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


531b37fa9e08ead34e37384e5add75eda290437d
[apple/xnu.git] / bsd / netinet / tcp_subr.c
1 /*
2 * Copyright (c) 2000-2012 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*
29 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
30 * The Regents of the University of California. All rights reserved.
31 *
32 * Redistribution and use in source and binary forms, with or without
33 * modification, are permitted provided that the following conditions
34 * are met:
35 * 1. Redistributions of source code must retain the above copyright
36 * notice, this list of conditions and the following disclaimer.
37 * 2. Redistributions in binary form must reproduce the above copyright
38 * notice, this list of conditions and the following disclaimer in the
39 * documentation and/or other materials provided with the distribution.
40 * 3. All advertising materials mentioning features or use of this software
41 * must display the following acknowledgement:
42 * This product includes software developed by the University of
43 * California, Berkeley and its contributors.
44 * 4. Neither the name of the University nor the names of its contributors
45 * may be used to endorse or promote products derived from this software
46 * without specific prior written permission.
47 *
48 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
49 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
50 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
51 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
52 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
53 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
54 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
55 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
56 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
57 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
58 * SUCH DAMAGE.
59 *
60 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
61 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.22 2001/08/22 00:59:12 silby Exp $
62 */
63 /*
64 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
65 * support for mandatory and extensible security protections. This notice
66 * is included in support of clause 2.2 (b) of the Apple Public License,
67 * Version 2.0.
68 */
69
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/callout.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/domain.h>
78 #include <sys/proc.h>
79 #include <sys/kauth.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/protosw.h>
83 #include <sys/random.h>
84 #include <sys/syslog.h>
85 #include <sys/mcache.h>
86 #include <kern/locks.h>
87 #include <kern/zalloc.h>
88
89 #include <net/route.h>
90 #include <net/if.h>
91
92 #define tcp_minmssoverload fring
93 #define _IP_VHL
94 #include <netinet/in.h>
95 #include <netinet/in_systm.h>
96 #include <netinet/ip.h>
97 #include <netinet/ip_icmp.h>
98 #if INET6
99 #include <netinet/ip6.h>
100 #endif
101 #include <netinet/in_pcb.h>
102 #if INET6
103 #include <netinet6/in6_pcb.h>
104 #endif
105 #include <netinet/in_var.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/icmp_var.h>
108 #if INET6
109 #include <netinet6/ip6_var.h>
110 #endif
111 #include <netinet/tcp.h>
112 #include <netinet/tcp_fsm.h>
113 #include <netinet/tcp_seq.h>
114 #include <netinet/tcp_timer.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet/tcp_cc.h>
117 #include <kern/thread_call.h>
118
119 #if INET6
120 #include <netinet6/tcp6_var.h>
121 #endif
122 #include <netinet/tcpip.h>
123 #if TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126 #include <netinet6/ip6protosw.h>
127
128 #if IPSEC
129 #include <netinet6/ipsec.h>
130 #if INET6
131 #include <netinet6/ipsec6.h>
132 #endif
133 #endif /*IPSEC*/
134
135 #undef tcp_minmssoverload
136
137 #if CONFIG_MACF_NET
138 #include <security/mac_framework.h>
139 #endif /* MAC_NET */
140
141 #include <libkern/crypto/md5.h>
142 #include <sys/kdebug.h>
143 #include <mach/sdt.h>
144
145 #include <netinet/lro_ext.h>
146
147 #define DBG_FNC_TCP_CLOSE NETDBG_CODE(DBG_NETTCP, ((5 << 8) | 2))
148
149 extern int tcp_lq_overflow;
150
151 /* temporary: for testing */
152 #if IPSEC
153 extern int ipsec_bypass;
154 #endif
155
156 int tcp_mssdflt = TCP_MSS;
157 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW | CTLFLAG_LOCKED,
158 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
159
160 #if INET6
161 int tcp_v6mssdflt = TCP6_MSS;
162 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
163 CTLFLAG_RW | CTLFLAG_LOCKED, &tcp_v6mssdflt , 0,
164 "Default TCP Maximum Segment Size for IPv6");
165 #endif
166
167 extern int tcp_do_autorcvbuf;
168
169 /*
170 * Minimum MSS we accept and use. This prevents DoS attacks where
171 * we are forced to a ridiculous low MSS like 20 and send hundreds
172 * of packets instead of one. The effect scales with the available
173 * bandwidth and quickly saturates the CPU and network interface
174 * with packet generation and sending. Set to zero to disable MINMSS
175 * checking. This setting prevents us from sending too small packets.
176 */
177 int tcp_minmss = TCP_MINMSS;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW | CTLFLAG_LOCKED,
179 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
180
181 /*
182 * Number of TCP segments per second we accept from remote host
183 * before we start to calculate average segment size. If average
184 * segment size drops below the minimum TCP MSS we assume a DoS
185 * attack and reset+drop the connection. Care has to be taken not to
186 * set this value too small to not kill interactive type connections
187 * (telnet, SSH) which send many small packets.
188 */
189 #ifdef FIX_WORKAROUND_FOR_3894301
190 __private_extern__ int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
191 #else
192 __private_extern__ int tcp_minmssoverload = 0;
193 #endif
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW | CTLFLAG_LOCKED,
195 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
196 "be under the MINMSS Size");
197
198 static int tcp_do_rfc1323 = 1;
199 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW | CTLFLAG_LOCKED,
200 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
201
202 // Not used
203 static int tcp_do_rfc1644 = 0;
204 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW | CTLFLAG_LOCKED,
205 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
206
207 static int do_tcpdrain = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW | CTLFLAG_LOCKED, &do_tcpdrain, 0,
209 "Enable tcp_drain routine for extra help when low on mbufs");
210
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD | CTLFLAG_LOCKED,
212 &tcbinfo.ipi_count, 0, "Number of active PCBs");
213
214 static int icmp_may_rst = 1;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW | CTLFLAG_LOCKED, &icmp_may_rst, 0,
216 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
217
218 static int tcp_strict_rfc1948 = 0;
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, strict_rfc1948, CTLFLAG_RW | CTLFLAG_LOCKED,
220 &tcp_strict_rfc1948, 0, "Determines if RFC1948 is followed exactly");
221
222 static int tcp_isn_reseed_interval = 0;
223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW | CTLFLAG_LOCKED,
224 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
225 static int tcp_background_io_enabled = 1;
226 SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_io_enabled, CTLFLAG_RW | CTLFLAG_LOCKED,
227 &tcp_background_io_enabled, 0, "Background IO Enabled");
228
229 int tcp_TCPTV_MIN = 100; /* 100ms minimum RTT */
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rtt_min, CTLFLAG_RW | CTLFLAG_LOCKED,
231 &tcp_TCPTV_MIN, 0, "min rtt value allowed");
232
233 int tcp_rexmt_slop = TCPTV_REXMTSLOP;
234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rexmt_slop, CTLFLAG_RW,
235 &tcp_rexmt_slop, 0, "Slop added to retransmit timeout");
236
237 __private_extern__ int tcp_use_randomport = 0;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, randomize_ports, CTLFLAG_RW | CTLFLAG_LOCKED,
239 &tcp_use_randomport, 0, "Randomize TCP port numbers");
240
241 extern struct tcp_cc_algo tcp_cc_newreno;
242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newreno_sockets, CTLFLAG_RD | CTLFLAG_LOCKED,
243 &tcp_cc_newreno.num_sockets, 0, "Number of sockets using newreno");
244
245 extern struct tcp_cc_algo tcp_cc_ledbat;
246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, background_sockets, CTLFLAG_RD | CTLFLAG_LOCKED,
247 &tcp_cc_ledbat.num_sockets, 0, "Number of sockets using background transport");
248
249 __private_extern__ int tcp_win_scale = 3;
250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, win_scale_factor, CTLFLAG_RW | CTLFLAG_LOCKED,
251 &tcp_win_scale, 0, "Window scaling factor");
252
253 static void tcp_cleartaocache(void);
254 static void tcp_notify(struct inpcb *, int);
255 static void tcp_cc_init(void);
256
257 struct zone *sack_hole_zone;
258 struct zone *tcp_reass_zone;
259 struct zone *tcp_bwmeas_zone;
260
261 /* The array containing pointers to currently implemented TCP CC algorithms */
262 struct tcp_cc_algo* tcp_cc_algo_list[TCP_CC_ALGO_COUNT];
263
264 extern int slowlink_wsize; /* window correction for slow links */
265 extern int path_mtu_discovery;
266
267 extern u_int32_t tcp_autorcvbuf_max;
268 extern u_int32_t tcp_autorcvbuf_inc_shift;
269 static void tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb);
270
271 #define TCP_BWMEAS_BURST_MINSIZE 6
272 #define TCP_BWMEAS_BURST_MAXSIZE 25
273
274 static uint32_t bwmeas_elm_size;
275
276 /*
277 * Target size of TCP PCB hash tables. Must be a power of two.
278 *
279 * Note that this can be overridden by the kernel environment
280 * variable net.inet.tcp.tcbhashsize
281 */
282 #ifndef TCBHASHSIZE
283 #define TCBHASHSIZE CONFIG_TCBHASHSIZE
284 #endif
285
286 __private_extern__ int tcp_tcbhashsize = TCBHASHSIZE;
287 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD | CTLFLAG_LOCKED,
288 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
289
290 /*
291 * This is the actual shape of what we allocate using the zone
292 * allocator. Doing it this way allows us to protect both structures
293 * using the same generation count, and also eliminates the overhead
294 * of allocating tcpcbs separately. By hiding the structure here,
295 * we avoid changing most of the rest of the code (although it needs
296 * to be changed, eventually, for greater efficiency).
297 */
298 #define ALIGNMENT 32
299 struct inp_tp {
300 struct inpcb inp;
301 struct tcpcb tcb __attribute__((aligned(ALIGNMENT)));
302 };
303 #undef ALIGNMENT
304
305 extern struct inpcbhead time_wait_slots[];
306 extern struct tcptimerlist tcp_timer_list;
307
308 int get_inpcb_str_size(void);
309 int get_tcp_str_size(void);
310
311 static void tcpcb_to_otcpcb(struct tcpcb *, struct otcpcb *);
312
313 static lck_attr_t *tcp_uptime_mtx_attr = NULL; /* mutex attributes */
314 static lck_grp_t *tcp_uptime_mtx_grp = NULL; /* mutex group definition */
315 static lck_grp_attr_t *tcp_uptime_mtx_grp_attr = NULL; /* mutex group attributes */
316 int tcp_notsent_lowat_check(struct socket *so);
317
318
319 int get_inpcb_str_size(void)
320 {
321 return sizeof(struct inpcb);
322 }
323
324
325 int get_tcp_str_size(void)
326 {
327 return sizeof(struct tcpcb);
328 }
329
330 int tcp_freeq(struct tcpcb *tp);
331
332 /*
333 * Initialize TCP congestion control algorithms.
