1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58
59 #include <vm/uma.h>
60
61 #include <net/route.h>
62 #include <net/if.h>
63
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #include <netinet6/nd6.h>
80 #endif
81 #include <netinet/ip_icmp.h>
82 #include <netinet/tcp.h>
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #include <netinet/tcp_offload.h>
89 #ifdef INET6
90 #include <netinet6/tcp6_var.h>
91 #endif
92 #include <netinet/tcpip.h>
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #include <netinet6/ip6protosw.h>
97
98 #ifdef IPSEC
99 #include <netipsec/ipsec.h>
100 #include <netipsec/xform.h>
101 #ifdef INET6
102 #include <netipsec/ipsec6.h>
103 #endif
104 #include <netipsec/key.h>
105 #endif /*IPSEC*/
106
107 #include <machine/in_cksum.h>
108 #include <sys/md5.h>
109
110 #include <security/mac/mac_framework.h>
111
112 int tcp_mssdflt = TCP_MSS;
113 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
114 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
115
116 #ifdef INET6
117 int tcp_v6mssdflt = TCP6_MSS;
118 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
119 CTLFLAG_RW, &tcp_v6mssdflt , 0,
120 "Default TCP Maximum Segment Size for IPv6");
121 #endif
122
123 /*
124 * Minimum MSS we accept and use. This prevents DoS attacks where
125 * we are forced to a ridiculous low MSS like 20 and send hundreds
126 * of packets instead of one. The effect scales with the available
127 * bandwidth and quickly saturates the CPU and network interface
128 * with packet generation and sending. Set to zero to disable MINMSS
129 * checking. This setting prevents us from sending too small packets.
130 */
131 int tcp_minmss = TCP_MINMSS;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
133 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
134
135 int tcp_do_rfc1323 = 1;
136 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
137 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
138
139 static int tcp_log_debug = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
141 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
142
143 static int tcp_tcbhashsize = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
145 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
146
147 static int do_tcpdrain = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW,
149 &do_tcpdrain, 0,
150 "Enable tcp_drain routine for extra help when low on mbufs");
151
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
153 &tcbinfo.ipi_count, 0, "Number of active PCBs");
154
155 static int icmp_may_rst = 1;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
157 &icmp_may_rst, 0,
158 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
159
160 static int tcp_isn_reseed_interval = 0;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
162 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
163
164 /*
165 * TCP bandwidth limiting sysctls. Note that the default lower bound of
166 * 1024 exists only for debugging. A good production default would be
167 * something like 6100.
168 */
169 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
170 "TCP inflight data limiting");
171
172 static int tcp_inflight_enable = 1;
173 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
174 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
175
176 static int tcp_inflight_debug = 0;
177 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
178 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
179
180 static int tcp_inflight_rttthresh;
181 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
182 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
183 "RTT threshold below which inflight will deactivate itself");
184
185 static int tcp_inflight_min = 6144;
186 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
187 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
188
189 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
190 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
191 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
192
193 static int tcp_inflight_stab = 20;
194 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
195 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
196
197 uma_zone_t sack_hole_zone;
198
199 static struct inpcb *tcp_notify(struct inpcb *, int);
200 static void tcp_isn_tick(void *);
201
202 /*
203 * Target size of TCP PCB hash tables. Must be a power of two.
204 *
205 * Note that this can be overridden by the kernel environment
206 * variable net.inet.tcp.tcbhashsize
207 */
208 #ifndef TCBHASHSIZE
209 #define TCBHASHSIZE 512
210 #endif
211
212 /*
213 * XXX
214 * Callouts should be moved into struct tcp directly. They are currently
215 * separate because the tcpcb structure is exported to userland for sysctl
216 * parsing purposes, which do not know about callouts.
217 */
218 struct tcpcb_mem {
219 struct tcpcb tcb;
220 struct tcp_timer tt;
221 };
222
223 static uma_zone_t tcpcb_zone;
224 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
225 struct callout isn_callout;
226 static struct mtx isn_mtx;
227
228 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
229 #define ISN_LOCK() mtx_lock(&isn_mtx)
230 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
231
232 /*
233 * TCP initialization.
234 */
235 static void
236 tcp_zone_change(void *tag)
237 {
238
239 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
240 uma_zone_set_max(tcpcb_zone, maxsockets);
241 tcp_tw_zone_change();
242 }
243
244 static int
245 tcp_inpcb_init(void *mem, int size, int flags)
246 {
247 struct inpcb *inp = mem;
248
249 INP_LOCK_INIT(inp, "inp", "tcpinp");
250 return (0);
251 }
252
253 void
254 tcp_init(void)
255 {
256
257 int hashsize = TCBHASHSIZE;
258 tcp_delacktime = TCPTV_DELACK;
259 tcp_keepinit = TCPTV_KEEP_INIT;
260 tcp_keepidle = TCPTV_KEEP_IDLE;
261 tcp_keepintvl = TCPTV_KEEPINTVL;
262 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
263 tcp_msl = TCPTV_MSL;
264 tcp_rexmit_min = TCPTV_MIN;
265 if (tcp_rexmit_min < 1)
266 tcp_rexmit_min = 1;
267 tcp_rexmit_slop = TCPTV_CPU_VAR;
268 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
269 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
270
271 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
272 LIST_INIT(&tcb);
273 tcbinfo.ipi_listhead = &tcb;
274 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
275 if (!powerof2(hashsize)) {
276 printf("WARNING: TCB hash size not a power of 2\n");
277 hashsize = 512; /* safe default */
278 }
279 tcp_tcbhashsize = hashsize;
280 tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
281 &tcbinfo.ipi_hashmask);
282 tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
283 &tcbinfo.ipi_porthashmask);
284 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
285 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
286 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
287 #ifdef INET6
288 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
289 #else /* INET6 */
290 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
291 #endif /* INET6 */
292 if (max_protohdr < TCP_MINPROTOHDR)
293 max_protohdr = TCP_MINPROTOHDR;
294 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
295 panic("tcp_init");
296 #undef TCP_MINPROTOHDR
297 /*
298 * These have to be type stable for the benefit of the timers.
299 */
300 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
301 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
302 uma_zone_set_max(tcpcb_zone, maxsockets);
303 tcp_tw_init();
304 syncache_init();
305 tcp_hc_init();
306 tcp_reass_init();
307 ISN_LOCK_INIT();
308 callout_init(&isn_callout, CALLOUT_MPSAFE);
309 tcp_isn_tick(NULL);
310 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
311 SHUTDOWN_PRI_DEFAULT);
312 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
313 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
315 EVENTHANDLER_PRI_ANY);
316 }
317
318 void
319 tcp_fini(void *xtp)
320 {
321
322 callout_stop(&isn_callout);
323 }
324
325 /*
326 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
327 * tcp_template used to store this data in mbufs, but we now recopy it out
328 * of the tcpcb each time to conserve mbufs.
329 */
330 void
331 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
332 {
333 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
334
335 INP_WLOCK_ASSERT(inp);
336
337 #ifdef INET6
338 if ((inp->inp_vflag & INP_IPV6) != 0) {
339 struct ip6_hdr *ip6;
340
341 ip6 = (struct ip6_hdr *)ip_ptr;
342 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
343 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
344 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
345 (IPV6_VERSION & IPV6_VERSION_MASK);
346 ip6->ip6_nxt = IPPROTO_TCP;
347 ip6->ip6_plen = htons(sizeof(struct tcphdr));
348 ip6->ip6_src = inp->in6p_laddr;
349 ip6->ip6_dst = inp->in6p_faddr;
350 } else
351 #endif
352 {
353 struct ip *ip;
354
355 ip = (struct ip *)ip_ptr;
356 ip->ip_v = IPVERSION;
357 ip->ip_hl = 5;
358 ip->ip_tos = inp->inp_ip_tos;
359 ip->ip_len = 0;
360 ip->ip_id = 0;
361 ip->ip_off = 0;
362 ip->ip_ttl = inp->inp_ip_ttl;
363 ip->ip_sum = 0;
364 ip->ip_p = IPPROTO_TCP;
365 ip->ip_src = inp->inp_laddr;
366 ip->ip_dst = inp->inp_faddr;
367 }
368 th->th_sport = inp->inp_lport;
369 th->th_dport = inp->inp_fport;
370 th->th_seq = 0;
371 th->th_ack = 0;
372 th->th_x2 = 0;
373 th->th_off = 5;
374 th->th_flags = 0;
375 th->th_win = 0;
376 th->th_urp = 0;
377 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
378 }
379
380 /*
381 * Create template to be used to send tcp packets on a connection.
