Deleted Added
sdiff udiff text old ( 133591 ) new ( 133874 )
full compact
tcp_timewait.c (133591) tcp_timewait.c (133874)
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

--- 13 unchanged lines hidden (view full) ---

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
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

--- 13 unchanged lines hidden (view full) ---

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 * $FreeBSD: head/sys/netinet/tcp_timewait.c 133591 2004-08-12 18:19:36Z dwmalone $
30 * $FreeBSD: head/sys/netinet/tcp_timewait.c 133874 2004-08-16 18:32:07Z rwatson $
31 */
32
33#include "opt_compat.h"
34#include "opt_inet.h"
35#include "opt_inet6.h"
36#include "opt_ipsec.h"
37#include "opt_mac.h"
38#include "opt_tcpdebug.h"

--- 66 unchanged lines hidden (view full) ---

105#endif
106#include <netipsec/key.h>
107#define IPSEC
108#endif /*FAST_IPSEC*/
109
110#include <machine/in_cksum.h>
111#include <sys/md5.h>
112
31 */
32
33#include "opt_compat.h"
34#include "opt_inet.h"
35#include "opt_inet6.h"
36#include "opt_ipsec.h"
37#include "opt_mac.h"
38#include "opt_tcpdebug.h"

--- 66 unchanged lines hidden (view full) ---

105#endif
106#include <netipsec/key.h>
107#define IPSEC
108#endif /*FAST_IPSEC*/
109
110#include <machine/in_cksum.h>
111#include <sys/md5.h>
112
113int tcp_mssdflt = TCP_MSS;
114SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
113int tcp_mssdflt = TCP_MSS;
114SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
115 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
116
117#ifdef INET6
118int tcp_v6mssdflt = TCP6_MSS;
119SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
120 CTLFLAG_RW, &tcp_v6mssdflt , 0,
121 "Default TCP Maximum Segment Size for IPv6");
122#endif

--- 18 unchanged lines hidden (view full) ---

141 * (telnet, SSH) which send many small packets.
142 */
143int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
144SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
145 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
146 "be under the MINMSS Size");
147
148#if 0
115 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
116
117#ifdef INET6
118int tcp_v6mssdflt = TCP6_MSS;
119SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
120 CTLFLAG_RW, &tcp_v6mssdflt , 0,
121 "Default TCP Maximum Segment Size for IPv6");
122#endif

--- 18 unchanged lines hidden (view full) ---

141 * (telnet, SSH) which send many small packets.
142 */
143int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
144SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
145 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
146 "be under the MINMSS Size");
147
148#if 0
149static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
150SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
149static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
150SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
151 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
152#endif
153
154int tcp_do_rfc1323 = 1;
151 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
152#endif
153
154int tcp_do_rfc1323 = 1;
155SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
155SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
156 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
157
158int tcp_do_rfc1644 = 0;
156 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
157
158int tcp_do_rfc1644 = 0;
159SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
159SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
160 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
161
162static int tcp_tcbhashsize = 0;
163SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
164 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
165
166static int do_tcpdrain = 1;
167SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
168 "Enable tcp_drain routine for extra help when low on mbufs");
169
160 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
161
162static int tcp_tcbhashsize = 0;
163SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
164 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
165
166static int do_tcpdrain = 1;
167SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
168 "Enable tcp_drain routine for extra help when low on mbufs");
169
170SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
170SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
171 &tcbinfo.ipi_count, 0, "Number of active PCBs");
172
173static int icmp_may_rst = 1;
171 &tcbinfo.ipi_count, 0, "Number of active PCBs");
172
173static int icmp_may_rst = 1;
174SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
174SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
175 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
176
177static int tcp_isn_reseed_interval = 0;
178SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
179 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
180
181/*
175 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
176
177static int tcp_isn_reseed_interval = 0;
178SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
179 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
180
181/*
182 * TCP bandwidth limiting sysctls. Note that the default lower bound of
183 * 1024 exists only for debugging. A good production default would be
182 * TCP bandwidth limiting sysctls. Note that the default lower bound of
183 * 1024 exists only for debugging. A good production default would be
184 * something like 6100.
185 */
186SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
187 "TCP inflight data limiting");
188
189static int tcp_inflight_enable = 1;
190SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
191 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");

--- 53 unchanged lines hidden (view full) ---

245
246/*
247 * Tcp initialization
248 */
249void
250tcp_init()
251{
252 int hashsize = TCBHASHSIZE;
184 * something like 6100.
185 */
186SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
187 "TCP inflight data limiting");
188
189static int tcp_inflight_enable = 1;
190SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
191 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");

--- 53 unchanged lines hidden (view full) ---

245
246/*
247 * Tcp initialization
248 */
249void
250tcp_init()
251{
252 int hashsize = TCBHASHSIZE;
253
253
254 tcp_ccgen = 1;
255
256 tcp_delacktime = TCPTV_DELACK;
257 tcp_keepinit = TCPTV_KEEP_INIT;
258 tcp_keepidle = TCPTV_KEEP_IDLE;
259 tcp_keepintvl = TCPTV_KEEPINTVL;
260 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
261 tcp_msl = TCPTV_MSL;

