kernel - Improve TCP socket handling at high speeds

[dragonfly.git] / sys / netinet / tcp_subr.c

/*
 * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
 * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
 *
 * This code is derived from software contributed to The DragonFly Project
 * by Jeffrey M. Hsu.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

/*
 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
 *      The Regents of the University of California.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 *
 *      @(#)tcp_subr.c  8.2 (Berkeley) 5/24/95
 * $FreeBSD: src/sys/netinet/tcp_subr.c,v 1.73.2.31 2003/01/24 05:11:34 sam Exp $
 */

#include "opt_compat.h"
#include "opt_inet.h"
#include "opt_inet6.h"
#include "opt_ipsec.h"
#include "opt_tcpdebug.h"

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/callout.h>
#include <sys/kernel.h>
#include <sys/sysctl.h>
#include <sys/malloc.h>
#include <sys/mpipe.h>
#include <sys/mbuf.h>
#ifdef INET6
#include <sys/domain.h>
#endif
#include <sys/proc.h>
#include <sys/priv.h>
#include <sys/socket.h>
#include <sys/socketops.h>
#include <sys/socketvar.h>
#include <sys/protosw.h>
#include <sys/random.h>
#include <sys/in_cksum.h>
#include <sys/ktr.h>

#include <net/route.h>
#include <net/if.h>
#include <net/netisr2.h>

#define _IP_VHL
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>
#include <netinet/in_pcb.h>
#include <netinet6/in6_pcb.h>
#include <netinet/in_var.h>
#include <netinet/ip_var.h>
#include <netinet6/ip6_var.h>
#include <netinet/ip_icmp.h>
#ifdef INET6
#include <netinet/icmp6.h>
#endif
#include <netinet/tcp.h>
#include <netinet/tcp_fsm.h>
#include <netinet/tcp_seq.h>
#include <netinet/tcp_timer.h>
#include <netinet/tcp_timer2.h>
#include <netinet/tcp_var.h>
#include <netinet6/tcp6_var.h>
#include <netinet/tcpip.h>
#ifdef TCPDEBUG
#include <netinet/tcp_debug.h>
#endif
#include <netinet6/ip6protosw.h>

#ifdef IPSEC
#include <netinet6/ipsec.h>
#include <netproto/key/key.h>
#ifdef INET6
#include <netinet6/ipsec6.h>
#endif
#endif

#ifdef FAST_IPSEC
#include <netproto/ipsec/ipsec.h>
#ifdef INET6
#include <netproto/ipsec/ipsec6.h>
#endif
#define IPSEC
#endif

#include <sys/md5.h>
#include <machine/smp.h>

#include <sys/msgport2.h>
#include <sys/mplock2.h>
#include <net/netmsg2.h>

#if !defined(KTR_TCP)
#define KTR_TCP         KTR_ALL
#endif
/*
KTR_INFO_MASTER(tcp);
KTR_INFO(KTR_TCP, tcp, rxmsg, 0, "tcp getmsg", 0);
KTR_INFO(KTR_TCP, tcp, wait, 1, "tcp waitmsg", 0);
KTR_INFO(KTR_TCP, tcp, delayed, 2, "tcp execute delayed ops", 0);
#define logtcp(name)    KTR_LOG(tcp_ ## name)
*/

#define TCP_IW_MAXSEGS_DFLT     4
#define TCP_IW_CAPSEGS_DFLT     3

struct inpcbinfo tcbinfo[MAXCPU];
struct tcpcbackqhead tcpcbackq[MAXCPU];

int tcp_mssdflt = TCP_MSS;
SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
    &tcp_mssdflt, 0, "Default TCP Maximum Segment Size");

#ifdef INET6
int tcp_v6mssdflt = TCP6_MSS;
SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLFLAG_RW,
    &tcp_v6mssdflt, 0, "Default TCP Maximum Segment Size for IPv6");
#endif

/*
 * Minimum MSS we accept and use. This prevents DoS attacks where
 * we are forced to a ridiculous low MSS like 20 and send hundreds
 * of packets instead of one. The effect scales with the available
 * bandwidth and quickly saturates the CPU and network interface
 * with packet generation and sending. Set to zero to disable MINMSS
 * checking. This setting prevents us from sending too small packets.
 */
int tcp_minmss = TCP_MINMSS;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
    &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");

#if 0
static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
    &tcp_rttdflt, 0, "Default maximum TCP Round Trip Time");
#endif

int tcp_do_rfc1323 = 1;
SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
    &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");

static int tcp_tcbhashsize = 0;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RD,
     &tcp_tcbhashsize, 0, "Size of TCP control block hashtable");

static int do_tcpdrain = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
     "Enable tcp_drain routine for extra help when low on mbufs");

static int icmp_may_rst = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
    "Certain ICMP unreachable messages may abort connections in SYN_SENT");

static int tcp_isn_reseed_interval = 0;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
    &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");

/*
 * TCP bandwidth limiting sysctls.  The inflight limiter is now turned on
 * by default, but with generous values which should allow maximal
 * bandwidth.  In particular, the slop defaults to 50 (5 packets).
 *
 * The reason for doing this is that the limiter is the only mechanism we
 * have which seems to do a really good job preventing receiver RX rings
 * on network interfaces from getting blown out.  Even though GigE/10GigE
 * is supposed to flow control it looks like either it doesn't actually
 * do it or Open Source drivers do not properly enable it.
 *
 * People using the limiter to reduce bottlenecks on slower WAN connections
 * should set the slop to 20 (2 packets).
 */
static int tcp_inflight_enable = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_enable, CTLFLAG_RW,
    &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");

static int tcp_inflight_debug = 0;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_debug, CTLFLAG_RW,
    &tcp_inflight_debug, 0, "Debug TCP inflight calculations");

static int tcp_inflight_min = 6144;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_min, CTLFLAG_RW,
    &tcp_inflight_min, 0, "Lower bound for TCP inflight window");

static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_max, CTLFLAG_RW,
    &tcp_inflight_max, 0, "Upper bound for TCP inflight window");

static int tcp_inflight_stab = 50;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_stab, CTLFLAG_RW,
    &tcp_inflight_stab, 0, "Fudge bw 1/10% (50=5%)");

static int tcp_inflight_adjrtt = 2;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, inflight_adjrtt, CTLFLAG_RW,
    &tcp_inflight_adjrtt, 0, "Slop for rtt 1/(hz*32)");

static int tcp_do_rfc3390 = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
    &tcp_do_rfc3390, 0,
    "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");

static u_long tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT;
SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwmaxsegs, CTLFLAG_RW,
    &tcp_iw_maxsegs, 0, "TCP IW segments max");

static u_long tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT;
SYSCTL_ULONG(_net_inet_tcp, OID_AUTO, iwcapsegs, CTLFLAG_RW,
    &tcp_iw_capsegs, 0, "TCP IW segments");

int tcp_low_rtobase = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, low_rtobase, CTLFLAG_RW,
    &tcp_low_rtobase, 0, "Lowering the Initial RTO (RFC 6298)");

static int tcp_do_ncr = 1;
SYSCTL_INT(_net_inet_tcp, OID_AUTO, ncr, CTLFLAG_RW,
    &tcp_do_ncr, 0, "Non-Congestion Robustness (RFC 4653)");

static MALLOC_DEFINE(M_TCPTEMP, "tcptemp", "TCP Templates for Keepalives");
static struct malloc_pipe tcptemp_mpipe;

static void tcp_willblock(void);
static void tcp_notify (struct inpcb *, int);

struct tcp_stats tcpstats_percpu[MAXCPU] __cachealign;

static int
sysctl_tcpstats(SYSCTL_HANDLER_ARGS)
{
        int cpu, error = 0;

        for (cpu = 0; cpu < ncpus2; ++cpu) {
                if ((error = SYSCTL_OUT(req, &tcpstats_percpu[cpu],
                                        sizeof(struct tcp_stats))))
                        break;
                if ((error = SYSCTL_IN(req, &tcpstats_percpu[cpu],
                                       sizeof(struct tcp_stats))))
                        break;
        }

        return (error);
}
SYSCTL_PROC(_net_inet_tcp, TCPCTL_STATS, stats, (CTLTYPE_OPAQUE | CTLFLAG_RW),
    0, 0, sysctl_tcpstats, "S,tcp_stats", "TCP statistics");

/*
 * Target size of TCP PCB hash tables. Must be a power of two.
 *
 * Note that this can be overridden by the kernel environment
 * variable net.inet.tcp.tcbhashsize
 */
#ifndef TCBHASHSIZE
#define TCBHASHSIZE     512
#endif

/*
 * This is the actual shape of what we allocate using the zone
 * allocator.  Doing it this way allows us to protect both structures
 * using the same generation count, and also eliminates the overhead
 * of allocating tcpcbs separately.  By hiding the structure here,
 * we avoid changing most of the rest of the code (although it needs
 * to be changed, eventually, for greater efficiency).
 */
#define ALIGNMENT       32
#define ALIGNM1         (ALIGNMENT - 1)
struct  inp_tp {
        union {
                struct  inpcb inp;
                char    align[(sizeof(struct inpcb) + ALIGNM1) & ~ALIGNM1];
        } inp_tp_u;
        struct  tcpcb tcb;
        struct  tcp_callout inp_tp_rexmt;
        struct  tcp_callout inp_tp_persist;
        struct  tcp_callout inp_tp_keep;
        struct  tcp_callout inp_tp_2msl;
        struct  tcp_callout inp_tp_delack;
        struct  netmsg_tcp_timer inp_tp_timermsg;
        struct  netmsg_base inp_tp_sndmore;
};
#undef ALIGNMENT
#undef ALIGNM1