382 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
383 * use for this function is in keepalives, which use tcp_respond.
384 */
385 struct tcptemp *
386 tcpip_maketemplate(struct inpcb *inp)
387 {
388 struct tcptemp *t;
389
390 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
391 if (t == NULL)
392 return (NULL);
393 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
394 return (t);
395 }
396
397 /*
398 * Send a single message to the TCP at address specified by
399 * the given TCP/IP header. If m == NULL, then we make a copy
400 * of the tcpiphdr at ti and send directly to the addressed host.
401 * This is used to force keep alive messages out using the TCP
402 * template for a connection. If flags are given then we send
403 * a message back to the TCP which originated the * segment ti,
404 * and discard the mbuf containing it and any other attached mbufs.
405 *
406 * In any case the ack and sequence number of the transmitted
407 * segment are as specified by the parameters.
408 *
409 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
410 */
411 void
412 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
413 tcp_seq ack, tcp_seq seq, int flags)
414 {
415 int tlen;
416 int win = 0;
417 struct ip *ip;
418 struct tcphdr *nth;
419 #ifdef INET6
420 struct ip6_hdr *ip6;
421 int isipv6;
422 #endif /* INET6 */
423 int ipflags = 0;
424 struct inpcb *inp;
425
426 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
427
428 #ifdef INET6
429 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
430 ip6 = ipgen;
431 #endif /* INET6 */
432 ip = ipgen;
433
434 if (tp != NULL) {
435 inp = tp->t_inpcb;
436 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
437 INP_WLOCK_ASSERT(inp);
438 } else
439 inp = NULL;
440
441 if (tp != NULL) {
442 if (!(flags & TH_RST)) {
443 win = sbspace(&inp->inp_socket->so_rcv);
444 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
445 win = (long)TCP_MAXWIN << tp->rcv_scale;
446 }
447 }
448 if (m == NULL) {
449 m = m_gethdr(M_DONTWAIT, MT_DATA);
450 if (m == NULL)
451 return;
452 tlen = 0;
453 m->m_data += max_linkhdr;
454 #ifdef INET6
455 if (isipv6) {
456 bcopy((caddr_t)ip6, mtod(m, caddr_t),
457 sizeof(struct ip6_hdr));
458 ip6 = mtod(m, struct ip6_hdr *);
459 nth = (struct tcphdr *)(ip6 + 1);
460 } else
461 #endif /* INET6 */
462 {
463 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
464 ip = mtod(m, struct ip *);
465 nth = (struct tcphdr *)(ip + 1);
466 }
467 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
468 flags = TH_ACK;
469 } else {
470 /*
471 * reuse the mbuf.
472 * XXX MRT We inherrit the FIB, which is lucky.
473 */
474 m_freem(m->m_next);
475 m->m_next = NULL;
476 m->m_data = (caddr_t)ipgen;
477 /* m_len is set later */
478 tlen = 0;
479 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
480 #ifdef INET6
481 if (isipv6) {
482 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
483 nth = (struct tcphdr *)(ip6 + 1);
484 } else
485 #endif /* INET6 */
486 {
487 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
488 nth = (struct tcphdr *)(ip + 1);
489 }
490 if (th != nth) {
491 /*
492 * this is usually a case when an extension header
493 * exists between the IPv6 header and the
494 * TCP header.
495 */
496 nth->th_sport = th->th_sport;
497 nth->th_dport = th->th_dport;
498 }
499 xchg(nth->th_dport, nth->th_sport, n_short);
500 #undef xchg
501 }
502 #ifdef INET6
503 if (isipv6) {
504 ip6->ip6_flow = 0;
505 ip6->ip6_vfc = IPV6_VERSION;
506 ip6->ip6_nxt = IPPROTO_TCP;
507 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
508 tlen));
509 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
510 } else
511 #endif
512 {
513 tlen += sizeof (struct tcpiphdr);
514 ip->ip_len = tlen;
515 ip->ip_ttl = ip_defttl;
516 if (path_mtu_discovery)
517 ip->ip_off |= IP_DF;
518 }
519 m->m_len = tlen;
520 m->m_pkthdr.len = tlen;
521 m->m_pkthdr.rcvif = NULL;
522 #ifdef MAC
523 if (inp != NULL) {
524 /*
525 * Packet is associated with a socket, so allow the
526 * label of the response to reflect the socket label.
527 */
528 INP_WLOCK_ASSERT(inp);
529 mac_create_mbuf_from_inpcb(inp, m);
530 } else {
531 /*
532 * Packet is not associated with a socket, so possibly
533 * update the label in place.
534 */
535 mac_reflect_mbuf_tcp(m);
536 }
537 #endif
538 nth->th_seq = htonl(seq);
539 nth->th_ack = htonl(ack);
540 nth->th_x2 = 0;
541 nth->th_off = sizeof (struct tcphdr) >> 2;
542 nth->th_flags = flags;
543 if (tp != NULL)
544 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
545 else
546 nth->th_win = htons((u_short)win);
547 nth->th_urp = 0;
548 #ifdef INET6
549 if (isipv6) {
550 nth->th_sum = 0;
551 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
552 sizeof(struct ip6_hdr),
553 tlen - sizeof(struct ip6_hdr));
554 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
555 NULL, NULL);
556 } else
557 #endif /* INET6 */
558 {
559 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
560 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
561 m->m_pkthdr.csum_flags = CSUM_TCP;
562 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
563 }
564 #ifdef TCPDEBUG
565 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
566 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
567 #endif
568 #ifdef INET6
569 if (isipv6)
570 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
571 else
572 #endif /* INET6 */
573 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
574 }
575
576 /*
577 * Create a new TCP control block, making an
578 * empty reassembly queue and hooking it to the argument
579 * protocol control block. The `inp' parameter must have
580 * come from the zone allocator set up in tcp_init().
581 */
582 struct tcpcb *
583 tcp_newtcpcb(struct inpcb *inp)
584 {
585 struct tcpcb_mem *tm;
586 struct tcpcb *tp;
587 #ifdef INET6
588 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
589 #endif /* INET6 */
590
591 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
592 if (tm == NULL)
593 return (NULL);
594 tp = &tm->tcb;
595 tp->t_timers = &tm->tt;
596 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
597 tp->t_maxseg = tp->t_maxopd =
598 #ifdef INET6
599 isipv6 ? tcp_v6mssdflt :
600 #endif /* INET6 */
601 tcp_mssdflt;
602
603 /* Set up our timeouts. */
604 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
605 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
606 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
607 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
608 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
609
610 if (tcp_do_rfc1323)
611 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
612 if (tcp_do_sack)
613 tp->t_flags |= TF_SACK_PERMIT;
614 TAILQ_INIT(&tp->snd_holes);
615 tp->t_inpcb = inp; /* XXX */
616 /*
617 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
618 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
619 * reasonable initial retransmit time.
620 */
621 tp->t_srtt = TCPTV_SRTTBASE;
622 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
623 tp->t_rttmin = tcp_rexmit_min;
624 tp->t_rxtcur = TCPTV_RTOBASE;
625 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
626 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
627 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
628 tp->t_rcvtime = ticks;
629 tp->t_bw_rtttime = ticks;
630 /*
631 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
632 * because the socket may be bound to an IPv6 wildcard address,
633 * which may match an IPv4-mapped IPv6 address.
634 */
635 inp->inp_ip_ttl = ip_defttl;
636 inp->inp_ppcb = tp;
637 return (tp); /* XXX */
638 }
639
640 /*
641 * Drop a TCP connection, reporting
642 * the specified error. If connection is synchronized,
643 * then send a RST to peer.
644 */
645 struct tcpcb *
646 tcp_drop(struct tcpcb *tp, int errno)
647 {
648 struct socket *so = tp->t_inpcb->inp_socket;
649
650 INP_INFO_WLOCK_ASSERT(&tcbinfo);
651 INP_WLOCK_ASSERT(tp->t_inpcb);
652
653 if (TCPS_HAVERCVDSYN(tp->t_state)) {
654 tp->t_state = TCPS_CLOSED;
655 (void) tcp_output_reset(tp);
656 tcpstat.tcps_drops++;
657 } else
658 tcpstat.tcps_conndrops++;
659 if (errno == ETIMEDOUT && tp->t_softerror)
660 errno = tp->t_softerror;
661 so->so_error = errno;
662 return (tcp_close(tp));
663 }
664
665 void
666 tcp_discardcb(struct tcpcb *tp)
667 {
668 struct tseg_qent *q;
669 struct inpcb *inp = tp->t_inpcb;
670 struct socket *so = inp->inp_socket;
671 #ifdef INET6
672 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
673 #endif /* INET6 */
674
675 INP_WLOCK_ASSERT(inp);
676
677 /*
678 * Make sure that all of our timers are stopped before we
679 * delete the PCB.
