2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 4. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
62 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
63 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $
66 #include "opt_compat.h"
68 #include "opt_inet6.h"
69 #include "opt_ipsec.h"
70 #include "opt_tcpdebug.h"
72 #include <sys/param.h>
73 #include <sys/systm.h>
74 #include <sys/callout.h>
75 #include <sys/kernel.h>
76 #include <sys/sysctl.h>
77 #include <sys/malloc.h>
78 #include <sys/mpipe.h>
81 #include <sys/domain.h>
85 #include <sys/socket.h>
86 #include <sys/socketops.h>
87 #include <sys/socketvar.h>
88 #include <sys/protosw.h>
89 #include <sys/random.h>
90 #include <sys/in_cksum.h>
93 #include <net/route.h>
95 #include <net/netisr2.h>
98 #include <netinet/in.h>
99 #include <netinet/in_systm.h>
100 #include <netinet/ip.h>
101 #include <netinet/ip6.h>
102 #include <netinet/in_pcb.h>
103 #include <netinet6/in6_pcb.h>
104 #include <netinet/in_var.h>
105 #include <netinet/ip_var.h>
106 #include <netinet6/ip6_var.h>
107 #include <netinet/ip_icmp.h>
109 #include <netinet/icmp6.h>
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_timer2.h>
116 #include <netinet/tcp_var.h>
117 #include <netinet6/tcp6_var.h>
118 #include <netinet/tcpip.h>
120 #include <netinet/tcp_debug.h>
122 #include <netinet6/ip6protosw.h>
125 #include <netinet6/ipsec.h>
126 #include <netproto/key/key.h>
128 #include <netinet6/ipsec6.h>
133 #include <netproto/ipsec/ipsec.h>
135 #include <netproto/ipsec/ipsec6.h>
141 #include <machine/smp.h>
143 #include <sys/msgport2.h>
144 #include <sys/mplock2.h>
145 #include <net/netmsg2.h>
147 #if !defined(KTR_TCP)
148 #define KTR_TCP KTR_ALL
151 KTR_INFO_MASTER(tcp);
152 KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0);
153 KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0);
154 KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0);
155 #define logtcp(name) KTR_LOG(tcp_ ## name)
158 #define TCP_IW_MAXSEGS_DFLT 4
159 #define TCP_IW_CAPSEGS_DFLT 3
161 struct inpcbinfo tcbinfo[MAXCPU];
162 struct tcpcbackqhead tcpcbackq[MAXCPU];
164 static struct lwkt_token tcp_port_token =
165 LWKT_TOKEN_INITIALIZER(tcp_port_token);
167 int tcp_mssdflt = TCP_MSS;
168 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
169 &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");
172 int tcp_v6mssdflt = TCP6_MSS;
173 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW,
174 &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6");
178 * Minimum MSS we accept and use. This prevents DoS attacks where
179 * we are forced to a ridiculous low MSS like 20 and send hundreds
180 * of packets instead of one. The effect scales with the available
181 * bandwidth and quickly saturates the CPU and network interface
182 * with packet generation and sending. Set to zero to disable MINMSS
183 * checking. This setting prevents us from sending too small packets.
185 int tcp_minmss = TCP_MINMSS;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
187 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
190 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
191 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
192 &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time");
195 int tcp_do_rfc1323 = 1;
196 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
197 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
199 static int tcp_tcbhashsize = 0;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
201 &tcp_tcbhashsize, 0, "Size of TCP control block hashtable");
203 static int do_tcpdrain = 1;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
205 "Enable tcp_drain routine for extra help when low on mbufs");
207 static int icmp_may_rst = 1;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
209 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
211 static int tcp_isn_reseed_interval = 0;
212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
213 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
216 * TCP bandwidth limiting sysctls. The inflight limiter is now turned on
217 * by default, but with generous values which should allow maximal
218 * bandwidth. In particular, the slop defaults to 50 (5 packets).
220 * The reason for doing this is that the limiter is the only mechanism we
221 * have which seems to do a really good job preventing receiver RX rings
222 * on network interfaces from getting blown out. Even though GigE/10GigE
223 * is supposed to flow control it looks like either it doesn't actually
224 * do it or Open Source drivers do not properly enable it.
226 * People using the limiter to reduce bottlenecks on slower WAN connections
227 * should set the slop to 20 (2 packets).
229 static int tcp_inflight_enable = 1;
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW,
231 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
233 static int tcp_inflight_debug = 0;
234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW,
235 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
237 static int tcp_inflight_min = 6144;
238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW,
239 &tcp_inflight_min, 0, "Lower bound for TCP inflight window");
241 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW,
243 &tcp_inflight_max, 0, "Upper bound for TCP inflight window");
245 static int tcp_inflight_stab = 50;
246 SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW,
247 &tcp_inflight_stab, 0, "Slop in maximal packets / 10 (20 = 3 packets)");
249 static int tcp_do_rfc3390 = 1;
250 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
252 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
254 static u_long tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT;
255 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwmaxsegs, CTLFLAG_RW,
256 &tcp_iw_maxsegs, 0, "TCP IW segments max");
258 static u_long tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT;
259 SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwcapsegs, CTLFLAG_RW,
260 &tcp_iw_capsegs, 0, "TCP IW segments");
262 int tcp_low_rtobase = 1;
263 SYSCTL_INT(_net_inet_tcp, OID_AUTO, low_rtobase, CTLFLAG_RW,
264 &tcp_low_rtobase, 0, "Lowering the Initial RTO (RFC 6298)");
266 static int tcp_do_ncr = 1;
267 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr, CTLFLAG_RW,
268 &tcp_do_ncr, 0, "Non-Congestion Robustness (RFC 4653)");
270 static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives");
271 static struct malloc_pipe tcptemp_mpipe;
273 static void tcp_willblock(void);
274 static void tcp_notify (struct inpcb *, int);
276 struct tcp_stats tcpstats_percpu[MAXCPU] __cachealign;
279 sysctl_tcpstats(SYSCTL_HANDLER_ARGS)
283 for (cpu = 0; cpu < ncpus; ++cpu) {
284 if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu],
285 sizeof(struct tcp_stats))))
287 if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu],
288 sizeof(struct tcp_stats))))
294 SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
295 0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics");
298 * Target size of TCP PCB hash tables. Must be a power of two.
300 * Note that this can be overridden by the kernel environment
301 * variable net.inet.tcp.tcbhashsize
304 #define TCBHASHSIZE 512
308 * This is the actual shape of what we allocate using the zone
309 * allocator. Doing it this way allows us to protect both structures
310 * using the same generation count, and also eliminates the overhead
311 * of allocating tcpcbs separately. By hiding the structure here,
312 * we avoid changing most of the rest of the code (although it needs
313 * to be changed, eventually, for greater efficiency).
316 #define ALIGNM1 (ALIGNMENT - 1)
320 char align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
323 struct tcp_callout inp_tp_rexmt;
324 struct tcp_callout inp_tp_persist;
325 struct tcp_callout inp_tp_keep;
326 struct tcp_callout inp_tp_2msl;
327 struct tcp_callout inp_tp_delack;
328 struct netmsg_tcp_timer inp_tp_timermsg;
329 struct netmsg_base inp_tp_sndmore;
340 struct inpcbporthead *porthashbase;
341 struct inpcbinfo *ticb;
343 int hashsize = TCBHASHSIZE;
347 * note: tcptemp is used for keepalives, and it is ok for an
348 * allocation to fail so do not specify MPF_INT.