334 */
335
336 void
337 tcp_cc_init(void)
338 {
339 bzero(&tcp_cc_algo_list, sizeof(tcp_cc_algo_list));
340 tcp_cc_algo_list[TCP_CC_ALGO_NEWRENO_INDEX] = &tcp_cc_newreno;
341 tcp_cc_algo_list[TCP_CC_ALGO_BACKGROUND_INDEX] = &tcp_cc_ledbat;
342 }
343
344 /*
345 * Tcp initialization
346 */
347 void
348 tcp_init()
349 {
350 vm_size_t str_size;
351 int i;
352 struct inpcbinfo *pcbinfo;
353
354 tcp_ccgen = 1;
355 tcp_cleartaocache();
356
357 tcp_keepinit = TCPTV_KEEP_INIT;
358 tcp_keepidle = TCPTV_KEEP_IDLE;
359 tcp_keepintvl = TCPTV_KEEPINTVL;
360 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
361 tcp_msl = TCPTV_MSL;
362
363 microuptime(&tcp_uptime);
364 read_random(&tcp_now, sizeof(tcp_now));
365 tcp_now = tcp_now & 0x3fffffff; /* Starts tcp internal clock at a random value */
366
367 LIST_INIT(&tcb);
368 tcbinfo.listhead = &tcb;
369 pcbinfo = &tcbinfo;
370 if (!powerof2(tcp_tcbhashsize)) {
371 printf("WARNING: TCB hash size not a power of 2\n");
372 tcp_tcbhashsize = 512; /* safe default */
373 }
374 tcbinfo.hashsize = tcp_tcbhashsize;
375 tcbinfo.hashbase = hashinit(tcp_tcbhashsize, M_PCB, &tcbinfo.hashmask);
376 tcbinfo.porthashbase = hashinit(tcp_tcbhashsize, M_PCB,
377 &tcbinfo.porthashmask);
378 str_size = P2ROUNDUP(sizeof(struct inp_tp), sizeof(u_int64_t));
379 tcbinfo.ipi_zone = (void *) zinit(str_size, 120000*str_size, 8192, "tcpcb");
380 zone_change(tcbinfo.ipi_zone, Z_CALLERACCT, FALSE);
381 zone_change(tcbinfo.ipi_zone, Z_EXPAND, TRUE);
382
383 str_size = P2ROUNDUP(sizeof(struct sackhole), sizeof(u_int64_t));
384 sack_hole_zone = zinit(str_size, 120000*str_size, 8192, "sack_hole zone");
385 zone_change(sack_hole_zone, Z_CALLERACCT, FALSE);
386 zone_change(sack_hole_zone, Z_EXPAND, TRUE);
387
388 tcp_reass_maxseg = nmbclusters / 16;
389 str_size = P2ROUNDUP(sizeof(struct tseg_qent), sizeof(u_int64_t));
390 tcp_reass_zone = zinit(str_size, (tcp_reass_maxseg + 1) * str_size,
391 0, "tcp_reass_zone");
392 if (tcp_reass_zone == NULL) {
393 panic("%s: failed allocating tcp_reass_zone", __func__);
394 /* NOTREACHED */
395 }
396 zone_change(tcp_reass_zone, Z_CALLERACCT, FALSE);
397 zone_change(tcp_reass_zone, Z_EXPAND, TRUE);
398
399 bwmeas_elm_size = P2ROUNDUP(sizeof(struct bwmeas), sizeof(u_int64_t));
400 tcp_bwmeas_zone = zinit(bwmeas_elm_size, (100 * bwmeas_elm_size), 0, "tcp_bwmeas_zone");
401 if (tcp_bwmeas_zone == NULL) {
402 panic("%s: failed allocating tcp_bwmeas_zone", __func__);
403 /* NOTREACHED */
404 }
405 zone_change(tcp_bwmeas_zone, Z_CALLERACCT, FALSE);
406 zone_change(tcp_bwmeas_zone, Z_EXPAND, TRUE);
407
408 #if INET6
409 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
410 #else /* INET6 */
411 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
412 #endif /* INET6 */
413 if (max_protohdr < TCP_MINPROTOHDR) {
414 _max_protohdr = TCP_MINPROTOHDR;
415 _max_protohdr = max_protohdr; /* round it up */
416 }
417 if (max_linkhdr + max_protohdr > MCLBYTES)
418 panic("tcp_init");
419 #undef TCP_MINPROTOHDR
420
421 /*
422 * allocate lock group attribute and group for tcp pcb mutexes
423 */
424 pcbinfo->mtx_grp_attr = lck_grp_attr_alloc_init();
425 pcbinfo->mtx_grp = lck_grp_alloc_init("tcppcb", pcbinfo->mtx_grp_attr);
426
427 /*
428 * allocate the lock attribute for tcp pcb mutexes
429 */
430 pcbinfo->mtx_attr = lck_attr_alloc_init();
431
432 if ((pcbinfo->mtx = lck_rw_alloc_init(pcbinfo->mtx_grp, pcbinfo->mtx_attr)) == NULL) {
433 printf("tcp_init: mutex not alloced!\n");
434 return; /* pretty much dead if this fails... */
435 }
436
437 for (i=0; i < N_TIME_WAIT_SLOTS; i++) {
438 LIST_INIT(&time_wait_slots[i]);
439 }
440
441 bzero(&tcp_timer_list, sizeof(tcp_timer_list));
442 LIST_INIT(&tcp_timer_list.lhead);
443 /*
444 * allocate lock group attribute, group and attribute for the tcp timer list
445 */
446 tcp_timer_list.mtx_grp_attr = lck_grp_attr_alloc_init();
447 tcp_timer_list.mtx_grp = lck_grp_alloc_init("tcptimerlist", tcp_timer_list.mtx_grp_attr);
448 tcp_timer_list.mtx_attr = lck_attr_alloc_init();
449 if ((tcp_timer_list.mtx = lck_mtx_alloc_init(tcp_timer_list.mtx_grp, tcp_timer_list.mtx_attr)) == NULL) {
450 panic("failed to allocate memory for tcp_timer_list.mtx\n");
451 };
452 tcp_timer_list.fast_quantum = TCP_FASTTIMER_QUANTUM;
453 tcp_timer_list.slow_quantum = TCP_SLOWTIMER_QUANTUM;
454 if ((tcp_timer_list.call = thread_call_allocate(tcp_run_timerlist, NULL)) == NULL) {
455 panic("failed to allocate call entry 1 in tcp_init\n");
456 }
457
458 /*
459 * allocate lock group attribute, group and attribute for tcp_uptime_lock
460 */
461 tcp_uptime_mtx_grp_attr = lck_grp_attr_alloc_init();
462 tcp_uptime_mtx_grp = lck_grp_alloc_init("tcpuptime", tcp_uptime_mtx_grp_attr);
463 tcp_uptime_mtx_attr = lck_attr_alloc_init();
464 tcp_uptime_lock = lck_spin_alloc_init(tcp_uptime_mtx_grp, tcp_uptime_mtx_attr);
465
466 /* Initialize TCP congestion control algorithms list */
467 tcp_cc_init();
468
469 /* Initialize TCP LRO data structures */
470 tcp_lro_init();
471 }
472
473 /*
474 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
475 * tcp_template used to store this data in mbufs, but we now recopy it out
476 * of the tcpcb each time to conserve mbufs.
477 */
478 void
479 tcp_fillheaders(tp, ip_ptr, tcp_ptr)
480 struct tcpcb *tp;
481 void *ip_ptr;
482 void *tcp_ptr;
483 {
484 struct inpcb *inp = tp->t_inpcb;
485 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
486
487 #if INET6
488 if ((inp->inp_vflag & INP_IPV6) != 0) {
489 struct ip6_hdr *ip6;
490
491 ip6 = (struct ip6_hdr *)ip_ptr;
492 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
493 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
494 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
495 (IPV6_VERSION & IPV6_VERSION_MASK);
496 ip6->ip6_nxt = IPPROTO_TCP;
497 ip6->ip6_plen = sizeof(struct tcphdr);
498 ip6->ip6_src = inp->in6p_laddr;
499 ip6->ip6_dst = inp->in6p_faddr;
500 tcp_hdr->th_sum = in6_cksum_phdr(&inp->in6p_laddr,
501 &inp->in6p_faddr, htonl(sizeof(struct tcphdr)),
502 htonl(IPPROTO_TCP));
503 } else
504 #endif
505 {
506 struct ip *ip = (struct ip *) ip_ptr;
507
508 ip->ip_vhl = IP_VHL_BORING;
509 ip->ip_tos = 0;
510 ip->ip_len = 0;
511 ip->ip_id = 0;
512 ip->ip_off = 0;
513 ip->ip_ttl = 0;
514 ip->ip_sum = 0;
515 ip->ip_p = IPPROTO_TCP;
516 ip->ip_src = inp->inp_laddr;
517 ip->ip_dst = inp->inp_faddr;
518 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
519 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
520 }
521
522 tcp_hdr->th_sport = inp->inp_lport;
523 tcp_hdr->th_dport = inp->inp_fport;
524 tcp_hdr->th_seq = 0;
525 tcp_hdr->th_ack = 0;
526 tcp_hdr->th_x2 = 0;
527 tcp_hdr->th_off = 5;
528 tcp_hdr->th_flags = 0;
529 tcp_hdr->th_win = 0;
530 tcp_hdr->th_urp = 0;
531 }
532
533 /*
534 * Create template to be used to send tcp packets on a connection.
535 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
536 * use for this function is in keepalives, which use tcp_respond.
537 */
538 struct tcptemp *
539 tcp_maketemplate(tp)
540 struct tcpcb *tp;
541 {
542 struct mbuf *m;
543 struct tcptemp *n;
544
545 m = m_get(M_DONTWAIT, MT_HEADER);
546 if (m == NULL)
547 return (0);
548 m->m_len = sizeof(struct tcptemp);
549 n = mtod(m, struct tcptemp *);
550
551 tcp_fillheaders(tp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
552 return (n);
553 }
554
555 /*
556 * Send a single message to the TCP at address specified by
557 * the given TCP/IP header. If m == 0, then we make a copy
558 * of the tcpiphdr at ti and send directly to the addressed host.
559 * This is used to force keep alive messages out using the TCP
560 * template for a connection. If flags are given then we send
561 * a message back to the TCP which originated the * segment ti,
562 * and discard the mbuf containing it and any other attached mbufs.
563 *
564 * In any case the ack and sequence number of the transmitted
565 * segment are as specified by the parameters.
566 *
567 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
568 */
569 void
570 tcp_respond(
571 struct tcpcb *tp,
572 void *ipgen,
573 register struct tcphdr *th,
574 register struct mbuf *m,
575 tcp_seq ack,
576 tcp_seq seq,
577 int flags,
578 unsigned int ifscope,
579 unsigned int nocell
580 )
581 {
582 register int tlen;
583 int win = 0;
584 struct route *ro = 0;
585 struct route sro;
586 struct ip *ip;
587 struct tcphdr *nth;
588 #if INET6
589 struct route_in6 *ro6 = 0;
590 struct route_in6 sro6;
591 struct ip6_hdr *ip6;
592 int isipv6;
593 #endif /* INET6 */
594 struct ifnet *outif;
595
596 #if INET6
597 isipv6 = IP_VHL_V(((struct ip *)ipgen)->ip_vhl) == 6;
598 ip6 = ipgen;
599 #endif /* INET6 */
600 ip = ipgen;
601
602 if (tp) {
603 if (!(flags & TH_RST)) {
604 win = tcp_sbspace(tp);
605 if (win > (int32_t)TCP_MAXWIN << tp->rcv_scale)
606 win = (int32_t)TCP_MAXWIN << tp->rcv_scale;
607 }
608 #if INET6
609 if (isipv6)
610 ro6 = &tp->t_inpcb->in6p_route;
611 else
612 #endif /* INET6 */
613 ro = &tp->t_inpcb->inp_route;
614 } else {
615 #if INET6
616 if (isipv6) {
617 ro6 = &sro6;
618 bzero(ro6, sizeof *ro6);
619 } else
620 #endif /* INET6 */
621 {
622 ro = &sro;
623 bzero(ro, sizeof *ro);
624 }
625 }
626 if (m == 0) {
627 m = m_gethdr(M_DONTWAIT, MT_HEADER); /* MAC-OK */
628 if (m == NULL)
629 return;
630 tlen = 0;
631 m->m_data += max_linkhdr;
632 #if INET6
633 if (isipv6) {
634 bcopy((caddr_t)ip6, mtod(m, caddr_t),
635 sizeof(struct ip6_hdr));
636 ip6 = mtod(m, struct ip6_hdr *);
637 nth = (struct tcphdr *)(void *)(ip6 + 1);
638 } else
639 #endif /* INET6 */
640 {
641 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
642 ip = mtod(m, struct ip *);
643 nth = (struct tcphdr *)(void *)(ip + 1);
644 }
645 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
646 flags = TH_ACK;
647 } else {
648 m_freem(m->m_next);
649 m->m_next = 0;
650 m->m_data = (caddr_t)ipgen;
651 /* m_len is set later */
652 tlen = 0;
653 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
654 #if INET6
655 if (isipv6) {
656 /* Expect 32-bit aligned IP on strict-align platforms */
657 IP6_HDR_STRICT_ALIGNMENT_CHECK(ip6);
658 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
659 nth = (struct tcphdr *)(void *)(ip6 + 1);
660 } else
661 #endif /* INET6 */
662 {
663 /* Expect 32-bit aligned IP on strict-align platforms */
664 IP_HDR_STRICT_ALIGNMENT_CHECK(ip);
665 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
666 nth = (struct tcphdr *)(void *)(ip + 1);
667 }
668 if (th != nth) {
669 /*
670 * this is usually a case when an extension header
671 * exists between the IPv6 header and the
672 * TCP header.