/*
 * Tcp initialization
 */
void
tcp_init(void)
{
        struct inpcbportinfo *portinfo;
        struct inpcbinfo *ticb;
        int hashsize = TCBHASHSIZE;
        int cpu;

        /*
         * note: tcptemp is used for keepalives, and it is ok for an
         * allocation to fail so do not specify MPF_INT.
         */
        mpipe_init(&tcptemp_mpipe, M_TCPTEMP, sizeof(struct tcptemp),
                    25, -1, 0, NULL, NULL, NULL);

        tcp_delacktime = TCPTV_DELACK;
        tcp_keepinit = TCPTV_KEEP_INIT;
        tcp_keepidle = TCPTV_KEEP_IDLE;
        tcp_keepintvl = TCPTV_KEEPINTVL;
        tcp_maxpersistidle = TCPTV_KEEP_IDLE;
        tcp_msl = TCPTV_MSL;
        tcp_rexmit_min = TCPTV_MIN;
        tcp_rexmit_slop = TCPTV_CPU_VAR;

        TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
        if (!powerof2(hashsize)) {
                kprintf("WARNING: TCB hash size not a power of 2\n");
                hashsize = 512; /* safe default */
        }
        tcp_tcbhashsize = hashsize;

        portinfo = kmalloc_cachealign(sizeof(*portinfo) * ncpus2, M_PCB,
            M_WAITOK);

        for (cpu = 0; cpu < ncpus2; cpu++) {
                ticb = &tcbinfo[cpu];
                in_pcbinfo_init(ticb);
                ticb->cpu = cpu;
                ticb->hashbase = hashinit(hashsize, M_PCB,
                                          &ticb->hashmask);
                in_pcbportinfo_init(&portinfo[cpu], hashsize, TRUE, cpu);
                ticb->portinfo = portinfo;
                ticb->portinfo_mask = ncpus2_mask;
                ticb->wildcardhashbase = hashinit(hashsize, M_PCB,
                                                  &ticb->wildcardhashmask);
                ticb->localgrphashbase = hashinit(hashsize, M_PCB,
                                                  &ticb->localgrphashmask);
                ticb->ipi_size = sizeof(struct inp_tp);
                TAILQ_INIT(&tcpcbackq[cpu]);
        }

        tcp_reass_maxseg = nmbclusters / 16;
        TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", &tcp_reass_maxseg);

#ifdef INET6
#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
#else
#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
#endif
        if (max_protohdr < TCP_MINPROTOHDR)
                max_protohdr = TCP_MINPROTOHDR;
        if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
                panic("tcp_init");
#undef TCP_MINPROTOHDR

        /*
         * Initialize TCP statistics counters for each CPU.
         */
        for (cpu = 0; cpu < ncpus2; ++cpu)
                bzero(&tcpstats_percpu[cpu], sizeof(struct tcp_stats));

        syncache_init();
        netisr_register_rollup(tcp_willblock, NETISR_ROLLUP_PRIO_TCP);
}

static void
tcp_willblock(void)
{
        struct tcpcb *tp;
        int cpu = mycpu->gd_cpuid;

        while ((tp = TAILQ_FIRST(&tcpcbackq[cpu])) != NULL) {
                KKASSERT(tp->t_flags & TF_ONOUTPUTQ);
                tp->t_flags &= ~TF_ONOUTPUTQ;
                TAILQ_REMOVE(&tcpcbackq[cpu], tp, t_outputq);
                tcp_output(tp);
        }
}

/*
 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
 * tcp_template used to store this data in mbufs, but we now recopy it out
 * of the tcpcb each time to conserve mbufs.
 */
void
tcp_fillheaders(struct tcpcb *tp, void *ip_ptr, void *tcp_ptr, boolean_t tso)
{
        struct inpcb *inp = tp->t_inpcb;
        struct tcphdr *tcp_hdr = (struct tcphdr *)tcp_ptr;

#ifdef INET6
        if (inp->inp_vflag & INP_IPV6) {
                struct ip6_hdr *ip6;

                ip6 = (struct ip6_hdr *)ip_ptr;
                ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
                        (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
                ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
                        (IPV6_VERSION & IPV6_VERSION_MASK);
                ip6->ip6_nxt = IPPROTO_TCP;
                ip6->ip6_plen = sizeof(struct tcphdr);
                ip6->ip6_src = inp->in6p_laddr;
                ip6->ip6_dst = inp->in6p_faddr;
                tcp_hdr->th_sum = 0;
        } else
#endif
        {
                struct ip *ip = (struct ip *) ip_ptr;
                u_int plen;

                ip->ip_vhl = IP_VHL_BORING;
                ip->ip_tos = 0;
                ip->ip_len = 0;
                ip->ip_id = 0;
                ip->ip_off = 0;
                ip->ip_ttl = 0;
                ip->ip_sum = 0;
                ip->ip_p = IPPROTO_TCP;
                ip->ip_src = inp->inp_laddr;
                ip->ip_dst = inp->inp_faddr;

                if (tso)
                        plen = htons(IPPROTO_TCP);
                else
                        plen = htons(sizeof(struct tcphdr) + IPPROTO_TCP);
                tcp_hdr->th_sum = in_pseudo(ip->ip_src.s_addr,
                    ip->ip_dst.s_addr, plen);
        }

        tcp_hdr->th_sport = inp->inp_lport;
        tcp_hdr->th_dport = inp->inp_fport;
        tcp_hdr->th_seq = 0;
        tcp_hdr->th_ack = 0;
        tcp_hdr->th_x2 = 0;
        tcp_hdr->th_off = 5;
        tcp_hdr->th_flags = 0;
        tcp_hdr->th_win = 0;
        tcp_hdr->th_urp = 0;
}

/*
 * Create template to be used to send tcp packets on a connection.
 * Allocates an mbuf and fills in a skeletal tcp/ip header.  The only
 * use for this function is in keepalives, which use tcp_respond.
 */
struct tcptemp *
tcp_maketemplate(struct tcpcb *tp)
{
        struct tcptemp *tmp;

        if ((tmp = mpipe_alloc_nowait(&tcptemp_mpipe)) == NULL)
                return (NULL);
        tcp_fillheaders(tp, &tmp->tt_ipgen, &tmp->tt_t, FALSE);
        return (tmp);
}

void
tcp_freetemplate(struct tcptemp *tmp)
{
        mpipe_free(&tcptemp_mpipe, tmp);
}

/*
 * Send a single message to the TCP at address specified by
 * the given TCP/IP header.  If m == NULL, then we make a copy
 * of the tcpiphdr at ti and send directly to the addressed host.
 * This is used to force keep alive messages out using the TCP
 * template for a connection.  If flags are given then we send
 * a message back to the TCP which originated the * segment ti,
 * and discard the mbuf containing it and any other attached mbufs.
 *
 * In any case the ack and sequence number of the transmitted
 * segment are as specified by the parameters.
 *
 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
 */
void
tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
            tcp_seq ack, tcp_seq seq, int flags)
{
        int tlen;
        int win = 0;
        struct route *ro = NULL;
        struct route sro;
        struct ip *ip = ipgen;
        struct tcphdr *nth;
        int ipflags = 0;
        struct route_in6 *ro6 = NULL;
        struct route_in6 sro6;
        struct ip6_hdr *ip6 = ipgen;
        boolean_t use_tmpro = TRUE;
#ifdef INET6
        boolean_t isipv6 = (IP_VHL_V(ip->ip_vhl) == 6);
#else
        const boolean_t isipv6 = FALSE;
#endif