--- 7 unchanged lines hidden (view full) ---

269 if (!powerof2(hashsize)) {
270 printf("WARNING: TCB hash size not a power of 2\n");
271 hashsize = 512; /* safe default */
272 }
273 tcp_tcbhashsize = hashsize;
274 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
275 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
276 &tcbinfo.porthashmask);
254 tcp_ccgen = 1;
255
256 tcp_delacktime = TCPTV_DELACK;
257 tcp_keepinit = TCPTV_KEEP_INIT;
258 tcp_keepidle = TCPTV_KEEP_IDLE;
259 tcp_keepintvl = TCPTV_KEEPINTVL;
260 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
261 tcp_msl = TCPTV_MSL;

--- 7 unchanged lines hidden (view full) ---

269 if (!powerof2(hashsize)) {
270 printf("WARNING: TCB hash size not a power of 2\n");
271 hashsize = 512; /* safe default */
272 }
273 tcp_tcbhashsize = hashsize;
274 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
275 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
276 &tcbinfo.porthashmask);
277 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
277 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
278 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
279 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
280#ifdef INET6
281#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
282#else /* INET6 */
283#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
284#endif /* INET6 */
285 if (max_protohdr < TCP_MINPROTOHDR)
286 max_protohdr = TCP_MINPROTOHDR;
287 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
288 panic("tcp_init");
289#undef TCP_MINPROTOHDR
290 /*
291 * These have to be type stable for the benefit of the timers.
292 */
278 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
279 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
280#ifdef INET6
281#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
282#else /* INET6 */
283#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
284#endif /* INET6 */
285 if (max_protohdr < TCP_MINPROTOHDR)
286 max_protohdr = TCP_MINPROTOHDR;
287 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
288 panic("tcp_init");
289#undef TCP_MINPROTOHDR
290 /*
291 * These have to be type stable for the benefit of the timers.
292 */
293 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
293 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 uma_zone_set_max(tcpcb_zone, maxsockets);
294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 uma_zone_set_max(tcpcb_zone, maxsockets);
296 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
296 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
297 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
298 uma_zone_set_max(tcptw_zone, maxsockets / 5);
299 tcp_timer_init();
300 syncache_init();
301 tcp_hc_init();
302 tcp_reass_init();
303 callout_init(&isn_callout, CALLOUT_MPSAFE);
304 tcp_isn_tick(NULL);
305 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
306 SHUTDOWN_PRI_DEFAULT);
297 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
298 uma_zone_set_max(tcptw_zone, maxsockets / 5);
299 tcp_timer_init();
300 syncache_init();
301 tcp_hc_init();
302 tcp_reass_init();
303 callout_init(&isn_callout, CALLOUT_MPSAFE);
304 tcp_isn_tick(NULL);
305 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
306 SHUTDOWN_PRI_DEFAULT);
307 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
307 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
308 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
309}
310
311void
312tcp_fini(xtp)
313 void *xtp;
314{
315 callout_stop(&isn_callout);

--- 138 unchanged lines hidden (view full) ---

454 if (m == NULL) {
455 m = m_gethdr(M_DONTWAIT, MT_HEADER);
456 if (m == NULL)
457 return;
458 tlen = 0;
459 m->m_data += max_linkhdr;
460#ifdef INET6
461 if (isipv6) {
308 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
309}
310
311void
312tcp_fini(xtp)
313 void *xtp;
314{
315 callout_stop(&isn_callout);

--- 138 unchanged lines hidden (view full) ---

454 if (m == NULL) {
455 m = m_gethdr(M_DONTWAIT, MT_HEADER);
456 if (m == NULL)
457 return;
458 tlen = 0;
459 m->m_data += max_linkhdr;
460#ifdef INET6
461 if (isipv6) {
462 bcopy((caddr_t)ip6, mtod(m, caddr_t),
462 bcopy((caddr_t)ip6, mtod(m, caddr_t),
463 sizeof(struct ip6_hdr));
464 ip6 = mtod(m, struct ip6_hdr *);
465 nth = (struct tcphdr *)(ip6 + 1);
466 } else
467#endif /* INET6 */
468 {
469 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
470 ip = mtod(m, struct ip *);

--- 35 unchanged lines hidden (view full) ---

506 ip6->ip6_flow = 0;
507 ip6->ip6_vfc = IPV6_VERSION;
508 ip6->ip6_nxt = IPPROTO_TCP;
509 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
510 tlen));
511 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
512 } else
513#endif
463 sizeof(struct ip6_hdr));
464 ip6 = mtod(m, struct ip6_hdr *);
465 nth = (struct tcphdr *)(ip6 + 1);
466 } else
467#endif /* INET6 */
468 {
469 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
470 ip = mtod(m, struct ip *);

