/*

 * 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 (c) 2010, Oracle and/or its affiliates. All rights reserved.

 */

#include <sys/types.h>

#include <sys/stream.h>

#define	_SUN_TPI_VERSION 2

#include <sys/tihdr.h>

#include <sys/socket.h>

#include <sys/xti_xtiopt.h>

#include <sys/xti_inet.h>

#include <sys/policy.h>

#include <inet/common.h>

#include <netinet/ip6.h>

#include <inet/ip.h>

#include <netinet/in.h>

#include <netinet/tcp.h>

#include <inet/optcom.h>

#include <inet/proto_set.h>

#include <inet/tcp_impl.h>

/*

 * Table of all known options handled on a TCP protocol stack.

 *

 * Note: This table contains options processed by both TCP and IP levels

 *       and is the superset of options that can be performed on a TCP over IP

 *       stack.

 */

opdes_t	tcp_opt_arr[] = {

{ SO_LINGER,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0,

	sizeof (struct linger), 0 },

{ SO_DEBUG,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_KEEPALIVE,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_DONTROUTE,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_USELOOPBACK, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0

	},

{ SO_BROADCAST,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_REUSEADDR, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_OOBINLINE, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_TYPE,	SOL_SOCKET, OA_R, OA_R, OP_NP, 0, sizeof (int), 0 },

{ SO_SNDBUF,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_RCVBUF,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_SNDTIMEO,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0,

	sizeof (struct timeval), 0 },

{ SO_RCVTIMEO,	SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0,

	sizeof (struct timeval), 0 },

{ SO_DGRAM_ERRIND, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0

	},

{ SO_SND_COPYAVOID, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_ANON_MLP, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int),

	0 },

{ SO_MAC_EXEMPT, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int),

	0 },

{ SO_MAC_IMPLICIT, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int),

	0 },

{ SO_ALLZONES, SOL_SOCKET, OA_R, OA_RW, OP_CONFIG, 0, sizeof (int),

	0 },

{ SO_EXCLBIND, SOL_SOCKET, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ SO_DOMAIN,	SOL_SOCKET, OA_R, OA_R, OP_NP, 0, sizeof (int), 0 },

{ SO_PROTOTYPE,	SOL_SOCKET, OA_R, OA_R, OP_NP, 0, sizeof (int), 0 },

{ TCP_NODELAY,	IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0

	},

{ TCP_MAXSEG,	IPPROTO_TCP, OA_R, OA_R, OP_NP, 0, sizeof (uint_t),

	536 },

{ TCP_NOTIFY_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP,

	OP_DEF_FN, sizeof (int), -1 /* not initialized */ },

{ TCP_ABORT_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP,

	OP_DEF_FN, sizeof (int), -1 /* not initialized */ },

{ TCP_CONN_NOTIFY_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP,

	OP_DEF_FN, sizeof (int), -1 /* not initialized */ },

{ TCP_CONN_ABORT_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP,

	OP_DEF_FN, sizeof (int), -1 /* not initialized */ },

{ TCP_RECVDSTADDR, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int),

	0 },

{ TCP_ANONPRIVBIND, IPPROTO_TCP, OA_R, OA_RW, OP_PRIVPORT, 0,

	sizeof (int), 0 },

{ TCP_EXCLBIND, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0

	},

{ TCP_INIT_CWND, IPPROTO_TCP, OA_RW, OA_RW, OP_CONFIG, 0,

	sizeof (int), 0 },

{ TCP_KEEPALIVE_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0	},

{ TCP_KEEPALIVE_ABORT_THRESHOLD, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0	},

{ TCP_CORK, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ TCP_RTO_INITIAL, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (uint32_t), 0 },

{ TCP_RTO_MIN, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (uint32_t), 0 },

{ TCP_RTO_MAX, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (uint32_t), 0 },

{ TCP_LINGER2, IPPROTO_TCP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ IP_OPTIONS,	IPPROTO_IP, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT),