        if (tp != NULL) {
                if (!(flags & TH_RST)) {
                        win = ssb_space(&tp->t_inpcb->inp_socket->so_rcv);
                        if (win < 0)
                                win = 0;
                        if (win > (long)TCP_MAXWIN << tp->rcv_scale)
                                win = (long)TCP_MAXWIN << tp->rcv_scale;
                }
                /*
                 * Don't use the route cache of a listen socket,
                 * it is not MPSAFE; use temporary route cache.
                 */
                if (tp->t_state != TCPS_LISTEN) {
                        if (isipv6)
                                ro6 = &tp->t_inpcb->in6p_route;
                        else
                                ro = &tp->t_inpcb->inp_route;
                        use_tmpro = FALSE;
                }
        }
        if (use_tmpro) {
                if (isipv6) {
                        ro6 = &sro6;
                        bzero(ro6, sizeof *ro6);
                } else {
                        ro = &sro;
                        bzero(ro, sizeof *ro);
                }
        }
        if (m == NULL) {
                m = m_gethdr(MB_DONTWAIT, MT_HEADER);
                if (m == NULL)
                        return;
                tlen = 0;
                m->m_data += max_linkhdr;
                if (isipv6) {
                        bcopy(ip6, mtod(m, caddr_t), sizeof(struct ip6_hdr));
                        ip6 = mtod(m, struct ip6_hdr *);
                        nth = (struct tcphdr *)(ip6 + 1);
                } else {
                        bcopy(ip, mtod(m, caddr_t), sizeof(struct ip));
                        ip = mtod(m, struct ip *);
                        nth = (struct tcphdr *)(ip + 1);
                }
                bcopy(th, nth, sizeof(struct tcphdr));
                flags = TH_ACK;
        } else {
                m_freem(m->m_next);
                m->m_next = NULL;
                m->m_data = (caddr_t)ipgen;
                /* m_len is set later */
                tlen = 0;
#define xchg(a, b, type) { type t; t = a; a = b; b = t; }
                if (isipv6) {
                        xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
                        nth = (struct tcphdr *)(ip6 + 1);
                } else {
                        xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
                        nth = (struct tcphdr *)(ip + 1);
                }
                if (th != nth) {
                        /*
                         * this is usually a case when an extension header
                         * exists between the IPv6 header and the
                         * TCP header.
                         */
                        nth->th_sport = th->th_sport;
                        nth->th_dport = th->th_dport;
                }
                xchg(nth->th_dport, nth->th_sport, n_short);
#undef xchg
        }
        if (isipv6) {
                ip6->ip6_flow = 0;
                ip6->ip6_vfc = IPV6_VERSION;
                ip6->ip6_nxt = IPPROTO_TCP;
                ip6->ip6_plen = htons((u_short)(sizeof(struct tcphdr) + tlen));
                tlen += sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
        } else {
                tlen += sizeof(struct tcpiphdr);
                ip->ip_len = tlen;
                ip->ip_ttl = ip_defttl;
        }
        m->m_len = tlen;
        m->m_pkthdr.len = tlen;
        m->m_pkthdr.rcvif = NULL;
        nth->th_seq = htonl(seq);
        nth->th_ack = htonl(ack);
        nth->th_x2 = 0;
        nth->th_off = sizeof(struct tcphdr) >> 2;
        nth->th_flags = flags;
        if (tp != NULL)
                nth->th_win = htons((u_short) (win >> tp->rcv_scale));
        else
                nth->th_win = htons((u_short)win);
        nth->th_urp = 0;
        if (isipv6) {
                nth->th_sum = 0;
                nth->th_sum = in6_cksum(m, IPPROTO_TCP,
                                        sizeof(struct ip6_hdr),
                                        tlen - sizeof(struct ip6_hdr));
                ip6->ip6_hlim = in6_selecthlim(tp ? tp->t_inpcb : NULL,
                                               (ro6 && ro6->ro_rt) ?
                                                ro6->ro_rt->rt_ifp : NULL);
        } else {
                nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
                    htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
                m->m_pkthdr.csum_flags = CSUM_TCP;
                m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
                m->m_pkthdr.csum_thlen = sizeof(struct tcphdr);
        }
#ifdef TCPDEBUG
        if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
                tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
#endif
        if (isipv6) {
                ip6_output(m, NULL, ro6, ipflags, NULL, NULL,
                           tp ? tp->t_inpcb : NULL);
                if ((ro6 == &sro6) && (ro6->ro_rt != NULL)) {
                        RTFREE(ro6->ro_rt);
                        ro6->ro_rt = NULL;
                }
        } else {
                ipflags |= IP_DEBUGROUTE;
                ip_output(m, NULL, ro, ipflags, NULL, tp ? tp->t_inpcb : NULL);
                if ((ro == &sro) && (ro->ro_rt != NULL)) {
                        RTFREE(ro->ro_rt);
                        ro->ro_rt = NULL;
                }
        }
}

/*
 * Create a new TCP control block, making an
 * empty reassembly queue and hooking it to the argument
 * protocol control block.  The `inp' parameter must have
 * come from the zone allocator set up in tcp_init().
 */
struct tcpcb *
tcp_newtcpcb(struct inpcb *inp)
{
        struct inp_tp *it;
        struct tcpcb *tp;
#ifdef INET6
        boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
#else
        const boolean_t isipv6 = FALSE;
#endif

        it = (struct inp_tp *)inp;
        tp = &it->tcb;
        bzero(tp, sizeof(struct tcpcb));
        TAILQ_INIT(&tp->t_segq);
        tp->t_maxseg = tp->t_maxopd = isipv6 ? tcp_v6mssdflt : tcp_mssdflt;
        tp->t_rxtthresh = tcprexmtthresh;

        /* Set up our timeouts. */
        tp->tt_rexmt = &it->inp_tp_rexmt;
        tp->tt_persist = &it->inp_tp_persist;
        tp->tt_keep = &it->inp_tp_keep;
        tp->tt_2msl = &it->inp_tp_2msl;
        tp->tt_delack = &it->inp_tp_delack;
        tcp_inittimers(tp);

        /*
         * Zero out timer message.  We don't create it here,
         * since the current CPU may not be the owner of this
         * inpcb.
         */
        tp->tt_msg = &it->inp_tp_timermsg;
        bzero(tp->tt_msg, sizeof(*tp->tt_msg));

        tp->t_keepinit = tcp_keepinit;
        tp->t_keepidle = tcp_keepidle;
        tp->t_keepintvl = tcp_keepintvl;
        tp->t_keepcnt = tcp_keepcnt;
        tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;

        if (tcp_do_ncr)
                tp->t_flags |= TF_NCR;
        if (tcp_do_rfc1323)
                tp->t_flags |= (TF_REQ_SCALE | TF_REQ_TSTMP);

        tp->t_inpcb = inp;      /* XXX */
        tp->t_state = TCPS_CLOSED;
        /*
         * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
         * rtt estimate.  Set rttvar so that srtt + 4 * rttvar gives
         * reasonable initial retransmit time.
         */
        tp->t_srtt = TCPTV_SRTTBASE;
        tp->t_rttvar =
            ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
        tp->t_rttmin = tcp_rexmit_min;
        tp->t_rxtcur = TCPTV_RTOBASE;
        tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
        tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
        tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
        tp->snd_last = ticks;
        tp->t_rcvtime = ticks;
        /*
         * IPv4 TTL initialization is necessary for an IPv6 socket as well,
         * because the socket may be bound to an IPv6 wildcard address,
         * which may match an IPv4-mapped IPv6 address.
         */
        inp->inp_ip_ttl = ip_defttl;
        inp->inp_ppcb = tp;
        tcp_sack_tcpcb_init(tp);

        tp->tt_sndmore = &it->inp_tp_sndmore;
        tcp_output_init(tp);

        return (tp);            /* XXX */
}

/*
 * Drop a TCP connection, reporting the specified error.
 * If connection is synchronized, then send a RST to peer.
 */
struct tcpcb *
tcp_drop(struct tcpcb *tp, int error)
{
        struct socket *so = tp->t_inpcb->inp_socket;

        if (TCPS_HAVERCVDSYN(tp->t_state)) {
                tp->t_state = TCPS_CLOSED;
                tcp_output(tp);
                tcpstat.tcps_drops++;
        } else
                tcpstat.tcps_conndrops++;
        if (error == ETIMEDOUT && tp->t_softerror)
                error = tp->t_softerror;
        so->so_error = error;
        return (tcp_close(tp));
}

struct netmsg_listen_detach {
        struct netmsg_base      base;
        struct tcpcb            *nm_tp;
        struct tcpcb            *nm_tp_inh;
};

static void
tcp_listen_detach_handler(netmsg_t msg)
{
        struct netmsg_listen_detach *nmsg = (struct netmsg_listen_detach *)msg;
        struct tcpcb *tp = nmsg->nm_tp;
        int cpu = mycpuid, nextcpu;

        if (tp->t_flags & TF_LISTEN)
                syncache_destroy(tp, nmsg->nm_tp_inh);

        in_pcbremwildcardhash_oncpu(tp->t_inpcb, &tcbinfo[cpu]);

        nextcpu = cpu + 1;
        if (nextcpu < ncpus2)
                lwkt_forwardmsg(netisr_cpuport(nextcpu), &nmsg->base.lmsg);
        else
                lwkt_replymsg(&nmsg->base.lmsg, 0);
}

/*
 * Close a TCP control block:
 *      discard all space held by the tcp
 *      discard internet protocol block
 *      wake up any sleepers
 */
struct tcpcb *
tcp_close(struct tcpcb *tp)
{
        struct tseg_qent *q;
        struct inpcb *inp = tp->t_inpcb;
        struct inpcb *inp_inh = NULL;
        struct tcpcb *tp_inh = NULL;
        struct socket *so = inp->inp_socket;
        struct rtentry *rt;
        boolean_t dosavessthresh;
#ifdef INET6
        boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) != 0);
        boolean_t isafinet6 = (INP_CHECK_SOCKAF(so, AF_INET6) != 0);
#else
        const boolean_t isipv6 = FALSE;
#endif

        if (tp->t_flags & TF_LISTEN) {
                /*
                 * Pending socket/syncache inheritance
                 *
                 * If this is a listen(2) socket, find another listen(2)
                 * socket in the same local group, which could inherit
                 * the syncache and sockets pending on the completion
                 * and incompletion queues.
                 *
                 * NOTE:
                 * Currently the inheritance could only happen on the
                 * listen(2) sockets w/ SO_REUSEPORT set.
                 */
                KASSERT(&curthread->td_msgport == netisr_cpuport(0),
                    ("listen socket close not in netisr0"));
                inp_inh = in_pcblocalgroup_last(&tcbinfo[0], inp);
                if (inp_inh != NULL)
                        tp_inh = intotcpcb(inp_inh);
        }

        /*
         * INP_WILDCARD_MP indicates that listen(2) has been called on
         * this socket.  This implies:
         * - A wildcard inp's hash is replicated for each protocol thread.
         * - Syncache for this inp grows independently in each protocol
         *   thread.
         * - There is more than one cpu
         *
         * We have to chain a message to the rest of the protocol threads
         * to cleanup the wildcard hash and the syncache.  The cleanup
         * in the current protocol thread is defered till the end of this
         * function.
         *
         * NOTE:
         * After cleanup the inp's hash and syncache entries, this inp will
         * no longer be available to the rest of the protocol threads, so we
         * are safe to whack the inp in the following code.
         */
        if (inp->inp_flags & INP_WILDCARD_MP) {
                struct netmsg_listen_detach nmsg;