--- 35 unchanged lines hidden (view full) ---

506 ip6->ip6_flow = 0;
507 ip6->ip6_vfc = IPV6_VERSION;
508 ip6->ip6_nxt = IPPROTO_TCP;
509 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
510 tlen));
511 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
512 } else
513#endif
514 {
515 tlen += sizeof (struct tcpiphdr);
516 ip->ip_len = tlen;
517 ip->ip_ttl = ip_defttl;
518 if (path_mtu_discovery)
519 ip->ip_off |= IP_DF;
520 }
514 {
515 tlen += sizeof (struct tcpiphdr);
516 ip->ip_len = tlen;
517 ip->ip_ttl = ip_defttl;
518 if (path_mtu_discovery)
519 ip->ip_off |= IP_DF;
520 }
521 m->m_len = tlen;
522 m->m_pkthdr.len = tlen;
523 m->m_pkthdr.rcvif = NULL;
524#ifdef MAC
525 if (inp != NULL) {
526 /*
527 * Packet is associated with a socket, so allow the
528 * label of the response to reflect the socket label.

--- 23 unchanged lines hidden (view full) ---

552 nth->th_sum = 0;
553 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
554 sizeof(struct ip6_hdr),
555 tlen - sizeof(struct ip6_hdr));
556 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
557 NULL, NULL);
558 } else
559#endif /* INET6 */
521 m->m_len = tlen;
522 m->m_pkthdr.len = tlen;
523 m->m_pkthdr.rcvif = NULL;
524#ifdef MAC
525 if (inp != NULL) {
526 /*
527 * Packet is associated with a socket, so allow the
528 * label of the response to reflect the socket label.

--- 23 unchanged lines hidden (view full) ---

552 nth->th_sum = 0;
553 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
554 sizeof(struct ip6_hdr),
555 tlen - sizeof(struct ip6_hdr));
556 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
557 NULL, NULL);
558 } else
559#endif /* INET6 */
560 {
561 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
562 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
563 m->m_pkthdr.csum_flags = CSUM_TCP;
564 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
565 }
560 {
561 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
562 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
563 m->m_pkthdr.csum_flags = CSUM_TCP;
564 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
565 }
566#ifdef TCPDEBUG
567 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
568 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
569#endif
570#ifdef INET6
571 if (isipv6)
572 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
573 else

--- 57 unchanged lines hidden (view full) ---

631 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
632 tp->t_rttmin = tcp_rexmit_min;
633 tp->t_rxtcur = TCPTV_RTOBASE;
634 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
635 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
636 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
637 tp->t_rcvtime = ticks;
638 tp->t_bw_rtttime = ticks;
566#ifdef TCPDEBUG
567 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
568 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
569#endif
570#ifdef INET6
571 if (isipv6)
572 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
573 else

--- 57 unchanged lines hidden (view full) ---

631 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
632 tp->t_rttmin = tcp_rexmit_min;
633 tp->t_rxtcur = TCPTV_RTOBASE;
634 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
635 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
636 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
637 tp->t_rcvtime = ticks;
638 tp->t_bw_rtttime = ticks;
639 /*
639 /*
640 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
641 * because the socket may be bound to an IPv6 wildcard address,
642 * which may match an IPv4-mapped IPv6 address.
643 */
644 inp->inp_ip_ttl = ip_defttl;
645 inp->inp_ppcb = (caddr_t)tp;
646 return (tp); /* XXX */
647}

--- 87 unchanged lines hidden (view full) ---

735 ssthresh = 0;
736 metrics.rmx_ssthresh = ssthresh;
737
738 metrics.rmx_rtt = tp->t_srtt;
739 metrics.rmx_rttvar = tp->t_rttvar;
740 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
741 metrics.rmx_bandwidth = tp->snd_bandwidth;
742 metrics.rmx_cwnd = tp->snd_cwnd;
640 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
641 * because the socket may be bound to an IPv6 wildcard address,
642 * which may match an IPv4-mapped IPv6 address.
643 */
644 inp->inp_ip_ttl = ip_defttl;
645 inp->inp_ppcb = (caddr_t)tp;
646 return (tp); /* XXX */
647}

--- 87 unchanged lines hidden (view full) ---

735 ssthresh = 0;
736 metrics.rmx_ssthresh = ssthresh;
737
738 metrics.rmx_rtt = tp->t_srtt;
739 metrics.rmx_rttvar = tp->t_rttvar;
740 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
741 metrics.rmx_bandwidth = tp->snd_bandwidth;
742 metrics.rmx_cwnd = tp->snd_cwnd;
743 metrics.rmx_sendpipe = 0;
743 metrics.rmx_sendpipe = 0;
744 metrics.rmx_recvpipe = 0;
745
746 tcp_hc_update(&inp->inp_inc, &metrics);
747 }
748
749 /* free the reassembly queue, if any */
750 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
751 LIST_REMOVE(q, tqe_q);