673 */
674 nth->th_sport = th->th_sport;
675 nth->th_dport = th->th_dport;
676 }
677 xchg(nth->th_dport, nth->th_sport, n_short);
678 #undef xchg
679 }
680 #if INET6
681 if (isipv6) {
682 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
683 tlen));
684 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
685 } else
686 #endif
687 {
688 tlen += sizeof (struct tcpiphdr);
689 ip->ip_len = tlen;
690 ip->ip_ttl = ip_defttl;
691 }
692 m->m_len = tlen;
693 m->m_pkthdr.len = tlen;
694 m->m_pkthdr.rcvif = 0;
695 #if CONFIG_MACF_NET
696 if (tp != NULL && tp->t_inpcb != NULL) {
697 /*
698 * Packet is associated with a socket, so allow the
699 * label of the response to reflect the socket label.
700 */
701 mac_mbuf_label_associate_inpcb(tp->t_inpcb, m);
702 } else {
703 /*
704 * Packet is not associated with a socket, so possibly
705 * update the label in place.
706 */
707 mac_netinet_tcp_reply(m);
708 }
709 #endif
710
711 nth->th_seq = htonl(seq);
712 nth->th_ack = htonl(ack);
713 nth->th_x2 = 0;
714 nth->th_off = sizeof (struct tcphdr) >> 2;
715 nth->th_flags = flags;
716 if (tp)
717 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
718 else
719 nth->th_win = htons((u_short)win);
720 nth->th_urp = 0;
721 #if INET6
722 if (isipv6) {
723 nth->th_sum = 0;
724 nth->th_sum = in6_cksum_phdr(&ip6->ip6_src,
725 &ip6->ip6_dst, htons((u_short)(tlen - sizeof(struct ip6_hdr))),
726 htonl(IPPROTO_TCP));
727 m->m_pkthdr.csum_flags = CSUM_TCPIPV6;
728 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
729 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
730 ro6 && ro6->ro_rt ?
731 ro6->ro_rt->rt_ifp :
732 NULL);
733 } else
734 #endif /* INET6 */
735 {
736 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
737 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
738 m->m_pkthdr.csum_flags = CSUM_TCP;
739 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
740 }
741 #if TCPDEBUG
742 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
743 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
744 #endif
745 #if IPSEC
746 if (ipsec_bypass == 0 && ipsec_setsocket(m, tp ? tp->t_inpcb->inp_socket : NULL) != 0) {
747 m_freem(m);
748 return;
749 }
750 #endif
751
752 if (tp != NULL) {
753 u_int32_t svc_flags = 0;
754 if (isipv6) {
755 svc_flags |= PKT_SCF_IPV6;
756 }
757 set_packet_service_class(m, tp->t_inpcb->inp_socket,
758 MBUF_SC_UNSPEC, svc_flags);
759
760 /* Embed flowhash and flow control flags */
761 m->m_pkthdr.m_flowhash = tp->t_inpcb->inp_flowhash;
762 m->m_pkthdr.m_fhflags |=
763 (PF_TAG_TCP | PF_TAG_FLOWHASH | PF_TAG_FLOWADV);
764 }
765
766 #if INET6
767 if (isipv6) {
768 struct ip6_out_args ip6oa = { ifscope, { 0 },
769 IP6OAF_SELECT_SRCIF | IP6OAF_BOUND_SRCADDR };
770
771 if (ifscope != IFSCOPE_NONE)
772 ip6oa.ip6oa_flags |= IP6OAF_BOUND_IF;
773 if (nocell)
774 ip6oa.ip6oa_flags |= IP6OAF_NO_CELLULAR;
775
776 (void) ip6_output(m, NULL, ro6, IPV6_OUTARGS, NULL,
777 NULL, &ip6oa);
778 if (ro6->ro_rt != NULL) {
779 if (ro6 == &sro6) {
780 rtfree(ro6->ro_rt);
781 ro6->ro_rt = NULL;
782 } else if ((outif = ro6->ro_rt->rt_ifp) !=
783 tp->t_inpcb->in6p_last_outifp) {
784 tp->t_inpcb->in6p_last_outifp = outif;
785 }
786 }
787 } else
788 #endif /* INET6 */
789 {
790 struct ip_out_args ipoa = { ifscope, { 0 },
791 IPOAF_SELECT_SRCIF | IPOAF_BOUND_SRCADDR };
792
793 if (ifscope != IFSCOPE_NONE)
794 ipoa.ipoa_flags |= IPOAF_BOUND_IF;
795 if (nocell)
796 ipoa.ipoa_flags |= IPOAF_NO_CELLULAR;
797
798 if (ro != &sro) {
799 /* Copy the cached route and take an extra reference */
800 inp_route_copyout(tp->t_inpcb, &sro);
801 }
802 /*
803 * For consistency, pass a local route copy.
804 */
805 (void) ip_output(m, NULL, &sro, IP_OUTARGS, NULL, &ipoa);
806
807 if (ro != &sro) {
808 if (sro.ro_rt != NULL &&
809 (outif = sro.ro_rt->rt_ifp) !=
810 tp->t_inpcb->inp_last_outifp)
811 tp->t_inpcb->inp_last_outifp = outif;
812 /* Synchronize cached PCB route */
813 inp_route_copyin(tp->t_inpcb, &sro);
814 } else if (sro.ro_rt != NULL) {
815 rtfree(sro.ro_rt);
816 }
817 }
818 }
819
820 /*
821 * Create a new TCP control block, making an
822 * empty reassembly queue and hooking it to the argument
823 * protocol control block. The `inp' parameter must have
824 * come from the zone allocator set up in tcp_init().
825 */
826 struct tcpcb *
827 tcp_newtcpcb(inp)
828 struct inpcb *inp;
829 {
830 struct inp_tp *it;
831 register struct tcpcb *tp;
832 register struct socket *so = inp->inp_socket;
833 #if INET6
834 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
835 #endif /* INET6 */
836
837 calculate_tcp_clock();
838
839 if (so->cached_in_sock_layer == 0) {
840 it = (struct inp_tp *)(void *)inp;
841 tp = &it->tcb;
842 }
843 else
844 tp = (struct tcpcb *)(void *)inp->inp_saved_ppcb;
845
846 bzero((char *) tp, sizeof(struct tcpcb));
847 LIST_INIT(&tp->t_segq);
848 tp->t_maxseg = tp->t_maxopd =
849 #if INET6
850 isipv6 ? tcp_v6mssdflt :
851 #endif /* INET6 */
852 tcp_mssdflt;
853
854 if (tcp_do_rfc1323)
855 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
856 tp->sack_enable = tcp_do_sack;
857 TAILQ_INIT(&tp->snd_holes);
858 tp->t_inpcb = inp; /* XXX */
859 /*
860 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
861 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
862 * reasonable initial retransmit time.
863 */
864 tp->t_srtt = TCPTV_SRTTBASE;
865 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
866 tp->t_rttmin = tcp_TCPTV_MIN;
867 tp->t_rxtcur = TCPTV_RTOBASE;
868
869 /* Initialize congestion control algorithm for this connection
870 * to newreno by default
871 */
872 tp->tcp_cc_index = TCP_CC_ALGO_NEWRENO_INDEX;
873 if (CC_ALGO(tp)->init != NULL) {
874 CC_ALGO(tp)->init(tp);
875 }
876
877 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
878 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
879 tp->snd_ssthresh_prev = TCP_MAXWIN << TCP_MAX_WINSHIFT;
880 tp->t_rcvtime = tcp_now;
881 tp->tentry.timer_start = tcp_now;
882 tp->t_persist_timeout = tcp_max_persist_timeout;
883 tp->t_persist_stop = 0;
884 tp->t_flagsext |= TF_RCVUNACK_WAITSS;
885 /*
886 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
887 * because the socket may be bound to an IPv6 wildcard address,
888 * which may match an IPv4-mapped IPv6 address.
889 */
890 inp->inp_ip_ttl = ip_defttl;
891 inp->inp_ppcb = (caddr_t)tp;
892 return (tp); /* XXX */
893 }
894
895 /*
896 * Drop a TCP connection, reporting
897 * the specified error. If connection is synchronized,
898 * then send a RST to peer.
899 */
900 struct tcpcb *
901 tcp_drop(tp, errno)
902 register struct tcpcb *tp;
903 int errno;
904 {
905 struct socket *so = tp->t_inpcb->inp_socket;
906 #if CONFIG_DTRACE
907 struct inpcb *inp = tp->t_inpcb;
908 #endif
909
910 if (TCPS_HAVERCVDSYN(tp->t_state)) {
911 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
912 struct tcpcb *, tp, int32_t, TCPS_CLOSED);
913 tp->t_state = TCPS_CLOSED;
914 (void) tcp_output(tp);
915 tcpstat.tcps_drops++;
916 } else
917 tcpstat.tcps_conndrops++;
918 if (errno == ETIMEDOUT && tp->t_softerror)
919 errno = tp->t_softerror;
920 so->so_error = errno;
921 return (tcp_close(tp));
922 }
923
924 void
925 tcp_getrt_rtt(struct tcpcb *tp, struct rtentry *rt)
926 {
927 u_int32_t rtt = rt->rt_rmx.rmx_rtt;
928 int isnetlocal = (tp->t_flags & TF_LOCAL);
929
930 if (rtt != 0) {
931 /*
932 * XXX the lock bit for RTT indicates that the value
933 * is also a minimum value; this is subject to time.
934 */
935 if (rt->rt_rmx.rmx_locks & RTV_RTT)
936 tp->t_rttmin = rtt / (RTM_RTTUNIT / TCP_RETRANSHZ);
937 else
938 tp->t_rttmin = isnetlocal ? tcp_TCPTV_MIN : TCPTV_REXMTMIN;
939 tp->t_srtt = rtt / (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
940 tcpstat.tcps_usedrtt++;
941 if (rt->rt_rmx.rmx_rttvar) {
942 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
943 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
944 tcpstat.tcps_usedrttvar++;
945 } else {
946 /* default variation is +- 1 rtt */
947 tp->t_rttvar =
948 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
949 }
950 TCPT_RANGESET(tp->t_rxtcur,
951 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
952 tp->t_rttmin, TCPTV_REXMTMAX,
953 TCP_ADD_REXMTSLOP(tp));
954 }
955 }
956
957 /*
958 * Close a TCP control block:
959 * discard all space held by the tcp
960 * discard internet protocol block
961 * wake up any sleepers
962 */
963 struct tcpcb *
964 tcp_close(tp)
965 register struct tcpcb *tp;
966 {
967 struct inpcb *inp = tp->t_inpcb;
968 struct socket *so = inp->inp_socket;
969 #if INET6
970 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
971 #endif /* INET6 */
972 struct rtentry *rt;
973 int dosavessthresh;
974
975 if ( inp->inp_ppcb == NULL) /* tcp_close was called previously, bail */
976 return(NULL);
977
978 tcp_canceltimers(tp);
979 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_START, tp,0,0,0,0);
980
981 /*
982 * If another thread for this tcp is currently in ip (indicated by
983 * the TF_SENDINPROG flag), defer the cleanup until after it returns
984 * back to tcp. This is done to serialize the close until after all
985 * pending output is finished, in order to avoid having the PCB be
986 * detached and the cached route cleaned, only for ip to cache the
987 * route back into the PCB again. Note that we've cleared all the
988 * timers at this point. Set TF_CLOSING to indicate to tcp_output()
989 * that is should call us again once it returns from ip; at that
990 * point both flags should be cleared and we can proceed further
991 * with the cleanup.
992 */
993 if ((tp->t_flags & TF_CLOSING) ||
994 inp->inp_sndinprog_cnt > 0) {
995 tp->t_flags |= TF_CLOSING;
996 return (NULL);
997 }
998
999 if (CC_ALGO(tp)->cleanup != NULL) {
1000 CC_ALGO(tp)->cleanup(tp);
1001 }
1002
1003 #if INET6
1004 rt = isipv6 ? inp->in6p_route.ro_rt : inp->inp_route.ro_rt;
1005 #else
1006 rt = inp->inp_route.ro_rt;
1007 #endif
1008 if (rt != NULL)
1009 RT_LOCK_SPIN(rt);
1010
1011 /*
1012 * If we got enough samples through the srtt filter,
1013 * save the rtt and rttvar in the routing entry.
1014 * 'Enough' is arbitrarily defined as the 16 samples.
1015 * 16 samples is enough for the srtt filter to converge
1016 * to within 5% of the correct value; fewer samples and
1017 * we could save a very bogus rtt.
1018 *
1019 * Don't update the default route's characteristics and don't
1020 * update anything that the user "locked".