350 mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp),
351 25, -1, 0, NULL, NULL, NULL);
353 tcp_delacktime = TCPTV_DELACK;
354 tcp_keepinit = TCPTV_KEEP_INIT;
355 tcp_keepidle = TCPTV_KEEP_IDLE;
356 tcp_keepintvl = TCPTV_KEEPINTVL;
357 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
359 tcp_rexmit_min = TCPTV_MIN;
360 tcp_rexmit_slop = TCPTV_CPU_VAR;
362 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
363 if (!powerof2(hashsize)) {
364 kprintf("WARNING: TCB hash size not a power of 2\n");
365 hashsize = 512; /* safe default */
367 tcp_tcbhashsize = hashsize;
368 porthashbase = hashinit(hashsize, M_PCB, &porthashmask);
370 for (cpu = 0; cpu < ncpus2; cpu++) {
371 ticb = &tcbinfo[cpu];
372 in_pcbinfo_init(ticb);
374 ticb->hashbase = hashinit(hashsize, M_PCB,
376 ticb->porthashbase = porthashbase;
377 ticb->porthashmask = porthashmask;
378 ticb->porttoken = &tcp_port_token;
380 ticb->porthashbase = hashinit(hashsize, M_PCB,
381 &ticb->porthashmask);
383 ticb->wildcardhashbase = hashinit(hashsize, M_PCB,
384 &ticb->wildcardhashmask);
385 ticb->ipi_size = sizeof(struct inp_tp);
386 TAILQ_INIT(&tcpcbackq[cpu]);
389 tcp_reass_maxseg = nmbclusters / 16;
390 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg);
393 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
395 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
397 if (max_protohdr < TCP_MINPROTOHDR)
398 max_protohdr = TCP_MINPROTOHDR;
399 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
401 #undef TCP_MINPROTOHDR
404 * Initialize TCP statistics counters for each CPU.
406 for (cpu = 0; cpu < ncpus; ++cpu) {
407 bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats));
411 netisr_register_rollup(tcp_willblock, NETISR_ROLLUP_PRIO_TCP);
418 int cpu = mycpu->gd_cpuid;
420 while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) {
421 KKASSERT(tp->t_flags & TF_ONOUTPUTQ);
422 tp->t_flags &= ~TF_ONOUTPUTQ;
423 TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq);
429 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
430 * tcp_template used to store this data in mbufs, but we now recopy it out
431 * of the tcpcb each time to conserve mbufs.
434 tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr, boolean_t tso)
436 struct inpcb *inp = tp->t_inpcb;
437 struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;
440 if (inp->inp_vflag & INP_IPV6) {
443 ip6 = (struct ip6_hdr *)ip_ptr;
444 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
445 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
446 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
447 (IPV6_VERSION & IPV6_VERSION_MASK);
448 ip6->ip6_nxt = IPPROTO_TCP;
449 ip6->ip6_plen = sizeof(struct tcphdr);
450 ip6->ip6_src = inp->in6p_laddr;
451 ip6->ip6_dst = inp->in6p_faddr;
456 struct ip *ip = (struct ip *) ip_ptr;
459 ip->ip_vhl = IP_VHL_BORING;
466 ip->ip_p = IPPROTO_TCP;
467 ip->ip_src = inp->inp_laddr;
468 ip->ip_dst = inp->inp_faddr;
471 plen = htons(IPPROTO_TCP);
473 plen = htons(sizeof(struct tcphdr) + IPPROTO_TCP);
474 tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr,
475 ip->ip_dst.s_addr, plen);
478 tcp_hdr->th_sport = inp->inp_lport;
479 tcp_hdr->th_dport = inp->inp_fport;
484 tcp_hdr->th_flags = 0;
490 * Create template to be used to send tcp packets on a connection.
491 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
492 * use for this function is in keepalives, which use tcp_respond.
495 tcp_maketemplate(struct tcpcb *tp)
499 if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL)
501 tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t, FALSE);
506 tcp_freetemplate(struct tcptemp *tmp)
508 mpipe_free(&tcptemp_mpipe, tmp);
512 * Send a single message to the TCP at address specified by
513 * the given TCP/IP header. If m == NULL, then we make a copy
514 * of the tcpiphdr at ti and send directly to the addressed host.
515 * This is used to force keep alive messages out using the TCP
516 * template for a connection. If flags are given then we send
517 * a message back to the TCP which originated the * segment ti,
518 * and discard the mbuf containing it and any other attached mbufs.
520 * In any case the ack and sequence number of the transmitted
521 * segment are as specified by the parameters.
523 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
526 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
527 tcp_seq ack, tcp_seq seq, int flags)
531 struct route *ro = NULL;
533 struct ip *ip = ipgen;
536 struct route_in6 *ro6 = NULL;
537 struct route_in6 sro6;
538 struct ip6_hdr *ip6 = ipgen;
539 boolean_t use_tmpro = TRUE;
541 boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6);
543 const boolean_t isipv6 = FALSE;
547 if (!(flags & TH_RST)) {
548 win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv);
551 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
552 win = (long)TCP_MAXWIN << tp->rcv_scale;
555 * Don't use the route cache of a listen socket,
556 * it is not MPSAFE; use temporary route cache.
558 if (tp->t_state != TCPS_LISTEN) {
560 ro6 = &tp->t_inpcb->in6p_route;
562 ro = &tp->t_inpcb->inp_route;
569 bzero(ro6, sizeof *ro6);
572 bzero(ro, sizeof *ro);
576 m = m_gethdr(MB_DONTWAIT, MT_HEADER);
580 m->m_data += max_linkhdr;
582 bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr));
583 ip6 = mtod(m, struct ip6_hdr *);
584 nth = (struct tcphdr *)(ip6 + 1);
586 bcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
587 ip = mtod(m, struct ip *);
588 nth = (struct tcphdr *)(ip + 1);
590 bcopy(th, nth, sizeof(struct tcphdr));
595 m->m_data = (caddr_t)ipgen;
596 /* m_len is set later */
598 #define xchg(a, b, type) { type t; t = a; a = b; b = t; }
600 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
601 nth = (struct tcphdr *)(ip6 + 1);
603 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
604 nth = (struct tcphdr *)(ip + 1);
608 * this is usually a case when an extension header
609 * exists between the IPv6 header and the
612 nth->th_sport = th->th_sport;
613 nth->th_dport = th->th_dport;
615 xchg(nth->th_dport, nth->th_sport, n_short);
620 ip6->ip6_vfc = IPV6_VERSION;
621 ip6->ip6_nxt = IPPROTO_TCP;
622 ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen));
623 tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
625 tlen += sizeof(struct tcpiphdr);
627 ip->ip_ttl = ip_defttl;
630 m->m_pkthdr.len = tlen;
631 m->m_pkthdr.rcvif = NULL;
632 nth->th_seq = htonl(seq);
633 nth->th_ack = htonl(ack);
635 nth->th_off = sizeof(struct tcphdr) >> 2;
636 nth->th_flags = flags;
638 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
640 nth->th_win = htons((u_short)win);
644 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
645 sizeof(struct ip6_hdr),
646 tlen - sizeof(struct ip6_hdr));
647 ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
648 (ro6 && ro6->ro_rt) ?