680 */
681 callout_stop(&tp->t_timers->tt_rexmt);
682 callout_stop(&tp->t_timers->tt_persist);
683 callout_stop(&tp->t_timers->tt_keep);
684 callout_stop(&tp->t_timers->tt_2msl);
685 callout_stop(&tp->t_timers->tt_delack);
686
687 /*
688 * If we got enough samples through the srtt filter,
689 * save the rtt and rttvar in the routing entry.
690 * 'Enough' is arbitrarily defined as 4 rtt samples.
691 * 4 samples is enough for the srtt filter to converge
692 * to within enough % of the correct value; fewer samples
693 * and we could save a bogus rtt. The danger is not high
694 * as tcp quickly recovers from everything.
695 * XXX: Works very well but needs some more statistics!
696 */
697 if (tp->t_rttupdated >= 4) {
698 struct hc_metrics_lite metrics;
699 u_long ssthresh;
700
701 bzero(&metrics, sizeof(metrics));
702 /*
703 * Update the ssthresh always when the conditions below
704 * are satisfied. This gives us better new start value
705 * for the congestion avoidance for new connections.
706 * ssthresh is only set if packet loss occured on a session.
707 *
708 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
709 * being torn down. Ideally this code would not use 'so'.
710 */
711 ssthresh = tp->snd_ssthresh;
712 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
713 /*
714 * convert the limit from user data bytes to
715 * packets then to packet data bytes.
716 */
717 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
718 if (ssthresh < 2)
719 ssthresh = 2;
720 ssthresh *= (u_long)(tp->t_maxseg +
721 #ifdef INET6
722 (isipv6 ? sizeof (struct ip6_hdr) +
723 sizeof (struct tcphdr) :
724 #endif
725 sizeof (struct tcpiphdr)
726 #ifdef INET6
727 )
728 #endif
729 );
730 } else
731 ssthresh = 0;
732 metrics.rmx_ssthresh = ssthresh;
733
734 metrics.rmx_rtt = tp->t_srtt;
735 metrics.rmx_rttvar = tp->t_rttvar;
736 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
737 metrics.rmx_bandwidth = tp->snd_bandwidth;
738 metrics.rmx_cwnd = tp->snd_cwnd;
739 metrics.rmx_sendpipe = 0;
740 metrics.rmx_recvpipe = 0;
741
742 tcp_hc_update(&inp->inp_inc, &metrics);
743 }
744
745 /* free the reassembly queue, if any */
746 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
747 LIST_REMOVE(q, tqe_q);
748 m_freem(q->tqe_m);
749 uma_zfree(tcp_reass_zone, q);
750 tp->t_segqlen--;
751 tcp_reass_qsize--;
752 }
753 /* Disconnect offload device, if any. */
754 tcp_offload_detach(tp);
755
756 tcp_free_sackholes(tp);
757 inp->inp_ppcb = NULL;
758 tp->t_inpcb = NULL;
759 uma_zfree(tcpcb_zone, tp);
760 }
761
762 /*
763 * Attempt to close a TCP control block, marking it as dropped, and freeing
764 * the socket if we hold the only reference.
765 */
766 struct tcpcb *
767 tcp_close(struct tcpcb *tp)
768 {
769 struct inpcb *inp = tp->t_inpcb;
770 struct socket *so;
771
772 INP_INFO_WLOCK_ASSERT(&tcbinfo);
773 INP_WLOCK_ASSERT(inp);
774
775 /* Notify any offload devices of listener close */
776 if (tp->t_state == TCPS_LISTEN)
777 tcp_offload_listen_close(tp);
778 in_pcbdrop(inp);
779 tcpstat.tcps_closed++;
780 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
781 so = inp->inp_socket;
782 soisdisconnected(so);
783 if (inp->inp_vflag & INP_SOCKREF) {
784 KASSERT(so->so_state & SS_PROTOREF,
785 ("tcp_close: !SS_PROTOREF"));
786 inp->inp_vflag &= ~INP_SOCKREF;
787 INP_WUNLOCK(inp);
788 ACCEPT_LOCK();
789 SOCK_LOCK(so);
790 so->so_state &= ~SS_PROTOREF;
791 sofree(so);
792 return (NULL);
793 }
794 return (tp);
795 }
796
797 void
798 tcp_drain(void)
799 {
800
801 if (do_tcpdrain) {
802 struct inpcb *inpb;
803 struct tcpcb *tcpb;
804 struct tseg_qent *te;
805
806 /*
807 * Walk the tcpbs, if existing, and flush the reassembly queue,
808 * if there is one...
809 * XXX: The "Net/3" implementation doesn't imply that the TCP
810 * reassembly queue should be flushed, but in a situation
811 * where we're really low on mbufs, this is potentially
812 * usefull.
813 */
814 INP_INFO_RLOCK(&tcbinfo);
815 LIST_FOREACH(inpb, tcbinfo.ipi_listhead, inp_list) {
816 if (inpb->inp_vflag & INP_TIMEWAIT)
817 continue;
818 INP_WLOCK(inpb);
819 if ((tcpb = intotcpcb(inpb)) != NULL) {
820 while ((te = LIST_FIRST(&tcpb->t_segq))
821 != NULL) {
822 LIST_REMOVE(te, tqe_q);
823 m_freem(te->tqe_m);
824 uma_zfree(tcp_reass_zone, te);
825 tcpb->t_segqlen--;
826 tcp_reass_qsize--;
827 }
828 tcp_clean_sackreport(tcpb);
829 }
830 INP_WUNLOCK(inpb);
831 }
832 INP_INFO_RUNLOCK(&tcbinfo);
833 }
834 }
835
836 /*
837 * Notify a tcp user of an asynchronous error;
838 * store error as soft error, but wake up user
839 * (for now, won't do anything until can select for soft error).
840 *
841 * Do not wake up user since there currently is no mechanism for
842 * reporting soft errors (yet - a kqueue filter may be added).
843 */
844 static struct inpcb *
845 tcp_notify(struct inpcb *inp, int error)
846 {
847 struct tcpcb *tp;
848
849 INP_INFO_WLOCK_ASSERT(&tcbinfo);
850 INP_WLOCK_ASSERT(inp);
851
852 if ((inp->inp_vflag & INP_TIMEWAIT) ||
853 (inp->inp_vflag & INP_DROPPED))
854 return (inp);
855
856 tp = intotcpcb(inp);
857 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
858
859 /*
860 * Ignore some errors if we are hooked up.
861 * If connection hasn't completed, has retransmitted several times,
862 * and receives a second error, give up now. This is better
863 * than waiting a long time to establish a connection that
864 * can never complete.
865 */
866 if (tp->t_state == TCPS_ESTABLISHED &&
867 (error == EHOSTUNREACH || error == ENETUNREACH ||
868 error == EHOSTDOWN)) {
869 return (inp);
870 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
871 tp->t_softerror) {
872 tp = tcp_drop(tp, error);
873 if (tp != NULL)
874 return (inp);
875 else
876 return (NULL);
877 } else {
878 tp->t_softerror = error;
879 return (inp);
880 }
881 #if 0
882 wakeup( &so->so_timeo);
883 sorwakeup(so);
884 sowwakeup(so);
885 #endif
886 }
887
888 static int
889 tcp_pcblist(SYSCTL_HANDLER_ARGS)
890 {
891 int error, i, m, n, pcb_count;
892 struct inpcb *inp, **inp_list;
893 inp_gen_t gencnt;
894 struct xinpgen xig;
895
896 /*
897 * The process of preparing the TCB list is too time-consuming and
898 * resource-intensive to repeat twice on every request.
899 */
900 if (req->oldptr == NULL) {
901 m = syncache_pcbcount();
902 n = tcbinfo.ipi_count;
903 req->oldidx = 2 * (sizeof xig)
904 + ((m + n) + n/8) * sizeof(struct xtcpcb);
905 return (0);
906 }
907
908 if (req->newptr != NULL)
909 return (EPERM);
910
911 /*
912 * OK, now we're committed to doing something.