1021 */
1022 if (tp->t_rttupdated >= 16) {
1023 register u_int32_t i = 0;
1024
1025 #if INET6
1026 if (isipv6) {
1027 struct sockaddr_in6 *sin6;
1028
1029 if (rt == NULL)
1030 goto no_valid_rt;
1031 sin6 = (struct sockaddr_in6 *)(void *)rt_key(rt);
1032 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
1033 goto no_valid_rt;
1034 }
1035 else
1036 #endif /* INET6 */
1037 if (rt == NULL || !(rt->rt_flags & RTF_UP) ||
1038 ((struct sockaddr_in *)(void *)rt_key(rt))->sin_addr.s_addr ==
1039 INADDR_ANY || rt->generation_id != route_generation) {
1040 if (tp->t_state >= TCPS_CLOSE_WAIT) {
1041 DTRACE_TCP4(state__change, void, NULL, struct inpcb *, inp,
1042 struct tcpcb *, tp, int32_t, TCPS_CLOSING);
1043 tp->t_state = TCPS_CLOSING;
1044 }
1045 goto no_valid_rt;
1046 }
1047
1048 RT_LOCK_ASSERT_HELD(rt);
1049 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1050 i = tp->t_srtt *
1051 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTT_SCALE));
1052 if (rt->rt_rmx.rmx_rtt && i)
1053 /*
1054 * filter this update to half the old & half
1055 * the new values, converting scale.
1056 * See route.h and tcp_var.h for a
1057 * description of the scaling constants.
1058 */
1059 rt->rt_rmx.rmx_rtt =
1060 (rt->rt_rmx.rmx_rtt + i) / 2;
1061 else
1062 rt->rt_rmx.rmx_rtt = i;
1063 tcpstat.tcps_cachedrtt++;
1064 }
1065 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1066 i = tp->t_rttvar *
1067 (RTM_RTTUNIT / (TCP_RETRANSHZ * TCP_RTTVAR_SCALE));
1068 if (rt->rt_rmx.rmx_rttvar && i)
1069 rt->rt_rmx.rmx_rttvar =
1070 (rt->rt_rmx.rmx_rttvar + i) / 2;
1071 else
1072 rt->rt_rmx.rmx_rttvar = i;
1073 tcpstat.tcps_cachedrttvar++;
1074 }
1075 /*
1076 * The old comment here said:
1077 * update the pipelimit (ssthresh) if it has been updated
1078 * already or if a pipesize was specified & the threshhold
1079 * got below half the pipesize. I.e., wait for bad news
1080 * before we start updating, then update on both good
1081 * and bad news.
1082 *
1083 * But we want to save the ssthresh even if no pipesize is
1084 * specified explicitly in the route, because such
1085 * connections still have an implicit pipesize specified
1086 * by the global tcp_sendspace. In the absence of a reliable
1087 * way to calculate the pipesize, it will have to do.
1088 */
1089 i = tp->snd_ssthresh;
1090 if (rt->rt_rmx.rmx_sendpipe != 0)
1091 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe / 2);
1092 else
1093 dosavessthresh = (i < so->so_snd.sb_hiwat / 2);
1094 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1095 i != 0 && rt->rt_rmx.rmx_ssthresh != 0)
1096 || dosavessthresh) {
1097 /*
1098 * convert the limit from user data bytes to
1099 * packets then to packet data bytes.
1100 */
1101 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
1102 if (i < 2)
1103 i = 2;
1104 i *= (u_int32_t)(tp->t_maxseg +
1105 #if INET6
1106 (isipv6 ? sizeof (struct ip6_hdr) +
1107 sizeof (struct tcphdr) :
1108 #endif
1109 sizeof (struct tcpiphdr)
1110 #if INET6
1111 )
1112 #endif
1113 );
1114 if (rt->rt_rmx.rmx_ssthresh)
1115 rt->rt_rmx.rmx_ssthresh =
1116 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1117 else
1118 rt->rt_rmx.rmx_ssthresh = i;
1119 tcpstat.tcps_cachedssthresh++;
1120 }
1121 }
1122
1123 /*
1124 * Mark route for deletion if no information is cached.
1125 */
1126 if (rt != NULL && (so->so_flags & SOF_OVERFLOW) && tcp_lq_overflow) {
1127 if (!(rt->rt_rmx.rmx_locks & RTV_RTT) &&
1128 rt->rt_rmx.rmx_rtt == 0) {
1129 rt->rt_flags |= RTF_DELCLONE;
1130 }
1131 }
1132
1133 no_valid_rt:
1134 if (rt != NULL)
1135 RT_UNLOCK(rt);
1136
1137 /* free the reassembly queue, if any */
1138 (void) tcp_freeq(tp);
1139
1140 tcp_free_sackholes(tp);
1141 if (tp->t_bwmeas != NULL) {
1142 tcp_bwmeas_free(tp);
1143 }
1144
1145 /* Free the packet list */
1146 if (tp->t_pktlist_head != NULL)
1147 m_freem_list(tp->t_pktlist_head);
1148 TCP_PKTLIST_CLEAR(tp);
1149
1150 #ifdef __APPLE__
1151 if (so->cached_in_sock_layer)
1152 inp->inp_saved_ppcb = (caddr_t) tp;
1153 #endif
1154 /* Issue a wakeup before detach so that we don't miss
1155 * a wakeup
1156 */
1157 sodisconnectwakeup(so);
1158
1159 /*
1160 * Clean up any LRO state
1161 */
1162 if (tp->t_flagsext & TF_LRO_OFFLOADED) {
1163 tcp_lro_remove_state(inp->inp_laddr, inp->inp_faddr,
1164 inp->inp_lport,
1165 inp->inp_fport);
1166 tp->t_flagsext &= ~TF_LRO_OFFLOADED;
1167 }
1168
1169 #if INET6
1170 if (INP_CHECK_SOCKAF(so, AF_INET6))
1171 in6_pcbdetach(inp);
1172 else
1173 #endif /* INET6 */
1174 in_pcbdetach(inp);
1175
1176 /* Call soisdisconnected after detach because it might unlock the socket */
1177 soisdisconnected(so);
1178 tcpstat.tcps_closed++;
1179 KERNEL_DEBUG(DBG_FNC_TCP_CLOSE | DBG_FUNC_END, tcpstat.tcps_closed,0,0,0,0);
1180 return(NULL);
1181 }
1182
1183 int
1184 tcp_freeq(tp)
1185 struct tcpcb *tp;
1186 {
1187
1188 register struct tseg_qent *q;
1189 int rv = 0;
1190
1191 while((q = LIST_FIRST(&tp->t_segq)) != NULL) {
1192 LIST_REMOVE(q, tqe_q);
1193 m_freem(q->tqe_m);
1194 zfree(tcp_reass_zone, q);
1195 tcp_reass_qsize--;
1196 rv = 1;
1197 }
1198 return (rv);
1199 }
1200
1201 void
1202 tcp_drain()
1203 {
1204 if (do_tcpdrain)
1205 {
1206 struct inpcb *inpb;
1207 struct tcpcb *tcpb;
1208 struct tseg_qent *te;
1209
1210 /*
1211 * Walk the tcpbs, if existing, and flush the reassembly queue,
1212 * if there is one...
1213 * XXX: The "Net/3" implementation doesn't imply that the TCP
1214 * reassembly queue should be flushed, but in a situation
1215 * where we're really low on mbufs, this is potentially
1216 * usefull.
1217 */
1218 if (!lck_rw_try_lock_exclusive(tcbinfo.mtx)) /* do it next time if the lock is in use */
1219 return;
1220
1221 for (inpb = LIST_FIRST(tcbinfo.listhead); inpb;
1222 inpb = LIST_NEXT(inpb, inp_list)) {
1223 if ((tcpb = intotcpcb(inpb))) {
1224 while ((te = LIST_FIRST(&tcpb->t_segq))
1225 != NULL) {
1226 LIST_REMOVE(te, tqe_q);
1227 m_freem(te->tqe_m);
1228 zfree(tcp_reass_zone, te);
1229 tcp_reass_qsize--;
1230 }
1231 }
1232 }
1233 lck_rw_done(tcbinfo.mtx);
1234
1235 }
1236 }
1237
1238 /*
1239 * Notify a tcp user of an asynchronous error;
1240 * store error as soft error, but wake up user
1241 * (for now, won't do anything until can select for soft error).
1242 *
1243 * Do not wake up user since there currently is no mechanism for
1244 * reporting soft errors (yet - a kqueue filter may be added).
1245 */
1246 static void
1247 tcp_notify(inp, error)
1248 struct inpcb *inp;
1249 int error;
1250 {
1251 struct tcpcb *tp;
1252
1253 if (inp == NULL || (inp->inp_state == INPCB_STATE_DEAD))
1254 return; /* pcb is gone already */
1255
1256 tp = (struct tcpcb *)inp->inp_ppcb;
1257
1258 /*
1259 * Ignore some errors if we are hooked up.
1260 * If connection hasn't completed, has retransmitted several times,
1261 * and receives a second error, give up now. This is better
1262 * than waiting a long time to establish a connection that
1263 * can never complete.
1264 */
1265 if (tp->t_state == TCPS_ESTABLISHED &&
1266 (error == EHOSTUNREACH || error == ENETUNREACH ||
1267 error == EHOSTDOWN)) {
1268 return;
1269 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1270 tp->t_softerror)
1271 tcp_drop(tp, error);
1272 else
1273 tp->t_softerror = error;
1274 #if 0
1275 wakeup((caddr_t) &so->so_timeo);
1276 sorwakeup(so);
1277 sowwakeup(so);
1278 #endif
1279 }
1280
1281 struct bwmeas*
1282 tcp_bwmeas_alloc(struct tcpcb *tp)
1283 {
1284 struct bwmeas *elm;
1285 elm = zalloc(tcp_bwmeas_zone);
1286 if (elm == NULL)
1287 return(elm);
1288
1289 bzero(elm, bwmeas_elm_size);
1290 elm->bw_minsizepkts = TCP_BWMEAS_BURST_MINSIZE;
1291 elm->bw_maxsizepkts = TCP_BWMEAS_BURST_MAXSIZE;
1292 elm->bw_minsize = elm->bw_minsizepkts * tp->t_maxseg;
1293 elm->bw_maxsize = elm->bw_maxsizepkts * tp->t_maxseg;
1294 return(elm);
1295 }
1296
1297 void
1298 tcp_bwmeas_free(struct tcpcb* tp)
1299 {
1300 zfree(tcp_bwmeas_zone, tp->t_bwmeas);
1301 tp->t_bwmeas = NULL;
1302 tp->t_flagsext &= ~(TF_MEASURESNDBW);
1303 }
1304
1305 /*
1306 * tcpcb_to_otcpcb copies specific bits of a tcpcb to a otcpcb format.
1307 * The otcpcb data structure is passed to user space and must not change.