649 ro6->ro_rt->rt_ifp : NULL);
651 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
652 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
653 m->m_pkthdr.csum_flags = CSUM_TCP;
654 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
655 m->m_pkthdr.csum_thlen = sizeof(struct tcphdr);
658 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
659 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
662 ip6_output(m, NULL, ro6, ipflags, NULL, NULL,
663 tp ? tp->t_inpcb : NULL);
664 if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) {
669 ipflags |= IP_DEBUGROUTE;
670 ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL);
671 if ((ro == &sro) && (ro->ro_rt != NULL)) {
679 * Create a new TCP control block, making an
680 * empty reassembly queue and hooking it to the argument
681 * protocol control block. The `inp' parameter must have
682 * come from the zone allocator set up in tcp_init().
685 tcp_newtcpcb(struct inpcb *inp)
690 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
692 const boolean_t isipv6 = FALSE;
695 it = (struct inp_tp *)inp;
697 bzero(tp, sizeof(struct tcpcb));
698 TAILQ_INIT(&tp->t_segq);
699 tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
700 tp->t_rxtthresh = tcprexmtthresh;
702 /* Set up our timeouts. */
703 tp->tt_rexmt = &it->inp_tp_rexmt;
704 tp->tt_persist = &it->inp_tp_persist;
705 tp->tt_keep = &it->inp_tp_keep;
706 tp->tt_2msl = &it->inp_tp_2msl;
707 tp->tt_delack = &it->inp_tp_delack;
711 * Zero out timer message. We don't create it here,
712 * since the current CPU may not be the owner of this
715 tp->tt_msg = &it->inp_tp_timermsg;
716 bzero(tp->tt_msg, sizeof(*tp->tt_msg));
718 tp->t_keepinit = tcp_keepinit;
719 tp->t_keepidle = tcp_keepidle;
720 tp->t_keepintvl = tcp_keepintvl;
721 tp->t_keepcnt = tcp_keepcnt;
722 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
725 tp->t_flags |= TF_NCR;
727 tp->t_flags |= (TF_REQ_SCALE | TF_REQ_TSTMP);
729 tp->t_inpcb = inp; /* XXX */
730 tp->t_state = TCPS_CLOSED;
732 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
733 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
734 * reasonable initial retransmit time.
736 tp->t_srtt = TCPTV_SRTTBASE;
738 ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
739 tp->t_rttmin = tcp_rexmit_min;
740 tp->t_rxtcur = TCPTV_RTOBASE;
741 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
742 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
743 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
744 tp->snd_last = ticks;
745 tp->t_rcvtime = ticks;
747 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
748 * because the socket may be bound to an IPv6 wildcard address,
749 * which may match an IPv4-mapped IPv6 address.
751 inp->inp_ip_ttl = ip_defttl;
753 tcp_sack_tcpcb_init(tp);
755 tp->tt_sndmore = &it->inp_tp_sndmore;
758 return (tp); /* XXX */
762 * Drop a TCP connection, reporting the specified error.
763 * If connection is synchronized, then send a RST to peer.
766 tcp_drop(struct tcpcb *tp, int error)
768 struct socket *so = tp->t_inpcb->inp_socket;
770 if (TCPS_HAVERCVDSYN(tp->t_state)) {
771 tp->t_state = TCPS_CLOSED;
773 tcpstat.tcps_drops++;
775 tcpstat.tcps_conndrops++;
776 if (error == ETIMEDOUT && tp->t_softerror)
777 error = tp->t_softerror;
778 so->so_error = error;
779 return (tcp_close(tp));
782 struct netmsg_listen_detach {
783 struct netmsg_base base;
788 tcp_listen_detach_handler(netmsg_t msg)
790 struct netmsg_listen_detach *nmsg = (struct netmsg_listen_detach *)msg;
791 struct tcpcb *tp = nmsg->nm_tp;
792 int cpu = mycpuid, nextcpu;
794 if (tp->t_flags & TF_LISTEN)
795 syncache_destroy(tp);
797 in_pcbremwildcardhash_oncpu(tp->t_inpcb, &tcbinfo[cpu]);
800 if (nextcpu < ncpus2)
801 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nmsg->base.lmsg);
803 lwkt_replymsg(&nmsg->base.lmsg, 0);
807 * Close a TCP control block:
808 * discard all space held by the tcp
809 * discard internet protocol block
810 * wake up any sleepers
813 tcp_close(struct tcpcb *tp)
816 struct inpcb *inp = tp->t_inpcb;
817 struct socket *so = inp->inp_socket;
819 boolean_t dosavessthresh;
821 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
822 boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0);
824 const boolean_t isipv6 = FALSE;
828 * INP_WILDCARD_MP indicates that listen(2) has been called on
829 * this socket. This implies:
830 * - A wildcard inp's hash is replicated for each protocol thread.
831 * - Syncache for this inp grows independently in each protocol
833 * - There is more than one cpu
835 * We have to chain a message to the rest of the protocol threads
836 * to cleanup the wildcard hash and the syncache. The cleanup
837 * in the current protocol thread is defered till the end of this
841 * After cleanup the inp's hash and syncache entries, this inp will
842 * no longer be available to the rest of the protocol threads, so we
843 * are safe to whack the inp in the following code.
845 if (inp->inp_flags & INP_WILDCARD_MP) {
846 struct netmsg_listen_detach nmsg;
848 KKASSERT(so->so_port == netisr_cpuport(0));
849 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
850 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
852 netmsg_init(&nmsg.base, NULL, &curthread->td_msgport,
853 MSGF_PRIORITY, tcp_listen_detach_handler);
855 lwkt_domsg(netisr_cpuport(1), &nmsg.base.lmsg, 0);
857 inp->inp_flags &= ~INP_WILDCARD_MP;
860 KKASSERT(tp->t_state != TCPS_TERMINATING);
861 tp->t_state = TCPS_TERMINATING;
864 * Make sure that all of our timers are stopped before we
865 * delete the PCB. For listen TCP socket (tp->tt_msg == NULL),
866 * timers are never used. If timer message is never created
867 * (tp->tt_msg->tt_tcb == NULL), timers are never used too.
869 if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) {
870 tcp_callout_stop(tp, tp->tt_rexmt);
871 tcp_callout_stop(tp, tp->tt_persist);
872 tcp_callout_stop(tp, tp->tt_keep);
873 tcp_callout_stop(tp, tp->tt_2msl);
874 tcp_callout_stop(tp, tp->tt_delack);
877 if (tp->t_flags & TF_ONOUTPUTQ) {
878 KKASSERT(tp->tt_cpu == mycpu->gd_cpuid);
879 TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq);
880 tp->t_flags &= ~TF_ONOUTPUTQ;
884 * If we got enough samples through the srtt filter,
885 * save the rtt and rttvar in the routing entry.
886 * 'Enough' is arbitrarily defined as the 16 samples.
887 * 16 samples is enough for the srtt filter to converge
888 * to within 5% of the correct value; fewer samples and
889 * we could save a very bogus rtt.
891 * Don't update the default route's characteristics and don't
892 * update anything that the user "locked".
