/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */

/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */
/* Copyright (c) 1990 Mentat Inc. */

#include <sys/types.h>
#include <sys/stream.h>
#include <sys/strsun.h>
#include <sys/strsubr.h>
#include <sys/stropts.h>
#include <sys/strlog.h>
#define	_SUN_TPI_VERSION 2
#include <sys/tihdr.h>
#include <sys/timod.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/suntpi.h>
#include <sys/xti_inet.h>
#include <sys/cmn_err.h>
#include <sys/debug.h>
#include <sys/sdt.h>
#include <sys/vtrace.h>
#include <sys/kmem.h>
#include <sys/ethernet.h>
#include <sys/cpuvar.h>
#include <sys/dlpi.h>
#include <sys/pattr.h>
#include <sys/policy.h>
#include <sys/priv.h>
#include <sys/zone.h>
#include <sys/sunldi.h>

#include <sys/errno.h>
#include <sys/signal.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/sockio.h>
#include <sys/isa_defs.h>
#include <sys/md5.h>
#include <sys/random.h>
#include <sys/uio.h>
#include <sys/systm.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <netinet/ip6.h>
#include <netinet/icmp6.h>
#include <net/if.h>
#include <net/route.h>
#include <inet/ipsec_impl.h>

#include <inet/common.h>
#include <inet/ip.h>
#include <inet/ip_impl.h>
#include <inet/ip6.h>
#include <inet/ip_ndp.h>
#include <inet/proto_set.h>
#include <inet/mib2.h>
#include <inet/nd.h>
#include <inet/optcom.h>
#include <inet/snmpcom.h>
#include <inet/kstatcom.h>
#include <inet/tcp.h>
#include <inet/tcp_impl.h>
#include <inet/udp_impl.h>
#include <net/pfkeyv2.h>
#include <inet/ipdrop.h>

#include <inet/ipclassifier.h>
#include <inet/ip_ire.h>
#include <inet/ip_ftable.h>
#include <inet/ip_if.h>
#include <inet/ipp_common.h>
#include <inet/ip_rts.h>
#include <inet/ip_netinfo.h>
#include <sys/squeue_impl.h>
#include <sys/squeue.h>
#include <inet/kssl/ksslapi.h>
#include <sys/tsol/label.h>
#include <sys/tsol/tnet.h>
#include <rpc/pmap_prot.h>
#include <sys/callo.h>

#include <sys/clock_impl.h>	/* For LBOLT_FASTPATH{,64} */

/*
 * TCP Notes: aka FireEngine Phase I (PSARC 2002/433)
 *
 * (Read the detailed design doc in PSARC case directory)
 *
 * The entire tcp state is contained in tcp_t and conn_t structure
 * which are allocated in tandem using ipcl_conn_create() and passing
 * IPCL_TCPCONN as a flag. We use 'conn_ref' and 'conn_lock' to protect
 * the references on the tcp_t. The tcp_t structure is never compressed
 * and packets always land on the correct TCP perimeter from the time
 * eager is created till the time tcp_t dies (as such the old mentat
 * TCP global queue is not used for detached state and no IPSEC checking
 * is required). The global queue is still allocated to send out resets
 * for connection which have no listeners and IP directly calls
 * tcp_xmit_listeners_reset() which does any policy check.
 *
 * Protection and Synchronisation mechanism:
 *
 * The tcp data structure does not use any kind of lock for protecting
 * its state but instead uses 'squeues' for mutual exclusion from various
 * read and write side threads. To access a tcp member, the thread should
 * always be behind squeue (via squeue_enter with flags as SQ_FILL, SQ_PROCESS,
 * or SQ_NODRAIN). Since the squeues allow a direct function call, caller
 * can pass any tcp function having prototype of edesc_t as argument
 * (different from traditional STREAMs model where packets come in only
 * designated entry points). The list of functions that can be directly
 * called via squeue are listed before the usual function prototype.
 *
 * Referencing:
 *
 * TCP is MT-Hot and we use a reference based scheme to make sure that the
 * tcp structure doesn't disappear when its needed. When the application
 * creates an outgoing connection or accepts an incoming connection, we
 * start out with 2 references on 'conn_ref'. One for TCP and one for IP.
 * The IP reference is just a symbolic reference since ip_tcpclose()
 * looks at tcp structure after tcp_close_output() returns which could
 * have dropped the last TCP reference. So as long as the connection is
 * in attached state i.e. !TCP_IS_DETACHED, we have 2 references on the
 * conn_t. The classifier puts its own reference when the connection is
 * inserted in listen or connected hash. Anytime a thread needs to enter
 * the tcp connection perimeter, it retrieves the conn/tcp from q->ptr
 * on write side or by doing a classify on read side and then puts a
 * reference on the conn before doing squeue_enter/tryenter/fill. For
 * read side, the classifier itself puts the reference under fanout lock
 * to make sure that tcp can't disappear before it gets processed. The
 * squeue will drop this reference automatically so the called function
 * doesn't have to do a DEC_REF.
 *
 * Opening a new connection:
 *
 * The outgoing connection open is pretty simple. tcp_open() does the
 * work in creating the conn/tcp structure and initializing it. The
 * squeue assignment is done based on the CPU the application
 * is running on. So for outbound connections, processing is always done
 * on application CPU which might be different from the incoming CPU
 * being interrupted by the NIC. An optimal way would be to figure out
 * the NIC <-> CPU binding at listen time, and assign the outgoing
 * connection to the squeue attached to the CPU that will be interrupted
 * for incoming packets (we know the NIC based on the bind IP address).
 * This might seem like a problem if more data is going out but the
 * fact is that in most cases the transmit is ACK driven transmit where
 * the outgoing data normally sits on TCP's xmit queue waiting to be
 * transmitted.
 *
 * Accepting a connection:
 *
 * This is a more interesting case because of various races involved in
 * establishing a eager in its own perimeter. Read the meta comment on
 * top of tcp_input_listener(). But briefly, the squeue is picked by
 * ip_fanout based on the ring or the sender (if loopback).
 *
 * Closing a connection:
 *
 * The close is fairly straight forward. tcp_close() calls tcp_close_output()
 * via squeue to do the close and mark the tcp as detached if the connection
 * was in state TCPS_ESTABLISHED or greater. In the later case, TCP keep its
 * reference but tcp_close() drop IP's reference always. So if tcp was
 * not killed, it is sitting in time_wait list with 2 reference - 1 for TCP
 * and 1 because it is in classifier's connected hash. This is the condition
 * we use to determine that its OK to clean up the tcp outside of squeue
 * when time wait expires (check the ref under fanout and conn_lock and
 * if it is 2, remove it from fanout hash and kill it).
 *
 * Although close just drops the necessary references and marks the
 * tcp_detached state, tcp_close needs to know the tcp_detached has been
 * set (under squeue) before letting the STREAM go away (because a
 * inbound packet might attempt to go up the STREAM while the close
 * has happened and tcp_detached is not set). So a special lock and
 * flag is used along with a condition variable (tcp_closelock, tcp_closed,
 * and tcp_closecv) to signal tcp_close that tcp_close_out() has marked
 * tcp_detached.
 *
 * Special provisions and fast paths:
 *
 * We make special provisions for sockfs by marking tcp_issocket
 * whenever we have only sockfs on top of TCP. This allows us to skip
 * putting the tcp in acceptor hash since a sockfs listener can never
 * become acceptor and also avoid allocating a tcp_t for acceptor STREAM
 * since eager has already been allocated and the accept now happens
 * on acceptor STREAM. There is a big blob of comment on top of
 * tcp_input_listener explaining the new accept. When socket is POP'd,
 * sockfs sends us an ioctl to mark the fact and we go back to old
 * behaviour. Once tcp_issocket is unset, its never set for the
 * life of that connection.
 *
 * IPsec notes :
 *
 * Since a packet is always executed on the correct TCP perimeter
 * all IPsec processing is defered to IP including checking new
 * connections and setting IPSEC policies for new connection. The
 * only exception is tcp_xmit_listeners_reset() which is called
 * directly from IP and needs to policy check to see if TH_RST
 * can be sent out.
 */

/*
 * Values for squeue switch:
 * 1: SQ_NODRAIN
 * 2: SQ_PROCESS
 * 3: SQ_FILL
 */
int tcp_squeue_wput = 2;	/* /etc/systems */
int tcp_squeue_flag;

/*
 * This controls how tiny a write must be before we try to copy it
 * into the mblk on the tail of the transmit queue.  Not much
 * speedup is observed for values larger than sixteen.  Zero will
 * disable the optimisation.
 */
int tcp_tx_pull_len = 16;

/*
 * TCP Statistics.
 *
 * How TCP statistics work.
 *
 * There are two types of statistics invoked by two macros.
 *
 * TCP_STAT(name) does non-atomic increment of a named stat counter. It is
 * supposed to be used in non MT-hot paths of the code.
 *
 * TCP_DBGSTAT(name) does atomic increment of a named stat counter. It is
 * supposed to be used for DEBUG purposes and may be used on a hot path.
 *
 * Both TCP_STAT and TCP_DBGSTAT counters are available using kstat
 * (use "kstat tcp" to get them).
 *
 * There is also additional debugging facility that marks tcp_clean_death()
 * instances and saves them in tcp_t structure. It is triggered by
 * TCP_TAG_CLEAN_DEATH define. Also, there is a global array of counters for
 * tcp_clean_death() calls that counts the number of times each tag was hit. It
 * is triggered by TCP_CLD_COUNTERS define.
 *
 * How to add new counters.
 *
 * 1) Add a field in the tcp_stat structure describing your counter.
 * 2) Add a line in the template in tcp_kstat2_init() with the name
 *    of the counter.
 *
 *    IMPORTANT!! - make sure that both are in sync !!
 * 3) Use either TCP_STAT or TCP_DBGSTAT with the name.
 *
 * Please avoid using private counters which are not kstat-exported.
 *
 * TCP_TAG_CLEAN_DEATH set to 1 enables tagging of tcp_clean_death() instances
 * in tcp_t structure.
 *
 * TCP_MAX_CLEAN_DEATH_TAG is the maximum number of possible clean death tags.
 */

#ifndef TCP_DEBUG_COUNTER
#ifdef DEBUG
#define	TCP_DEBUG_COUNTER 1
#else
#define	TCP_DEBUG_COUNTER 0
#endif
#endif

#define	TCP_CLD_COUNTERS 0

#define	TCP_TAG_CLEAN_DEATH 1
#define	TCP_MAX_CLEAN_DEATH_TAG 32

#ifdef lint
static int _lint_dummy_;
#endif

#if TCP_CLD_COUNTERS
static uint_t tcp_clean_death_stat[TCP_MAX_CLEAN_DEATH_TAG];
#define	TCP_CLD_STAT(x) tcp_clean_death_stat[x]++
#elif defined(lint)
#define	TCP_CLD_STAT(x) ASSERT(_lint_dummy_ == 0);
#else
#define	TCP_CLD_STAT(x)
#endif

#if TCP_DEBUG_COUNTER
#define	TCP_DBGSTAT(tcps, x)	\
	atomic_add_64(&((tcps)->tcps_statistics.x.value.ui64), 1)
#define	TCP_G_DBGSTAT(x)	\
	atomic_add_64(&(tcp_g_statistics.x.value.ui64), 1)
#elif defined(lint)
#define	TCP_DBGSTAT(tcps, x) ASSERT(_lint_dummy_ == 0);
#define	TCP_G_DBGSTAT(x) ASSERT(_lint_dummy_ == 0);
#else
#define	TCP_DBGSTAT(tcps, x)
#define	TCP_G_DBGSTAT(x)
#endif

#define	TCP_G_STAT(x)	(tcp_g_statistics.x.value.ui64++)

tcp_g_stat_t	tcp_g_statistics;
kstat_t		*tcp_g_kstat;

/* Macros for timestamp comparisons */
#define	TSTMP_GEQ(a, b)	((int32_t)((a)-(b)) >= 0)
#define	TSTMP_LT(a, b)	((int32_t)((a)-(b)) < 0)

/*
 * Parameters for TCP Initial Send Sequence number (ISS) generation.  When
 * tcp_strong_iss is set to 1, which is the default, the ISS is calculated
 * by adding three components: a time component which grows by 1 every 4096
 * nanoseconds (versus every 4 microseconds suggested by RFC 793, page 27);
 * a per-connection component which grows by 125000 for every new connection;
 * and an "extra" component that grows by a random amount centered
 * approximately on 64000.  This causes the ISS generator to cycle every
 * 4.89 hours if no TCP connections are made, and faster if connections are
 * made.
 *
 * When tcp_strong_iss is set to 0, ISS is calculated by adding two
 * components: a time component which grows by 250000 every second; and
 * a per-connection component which grows by 125000 for every new connections.
 *
 * A third method, when tcp_strong_iss is set to 2, for generating ISS is
 * prescribed by Steve Bellovin.  This involves adding time, the 125000 per
 * connection, and a one-way hash (MD5) of the connection ID <sport, dport,
 * src, dst>, a "truly" random (per RFC 1750) number, and a console-entered
 * password.
 */
#define	ISS_INCR	250000
#define	ISS_NSEC_SHT	12

static sin_t	sin_null;	/* Zero address for quick clears */
static sin6_t	sin6_null;	/* Zero address for quick clears */

/*
 * This implementation follows the 4.3BSD interpretation of the urgent
 * pointer and not RFC 1122. Switching to RFC 1122 behavior would cause
 * incompatible changes in protocols like telnet and rlogin.
 */
#define	TCP_OLD_URP_INTERPRETATION	1

/*
 * Since tcp_listener is not cleared atomically with tcp_detached
 * being cleared we need this extra bit to tell a detached connection
 * apart from one that is in the process of being accepted.
 */
#define	TCP_IS_DETACHED_NONEAGER(tcp)	\
	(TCP_IS_DETACHED(tcp) &&	\
	    (!(tcp)->tcp_hard_binding))

/*
 * TCP reassembly macros.  We hide starting and ending sequence numbers in
 * b_next and b_prev of messages on the reassembly queue.  The messages are
 * chained using b_cont.  These macros are used in tcp_reass() so we don't
 * have to see the ugly casts and assignments.
 */
#define	TCP_REASS_SEQ(mp)		((uint32_t)(uintptr_t)((mp)->b_next))
#define	TCP_REASS_SET_SEQ(mp, u)	((mp)->b_next = \
					(mblk_t *)(uintptr_t)(u))
#define	TCP_REASS_END(mp)		((uint32_t)(uintptr_t)((mp)->b_prev))
#define	TCP_REASS_SET_END(mp, u)	((mp)->b_prev = \
					(mblk_t *)(uintptr_t)(u))

/*
 * Implementation of TCP Timers.
 * =============================
 *
 * INTERFACE:
 *
 * There are two basic functions dealing with tcp timers:
 *
 *	timeout_id_t	tcp_timeout(connp, func, time)
 * 	clock_t		tcp_timeout_cancel(connp, timeout_id)
 *	TCP_TIMER_RESTART(tcp, intvl)
 *
 * tcp_timeout() starts a timer for the 'tcp' instance arranging to call 'func'
 * after 'time' ticks passed. The function called by timeout() must adhere to
 * the same restrictions as a driver soft interrupt handler - it must not sleep
 * or call other functions that might sleep. The value returned is the opaque
 * non-zero timeout identifier that can be passed to tcp_timeout_cancel() to
 * cancel the request. The call to tcp_timeout() may fail in which case it
 * returns zero. This is different from the timeout(9F) function which never
 * fails.
 *
 * The call-back function 'func' always receives 'connp' as its single
 * argument. It is always executed in the squeue corresponding to the tcp
 * structure. The tcp structure is guaranteed to be present at the time the
 * call-back is called.
 *
 * NOTE: The call-back function 'func' is never called if tcp is in
 * 	the TCPS_CLOSED state.
 *
 * tcp_timeout_cancel() attempts to cancel a pending tcp_timeout()
 * request. locks acquired by the call-back routine should not be held across
 * the call to tcp_timeout_cancel() or a deadlock may result.
 *
 * tcp_timeout_cancel() returns -1 if it can not cancel the timeout request.
 * Otherwise, it returns an integer value greater than or equal to 0. In
 * particular, if the call-back function is already placed on the squeue, it can
 * not be canceled.
 *
 * NOTE: both tcp_timeout() and tcp_timeout_cancel() should always be called
 * 	within squeue context corresponding to the tcp instance. Since the
 *	call-back is also called via the same squeue, there are no race
 *	conditions described in untimeout(9F) manual page since all calls are
 *	strictly serialized.
 *
 *      TCP_TIMER_RESTART() is a macro that attempts to cancel a pending timeout
 *	stored in tcp_timer_tid and starts a new one using
 *	MSEC_TO_TICK(intvl). It always uses tcp_timer() function as a call-back
 *	and stores the return value of tcp_timeout() in the tcp->tcp_timer_tid
 *	field.
 *
 * NOTE: since the timeout cancellation is not guaranteed, the cancelled
 *	call-back may still be called, so it is possible tcp_timer() will be
 *	called several times. This should not be a problem since tcp_timer()
 *	should always check the tcp instance state.
 *
 *
 * IMPLEMENTATION:
 *
 * TCP timers are implemented using three-stage process. The call to
 * tcp_timeout() uses timeout(9F) function to call tcp_timer_callback() function
 * when the timer expires. The tcp_timer_callback() arranges the call of the
 * tcp_timer_handler() function via squeue corresponding to the tcp
 * instance. The tcp_timer_handler() calls actual requested timeout call-back
 * and passes tcp instance as an argument to it. Information is passed between
 * stages using the tcp_timer_t structure which contains the connp pointer, the
 * tcp call-back to call and the timeout id returned by the timeout(9F).
 *
 * The tcp_timer_t structure is not used directly, it is embedded in an mblk_t -
 * like structure that is used to enter an squeue. The mp->b_rptr of this pseudo
 * mblk points to the beginning of tcp_timer_t structure. The tcp_timeout()
 * returns the pointer to this mblk.
 *
 * The pseudo mblk is allocated from a special tcp_timer_cache kmem cache. It
 * looks like a normal mblk without actual dblk attached to it.
 *
 * To optimize performance each tcp instance holds a small cache of timer
 * mblocks. In the current implementation it caches up to two timer mblocks per
 * tcp instance. The cache is preserved over tcp frees and is only freed when
 * the whole tcp structure is destroyed by its kmem destructor. Since all tcp
 * timer processing happens on a corresponding squeue, the cache manipulation
 * does not require any locks. Experiments show that majority of timer mblocks
 * allocations are satisfied from the tcp cache and do not involve kmem calls.
 *
 * The tcp_timeout() places a refhold on the connp instance which guarantees
 * that it will be present at the time the call-back function fires. The
 * tcp_timer_handler() drops the reference after calling the call-back, so the
 * call-back function does not need to manipulate the references explicitly.
 */

typedef struct tcp_timer_s {
	conn_t	*connp;
	void 	(*tcpt_proc)(void *);
	callout_id_t   tcpt_tid;
} tcp_timer_t;

static kmem_cache_t *tcp_timercache;
kmem_cache_t	*tcp_sack_info_cache;

/*
 * For scalability, we must not run a timer for every TCP connection
 * in TIME_WAIT state.  To see why, consider (for time wait interval of
 * 4 minutes):
 *	1000 connections/sec * 240 seconds/time wait = 240,000 active conn's
 *
 * This list is ordered by time, so you need only delete from the head
 * until you get to entries which aren't old enough to delete yet.
 * The list consists of only the detached TIME_WAIT connections.
 *
 * Note that the timer (tcp_time_wait_expire) is started when the tcp_t
 * becomes detached TIME_WAIT (either by changing the state and already
 * being detached or the other way around). This means that the TIME_WAIT
 * state can be extended (up to doubled) if the connection doesn't become
 * detached for a long time.
 *
 * The list manipulations (including tcp_time_wait_next/prev)
 * are protected by the tcp_time_wait_lock. The content of the
 * detached TIME_WAIT connections is protected by the normal perimeters.
 *
 * This list is per squeue and squeues are shared across the tcp_stack_t's.
 * Things on tcp_time_wait_head remain associated with the tcp_stack_t
 * and conn_netstack.
 * The tcp_t's that are added to tcp_free_list are disassociated and
 * have NULL tcp_tcps and conn_netstack pointers.
 */
typedef struct tcp_squeue_priv_s {
	kmutex_t	tcp_time_wait_lock;
	callout_id_t	tcp_time_wait_tid;
	tcp_t		*tcp_time_wait_head;
	tcp_t		*tcp_time_wait_tail;
	tcp_t		*tcp_free_list;
	uint_t		tcp_free_list_cnt;
} tcp_squeue_priv_t;

/*
 * TCP_TIME_WAIT_DELAY governs how often the time_wait_collector runs.
 * Running it every 5 seconds seems to give the best results.
 */
#define	TCP_TIME_WAIT_DELAY drv_usectohz(5000000)

/*
 * To prevent memory hog, limit the number of entries in tcp_free_list
 * to 1% of available memory / number of cpus
 */
uint_t tcp_free_list_max_cnt = 0;

#define	TCP_XMIT_LOWATER	4096
#define	TCP_XMIT_HIWATER	49152
#define	TCP_RECV_LOWATER	2048
#define	TCP_RECV_HIWATER	128000

/*
 *  PAWS needs a timer for 24 days.  This is the number of ticks in 24 days
 */
#define	PAWS_TIMEOUT	((clock_t)(24*24*60*60*hz))

#define	TIDUSZ	4096	/* transport interface data unit size */

/*
 * Bind hash list size and has function.  It has to be a power of 2 for
 * hashing.
 */
#define	TCP_BIND_FANOUT_SIZE	512
#define	TCP_BIND_HASH(lport) (ntohs(lport) & (TCP_BIND_FANOUT_SIZE - 1))
/*
 * Size of listen and acceptor hash list.  It has to be a power of 2 for
 * hashing.
 */
#define	TCP_FANOUT_SIZE		256

#ifdef	_ILP32
#define	TCP_ACCEPTOR_HASH(accid)					\
		(((uint_t)(accid) >> 8) & (TCP_FANOUT_SIZE - 1))
#else
#define	TCP_ACCEPTOR_HASH(accid)					\
		((uint_t)(accid) & (TCP_FANOUT_SIZE - 1))
#endif	/* _ILP32 */

#define	IP_ADDR_CACHE_SIZE	2048
#define	IP_ADDR_CACHE_HASH(faddr)					\
	(ntohl(faddr) & (IP_ADDR_CACHE_SIZE -1))

/*
 * TCP options struct returned from tcp_parse_options.
 */
typedef struct tcp_opt_s {
	uint32_t	tcp_opt_mss;
	uint32_t	tcp_opt_wscale;
	uint32_t	tcp_opt_ts_val;
	uint32_t	tcp_opt_ts_ecr;
	tcp_t		*tcp;
} tcp_opt_t;

/*
 * RFC1323-recommended phrasing of TSTAMP option, for easier parsing
 */

#ifdef _BIG_ENDIAN
#define	TCPOPT_NOP_NOP_TSTAMP ((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | \
	(TCPOPT_TSTAMP << 8) | 10)
#else
#define	TCPOPT_NOP_NOP_TSTAMP ((10 << 24) | (TCPOPT_TSTAMP << 16) | \
	(TCPOPT_NOP << 8) | TCPOPT_NOP)
#endif

/*
 * Flags returned from tcp_parse_options.
 */
#define	TCP_OPT_MSS_PRESENT	1
#define	TCP_OPT_WSCALE_PRESENT	2
#define	TCP_OPT_TSTAMP_PRESENT	4
#define	TCP_OPT_SACK_OK_PRESENT	8
#define	TCP_OPT_SACK_PRESENT	16

/* TCP option length */
#define	TCPOPT_NOP_LEN		1
#define	TCPOPT_MAXSEG_LEN	4
#define	TCPOPT_WS_LEN		3
#define	TCPOPT_REAL_WS_LEN	(TCPOPT_WS_LEN+1)
#define	TCPOPT_TSTAMP_LEN	10
#define	TCPOPT_REAL_TS_LEN	(TCPOPT_TSTAMP_LEN+2)
#define	TCPOPT_SACK_OK_LEN	2
#define	TCPOPT_REAL_SACK_OK_LEN	(TCPOPT_SACK_OK_LEN+2)
#define	TCPOPT_REAL_SACK_LEN	4
#define	TCPOPT_MAX_SACK_LEN	36
#define	TCPOPT_HEADER_LEN	2

/* TCP cwnd burst factor. */
#define	TCP_CWND_INFINITE	65535
#define	TCP_CWND_SS		3
#define	TCP_CWND_NORMAL		5

/* Maximum TCP initial cwin (start/restart). */
#define	TCP_MAX_INIT_CWND	8

/*
 * Initialize cwnd according to RFC 3390.  def_max_init_cwnd is
 * either tcp_slow_start_initial or tcp_slow_start_after idle
 * depending on the caller.  If the upper layer has not used the
 * TCP_INIT_CWND option to change the initial cwnd, tcp_init_cwnd
 * should be 0 and we use the formula in RFC 3390 to set tcp_cwnd.
 * If the upper layer has changed set the tcp_init_cwnd, just use
 * it to calculate the tcp_cwnd.
 */
#define	SET_TCP_INIT_CWND(tcp, mss, def_max_init_cwnd)			\
{									\
	if ((tcp)->tcp_init_cwnd == 0) {				\
		(tcp)->tcp_cwnd = MIN(def_max_init_cwnd * (mss),	\
		    MIN(4 * (mss), MAX(2 * (mss), 4380 / (mss) * (mss)))); \
	} else {							\
		(tcp)->tcp_cwnd = (tcp)->tcp_init_cwnd * (mss);		\
	}								\
	tcp->tcp_cwnd_cnt = 0;						\
}

/* TCP Timer control structure */
typedef struct tcpt_s {
	pfv_t	tcpt_pfv;	/* The routine we are to call */
	tcp_t	*tcpt_tcp;	/* The parameter we are to pass in */
} tcpt_t;

/*
 * Functions called directly via squeue having a prototype of edesc_t.
 */
void		tcp_input_listener(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *ira);
static void	tcp_wput_nondata(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
void		tcp_accept_finish(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static void	tcp_wput_ioctl(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static void	tcp_wput_proto(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
void		tcp_input_data(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *ira);
static void	tcp_close_output(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
void		tcp_output(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
void		tcp_output_urgent(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static void	tcp_rsrv_input(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static void	tcp_timer_handler(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static void	tcp_linger_interrupted(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);


/* Prototype for TCP functions */
static void	tcp_random_init(void);
int		tcp_random(void);
static void	tcp_tli_accept(tcp_t *tcp, mblk_t *mp);
static void	tcp_accept_swap(tcp_t *listener, tcp_t *acceptor,
		    tcp_t *eager);
static int	tcp_set_destination(tcp_t *tcp);
static in_port_t tcp_bindi(tcp_t *tcp, in_port_t port, const in6_addr_t *laddr,
    int reuseaddr, boolean_t quick_connect, boolean_t bind_to_req_port_only,
    boolean_t user_specified);
static void	tcp_closei_local(tcp_t *tcp);
static void	tcp_close_detached(tcp_t *tcp);
static boolean_t tcp_conn_con(tcp_t *tcp, uchar_t *iphdr,
		    mblk_t *idmp, mblk_t **defermp, ip_recv_attr_t *ira);
static void	tcp_tpi_connect(tcp_t *tcp, mblk_t *mp);
static int	tcp_connect_ipv4(tcp_t *tcp, ipaddr_t *dstaddrp,
		    in_port_t dstport, uint_t srcid);
static int	tcp_connect_ipv6(tcp_t *tcp, in6_addr_t *dstaddrp,
		    in_port_t dstport, uint32_t flowinfo,
		    uint_t srcid, uint32_t scope_id);
static int	tcp_clean_death(tcp_t *tcp, int err, uint8_t tag);
static void	tcp_disconnect(tcp_t *tcp, mblk_t *mp);
static char	*tcp_display(tcp_t *tcp, char *, char);
static boolean_t tcp_eager_blowoff(tcp_t *listener, t_scalar_t seqnum);
static void	tcp_eager_cleanup(tcp_t *listener, boolean_t q0_only);
static void	tcp_eager_unlink(tcp_t *tcp);
static void	tcp_err_ack(tcp_t *tcp, mblk_t *mp, int tlierr,
		    int unixerr);
static void	tcp_err_ack_prim(tcp_t *tcp, mblk_t *mp, int primitive,
		    int tlierr, int unixerr);
static int	tcp_extra_priv_ports_get(queue_t *q, mblk_t *mp, caddr_t cp,
		    cred_t *cr);
static int	tcp_extra_priv_ports_add(queue_t *q, mblk_t *mp,
		    char *value, caddr_t cp, cred_t *cr);
static int	tcp_extra_priv_ports_del(queue_t *q, mblk_t *mp,
		    char *value, caddr_t cp, cred_t *cr);
static int	tcp_tpistate(tcp_t *tcp);
static void	tcp_bind_hash_insert(tf_t *tf, tcp_t *tcp,
    int caller_holds_lock);
static void	tcp_bind_hash_remove(tcp_t *tcp);
static tcp_t	*tcp_acceptor_hash_lookup(t_uscalar_t id, tcp_stack_t *);
void		tcp_acceptor_hash_insert(t_uscalar_t id, tcp_t *tcp);
static void	tcp_acceptor_hash_remove(tcp_t *tcp);
static void	tcp_capability_req(tcp_t *tcp, mblk_t *mp);
static void	tcp_info_req(tcp_t *tcp, mblk_t *mp);
static void	tcp_addr_req(tcp_t *tcp, mblk_t *mp);
static void	tcp_init_values(tcp_t *tcp);
static void	tcp_ip_notify(tcp_t *tcp);
static void	tcp_iss_init(tcp_t *tcp);
static void	tcp_keepalive_killer(void *arg);
static int	tcp_parse_options(tcpha_t *tcpha, tcp_opt_t *tcpopt);
static void	tcp_mss_set(tcp_t *tcp, uint32_t size);
static int	tcp_conprim_opt_process(tcp_t *tcp, mblk_t *mp,
		    int *do_disconnectp, int *t_errorp, int *sys_errorp);
static boolean_t tcp_allow_connopt_set(int level, int name);
int		tcp_opt_default(queue_t *q, int level, int name, uchar_t *ptr);
static int	tcp_param_get(queue_t *q, mblk_t *mp, caddr_t cp, cred_t *cr);
static boolean_t tcp_param_register(IDP *ndp, tcpparam_t *tcppa, int cnt,
    tcp_stack_t *);
static int	tcp_param_set(queue_t *q, mblk_t *mp, char *value,
		    caddr_t cp, cred_t *cr);
static int	tcp_param_set_aligned(queue_t *q, mblk_t *mp, char *value,
		    caddr_t cp, cred_t *cr);
static void	tcp_iss_key_init(uint8_t *phrase, int len, tcp_stack_t *);
static int	tcp_1948_phrase_set(queue_t *q, mblk_t *mp, char *value,
		    caddr_t cp, cred_t *cr);
static void	tcp_process_shrunk_swnd(tcp_t *tcp, uint32_t shrunk_cnt);
static void	tcp_update_xmit_tail(tcp_t *tcp, uint32_t snxt);
static mblk_t	*tcp_reass(tcp_t *tcp, mblk_t *mp, uint32_t start);
static void	tcp_reass_elim_overlap(tcp_t *tcp, mblk_t *mp);
static void	tcp_reinit(tcp_t *tcp);
static void	tcp_reinit_values(tcp_t *tcp);

static uint_t	tcp_rwnd_reopen(tcp_t *tcp);
static uint_t	tcp_rcv_drain(tcp_t *tcp);
static void	tcp_sack_rxmit(tcp_t *tcp, uint_t *flags);
static boolean_t tcp_send_rst_chk(tcp_stack_t *);
static void	tcp_ss_rexmit(tcp_t *tcp);
static mblk_t	*tcp_input_add_ancillary(tcp_t *tcp, mblk_t *mp, ip_pkt_t *ipp,
    ip_recv_attr_t *);
static void	tcp_process_options(tcp_t *, tcpha_t *);
static void	tcp_rsrv(queue_t *q);
static int	tcp_snmp_state(tcp_t *tcp);
static void	tcp_timer(void *arg);
static void	tcp_timer_callback(void *);
static in_port_t tcp_update_next_port(in_port_t port, const tcp_t *tcp,
    boolean_t random);
static in_port_t tcp_get_next_priv_port(const tcp_t *);
static void	tcp_wput_sock(queue_t *q, mblk_t *mp);
static void	tcp_wput_fallback(queue_t *q, mblk_t *mp);
void		tcp_tpi_accept(queue_t *q, mblk_t *mp);
static void	tcp_wput_data(tcp_t *tcp, mblk_t *mp, boolean_t urgent);
static void	tcp_wput_flush(tcp_t *tcp, mblk_t *mp);
static void	tcp_wput_iocdata(tcp_t *tcp, mblk_t *mp);
static int	tcp_send(tcp_t *tcp, const int mss,
		    const int total_hdr_len, const int tcp_hdr_len,
		    const int num_sack_blk, int *usable, uint_t *snxt,
		    int *tail_unsent, mblk_t **xmit_tail, mblk_t *local_time);
static void	tcp_fill_header(tcp_t *tcp, uchar_t *rptr, clock_t now,
		    int num_sack_blk);
static void	tcp_wsrv(queue_t *q);
static int	tcp_xmit_end(tcp_t *tcp);
static void	tcp_ack_timer(void *arg);
static mblk_t	*tcp_ack_mp(tcp_t *tcp);
static void	tcp_xmit_early_reset(char *str, mblk_t *mp,
		    uint32_t seq, uint32_t ack, int ctl, ip_recv_attr_t *,
		    ip_stack_t *, conn_t *);
static void	tcp_xmit_ctl(char *str, tcp_t *tcp, uint32_t seq,
		    uint32_t ack, int ctl);
static void	tcp_set_rto(tcp_t *, time_t);
static void	tcp_icmp_input(void *, mblk_t *, void *, ip_recv_attr_t *);
static void	tcp_icmp_error_ipv6(tcp_t *, mblk_t *, ip_recv_attr_t *);
static boolean_t tcp_verifyicmp(conn_t *, void *, icmph_t *, icmp6_t *,
    ip_recv_attr_t *);
static int	tcp_build_hdrs(tcp_t *);
static void	tcp_time_wait_processing(tcp_t *tcp, mblk_t *mp,
    uint32_t seg_seq, uint32_t seg_ack, int seg_len, tcpha_t *tcpha,
    ip_recv_attr_t *ira);
boolean_t	tcp_paws_check(tcp_t *tcp, tcpha_t *tcpha, tcp_opt_t *tcpoptp);
static boolean_t tcp_zcopy_check(tcp_t *);
static void	tcp_zcopy_notify(tcp_t *);
static mblk_t	*tcp_zcopy_backoff(tcp_t *, mblk_t *, boolean_t);
static void	tcp_update_lso(tcp_t *tcp, ip_xmit_attr_t *ixa);
static void	tcp_update_pmtu(tcp_t *tcp, boolean_t decrease_only);
static void	tcp_update_zcopy(tcp_t *tcp);
static void	tcp_notify(void *, ip_xmit_attr_t *, ixa_notify_type_t,
    ixa_notify_arg_t);
static void	tcp_rexmit_after_error(tcp_t *tcp);
static void	tcp_send_data(tcp_t *, mblk_t *);
extern mblk_t	*tcp_timermp_alloc(int);
extern void	tcp_timermp_free(tcp_t *);
static void	tcp_timer_free(tcp_t *tcp, mblk_t *mp);
static void	tcp_stop_lingering(tcp_t *tcp);
static void	tcp_close_linger_timeout(void *arg);
static void	*tcp_stack_init(netstackid_t stackid, netstack_t *ns);
static void	tcp_stack_fini(netstackid_t stackid, void *arg);
static void	*tcp_g_kstat_init(tcp_g_stat_t *);
static void	tcp_g_kstat_fini(kstat_t *);
static void	*tcp_kstat_init(netstackid_t, tcp_stack_t *);
static void	tcp_kstat_fini(netstackid_t, kstat_t *);
static void	*tcp_kstat2_init(netstackid_t, tcp_stat_t *);
static void	tcp_kstat2_fini(netstackid_t, kstat_t *);
static int	tcp_kstat_update(kstat_t *kp, int rw);
static mblk_t	*tcp_conn_create_v6(conn_t *lconnp, conn_t *connp, mblk_t *mp,
    ip_recv_attr_t *ira);
static mblk_t	*tcp_conn_create_v4(conn_t *lconnp, conn_t *connp, mblk_t *mp,
    ip_recv_attr_t *ira);
static int	tcp_squeue_switch(int);

static int	tcp_open(queue_t *, dev_t *, int, int, cred_t *, boolean_t);
static int	tcp_openv4(queue_t *, dev_t *, int, int, cred_t *);
static int	tcp_openv6(queue_t *, dev_t *, int, int, cred_t *);
static int	tcp_tpi_close(queue_t *, int);
static int	tcp_tpi_close_accept(queue_t *);

static void	tcp_squeue_add(squeue_t *);
static void	tcp_setcred_data(mblk_t *, ip_recv_attr_t *);

extern void	tcp_kssl_input(tcp_t *, mblk_t *, cred_t *);

void tcp_eager_kill(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy);
void tcp_clean_death_wrapper(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);

static int tcp_accept(sock_lower_handle_t, sock_lower_handle_t,
	    sock_upper_handle_t, cred_t *);
static int tcp_listen(sock_lower_handle_t, int, cred_t *);
static int tcp_do_listen(conn_t *, struct sockaddr *, socklen_t, int, cred_t *,
    boolean_t);
static int tcp_do_connect(conn_t *, const struct sockaddr *, socklen_t,
    cred_t *, pid_t);
static int tcp_do_bind(conn_t *, struct sockaddr *, socklen_t, cred_t *,
    boolean_t);
static int tcp_do_unbind(conn_t *);
static int tcp_bind_check(conn_t *, struct sockaddr *, socklen_t, cred_t *,
    boolean_t);

static void tcp_ulp_newconn(conn_t *, conn_t *, mblk_t *);

