 ip = ctl_param->ipc_icmp_ip;
2409 icp = ctl_param->ipc_icmp;
2410 m = ctl_param->ipc_m;
2411 off = ctl_param->ipc_off;
2412 } else {
2413 ip = NULL;
2414 icp = NULL;
2415 m = NULL;
2416 off = 0;
2417 }
2418
2419 faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr;
2420 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY) {
2421 return;
2422 }
2423
2424 if ((unsigned)cmd >= PRC_NCMDS) {
2425 return;
2426 }
2427
2428 /* Source quench is deprecated */
2429 if (cmd == PRC_QUENCH) {
2430 return;
2431 }
2432
2433 if (cmd == PRC_MSGSIZE) {
2434 notify = tcp_mtudisc;
2435 } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2436 cmd == PRC_UNREACH_PORT || cmd == PRC_UNREACH_PROTOCOL ||
2437 cmd == PRC_TIMXCEED_INTRANS) && ip) {
2438 notify = tcp_drop_syn_sent;
2439 }
2440 /*
2441 * Hostdead is ugly because it goes linearly through all PCBs.
2442 * XXX: We never get this from ICMP, otherwise it makes an
2443 * excellent DoS attack on machines with many connections.
2444 */
2445 else if (cmd == PRC_HOSTDEAD) {
2446 ip = NULL;
2447 } else if (inetctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) {
2448 return;
2449 }
2450
2451 #if SKYWALK
2452 bzero(&prctl_ev_val, sizeof(prctl_ev_val));
2453 bzero(&sock_laddr, sizeof(sock_laddr));
2454 #endif /* SKYWALK */
2455
2456 if (ip == NULL) {
2457 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
2458 #if SKYWALK
2459 protoctl_event_enqueue_nwk_wq_entry(ifp, NULL,
2460 sa, 0, 0, IPPROTO_TCP, cmd, NULL);
2461 #endif /* SKYWALK */
2462 return;
2463 }
2464
2465 /* Check if we can safely get the sport, dport and the sequence number from the tcp header. */
2466 if (m == NULL ||
2467 (m->m_len < off + (sizeof(unsigned short) + sizeof(unsigned short) + sizeof(tcp_seq)))) {
2468 /* Insufficient length */
2469 return;
2470 }
2471
2472 th = (struct tcphdr*)(void*)(mtod(m, uint8_t*) + off);
2473 icmp_tcp_seq = ntohl(th->th_seq);
2474
2475 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
2476 ip->ip_src, th->th_sport, 0, NULL);
2477
2478 if (inp == NULL ||
2479 inp->inp_socket == NULL) {
2480 #if SKYWALK
2481 if (cmd == PRC_MSGSIZE) {
2482 prctl_ev_val.val = ntohs(icp->icmp_nextmtu);
2483 }
2484 prctl_ev_val.tcp_seq_number = icmp_tcp_seq;
2485
2486 sock_laddr.sin.sin_family = AF_INET;
2487 sock_laddr.sin.sin_len = sizeof(sock_laddr.sin);
2488 sock_laddr.sin.sin_addr = ip->ip_src;
2489
2490 protoctl_event_enqueue_nwk_wq_entry(ifp,
2491 (struct sockaddr *)&sock_laddr, sa,
2492 th->th_sport, th->th_dport, IPPROTO_TCP,
2493 cmd, &prctl_ev_val);
2494 #endif /* SKYWALK */
2495 return;
2496 }
2497
2498 socket_lock(inp->inp_socket, 1);
2499 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) ==
2500 WNT_STOPUSING) {
2501 socket_unlock(inp->inp_socket, 1);
2502 return;
2503 }
2504
2505 if (PRC_IS_REDIRECT(cmd)) {
2506 /* signal EHOSTDOWN, as it flushes the cached route */
2507 (*notify)(inp, EHOSTDOWN);
2508 } else {
2509 tp = intotcpcb(inp);
2510 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2511 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2512 if (cmd == PRC_MSGSIZE) {
2513 tcp_handle_msgsize(ip, inp);
2514 }
2515
2516 (*notify)(inp, inetctlerrmap[cmd]);
2517 }
2518 }
2519 socket_unlock(inp->inp_socket, 1);
2520 }
2521
2522 void
tcp6_ctlinput(int cmd,struct sockaddr * sa,void * d,__unused struct ifnet * ifp)2523 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d, __unused struct ifnet *ifp)
2524 {
2525 tcp_seq icmp_tcp_seq;
2526 struct in6_addr *dst;
2527 void (*notify)(struct inpcb *, int) = tcp_notify;
2528 struct ip6_hdr *ip6;
2529 struct mbuf *m;
2530 struct inpcb *inp;
2531 struct tcpcb *tp;
2532 struct icmp6_hdr *icmp6;
2533 struct ip6ctlparam *ip6cp = NULL;
2534 const struct sockaddr_in6 *sa6_src = NULL;
2535 unsigned int mtu;
2536 unsigned int off;
2537
2538 struct tcp_ports {
2539 uint16_t th_sport;
2540 uint16_t th_dport;
2541 } t_ports;
2542 #if SKYWALK
2543 union sockaddr_in_4_6 sock_laddr;
2544 struct protoctl_ev_val prctl_ev_val;
2545 #endif /* SKYWALK */
2546
2547 if (sa->sa_family != AF_INET6 ||
2548 sa->sa_len != sizeof(struct sockaddr_in6)) {
2549 return;
2550 }
2551
2552 /* Source quench is deprecated */
2553 if (cmd == PRC_QUENCH) {
2554 return;
2555 }
2556
2557 if ((unsigned)cmd >= PRC_NCMDS) {
2558 return;
2559 }
2560
2561 /* if the parameter is from icmp6, decode it. */
2562 if (d != NULL) {
2563 ip6cp = (struct ip6ctlparam *)d;
2564 icmp6 = ip6cp->ip6c_icmp6;
2565 m = ip6cp->ip6c_m;
2566 ip6 = ip6cp->ip6c_ip6;
2567 off = ip6cp->ip6c_off;
2568 sa6_src = ip6cp->ip6c_src;
2569 dst = ip6cp->ip6c_finaldst;
2570 } else {
2571 m = NULL;
2572 ip6 = NULL;
2573 off = 0; /* fool gcc */
2574 sa6_src = &sa6_any;
2575 dst = NULL;
2576 }
2577
2578 if (cmd == PRC_MSGSIZE) {
2579 notify = tcp_mtudisc;
2580 } else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
2581 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) &&
2582 ip6 != NULL) {
2583 notify = tcp_drop_syn_sent;
2584 }
2585 /*
2586 * Hostdead is ugly because it goes linearly through all PCBs.
2587 * XXX: We never get this from ICMP, otherwise it makes an
2588 * excellent DoS attack on machines with many connections.
2589 */
2590 else if (cmd == PRC_HOSTDEAD) {
2591 ip6 = NULL;
2592 } else if (inet6ctlerrmap[cmd] == 0 && !PRC_IS_REDIRECT(cmd)) {
2593 return;
2594 }
2595
2596 #if SKYWALK
2597 bzero(&prctl_ev_val, sizeof(prctl_ev_val));
2598 bzero(&sock_laddr, sizeof(sock_laddr));
2599 #endif /* SKYWALK */
2600
2601 if (ip6 == NULL) {
2602 in6_pcbnotify(&tcbinfo, sa, 0, (struct sockaddr *)(size_t)sa6_src,
2603 0, cmd, NULL, notify);
2604 #if SKYWALK
2605 protoctl_event_enqueue_nwk_wq_entry(ifp, NULL, sa,
2606 0, 0, IPPROTO_TCP, cmd, NULL);
2607 #endif /* SKYWALK */
2608 return;
2609 }
2610
2611 /* Check if we can safely get the ports from the tcp hdr */
2612 if (m == NULL ||
2613 (m->m_pkthdr.len <
2614 (int32_t) (off + sizeof(struct tcp_ports)))) {
2615 return;
2616 }
2617 bzero(&t_ports, sizeof(struct tcp_ports));
2618 m_copydata(m, off, sizeof(struct tcp_ports), (caddr_t)&t_ports);
2619
2620 off += sizeof(struct tcp_ports);
2621 if (m->m_pkthdr.len < (int32_t) (off + sizeof(tcp_seq))) {
2622 return;
2623 }
2624 m_copydata(m, off, sizeof(tcp_seq), (caddr_t)&icmp_tcp_seq);
2625 icmp_tcp_seq = ntohl(icmp_tcp_seq);
2626
2627 if (cmd == PRC_MSGSIZE) {
2628 mtu = ntohl(icmp6->icmp6_mtu);
2629 /*
2630 * If no alternative MTU was proposed, or the proposed
2631 * MTU was too small, set to the min.
2632 */
2633 if (mtu < IPV6_MMTU) {
2634 mtu = IPV6_MMTU - 8;
2635 }
2636 }
2637
2638 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_dst, t_ports.th_dport, ip6_input_getdstifscope(m),
2639 &ip6->ip6_src, t_ports.th_sport, ip6_input_getsrcifscope(m), 0, NULL);
2640
2641 if (inp == NULL ||
2642 inp->inp_socket == NULL) {
2643 #if SKYWALK
2644 if (cmd == PRC_MSGSIZE) {
2645 prctl_ev_val.val = mtu;
2646 }
2647 prctl_ev_val.tcp_seq_number = icmp_tcp_seq;
2648
2649 sock_laddr.sin6.sin6_family = AF_INET6;
2650 sock_laddr.sin6.sin6_len = sizeof(sock_laddr.sin6);
2651 sock_laddr.sin6.sin6_addr = ip6->ip6_src;
2652
2653 protoctl_event_enqueue_nwk_wq_entry(ifp,
2654 (struct sockaddr *)&sock_laddr, sa,
2655 t_ports.th_sport, t_ports.th_dport, IPPROTO_TCP,
2656 cmd, &prctl_ev_val);
2657 #endif /* SKYWALK */
2658 return;
2659 }
2660
2661 socket_lock(inp->inp_socket, 1);
2662 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) ==
2663 WNT_STOPUSING) {
2664 socket_unlock(inp->inp_socket, 1);
2665 return;
2666 }
2667
2668 if (PRC_IS_REDIRECT(cmd)) {
2669 /* signal EHOSTDOWN, as it flushes the cached route */
2670 (*notify)(inp, EHOSTDOWN);
2671 } else {
2672 tp = intotcpcb(inp);
2673 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
2674 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
2675 if (cmd == PRC_MSGSIZE) {
2676 /*
2677 * Only process the offered MTU if it
2678 * is smaller than the current one.
2679 */
2680 if (mtu < tp->t_maxseg +
2681 (sizeof(struct tcphdr) + sizeof(struct ip6_hdr))) {
2682 (*notify)(inp, inetctlerrmap[cmd]);
2683 }
2684 } else {
2685 (*notify)(inp, inetctlerrmap[cmd]);
2686 }
2687 }
2688 }
2689 socket_unlock(inp->inp_socket, 1);
2690 }
2691
2692
2693 /*
2694 * Following is where TCP initial sequence number generation occurs.
2695 *
2696 * There are two places where we must use initial sequence numbers:
2697 * 1. In SYN-ACK packets.
2698 * 2. In SYN packets.