                KKASSERT(so->so_port == netisr_cpuport(0));
                KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
                KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);

                netmsg_init(&nmsg.base, NULL, &curthread->td_msgport,
                            MSGF_PRIORITY, tcp_listen_detach_handler);
                nmsg.nm_tp = tp;
                nmsg.nm_tp_inh = tp_inh;
                lwkt_domsg(netisr_cpuport(1), &nmsg.base.lmsg, 0);

                inp->inp_flags &= ~INP_WILDCARD_MP;
        }

        KKASSERT(tp->t_state != TCPS_TERMINATING);
        tp->t_state = TCPS_TERMINATING;

        /*
         * Make sure that all of our timers are stopped before we
         * delete the PCB.  For listen TCP socket (tp->tt_msg == NULL),
         * timers are never used.  If timer message is never created
         * (tp->tt_msg->tt_tcb == NULL), timers are never used too.
         */
        if (tp->tt_msg != NULL && tp->tt_msg->tt_tcb != NULL) {
                tcp_callout_stop(tp, tp->tt_rexmt);
                tcp_callout_stop(tp, tp->tt_persist);
                tcp_callout_stop(tp, tp->tt_keep);
                tcp_callout_stop(tp, tp->tt_2msl);
                tcp_callout_stop(tp, tp->tt_delack);
        }

        if (tp->t_flags & TF_ONOUTPUTQ) {
                KKASSERT(tp->tt_cpu == mycpu->gd_cpuid);
                TAILQ_REMOVE(&tcpcbackq[tp->tt_cpu], tp, t_outputq);
                tp->t_flags &= ~TF_ONOUTPUTQ;
        }

        /*
         * If we got enough samples through the srtt filter,
         * save the rtt and rttvar in the routing entry.
         * 'Enough' is arbitrarily defined as the 16 samples.
         * 16 samples is enough for the srtt filter to converge
         * to within 5% of the correct value; fewer samples and
         * we could save a very bogus rtt.
         *
         * Don't update the default route's characteristics and don't
         * update anything that the user "locked".
         */
        if (tp->t_rttupdated >= 16) {
                u_long i = 0;

                if (isipv6) {
                        struct sockaddr_in6 *sin6;

                        if ((rt = inp->in6p_route.ro_rt) == NULL)
                                goto no_valid_rt;
                        sin6 = (struct sockaddr_in6 *)rt_key(rt);
                        if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
                                goto no_valid_rt;
                } else
                        if ((rt = inp->inp_route.ro_rt) == NULL ||
                            ((struct sockaddr_in *)rt_key(rt))->
                             sin_addr.s_addr == INADDR_ANY)
                                goto no_valid_rt;

                if (!(rt->rt_rmx.rmx_locks & RTV_RTT)) {
                        i = tp->t_srtt * (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
                        if (rt->rt_rmx.rmx_rtt && i)
                                /*
                                 * filter this update to half the old & half
                                 * the new values, converting scale.
                                 * See route.h and tcp_var.h for a
                                 * description of the scaling constants.
                                 */
                                rt->rt_rmx.rmx_rtt =
                                    (rt->rt_rmx.rmx_rtt + i) / 2;
                        else
                                rt->rt_rmx.rmx_rtt = i;
                        tcpstat.tcps_cachedrtt++;
                }
                if (!(rt->rt_rmx.rmx_locks & RTV_RTTVAR)) {
                        i = tp->t_rttvar *
                            (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
                        if (rt->rt_rmx.rmx_rttvar && i)
                                rt->rt_rmx.rmx_rttvar =
                                    (rt->rt_rmx.rmx_rttvar + i) / 2;
                        else
                                rt->rt_rmx.rmx_rttvar = i;
                        tcpstat.tcps_cachedrttvar++;
                }
                /*
                 * The old comment here said:
                 * update the pipelimit (ssthresh) if it has been updated
                 * already or if a pipesize was specified & the threshhold
                 * got below half the pipesize.  I.e., wait for bad news
                 * before we start updating, then update on both good
                 * and bad news.
                 *
                 * But we want to save the ssthresh even if no pipesize is
                 * specified explicitly in the route, because such
                 * connections still have an implicit pipesize specified
                 * by the global tcp_sendspace.  In the absence of a reliable
                 * way to calculate the pipesize, it will have to do.
                 */
                i = tp->snd_ssthresh;
                if (rt->rt_rmx.rmx_sendpipe != 0)
                        dosavessthresh = (i < rt->rt_rmx.rmx_sendpipe/2);
                else
                        dosavessthresh = (i < so->so_snd.ssb_hiwat/2);
                if (dosavessthresh ||
                    (!(rt->rt_rmx.rmx_locks & RTV_SSTHRESH) && (i != 0) &&
                     (rt->rt_rmx.rmx_ssthresh != 0))) {
                        /*
                         * convert the limit from user data bytes to
                         * packets then to packet data bytes.
                         */
                        i = (i + tp->t_maxseg / 2) / tp->t_maxseg;
                        if (i < 2)
                                i = 2;
                        i *= tp->t_maxseg +
                             (isipv6 ?
                              sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
                              sizeof(struct tcpiphdr));
                        if (rt->rt_rmx.rmx_ssthresh)
                                rt->rt_rmx.rmx_ssthresh =
                                    (rt->rt_rmx.rmx_ssthresh + i) / 2;
                        else
                                rt->rt_rmx.rmx_ssthresh = i;
                        tcpstat.tcps_cachedssthresh++;
                }
        }

no_valid_rt:
        /* free the reassembly queue, if any */
        while((q = TAILQ_FIRST(&tp->t_segq)) != NULL) {
                TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
                m_freem(q->tqe_m);
                kfree(q, M_TSEGQ);
                atomic_add_int(&tcp_reass_qsize, -1);
        }
        /* throw away SACK blocks in scoreboard*/
        if (TCP_DO_SACK(tp))
                tcp_sack_destroy(&tp->scb);

        inp->inp_ppcb = NULL;
        soisdisconnected(so);
        /* note: pcb detached later on */

        tcp_destroy_timermsg(tp);
        tcp_output_cancel(tp);

        if (tp->t_flags & TF_LISTEN) {
                syncache_destroy(tp, tp_inh);
                if (inp_inh != NULL && inp_inh->inp_socket != NULL) {
                        /*
                         * Pending sockets inheritance only needs
                         * to be done once in the current thread,
                         * i.e. netisr0.
                         */
                        soinherit(so, inp_inh->inp_socket);
                }
        }

        so_async_rcvd_drop(so);
        /* Drop the reference for the asynchronized pru_rcvd */
        sofree(so);

        /*
         * NOTE:
         * pcbdetach removes any wildcard hash entry on the current CPU.
         */
#ifdef INET6
        if (isafinet6)
                in6_pcbdetach(inp);
        else
#endif
                in_pcbdetach(inp);

        tcpstat.tcps_closed++;
        return (NULL);
}

static __inline void
tcp_drain_oncpu(struct inpcbhead *head)
{
        struct inpcb *marker;
        struct inpcb *inpb;
        struct tcpcb *tcpb;
        struct tseg_qent *te;

        /*
         * Allows us to block while running the list
         */
        marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
        marker->inp_flags |= INP_PLACEMARKER;
        LIST_INSERT_HEAD(head, marker, inp_list);

        while ((inpb = LIST_NEXT(marker, inp_list)) != NULL) {
                if ((inpb->inp_flags & INP_PLACEMARKER) == 0 &&
                    (tcpb = intotcpcb(inpb)) != NULL &&
                    (te = TAILQ_FIRST(&tcpb->t_segq)) != NULL) {
                        TAILQ_REMOVE(&tcpb->t_segq, te, tqe_q);
                        if (te->tqe_th->th_flags & TH_FIN)
                                tcpb->t_flags &= ~TF_QUEDFIN;
                        m_freem(te->tqe_m);
                        kfree(te, M_TSEGQ);
                        atomic_add_int(&tcp_reass_qsize, -1);
                        /* retry */
                } else {
                        LIST_REMOVE(marker, inp_list);
                        LIST_INSERT_AFTER(inpb, marker, inp_list);
                }
        }
        LIST_REMOVE(marker, inp_list);
        kfree(marker, M_TEMP);
}

struct netmsg_tcp_drain {
        struct netmsg_base      base;
        struct inpcbhead        *nm_head;
};

static void
tcp_drain_handler(netmsg_t msg)
{
        struct netmsg_tcp_drain *nm = (void *)msg;

        tcp_drain_oncpu(nm->nm_head);
        lwkt_replymsg(&nm->base.lmsg, 0);
}

void
tcp_drain(void)
{
        int cpu;

        if (!do_tcpdrain)
                return;

        /*
         * Walk the tcpbs, if existing, and flush the reassembly queue,
         * if there is one...
         * XXX: The "Net/3" implementation doesn't imply that the TCP
         *      reassembly queue should be flushed, but in a situation
         *      where we're really low on mbufs, this is potentially
         *      useful.
         */
        for (cpu = 0; cpu < ncpus2; cpu++) {
                struct netmsg_tcp_drain *nm;

                if (cpu == mycpu->gd_cpuid) {
                        tcp_drain_oncpu(&tcbinfo[cpu].pcblisthead);
                } else {
                        nm = kmalloc(sizeof(struct netmsg_tcp_drain),
                                     M_LWKTMSG, M_NOWAIT);
                        if (nm == NULL)
                                continue;
                        netmsg_init(&nm->base, NULL, &netisr_afree_rport,
                                    0, tcp_drain_handler);
                        nm->nm_head = &tcbinfo[cpu].pcblisthead;
                        lwkt_sendmsg(netisr_cpuport(cpu), &nm->base.lmsg);
                }
        }
}