/*
 * Routines related to the TCP_IOC_ABORT_CONN ioctl command.
 *
 * TCP_IOC_ABORT_CONN is a non-transparent ioctl command used for aborting
 * TCP connections. To invoke this ioctl, a tcp_ioc_abort_conn_t structure
 * (defined in tcp.h) needs to be filled in and passed into the kernel
 * via an I_STR ioctl command (see streamio(7I)). The tcp_ioc_abort_conn_t
 * structure contains the four-tuple of a TCP connection and a range of TCP
 * states (specified by ac_start and ac_end). The use of wildcard addresses
 * and ports is allowed. Connections with a matching four tuple and a state
 * within the specified range will be aborted. The valid states for the
 * ac_start and ac_end fields are in the range TCPS_SYN_SENT to TCPS_TIME_WAIT,
 * inclusive.
 *
 * An application which has its connection aborted by this ioctl will receive
 * an error that is dependent on the connection state at the time of the abort.
 * If the connection state is < TCPS_TIME_WAIT, an application should behave as
 * though a RST packet has been received.  If the connection state is equal to
 * TCPS_TIME_WAIT, the 2MSL timeout will immediately be canceled by the kernel
 * and all resources associated with the connection will be freed.
 */
static mblk_t	*tcp_ioctl_abort_build_msg(tcp_ioc_abort_conn_t *, tcp_t *);
static void	tcp_ioctl_abort_dump(tcp_ioc_abort_conn_t *);
static void	tcp_ioctl_abort_handler(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy);
static int	tcp_ioctl_abort(tcp_ioc_abort_conn_t *, tcp_stack_t *tcps);
static void	tcp_ioctl_abort_conn(queue_t *, mblk_t *);
static int	tcp_ioctl_abort_bucket(tcp_ioc_abort_conn_t *, int, int *,
    boolean_t, tcp_stack_t *);

static struct module_info tcp_rinfo =  {
	TCP_MOD_ID, TCP_MOD_NAME, 0, INFPSZ, TCP_RECV_HIWATER, TCP_RECV_LOWATER
};

static struct module_info tcp_winfo =  {
	TCP_MOD_ID, TCP_MOD_NAME, 0, INFPSZ, 127, 16
};

/*
 * Entry points for TCP as a device. The normal case which supports
 * the TCP functionality.
 * We have separate open functions for the /dev/tcp and /dev/tcp6 devices.
 */
struct qinit tcp_rinitv4 = {
	NULL, (pfi_t)tcp_rsrv, tcp_openv4, tcp_tpi_close, NULL, &tcp_rinfo
};

struct qinit tcp_rinitv6 = {
	NULL, (pfi_t)tcp_rsrv, tcp_openv6, tcp_tpi_close, NULL, &tcp_rinfo
};

struct qinit tcp_winit = {
	(pfi_t)tcp_wput, (pfi_t)tcp_wsrv, NULL, NULL, NULL, &tcp_winfo
};

/* Initial entry point for TCP in socket mode. */
struct qinit tcp_sock_winit = {
	(pfi_t)tcp_wput_sock, (pfi_t)tcp_wsrv, NULL, NULL, NULL, &tcp_winfo
};

/* TCP entry point during fallback */
struct qinit tcp_fallback_sock_winit = {
	(pfi_t)tcp_wput_fallback, NULL, NULL, NULL, NULL, &tcp_winfo
};

/*
 * Entry points for TCP as a acceptor STREAM opened by sockfs when doing
 * an accept. Avoid allocating data structures since eager has already
 * been created.
 */
struct qinit tcp_acceptor_rinit = {
	NULL, (pfi_t)tcp_rsrv, NULL, tcp_tpi_close_accept, NULL, &tcp_winfo
};

struct qinit tcp_acceptor_winit = {
	(pfi_t)tcp_tpi_accept, NULL, NULL, NULL, NULL, &tcp_winfo
};

/* For AF_INET aka /dev/tcp */
struct streamtab tcpinfov4 = {
	&tcp_rinitv4, &tcp_winit
};

/* For AF_INET6 aka /dev/tcp6 */
struct streamtab tcpinfov6 = {
	&tcp_rinitv6, &tcp_winit
};

sock_downcalls_t sock_tcp_downcalls;

/* Setable only in /etc/system. Move to ndd? */
boolean_t tcp_icmp_source_quench = B_FALSE;

/*
 * Following assumes TPI alignment requirements stay along 32 bit
 * boundaries
 */
#define	ROUNDUP32(x) \
	(((x) + (sizeof (int32_t) - 1)) & ~(sizeof (int32_t) - 1))

/* Template for response to info request. */
static struct T_info_ack tcp_g_t_info_ack = {
	T_INFO_ACK,		/* PRIM_type */
	0,			/* TSDU_size */
	T_INFINITE,		/* ETSDU_size */
	T_INVALID,		/* CDATA_size */
	T_INVALID,		/* DDATA_size */
	sizeof (sin_t),		/* ADDR_size */
	0,			/* OPT_size - not initialized here */
	TIDUSZ,			/* TIDU_size */
	T_COTS_ORD,		/* SERV_type */
	TCPS_IDLE,		/* CURRENT_state */
	(XPG4_1|EXPINLINE)	/* PROVIDER_flag */
};

static struct T_info_ack tcp_g_t_info_ack_v6 = {
	T_INFO_ACK,		/* PRIM_type */
	0,			/* TSDU_size */
	T_INFINITE,		/* ETSDU_size */
	T_INVALID,		/* CDATA_size */
	T_INVALID,		/* DDATA_size */
	sizeof (sin6_t),	/* ADDR_size */
	0,			/* OPT_size - not initialized here */
	TIDUSZ,		/* TIDU_size */
	T_COTS_ORD,		/* SERV_type */
	TCPS_IDLE,		/* CURRENT_state */
	(XPG4_1|EXPINLINE)	/* PROVIDER_flag */
};

#define	MS	1L
#define	SECONDS	(1000 * MS)
#define	MINUTES	(60 * SECONDS)
#define	HOURS	(60 * MINUTES)
#define	DAYS	(24 * HOURS)

#define	PARAM_MAX (~(uint32_t)0)

/* Max size IP datagram is 64k - 1 */
#define	TCP_MSS_MAX_IPV4 (IP_MAXPACKET - (sizeof (ipha_t) + sizeof (tcpha_t)))
#define	TCP_MSS_MAX_IPV6 (IP_MAXPACKET - (sizeof (ip6_t) + sizeof (tcpha_t)))
/* Max of the above */
#define	TCP_MSS_MAX	TCP_MSS_MAX_IPV4

/* Largest TCP port number */
#define	TCP_MAX_PORT	(64 * 1024 - 1)

/*
 * tcp_wroff_xtra is the extra space in front of TCP/IP header for link
 * layer header.  It has to be a multiple of 4.
 */
static tcpparam_t lcl_tcp_wroff_xtra_param = { 0, 256, 32, "tcp_wroff_xtra" };
#define	tcps_wroff_xtra	tcps_wroff_xtra_param->tcp_param_val

/*
 * All of these are alterable, within the min/max values given, at run time.
 * Note that the default value of "tcp_time_wait_interval" is four minutes,
 * per the TCP spec.
 */
/* BEGIN CSTYLED */
static tcpparam_t	lcl_tcp_param_arr[] = {
 /*min		max		value		name */
 { 1*SECONDS,	10*MINUTES,	1*MINUTES,	"tcp_time_wait_interval"},
 { 1,		PARAM_MAX,	128,		"tcp_conn_req_max_q" },
 { 0,		PARAM_MAX,	1024,		"tcp_conn_req_max_q0" },
 { 1,		1024,		1,		"tcp_conn_req_min" },
 { 0*MS,	20*SECONDS,	0*MS,		"tcp_conn_grace_period" },
 { 128,		(1<<30),	1024*1024,	"tcp_cwnd_max" },
 { 0,		10,		0,		"tcp_debug" },
 { 1024,	(32*1024),	1024,		"tcp_smallest_nonpriv_port"},
 { 1*SECONDS,	PARAM_MAX,	3*MINUTES,	"tcp_ip_abort_cinterval"},
 { 1*SECONDS,	PARAM_MAX,	3*MINUTES,	"tcp_ip_abort_linterval"},
 { 500*MS,	PARAM_MAX,	8*MINUTES,	"tcp_ip_abort_interval"},
 { 1*SECONDS,	PARAM_MAX,	10*SECONDS,	"tcp_ip_notify_cinterval"},
 { 500*MS,	PARAM_MAX,	10*SECONDS,	"tcp_ip_notify_interval"},
 { 1,		255,		64,		"tcp_ipv4_ttl"},
 { 10*SECONDS,	10*DAYS,	2*HOURS,	"tcp_keepalive_interval"},
 { 0,		100,		10,		"tcp_maxpsz_multiplier" },
 { 1,		TCP_MSS_MAX_IPV4, 536,		"tcp_mss_def_ipv4"},
 { 1,		TCP_MSS_MAX_IPV4, TCP_MSS_MAX_IPV4, "tcp_mss_max_ipv4"},
 { 1,		TCP_MSS_MAX,	108,		"tcp_mss_min"},
 { 1,		(64*1024)-1,	(4*1024)-1,	"tcp_naglim_def"},
 { 1*MS,	20*SECONDS,	3*SECONDS,	"tcp_rexmit_interval_initial"},
 { 1*MS,	2*HOURS,	60*SECONDS,	"tcp_rexmit_interval_max"},
 { 1*MS,	2*HOURS,	400*MS,		"tcp_rexmit_interval_min"},
 { 1*MS,	1*MINUTES,	100*MS,		"tcp_deferred_ack_interval" },
 { 0,		16,		0,		"tcp_snd_lowat_fraction" },
 { 0,		128000,		0,		"tcp_sth_rcv_hiwat" },
 { 0,		128000,		0,		"tcp_sth_rcv_lowat" },
 { 1,		10000,		3,		"tcp_dupack_fast_retransmit" },
 { 0,		1,		0,		"tcp_ignore_path_mtu" },
 { 1024,	TCP_MAX_PORT,	32*1024,	"tcp_smallest_anon_port"},
 { 1024,	TCP_MAX_PORT,	TCP_MAX_PORT,	"tcp_largest_anon_port"},
 { TCP_XMIT_LOWATER, (1<<30), TCP_XMIT_HIWATER,"tcp_xmit_hiwat"},
 { TCP_XMIT_LOWATER, (1<<30), TCP_XMIT_LOWATER,"tcp_xmit_lowat"},
 { TCP_RECV_LOWATER, (1<<30), TCP_RECV_HIWATER,"tcp_recv_hiwat"},
 { 1,		65536,		4,		"tcp_recv_hiwat_minmss"},
 { 1*SECONDS,	PARAM_MAX,	675*SECONDS,	"tcp_fin_wait_2_flush_interval"},
 { 8192,	(1<<30),	1024*1024,	"tcp_max_buf"},
/*
 * Question:  What default value should I set for tcp_strong_iss?
 */
 { 0,		2,		1,		"tcp_strong_iss"},
 { 0,		65536,		20,		"tcp_rtt_updates"},
 { 0,		1,		1,		"tcp_wscale_always"},
 { 0,		1,		0,		"tcp_tstamp_always"},
 { 0,		1,		1,		"tcp_tstamp_if_wscale"},
 { 0*MS,	2*HOURS,	0*MS,		"tcp_rexmit_interval_extra"},
 { 0,		16,		2,		"tcp_deferred_acks_max"},
 { 1,		16384,		4,		"tcp_slow_start_after_idle"},
 { 1,		4,		4,		"tcp_slow_start_initial"},
 { 0,		2,		2,		"tcp_sack_permitted"},
 { 0,		1,		1,		"tcp_compression_enabled"},
 { 0,		IPV6_MAX_HOPS,	IPV6_DEFAULT_HOPS,	"tcp_ipv6_hoplimit"},
 { 1,		TCP_MSS_MAX_IPV6, 1220,		"tcp_mss_def_ipv6"},
 { 1,		TCP_MSS_MAX_IPV6, TCP_MSS_MAX_IPV6, "tcp_mss_max_ipv6"},
 { 0,		1,		0,		"tcp_rev_src_routes"},
 { 10*MS,	500*MS,		50*MS,		"tcp_local_dack_interval"},
 { 0,		16,		8,		"tcp_local_dacks_max"},
 { 0,		2,		1,		"tcp_ecn_permitted"},
 { 0,		1,		1,		"tcp_rst_sent_rate_enabled"},
 { 0,		PARAM_MAX,	40,		"tcp_rst_sent_rate"},
 { 0,		100*MS,		50*MS,		"tcp_push_timer_interval"},
 { 0,		1,		0,		"tcp_use_smss_as_mss_opt"},
 { 0,		PARAM_MAX,	8*MINUTES,	"tcp_keepalive_abort_interval"},
 { 0,		1,		0,		"tcp_dev_flow_ctl"},
};
/* END CSTYLED */

/* Round up the value to the nearest mss. */
#define	MSS_ROUNDUP(value, mss)		((((value) - 1) / (mss) + 1) * (mss))

/*
 * Set ECN capable transport (ECT) code point in IP header.
 *
 * Note that there are 2 ECT code points '01' and '10', which are called
 * ECT(1) and ECT(0) respectively.  Here we follow the original ECT code
 * point ECT(0) for TCP as described in RFC 2481.
 */
#define	SET_ECT(tcp, iph) \
	if ((tcp)->tcp_connp->conn_ipversion == IPV4_VERSION) { \
		/* We need to clear the code point first. */ \
		((ipha_t *)(iph))->ipha_type_of_service &= 0xFC; \
		((ipha_t *)(iph))->ipha_type_of_service |= IPH_ECN_ECT0; \
	} else { \
		((ip6_t *)(iph))->ip6_vcf &= htonl(0xFFCFFFFF); \
		((ip6_t *)(iph))->ip6_vcf |= htonl(IPH_ECN_ECT0 << 20); \
	}

/*
 * The format argument to pass to tcp_display().
 * DISP_PORT_ONLY means that the returned string has only port info.
 * DISP_ADDR_AND_PORT means that the returned string also contains the
 * remote and local IP address.
 */
#define	DISP_PORT_ONLY		1
#define	DISP_ADDR_AND_PORT	2

#define	IS_VMLOANED_MBLK(mp) \
	(((mp)->b_datap->db_struioflag & STRUIO_ZC) != 0)

uint32_t do_tcpzcopy = 1;		/* 0: disable, 1: enable, 2: force */

/*
 * Forces all connections to obey the value of the tcps_maxpsz_multiplier
 * tunable settable via NDD.  Otherwise, the per-connection behavior is
 * determined dynamically during tcp_set_destination(), which is the default.
 */
boolean_t tcp_static_maxpsz = B_FALSE;

/* Setable in /etc/system */
/* If set to 0, pick ephemeral port sequentially; otherwise randomly. */
uint32_t tcp_random_anon_port = 1;

/*
 * To reach to an eager in Q0 which can be dropped due to an incoming
 * new SYN request when Q0 is full, a new doubly linked list is
 * introduced. This list allows to select an eager from Q0 in O(1) time.
 * This is needed to avoid spending too much time walking through the
 * long list of eagers in Q0 when tcp_drop_q0() is called. Each member of
 * this new list has to be a member of Q0.
 * This list is headed by listener's tcp_t. When the list is empty,
 * both the pointers - tcp_eager_next_drop_q0 and tcp_eager_prev_drop_q0,
 * of listener's tcp_t point to listener's tcp_t itself.
 *
 * Given an eager in Q0 and a listener, MAKE_DROPPABLE() puts the eager
 * in the list. MAKE_UNDROPPABLE() takes the eager out of the list.
 * These macros do not affect the eager's membership to Q0.
 */


#define	MAKE_DROPPABLE(listener, eager)					\
	if ((eager)->tcp_eager_next_drop_q0 == NULL) {			\
		(listener)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0\
		    = (eager);						\
		(eager)->tcp_eager_prev_drop_q0 = (listener);		\
		(eager)->tcp_eager_next_drop_q0 =			\
		    (listener)->tcp_eager_next_drop_q0;			\
		(listener)->tcp_eager_next_drop_q0 = (eager);		\
	}

#define	MAKE_UNDROPPABLE(eager)						\
	if ((eager)->tcp_eager_next_drop_q0 != NULL) {			\
		(eager)->tcp_eager_next_drop_q0->tcp_eager_prev_drop_q0	\
		    = (eager)->tcp_eager_prev_drop_q0;			\
		(eager)->tcp_eager_prev_drop_q0->tcp_eager_next_drop_q0	\
		    = (eager)->tcp_eager_next_drop_q0;			\
		(eager)->tcp_eager_prev_drop_q0 = NULL;			\
		(eager)->tcp_eager_next_drop_q0 = NULL;			\
	}

/*
 * If tcp_drop_ack_unsent_cnt is greater than 0, when TCP receives more
 * than tcp_drop_ack_unsent_cnt number of ACKs which acknowledge unsent
 * data, TCP will not respond with an ACK.  RFC 793 requires that
 * TCP responds with an ACK for such a bogus ACK.  By not following
 * the RFC, we prevent TCP from getting into an ACK storm if somehow
 * an attacker successfully spoofs an acceptable segment to our
 * peer; or when our peer is "confused."
 */
uint32_t tcp_drop_ack_unsent_cnt = 10;

/*
 * Hook functions to enable cluster networking
 * On non-clustered systems these vectors must always be NULL.
 */

void (*cl_inet_listen)(netstackid_t stack_id, uint8_t protocol,
			    sa_family_t addr_family, uint8_t *laddrp,
			    in_port_t lport, void *args) = NULL;
void (*cl_inet_unlisten)(netstackid_t stack_id, uint8_t protocol,
			    sa_family_t addr_family, uint8_t *laddrp,
			    in_port_t lport, void *args) = NULL;

int (*cl_inet_connect2)(netstackid_t stack_id, uint8_t protocol,
			    boolean_t is_outgoing,
			    sa_family_t addr_family,
			    uint8_t *laddrp, in_port_t lport,
			    uint8_t *faddrp, in_port_t fport,
			    void *args) = NULL;
void (*cl_inet_disconnect)(netstackid_t stack_id, uint8_t protocol,
			    sa_family_t addr_family, uint8_t *laddrp,
			    in_port_t lport, uint8_t *faddrp,
			    in_port_t fport, void *args) = NULL;


/*
 * int CL_INET_CONNECT(conn_t *cp, tcp_t *tcp, boolean_t is_outgoing, int err)
 */
#define	CL_INET_CONNECT(connp, is_outgoing, err) {		\
	(err) = 0;						\
	if (cl_inet_connect2 != NULL) {				\
		/*						\
		 * Running in cluster mode - register active connection	\
		 * information						\
		 */							\
		if ((connp)->conn_ipversion == IPV4_VERSION) {		\
			if ((connp)->conn_laddr_v4 != 0) {		\
				(err) = (*cl_inet_connect2)(		\
				    (connp)->conn_netstack->netstack_stackid,\
				    IPPROTO_TCP, is_outgoing, AF_INET,	\
				    (uint8_t *)(&((connp)->conn_laddr_v4)),\
				    (in_port_t)(connp)->conn_lport,	\
				    (uint8_t *)(&((connp)->conn_faddr_v4)),\
				    (in_port_t)(connp)->conn_fport, NULL); \
			}						\
		} else {						\
			if (!IN6_IS_ADDR_UNSPECIFIED(			\
			    &(connp)->conn_laddr_v6)) {			\
				(err) = (*cl_inet_connect2)(		\
				    (connp)->conn_netstack->netstack_stackid,\
				    IPPROTO_TCP, is_outgoing, AF_INET6,	\
				    (uint8_t *)(&((connp)->conn_laddr_v6)),\
				    (in_port_t)(connp)->conn_lport,	\
				    (uint8_t *)(&((connp)->conn_faddr_v6)), \
				    (in_port_t)(connp)->conn_fport, NULL); \
			}						\
		}							\
	}								\
}

#define	CL_INET_DISCONNECT(connp)	{				\
	if (cl_inet_disconnect != NULL) {				\
		/*							\
		 * Running in cluster mode - deregister active		\
		 * connection information				\
		 */							\
		if ((connp)->conn_ipversion == IPV4_VERSION) {		\
			if ((connp)->conn_laddr_v4 != 0) {		\
				(*cl_inet_disconnect)(			\
				    (connp)->conn_netstack->netstack_stackid,\
				    IPPROTO_TCP, AF_INET,		\
				    (uint8_t *)(&((connp)->conn_laddr_v4)),\
				    (in_port_t)(connp)->conn_lport,	\
				    (uint8_t *)(&((connp)->conn_faddr_v4)),\
				    (in_port_t)(connp)->conn_fport, NULL); \
			}						\
		} else {						\
			if (!IN6_IS_ADDR_UNSPECIFIED(			\
			    &(connp)->conn_laddr_v6)) {			\
				(*cl_inet_disconnect)(			\
				    (connp)->conn_netstack->netstack_stackid,\
				    IPPROTO_TCP, AF_INET6,		\
				    (uint8_t *)(&((connp)->conn_laddr_v6)),\
				    (in_port_t)(connp)->conn_lport,	\
				    (uint8_t *)(&((connp)->conn_faddr_v6)), \
				    (in_port_t)(connp)->conn_fport, NULL); \
			}						\
		}							\
	}								\
}

/*
 * Cluster networking hook for traversing current connection list.
 * This routine is used to extract the current list of live connections
 * which must continue to to be dispatched to this node.
 */
int cl_tcp_walk_list(netstackid_t stack_id,
    int (*callback)(cl_tcp_info_t *, void *), void *arg);

static int cl_tcp_walk_list_stack(int (*callback)(cl_tcp_info_t *, void *),
    void *arg, tcp_stack_t *tcps);

static void
tcp_set_recv_threshold(tcp_t *tcp, uint32_t new_rcvthresh)
{
	uint32_t default_threshold = SOCKET_RECVHIWATER >> 3;

	if (IPCL_IS_NONSTR(tcp->tcp_connp)) {
		conn_t *connp = tcp->tcp_connp;
		struct sock_proto_props sopp;

		/*
		 * only increase rcvthresh upto default_threshold
		 */
		if (new_rcvthresh > default_threshold)
			new_rcvthresh = default_threshold;

		sopp.sopp_flags = SOCKOPT_RCVTHRESH;
		sopp.sopp_rcvthresh = new_rcvthresh;

		(*connp->conn_upcalls->su_set_proto_props)
		    (connp->conn_upper_handle, &sopp);
	}
}
/*
 * Figure out the value of window scale opton.  Note that the rwnd is
 * ASSUMED to be rounded up to the nearest MSS before the calculation.
 * We cannot find the scale value and then do a round up of tcp_rwnd
 * because the scale value may not be correct after that.
 *
 * Set the compiler flag to make this function inline.
 */
static void
tcp_set_ws_value(tcp_t *tcp)
{
	int i;
	uint32_t rwnd = tcp->tcp_rwnd;

	for (i = 0; rwnd > TCP_MAXWIN && i < TCP_MAX_WINSHIFT;
	    i++, rwnd >>= 1)
		;
	tcp->tcp_rcv_ws = i;
}

/*
 * Remove a connection from the list of detached TIME_WAIT connections.
 * It returns B_FALSE if it can't remove the connection from the list
 * as the connection has already been removed from the list due to an
 * earlier call to tcp_time_wait_remove(); otherwise it returns B_TRUE.
 */
static boolean_t
tcp_time_wait_remove(tcp_t *tcp, tcp_squeue_priv_t *tcp_time_wait)
{
	boolean_t	locked = B_FALSE;

	if (tcp_time_wait == NULL) {
		tcp_time_wait = *((tcp_squeue_priv_t **)
		    squeue_getprivate(tcp->tcp_connp->conn_sqp, SQPRIVATE_TCP));
		mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
		locked = B_TRUE;
	} else {
		ASSERT(MUTEX_HELD(&tcp_time_wait->tcp_time_wait_lock));
	}

	if (tcp->tcp_time_wait_expire == 0) {
		ASSERT(tcp->tcp_time_wait_next == NULL);
		ASSERT(tcp->tcp_time_wait_prev == NULL);
		if (locked)
			mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
		return (B_FALSE);
	}
	ASSERT(TCP_IS_DETACHED(tcp));
	ASSERT(tcp->tcp_state == TCPS_TIME_WAIT);

	if (tcp == tcp_time_wait->tcp_time_wait_head) {
		ASSERT(tcp->tcp_time_wait_prev == NULL);
		tcp_time_wait->tcp_time_wait_head = tcp->tcp_time_wait_next;
		if (tcp_time_wait->tcp_time_wait_head != NULL) {
			tcp_time_wait->tcp_time_wait_head->tcp_time_wait_prev =
			    NULL;
		} else {
			tcp_time_wait->tcp_time_wait_tail = NULL;
		}
	} else if (tcp == tcp_time_wait->tcp_time_wait_tail) {
		ASSERT(tcp != tcp_time_wait->tcp_time_wait_head);
		ASSERT(tcp->tcp_time_wait_next == NULL);
		tcp_time_wait->tcp_time_wait_tail = tcp->tcp_time_wait_prev;
		ASSERT(tcp_time_wait->tcp_time_wait_tail != NULL);
		tcp_time_wait->tcp_time_wait_tail->tcp_time_wait_next = NULL;
	} else {
		ASSERT(tcp->tcp_time_wait_prev->tcp_time_wait_next == tcp);
		ASSERT(tcp->tcp_time_wait_next->tcp_time_wait_prev == tcp);
		tcp->tcp_time_wait_prev->tcp_time_wait_next =
		    tcp->tcp_time_wait_next;
		tcp->tcp_time_wait_next->tcp_time_wait_prev =
		    tcp->tcp_time_wait_prev;
	}
	tcp->tcp_time_wait_next = NULL;
	tcp->tcp_time_wait_prev = NULL;
	tcp->tcp_time_wait_expire = 0;

	if (locked)
		mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
	return (B_TRUE);
}

/*
 * Add a connection to the list of detached TIME_WAIT connections
 * and set its time to expire.
 */
static void
tcp_time_wait_append(tcp_t *tcp)
{
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	tcp_squeue_priv_t *tcp_time_wait =
	    *((tcp_squeue_priv_t **)squeue_getprivate(tcp->tcp_connp->conn_sqp,
	    SQPRIVATE_TCP));

	tcp_timers_stop(tcp);

	/* Freed above */
	ASSERT(tcp->tcp_timer_tid == 0);
	ASSERT(tcp->tcp_ack_tid == 0);

	/* must have happened at the time of detaching the tcp */
	ASSERT(tcp->tcp_ptpahn == NULL);
	ASSERT(tcp->tcp_flow_stopped == 0);
	ASSERT(tcp->tcp_time_wait_next == NULL);
	ASSERT(tcp->tcp_time_wait_prev == NULL);
	ASSERT(tcp->tcp_time_wait_expire == NULL);
	ASSERT(tcp->tcp_listener == NULL);

	tcp->tcp_time_wait_expire = ddi_get_lbolt();
	/*
	 * The value computed below in tcp->tcp_time_wait_expire may
	 * appear negative or wrap around. That is ok since our
	 * interest is only in the difference between the current lbolt
	 * value and tcp->tcp_time_wait_expire. But the value should not
	 * be zero, since it means the tcp is not in the TIME_WAIT list.
	 * The corresponding comparison in tcp_time_wait_collector() uses
	 * modular arithmetic.
	 */
	tcp->tcp_time_wait_expire +=
	    drv_usectohz(tcps->tcps_time_wait_interval * 1000);
	if (tcp->tcp_time_wait_expire == 0)
		tcp->tcp_time_wait_expire = 1;

	ASSERT(TCP_IS_DETACHED(tcp));
	ASSERT(tcp->tcp_state == TCPS_TIME_WAIT);
	ASSERT(tcp->tcp_time_wait_next == NULL);
	ASSERT(tcp->tcp_time_wait_prev == NULL);
	TCP_DBGSTAT(tcps, tcp_time_wait);

	mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
	if (tcp_time_wait->tcp_time_wait_head == NULL) {
		ASSERT(tcp_time_wait->tcp_time_wait_tail == NULL);
		tcp_time_wait->tcp_time_wait_head = tcp;
	} else {
		ASSERT(tcp_time_wait->tcp_time_wait_tail != NULL);
		ASSERT(tcp_time_wait->tcp_time_wait_tail->tcp_state ==
		    TCPS_TIME_WAIT);
		tcp_time_wait->tcp_time_wait_tail->tcp_time_wait_next = tcp;
		tcp->tcp_time_wait_prev = tcp_time_wait->tcp_time_wait_tail;
	}
	tcp_time_wait->tcp_time_wait_tail = tcp;
	mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
}

/* ARGSUSED */
void
tcp_timewait_output(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
{
	conn_t	*connp = (conn_t *)arg;
	tcp_t	*tcp = connp->conn_tcp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	ASSERT(tcp != NULL);
	if (tcp->tcp_state == TCPS_CLOSED) {
		return;
	}

	ASSERT((connp->conn_family == AF_INET &&
	    connp->conn_ipversion == IPV4_VERSION) ||
	    (connp->conn_family == AF_INET6 &&
	    (connp->conn_ipversion == IPV4_VERSION ||
	    connp->conn_ipversion == IPV6_VERSION)));
	ASSERT(!tcp->tcp_listener);

	TCP_STAT(tcps, tcp_time_wait_reap);
	ASSERT(TCP_IS_DETACHED(tcp));

	/*
	 * Because they have no upstream client to rebind or tcp_close()
	 * them later, we axe the connection here and now.
	 */
	tcp_close_detached(tcp);
}

/*
 * Remove cached/latched IPsec references.
 */
void
tcp_ipsec_cleanup(tcp_t *tcp)
{
	conn_t		*connp = tcp->tcp_connp;

	ASSERT(connp->conn_flags & IPCL_TCPCONN);

	if (connp->conn_latch != NULL) {
		IPLATCH_REFRELE(connp->conn_latch);
		connp->conn_latch = NULL;
	}
	if (connp->conn_latch_in_policy != NULL) {
		IPPOL_REFRELE(connp->conn_latch_in_policy);
		connp->conn_latch_in_policy = NULL;
	}
	if (connp->conn_latch_in_action != NULL) {
		IPACT_REFRELE(connp->conn_latch_in_action);
		connp->conn_latch_in_action = NULL;
	}
	if (connp->conn_policy != NULL) {
		IPPH_REFRELE(connp->conn_policy, connp->conn_netstack);
		connp->conn_policy = NULL;
	}
}

/*
 * Cleaup before placing on free list.
 * Disassociate from the netstack/tcp_stack_t since the freelist
 * is per squeue and not per netstack.
 */
void
tcp_cleanup(tcp_t *tcp)
{
	mblk_t		*mp;
	tcp_sack_info_t	*tcp_sack_info;
	conn_t		*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	netstack_t	*ns = tcps->tcps_netstack;
	mblk_t		*tcp_rsrv_mp;

	tcp_bind_hash_remove(tcp);

	/* Cleanup that which needs the netstack first */
	tcp_ipsec_cleanup(tcp);
	ixa_cleanup(connp->conn_ixa);

	if (connp->conn_ht_iphc != NULL) {
		kmem_free(connp->conn_ht_iphc, connp->conn_ht_iphc_allocated);
		connp->conn_ht_iphc = NULL;
		connp->conn_ht_iphc_allocated = 0;
		connp->conn_ht_iphc_len = 0;
		connp->conn_ht_ulp = NULL;
		connp->conn_ht_ulp_len = 0;
		tcp->tcp_ipha = NULL;
		tcp->tcp_ip6h = NULL;
		tcp->tcp_tcpha = NULL;
	}

	/* We clear any IP_OPTIONS and extension headers */
	ip_pkt_free(&connp->conn_xmit_ipp);

	tcp_free(tcp);

	/* Release any SSL context */
	if (tcp->tcp_kssl_ent != NULL) {
		kssl_release_ent(tcp->tcp_kssl_ent, NULL, KSSL_NO_PROXY);
		tcp->tcp_kssl_ent = NULL;
	}

	if (tcp->tcp_kssl_ctx != NULL) {
		kssl_release_ctx(tcp->tcp_kssl_ctx);
		tcp->tcp_kssl_ctx = NULL;
	}
	tcp->tcp_kssl_pending = B_FALSE;

	/*
	 * Since we will bzero the entire structure, we need to
	 * remove it and reinsert it in global hash list. We
	 * know the walkers can't get to this conn because we
	 * had set CONDEMNED flag earlier and checked reference
	 * under conn_lock so walker won't pick it and when we
	 * go the ipcl_globalhash_remove() below, no walker
	 * can get to it.
	 */
	ipcl_globalhash_remove(connp);

	/* Save some state */
	mp = tcp->tcp_timercache;

	tcp_sack_info = tcp->tcp_sack_info;
	tcp_rsrv_mp = tcp->tcp_rsrv_mp;

	if (connp->conn_cred != NULL) {
		crfree(connp->conn_cred);
		connp->conn_cred = NULL;
	}
	ipcl_conn_cleanup(connp);
	connp->conn_flags = IPCL_TCPCONN;

	/*
	 * Now it is safe to decrement the reference counts.
	 * This might be the last reference on the netstack
	 * in which case it will cause the freeing of the IP Instance.
	 */
	connp->conn_netstack = NULL;
	connp->conn_ixa->ixa_ipst = NULL;
	netstack_rele(ns);
	ASSERT(tcps != NULL);
	tcp->tcp_tcps = NULL;

	bzero(tcp, sizeof (tcp_t));

	/* restore the state */
	tcp->tcp_timercache = mp;

	tcp->tcp_sack_info = tcp_sack_info;
	tcp->tcp_rsrv_mp = tcp_rsrv_mp;

	tcp->tcp_connp = connp;

	ASSERT(connp->conn_tcp == tcp);
	ASSERT(connp->conn_flags & IPCL_TCPCONN);
	connp->conn_state_flags = CONN_INCIPIENT;
	ASSERT(connp->conn_proto == IPPROTO_TCP);
	ASSERT(connp->conn_ref == 1);
}

/*
 * Blows away all tcps whose TIME_WAIT has expired. List traversal
 * is done forwards from the head.
 * This walks all stack instances since
 * tcp_time_wait remains global across all stacks.
 */
/* ARGSUSED */
void
tcp_time_wait_collector(void *arg)
{
	tcp_t *tcp;
	clock_t now;
	mblk_t *mp;
	conn_t *connp;
	kmutex_t *lock;
	boolean_t removed;

	squeue_t *sqp = (squeue_t *)arg;
	tcp_squeue_priv_t *tcp_time_wait =
	    *((tcp_squeue_priv_t **)squeue_getprivate(sqp, SQPRIVATE_TCP));

	mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
	tcp_time_wait->tcp_time_wait_tid = 0;

	if (tcp_time_wait->tcp_free_list != NULL &&
	    tcp_time_wait->tcp_free_list->tcp_in_free_list == B_TRUE) {
		TCP_G_STAT(tcp_freelist_cleanup);
		while ((tcp = tcp_time_wait->tcp_free_list) != NULL) {
			tcp_time_wait->tcp_free_list = tcp->tcp_time_wait_next;
			tcp->tcp_time_wait_next = NULL;
			tcp_time_wait->tcp_free_list_cnt--;
			ASSERT(tcp->tcp_tcps == NULL);
			CONN_DEC_REF(tcp->tcp_connp);
		}
		ASSERT(tcp_time_wait->tcp_free_list_cnt == 0);
	}