2699 *
2700 * The ISNs in SYN-ACK packets have no monotonicity requirement,
2701 * and should be as unpredictable as possible to avoid the possibility
2702 * of spoofing and/or connection hijacking. To satisfy this
2703 * requirement, SYN-ACK ISNs are generated via the arc4random()
2704 * function. If exact RFC 1948 compliance is requested via sysctl,
2705 * these ISNs will be generated just like those in SYN packets.
2706 *
2707 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
2708 * depends on this property. In addition, these ISNs should be
2709 * unguessable so as to prevent connection hijacking. To satisfy
2710 * the requirements of this situation, the algorithm outlined in
2711 * RFC 1948 is used to generate sequence numbers.
2712 *
2713 * For more information on the theory of operation, please see
2714 * RFC 1948.
2715 *
2716 * Implementation details:
2717 *
2718 * Time is based off the system timer, and is corrected so that it
2719 * increases by one megabyte per second. This allows for proper
2720 * recycling on high speed LANs while still leaving over an hour
2721 * before rollover.
2722 *
2723 * Two sysctls control the generation of ISNs:
2724 *
2725 * net.inet.tcp.isn_reseed_interval controls the number of seconds
2726 * between seeding of isn_secret. This is normally set to zero,
2727 * as reseeding should not be necessary.
2728 *
2729 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
2730 * strictly. When strict compliance is requested, reseeding is
2731 * disabled and SYN-ACKs will be generated in the same manner as
2732 * SYNs. Strict mode is disabled by default.
2733 *
2734 */
2735
2736 #define ISN_BYTES_PER_SECOND 1048576
2737
2738 tcp_seq
tcp_new_isn(struct tcpcb * tp)2739 tcp_new_isn(struct tcpcb *tp)
2740 {
2741 u_int32_t md5_buffer[4];
2742 tcp_seq new_isn;
2743 struct timeval timenow;
2744 u_char isn_secret[32];
2745 long isn_last_reseed = 0;
2746 MD5_CTX isn_ctx;
2747
2748 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
2749 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT)) &&
2750 tcp_strict_rfc1948 == 0)
2751 #ifdef __APPLE__
2752 { return RandomULong(); }
2753 #else
2754 { return arc4random(); }
2755 #endif
2756 getmicrotime(&timenow);
2757
2758 /* Seed if this is the first use, reseed if requested. */
2759 if ((isn_last_reseed == 0) ||
2760 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
2761 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval * hz)
2762 < (u_int)timenow.tv_sec))) {
2763 #ifdef __APPLE__
2764 read_frandom(&isn_secret, sizeof(isn_secret));
2765 #else
2766 read_random_unlimited(&isn_secret, sizeof(isn_secret));
2767 #endif
2768 isn_last_reseed = timenow.tv_sec;
2769 }
2770
2771 /* Compute the md5 hash and return the ISN. */
2772 MD5Init(&isn_ctx);
2773 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport,
2774 sizeof(u_short));
2775 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport,
2776 sizeof(u_short));
2777 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
2778 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
2779 sizeof(struct in6_addr));
2780 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
2781 sizeof(struct in6_addr));
2782 } else {
2783 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
2784 sizeof(struct in_addr));
2785 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
2786 sizeof(struct in_addr));
2787 }
2788 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
2789 MD5Final((u_char *) &md5_buffer, &isn_ctx);
2790 new_isn = (tcp_seq) md5_buffer[0];
2791 new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz);
2792 return new_isn;
2793 }
2794
2795
2796 /*
2797 * When a specific ICMP unreachable message is received and the
2798 * connection state is SYN-SENT, drop the connection. This behavior
2799 * is controlled by the icmp_may_rst sysctl.
2800 */
2801 void
tcp_drop_syn_sent(struct inpcb * inp,int errno)2802 tcp_drop_syn_sent(struct inpcb *inp, int errno)
2803 {
2804 struct tcpcb *tp = intotcpcb(inp);
2805
2806 if (tp && tp->t_state == TCPS_SYN_SENT) {
2807 tcp_drop(tp, errno);
2808 }
2809 }
2810
2811 /*
2812 * When `need fragmentation' ICMP is received, update our idea of the MSS
2813 * based on the new value in the route. Also nudge TCP to send something,
2814 * since we know the packet we just sent was dropped.
2815 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2816 */
2817 void
tcp_mtudisc(struct inpcb * inp,__unused int errno)2818 tcp_mtudisc(struct inpcb *inp, __unused int errno)
2819 {
2820 struct tcpcb *tp = intotcpcb(inp);
2821 struct rtentry *rt;
2822 struct socket *so = inp->inp_socket;
2823 int mss;
2824 u_int32_t mtu;
2825 u_int32_t protoHdrOverhead = sizeof(struct tcpiphdr);
2826 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
2827
2828 /*
2829 * Nothing left to send after the socket is defunct or TCP is in the closed state
2830 */
2831 if ((so->so_state & SS_DEFUNCT) || (tp != NULL && tp->t_state == TCPS_CLOSED)) {
2832 return;
2833 }
2834
2835 if (isipv6) {
2836 protoHdrOverhead = sizeof(struct ip6_hdr) +
2837 sizeof(struct tcphdr);
2838 }
2839
2840 if (tp != NULL) {
2841 if (isipv6) {
2842 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2843 } else {
2844 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2845 }
2846 if (!rt || !rt->rt_rmx.rmx_mtu) {
2847 tp->t_maxopd = tp->t_maxseg =
2848 isipv6 ? tcp_v6mssdflt :
2849 tcp_mssdflt;
2850
2851 /* Route locked during lookup above */
2852 if (rt != NULL) {
2853 RT_UNLOCK(rt);
2854 }
2855 return;
2856 }
2857 mtu = rt->rt_rmx.rmx_mtu;
2858
2859 /* Route locked during lookup above */
2860 RT_UNLOCK(rt);
2861
2862 #if NECP
2863 // Adjust MTU if necessary.
2864 mtu = necp_socket_get_effective_mtu(inp, mtu);
2865 #endif /* NECP */
2866 mss = mtu - protoHdrOverhead;
2867
2868 if (tp->t_maxopd) {
2869 mss = min(mss, tp->t_maxopd);
2870 }
2871 /*
2872 * XXX - The above conditional probably violates the TCP
2873 * spec. The problem is that, since we don't know the
2874 * other end's MSS, we are supposed to use a conservative
2875 * default. But, if we do that, then MTU discovery will
2876 * never actually take place, because the conservative
2877 * default is much less than the MTUs typically seen
2878 * on the Internet today. For the moment, we'll sweep
2879 * this under the carpet.
2880 *
2881 * The conservative default might not actually be a problem
2882 * if the only case this occurs is when sending an initial
2883 * SYN with options and data to a host we've never talked
2884 * to before. Then, they will reply with an MSS value which
2885 * will get recorded and the new parameters should get
2886 * recomputed. For Further Study.
2887 */
2888 if (tp->t_maxopd <= mss) {
2889 return;
2890 }
2891 tp->t_maxopd = mss;
2892
2893 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2894 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) {
2895 mss -= TCPOLEN_TSTAMP_APPA;
2896 }
2897
2898 #if MPTCP
2899 mss -= mptcp_adj_mss(tp, TRUE);
2900 #endif
2901 if (so->so_snd.sb_hiwat < mss) {
2902 mss = so->so_snd.sb_hiwat;
2903 }
2904
2905 tp->t_maxseg = mss;
2906
2907 ASSERT(tp->t_maxseg);
2908
2909 /*
2910 * Reset the slow-start flight size as it may depends on the
2911 * new MSS
2912 */
2913 if (CC_ALGO(tp)->cwnd_init != NULL) {
2914 CC_ALGO(tp)->cwnd_init(tp);
2915 }
2916
2917 if (TCP_USE_RLEDBAT(tp, so) && tcp_cc_rledbat.rwnd_init != NULL) {
2918 tcp_cc_rledbat.rwnd_init(tp);
2919 }
2920
2921 tcpstat.tcps_mturesent++;
2922 tp->t_rtttime = 0;
2923 tp->snd_nxt = tp->snd_una;
2924 tcp_output(tp);
2925 }
2926 }
2927
2928 /*
2929 * Look-up the routing entry to the peer of this inpcb. If no route
2930 * is found and it cannot be allocated the return NULL. This routine
2931 * is called by TCP routines that access the rmx structure and by tcp_mss
2932 * to get the interface MTU. If a route is found, this routine will
2933 * hold the rtentry lock; the caller is responsible for unlocking.
2934 */
2935 struct rtentry *
tcp_rtlookup(struct inpcb * inp,unsigned int input_ifscope)2936 tcp_rtlookup(struct inpcb *inp, unsigned int input_ifscope)
2937 {
2938 struct route *ro;
2939 struct rtentry *rt;
2940 struct tcpcb *tp;
2941
2942 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2943
2944 ro = &inp->inp_route;
2945 if ((rt = ro->ro_rt) != NULL) {
2946 RT_LOCK(rt);
2947 }
2948
2949 if (ROUTE_UNUSABLE(ro)) {
2950 if (rt != NULL) {
2951 RT_UNLOCK(rt);
2952 rt = NULL;
2953 }
2954 ROUTE_RELEASE(ro);
2955 /* No route yet, so try to acquire one */
2956 if (inp->inp_faddr.s_addr != INADDR_ANY) {
2957 unsigned int ifscope;
2958
2959 ro->ro_dst.sa_family = AF_INET;
2960 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
2961 ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr =
2962 inp->inp_faddr;
2963
2964 /*
2965 * If the socket was bound to an interface, then
2966 * the bound-to-interface takes precedence over
2967 * the inbound interface passed in by the caller
2968 * (if we get here as part of the output path then
2969 * input_ifscope is IFSCOPE_NONE).
2970 */
2971 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2972 inp->inp_boundifp->if_index : input_ifscope;
2973
2974 rtalloc_scoped(ro, ifscope);
2975 if ((rt = ro->ro_rt) != NULL) {
2976 RT_LOCK(rt);
2977 }
2978 }
2979 }
2980 if (rt != NULL) {
2981 RT_LOCK_ASSERT_HELD(rt);
2982 }
2983
2984 /*
2985 * Update MTU discovery determination. Don't do it if:
2986 * 1) it is disabled via the sysctl
2987 * 2) the route isn't up
2988 * 3) the MTU is locked (if it is, then discovery has been
2989 * disabled)
2990 */
2991
2992 tp = intotcpcb(inp);
2993
2994 if (!path_mtu_discovery || ((rt != NULL) &&
2995 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) {
2996 tp->t_flags &= ~TF_PMTUD;
2997 } else {
2998 tp->t_flags |= TF_PMTUD;
2999 }
3000
3001 if (rt != NULL && rt->rt_ifp != NULL) {
3002 somultipages(inp->inp_socket,
3003 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
3004 tcp_set_tso(tp, rt->rt_ifp);
3005 soif2kcl(inp->inp_socket,
3006 (rt->rt_ifp->if_eflags & IFEF_2KCL));
3007 tcp_set_ecn(tp, rt->rt_ifp);
3008 if (inp->inp_last_outifp == NULL) {
3009 inp->inp_last_outifp = rt->rt_ifp;
3010 #if SKYWALK
3011 if (NETNS_TOKEN_VALID(&inp->inp_netns_token)) {
3012 netns_set_ifnet(&inp->inp_netns_token,
3013 inp->inp_last_outifp);
3014 }
3015 #endif /* SKYWALK */
3016 }
3017 }
3018
3019 /* Note if the peer is local */
3020 if (rt != NULL && !(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
3021 (rt->rt_gateway->sa_family == AF_LINK ||
3022 rt->rt_ifp->if_flags & IFF_LOOPBACK ||
3023 in_localaddr(inp->inp_faddr))) {
3024 tp->t_flags |= TF_LOCAL;
3025 }
3026
3027 /*
3028 * Caller needs to call RT_UNLOCK(rt).