894 if (tp->t_rttupdated >= 16) {
898 struct sockaddr_in6 *sin6;
900 if ((rt = inp->in6p_route.ro_rt) == NULL)
902 sin6 = (struct sockaddr_in6 *)rt_key(rt);
903 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
906 if ((rt = inp->inp_route.ro_rt) == NULL ||
907 ((struct sockaddr_in *)rt_key(rt))->
908 sin_addr.s_addr == INADDR_ANY)
911 if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) {
912 i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
913 if (rt->rt_rmx.rmx_rtt && i)
915 * filter this update to half the old & half
916 * the new values, converting scale.
917 * See route.h and tcp_var.h for a
918 * description of the scaling constants.
921 (rt->rt_rmx.rmx_rtt + i) / 2;
923 rt->rt_rmx.rmx_rtt = i;
924 tcpstat.tcps_cachedrtt++;
926 if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) {
928 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
929 if (rt->rt_rmx.rmx_rttvar && i)
930 rt->rt_rmx.rmx_rttvar =
931 (rt->rt_rmx.rmx_rttvar + i) / 2;
933 rt->rt_rmx.rmx_rttvar = i;
934 tcpstat.tcps_cachedrttvar++;
937 * The old comment here said:
938 * update the pipelimit (ssthresh) if it has been updated
939 * already or if a pipesize was specified & the threshhold
940 * got below half the pipesize. I.e., wait for bad news
941 * before we start updating, then update on both good
944 * But we want to save the ssthresh even if no pipesize is
945 * specified explicitly in the route, because such
946 * connections still have an implicit pipesize specified
947 * by the global tcp_sendspace. In the absence of a reliable
948 * way to calculate the pipesize, it will have to do.
950 i = tp->snd_ssthresh;
951 if (rt->rt_rmx.rmx_sendpipe != 0)
952 dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2);
954 dosavessthresh = (i < so->so_snd.ssb_hiwat/2);
955 if (dosavessthresh ||
956 (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) &&
957 (rt->rt_rmx.rmx_ssthresh != 0))) {
959 * convert the limit from user data bytes to
960 * packets then to packet data bytes.
962 i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
967 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
968 sizeof(struct tcpiphdr));
969 if (rt->rt_rmx.rmx_ssthresh)
970 rt->rt_rmx.rmx_ssthresh =
971 (rt->rt_rmx.rmx_ssthresh + i) / 2;
973 rt->rt_rmx.rmx_ssthresh = i;
974 tcpstat.tcps_cachedssthresh++;
979 /* free the reassembly queue, if any */
980 while((q = TAILQ_FIRST(&tp->t_segq)) != NULL) {
981 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
984 atomic_add_int(&tcp_reass_qsize, -1);
986 /* throw away SACK blocks in scoreboard*/
988 tcp_sack_destroy(&tp->scb);
990 inp->inp_ppcb = NULL;
991 soisdisconnected(so);
992 /* note: pcb detached later on */
994 tcp_destroy_timermsg(tp);
995 tcp_output_cancel(tp);
997 if (tp->t_flags & TF_LISTEN)
998 syncache_destroy(tp);
1000 so_async_rcvd_drop(so);
1001 /* Drop the reference for the asynchronized pru_rcvd */
1006 * pcbdetach removes any wildcard hash entry on the current CPU.
1015 tcpstat.tcps_closed++;
1019 static __inline void
1020 tcp_drain_oncpu(struct inpcbhead *head)
1022 struct inpcb *marker;
1025 struct tseg_qent *te;
1028 * Allows us to block while running the list
1030 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
1031 marker->inp_flags |= INP_PLACEMARKER;
1032 LIST_INSERT_HEAD(head, marker, inp_list);
1034 while ((inpb = LIST_NEXT(marker, inp_list)) != NULL) {
1035 if ((inpb->inp_flags & INP_PLACEMARKER) == 0 &&
1036 (tcpb = intotcpcb(inpb)) != NULL &&
1037 (te = TAILQ_FIRST(&tcpb->t_segq)) != NULL) {
1038 TAILQ_REMOVE(&tcpb->t_segq, te, tqe_q);
1039 if (te->tqe_th->th_flags & TH_FIN)
1040 tcpb->t_flags &= ~TF_QUEDFIN;
1043 atomic_add_int(&tcp_reass_qsize, -1);
1046 LIST_REMOVE(marker, inp_list);
1047 LIST_INSERT_AFTER(inpb, marker, inp_list);
1050 LIST_REMOVE(marker, inp_list);
1051 kfree(marker, M_TEMP);
1054 struct netmsg_tcp_drain {
1055 struct netmsg_base base;
1056 struct inpcbhead *nm_head;
1060 tcp_drain_handler(netmsg_t msg)
1062 struct netmsg_tcp_drain *nm = (void *)msg;
1064 tcp_drain_oncpu(nm->nm_head);
1065 lwkt_replymsg(&nm->base.lmsg, 0);
1077 * Walk the tcpbs, if existing, and flush the reassembly queue,
1078 * if there is one...
1079 * XXX: The "Net/3" implementation doesn't imply that the TCP
1080 * reassembly queue should be flushed, but in a situation
1081 * where we're really low on mbufs, this is potentially
1084 for (cpu = 0; cpu < ncpus2; cpu++) {
1085 struct netmsg_tcp_drain *nm;
1087 if (cpu == mycpu->gd_cpuid) {
1088 tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead);
1090 nm = kmalloc(sizeof(struct netmsg_tcp_drain),
1091 M_LWKTMSG, M_NOWAIT);
1094 netmsg_init(&nm->base, NULL, &netisr_afree_rport,
1095 0, tcp_drain_handler);
1096 nm->nm_head = &tcbinfo[cpu].pcblisthead;
1097 lwkt_sendmsg(netisr_cpuport(cpu), &nm->base.lmsg);
1103 * Notify a tcp user of an asynchronous error;
1104 * store error as soft error, but wake up user
1105 * (for now, won't do anything until can select for soft error).
1107 * Do not wake up user since there currently is no mechanism for
1108 * reporting soft errors (yet - a kqueue filter may be added).
1111 tcp_notify(struct inpcb *inp, int error)
1113 struct tcpcb *tp = intotcpcb(inp);
1116 * Ignore some errors if we are hooked up.
1117 * If connection hasn't completed, has retransmitted several times,
1118 * and receives a second error, give up now. This is better
1119 * than waiting a long time to establish a connection that
1120 * can never complete.
1122 if (tp->t_state == TCPS_ESTABLISHED &&
1123 (error == EHOSTUNREACH || error == ENETUNREACH ||
1124 error == EHOSTDOWN)) {
1126 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1128 tcp_drop(tp, error);
1130 tp->t_softerror = error;
1132 wakeup(&so->so_timeo);
1139 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1142 struct inpcb *marker;
1151 * The process of preparing the TCB list is too time-consuming and
1152 * resource-intensive to repeat twice on every request.
1154 if (req->oldptr == NULL) {
1155 for (ccpu = 0; ccpu < ncpus; ++ccpu) {
1156 gd = globaldata_find(ccpu);
1157 n += tcbinfo[gd->gd_cpuid].ipi_count;
1159 req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb);
1163 if (req->newptr != NULL)
1166 marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
1167 marker->inp_flags |= INP_PLACEMARKER;
1170 * OK, now we're committed to doing something. Run the inpcb list
1171 * for each cpu in the system and construct the output. Use a
1172 * list placemarker to deal with list changes occuring during
1173 * copyout blockages (but otherwise depend on being on the correct
1174 * cpu to avoid races).