913 */
914 INP_INFO_RLOCK(&tcbinfo);
915 gencnt = tcbinfo.ipi_gencnt;
916 n = tcbinfo.ipi_count;
917 INP_INFO_RUNLOCK(&tcbinfo);
918
919 m = syncache_pcbcount();
920
921 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
922 + (n + m) * sizeof(struct xtcpcb));
923 if (error != 0)
924 return (error);
925
926 xig.xig_len = sizeof xig;
927 xig.xig_count = n + m;
928 xig.xig_gen = gencnt;
929 xig.xig_sogen = so_gencnt;
930 error = SYSCTL_OUT(req, &xig, sizeof xig);
931 if (error)
932 return (error);
933
934 error = syncache_pcblist(req, m, &pcb_count);
935 if (error)
936 return (error);
937
938 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
939 if (inp_list == NULL)
940 return (ENOMEM);
941
942 INP_INFO_RLOCK(&tcbinfo);
943 for (inp = LIST_FIRST(tcbinfo.ipi_listhead), i = 0; inp != NULL && i
944 < n; inp = LIST_NEXT(inp, inp_list)) {
945 INP_RLOCK(inp);
946 if (inp->inp_gencnt <= gencnt) {
947 /*
948 * XXX: This use of cr_cansee(), introduced with
949 * TCP state changes, is not quite right, but for
950 * now, better than nothing.
951 */
952 if (inp->inp_vflag & INP_TIMEWAIT) {
953 if (intotw(inp) != NULL)
954 error = cr_cansee(req->td->td_ucred,
955 intotw(inp)->tw_cred);
956 else
957 error = EINVAL; /* Skip this inp. */
958 } else
959 error = cr_canseeinpcb(req->td->td_ucred, inp);
960 if (error == 0)
961 inp_list[i++] = inp;
962 }
963 INP_RUNLOCK(inp);
964 }
965 INP_INFO_RUNLOCK(&tcbinfo);
966 n = i;
967
968 error = 0;
969 for (i = 0; i < n; i++) {
970 inp = inp_list[i];
971 INP_RLOCK(inp);
972 if (inp->inp_gencnt <= gencnt) {
973 struct xtcpcb xt;
974 void *inp_ppcb;
975
976 bzero(&xt, sizeof(xt));
977 xt.xt_len = sizeof xt;
978 /* XXX should avoid extra copy */
979 bcopy(inp, &xt.xt_inp, sizeof *inp);
980 inp_ppcb = inp->inp_ppcb;
981 if (inp_ppcb == NULL)
982 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
983 else if (inp->inp_vflag & INP_TIMEWAIT) {
984 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
985 xt.xt_tp.t_state = TCPS_TIME_WAIT;
986 } else
987 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
988 if (inp->inp_socket != NULL)
989 sotoxsocket(inp->inp_socket, &xt.xt_socket);
990 else {
991 bzero(&xt.xt_socket, sizeof xt.xt_socket);
992 xt.xt_socket.xso_protocol = IPPROTO_TCP;
993 }
994 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
995 INP_RUNLOCK(inp);
996 error = SYSCTL_OUT(req, &xt, sizeof xt);
997 } else
998 INP_RUNLOCK(inp);
999
1000 }
1001 if (!error) {
1002 /*
1003 * Give the user an updated idea of our state.
1004 * If the generation differs from what we told
1005 * her before, she knows that something happened
1006 * while we were processing this request, and it
1007 * might be necessary to retry.
1008 */
1009 INP_INFO_RLOCK(&tcbinfo);
1010 xig.xig_gen = tcbinfo.ipi_gencnt;
1011 xig.xig_sogen = so_gencnt;
1012 xig.xig_count = tcbinfo.ipi_count + pcb_count;
1013 INP_INFO_RUNLOCK(&tcbinfo);
1014 error = SYSCTL_OUT(req, &xig, sizeof xig);
1015 }
1016 free(inp_list, M_TEMP);
1017 return (error);
1018 }
1019
1020 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1021 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1022
1023 static int
1024 tcp_getcred(SYSCTL_HANDLER_ARGS)
1025 {
1026 struct xucred xuc;
1027 struct sockaddr_in addrs[2];
1028 struct inpcb *inp;
1029 int error;
1030
1031 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1032 if (error)
1033 return (error);
1034 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1035 if (error)
1036 return (error);
1037 INP_INFO_RLOCK(&tcbinfo);
1038 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1039 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1040 if (inp != NULL) {
1041 INP_RLOCK(inp);
1042 INP_INFO_RUNLOCK(&tcbinfo);
1043 if (inp->inp_socket == NULL)
1044 error = ENOENT;
1045 if (error == 0)
1046 error = cr_canseeinpcb(req->td->td_ucred, inp);
1047 if (error == 0)
1048 cru2x(inp->inp_cred, &xuc);
1049 INP_RUNLOCK(inp);
1050 } else {
1051 INP_INFO_RUNLOCK(&tcbinfo);
1052 error = ENOENT;
1053 }
1054 if (error == 0)
1055 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1056 return (error);
1057 }
1058
1059 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1060 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1061 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1062
1063 #ifdef INET6
1064 static int
1065 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1066 {
1067 struct xucred xuc;
1068 struct sockaddr_in6 addrs[2];
1069 struct inpcb *inp;
1070 int error, mapped = 0;
1071
1072 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1073 if (error)
1074 return (error);
1075 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1076 if (error)
1077 return (error);
1078 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1079 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1080 return (error);
1081 }
1082 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1083 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1084 mapped = 1;
1085 else
1086 return (EINVAL);
1087 }
1088
1089 INP_INFO_RLOCK(&tcbinfo);
1090 if (mapped == 1)
1091 inp = in_pcblookup_hash(&tcbinfo,
1092 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1093 addrs[1].sin6_port,
1094 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1095 addrs[0].sin6_port,
1096 0, NULL);
1097 else
1098 inp = in6_pcblookup_hash(&tcbinfo,
1099 &addrs[1].sin6_addr, addrs[1].sin6_port,
1100 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1101 if (inp != NULL) {
1102 INP_RLOCK(inp);
1103 INP_INFO_RUNLOCK(&tcbinfo);
1104 if (inp->inp_socket == NULL)
1105 error = ENOENT;
1106 if (error == 0)
1107 error = cr_canseeinpcb(req->td->td_ucred, inp);
1108 if (error == 0)
1109 cru2x(inp->inp_cred, &xuc);
1110 INP_RUNLOCK(inp);
1111 } else {
1112 INP_INFO_RUNLOCK(&tcbinfo);
1113 error = ENOENT;
1114 }
1115 if (error == 0)
1116 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1117 return (error);
1118 }
1119
1120 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1121 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1122 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1123 #endif
1124
1125
1126 void
1127 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1128 {
1129 struct ip *ip = vip;
1130 struct tcphdr *th;
1131 struct in_addr faddr;
1132 struct inpcb *inp;
1133 struct tcpcb *tp;
1134 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1135 struct icmp *icp;
1136 struct in_conninfo inc;
1137 tcp_seq icmp_tcp_seq;
1138 int mtu;
1139
1140 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1141 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1142 return;
1143
1144 if (cmd == PRC_MSGSIZE)
1145 notify = tcp_mtudisc;
1146 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1147 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1148 notify = tcp_drop_syn_sent;
1149 /*
1150 * Redirects don't need to be handled up here.
1151 */
1152 else if (PRC_IS_REDIRECT(cmd))
1153 return;
1154 /*
1155 * Source quench is depreciated.
1156 */
1157 else if (cmd == PRC_QUENCH)
1158 return;
1159 /*
1160 * Hostdead is ugly because it goes linearly through all PCBs.
1161 * XXX: We never get this from ICMP, otherwise it makes an
1162 * excellent DoS attack on machines with many connections.
1163 */
1164 else if (cmd == PRC_HOSTDEAD)
1165 ip = NULL;
1166 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1167 return;
1168 if (ip != NULL) {
1169 icp = (struct icmp *)((caddr_t)ip
1170 - offsetof(struct icmp, icmp_ip));
1171 th = (struct tcphdr *)((caddr_t)ip
1172 + (ip->ip_hl << 2));
1173 INP_INFO_WLOCK(&tcbinfo);
1174 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1175 ip->ip_src, th->th_sport, 0, NULL);
1176 if (inp != NULL) {
1177 INP_WLOCK(inp);
1178 if (!(inp->inp_vflag & INP_TIMEWAIT) &&
1179 !(inp->inp_vflag & INP_DROPPED) &&
1180 !(inp->inp_socket == NULL)) {
1181 icmp_tcp_seq = htonl(th->th_seq);
1182 tp = intotcpcb(inp);
1183 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1184 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1185 if (cmd == PRC_MSGSIZE) {
1186 /*
1187 * MTU discovery:
1188 * If we got a needfrag set the MTU
1189 * in the route to the suggested new
1190 * value (if given) and then notify.