3029 */
3030 return rt;
3031 }
3032
3033 struct rtentry *
tcp_rtlookup6(struct inpcb * inp,unsigned int input_ifscope)3034 tcp_rtlookup6(struct inpcb *inp, unsigned int input_ifscope)
3035 {
3036 struct route_in6 *ro6;
3037 struct rtentry *rt;
3038 struct tcpcb *tp;
3039
3040 LCK_MTX_ASSERT(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
3041
3042 ro6 = &inp->in6p_route;
3043 if ((rt = ro6->ro_rt) != NULL) {
3044 RT_LOCK(rt);
3045 }
3046
3047 if (ROUTE_UNUSABLE(ro6)) {
3048 if (rt != NULL) {
3049 RT_UNLOCK(rt);
3050 rt = NULL;
3051 }
3052 ROUTE_RELEASE(ro6);
3053 /* No route yet, so try to acquire one */
3054 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
3055 struct sockaddr_in6 *dst6;
3056 unsigned int ifscope;
3057
3058 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst;
3059 dst6->sin6_family = AF_INET6;
3060 dst6->sin6_len = sizeof(*dst6);
3061 dst6->sin6_addr = inp->in6p_faddr;
3062
3063 /*
3064 * If the socket was bound to an interface, then
3065 * the bound-to-interface takes precedence over
3066 * the inbound interface passed in by the caller
3067 * (if we get here as part of the output path then
3068 * input_ifscope is IFSCOPE_NONE).
3069 */
3070 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
3071 inp->inp_boundifp->if_index : input_ifscope;
3072
3073 rtalloc_scoped((struct route *)ro6, ifscope);
3074 if ((rt = ro6->ro_rt) != NULL) {
3075 RT_LOCK(rt);
3076 }
3077 }
3078 }
3079 if (rt != NULL) {
3080 RT_LOCK_ASSERT_HELD(rt);
3081 }
3082
3083 /*
3084 * Update path MTU Discovery determination
3085 * while looking up the route:
3086 * 1) we have a valid route to the destination
3087 * 2) the MTU is not locked (if it is, then discovery has been
3088 * disabled)
3089 */
3090
3091
3092 tp = intotcpcb(inp);
3093
3094 /*
3095 * Update MTU discovery determination. Don't do it if:
3096 * 1) it is disabled via the sysctl
3097 * 2) the route isn't up
3098 * 3) the MTU is locked (if it is, then discovery has been
3099 * disabled)
3100 */
3101
3102 if (!path_mtu_discovery || ((rt != NULL) &&
3103 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU)))) {
3104 tp->t_flags &= ~TF_PMTUD;
3105 } else {
3106 tp->t_flags |= TF_PMTUD;
3107 }
3108
3109 if (rt != NULL && rt->rt_ifp != NULL) {
3110 somultipages(inp->inp_socket,
3111 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
3112 tcp_set_tso(tp, rt->rt_ifp);
3113 soif2kcl(inp->inp_socket,
3114 (rt->rt_ifp->if_eflags & IFEF_2KCL));
3115 tcp_set_ecn(tp, rt->rt_ifp);
3116 if (inp->inp_last_outifp == NULL) {
3117 inp->inp_last_outifp = rt->rt_ifp;
3118 #if SKYWALK
3119 if (NETNS_TOKEN_VALID(&inp->inp_netns_token)) {
3120 netns_set_ifnet(&inp->inp_netns_token,
3121 inp->inp_last_outifp);
3122 }
3123 #endif /* SKYWALK */
3124 }
3125
3126 /* Note if the peer is local */
3127 if (!(rt->rt_ifp->if_flags & IFF_POINTOPOINT) &&
3128 (IN6_IS_ADDR_LOOPBACK(&inp->in6p_faddr) ||
3129 IN6_IS_ADDR_LINKLOCAL(&inp->in6p_faddr) ||
3130 rt->rt_gateway->sa_family == AF_LINK ||
3131 in6_localaddr(&inp->in6p_faddr))) {
3132 tp->t_flags |= TF_LOCAL;
3133 }
3134 }
3135
3136 /*
3137 * Caller needs to call RT_UNLOCK(rt).
3138 */
3139 return rt;
3140 }
3141
3142 #if IPSEC
3143 /* compute ESP/AH header size for TCP, including outer IP header. */
3144 size_t
ipsec_hdrsiz_tcp(struct tcpcb * tp)3145 ipsec_hdrsiz_tcp(struct tcpcb *tp)
3146 {
3147 struct inpcb *inp;
3148 struct mbuf *m;
3149 size_t hdrsiz;
3150 struct ip *ip;
3151 struct ip6_hdr *ip6 = NULL;
3152 struct tcphdr *th;
3153
3154 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL)) {
3155 return 0;
3156 }
3157 MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */
3158 if (!m) {
3159 return 0;
3160 }
3161
3162 if ((inp->inp_vflag & INP_IPV6) != 0) {
3163 ip6 = mtod(m, struct ip6_hdr *);
3164 th = (struct tcphdr *)(void *)(ip6 + 1);
3165 m->m_pkthdr.len = m->m_len =
3166 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3167 tcp_fillheaders(m, tp, ip6, th);
3168 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
3169 } else {
3170 ip = mtod(m, struct ip *);
3171 th = (struct tcphdr *)(ip + 1);
3172 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
3173 tcp_fillheaders(m, tp, ip, th);
3174 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
3175 }
3176 m_free(m);
3177 return hdrsiz;
3178 }
3179 #endif /* IPSEC */
3180
3181 int
tcp_lock(struct socket * so,int refcount,void * lr)3182 tcp_lock(struct socket *so, int refcount, void *lr)
3183 {
3184 void *lr_saved;
3185
3186 if (lr == NULL) {
3187 lr_saved = __builtin_return_address(0);
3188 } else {
3189 lr_saved = lr;
3190 }
3191
3192 retry:
3193 if (so->so_pcb != NULL) {
3194 if (so->so_flags & SOF_MP_SUBFLOW) {
3195 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3196 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3197
3198 socket_lock(mp_so, refcount);
3199
3200 /*
3201 * Check if we became non-MPTCP while waiting for the lock.
3202 * If yes, we have to retry to grab the right lock.
3203 */
3204 if (!(so->so_flags & SOF_MP_SUBFLOW)) {
3205 socket_unlock(mp_so, refcount);
3206 goto retry;
3207 }
3208 } else {
3209 lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3210
3211 if (so->so_flags & SOF_MP_SUBFLOW) {
3212 /*
3213 * While waiting for the lock, we might have
3214 * become MPTCP-enabled (see mptcp_subflow_socreate).
3215 */
3216 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3217 goto retry;
3218 }
3219 }
3220 } else {
3221 panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s",
3222 so, lr_saved, solockhistory_nr(so));
3223 /* NOTREACHED */
3224 }
3225
3226 if (so->so_usecount < 0) {
3227 panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s",
3228 so, so->so_pcb, lr_saved, so->so_usecount,
3229 solockhistory_nr(so));
3230 /* NOTREACHED */
3231 }
3232 if (refcount) {
3233 so->so_usecount++;
3234 }
3235 so->lock_lr[so->next_lock_lr] = lr_saved;
3236 so->next_lock_lr = (so->next_lock_lr + 1) % SO_LCKDBG_MAX;
3237 return 0;
3238 }
3239
3240 int
tcp_unlock(struct socket * so,int refcount,void * lr)3241 tcp_unlock(struct socket *so, int refcount, void *lr)
3242 {
3243 void *lr_saved;
3244
3245 if (lr == NULL) {
3246 lr_saved = __builtin_return_address(0);
3247 } else {
3248 lr_saved = lr;
3249 }
3250
3251 #ifdef MORE_TCPLOCK_DEBUG
3252 printf("tcp_unlock: so=0x%llx sopcb=0x%llx lock=0x%llx ref=%x "
3253 "lr=0x%llx\n", (uint64_t)VM_KERNEL_ADDRPERM(so),
3254 (uint64_t)VM_KERNEL_ADDRPERM(so->so_pcb),
3255 (uint64_t)VM_KERNEL_ADDRPERM(&(sotoinpcb(so)->inpcb_mtx)),
3256 so->so_usecount, (uint64_t)VM_KERNEL_ADDRPERM(lr_saved));
3257 #endif
3258 if (refcount) {
3259 so->so_usecount--;
3260 }
3261
3262 if (so->so_usecount < 0) {
3263 panic("tcp_unlock: so=%p usecount=%x lrh= %s",
3264 so, so->so_usecount, solockhistory_nr(so));
3265 /* NOTREACHED */
3266 }
3267 if (so->so_pcb == NULL) {
3268 panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s",
3269 so, so->so_usecount, lr_saved, solockhistory_nr(so));
3270 /* NOTREACHED */
3271 } else {
3272 so->unlock_lr[so->next_unlock_lr] = lr_saved;
3273 so->next_unlock_lr = (so->next_unlock_lr + 1) % SO_LCKDBG_MAX;
3274
3275 if (so->so_flags & SOF_MP_SUBFLOW) {
3276 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3277 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3278
3279 socket_lock_assert_owned(mp_so);
3280
3281 socket_unlock(mp_so, refcount);
3282 } else {
3283 LCK_MTX_ASSERT(&((struct inpcb *)so->so_pcb)->inpcb_mtx,
3284 LCK_MTX_ASSERT_OWNED);
3285 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
3286 }
3287 }
3288 return 0;
3289 }
3290
3291 lck_mtx_t *
tcp_getlock(struct socket * so,int flags)3292 tcp_getlock(struct socket *so, int flags)
3293 {
3294 struct inpcb *inp = sotoinpcb(so);
3295
3296 if (so->so_pcb) {
3297 if (so->so_usecount < 0) {
3298 panic("tcp_getlock: so=%p usecount=%x lrh= %s",
3299 so, so->so_usecount, solockhistory_nr(so));
3300 }
3301
3302 if (so->so_flags & SOF_MP_SUBFLOW) {
3303 struct mptcb *mp_tp = tptomptp(sototcpcb(so));
3304 struct socket *mp_so = mptetoso(mp_tp->mpt_mpte);
3305
3306 return mp_so->so_proto->pr_getlock(mp_so, flags);
3307 } else {
3308 return &inp->inpcb_mtx;
3309 }
3310 } else {
3311 panic("tcp_getlock: so=%p NULL so_pcb %s",
3312 so, solockhistory_nr(so));
3313 return so->so_proto->pr_domain->dom_mtx;
3314 }
3315 }
3316
3317 /*
3318 * Determine if we can grow the recieve socket buffer to avoid sending
3319 * a zero window update to the peer. We allow even socket buffers that
3320 * have fixed size (set by the application) to grow if the resource
3321 * constraints are met. They will also be trimmed after the application
3322 * reads data.