1176 origcpu = mycpu->gd_cpuid;
1177 for (ccpu = 1; ccpu <= ncpus && error == 0; ++ccpu) {
1183 cpu_id = (origcpu + ccpu) % ncpus;
1184 if ((smp_active_mask & CPUMASK(cpu_id)) == 0)
1186 rgd = globaldata_find(cpu_id);
1187 lwkt_setcpu_self(rgd);
1189 n = tcbinfo[cpu_id].ipi_count;
1191 LIST_INSERT_HEAD(&tcbinfo[cpu_id].pcblisthead, marker, inp_list);
1193 while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
1195 * process a snapshot of pcbs, ignoring placemarkers
1196 * and using our own to allow SYSCTL_OUT to block.
1198 LIST_REMOVE(marker, inp_list);
1199 LIST_INSERT_AFTER(inp, marker, inp_list);
1201 if (inp->inp_flags & INP_PLACEMARKER)
1203 if (prison_xinpcb(req->td, inp))
1206 xt.xt_len = sizeof xt;
1207 bcopy(inp, &xt.xt_inp, sizeof *inp);
1208 inp_ppcb = inp->inp_ppcb;
1209 if (inp_ppcb != NULL)
1210 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1212 bzero(&xt.xt_tp, sizeof xt.xt_tp);
1213 if (inp->inp_socket)
1214 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1215 if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0)
1219 LIST_REMOVE(marker, inp_list);
1220 if (error == 0 && i < n) {
1221 bzero(&xt, sizeof xt);
1222 xt.xt_len = sizeof xt;
1224 error = SYSCTL_OUT(req, &xt, sizeof xt);
1233 * Make sure we are on the same cpu we were on originally, since
1234 * higher level callers expect this. Also don't pollute caches with
1235 * migrated userland data by (eventually) returning to userland
1236 * on a different cpu.
1238 lwkt_setcpu_self(globaldata_find(origcpu));
1239 kfree(marker, M_TEMP);
1243 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1244 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1247 tcp_getcred(SYSCTL_HANDLER_ARGS)
1249 struct sockaddr_in addrs[2];
1254 error = priv_check(req->td, PRIV_ROOT);
1257 error = SYSCTL_IN(req, addrs, sizeof addrs);
1261 cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
1262 addrs[0].sin_addr.s_addr, addrs[0].sin_port);
1263 inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr,
1264 addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1265 if (inp == NULL || inp->inp_socket == NULL) {
1269 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
1275 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
1276 0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");
1280 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1282 struct sockaddr_in6 addrs[2];
1285 boolean_t mapped = FALSE;
1287 error = priv_check(req->td, PRIV_ROOT);
1290 error = SYSCTL_IN(req, addrs, sizeof addrs);
1293 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1294 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1301 inp = in_pcblookup_hash(&tcbinfo[0],
1302 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1304 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1308 inp = in6_pcblookup_hash(&tcbinfo[0],
1309 &addrs[1].sin6_addr, addrs[1].sin6_port,
1310 &addrs[0].sin6_addr, addrs[0].sin6_port,
1313 if (inp == NULL || inp->inp_socket == NULL) {
1317 error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
1323 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
1325 tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
1328 struct netmsg_tcp_notify {
1329 struct netmsg_base base;
1330 void (*nm_notify)(struct inpcb *, int);
1331 struct in_addr nm_faddr;
1336 tcp_notifyall_oncpu(netmsg_t msg)
1338 struct netmsg_tcp_notify *nm = (struct netmsg_tcp_notify *)msg;
1341 in_pcbnotifyall(&tcbinfo[mycpuid].pcblisthead, nm->nm_faddr,
1342 nm->nm_arg, nm->nm_notify);
1344 nextcpu = mycpuid + 1;
1345 if (nextcpu < ncpus2)
1346 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
1348 lwkt_replymsg(&nm->base.lmsg, 0);
1352 tcp_ctlinput(netmsg_t msg)
1354 int cmd = msg->ctlinput.nm_cmd;
1355 struct sockaddr *sa = msg->ctlinput.nm_arg;
1356 struct ip *ip = msg->ctlinput.nm_extra;
1358 struct in_addr faddr;
1361 void (*notify)(struct inpcb *, int) = tcp_notify;
1365 if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
1369 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1370 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1373 arg = inetctlerrmap[cmd];
1374 if (cmd == PRC_QUENCH) {
1375 notify = tcp_quench;
1376 } else if (icmp_may_rst &&
1377 (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1378 cmd == PRC_UNREACH_PORT ||
1379 cmd == PRC_TIMXCEED_INTRANS) &&
1381 notify = tcp_drop_syn_sent;
1382 } else if (cmd == PRC_MSGSIZE) {
1383 struct icmp *icmp = (struct icmp *)
1384 ((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1386 arg = ntohs(icmp->icmp_nextmtu);
1387 notify = tcp_mtudisc;
1388 } else if (PRC_IS_REDIRECT(cmd)) {
1390 notify = in_rtchange;
1391 } else if (cmd == PRC_HOSTDEAD) {
1397 th = (struct tcphdr *)((caddr_t)ip +
1398 (IP_VHL_HL(ip->ip_vhl) << 2));
1399 cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
1400 ip->ip_src.s_addr, th->th_sport);
1401 inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport,
1402 ip->ip_src, th->th_sport, 0, NULL);
1403 if ((inp != NULL) && (inp->inp_socket != NULL)) {
1404 icmpseq = htonl(th->th_seq);
1405 tp = intotcpcb(inp);
1406 if (SEQ_GEQ(icmpseq, tp->snd_una) &&
1407 SEQ_LT(icmpseq, tp->snd_max))
1408 (*notify)(inp, arg);
1410 struct in_conninfo inc;
1412 inc.inc_fport = th->th_dport;
1413 inc.inc_lport = th->th_sport;
1414 inc.inc_faddr = faddr;
1415 inc.inc_laddr = ip->ip_src;
1419 syncache_unreach(&inc, th);
1423 struct netmsg_tcp_notify *nm;
1425 KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
1426 nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
1427 netmsg_init(&nm->base, NULL, &netisr_afree_rport,
1428 0, tcp_notifyall_oncpu);
1429 nm->nm_faddr = faddr;
1431 nm->nm_notify = notify;
1433 lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
1436 lwkt_replymsg(&msg->lmsg, 0);
1442 tcp6_ctlinput(netmsg_t msg)
1444 int cmd = msg->ctlinput.nm_cmd;
1445 struct sockaddr *sa = msg->ctlinput.nm_arg;
1446 void *d = msg->ctlinput.nm_extra;
1448 void (*notify) (struct inpcb *, int) = tcp_notify;
1449 struct ip6_hdr *ip6;
1451 struct ip6ctlparam *ip6cp = NULL;
1452 const struct sockaddr_in6 *sa6_src = NULL;
1454 struct tcp_portonly {
1460 if (sa->sa_family != AF_INET6 ||
1461 sa->sa_len != sizeof(struct sockaddr_in6)) {
1466 if (cmd == PRC_QUENCH)
1467 notify = tcp_quench;
1468 else if (cmd == PRC_MSGSIZE) {
1469 struct ip6ctlparam *ip6cp = d;
1470 struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6;
1472 arg = ntohl(icmp6->icmp6_mtu);
1473 notify = tcp_mtudisc;
1474 } else if (!PRC_IS_REDIRECT(cmd) &&
1475 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) {
1479 /* if the parameter is from icmp6, decode it. */
1481 ip6cp = (struct ip6ctlparam *)d;
1483 ip6 = ip6cp->ip6c_ip6;
1484 off = ip6cp->ip6c_off;
1485 sa6_src = ip6cp->ip6c_src;
1489 off = 0; /* fool gcc */
1494 struct in_conninfo inc;
1496 * XXX: We assume that when IPV6 is non NULL,
1497 * M and OFF are valid.