--- 44 unchanged lines hidden (view full) ---

796 struct tcpcb *tcpb;
797 struct tseg_qent *te;
798
799 /*
800 * Walk the tcpbs, if existing, and flush the reassembly queue,
801 * if there is one...
802 * XXX: The "Net/3" implementation doesn't imply that the TCP
803 * reassembly queue should be flushed, but in a situation
744 metrics.rmx_recvpipe = 0;
745
746 tcp_hc_update(&inp->inp_inc, &metrics);
747 }
748
749 /* free the reassembly queue, if any */
750 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
751 LIST_REMOVE(q, tqe_q);

--- 44 unchanged lines hidden (view full) ---

796 struct tcpcb *tcpb;
797 struct tseg_qent *te;
798
799 /*
800 * Walk the tcpbs, if existing, and flush the reassembly queue,
801 * if there is one...
802 * XXX: The "Net/3" implementation doesn't imply that the TCP
803 * reassembly queue should be flushed, but in a situation
804 * where we're really low on mbufs, this is potentially
805 * usefull.
804 * where we're really low on mbufs, this is potentially
805 * usefull.
806 */
807 INP_INFO_RLOCK(&tcbinfo);
808 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
809 if (inpb->inp_vflag & INP_TIMEWAIT)
810 continue;
811 INP_LOCK(inpb);
812 if ((tcpb = intotcpcb(inpb)) != NULL) {
813 while ((te = LIST_FIRST(&tcpb->t_segq))

--- 95 unchanged lines hidden (view full) ---

909 xig.xig_sogen = so_gencnt;
910 error = SYSCTL_OUT(req, &xig, sizeof xig);
911 if (error)
912 return error;
913
914 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
915 if (inp_list == NULL)
916 return ENOMEM;
806 */
807 INP_INFO_RLOCK(&tcbinfo);
808 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
809 if (inpb->inp_vflag & INP_TIMEWAIT)
810 continue;
811 INP_LOCK(inpb);
812 if ((tcpb = intotcpcb(inpb)) != NULL) {
813 while ((te = LIST_FIRST(&tcpb->t_segq))

--- 95 unchanged lines hidden (view full) ---

909 xig.xig_sogen = so_gencnt;
910 error = SYSCTL_OUT(req, &xig, sizeof xig);
911 if (error)
912 return error;
913
914 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
915 if (inp_list == NULL)
916 return ENOMEM;
917
917
918 s = splnet();
919 INP_INFO_RLOCK(&tcbinfo);
920 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
921 inp = LIST_NEXT(inp, inp_list)) {
922 INP_LOCK(inp);
923 if (inp->inp_gencnt <= gencnt) {
924 /*
925 * XXX: This use of cr_cansee(), introduced with

--- 212 unchanged lines hidden (view full) ---

1138 * excellent DoS attack on machines with many connections.
1139 */
1140 else if (cmd == PRC_HOSTDEAD)
1141 ip = NULL;
1142 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1143 return;
1144 if (ip != NULL) {
1145 s = splnet();
918 s = splnet();
919 INP_INFO_RLOCK(&tcbinfo);
920 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
921 inp = LIST_NEXT(inp, inp_list)) {
922 INP_LOCK(inp);
923 if (inp->inp_gencnt <= gencnt) {
924 /*
925 * XXX: This use of cr_cansee(), introduced with

--- 212 unchanged lines hidden (view full) ---

1138 * excellent DoS attack on machines with many connections.
1139 */
1140 else if (cmd == PRC_HOSTDEAD)
1141 ip = NULL;
1142 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1143 return;
1144 if (ip != NULL) {
1145 s = splnet();
1146 th = (struct tcphdr *)((caddr_t)ip
1146 th = (struct tcphdr *)((caddr_t)ip
1147 + (ip->ip_hl << 2));
1148 INP_INFO_WLOCK(&tcbinfo);
1149 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1150 ip->ip_src, th->th_sport, 0, NULL);
1151 if (inp != NULL) {
1152 INP_LOCK(inp);
1153 if (inp->inp_socket != NULL) {
1154 icmp_seq = htonl(th->th_seq);
1155 tp = intotcpcb(inp);
1156 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1147 + (ip->ip_hl << 2));
1148 INP_INFO_WLOCK(&tcbinfo);
1149 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1150 ip->ip_src, th->th_sport, 0, NULL);
1151 if (inp != NULL) {
1152 INP_LOCK(inp);
1153 if (inp->inp_socket != NULL) {
1154 icmp_seq = htonl(th->th_seq);
1155 tp = intotcpcb(inp);
1156 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1157 SEQ_LT(icmp_seq, tp->snd_max))
1157 SEQ_LT(icmp_seq, tp->snd_max))
1158 inp = (*notify)(inp, inetctlerrmap[cmd]);
1159 }
1160 if (inp != NULL)
1161 INP_UNLOCK(inp);
1162 } else {
1163 struct in_conninfo inc;
1164
1165 inc.inc_fport = th->th_dport;