	IP_MAX_OPT_LENGTH + IP_ADDR_LEN, -1 /* not initialized */ },

{ T_IP_OPTIONS,	IPPROTO_IP, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT),

	IP_MAX_OPT_LENGTH + IP_ADDR_LEN, -1 /* not initialized */ },

{ IP_TOS,	IPPROTO_IP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ T_IP_TOS,	IPPROTO_IP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ IP_TTL,	IPPROTO_IP, OA_RW, OA_RW, OP_NP, OP_DEF_FN,

	sizeof (int), -1 /* not initialized */ },

{ IP_SEC_OPT, IPPROTO_IP, OA_RW, OA_RW, OP_NP, OP_NODEFAULT,

	sizeof (ipsec_req_t), -1 /* not initialized */ },

{ IP_BOUND_IF, IPPROTO_IP, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int),	0 /* no ifindex */ },

{ IP_UNSPEC_SRC, IPPROTO_IP, OA_R, OA_RW, OP_RAW, 0,

	sizeof (int), 0 },

{ IPV6_UNICAST_HOPS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, OP_DEF_FN,

	sizeof (int), -1 /* not initialized */ },

{ IPV6_BOUND_IF, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int),	0 /* no ifindex */ },

{ IP_DONTFRAG, IPPROTO_IP, OA_RW, OA_RW, OP_NP, 0, sizeof (int), 0 },

{ IP_NEXTHOP, IPPROTO_IP, OA_R, OA_RW, OP_CONFIG, 0,

	sizeof (in_addr_t),	-1 /* not initialized  */ },

{ IPV6_UNSPEC_SRC, IPPROTO_IPV6, OA_R, OA_RW, OP_RAW, 0,

	sizeof (int), 0 },

{ IPV6_PKTINFO, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	(OP_NODEFAULT|OP_VARLEN),

	sizeof (struct in6_pktinfo), -1 /* not initialized */ },

{ IPV6_NEXTHOP, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	OP_NODEFAULT,

	sizeof (sin6_t), -1 /* not initialized */ },

{ IPV6_HOPOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT), 255*8,

	-1 /* not initialized */ },

{ IPV6_DSTOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT), 255*8,

	-1 /* not initialized */ },

{ IPV6_RTHDRDSTOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT), 255*8,

	-1 /* not initialized */ },

{ IPV6_RTHDR, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	(OP_VARLEN|OP_NODEFAULT), 255*8,

	-1 /* not initialized */ },

{ IPV6_TCLASS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	OP_NODEFAULT,

	sizeof (int), -1 /* not initialized */ },

{ IPV6_PATHMTU, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP,

	OP_NODEFAULT,

	sizeof (struct ip6_mtuinfo), -1 /* not initialized */ },

{ IPV6_DONTFRAG, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_USE_MIN_MTU, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_V6ONLY, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

/* Enable receipt of ancillary data */

{ IPV6_RECVPKTINFO, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVHOPLIMIT, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVHOPOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ _OLD_IPV6_RECVDSTOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVDSTOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVRTHDR, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVRTHDRDSTOPTS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_RECVTCLASS, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (int), 0 },

{ IPV6_SEC_OPT, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, OP_NODEFAULT,

	sizeof (ipsec_req_t), -1 /* not initialized */ },

{ IPV6_SRC_PREFERENCES, IPPROTO_IPV6, OA_RW, OA_RW, OP_NP, 0,

	sizeof (uint32_t), IPV6_PREFER_SRC_DEFAULT },

};

/*

 * Table of all supported levels

 * Note: Some levels (e.g. XTI_GENERIC) may be valid but may not have

 * any supported options so we need this info separately.

 *

 * This is needed only for topmost tpi providers and is used only by

 * XTI interfaces.