1308 */
1309 static void
1310 tcpcb_to_otcpcb(struct tcpcb *tp, struct otcpcb *otp)
1311 {
1312 int i;
1313
1314 otp->t_segq = (u_int32_t)(uintptr_t)tp->t_segq.lh_first;
1315 otp->t_dupacks = tp->t_dupacks;
1316 for (i = 0; i < TCPT_NTIMERS_EXT; i++)
1317 otp->t_timer[i] = tp->t_timer[i];
1318 otp->t_inpcb = (_TCPCB_PTR(struct inpcb *))(uintptr_t)tp->t_inpcb;
1319 otp->t_state = tp->t_state;
1320 otp->t_flags = tp->t_flags;
1321 otp->t_force = tp->t_force;
1322 otp->snd_una = tp->snd_una;
1323 otp->snd_max = tp->snd_max;
1324 otp->snd_nxt = tp->snd_nxt;
1325 otp->snd_up = tp->snd_up;
1326 otp->snd_wl1 = tp->snd_wl1;
1327 otp->snd_wl2 = tp->snd_wl2;
1328 otp->iss = tp->iss;
1329 otp->irs = tp->irs;
1330 otp->rcv_nxt = tp->rcv_nxt;
1331 otp->rcv_adv = tp->rcv_adv;
1332 otp->rcv_wnd = tp->rcv_wnd;
1333 otp->rcv_up = tp->rcv_up;
1334 otp->snd_wnd = tp->snd_wnd;
1335 otp->snd_cwnd = tp->snd_cwnd;
1336 otp->snd_ssthresh = tp->snd_ssthresh;
1337 otp->t_maxopd = tp->t_maxopd;
1338 otp->t_rcvtime = tp->t_rcvtime;
1339 otp->t_starttime = tp->t_starttime;
1340 otp->t_rtttime = tp->t_rtttime;
1341 otp->t_rtseq = tp->t_rtseq;
1342 otp->t_rxtcur = tp->t_rxtcur;
1343 otp->t_maxseg = tp->t_maxseg;
1344 otp->t_srtt = tp->t_srtt;
1345 otp->t_rttvar = tp->t_rttvar;
1346 otp->t_rxtshift = tp->t_rxtshift;
1347 otp->t_rttmin = tp->t_rttmin;
1348 otp->t_rttupdated = tp->t_rttupdated;
1349 otp->max_sndwnd = tp->max_sndwnd;
1350 otp->t_softerror = tp->t_softerror;
1351 otp->t_oobflags = tp->t_oobflags;
1352 otp->t_iobc = tp->t_iobc;
1353 otp->snd_scale = tp->snd_scale;
1354 otp->rcv_scale = tp->rcv_scale;
1355 otp->request_r_scale = tp->request_r_scale;
1356 otp->requested_s_scale = tp->requested_s_scale;
1357 otp->ts_recent = tp->ts_recent;
1358 otp->ts_recent_age = tp->ts_recent_age;
1359 otp->last_ack_sent = tp->last_ack_sent;
1360 otp->cc_send = tp->cc_send;
1361 otp->cc_recv = tp->cc_recv;
1362 otp->snd_recover = tp->snd_recover;
1363 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1364 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1365 otp->t_badrxtwin = tp->t_badrxtwin;
1366 }
1367
1368 static int
1369 tcp_pcblist SYSCTL_HANDLER_ARGS
1370 {
1371 #pragma unused(oidp, arg1, arg2)
1372 int error, i, n;
1373 struct inpcb *inp, **inp_list;
1374 inp_gen_t gencnt;
1375 struct xinpgen xig;
1376 int slot;
1377
1378 /*
1379 * The process of preparing the TCB list is too time-consuming and
1380 * resource-intensive to repeat twice on every request.
1381 */
1382 lck_rw_lock_shared(tcbinfo.mtx);
1383 if (req->oldptr == USER_ADDR_NULL) {
1384 n = tcbinfo.ipi_count;
1385 req->oldidx = 2 * (sizeof xig)
1386 + (n + n/8) * sizeof(struct xtcpcb);
1387 lck_rw_done(tcbinfo.mtx);
1388 return 0;
1389 }
1390
1391 if (req->newptr != USER_ADDR_NULL) {
1392 lck_rw_done(tcbinfo.mtx);
1393 return EPERM;
1394 }
1395
1396 /*
1397 * OK, now we're committed to doing something.
1398 */
1399 gencnt = tcbinfo.ipi_gencnt;
1400 n = tcbinfo.ipi_count;
1401
1402 bzero(&xig, sizeof(xig));
1403 xig.xig_len = sizeof xig;
1404 xig.xig_count = n;
1405 xig.xig_gen = gencnt;
1406 xig.xig_sogen = so_gencnt;
1407 error = SYSCTL_OUT(req, &xig, sizeof xig);
1408 if (error) {
1409 lck_rw_done(tcbinfo.mtx);
1410 return error;
1411 }
1412 /*
1413 * We are done if there is no pcb
1414 */
1415 if (n == 0) {
1416 lck_rw_done(tcbinfo.mtx);
1417 return 0;
1418 }
1419
1420 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1421 if (inp_list == 0) {
1422 lck_rw_done(tcbinfo.mtx);
1423 return ENOMEM;
1424 }
1425
1426 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
1427 inp = LIST_NEXT(inp, inp_list)) {
1428 #ifdef __APPLE__
1429 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD)
1430 #else
1431 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
1432 #endif
1433 inp_list[i++] = inp;
1434 }
1435
1436 for (slot = 0; slot < N_TIME_WAIT_SLOTS; slot++) {
1437 struct inpcb *inpnxt;
1438
1439 for (inp = time_wait_slots[slot].lh_first; inp && i < n; inp = inpnxt) {
1440 inpnxt = inp->inp_list.le_next;
1441 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD)
1442 inp_list[i++] = inp;
1443 }
1444 }
1445
1446 n = i;
1447
1448 error = 0;
1449 for (i = 0; i < n; i++) {
1450 inp = inp_list[i];
1451 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) {
1452 struct xtcpcb xt;
1453 caddr_t inp_ppcb;
1454
1455 bzero(&xt, sizeof(xt));
1456 xt.xt_len = sizeof xt;
1457 /* XXX should avoid extra copy */
1458 inpcb_to_compat(inp, &xt.xt_inp);
1459 inp_ppcb = inp->inp_ppcb;
1460 if (inp_ppcb != NULL) {
1461 tcpcb_to_otcpcb((struct tcpcb *)(void *)inp_ppcb,
1462 &xt.xt_tp);
1463 } else {
1464 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1465 }
1466 if (inp->inp_socket)
1467 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1468 error = SYSCTL_OUT(req, &xt, sizeof xt);
1469 }
1470 }
1471 if (!error) {
1472 /*
1473 * Give the user an updated idea of our state.
1474 * If the generation differs from what we told
1475 * her before, she knows that something happened
1476 * while we were processing this request, and it
1477 * might be necessary to retry.
1478 */
1479 bzero(&xig, sizeof(xig));
1480 xig.xig_len = sizeof xig;
1481 xig.xig_gen = tcbinfo.ipi_gencnt;
1482 xig.xig_sogen = so_gencnt;
1483 xig.xig_count = tcbinfo.ipi_count;
1484 error = SYSCTL_OUT(req, &xig, sizeof xig);
1485 }
1486 FREE(inp_list, M_TEMP);
1487 lck_rw_done(tcbinfo.mtx);
1488 return error;
1489 }
1490
1491 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1492 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1493
1494 #if !CONFIG_EMBEDDED
1495
1496 static void
1497 tcpcb_to_xtcpcb64(struct tcpcb *tp, struct xtcpcb64 *otp)
1498 {
1499 int i;
1500
1501 otp->t_segq = (u_int32_t)(uintptr_t)tp->t_segq.lh_first;
1502 otp->t_dupacks = tp->t_dupacks;
1503 for (i = 0; i < TCPT_NTIMERS_EXT; i++)
1504 otp->t_timer[i] = tp->t_timer[i];
1505 otp->t_state = tp->t_state;
1506 otp->t_flags = tp->t_flags;
1507 otp->t_force = tp->t_force;
1508 otp->snd_una = tp->snd_una;
1509 otp->snd_max = tp->snd_max;
1510 otp->snd_nxt = tp->snd_nxt;
1511 otp->snd_up = tp->snd_up;
1512 otp->snd_wl1 = tp->snd_wl1;
1513 otp->snd_wl2 = tp->snd_wl2;
1514 otp->iss = tp->iss;
1515 otp->irs = tp->irs;
1516 otp->rcv_nxt = tp->rcv_nxt;
1517 otp->rcv_adv = tp->rcv_adv;
1518 otp->rcv_wnd = tp->rcv_wnd;
1519 otp->rcv_up = tp->rcv_up;
1520 otp->snd_wnd = tp->snd_wnd;
1521 otp->snd_cwnd = tp->snd_cwnd;
1522 otp->snd_ssthresh = tp->snd_ssthresh;
1523 otp->t_maxopd = tp->t_maxopd;
1524 otp->t_rcvtime = tp->t_rcvtime;
1525 otp->t_starttime = tp->t_starttime;
1526 otp->t_rtttime = tp->t_rtttime;
1527 otp->t_rtseq = tp->t_rtseq;
1528 otp->t_rxtcur = tp->t_rxtcur;
1529 otp->t_maxseg = tp->t_maxseg;
1530 otp->t_srtt = tp->t_srtt;
1531 otp->t_rttvar = tp->t_rttvar;
1532 otp->t_rxtshift = tp->t_rxtshift;
1533 otp->t_rttmin = tp->t_rttmin;
1534 otp->t_rttupdated = tp->t_rttupdated;
1535 otp->max_sndwnd = tp->max_sndwnd;
1536 otp->t_softerror = tp->t_softerror;
1537 otp->t_oobflags = tp->t_oobflags;
1538 otp->t_iobc = tp->t_iobc;
1539 otp->snd_scale = tp->snd_scale;
1540 otp->rcv_scale = tp->rcv_scale;
1541 otp->request_r_scale = tp->request_r_scale;
1542 otp->requested_s_scale = tp->requested_s_scale;
1543 otp->ts_recent = tp->ts_recent;
1544 otp->ts_recent_age = tp->ts_recent_age;
1545 otp->last_ack_sent = tp->last_ack_sent;
1546 otp->cc_send = tp->cc_send;
1547 otp->cc_recv = tp->cc_recv;
1548 otp->snd_recover = tp->snd_recover;
1549 otp->snd_cwnd_prev = tp->snd_cwnd_prev;
1550 otp->snd_ssthresh_prev = tp->snd_ssthresh_prev;
1551 otp->t_badrxtwin = tp->t_badrxtwin;
1552 }
1553
1554
1555 static int
1556 tcp_pcblist64 SYSCTL_HANDLER_ARGS
1557 {
1558 #pragma unused(oidp, arg1, arg2)
1559 int error, i, n;
1560 struct inpcb *inp, **inp_list;
1561 inp_gen_t gencnt;
1562 struct xinpgen xig;
1563 int slot;
1564
1565 /*
1566 * The process of preparing the TCB list is too time-consuming and
1567 * resource-intensive to repeat twice on every request.
1568 */
1569 lck_rw_lock_shared(tcbinfo.mtx);
1570 if (req->oldptr == USER_ADDR_NULL) {
1571 n = tcbinfo.ipi_count;
1572 req->oldidx = 2 * (sizeof xig)
1573 + (n + n/8) * sizeof(struct xtcpcb64);
1574 lck_rw_done(tcbinfo.mtx);
1575 return 0;
1576 }
1577
1578 if (req->newptr != USER_ADDR_NULL) {
1579 lck_rw_done(tcbinfo.mtx);
1580 return EPERM;
1581 }
1582
1583 /*
1584 * OK, now we're committed to doing something.
1585 */
1586 gencnt = tcbinfo.ipi_gencnt;
1587 n = tcbinfo.ipi_count;
1588
1589 bzero(&xig, sizeof(xig));
1590 xig.xig_len = sizeof xig;
1591 xig.xig_count = n;
1592 xig.xig_gen = gencnt;
1593 xig.xig_sogen = so_gencnt;
1594 error = SYSCTL_OUT(req, &xig, sizeof xig);
1595 if (error) {
1596 lck_rw_done(tcbinfo.mtx);
1597 return error;
1598 }
1599 /*
1600 * We are done if there is no pcb
1601 */
1602 if (n == 0) {
1603 lck_rw_done(tcbinfo.mtx);
1604 return 0;
1605 }
1606
1607 inp_list = _MALLOC(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1608 if (inp_list == 0) {
1609 lck_rw_done(tcbinfo.mtx);
1610 return ENOMEM;
1611 }
1612
1613 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp && i < n;
1614 inp = LIST_NEXT(inp, inp_list)) {
1615 #ifdef __APPLE__
1616 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD)
1617 #else
1618 if (inp->inp_gencnt <= gencnt && !prison_xinpcb(req->p, inp))
1619 #endif
1620 inp_list[i++] = inp;
1621 }
1622
1623 for (slot = 0; slot < N_TIME_WAIT_SLOTS; slot++) {
1624 struct inpcb *inpnxt;
1625
1626 for (inp = time_wait_slots[slot].lh_first; inp && i < n; inp = inpnxt) {
1627 inpnxt = inp->inp_list.le_next;
1628 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD)
1629 inp_list[i++] = inp;
1630 }
1631 }
1632
1633 n = i;
1634
1635 error = 0;
1636 for (i = 0; i < n; i++) {
1637 inp = inp_list[i];
1638 if (inp->inp_gencnt <= gencnt && inp->inp_state != INPCB_STATE_DEAD) {
1639 struct xtcpcb64 xt;
1640
1641 bzero(&xt, sizeof(xt));
1642 xt.xt_len = sizeof xt;
1643 inpcb_to_xinpcb64(inp, &xt.xt_inpcb);
1644 xt.xt_inpcb.inp_ppcb = (u_int64_t)(uintptr_t)inp->inp_ppcb;
1645 if (inp->inp_ppcb != NULL)
1646 tcpcb_to_xtcpcb64((struct tcpcb *)inp->inp_ppcb, &xt);
1647 if (inp->inp_socket)
1648 sotoxsocket64(inp->inp_socket, &xt.xt_inpcb.xi_socket);
1649 error = SYSCTL_OUT(req, &xt, sizeof xt);
1650 }
1651 }
1652 if (!error) {
1653 /*
1654 * Give the user an updated idea of our state.