3323 */
3324 static void
tcp_sbrcv_grow_rwin(struct tcpcb * tp,struct sockbuf * sb)3325 tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb)
3326 {
3327 u_int32_t rcvbufinc = tp->t_maxseg << 4;
3328 u_int32_t rcvbuf = sb->sb_hiwat;
3329 struct socket *so = tp->t_inpcb->inp_socket;
3330
3331 if (tcp_recv_bg == 1 || IS_TCP_RECV_BG(so)) {
3332 return;
3333 }
3334
3335 if (tcp_do_autorcvbuf == 1 &&
3336 tcp_cansbgrow(sb) &&
3337 (tp->t_flags & TF_SLOWLINK) == 0 &&
3338 (so->so_flags1 & SOF1_EXTEND_BK_IDLE_WANTED) == 0 &&
3339 (rcvbuf - sb->sb_cc) < rcvbufinc &&
3340 rcvbuf < tcp_autorcvbuf_max &&
3341 (sb->sb_idealsize > 0 &&
3342 sb->sb_hiwat <= (sb->sb_idealsize + rcvbufinc))) {
3343 sbreserve(sb,
3344 min((sb->sb_hiwat + rcvbufinc), tcp_autorcvbuf_max));
3345 }
3346 }
3347
3348 int32_t
tcp_sbspace(struct tcpcb * tp)3349 tcp_sbspace(struct tcpcb *tp)
3350 {
3351 struct socket *so = tp->t_inpcb->inp_socket;
3352 struct sockbuf *sb = &so->so_rcv;
3353 u_int32_t rcvbuf;
3354 int32_t space;
3355 int32_t pending = 0;
3356
3357 if (so->so_flags & SOF_MP_SUBFLOW) {
3358 /* We still need to grow TCP's buffer to have a BDP-estimate */
3359 tcp_sbrcv_grow_rwin(tp, sb);
3360
3361 return mptcp_sbspace(tptomptp(tp));
3362 }
3363
3364 tcp_sbrcv_grow_rwin(tp, sb);
3365
3366 /* hiwat might have changed */
3367 rcvbuf = sb->sb_hiwat;
3368
3369 space = ((int32_t) imin((rcvbuf - sb->sb_cc),
3370 (sb->sb_mbmax - sb->sb_mbcnt)));
3371 if (space < 0) {
3372 space = 0;
3373 }
3374
3375 #if CONTENT_FILTER
3376 /* Compensate for data being processed by content filters */
3377 pending = cfil_sock_data_space(sb);
3378 #endif /* CONTENT_FILTER */
3379 if (pending > space) {
3380 space = 0;
3381 } else {
3382 space -= pending;
3383 }
3384
3385 /*
3386 * Avoid increasing window size if the current window
3387 * is already very low, we could be in "persist" mode and
3388 * we could break some apps (see rdar://5409343)
3389 */
3390
3391 if (space < tp->t_maxseg) {
3392 return space;
3393 }
3394
3395 /* Clip window size for slower link */
3396
3397 if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0) {
3398 return imin(space, slowlink_wsize);
3399 }
3400
3401 return space;
3402 }
3403 /*
3404 * Checks TCP Segment Offloading capability for a given connection
3405 * and interface pair.
3406 */
3407 void
tcp_set_tso(struct tcpcb * tp,struct ifnet * ifp)3408 tcp_set_tso(struct tcpcb *tp, struct ifnet *ifp)
3409 {
3410 struct inpcb *inp;
3411 int isipv6;
3412 struct ifnet *tunnel_ifp = NULL;
3413 #define IFNET_TSO_MASK (IFNET_TSO_IPV6 | IFNET_TSO_IPV4)
3414
3415 tp->t_flags &= ~TF_TSO;
3416
3417 /*
3418 * Bail if there's a non-TSO-capable filter on the interface.
3419 */
3420 if (ifp == NULL || ifp->if_flt_no_tso_count > 0) {
3421 return;
3422 }
3423
3424 inp = tp->t_inpcb;
3425 isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
3426
3427 #if MPTCP
3428 /*
3429 * We can't use TSO if this tcpcb belongs to an MPTCP session.
3430 */
3431 if (inp->inp_socket->so_flags & SOF_MP_SUBFLOW) {
3432 return;
3433 }
3434 #endif
3435 /*
3436 * We can't use TSO if the TSO capability of the tunnel interface does
3437 * not match the capability of another interface known by TCP
3438 */
3439 if (inp->inp_policyresult.results.result == NECP_KERNEL_POLICY_RESULT_IP_TUNNEL) {
3440 u_int tunnel_if_index = inp->inp_policyresult.results.result_parameter.tunnel_interface_index;
3441
3442 if (tunnel_if_index != 0) {
3443 ifnet_head_lock_shared();
3444 tunnel_ifp = ifindex2ifnet[tunnel_if_index];
3445 ifnet_head_done();
3446 }
3447
3448 if (tunnel_ifp == NULL) {
3449 return;
3450 }
3451
3452 if ((ifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3453 if (tso_debug > 0) {
3454 os_log(OS_LOG_DEFAULT,
3455 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with ifp %s",
3456 __func__,
3457 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3458 tunnel_ifp->if_xname, ifp->if_xname);
3459 }
3460 return;
3461 }
3462 if (inp->inp_last_outifp != NULL &&
3463 (inp->inp_last_outifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3464 if (tso_debug > 0) {
3465 os_log(OS_LOG_DEFAULT,
3466 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_last_outifp %s",
3467 __func__,
3468 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3469 tunnel_ifp->if_xname, inp->inp_last_outifp->if_xname);
3470 }
3471 return;
3472 }
3473 if ((inp->inp_flags & INP_BOUND_IF) && inp->inp_boundifp != NULL &&
3474 (inp->inp_boundifp->if_hwassist & IFNET_TSO_MASK) != (tunnel_ifp->if_hwassist & IFNET_TSO_MASK)) {
3475 if (tso_debug > 0) {
3476 os_log(OS_LOG_DEFAULT,
3477 "%s: %u > %u TSO 0 tunnel_ifp %s hwassist mismatch with inp_boundifp %s",
3478 __func__,
3479 ntohs(tp->t_inpcb->inp_lport), ntohs(tp->t_inpcb->inp_fport),
3480 tunnel_ifp->if_xname, inp->inp_boundifp->if_xname);
3481 }
3482 return;
3483 }
3484 }
3485
3486 if (isipv6) {
3487 if (ifp->if_hwassist & IFNET_TSO_IPV6) {
3488 tp->t_flags |= TF_TSO;
3489 if (ifp->if_tso_v6_mtu != 0) {
3490 tp->tso_max_segment_size = ifp->if_tso_v6_mtu;
3491 } else {
3492 tp->tso_max_segment_size = TCP_MAXWIN;
3493 }
3494 }
3495 } else {
3496 if (ifp->if_hwassist & IFNET_TSO_IPV4) {
3497 tp->t_flags |= TF_TSO;
3498 if (ifp->if_tso_v4_mtu != 0) {
3499 tp->tso_max_segment_size = ifp->if_tso_v4_mtu;
3500 } else {
3501 tp->tso_max_segment_size = TCP_MAXWIN;
3502 }
3503 if (INTF_ADJUST_MTU_FOR_CLAT46(ifp)) {
3504 tp->tso_max_segment_size -=
3505 CLAT46_HDR_EXPANSION_OVERHD;
3506 }
3507 }
3508 }
3509
3510 if (tso_debug > 1) {
3511 os_log(OS_LOG_DEFAULT, "%s: %u > %u TSO %d ifp %s",
3512 __func__,
3513 ntohs(tp->t_inpcb->inp_lport),
3514 ntohs(tp->t_inpcb->inp_fport),
3515 (tp->t_flags & TF_TSO) != 0,
3516 ifp != NULL ? ifp->if_xname : "<NULL>");
3517 }
3518 }
3519
3520 #define TIMEVAL_TO_TCPHZ(_tv_) ((uint32_t)((_tv_).tv_sec * TCP_RETRANSHZ + \
3521 (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC))
3522
3523 /*
3524 * Function to calculate the tcp clock. The tcp clock will get updated
3525 * at the boundaries of the tcp layer. This is done at 3 places:
3526 * 1. Right before processing an input tcp packet
3527 * 2. Whenever a connection wants to access the network using tcp_usrreqs
3528 * 3. When a tcp timer fires or before tcp slow timeout
3529 *
3530 */
3531
3532 void
calculate_tcp_clock(void)3533 calculate_tcp_clock(void)
3534 {
3535 struct timeval tv = tcp_uptime;
3536 struct timeval interval = {.tv_sec = 0, .tv_usec = TCP_RETRANSHZ_TO_USEC};
3537 struct timeval now, hold_now;
3538 uint32_t incr = 0;
3539
3540 microuptime(&now);
3541
3542 /*
3543 * Update coarse-grained networking timestamp (in sec.); the idea
3544 * is to update the counter returnable via net_uptime() when
3545 * we read time.
3546 */
3547 net_update_uptime_with_time(&now);
3548
3549 timevaladd(&tv, &interval);
3550 if (timevalcmp(&now, &tv, >)) {
3551 /* time to update the clock */
3552 lck_spin_lock(&tcp_uptime_lock);
3553 if (timevalcmp(&tcp_uptime, &now, >=)) {
3554 /* clock got updated while waiting for the lock */
3555 lck_spin_unlock(&tcp_uptime_lock);
3556 return;
3557 }
3558
3559 microuptime(&now);
3560 hold_now = now;
3561 tv = tcp_uptime;
3562 timevalsub(&now, &tv);
3563
3564 incr = TIMEVAL_TO_TCPHZ(now);
3565
3566 /* Account for the previous remainder */
3567 uint32_t remaining_us = (now.tv_usec % TCP_RETRANSHZ_TO_USEC) +
3568 tcp_now_remainder_us;
3569 if (remaining_us >= TCP_RETRANSHZ_TO_USEC) {
3570 incr += (remaining_us / TCP_RETRANSHZ_TO_USEC);
3571 }
3572
3573 if (incr > 0) {
3574 tcp_uptime = hold_now;
3575 tcp_now_remainder_us = remaining_us % TCP_RETRANSHZ_TO_USEC;
3576 tcp_now += incr;
3577 }
3578
3579 lck_spin_unlock(&tcp_uptime_lock);
3580 }
3581 }
3582
3583 /*
3584 * Compute receive window scaling that we are going to request
3585 * for this connection based on sb_hiwat. Try to leave some
3586 * room to potentially increase the window size upto a maximum
3587 * defined by the constant tcp_autorcvbuf_max.
3588 */
3589 void
tcp_set_max_rwinscale(struct tcpcb * tp,struct socket * so)3590 tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so)
3591 {
3592 uint32_t maxsockbufsize;
3593
3594 tp->request_r_scale = MAX((uint8_t)tcp_win_scale, tp->request_r_scale);
3595 maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ?
3596 so->so_rcv.sb_hiwat : tcp_autorcvbuf_max;
3597
3598 /*
3599 * Window scale should not exceed what is needed
3600 * to send the max receive window size; adding 1 to TCP_MAXWIN
3601 * ensures that.