	/*
	 * In order to reap time waits reliably, we should use a
	 * source of time that is not adjustable by the user -- hence
	 * the call to ddi_get_lbolt().
	 */
	now = ddi_get_lbolt();
	while ((tcp = tcp_time_wait->tcp_time_wait_head) != NULL) {
		/*
		 * Compare times using modular arithmetic, since
		 * lbolt can wrapover.
		 */
		if ((now - tcp->tcp_time_wait_expire) < 0) {
			break;
		}

		removed = tcp_time_wait_remove(tcp, tcp_time_wait);
		ASSERT(removed);

		connp = tcp->tcp_connp;
		ASSERT(connp->conn_fanout != NULL);
		lock = &connp->conn_fanout->connf_lock;
		/*
		 * This is essentially a TW reclaim fast path optimization for
		 * performance where the timewait collector checks under the
		 * fanout lock (so that no one else can get access to the
		 * conn_t) that the refcnt is 2 i.e. one for TCP and one for
		 * the classifier hash list. If ref count is indeed 2, we can
		 * just remove the conn under the fanout lock and avoid
		 * cleaning up the conn under the squeue, provided that
		 * clustering callbacks are not enabled. If clustering is
		 * enabled, we need to make the clustering callback before
		 * setting the CONDEMNED flag and after dropping all locks and
		 * so we forego this optimization and fall back to the slow
		 * path. Also please see the comments in tcp_closei_local
		 * regarding the refcnt logic.
		 *
		 * Since we are holding the tcp_time_wait_lock, its better
		 * not to block on the fanout_lock because other connections
		 * can't add themselves to time_wait list. So we do a
		 * tryenter instead of mutex_enter.
		 */
		if (mutex_tryenter(lock)) {
			mutex_enter(&connp->conn_lock);
			if ((connp->conn_ref == 2) &&
			    (cl_inet_disconnect == NULL)) {
				ipcl_hash_remove_locked(connp,
				    connp->conn_fanout);
				/*
				 * Set the CONDEMNED flag now itself so that
				 * the refcnt cannot increase due to any
				 * walker.
				 */
				connp->conn_state_flags |= CONN_CONDEMNED;
				mutex_exit(lock);
				mutex_exit(&connp->conn_lock);
				if (tcp_time_wait->tcp_free_list_cnt <
				    tcp_free_list_max_cnt) {
					/* Add to head of tcp_free_list */
					mutex_exit(
					    &tcp_time_wait->tcp_time_wait_lock);
					tcp_cleanup(tcp);
					ASSERT(connp->conn_latch == NULL);
					ASSERT(connp->conn_policy == NULL);
					ASSERT(tcp->tcp_tcps == NULL);
					ASSERT(connp->conn_netstack == NULL);

					mutex_enter(
					    &tcp_time_wait->tcp_time_wait_lock);
					tcp->tcp_time_wait_next =
					    tcp_time_wait->tcp_free_list;
					tcp_time_wait->tcp_free_list = tcp;
					tcp_time_wait->tcp_free_list_cnt++;
					continue;
				} else {
					/* Do not add to tcp_free_list */
					mutex_exit(
					    &tcp_time_wait->tcp_time_wait_lock);
					tcp_bind_hash_remove(tcp);
					ixa_cleanup(tcp->tcp_connp->conn_ixa);
					tcp_ipsec_cleanup(tcp);
					CONN_DEC_REF(tcp->tcp_connp);
				}
			} else {
				CONN_INC_REF_LOCKED(connp);
				mutex_exit(lock);
				mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
				mutex_exit(&connp->conn_lock);
				/*
				 * We can reuse the closemp here since conn has
				 * detached (otherwise we wouldn't even be in
				 * time_wait list). tcp_closemp_used can safely
				 * be changed without taking a lock as no other
				 * thread can concurrently access it at this
				 * point in the connection lifecycle.
				 */

				if (tcp->tcp_closemp.b_prev == NULL)
					tcp->tcp_closemp_used = B_TRUE;
				else
					cmn_err(CE_PANIC,
					    "tcp_timewait_collector: "
					    "concurrent use of tcp_closemp: "
					    "connp %p tcp %p\n", (void *)connp,
					    (void *)tcp);

				TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15);
				mp = &tcp->tcp_closemp;
				SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
				    tcp_timewait_output, connp, NULL,
				    SQ_FILL, SQTAG_TCP_TIMEWAIT);
			}
		} else {
			mutex_enter(&connp->conn_lock);
			CONN_INC_REF_LOCKED(connp);
			mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
			mutex_exit(&connp->conn_lock);
			/*
			 * We can reuse the closemp here since conn has
			 * detached (otherwise we wouldn't even be in
			 * time_wait list). tcp_closemp_used can safely
			 * be changed without taking a lock as no other
			 * thread can concurrently access it at this
			 * point in the connection lifecycle.
			 */

			if (tcp->tcp_closemp.b_prev == NULL)
				tcp->tcp_closemp_used = B_TRUE;
			else
				cmn_err(CE_PANIC, "tcp_timewait_collector: "
				    "concurrent use of tcp_closemp: "
				    "connp %p tcp %p\n", (void *)connp,
				    (void *)tcp);

			TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15);
			mp = &tcp->tcp_closemp;
			SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
			    tcp_timewait_output, connp, NULL,
			    SQ_FILL, SQTAG_TCP_TIMEWAIT);
		}
		mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
	}

	if (tcp_time_wait->tcp_free_list != NULL)
		tcp_time_wait->tcp_free_list->tcp_in_free_list = B_TRUE;

	tcp_time_wait->tcp_time_wait_tid =
	    timeout_generic(CALLOUT_NORMAL, tcp_time_wait_collector, sqp,
	    TICK_TO_NSEC(TCP_TIME_WAIT_DELAY), CALLOUT_TCP_RESOLUTION,
	    CALLOUT_FLAG_ROUNDUP);
	mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
}

/*
 * Reply to a clients T_CONN_RES TPI message. This function
 * is used only for TLI/XTI listener. Sockfs sends T_CONN_RES
 * on the acceptor STREAM and processed in tcp_accept_common().
 * Read the block comment on top of tcp_input_listener().
 */
static void
tcp_tli_accept(tcp_t *listener, mblk_t *mp)
{
	tcp_t		*acceptor;
	tcp_t		*eager;
	tcp_t   	*tcp;
	struct T_conn_res	*tcr;
	t_uscalar_t	acceptor_id;
	t_scalar_t	seqnum;
	mblk_t		*discon_mp = NULL;
	mblk_t		*ok_mp;
	mblk_t		*mp1;
	tcp_stack_t	*tcps = listener->tcp_tcps;
	conn_t		*econnp;

	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tcr)) {
		tcp_err_ack(listener, mp, TPROTO, 0);
		return;
	}
	tcr = (struct T_conn_res *)mp->b_rptr;

	/*
	 * Under ILP32 the stream head points tcr->ACCEPTOR_id at the
	 * read side queue of the streams device underneath us i.e. the
	 * read side queue of 'ip'. Since we can't deference QUEUE_ptr we
	 * look it up in the queue_hash.  Under LP64 it sends down the
	 * minor_t of the accepting endpoint.
	 *
	 * Once the acceptor/eager are modified (in tcp_accept_swap) the
	 * fanout hash lock is held.
	 * This prevents any thread from entering the acceptor queue from
	 * below (since it has not been hard bound yet i.e. any inbound
	 * packets will arrive on the listener conn_t and
	 * go through the classifier).
	 * The CONN_INC_REF will prevent the acceptor from closing.
	 *
	 * XXX It is still possible for a tli application to send down data
	 * on the accepting stream while another thread calls t_accept.
	 * This should not be a problem for well-behaved applications since
	 * the T_OK_ACK is sent after the queue swapping is completed.
	 *
	 * If the accepting fd is the same as the listening fd, avoid
	 * queue hash lookup since that will return an eager listener in a
	 * already established state.
	 */
	acceptor_id = tcr->ACCEPTOR_id;
	mutex_enter(&listener->tcp_eager_lock);
	if (listener->tcp_acceptor_id == acceptor_id) {
		eager = listener->tcp_eager_next_q;
		/* only count how many T_CONN_INDs so don't count q0 */
		if ((listener->tcp_conn_req_cnt_q != 1) ||
		    (eager->tcp_conn_req_seqnum != tcr->SEQ_number)) {
			mutex_exit(&listener->tcp_eager_lock);
			tcp_err_ack(listener, mp, TBADF, 0);
			return;
		}
		if (listener->tcp_conn_req_cnt_q0 != 0) {
			/* Throw away all the eagers on q0. */
			tcp_eager_cleanup(listener, 1);
		}
		if (listener->tcp_syn_defense) {
			listener->tcp_syn_defense = B_FALSE;
			if (listener->tcp_ip_addr_cache != NULL) {
				kmem_free(listener->tcp_ip_addr_cache,
				    IP_ADDR_CACHE_SIZE * sizeof (ipaddr_t));
				listener->tcp_ip_addr_cache = NULL;
			}
		}
		/*
		 * Transfer tcp_conn_req_max to the eager so that when
		 * a disconnect occurs we can revert the endpoint to the
		 * listen state.
		 */
		eager->tcp_conn_req_max = listener->tcp_conn_req_max;
		ASSERT(listener->tcp_conn_req_cnt_q0 == 0);
		/*
		 * Get a reference on the acceptor just like the
		 * tcp_acceptor_hash_lookup below.
		 */
		acceptor = listener;
		CONN_INC_REF(acceptor->tcp_connp);
	} else {
		acceptor = tcp_acceptor_hash_lookup(acceptor_id, tcps);
		if (acceptor == NULL) {
			if (listener->tcp_connp->conn_debug) {
				(void) strlog(TCP_MOD_ID, 0, 1,
				    SL_ERROR|SL_TRACE,
				    "tcp_accept: did not find acceptor 0x%x\n",
				    acceptor_id);
			}
			mutex_exit(&listener->tcp_eager_lock);
			tcp_err_ack(listener, mp, TPROVMISMATCH, 0);
			return;
		}
		/*
		 * Verify acceptor state. The acceptable states for an acceptor
		 * include TCPS_IDLE and TCPS_BOUND.
		 */
		switch (acceptor->tcp_state) {
		case TCPS_IDLE:
			/* FALLTHRU */
		case TCPS_BOUND:
			break;
		default:
			CONN_DEC_REF(acceptor->tcp_connp);
			mutex_exit(&listener->tcp_eager_lock);
			tcp_err_ack(listener, mp, TOUTSTATE, 0);
			return;
		}
	}

	/* The listener must be in TCPS_LISTEN */
	if (listener->tcp_state != TCPS_LISTEN) {
		CONN_DEC_REF(acceptor->tcp_connp);
		mutex_exit(&listener->tcp_eager_lock);
		tcp_err_ack(listener, mp, TOUTSTATE, 0);
		return;
	}

	/*
	 * Rendezvous with an eager connection request packet hanging off
	 * 'tcp' that has the 'seqnum' tag.  We tagged the detached open
	 * tcp structure when the connection packet arrived in
	 * tcp_input_listener().
	 */
	seqnum = tcr->SEQ_number;
	eager = listener;
	do {
		eager = eager->tcp_eager_next_q;
		if (eager == NULL) {
			CONN_DEC_REF(acceptor->tcp_connp);
			mutex_exit(&listener->tcp_eager_lock);
			tcp_err_ack(listener, mp, TBADSEQ, 0);
			return;
		}
	} while (eager->tcp_conn_req_seqnum != seqnum);
	mutex_exit(&listener->tcp_eager_lock);

	/*
	 * At this point, both acceptor and listener have 2 ref
	 * that they begin with. Acceptor has one additional ref
	 * we placed in lookup while listener has 3 additional
	 * ref for being behind the squeue (tcp_accept() is
	 * done on listener's squeue); being in classifier hash;
	 * and eager's ref on listener.
	 */
	ASSERT(listener->tcp_connp->conn_ref >= 5);
	ASSERT(acceptor->tcp_connp->conn_ref >= 3);

	/*
	 * The eager at this point is set in its own squeue and
	 * could easily have been killed (tcp_accept_finish will
	 * deal with that) because of a TH_RST so we can only
	 * ASSERT for a single ref.
	 */
	ASSERT(eager->tcp_connp->conn_ref >= 1);

	/*
	 * Pre allocate the discon_ind mblk also. tcp_accept_finish will
	 * use it if something failed.
	 */
	discon_mp = allocb(MAX(sizeof (struct T_discon_ind),
	    sizeof (struct stroptions)), BPRI_HI);
	if (discon_mp == NULL) {
		CONN_DEC_REF(acceptor->tcp_connp);
		CONN_DEC_REF(eager->tcp_connp);
		tcp_err_ack(listener, mp, TSYSERR, ENOMEM);
		return;
	}

	econnp = eager->tcp_connp;

	/* Hold a copy of mp, in case reallocb fails */
	if ((mp1 = copymsg(mp)) == NULL) {
		CONN_DEC_REF(acceptor->tcp_connp);
		CONN_DEC_REF(eager->tcp_connp);
		freemsg(discon_mp);
		tcp_err_ack(listener, mp, TSYSERR, ENOMEM);
		return;
	}

	tcr = (struct T_conn_res *)mp1->b_rptr;

	/*
	 * This is an expanded version of mi_tpi_ok_ack_alloc()
	 * which allocates a larger mblk and appends the new
	 * local address to the ok_ack.  The address is copied by
	 * soaccept() for getsockname().
	 */
	{
		int extra;

		extra = (econnp->conn_family == AF_INET) ?
		    sizeof (sin_t) : sizeof (sin6_t);

		/*
		 * Try to re-use mp, if possible.  Otherwise, allocate
		 * an mblk and return it as ok_mp.  In any case, mp
		 * is no longer usable upon return.
		 */
		if ((ok_mp = mi_tpi_ok_ack_alloc_extra(mp, extra)) == NULL) {
			CONN_DEC_REF(acceptor->tcp_connp);
			CONN_DEC_REF(eager->tcp_connp);
			freemsg(discon_mp);
			/* Original mp has been freed by now, so use mp1 */
			tcp_err_ack(listener, mp1, TSYSERR, ENOMEM);
			return;
		}

		mp = NULL;	/* We should never use mp after this point */

		switch (extra) {
		case sizeof (sin_t): {
			sin_t *sin = (sin_t *)ok_mp->b_wptr;

			ok_mp->b_wptr += extra;
			sin->sin_family = AF_INET;
			sin->sin_port = econnp->conn_lport;
			sin->sin_addr.s_addr = econnp->conn_laddr_v4;
			break;
		}
		case sizeof (sin6_t): {
			sin6_t *sin6 = (sin6_t *)ok_mp->b_wptr;

			ok_mp->b_wptr += extra;
			sin6->sin6_family = AF_INET6;
			sin6->sin6_port = econnp->conn_lport;
			sin6->sin6_addr = econnp->conn_laddr_v6;
			sin6->sin6_flowinfo = econnp->conn_flowinfo;
			if (IN6_IS_ADDR_LINKSCOPE(&econnp->conn_laddr_v6) &&
			    (econnp->conn_ixa->ixa_flags & IXAF_SCOPEID_SET)) {
				sin6->sin6_scope_id =
				    econnp->conn_ixa->ixa_scopeid;
			} else {
				sin6->sin6_scope_id = 0;
			}
			sin6->__sin6_src_id = 0;
			break;
		}
		default:
			break;
		}
		ASSERT(ok_mp->b_wptr <= ok_mp->b_datap->db_lim);
	}

	/*
	 * If there are no options we know that the T_CONN_RES will
	 * succeed. However, we can't send the T_OK_ACK upstream until
	 * the tcp_accept_swap is done since it would be dangerous to
	 * let the application start using the new fd prior to the swap.
	 */
	tcp_accept_swap(listener, acceptor, eager);

	/*
	 * tcp_accept_swap unlinks eager from listener but does not drop
	 * the eager's reference on the listener.
	 */
	ASSERT(eager->tcp_listener == NULL);
	ASSERT(listener->tcp_connp->conn_ref >= 5);

	/*
	 * The eager is now associated with its own queue. Insert in
	 * the hash so that the connection can be reused for a future
	 * T_CONN_RES.
	 */
	tcp_acceptor_hash_insert(acceptor_id, eager);

	/*
	 * We now do the processing of options with T_CONN_RES.
	 * We delay till now since we wanted to have queue to pass to
	 * option processing routines that points back to the right
	 * instance structure which does not happen until after
	 * tcp_accept_swap().
	 *
	 * Note:
	 * The sanity of the logic here assumes that whatever options
	 * are appropriate to inherit from listner=>eager are done
	 * before this point, and whatever were to be overridden (or not)
	 * in transfer logic from eager=>acceptor in tcp_accept_swap().
	 * [ Warning: acceptor endpoint can have T_OPTMGMT_REQ done to it
	 *   before its ACCEPTOR_id comes down in T_CONN_RES ]
	 * This may not be true at this point in time but can be fixed
	 * independently. This option processing code starts with
	 * the instantiated acceptor instance and the final queue at
	 * this point.
	 */

	if (tcr->OPT_length != 0) {
		/* Options to process */
		int t_error = 0;
		int sys_error = 0;
		int do_disconnect = 0;

		if (tcp_conprim_opt_process(eager, mp1,
		    &do_disconnect, &t_error, &sys_error) < 0) {
			eager->tcp_accept_error = 1;
			if (do_disconnect) {
				/*
				 * An option failed which does not allow
				 * connection to be accepted.
				 *
				 * We allow T_CONN_RES to succeed and
				 * put a T_DISCON_IND on the eager queue.
				 */
				ASSERT(t_error == 0 && sys_error == 0);
				eager->tcp_send_discon_ind = 1;
			} else {
				ASSERT(t_error != 0);
				freemsg(ok_mp);
				/*
				 * Original mp was either freed or set
				 * to ok_mp above, so use mp1 instead.
				 */
				tcp_err_ack(listener, mp1, t_error, sys_error);
				goto finish;
			}
		}
		/*
		 * Most likely success in setting options (except if
		 * eager->tcp_send_discon_ind set).
		 * mp1 option buffer represented by OPT_length/offset
		 * potentially modified and contains results of setting
		 * options at this point
		 */
	}

	/* We no longer need mp1, since all options processing has passed */
	freemsg(mp1);

	putnext(listener->tcp_connp->conn_rq, ok_mp);

	mutex_enter(&listener->tcp_eager_lock);
	if (listener->tcp_eager_prev_q0->tcp_conn_def_q0) {
		tcp_t	*tail;
		mblk_t	*conn_ind;

		/*
		 * This path should not be executed if listener and
		 * acceptor streams are the same.
		 */
		ASSERT(listener != acceptor);

		tcp = listener->tcp_eager_prev_q0;
		/*
		 * listener->tcp_eager_prev_q0 points to the TAIL of the
		 * deferred T_conn_ind queue. We need to get to the head of
		 * the queue in order to send up T_conn_ind the same order as
		 * how the 3WHS is completed.
		 */
		while (tcp != listener) {
			if (!tcp->tcp_eager_prev_q0->tcp_conn_def_q0)
				break;
			else
				tcp = tcp->tcp_eager_prev_q0;
		}
		ASSERT(tcp != listener);
		conn_ind = tcp->tcp_conn.tcp_eager_conn_ind;
		ASSERT(conn_ind != NULL);
		tcp->tcp_conn.tcp_eager_conn_ind = NULL;

		/* Move from q0 to q */
		ASSERT(listener->tcp_conn_req_cnt_q0 > 0);
		listener->tcp_conn_req_cnt_q0--;
		listener->tcp_conn_req_cnt_q++;
		tcp->tcp_eager_next_q0->tcp_eager_prev_q0 =
		    tcp->tcp_eager_prev_q0;
		tcp->tcp_eager_prev_q0->tcp_eager_next_q0 =
		    tcp->tcp_eager_next_q0;
		tcp->tcp_eager_prev_q0 = NULL;
		tcp->tcp_eager_next_q0 = NULL;
		tcp->tcp_conn_def_q0 = B_FALSE;

		/* Make sure the tcp isn't in the list of droppables */
		ASSERT(tcp->tcp_eager_next_drop_q0 == NULL &&
		    tcp->tcp_eager_prev_drop_q0 == NULL);

		/*
		 * Insert at end of the queue because sockfs sends
		 * down T_CONN_RES in chronological order. Leaving
		 * the older conn indications at front of the queue
		 * helps reducing search time.
		 */
		tail = listener->tcp_eager_last_q;
		if (tail != NULL)
			tail->tcp_eager_next_q = tcp;
		else
			listener->tcp_eager_next_q = tcp;
		listener->tcp_eager_last_q = tcp;
		tcp->tcp_eager_next_q = NULL;
		mutex_exit(&listener->tcp_eager_lock);
		putnext(tcp->tcp_connp->conn_rq, conn_ind);
	} else {
		mutex_exit(&listener->tcp_eager_lock);
	}

	/*
	 * Done with the acceptor - free it
	 *
	 * Note: from this point on, no access to listener should be made
	 * as listener can be equal to acceptor.
	 */
finish:
	ASSERT(acceptor->tcp_detached);
	acceptor->tcp_connp->conn_rq = NULL;
	ASSERT(!IPCL_IS_NONSTR(acceptor->tcp_connp));
	acceptor->tcp_connp->conn_wq = NULL;
	(void) tcp_clean_death(acceptor, 0, 2);
	CONN_DEC_REF(acceptor->tcp_connp);

	/*
	 * We pass discon_mp to tcp_accept_finish to get on the right squeue.
	 *
	 * It will update the setting for sockfs/stream head and also take
	 * care of any data that arrived before accept() wad called.
	 * In case we already received a FIN then tcp_accept_finish will send up
	 * the ordrel. It will also send up a window update if the window
	 * has opened up.
	 */

	/*
	 * XXX: we currently have a problem if XTI application closes the
	 * acceptor stream in between. This problem exists in on10-gate also
	 * and is well know but nothing can be done short of major rewrite
	 * to fix it. Now it is possible to take care of it by assigning TLI/XTI
	 * eager same squeue as listener (we can distinguish non socket
	 * listeners at the time of handling a SYN in tcp_input_listener)
	 * and do most of the work that tcp_accept_finish does here itself
	 * and then get behind the acceptor squeue to access the acceptor
	 * queue.
	 */
	/*
	 * We already have a ref on tcp so no need to do one before squeue_enter
	 */
	SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, discon_mp,
	    tcp_accept_finish, eager->tcp_connp, NULL, SQ_FILL,
	    SQTAG_TCP_ACCEPT_FINISH);
}

/*
 * Swap information between the eager and acceptor for a TLI/XTI client.
 * The sockfs accept is done on the acceptor stream and control goes
 * through tcp_tli_accept() and tcp_accept()/tcp_accept_swap() is not
 * called. In either case, both the eager and listener are in their own
 * perimeter (squeue) and the code has to deal with potential race.
 *
 * See the block comment on top of tcp_accept() and tcp_tli_accept().
 */
static void
tcp_accept_swap(tcp_t *listener, tcp_t *acceptor, tcp_t *eager)
{
	conn_t	*econnp, *aconnp;

	ASSERT(eager->tcp_connp->conn_rq == listener->tcp_connp->conn_rq);
	ASSERT(eager->tcp_detached && !acceptor->tcp_detached);
	ASSERT(!TCP_IS_SOCKET(acceptor));
	ASSERT(!TCP_IS_SOCKET(eager));
	ASSERT(!TCP_IS_SOCKET(listener));

	/*
	 * Trusted Extensions may need to use a security label that is
	 * different from the acceptor's label on MLP and MAC-Exempt
	 * sockets. If this is the case, the required security label
	 * already exists in econnp->conn_ixa->ixa_tsl. Since we make the
	 * acceptor stream refer to econnp we atomatically get that label.
	 */

	acceptor->tcp_detached = B_TRUE;
	/*
	 * To permit stream re-use by TLI/XTI, the eager needs a copy of
	 * the acceptor id.
	 */
	eager->tcp_acceptor_id = acceptor->tcp_acceptor_id;

	/* remove eager from listen list... */
	mutex_enter(&listener->tcp_eager_lock);
	tcp_eager_unlink(eager);
	ASSERT(eager->tcp_eager_next_q == NULL &&
	    eager->tcp_eager_last_q == NULL);
	ASSERT(eager->tcp_eager_next_q0 == NULL &&
	    eager->tcp_eager_prev_q0 == NULL);
	mutex_exit(&listener->tcp_eager_lock);

	econnp = eager->tcp_connp;
	aconnp = acceptor->tcp_connp;
	econnp->conn_rq = aconnp->conn_rq;
	econnp->conn_wq = aconnp->conn_wq;
	econnp->conn_rq->q_ptr = econnp;
	econnp->conn_wq->q_ptr = econnp;

	/*
	 * In the TLI/XTI loopback case, we are inside the listener's squeue,
	 * which might be a different squeue from our peer TCP instance.
	 * For TCP Fusion, the peer expects that whenever tcp_detached is
	 * clear, our TCP queues point to the acceptor's queues.  Thus, use
	 * membar_producer() to ensure that the assignments of conn_rq/conn_wq
	 * above reach global visibility prior to the clearing of tcp_detached.
	 */
	membar_producer();
	eager->tcp_detached = B_FALSE;

	ASSERT(eager->tcp_ack_tid == 0);

	econnp->conn_dev = aconnp->conn_dev;
	econnp->conn_minor_arena = aconnp->conn_minor_arena;

	ASSERT(econnp->conn_minor_arena != NULL);
	if (econnp->conn_cred != NULL)
		crfree(econnp->conn_cred);
	econnp->conn_cred = aconnp->conn_cred;
	aconnp->conn_cred = NULL;
	econnp->conn_cpid = aconnp->conn_cpid;
	ASSERT(econnp->conn_netstack == aconnp->conn_netstack);
	ASSERT(eager->tcp_tcps == acceptor->tcp_tcps);

	econnp->conn_zoneid = aconnp->conn_zoneid;
	econnp->conn_allzones = aconnp->conn_allzones;
	econnp->conn_ixa->ixa_zoneid = aconnp->conn_ixa->ixa_zoneid;

	econnp->conn_mac_mode = aconnp->conn_mac_mode;
	econnp->conn_zone_is_global = aconnp->conn_zone_is_global;
	aconnp->conn_mac_mode = CONN_MAC_DEFAULT;

	/* Do the IPC initialization */
	CONN_INC_REF(econnp);

	/* Done with old IPC. Drop its ref on its connp */
	CONN_DEC_REF(aconnp);
}


/*
 * Adapt to the information, such as rtt and rtt_sd, provided from the
 * DCE and IRE maintained by IP.
 *
 * Checks for multicast and broadcast destination address.
 * Returns zero if ok; an errno on failure.
 *
 * Note that the MSS calculation here is based on the info given in
 * the DCE and IRE.  We do not do any calculation based on TCP options.  They
 * will be handled in tcp_input_data() when TCP knows which options to use.
 *
 * Note on how TCP gets its parameters for a connection.
 *
 * When a tcp_t structure is allocated, it gets all the default parameters.
 * In tcp_set_destination(), it gets those metric parameters, like rtt, rtt_sd,
 * spipe, rpipe, ... from the route metrics.  Route metric overrides the
 * default.
 *
 * An incoming SYN with a multicast or broadcast destination address is dropped
 * in ip_fanout_v4/v6.
 *
 * An incoming SYN with a multicast or broadcast source address is always
 * dropped in tcp_set_destination, since IPDF_ALLOW_MCBC is not set in
 * conn_connect.
 * The same logic in tcp_set_destination also serves to
 * reject an attempt to connect to a broadcast or multicast (destination)
 * address.
 */
static int
tcp_set_destination(tcp_t *tcp)
{
	uint32_t	mss_max;
	uint32_t	mss;
	boolean_t	tcp_detached = TCP_IS_DETACHED(tcp);
	conn_t		*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	iulp_t		uinfo;
	int		error;
	uint32_t	flags;

	flags = IPDF_LSO | IPDF_ZCOPY;
	/*
	 * Make sure we have a dce for the destination to avoid dce_ident
	 * contention for connected sockets.
	 */
	flags |= IPDF_UNIQUE_DCE;

	if (!tcps->tcps_ignore_path_mtu)
		connp->conn_ixa->ixa_flags |= IXAF_PMTU_DISCOVERY;

	/* Use conn_lock to satify ASSERT; tcp is already serialized */
	mutex_enter(&connp->conn_lock);
	error = conn_connect(connp, &uinfo, flags);
	mutex_exit(&connp->conn_lock);
	if (error != 0)
		return (error);

	error = tcp_build_hdrs(tcp);
	if (error != 0)
		return (error);

	tcp->tcp_localnet = uinfo.iulp_localnet;

	if (uinfo.iulp_rtt != 0) {
		clock_t	rto;

		tcp->tcp_rtt_sa = uinfo.iulp_rtt;
		tcp->tcp_rtt_sd = uinfo.iulp_rtt_sd;
		rto = (tcp->tcp_rtt_sa >> 3) + tcp->tcp_rtt_sd +
		    tcps->tcps_rexmit_interval_extra +
		    (tcp->tcp_rtt_sa >> 5);

		if (rto > tcps->tcps_rexmit_interval_max) {
			tcp->tcp_rto = tcps->tcps_rexmit_interval_max;
		} else if (rto < tcps->tcps_rexmit_interval_min) {
			tcp->tcp_rto = tcps->tcps_rexmit_interval_min;
		} else {
			tcp->tcp_rto = rto;
		}
	}
	if (uinfo.iulp_ssthresh != 0)
		tcp->tcp_cwnd_ssthresh = uinfo.iulp_ssthresh;
	else
		tcp->tcp_cwnd_ssthresh = TCP_MAX_LARGEWIN;
	if (uinfo.iulp_spipe > 0) {
		connp->conn_sndbuf = MIN(uinfo.iulp_spipe,
		    tcps->tcps_max_buf);
		if (tcps->tcps_snd_lowat_fraction != 0) {
			connp->conn_sndlowat = connp->conn_sndbuf /
			    tcps->tcps_snd_lowat_fraction;
		}
		(void) tcp_maxpsz_set(tcp, B_TRUE);
	}
	/*
	 * Note that up till now, acceptor always inherits receive
	 * window from the listener.  But if there is a metrics
	 * associated with a host, we should use that instead of
	 * inheriting it from listener. Thus we need to pass this
	 * info back to the caller.
	 */
	if (uinfo.iulp_rpipe > 0) {
		tcp->tcp_rwnd = MIN(uinfo.iulp_rpipe,
		    tcps->tcps_max_buf);
	}

	if (uinfo.iulp_rtomax > 0) {
		tcp->tcp_second_timer_threshold =
		    uinfo.iulp_rtomax;
	}

	/*
	 * Use the metric option settings, iulp_tstamp_ok and
	 * iulp_wscale_ok, only for active open. What this means
	 * is that if the other side uses timestamp or window
	 * scale option, TCP will also use those options. That
	 * is for passive open.  If the application sets a
	 * large window, window scale is enabled regardless of
	 * the value in iulp_wscale_ok.  This is the behavior
	 * since 2.6.  So we keep it.
	 * The only case left in passive open processing is the
	 * check for SACK.
	 * For ECN, it should probably be like SACK.  But the
	 * current value is binary, so we treat it like the other
	 * cases.  The metric only controls active open.For passive
	 * open, the ndd param, tcp_ecn_permitted, controls the
	 * behavior.
	 */
	if (!tcp_detached) {
		/*
		 * The if check means that the following can only
		 * be turned on by the metrics only IRE, but not off.
		 */
		if (uinfo.iulp_tstamp_ok)
			tcp->tcp_snd_ts_ok = B_TRUE;
		if (uinfo.iulp_wscale_ok)
			tcp->tcp_snd_ws_ok = B_TRUE;
		if (uinfo.iulp_sack == 2)
			tcp->tcp_snd_sack_ok = B_TRUE;
		if (uinfo.iulp_ecn_ok)
			tcp->tcp_ecn_ok = B_TRUE;
	} else {
		/*
		 * Passive open.
		 *
		 * As above, the if check means that SACK can only be
		 * turned on by the metric only IRE.
		 */
		if (uinfo.iulp_sack > 0) {
			tcp->tcp_snd_sack_ok = B_TRUE;
		}
	}

	/*
	 * XXX Note that currently, iulp_mtu can be as small as 68
	 * because of PMTUd.  So tcp_mss may go to negative if combined
	 * length of all those options exceeds 28 bytes.  But because
	 * of the tcp_mss_min check below, we may not have a problem if
	 * tcp_mss_min is of a reasonable value.  The default is 1 so
	 * the negative problem still exists.  And the check defeats PMTUd.
	 * In fact, if PMTUd finds that the MSS should be smaller than
	 * tcp_mss_min, TCP should turn off PMUTd and use the tcp_mss_min
	 * value.
	 *
	 * We do not deal with that now.  All those problems related to
	 * PMTUd will be fixed later.
	 */
	ASSERT(uinfo.iulp_mtu != 0);
	mss = tcp->tcp_initial_pmtu = uinfo.iulp_mtu;

	/* Sanity check for MSS value. */
	if (connp->conn_ipversion == IPV4_VERSION)
		mss_max = tcps->tcps_mss_max_ipv4;
	else
		mss_max = tcps->tcps_mss_max_ipv6;

	if (tcp->tcp_ipsec_overhead == 0)
		tcp->tcp_ipsec_overhead = conn_ipsec_length(connp);

	mss -= tcp->tcp_ipsec_overhead;

	if (mss < tcps->tcps_mss_min)
		mss = tcps->tcps_mss_min;
	if (mss > mss_max)
		mss = mss_max;

	/* Note that this is the maximum MSS, excluding all options. */
	tcp->tcp_mss = mss;

	/*
	 * Update the tcp connection with LSO capability.
	 */
	tcp_update_lso(tcp, connp->conn_ixa);

	/*
	 * Initialize the ISS here now that we have the full connection ID.
	 * The RFC 1948 method of initial sequence number generation requires
	 * knowledge of the full connection ID before setting the ISS.
	 */
	tcp_iss_init(tcp);

	tcp->tcp_loopback = (uinfo.iulp_loopback | uinfo.iulp_local);

	/*
	 * Make sure that conn is not marked incipient
	 * for incoming connections. A blind
	 * removal of incipient flag is cheaper than
	 * check and removal.
	 */
	mutex_enter(&connp->conn_lock);
	connp->conn_state_flags &= ~CONN_INCIPIENT;
	mutex_exit(&connp->conn_lock);
	return (0);
}

static void
tcp_tpi_bind(tcp_t *tcp, mblk_t *mp)
{
	int	error;
	conn_t	*connp = tcp->tcp_connp;
	struct sockaddr	*sa;
	mblk_t  *mp1;
	struct T_bind_req *tbr;
	int	backlog;
	socklen_t	len;
	sin_t	*sin;
	sin6_t	*sin6;
	cred_t		*cr;

	/*
	 * All Solaris components should pass a db_credp
	 * for this TPI message, hence we ASSERT.
	 * But in case there is some other M_PROTO that looks
	 * like a TPI message sent by some other kernel
	 * component, we check and return an error.
	 */
	cr = msg_getcred(mp, NULL);
	ASSERT(cr != NULL);
	if (cr == NULL) {
		tcp_err_ack(tcp, mp, TSYSERR, EINVAL);
		return;
	}

	ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <= (uintptr_t)INT_MAX);
	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tbr)) {
		if (connp->conn_debug) {
			(void) strlog(TCP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
			    "tcp_tpi_bind: bad req, len %u",
			    (uint_t)(mp->b_wptr - mp->b_rptr));
		}
		tcp_err_ack(tcp, mp, TPROTO, 0);
		return;
	}
	/* Make sure the largest address fits */
	mp1 = reallocb(mp, sizeof (struct T_bind_ack) + sizeof (sin6_t), 1);
	if (mp1 == NULL) {
		tcp_err_ack(tcp, mp, TSYSERR, ENOMEM);
		return;
	}
	mp = mp1;
	tbr = (struct T_bind_req *)mp->b_rptr;

	backlog = tbr->CONIND_number;
	len = tbr->ADDR_length;

	switch (len) {
	case 0:		/* request for a generic port */
		tbr->ADDR_offset = sizeof (struct T_bind_req);
		if (connp->conn_family == AF_INET) {
			tbr->ADDR_length = sizeof (sin_t);
			sin = (sin_t *)&tbr[1];
			*sin = sin_null;
			sin->sin_family = AF_INET;
			sa = (struct sockaddr *)sin;
			len = sizeof (sin_t);
			mp->b_wptr = (uchar_t *)&sin[1];
		} else {
			ASSERT(connp->conn_family == AF_INET6);
			tbr->ADDR_length = sizeof (sin6_t);
			sin6 = (sin6_t *)&tbr[1];
			*sin6 = sin6_null;
			sin6->sin6_family = AF_INET6;
			sa = (struct sockaddr *)sin6;
			len = sizeof (sin6_t);
			mp->b_wptr = (uchar_t *)&sin6[1];
		}
		break;

	case sizeof (sin_t):    /* Complete IPv4 address */
		sa = (struct sockaddr *)mi_offset_param(mp, tbr->ADDR_offset,
		    sizeof (sin_t));
		break;

	case sizeof (sin6_t): /* Complete IPv6 address */
		sa = (struct sockaddr *)mi_offset_param(mp,
		    tbr->ADDR_offset, sizeof (sin6_t));
		break;

	default:
		if (connp->conn_debug) {
			(void) strlog(TCP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
			    "tcp_tpi_bind: bad address length, %d",
			    tbr->ADDR_length);
		}
		tcp_err_ack(tcp, mp, TBADADDR, 0);
		return;
	}

	if (backlog > 0) {
		error = tcp_do_listen(connp, sa, len, backlog, DB_CRED(mp),
		    tbr->PRIM_type != O_T_BIND_REQ);
	} else {
		error = tcp_do_bind(connp, sa, len, DB_CRED(mp),
		    tbr->PRIM_type != O_T_BIND_REQ);
	}
done:
	if (error > 0) {
		tcp_err_ack(tcp, mp, TSYSERR, error);
	} else if (error < 0) {
		tcp_err_ack(tcp, mp, -error, 0);
	} else {
		/*
		 * Update port information as sockfs/tpi needs it for checking
		 */
		if (connp->conn_family == AF_INET) {
			sin = (sin_t *)sa;
			sin->sin_port = connp->conn_lport;
		} else {
			sin6 = (sin6_t *)sa;
			sin6->sin6_port = connp->conn_lport;
		}
		mp->b_datap->db_type = M_PCPROTO;
		tbr->PRIM_type = T_BIND_ACK;
		putnext(connp->conn_rq, mp);
	}
}