/*
 * Notify a tcp user of an asynchronous error;
 * store error as soft error, but wake up user
 * (for now, won't do anything until can select for soft error).
 *
 * Do not wake up user since there currently is no mechanism for
 * reporting soft errors (yet - a kqueue filter may be added).
 */
static void
tcp_notify(struct inpcb *inp, int error)
{
        struct tcpcb *tp = intotcpcb(inp);

        /*
         * Ignore some errors if we are hooked up.
         * If connection hasn't completed, has retransmitted several times,
         * and receives a second error, give up now.  This is better
         * than waiting a long time to establish a connection that
         * can never complete.
         */
        if (tp->t_state == TCPS_ESTABLISHED &&
             (error == EHOSTUNREACH || error == ENETUNREACH ||
              error == EHOSTDOWN)) {
                return;
        } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
            tp->t_softerror)
                tcp_drop(tp, error);
        else
                tp->t_softerror = error;
#if 0
        wakeup(&so->so_timeo);
        sorwakeup(so);
        sowwakeup(so);
#endif
}

static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)
{
        int error, i, n;
        struct inpcb *marker;
        struct inpcb *inp;
        globaldata_t gd;
        int origcpu, ccpu;

        error = 0;
        n = 0;

        /*
         * The process of preparing the TCB list is too time-consuming and
         * resource-intensive to repeat twice on every request.
         */
        if (req->oldptr == NULL) {
                for (ccpu = 0; ccpu < ncpus2; ++ccpu) {
                        gd = globaldata_find(ccpu);
                        n += tcbinfo[gd->gd_cpuid].ipi_count;
                }
                req->oldidx = (n + n/8 + 10) * sizeof(struct xtcpcb);
                return (0);
        }

        if (req->newptr != NULL)
                return (EPERM);

        marker = kmalloc(sizeof(struct inpcb), M_TEMP, M_WAITOK|M_ZERO);
        marker->inp_flags |= INP_PLACEMARKER;

        /*
         * OK, now we're committed to doing something.  Run the inpcb list
         * for each cpu in the system and construct the output.  Use a
         * list placemarker to deal with list changes occuring during
         * copyout blockages (but otherwise depend on being on the correct
         * cpu to avoid races).
         */
        origcpu = mycpu->gd_cpuid;
        for (ccpu = 0; ccpu < ncpus2 && error == 0; ++ccpu) {
                caddr_t inp_ppcb;
                struct xtcpcb xt;

                lwkt_migratecpu(ccpu);

                n = tcbinfo[ccpu].ipi_count;

                LIST_INSERT_HEAD(&tcbinfo[ccpu].pcblisthead, marker, inp_list);
                i = 0;
                while ((inp = LIST_NEXT(marker, inp_list)) != NULL && i < n) {
                        /*
                         * process a snapshot of pcbs, ignoring placemarkers
                         * and using our own to allow SYSCTL_OUT to block.
                         */
                        LIST_REMOVE(marker, inp_list);
                        LIST_INSERT_AFTER(inp, marker, inp_list);

                        if (inp->inp_flags & INP_PLACEMARKER)
                                continue;
                        if (prison_xinpcb(req->td, inp))
                                continue;

                        xt.xt_len = sizeof xt;
                        bcopy(inp, &xt.xt_inp, sizeof *inp);
                        inp_ppcb = inp->inp_ppcb;
                        if (inp_ppcb != NULL)
                                bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
                        else
                                bzero(&xt.xt_tp, sizeof xt.xt_tp);
                        if (inp->inp_socket)
                                sotoxsocket(inp->inp_socket, &xt.xt_socket);
                        if ((error = SYSCTL_OUT(req, &xt, sizeof xt)) != 0)
                                break;
                        ++i;
                }
                LIST_REMOVE(marker, inp_list);
                if (error == 0 && i < n) {
                        bzero(&xt, sizeof xt);
                        xt.xt_len = sizeof xt;
                        while (i < n) {
                                error = SYSCTL_OUT(req, &xt, sizeof xt);
                                if (error)
                                        break;
                                ++i;
                        }
                }
        }

        /*
         * Make sure we are on the same cpu we were on originally, since
         * higher level callers expect this.  Also don't pollute caches with
         * migrated userland data by (eventually) returning to userland
         * on a different cpu.
         */
        lwkt_migratecpu(origcpu);
        kfree(marker, M_TEMP);
        return (error);
}

SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
            tcp_pcblist, "S,xtcpcb", "List of active TCP connections");

static int
tcp_getcred(SYSCTL_HANDLER_ARGS)
{
        struct sockaddr_in addrs[2];
        struct inpcb *inp;
        int cpu;
        int error;

        error = priv_check(req->td, PRIV_ROOT);
        if (error != 0)
                return (error);
        error = SYSCTL_IN(req, addrs, sizeof addrs);
        if (error != 0)
                return (error);
        crit_enter();
        cpu = tcp_addrcpu(addrs[1].sin_addr.s_addr, addrs[1].sin_port,
            addrs[0].sin_addr.s_addr, addrs[0].sin_port);
        inp = in_pcblookup_hash(&tcbinfo[cpu], addrs[1].sin_addr,
            addrs[1].sin_port, addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
        if (inp == NULL || inp->inp_socket == NULL) {
                error = ENOENT;
                goto out;
        }
        error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
out:
        crit_exit();
        return (error);
}

SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
    0, 0, tcp_getcred, "S,ucred", "Get the ucred of a TCP connection");

#ifdef INET6
static int
tcp6_getcred(SYSCTL_HANDLER_ARGS)
{
        struct sockaddr_in6 addrs[2];
        struct inpcb *inp;
        int error;
        boolean_t mapped = FALSE;

        error = priv_check(req->td, PRIV_ROOT);
        if (error != 0)
                return (error);
        error = SYSCTL_IN(req, addrs, sizeof addrs);
        if (error != 0)
                return (error);
        if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
                if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
                        mapped = TRUE;
                else
                        return (EINVAL);
        }
        crit_enter();
        if (mapped) {
                inp = in_pcblookup_hash(&tcbinfo[0],
                    *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
                    addrs[1].sin6_port,
                    *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
                    addrs[0].sin6_port,
                    0, NULL);
        } else {
                inp = in6_pcblookup_hash(&tcbinfo[0],
                    &addrs[1].sin6_addr, addrs[1].sin6_port,
                    &addrs[0].sin6_addr, addrs[0].sin6_port,
                    0, NULL);
        }
        if (inp == NULL || inp->inp_socket == NULL) {
                error = ENOENT;
                goto out;
        }
        error = SYSCTL_OUT(req, inp->inp_socket->so_cred, sizeof(struct ucred));
out:
        crit_exit();
        return (error);
}

SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred, (CTLTYPE_OPAQUE | CTLFLAG_RW),
            0, 0,
            tcp6_getcred, "S,ucred", "Get the ucred of a TCP6 connection");
#endif

struct netmsg_tcp_notify {
        struct netmsg_base base;
        void            (*nm_notify)(struct inpcb *, int);
        struct in_addr  nm_faddr;
        int             nm_arg;
};

static void
tcp_notifyall_oncpu(netmsg_t msg)
{
        struct netmsg_tcp_notify *nm = (struct netmsg_tcp_notify *)msg;
        int nextcpu;

        in_pcbnotifyall(&tcbinfo[mycpuid].pcblisthead, nm->nm_faddr,
                        nm->nm_arg, nm->nm_notify);

        nextcpu = mycpuid + 1;
        if (nextcpu < ncpus2)
                lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
        else
                lwkt_replymsg(&nm->base.lmsg, 0);
}

void
tcp_ctlinput(netmsg_t msg)
{
        int cmd = msg->ctlinput.nm_cmd;
        struct sockaddr *sa = msg->ctlinput.nm_arg;
        struct ip *ip = msg->ctlinput.nm_extra;
        struct tcphdr *th;
        struct in_addr faddr;
        struct inpcb *inp;
        struct tcpcb *tp;
        void (*notify)(struct inpcb *, int) = tcp_notify;
        tcp_seq icmpseq;
        int arg, cpu;

        if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0) {
                goto done;
        }

        faddr = ((struct sockaddr_in *)sa)->sin_addr;
        if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
                goto done;

        arg = inetctlerrmap[cmd];
        if (cmd == PRC_QUENCH) {
                notify = tcp_quench;
        } else if (icmp_may_rst &&
                   (cmd == PRC_UNREACH_ADMIN_PROHIB ||
                    cmd == PRC_UNREACH_PORT ||
                    cmd == PRC_TIMXCEED_INTRANS) &&
                   ip != NULL) {
                notify = tcp_drop_syn_sent;
        } else if (cmd == PRC_MSGSIZE) {
                struct icmp *icmp = (struct icmp *)
                    ((caddr_t)ip - offsetof(struct icmp, icmp_ip));

                arg = ntohs(icmp->icmp_nextmtu);
                notify = tcp_mtudisc;
        } else if (PRC_IS_REDIRECT(cmd)) {
                ip = NULL;
                notify = in_rtchange;
        } else if (cmd == PRC_HOSTDEAD) {
                ip = NULL;
        }

        if (ip != NULL) {
                crit_enter();
                th = (struct tcphdr *)((caddr_t)ip +
                                       (IP_VHL_HL(ip->ip_vhl) << 2));
                cpu = tcp_addrcpu(faddr.s_addr, th->th_dport,
                                  ip->ip_src.s_addr, th->th_sport);
                inp = in_pcblookup_hash(&tcbinfo[cpu], faddr, th->th_dport,
                                        ip->ip_src, th->th_sport, 0, NULL);
                if ((inp != NULL) && (inp->inp_socket != NULL)) {
                        icmpseq = htonl(th->th_seq);
                        tp = intotcpcb(inp);
                        if (SEQ_GEQ(icmpseq, tp->snd_una) &&
                            SEQ_LT(icmpseq, tp->snd_max))
                                (*notify)(inp, arg);
                } else {
                        struct in_conninfo inc;