--- 98 unchanged lines hidden (view full) ---

1264 *
1265 * All ISNs for SYN-ACK packets are generated by the syncache. See
1266 * tcp_syncache.c for details.
1267 *
1268 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1269 * depends on this property. In addition, these ISNs should be
1270 * unguessable so as to prevent connection hijacking. To satisfy
1271 * the requirements of this situation, the algorithm outlined in
1158 inp = (*notify)(inp, inetctlerrmap[cmd]);
1159 }
1160 if (inp != NULL)
1161 INP_UNLOCK(inp);
1162 } else {
1163 struct in_conninfo inc;
1164
1165 inc.inc_fport = th->th_dport;

--- 98 unchanged lines hidden (view full) ---

1264 *
1265 * All ISNs for SYN-ACK packets are generated by the syncache. See
1266 * tcp_syncache.c for details.
1267 *
1268 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1269 * depends on this property. In addition, these ISNs should be
1270 * unguessable so as to prevent connection hijacking. To satisfy
1271 * the requirements of this situation, the algorithm outlined in
1272 * RFC 1948 is used, with only small modifications.
1272 * RFC 1948 is used, with only small modifications.
1273 *
1274 * Implementation details:
1275 *
1276 * Time is based off the system timer, and is corrected so that it
1277 * increases by one megabyte per second. This allows for proper
1278 * recycling on high speed LANs while still leaving over an hour
1279 * before rollover.
1280 *

--- 33 unchanged lines hidden (view full) ---

1314
1315 /* Seed if this is the first use, reseed if requested. */
1316 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1317 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1318 < (u_int)ticks))) {
1319 read_random(&isn_secret, sizeof(isn_secret));
1320 isn_last_reseed = ticks;
1321 }
1273 *
1274 * Implementation details:
1275 *
1276 * Time is based off the system timer, and is corrected so that it
1277 * increases by one megabyte per second. This allows for proper
1278 * recycling on high speed LANs while still leaving over an hour
1279 * before rollover.
1280 *

--- 33 unchanged lines hidden (view full) ---

1314
1315 /* Seed if this is the first use, reseed if requested. */
1316 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1317 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1318 < (u_int)ticks))) {
1319 read_random(&isn_secret, sizeof(isn_secret));
1320 isn_last_reseed = ticks;
1321 }
1322
1322
1323 /* Compute the md5 hash and return the ISN. */
1324 MD5Init(&isn_ctx);
1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1326 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1327#ifdef INET6
1328 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1329 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1330 sizeof(struct in6_addr));

--- 21 unchanged lines hidden (view full) ---

1352 * to keep time flowing at a relatively constant rate. If the random
1353 * increments have already pushed us past the projected offset, do nothing.
1354 */
1355static void
1356tcp_isn_tick(xtp)
1357 void *xtp;
1358{
1359 u_int32_t projected_offset;
1323 /* Compute the md5 hash and return the ISN. */
1324 MD5Init(&isn_ctx);
1325 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1326 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1327#ifdef INET6
1328 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1329 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1330 sizeof(struct in6_addr));

--- 21 unchanged lines hidden (view full) ---

1352 * to keep time flowing at a relatively constant rate. If the random
1353 * increments have already pushed us past the projected offset, do nothing.
1354 */
1355static void
1356tcp_isn_tick(xtp)
1357 void *xtp;
1358{
1359 u_int32_t projected_offset;
1360
1360
1361 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz;
1362
1363 if (projected_offset > isn_offset)
1364 isn_offset = projected_offset;
1365
1366 isn_offset_old = isn_offset;
1367 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL);
1368}

--- 138 unchanged lines hidden (view full) ---

1507}
1508
1509/*
1510 * Look-up the routing entry to the peer of this inpcb. If no route
1511 * is found and it cannot be allocated, then return NULL. This routine
1512 * is called by TCP routines that access the rmx structure and by tcp_mss
1513 * to get the interface MTU.
1514 */
1361 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / hz;
1362
1363 if (projected_offset > isn_offset)
1364 isn_offset = projected_offset;
1365
1366 isn_offset_old = isn_offset;
1367 callout_reset(&isn_callout, 1, tcp_isn_tick, NULL);
1368}

--- 138 unchanged lines hidden (view full) ---

1507}
1508
1509/*
1510 * Look-up the routing entry to the peer of this inpcb. If no route
1511 * is found and it cannot be allocated, then return NULL. This routine
1512 * is called by TCP routines that access the rmx structure and by tcp_mss
1513 * to get the interface MTU.
1514 */
1515u_long
1515u_long
1516tcp_maxmtu(inc)
1517 struct in_conninfo *inc;
1518{
1519 struct route sro;
1520 struct sockaddr_in *dst;
1521 struct ifnet *ifp;
1522 u_long maxmtu = 0;
1523

--- 76 unchanged lines hidden (view full) ---

1600 ip6 = mtod(m, struct ip6_hdr *);
1601 th = (struct tcphdr *)(ip6 + 1);
1602 m->m_pkthdr.len = m->m_len =
1603 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1604 tcpip_fillheaders(inp, ip6, th);
1605 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1606 } else
1607#endif /* INET6 */
1516tcp_maxmtu(inc)
1517 struct in_conninfo *inc;
1518{
1519 struct route sro;
1520 struct sockaddr_in *dst;
1521 struct ifnet *ifp;
1522 u_long maxmtu = 0;
1523