1191 */
1192 bzero(&inc, sizeof(inc));
1193 inc.inc_flags = 0; /* IPv4 */
1194 inc.inc_faddr = faddr;
1195 inc.inc_fibnum =
1196 inp->inp_inc.inc_fibnum;
1197
1198 mtu = ntohs(icp->icmp_nextmtu);
1199 /*
1200 * If no alternative MTU was
1201 * proposed, try the next smaller
1202 * one. ip->ip_len has already
1203 * been swapped in icmp_input().
1204 */
1205 if (!mtu)
1206 mtu = ip_next_mtu(ip->ip_len,
1207 1);
1208 if (mtu < max(296, (tcp_minmss)
1209 + sizeof(struct tcpiphdr)))
1210 mtu = 0;
1211 if (!mtu)
1212 mtu = tcp_mssdflt
1213 + sizeof(struct tcpiphdr);
1214 /*
1215 * Only cache the the MTU if it
1216 * is smaller than the interface
1217 * or route MTU. tcp_mtudisc()
1218 * will do right thing by itself.
1219 */
1220 if (mtu <= tcp_maxmtu(&inc, NULL))
1221 tcp_hc_updatemtu(&inc, mtu);
1222 }
1223
1224 inp = (*notify)(inp, inetctlerrmap[cmd]);
1225 }
1226 }
1227 if (inp != NULL)
1228 INP_WUNLOCK(inp);
1229 } else {
1230 inc.inc_fport = th->th_dport;
1231 inc.inc_lport = th->th_sport;
1232 inc.inc_faddr = faddr;
1233 inc.inc_laddr = ip->ip_src;
1234 #ifdef INET6
1235 inc.inc_isipv6 = 0;
1236 #endif
1237 syncache_unreach(&inc, th);
1238 }
1239 INP_INFO_WUNLOCK(&tcbinfo);
1240 } else
1241 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1242 }
1243
1244 #ifdef INET6
1245 void
1246 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1247 {
1248 struct tcphdr th;
1249 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1250 struct ip6_hdr *ip6;
1251 struct mbuf *m;
1252 struct ip6ctlparam *ip6cp = NULL;
1253 const struct sockaddr_in6 *sa6_src = NULL;
1254 int off;
1255 struct tcp_portonly {
1256 u_int16_t th_sport;
1257 u_int16_t th_dport;
1258 } *thp;
1259
1260 if (sa->sa_family != AF_INET6 ||
1261 sa->sa_len != sizeof(struct sockaddr_in6))
1262 return;
1263
1264 if (cmd == PRC_MSGSIZE)
1265 notify = tcp_mtudisc;
1266 else if (!PRC_IS_REDIRECT(cmd) &&
1267 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1268 return;
1269 /* Source quench is depreciated. */
1270 else if (cmd == PRC_QUENCH)
1271 return;
1272
1273 /* if the parameter is from icmp6, decode it. */
1274 if (d != NULL) {
1275 ip6cp = (struct ip6ctlparam *)d;
1276 m = ip6cp->ip6c_m;
1277 ip6 = ip6cp->ip6c_ip6;
1278 off = ip6cp->ip6c_off;
1279 sa6_src = ip6cp->ip6c_src;
1280 } else {
1281 m = NULL;
1282 ip6 = NULL;
1283 off = 0; /* fool gcc */
1284 sa6_src = &sa6_any;
1285 }
1286
1287 if (ip6 != NULL) {
1288 struct in_conninfo inc;
1289 /*
1290 * XXX: We assume that when IPV6 is non NULL,
1291 * M and OFF are valid.
1292 */
1293
1294 /* check if we can safely examine src and dst ports */
1295 if (m->m_pkthdr.len < off + sizeof(*thp))
1296 return;
1297
1298 bzero(&th, sizeof(th));
1299 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1300
1301 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1302 (struct sockaddr *)ip6cp->ip6c_src,
1303 th.th_sport, cmd, NULL, notify);
1304
1305 inc.inc_fport = th.th_dport;
1306 inc.inc_lport = th.th_sport;
1307 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1308 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1309 inc.inc_isipv6 = 1;
1310 INP_INFO_WLOCK(&tcbinfo);
1311 syncache_unreach(&inc, &th);
1312 INP_INFO_WUNLOCK(&tcbinfo);
1313 } else
1314 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1315 0, cmd, NULL, notify);
1316 }
1317 #endif /* INET6 */
1318
1319
1320 /*
1321 * Following is where TCP initial sequence number generation occurs.
1322 *
1323 * There are two places where we must use initial sequence numbers:
1324 * 1. In SYN-ACK packets.
1325 * 2. In SYN packets.
1326 *
1327 * All ISNs for SYN-ACK packets are generated by the syncache. See
1328 * tcp_syncache.c for details.
1329 *
1330 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1331 * depends on this property. In addition, these ISNs should be
1332 * unguessable so as to prevent connection hijacking. To satisfy
1333 * the requirements of this situation, the algorithm outlined in
1334 * RFC 1948 is used, with only small modifications.
1335 *
1336 * Implementation details:
1337 *
1338 * Time is based off the system timer, and is corrected so that it
1339 * increases by one megabyte per second. This allows for proper
1340 * recycling on high speed LANs while still leaving over an hour
1341 * before rollover.
1342 *
1343 * As reading the *exact* system time is too expensive to be done
1344 * whenever setting up a TCP connection, we increment the time
1345 * offset in two ways. First, a small random positive increment
1346 * is added to isn_offset for each connection that is set up.
1347 * Second, the function tcp_isn_tick fires once per clock tick
1348 * and increments isn_offset as necessary so that sequence numbers
1349 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1350 * random positive increments serve only to ensure that the same
1351 * exact sequence number is never sent out twice (as could otherwise
1352 * happen when a port is recycled in less than the system tick
1353 * interval.)
1354 *
1355 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1356 * between seeding of isn_secret. This is normally set to zero,
1357 * as reseeding should not be necessary.
1358 *
1359 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1360 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1361 * general, this means holding an exclusive (write) lock.
1362 */
1363
1364 #define ISN_BYTES_PER_SECOND 1048576
1365 #define ISN_STATIC_INCREMENT 4096
1366 #define ISN_RANDOM_INCREMENT (4096 - 1)
1367
1368 static u_char isn_secret[32];
1369 static int isn_last_reseed;
1370 static u_int32_t isn_offset, isn_offset_old;
1371 static MD5_CTX isn_ctx;
1372
1373 tcp_seq
1374 tcp_new_isn(struct tcpcb *tp)
1375 {
1376 u_int32_t md5_buffer[4];
1377 tcp_seq new_isn;
1378
1379 INP_WLOCK_ASSERT(tp->t_inpcb);
1380
1381 ISN_LOCK();
1382 /* Seed if this is the first use, reseed if requested. */
1383 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1384 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1385 < (u_int)ticks))) {
1386 read_random(&isn_secret, sizeof(isn_secret));
1387 isn_last_reseed = ticks;
1388 }
1389
1390 /* Compute the md5 hash and return the ISN. */
1391 MD5Init(&isn_ctx);
1392 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1393 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1394 #ifdef INET6
1395 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1396 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1397 sizeof(struct in6_addr));
1398 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1399 sizeof(struct in6_addr));
1400 } else
1401 #endif
1402 {
1403 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1404 sizeof(struct in_addr));
1405 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1406 sizeof(struct in_addr));
1407 }
1408 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1409 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1410 new_isn = (tcp_seq) md5_buffer[0];
1411 isn_offset += ISN_STATIC_INCREMENT +
1412 (arc4random() & ISN_RANDOM_INCREMENT);
1413 new_isn += isn_offset;
1414 ISN_UNLOCK();
1415 return (new_isn);
1416 }
1417
1418 /*
1419 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1420 * to keep time flowing at a relatively constant rate. If the random
1421 * increments have already pushed us past the projected offset, do nothing.