                        inc.inc_fport = th->th_dport;
                        inc.inc_lport = th->th_sport;
                        inc.inc_faddr = faddr;
                        inc.inc_laddr = ip->ip_src;
#ifdef INET6
                        inc.inc_isipv6 = 0;
#endif
                        syncache_unreach(&inc, th);
                }
                crit_exit();
        } else {
                struct netmsg_tcp_notify *nm;

                KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
                nm = kmalloc(sizeof(*nm), M_LWKTMSG, M_INTWAIT);
                netmsg_init(&nm->base, NULL, &netisr_afree_rport,
                            0, tcp_notifyall_oncpu);
                nm->nm_faddr = faddr;
                nm->nm_arg = arg;
                nm->nm_notify = notify;

                lwkt_sendmsg(netisr_cpuport(0), &nm->base.lmsg);
        }
done:
        lwkt_replymsg(&msg->lmsg, 0);
}

#ifdef INET6

void
tcp6_ctlinput(netmsg_t msg)
{
        int cmd = msg->ctlinput.nm_cmd;
        struct sockaddr *sa = msg->ctlinput.nm_arg;
        void *d = msg->ctlinput.nm_extra;
        struct tcphdr th;
        void (*notify) (struct inpcb *, int) = tcp_notify;
        struct ip6_hdr *ip6;
        struct mbuf *m;
        struct ip6ctlparam *ip6cp = NULL;
        const struct sockaddr_in6 *sa6_src = NULL;
        int off;
        struct tcp_portonly {
                u_int16_t th_sport;
                u_int16_t th_dport;
        } *thp;
        int arg;

        if (sa->sa_family != AF_INET6 ||
            sa->sa_len != sizeof(struct sockaddr_in6)) {
                goto out;
        }

        arg = 0;
        if (cmd == PRC_QUENCH)
                notify = tcp_quench;
        else if (cmd == PRC_MSGSIZE) {
                struct ip6ctlparam *ip6cp = d;
                struct icmp6_hdr *icmp6 = ip6cp->ip6c_icmp6;

                arg = ntohl(icmp6->icmp6_mtu);
                notify = tcp_mtudisc;
        } else if (!PRC_IS_REDIRECT(cmd) &&
                 ((unsigned)cmd > PRC_NCMDS || inet6ctlerrmap[cmd] == 0)) {
                goto out;
        }

        /* if the parameter is from icmp6, decode it. */
        if (d != NULL) {
                ip6cp = (struct ip6ctlparam *)d;
                m = ip6cp->ip6c_m;
                ip6 = ip6cp->ip6c_ip6;
                off = ip6cp->ip6c_off;
                sa6_src = ip6cp->ip6c_src;
        } else {
                m = NULL;
                ip6 = NULL;
                off = 0;        /* fool gcc */
                sa6_src = &sa6_any;
        }

        if (ip6 != NULL) {
                struct in_conninfo inc;
                /*
                 * XXX: We assume that when IPV6 is non NULL,
                 * M and OFF are valid.
                 */

                /* check if we can safely examine src and dst ports */
                if (m->m_pkthdr.len < off + sizeof *thp)
                        goto out;

                bzero(&th, sizeof th);
                m_copydata(m, off, sizeof *thp, (caddr_t)&th);

                in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, th.th_dport,
                    (struct sockaddr *)ip6cp->ip6c_src,
                    th.th_sport, cmd, arg, notify);

                inc.inc_fport = th.th_dport;
                inc.inc_lport = th.th_sport;
                inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
                inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
                inc.inc_isipv6 = 1;
                syncache_unreach(&inc, &th);
        } else {
                in6_pcbnotify(&tcbinfo[0].pcblisthead, sa, 0,
                    (const struct sockaddr *)sa6_src, 0, cmd, arg, notify);
        }
out:
        lwkt_replymsg(&msg->ctlinput.base.lmsg, 0);
}

#endif

/*
 * Following is where TCP initial sequence number generation occurs.
 *
 * There are two places where we must use initial sequence numbers:
 * 1.  In SYN-ACK packets.
 * 2.  In SYN packets.
 *
 * All ISNs for SYN-ACK packets are generated by the syncache.  See
 * tcp_syncache.c for details.
 *
 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
 * depends on this property.  In addition, these ISNs should be
 * unguessable so as to prevent connection hijacking.  To satisfy
 * the requirements of this situation, the algorithm outlined in
 * RFC 1948 is used to generate sequence numbers.
 *
 * Implementation details:
 *
 * Time is based off the system timer, and is corrected so that it
 * increases by one megabyte per second.  This allows for proper
 * recycling on high speed LANs while still leaving over an hour
 * before rollover.
 *
 * net.inet.tcp.isn_reseed_interval controls the number of seconds
 * between seeding of isn_secret.  This is normally set to zero,
 * as reseeding should not be necessary.
 *
 */

#define ISN_BYTES_PER_SECOND 1048576

u_char isn_secret[32];
int isn_last_reseed;
MD5_CTX isn_ctx;

tcp_seq
tcp_new_isn(struct tcpcb *tp)
{
        u_int32_t md5_buffer[4];
        tcp_seq new_isn;

        /* Seed if this is the first use, reseed if requested. */
        if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
             (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
                < (u_int)ticks))) {
                read_random_unlimited(&isn_secret, sizeof isn_secret);
                isn_last_reseed = ticks;
        }

        /* Compute the md5 hash and return the ISN. */
        MD5Init(&isn_ctx);
        MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_fport, sizeof(u_short));
        MD5Update(&isn_ctx, (u_char *)&tp->t_inpcb->inp_lport, sizeof(u_short));
#ifdef INET6
        if (tp->t_inpcb->inp_vflag & INP_IPV6) {
                MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
                          sizeof(struct in6_addr));
                MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
                          sizeof(struct in6_addr));
        } else
#endif
        {
                MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
                          sizeof(struct in_addr));
                MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
                          sizeof(struct in_addr));
        }
        MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
        MD5Final((u_char *) &md5_buffer, &isn_ctx);
        new_isn = (tcp_seq) md5_buffer[0];
        new_isn += ticks * (ISN_BYTES_PER_SECOND / hz);
        return (new_isn);
}

/*
 * When a source quench is received, close congestion window
 * to one segment.  We will gradually open it again as we proceed.
 */
void
tcp_quench(struct inpcb *inp, int error)
{
        struct tcpcb *tp = intotcpcb(inp);

        if (tp != NULL) {
                tp->snd_cwnd = tp->t_maxseg;
                tp->snd_wacked = 0;
        }
}

/*
 * When a specific ICMP unreachable message is received and the
 * connection state is SYN-SENT, drop the connection.  This behavior
 * is controlled by the icmp_may_rst sysctl.
 */
void
tcp_drop_syn_sent(struct inpcb *inp, int error)
{
        struct tcpcb *tp = intotcpcb(inp);

        if ((tp != NULL) && (tp->t_state == TCPS_SYN_SENT))
                tcp_drop(tp, error);
}

/*
 * When a `need fragmentation' ICMP is received, update our idea of the MSS
 * based on the new value in the route.  Also nudge TCP to send something,
 * since we know the packet we just sent was dropped.
 * This duplicates some code in the tcp_mss() function in tcp_input.c.
 */
void
tcp_mtudisc(struct inpcb *inp, int mtu)
{
        struct tcpcb *tp = intotcpcb(inp);
        struct rtentry *rt;
        struct socket *so = inp->inp_socket;
        int maxopd, mss;
#ifdef INET6
        boolean_t isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0);
#else
        const boolean_t isipv6 = FALSE;
#endif

        if (tp == NULL)
                return;

        /*
         * If no MTU is provided in the ICMP message, use the
         * next lower likely value, as specified in RFC 1191.
         */
        if (mtu == 0) {
                int oldmtu;

                oldmtu = tp->t_maxopd + 
                    (isipv6 ?
                     sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
                     sizeof(struct tcpiphdr));
                mtu = ip_next_mtu(oldmtu, 0);
        }

        if (isipv6)
                rt = tcp_rtlookup6(&inp->inp_inc);
        else
                rt = tcp_rtlookup(&inp->inp_inc);
        if (rt != NULL) {
                if (rt->rt_rmx.rmx_mtu != 0 && rt->rt_rmx.rmx_mtu < mtu)
                        mtu = rt->rt_rmx.rmx_mtu;

                maxopd = mtu -
                    (isipv6 ?
                     sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
                     sizeof(struct tcpiphdr));

                /*
                 * XXX - The following conditional probably violates the TCP
                 * spec.  The problem is that, since we don't know the
                 * other end's MSS, we are supposed to use a conservative
                 * default.  But, if we do that, then MTU discovery will
                 * never actually take place, because the conservative
                 * default is much less than the MTUs typically seen
                 * on the Internet today.  For the moment, we'll sweep
                 * this under the carpet.
                 *
                 * The conservative default might not actually be a problem
                 * if the only case this occurs is when sending an initial
                 * SYN with options and data to a host we've never talked
                 * to before.  Then, they will reply with an MSS value which
                 * will get recorded and the new parameters should get
                 * recomputed.  For Further Study.
                 */
                if (rt->rt_rmx.rmx_mssopt  && rt->rt_rmx.rmx_mssopt < maxopd)
                        maxopd = rt->rt_rmx.rmx_mssopt;
        } else
                maxopd = mtu -
                    (isipv6 ?
                     sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
                     sizeof(struct tcpiphdr));

        if (tp->t_maxopd <= maxopd)
                return;
        tp->t_maxopd = maxopd;

        mss = maxopd;
        if ((tp->t_flags & (TF_REQ_TSTMP | TF_RCVD_TSTMP | TF_NOOPT)) ==
                           (TF_REQ_TSTMP | TF_RCVD_TSTMP))
                mss -= TCPOLEN_TSTAMP_APPA;