/*
 * If the "bind_to_req_port_only" parameter is set, if the requested port
 * number is available, return it, If not return 0
 *
 * If "bind_to_req_port_only" parameter is not set and
 * If the requested port number is available, return it.  If not, return
 * the first anonymous port we happen across.  If no anonymous ports are
 * available, return 0. addr is the requested local address, if any.
 *
 * In either case, when succeeding update the tcp_t to record the port number
 * and insert it in the bind hash table.
 *
 * Note that TCP over IPv4 and IPv6 sockets can use the same port number
 * without setting SO_REUSEADDR. This is needed so that they
 * can be viewed as two independent transport protocols.
 */
static in_port_t
tcp_bindi(tcp_t *tcp, in_port_t port, const in6_addr_t *laddr,
    int reuseaddr, boolean_t quick_connect,
    boolean_t bind_to_req_port_only, boolean_t user_specified)
{
	/* number of times we have run around the loop */
	int count = 0;
	/* maximum number of times to run around the loop */
	int loopmax;
	conn_t *connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	/*
	 * Lookup for free addresses is done in a loop and "loopmax"
	 * influences how long we spin in the loop
	 */
	if (bind_to_req_port_only) {
		/*
		 * If the requested port is busy, don't bother to look
		 * for a new one. Setting loop maximum count to 1 has
		 * that effect.
		 */
		loopmax = 1;
	} else {
		/*
		 * If the requested port is busy, look for a free one
		 * in the anonymous port range.
		 * Set loopmax appropriately so that one does not look
		 * forever in the case all of the anonymous ports are in use.
		 */
		if (connp->conn_anon_priv_bind) {
			/*
			 * loopmax =
			 * 	(IPPORT_RESERVED-1) - tcp_min_anonpriv_port + 1
			 */
			loopmax = IPPORT_RESERVED -
			    tcps->tcps_min_anonpriv_port;
		} else {
			loopmax = (tcps->tcps_largest_anon_port -
			    tcps->tcps_smallest_anon_port + 1);
		}
	}
	do {
		uint16_t	lport;
		tf_t		*tbf;
		tcp_t		*ltcp;
		conn_t		*lconnp;

		lport = htons(port);

		/*
		 * Ensure that the tcp_t is not currently in the bind hash.
		 * Hold the lock on the hash bucket to ensure that
		 * the duplicate check plus the insertion is an atomic
		 * operation.
		 *
		 * This function does an inline lookup on the bind hash list
		 * Make sure that we access only members of tcp_t
		 * and that we don't look at tcp_tcp, since we are not
		 * doing a CONN_INC_REF.
		 */
		tcp_bind_hash_remove(tcp);
		tbf = &tcps->tcps_bind_fanout[TCP_BIND_HASH(lport)];
		mutex_enter(&tbf->tf_lock);
		for (ltcp = tbf->tf_tcp; ltcp != NULL;
		    ltcp = ltcp->tcp_bind_hash) {
			if (lport == ltcp->tcp_connp->conn_lport)
				break;
		}

		for (; ltcp != NULL; ltcp = ltcp->tcp_bind_hash_port) {
			boolean_t not_socket;
			boolean_t exclbind;

			lconnp = ltcp->tcp_connp;

			/*
			 * On a labeled system, we must treat bindings to ports
			 * on shared IP addresses by sockets with MAC exemption
			 * privilege as being in all zones, as there's
			 * otherwise no way to identify the right receiver.
			 */
			if (!IPCL_BIND_ZONE_MATCH(lconnp, connp))
				continue;

			/*
			 * If TCP_EXCLBIND is set for either the bound or
			 * binding endpoint, the semantics of bind
			 * is changed according to the following.
			 *
			 * spec = specified address (v4 or v6)
			 * unspec = unspecified address (v4 or v6)
			 * A = specified addresses are different for endpoints
			 *
			 * bound	bind to		allowed
			 * -------------------------------------
			 * unspec	unspec		no
			 * unspec	spec		no
			 * spec		unspec		no
			 * spec		spec		yes if A
			 *
			 * For labeled systems, SO_MAC_EXEMPT behaves the same
			 * as TCP_EXCLBIND, except that zoneid is ignored.
			 *
			 * Note:
			 *
			 * 1. Because of TLI semantics, an endpoint can go
			 * back from, say TCP_ESTABLISHED to TCPS_LISTEN or
			 * TCPS_BOUND, depending on whether it is originally
			 * a listener or not.  That is why we need to check
			 * for states greater than or equal to TCPS_BOUND
			 * here.
			 *
			 * 2. Ideally, we should only check for state equals
			 * to TCPS_LISTEN. And the following check should be
			 * added.
			 *
			 * if (ltcp->tcp_state == TCPS_LISTEN ||
			 *	!reuseaddr || !lconnp->conn_reuseaddr) {
			 *		...
			 * }
			 *
			 * The semantics will be changed to this.  If the
			 * endpoint on the list is in state not equal to
			 * TCPS_LISTEN and both endpoints have SO_REUSEADDR
			 * set, let the bind succeed.
			 *
			 * Because of (1), we cannot do that for TLI
			 * endpoints.  But we can do that for socket endpoints.
			 * If in future, we can change this going back
			 * semantics, we can use the above check for TLI also.
			 */
			not_socket = !(TCP_IS_SOCKET(ltcp) &&
			    TCP_IS_SOCKET(tcp));
			exclbind = lconnp->conn_exclbind ||
			    connp->conn_exclbind;

			if ((lconnp->conn_mac_mode != CONN_MAC_DEFAULT) ||
			    (connp->conn_mac_mode != CONN_MAC_DEFAULT) ||
			    (exclbind && (not_socket ||
			    ltcp->tcp_state <= TCPS_ESTABLISHED))) {
				if (V6_OR_V4_INADDR_ANY(
				    lconnp->conn_bound_addr_v6) ||
				    V6_OR_V4_INADDR_ANY(*laddr) ||
				    IN6_ARE_ADDR_EQUAL(laddr,
				    &lconnp->conn_bound_addr_v6)) {
					break;
				}
				continue;
			}

			/*
			 * Check ipversion to allow IPv4 and IPv6 sockets to
			 * have disjoint port number spaces, if *_EXCLBIND
			 * is not set and only if the application binds to a
			 * specific port. We use the same autoassigned port
			 * number space for IPv4 and IPv6 sockets.
			 */
			if (connp->conn_ipversion != lconnp->conn_ipversion &&
			    bind_to_req_port_only)
				continue;

			/*
			 * Ideally, we should make sure that the source
			 * address, remote address, and remote port in the
			 * four tuple for this tcp-connection is unique.
			 * However, trying to find out the local source
			 * address would require too much code duplication
			 * with IP, since IP needs needs to have that code
			 * to support userland TCP implementations.
			 */
			if (quick_connect &&
			    (ltcp->tcp_state > TCPS_LISTEN) &&
			    ((connp->conn_fport != lconnp->conn_fport) ||
			    !IN6_ARE_ADDR_EQUAL(&connp->conn_faddr_v6,
			    &lconnp->conn_faddr_v6)))
				continue;

			if (!reuseaddr) {
				/*
				 * No socket option SO_REUSEADDR.
				 * If existing port is bound to
				 * a non-wildcard IP address
				 * and the requesting stream is
				 * bound to a distinct
				 * different IP addresses
				 * (non-wildcard, also), keep
				 * going.
				 */
				if (!V6_OR_V4_INADDR_ANY(*laddr) &&
				    !V6_OR_V4_INADDR_ANY(
				    lconnp->conn_bound_addr_v6) &&
				    !IN6_ARE_ADDR_EQUAL(laddr,
				    &lconnp->conn_bound_addr_v6))
					continue;
				if (ltcp->tcp_state >= TCPS_BOUND) {
					/*
					 * This port is being used and
					 * its state is >= TCPS_BOUND,
					 * so we can't bind to it.
					 */
					break;
				}
			} else {
				/*
				 * socket option SO_REUSEADDR is set on the
				 * binding tcp_t.
				 *
				 * If two streams are bound to
				 * same IP address or both addr
				 * and bound source are wildcards
				 * (INADDR_ANY), we want to stop
				 * searching.
				 * We have found a match of IP source
				 * address and source port, which is
				 * refused regardless of the
				 * SO_REUSEADDR setting, so we break.
				 */
				if (IN6_ARE_ADDR_EQUAL(laddr,
				    &lconnp->conn_bound_addr_v6) &&
				    (ltcp->tcp_state == TCPS_LISTEN ||
				    ltcp->tcp_state == TCPS_BOUND))
					break;
			}
		}
		if (ltcp != NULL) {
			/* The port number is busy */
			mutex_exit(&tbf->tf_lock);
		} else {
			/*
			 * This port is ours. Insert in fanout and mark as
			 * bound to prevent others from getting the port
			 * number.
			 */
			tcp->tcp_state = TCPS_BOUND;
			connp->conn_lport = htons(port);

			ASSERT(&tcps->tcps_bind_fanout[TCP_BIND_HASH(
			    connp->conn_lport)] == tbf);
			tcp_bind_hash_insert(tbf, tcp, 1);

			mutex_exit(&tbf->tf_lock);

			/*
			 * We don't want tcp_next_port_to_try to "inherit"
			 * a port number supplied by the user in a bind.
			 */
			if (user_specified)
				return (port);

			/*
			 * This is the only place where tcp_next_port_to_try
			 * is updated. After the update, it may or may not
			 * be in the valid range.
			 */
			if (!connp->conn_anon_priv_bind)
				tcps->tcps_next_port_to_try = port + 1;
			return (port);
		}

		if (connp->conn_anon_priv_bind) {
			port = tcp_get_next_priv_port(tcp);
		} else {
			if (count == 0 && user_specified) {
				/*
				 * We may have to return an anonymous port. So
				 * get one to start with.
				 */
				port =
				    tcp_update_next_port(
				    tcps->tcps_next_port_to_try,
				    tcp, B_TRUE);
				user_specified = B_FALSE;
			} else {
				port = tcp_update_next_port(port + 1, tcp,
				    B_FALSE);
			}
		}
		if (port == 0)
			break;

		/*
		 * Don't let this loop run forever in the case where
		 * all of the anonymous ports are in use.
		 */
	} while (++count < loopmax);
	return (0);
}

/*
 * tcp_clean_death / tcp_close_detached must not be called more than once
 * on a tcp. Thus every function that potentially calls tcp_clean_death
 * must check for the tcp state before calling tcp_clean_death.
 * Eg. tcp_input_data, tcp_eager_kill, tcp_clean_death_wrapper,
 * tcp_timer_handler, all check for the tcp state.
 */
/* ARGSUSED */
void
tcp_clean_death_wrapper(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *dummy)
{
	tcp_t	*tcp = ((conn_t *)arg)->conn_tcp;

	freemsg(mp);
	if (tcp->tcp_state > TCPS_BOUND)
		(void) tcp_clean_death(((conn_t *)arg)->conn_tcp,
		    ETIMEDOUT, 5);
}

/*
 * We are dying for some reason.  Try to do it gracefully.  (May be called
 * as writer.)
 *
 * Return -1 if the structure was not cleaned up (if the cleanup had to be
 * done by a service procedure).
 * TBD - Should the return value distinguish between the tcp_t being
 * freed and it being reinitialized?
 */
static int
tcp_clean_death(tcp_t *tcp, int err, uint8_t tag)
{
	mblk_t	*mp;
	queue_t	*q;
	conn_t	*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	TCP_CLD_STAT(tag);

#if TCP_TAG_CLEAN_DEATH
	tcp->tcp_cleandeathtag = tag;
#endif

	if (tcp->tcp_fused)
		tcp_unfuse(tcp);

	if (tcp->tcp_linger_tid != 0 &&
	    TCP_TIMER_CANCEL(tcp, tcp->tcp_linger_tid) >= 0) {
		tcp_stop_lingering(tcp);
	}

	ASSERT(tcp != NULL);
	ASSERT((connp->conn_family == AF_INET &&
	    connp->conn_ipversion == IPV4_VERSION) ||
	    (connp->conn_family == AF_INET6 &&
	    (connp->conn_ipversion == IPV4_VERSION ||
	    connp->conn_ipversion == IPV6_VERSION)));

	if (TCP_IS_DETACHED(tcp)) {
		if (tcp->tcp_hard_binding) {
			/*
			 * Its an eager that we are dealing with. We close the
			 * eager but in case a conn_ind has already gone to the
			 * listener, let tcp_accept_finish() send a discon_ind
			 * to the listener and drop the last reference. If the
			 * listener doesn't even know about the eager i.e. the
			 * conn_ind hasn't gone up, blow away the eager and drop
			 * the last reference as well. If the conn_ind has gone
			 * up, state should be BOUND. tcp_accept_finish
			 * will figure out that the connection has received a
			 * RST and will send a DISCON_IND to the application.
			 */
			tcp_closei_local(tcp);
			if (!tcp->tcp_tconnind_started) {
				CONN_DEC_REF(connp);
			} else {
				tcp->tcp_state = TCPS_BOUND;
			}
		} else {
			tcp_close_detached(tcp);
		}
		return (0);
	}

	TCP_STAT(tcps, tcp_clean_death_nondetached);

	q = connp->conn_rq;

	/* Trash all inbound data */
	if (!IPCL_IS_NONSTR(connp)) {
		ASSERT(q != NULL);
		flushq(q, FLUSHALL);
	}

	/*
	 * If we are at least part way open and there is error
	 * (err==0 implies no error)
	 * notify our client by a T_DISCON_IND.
	 */
	if ((tcp->tcp_state >= TCPS_SYN_SENT) && err) {
		if (tcp->tcp_state >= TCPS_ESTABLISHED &&
		    !TCP_IS_SOCKET(tcp)) {
			/*
			 * Send M_FLUSH according to TPI. Because sockets will
			 * (and must) ignore FLUSHR we do that only for TPI
			 * endpoints and sockets in STREAMS mode.
			 */
			(void) putnextctl1(q, M_FLUSH, FLUSHR);
		}
		if (connp->conn_debug) {
			(void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE|SL_ERROR,
			    "tcp_clean_death: discon err %d", err);
		}
		if (IPCL_IS_NONSTR(connp)) {
			/* Direct socket, use upcall */
			(*connp->conn_upcalls->su_disconnected)(
			    connp->conn_upper_handle, tcp->tcp_connid, err);
		} else {
			mp = mi_tpi_discon_ind(NULL, err, 0);
			if (mp != NULL) {
				putnext(q, mp);
			} else {
				if (connp->conn_debug) {
					(void) strlog(TCP_MOD_ID, 0, 1,
					    SL_ERROR|SL_TRACE,
					    "tcp_clean_death, sending M_ERROR");
				}
				(void) putnextctl1(q, M_ERROR, EPROTO);
			}
		}
		if (tcp->tcp_state <= TCPS_SYN_RCVD) {
			/* SYN_SENT or SYN_RCVD */
			BUMP_MIB(&tcps->tcps_mib, tcpAttemptFails);
		} else if (tcp->tcp_state <= TCPS_CLOSE_WAIT) {
			/* ESTABLISHED or CLOSE_WAIT */
			BUMP_MIB(&tcps->tcps_mib, tcpEstabResets);
		}
	}

	tcp_reinit(tcp);
	if (IPCL_IS_NONSTR(connp))
		(void) tcp_do_unbind(connp);

	return (-1);
}

/*
 * In case tcp is in the "lingering state" and waits for the SO_LINGER timeout
 * to expire, stop the wait and finish the close.
 */
static void
tcp_stop_lingering(tcp_t *tcp)
{
	clock_t	delta = 0;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

	tcp->tcp_linger_tid = 0;
	if (tcp->tcp_state > TCPS_LISTEN) {
		tcp_acceptor_hash_remove(tcp);
		mutex_enter(&tcp->tcp_non_sq_lock);
		if (tcp->tcp_flow_stopped) {
			tcp_clrqfull(tcp);
		}
		mutex_exit(&tcp->tcp_non_sq_lock);

		if (tcp->tcp_timer_tid != 0) {
			delta = TCP_TIMER_CANCEL(tcp, tcp->tcp_timer_tid);
			tcp->tcp_timer_tid = 0;
		}
		/*
		 * Need to cancel those timers which will not be used when
		 * TCP is detached.  This has to be done before the conn_wq
		 * is cleared.
		 */
		tcp_timers_stop(tcp);

		tcp->tcp_detached = B_TRUE;
		connp->conn_rq = NULL;
		connp->conn_wq = NULL;

		if (tcp->tcp_state == TCPS_TIME_WAIT) {
			tcp_time_wait_append(tcp);
			TCP_DBGSTAT(tcps, tcp_detach_time_wait);
			goto finish;
		}

		/*
		 * If delta is zero the timer event wasn't executed and was
		 * successfully canceled. In this case we need to restart it
		 * with the minimal delta possible.
		 */
		if (delta >= 0) {
			tcp->tcp_timer_tid = TCP_TIMER(tcp, tcp_timer,
			    delta ? delta : 1);
		}
	} else {
		tcp_closei_local(tcp);
		CONN_DEC_REF(connp);
	}
finish:
	/* Signal closing thread that it can complete close */
	mutex_enter(&tcp->tcp_closelock);
	tcp->tcp_detached = B_TRUE;
	connp->conn_rq = NULL;
	connp->conn_wq = NULL;

	tcp->tcp_closed = 1;
	cv_signal(&tcp->tcp_closecv);
	mutex_exit(&tcp->tcp_closelock);
}

/*
 * Handle lingering timeouts. This function is called when the SO_LINGER timeout
 * expires.
 */
static void
tcp_close_linger_timeout(void *arg)
{
	conn_t	*connp = (conn_t *)arg;
	tcp_t 	*tcp = connp->conn_tcp;

	tcp->tcp_client_errno = ETIMEDOUT;
	tcp_stop_lingering(tcp);
}

static void
tcp_close_common(conn_t *connp, int flags)
{
	tcp_t		*tcp = connp->conn_tcp;
	mblk_t 		*mp = &tcp->tcp_closemp;
	boolean_t	conn_ioctl_cleanup_reqd = B_FALSE;
	mblk_t		*bp;

	ASSERT(connp->conn_ref >= 2);

	/*
	 * Mark the conn as closing. ipsq_pending_mp_add will not
	 * add any mp to the pending mp list, after this conn has
	 * started closing.
	 */
	mutex_enter(&connp->conn_lock);
	connp->conn_state_flags |= CONN_CLOSING;
	if (connp->conn_oper_pending_ill != NULL)
		conn_ioctl_cleanup_reqd = B_TRUE;
	CONN_INC_REF_LOCKED(connp);
	mutex_exit(&connp->conn_lock);
	tcp->tcp_closeflags = (uint8_t)flags;
	ASSERT(connp->conn_ref >= 3);

	/*
	 * tcp_closemp_used is used below without any protection of a lock
	 * as we don't expect any one else to use it concurrently at this
	 * point otherwise it would be a major defect.
	 */

	if (mp->b_prev == NULL)
		tcp->tcp_closemp_used = B_TRUE;
	else
		cmn_err(CE_PANIC, "tcp_close: concurrent use of tcp_closemp: "
		    "connp %p tcp %p\n", (void *)connp, (void *)tcp);

	TCP_DEBUG_GETPCSTACK(tcp->tcmp_stk, 15);

	SQUEUE_ENTER_ONE(connp->conn_sqp, mp, tcp_close_output, connp,
	    NULL, tcp_squeue_flag, SQTAG_IP_TCP_CLOSE);

	mutex_enter(&tcp->tcp_closelock);
	while (!tcp->tcp_closed) {
		if (!cv_wait_sig(&tcp->tcp_closecv, &tcp->tcp_closelock)) {
			/*
			 * The cv_wait_sig() was interrupted. We now do the
			 * following:
			 *
			 * 1) If the endpoint was lingering, we allow this
			 * to be interrupted by cancelling the linger timeout
			 * and closing normally.
			 *
			 * 2) Revert to calling cv_wait()
			 *
			 * We revert to using cv_wait() to avoid an
			 * infinite loop which can occur if the calling
			 * thread is higher priority than the squeue worker
			 * thread and is bound to the same cpu.
			 */
			if (connp->conn_linger && connp->conn_lingertime > 0) {
				mutex_exit(&tcp->tcp_closelock);
				/* Entering squeue, bump ref count. */
				CONN_INC_REF(connp);
				bp = allocb_wait(0, BPRI_HI, STR_NOSIG, NULL);
				SQUEUE_ENTER_ONE(connp->conn_sqp, bp,
				    tcp_linger_interrupted, connp, NULL,
				    tcp_squeue_flag, SQTAG_IP_TCP_CLOSE);
				mutex_enter(&tcp->tcp_closelock);
			}
			break;
		}
	}
	while (!tcp->tcp_closed)
		cv_wait(&tcp->tcp_closecv, &tcp->tcp_closelock);
	mutex_exit(&tcp->tcp_closelock);

	/*
	 * In the case of listener streams that have eagers in the q or q0
	 * we wait for the eagers to drop their reference to us. conn_rq and
	 * conn_wq of the eagers point to our queues. By waiting for the
	 * refcnt to drop to 1, we are sure that the eagers have cleaned
	 * up their queue pointers and also dropped their references to us.
	 */
	if (tcp->tcp_wait_for_eagers) {
		mutex_enter(&connp->conn_lock);
		while (connp->conn_ref != 1) {
			cv_wait(&connp->conn_cv, &connp->conn_lock);
		}
		mutex_exit(&connp->conn_lock);
	}
	/*
	 * ioctl cleanup. The mp is queued in the ipx_pending_mp.
	 */
	if (conn_ioctl_cleanup_reqd)
		conn_ioctl_cleanup(connp);

	connp->conn_cpid = NOPID;
}

static int
tcp_tpi_close(queue_t *q, int flags)
{
	conn_t		*connp;

	ASSERT(WR(q)->q_next == NULL);

	if (flags & SO_FALLBACK) {
		/*
		 * stream is being closed while in fallback
		 * simply free the resources that were allocated
		 */
		inet_minor_free(WR(q)->q_ptr, (dev_t)(RD(q)->q_ptr));
		qprocsoff(q);
		goto done;
	}

	connp = Q_TO_CONN(q);
	/*
	 * We are being closed as /dev/tcp or /dev/tcp6.
	 */
	tcp_close_common(connp, flags);

	qprocsoff(q);
	inet_minor_free(connp->conn_minor_arena, connp->conn_dev);

	/*
	 * Drop IP's reference on the conn. This is the last reference
	 * on the connp if the state was less than established. If the
	 * connection has gone into timewait state, then we will have
	 * one ref for the TCP and one more ref (total of two) for the
	 * classifier connected hash list (a timewait connections stays
	 * in connected hash till closed).
	 *
	 * We can't assert the references because there might be other
	 * transient reference places because of some walkers or queued
	 * packets in squeue for the timewait state.
	 */
	CONN_DEC_REF(connp);
done:
	q->q_ptr = WR(q)->q_ptr = NULL;
	return (0);
}

static int
tcp_tpi_close_accept(queue_t *q)
{
	vmem_t	*minor_arena;
	dev_t	conn_dev;

	ASSERT(WR(q)->q_qinfo == &tcp_acceptor_winit);

	/*
	 * We had opened an acceptor STREAM for sockfs which is
	 * now being closed due to some error.
	 */
	qprocsoff(q);

	minor_arena = (vmem_t *)WR(q)->q_ptr;
	conn_dev = (dev_t)RD(q)->q_ptr;
	ASSERT(minor_arena != NULL);
	ASSERT(conn_dev != 0);
	inet_minor_free(minor_arena, conn_dev);
	q->q_ptr = WR(q)->q_ptr = NULL;
	return (0);
}

/*
 * Called by tcp_close() routine via squeue when lingering is
 * interrupted by a signal.
 */

/* ARGSUSED */
static void
tcp_linger_interrupted(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
{
	conn_t	*connp = (conn_t *)arg;
	tcp_t	*tcp = connp->conn_tcp;

	freeb(mp);
	if (tcp->tcp_linger_tid != 0 &&
	    TCP_TIMER_CANCEL(tcp, tcp->tcp_linger_tid) >= 0) {
		tcp_stop_lingering(tcp);
		tcp->tcp_client_errno = EINTR;
	}
}

/*
 * Called by streams close routine via squeues when our client blows off her
 * descriptor, we take this to mean: "close the stream state NOW, close the tcp
 * connection politely" When SO_LINGER is set (with a non-zero linger time and
 * it is not a nonblocking socket) then this routine sleeps until the FIN is
 * acked.
 *
 * NOTE: tcp_close potentially returns error when lingering.
 * However, the stream head currently does not pass these errors
 * to the application. 4.4BSD only returns EINTR and EWOULDBLOCK
 * errors to the application (from tsleep()) and not errors
 * like ECONNRESET caused by receiving a reset packet.
 */

/* ARGSUSED */
static void
tcp_close_output(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
{
	char	*msg;
	conn_t	*connp = (conn_t *)arg;
	tcp_t	*tcp = connp->conn_tcp;
	clock_t	delta = 0;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	ASSERT((connp->conn_fanout != NULL && connp->conn_ref >= 4) ||
	    (connp->conn_fanout == NULL && connp->conn_ref >= 3));

	mutex_enter(&tcp->tcp_eager_lock);
	if (tcp->tcp_conn_req_cnt_q0 != 0 || tcp->tcp_conn_req_cnt_q != 0) {
		/* Cleanup for listener */
		tcp_eager_cleanup(tcp, 0);
		tcp->tcp_wait_for_eagers = 1;
	}
	mutex_exit(&tcp->tcp_eager_lock);

	tcp->tcp_lso = B_FALSE;

	msg = NULL;
	switch (tcp->tcp_state) {
	case TCPS_CLOSED:
	case TCPS_IDLE:
	case TCPS_BOUND:
	case TCPS_LISTEN:
		break;
	case TCPS_SYN_SENT:
		msg = "tcp_close, during connect";
		break;
	case TCPS_SYN_RCVD:
		/*
		 * Close during the connect 3-way handshake
		 * but here there may or may not be pending data
		 * already on queue. Process almost same as in
		 * the ESTABLISHED state.
		 */
		/* FALLTHRU */
	default:
		if (tcp->tcp_fused)
			tcp_unfuse(tcp);

		/*
		 * If SO_LINGER has set a zero linger time, abort the
		 * connection with a reset.
		 */
		if (connp->conn_linger && connp->conn_lingertime == 0) {
			msg = "tcp_close, zero lingertime";
			break;
		}

		/*
		 * Abort connection if there is unread data queued.
		 */
		if (tcp->tcp_rcv_list || tcp->tcp_reass_head) {
			msg = "tcp_close, unread data";
			break;
		}
		/*
		 * We have done a qwait() above which could have possibly
		 * drained more messages in turn causing transition to a
		 * different state. Check whether we have to do the rest
		 * of the processing or not.
		 */
		if (tcp->tcp_state <= TCPS_LISTEN)
			break;

		/*
		 * Transmit the FIN before detaching the tcp_t.
		 * After tcp_detach returns this queue/perimeter
		 * no longer owns the tcp_t thus others can modify it.
		 */
		(void) tcp_xmit_end(tcp);

		/*
		 * If lingering on close then wait until the fin is acked,
		 * the SO_LINGER time passes, or a reset is sent/received.
		 */
		if (connp->conn_linger && connp->conn_lingertime > 0 &&
		    !(tcp->tcp_fin_acked) &&
		    tcp->tcp_state >= TCPS_ESTABLISHED) {
			if (tcp->tcp_closeflags & (FNDELAY|FNONBLOCK)) {
				tcp->tcp_client_errno = EWOULDBLOCK;
			} else if (tcp->tcp_client_errno == 0) {

				ASSERT(tcp->tcp_linger_tid == 0);

				tcp->tcp_linger_tid = TCP_TIMER(tcp,
				    tcp_close_linger_timeout,
				    connp->conn_lingertime * hz);

				/* tcp_close_linger_timeout will finish close */
				if (tcp->tcp_linger_tid == 0)
					tcp->tcp_client_errno = ENOSR;
				else
					return;
			}

			/*
			 * Check if we need to detach or just close
			 * the instance.
			 */
			if (tcp->tcp_state <= TCPS_LISTEN)
				break;
		}

		/*
		 * Make sure that no other thread will access the conn_rq of
		 * this instance (through lookups etc.) as conn_rq will go
		 * away shortly.
		 */
		tcp_acceptor_hash_remove(tcp);

		mutex_enter(&tcp->tcp_non_sq_lock);
		if (tcp->tcp_flow_stopped) {
			tcp_clrqfull(tcp);
		}
		mutex_exit(&tcp->tcp_non_sq_lock);

		if (tcp->tcp_timer_tid != 0) {
			delta = TCP_TIMER_CANCEL(tcp, tcp->tcp_timer_tid);
			tcp->tcp_timer_tid = 0;
		}
		/*
		 * Need to cancel those timers which will not be used when
		 * TCP is detached.  This has to be done before the conn_wq
		 * is set to NULL.
		 */
		tcp_timers_stop(tcp);

		tcp->tcp_detached = B_TRUE;
		if (tcp->tcp_state == TCPS_TIME_WAIT) {
			tcp_time_wait_append(tcp);
			TCP_DBGSTAT(tcps, tcp_detach_time_wait);
			ASSERT(connp->conn_ref >= 3);
			goto finish;
		}

		/*
		 * If delta is zero the timer event wasn't executed and was
		 * successfully canceled. In this case we need to restart it
		 * with the minimal delta possible.
		 */
		if (delta >= 0)
			tcp->tcp_timer_tid = TCP_TIMER(tcp, tcp_timer,
			    delta ? delta : 1);

		ASSERT(connp->conn_ref >= 3);
		goto finish;
	}

	/* Detach did not complete. Still need to remove q from stream. */
	if (msg) {
		if (tcp->tcp_state == TCPS_ESTABLISHED ||
		    tcp->tcp_state == TCPS_CLOSE_WAIT)
			BUMP_MIB(&tcps->tcps_mib, tcpEstabResets);
		if (tcp->tcp_state == TCPS_SYN_SENT ||
		    tcp->tcp_state == TCPS_SYN_RCVD)
			BUMP_MIB(&tcps->tcps_mib, tcpAttemptFails);
		tcp_xmit_ctl(msg, tcp,  tcp->tcp_snxt, 0, TH_RST);
	}

	tcp_closei_local(tcp);
	CONN_DEC_REF(connp);
	ASSERT(connp->conn_ref >= 2);

finish:
	mutex_enter(&tcp->tcp_closelock);
	/*
	 * Don't change the queues in the case of a listener that has
	 * eagers in its q or q0. It could surprise the eagers.
	 * Instead wait for the eagers outside the squeue.
	 */
	if (!tcp->tcp_wait_for_eagers) {
		tcp->tcp_detached = B_TRUE;
		connp->conn_rq = NULL;
		connp->conn_wq = NULL;
	}

	/* Signal tcp_close() to finish closing. */
	tcp->tcp_closed = 1;
	cv_signal(&tcp->tcp_closecv);
	mutex_exit(&tcp->tcp_closelock);
}

/*
 * Clean up the b_next and b_prev fields of every mblk pointed at by *mpp.
 * Some stream heads get upset if they see these later on as anything but NULL.
 */
static void
tcp_close_mpp(mblk_t **mpp)
{
	mblk_t	*mp;

	if ((mp = *mpp) != NULL) {
		do {
			mp->b_next = NULL;
			mp->b_prev = NULL;
		} while ((mp = mp->b_cont) != NULL);

		mp = *mpp;
		*mpp = NULL;
		freemsg(mp);
	}
}

/* Do detached close. */
static void
tcp_close_detached(tcp_t *tcp)
{
	if (tcp->tcp_fused)
		tcp_unfuse(tcp);

	/*
	 * Clustering code serializes TCP disconnect callbacks and
	 * cluster tcp list walks by blocking a TCP disconnect callback
	 * if a cluster tcp list walk is in progress. This ensures
	 * accurate accounting of TCPs in the cluster code even though
	 * the TCP list walk itself is not atomic.
	 */
	tcp_closei_local(tcp);
	CONN_DEC_REF(tcp->tcp_connp);
}

/*
 * Stop all TCP timers, and free the timer mblks if requested.
 */
void
tcp_timers_stop(tcp_t *tcp)
{
	if (tcp->tcp_timer_tid != 0) {
		(void) TCP_TIMER_CANCEL(tcp, tcp->tcp_timer_tid);
		tcp->tcp_timer_tid = 0;
	}
	if (tcp->tcp_ka_tid != 0) {
		(void) TCP_TIMER_CANCEL(tcp, tcp->tcp_ka_tid);
		tcp->tcp_ka_tid = 0;
	}
	if (tcp->tcp_ack_tid != 0) {
		(void) TCP_TIMER_CANCEL(tcp, tcp->tcp_ack_tid);
		tcp->tcp_ack_tid = 0;
	}
	if (tcp->tcp_push_tid != 0) {
		(void) TCP_TIMER_CANCEL(tcp, tcp->tcp_push_tid);
		tcp->tcp_push_tid = 0;
	}
}

/*
 * The tcp_t is going away. Remove it from all lists and set it
 * to TCPS_CLOSED. The freeing up of memory is deferred until
 * tcp_inactive. This is needed since a thread in tcp_rput might have
 * done a CONN_INC_REF on this structure before it was removed from the
 * hashes.
 */
static void
tcp_closei_local(tcp_t *tcp)
{
	conn_t		*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	if (!TCP_IS_SOCKET(tcp))
		tcp_acceptor_hash_remove(tcp);

	UPDATE_MIB(&tcps->tcps_mib, tcpHCInSegs, tcp->tcp_ibsegs);
	tcp->tcp_ibsegs = 0;
	UPDATE_MIB(&tcps->tcps_mib, tcpHCOutSegs, tcp->tcp_obsegs);
	tcp->tcp_obsegs = 0;

	/*
	 * If we are an eager connection hanging off a listener that
	 * hasn't formally accepted the connection yet, get off his
	 * list and blow off any data that we have accumulated.
	 */
	if (tcp->tcp_listener != NULL) {
		tcp_t	*listener = tcp->tcp_listener;
		mutex_enter(&listener->tcp_eager_lock);
		/*
		 * tcp_tconnind_started == B_TRUE means that the
		 * conn_ind has already gone to listener. At
		 * this point, eager will be closed but we
		 * leave it in listeners eager list so that
		 * if listener decides to close without doing
		 * accept, we can clean this up. In tcp_tli_accept
		 * we take care of the case of accept on closed
		 * eager.
		 */
		if (!tcp->tcp_tconnind_started) {
			tcp_eager_unlink(tcp);
			mutex_exit(&listener->tcp_eager_lock);
			/*
			 * We don't want to have any pointers to the
			 * listener queue, after we have released our
			 * reference on the listener
			 */
			ASSERT(tcp->tcp_detached);
			connp->conn_rq = NULL;
			connp->conn_wq = NULL;
			CONN_DEC_REF(listener->tcp_connp);
		} else {
			mutex_exit(&listener->tcp_eager_lock);
		}
	}

	/* Stop all the timers */
	tcp_timers_stop(tcp);

	if (tcp->tcp_state == TCPS_LISTEN) {
		if (tcp->tcp_ip_addr_cache) {
			kmem_free((void *)tcp->tcp_ip_addr_cache,
			    IP_ADDR_CACHE_SIZE * sizeof (ipaddr_t));
			tcp->tcp_ip_addr_cache = NULL;
		}
	}
	mutex_enter(&tcp->tcp_non_sq_lock);
	if (tcp->tcp_flow_stopped)
		tcp_clrqfull(tcp);
	mutex_exit(&tcp->tcp_non_sq_lock);

	tcp_bind_hash_remove(tcp);
	/*
	 * If the tcp_time_wait_collector (which runs outside the squeue)
	 * is trying to remove this tcp from the time wait list, we will
	 * block in tcp_time_wait_remove while trying to acquire the
	 * tcp_time_wait_lock. The logic in tcp_time_wait_collector also
	 * requires the ipcl_hash_remove to be ordered after the
	 * tcp_time_wait_remove for the refcnt checks to work correctly.
	 */
	if (tcp->tcp_state == TCPS_TIME_WAIT)
		(void) tcp_time_wait_remove(tcp, NULL);
	CL_INET_DISCONNECT(connp);
	ipcl_hash_remove(connp);
	ixa_cleanup(connp->conn_ixa);

	/*
	 * Mark the conn as CONDEMNED
	 */
	mutex_enter(&connp->conn_lock);
	connp->conn_state_flags |= CONN_CONDEMNED;
	mutex_exit(&connp->conn_lock);

	/* Need to cleanup any pending ioctls */
	ASSERT(tcp->tcp_time_wait_next == NULL);
	ASSERT(tcp->tcp_time_wait_prev == NULL);
	ASSERT(tcp->tcp_time_wait_expire == 0);
	tcp->tcp_state = TCPS_CLOSED;

	/* Release any SSL context */
	if (tcp->tcp_kssl_ent != NULL) {
		kssl_release_ent(tcp->tcp_kssl_ent, NULL, KSSL_NO_PROXY);
		tcp->tcp_kssl_ent = NULL;
	}
	if (tcp->tcp_kssl_ctx != NULL) {
		kssl_release_ctx(tcp->tcp_kssl_ctx);
		tcp->tcp_kssl_ctx = NULL;
	}
	tcp->tcp_kssl_pending = B_FALSE;

	tcp_ipsec_cleanup(tcp);
}

/*
 * tcp is dying (called from ipcl_conn_destroy and error cases).
 * Free the tcp_t in either case.
 */
void
tcp_free(tcp_t *tcp)
{
	mblk_t		*mp;
	conn_t		*connp = tcp->tcp_connp;