1422 */
1423 static void
1424 tcp_isn_tick(void *xtp)
1425 {
1426 u_int32_t projected_offset;
1427
1428 ISN_LOCK();
1429 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1430
1431 if (SEQ_GT(projected_offset, isn_offset))
1432 isn_offset = projected_offset;
1433
1434 isn_offset_old = isn_offset;
1435 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1436 ISN_UNLOCK();
1437 }
1438
1439 /*
1440 * When a specific ICMP unreachable message is received and the
1441 * connection state is SYN-SENT, drop the connection. This behavior
1442 * is controlled by the icmp_may_rst sysctl.
1443 */
1444 struct inpcb *
1445 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1446 {
1447 struct tcpcb *tp;
1448
1449 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1450 INP_WLOCK_ASSERT(inp);
1451
1452 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1453 (inp->inp_vflag & INP_DROPPED))
1454 return (inp);
1455
1456 tp = intotcpcb(inp);
1457 if (tp->t_state != TCPS_SYN_SENT)
1458 return (inp);
1459
1460 tp = tcp_drop(tp, errno);
1461 if (tp != NULL)
1462 return (inp);
1463 else
1464 return (NULL);
1465 }
1466
1467 /*
1468 * When `need fragmentation' ICMP is received, update our idea of the MSS
1469 * based on the new value in the route. Also nudge TCP to send something,
1470 * since we know the packet we just sent was dropped.
1471 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1472 */
1473 struct inpcb *
1474 tcp_mtudisc(struct inpcb *inp, int errno)
1475 {
1476 struct tcpcb *tp;
1477 struct socket *so = inp->inp_socket;
1478 u_int maxmtu;
1479 u_int romtu;
1480 int mss;
1481 #ifdef INET6
1482 int isipv6;
1483 #endif /* INET6 */
1484
1485 INP_WLOCK_ASSERT(inp);
1486 if ((inp->inp_vflag & INP_TIMEWAIT) ||
1487 (inp->inp_vflag & INP_DROPPED))
1488 return (inp);
1489
1490 tp = intotcpcb(inp);
1491 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1492
1493 #ifdef INET6
1494 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1495 #endif
1496 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1497 romtu =
1498 #ifdef INET6
1499 isipv6 ? tcp_maxmtu6(&inp->inp_inc, NULL) :
1500 #endif /* INET6 */
1501 tcp_maxmtu(&inp->inp_inc, NULL);
1502 if (!maxmtu)
1503 maxmtu = romtu;
1504 else
1505 maxmtu = min(maxmtu, romtu);
1506 if (!maxmtu) {
1507 tp->t_maxopd = tp->t_maxseg =
1508 #ifdef INET6
1509 isipv6 ? tcp_v6mssdflt :
1510 #endif /* INET6 */
1511 tcp_mssdflt;
1512 return (inp);
1513 }
1514 mss = maxmtu -
1515 #ifdef INET6
1516 (isipv6 ? sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1517 #endif /* INET6 */
1518 sizeof(struct tcpiphdr)
1519 #ifdef INET6
1520 )
1521 #endif /* INET6 */
1522 ;
1523
1524 /*
1525 * XXX - The above conditional probably violates the TCP
1526 * spec. The problem is that, since we don't know the
1527 * other end's MSS, we are supposed to use a conservative
1528 * default. But, if we do that, then MTU discovery will
1529 * never actually take place, because the conservative
1530 * default is much less than the MTUs typically seen
1531 * on the Internet today. For the moment, we'll sweep
1532 * this under the carpet.
1533 *
1534 * The conservative default might not actually be a problem
1535 * if the only case this occurs is when sending an initial
1536 * SYN with options and data to a host we've never talked
1537 * to before. Then, they will reply with an MSS value which
1538 * will get recorded and the new parameters should get
1539 * recomputed. For Further Study.
1540 */
1541 if (tp->t_maxopd <= mss)
1542 return (inp);
1543 tp->t_maxopd = mss;
1544
1545 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1546 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1547 mss -= TCPOLEN_TSTAMP_APPA;
1548 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1549 if (mss > MCLBYTES)
1550 mss &= ~(MCLBYTES-1);
1551 #else
1552 if (mss > MCLBYTES)
1553 mss = mss / MCLBYTES * MCLBYTES;
1554 #endif
1555 if (so->so_snd.sb_hiwat < mss)
1556 mss = so->so_snd.sb_hiwat;
1557
1558 tp->t_maxseg = mss;
1559
1560 tcpstat.tcps_mturesent++;
1561 tp->t_rtttime = 0;
1562 tp->snd_nxt = tp->snd_una;
1563 tcp_free_sackholes(tp);
1564 tp->snd_recover = tp->snd_max;
1565 if (tp->t_flags & TF_SACK_PERMIT)
1566 EXIT_FASTRECOVERY(tp);
1567 tcp_output_send(tp);
1568 return (inp);
1569 }
1570
1571 /*
1572 * Look-up the routing entry to the peer of this inpcb. If no route
1573 * is found and it cannot be allocated, then return NULL. This routine
1574 * is called by TCP routines that access the rmx structure and by tcp_mss
1575 * to get the interface MTU.
1576 */
1577 u_long
1578 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1579 {
1580 struct route sro;
1581 struct sockaddr_in *dst;
1582 struct ifnet *ifp;
1583 u_long maxmtu = 0;
1584
1585 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1586
1587 bzero(&sro, sizeof(sro));
1588 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1589 dst = (struct sockaddr_in *)&sro.ro_dst;
1590 dst->sin_family = AF_INET;
1591 dst->sin_len = sizeof(*dst);
1592 dst->sin_addr = inc->inc_faddr;
1593 in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1594 }
1595 if (sro.ro_rt != NULL) {
1596 ifp = sro.ro_rt->rt_ifp;
1597 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1598 maxmtu = ifp->if_mtu;
1599 else
1600 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1601
1602 /* Report additional interface capabilities. */
1603 if (flags != NULL) {
1604 if (ifp->if_capenable & IFCAP_TSO4 &&
1605 ifp->if_hwassist & CSUM_TSO)
1606 *flags |= CSUM_TSO;
1607 }
1608 RTFREE(sro.ro_rt);
1609 }
1610 return (maxmtu);
1611 }
1612
1613 #ifdef INET6
1614 u_long
1615 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1616 {
1617 struct route_in6 sro6;
1618 struct ifnet *ifp;
1619 u_long maxmtu = 0;
1620
1621 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1622
1623 bzero(&sro6, sizeof(sro6));
1624 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1625 sro6.ro_dst.sin6_family = AF_INET6;
1626 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1627 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1628 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1629 }
1630 if (sro6.ro_rt != NULL) {
1631 ifp = sro6.ro_rt->rt_ifp;
1632 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1633 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1634 else
1635 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1636 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1637
1638 /* Report additional interface capabilities. */
1639 if (flags != NULL) {
1640 if (ifp->if_capenable & IFCAP_TSO6 &&
1641 ifp->if_hwassist & CSUM_TSO)
1642 *flags |= CSUM_TSO;
1643 }
1644 RTFREE(sro6.ro_rt);
1645 }
1646
1647 return (maxmtu);
1648 }
1649 #endif /* INET6 */
1650
1651 #ifdef IPSEC
1652 /* compute ESP/AH header size for TCP, including outer IP header. */
1653 size_t
1654 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1655 {
1656 struct inpcb *inp;
1657 struct mbuf *m;
1658 size_t hdrsiz;
1659 struct ip *ip;
1660 #ifdef INET6
1661 struct ip6_hdr *ip6;
1662 #endif
1663 struct tcphdr *th;
1664
1665 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1666 return (0);
1667 MGETHDR(m, M_DONTWAIT, MT_DATA);
1668 if (!m)
1669 return (0);
1670
1671 #ifdef INET6
1672 if ((inp->inp_vflag & INP_IPV6) != 0) {
1673 ip6 = mtod(m, struct ip6_hdr *);
1674 th = (struct tcphdr *)(ip6 + 1);
1675 m->m_pkthdr.len = m->m_len =
1676 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1677 tcpip_fillheaders(inp, ip6, th);
1678 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1679 } else
1680 #endif /* INET6 */
1681 {
1682 ip = mtod(m, struct ip *);
1683 th = (struct tcphdr *)(ip + 1);
1684 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1685 tcpip_fillheaders(inp, ip, th);
1686 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1687 }
1688
1689 m_free(m);
1690 return (hdrsiz);
1691 }
1692 #endif /* IPSEC */
1693
1694 /*
1695 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1696 *
1697 * This code attempts to calculate the bandwidth-delay product as a
1698 * means of determining the optimal window size to maximize bandwidth,
1699 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1700 * routers. This code also does a fairly good job keeping RTTs in check
1701 * across slow links like modems. We implement an algorithm which is very
1702 * similar (but not meant to be) TCP/Vegas. The code operates on the
1703 * transmitter side of a TCP connection and so only effects the transmit
1704 * side of the connection.