1500 /* check if we can safely examine src and dst ports */
1501 if (m->m_pkthdr.len < off + sizeof *thp)
1504 bzero(&th, sizeof th);
1505 m_copydata(m, off, sizeof *thp, (caddr_t)&th);
1507 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, th.th_dport,
1508 (struct sockaddr *)ip6cp->ip6c_src,
1509 th.th_sport, cmd, arg, notify);
1511 inc.inc_fport = th.th_dport;
1512 inc.inc_lport = th.th_sport;
1513 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1514 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1516 syncache_unreach(&inc, &th);
1518 in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, 0,
1519 (const struct sockaddr *)sa6_src, 0, cmd, arg, notify);
1522 lwkt_replymsg(&msg->ctlinput.base.lmsg, 0);
1528 * Following is where TCP initial sequence number generation occurs.
1530 * There are two places where we must use initial sequence numbers:
1531 * 1. In SYN-ACK packets.
1532 * 2. In SYN packets.
1534 * All ISNs for SYN-ACK packets are generated by the syncache. See
1535 * tcp_syncache.c for details.
1537 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1538 * depends on this property. In addition, these ISNs should be
1539 * unguessable so as to prevent connection hijacking. To satisfy
1540 * the requirements of this situation, the algorithm outlined in
1541 * RFC 1948 is used to generate sequence numbers.
1543 * Implementation details:
1545 * Time is based off the system timer, and is corrected so that it
1546 * increases by one megabyte per second. This allows for proper
1547 * recycling on high speed LANs while still leaving over an hour
1550 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1551 * between seeding of isn_secret. This is normally set to zero,
1552 * as reseeding should not be necessary.
1556 #define ISN_BYTES_PER_SECOND 1048576
1558 u_char isn_secret[32];
1559 int isn_last_reseed;
1563 tcp_new_isn(struct tcpcb *tp)
1565 u_int32_t md5_buffer[4];
1568 /* Seed if this is the first use, reseed if requested. */
1569 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1570 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1572 read_random_unlimited(&isn_secret, sizeof isn_secret);
1573 isn_last_reseed = ticks;
1576 /* Compute the md5 hash and return the ISN. */
1578 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short));
1579 MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short));
1581 if (tp->t_inpcb->inp_vflag & INP_IPV6) {
1582 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1583 sizeof(struct in6_addr));
1584 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1585 sizeof(struct in6_addr));
1589 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1590 sizeof(struct in_addr));
1591 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1592 sizeof(struct in_addr));
1594 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1595 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1596 new_isn = (tcp_seq) md5_buffer[0];
1597 new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
1602 * When a source quench is received, close congestion window
1603 * to one segment. We will gradually open it again as we proceed.
1606 tcp_quench(struct inpcb *inp, int error)
1608 struct tcpcb *tp = intotcpcb(inp);
1611 tp->snd_cwnd = tp->t_maxseg;
1617 * When a specific ICMP unreachable message is received and the
1618 * connection state is SYN-SENT, drop the connection. This behavior
1619 * is controlled by the icmp_may_rst sysctl.
1622 tcp_drop_syn_sent(struct inpcb *inp, int error)
1624 struct tcpcb *tp = intotcpcb(inp);
1626 if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT))
1627 tcp_drop(tp, error);
1631 * When a `need fragmentation' ICMP is received, update our idea of the MSS
1632 * based on the new value in the route. Also nudge TCP to send something,
1633 * since we know the packet we just sent was dropped.
1634 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1637 tcp_mtudisc(struct inpcb *inp, int mtu)
1639 struct tcpcb *tp = intotcpcb(inp);
1641 struct socket *so = inp->inp_socket;
1644 boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0);
1646 const boolean_t isipv6 = FALSE;
1653 * If no MTU is provided in the ICMP message, use the
1654 * next lower likely value, as specified in RFC 1191.
1659 oldmtu = tp->t_maxopd +
1661 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1662 sizeof(struct tcpiphdr));
1663 mtu = ip_next_mtu(oldmtu, 0);
1667 rt = tcp_rtlookup6(&inp->inp_inc);
1669 rt = tcp_rtlookup(&inp->inp_inc);
1671 if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu)
1672 mtu = rt->rt_rmx.rmx_mtu;
1676 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1677 sizeof(struct tcpiphdr));
1680 * XXX - The following conditional probably violates the TCP
1681 * spec. The problem is that, since we don't know the
1682 * other end's MSS, we are supposed to use a conservative
1683 * default. But, if we do that, then MTU discovery will
1684 * never actually take place, because the conservative
1685 * default is much less than the MTUs typically seen
1686 * on the Internet today. For the moment, we'll sweep
1687 * this under the carpet.
1689 * The conservative default might not actually be a problem
1690 * if the only case this occurs is when sending an initial
1691 * SYN with options and data to a host we've never talked
1692 * to before. Then, they will reply with an MSS value which
1693 * will get recorded and the new parameters should get
1694 * recomputed. For Further Study.
1696 if (rt->rt_rmx.rmx_mssopt && rt->rt_rmx.rmx_mssopt < maxopd)
1697 maxopd = rt->rt_rmx.rmx_mssopt;
1701 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1702 sizeof(struct tcpiphdr));
1704 if (tp->t_maxopd <= maxopd)
1706 tp->t_maxopd = maxopd;
1709 if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) ==
1710 (TF_REQ_TSTMP | TF_RCVD_TSTMP))
1711 mss -= TCPOLEN_TSTAMP_APPA;
1713 /* round down to multiple of MCLBYTES */
1714 #if (MCLBYTES & (MCLBYTES - 1)) == 0 /* test if MCLBYTES power of 2 */
1716 mss &= ~(MCLBYTES - 1);
1719 mss = (mss / MCLBYTES) * MCLBYTES;
1722 if (so->so_snd.ssb_hiwat < mss)
1723 mss = so->so_snd.ssb_hiwat;
1727 tp->snd_nxt = tp->snd_una;
1729 tcpstat.tcps_mturesent++;
1733 * Look-up the routing entry to the peer of this inpcb. If no route
1734 * is found and it cannot be allocated the return NULL. This routine
1735 * is called by TCP routines that access the rmx structure and by tcp_mss
1736 * to get the interface MTU.