 */

optlevel_t	tcp_valid_levels_arr[] = {

	XTI_GENERIC,

	SOL_SOCKET,

	IPPROTO_TCP,

	IPPROTO_IP,

	IPPROTO_IPV6

};

#define	TCP_OPT_ARR_CNT		A_CNT(tcp_opt_arr)

#define	TCP_VALID_LEVELS_CNT	A_CNT(tcp_valid_levels_arr)

uint_t tcp_max_optsize; /* initialized when TCP driver is loaded */

/*

 * Initialize option database object for TCP

 *

 * This object represents database of options to search passed to

 * {sock,tpi}optcom_req() interface routine to take care of option

 * management and associated methods.

 */

optdb_obj_t tcp_opt_obj = {

	tcp_opt_default,	/* TCP default value function pointer */

	tcp_tpi_opt_get,	/* TCP get function pointer */

	tcp_tpi_opt_set,	/* TCP set function pointer */

	TCP_OPT_ARR_CNT,	/* TCP option database count of entries */

	tcp_opt_arr,		/* TCP option database */

	TCP_VALID_LEVELS_CNT,	/* TCP valid level count of entries */

	tcp_valid_levels_arr	/* TCP valid level array */

};

/* Maximum TCP initial cwin (start/restart). */

#define	TCP_MAX_INIT_CWND	16

static int tcp_max_init_cwnd = TCP_MAX_INIT_CWND;

/*

 * Some TCP options can be "set" by requesting them in the option

 * buffer. This is needed for XTI feature test though we do not

 * allow it in general. We interpret that this mechanism is more

 * applicable to OSI protocols and need not be allowed in general.

 * This routine filters out options for which it is not allowed (most)

 * and lets through those (few) for which it is. [ The XTI interface

 * test suite specifics will imply that any XTI_GENERIC level XTI_* if

 * ever implemented will have to be allowed here ].

 */

static boolean_t

tcp_allow_connopt_set(int level, int name)

{

	switch (level) {

	case IPPROTO_TCP:

		switch (name) {

		case TCP_NODELAY:

			return (B_TRUE);

		default:

			return (B_FALSE);

		}

		/*NOTREACHED*/

	default:

		return (B_FALSE);

	}

	/*NOTREACHED*/

}

/*

 * This routine gets default values of certain options whose default

 * values are maintained by protocol specific code

 */

/* ARGSUSED */

int

tcp_opt_default(queue_t *q, int level, int name, uchar_t *ptr)

{

	int32_t	*i1 = (int32_t *)ptr;

	tcp_stack_t	*tcps = Q_TO_TCP(q)->tcp_tcps;

	switch (level) {

	case IPPROTO_TCP:

		switch (name) {

		case TCP_NOTIFY_THRESHOLD:

			*i1 = tcps->tcps_ip_notify_interval;

			break;

		case TCP_ABORT_THRESHOLD:

			*i1 = tcps->tcps_ip_abort_interval;

			break;

		case TCP_CONN_NOTIFY_THRESHOLD:

			*i1 = tcps->tcps_ip_notify_cinterval;

			break;

		case TCP_CONN_ABORT_THRESHOLD:

			*i1 = tcps->tcps_ip_abort_cinterval;

			break;

		default:

			return (-1);

		}

		break;

	case IPPROTO_IP:

		switch (name) {

		case IP_TTL:

			*i1 = tcps->tcps_ipv4_ttl;

			break;

		default:

			return (-1);

		}

		break;

	case IPPROTO_IPV6:

		switch (name) {

		case IPV6_UNICAST_HOPS:

			*i1 = tcps->tcps_ipv6_hoplimit;

			break;

		default:

			return (-1);

		}

		break;

	default:

		return (-1);

	}

	return (sizeof (int));

}

/*

 * TCP routine to get the values of options.