3602 */
3603 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
3604 ((TCP_MAXWIN + 1) << tp->request_r_scale) < maxsockbufsize) {
3605 tp->request_r_scale++;
3606 }
3607 tp->request_r_scale = MIN(tp->request_r_scale, TCP_MAX_WINSHIFT);
3608 }
3609
3610 int
tcp_notsent_lowat_check(struct socket * so)3611 tcp_notsent_lowat_check(struct socket *so)
3612 {
3613 struct inpcb *inp = sotoinpcb(so);
3614 struct tcpcb *tp = NULL;
3615 int notsent = 0;
3616
3617 if (inp != NULL) {
3618 tp = intotcpcb(inp);
3619 }
3620
3621 if (tp == NULL) {
3622 return 0;
3623 }
3624
3625 notsent = so->so_snd.sb_cc -
3626 (tp->snd_nxt - tp->snd_una);
3627
3628 /*
3629 * When we send a FIN or SYN, not_sent can be negative.
3630 * In that case also we need to send a write event to the
3631 * process if it is waiting. In the FIN case, it will
3632 * get an error from send because cantsendmore will be set.
3633 */
3634 if (notsent <= tp->t_notsent_lowat) {
3635 return 1;
3636 }
3637
3638 /*
3639 * When Nagle's algorithm is not disabled, it is better
3640 * to wakeup the client until there is atleast one
3641 * maxseg of data to write.
3642 */
3643 if ((tp->t_flags & TF_NODELAY) == 0 &&
3644 notsent > 0 && notsent < tp->t_maxseg) {
3645 return 1;
3646 }
3647 return 0;
3648 }
3649
3650 void
tcp_rxtseg_insert(struct tcpcb * tp,tcp_seq start,tcp_seq end)3651 tcp_rxtseg_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3652 {
3653 struct tcp_rxt_seg *rxseg = NULL, *prev = NULL, *next = NULL;
3654 uint16_t rxcount = 0;
3655
3656 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3657 tp->t_dsack_lastuna = tp->snd_una;
3658 }
3659 /*
3660 * First check if there is a segment already existing for this
3661 * sequence space.
3662 */
3663
3664 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3665 if (SEQ_GT(rxseg->rx_start, start)) {
3666 break;
3667 }
3668 prev = rxseg;
3669 }
3670 next = rxseg;
3671
3672 /* check if prev seg is for this sequence */
3673 if (prev != NULL && SEQ_LEQ(prev->rx_start, start) &&
3674 SEQ_GEQ(prev->rx_end, end)) {
3675 prev->rx_count++;
3676 return;
3677 }
3678
3679 /*
3680 * There are a couple of possibilities at this point.
3681 * 1. prev overlaps with the beginning of this sequence
3682 * 2. next overlaps with the end of this sequence
3683 * 3. there is no overlap.
3684 */
3685
3686 if (prev != NULL && SEQ_GT(prev->rx_end, start)) {
3687 if (prev->rx_start == start && SEQ_GT(end, prev->rx_end)) {
3688 start = prev->rx_end + 1;
3689 prev->rx_count++;
3690 } else {
3691 prev->rx_end = (start - 1);
3692 rxcount = prev->rx_count;
3693 }
3694 }
3695
3696 if (next != NULL && SEQ_LT(next->rx_start, end)) {
3697 if (SEQ_LEQ(next->rx_end, end)) {
3698 end = next->rx_start - 1;
3699 next->rx_count++;
3700 } else {
3701 next->rx_start = end + 1;
3702 rxcount = next->rx_count;
3703 }
3704 }
3705 if (!SEQ_LT(start, end)) {
3706 return;
3707 }
3708
3709 if (tcp_rxt_seg_max > 0 && tp->t_rxt_seg_count >= tcp_rxt_seg_max) {
3710 rxseg = SLIST_FIRST(&tp->t_rxt_segments);
3711 if (prev == rxseg) {
3712 prev = NULL;
3713 }
3714 SLIST_REMOVE(&tp->t_rxt_segments, rxseg,
3715 tcp_rxt_seg, rx_link);
3716
3717 tcp_rxt_seg_drop++;
3718 tp->t_rxt_seg_drop++;
3719 TCP_LOG(tp, "removed rxseg list overflow %u:%u ",
3720 rxseg->rx_start, rxseg->rx_end);
3721 zfree(tcp_rxt_seg_zone, rxseg);
3722
3723 tp->t_rxt_seg_count -= 1;
3724 }
3725
3726 rxseg = zalloc_flags(tcp_rxt_seg_zone, Z_WAITOK | Z_ZERO | Z_NOFAIL);
3727 rxseg->rx_start = start;
3728 rxseg->rx_end = end;
3729 rxseg->rx_count = rxcount + 1;
3730
3731 if (prev != NULL) {
3732 SLIST_INSERT_AFTER(prev, rxseg, rx_link);
3733 } else {
3734 SLIST_INSERT_HEAD(&tp->t_rxt_segments, rxseg, rx_link);
3735 }
3736 tp->t_rxt_seg_count += 1;
3737 }
3738
3739 struct tcp_rxt_seg *
tcp_rxtseg_find(struct tcpcb * tp,tcp_seq start,tcp_seq end)3740 tcp_rxtseg_find(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3741 {
3742 struct tcp_rxt_seg *rxseg;
3743
3744 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3745 return NULL;
3746 }
3747
3748 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3749 if (SEQ_LEQ(rxseg->rx_start, start) &&
3750 SEQ_GEQ(rxseg->rx_end, end)) {
3751 return rxseg;
3752 }
3753 if (SEQ_GT(rxseg->rx_start, start)) {
3754 break;
3755 }
3756 }
3757 return NULL;
3758 }
3759
3760 void
tcp_rxtseg_set_spurious(struct tcpcb * tp,tcp_seq start,tcp_seq end)3761 tcp_rxtseg_set_spurious(struct tcpcb *tp, tcp_seq start, tcp_seq end)
3762 {
3763 struct tcp_rxt_seg *rxseg;
3764
3765 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3766 return;
3767 }
3768
3769 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3770 if (SEQ_GEQ(rxseg->rx_start, start) &&
3771 SEQ_LEQ(rxseg->rx_end, end)) {
3772 /*
3773 * If the segment was retransmitted only once, mark it as
3774 * spurious.
3775 */
3776 if (rxseg->rx_count == 1) {
3777 rxseg->rx_flags |= TCP_RXT_SPURIOUS;
3778 }
3779 }
3780
3781 if (SEQ_GEQ(rxseg->rx_start, end)) {
3782 break;
3783 }
3784 }
3785 return;
3786 }
3787
3788 void
tcp_rxtseg_clean(struct tcpcb * tp)3789 tcp_rxtseg_clean(struct tcpcb *tp)
3790 {
3791 struct tcp_rxt_seg *rxseg, *next;
3792
3793 SLIST_FOREACH_SAFE(rxseg, &tp->t_rxt_segments, rx_link, next) {
3794 SLIST_REMOVE(&tp->t_rxt_segments, rxseg,
3795 tcp_rxt_seg, rx_link);
3796 zfree(tcp_rxt_seg_zone, rxseg);
3797 }
3798 tp->t_rxt_seg_count = 0;
3799 tp->t_dsack_lastuna = tp->snd_max;
3800 }
3801
3802 boolean_t
tcp_rxtseg_detect_bad_rexmt(struct tcpcb * tp,tcp_seq th_ack)3803 tcp_rxtseg_detect_bad_rexmt(struct tcpcb *tp, tcp_seq th_ack)
3804 {
3805 boolean_t bad_rexmt;
3806 struct tcp_rxt_seg *rxseg;
3807
3808 if (SLIST_EMPTY(&tp->t_rxt_segments)) {
3809 return FALSE;
3810 }
3811
3812 /*
3813 * If all of the segments in this window are not cumulatively
3814 * acknowledged, then there can still be undetected packet loss.
3815 * Do not restore congestion window in that case.