1739 tcp_rtlookup(struct in_conninfo *inc)
1741 struct route *ro = &inc->inc_route;
1743 if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) {
1744 /* No route yet, so try to acquire one */
1745 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1747 * unused portions of the structure MUST be zero'd
1748 * out because rtalloc() treats it as opaque data
1750 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1751 ro->ro_dst.sa_family = AF_INET;
1752 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1753 ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
1763 tcp_rtlookup6(struct in_conninfo *inc)
1765 struct route_in6 *ro6 = &inc->inc6_route;
1767 if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) {
1768 /* No route yet, so try to acquire one */
1769 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1771 * unused portions of the structure MUST be zero'd
1772 * out because rtalloc() treats it as opaque data
1774 bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6));
1775 ro6->ro_dst.sin6_family = AF_INET6;
1776 ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1777 ro6->ro_dst.sin6_addr = inc->inc6_faddr;
1778 rtalloc((struct route *)ro6);
1781 return (ro6->ro_rt);
1786 /* compute ESP/AH header size for TCP, including outer IP header. */
1788 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1796 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1798 MGETHDR(m, MB_DONTWAIT, MT_DATA);
1803 if (inp->inp_vflag & INP_IPV6) {
1804 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1806 th = (struct tcphdr *)(ip6 + 1);
1807 m->m_pkthdr.len = m->m_len =
1808 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1809 tcp_fillheaders(tp, ip6, th, FALSE);
1810 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1814 ip = mtod(m, struct ip *);
1815 th = (struct tcphdr *)(ip + 1);
1816 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1817 tcp_fillheaders(tp, ip, th, FALSE);
1818 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1827 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1829 * This code attempts to calculate the bandwidth-delay product as a
1830 * means of determining the optimal window size to maximize bandwidth,
1831 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1832 * routers. This code also does a fairly good job keeping RTTs in check
1833 * across slow links like modems. We implement an algorithm which is very
1834 * similar (but not meant to be) TCP/Vegas. The code operates on the
1835 * transmitter side of a TCP connection and so only effects the transmit
1836 * side of the connection.
1838 * BACKGROUND: TCP makes no provision for the management of buffer space
1839 * at the end points or at the intermediate routers and switches. A TCP
1840 * stream, whether using NewReno or not, will eventually buffer as
1841 * many packets as it is able and the only reason this typically works is
1842 * due to the fairly small default buffers made available for a connection
1843 * (typicaly 16K or 32K). As machines use larger windows and/or window
1844 * scaling it is now fairly easy for even a single TCP connection to blow-out
1845 * all available buffer space not only on the local interface, but on
1846 * intermediate routers and switches as well. NewReno makes a misguided
1847 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1848 * then backing off, then steadily increasing the window again until another
1849 * failure occurs, ad-infinitum. This results in terrible oscillation that
1850 * is only made worse as network loads increase and the idea of intentionally
1851 * blowing out network buffers is, frankly, a terrible way to manage network
1854 * It is far better to limit the transmit window prior to the failure
1855 * condition being achieved. There are two general ways to do this: First
1856 * you can 'scan' through different transmit window sizes and locate the
1857 * point where the RTT stops increasing, indicating that you have filled the
1858 * pipe, then scan backwards until you note that RTT stops decreasing, then
1859 * repeat ad-infinitum. This method works in principle but has severe
1860 * implementation issues due to RTT variances, timer granularity, and
1861 * instability in the algorithm which can lead to many false positives and
1862 * create oscillations as well as interact badly with other TCP streams
1863 * implementing the same algorithm.
1865 * The second method is to limit the window to the bandwidth delay product
1866 * of the link. This is the method we implement. RTT variances and our
1867 * own manipulation of the congestion window, bwnd, can potentially
1868 * destabilize the algorithm. For this reason we have to stabilize the
1869 * elements used to calculate the window. We do this by using the minimum
1870 * observed RTT, the long term average of the observed bandwidth, and
1871 * by adding two segments worth of slop. It isn't perfect but it is able
1872 * to react to changing conditions and gives us a very stable basis on
1873 * which to extend the algorithm.
1876 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1884 * If inflight_enable is disabled in the middle of a tcp connection,
1885 * make sure snd_bwnd is effectively disabled.
1887 if (!tcp_inflight_enable) {
1888 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1889 tp->snd_bandwidth = 0;
1894 * Validate the delta time. If a connection is new or has been idle
1895 * a long time we have to reset the bandwidth calculator.
1898 delta_ticks = save_ticks - tp->t_bw_rtttime;
1899 if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) {
1900 tp->t_bw_rtttime = ticks;
1901 tp->t_bw_rtseq = ack_seq;
1902 if (tp->snd_bandwidth == 0)
1903 tp->snd_bandwidth = tcp_inflight_min;
1906 if (delta_ticks == 0)
1910 * Sanity check, plus ignore pure window update acks.
1912 if ((int)(ack_seq - tp->t_bw_rtseq) <= 0)
1916 * Figure out the bandwidth. Due to the tick granularity this
1917 * is a very rough number and it MUST be averaged over a fairly
1918 * long period of time. XXX we need to take into account a link
1919 * that is not using all available bandwidth, but for now our
1920 * slop will ramp us up if this case occurs and the bandwidth later
1923 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks;
1924 tp->t_bw_rtttime = save_ticks;
1925 tp->t_bw_rtseq = ack_seq;
1926 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1928 tp->snd_bandwidth = bw;
1931 * Calculate the semi-static bandwidth delay product, plus two maximal
1932 * segments. The additional slop puts us squarely in the sweet
1933 * spot and also handles the bandwidth run-up case. Without the
1934 * slop we could be locking ourselves into a lower bandwidth.
1936 * Situations Handled:
1937 * (1) Prevents over-queueing of packets on LANs, especially on
1938 * high speed LANs, allowing larger TCP buffers to be
1939 * specified, and also does a good job preventing
1940 * over-queueing of packets over choke points like modems
1941 * (at least for the transmit side).
1943 * (2) Is able to handle changing network loads (bandwidth
1944 * drops so bwnd drops, bandwidth increases so bwnd
1947 * (3) Theoretically should stabilize in the face of multiple
1948 * connections implementing the same algorithm (this may need
1951 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1952 * be adjusted with a sysctl but typically only needs to be on
1953 * very slow connections. A value no smaller then 5 should
1954 * be used, but only reduce this default if you have no other
1958 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1959 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) +
1960 tcp_inflight_stab * (int)tp->t_maxseg / 10;
1963 if (tcp_inflight_debug > 0) {
1965 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1967 kprintf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1968 tp, bw, tp->t_rttbest, tp->t_srtt, bwnd);
1971 if ((long)bwnd < tcp_inflight_min)
1972 bwnd = tcp_inflight_min;
1973 if (bwnd > tcp_inflight_max)
1974 bwnd = tcp_inflight_max;
1975 if ((long)bwnd < tp->t_maxseg * 2)
1976 bwnd = tp->t_maxseg * 2;
1977 tp->snd_bwnd = bwnd;
1981 tcp_rmx_iwsegs(struct tcpcb *tp, u_long *maxsegs, u_long *capsegs)
1984 struct inpcb *inp = tp->t_inpcb;
1986 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
1988 const boolean_t isipv6 = FALSE;
1992 if (tcp_iw_maxsegs < TCP_IW_MAXSEGS_DFLT)
1993 tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT;
1994 if (tcp_iw_capsegs < TCP_IW_CAPSEGS_DFLT)
1995 tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT;
1998 rt = tcp_rtlookup6(&inp->inp_inc);
2000 rt = tcp_rtlookup(&inp->inp_inc);
2002 rt->rt_rmx.rmx_iwmaxsegs < TCP_IW_MAXSEGS_DFLT ||
2003 rt->rt_rmx.rmx_iwcapsegs < TCP_IW_CAPSEGS_DFLT) {
2004 *maxsegs = tcp_iw_maxsegs;
2005 *capsegs = tcp_iw_capsegs;
2008 *maxsegs = rt->rt_rmx.rmx_iwmaxsegs;
2009 *capsegs = rt->rt_rmx.rmx_iwcapsegs;
2013 tcp_initial_window(struct tcpcb *tp)
2015 if (tcp_do_rfc3390) {
2018 * "If the SYN or SYN/ACK is lost, the initial window
2019 * used by a sender after a correctly transmitted SYN
2020 * MUST be one segment consisting of MSS bytes."
