diff -r 000000000000 -r af10295192d8 networkprotocols/tcpipv4v6prt/inc/tcpip_ini.h
--- /dev/null Thu Jan 01 00:00:00 1970 +0000
+++ b/networkprotocols/tcpipv4v6prt/inc/tcpip_ini.h Tue Jan 26 15:23:49 2010 +0200
@@ -0,0 +1,741 @@
+// Copyright (c) 2004-2009 Nokia Corporation and/or its subsidiary(-ies).
+// All rights reserved.
+// This component and the accompanying materials are made available
+// under the terms of "Eclipse Public License v1.0"
+// which accompanies this distribution, and is available
+// at the URL "http://www.eclipse.org/legal/epl-v10.html".
+//
+// Initial Contributors:
+// Nokia Corporation - initial contribution.
+//
+// Contributors:
+//
+// Description:
+// tcpip_ini.h - INI File literals
+//
+
+#ifndef __TCPIP_INI_H__
+#define __TCPIP_INI_H__
+
+/**
+* @file tcpip_ini.h
+* @ingroup tcpip_ini_parameters
+* Define TCPIP.INI configuration parameters
+* @internalComponent
+*/
+
+/**
+* @defgroup tcpip_ini_parameters TCPIP.INI configuration parameters
+*
+* @{
+*/
+
+//
+// INI File literals
+// *****************
+//
+
+/** The name of the ini-file */
+_LIT(TCPIP_INI_DATA, "tcpip.ini");
+
+/**
+* @name Start Section
+*
+* When the protocol stack is loaded into memory, it can automatically start
+* a number of system daemons. The DNS resolver daemon (dnd.exe) is an example
+* of such system damoens. When the stack is unloaded, any running daemons are
+* automatically killed.
+*
+* Each daemon has its own section in the TCPIP.INI file. The daemons that
+* should be started up are listed in the [start] section under the daemons
+* key, separated by commas. Each daemon specific section contains the
+* initialization parameters for the daemon. Currently, the only supported
+* parameter key is filename, which specifies the location of the executable.
+* For example:
+@verbatim
+[start]
+daemons= dnd,httpserver
+
+[dnd]
+filename= z:\system\programs\dnd.exe
+
+[httpserver]
+filename= d:\system\programs\httpserver.exe
+@endverbatim
+* @{
+*/
+/** The section name [start] */
+_LIT(TCPIP_INI_START, "start");
+/** The daemons list. daemons= dnd,kmd */
+_LIT(TCPIP_INI_DAEMONS, "daemons");
+/** The filename daemon. [dnd]
filename= dnd.exe */
+_LIT(TCPIP_INI_FILENAME, "filename");
+/** @} */
+
+/**
+* @name Network layer parameters
+* @{
+*/
+/** The section name [ip] */
+_LIT(TCPIP_INI_IP, "ip"); // [ip]
+/**
+* Default TTL (time to live) value for IP packets.
+*
+* Default: #KTcpipIni_Maxttl
+*/
+_LIT(TCPIP_INI_MAXTTL, "maxttl"); // maxttl= 69
+/**
+* Default TTL for link local destinations.(IPv4 and IPv6).
+* If negative, then default is same as maxttl.
+*
+* - Default: #KTcpipIni_LinkLocalttl
+* - Range: [-1..255]
+*/
+_LIT(TCPIP_INI_LINKLOCALTTL, "linklocalttl");
+/**
+* Maximum number of simultaneous dial attempts via NifMan.
+*
+* - Default: #KTcpipIni_Maxdials
+* - Range: >= 0
+* @deprecated Not used
+* @since 7.0
+*/
+_LIT(TCPIP_INI_MAXDIALS, "maxdials");
+/**
+* Packet forwarding.
+* - Default: #KTcpipIni_Forward
+* - Range: 0 = disabled, > 0 = enabled, and the number of cached flows to use.
+*/
+_LIT(TCPIP_INI_FORWARD, "forward");
+/**
+* Enable IPv4 link-local configuration.
+* When set, the stack will autoconfigure interface for IPv4 linklocal
+* address space (169.254.0.0/16). This happens in addition to
+* any other configuration if present unless option 3 is specified. Possible
+* values and their meanings:
+*
+* Default: #KTcpipIni_Ipv4Linklocal
+*
+* Valid values:
+* - 0= No IPv4 link-local addresses,
+* - 1= Always use IPv4 link-local addresses,
+* - 2= Only use IPv4 link-local addresses when no global address has been configured,
+* - 3= Use link-local address if no IPv4 address is read from Nif or configuration daemon (e.g., DHCP).