--- 76 unchanged lines hidden (view full) ---

1600 ip6 = mtod(m, struct ip6_hdr *);
1601 th = (struct tcphdr *)(ip6 + 1);
1602 m->m_pkthdr.len = m->m_len =
1603 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1604 tcpip_fillheaders(inp, ip6, th);
1605 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1606 } else
1607#endif /* INET6 */
1608 {
1609 ip = mtod(m, struct ip *);
1610 th = (struct tcphdr *)(ip + 1);
1611 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1612 tcpip_fillheaders(inp, ip, th);
1613 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1614 }
1608 {
1609 ip = mtod(m, struct ip *);
1610 th = (struct tcphdr *)(ip + 1);
1611 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1612 tcpip_fillheaders(inp, ip, th);
1613 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1614 }
1615
1616 m_free(m);
1617 return hdrsiz;
1618}
1619#endif /*IPSEC*/
1620
1621/*
1622 * Move a TCP connection into TIME_WAIT state.

--- 23 unchanged lines hidden (view full) ---

1646 /*
1647 * Recover last window size sent.
1648 */
1649 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1650
1651 /*
1652 * Set t_recent if timestamps are used on the connection.
1653 */
1615
1616 m_free(m);
1617 return hdrsiz;
1618}
1619#endif /*IPSEC*/
1620
1621/*
1622 * Move a TCP connection into TIME_WAIT state.

--- 23 unchanged lines hidden (view full) ---

1646 /*
1647 * Recover last window size sent.
1648 */
1649 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1650
1651 /*
1652 * Set t_recent if timestamps are used on the connection.
1653 */
1654 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1655 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1654 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1655 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1656 tw->t_recent = tp->ts_recent;
1657 else
1658 tw->t_recent = 0;
1659
1660 tw->snd_nxt = tp->snd_nxt;
1661 tw->rcv_nxt = tp->rcv_nxt;
1662 tw->iss = tp->iss;
1663 tw->irs = tp->irs;

--- 50 unchanged lines hidden (view full) ---

1714int
1715tcp_twrecycleable(struct tcptw *tw)
1716{
1717 tcp_seq new_iss = tw->iss;
1718 tcp_seq new_irs = tw->irs;
1719
1720 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1721 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1656 tw->t_recent = tp->ts_recent;
1657 else
1658 tw->t_recent = 0;
1659
1660 tw->snd_nxt = tp->snd_nxt;
1661 tw->rcv_nxt = tp->rcv_nxt;
1662 tw->iss = tp->iss;
1663 tw->irs = tp->irs;

--- 50 unchanged lines hidden (view full) ---

1714int
1715tcp_twrecycleable(struct tcptw *tw)
1716{
1717 tcp_seq new_iss = tw->iss;
1718 tcp_seq new_irs = tw->irs;
1719
1720 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1721 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1722
1722
1723 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1724 return 1;
1725 else
1726 return 0;
1727}
1728
1729struct tcptw *
1730tcp_twclose(struct tcptw *tw, int reuse)

--- 53 unchanged lines hidden (view full) ---

1784#endif
1785 {
1786 hdrlen = sizeof(struct tcpiphdr);
1787 ip = mtod(m, struct ip *);
1788 th = (struct tcphdr *)(ip + 1);
1789 tcpip_fillheaders(inp, ip, th);
1790 }
1791 optp = (u_int8_t *)(th + 1);
1723 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1724 return 1;
1725 else
1726 return 0;
1727}
1728
1729struct tcptw *
1730tcp_twclose(struct tcptw *tw, int reuse)

--- 53 unchanged lines hidden (view full) ---

1784#endif
1785 {
1786 hdrlen = sizeof(struct tcpiphdr);
1787 ip = mtod(m, struct ip *);
1788 th = (struct tcphdr *)(ip + 1);
1789 tcpip_fillheaders(inp, ip, th);
1790 }
1791 optp = (u_int8_t *)(th + 1);
1792
1793 /*
1792
1793 /*
1794 * Send a timestamp and echo-reply if both our side and our peer
1795 * have sent timestamps in our SYN's and this is not a RST.
1794 * Send a timestamp and echo-reply if both our side and our peer
1795 * have sent timestamps in our SYN's and this is not a RST.
1796 */
1796 */
1797 if (tw->t_recent && flags == TH_ACK) {
1798 u_int32_t *lp = (u_int32_t *)optp;
1799
1797 if (tw->t_recent && flags == TH_ACK) {
1798 u_int32_t *lp = (u_int32_t *)optp;
1799
1800 /* Form timestamp option as shown in appendix A of RFC 1323. */
1801 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1802 *lp++ = htonl(ticks);
1803 *lp = htonl(tw->t_recent);
1804 optp += TCPOLEN_TSTAMP_APPA;
1805 }
1800 /* Form timestamp option as shown in appendix A of RFC 1323. */
1801 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1802 *lp++ = htonl(ticks);
1803 *lp = htonl(tw->t_recent);
1804 optp += TCPOLEN_TSTAMP_APPA;
1805 }
1806
1806
1807 /*
1807 /*
1808 * Send `CC-family' options if needed, and it's not a RST.
1808 * Send `CC-family' options if needed, and it's not a RST.
1809 */
1809 */
1810 if (tw->cc_recv != 0 && flags == TH_ACK) {
1811 u_int32_t *lp = (u_int32_t *)optp;
1812
1813 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1814 *lp = htonl(tw->cc_send);
1815 optp += TCPOLEN_CC_APPA;
1810 if (tw->cc_recv != 0 && flags == TH_ACK) {
1811 u_int32_t *lp = (u_int32_t *)optp;
1812
1813 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1814 *lp = htonl(tw->cc_send);
1815 optp += TCPOLEN_CC_APPA;
1816 }
1816 }
1817 optlen = optp - (u_int8_t *)(th + 1);
1818
1819 m->m_len = hdrlen + optlen;
1820 m->m_pkthdr.len = m->m_len;
1821
1822 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1823
1824 th->th_seq = htonl(tw->snd_nxt);