	ASSERT(tcp != NULL);
	ASSERT(tcp->tcp_ptpahn == NULL && tcp->tcp_acceptor_hash == NULL);

	connp->conn_rq = NULL;
	connp->conn_wq = NULL;

	tcp_close_mpp(&tcp->tcp_xmit_head);
	tcp_close_mpp(&tcp->tcp_reass_head);
	if (tcp->tcp_rcv_list != NULL) {
		/* Free b_next chain */
		tcp_close_mpp(&tcp->tcp_rcv_list);
	}
	if ((mp = tcp->tcp_urp_mp) != NULL) {
		freemsg(mp);
	}
	if ((mp = tcp->tcp_urp_mark_mp) != NULL) {
		freemsg(mp);
	}

	if (tcp->tcp_fused_sigurg_mp != NULL) {
		ASSERT(!IPCL_IS_NONSTR(tcp->tcp_connp));
		freeb(tcp->tcp_fused_sigurg_mp);
		tcp->tcp_fused_sigurg_mp = NULL;
	}

	if (tcp->tcp_ordrel_mp != NULL) {
		ASSERT(!IPCL_IS_NONSTR(tcp->tcp_connp));
		freeb(tcp->tcp_ordrel_mp);
		tcp->tcp_ordrel_mp = NULL;
	}

	if (tcp->tcp_sack_info != NULL) {
		if (tcp->tcp_notsack_list != NULL) {
			TCP_NOTSACK_REMOVE_ALL(tcp->tcp_notsack_list,
			    tcp);
		}
		bzero(tcp->tcp_sack_info, sizeof (tcp_sack_info_t));
	}

	if (tcp->tcp_hopopts != NULL) {
		mi_free(tcp->tcp_hopopts);
		tcp->tcp_hopopts = NULL;
		tcp->tcp_hopoptslen = 0;
	}
	ASSERT(tcp->tcp_hopoptslen == 0);
	if (tcp->tcp_dstopts != NULL) {
		mi_free(tcp->tcp_dstopts);
		tcp->tcp_dstopts = NULL;
		tcp->tcp_dstoptslen = 0;
	}
	ASSERT(tcp->tcp_dstoptslen == 0);
	if (tcp->tcp_rthdrdstopts != NULL) {
		mi_free(tcp->tcp_rthdrdstopts);
		tcp->tcp_rthdrdstopts = NULL;
		tcp->tcp_rthdrdstoptslen = 0;
	}
	ASSERT(tcp->tcp_rthdrdstoptslen == 0);
	if (tcp->tcp_rthdr != NULL) {
		mi_free(tcp->tcp_rthdr);
		tcp->tcp_rthdr = NULL;
		tcp->tcp_rthdrlen = 0;
	}
	ASSERT(tcp->tcp_rthdrlen == 0);

	/*
	 * Following is really a blowing away a union.
	 * It happens to have exactly two members of identical size
	 * the following code is enough.
	 */
	tcp_close_mpp(&tcp->tcp_conn.tcp_eager_conn_ind);
}


/*
 * Put a connection confirmation message upstream built from the
 * address/flowid information with the conn and iph. Report our success or
 * failure.
 */
static boolean_t
tcp_conn_con(tcp_t *tcp, uchar_t *iphdr, mblk_t *idmp,
    mblk_t **defermp, ip_recv_attr_t *ira)
{
	sin_t	sin;
	sin6_t	sin6;
	mblk_t	*mp;
	char	*optp = NULL;
	int	optlen = 0;
	conn_t	*connp = tcp->tcp_connp;

	if (defermp != NULL)
		*defermp = NULL;

	if (tcp->tcp_conn.tcp_opts_conn_req != NULL) {
		/*
		 * Return in T_CONN_CON results of option negotiation through
		 * the T_CONN_REQ. Note: If there is an real end-to-end option
		 * negotiation, then what is received from remote end needs
		 * to be taken into account but there is no such thing (yet?)
		 * in our TCP/IP.
		 * Note: We do not use mi_offset_param() here as
		 * tcp_opts_conn_req contents do not directly come from
		 * an application and are either generated in kernel or
		 * from user input that was already verified.
		 */
		mp = tcp->tcp_conn.tcp_opts_conn_req;
		optp = (char *)(mp->b_rptr +
		    ((struct T_conn_req *)mp->b_rptr)->OPT_offset);
		optlen = (int)
		    ((struct T_conn_req *)mp->b_rptr)->OPT_length;
	}

	if (IPH_HDR_VERSION(iphdr) == IPV4_VERSION) {

		/* packet is IPv4 */
		if (connp->conn_family == AF_INET) {
			sin = sin_null;
			sin.sin_addr.s_addr = connp->conn_faddr_v4;
			sin.sin_port = connp->conn_fport;
			sin.sin_family = AF_INET;
			mp = mi_tpi_conn_con(NULL, (char *)&sin,
			    (int)sizeof (sin_t), optp, optlen);
		} else {
			sin6 = sin6_null;
			sin6.sin6_addr = connp->conn_faddr_v6;
			sin6.sin6_port = connp->conn_fport;
			sin6.sin6_family = AF_INET6;
			mp = mi_tpi_conn_con(NULL, (char *)&sin6,
			    (int)sizeof (sin6_t), optp, optlen);

		}
	} else {
		ip6_t	*ip6h = (ip6_t *)iphdr;

		ASSERT(IPH_HDR_VERSION(iphdr) == IPV6_VERSION);
		ASSERT(connp->conn_family == AF_INET6);
		sin6 = sin6_null;
		sin6.sin6_addr = connp->conn_faddr_v6;
		sin6.sin6_port = connp->conn_fport;
		sin6.sin6_family = AF_INET6;
		sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
		mp = mi_tpi_conn_con(NULL, (char *)&sin6,
		    (int)sizeof (sin6_t), optp, optlen);
	}

	if (!mp)
		return (B_FALSE);

	mblk_copycred(mp, idmp);

	if (defermp == NULL) {
		conn_t *connp = tcp->tcp_connp;
		if (IPCL_IS_NONSTR(connp)) {
			(*connp->conn_upcalls->su_connected)
			    (connp->conn_upper_handle, tcp->tcp_connid,
			    ira->ira_cred, ira->ira_cpid);
			freemsg(mp);
		} else {
			if (ira->ira_cred != NULL) {
				/* So that getpeerucred works for TPI sockfs */
				mblk_setcred(mp, ira->ira_cred, ira->ira_cpid);
			}
			putnext(connp->conn_rq, mp);
		}
	} else {
		*defermp = mp;
	}

	if (tcp->tcp_conn.tcp_opts_conn_req != NULL)
		tcp_close_mpp(&tcp->tcp_conn.tcp_opts_conn_req);
	return (B_TRUE);
}

/*
 * Defense for the SYN attack -
 * 1. When q0 is full, drop from the tail (tcp_eager_prev_drop_q0) the oldest
 *    one from the list of droppable eagers. This list is a subset of q0.
 *    see comments before the definition of MAKE_DROPPABLE().
 * 2. Don't drop a SYN request before its first timeout. This gives every
 *    request at least til the first timeout to complete its 3-way handshake.
 * 3. Maintain tcp_syn_rcvd_timeout as an accurate count of how many
 *    requests currently on the queue that has timed out. This will be used
 *    as an indicator of whether an attack is under way, so that appropriate
 *    actions can be taken. (It's incremented in tcp_timer() and decremented
 *    either when eager goes into ESTABLISHED, or gets freed up.)
 * 4. The current threshold is - # of timeout > q0len/4 => SYN alert on
 *    # of timeout drops back to <= q0len/32 => SYN alert off
 */
static boolean_t
tcp_drop_q0(tcp_t *tcp)
{
	tcp_t	*eager;
	mblk_t	*mp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	ASSERT(MUTEX_HELD(&tcp->tcp_eager_lock));
	ASSERT(tcp->tcp_eager_next_q0 != tcp->tcp_eager_prev_q0);

	/* Pick oldest eager from the list of droppable eagers */
	eager = tcp->tcp_eager_prev_drop_q0;

	/* If list is empty. return B_FALSE */
	if (eager == tcp) {
		return (B_FALSE);
	}

	/* If allocated, the mp will be freed in tcp_clean_death_wrapper() */
	if ((mp = allocb(0, BPRI_HI)) == NULL)
		return (B_FALSE);

	/*
	 * Take this eager out from the list of droppable eagers since we are
	 * going to drop it.
	 */
	MAKE_UNDROPPABLE(eager);

	if (tcp->tcp_connp->conn_debug) {
		(void) strlog(TCP_MOD_ID, 0, 3, SL_TRACE,
		    "tcp_drop_q0: listen half-open queue (max=%d) overflow"
		    " (%d pending) on %s, drop one", tcps->tcps_conn_req_max_q0,
		    tcp->tcp_conn_req_cnt_q0,
		    tcp_display(tcp, NULL, DISP_PORT_ONLY));
	}

	BUMP_MIB(&tcps->tcps_mib, tcpHalfOpenDrop);

	/* Put a reference on the conn as we are enqueueing it in the sqeue */
	CONN_INC_REF(eager->tcp_connp);

	SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
	    tcp_clean_death_wrapper, eager->tcp_connp, NULL,
	    SQ_FILL, SQTAG_TCP_DROP_Q0);

	return (B_TRUE);
}

/*
 * Handle a SYN on an AF_INET6 socket; can be either IPv4 or IPv6
 */
static mblk_t *
tcp_conn_create_v6(conn_t *lconnp, conn_t *connp, mblk_t *mp,
    ip_recv_attr_t *ira)
{
	tcp_t 		*ltcp = lconnp->conn_tcp;
	tcp_t		*tcp = connp->conn_tcp;
	mblk_t		*tpi_mp;
	ipha_t		*ipha;
	ip6_t		*ip6h;
	sin6_t 		sin6;
	uint_t		ifindex = ira->ira_ruifindex;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	if (ira->ira_flags & IRAF_IS_IPV4) {
		ipha = (ipha_t *)mp->b_rptr;

		connp->conn_ipversion = IPV4_VERSION;
		IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &connp->conn_laddr_v6);
		IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &connp->conn_faddr_v6);
		connp->conn_saddr_v6 = connp->conn_laddr_v6;

		sin6 = sin6_null;
		sin6.sin6_addr = connp->conn_faddr_v6;
		sin6.sin6_port = connp->conn_fport;
		sin6.sin6_family = AF_INET6;
		sin6.__sin6_src_id = ip_srcid_find_addr(&connp->conn_laddr_v6,
		    IPCL_ZONEID(lconnp), tcps->tcps_netstack);

		if (connp->conn_recv_ancillary.crb_recvdstaddr) {
			sin6_t	sin6d;

			sin6d = sin6_null;
			sin6d.sin6_addr = connp->conn_laddr_v6;
			sin6d.sin6_port = connp->conn_lport;
			sin6d.sin6_family = AF_INET;
			tpi_mp = mi_tpi_extconn_ind(NULL,
			    (char *)&sin6d, sizeof (sin6_t),
			    (char *)&tcp,
			    (t_scalar_t)sizeof (intptr_t),
			    (char *)&sin6d, sizeof (sin6_t),
			    (t_scalar_t)ltcp->tcp_conn_req_seqnum);
		} else {
			tpi_mp = mi_tpi_conn_ind(NULL,
			    (char *)&sin6, sizeof (sin6_t),
			    (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
			    (t_scalar_t)ltcp->tcp_conn_req_seqnum);
		}
	} else {
		ip6h = (ip6_t *)mp->b_rptr;

		connp->conn_ipversion = IPV6_VERSION;
		connp->conn_laddr_v6 = ip6h->ip6_dst;
		connp->conn_faddr_v6 = ip6h->ip6_src;
		connp->conn_saddr_v6 = connp->conn_laddr_v6;

		sin6 = sin6_null;
		sin6.sin6_addr = connp->conn_faddr_v6;
		sin6.sin6_port = connp->conn_fport;
		sin6.sin6_family = AF_INET6;
		sin6.sin6_flowinfo = ip6h->ip6_vcf & ~IPV6_VERS_AND_FLOW_MASK;
		sin6.__sin6_src_id = ip_srcid_find_addr(&connp->conn_laddr_v6,
		    IPCL_ZONEID(lconnp), tcps->tcps_netstack);

		if (IN6_IS_ADDR_LINKSCOPE(&ip6h->ip6_src)) {
			/* Pass up the scope_id of remote addr */
			sin6.sin6_scope_id = ifindex;
		} else {
			sin6.sin6_scope_id = 0;
		}
		if (connp->conn_recv_ancillary.crb_recvdstaddr) {
			sin6_t	sin6d;

			sin6d = sin6_null;
			sin6.sin6_addr = connp->conn_laddr_v6;
			sin6d.sin6_port = connp->conn_lport;
			sin6d.sin6_family = AF_INET6;
			if (IN6_IS_ADDR_LINKSCOPE(&connp->conn_laddr_v6))
				sin6d.sin6_scope_id = ifindex;

			tpi_mp = mi_tpi_extconn_ind(NULL,
			    (char *)&sin6d, sizeof (sin6_t),
			    (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
			    (char *)&sin6d, sizeof (sin6_t),
			    (t_scalar_t)ltcp->tcp_conn_req_seqnum);
		} else {
			tpi_mp = mi_tpi_conn_ind(NULL,
			    (char *)&sin6, sizeof (sin6_t),
			    (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
			    (t_scalar_t)ltcp->tcp_conn_req_seqnum);
		}
	}

	tcp->tcp_mss = tcps->tcps_mss_def_ipv6;
	return (tpi_mp);
}

/* Handle a SYN on an AF_INET socket */
mblk_t *
tcp_conn_create_v4(conn_t *lconnp, conn_t *connp, mblk_t *mp,
    ip_recv_attr_t *ira)
{
	tcp_t 		*ltcp = lconnp->conn_tcp;
	tcp_t		*tcp = connp->conn_tcp;
	sin_t		sin;
	mblk_t		*tpi_mp = NULL;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	ipha_t		*ipha;

	ASSERT(ira->ira_flags & IRAF_IS_IPV4);
	ipha = (ipha_t *)mp->b_rptr;

	connp->conn_ipversion = IPV4_VERSION;
	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &connp->conn_laddr_v6);
	IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &connp->conn_faddr_v6);
	connp->conn_saddr_v6 = connp->conn_laddr_v6;

	sin = sin_null;
	sin.sin_addr.s_addr = connp->conn_faddr_v4;
	sin.sin_port = connp->conn_fport;
	sin.sin_family = AF_INET;
	if (lconnp->conn_recv_ancillary.crb_recvdstaddr) {
		sin_t	sind;

		sind = sin_null;
		sind.sin_addr.s_addr = connp->conn_laddr_v4;
		sind.sin_port = connp->conn_lport;
		sind.sin_family = AF_INET;
		tpi_mp = mi_tpi_extconn_ind(NULL,
		    (char *)&sind, sizeof (sin_t), (char *)&tcp,
		    (t_scalar_t)sizeof (intptr_t), (char *)&sind,
		    sizeof (sin_t), (t_scalar_t)ltcp->tcp_conn_req_seqnum);
	} else {
		tpi_mp = mi_tpi_conn_ind(NULL,
		    (char *)&sin, sizeof (sin_t),
		    (char *)&tcp, (t_scalar_t)sizeof (intptr_t),
		    (t_scalar_t)ltcp->tcp_conn_req_seqnum);
	}

	tcp->tcp_mss = tcps->tcps_mss_def_ipv4;
	return (tpi_mp);
}

/*
 * tcp_get_conn/tcp_free_conn
 *
 * tcp_get_conn is used to get a clean tcp connection structure.
 * It tries to reuse the connections put on the freelist by the
 * time_wait_collector failing which it goes to kmem_cache. This
 * way has two benefits compared to just allocating from and
 * freeing to kmem_cache.
 * 1) The time_wait_collector can free (which includes the cleanup)
 * outside the squeue. So when the interrupt comes, we have a clean
 * connection sitting in the freelist. Obviously, this buys us
 * performance.
 *
 * 2) Defence against DOS attack. Allocating a tcp/conn in tcp_input_listener
 * has multiple disadvantages - tying up the squeue during alloc.
 * But allocating the conn/tcp in IP land is also not the best since
 * we can't check the 'q' and 'q0' which are protected by squeue and
 * blindly allocate memory which might have to be freed here if we are
 * not allowed to accept the connection. By using the freelist and
 * putting the conn/tcp back in freelist, we don't pay a penalty for
 * allocating memory without checking 'q/q0' and freeing it if we can't
 * accept the connection.
 *
 * Care should be taken to put the conn back in the same squeue's freelist
 * from which it was allocated. Best results are obtained if conn is
 * allocated from listener's squeue and freed to the same. Time wait
 * collector will free up the freelist is the connection ends up sitting
 * there for too long.
 */
void *
tcp_get_conn(void *arg, tcp_stack_t *tcps)
{
	tcp_t			*tcp = NULL;
	conn_t			*connp = NULL;
	squeue_t		*sqp = (squeue_t *)arg;
	tcp_squeue_priv_t 	*tcp_time_wait;
	netstack_t		*ns;
	mblk_t			*tcp_rsrv_mp = NULL;

	tcp_time_wait =
	    *((tcp_squeue_priv_t **)squeue_getprivate(sqp, SQPRIVATE_TCP));

	mutex_enter(&tcp_time_wait->tcp_time_wait_lock);
	tcp = tcp_time_wait->tcp_free_list;
	ASSERT((tcp != NULL) ^ (tcp_time_wait->tcp_free_list_cnt == 0));
	if (tcp != NULL) {
		tcp_time_wait->tcp_free_list = tcp->tcp_time_wait_next;
		tcp_time_wait->tcp_free_list_cnt--;
		mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
		tcp->tcp_time_wait_next = NULL;
		connp = tcp->tcp_connp;
		connp->conn_flags |= IPCL_REUSED;

		ASSERT(tcp->tcp_tcps == NULL);
		ASSERT(connp->conn_netstack == NULL);
		ASSERT(tcp->tcp_rsrv_mp != NULL);
		ns = tcps->tcps_netstack;
		netstack_hold(ns);
		connp->conn_netstack = ns;
		connp->conn_ixa->ixa_ipst = ns->netstack_ip;
		tcp->tcp_tcps = tcps;
		ipcl_globalhash_insert(connp);

		connp->conn_ixa->ixa_notify_cookie = tcp;
		ASSERT(connp->conn_ixa->ixa_notify == tcp_notify);
		connp->conn_recv = tcp_input_data;
		ASSERT(connp->conn_recvicmp == tcp_icmp_input);
		ASSERT(connp->conn_verifyicmp == tcp_verifyicmp);
		return ((void *)connp);
	}
	mutex_exit(&tcp_time_wait->tcp_time_wait_lock);
	/*
	 * Pre-allocate the tcp_rsrv_mp. This mblk will not be freed until
	 * this conn_t/tcp_t is freed at ipcl_conn_destroy().
	 */
	tcp_rsrv_mp = allocb(0, BPRI_HI);
	if (tcp_rsrv_mp == NULL)
		return (NULL);

	if ((connp = ipcl_conn_create(IPCL_TCPCONN, KM_NOSLEEP,
	    tcps->tcps_netstack)) == NULL) {
		freeb(tcp_rsrv_mp);
		return (NULL);
	}

	tcp = connp->conn_tcp;
	tcp->tcp_rsrv_mp = tcp_rsrv_mp;
	mutex_init(&tcp->tcp_rsrv_mp_lock, NULL, MUTEX_DEFAULT, NULL);

	tcp->tcp_tcps = tcps;

	connp->conn_recv = tcp_input_data;
	connp->conn_recvicmp = tcp_icmp_input;
	connp->conn_verifyicmp = tcp_verifyicmp;

	/*
	 * Register tcp_notify to listen to capability changes detected by IP.
	 * This upcall is made in the context of the call to conn_ip_output
	 * thus it is inside the squeue.
	 */
	connp->conn_ixa->ixa_notify = tcp_notify;
	connp->conn_ixa->ixa_notify_cookie = tcp;

	return ((void *)connp);
}

/* BEGIN CSTYLED */
/*
 *
 * The sockfs ACCEPT path:
 * =======================
 *
 * The eager is now established in its own perimeter as soon as SYN is
 * received in tcp_input_listener(). When sockfs receives conn_ind, it
 * completes the accept processing on the acceptor STREAM. The sending
 * of conn_ind part is common for both sockfs listener and a TLI/XTI
 * listener but a TLI/XTI listener completes the accept processing
 * on the listener perimeter.
 *
 * Common control flow for 3 way handshake:
 * ----------------------------------------
 *
 * incoming SYN (listener perimeter)	-> tcp_input_listener()
 *
 * incoming SYN-ACK-ACK (eager perim) 	-> tcp_input_data()
 * send T_CONN_IND (listener perim)	-> tcp_send_conn_ind()
 *
 * Sockfs ACCEPT Path:
 * -------------------
 *
 * open acceptor stream (tcp_open allocates tcp_tli_accept()
 * as STREAM entry point)
 *
 * soaccept() sends T_CONN_RES on the acceptor STREAM to tcp_tli_accept()
 *
 * tcp_tli_accept() extracts the eager and makes the q->q_ptr <-> eager
 * association (we are not behind eager's squeue but sockfs is protecting us
 * and no one knows about this stream yet. The STREAMS entry point q->q_info
 * is changed to point at tcp_wput().
 *
 * tcp_accept_common() sends any deferred eagers via tcp_send_pending() to
 * listener (done on listener's perimeter).
 *
 * tcp_tli_accept() calls tcp_accept_finish() on eagers perimeter to finish
 * accept.
 *
 * TLI/XTI client ACCEPT path:
 * ---------------------------
 *
 * soaccept() sends T_CONN_RES on the listener STREAM.
 *
 * tcp_tli_accept() -> tcp_accept_swap() complete the processing and send
 * a M_SETOPS mblk to eager perimeter to finish accept (tcp_accept_finish()).
 *
 * Locks:
 * ======
 *
 * listener->tcp_eager_lock protects the listeners->tcp_eager_next_q0 and
 * and listeners->tcp_eager_next_q.
 *
 * Referencing:
 * ============
 *
 * 1) We start out in tcp_input_listener by eager placing a ref on
 * listener and listener adding eager to listeners->tcp_eager_next_q0.
 *
 * 2) When a SYN-ACK-ACK arrives, we send the conn_ind to listener. Before
 * doing so we place a ref on the eager. This ref is finally dropped at the
 * end of tcp_accept_finish() while unwinding from the squeue, i.e. the
 * reference is dropped by the squeue framework.
 *
 * 3) The ref on listener placed in 1 above is dropped in tcp_accept_finish
 *
 * The reference must be released by the same entity that added the reference
 * In the above scheme, the eager is the entity that adds and releases the
 * references. Note that tcp_accept_finish executes in the squeue of the eager
 * (albeit after it is attached to the acceptor stream). Though 1. executes
 * in the listener's squeue, the eager is nascent at this point and the
 * reference can be considered to have been added on behalf of the eager.
 *
 * Eager getting a Reset or listener closing:
 * ==========================================
 *
 * Once the listener and eager are linked, the listener never does the unlink.
 * If the listener needs to close, tcp_eager_cleanup() is called which queues
 * a message on all eager perimeter. The eager then does the unlink, clears
 * any pointers to the listener's queue and drops the reference to the
 * listener. The listener waits in tcp_close outside the squeue until its
 * refcount has dropped to 1. This ensures that the listener has waited for
 * all eagers to clear their association with the listener.
 *
 * Similarly, if eager decides to go away, it can unlink itself and close.
 * When the T_CONN_RES comes down, we check if eager has closed. Note that
 * the reference to eager is still valid because of the extra ref we put
 * in tcp_send_conn_ind.
 *
 * Listener can always locate the eager under the protection
 * of the listener->tcp_eager_lock, and then do a refhold
 * on the eager during the accept processing.
 *
 * The acceptor stream accesses the eager in the accept processing
 * based on the ref placed on eager before sending T_conn_ind.
 * The only entity that can negate this refhold is a listener close
 * which is mutually exclusive with an active acceptor stream.
 *
 * Eager's reference on the listener
 * ===================================
 *
 * If the accept happens (even on a closed eager) the eager drops its
 * reference on the listener at the start of tcp_accept_finish. If the
 * eager is killed due to an incoming RST before the T_conn_ind is sent up,
 * the reference is dropped in tcp_closei_local. If the listener closes,
 * the reference is dropped in tcp_eager_kill. In all cases the reference
 * is dropped while executing in the eager's context (squeue).
 */
/* END CSTYLED */

/* Process the SYN packet, mp, directed at the listener 'tcp' */

/*
 * THIS FUNCTION IS DIRECTLY CALLED BY IP VIA SQUEUE FOR SYN.
 * tcp_input_data will not see any packets for listeners since the listener
 * has conn_recv set to tcp_input_listener.
 */
/* ARGSUSED */
void
tcp_input_listener(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
{
	tcpha_t		*tcpha;
	uint32_t	seg_seq;
	tcp_t		*eager;
	int		err;
	conn_t		*econnp = NULL;
	squeue_t	*new_sqp;
	mblk_t		*mp1;
	uint_t 		ip_hdr_len;
	conn_t		*lconnp = (conn_t *)arg;
	tcp_t		*listener = lconnp->conn_tcp;
	tcp_stack_t	*tcps = listener->tcp_tcps;
	ip_stack_t	*ipst = tcps->tcps_netstack->netstack_ip;
	uint_t		flags;
	mblk_t		*tpi_mp;
	uint_t		ifindex = ira->ira_ruifindex;

	ip_hdr_len = ira->ira_ip_hdr_length;
	tcpha = (tcpha_t *)&mp->b_rptr[ip_hdr_len];
	flags = (unsigned int)tcpha->tha_flags & 0xFF;

	if (!(flags & TH_SYN)) {
		if ((flags & TH_RST) || (flags & TH_URG)) {
			freemsg(mp);
			return;
		}
		if (flags & TH_ACK) {
			/* Note this executes in listener's squeue */
			tcp_xmit_listeners_reset(mp, ira, ipst, lconnp);
			return;
		}

		freemsg(mp);
		return;
	}

	if (listener->tcp_state != TCPS_LISTEN)
		goto error2;

	ASSERT(IPCL_IS_BOUND(lconnp));

	mutex_enter(&listener->tcp_eager_lock);
	if (listener->tcp_conn_req_cnt_q >= listener->tcp_conn_req_max) {
		mutex_exit(&listener->tcp_eager_lock);
		TCP_STAT(tcps, tcp_listendrop);
		BUMP_MIB(&tcps->tcps_mib, tcpListenDrop);
		if (lconnp->conn_debug) {
			(void) strlog(TCP_MOD_ID, 0, 1, SL_TRACE|SL_ERROR,
			    "tcp_input_listener: listen backlog (max=%d) "
			    "overflow (%d pending) on %s",
			    listener->tcp_conn_req_max,
			    listener->tcp_conn_req_cnt_q,
			    tcp_display(listener, NULL, DISP_PORT_ONLY));
		}
		goto error2;
	}

	if (listener->tcp_conn_req_cnt_q0 >=
	    listener->tcp_conn_req_max + tcps->tcps_conn_req_max_q0) {
		/*
		 * Q0 is full. Drop a pending half-open req from the queue
		 * to make room for the new SYN req. Also mark the time we
		 * drop a SYN.
		 *
		 * A more aggressive defense against SYN attack will
		 * be to set the "tcp_syn_defense" flag now.
		 */
		TCP_STAT(tcps, tcp_listendropq0);
		listener->tcp_last_rcv_lbolt = ddi_get_lbolt64();
		if (!tcp_drop_q0(listener)) {
			mutex_exit(&listener->tcp_eager_lock);
			BUMP_MIB(&tcps->tcps_mib, tcpListenDropQ0);
			if (lconnp->conn_debug) {
				(void) strlog(TCP_MOD_ID, 0, 3, SL_TRACE,
				    "tcp_input_listener: listen half-open "
				    "queue (max=%d) full (%d pending) on %s",
				    tcps->tcps_conn_req_max_q0,
				    listener->tcp_conn_req_cnt_q0,
				    tcp_display(listener, NULL,
				    DISP_PORT_ONLY));
			}
			goto error2;
		}
	}
	mutex_exit(&listener->tcp_eager_lock);

	/*
	 * IP sets ira_sqp to either the senders conn_sqp (for loopback)
	 * or based on the ring (for packets from GLD). Otherwise it is
	 * set based on lbolt i.e., a somewhat random number.
	 */
	ASSERT(ira->ira_sqp != NULL);
	new_sqp = ira->ira_sqp;

	econnp = (conn_t *)tcp_get_conn(arg2, tcps);
	if (econnp == NULL)
		goto error2;

	ASSERT(econnp->conn_netstack == lconnp->conn_netstack);
	econnp->conn_sqp = new_sqp;
	econnp->conn_initial_sqp = new_sqp;
	econnp->conn_ixa->ixa_sqp = new_sqp;

	econnp->conn_fport = tcpha->tha_lport;
	econnp->conn_lport = tcpha->tha_fport;

	err = conn_inherit_parent(lconnp, econnp);
	if (err != 0)
		goto error3;

	ASSERT(OK_32PTR(mp->b_rptr));
	ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION ||
	    IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);

	if (lconnp->conn_family == AF_INET) {
		ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION);
		tpi_mp = tcp_conn_create_v4(lconnp, econnp, mp, ira);
	} else {
		tpi_mp = tcp_conn_create_v6(lconnp, econnp, mp, ira);
	}

	if (tpi_mp == NULL)
		goto error3;

	eager = econnp->conn_tcp;
	eager->tcp_detached = B_TRUE;
	SOCK_CONNID_INIT(eager->tcp_connid);

	tcp_init_values(eager);

	ASSERT((econnp->conn_ixa->ixa_flags &
	    (IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
	    IXAF_VERIFY_PMTU | IXAF_VERIFY_LSO)) ==
	    (IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
	    IXAF_VERIFY_PMTU | IXAF_VERIFY_LSO));

	if (!tcps->tcps_dev_flow_ctl)
		econnp->conn_ixa->ixa_flags |= IXAF_NO_DEV_FLOW_CTL;

	/* Prepare for diffing against previous packets */
	eager->tcp_recvifindex = 0;
	eager->tcp_recvhops = 0xffffffffU;

	if (!(ira->ira_flags & IRAF_IS_IPV4) && econnp->conn_bound_if == 0) {
		if (IN6_IS_ADDR_LINKSCOPE(&econnp->conn_faddr_v6) ||
		    IN6_IS_ADDR_LINKSCOPE(&econnp->conn_laddr_v6)) {
			econnp->conn_incoming_ifindex = ifindex;
			econnp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
			econnp->conn_ixa->ixa_scopeid = ifindex;
		}
	}

	if ((ira->ira_flags & (IRAF_IS_IPV4|IRAF_IPV4_OPTIONS)) ==
	    (IRAF_IS_IPV4|IRAF_IPV4_OPTIONS) &&
	    tcps->tcps_rev_src_routes) {
		ipha_t *ipha = (ipha_t *)mp->b_rptr;
		ip_pkt_t *ipp = &econnp->conn_xmit_ipp;

		/* Source routing option copyover (reverse it) */
		err = ip_find_hdr_v4(ipha, ipp, B_TRUE);
		if (err != 0) {
			freemsg(tpi_mp);
			goto error3;
		}
		ip_pkt_source_route_reverse_v4(ipp);
	}

	ASSERT(eager->tcp_conn.tcp_eager_conn_ind == NULL);
	ASSERT(!eager->tcp_tconnind_started);
	/*
	 * If the SYN came with a credential, it's a loopback packet or a
	 * labeled packet; attach the credential to the TPI message.
	 */
	if (ira->ira_cred != NULL)
		mblk_setcred(tpi_mp, ira->ira_cred, ira->ira_cpid);

	eager->tcp_conn.tcp_eager_conn_ind = tpi_mp;

	/* Inherit the listener's SSL protection state */
	if ((eager->tcp_kssl_ent = listener->tcp_kssl_ent) != NULL) {
		kssl_hold_ent(eager->tcp_kssl_ent);
		eager->tcp_kssl_pending = B_TRUE;
	}

	/* Inherit the listener's non-STREAMS flag */
	if (IPCL_IS_NONSTR(lconnp)) {
		econnp->conn_flags |= IPCL_NONSTR;
	}

	ASSERT(eager->tcp_ordrel_mp == NULL);

	if (!IPCL_IS_NONSTR(econnp)) {
		/*
		 * Pre-allocate the T_ordrel_ind mblk for TPI socket so that
		 * at close time, we will always have that to send up.
		 * Otherwise, we need to do special handling in case the
		 * allocation fails at that time.
		 */
		if ((eager->tcp_ordrel_mp = mi_tpi_ordrel_ind()) == NULL)
			goto error3;
	}
	/*
	 * Now that the IP addresses and ports are setup in econnp we
	 * can do the IPsec policy work.
	 */
	if (ira->ira_flags & IRAF_IPSEC_SECURE) {
		if (lconnp->conn_policy != NULL) {
			/*
			 * Inherit the policy from the listener; use
			 * actions from ira
			 */
			if (!ip_ipsec_policy_inherit(econnp, lconnp, ira)) {
				CONN_DEC_REF(econnp);
				freemsg(mp);
				goto error3;
			}
		}
	}

	/* Inherit various TCP parameters from the listener */
	eager->tcp_naglim = listener->tcp_naglim;
	eager->tcp_first_timer_threshold = listener->tcp_first_timer_threshold;
	eager->tcp_second_timer_threshold =
	    listener->tcp_second_timer_threshold;
	eager->tcp_first_ctimer_threshold =
	    listener->tcp_first_ctimer_threshold;
	eager->tcp_second_ctimer_threshold =
	    listener->tcp_second_ctimer_threshold;

	/*
	 * tcp_set_destination() may set tcp_rwnd according to the route
	 * metrics. If it does not, the eager's receive window will be set
	 * to the listener's receive window later in this function.
	 */
	eager->tcp_rwnd = 0;

	/*
	 * Inherit listener's tcp_init_cwnd.  Need to do this before
	 * calling tcp_process_options() which set the initial cwnd.
	 */
	eager->tcp_init_cwnd = listener->tcp_init_cwnd;

	if (is_system_labeled()) {
		ip_xmit_attr_t *ixa = econnp->conn_ixa;

		ASSERT(ira->ira_tsl != NULL);
		/* Discard any old label */
		if (ixa->ixa_free_flags & IXA_FREE_TSL) {
			ASSERT(ixa->ixa_tsl != NULL);
			label_rele(ixa->ixa_tsl);
			ixa->ixa_free_flags &= ~IXA_FREE_TSL;
			ixa->ixa_tsl = NULL;
		}
		if ((lconnp->conn_mlp_type != mlptSingle ||
		    lconnp->conn_mac_mode != CONN_MAC_DEFAULT) &&
		    ira->ira_tsl != NULL) {
			/*
			 * If this is an MLP connection or a MAC-Exempt
			 * connection with an unlabeled node, packets are to be
			 * exchanged using the security label of the received
			 * SYN packet instead of the server application's label.
			 * tsol_check_dest called from ip_set_destination
			 * might later update TSF_UNLABELED by replacing
			 * ixa_tsl with a new label.
			 */
			label_hold(ira->ira_tsl);
			ip_xmit_attr_replace_tsl(ixa, ira->ira_tsl);
			DTRACE_PROBE2(mlp_syn_accept, conn_t *,
			    econnp, ts_label_t *, ixa->ixa_tsl)
		} else {
			ixa->ixa_tsl = crgetlabel(econnp->conn_cred);
			DTRACE_PROBE2(syn_accept, conn_t *,
			    econnp, ts_label_t *, ixa->ixa_tsl)
		}
		/*
		 * conn_connect() called from tcp_set_destination will verify
		 * the destination is allowed to receive packets at the
		 * security label of the SYN-ACK we are generating. As part of
		 * that, tsol_check_dest() may create a new effective label for
		 * this connection.
		 * Finally conn_connect() will call conn_update_label.
		 * All that remains for TCP to do is to call
		 * conn_build_hdr_template which is done as part of
		 * tcp_set_destination.
		 */
	}

	/*
	 * Since we will clear tcp_listener before we clear tcp_detached
	 * in the accept code we need tcp_hard_binding aka tcp_accept_inprogress
	 * so we can tell a TCP_DETACHED_NONEAGER apart.
	 */
	eager->tcp_hard_binding = B_TRUE;

	tcp_bind_hash_insert(&tcps->tcps_bind_fanout[
	    TCP_BIND_HASH(econnp->conn_lport)], eager, 0);

	CL_INET_CONNECT(econnp, B_FALSE, err);
	if (err != 0) {
		tcp_bind_hash_remove(eager);
		goto error3;
	}

	/*
	 * No need to check for multicast destination since ip will only pass
	 * up multicasts to those that have expressed interest
	 * TODO: what about rejecting broadcasts?
	 * Also check that source is not a multicast or broadcast address.
	 */
	eager->tcp_state = TCPS_SYN_RCVD;
	SOCK_CONNID_BUMP(eager->tcp_connid);

	/*
	 * Adapt our mss, ttl, ... based on the remote address.
	 */

	if (tcp_set_destination(eager) != 0) {
		BUMP_MIB(&tcps->tcps_mib, tcpAttemptFails);
		/* Undo the bind_hash_insert */
		tcp_bind_hash_remove(eager);
		goto error3;
	}

	/* Process all TCP options. */
	tcp_process_options(eager, tcpha);

	/* Is the other end ECN capable? */
	if (tcps->tcps_ecn_permitted >= 1 &&
	    (tcpha->tha_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) {
		eager->tcp_ecn_ok = B_TRUE;
	}

	/*
	 * The listener's conn_rcvbuf should be the default window size or a
	 * window size changed via SO_RCVBUF option. First round up the
	 * eager's tcp_rwnd to the nearest MSS. Then find out the window
	 * scale option value if needed. Call tcp_rwnd_set() to finish the
	 * setting.
	 *
	 * Note if there is a rpipe metric associated with the remote host,
	 * we should not inherit receive window size from listener.
	 */
	eager->tcp_rwnd = MSS_ROUNDUP(
	    (eager->tcp_rwnd == 0 ? econnp->conn_rcvbuf :
	    eager->tcp_rwnd), eager->tcp_mss);
	if (eager->tcp_snd_ws_ok)
		tcp_set_ws_value(eager);
	/*
	 * Note that this is the only place tcp_rwnd_set() is called for
	 * accepting a connection.  We need to call it here instead of
	 * after the 3-way handshake because we need to tell the other
	 * side our rwnd in the SYN-ACK segment.
	 */
	(void) tcp_rwnd_set(eager, eager->tcp_rwnd);