1655 * If the generation differs from what we told
1656 * her before, she knows that something happened
1657 * while we were processing this request, and it
1658 * might be necessary to retry.
1659 */
1660 bzero(&xig, sizeof(xig));
1661 xig.xig_len = sizeof xig;
1662 xig.xig_gen = tcbinfo.ipi_gencnt;
1663 xig.xig_sogen = so_gencnt;
1664 xig.xig_count = tcbinfo.ipi_count;
1665 error = SYSCTL_OUT(req, &xig, sizeof xig);
1666 }
1667 FREE(inp_list, M_TEMP);
1668 lck_rw_done(tcbinfo.mtx);
1669 return error;
1670 }
1671
1672 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist64, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1673 tcp_pcblist64, "S,xtcpcb64", "List of active TCP connections");
1674
1675 #endif /* !CONFIG_EMBEDDED */
1676
1677 static int
1678 tcp_pcblist_n SYSCTL_HANDLER_ARGS
1679 {
1680 #pragma unused(oidp, arg1, arg2)
1681 int error = 0;
1682
1683 error = get_pcblist_n(IPPROTO_TCP, req, &tcbinfo);
1684
1685 return error;
1686 }
1687
1688
1689 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, pcblist_n, CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0,
1690 tcp_pcblist_n, "S,xtcpcb_n", "List of active TCP connections");
1691
1692
1693 __private_extern__ void
1694 tcp_get_ports_used(unsigned int ifindex, uint8_t *bitfield)
1695 {
1696 inpcb_get_ports_used(ifindex, bitfield, &tcbinfo);
1697 }
1698
1699 __private_extern__ uint32_t
1700 tcp_count_opportunistic(unsigned int ifindex, u_int32_t flags)
1701 {
1702 return inpcb_count_opportunistic(ifindex, &tcbinfo, flags);
1703 }
1704
1705 void
1706 tcp_ctlinput(cmd, sa, vip)
1707 int cmd;
1708 struct sockaddr *sa;
1709 void *vip;
1710 {
1711 tcp_seq icmp_tcp_seq;
1712 struct ip *ip = vip;
1713 struct in_addr faddr;
1714 struct inpcb *inp;
1715 struct tcpcb *tp;
1716
1717 void (*notify)(struct inpcb *, int) = tcp_notify;
1718
1719 faddr = ((struct sockaddr_in *)(void *)sa)->sin_addr;
1720 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1721 return;
1722
1723 if (cmd == PRC_MSGSIZE)
1724 notify = tcp_mtudisc;
1725 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1726 cmd == PRC_UNREACH_PORT) && ip)
1727 notify = tcp_drop_syn_sent;
1728 else if (PRC_IS_REDIRECT(cmd)) {
1729 ip = 0;
1730 notify = in_rtchange;
1731 } else if (cmd == PRC_HOSTDEAD)
1732 ip = 0;
1733 /* Source quench is deprecated */
1734 else if (cmd == PRC_QUENCH)
1735 return;
1736 else if ((unsigned)cmd > PRC_NCMDS || inetctlerrmap[cmd] == 0)
1737 return;
1738 if (ip) {
1739 struct tcphdr th;
1740 struct icmp *icp;
1741
1742 icp = (struct icmp *)(void *)
1743 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1744 bcopy(((caddr_t)ip + (IP_VHL_HL(ip->ip_vhl) << 2)),
1745 &th, sizeof (th));
1746 inp = in_pcblookup_hash(&tcbinfo, faddr, th.th_dport,
1747 ip->ip_src, th.th_sport, 0, NULL);
1748 if (inp != NULL && inp->inp_socket != NULL) {
1749 tcp_lock(inp->inp_socket, 1, 0);
1750 if (in_pcb_checkstate(inp, WNT_RELEASE, 1) == WNT_STOPUSING) {
1751 tcp_unlock(inp->inp_socket, 1, 0);
1752 return;
1753 }
1754 icmp_tcp_seq = htonl(th.th_seq);
1755 tp = intotcpcb(inp);
1756 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1757 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1758 if (cmd == PRC_MSGSIZE) {
1759
1760 /*
1761 * MTU discovery:
1762 * If we got a needfrag and there is a host route to the
1763 * original destination, and the MTU is not locked, then
1764 * set the MTU in the route to the suggested new value
1765 * (if given) and then notify as usual. The ULPs will
1766 * notice that the MTU has changed and adapt accordingly.
1767 * If no new MTU was suggested, then we guess a new one
1768 * less than the current value. If the new MTU is
1769 * unreasonably small (defined by sysctl tcp_minmss), then
1770 * we reset the MTU to the interface value and enable the
1771 * lock bit, indicating that we are no longer doing MTU
1772 * discovery.
1773 */
1774 struct rtentry *rt;
1775 int mtu;
1776 struct sockaddr_in icmpsrc = { sizeof (struct sockaddr_in), AF_INET,
1777 0 , { 0 }, { 0,0,0,0,0,0,0,0 } };
1778 icmpsrc.sin_addr = icp->icmp_ip.ip_dst;
1779
1780 rt = rtalloc1((struct sockaddr *)&icmpsrc, 0,
1781 RTF_CLONING | RTF_PRCLONING);
1782 if (rt != NULL) {
1783 RT_LOCK(rt);
1784 if ((rt->rt_flags & RTF_HOST) &&
1785 !(rt->rt_rmx.rmx_locks & RTV_MTU)) {
1786 mtu = ntohs(icp->icmp_nextmtu);
1787 if (!mtu)
1788 mtu = ip_next_mtu(rt->rt_rmx.
1789 rmx_mtu, 1);
1790 #if DEBUG_MTUDISC
1791 printf("MTU for %s reduced to %d\n",
1792 inet_ntop(AF_INET,
1793 &icmpsrc.sin_addr, ipv4str,
1794 sizeof (ipv4str)), mtu);
1795 #endif
1796 if (mtu < max(296, (tcp_minmss +
1797 sizeof (struct tcpiphdr)))) {
1798 /* rt->rt_rmx.rmx_mtu =
1799 rt->rt_ifp->if_mtu; */
1800 rt->rt_rmx.rmx_locks |= RTV_MTU;
1801 } else if (rt->rt_rmx.rmx_mtu > mtu) {
1802 rt->rt_rmx.rmx_mtu = mtu;
1803 }
1804 }
1805 RT_UNLOCK(rt);
1806 rtfree(rt);
1807 }
1808 }
1809
1810 (*notify)(inp, inetctlerrmap[cmd]);
1811 }
1812 tcp_unlock(inp->inp_socket, 1, 0);
1813 }
1814 } else
1815 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1816 }
1817
1818 #if INET6
1819 void
1820 tcp6_ctlinput(cmd, sa, d)
1821 int cmd;
1822 struct sockaddr *sa;
1823 void *d;
1824 {
1825 struct tcphdr th;
1826 void (*notify)(struct inpcb *, int) = tcp_notify;
1827 struct ip6_hdr *ip6;
1828 struct mbuf *m;
1829 struct ip6ctlparam *ip6cp = NULL;
1830 const struct sockaddr_in6 *sa6_src = NULL;
1831 int off;
1832 struct tcp_portonly {
1833 u_int16_t th_sport;
1834 u_int16_t th_dport;
1835 } *thp;
1836
1837 if (sa->sa_family != AF_INET6 ||
1838 sa->sa_len != sizeof(struct sockaddr_in6))
1839 return;
1840
1841 if (cmd == PRC_MSGSIZE)
1842 notify = tcp_mtudisc;
1843 else if (!PRC_IS_REDIRECT(cmd) &&
1844 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1845 return;
1846 /* Source quench is deprecated */
1847 else if (cmd == PRC_QUENCH)
1848 return;
1849
1850 /* if the parameter is from icmp6, decode it. */
1851 if (d != NULL) {
1852 ip6cp = (struct ip6ctlparam *)d;
1853 m = ip6cp->ip6c_m;
1854 ip6 = ip6cp->ip6c_ip6;
1855 off = ip6cp->ip6c_off;
1856 sa6_src = ip6cp->ip6c_src;
1857 } else {
1858 m = NULL;
1859 ip6 = NULL;
1860 off = 0; /* fool gcc */
1861 sa6_src = &sa6_any;
1862 }
1863
1864 if (ip6) {
1865 /*
1866 * XXX: We assume that when IPV6 is non NULL,
1867 * M and OFF are valid.
1868 */
1869
1870 /* check if we can safely examine src and dst ports */
1871 if (m->m_pkthdr.len < off + sizeof(*thp))
1872 return;
1873
1874 bzero(&th, sizeof(th));
1875 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1876
1877 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1878 (struct sockaddr *)ip6cp->ip6c_src,
1879 th.th_sport, cmd, NULL, notify);
1880 } else {
1881 in6_pcbnotify(&tcbinfo, sa, 0,
1882 (struct sockaddr *)(size_t)sa6_src, 0, cmd, NULL, notify);
1883 }
1884 }
1885 #endif /* INET6 */
1886
1887
1888 /*
1889 * Following is where TCP initial sequence number generation occurs.
1890 *
1891 * There are two places where we must use initial sequence numbers:
1892 * 1. In SYN-ACK packets.
1893 * 2. In SYN packets.
1894 *
1895 * The ISNs in SYN-ACK packets have no monotonicity requirement,
1896 * and should be as unpredictable as possible to avoid the possibility
1897 * of spoofing and/or connection hijacking. To satisfy this
1898 * requirement, SYN-ACK ISNs are generated via the arc4random()
1899 * function. If exact RFC 1948 compliance is requested via sysctl,
1900 * these ISNs will be generated just like those in SYN packets.
1901 *
1902 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1903 * depends on this property. In addition, these ISNs should be
1904 * unguessable so as to prevent connection hijacking. To satisfy
1905 * the requirements of this situation, the algorithm outlined in
1906 * RFC 1948 is used to generate sequence numbers.
1907 *
1908 * For more information on the theory of operation, please see
1909 * RFC 1948.
1910 *
1911 * Implementation details:
1912 *
1913 * Time is based off the system timer, and is corrected so that it
1914 * increases by one megabyte per second. This allows for proper
1915 * recycling on high speed LANs while still leaving over an hour
1916 * before rollover.
1917 *
1918 * Two sysctls control the generation of ISNs:
1919 *
1920 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1921 * between seeding of isn_secret. This is normally set to zero,
1922 * as reseeding should not be necessary.
1923 *
1924 * net.inet.tcp.strict_rfc1948 controls whether RFC 1948 is followed
1925 * strictly. When strict compliance is requested, reseeding is
1926 * disabled and SYN-ACKs will be generated in the same manner as
1927 * SYNs. Strict mode is disabled by default.
1928 *
1929 */
1930
1931 #define ISN_BYTES_PER_SECOND 1048576
1932
1933 tcp_seq
1934 tcp_new_isn(tp)
1935 struct tcpcb *tp;
1936 {
1937 u_int32_t md5_buffer[4];
1938 tcp_seq new_isn;
1939 struct timeval timenow;
1940 u_char isn_secret[32];
1941 int isn_last_reseed = 0;
1942 MD5_CTX isn_ctx;
1943
1944 /* Use arc4random for SYN-ACKs when not in exact RFC1948 mode. */
1945 if (((tp->t_state == TCPS_LISTEN) || (tp->t_state == TCPS_TIME_WAIT))
1946 && tcp_strict_rfc1948 == 0)
1947 #ifdef __APPLE__
1948 return random();
1949 #else
1950 return arc4random();
1951 #endif
1952 getmicrotime(&timenow);
1953
1954 /* Seed if this is the first use, reseed if requested. */
1955 if ((isn_last_reseed == 0) ||
1956 ((tcp_strict_rfc1948 == 0) && (tcp_isn_reseed_interval > 0) &&
1957 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1958 < (u_int)timenow.tv_sec))) {
1959 #ifdef __APPLE__
1960 read_random(&isn_secret, sizeof(isn_secret));
1961 #else
1962 read_random_unlimited(&isn_secret, sizeof(isn_secret));
1963 #endif
1964 isn_last_reseed = timenow.tv_sec;
1965 }
1966
1967 /* Compute the md5 hash and return the ISN. */
1968 MD5Init(&isn_ctx);
1969 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1970 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1971 #if INET6
1972 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1973 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1974 sizeof(struct in6_addr));
1975 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1976 sizeof(struct in6_addr));
1977 } else
1978 #endif
1979 {
1980 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1981 sizeof(struct in_addr));
1982 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1983 sizeof(struct in_addr));
1984 }
1985 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1986 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1987 new_isn = (tcp_seq) md5_buffer[0];
1988 new_isn += timenow.tv_sec * (ISN_BYTES_PER_SECOND / hz);
1989 return new_isn;
1990 }
1991
1992
1993 /*
1994 * When a specific ICMP unreachable message is received and the
1995 * connection state is SYN-SENT, drop the connection. This behavior
1996 * is controlled by the icmp_may_rst sysctl.