1705 *
1706 * BACKGROUND: TCP makes no provision for the management of buffer space
1707 * at the end points or at the intermediate routers and switches. A TCP
1708 * stream, whether using NewReno or not, will eventually buffer as
1709 * many packets as it is able and the only reason this typically works is
1710 * due to the fairly small default buffers made available for a connection
1711 * (typicaly 16K or 32K). As machines use larger windows and/or window
1712 * scaling it is now fairly easy for even a single TCP connection to blow-out
1713 * all available buffer space not only on the local interface, but on
1714 * intermediate routers and switches as well. NewReno makes a misguided
1715 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1716 * then backing off, then steadily increasing the window again until another
1717 * failure occurs, ad-infinitum. This results in terrible oscillation that
1718 * is only made worse as network loads increase and the idea of intentionally
1719 * blowing out network buffers is, frankly, a terrible way to manage network
1720 * resources.
1721 *
1722 * It is far better to limit the transmit window prior to the failure
1723 * condition being achieved. There are two general ways to do this: First
1724 * you can 'scan' through different transmit window sizes and locate the
1725 * point where the RTT stops increasing, indicating that you have filled the
1726 * pipe, then scan backwards until you note that RTT stops decreasing, then
1727 * repeat ad-infinitum. This method works in principle but has severe
1728 * implementation issues due to RTT variances, timer granularity, and
1729 * instability in the algorithm which can lead to many false positives and
1730 * create oscillations as well as interact badly with other TCP streams
1731 * implementing the same algorithm.
1732 *
1733 * The second method is to limit the window to the bandwidth delay product
1734 * of the link. This is the method we implement. RTT variances and our
1735 * own manipulation of the congestion window, bwnd, can potentially
1736 * destabilize the algorithm. For this reason we have to stabilize the
1737 * elements used to calculate the window. We do this by using the minimum
1738 * observed RTT, the long term average of the observed bandwidth, and
1739 * by adding two segments worth of slop. It isn't perfect but it is able
1740 * to react to changing conditions and gives us a very stable basis on
1741 * which to extend the algorithm.
1742 */
1743 void
1744 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1745 {
1746 u_long bw;
1747 u_long bwnd;
1748 int save_ticks;
1749
1750 INP_WLOCK_ASSERT(tp->t_inpcb);
1751
1752 /*
1753 * If inflight_enable is disabled in the middle of a tcp connection,
1754 * make sure snd_bwnd is effectively disabled.
1755 */
1756 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
1757 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1758 tp->snd_bandwidth = 0;
1759 return;
1760 }
1761
1762 /*
1763 * Figure out the bandwidth. Due to the tick granularity this
1764 * is a very rough number and it MUST be averaged over a fairly
1765 * long period of time. XXX we need to take into account a link
1766 * that is not using all available bandwidth, but for now our
1767 * slop will ramp us up if this case occurs and the bandwidth later
1768 * increases.
1769 *
1770 * Note: if ticks rollover 'bw' may wind up negative. We must
1771 * effectively reset t_bw_rtttime for this case.
1772 */
1773 save_ticks = ticks;
1774 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1775 return;
1776
1777 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1778 (save_ticks - tp->t_bw_rtttime);
1779 tp->t_bw_rtttime = save_ticks;
1780 tp->t_bw_rtseq = ack_seq;
1781 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1782 return;
1783 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1784
1785 tp->snd_bandwidth = bw;
1786
1787 /*
1788 * Calculate the semi-static bandwidth delay product, plus two maximal
1789 * segments. The additional slop puts us squarely in the sweet
1790 * spot and also handles the bandwidth run-up case and stabilization.
1791 * Without the slop we could be locking ourselves into a lower
1792 * bandwidth.
1793 *
1794 * Situations Handled:
1795 * (1) Prevents over-queueing of packets on LANs, especially on
1796 * high speed LANs, allowing larger TCP buffers to be
1797 * specified, and also does a good job preventing
1798 * over-queueing of packets over choke points like modems
1799 * (at least for the transmit side).
1800 *
1801 * (2) Is able to handle changing network loads (bandwidth
1802 * drops so bwnd drops, bandwidth increases so bwnd
1803 * increases).
1804 *
1805 * (3) Theoretically should stabilize in the face of multiple
1806 * connections implementing the same algorithm (this may need
1807 * a little work).
1808 *
1809 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1810 * be adjusted with a sysctl but typically only needs to be
1811 * on very slow connections. A value no smaller then 5
1812 * should be used, but only reduce this default if you have
1813 * no other choice.
1814 */
1815 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1816 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
1817 #undef USERTT
1818
1819 if (tcp_inflight_debug > 0) {
1820 static int ltime;
1821 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1822 ltime = ticks;
1823 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1824 tp,
1825 bw,
1826 tp->t_rttbest,
1827 tp->t_srtt,
1828 bwnd
1829 );
1830 }
1831 }
1832 if ((long)bwnd < tcp_inflight_min)
1833 bwnd = tcp_inflight_min;
1834 if (bwnd > tcp_inflight_max)
1835 bwnd = tcp_inflight_max;
1836 if ((long)bwnd < tp->t_maxseg * 2)
1837 bwnd = tp->t_maxseg * 2;
1838 tp->snd_bwnd = bwnd;
1839 }
1840
1841 #ifdef TCP_SIGNATURE
1842 /*
1843 * Callback function invoked by m_apply() to digest TCP segment data
1844 * contained within an mbuf chain.
1845 */
1846 static int
1847 tcp_signature_apply(void *fstate, void *data, u_int len)
1848 {
1849
1850 MD5Update(fstate, (u_char *)data, len);
1851 return (0);
1852 }
1853
1854 /*
1855 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
1856 *
1857 * Parameters:
1858 * m pointer to head of mbuf chain
1859 * off0 offset to TCP header within the mbuf chain
1860 * len length of TCP segment data, excluding options
1861 * optlen length of TCP segment options
1862 * buf pointer to storage for computed MD5 digest
1863 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1864 *
1865 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1866 * When called from tcp_input(), we can be sure that th_sum has been
1867 * zeroed out and verified already.
1868 *
1869 * This function is for IPv4 use only. Calling this function with an
1870 * IPv6 packet in the mbuf chain will yield undefined results.
1871 *
1872 * Return 0 if successful, otherwise return -1.
1873 *
1874 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1875 * search with the destination IP address, and a 'magic SPI' to be
1876 * determined by the application. This is hardcoded elsewhere to 1179
1877 * right now. Another branch of this code exists which uses the SPD to
1878 * specify per-application flows but it is unstable.
1879 */
1880 int
1881 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
1882 u_char *buf, u_int direction)
1883 {
1884 union sockaddr_union dst;
1885 struct ippseudo ippseudo;
1886 MD5_CTX ctx;
1887 int doff;
1888 struct ip *ip;
1889 struct ipovly *ipovly;
1890 struct secasvar *sav;
1891 struct tcphdr *th;
1892 u_short savecsum;
1893
1894 KASSERT(m != NULL, ("NULL mbuf chain"));
1895 KASSERT(buf != NULL, ("NULL signature pointer"));
1896
1897 /* Extract the destination from the IP header in the mbuf. */
1898 ip = mtod(m, struct ip *);
1899 bzero(&dst, sizeof(union sockaddr_union));
1900 dst.sa.sa_len = sizeof(struct sockaddr_in);
1901 dst.sa.sa_family = AF_INET;
1902 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1903 ip->ip_src : ip->ip_dst;
1904
1905 /* Look up an SADB entry which matches the address of the peer. */
1906 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1907 if (sav == NULL) {
1908 printf("%s: SADB lookup failed for %s\n", __func__,
1909 inet_ntoa(dst.sin.sin_addr));
1910 return (EINVAL);
1911 }
1912
1913 MD5Init(&ctx);
1914 ipovly = (struct ipovly *)ip;
1915 th = (struct tcphdr *)((u_char *)ip + off0);
1916 doff = off0 + sizeof(struct tcphdr) + optlen;
1917
1918 /*
1919 * Step 1: Update MD5 hash with IP pseudo-header.