--- 8 unchanged lines hidden (view full) ---

1833 sizeof(struct tcphdr) + optlen);
1834 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1835 error = ip6_output(m, inp->in6p_outputopts, NULL,
1836 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1837 } else
1838#endif
1839 {
1840 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1817 optlen = optp - (u_int8_t *)(th + 1);
1818
1819 m->m_len = hdrlen + optlen;
1820 m->m_pkthdr.len = m->m_len;
1821
1822 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1823
1824 th->th_seq = htonl(tw->snd_nxt);

--- 8 unchanged lines hidden (view full) ---

1833 sizeof(struct tcphdr) + optlen);
1834 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1835 error = ip6_output(m, inp->in6p_outputopts, NULL,
1836 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1837 } else
1838#endif
1839 {
1840 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1841 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1841 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1842 m->m_pkthdr.csum_flags = CSUM_TCP;
1843 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1844 ip->ip_len = m->m_pkthdr.len;
1845 if (path_mtu_discovery)
1846 ip->ip_off |= IP_DF;
1847 error = ip_output(m, inp->inp_options, NULL,
1848 (tw->tw_so_options & SO_DONTROUTE), NULL, inp);
1849 }

--- 13 unchanged lines hidden (view full) ---

1863 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1864 * routers. This code also does a fairly good job keeping RTTs in check
1865 * across slow links like modems. We implement an algorithm which is very
1866 * similar (but not meant to be) TCP/Vegas. The code operates on the
1867 * transmitter side of a TCP connection and so only effects the transmit
1868 * side of the connection.
1869 *
1870 * BACKGROUND: TCP makes no provision for the management of buffer space
1842 m->m_pkthdr.csum_flags = CSUM_TCP;
1843 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1844 ip->ip_len = m->m_pkthdr.len;
1845 if (path_mtu_discovery)
1846 ip->ip_off |= IP_DF;
1847 error = ip_output(m, inp->inp_options, NULL,
1848 (tw->tw_so_options & SO_DONTROUTE), NULL, inp);
1849 }

--- 13 unchanged lines hidden (view full) ---

1863 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1864 * routers. This code also does a fairly good job keeping RTTs in check
1865 * across slow links like modems. We implement an algorithm which is very
1866 * similar (but not meant to be) TCP/Vegas. The code operates on the
1867 * transmitter side of a TCP connection and so only effects the transmit
1868 * side of the connection.
1869 *
1870 * BACKGROUND: TCP makes no provision for the management of buffer space
1871 * at the end points or at the intermediate routers and switches. A TCP
1871 * at the end points or at the intermediate routers and switches. A TCP
1872 * stream, whether using NewReno or not, will eventually buffer as
1873 * many packets as it is able and the only reason this typically works is
1874 * due to the fairly small default buffers made available for a connection
1875 * (typicaly 16K or 32K). As machines use larger windows and/or window
1876 * scaling it is now fairly easy for even a single TCP connection to blow-out
1872 * stream, whether using NewReno or not, will eventually buffer as
1873 * many packets as it is able and the only reason this typically works is
1874 * due to the fairly small default buffers made available for a connection
1875 * (typicaly 16K or 32K). As machines use larger windows and/or window
1876 * scaling it is now fairly easy for even a single TCP connection to blow-out
1877 * all available buffer space not only on the local interface, but on
1877 * all available buffer space not only on the local interface, but on
1878 * intermediate routers and switches as well. NewReno makes a misguided
1879 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1880 * then backing off, then steadily increasing the window again until another
1881 * failure occurs, ad-infinitum. This results in terrible oscillation that
1882 * is only made worse as network loads increase and the idea of intentionally
1883 * blowing out network buffers is, frankly, a terrible way to manage network
1884 * resources.
1885 *