2022 * However, we do something a little bit more aggressive
2023 * then RFC3390 here:
2024 * - Only if time spent in the SYN or SYN|ACK retransmition
2025 * >= 3 seconds, the IW is reduced. We do this mainly
2026 * because when RFC3390 is published, the initial RTO is
2027 * still 3 seconds (the threshold we test here), while
2028 * after RFC6298, the initial RTO is 1 second. This
2029 * behaviour probably still falls within the spirit of
2031 * - When IW is reduced, 2*MSS is used instead of 1*MSS.
2032 * Mainly to avoid sender and receiver deadlock until
2033 * delayed ACK timer expires. And even RFC2581 does not
2034 * try to reduce IW upon SYN or SYN|ACK retransmition
2038 * http://tools.ietf.org/html/draft-ietf-tcpm-initcwnd-03
2040 if (tp->t_rxtsyn >= TCPTV_RTOBASE3) {
2041 return (2 * tp->t_maxseg);
2043 u_long maxsegs, capsegs;
2045 tcp_rmx_iwsegs(tp, &maxsegs, &capsegs);
2046 return min(maxsegs * tp->t_maxseg,
2047 max(2 * tp->t_maxseg, capsegs * 1460));
2051 * Even RFC2581 (back to 1999) allows 2*SMSS IW.
2053 * Mainly to avoid sender and receiver deadlock
2054 * until delayed ACK timer expires.
2056 return (2 * tp->t_maxseg);
2060 #ifdef TCP_SIGNATURE
2062 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2064 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2065 * When called from tcp_input(), we can be sure that th_sum has been
2066 * zeroed out and verified already.
2068 * Return 0 if successful, otherwise return -1.
2070 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2071 * search with the destination IP address, and a 'magic SPI' to be
2072 * determined by the application. This is hardcoded elsewhere to 1179
2073 * right now. Another branch of this code exists which uses the SPD to
2074 * specify per-application flows but it is unstable.
2077 tcpsignature_compute(
2078 struct mbuf *m, /* mbuf chain */
2079 int len, /* length of TCP data */
2080 int optlen, /* length of TCP options */
2081 u_char *buf, /* storage for MD5 digest */
2082 u_int direction) /* direction of flow */
2084 struct ippseudo ippseudo;
2088 struct ipovly *ipovly;
2089 struct secasvar *sav;
2092 struct ip6_hdr *ip6;
2093 struct in6_addr in6;
2099 KASSERT(m != NULL, ("passed NULL mbuf. Game over."));
2100 KASSERT(buf != NULL, ("passed NULL storage pointer for MD5 signature"));
2102 * Extract the destination from the IP header in the mbuf.
2104 ip = mtod(m, struct ip *);
2106 ip6 = NULL; /* Make the compiler happy. */
2109 * Look up an SADB entry which matches the address found in
2112 switch (IP_VHL_V(ip->ip_vhl)) {
2114 sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst,
2115 IPPROTO_TCP, htonl(TCP_SIG_SPI));
2118 case (IPV6_VERSION >> 4):
2119 ip6 = mtod(m, struct ip6_hdr *);
2120 sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst,
2121 IPPROTO_TCP, htonl(TCP_SIG_SPI));
2130 kprintf("%s: SADB lookup failed\n", __func__);
2136 * Step 1: Update MD5 hash with IP pseudo-header.
2138 * XXX The ippseudo header MUST be digested in network byte order,
2139 * or else we'll fail the regression test. Assume all fields we've
2140 * been doing arithmetic on have been in host byte order.
2141 * XXX One cannot depend on ipovly->ih_len here. When called from
2142 * tcp_output(), the underlying ip_len member has not yet been set.
2144 switch (IP_VHL_V(ip->ip_vhl)) {
2146 ipovly = (struct ipovly *)ip;
2147 ippseudo.ippseudo_src = ipovly->ih_src;
2148 ippseudo.ippseudo_dst = ipovly->ih_dst;
2149 ippseudo.ippseudo_pad = 0;
2150 ippseudo.ippseudo_p = IPPROTO_TCP;
2151 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2152 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2153 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2154 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2158 * RFC 2385, 2.0 Proposal
2159 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2160 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2161 * extended next header value (to form 32 bits), and 32-bit segment
2163 * Note: Upper-Layer Packet Length comes before Next Header.
2165 case (IPV6_VERSION >> 4):
2167 in6_clearscope(&in6);
2168 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2170 in6_clearscope(&in6);
2171 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2172 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2173 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2175 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2176 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2177 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2179 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2180 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2181 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2190 * Step 2: Update MD5 hash with TCP header, excluding options.
2191 * The TCP checksum must be set to zero.
2193 savecsum = th->th_sum;
2195 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2196 th->th_sum = savecsum;
2198 * Step 3: Update MD5 hash with TCP segment data.
2199 * Use m_apply() to avoid an early m_pullup().
2202 m_apply(m, doff, len, tcpsignature_apply, &ctx);
2204 * Step 4: Update MD5 hash with shared secret.
2206 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2207 MD5Final(buf, &ctx);
2208 key_sa_recordxfer(sav, m);
2214 tcpsignature_apply(void *fstate, void *data, unsigned int len)
2217 MD5Update((MD5_CTX *)fstate, (unsigned char *)data, len);
2220 #endif /* TCP_SIGNATURE */
2223 tcp_tso_pullup(struct mbuf **mp, int hoff, struct ip **ip0, int *iphlen0,
2224 struct tcphdr **th0, int *thoff0)
2226 struct mbuf *m = *mp;
2232 len = hoff + sizeof(struct ip);
2234 /* The fixed IP header must reside completely in the first mbuf. */
2235 if (m->m_len < len) {
2236 m = m_pullup(m, len);
2241 ip = mtodoff(m, struct ip *, hoff);
2242 iphlen = IP_VHL_HL(ip->ip_vhl) << 2;
2244 /* The full IP header must reside completely in the one mbuf. */
2245 if (m->m_len < hoff + iphlen) {
2246 m = m_pullup(m, hoff + iphlen);
2249 ip = mtodoff(m, struct ip *, hoff);
2252 KASSERT(ip->ip_p == IPPROTO_TCP, ("not tcp %d", ip->ip_p));
2254 if (m->m_len < hoff + iphlen + sizeof(struct tcphdr)) {
2255 m = m_pullup(m, hoff + iphlen + sizeof(struct tcphdr));
2258 ip = mtodoff(m, struct ip *, hoff);
2261 th = (struct tcphdr *)((caddr_t)ip + iphlen);
2262 thoff = th->th_off << 2;
2264 if (m->m_len < hoff + iphlen + thoff) {
2265 m = m_pullup(m, hoff + iphlen + thoff);
2268 ip = mtodoff(m, struct ip *, hoff);
2269 th = (struct tcphdr *)((caddr_t)ip + iphlen);