	ASSERT(eager->tcp_connp->conn_rcvbuf != 0 &&
	    eager->tcp_connp->conn_rcvbuf == eager->tcp_rwnd);

	ASSERT(econnp->conn_rcvbuf != 0 &&
	    econnp->conn_rcvbuf == eager->tcp_rwnd);

	/* Put a ref on the listener for the eager. */
	CONN_INC_REF(lconnp);
	mutex_enter(&listener->tcp_eager_lock);
	listener->tcp_eager_next_q0->tcp_eager_prev_q0 = eager;
	eager->tcp_eager_next_q0 = listener->tcp_eager_next_q0;
	listener->tcp_eager_next_q0 = eager;
	eager->tcp_eager_prev_q0 = listener;

	/* Set tcp_listener before adding it to tcp_conn_fanout */
	eager->tcp_listener = listener;
	eager->tcp_saved_listener = listener;

	/*
	 * Tag this detached tcp vector for later retrieval
	 * by our listener client in tcp_accept().
	 */
	eager->tcp_conn_req_seqnum = listener->tcp_conn_req_seqnum;
	listener->tcp_conn_req_cnt_q0++;
	if (++listener->tcp_conn_req_seqnum == -1) {
		/*
		 * -1 is "special" and defined in TPI as something
		 * that should never be used in T_CONN_IND
		 */
		++listener->tcp_conn_req_seqnum;
	}
	mutex_exit(&listener->tcp_eager_lock);

	if (listener->tcp_syn_defense) {
		/* Don't drop the SYN that comes from a good IP source */
		ipaddr_t *addr_cache;

		addr_cache = (ipaddr_t *)(listener->tcp_ip_addr_cache);
		if (addr_cache != NULL && econnp->conn_faddr_v4 ==
		    addr_cache[IP_ADDR_CACHE_HASH(econnp->conn_faddr_v4)]) {
			eager->tcp_dontdrop = B_TRUE;
		}
	}

	/*
	 * We need to insert the eager in its own perimeter but as soon
	 * as we do that, we expose the eager to the classifier and
	 * should not touch any field outside the eager's perimeter.
	 * So do all the work necessary before inserting the eager
	 * in its own perimeter. Be optimistic that conn_connect()
	 * will succeed but undo everything if it fails.
	 */
	seg_seq = ntohl(tcpha->tha_seq);
	eager->tcp_irs = seg_seq;
	eager->tcp_rack = seg_seq;
	eager->tcp_rnxt = seg_seq + 1;
	eager->tcp_tcpha->tha_ack = htonl(eager->tcp_rnxt);
	BUMP_MIB(&tcps->tcps_mib, tcpPassiveOpens);
	eager->tcp_state = TCPS_SYN_RCVD;
	mp1 = tcp_xmit_mp(eager, eager->tcp_xmit_head, eager->tcp_mss,
	    NULL, NULL, eager->tcp_iss, B_FALSE, NULL, B_FALSE);
	if (mp1 == NULL) {
		/*
		 * Increment the ref count as we are going to
		 * enqueueing an mp in squeue
		 */
		CONN_INC_REF(econnp);
		goto error;
	}

	/*
	 * We need to start the rto timer. In normal case, we start
	 * the timer after sending the packet on the wire (or at
	 * least believing that packet was sent by waiting for
	 * conn_ip_output() to return). Since this is the first packet
	 * being sent on the wire for the eager, our initial tcp_rto
	 * is at least tcp_rexmit_interval_min which is a fairly
	 * large value to allow the algorithm to adjust slowly to large
	 * fluctuations of RTT during first few transmissions.
	 *
	 * Starting the timer first and then sending the packet in this
	 * case shouldn't make much difference since tcp_rexmit_interval_min
	 * is of the order of several 100ms and starting the timer
	 * first and then sending the packet will result in difference
	 * of few micro seconds.
	 *
	 * Without this optimization, we are forced to hold the fanout
	 * lock across the ipcl_bind_insert() and sending the packet
	 * so that we don't race against an incoming packet (maybe RST)
	 * for this eager.
	 *
	 * It is necessary to acquire an extra reference on the eager
	 * at this point and hold it until after tcp_send_data() to
	 * ensure against an eager close race.
	 */

	CONN_INC_REF(econnp);

	TCP_TIMER_RESTART(eager, eager->tcp_rto);

	/*
	 * Insert the eager in its own perimeter now. We are ready to deal
	 * with any packets on eager.
	 */
	if (ipcl_conn_insert(econnp) != 0)
		goto error;

	/*
	 * Send the SYN-ACK. Can't use tcp_send_data since we can't update
	 * pmtu etc; we are not on the eager's squeue
	 */
	ASSERT(econnp->conn_ixa->ixa_notify_cookie == econnp->conn_tcp);
	(void) conn_ip_output(mp1, econnp->conn_ixa);
	CONN_DEC_REF(econnp);
	freemsg(mp);

	return;
error:
	freemsg(mp1);
	eager->tcp_closemp_used = B_TRUE;
	TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
	mp1 = &eager->tcp_closemp;
	SQUEUE_ENTER_ONE(econnp->conn_sqp, mp1, tcp_eager_kill,
	    econnp, NULL, SQ_FILL, SQTAG_TCP_CONN_REQ_2);

	/*
	 * If a connection already exists, send the mp to that connections so
	 * that it can be appropriately dealt with.
	 */
	ipst = tcps->tcps_netstack->netstack_ip;

	if ((econnp = ipcl_classify(mp, ira, ipst)) != NULL) {
		if (!IPCL_IS_CONNECTED(econnp)) {
			/*
			 * Something bad happened. ipcl_conn_insert()
			 * failed because a connection already existed
			 * in connected hash but we can't find it
			 * anymore (someone blew it away). Just
			 * free this message and hopefully remote
			 * will retransmit at which time the SYN can be
			 * treated as a new connection or dealth with
			 * a TH_RST if a connection already exists.
			 */
			CONN_DEC_REF(econnp);
			freemsg(mp);
		} else {
			SQUEUE_ENTER_ONE(econnp->conn_sqp, mp, tcp_input_data,
			    econnp, ira, SQ_FILL, SQTAG_TCP_CONN_REQ_1);
		}
	} else {
		/* Nobody wants this packet */
		freemsg(mp);
	}
	return;
error3:
	CONN_DEC_REF(econnp);
error2:
	freemsg(mp);
}

/*
 * In an ideal case of vertical partition in NUMA architecture, its
 * beneficial to have the listener and all the incoming connections
 * tied to the same squeue. The other constraint is that incoming
 * connections should be tied to the squeue attached to interrupted
 * CPU for obvious locality reason so this leaves the listener to
 * be tied to the same squeue. Our only problem is that when listener
 * is binding, the CPU that will get interrupted by the NIC whose
 * IP address the listener is binding to is not even known. So
 * the code below allows us to change that binding at the time the
 * CPU is interrupted by virtue of incoming connection's squeue.
 *
 * This is usefull only in case of a listener bound to a specific IP
 * address. For other kind of listeners, they get bound the
 * very first time and there is no attempt to rebind them.
 */
void
tcp_input_listener_unbound(void *arg, mblk_t *mp, void *arg2,
    ip_recv_attr_t *ira)
{
	conn_t		*connp = (conn_t *)arg;
	squeue_t	*sqp = (squeue_t *)arg2;
	squeue_t	*new_sqp;
	uint32_t	conn_flags;

	/*
	 * IP sets ira_sqp to either the senders conn_sqp (for loopback)
	 * or based on the ring (for packets from GLD). Otherwise it is
	 * set based on lbolt i.e., a somewhat random number.
	 */
	ASSERT(ira->ira_sqp != NULL);
	new_sqp = ira->ira_sqp;

	if (connp->conn_fanout == NULL)
		goto done;

	if (!(connp->conn_flags & IPCL_FULLY_BOUND)) {
		mutex_enter(&connp->conn_fanout->connf_lock);
		mutex_enter(&connp->conn_lock);
		/*
		 * No one from read or write side can access us now
		 * except for already queued packets on this squeue.
		 * But since we haven't changed the squeue yet, they
		 * can't execute. If they are processed after we have
		 * changed the squeue, they are sent back to the
		 * correct squeue down below.
		 * But a listner close can race with processing of
		 * incoming SYN. If incoming SYN processing changes
		 * the squeue then the listener close which is waiting
		 * to enter the squeue would operate on the wrong
		 * squeue. Hence we don't change the squeue here unless
		 * the refcount is exactly the minimum refcount. The
		 * minimum refcount of 4 is counted as - 1 each for
		 * TCP and IP, 1 for being in the classifier hash, and
		 * 1 for the mblk being processed.
		 */

		if (connp->conn_ref != 4 ||
		    connp->conn_tcp->tcp_state != TCPS_LISTEN) {
			mutex_exit(&connp->conn_lock);
			mutex_exit(&connp->conn_fanout->connf_lock);
			goto done;
		}
		if (connp->conn_sqp != new_sqp) {
			while (connp->conn_sqp != new_sqp)
				(void) casptr(&connp->conn_sqp, sqp, new_sqp);
			/* No special MT issues for outbound ixa_sqp hint */
			connp->conn_ixa->ixa_sqp = new_sqp;
		}

		do {
			conn_flags = connp->conn_flags;
			conn_flags |= IPCL_FULLY_BOUND;
			(void) cas32(&connp->conn_flags, connp->conn_flags,
			    conn_flags);
		} while (!(connp->conn_flags & IPCL_FULLY_BOUND));

		mutex_exit(&connp->conn_fanout->connf_lock);
		mutex_exit(&connp->conn_lock);

		/*
		 * Assume we have picked a good squeue for the listener. Make
		 * subsequent SYNs not try to change the squeue.
		 */
		connp->conn_recv = tcp_input_listener;
	}

done:
	if (connp->conn_sqp != sqp) {
		CONN_INC_REF(connp);
		SQUEUE_ENTER_ONE(connp->conn_sqp, mp, connp->conn_recv, connp,
		    ira, SQ_FILL, SQTAG_TCP_CONN_REQ_UNBOUND);
	} else {
		tcp_input_listener(connp, mp, sqp, ira);
	}
}

/*
 * Successful connect request processing begins when our client passes
 * a T_CONN_REQ message into tcp_wput(), which performs function calls into
 * IP and the passes a T_OK_ACK (or T_ERROR_ACK upstream).
 *
 * After various error checks are completed, tcp_tpi_connect() lays
 * the target address and port into the composite header template.
 * Then we ask IP for information, including a source address if we didn't
 * already have one. Finally we prepare to send the SYN packet, and then
 * send up the T_OK_ACK reply message.
 */
static void
tcp_tpi_connect(tcp_t *tcp, mblk_t *mp)
{
	sin_t		*sin;
	struct T_conn_req	*tcr;
	struct sockaddr	*sa;
	socklen_t	len;
	int		error;
	cred_t		*cr;
	pid_t		cpid;
	conn_t		*connp = tcp->tcp_connp;
	queue_t		*q = connp->conn_wq;

	/*
	 * All Solaris components should pass a db_credp
	 * for this TPI message, hence we ASSERT.
	 * But in case there is some other M_PROTO that looks
	 * like a TPI message sent by some other kernel
	 * component, we check and return an error.
	 */
	cr = msg_getcred(mp, &cpid);
	ASSERT(cr != NULL);
	if (cr == NULL) {
		tcp_err_ack(tcp, mp, TSYSERR, EINVAL);
		return;
	}

	tcr = (struct T_conn_req *)mp->b_rptr;

	ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <= (uintptr_t)INT_MAX);
	if ((mp->b_wptr - mp->b_rptr) < sizeof (*tcr)) {
		tcp_err_ack(tcp, mp, TPROTO, 0);
		return;
	}

	/*
	 * Pre-allocate the T_ordrel_ind mblk so that at close time, we
	 * will always have that to send up.  Otherwise, we need to do
	 * special handling in case the allocation fails at that time.
	 * If the end point is TPI, the tcp_t can be reused and the
	 * tcp_ordrel_mp may be allocated already.
	 */
	if (tcp->tcp_ordrel_mp == NULL) {
		if ((tcp->tcp_ordrel_mp = mi_tpi_ordrel_ind()) == NULL) {
			tcp_err_ack(tcp, mp, TSYSERR, ENOMEM);
			return;
		}
	}

	/*
	 * Determine packet type based on type of address passed in
	 * the request should contain an IPv4 or IPv6 address.
	 * Make sure that address family matches the type of
	 * family of the address passed down.
	 */
	switch (tcr->DEST_length) {
	default:
		tcp_err_ack(tcp, mp, TBADADDR, 0);
		return;

	case (sizeof (sin_t) - sizeof (sin->sin_zero)): {
		/*
		 * XXX: The check for valid DEST_length was not there
		 * in earlier releases and some buggy
		 * TLI apps (e.g Sybase) got away with not feeding
		 * in sin_zero part of address.
		 * We allow that bug to keep those buggy apps humming.
		 * Test suites require the check on DEST_length.
		 * We construct a new mblk with valid DEST_length
		 * free the original so the rest of the code does
		 * not have to keep track of this special shorter
		 * length address case.
		 */
		mblk_t *nmp;
		struct T_conn_req *ntcr;
		sin_t *nsin;

		nmp = allocb(sizeof (struct T_conn_req) + sizeof (sin_t) +
		    tcr->OPT_length, BPRI_HI);
		if (nmp == NULL) {
			tcp_err_ack(tcp, mp, TSYSERR, ENOMEM);
			return;
		}
		ntcr = (struct T_conn_req *)nmp->b_rptr;
		bzero(ntcr, sizeof (struct T_conn_req)); /* zero fill */
		ntcr->PRIM_type = T_CONN_REQ;
		ntcr->DEST_length = sizeof (sin_t);
		ntcr->DEST_offset = sizeof (struct T_conn_req);

		nsin = (sin_t *)((uchar_t *)ntcr + ntcr->DEST_offset);
		*nsin = sin_null;
		/* Get pointer to shorter address to copy from original mp */
		sin = (sin_t *)mi_offset_param(mp, tcr->DEST_offset,
		    tcr->DEST_length); /* extract DEST_length worth of sin_t */
		if (sin == NULL || !OK_32PTR((char *)sin)) {
			freemsg(nmp);
			tcp_err_ack(tcp, mp, TSYSERR, EINVAL);
			return;
		}
		nsin->sin_family = sin->sin_family;
		nsin->sin_port = sin->sin_port;
		nsin->sin_addr = sin->sin_addr;
		/* Note:nsin->sin_zero zero-fill with sin_null assign above */
		nmp->b_wptr = (uchar_t *)&nsin[1];
		if (tcr->OPT_length != 0) {
			ntcr->OPT_length = tcr->OPT_length;
			ntcr->OPT_offset = nmp->b_wptr - nmp->b_rptr;
			bcopy((uchar_t *)tcr + tcr->OPT_offset,
			    (uchar_t *)ntcr + ntcr->OPT_offset,
			    tcr->OPT_length);
			nmp->b_wptr += tcr->OPT_length;
		}
		freemsg(mp);	/* original mp freed */
		mp = nmp;	/* re-initialize original variables */
		tcr = ntcr;
	}
	/* FALLTHRU */

	case sizeof (sin_t):
		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
		    sizeof (sin_t));
		len = sizeof (sin_t);
		break;

	case sizeof (sin6_t):
		sa = (struct sockaddr *)mi_offset_param(mp, tcr->DEST_offset,
		    sizeof (sin6_t));
		len = sizeof (sin6_t);
		break;
	}

	error = proto_verify_ip_addr(connp->conn_family, sa, len);
	if (error != 0) {
		tcp_err_ack(tcp, mp, TSYSERR, error);
		return;
	}

	/*
	 * TODO: If someone in TCPS_TIME_WAIT has this dst/port we
	 * should key on their sequence number and cut them loose.
	 */

	/*
	 * If options passed in, feed it for verification and handling
	 */
	if (tcr->OPT_length != 0) {
		mblk_t	*ok_mp;
		mblk_t	*discon_mp;
		mblk_t  *conn_opts_mp;
		int t_error, sys_error, do_disconnect;

		conn_opts_mp = NULL;

		if (tcp_conprim_opt_process(tcp, mp,
		    &do_disconnect, &t_error, &sys_error) < 0) {
			if (do_disconnect) {
				ASSERT(t_error == 0 && sys_error == 0);
				discon_mp = mi_tpi_discon_ind(NULL,
				    ECONNREFUSED, 0);
				if (!discon_mp) {
					tcp_err_ack_prim(tcp, mp, T_CONN_REQ,
					    TSYSERR, ENOMEM);
					return;
				}
				ok_mp = mi_tpi_ok_ack_alloc(mp);
				if (!ok_mp) {
					tcp_err_ack_prim(tcp, NULL, T_CONN_REQ,
					    TSYSERR, ENOMEM);
					return;
				}
				qreply(q, ok_mp);
				qreply(q, discon_mp); /* no flush! */
			} else {
				ASSERT(t_error != 0);
				tcp_err_ack_prim(tcp, mp, T_CONN_REQ, t_error,
				    sys_error);
			}
			return;
		}
		/*
		 * Success in setting options, the mp option buffer represented
		 * by OPT_length/offset has been potentially modified and
		 * contains results of option processing. We copy it in
		 * another mp to save it for potentially influencing returning
		 * it in T_CONN_CONN.
		 */
		if (tcr->OPT_length != 0) { /* there are resulting options */
			conn_opts_mp = copyb(mp);
			if (!conn_opts_mp) {
				tcp_err_ack_prim(tcp, mp, T_CONN_REQ,
				    TSYSERR, ENOMEM);
				return;
			}
			ASSERT(tcp->tcp_conn.tcp_opts_conn_req == NULL);
			tcp->tcp_conn.tcp_opts_conn_req = conn_opts_mp;
			/*
			 * Note:
			 * These resulting option negotiation can include any
			 * end-to-end negotiation options but there no such
			 * thing (yet?) in our TCP/IP.
			 */
		}
	}

	/* call the non-TPI version */
	error = tcp_do_connect(tcp->tcp_connp, sa, len, cr, cpid);
	if (error < 0) {
		mp = mi_tpi_err_ack_alloc(mp, -error, 0);
	} else if (error > 0) {
		mp = mi_tpi_err_ack_alloc(mp, TSYSERR, error);
	} else {
		mp = mi_tpi_ok_ack_alloc(mp);
	}

	/*
	 * Note: Code below is the "failure" case
	 */
	/* return error ack and blow away saved option results if any */
connect_failed:
	if (mp != NULL)
		putnext(connp->conn_rq, mp);
	else {
		tcp_err_ack_prim(tcp, NULL, T_CONN_REQ,
		    TSYSERR, ENOMEM);
	}
}

/*
 * Handle connect to IPv4 destinations, including connections for AF_INET6
 * sockets connecting to IPv4 mapped IPv6 destinations.
 * Returns zero if OK, a positive errno, or a negative TLI error.
 */
static int
tcp_connect_ipv4(tcp_t *tcp, ipaddr_t *dstaddrp, in_port_t dstport,
    uint_t srcid)
{
	ipaddr_t 	dstaddr = *dstaddrp;
	uint16_t 	lport;
	conn_t		*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	int		error;

	ASSERT(connp->conn_ipversion == IPV4_VERSION);

	/* Check for attempt to connect to INADDR_ANY */
	if (dstaddr == INADDR_ANY)  {
		/*
		 * SunOS 4.x and 4.3 BSD allow an application
		 * to connect a TCP socket to INADDR_ANY.
		 * When they do this, the kernel picks the
		 * address of one interface and uses it
		 * instead.  The kernel usually ends up
		 * picking the address of the loopback
		 * interface.  This is an undocumented feature.
		 * However, we provide the same thing here
		 * in order to have source and binary
		 * compatibility with SunOS 4.x.
		 * Update the T_CONN_REQ (sin/sin6) since it is used to
		 * generate the T_CONN_CON.
		 */
		dstaddr = htonl(INADDR_LOOPBACK);
		*dstaddrp = dstaddr;
	}

	/* Handle __sin6_src_id if socket not bound to an IP address */
	if (srcid != 0 && connp->conn_laddr_v4 == INADDR_ANY) {
		ip_srcid_find_id(srcid, &connp->conn_laddr_v6,
		    IPCL_ZONEID(connp), tcps->tcps_netstack);
		connp->conn_saddr_v6 = connp->conn_laddr_v6;
	}

	IN6_IPADDR_TO_V4MAPPED(dstaddr, &connp->conn_faddr_v6);
	connp->conn_fport = dstport;

	/*
	 * At this point the remote destination address and remote port fields
	 * in the tcp-four-tuple have been filled in the tcp structure. Now we
	 * have to see which state tcp was in so we can take appropriate action.
	 */
	if (tcp->tcp_state == TCPS_IDLE) {
		/*
		 * We support a quick connect capability here, allowing
		 * clients to transition directly from IDLE to SYN_SENT
		 * tcp_bindi will pick an unused port, insert the connection
		 * in the bind hash and transition to BOUND state.
		 */
		lport = tcp_update_next_port(tcps->tcps_next_port_to_try,
		    tcp, B_TRUE);
		lport = tcp_bindi(tcp, lport, &connp->conn_laddr_v6, 0, B_TRUE,
		    B_FALSE, B_FALSE);
		if (lport == 0)
			return (-TNOADDR);
	}

	/*
	 * Lookup the route to determine a source address and the uinfo.
	 * Setup TCP parameters based on the metrics/DCE.
	 */
	error = tcp_set_destination(tcp);
	if (error != 0)
		return (error);

	/*
	 * Don't let an endpoint connect to itself.
	 */
	if (connp->conn_faddr_v4 == connp->conn_laddr_v4 &&
	    connp->conn_fport == connp->conn_lport)
		return (-TBADADDR);

	tcp->tcp_state = TCPS_SYN_SENT;

	return (ipcl_conn_insert_v4(connp));
}

/*
 * Handle connect to IPv6 destinations.
 * Returns zero if OK, a positive errno, or a negative TLI error.
 */
static int
tcp_connect_ipv6(tcp_t *tcp, in6_addr_t *dstaddrp, in_port_t dstport,
    uint32_t flowinfo, uint_t srcid, uint32_t scope_id)
{
	uint16_t 	lport;
	conn_t		*connp = tcp->tcp_connp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	int		error;

	ASSERT(connp->conn_family == AF_INET6);

	/*
	 * If we're here, it means that the destination address is a native
	 * IPv6 address.  Return an error if conn_ipversion is not IPv6.  A
	 * reason why it might not be IPv6 is if the socket was bound to an
	 * IPv4-mapped IPv6 address.
	 */
	if (connp->conn_ipversion != IPV6_VERSION)
		return (-TBADADDR);

	/*
	 * Interpret a zero destination to mean loopback.
	 * Update the T_CONN_REQ (sin/sin6) since it is used to
	 * generate the T_CONN_CON.
	 */
	if (IN6_IS_ADDR_UNSPECIFIED(dstaddrp))
		*dstaddrp = ipv6_loopback;

	/* Handle __sin6_src_id if socket not bound to an IP address */
	if (srcid != 0 && IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) {
		ip_srcid_find_id(srcid, &connp->conn_laddr_v6,
		    IPCL_ZONEID(connp), tcps->tcps_netstack);
		connp->conn_saddr_v6 = connp->conn_laddr_v6;
	}

	/*
	 * Take care of the scope_id now.
	 */
	if (scope_id != 0 && IN6_IS_ADDR_LINKSCOPE(dstaddrp)) {
		connp->conn_ixa->ixa_flags |= IXAF_SCOPEID_SET;
		connp->conn_ixa->ixa_scopeid = scope_id;
	} else {
		connp->conn_ixa->ixa_flags &= ~IXAF_SCOPEID_SET;
	}

	connp->conn_flowinfo = flowinfo;
	connp->conn_faddr_v6 = *dstaddrp;
	connp->conn_fport = dstport;

	/*
	 * At this point the remote destination address and remote port fields
	 * in the tcp-four-tuple have been filled in the tcp structure. Now we
	 * have to see which state tcp was in so we can take appropriate action.
	 */
	if (tcp->tcp_state == TCPS_IDLE) {
		/*
		 * We support a quick connect capability here, allowing
		 * clients to transition directly from IDLE to SYN_SENT
		 * tcp_bindi will pick an unused port, insert the connection
		 * in the bind hash and transition to BOUND state.
		 */
		lport = tcp_update_next_port(tcps->tcps_next_port_to_try,
		    tcp, B_TRUE);
		lport = tcp_bindi(tcp, lport, &connp->conn_laddr_v6, 0, B_TRUE,
		    B_FALSE, B_FALSE);
		if (lport == 0)
			return (-TNOADDR);
	}

	/*
	 * Lookup the route to determine a source address and the uinfo.
	 * Setup TCP parameters based on the metrics/DCE.
	 */
	error = tcp_set_destination(tcp);
	if (error != 0)
		return (error);

	/*
	 * Don't let an endpoint connect to itself.
	 */
	if (IN6_ARE_ADDR_EQUAL(&connp->conn_faddr_v6, &connp->conn_laddr_v6) &&
	    connp->conn_fport == connp->conn_lport)
		return (-TBADADDR);

	tcp->tcp_state = TCPS_SYN_SENT;

	return (ipcl_conn_insert_v6(connp));
}

/*
 * Disconnect
 * Note that unlike other functions this returns a positive tli error
 * when it fails; it never returns an errno.
 */
static int
tcp_disconnect_common(tcp_t *tcp, t_scalar_t seqnum)
{
	conn_t		*lconnp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

	/*
	 * Right now, upper modules pass down a T_DISCON_REQ to TCP,
	 * when the stream is in BOUND state. Do not send a reset,
	 * since the destination IP address is not valid, and it can
	 * be the initialized value of all zeros (broadcast address).
	 */
	if (tcp->tcp_state <= TCPS_BOUND) {
		if (connp->conn_debug) {
			(void) strlog(TCP_MOD_ID, 0, 1, SL_ERROR|SL_TRACE,
			    "tcp_disconnect: bad state, %d", tcp->tcp_state);
		}
		return (TOUTSTATE);
	}


	if (seqnum == -1 || tcp->tcp_conn_req_max == 0) {

		/*
		 * According to TPI, for non-listeners, ignore seqnum
		 * and disconnect.
		 * Following interpretation of -1 seqnum is historical
		 * and implied TPI ? (TPI only states that for T_CONN_IND,
		 * a valid seqnum should not be -1).
		 *
		 *	-1 means disconnect everything
		 *	regardless even on a listener.
		 */

		int old_state = tcp->tcp_state;
		ip_stack_t *ipst = tcps->tcps_netstack->netstack_ip;

		/*
		 * The connection can't be on the tcp_time_wait_head list
		 * since it is not detached.
		 */
		ASSERT(tcp->tcp_time_wait_next == NULL);
		ASSERT(tcp->tcp_time_wait_prev == NULL);
		ASSERT(tcp->tcp_time_wait_expire == 0);
		/*
		 * If it used to be a listener, check to make sure no one else
		 * has taken the port before switching back to LISTEN state.
		 */
		if (connp->conn_ipversion == IPV4_VERSION) {
			lconnp = ipcl_lookup_listener_v4(connp->conn_lport,
			    connp->conn_laddr_v4, IPCL_ZONEID(connp), ipst);
		} else {
			uint_t ifindex = 0;

			if (connp->conn_ixa->ixa_flags & IXAF_SCOPEID_SET)
				ifindex = connp->conn_ixa->ixa_scopeid;

			/* Allow conn_bound_if listeners? */
			lconnp = ipcl_lookup_listener_v6(connp->conn_lport,
			    &connp->conn_laddr_v6, ifindex, IPCL_ZONEID(connp),
			    ipst);
		}
		if (tcp->tcp_conn_req_max && lconnp == NULL) {
			tcp->tcp_state = TCPS_LISTEN;
		} else if (old_state > TCPS_BOUND) {
			tcp->tcp_conn_req_max = 0;
			tcp->tcp_state = TCPS_BOUND;
		}
		if (lconnp != NULL)
			CONN_DEC_REF(lconnp);
		if (old_state == TCPS_SYN_SENT || old_state == TCPS_SYN_RCVD) {
			BUMP_MIB(&tcps->tcps_mib, tcpAttemptFails);
		} else if (old_state == TCPS_ESTABLISHED ||
		    old_state == TCPS_CLOSE_WAIT) {
			BUMP_MIB(&tcps->tcps_mib, tcpEstabResets);
		}

		if (tcp->tcp_fused)
			tcp_unfuse(tcp);

		mutex_enter(&tcp->tcp_eager_lock);
		if ((tcp->tcp_conn_req_cnt_q0 != 0) ||
		    (tcp->tcp_conn_req_cnt_q != 0)) {
			tcp_eager_cleanup(tcp, 0);
		}
		mutex_exit(&tcp->tcp_eager_lock);

		tcp_xmit_ctl("tcp_disconnect", tcp, tcp->tcp_snxt,
		    tcp->tcp_rnxt, TH_RST | TH_ACK);

		tcp_reinit(tcp);

		return (0);
	} else if (!tcp_eager_blowoff(tcp, seqnum)) {
		return (TBADSEQ);
	}
	return (0);
}

/*
 * Our client hereby directs us to reject the connection request
 * that tcp_input_listener() marked with 'seqnum'.  Rejection consists
 * of sending the appropriate RST, not an ICMP error.
 */
static void
tcp_disconnect(tcp_t *tcp, mblk_t *mp)
{
	t_scalar_t seqnum;
	int	error;
	conn_t	*connp = tcp->tcp_connp;

	ASSERT((uintptr_t)(mp->b_wptr - mp->b_rptr) <= (uintptr_t)INT_MAX);
	if ((mp->b_wptr - mp->b_rptr) < sizeof (struct T_discon_req)) {
		tcp_err_ack(tcp, mp, TPROTO, 0);
		return;
	}
	seqnum = ((struct T_discon_req *)mp->b_rptr)->SEQ_number;
	error = tcp_disconnect_common(tcp, seqnum);
	if (error != 0)
		tcp_err_ack(tcp, mp, error, 0);
	else {
		if (tcp->tcp_state >= TCPS_ESTABLISHED) {
			/* Send M_FLUSH according to TPI */
			(void) putnextctl1(connp->conn_rq, M_FLUSH, FLUSHRW);
		}
		mp = mi_tpi_ok_ack_alloc(mp);
		if (mp != NULL)
			putnext(connp->conn_rq, mp);
	}
}

/*
 * Diagnostic routine used to return a string associated with the tcp state.
 * Note that if the caller does not supply a buffer, it will use an internal
 * static string.  This means that if multiple threads call this function at
 * the same time, output can be corrupted...  Note also that this function
 * does not check the size of the supplied buffer.  The caller has to make
 * sure that it is big enough.
 */
static char *
tcp_display(tcp_t *tcp, char *sup_buf, char format)
{
	char		buf1[30];
	static char	priv_buf[INET6_ADDRSTRLEN * 2 + 80];
	char		*buf;
	char		*cp;
	in6_addr_t	local, remote;
	char		local_addrbuf[INET6_ADDRSTRLEN];
	char		remote_addrbuf[INET6_ADDRSTRLEN];
	conn_t		*connp;

	if (sup_buf != NULL)
		buf = sup_buf;
	else
		buf = priv_buf;

	if (tcp == NULL)
		return ("NULL_TCP");

	connp = tcp->tcp_connp;
	switch (tcp->tcp_state) {
	case TCPS_CLOSED:
		cp = "TCP_CLOSED";
		break;
	case TCPS_IDLE:
		cp = "TCP_IDLE";
		break;
	case TCPS_BOUND:
		cp = "TCP_BOUND";
		break;
	case TCPS_LISTEN:
		cp = "TCP_LISTEN";
		break;
	case TCPS_SYN_SENT:
		cp = "TCP_SYN_SENT";
		break;
	case TCPS_SYN_RCVD:
		cp = "TCP_SYN_RCVD";
		break;
	case TCPS_ESTABLISHED:
		cp = "TCP_ESTABLISHED";
		break;
	case TCPS_CLOSE_WAIT:
		cp = "TCP_CLOSE_WAIT";
		break;
	case TCPS_FIN_WAIT_1:
		cp = "TCP_FIN_WAIT_1";
		break;
	case TCPS_CLOSING:
		cp = "TCP_CLOSING";
		break;
	case TCPS_LAST_ACK:
		cp = "TCP_LAST_ACK";
		break;
	case TCPS_FIN_WAIT_2:
		cp = "TCP_FIN_WAIT_2";
		break;
	case TCPS_TIME_WAIT:
		cp = "TCP_TIME_WAIT";
		break;
	default:
		(void) mi_sprintf(buf1, "TCPUnkState(%d)", tcp->tcp_state);
		cp = buf1;
		break;
	}
	switch (format) {
	case DISP_ADDR_AND_PORT:
		if (connp->conn_ipversion == IPV4_VERSION) {
			/*
			 * Note that we use the remote address in the tcp_b
			 * structure.  This means that it will print out
			 * the real destination address, not the next hop's
			 * address if source routing is used.
			 */
			IN6_IPADDR_TO_V4MAPPED(connp->conn_laddr_v4, &local);
			IN6_IPADDR_TO_V4MAPPED(connp->conn_faddr_v4, &remote);

		} else {
			local = connp->conn_laddr_v6;
			remote = connp->conn_faddr_v6;
		}
		(void) inet_ntop(AF_INET6, &local, local_addrbuf,
		    sizeof (local_addrbuf));
		(void) inet_ntop(AF_INET6, &remote, remote_addrbuf,
		    sizeof (remote_addrbuf));
		(void) mi_sprintf(buf, "[%s.%u, %s.%u] %s",
		    local_addrbuf, ntohs(connp->conn_lport), remote_addrbuf,
		    ntohs(connp->conn_fport), cp);
		break;
	case DISP_PORT_ONLY:
	default:
		(void) mi_sprintf(buf, "[%u, %u] %s",
		    ntohs(connp->conn_lport), ntohs(connp->conn_fport), cp);
		break;
	}

	return (buf);
}

/*
 * Called via squeue to get on to eager's perimeter. It sends a
 * TH_RST if eager is in the fanout table. The listener wants the
 * eager to disappear either by means of tcp_eager_blowoff() or
 * tcp_eager_cleanup() being called. tcp_eager_kill() can also be
 * called (via squeue) if the eager cannot be inserted in the
 * fanout table in tcp_input_listener().
 */
/* ARGSUSED */
void
tcp_eager_kill(void *arg, mblk_t *mp, void *arg2, ip_recv_attr_t *dummy)
{
	conn_t	*econnp = (conn_t *)arg;
	tcp_t	*eager = econnp->conn_tcp;
	tcp_t	*listener = eager->tcp_listener;

	/*
	 * We could be called because listener is closing. Since
	 * the eager was using listener's queue's, we avoid
	 * using the listeners queues from now on.
	 */
	ASSERT(eager->tcp_detached);
	econnp->conn_rq = NULL;
	econnp->conn_wq = NULL;

	/*
	 * An eager's conn_fanout will be NULL if it's a duplicate
	 * for an existing 4-tuples in the conn fanout table.
	 * We don't want to send an RST out in such case.
	 */
	if (econnp->conn_fanout != NULL && eager->tcp_state > TCPS_LISTEN) {
		tcp_xmit_ctl("tcp_eager_kill, can't wait",
		    eager, eager->tcp_snxt, 0, TH_RST);
	}

	/* We are here because listener wants this eager gone */
	if (listener != NULL) {
		mutex_enter(&listener->tcp_eager_lock);
		tcp_eager_unlink(eager);
		if (eager->tcp_tconnind_started) {
			/*
			 * The eager has sent a conn_ind up to the
			 * listener but listener decides to close
			 * instead. We need to drop the extra ref
			 * placed on eager in tcp_input_data() before
			 * sending the conn_ind to listener.
			 */
			CONN_DEC_REF(econnp);
		}
		mutex_exit(&listener->tcp_eager_lock);
		CONN_DEC_REF(listener->tcp_connp);
	}

	if (eager->tcp_state != TCPS_CLOSED)
		tcp_close_detached(eager);
}

/*
 * Reset any eager connection hanging off this listener marked
 * with 'seqnum' and then reclaim it's resources.
 */
static boolean_t
tcp_eager_blowoff(tcp_t	*listener, t_scalar_t seqnum)
{
	tcp_t	*eager;
	mblk_t 	*mp;
	tcp_stack_t	*tcps = listener->tcp_tcps;

	TCP_STAT(tcps, tcp_eager_blowoff_calls);
	eager = listener;
	mutex_enter(&listener->tcp_eager_lock);
	do {
		eager = eager->tcp_eager_next_q;
		if (eager == NULL) {
			mutex_exit(&listener->tcp_eager_lock);
			return (B_FALSE);
		}
	} while (eager->tcp_conn_req_seqnum != seqnum);

	if (eager->tcp_closemp_used) {
		mutex_exit(&listener->tcp_eager_lock);
		return (B_TRUE);
	}
	eager->tcp_closemp_used = B_TRUE;
	TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
	CONN_INC_REF(eager->tcp_connp);
	mutex_exit(&listener->tcp_eager_lock);
	mp = &eager->tcp_closemp;
	SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp, tcp_eager_kill,
	    eager->tcp_connp, NULL, SQ_FILL, SQTAG_TCP_EAGER_BLOWOFF);
	return (B_TRUE);
}