3816 */
3817 if (SEQ_LT(th_ack, tp->snd_recover)) {
3818 return FALSE;
3819 }
3820
3821 bad_rexmt = TRUE;
3822 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3823 if (!(rxseg->rx_flags & TCP_RXT_SPURIOUS)) {
3824 bad_rexmt = FALSE;
3825 break;
3826 }
3827 }
3828 return bad_rexmt;
3829 }
3830
3831 u_int32_t
tcp_rxtseg_total_size(struct tcpcb * tp)3832 tcp_rxtseg_total_size(struct tcpcb *tp)
3833 {
3834 struct tcp_rxt_seg *rxseg;
3835 u_int32_t total_size = 0;
3836
3837 SLIST_FOREACH(rxseg, &tp->t_rxt_segments, rx_link) {
3838 total_size += (rxseg->rx_end - rxseg->rx_start) + 1;
3839 }
3840 return total_size;
3841 }
3842
3843 void
tcp_get_connectivity_status(struct tcpcb * tp,struct tcp_conn_status * connstatus)3844 tcp_get_connectivity_status(struct tcpcb *tp,
3845 struct tcp_conn_status *connstatus)
3846 {
3847 if (tp == NULL || connstatus == NULL) {
3848 return;
3849 }
3850 bzero(connstatus, sizeof(*connstatus));
3851 if (tp->t_rxtshift >= TCP_CONNECTIVITY_PROBES_MAX) {
3852 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3853 connstatus->write_probe_failed = 1;
3854 } else {
3855 connstatus->conn_probe_failed = 1;
3856 }
3857 }
3858 if (tp->t_rtimo_probes >= TCP_CONNECTIVITY_PROBES_MAX) {
3859 connstatus->read_probe_failed = 1;
3860 }
3861 if (tp->t_inpcb != NULL && tp->t_inpcb->inp_last_outifp != NULL &&
3862 (tp->t_inpcb->inp_last_outifp->if_eflags & IFEF_PROBE_CONNECTIVITY)) {
3863 connstatus->probe_activated = 1;
3864 }
3865 }
3866
3867 boolean_t
tfo_enabled(const struct tcpcb * tp)3868 tfo_enabled(const struct tcpcb *tp)
3869 {
3870 return (tp->t_flagsext & TF_FASTOPEN)? TRUE : FALSE;
3871 }
3872
3873 void
tcp_disable_tfo(struct tcpcb * tp)3874 tcp_disable_tfo(struct tcpcb *tp)
3875 {
3876 tp->t_flagsext &= ~TF_FASTOPEN;
3877 }
3878
3879 static struct mbuf *
tcp_make_keepalive_frame(struct tcpcb * tp,struct ifnet * ifp,boolean_t is_probe)3880 tcp_make_keepalive_frame(struct tcpcb *tp, struct ifnet *ifp,
3881 boolean_t is_probe)
3882 {
3883 struct inpcb *inp = tp->t_inpcb;
3884 struct tcphdr *th;
3885 u_int8_t *data;
3886 int win = 0;
3887 struct mbuf *m;
3888
3889 /*
3890 * The code assumes the IP + TCP headers fit in an mbuf packet header
3891 */
3892 _CASSERT(sizeof(struct ip) + sizeof(struct tcphdr) <= _MHLEN);
3893 _CASSERT(sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= _MHLEN);
3894
3895 MGETHDR(m, M_WAIT, MT_HEADER);
3896 if (m == NULL) {
3897 return NULL;
3898 }
3899 m->m_pkthdr.pkt_proto = IPPROTO_TCP;
3900
3901 data = mbuf_datastart(m);
3902
3903 if (inp->inp_vflag & INP_IPV4) {
3904 bzero(data, sizeof(struct ip) + sizeof(struct tcphdr));
3905 th = (struct tcphdr *)(void *) (data + sizeof(struct ip));
3906 m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
3907 m->m_pkthdr.len = m->m_len;
3908 } else {
3909 VERIFY(inp->inp_vflag & INP_IPV6);
3910
3911 bzero(data, sizeof(struct ip6_hdr)
3912 + sizeof(struct tcphdr));
3913 th = (struct tcphdr *)(void *)(data + sizeof(struct ip6_hdr));
3914 m->m_len = sizeof(struct ip6_hdr) +
3915 sizeof(struct tcphdr);
3916 m->m_pkthdr.len = m->m_len;
3917 }
3918
3919 tcp_fillheaders(m, tp, data, th);
3920
3921 if (inp->inp_vflag & INP_IPV4) {
3922 struct ip *ip;
3923
3924 ip = (__typeof__(ip))(void *)data;
3925
3926 ip->ip_id = rfc6864 ? 0 : ip_randomid((uint64_t)m);
3927 ip->ip_off = htons(IP_DF);
3928 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct tcphdr));
3929 ip->ip_ttl = inp->inp_ip_ttl;
3930 ip->ip_tos |= (inp->inp_ip_tos & ~IPTOS_ECN_MASK);
3931 ip->ip_sum = in_cksum_hdr(ip);
3932 } else {
3933 struct ip6_hdr *ip6;
3934
3935 ip6 = (__typeof__(ip6))(void *)data;
3936
3937 ip6->ip6_plen = htons(sizeof(struct tcphdr));
3938 ip6->ip6_hlim = in6_selecthlim(inp, ifp);
3939 ip6->ip6_flow = ip6->ip6_flow & ~IPV6_FLOW_ECN_MASK;
3940
3941 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_src)) {
3942 ip6->ip6_src.s6_addr16[1] = 0;
3943 }
3944 if (IN6_IS_SCOPE_EMBED(&ip6->ip6_dst)) {
3945 ip6->ip6_dst.s6_addr16[1] = 0;
3946 }
3947 }
3948 th->th_flags = TH_ACK;
3949
3950 win = tcp_sbspace(tp);
3951 if (win > ((int32_t)TCP_MAXWIN << tp->rcv_scale)) {
3952 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
3953 }
3954 th->th_win = htons((u_short) (win >> tp->rcv_scale));
3955
3956 if (is_probe) {
3957 th->th_seq = htonl(tp->snd_una - 1);
3958 } else {
3959 th->th_seq = htonl(tp->snd_una);
3960 }
3961 th->th_ack = htonl(tp->rcv_nxt);
3962
3963 /* Force recompute TCP checksum to be the final value */
3964 th->th_sum = 0;
3965 if (inp->inp_vflag & INP_IPV4) {
3966 th->th_sum = inet_cksum(m, IPPROTO_TCP,
3967 sizeof(struct ip), sizeof(struct tcphdr));
3968 } else {
3969 th->th_sum = inet6_cksum(m, IPPROTO_TCP,
3970 sizeof(struct ip6_hdr), sizeof(struct tcphdr));
3971 }
3972
3973 return m;
3974 }
3975
3976 void
tcp_fill_keepalive_offload_frames(ifnet_t ifp,struct ifnet_keepalive_offload_frame * frames_array,u_int32_t frames_array_count,size_t frame_data_offset,u_int32_t * used_frames_count)3977 tcp_fill_keepalive_offload_frames(ifnet_t ifp,
3978 struct ifnet_keepalive_offload_frame *frames_array,
3979 u_int32_t frames_array_count, size_t frame_data_offset,
3980 u_int32_t *used_frames_count)
3981 {
3982 struct inpcb *inp;
3983 inp_gen_t gencnt;
3984 u_int32_t frame_index = *used_frames_count;
3985
3986 if (ifp == NULL || frames_array == NULL ||
3987 frames_array_count == 0 ||
3988 frame_index >= frames_array_count ||
3989 frame_data_offset >= IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
3990 return;
3991 }
3992
3993 /*
3994 * This function is called outside the regular TCP processing
3995 * so we need to update the TCP clock.
3996 */
3997 calculate_tcp_clock();
3998
3999 lck_rw_lock_shared(&tcbinfo.ipi_lock);
4000 gencnt = tcbinfo.ipi_gencnt;
4001 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
4002 struct socket *so;
4003 struct ifnet_keepalive_offload_frame *frame;
4004 struct mbuf *m = NULL;
4005 struct tcpcb *tp = intotcpcb(inp);
4006
4007 if (frame_index >= frames_array_count) {
4008 break;
4009 }
4010
4011 if (inp->inp_gencnt > gencnt ||
4012 inp->inp_state == INPCB_STATE_DEAD) {
4013 continue;
4014 }
4015
4016 if ((so = inp->inp_socket) == NULL ||
4017 (so->so_state & SS_DEFUNCT)) {
4018 continue;
4019 }
4020 /*
4021 * check for keepalive offload flag without socket
4022 * lock to avoid a deadlock
4023 */
4024 if (!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD)) {
4025 continue;
4026 }
4027
4028 if (!(inp->inp_vflag & (INP_IPV4 | INP_IPV6))) {
4029 continue;
4030 }
4031 if (inp->inp_ppcb == NULL ||
4032 in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
4033 continue;
4034 }
4035 socket_lock(so, 1);
4036 /* Release the want count */
4037 if (inp->inp_ppcb == NULL ||
4038 (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING)) {
4039 socket_unlock(so, 1);
4040 continue;
4041 }
4042 if ((inp->inp_vflag & INP_IPV4) &&
4043 (inp->inp_laddr.s_addr == INADDR_ANY ||
4044 inp->inp_faddr.s_addr == INADDR_ANY)) {
4045 socket_unlock(so, 1);
4046 continue;
4047 }
4048 if ((inp->inp_vflag & INP_IPV6) &&
4049 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ||
4050 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))) {
4051 socket_unlock(so, 1);
4052 continue;
4053 }
4054 if (inp->inp_lport == 0 || inp->inp_fport == 0) {
4055 socket_unlock(so, 1);
4056 continue;
4057 }
4058 if (inp->inp_last_outifp == NULL ||
4059 inp->inp_last_outifp->if_index != ifp->if_index) {
4060 socket_unlock(so, 1);
4061 continue;
4062 }
4063 if ((inp->inp_vflag & INP_IPV4) && frame_data_offset +
4064 sizeof(struct ip) + sizeof(struct tcphdr) >
4065 IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
4066 socket_unlock(so, 1);
4067 continue;
4068 } else if (!(inp->inp_vflag & INP_IPV4) && frame_data_offset +
4069 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) >
4070 IFNET_KEEPALIVE_OFFLOAD_FRAME_DATA_SIZE) {
4071 socket_unlock(so, 1);
4072 continue;
4073 }
4074 /*
4075 * There is no point in waking up the device for connections
4076 * that are not established. Long lived connection are meant
4077 * for processes that will sent and receive data
4078 */
4079 if (tp->t_state != TCPS_ESTABLISHED) {
4080 socket_unlock(so, 1);
4081 continue;
4082 }
4083 /*
4084 * This inp has all the information that is needed to
4085 * generate an offload frame.
4086 */
4087 frame = &frames_array[frame_index];
4088 frame->type = IFNET_KEEPALIVE_OFFLOAD_FRAME_TCP;
4089 frame->ether_type = (inp->inp_vflag & INP_IPV4) ?
4090 IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV4 :
4091 IFNET_KEEPALIVE_OFFLOAD_FRAME_ETHERTYPE_IPV6;
4092 frame->interval = (uint16_t)(tp->t_keepidle > 0 ? tp->t_keepidle :
4093 tcp_keepidle);
4094 frame->keep_cnt = (uint8_t)TCP_CONN_KEEPCNT(tp);
4095 frame->keep_retry = (uint16_t)TCP_CONN_KEEPINTVL(tp);
4096 if (so->so_options & SO_NOWAKEFROMSLEEP) {
4097 frame->flags |=
4098 IFNET_KEEPALIVE_OFFLOAD_FLAG_NOWAKEFROMSLEEP;
4099 }
4100 frame->local_port = ntohs(inp->inp_lport);
4101 frame->remote_port = ntohs(inp->inp_fport);
4102 frame->local_seq = tp->snd_nxt;
4103 frame->remote_seq = tp->rcv_nxt;