 */

int

tcp_opt_get(conn_t *connp, int level, int name, uchar_t *ptr)

{

	int		*i1 = (int *)ptr;

	tcp_t		*tcp = connp->conn_tcp;

	conn_opt_arg_t	coas;

	int		retval;

	coas.coa_connp = connp;

	coas.coa_ixa = connp->conn_ixa;

	coas.coa_ipp = &connp->conn_xmit_ipp;

	coas.coa_ancillary = B_FALSE;

	coas.coa_changed = 0;

	switch (level) {

	case SOL_SOCKET:

		switch (name) {

		case SO_SND_COPYAVOID:

			*i1 = tcp->tcp_snd_zcopy_on ?

			    SO_SND_COPYAVOID : 0;

			return (sizeof (int));

		case SO_ACCEPTCONN:

			*i1 = (tcp->tcp_state == TCPS_LISTEN);

			return (sizeof (int));

		}

		break;

	case IPPROTO_TCP:

		switch (name) {

		case TCP_NODELAY:

			*i1 = (tcp->tcp_naglim == 1) ? TCP_NODELAY : 0;

			return (sizeof (int));

		case TCP_MAXSEG:

			*i1 = tcp->tcp_mss;

			return (sizeof (int));

		case TCP_NOTIFY_THRESHOLD:

			*i1 = (int)tcp->tcp_first_timer_threshold;

			return (sizeof (int));

		case TCP_ABORT_THRESHOLD:

			*i1 = tcp->tcp_second_timer_threshold;

			return (sizeof (int));

		case TCP_CONN_NOTIFY_THRESHOLD:

			*i1 = tcp->tcp_first_ctimer_threshold;

			return (sizeof (int));

		case TCP_CONN_ABORT_THRESHOLD:

			*i1 = tcp->tcp_second_ctimer_threshold;

			return (sizeof (int));

		case TCP_INIT_CWND:

			*i1 = tcp->tcp_init_cwnd;

			return (sizeof (int));

		case TCP_KEEPALIVE_THRESHOLD:

			*i1 = tcp->tcp_ka_interval;

			return (sizeof (int));

		case TCP_KEEPALIVE_ABORT_THRESHOLD:

			*i1 = tcp->tcp_ka_abort_thres;

			return (sizeof (int));

		case TCP_CORK:

			*i1 = tcp->tcp_cork;

			return (sizeof (int));

		case TCP_RTO_INITIAL:

			*i1 = tcp->tcp_rto_initial;

			return (sizeof (uint32_t));

		case TCP_RTO_MIN:

			*i1 = tcp->tcp_rto_min;

			return (sizeof (uint32_t));

		case TCP_RTO_MAX:

			*i1 = tcp->tcp_rto_max;

			return (sizeof (uint32_t));

		case TCP_LINGER2:

			*i1 = tcp->tcp_fin_wait_2_flush_interval / SECONDS;

			return (sizeof (int));

		}

		break;

	case IPPROTO_IP:

		if (connp->conn_family != AF_INET)

			return (-1);

		switch (name) {

		case IP_OPTIONS:

		case T_IP_OPTIONS:

			/* Caller ensures enough space */

			return (ip_opt_get_user(connp, ptr));

		default:

			break;

		}

		break;

	case IPPROTO_IPV6:

		/*

		 * IPPROTO_IPV6 options are only supported for sockets

		 * that are using IPv6 on the wire.

		 */

		if (connp->conn_ipversion != IPV6_VERSION) {

			return (-1);

		}

		switch (name) {

		case IPV6_PATHMTU:

			if (tcp->tcp_state < TCPS_ESTABLISHED)

				return (-1);

			break;

		}

		break;

	}

	mutex_enter(&connp->conn_lock);

	retval = conn_opt_get(&coas, level, name, ptr);

	mutex_exit(&connp->conn_lock);

	return (retval);

}

/*

 * We declare as 'int' rather than 'void' to satisfy pfi_t arg requirements.

 * Parameters are assumed to be verified by the caller.

 */

/* ARGSUSED */

int

tcp_opt_set(conn_t *connp, uint_t optset_context, int level, int name,

    uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,

    void *thisdg_attrs, cred_t *cr)

{

	tcp_t	*tcp = connp->conn_tcp;