/*
 * Reset any eager connection hanging off this listener
 * and then reclaim it's resources.
 */
static void
tcp_eager_cleanup(tcp_t *listener, boolean_t q0_only)
{
	tcp_t	*eager;
	mblk_t	*mp;
	tcp_stack_t	*tcps = listener->tcp_tcps;

	ASSERT(MUTEX_HELD(&listener->tcp_eager_lock));

	if (!q0_only) {
		/* First cleanup q */
		TCP_STAT(tcps, tcp_eager_blowoff_q);
		eager = listener->tcp_eager_next_q;
		while (eager != NULL) {
			if (!eager->tcp_closemp_used) {
				eager->tcp_closemp_used = B_TRUE;
				TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
				CONN_INC_REF(eager->tcp_connp);
				mp = &eager->tcp_closemp;
				SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
				    tcp_eager_kill, eager->tcp_connp, NULL,
				    SQ_FILL, SQTAG_TCP_EAGER_CLEANUP);
			}
			eager = eager->tcp_eager_next_q;
		}
	}
	/* Then cleanup q0 */
	TCP_STAT(tcps, tcp_eager_blowoff_q0);
	eager = listener->tcp_eager_next_q0;
	while (eager != listener) {
		if (!eager->tcp_closemp_used) {
			eager->tcp_closemp_used = B_TRUE;
			TCP_DEBUG_GETPCSTACK(eager->tcmp_stk, 15);
			CONN_INC_REF(eager->tcp_connp);
			mp = &eager->tcp_closemp;
			SQUEUE_ENTER_ONE(eager->tcp_connp->conn_sqp, mp,
			    tcp_eager_kill, eager->tcp_connp, NULL, SQ_FILL,
			    SQTAG_TCP_EAGER_CLEANUP_Q0);
		}
		eager = eager->tcp_eager_next_q0;
	}
}

/*
 * If we are an eager connection hanging off a listener that hasn't
 * formally accepted the connection yet, get off his list and blow off
 * any data that we have accumulated.
 */
static void
tcp_eager_unlink(tcp_t *tcp)
{
	tcp_t	*listener = tcp->tcp_listener;

	ASSERT(MUTEX_HELD(&listener->tcp_eager_lock));
	ASSERT(listener != NULL);
	if (tcp->tcp_eager_next_q0 != NULL) {
		ASSERT(tcp->tcp_eager_prev_q0 != NULL);

		/* Remove the eager tcp from q0 */
		tcp->tcp_eager_next_q0->tcp_eager_prev_q0 =
		    tcp->tcp_eager_prev_q0;
		tcp->tcp_eager_prev_q0->tcp_eager_next_q0 =
		    tcp->tcp_eager_next_q0;
		ASSERT(listener->tcp_conn_req_cnt_q0 > 0);
		listener->tcp_conn_req_cnt_q0--;

		tcp->tcp_eager_next_q0 = NULL;
		tcp->tcp_eager_prev_q0 = NULL;

		/*
		 * Take the eager out, if it is in the list of droppable
		 * eagers.
		 */
		MAKE_UNDROPPABLE(tcp);

		if (tcp->tcp_syn_rcvd_timeout != 0) {
			/* we have timed out before */
			ASSERT(listener->tcp_syn_rcvd_timeout > 0);
			listener->tcp_syn_rcvd_timeout--;
		}
	} else {
		tcp_t   **tcpp = &listener->tcp_eager_next_q;
		tcp_t	*prev = NULL;

		for (; tcpp[0]; tcpp = &tcpp[0]->tcp_eager_next_q) {
			if (tcpp[0] == tcp) {
				if (listener->tcp_eager_last_q == tcp) {
					/*
					 * If we are unlinking the last
					 * element on the list, adjust
					 * tail pointer. Set tail pointer
					 * to nil when list is empty.
					 */
					ASSERT(tcp->tcp_eager_next_q == NULL);
					if (listener->tcp_eager_last_q ==
					    listener->tcp_eager_next_q) {
						listener->tcp_eager_last_q =
						    NULL;
					} else {
						/*
						 * We won't get here if there
						 * is only one eager in the
						 * list.
						 */
						ASSERT(prev != NULL);
						listener->tcp_eager_last_q =
						    prev;
					}
				}
				tcpp[0] = tcp->tcp_eager_next_q;
				tcp->tcp_eager_next_q = NULL;
				tcp->tcp_eager_last_q = NULL;
				ASSERT(listener->tcp_conn_req_cnt_q > 0);
				listener->tcp_conn_req_cnt_q--;
				break;
			}
			prev = tcpp[0];
		}
	}
	tcp->tcp_listener = NULL;
}

/* Shorthand to generate and send TPI error acks to our client */
static void
tcp_err_ack(tcp_t *tcp, mblk_t *mp, int t_error, int sys_error)
{
	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
		putnext(tcp->tcp_connp->conn_rq, mp);
}

/* Shorthand to generate and send TPI error acks to our client */
static void
tcp_err_ack_prim(tcp_t *tcp, mblk_t *mp, int primitive,
    int t_error, int sys_error)
{
	struct T_error_ack	*teackp;

	if ((mp = tpi_ack_alloc(mp, sizeof (struct T_error_ack),
	    M_PCPROTO, T_ERROR_ACK)) != NULL) {
		teackp = (struct T_error_ack *)mp->b_rptr;
		teackp->ERROR_prim = primitive;
		teackp->TLI_error = t_error;
		teackp->UNIX_error = sys_error;
		putnext(tcp->tcp_connp->conn_rq, mp);
	}
}

/*
 * Note: No locks are held when inspecting tcp_g_*epriv_ports
 * but instead the code relies on:
 * - the fact that the address of the array and its size never changes
 * - the atomic assignment of the elements of the array
 */
/* ARGSUSED */
static int
tcp_extra_priv_ports_get(queue_t *q, mblk_t *mp, caddr_t cp, cred_t *cr)
{
	int i;
	tcp_stack_t	*tcps = Q_TO_TCP(q)->tcp_tcps;

	for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
		if (tcps->tcps_g_epriv_ports[i] != 0)
			(void) mi_mpprintf(mp, "%d ",
			    tcps->tcps_g_epriv_ports[i]);
	}
	return (0);
}

/*
 * Hold a lock while changing tcp_g_epriv_ports to prevent multiple
 * threads from changing it at the same time.
 */
/* ARGSUSED */
static int
tcp_extra_priv_ports_add(queue_t *q, mblk_t *mp, char *value, caddr_t cp,
    cred_t *cr)
{
	long	new_value;
	int	i;
	tcp_stack_t	*tcps = Q_TO_TCP(q)->tcp_tcps;

	/*
	 * Fail the request if the new value does not lie within the
	 * port number limits.
	 */
	if (ddi_strtol(value, NULL, 10, &new_value) != 0 ||
	    new_value <= 0 || new_value >= 65536) {
		return (EINVAL);
	}

	mutex_enter(&tcps->tcps_epriv_port_lock);
	/* Check if the value is already in the list */
	for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
		if (new_value == tcps->tcps_g_epriv_ports[i]) {
			mutex_exit(&tcps->tcps_epriv_port_lock);
			return (EEXIST);
		}
	}
	/* Find an empty slot */
	for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
		if (tcps->tcps_g_epriv_ports[i] == 0)
			break;
	}
	if (i == tcps->tcps_g_num_epriv_ports) {
		mutex_exit(&tcps->tcps_epriv_port_lock);
		return (EOVERFLOW);
	}
	/* Set the new value */
	tcps->tcps_g_epriv_ports[i] = (uint16_t)new_value;
	mutex_exit(&tcps->tcps_epriv_port_lock);
	return (0);
}

/*
 * Hold a lock while changing tcp_g_epriv_ports to prevent multiple
 * threads from changing it at the same time.
 */
/* ARGSUSED */
static int
tcp_extra_priv_ports_del(queue_t *q, mblk_t *mp, char *value, caddr_t cp,
    cred_t *cr)
{
	long	new_value;
	int	i;
	tcp_stack_t	*tcps = Q_TO_TCP(q)->tcp_tcps;

	/*
	 * Fail the request if the new value does not lie within the
	 * port number limits.
	 */
	if (ddi_strtol(value, NULL, 10, &new_value) != 0 || new_value <= 0 ||
	    new_value >= 65536) {
		return (EINVAL);
	}

	mutex_enter(&tcps->tcps_epriv_port_lock);
	/* Check that the value is already in the list */
	for (i = 0; i < tcps->tcps_g_num_epriv_ports; i++) {
		if (tcps->tcps_g_epriv_ports[i] == new_value)
			break;
	}
	if (i == tcps->tcps_g_num_epriv_ports) {
		mutex_exit(&tcps->tcps_epriv_port_lock);
		return (ESRCH);
	}
	/* Clear the value */
	tcps->tcps_g_epriv_ports[i] = 0;
	mutex_exit(&tcps->tcps_epriv_port_lock);
	return (0);
}

/* Return the TPI/TLI equivalent of our current tcp_state */
static int
tcp_tpistate(tcp_t *tcp)
{
	switch (tcp->tcp_state) {
	case TCPS_IDLE:
		return (TS_UNBND);
	case TCPS_LISTEN:
		/*
		 * Return whether there are outstanding T_CONN_IND waiting
		 * for the matching T_CONN_RES. Therefore don't count q0.
		 */
		if (tcp->tcp_conn_req_cnt_q > 0)
			return (TS_WRES_CIND);
		else
			return (TS_IDLE);
	case TCPS_BOUND:
		return (TS_IDLE);
	case TCPS_SYN_SENT:
		return (TS_WCON_CREQ);
	case TCPS_SYN_RCVD:
		/*
		 * Note: assumption: this has to the active open SYN_RCVD.
		 * The passive instance is detached in SYN_RCVD stage of
		 * incoming connection processing so we cannot get request
		 * for T_info_ack on it.
		 */
		return (TS_WACK_CRES);
	case TCPS_ESTABLISHED:
		return (TS_DATA_XFER);
	case TCPS_CLOSE_WAIT:
		return (TS_WREQ_ORDREL);
	case TCPS_FIN_WAIT_1:
		return (TS_WIND_ORDREL);
	case TCPS_FIN_WAIT_2:
		return (TS_WIND_ORDREL);

	case TCPS_CLOSING:
	case TCPS_LAST_ACK:
	case TCPS_TIME_WAIT:
	case TCPS_CLOSED:
		/*
		 * Following TS_WACK_DREQ7 is a rendition of "not
		 * yet TS_IDLE" TPI state. There is no best match to any
		 * TPI state for TCPS_{CLOSING, LAST_ACK, TIME_WAIT} but we
		 * choose a value chosen that will map to TLI/XTI level
		 * state of TSTATECHNG (state is process of changing) which
		 * captures what this dummy state represents.
		 */
		return (TS_WACK_DREQ7);
	default:
		cmn_err(CE_WARN, "tcp_tpistate: strange state (%d) %s",
		    tcp->tcp_state, tcp_display(tcp, NULL,
		    DISP_PORT_ONLY));
		return (TS_UNBND);
	}
}

static void
tcp_copy_info(struct T_info_ack *tia, tcp_t *tcp)
{
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

	if (connp->conn_family == AF_INET6)
		*tia = tcp_g_t_info_ack_v6;
	else
		*tia = tcp_g_t_info_ack;
	tia->CURRENT_state = tcp_tpistate(tcp);
	tia->OPT_size = tcp_max_optsize;
	if (tcp->tcp_mss == 0) {
		/* Not yet set - tcp_open does not set mss */
		if (connp->conn_ipversion == IPV4_VERSION)
			tia->TIDU_size = tcps->tcps_mss_def_ipv4;
		else
			tia->TIDU_size = tcps->tcps_mss_def_ipv6;
	} else {
		tia->TIDU_size = tcp->tcp_mss;
	}
	/* TODO: Default ETSDU is 1.  Is that correct for tcp? */
}

static void
tcp_do_capability_ack(tcp_t *tcp, struct T_capability_ack *tcap,
    t_uscalar_t cap_bits1)
{
	tcap->CAP_bits1 = 0;

	if (cap_bits1 & TC1_INFO) {
		tcp_copy_info(&tcap->INFO_ack, tcp);
		tcap->CAP_bits1 |= TC1_INFO;
	}

	if (cap_bits1 & TC1_ACCEPTOR_ID) {
		tcap->ACCEPTOR_id = tcp->tcp_acceptor_id;
		tcap->CAP_bits1 |= TC1_ACCEPTOR_ID;
	}

}

/*
 * This routine responds to T_CAPABILITY_REQ messages.  It is called by
 * tcp_wput.  Much of the T_CAPABILITY_ACK information is copied from
 * tcp_g_t_info_ack.  The current state of the stream is copied from
 * tcp_state.
 */
static void
tcp_capability_req(tcp_t *tcp, mblk_t *mp)
{
	t_uscalar_t		cap_bits1;
	struct T_capability_ack	*tcap;

	if (MBLKL(mp) < sizeof (struct T_capability_req)) {
		freemsg(mp);
		return;
	}

	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;

	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
	    mp->b_datap->db_type, T_CAPABILITY_ACK);
	if (mp == NULL)
		return;

	tcap = (struct T_capability_ack *)mp->b_rptr;
	tcp_do_capability_ack(tcp, tcap, cap_bits1);

	putnext(tcp->tcp_connp->conn_rq, mp);
}

/*
 * This routine responds to T_INFO_REQ messages.  It is called by tcp_wput.
 * Most of the T_INFO_ACK information is copied from tcp_g_t_info_ack.
 * The current state of the stream is copied from tcp_state.
 */
static void
tcp_info_req(tcp_t *tcp, mblk_t *mp)
{
	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
	    T_INFO_ACK);
	if (!mp) {
		tcp_err_ack(tcp, mp, TSYSERR, ENOMEM);
		return;
	}
	tcp_copy_info((struct T_info_ack *)mp->b_rptr, tcp);
	putnext(tcp->tcp_connp->conn_rq, mp);
}

/* Respond to the TPI addr request */
static void
tcp_addr_req(tcp_t *tcp, mblk_t *mp)
{
	struct sockaddr *sa;
	mblk_t	*ackmp;
	struct T_addr_ack *taa;
	conn_t	*connp = tcp->tcp_connp;
	uint_t	addrlen;

	/* Make it large enough for worst case */
	ackmp = reallocb(mp, sizeof (struct T_addr_ack) +
	    2 * sizeof (sin6_t), 1);
	if (ackmp == NULL) {
		tcp_err_ack(tcp, mp, TSYSERR, ENOMEM);
		return;
	}

	taa = (struct T_addr_ack *)ackmp->b_rptr;

	bzero(taa, sizeof (struct T_addr_ack));
	ackmp->b_wptr = (uchar_t *)&taa[1];

	taa->PRIM_type = T_ADDR_ACK;
	ackmp->b_datap->db_type = M_PCPROTO;

	if (connp->conn_family == AF_INET)
		addrlen = sizeof (sin_t);
	else
		addrlen = sizeof (sin6_t);

	/*
	 * Note: Following code assumes 32 bit alignment of basic
	 * data structures like sin_t and struct T_addr_ack.
	 */
	if (tcp->tcp_state >= TCPS_BOUND) {
		/*
		 * Fill in local address first
		 */
		taa->LOCADDR_offset = sizeof (*taa);
		taa->LOCADDR_length = addrlen;
		sa = (struct sockaddr *)&taa[1];
		(void) conn_getsockname(connp, sa, &addrlen);
		ackmp->b_wptr += addrlen;
	}
	if (tcp->tcp_state >= TCPS_SYN_RCVD) {
		/*
		 * Fill in Remote address
		 */
		taa->REMADDR_length = addrlen;
		/* assumed 32-bit alignment */
		taa->REMADDR_offset = taa->LOCADDR_offset + taa->LOCADDR_length;
		sa = (struct sockaddr *)(ackmp->b_rptr + taa->REMADDR_offset);
		(void) conn_getpeername(connp, sa, &addrlen);
		ackmp->b_wptr += addrlen;
	}
	ASSERT(ackmp->b_wptr <= ackmp->b_datap->db_lim);
	putnext(tcp->tcp_connp->conn_rq, ackmp);
}

/*
 * Handle reinitialization of a tcp structure.
 * Maintain "binding state" resetting the state to BOUND, LISTEN, or IDLE.
 */
static void
tcp_reinit(tcp_t *tcp)
{
	mblk_t		*mp;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp  = tcp->tcp_connp;

	TCP_STAT(tcps, tcp_reinit_calls);

	/* tcp_reinit should never be called for detached tcp_t's */
	ASSERT(tcp->tcp_listener == NULL);
	ASSERT((connp->conn_family == AF_INET &&
	    connp->conn_ipversion == IPV4_VERSION) ||
	    (connp->conn_family == AF_INET6 &&
	    (connp->conn_ipversion == IPV4_VERSION ||
	    connp->conn_ipversion == IPV6_VERSION)));

	/* Cancel outstanding timers */
	tcp_timers_stop(tcp);

	/*
	 * Reset everything in the state vector, after updating global
	 * MIB data from instance counters.
	 */
	UPDATE_MIB(&tcps->tcps_mib, tcpHCInSegs, tcp->tcp_ibsegs);
	tcp->tcp_ibsegs = 0;
	UPDATE_MIB(&tcps->tcps_mib, tcpHCOutSegs, tcp->tcp_obsegs);
	tcp->tcp_obsegs = 0;

	tcp_close_mpp(&tcp->tcp_xmit_head);
	if (tcp->tcp_snd_zcopy_aware)
		tcp_zcopy_notify(tcp);
	tcp->tcp_xmit_last = tcp->tcp_xmit_tail = NULL;
	tcp->tcp_unsent = tcp->tcp_xmit_tail_unsent = 0;
	mutex_enter(&tcp->tcp_non_sq_lock);
	if (tcp->tcp_flow_stopped &&
	    TCP_UNSENT_BYTES(tcp) <= connp->conn_sndlowat) {
		tcp_clrqfull(tcp);
	}
	mutex_exit(&tcp->tcp_non_sq_lock);
	tcp_close_mpp(&tcp->tcp_reass_head);
	tcp->tcp_reass_tail = NULL;
	if (tcp->tcp_rcv_list != NULL) {
		/* Free b_next chain */
		tcp_close_mpp(&tcp->tcp_rcv_list);
		tcp->tcp_rcv_last_head = NULL;
		tcp->tcp_rcv_last_tail = NULL;
		tcp->tcp_rcv_cnt = 0;
	}
	tcp->tcp_rcv_last_tail = NULL;

	if ((mp = tcp->tcp_urp_mp) != NULL) {
		freemsg(mp);
		tcp->tcp_urp_mp = NULL;
	}
	if ((mp = tcp->tcp_urp_mark_mp) != NULL) {
		freemsg(mp);
		tcp->tcp_urp_mark_mp = NULL;
	}
	if (tcp->tcp_fused_sigurg_mp != NULL) {
		ASSERT(!IPCL_IS_NONSTR(tcp->tcp_connp));
		freeb(tcp->tcp_fused_sigurg_mp);
		tcp->tcp_fused_sigurg_mp = NULL;
	}
	if (tcp->tcp_ordrel_mp != NULL) {
		ASSERT(!IPCL_IS_NONSTR(tcp->tcp_connp));
		freeb(tcp->tcp_ordrel_mp);
		tcp->tcp_ordrel_mp = NULL;
	}

	/*
	 * Following is a union with two members which are
	 * identical types and size so the following cleanup
	 * is enough.
	 */
	tcp_close_mpp(&tcp->tcp_conn.tcp_eager_conn_ind);

	CL_INET_DISCONNECT(connp);

	/*
	 * The connection can't be on the tcp_time_wait_head list
	 * since it is not detached.
	 */
	ASSERT(tcp->tcp_time_wait_next == NULL);
	ASSERT(tcp->tcp_time_wait_prev == NULL);
	ASSERT(tcp->tcp_time_wait_expire == 0);

	if (tcp->tcp_kssl_pending) {
		tcp->tcp_kssl_pending = B_FALSE;

		/* Don't reset if the initialized by bind. */
		if (tcp->tcp_kssl_ent != NULL) {
			kssl_release_ent(tcp->tcp_kssl_ent, NULL,
			    KSSL_NO_PROXY);
		}
	}
	if (tcp->tcp_kssl_ctx != NULL) {
		kssl_release_ctx(tcp->tcp_kssl_ctx);
		tcp->tcp_kssl_ctx = NULL;
	}

	/*
	 * Reset/preserve other values
	 */
	tcp_reinit_values(tcp);
	ipcl_hash_remove(connp);
	ixa_cleanup(connp->conn_ixa);
	tcp_ipsec_cleanup(tcp);

	connp->conn_laddr_v6 = connp->conn_bound_addr_v6;
	connp->conn_saddr_v6 = connp->conn_bound_addr_v6;

	if (tcp->tcp_conn_req_max != 0) {
		/*
		 * This is the case when a TLI program uses the same
		 * transport end point to accept a connection.  This
		 * makes the TCP both a listener and acceptor.  When
		 * this connection is closed, we need to set the state
		 * back to TCPS_LISTEN.  Make sure that the eager list
		 * is reinitialized.
		 *
		 * Note that this stream is still bound to the four
		 * tuples of the previous connection in IP.  If a new
		 * SYN with different foreign address comes in, IP will
		 * not find it and will send it to the global queue.  In
		 * the global queue, TCP will do a tcp_lookup_listener()
		 * to find this stream.  This works because this stream
		 * is only removed from connected hash.
		 *
		 */
		tcp->tcp_state = TCPS_LISTEN;
		tcp->tcp_eager_next_q0 = tcp->tcp_eager_prev_q0 = tcp;
		tcp->tcp_eager_next_drop_q0 = tcp;
		tcp->tcp_eager_prev_drop_q0 = tcp;
		/*
		 * Initially set conn_recv to tcp_input_listener_unbound to try
		 * to pick a good squeue for the listener when the first SYN
		 * arrives. tcp_input_listener_unbound sets it to
		 * tcp_input_listener on that first SYN.
		 */
		connp->conn_recv = tcp_input_listener_unbound;

		connp->conn_proto = IPPROTO_TCP;
		connp->conn_faddr_v6 = ipv6_all_zeros;
		connp->conn_fport = 0;

		(void) ipcl_bind_insert(connp);
	} else {
		tcp->tcp_state = TCPS_BOUND;
	}

	/*
	 * Initialize to default values
	 */
	tcp_init_values(tcp);

	ASSERT(tcp->tcp_ptpbhn != NULL);
	tcp->tcp_rwnd = connp->conn_rcvbuf;
	tcp->tcp_mss = connp->conn_ipversion != IPV4_VERSION ?
	    tcps->tcps_mss_def_ipv6 : tcps->tcps_mss_def_ipv4;
}

/*
 * Force values to zero that need be zero.
 * Do not touch values asociated with the BOUND or LISTEN state
 * since the connection will end up in that state after the reinit.
 * NOTE: tcp_reinit_values MUST have a line for each field in the tcp_t
 * structure!
 */
static void
tcp_reinit_values(tcp)
	tcp_t *tcp;
{
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

#ifndef	lint
#define	DONTCARE(x)
#define	PRESERVE(x)
#else
#define	DONTCARE(x)	((x) = (x))
#define	PRESERVE(x)	((x) = (x))
#endif	/* lint */

	PRESERVE(tcp->tcp_bind_hash_port);
	PRESERVE(tcp->tcp_bind_hash);
	PRESERVE(tcp->tcp_ptpbhn);
	PRESERVE(tcp->tcp_acceptor_hash);
	PRESERVE(tcp->tcp_ptpahn);

	/* Should be ASSERT NULL on these with new code! */
	ASSERT(tcp->tcp_time_wait_next == NULL);
	ASSERT(tcp->tcp_time_wait_prev == NULL);
	ASSERT(tcp->tcp_time_wait_expire == 0);
	PRESERVE(tcp->tcp_state);
	PRESERVE(connp->conn_rq);
	PRESERVE(connp->conn_wq);

	ASSERT(tcp->tcp_xmit_head == NULL);
	ASSERT(tcp->tcp_xmit_last == NULL);
	ASSERT(tcp->tcp_unsent == 0);
	ASSERT(tcp->tcp_xmit_tail == NULL);
	ASSERT(tcp->tcp_xmit_tail_unsent == 0);

	tcp->tcp_snxt = 0;			/* Displayed in mib */
	tcp->tcp_suna = 0;			/* Displayed in mib */
	tcp->tcp_swnd = 0;
	DONTCARE(tcp->tcp_cwnd);	/* Init in tcp_process_options */

	ASSERT(tcp->tcp_ibsegs == 0);
	ASSERT(tcp->tcp_obsegs == 0);

	if (connp->conn_ht_iphc != NULL) {
		kmem_free(connp->conn_ht_iphc, connp->conn_ht_iphc_allocated);
		connp->conn_ht_iphc = NULL;
		connp->conn_ht_iphc_allocated = 0;
		connp->conn_ht_iphc_len = 0;
		connp->conn_ht_ulp = NULL;
		connp->conn_ht_ulp_len = 0;
		tcp->tcp_ipha = NULL;
		tcp->tcp_ip6h = NULL;
		tcp->tcp_tcpha = NULL;
	}

	/* We clear any IP_OPTIONS and extension headers */
	ip_pkt_free(&connp->conn_xmit_ipp);

	DONTCARE(tcp->tcp_naglim);		/* Init in tcp_init_values */
	DONTCARE(tcp->tcp_ipha);
	DONTCARE(tcp->tcp_ip6h);
	DONTCARE(tcp->tcp_tcpha);
	tcp->tcp_valid_bits = 0;

	DONTCARE(tcp->tcp_timer_backoff);	/* Init in tcp_init_values */
	DONTCARE(tcp->tcp_last_recv_time);	/* Init in tcp_init_values */
	tcp->tcp_last_rcv_lbolt = 0;

	tcp->tcp_init_cwnd = 0;

	tcp->tcp_urp_last_valid = 0;
	tcp->tcp_hard_binding = 0;

	tcp->tcp_fin_acked = 0;
	tcp->tcp_fin_rcvd = 0;
	tcp->tcp_fin_sent = 0;
	tcp->tcp_ordrel_done = 0;

	tcp->tcp_detached = 0;

	tcp->tcp_snd_ws_ok = B_FALSE;
	tcp->tcp_snd_ts_ok = B_FALSE;
	tcp->tcp_zero_win_probe = 0;

	tcp->tcp_loopback = 0;
	tcp->tcp_localnet = 0;
	tcp->tcp_syn_defense = 0;
	tcp->tcp_set_timer = 0;

	tcp->tcp_active_open = 0;
	tcp->tcp_rexmit = B_FALSE;
	tcp->tcp_xmit_zc_clean = B_FALSE;

	tcp->tcp_snd_sack_ok = B_FALSE;
	tcp->tcp_hwcksum = B_FALSE;

	DONTCARE(tcp->tcp_maxpsz_multiplier);	/* Init in tcp_init_values */

	tcp->tcp_conn_def_q0 = 0;
	tcp->tcp_ip_forward_progress = B_FALSE;
	tcp->tcp_ecn_ok = B_FALSE;

	tcp->tcp_cwr = B_FALSE;
	tcp->tcp_ecn_echo_on = B_FALSE;
	tcp->tcp_is_wnd_shrnk = B_FALSE;

	if (tcp->tcp_sack_info != NULL) {
		if (tcp->tcp_notsack_list != NULL) {
			TCP_NOTSACK_REMOVE_ALL(tcp->tcp_notsack_list,
			    tcp);
		}
		kmem_cache_free(tcp_sack_info_cache, tcp->tcp_sack_info);
		tcp->tcp_sack_info = NULL;
	}

	tcp->tcp_rcv_ws = 0;
	tcp->tcp_snd_ws = 0;
	tcp->tcp_ts_recent = 0;
	tcp->tcp_rnxt = 0;			/* Displayed in mib */
	DONTCARE(tcp->tcp_rwnd);		/* Set in tcp_reinit() */
	tcp->tcp_initial_pmtu = 0;

	ASSERT(tcp->tcp_reass_head == NULL);
	ASSERT(tcp->tcp_reass_tail == NULL);

	tcp->tcp_cwnd_cnt = 0;

	ASSERT(tcp->tcp_rcv_list == NULL);
	ASSERT(tcp->tcp_rcv_last_head == NULL);
	ASSERT(tcp->tcp_rcv_last_tail == NULL);
	ASSERT(tcp->tcp_rcv_cnt == 0);

	DONTCARE(tcp->tcp_cwnd_ssthresh); /* Init in tcp_set_destination */
	DONTCARE(tcp->tcp_cwnd_max);		/* Init in tcp_init_values */
	tcp->tcp_csuna = 0;

	tcp->tcp_rto = 0;			/* Displayed in MIB */
	DONTCARE(tcp->tcp_rtt_sa);		/* Init in tcp_init_values */
	DONTCARE(tcp->tcp_rtt_sd);		/* Init in tcp_init_values */
	tcp->tcp_rtt_update = 0;

	DONTCARE(tcp->tcp_swl1); /* Init in case TCPS_LISTEN/TCPS_SYN_SENT */
	DONTCARE(tcp->tcp_swl2); /* Init in case TCPS_LISTEN/TCPS_SYN_SENT */

	tcp->tcp_rack = 0;			/* Displayed in mib */
	tcp->tcp_rack_cnt = 0;
	tcp->tcp_rack_cur_max = 0;
	tcp->tcp_rack_abs_max = 0;

	tcp->tcp_max_swnd = 0;

	ASSERT(tcp->tcp_listener == NULL);

	DONTCARE(tcp->tcp_irs);			/* tcp_valid_bits cleared */
	DONTCARE(tcp->tcp_iss);			/* tcp_valid_bits cleared */
	DONTCARE(tcp->tcp_fss);			/* tcp_valid_bits cleared */
	DONTCARE(tcp->tcp_urg);			/* tcp_valid_bits cleared */

	ASSERT(tcp->tcp_conn_req_cnt_q == 0);
	ASSERT(tcp->tcp_conn_req_cnt_q0 == 0);
	PRESERVE(tcp->tcp_conn_req_max);
	PRESERVE(tcp->tcp_conn_req_seqnum);

	DONTCARE(tcp->tcp_first_timer_threshold); /* Init in tcp_init_values */
	DONTCARE(tcp->tcp_second_timer_threshold); /* Init in tcp_init_values */
	DONTCARE(tcp->tcp_first_ctimer_threshold); /* Init in tcp_init_values */
	DONTCARE(tcp->tcp_second_ctimer_threshold); /* in tcp_init_values */

	DONTCARE(tcp->tcp_urp_last);	/* tcp_urp_last_valid is cleared */
	ASSERT(tcp->tcp_urp_mp == NULL);
	ASSERT(tcp->tcp_urp_mark_mp == NULL);
	ASSERT(tcp->tcp_fused_sigurg_mp == NULL);

	ASSERT(tcp->tcp_eager_next_q == NULL);
	ASSERT(tcp->tcp_eager_last_q == NULL);
	ASSERT((tcp->tcp_eager_next_q0 == NULL &&
	    tcp->tcp_eager_prev_q0 == NULL) ||
	    tcp->tcp_eager_next_q0 == tcp->tcp_eager_prev_q0);
	ASSERT(tcp->tcp_conn.tcp_eager_conn_ind == NULL);

	ASSERT((tcp->tcp_eager_next_drop_q0 == NULL &&
	    tcp->tcp_eager_prev_drop_q0 == NULL) ||
	    tcp->tcp_eager_next_drop_q0 == tcp->tcp_eager_prev_drop_q0);

	tcp->tcp_client_errno = 0;

	DONTCARE(connp->conn_sum);		/* Init in tcp_init_values */

	connp->conn_faddr_v6 = ipv6_all_zeros;	/* Displayed in MIB */

	PRESERVE(connp->conn_bound_addr_v6);
	tcp->tcp_last_sent_len = 0;
	tcp->tcp_dupack_cnt = 0;

	connp->conn_fport = 0;			/* Displayed in MIB */
	PRESERVE(connp->conn_lport);

	PRESERVE(tcp->tcp_acceptor_lockp);

	ASSERT(tcp->tcp_ordrel_mp == NULL);
	PRESERVE(tcp->tcp_acceptor_id);
	DONTCARE(tcp->tcp_ipsec_overhead);

	PRESERVE(connp->conn_family);
	/* Remove any remnants of mapped address binding */
	if (connp->conn_family == AF_INET6) {
		connp->conn_ipversion = IPV6_VERSION;
		tcp->tcp_mss = tcps->tcps_mss_def_ipv6;
	} else {
		connp->conn_ipversion = IPV4_VERSION;
		tcp->tcp_mss = tcps->tcps_mss_def_ipv4;
	}

	connp->conn_bound_if = 0;
	connp->conn_recv_ancillary.crb_all = 0;
	tcp->tcp_recvifindex = 0;
	tcp->tcp_recvhops = 0;
	tcp->tcp_closed = 0;
	tcp->tcp_cleandeathtag = 0;
	if (tcp->tcp_hopopts != NULL) {
		mi_free(tcp->tcp_hopopts);
		tcp->tcp_hopopts = NULL;
		tcp->tcp_hopoptslen = 0;
	}
	ASSERT(tcp->tcp_hopoptslen == 0);
	if (tcp->tcp_dstopts != NULL) {
		mi_free(tcp->tcp_dstopts);
		tcp->tcp_dstopts = NULL;
		tcp->tcp_dstoptslen = 0;
	}
	ASSERT(tcp->tcp_dstoptslen == 0);
	if (tcp->tcp_rthdrdstopts != NULL) {
		mi_free(tcp->tcp_rthdrdstopts);
		tcp->tcp_rthdrdstopts = NULL;
		tcp->tcp_rthdrdstoptslen = 0;
	}
	ASSERT(tcp->tcp_rthdrdstoptslen == 0);
	if (tcp->tcp_rthdr != NULL) {
		mi_free(tcp->tcp_rthdr);
		tcp->tcp_rthdr = NULL;
		tcp->tcp_rthdrlen = 0;
	}
	ASSERT(tcp->tcp_rthdrlen == 0);

	/* Reset fusion-related fields */
	tcp->tcp_fused = B_FALSE;
	tcp->tcp_unfusable = B_FALSE;
	tcp->tcp_fused_sigurg = B_FALSE;
	tcp->tcp_loopback_peer = NULL;

	tcp->tcp_lso = B_FALSE;

	tcp->tcp_in_ack_unsent = 0;
	tcp->tcp_cork = B_FALSE;
	tcp->tcp_tconnind_started = B_FALSE;

	PRESERVE(tcp->tcp_squeue_bytes);

	ASSERT(tcp->tcp_kssl_ctx == NULL);
	ASSERT(!tcp->tcp_kssl_pending);
	PRESERVE(tcp->tcp_kssl_ent);

	tcp->tcp_closemp_used = B_FALSE;

	PRESERVE(tcp->tcp_rsrv_mp);
	PRESERVE(tcp->tcp_rsrv_mp_lock);

#ifdef DEBUG
	DONTCARE(tcp->tcmp_stk[0]);
#endif

	PRESERVE(tcp->tcp_connid);


#undef	DONTCARE
#undef	PRESERVE
}

static void
tcp_init_values(tcp_t *tcp)
{
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

	ASSERT((connp->conn_family == AF_INET &&
	    connp->conn_ipversion == IPV4_VERSION) ||
	    (connp->conn_family == AF_INET6 &&
	    (connp->conn_ipversion == IPV4_VERSION ||
	    connp->conn_ipversion == IPV6_VERSION)));

	/*
	 * Initialize tcp_rtt_sa and tcp_rtt_sd so that the calculated RTO
	 * will be close to tcp_rexmit_interval_initial.  By doing this, we
	 * allow the algorithm to adjust slowly to large fluctuations of RTT
	 * during first few transmissions of a connection as seen in slow
	 * links.
	 */
	tcp->tcp_rtt_sa = tcps->tcps_rexmit_interval_initial << 2;
	tcp->tcp_rtt_sd = tcps->tcps_rexmit_interval_initial >> 1;
	tcp->tcp_rto = (tcp->tcp_rtt_sa >> 3) + tcp->tcp_rtt_sd +
	    tcps->tcps_rexmit_interval_extra + (tcp->tcp_rtt_sa >> 5) +
	    tcps->tcps_conn_grace_period;
	if (tcp->tcp_rto < tcps->tcps_rexmit_interval_min)
		tcp->tcp_rto = tcps->tcps_rexmit_interval_min;
	tcp->tcp_timer_backoff = 0;
	tcp->tcp_ms_we_have_waited = 0;
	tcp->tcp_last_recv_time = ddi_get_lbolt();
	tcp->tcp_cwnd_max = tcps->tcps_cwnd_max_;
	tcp->tcp_cwnd_ssthresh = TCP_MAX_LARGEWIN;
	tcp->tcp_snd_burst = TCP_CWND_INFINITE;

	tcp->tcp_maxpsz_multiplier = tcps->tcps_maxpsz_multiplier;

	tcp->tcp_first_timer_threshold = tcps->tcps_ip_notify_interval;
	tcp->tcp_first_ctimer_threshold = tcps->tcps_ip_notify_cinterval;
	tcp->tcp_second_timer_threshold = tcps->tcps_ip_abort_interval;
	/*
	 * Fix it to tcp_ip_abort_linterval later if it turns out to be a
	 * passive open.
	 */
	tcp->tcp_second_ctimer_threshold = tcps->tcps_ip_abort_cinterval;

	tcp->tcp_naglim = tcps->tcps_naglim_def;