4104 if (inp->inp_vflag & INP_IPV4) {
4105 ASSERT(frame_data_offset + sizeof(struct ip) + sizeof(struct tcphdr) <= UINT8_MAX);
4106 frame->length = (uint8_t)(frame_data_offset +
4107 sizeof(struct ip) + sizeof(struct tcphdr));
4108 frame->reply_length = frame->length;
4109
4110 frame->addr_length = sizeof(struct in_addr);
4111 bcopy(&inp->inp_laddr, frame->local_addr,
4112 sizeof(struct in_addr));
4113 bcopy(&inp->inp_faddr, frame->remote_addr,
4114 sizeof(struct in_addr));
4115 } else {
4116 struct in6_addr *ip6;
4117
4118 ASSERT(frame_data_offset + sizeof(struct ip6_hdr) + sizeof(struct tcphdr) <= UINT8_MAX);
4119 frame->length = (uint8_t)(frame_data_offset +
4120 sizeof(struct ip6_hdr) + sizeof(struct tcphdr));
4121 frame->reply_length = frame->length;
4122
4123 frame->addr_length = sizeof(struct in6_addr);
4124 ip6 = (struct in6_addr *)(void *)frame->local_addr;
4125 bcopy(&inp->in6p_laddr, ip6, sizeof(struct in6_addr));
4126 if (IN6_IS_SCOPE_EMBED(ip6)) {
4127 ip6->s6_addr16[1] = 0;
4128 }
4129
4130 ip6 = (struct in6_addr *)(void *)frame->remote_addr;
4131 bcopy(&inp->in6p_faddr, ip6, sizeof(struct in6_addr));
4132 if (IN6_IS_SCOPE_EMBED(ip6)) {
4133 ip6->s6_addr16[1] = 0;
4134 }
4135 }
4136
4137 /*
4138 * First the probe
4139 */
4140 m = tcp_make_keepalive_frame(tp, ifp, TRUE);
4141 if (m == NULL) {
4142 socket_unlock(so, 1);
4143 continue;
4144 }
4145 bcopy(m->m_data, frame->data + frame_data_offset,
4146 m->m_len);
4147 m_freem(m);
4148
4149 /*
4150 * Now the response packet to incoming probes
4151 */
4152 m = tcp_make_keepalive_frame(tp, ifp, FALSE);
4153 if (m == NULL) {
4154 socket_unlock(so, 1);
4155 continue;
4156 }
4157 bcopy(m->m_data, frame->reply_data + frame_data_offset,
4158 m->m_len);
4159 m_freem(m);
4160
4161 frame_index++;
4162 socket_unlock(so, 1);
4163 }
4164 lck_rw_done(&tcbinfo.ipi_lock);
4165 *used_frames_count = frame_index;
4166 }
4167
4168 static bool
inp_matches_kao_frame(ifnet_t ifp,struct ifnet_keepalive_offload_frame * frame,struct inpcb * inp)4169 inp_matches_kao_frame(ifnet_t ifp, struct ifnet_keepalive_offload_frame *frame,
4170 struct inpcb *inp)
4171 {
4172 if (inp->inp_ppcb == NULL) {
4173 return false;
4174 }
4175 /* Release the want count */
4176 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
4177 return false;
4178 }
4179 if (inp->inp_last_outifp == NULL ||
4180 inp->inp_last_outifp->if_index != ifp->if_index) {
4181 return false;
4182 }
4183 if (frame->local_port != ntohs(inp->inp_lport) ||
4184 frame->remote_port != ntohs(inp->inp_fport)) {
4185 return false;
4186 }
4187 if (inp->inp_vflag & INP_IPV4) {
4188 if (memcmp(&inp->inp_laddr, frame->local_addr,
4189 sizeof(struct in_addr)) != 0 ||
4190 memcmp(&inp->inp_faddr, frame->remote_addr,
4191 sizeof(struct in_addr)) != 0) {
4192 return false;
4193 }
4194 } else if (inp->inp_vflag & INP_IPV6) {
4195 if (memcmp(&inp->inp_laddr, frame->local_addr,
4196 sizeof(struct in6_addr)) != 0 ||
4197 memcmp(&inp->inp_faddr, frame->remote_addr,
4198 sizeof(struct in6_addr)) != 0) {
4199 return false;
4200 }
4201 } else {
4202 return false;
4203 }
4204 return true;
4205 }
4206
4207 int
tcp_notify_kao_timeout(ifnet_t ifp,struct ifnet_keepalive_offload_frame * frame)4208 tcp_notify_kao_timeout(ifnet_t ifp,
4209 struct ifnet_keepalive_offload_frame *frame)
4210 {
4211 struct inpcb *inp = NULL;
4212 struct socket *so = NULL;
4213 bool found = false;
4214
4215 /*
4216 * Unlock the list before posting event on the matching socket
4217 */
4218 lck_rw_lock_shared(&tcbinfo.ipi_lock);
4219
4220 LIST_FOREACH(inp, tcbinfo.ipi_listhead, inp_list) {
4221 if ((so = inp->inp_socket) == NULL ||
4222 (so->so_state & SS_DEFUNCT)) {
4223 continue;
4224 }
4225 if (!(inp->inp_flags2 & INP2_KEEPALIVE_OFFLOAD)) {
4226 continue;
4227 }
4228 if (!(inp->inp_vflag & (INP_IPV4 | INP_IPV6))) {
4229 continue;
4230 }
4231 if (inp->inp_ppcb == NULL ||
4232 in_pcb_checkstate(inp, WNT_ACQUIRE, 0) == WNT_STOPUSING) {
4233 continue;
4234 }
4235 socket_lock(so, 1);
4236 if (inp_matches_kao_frame(ifp, frame, inp)) {
4237 /*
4238 * Keep the matching socket locked
4239 */
4240 found = true;
4241 break;
4242 }
4243 socket_unlock(so, 1);
4244 }
4245 lck_rw_done(&tcbinfo.ipi_lock);
4246
4247 if (found) {
4248 ASSERT(inp != NULL);
4249 ASSERT(so != NULL);
4250 ASSERT(so == inp->inp_socket);
4251 /*
4252 * Drop the TCP connection like tcptimers() does
4253 */
4254 struct tcpcb *tp = inp->inp_ppcb;
4255
4256 tcpstat.tcps_keepdrops++;
4257 soevent(so,
4258 (SO_FILT_HINT_LOCKED | SO_FILT_HINT_TIMEOUT));
4259 tp = tcp_drop(tp, ETIMEDOUT);
4260
4261 tcpstat.tcps_ka_offload_drops++;
4262 os_log_info(OS_LOG_DEFAULT, "%s: dropped lport %u fport %u\n",
4263 __func__, frame->local_port, frame->remote_port);
4264
4265 socket_unlock(so, 1);
4266 }
4267
4268 return 0;
4269 }
4270
4271 errno_t
tcp_notify_ack_id_valid(struct tcpcb * tp,struct socket * so,u_int32_t notify_id)4272 tcp_notify_ack_id_valid(struct tcpcb *tp, struct socket *so,
4273 u_int32_t notify_id)
4274 {
4275 struct tcp_notify_ack_marker *elm;
4276
4277 if (so->so_snd.sb_cc == 0) {
4278 return ENOBUFS;
4279 }
4280
4281 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4282 /* Duplicate id is not allowed */
4283 if (elm->notify_id == notify_id) {
4284 return EINVAL;
4285 }
4286 /* Duplicate position is not allowed */
4287 if (elm->notify_snd_una == tp->snd_una + so->so_snd.sb_cc) {
4288 return EINVAL;
4289 }
4290 }
4291 return 0;
4292 }
4293
4294 errno_t
tcp_add_notify_ack_marker(struct tcpcb * tp,u_int32_t notify_id)4295 tcp_add_notify_ack_marker(struct tcpcb *tp, u_int32_t notify_id)
4296 {
4297 struct tcp_notify_ack_marker *nm, *elm = NULL;
4298 struct socket *so = tp->t_inpcb->inp_socket;
4299
4300 nm = kalloc_type(struct tcp_notify_ack_marker, M_WAIT | Z_ZERO);
4301 if (nm == NULL) {
4302 return ENOMEM;
4303 }
4304 nm->notify_id = notify_id;
4305 nm->notify_snd_una = tp->snd_una + so->so_snd.sb_cc;
4306
4307 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4308 if (SEQ_GT(nm->notify_snd_una, elm->notify_snd_una)) {
4309 break;
4310 }
4311 }
4312
4313 if (elm == NULL) {
4314 VERIFY(SLIST_EMPTY(&tp->t_notify_ack));
4315 SLIST_INSERT_HEAD(&tp->t_notify_ack, nm, notify_next);
4316 } else {
4317 SLIST_INSERT_AFTER(elm, nm, notify_next);
4318 }
4319 tp->t_notify_ack_count++;
4320 return 0;
4321 }
4322
4323 void
tcp_notify_ack_free(struct tcpcb * tp)4324 tcp_notify_ack_free(struct tcpcb *tp)
4325 {
4326 struct tcp_notify_ack_marker *elm, *next;
4327 if (SLIST_EMPTY(&tp->t_notify_ack)) {
4328 return;
4329 }
4330
4331 SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) {
4332 SLIST_REMOVE(&tp->t_notify_ack, elm, tcp_notify_ack_marker,
4333 notify_next);
4334 kfree_type(struct tcp_notify_ack_marker, elm);
4335 }
4336 SLIST_INIT(&tp->t_notify_ack);
4337 tp->t_notify_ack_count = 0;
4338 }
4339
4340 inline void
tcp_notify_acknowledgement(struct tcpcb * tp,struct socket * so)4341 tcp_notify_acknowledgement(struct tcpcb *tp, struct socket *so)
4342 {
4343 struct tcp_notify_ack_marker *elm;
4344
4345 elm = SLIST_FIRST(&tp->t_notify_ack);
4346 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4347 soevent(so, SO_FILT_HINT_LOCKED | SO_FILT_HINT_NOTIFY_ACK);
4348 }
4349 }
4350
4351 void
tcp_get_notify_ack_count(struct tcpcb * tp,struct tcp_notify_ack_complete * retid)4352 tcp_get_notify_ack_count(struct tcpcb *tp,
4353 struct tcp_notify_ack_complete *retid)
4354 {
4355 struct tcp_notify_ack_marker *elm;
4356 uint32_t complete = 0;
4357
4358 SLIST_FOREACH(elm, &tp->t_notify_ack, notify_next) {
4359 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4360 ASSERT(complete < UINT32_MAX);
4361 complete++;
4362 } else {
4363 break;
4364 }
4365 }
4366 retid->notify_pending = tp->t_notify_ack_count - complete;
4367 retid->notify_complete_count = min(TCP_MAX_NOTIFY_ACK, complete);
4368 }
4369
4370 void
tcp_get_notify_ack_ids(struct tcpcb * tp,struct tcp_notify_ack_complete * retid)4371 tcp_get_notify_ack_ids(struct tcpcb *tp,
4372 struct tcp_notify_ack_complete *retid)
4373 {
4374 size_t i = 0;
4375 struct tcp_notify_ack_marker *elm, *next;
4376
4377 SLIST_FOREACH_SAFE(elm, &tp->t_notify_ack, notify_next, next) {
4378 if (i >= retid->notify_complete_count) {
4379 break;
4380 }
4381 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4382 retid->notify_complete_id[i++] = elm->notify_id;
4383 SLIST_REMOVE(&tp->t_notify_ack, elm,
4384 tcp_notify_ack_marker, notify_next);
4385 kfree_type(struct tcp_notify_ack_marker, elm);
4386 tp->t_notify_ack_count--;
4387 } else {
4388 break;
4389 }
4390 }
4391 }
4392
4393 bool
tcp_notify_ack_active(struct socket * so)4394 tcp_notify_ack_active(struct socket *so)
4395 {
4396 if ((SOCK_DOM(so) == PF_INET || SOCK_DOM(so) == PF_INET6) &&
4397 SOCK_TYPE(so) == SOCK_STREAM) {
4398 struct tcpcb *tp = intotcpcb(sotoinpcb(so));
4399
4400 if (!SLIST_EMPTY(&tp->t_notify_ack)) {
4401 struct tcp_notify_ack_marker *elm;
4402 elm = SLIST_FIRST(&tp->t_notify_ack);
4403 if (SEQ_GEQ(tp->snd_una, elm->notify_snd_una)) {
4404 return true;
4405 }
4406 }
4407 }
4408 return false;