--- 5 unchanged lines hidden (view full) ---

1891 * repeat ad-infinitum. This method works in principle but has severe
1892 * implementation issues due to RTT variances, timer granularity, and
1893 * instability in the algorithm which can lead to many false positives and
1894 * create oscillations as well as interact badly with other TCP streams
1895 * implementing the same algorithm.
1896 *
1897 * The second method is to limit the window to the bandwidth delay product
1898 * of the link. This is the method we implement. RTT variances and our
1878 * intermediate routers and switches as well. NewReno makes a misguided
1879 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1880 * then backing off, then steadily increasing the window again until another
1881 * failure occurs, ad-infinitum. This results in terrible oscillation that
1882 * is only made worse as network loads increase and the idea of intentionally
1883 * blowing out network buffers is, frankly, a terrible way to manage network
1884 * resources.
1885 *

--- 5 unchanged lines hidden (view full) ---

1891 * repeat ad-infinitum. This method works in principle but has severe
1892 * implementation issues due to RTT variances, timer granularity, and
1893 * instability in the algorithm which can lead to many false positives and
1894 * create oscillations as well as interact badly with other TCP streams
1895 * implementing the same algorithm.
1896 *
1897 * The second method is to limit the window to the bandwidth delay product
1898 * of the link. This is the method we implement. RTT variances and our
1899 * own manipulation of the congestion window, bwnd, can potentially
1899 * own manipulation of the congestion window, bwnd, can potentially
1900 * destabilize the algorithm. For this reason we have to stabilize the
1901 * elements used to calculate the window. We do this by using the minimum
1902 * observed RTT, the long term average of the observed bandwidth, and
1903 * by adding two segments worth of slop. It isn't perfect but it is able
1904 * to react to changing conditions and gives us a very stable basis on
1905 * which to extend the algorithm.
1906 */
1907void

--- 23 unchanged lines hidden (view full) ---

1931 *
1932 * Note: if ticks rollover 'bw' may wind up negative. We must
1933 * effectively reset t_bw_rtttime for this case.
1934 */
1935 save_ticks = ticks;
1936 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1937 return;
1938
1900 * destabilize the algorithm. For this reason we have to stabilize the
1901 * elements used to calculate the window. We do this by using the minimum
1902 * observed RTT, the long term average of the observed bandwidth, and
1903 * by adding two segments worth of slop. It isn't perfect but it is able
1904 * to react to changing conditions and gives us a very stable basis on
1905 * which to extend the algorithm.
1906 */
1907void

--- 23 unchanged lines hidden (view full) ---

1931 *
1932 * Note: if ticks rollover 'bw' may wind up negative. We must
1933 * effectively reset t_bw_rtttime for this case.
1934 */
1935 save_ticks = ticks;
1936 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1937 return;
1938
1939 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1939 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1940 (save_ticks - tp->t_bw_rtttime);
1941 tp->t_bw_rtttime = save_ticks;
1942 tp->t_bw_rtseq = ack_seq;
1943 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1944 return;
1945 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1946
1947 tp->snd_bandwidth = bw;
1948
1949 /*
1950 * Calculate the semi-static bandwidth delay product, plus two maximal
1951 * segments. The additional slop puts us squarely in the sweet
1952 * spot and also handles the bandwidth run-up case and stabilization.
1953 * Without the slop we could be locking ourselves into a lower
1954 * bandwidth.
1955 *
1956 * Situations Handled:
1957 * (1) Prevents over-queueing of packets on LANs, especially on
1958 * high speed LANs, allowing larger TCP buffers to be
1940 (save_ticks - tp->t_bw_rtttime);
1941 tp->t_bw_rtttime = save_ticks;
1942 tp->t_bw_rtseq = ack_seq;
1943 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1944 return;
1945 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1946
1947 tp->snd_bandwidth = bw;
1948
1949 /*
1950 * Calculate the semi-static bandwidth delay product, plus two maximal
1951 * segments. The additional slop puts us squarely in the sweet
1952 * spot and also handles the bandwidth run-up case and stabilization.
1953 * Without the slop we could be locking ourselves into a lower
1954 * bandwidth.
1955 *
1956 * Situations Handled:
1957 * (1) Prevents over-queueing of packets on LANs, especially on
1958 * high speed LANs, allowing larger TCP buffers to be
1959 * specified, and also does a good job preventing
1959 * specified, and also does a good job preventing
1960 * over-queueing of packets over choke points like modems
1961 * (at least for the transmit side).
1962 *
1963 * (2) Is able to handle changing network loads (bandwidth
1964 * drops so bwnd drops, bandwidth increases so bwnd
1965 * increases).
1966 *
1967 * (3) Theoretically should stabilize in the face of multiple

--- 155 unchanged lines hidden --- 		1960 * over-queueing of packets over choke points like modems
1961 * (at least for the transmit side).
1962 *
1963 * (2) Is able to handle changing network loads (bandwidth
1964 * drops so bwnd drops, bandwidth increases so bwnd
1965 * increases).
1966 *
1967 * (3) Theoretically should stabilize in the face of multiple

--- 155 unchanged lines hidden ---