	/* NOTE:  ISS is now set in tcp_set_destination(). */

	/* Reset fusion-related fields */
	tcp->tcp_fused = B_FALSE;
	tcp->tcp_unfusable = B_FALSE;
	tcp->tcp_fused_sigurg = B_FALSE;
	tcp->tcp_loopback_peer = NULL;

	/* We rebuild the header template on the next connect/conn_request */

	connp->conn_mlp_type = mlptSingle;

	/*
	 * Init the window scale to the max so tcp_rwnd_set() won't pare
	 * down tcp_rwnd. tcp_set_destination() will set the right value later.
	 */
	tcp->tcp_rcv_ws = TCP_MAX_WINSHIFT;
	tcp->tcp_rwnd = connp->conn_rcvbuf;

	tcp->tcp_cork = B_FALSE;
	/*
	 * Init the tcp_debug option if it wasn't already set.  This value
	 * determines whether TCP
	 * calls strlog() to print out debug messages.  Doing this
	 * initialization here means that this value is not inherited thru
	 * tcp_reinit().
	 */
	if (!connp->conn_debug)
		connp->conn_debug = tcps->tcps_dbg;

	tcp->tcp_ka_interval = tcps->tcps_keepalive_interval;
	tcp->tcp_ka_abort_thres = tcps->tcps_keepalive_abort_interval;
}

/* At minimum we need 8 bytes in the TCP header for the lookup */
#define	ICMP_MIN_TCP_HDR	8

/*
 * tcp_icmp_input is called as conn_recvicmp to process ICMP error messages
 * passed up by IP. The message is always received on the correct tcp_t.
 * Assumes that IP has pulled up everything up to and including the ICMP header.
 */
/* ARGSUSED2 */
static void
tcp_icmp_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
{
	conn_t		*connp = (conn_t *)arg1;
	icmph_t		*icmph;
	ipha_t		*ipha;
	int		iph_hdr_length;
	tcpha_t		*tcpha;
	uint32_t	seg_seq;
	tcp_t		*tcp = connp->conn_tcp;

	/* Assume IP provides aligned packets */
	ASSERT(OK_32PTR(mp->b_rptr));
	ASSERT((MBLKL(mp) >= sizeof (ipha_t)));

	/*
	 * Verify IP version. Anything other than IPv4 or IPv6 packet is sent
	 * upstream. ICMPv6 is handled in tcp_icmp_error_ipv6.
	 */
	if (!(ira->ira_flags & IRAF_IS_IPV4)) {
		tcp_icmp_error_ipv6(tcp, mp, ira);
		return;
	}

	/* Skip past the outer IP and ICMP headers */
	iph_hdr_length = ira->ira_ip_hdr_length;
	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
	/*
	 * If we don't have the correct outer IP header length
	 * or if we don't have a complete inner IP header
	 * drop it.
	 */
	if (iph_hdr_length < sizeof (ipha_t) ||
	    (ipha_t *)&icmph[1] + 1 > (ipha_t *)mp->b_wptr) {
noticmpv4:
		freemsg(mp);
		return;
	}
	ipha = (ipha_t *)&icmph[1];

	/* Skip past the inner IP and find the ULP header */
	iph_hdr_length = IPH_HDR_LENGTH(ipha);
	tcpha = (tcpha_t *)((char *)ipha + iph_hdr_length);
	/*
	 * If we don't have the correct inner IP header length or if the ULP
	 * is not IPPROTO_TCP or if we don't have at least ICMP_MIN_TCP_HDR
	 * bytes of TCP header, drop it.
	 */
	if (iph_hdr_length < sizeof (ipha_t) ||
	    ipha->ipha_protocol != IPPROTO_TCP ||
	    (uchar_t *)tcpha + ICMP_MIN_TCP_HDR > mp->b_wptr) {
		goto noticmpv4;
	}

	seg_seq = ntohl(tcpha->tha_seq);
	switch (icmph->icmph_type) {
	case ICMP_DEST_UNREACHABLE:
		switch (icmph->icmph_code) {
		case ICMP_FRAGMENTATION_NEEDED:
			/*
			 * Update Path MTU, then try to send something out.
			 */
			tcp_update_pmtu(tcp, B_TRUE);
			tcp_rexmit_after_error(tcp);
			break;
		case ICMP_PORT_UNREACHABLE:
		case ICMP_PROTOCOL_UNREACHABLE:
			switch (tcp->tcp_state) {
			case TCPS_SYN_SENT:
			case TCPS_SYN_RCVD:
				/*
				 * ICMP can snipe away incipient
				 * TCP connections as long as
				 * seq number is same as initial
				 * send seq number.
				 */
				if (seg_seq == tcp->tcp_iss) {
					(void) tcp_clean_death(tcp,
					    ECONNREFUSED, 6);
				}
				break;
			}
			break;
		case ICMP_HOST_UNREACHABLE:
		case ICMP_NET_UNREACHABLE:
			/* Record the error in case we finally time out. */
			if (icmph->icmph_code == ICMP_HOST_UNREACHABLE)
				tcp->tcp_client_errno = EHOSTUNREACH;
			else
				tcp->tcp_client_errno = ENETUNREACH;
			if (tcp->tcp_state == TCPS_SYN_RCVD) {
				if (tcp->tcp_listener != NULL &&
				    tcp->tcp_listener->tcp_syn_defense) {
					/*
					 * Ditch the half-open connection if we
					 * suspect a SYN attack is under way.
					 */
					(void) tcp_clean_death(tcp,
					    tcp->tcp_client_errno, 7);
				}
			}
			break;
		default:
			break;
		}
		break;
	case ICMP_SOURCE_QUENCH: {
		/*
		 * use a global boolean to control
		 * whether TCP should respond to ICMP_SOURCE_QUENCH.
		 * The default is false.
		 */
		if (tcp_icmp_source_quench) {
			/*
			 * Reduce the sending rate as if we got a
			 * retransmit timeout
			 */
			uint32_t npkt;

			npkt = ((tcp->tcp_snxt - tcp->tcp_suna) >> 1) /
			    tcp->tcp_mss;
			tcp->tcp_cwnd_ssthresh = MAX(npkt, 2) * tcp->tcp_mss;
			tcp->tcp_cwnd = tcp->tcp_mss;
			tcp->tcp_cwnd_cnt = 0;
		}
		break;
	}
	}
	freemsg(mp);
}

/*
 * CALLED OUTSIDE OF SQUEUE! It can not follow any pointers that tcp might
 * change. But it can refer to fields like tcp_suna and tcp_snxt.
 *
 * Function tcp_verifyicmp is called as conn_verifyicmp to verify the ICMP
 * error messages received by IP. The message is always received on the correct
 * tcp_t.
 */
/* ARGSUSED */
static boolean_t
tcp_verifyicmp(conn_t *connp, void *arg2, icmph_t *icmph, icmp6_t *icmp6,
    ip_recv_attr_t *ira)
{
	tcpha_t		*tcpha = (tcpha_t *)arg2;
	uint32_t	seq = ntohl(tcpha->tha_seq);
	tcp_t		*tcp = connp->conn_tcp;

	/*
	 * TCP sequence number contained in payload of the ICMP error message
	 * should be within the range SND.UNA <= SEG.SEQ < SND.NXT. Otherwise,
	 * the message is either a stale ICMP error, or an attack from the
	 * network. Fail the verification.
	 */
	if (SEQ_LT(seq, tcp->tcp_suna) || SEQ_GEQ(seq, tcp->tcp_snxt))
		return (B_FALSE);

	/* For "too big" we also check the ignore flag */
	if (ira->ira_flags & IRAF_IS_IPV4) {
		ASSERT(icmph != NULL);
		if (icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
		    icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED &&
		    tcp->tcp_tcps->tcps_ignore_path_mtu)
			return (B_FALSE);
	} else {
		ASSERT(icmp6 != NULL);
		if (icmp6->icmp6_type == ICMP6_PACKET_TOO_BIG &&
		    tcp->tcp_tcps->tcps_ignore_path_mtu)
			return (B_FALSE);
	}
	return (B_TRUE);
}

/*
 * Update the TCP connection according to change of PMTU.
 *
 * Path MTU might have changed by either increase or decrease, so need to
 * adjust the MSS based on the value of ixa_pmtu. No need to handle tiny
 * or negative MSS, since tcp_mss_set() will do it.
 */
static void
tcp_update_pmtu(tcp_t *tcp, boolean_t decrease_only)
{
	uint32_t	pmtu;
	int32_t		mss;
	conn_t		*connp = tcp->tcp_connp;
	ip_xmit_attr_t	*ixa = connp->conn_ixa;
	iaflags_t	ixaflags;

	if (tcp->tcp_tcps->tcps_ignore_path_mtu)
		return;

	if (tcp->tcp_state < TCPS_ESTABLISHED)
		return;

	/*
	 * Always call ip_get_pmtu() to make sure that IP has updated
	 * ixa_flags properly.
	 */
	pmtu = ip_get_pmtu(ixa);
	ixaflags = ixa->ixa_flags;

	/*
	 * Calculate the MSS by decreasing the PMTU by conn_ht_iphc_len and
	 * IPsec overhead if applied. Make sure to use the most recent
	 * IPsec information.
	 */
	mss = pmtu - connp->conn_ht_iphc_len - conn_ipsec_length(connp);

	/*
	 * Nothing to change, so just return.
	 */
	if (mss == tcp->tcp_mss)
		return;

	/*
	 * Currently, for ICMP errors, only PMTU decrease is handled.
	 */
	if (mss > tcp->tcp_mss && decrease_only)
		return;

	DTRACE_PROBE2(tcp_update_pmtu, int32_t, tcp->tcp_mss, uint32_t, mss);

	/*
	 * Update ixa_fragsize and ixa_pmtu.
	 */
	ixa->ixa_fragsize = ixa->ixa_pmtu = pmtu;

	/*
	 * Adjust MSS and all relevant variables.
	 */
	tcp_mss_set(tcp, mss);

	/*
	 * If the PMTU is below the min size maintained by IP, then ip_get_pmtu
	 * has set IXAF_PMTU_TOO_SMALL and cleared IXAF_PMTU_IPV4_DF. Since TCP
	 * has a (potentially different) min size we do the same. Make sure to
	 * clear IXAF_DONTFRAG, which is used by IP to decide whether to
	 * fragment the packet.
	 *
	 * LSO over IPv6 can not be fragmented. So need to disable LSO
	 * when IPv6 fragmentation is needed.
	 */
	if (mss < tcp->tcp_tcps->tcps_mss_min)
		ixaflags |= IXAF_PMTU_TOO_SMALL;

	if (ixaflags & IXAF_PMTU_TOO_SMALL)
		ixaflags &= ~(IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);

	if ((connp->conn_ipversion == IPV4_VERSION) &&
	    !(ixaflags & IXAF_PMTU_IPV4_DF)) {
		tcp->tcp_ipha->ipha_fragment_offset_and_flags = 0;
	}
	ixa->ixa_flags = ixaflags;
}

/*
 * Do slow start retransmission after ICMP errors of PMTU changes.
 */
static void
tcp_rexmit_after_error(tcp_t *tcp)
{
	/*
	 * All sent data has been acknowledged or no data left to send, just
	 * to return.
	 */
	if (!SEQ_LT(tcp->tcp_suna, tcp->tcp_snxt) ||
	    (tcp->tcp_xmit_head == NULL))
		return;

	if ((tcp->tcp_valid_bits & TCP_FSS_VALID) && (tcp->tcp_unsent == 0))
		tcp->tcp_rexmit_max = tcp->tcp_fss;
	else
		tcp->tcp_rexmit_max = tcp->tcp_snxt;

	tcp->tcp_rexmit_nxt = tcp->tcp_suna;
	tcp->tcp_rexmit = B_TRUE;
	tcp->tcp_dupack_cnt = 0;
	tcp->tcp_snd_burst = TCP_CWND_SS;
	tcp_ss_rexmit(tcp);
}

/*
 * tcp_icmp_error_ipv6 is called from tcp_icmp_input to process ICMPv6
 * error messages passed up by IP.
 * Assumes that IP has pulled up all the extension headers as well
 * as the ICMPv6 header.
 */
static void
tcp_icmp_error_ipv6(tcp_t *tcp, mblk_t *mp, ip_recv_attr_t *ira)
{
	icmp6_t		*icmp6;
	ip6_t		*ip6h;
	uint16_t	iph_hdr_length = ira->ira_ip_hdr_length;
	tcpha_t		*tcpha;
	uint8_t		*nexthdrp;
	uint32_t	seg_seq;

	/*
	 * Verify that we have a complete IP header.
	 */
	ASSERT((MBLKL(mp) >= sizeof (ip6_t)));

	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
	ip6h = (ip6_t *)&icmp6[1];
	/*
	 * Verify if we have a complete ICMP and inner IP header.
	 */
	if ((uchar_t *)&ip6h[1] > mp->b_wptr) {
noticmpv6:
		freemsg(mp);
		return;
	}

	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp))
		goto noticmpv6;
	tcpha = (tcpha_t *)((char *)ip6h + iph_hdr_length);
	/*
	 * Validate inner header. If the ULP is not IPPROTO_TCP or if we don't
	 * have at least ICMP_MIN_TCP_HDR bytes of  TCP header drop the
	 * packet.
	 */
	if ((*nexthdrp != IPPROTO_TCP) ||
	    ((uchar_t *)tcpha + ICMP_MIN_TCP_HDR) > mp->b_wptr) {
		goto noticmpv6;
	}

	seg_seq = ntohl(tcpha->tha_seq);
	switch (icmp6->icmp6_type) {
	case ICMP6_PACKET_TOO_BIG:
		/*
		 * Update Path MTU, then try to send something out.
		 */
		tcp_update_pmtu(tcp, B_TRUE);
		tcp_rexmit_after_error(tcp);
		break;
	case ICMP6_DST_UNREACH:
		switch (icmp6->icmp6_code) {
		case ICMP6_DST_UNREACH_NOPORT:
			if (((tcp->tcp_state == TCPS_SYN_SENT) ||
			    (tcp->tcp_state == TCPS_SYN_RCVD)) &&
			    (seg_seq == tcp->tcp_iss)) {
				(void) tcp_clean_death(tcp,
				    ECONNREFUSED, 8);
			}
			break;
		case ICMP6_DST_UNREACH_ADMIN:
		case ICMP6_DST_UNREACH_NOROUTE:
		case ICMP6_DST_UNREACH_BEYONDSCOPE:
		case ICMP6_DST_UNREACH_ADDR:
			/* Record the error in case we finally time out. */
			tcp->tcp_client_errno = EHOSTUNREACH;
			if (((tcp->tcp_state == TCPS_SYN_SENT) ||
			    (tcp->tcp_state == TCPS_SYN_RCVD)) &&
			    (seg_seq == tcp->tcp_iss)) {
				if (tcp->tcp_listener != NULL &&
				    tcp->tcp_listener->tcp_syn_defense) {
					/*
					 * Ditch the half-open connection if we
					 * suspect a SYN attack is under way.
					 */
					(void) tcp_clean_death(tcp,
					    tcp->tcp_client_errno, 9);
				}
			}


			break;
		default:
			break;
		}
		break;
	case ICMP6_PARAM_PROB:
		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
		    (uchar_t *)nexthdrp) {
			if (tcp->tcp_state == TCPS_SYN_SENT ||
			    tcp->tcp_state == TCPS_SYN_RCVD) {
				(void) tcp_clean_death(tcp,
				    ECONNREFUSED, 10);
			}
			break;
		}
		break;

	case ICMP6_TIME_EXCEEDED:
	default:
		break;
	}
	freemsg(mp);
}

/*
 * Notify IP that we are having trouble with this connection.  IP should
 * make note so it can potentially use a different IRE.
 */
static void
tcp_ip_notify(tcp_t *tcp)
{
	conn_t		*connp = tcp->tcp_connp;
	ire_t		*ire;

	/*
	 * Note: in the case of source routing we want to blow away the
	 * route to the first source route hop.
	 */
	ire = connp->conn_ixa->ixa_ire;
	if (ire != NULL && !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
		if (ire->ire_ipversion == IPV4_VERSION) {
			/*
			 * As per RFC 1122, we send an RTM_LOSING to inform
			 * routing protocols.
			 */
			ip_rts_change(RTM_LOSING, ire->ire_addr,
			    ire->ire_gateway_addr, ire->ire_mask,
			    connp->conn_laddr_v4,  0, 0, 0,
			    (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_IFA),
			    ire->ire_ipst);
		}
		(void) ire_no_good(ire);
	}
}

#pragma inline(tcp_send_data)

/*
 * Timer callback routine for keepalive probe.  We do a fake resend of
 * last ACKed byte.  Then set a timer using RTO.  When the timer expires,
 * check to see if we have heard anything from the other end for the last
 * RTO period.  If we have, set the timer to expire for another
 * tcp_keepalive_intrvl and check again.  If we have not, set a timer using
 * RTO << 1 and check again when it expires.  Keep exponentially increasing
 * the timeout if we have not heard from the other side.  If for more than
 * (tcp_ka_interval + tcp_ka_abort_thres) we have not heard anything,
 * kill the connection unless the keepalive abort threshold is 0.  In
 * that case, we will probe "forever."
 */
static void
tcp_keepalive_killer(void *arg)
{
	mblk_t	*mp;
	conn_t	*connp = (conn_t *)arg;
	tcp_t  	*tcp = connp->conn_tcp;
	int32_t	firetime;
	int32_t	idletime;
	int32_t	ka_intrvl;
	tcp_stack_t	*tcps = tcp->tcp_tcps;

	tcp->tcp_ka_tid = 0;

	if (tcp->tcp_fused)
		return;

	BUMP_MIB(&tcps->tcps_mib, tcpTimKeepalive);
	ka_intrvl = tcp->tcp_ka_interval;

	/*
	 * Keepalive probe should only be sent if the application has not
	 * done a close on the connection.
	 */
	if (tcp->tcp_state > TCPS_CLOSE_WAIT) {
		return;
	}
	/* Timer fired too early, restart it. */
	if (tcp->tcp_state < TCPS_ESTABLISHED) {
		tcp->tcp_ka_tid = TCP_TIMER(tcp, tcp_keepalive_killer,
		    MSEC_TO_TICK(ka_intrvl));
		return;
	}

	idletime = TICK_TO_MSEC(ddi_get_lbolt() - tcp->tcp_last_recv_time);
	/*
	 * If we have not heard from the other side for a long
	 * time, kill the connection unless the keepalive abort
	 * threshold is 0.  In that case, we will probe "forever."
	 */
	if (tcp->tcp_ka_abort_thres != 0 &&
	    idletime > (ka_intrvl + tcp->tcp_ka_abort_thres)) {
		BUMP_MIB(&tcps->tcps_mib, tcpTimKeepaliveDrop);
		(void) tcp_clean_death(tcp, tcp->tcp_client_errno ?
		    tcp->tcp_client_errno : ETIMEDOUT, 11);
		return;
	}

	if (tcp->tcp_snxt == tcp->tcp_suna &&
	    idletime >= ka_intrvl) {
		/* Fake resend of last ACKed byte. */
		mblk_t	*mp1 = allocb(1, BPRI_LO);

		if (mp1 != NULL) {
			*mp1->b_wptr++ = '\0';
			mp = tcp_xmit_mp(tcp, mp1, 1, NULL, NULL,
			    tcp->tcp_suna - 1, B_FALSE, NULL, B_TRUE);
			freeb(mp1);
			/*
			 * if allocation failed, fall through to start the
			 * timer back.
			 */
			if (mp != NULL) {
				tcp_send_data(tcp, mp);
				BUMP_MIB(&tcps->tcps_mib,
				    tcpTimKeepaliveProbe);
				if (tcp->tcp_ka_last_intrvl != 0) {
					int max;
					/*
					 * We should probe again at least
					 * in ka_intrvl, but not more than
					 * tcp_rexmit_interval_max.
					 */
					max = tcps->tcps_rexmit_interval_max;
					firetime = MIN(ka_intrvl - 1,
					    tcp->tcp_ka_last_intrvl << 1);
					if (firetime > max)
						firetime = max;
				} else {
					firetime = tcp->tcp_rto;
				}
				tcp->tcp_ka_tid = TCP_TIMER(tcp,
				    tcp_keepalive_killer,
				    MSEC_TO_TICK(firetime));
				tcp->tcp_ka_last_intrvl = firetime;
				return;
			}
		}
	} else {
		tcp->tcp_ka_last_intrvl = 0;
	}

	/* firetime can be negative if (mp1 == NULL || mp == NULL) */
	if ((firetime = ka_intrvl - idletime) < 0) {
		firetime = ka_intrvl;
	}
	tcp->tcp_ka_tid = TCP_TIMER(tcp, tcp_keepalive_killer,
	    MSEC_TO_TICK(firetime));
}

int
tcp_maxpsz_set(tcp_t *tcp, boolean_t set_maxblk)
{
	conn_t	*connp = tcp->tcp_connp;
	queue_t	*q = connp->conn_rq;
	int32_t	mss = tcp->tcp_mss;
	int	maxpsz;

	if (TCP_IS_DETACHED(tcp))
		return (mss);
	if (tcp->tcp_fused) {
		maxpsz = tcp_fuse_maxpsz(tcp);
		mss = INFPSZ;
	} else if (tcp->tcp_maxpsz_multiplier == 0) {
		/*
		 * Set the sd_qn_maxpsz according to the socket send buffer
		 * size, and sd_maxblk to INFPSZ (-1).  This will essentially
		 * instruct the stream head to copyin user data into contiguous
		 * kernel-allocated buffers without breaking it up into smaller
		 * chunks.  We round up the buffer size to the nearest SMSS.
		 */
		maxpsz = MSS_ROUNDUP(connp->conn_sndbuf, mss);
		if (tcp->tcp_kssl_ctx == NULL)
			mss = INFPSZ;
		else
			mss = SSL3_MAX_RECORD_LEN;
	} else {
		/*
		 * Set sd_qn_maxpsz to approx half the (receivers) buffer
		 * (and a multiple of the mss).  This instructs the stream
		 * head to break down larger than SMSS writes into SMSS-
		 * size mblks, up to tcp_maxpsz_multiplier mblks at a time.
		 */
		maxpsz = tcp->tcp_maxpsz_multiplier * mss;
		if (maxpsz > connp->conn_sndbuf / 2) {
			maxpsz = connp->conn_sndbuf / 2;
			/* Round up to nearest mss */
			maxpsz = MSS_ROUNDUP(maxpsz, mss);
		}
	}

	(void) proto_set_maxpsz(q, connp, maxpsz);
	if (!(IPCL_IS_NONSTR(connp)))
		connp->conn_wq->q_maxpsz = maxpsz;
	if (set_maxblk)
		(void) proto_set_tx_maxblk(q, connp, mss);
	return (mss);
}

/*
 * Extract option values from a tcp header.  We put any found values into the
 * tcpopt struct and return a bitmask saying which options were found.
 */
static int
tcp_parse_options(tcpha_t *tcpha, tcp_opt_t *tcpopt)
{
	uchar_t		*endp;
	int		len;
	uint32_t	mss;
	uchar_t		*up = (uchar_t *)tcpha;
	int		found = 0;
	int32_t		sack_len;
	tcp_seq		sack_begin, sack_end;
	tcp_t		*tcp;

	endp = up + TCP_HDR_LENGTH(tcpha);
	up += TCP_MIN_HEADER_LENGTH;
	while (up < endp) {
		len = endp - up;
		switch (*up) {
		case TCPOPT_EOL:
			break;

		case TCPOPT_NOP:
			up++;
			continue;

		case TCPOPT_MAXSEG:
			if (len < TCPOPT_MAXSEG_LEN ||
			    up[1] != TCPOPT_MAXSEG_LEN)
				break;

			mss = BE16_TO_U16(up+2);
			/* Caller must handle tcp_mss_min and tcp_mss_max_* */
			tcpopt->tcp_opt_mss = mss;
			found |= TCP_OPT_MSS_PRESENT;

			up += TCPOPT_MAXSEG_LEN;
			continue;

		case TCPOPT_WSCALE:
			if (len < TCPOPT_WS_LEN || up[1] != TCPOPT_WS_LEN)
				break;

			if (up[2] > TCP_MAX_WINSHIFT)
				tcpopt->tcp_opt_wscale = TCP_MAX_WINSHIFT;
			else
				tcpopt->tcp_opt_wscale = up[2];
			found |= TCP_OPT_WSCALE_PRESENT;

			up += TCPOPT_WS_LEN;
			continue;

		case TCPOPT_SACK_PERMITTED:
			if (len < TCPOPT_SACK_OK_LEN ||
			    up[1] != TCPOPT_SACK_OK_LEN)
				break;
			found |= TCP_OPT_SACK_OK_PRESENT;
			up += TCPOPT_SACK_OK_LEN;
			continue;

		case TCPOPT_SACK:
			if (len <= 2 || up[1] <= 2 || len < up[1])
				break;

			/* If TCP is not interested in SACK blks... */
			if ((tcp = tcpopt->tcp) == NULL) {
				up += up[1];
				continue;
			}
			sack_len = up[1] - TCPOPT_HEADER_LEN;
			up += TCPOPT_HEADER_LEN;

			/*
			 * If the list is empty, allocate one and assume
			 * nothing is sack'ed.
			 */
			ASSERT(tcp->tcp_sack_info != NULL);
			if (tcp->tcp_notsack_list == NULL) {
				tcp_notsack_update(&(tcp->tcp_notsack_list),
				    tcp->tcp_suna, tcp->tcp_snxt,
				    &(tcp->tcp_num_notsack_blk),
				    &(tcp->tcp_cnt_notsack_list));

				/*
				 * Make sure tcp_notsack_list is not NULL.
				 * This happens when kmem_alloc(KM_NOSLEEP)
				 * returns NULL.
				 */
				if (tcp->tcp_notsack_list == NULL) {
					up += sack_len;
					continue;
				}
				tcp->tcp_fack = tcp->tcp_suna;
			}

			while (sack_len > 0) {
				if (up + 8 > endp) {
					up = endp;
					break;
				}
				sack_begin = BE32_TO_U32(up);
				up += 4;
				sack_end = BE32_TO_U32(up);
				up += 4;
				sack_len -= 8;
				/*
				 * Bounds checking.  Make sure the SACK
				 * info is within tcp_suna and tcp_snxt.
				 * If this SACK blk is out of bound, ignore
				 * it but continue to parse the following
				 * blks.
				 */
				if (SEQ_LEQ(sack_end, sack_begin) ||
				    SEQ_LT(sack_begin, tcp->tcp_suna) ||
				    SEQ_GT(sack_end, tcp->tcp_snxt)) {
					continue;
				}
				tcp_notsack_insert(&(tcp->tcp_notsack_list),
				    sack_begin, sack_end,
				    &(tcp->tcp_num_notsack_blk),
				    &(tcp->tcp_cnt_notsack_list));
				if (SEQ_GT(sack_end, tcp->tcp_fack)) {
					tcp->tcp_fack = sack_end;
				}
			}
			found |= TCP_OPT_SACK_PRESENT;
			continue;

		case TCPOPT_TSTAMP:
			if (len < TCPOPT_TSTAMP_LEN ||
			    up[1] != TCPOPT_TSTAMP_LEN)
				break;

			tcpopt->tcp_opt_ts_val = BE32_TO_U32(up+2);
			tcpopt->tcp_opt_ts_ecr = BE32_TO_U32(up+6);

			found |= TCP_OPT_TSTAMP_PRESENT;

			up += TCPOPT_TSTAMP_LEN;
			continue;

		default:
			if (len <= 1 || len < (int)up[1] || up[1] == 0)
				break;
			up += up[1];
			continue;
		}
		break;
	}
	return (found);
}

/*
 * Set the MSS associated with a particular tcp based on its current value,
 * and a new one passed in. Observe minimums and maximums, and reset other
 * state variables that we want to view as multiples of MSS.
 *
 * The value of MSS could be either increased or descreased.
 */
static void
tcp_mss_set(tcp_t *tcp, uint32_t mss)
{
	uint32_t	mss_max;
	tcp_stack_t	*tcps = tcp->tcp_tcps;
	conn_t		*connp = tcp->tcp_connp;

	if (connp->conn_ipversion == IPV4_VERSION)
		mss_max = tcps->tcps_mss_max_ipv4;
	else
		mss_max = tcps->tcps_mss_max_ipv6;

	if (mss < tcps->tcps_mss_min)
		mss = tcps->tcps_mss_min;
	if (mss > mss_max)
		mss = mss_max;
	/*
	 * Unless naglim has been set by our client to
	 * a non-mss value, force naglim to track mss.
	 * This can help to aggregate small writes.
	 */
	if (mss < tcp->tcp_naglim || tcp->tcp_mss == tcp->tcp_naglim)
		tcp->tcp_naglim = mss;
	/*
	 * TCP should be able to buffer at least 4 MSS data for obvious
	 * performance reason.
	 */
	if ((mss << 2) > connp->conn_sndbuf)
		connp->conn_sndbuf = mss << 2;

	/*
	 * Set the send lowater to at least twice of MSS.
	 */
	if ((mss << 1) > connp->conn_sndlowat)
		connp->conn_sndlowat = mss << 1;

	/*
	 * Update tcp_cwnd according to the new value of MSS. Keep the
	 * previous ratio to preserve the transmit rate.
	 */
	tcp->tcp_cwnd = (tcp->tcp_cwnd / tcp->tcp_mss) * mss;
	tcp->tcp_cwnd_cnt = 0;

	tcp->tcp_mss = mss;
	(void) tcp_maxpsz_set(tcp, B_TRUE);
}

/* For /dev/tcp aka AF_INET open */
static int
tcp_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
{
	return (tcp_open(q, devp, flag, sflag, credp, B_FALSE));
}

/* For /dev/tcp6 aka AF_INET6 open */
static int
tcp_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
{
	return (tcp_open(q, devp, flag, sflag, credp, B_TRUE));
}

static conn_t *
tcp_create_common(cred_t *credp, boolean_t isv6, boolean_t issocket,
    int *errorp)
{
	tcp_t		*tcp = NULL;
	conn_t		*connp;
	zoneid_t	zoneid;
	tcp_stack_t	*tcps;
	squeue_t	*sqp;

	ASSERT(errorp != NULL);
	/*
	 * Find the proper zoneid and netstack.
	 */
	/*
	 * Special case for install: miniroot needs to be able to
	 * access files via NFS as though it were always in the
	 * global zone.
	 */
	if (credp == kcred && nfs_global_client_only != 0) {
		zoneid = GLOBAL_ZONEID;
		tcps = netstack_find_by_stackid(GLOBAL_NETSTACKID)->
		    netstack_tcp;
		ASSERT(tcps != NULL);
	} else {
		netstack_t *ns;

		ns = netstack_find_by_cred(credp);
		ASSERT(ns != NULL);
		tcps = ns->netstack_tcp;
		ASSERT(tcps != NULL);

		/*
		 * For exclusive stacks we set the zoneid to zero
		 * to make TCP operate as if in the global zone.
		 */
		if (tcps->tcps_netstack->netstack_stackid !=
		    GLOBAL_NETSTACKID)
			zoneid = GLOBAL_ZONEID;
		else
			zoneid = crgetzoneid(credp);
	}

	sqp = IP_SQUEUE_GET((uint_t)gethrtime());
	connp = (conn_t *)tcp_get_conn(sqp, tcps);
	/*
	 * Both tcp_get_conn and netstack_find_by_cred incremented refcnt,
	 * so we drop it by one.
	 */
	netstack_rele(tcps->tcps_netstack);
	if (connp == NULL) {
		*errorp = ENOSR;
		return (NULL);
	}
	ASSERT(connp->conn_ixa->ixa_protocol == connp->conn_proto);

	connp->conn_sqp = sqp;
	connp->conn_initial_sqp = connp->conn_sqp;
	connp->conn_ixa->ixa_sqp = connp->conn_sqp;
	tcp = connp->conn_tcp;

	/*
	 * Besides asking IP to set the checksum for us, have conn_ip_output
	 * to do the following checks when necessary:
	 *
	 * IXAF_VERIFY_SOURCE: drop packets when our outer source goes invalid
	 * IXAF_VERIFY_PMTU: verify PMTU changes
	 * IXAF_VERIFY_LSO: verify LSO capability changes
	 */
	connp->conn_ixa->ixa_flags |= IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
	    IXAF_VERIFY_PMTU | IXAF_VERIFY_LSO;

	if (!tcps->tcps_dev_flow_ctl)
		connp->conn_ixa->ixa_flags |= IXAF_NO_DEV_FLOW_CTL;

	if (isv6) {
		connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
		connp->conn_ipversion = IPV6_VERSION;
		connp->conn_family = AF_INET6;
		tcp->tcp_mss = tcps->tcps_mss_def_ipv6;
		connp->conn_default_ttl = tcps->tcps_ipv6_hoplimit;
	} else {
		connp->conn_ipversion = IPV4_VERSION;
		connp->conn_family = AF_INET;
		tcp->tcp_mss = tcps->tcps_mss_def_ipv4;
		connp->conn_default_ttl = tcps->tcps_ipv4_ttl;
	}
	connp->conn_xmit_ipp.ipp_unicast_hops = connp->conn_default_ttl;

	crhold(credp);
	connp->conn_cred = credp;
	connp->conn_cpid = curproc->p_pid;
	connp->conn_open_time = ddi_get_lbolt64();

	connp->conn_zoneid = zoneid;
	/* conn_allzones can not be set this early, hence no IPCL_ZONEID */
	connp->conn_ixa->ixa_zoneid = zoneid;
	connp->conn_mlp_type = mlptSingle;
	ASSERT(connp->conn_netstack == tcps->tcps_netstack);
	ASSERT(tcp->tcp_tcps == tcps);

	/*
	 * If the caller has the process-wide flag set, then default to MAC
	 * exempt mode.  This allows read-down to unlabeled hosts.
	 */
	if (getpflags(NET_MAC_AWARE, credp) != 0)
		connp->conn_mac_mode = CONN_MAC_AWARE;

	connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);

	if (issocket) {
		tcp->tcp_issocket = 1;
	}

	connp->conn_rcvbuf = tcps->tcps_recv_hiwat;
	connp->conn_sndbuf = tcps->tcps_xmit_hiwat;
	connp->conn_sndlowat = tcps->tcps_xmit_lowat;
	connp->conn_so_type = SOCK_STREAM;
	connp->conn_wroff = connp->conn_ht_iphc_allocated +
	    tcps->tcps_wroff_xtra;

	SOCK_CONNID_INIT(tcp->tcp_connid);
	tcp->tcp_state = TCPS_IDLE;
	tcp_init_values(tcp);
	return (connp);
}

static int
tcp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
    boolean_t isv6)
{
	tcp_t		*tcp = NULL;
	conn_t		*connp = NULL;
	int		err;
	vmem_t		*minor_arena = NULL;
	dev_t		conn_dev;
	boolean_t	issocket;

	if (q->q_ptr != NULL)
		return (0);

	if (sflag == MODOPEN)
		return (EINVAL);

	if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
	    ((conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
		minor_arena = ip_minor_arena_la;
	} else {
		/*
		 * Either minor numbers in the large arena were exhausted
		 * or a non socket application is doing the open.
		 * Try to allocate from the small arena.
		 */
		if ((conn_dev = inet_minor_alloc(ip_minor_arena_sa)) == 0) {
			return (EBUSY);
		}
		minor_arena = ip_minor_arena_sa;
	}

	ASSERT(minor_arena != NULL);

	*devp = makedevice(getmajor(*devp), (minor_t)conn_dev);

	if (flag & SO_FALLBACK) {
		/*
		 * Non streams socket needs a stream to fallback to
		 */
		RD(q)->q_ptr = (void *)conn_dev;
		WR(q)->q_qinfo = &tcp_fallback_sock_winit;
		WR(q)->q_ptr = (void *)minor_arena;
		qprocson(q);
		return (0);
	} else if (flag & SO_ACCEPTOR) {
		q->q_qinfo = &tcp_acceptor_rinit;
		/*
		 * the conn_dev and minor_arena will be subsequently used by
		 * tcp_tli_accept() and tcp_tpi_close_accept() to figure out
		 * the minor device number for this connection from the q_ptr.
		 */
		RD(q)->q_ptr = (void *)conn_dev;
		WR(q)->q_qinfo = &tcp_acceptor_winit;
		WR(q)->q_ptr = (void *)minor_arena;
		qprocson(q);
		return (0);
	}

	issocket = flag & SO_SOCKSTR;
	connp = tcp_create_common(credp, isv6, issocket, &err);

	if (connp == NULL) {
		inet_minor_free(minor_arena, conn_dev);
		q->q_ptr = WR(q)->q_ptr = NULL;
		return (err);
	}

	connp->conn_rq = q;
	connp->conn_wq = WR(q);
	q->q_ptr = WR(q)->q_ptr = connp;

	connp->conn_dev = conn_dev;
	connp->conn_minor_arena = minor_arena;

	ASSERT(q->q_qinfo == &tcp_rinitv4 || q->q_qinfo == &tcp_rinitv6);
	ASSERT(WR(q)->q_qinfo == &tcp_winit);

	tcp = connp->conn_tcp;

	if (issocket) {
		WR(q)->q_qinfo = &tcp_sock_winit;
	} else {
#ifdef  _ILP32
		tcp->tcp_acceptor_id = (t_uscalar_t)RD(q);
#else
		tcp->tcp_acceptor_id = conn_dev;
#endif  /* _ILP32 */
		tcp_acceptor_hash_insert(tcp->tcp_acceptor_id, tcp);
	}

	/*
	 * Put the ref for TCP. Ref for IP was already put
	 * by ipcl_conn_create. Also Make the conn_t globally
	 * visible to walkers
	 */
	mutex_enter(&