        /* round down to multiple of MCLBYTES */
#if     (MCLBYTES & (MCLBYTES - 1)) == 0    /* test if MCLBYTES power of 2 */
        if (mss > MCLBYTES)
                mss &= ~(MCLBYTES - 1); 
#else
        if (mss > MCLBYTES)
                mss = (mss / MCLBYTES) * MCLBYTES;
#endif

        if (so->so_snd.ssb_hiwat < mss)
                mss = so->so_snd.ssb_hiwat;

        tp->t_maxseg = mss;
        tp->t_rtttime = 0;
        tp->snd_nxt = tp->snd_una;
        tcp_output(tp);
        tcpstat.tcps_mturesent++;
}

/*
 * Look-up the routing entry to the peer of this inpcb.  If no route
 * is found and it cannot be allocated the return NULL.  This routine
 * is called by TCP routines that access the rmx structure and by tcp_mss
 * to get the interface MTU.
 */
struct rtentry *
tcp_rtlookup(struct in_conninfo *inc)
{
        struct route *ro = &inc->inc_route;

        if (ro->ro_rt == NULL || !(ro->ro_rt->rt_flags & RTF_UP)) {
                /* No route yet, so try to acquire one */
                if (inc->inc_faddr.s_addr != INADDR_ANY) {
                        /*
                         * unused portions of the structure MUST be zero'd
                         * out because rtalloc() treats it as opaque data
                         */
                        bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
                        ro->ro_dst.sa_family = AF_INET;
                        ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
                        ((struct sockaddr_in *) &ro->ro_dst)->sin_addr =
                            inc->inc_faddr;
                        rtalloc(ro);
                }
        }
        return (ro->ro_rt);
}

#ifdef INET6
struct rtentry *
tcp_rtlookup6(struct in_conninfo *inc)
{
        struct route_in6 *ro6 = &inc->inc6_route;

        if (ro6->ro_rt == NULL || !(ro6->ro_rt->rt_flags & RTF_UP)) {
                /* No route yet, so try to acquire one */
                if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
                        /*
                         * unused portions of the structure MUST be zero'd
                         * out because rtalloc() treats it as opaque data
                         */
                        bzero(&ro6->ro_dst, sizeof(struct sockaddr_in6));
                        ro6->ro_dst.sin6_family = AF_INET6;
                        ro6->ro_dst.sin6_len = sizeof(struct sockaddr_in6);
                        ro6->ro_dst.sin6_addr = inc->inc6_faddr;
                        rtalloc((struct route *)ro6);
                }
        }
        return (ro6->ro_rt);
}
#endif

#ifdef IPSEC
/* compute ESP/AH header size for TCP, including outer IP header. */
size_t
ipsec_hdrsiz_tcp(struct tcpcb *tp)
{
        struct inpcb *inp;
        struct mbuf *m;
        size_t hdrsiz;
        struct ip *ip;
        struct tcphdr *th;

        if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
                return (0);
        MGETHDR(m, MB_DONTWAIT, MT_DATA);
        if (!m)
                return (0);

#ifdef INET6
        if (inp->inp_vflag & INP_IPV6) {
                struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);

                th = (struct tcphdr *)(ip6 + 1);
                m->m_pkthdr.len = m->m_len =
                    sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
                tcp_fillheaders(tp, ip6, th, FALSE);
                hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
        } else
#endif
        {
                ip = mtod(m, struct ip *);
                th = (struct tcphdr *)(ip + 1);
                m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
                tcp_fillheaders(tp, ip, th, FALSE);
                hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
        }

        m_free(m);
        return (hdrsiz);
}
#endif

/*
 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
 *
 * This code attempts to calculate the bandwidth-delay product as a
 * means of determining the optimal window size to maximize bandwidth,
 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
 * routers.  This code also does a fairly good job keeping RTTs in check
 * across slow links like modems.  We implement an algorithm which is very
 * similar (but not meant to be) TCP/Vegas.  The code operates on the
 * transmitter side of a TCP connection and so only effects the transmit
 * side of the connection.
 *
 * BACKGROUND:  TCP makes no provision for the management of buffer space
 * at the end points or at the intermediate routers and switches.  A TCP
 * stream, whether using NewReno or not, will eventually buffer as
 * many packets as it is able and the only reason this typically works is
 * due to the fairly small default buffers made available for a connection
 * (typicaly 16K or 32K).  As machines use larger windows and/or window
 * scaling it is now fairly easy for even a single TCP connection to blow-out
 * all available buffer space not only on the local interface, but on
 * intermediate routers and switches as well.  NewReno makes a misguided
 * attempt to 'solve' this problem by waiting for an actual failure to occur,
 * then backing off, then steadily increasing the window again until another
 * failure occurs, ad-infinitum.  This results in terrible oscillation that
 * is only made worse as network loads increase and the idea of intentionally
 * blowing out network buffers is, frankly, a terrible way to manage network
 * resources.
 *
 * It is far better to limit the transmit window prior to the failure
 * condition being achieved.  There are two general ways to do this:  First
 * you can 'scan' through different transmit window sizes and locate the
 * point where the RTT stops increasing, indicating that you have filled the
 * pipe, then scan backwards until you note that RTT stops decreasing, then
 * repeat ad-infinitum.  This method works in principle but has severe
 * implementation issues due to RTT variances, timer granularity, and
 * instability in the algorithm which can lead to many false positives and
 * create oscillations as well as interact badly with other TCP streams
 * implementing the same algorithm.
 *
 * The second method is to limit the window to the bandwidth delay product
 * of the link.  This is the method we implement.  RTT variances and our
 * own manipulation of the congestion window, bwnd, can potentially
 * destabilize the algorithm.  For this reason we have to stabilize the
 * elements used to calculate the window.  We do this by using the minimum
 * observed RTT, the long term average of the observed bandwidth, and
 * by adding two segments worth of slop.  It isn't perfect but it is able
 * to react to changing conditions and gives us a very stable basis on
 * which to extend the algorithm.
 */
void
tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
{
        u_long bw;
        u_long ibw;
        u_long bwnd;
        int save_ticks;
        int delta_ticks;

        /*
         * If inflight_enable is disabled in the middle of a tcp connection,
         * make sure snd_bwnd is effectively disabled.
         */
        if (!tcp_inflight_enable) {
                tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
                tp->snd_bandwidth = 0;
                return;
        }

        /*
         * Validate the delta time.  If a connection is new or has been idle
         * a long time we have to reset the bandwidth calculator.
         */
        save_ticks = ticks;
        cpu_ccfence();
        delta_ticks = save_ticks - tp->t_bw_rtttime;
        if (tp->t_bw_rtttime == 0 || delta_ticks < 0 || delta_ticks > hz * 10) {
                tp->t_bw_rtttime = ticks;
                tp->t_bw_rtseq = ack_seq;
                if (tp->snd_bandwidth == 0)
                        tp->snd_bandwidth = tcp_inflight_min;
                return;
        }

        /*
         * A delta of at least 1 tick is required.  Waiting 2 ticks will
         * result in better (bw) accuracy.  More than that and the ramp-up
         * will be too slow.
         */
        if (delta_ticks == 0 || delta_ticks == 1)
                return;

        /*
         * Sanity check, plus ignore pure window update acks.
         */
        if ((int)(ack_seq - tp->t_bw_rtseq) <= 0)
                return;

        /*
         * Figure out the bandwidth.  Due to the tick granularity this
         * is a very rough number and it MUST be averaged over a fairly
         * long period of time.  XXX we need to take into account a link
         * that is not using all available bandwidth, but for now our
         * slop will ramp us up if this case occurs and the bandwidth later
         * increases.
         */
        ibw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz / delta_ticks;
        tp->t_bw_rtttime = save_ticks;
        tp->t_bw_rtseq = ack_seq;
        bw = ((int64_t)tp->snd_bandwidth * 15 + ibw) >> 4;

        tp->snd_bandwidth = bw;

        /*
         * Calculate the semi-static bandwidth delay product, plus two maximal
         * segments.  The additional slop puts us squarely in the sweet
         * spot and also handles the bandwidth run-up case.  Without the
         * slop we could be locking ourselves into a lower bandwidth.
         *
         * At very high speeds the bw calculation can become overly sensitive
         * and error prone when delta_ticks is low (e.g. usually 1).  To deal
         * with the problem the stab must be scaled to the bw.  A stab of 50
         * (the default) increases the bw for the purposes of the bwnd
         * calculation by 5%.
         *
         * Situations Handled:
         *      (1) Prevents over-queueing of packets on LANs, especially on
         *          high speed LANs, allowing larger TCP buffers to be
         *          specified, and also does a good job preventing
         *          over-queueing of packets over choke points like modems
         *          (at least for the transmit side).
         *
         *      (2) Is able to handle changing network loads (bandwidth
         *          drops so bwnd drops, bandwidth increases so bwnd
         *          increases).
         *
         *      (3) Theoretically should stabilize in the face of multiple
         *          connections implementing the same algorithm (this may need
         *          a little work).
         *
         *      (4) Stability value (defaults to 20 = 2 maximal packets) can
         *          be adjusted with a sysctl but typically only needs to be on
         *          very slow connections.  A value no smaller then 5 should
         *          be used, but only reduce this default if you have no other
         *          choice.
         */