1997 */
1998 void
1999 tcp_drop_syn_sent(inp, errno)
2000 struct inpcb *inp;
2001 int errno;
2002 {
2003 struct tcpcb *tp = intotcpcb(inp);
2004
2005 if (tp && tp->t_state == TCPS_SYN_SENT)
2006 tcp_drop(tp, errno);
2007 }
2008
2009 /*
2010 * When `need fragmentation' ICMP is received, update our idea of the MSS
2011 * based on the new value in the route. Also nudge TCP to send something,
2012 * since we know the packet we just sent was dropped.
2013 * This duplicates some code in the tcp_mss() function in tcp_input.c.
2014 */
2015 void
2016 tcp_mtudisc(
2017 struct inpcb *inp,
2018 __unused int errno
2019 )
2020 {
2021 struct tcpcb *tp = intotcpcb(inp);
2022 struct rtentry *rt;
2023 struct rmxp_tao *taop;
2024 struct socket *so = inp->inp_socket;
2025 int offered;
2026 int mss;
2027 #if INET6
2028 int isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
2029 #endif /* INET6 */
2030
2031 if (tp) {
2032 #if INET6
2033 if (isipv6)
2034 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2035 else
2036 #endif /* INET6 */
2037 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2038 if (!rt || !rt->rt_rmx.rmx_mtu) {
2039 tp->t_maxopd = tp->t_maxseg =
2040 #if INET6
2041 isipv6 ? tcp_v6mssdflt :
2042 #endif /* INET6 */
2043 tcp_mssdflt;
2044
2045 /* Route locked during lookup above */
2046 if (rt != NULL)
2047 RT_UNLOCK(rt);
2048 return;
2049 }
2050 taop = rmx_taop(rt->rt_rmx);
2051 offered = taop->tao_mssopt;
2052 mss = rt->rt_rmx.rmx_mtu -
2053 #if INET6
2054 (isipv6 ?
2055 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2056 #endif /* INET6 */
2057 sizeof(struct tcpiphdr)
2058 #if INET6
2059 )
2060 #endif /* INET6 */
2061 ;
2062
2063 /* Route locked during lookup above */
2064 RT_UNLOCK(rt);
2065
2066 if (offered)
2067 mss = min(mss, offered);
2068 /*
2069 * XXX - The above conditional probably violates the TCP
2070 * spec. The problem is that, since we don't know the
2071 * other end's MSS, we are supposed to use a conservative
2072 * default. But, if we do that, then MTU discovery will
2073 * never actually take place, because the conservative
2074 * default is much less than the MTUs typically seen
2075 * on the Internet today. For the moment, we'll sweep
2076 * this under the carpet.
2077 *
2078 * The conservative default might not actually be a problem
2079 * if the only case this occurs is when sending an initial
2080 * SYN with options and data to a host we've never talked
2081 * to before. Then, they will reply with an MSS value which
2082 * will get recorded and the new parameters should get
2083 * recomputed. For Further Study.
2084 */
2085 if (tp->t_maxopd <= mss)
2086 return;
2087 tp->t_maxopd = mss;
2088
2089 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2090 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2091 mss -= TCPOLEN_TSTAMP_APPA;
2092
2093 if (so->so_snd.sb_hiwat < mss)
2094 mss = so->so_snd.sb_hiwat;
2095
2096 tp->t_maxseg = mss;
2097
2098 /*
2099 * Reset the slow-start flight size as it may depends on the new MSS
2100 */
2101 if (CC_ALGO(tp)->cwnd_init != NULL)
2102 CC_ALGO(tp)->cwnd_init(tp);
2103 tcpstat.tcps_mturesent++;
2104 tp->t_rtttime = 0;
2105 tp->snd_nxt = tp->snd_una;
2106 tcp_output(tp);
2107 }
2108 }
2109
2110 /*
2111 * Look-up the routing entry to the peer of this inpcb. If no route
2112 * is found and it cannot be allocated the return NULL. This routine
2113 * is called by TCP routines that access the rmx structure and by tcp_mss
2114 * to get the interface MTU. If a route is found, this routine will
2115 * hold the rtentry lock; the caller is responsible for unlocking.
2116 */
2117 struct rtentry *
2118 tcp_rtlookup(inp, input_ifscope)
2119 struct inpcb *inp;
2120 unsigned int input_ifscope;
2121 {
2122 struct route *ro;
2123 struct rtentry *rt;
2124 struct tcpcb *tp;
2125
2126 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2127
2128 ro = &inp->inp_route;
2129 if ((rt = ro->ro_rt) != NULL)
2130 RT_LOCK(rt);
2131
2132 if (rt == NULL || !(rt->rt_flags & RTF_UP) ||
2133 rt->generation_id != route_generation) {
2134 /* No route yet, so try to acquire one */
2135 if (inp->inp_faddr.s_addr != INADDR_ANY) {
2136 unsigned int ifscope;
2137
2138 ro->ro_dst.sa_family = AF_INET;
2139 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
2140 ((struct sockaddr_in *)(void *)&ro->ro_dst)->sin_addr =
2141 inp->inp_faddr;
2142
2143 /*
2144 * If the socket was bound to an interface, then
2145 * the bound-to-interface takes precedence over
2146 * the inbound interface passed in by the caller
2147 * (if we get here as part of the output path then
2148 * input_ifscope is IFSCOPE_NONE).
2149 */
2150 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2151 inp->inp_boundifp->if_index : input_ifscope;
2152
2153 if (rt != NULL)
2154 RT_UNLOCK(rt);
2155 rtalloc_scoped(ro, ifscope);
2156 if ((rt = ro->ro_rt) != NULL)
2157 RT_LOCK(rt);
2158 }
2159 }
2160
2161 /*
2162 * Update MTU discovery determination. Don't do it if:
2163 * 1) it is disabled via the sysctl
2164 * 2) the route isn't up
2165 * 3) the MTU is locked (if it is, then discovery has been
2166 * disabled)
2167 */
2168
2169 tp = intotcpcb(inp);
2170
2171 if (!path_mtu_discovery || ((rt != NULL) &&
2172 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2173 tp->t_flags &= ~TF_PMTUD;
2174 else
2175 tp->t_flags |= TF_PMTUD;
2176
2177 #if CONFIG_IFEF_NOWINDOWSCALE
2178 if (tcp_obey_ifef_nowindowscale &&
2179 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2180 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2181 /* Window scaling is enabled on this interface */
2182 tp->t_flags &= ~TF_REQ_SCALE;
2183 }
2184 #endif
2185
2186 if (rt != NULL && rt->rt_ifp != NULL) {
2187 somultipages(inp->inp_socket,
2188 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2189 tcp_set_tso(tp, rt->rt_ifp);
2190 }
2191
2192 /*
2193 * Caller needs to call RT_UNLOCK(rt).
2194 */
2195 return rt;
2196 }
2197
2198 #if INET6
2199 struct rtentry *
2200 tcp_rtlookup6(inp, input_ifscope)
2201 struct inpcb *inp;
2202 unsigned int input_ifscope;
2203 {
2204 struct route_in6 *ro6;
2205 struct rtentry *rt;
2206 struct tcpcb *tp;
2207
2208 lck_mtx_assert(rnh_lock, LCK_MTX_ASSERT_NOTOWNED);
2209
2210 ro6 = &inp->in6p_route;
2211 if ((rt = ro6->ro_rt) != NULL)
2212 RT_LOCK(rt);
2213
2214 if (rt == NULL || !(rt->rt_flags & RTF_UP) ||
2215 rt->generation_id != route_generation) {
2216 /* No route yet, so try to acquire one */
2217 if (!IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr)) {
2218 struct sockaddr_in6 *dst6;
2219 unsigned int ifscope;
2220
2221 dst6 = (struct sockaddr_in6 *)&ro6->ro_dst;
2222 dst6->sin6_family = AF_INET6;
2223 dst6->sin6_len = sizeof(*dst6);
2224 dst6->sin6_addr = inp->in6p_faddr;
2225
2226 /*
2227 * If the socket was bound to an interface, then
2228 * the bound-to-interface takes precedence over
2229 * the inbound interface passed in by the caller
2230 * (if we get here as part of the output path then
2231 * input_ifscope is IFSCOPE_NONE).
2232 */
2233 ifscope = (inp->inp_flags & INP_BOUND_IF) ?
2234 inp->inp_boundifp->if_index : input_ifscope;
2235
2236 if (rt != NULL)
2237 RT_UNLOCK(rt);
2238 rtalloc_scoped((struct route *)ro6, ifscope);
2239 if ((rt = ro6->ro_rt) != NULL)
2240 RT_LOCK(rt);
2241 }
2242 }
2243 /*
2244 * Update path MTU Discovery determination
2245 * while looking up the route:
2246 * 1) we have a valid route to the destination
2247 * 2) the MTU is not locked (if it is, then discovery has been
2248 * disabled)
2249 */
2250
2251
2252 tp = intotcpcb(inp);
2253
2254 /*
2255 * Update MTU discovery determination. Don't do it if:
2256 * 1) it is disabled via the sysctl
2257 * 2) the route isn't up
2258 * 3) the MTU is locked (if it is, then discovery has been
2259 * disabled)
2260 */
2261
2262 if (!path_mtu_discovery || ((rt != NULL) &&
2263 (!(rt->rt_flags & RTF_UP) || (rt->rt_rmx.rmx_locks & RTV_MTU))))
2264 tp->t_flags &= ~TF_PMTUD;
2265 else
2266 tp->t_flags |= TF_PMTUD;
2267
2268 #if CONFIG_IFEF_NOWINDOWSCALE
2269 if (tcp_obey_ifef_nowindowscale &&
2270 tp->t_state == TCPS_SYN_SENT && rt != NULL && rt->rt_ifp != NULL &&
2271 (rt->rt_ifp->if_eflags & IFEF_NOWINDOWSCALE)) {
2272 /* Window scaling is not enabled on this interface */
2273 tp->t_flags &= ~TF_REQ_SCALE;
2274 }
2275 #endif
2276
2277 if (rt != NULL && rt->rt_ifp != NULL) {
2278 somultipages(inp->inp_socket,
2279 (rt->rt_ifp->if_hwassist & IFNET_MULTIPAGES));
2280 tcp_set_tso(tp, rt->rt_ifp);
2281 }
2282
2283 /*
2284 * Caller needs to call RT_UNLOCK(rt).
2285 */
2286 return rt;
2287 }
2288 #endif /* INET6 */
2289
2290 #if IPSEC
2291 /* compute ESP/AH header size for TCP, including outer IP header. */
2292 size_t
2293 ipsec_hdrsiz_tcp(tp)
2294 struct tcpcb *tp;
2295 {
2296 struct inpcb *inp;
2297 struct mbuf *m;
2298 size_t hdrsiz;
2299 struct ip *ip;
2300 #if INET6
2301 struct ip6_hdr *ip6 = NULL;
2302 #endif /* INET6 */
2303 struct tcphdr *th;
2304
2305 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
2306 return 0;
2307 MGETHDR(m, M_DONTWAIT, MT_DATA); /* MAC-OK */
2308 if (!m)
2309 return 0;
2310
2311 #if INET6
2312 if ((inp->inp_vflag & INP_IPV6) != 0) {
2313 ip6 = mtod(m, struct ip6_hdr *);
2314 th = (struct tcphdr *)(void *)(ip6 + 1);
2315 m->m_pkthdr.len = m->m_len =
2316 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2317 tcp_fillheaders(tp, ip6, th);
2318 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2319 } else
2320 #endif /* INET6 */
2321 {
2322 ip = mtod(m, struct ip *);
2323 th = (struct tcphdr *)(ip + 1);
2324 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2325 tcp_fillheaders(tp, ip, th);
2326 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2327 }
2328 m_free(m);
2329 return hdrsiz;
2330 }
2331 #endif /*IPSEC*/
2332
2333 /*
2334 * Return a pointer to the cached information about the remote host.