4409 }
4410
4411 inline int32_t
inp_get_sndbytes_allunsent(struct socket * so,u_int32_t th_ack)4412 inp_get_sndbytes_allunsent(struct socket *so, u_int32_t th_ack)
4413 {
4414 struct inpcb *inp = sotoinpcb(so);
4415 struct tcpcb *tp = intotcpcb(inp);
4416
4417 if ((so->so_snd.sb_flags & SB_SNDBYTE_CNT) &&
4418 so->so_snd.sb_cc > 0) {
4419 int32_t unsent, sent;
4420 sent = tp->snd_max - th_ack;
4421 if (tp->t_flags & TF_SENTFIN) {
4422 sent--;
4423 }
4424 unsent = so->so_snd.sb_cc - sent;
4425 return unsent;
4426 }
4427 return 0;
4428 }
4429
4430 uint8_t
tcp_get_ace(struct tcphdr * th)4431 tcp_get_ace(struct tcphdr *th)
4432 {
4433 uint8_t ace = 0;
4434 if (th->th_flags & TH_ECE) {
4435 ace += 1;
4436 }
4437 if (th->th_flags & TH_CWR) {
4438 ace += 2;
4439 }
4440 if (th->th_x2 & (TH_AE >> 8)) {
4441 ace += 4;
4442 }
4443
4444 return ace;
4445 }
4446
4447 #define IFP_PER_FLOW_STAT(_ipv4_, _stat_) { \
4448 if (_ipv4_) { \
4449 ifp->if_ipv4_stat->_stat_++; \
4450 } else { \
4451 ifp->if_ipv6_stat->_stat_++; \
4452 } \
4453 }
4454
4455 #define FLOW_ECN_ENABLED(_flags_) \
4456 ((_flags_ & (TE_ECN_ON)) == (TE_ECN_ON))
4457
4458 void
tcp_update_stats_per_flow(struct ifnet_stats_per_flow * ifs,struct ifnet * ifp)4459 tcp_update_stats_per_flow(struct ifnet_stats_per_flow *ifs,
4460 struct ifnet *ifp)
4461 {
4462 if (ifp == NULL || !IF_FULLY_ATTACHED(ifp)) {
4463 return;
4464 }
4465
4466 ifnet_lock_shared(ifp);
4467 if (ifs->ecn_flags & TE_SETUPSENT) {
4468 if (ifs->ecn_flags & TE_CLIENT_SETUP) {
4469 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_client_setup);
4470 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4471 IFP_PER_FLOW_STAT(ifs->ipv4,
4472 ecn_client_success);
4473 } else if (ifs->ecn_flags & TE_LOST_SYN) {
4474 IFP_PER_FLOW_STAT(ifs->ipv4,
4475 ecn_syn_lost);
4476 } else {
4477 IFP_PER_FLOW_STAT(ifs->ipv4,
4478 ecn_peer_nosupport);
4479 }
4480 } else {
4481 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_server_setup);
4482 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4483 IFP_PER_FLOW_STAT(ifs->ipv4,
4484 ecn_server_success);
4485 } else if (ifs->ecn_flags & TE_LOST_SYN) {
4486 IFP_PER_FLOW_STAT(ifs->ipv4,
4487 ecn_synack_lost);
4488 } else {
4489 IFP_PER_FLOW_STAT(ifs->ipv4,
4490 ecn_peer_nosupport);
4491 }
4492 }
4493 } else {
4494 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off_conn);
4495 }
4496 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4497 if (ifs->ecn_flags & TE_RECV_ECN_CE) {
4498 tcpstat.tcps_ecn_conn_recv_ce++;
4499 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ce);
4500 }
4501 if (ifs->ecn_flags & TE_RECV_ECN_ECE) {
4502 tcpstat.tcps_ecn_conn_recv_ece++;
4503 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_recv_ece);
4504 }
4505 if (ifs->ecn_flags & (TE_RECV_ECN_CE | TE_RECV_ECN_ECE)) {
4506 if (ifs->txretransmitbytes > 0 ||
4507 ifs->rxoutoforderbytes > 0) {
4508 tcpstat.tcps_ecn_conn_pl_ce++;
4509 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plce);
4510 } else {
4511 tcpstat.tcps_ecn_conn_nopl_ce++;
4512 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_noplce);
4513 }
4514 } else {
4515 if (ifs->txretransmitbytes > 0 ||
4516 ifs->rxoutoforderbytes > 0) {
4517 tcpstat.tcps_ecn_conn_plnoce++;
4518 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_conn_plnoce);
4519 }
4520 }
4521 }
4522
4523 /* Other stats are interesting for non-local connections only */
4524 if (ifs->local) {
4525 ifnet_lock_done(ifp);
4526 return;
4527 }
4528
4529 if (ifs->ipv4) {
4530 ifp->if_ipv4_stat->timestamp = net_uptime();
4531 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4532 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_on);
4533 } else {
4534 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv4_stat->ecn_off);
4535 }
4536 } else {
4537 ifp->if_ipv6_stat->timestamp = net_uptime();
4538 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4539 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_on);
4540 } else {
4541 tcp_flow_ecn_perf_stats(ifs, &ifp->if_ipv6_stat->ecn_off);
4542 }
4543 }
4544
4545 if (ifs->rxmit_drop) {
4546 if (FLOW_ECN_ENABLED(ifs->ecn_flags)) {
4547 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_on.rxmit_drop);
4548 } else {
4549 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_off.rxmit_drop);
4550 }
4551 }
4552 if (ifs->ecn_fallback_synloss) {
4553 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_synloss);
4554 }
4555 if (ifs->ecn_fallback_droprst) {
4556 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprst);
4557 }
4558 if (ifs->ecn_fallback_droprxmt) {
4559 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_droprxmt);
4560 }
4561 if (ifs->ecn_fallback_ce) {
4562 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_ce);
4563 }
4564 if (ifs->ecn_fallback_reorder) {
4565 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_fallback_reorder);
4566 }
4567 if (ifs->ecn_recv_ce > 0) {
4568 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ce);
4569 }
4570 if (ifs->ecn_recv_ece > 0) {
4571 IFP_PER_FLOW_STAT(ifs->ipv4, ecn_recv_ece);
4572 }
4573
4574 tcp_flow_lim_stats(ifs, &ifp->if_lim_stat);
4575 ifnet_lock_done(ifp);
4576 }
4577
4578 #if SKYWALK
4579
4580 #include <skywalk/core/skywalk_var.h>
4581 #include <skywalk/nexus/flowswitch/nx_flowswitch.h>
4582
4583 void
tcp_add_fsw_flow(struct tcpcb * tp,struct ifnet * ifp)4584 tcp_add_fsw_flow(struct tcpcb *tp, struct ifnet *ifp)
4585 {
4586 struct inpcb *inp = tp->t_inpcb;
4587 struct socket *so = inp->inp_socket;
4588 uuid_t fsw_uuid;
4589 struct nx_flow_req nfr;
4590 int err;
4591
4592 if (!NX_FSW_TCP_RX_AGG_ENABLED()) {
4593 return;
4594 }
4595
4596 if (ifp == NULL || kern_nexus_get_flowswitch_instance(ifp, fsw_uuid)) {
4597 TCP_LOG_FSW_FLOW(tp, "skip ifp no fsw");
4598 return;
4599 }
4600
4601 memset(&nfr, 0, sizeof(nfr));
4602
4603 if (inp->inp_vflag & INP_IPV4) {
4604 ASSERT(!(inp->inp_laddr.s_addr == INADDR_ANY ||
4605 inp->inp_faddr.s_addr == INADDR_ANY ||
4606 IN_MULTICAST(ntohl(inp->inp_laddr.s_addr)) ||
4607 IN_MULTICAST(ntohl(inp->inp_faddr.s_addr))));
4608 nfr.nfr_saddr.sin.sin_len = sizeof(struct sockaddr_in);
4609 nfr.nfr_saddr.sin.sin_family = AF_INET;
4610 nfr.nfr_saddr.sin.sin_port = inp->inp_lport;
4611 memcpy(&nfr.nfr_saddr.sin.sin_addr, &inp->inp_laddr,
4612 sizeof(struct in_addr));
4613 nfr.nfr_daddr.sin.sin_len = sizeof(struct sockaddr_in);
4614 nfr.nfr_daddr.sin.sin_family = AF_INET;
4615 nfr.nfr_daddr.sin.sin_port = inp->inp_fport;
4616 memcpy(&nfr.nfr_daddr.sin.sin_addr, &inp->inp_faddr,
4617 sizeof(struct in_addr));
4618 } else {
4619 ASSERT(!(IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ||
4620 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr) ||
4621 IN6_IS_ADDR_MULTICAST(&inp->in6p_laddr) ||
4622 IN6_IS_ADDR_MULTICAST(&inp->in6p_faddr)));
4623 nfr.nfr_saddr.sin6.sin6_len = sizeof(struct sockaddr_in6);
4624 nfr.nfr_saddr.sin6.sin6_family = AF_INET6;
4625 nfr.nfr_saddr.sin6.sin6_port = inp->inp_lport;
4626 memcpy(&nfr.nfr_saddr.sin6.sin6_addr, &inp->in6p_laddr,
4627 sizeof(struct in6_addr));
4628 nfr.nfr_daddr.sin6.sin6_len = sizeof(struct sockaddr_in6);
4629 nfr.nfr_daddr.sin.sin_family = AF_INET6;
4630 nfr.nfr_daddr.sin6.sin6_port = inp->inp_fport;
4631 memcpy(&nfr.nfr_daddr.sin6.sin6_addr, &inp->in6p_faddr,
4632 sizeof(struct in6_addr));
4633 /* clear embedded scope ID */
4634 if (IN6_IS_SCOPE_EMBED(&nfr.nfr_saddr.sin6.sin6_addr)) {
4635 nfr.nfr_saddr.sin6.sin6_addr.s6_addr16[1] = 0;
4636 }
4637 if (IN6_IS_SCOPE_EMBED(&nfr.nfr_daddr.sin6.sin6_addr)) {
4638 nfr.nfr_daddr.sin6.sin6_addr.s6_addr16[1] = 0;
4639 }
4640 }
4641
4642 nfr.nfr_nx_port = 1;
4643 nfr.nfr_ip_protocol = IPPROTO_TCP;
4644 nfr.nfr_transport_protocol = IPPROTO_TCP;
4645 nfr.nfr_flags = NXFLOWREQF_ASIS;
4646 nfr.nfr_epid = (so != NULL ? so->last_pid : 0);
4647 if (NETNS_TOKEN_VALID(&inp->inp_netns_token)) {
4648 nfr.nfr_port_reservation = inp->inp_netns_token;
4649 nfr.nfr_flags |= NXFLOWREQF_EXT_PORT_RSV;
4650 }
4651 ASSERT(inp->inp_flowhash != 0);
4652 nfr.nfr_inp_flowhash = inp->inp_flowhash;
4653
4654 uuid_generate_random(nfr.nfr_flow_uuid);
4655 err = kern_nexus_flow_add(kern_nexus_shared_controller(), fsw_uuid,
4656 &nfr, sizeof(nfr));
4657
4658 if (err == 0) {
4659 uuid_copy(tp->t_fsw_uuid, fsw_uuid);
4660 uuid_copy(tp->t_flow_uuid, nfr.nfr_flow_uuid);
4661 }
4662
4663 TCP_LOG_FSW_FLOW(tp, "add err %d\n", err);
4664 }
4665
4666 void
tcp_del_fsw_flow(struct tcpcb * tp)4667 tcp_del_fsw_flow(struct tcpcb *tp)
4668 {
4669 if (uuid_is_null(tp->t_fsw_uuid) || uuid_is_null(tp->t_flow_uuid)) {
4670 return;
4671 }
4672
4673 struct nx_flow_req nfr;
4674 uuid_copy(nfr.nfr_flow_uuid, tp->t_flow_uuid);
4675
4676 /* It's possible for this call to fail if the nexus has detached */
4677 int err = kern_nexus_flow_del(kern_nexus_shared_controller(),
4678 tp->t_fsw_uuid, &nfr, sizeof(nfr));
4679 VERIFY(err == 0 || err == ENOENT || err == ENXIO);
4680
4681 uuid_clear(tp->t_fsw_uuid);
4682 uuid_clear(tp->t_flow_uuid);
4683
4684 TCP_LOG_FSW_FLOW(tp, "del err %d\n", err);
4685 }
4686
4687 #endif /* SKYWALK */
4688