#define USERTT  (((tp->t_srtt + tp->t_rttbest) / 2) + tcp_inflight_adjrtt)
        bw += bw * tcp_inflight_stab / 1000;
        bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) +
               (int)tp->t_maxseg * 2;
#undef USERTT

        if (tcp_inflight_debug > 0) {
                static int ltime;
                if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
                        ltime = ticks;
                        kprintf("%p ibw %ld bw %ld rttbest %d srtt %d "
                                "bwnd %ld delta %d snd_win %ld\n",
                                tp, ibw, bw, tp->t_rttbest, tp->t_srtt,
                                bwnd, delta_ticks, tp->snd_wnd);
                }
        }
        if ((long)bwnd < tcp_inflight_min)
                bwnd = tcp_inflight_min;
        if (bwnd > tcp_inflight_max)
                bwnd = tcp_inflight_max;
        if ((long)bwnd < tp->t_maxseg * 2)
                bwnd = tp->t_maxseg * 2;
        tp->snd_bwnd = bwnd;
}

static void
tcp_rmx_iwsegs(struct tcpcb *tp, u_long *maxsegs, u_long *capsegs)
{
        struct rtentry *rt;
        struct inpcb *inp = tp->t_inpcb;
#ifdef INET6
        boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
#else
        const boolean_t isipv6 = FALSE;
#endif

        /* XXX */
        if (tcp_iw_maxsegs < TCP_IW_MAXSEGS_DFLT)
                tcp_iw_maxsegs = TCP_IW_MAXSEGS_DFLT;
        if (tcp_iw_capsegs < TCP_IW_CAPSEGS_DFLT)
                tcp_iw_capsegs = TCP_IW_CAPSEGS_DFLT;

        if (isipv6)
                rt = tcp_rtlookup6(&inp->inp_inc);
        else
                rt = tcp_rtlookup(&inp->inp_inc);
        if (rt == NULL ||
            rt->rt_rmx.rmx_iwmaxsegs < TCP_IW_MAXSEGS_DFLT ||
            rt->rt_rmx.rmx_iwcapsegs < TCP_IW_CAPSEGS_DFLT) {
                *maxsegs = tcp_iw_maxsegs;
                *capsegs = tcp_iw_capsegs;
                return;
        }
        *maxsegs = rt->rt_rmx.rmx_iwmaxsegs;
        *capsegs = rt->rt_rmx.rmx_iwcapsegs;
}

u_long
tcp_initial_window(struct tcpcb *tp)
{
        if (tcp_do_rfc3390) {
                /*
                 * RFC3390:
                 * "If the SYN or SYN/ACK is lost, the initial window
                 *  used by a sender after a correctly transmitted SYN
                 *  MUST be one segment consisting of MSS bytes."
                 *
                 * However, we do something a little bit more aggressive
                 * then RFC3390 here:
                 * - Only if time spent in the SYN or SYN|ACK retransmition
                 *   >= 3 seconds, the IW is reduced.  We do this mainly
                 *   because when RFC3390 is published, the initial RTO is
                 *   still 3 seconds (the threshold we test here), while
                 *   after RFC6298, the initial RTO is 1 second.  This
                 *   behaviour probably still falls within the spirit of
                 *   RFC3390.
                 * - When IW is reduced, 2*MSS is used instead of 1*MSS.
                 *   Mainly to avoid sender and receiver deadlock until
                 *   delayed ACK timer expires.  And even RFC2581 does not
                 *   try to reduce IW upon SYN or SYN|ACK retransmition
                 *   timeout.
                 *
                 * See also:
                 * http://tools.ietf.org/html/draft-ietf-tcpm-initcwnd-03
                 */
                if (tp->t_rxtsyn >= TCPTV_RTOBASE3) {
                        return (2 * tp->t_maxseg);
                } else {
                        u_long maxsegs, capsegs;

                        tcp_rmx_iwsegs(tp, &maxsegs, &capsegs);
                        return min(maxsegs * tp->t_maxseg,
                                   max(2 * tp->t_maxseg, capsegs * 1460));
                }
        } else {
                /*
                 * Even RFC2581 (back to 1999) allows 2*SMSS IW.
                 *
                 * Mainly to avoid sender and receiver deadlock
                 * until delayed ACK timer expires.
                 */
                return (2 * tp->t_maxseg);
        }
}

#ifdef TCP_SIGNATURE
/*
 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
 *
 * We do this over ip, tcphdr, segment data, and the key in the SADB.
 * When called from tcp_input(), we can be sure that th_sum has been
 * zeroed out and verified already.
 *
 * Return 0 if successful, otherwise return -1.
 *
 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
 * search with the destination IP address, and a 'magic SPI' to be
 * determined by the application. This is hardcoded elsewhere to 1179
 * right now. Another branch of this code exists which uses the SPD to
 * specify per-application flows but it is unstable.
 */
int
tcpsignature_compute(
        struct mbuf *m,         /* mbuf chain */
        int len,                /* length of TCP data */
        int optlen,             /* length of TCP options */
        u_char *buf,            /* storage for MD5 digest */
        u_int direction)        /* direction of flow */
{
        struct ippseudo ippseudo;
        MD5_CTX ctx;
        int doff;
        struct ip *ip;
        struct ipovly *ipovly;
        struct secasvar *sav;
        struct tcphdr *th;
#ifdef INET6
        struct ip6_hdr *ip6;
        struct in6_addr in6;
        uint32_t plen;
        uint16_t nhdr;
#endif /* INET6 */
        u_short savecsum;

        KASSERT(m != NULL, ("passed NULL mbuf. Game over."));
        KASSERT(buf != NULL, ("passed NULL storage pointer for MD5 signature"));
        /*
         * Extract the destination from the IP header in the mbuf.
         */
        ip = mtod(m, struct ip *);
#ifdef INET6
        ip6 = NULL;     /* Make the compiler happy. */
#endif /* INET6 */
        /*
         * Look up an SADB entry which matches the address found in
         * the segment.
         */
        switch (IP_VHL_V(ip->ip_vhl)) {
        case IPVERSION:
                sav = key_allocsa(AF_INET, (caddr_t)&ip->ip_src, (caddr_t)&ip->ip_dst,
                                IPPROTO_TCP, htonl(TCP_SIG_SPI));
                break;
#ifdef INET6
        case (IPV6_VERSION >> 4):
                ip6 = mtod(m, struct ip6_hdr *);
                sav = key_allocsa(AF_INET6, (caddr_t)&ip6->ip6_src, (caddr_t)&ip6->ip6_dst,
                                IPPROTO_TCP, htonl(TCP_SIG_SPI));
                break;
#endif /* INET6 */
        default:
                return (EINVAL);
                /* NOTREACHED */
                break;
        }
        if (sav == NULL) {
                kprintf("%s: SADB lookup failed\n", __func__);
                return (EINVAL);
        }
        MD5Init(&ctx);

        /*
         * Step 1: Update MD5 hash with IP pseudo-header.
         *
         * XXX The ippseudo header MUST be digested in network byte order,
         * or else we'll fail the regression test. Assume all fields we've
         * been doing arithmetic on have been in host byte order.
         * XXX One cannot depend on ipovly->ih_len here. When called from
         * tcp_output(), the underlying ip_len member has not yet been set.
         */
        switch (IP_VHL_V(ip->ip_vhl)) {
        case IPVERSION:
                ipovly = (struct ipovly *)ip;
                ippseudo.ippseudo_src = ipovly->ih_src;
                ippseudo.ippseudo_dst = ipovly->ih_dst;
                ippseudo.ippseudo_pad = 0;
                ippseudo.ippseudo_p = IPPROTO_TCP;
                ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
                MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
                th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
                doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
                break;
#ifdef INET6
        /*
         * RFC 2385, 2.0  Proposal
         * For IPv6, the pseudo-header is as described in RFC 2460, namely the
         * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
         * extended next header value (to form 32 bits), and 32-bit segment
         * length.
         * Note: Upper-Layer Packet Length comes before Next Header.
         */
        case (IPV6_VERSION >> 4):
                in6 = ip6->ip6_src;
                in6_clearscope(&in6);
                MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
                in6 = ip6->ip6_dst;
                in6_clearscope(&in6);
                MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
                plen = htonl(len + sizeof(struct tcphdr) + optlen);
                MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
                nhdr = 0;
                MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
                MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
                MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
                nhdr = IPPROTO_TCP;
                MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
                th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
                doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
                break;
#endif /* INET6 */
        default:
                return (EINVAL);
                /* NOTREACHED */
                break;
        }
        /*
         * Step 2: Update MD5 hash with TCP header, excluding options.
         * The TCP checksum must be set to zero.
         */
        savecsum = th->th_sum;
        th->th_sum = 0;
        MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
        th->th_sum = savecsum;
        /*
         * Step 3: Update MD5 hash with TCP segment data.
         *         Use m_apply() to avoid an early m_pullup().
         */
        if (len > 0)
                m_apply(m, doff, len, tcpsignature_apply, &ctx);
        /*
         * Step 4: Update MD5 hash with shared secret.
         */
        MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
        MD5Final(buf, &ctx);
        key_sa_recordxfer(sav, m);
        key_freesav(sav);
        return (0);
}

int
tcpsignature_apply(void *fstate, void *data, unsigned int len)
{

        MD5Update((MD5_CTX *)fstate, (unsigned char *)data, len);
        return (0);
}
#endif /* TCP_SIGNATURE */