2335 * The cached information is stored in the protocol specific part of
2336 * the route metrics.
2337 */
2338 struct rmxp_tao *
2339 tcp_gettaocache(inp)
2340 struct inpcb *inp;
2341 {
2342 struct rtentry *rt;
2343 struct rmxp_tao *taop;
2344
2345 #if INET6
2346 if ((inp->inp_vflag & INP_IPV6) != 0)
2347 rt = tcp_rtlookup6(inp, IFSCOPE_NONE);
2348 else
2349 #endif /* INET6 */
2350 rt = tcp_rtlookup(inp, IFSCOPE_NONE);
2351
2352 /* Make sure this is a host route and is up. */
2353 if (rt == NULL ||
2354 (rt->rt_flags & (RTF_UP|RTF_HOST)) != (RTF_UP|RTF_HOST)) {
2355 /* Route locked during lookup above */
2356 if (rt != NULL)
2357 RT_UNLOCK(rt);
2358 return NULL;
2359 }
2360
2361 taop = rmx_taop(rt->rt_rmx);
2362 /* Route locked during lookup above */
2363 RT_UNLOCK(rt);
2364 return (taop);
2365 }
2366
2367 /*
2368 * Clear all the TAO cache entries, called from tcp_init.
2369 *
2370 * XXX
2371 * This routine is just an empty one, because we assume that the routing
2372 * routing tables are initialized at the same time when TCP, so there is
2373 * nothing in the cache left over.
2374 */
2375 static void
2376 tcp_cleartaocache()
2377 {
2378 }
2379
2380 int
2381 tcp_lock(struct socket *so, int refcount, void *lr)
2382 {
2383 void *lr_saved;
2384
2385 if (lr == NULL)
2386 lr_saved = __builtin_return_address(0);
2387 else
2388 lr_saved = lr;
2389
2390 if (so->so_pcb != NULL) {
2391 lck_mtx_lock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2392 } else {
2393 panic("tcp_lock: so=%p NO PCB! lr=%p lrh= %s\n",
2394 so, lr_saved, solockhistory_nr(so));
2395 /* NOTREACHED */
2396 }
2397
2398 if (so->so_usecount < 0) {
2399 panic("tcp_lock: so=%p so_pcb=%p lr=%p ref=%x lrh= %s\n",
2400 so, so->so_pcb, lr_saved, so->so_usecount, solockhistory_nr(so));
2401 /* NOTREACHED */
2402 }
2403 if (refcount)
2404 so->so_usecount++;
2405 so->lock_lr[so->next_lock_lr] = lr_saved;
2406 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX;
2407 return (0);
2408 }
2409
2410 int
2411 tcp_unlock(struct socket *so, int refcount, void *lr)
2412 {
2413 void *lr_saved;
2414
2415 if (lr == NULL)
2416 lr_saved = __builtin_return_address(0);
2417 else
2418 lr_saved = lr;
2419
2420 #ifdef MORE_TCPLOCK_DEBUG
2421 printf("tcp_unlock: so=%p sopcb=%p lock=%p ref=%x lr=%p\n",
2422 so, so->so_pcb, &((struct inpcb *)so->so_pcb)->inpcb_mtx,
2423 so->so_usecount, lr_saved);
2424 #endif
2425 if (refcount)
2426 so->so_usecount--;
2427
2428 if (so->so_usecount < 0) {
2429 panic("tcp_unlock: so=%p usecount=%x lrh= %s\n",
2430 so, so->so_usecount, solockhistory_nr(so));
2431 /* NOTREACHED */
2432 }
2433 if (so->so_pcb == NULL) {
2434 panic("tcp_unlock: so=%p NO PCB usecount=%x lr=%p lrh= %s\n",
2435 so, so->so_usecount, lr_saved, solockhistory_nr(so));
2436 /* NOTREACHED */
2437 } else {
2438 lck_mtx_assert(&((struct inpcb *)so->so_pcb)->inpcb_mtx,
2439 LCK_MTX_ASSERT_OWNED);
2440 so->unlock_lr[so->next_unlock_lr] = lr_saved;
2441 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX;
2442 lck_mtx_unlock(&((struct inpcb *)so->so_pcb)->inpcb_mtx);
2443 }
2444 return (0);
2445 }
2446
2447 lck_mtx_t *
2448 tcp_getlock(
2449 struct socket *so,
2450 __unused int locktype)
2451 {
2452 struct inpcb *inp = sotoinpcb(so);
2453
2454 if (so->so_pcb) {
2455 if (so->so_usecount < 0)
2456 panic("tcp_getlock: so=%p usecount=%x lrh= %s\n",
2457 so, so->so_usecount, solockhistory_nr(so));
2458 return(&inp->inpcb_mtx);
2459 }
2460 else {
2461 panic("tcp_getlock: so=%p NULL so_pcb %s\n",
2462 so, solockhistory_nr(so));
2463 return (so->so_proto->pr_domain->dom_mtx);
2464 }
2465 }
2466
2467 /* Determine if we can grow the recieve socket buffer to avoid sending
2468 * a zero window update to the peer. We allow even socket buffers that
2469 * have fixed size (set by the application) to grow if the resource
2470 * constraints are met. They will also be trimmed after the application
2471 * reads data.
2472 */
2473 static void
2474 tcp_sbrcv_grow_rwin(struct tcpcb *tp, struct sockbuf *sb) {
2475 u_int32_t rcvbufinc = tp->t_maxseg << tcp_autorcvbuf_inc_shift;
2476 if (tcp_do_autorcvbuf == 1 &&
2477 tcp_cansbgrow(sb) &&
2478 (tp->t_flags & TF_SLOWLINK) == 0 &&
2479 (sb->sb_hiwat - sb->sb_cc) < rcvbufinc &&
2480 (sb->sb_hiwat < tcp_autorcvbuf_max)) {
2481 sbreserve(sb, (sb->sb_hiwat + rcvbufinc));
2482 }
2483 }
2484
2485 int32_t
2486 tcp_sbspace(struct tcpcb *tp)
2487 {
2488 struct sockbuf *sb = &tp->t_inpcb->inp_socket->so_rcv;
2489 int32_t space;
2490
2491 tcp_sbrcv_grow_rwin(tp, sb);
2492
2493 space = ((int32_t) imin((sb->sb_hiwat - sb->sb_cc),
2494 (sb->sb_mbmax - sb->sb_mbcnt)));
2495 if (space < 0)
2496 space = 0;
2497
2498 /* Avoid increasing window size if the current window
2499 * is already very low, we could be in "persist" mode and
2500 * we could break some apps (see rdar://5409343)
2501 */
2502
2503 if (space < tp->t_maxseg)
2504 return space;
2505
2506 /* Clip window size for slower link */
2507
2508 if (((tp->t_flags & TF_SLOWLINK) != 0) && slowlink_wsize > 0 )
2509 return imin(space, slowlink_wsize);
2510
2511 return space;
2512 }
2513 /*
2514 * Checks TCP Segment Offloading capability for a given connection and interface pair.
2515 */
2516 void
2517 tcp_set_tso(tp, ifp)
2518 struct tcpcb *tp;
2519 struct ifnet *ifp;
2520 {
2521 #if INET6
2522 struct inpcb *inp = tp->t_inpcb;
2523 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
2524
2525 if (isipv6) {
2526 if (ifp && ifp->if_hwassist & IFNET_TSO_IPV6) {
2527 tp->t_flags |= TF_TSO;
2528 if (ifp->if_tso_v6_mtu != 0)
2529 tp->tso_max_segment_size = ifp->if_tso_v6_mtu;
2530 else
2531 tp->tso_max_segment_size = TCP_MAXWIN;
2532 } else
2533 tp->t_flags &= ~TF_TSO;
2534
2535 } else
2536 #endif /* INET6 */
2537
2538 {
2539 if (ifp && ifp->if_hwassist & IFNET_TSO_IPV4) {
2540 tp->t_flags |= TF_TSO;
2541 if (ifp->if_tso_v4_mtu != 0)
2542 tp->tso_max_segment_size = ifp->if_tso_v4_mtu;
2543 else
2544 tp->tso_max_segment_size = TCP_MAXWIN;
2545 } else
2546 tp->t_flags &= ~TF_TSO;
2547 }
2548 }
2549
2550 #define TIMEVAL_TO_TCPHZ(_tv_) ((_tv_).tv_sec * TCP_RETRANSHZ + (_tv_).tv_usec / TCP_RETRANSHZ_TO_USEC)
2551
2552 /* Function to calculate the tcp clock. The tcp clock will get updated
2553 * at the boundaries of the tcp layer. This is done at 3 places:
2554 * 1. Right before processing an input tcp packet
2555 * 2. Whenever a connection wants to access the network using tcp_usrreqs
2556 * 3. When a tcp timer fires or before tcp slow timeout
2557 *
2558 */
2559
2560 void
2561 calculate_tcp_clock()
2562 {
2563 struct timeval tv = tcp_uptime;
2564 struct timeval interval = {0, TCP_RETRANSHZ_TO_USEC};
2565 struct timeval now, hold_now;
2566 uint32_t incr = 0;
2567
2568 timevaladd(&tv, &interval);
2569 microuptime(&now);
2570 if (timevalcmp(&now, &tv, >)) {
2571 /* time to update the clock */
2572 lck_spin_lock(tcp_uptime_lock);
2573 if (timevalcmp(&tcp_uptime, &now, >=)) {
2574 /* clock got updated while we were waiting for the lock */
2575 lck_spin_unlock(tcp_uptime_lock);
2576 return;
2577 }
2578
2579 microuptime(&now);
2580 hold_now = now;
2581 tv = tcp_uptime;
2582 timevalsub(&now, &tv);
2583
2584 incr = TIMEVAL_TO_TCPHZ(now);
2585 if (incr > 0) {
2586 tcp_uptime = hold_now;
2587 tcp_now += incr;
2588 }
2589
2590 lck_spin_unlock(tcp_uptime_lock);
2591 }
2592 return;
2593 }
2594
2595 /* Compute receive window scaling that we are going to request
2596 * for this connection based on sb_hiwat. Try to leave some
2597 * room to potentially increase the window size upto a maximum
2598 * defined by the constant tcp_autorcvbuf_max.
2599 */
2600 void
2601 tcp_set_max_rwinscale(struct tcpcb *tp, struct socket *so) {
2602 u_int32_t maxsockbufsize;
2603
2604 tp->request_r_scale = max(tcp_win_scale, tp->request_r_scale);
2605 maxsockbufsize = ((so->so_rcv.sb_flags & SB_USRSIZE) != 0) ?
2606 so->so_rcv.sb_hiwat : tcp_autorcvbuf_max;
2607
2608 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
2609 (TCP_MAXWIN << tp->request_r_scale) < maxsockbufsize)
2610 tp->request_r_scale++;
2611 tp->request_r_scale = min(tp->request_r_scale, TCP_MAX_WINSHIFT);
2612
2613 }
2614
2615 int
2616 tcp_notsent_lowat_check(struct socket *so) {
2617 struct inpcb *inp = sotoinpcb(so);
2618 struct tcpcb *tp = NULL;
2619 int notsent = 0;
2620 if (inp != NULL) {
2621 tp = intotcpcb(inp);
2622 }
2623
2624 notsent = so->so_snd.sb_cc -
2625 (tp->snd_nxt - tp->snd_una);
2626
2627 /* When we send a FIN or SYN, not_sent can be negative.
2628 * In that case also we need to send a write event to the
2629 * process if it is waiting. In the FIN case, it will
2630 * get an error from send because cantsendmore will be set.
2631 */
2632 if (notsent <= tp->t_notsent_lowat) {
2633 return(1);
2634 }
2635
2636 /* When Nagle's algorithm is not disabled, it is better
2637 * to wakeup the client until there is atleast one
2638 * maxseg of data to write.
2639 */
2640 if ((tp->t_flags & TF_NODELAY) == 0 &&
2641 notsent > 0 && notsent < tp->t_maxseg) {
2642 return(1);
2643 }
2644 return(0);
2645 }
2646
2647
2648 /* DSEP Review Done pl-20051213-v02 @3253,@3391,@3400 */
mirror of XNU