1920 *
1921 * XXX The ippseudo header MUST be digested in network byte order,
1922 * or else we'll fail the regression test. Assume all fields we've
1923 * been doing arithmetic on have been in host byte order.
1924 * XXX One cannot depend on ipovly->ih_len here. When called from
1925 * tcp_output(), the underlying ip_len member has not yet been set.
1926 */
1927 ippseudo.ippseudo_src = ipovly->ih_src;
1928 ippseudo.ippseudo_dst = ipovly->ih_dst;
1929 ippseudo.ippseudo_pad = 0;
1930 ippseudo.ippseudo_p = IPPROTO_TCP;
1931 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
1932 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1933
1934 /*
1935 * Step 2: Update MD5 hash with TCP header, excluding options.
1936 * The TCP checksum must be set to zero.
1937 */
1938 savecsum = th->th_sum;
1939 th->th_sum = 0;
1940 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1941 th->th_sum = savecsum;
1942
1943 /*
1944 * Step 3: Update MD5 hash with TCP segment data.
1945 * Use m_apply() to avoid an early m_pullup().
1946 */
1947 if (len > 0)
1948 m_apply(m, doff, len, tcp_signature_apply, &ctx);
1949
1950 /*
1951 * Step 4: Update MD5 hash with shared secret.
1952 */
1953 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1954 MD5Final(buf, &ctx);
1955
1956 key_sa_recordxfer(sav, m);
1957 KEY_FREESAV(&sav);
1958 return (0);
1959 }
1960 #endif /* TCP_SIGNATURE */
1961
1962 static int
1963 sysctl_drop(SYSCTL_HANDLER_ARGS)
1964 {
1965 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
1966 struct sockaddr_storage addrs[2];
1967 struct inpcb *inp;
1968 struct tcpcb *tp;
1969 struct tcptw *tw;
1970 struct sockaddr_in *fin, *lin;
1971 #ifdef INET6
1972 struct sockaddr_in6 *fin6, *lin6;
1973 struct in6_addr f6, l6;
1974 #endif
1975 int error;
1976
1977 inp = NULL;
1978 fin = lin = NULL;
1979 #ifdef INET6
1980 fin6 = lin6 = NULL;
1981 #endif
1982 error = 0;
1983
1984 if (req->oldptr != NULL || req->oldlen != 0)
1985 return (EINVAL);
1986 if (req->newptr == NULL)
1987 return (EPERM);
1988 if (req->newlen < sizeof(addrs))
1989 return (ENOMEM);
1990 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
1991 if (error)
1992 return (error);
1993
1994 switch (addrs[0].ss_family) {
1995 #ifdef INET6
1996 case AF_INET6:
1997 fin6 = (struct sockaddr_in6 *)&addrs[0];
1998 lin6 = (struct sockaddr_in6 *)&addrs[1];
1999 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2000 lin6->sin6_len != sizeof(struct sockaddr_in6))
2001 return (EINVAL);
2002 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2003 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2004 return (EINVAL);
2005 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2006 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2007 fin = (struct sockaddr_in *)&addrs[0];
2008 lin = (struct sockaddr_in *)&addrs[1];
2009 break;
2010 }
2011 error = sa6_embedscope(fin6, ip6_use_defzone);
2012 if (error)
2013 return (error);
2014 error = sa6_embedscope(lin6, ip6_use_defzone);
2015 if (error)
2016 return (error);
2017 break;
2018 #endif
2019 case AF_INET:
2020 fin = (struct sockaddr_in *)&addrs[0];
2021 lin = (struct sockaddr_in *)&addrs[1];
2022 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2023 lin->sin_len != sizeof(struct sockaddr_in))
2024 return (EINVAL);
2025 break;
2026 default:
2027 return (EINVAL);
2028 }
2029 INP_INFO_WLOCK(&tcbinfo);
2030 switch (addrs[0].ss_family) {
2031 #ifdef INET6
2032 case AF_INET6:
2033 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2034 &l6, lin6->sin6_port, 0, NULL);
2035 break;
2036 #endif
2037 case AF_INET:
2038 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2039 lin->sin_addr, lin->sin_port, 0, NULL);
2040 break;
2041 }
2042 if (inp != NULL) {
2043 INP_WLOCK(inp);
2044 if (inp->inp_vflag & INP_TIMEWAIT) {
2045 /*
2046 * XXXRW: There currently exists a state where an
2047 * inpcb is present, but its timewait state has been
2048 * discarded. For now, don't allow dropping of this
2049 * type of inpcb.
2050 */
2051 tw = intotw(inp);
2052 if (tw != NULL)
2053 tcp_twclose(tw, 0);
2054 else
2055 INP_WUNLOCK(inp);
2056 } else if (!(inp->inp_vflag & INP_DROPPED) &&
2057 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2058 tp = intotcpcb(inp);
2059 tp = tcp_drop(tp, ECONNABORTED);
2060 if (tp != NULL)
2061 INP_WUNLOCK(inp);
2062 } else
2063 INP_WUNLOCK(inp);
2064 } else
2065 error = ESRCH;
2066 INP_INFO_WUNLOCK(&tcbinfo);
2067 return (error);
2068 }
2069
2070 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2071 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2072 0, sysctl_drop, "", "Drop TCP connection");
2073
2074 /*
2075 * Generate a standardized TCP log line for use throughout the
2076 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2077 * allow use in the interrupt context.
2078 *
2079 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2080 * NB: The function may return NULL if memory allocation failed.
2081 *
2082 * Due to header inclusion and ordering limitations the struct ip
2083 * and ip6_hdr pointers have to be passed as void pointers.
2084 */
2085 char *
2086 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2087 const void *ip6hdr)
2088 {
2089 char *s, *sp;
2090 size_t size;
2091 struct ip *ip;
2092 #ifdef INET6
2093 const struct ip6_hdr *ip6;
2094
2095 ip6 = (const struct ip6_hdr *)ip6hdr;
2096 #endif /* INET6 */
2097 ip = (struct ip *)ip4hdr;
2098
2099 /*
2100 * The log line looks like this:
2101 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2102 */
2103 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2104 sizeof(PRINT_TH_FLAGS) + 1 +
2105 #ifdef INET6
2106 2 * INET6_ADDRSTRLEN;
2107 #else
2108 2 * INET_ADDRSTRLEN;
2109 #endif /* INET6 */
2110
2111 /* Is logging enabled? */
2112 if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2113 return (NULL);
2114
2115 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2116 if (s == NULL)
2117 return (NULL);
2118
2119 strcat(s, "TCP: [");
2120 sp = s + strlen(s);
2121
2122 if (inc && inc->inc_isipv6 == 0) {
2123 inet_ntoa_r(inc->inc_faddr, sp);
2124 sp = s + strlen(s);
2125 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2126 sp = s + strlen(s);
2127 inet_ntoa_r(inc->inc_laddr, sp);
2128 sp = s + strlen(s);
2129 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2130 #ifdef INET6
2131 } else if (inc) {
2132 ip6_sprintf(sp, &inc->inc6_faddr);
2133 sp = s + strlen(s);
2134 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2135 sp = s + strlen(s);
2136 ip6_sprintf(sp, &inc->inc6_laddr);
2137 sp = s + strlen(s);
2138 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2139 } else if (ip6 && th) {
2140 ip6_sprintf(sp, &ip6->ip6_src);
2141 sp = s + strlen(s);
2142 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2143 sp = s + strlen(s);
2144 ip6_sprintf(sp, &ip6->ip6_dst);
2145 sp = s + strlen(s);
2146 sprintf(sp, "]:%i", ntohs(th->th_dport));
2147 #endif /* INET6 */
2148 } else if (ip && th) {
2149 inet_ntoa_r(ip->ip_src, sp);
2150 sp = s + strlen(s);
2151 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2152 sp = s + strlen(s);
2153 inet_ntoa_r(ip->ip_dst, sp);
2154 sp = s + strlen(s);
2155 sprintf(sp, "]:%i", ntohs(th->th_dport));
2156 } else {
2157 free(s, M_TCPLOG);
2158 return (NULL);
2159 }
2160 sp = s + strlen(s);
2161 if (th)
2162 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2163 if (*(s + size - 1) != '\0')
2164 panic("%s: string too long", __func__);
2165 return (s);
2166 }
Cache object: b28b867a7141f22ffb1c8bce919b04b5
|