+*/
+_LIT(TCPIP_INI_IPV4LINKLOCAL, "ipv4linklocal");
+/**
+* Probe addresses without route from all links.
+* If set to 1, destination addresses for which no route can be found,
+* are probed with ND (ARP for IPv4) on all interfaces within scope.
+* This setting is ignored if disabled if the source address is an IPv4
+* link local or else the link local would not function except for
+* neighbours with routes already discovered and cached. This is
+* necessary for compliance with the ZEROCONF RFC.
+*
+* - Default: #KTcpipIni_ProbeAddress
+* - Range: 0=disabled, 1=enabled
+*/
+_LIT(TCPIP_INI_PROBEADDRESS, "probeaddress");
+/**
+* No id defence.
+*
+* Normally the node tries to keep assigned ID reserved for this node,
+* regardless of the prefix part. Enabling this option, disables this
+* feature (and allows other hosts use my id with different prefixes).
+*
+* - Default: #KTcpipIni_NoDefendId
+* - Range: 0 = disabled (defend id), 1 = enabled (do not defend id).
+*/
+_LIT(TCPIP_INI_NODEFENDID, "nodefendid");
+/**
+* No interface error to attaced flows.
+*
+* This is the default value for new sockets (flows).
+* The default can be changed by a socket option
+* #KSoNoInterfaceError.
+*
+* - Default: #KTcpipIni_Noiferror
+* - Range: 0= pass, 1= don't pass errors to flow
+*/
+_LIT(TCPIP_INI_NOIFERROR, "noiferror");
+/**
+* Max time to wait for route or interface.
+*
+* - Default: #KTcpipIni_Maxholdtime
+* - Range: [0..maxint]
+*/
+_LIT(TCPIP_INI_MAXHOLDTIME, "maxholdtime");
+/**
+* Shutdown delay.
+* How long to wait until to kill daemons after last user (SAP/NIF) exits.
+*
+* - Default: #KTcpipIni_ShutdownDelay seconds.
+* - Range: [0..maxint]
+*/
+_LIT(TCPIP_INI_SHUTDOWN_DELAY, "shutdown_delay");
+/**
+* Keep interface up.
+*
+* This is the default value for new sockets (flows).
+* The default can be changed by a socket option
+* #KSoKeepInterfaceUp.
+*
+* When enabled (1), attaching the flow to the interface
+* increments the flow count on the interface. This
+* count is part of the process that controls how
+* long idle interfaces are kept up.
+*
+* When disabled (0), the flow is not counted.
+*
+* - Default: #KTcpipIni_KeepInterfaceUp
+* - Range: 0 = disabled, 1 = enabled
+*/
+_LIT(TCPIP_INI_KEEP_INTERFACE_UP, "keepinterfaceup");
+/**
+* Max Fragment assemblies.
+*
+* Max number of simultaneous incomplete fragment assembly (currently per
+* protocol version: IPv4 and IPv6). This is a safeguard against denial of
+* service (DOS) attacks. Prevents overusing the heap space for the
+* assembly control blocks.
+*
+* - Default: #KTcpipIni_FragCount
+* - Range: [0..maxint]
+*/
+_LIT(TCPIP_INI_FRAG_COUNT, "frag_count");
+/**
+* Max Fragment buffering.
+*
+* Total amount buffer space that can be allocated (RMBufs) for incomplete
+* assemblies (per protocol version). This is a safeguard against denial of
+* service (DOS) attacks. Prevents overusing the RMBuf pool for the
+* packet content.
+*
+* - Default: #KTcpipIni_FragTotal (bytes)
+* - Range: [0..maxint]
+*/
+_LIT(TCPIP_INI_FRAG_TOTAL, "frag_total");
+/**
+* Destination cache.
+* - Default: 0 (disabled)
+* - Range: 0 = disabled,
+* 1 = enabled / one entry per address,
+* 2 = enabled / one entry per network
+*
+* @see MDestinationCache for more information.
+*/
+_LIT(TCPIP_INI_DSTCACHE, "dstcache");
+/**
+* Lifetime of dest. cache entry.
+*
+* - Default: 600 (secs)
+*/
+_LIT(TCPIP_INI_DST_LIFETIME, "dst_lifetime");
+/**
+* Max total size of dest. cache.
+*
+* - Default: 2048 (bytes)
+*/
+_LIT(TCPIP_INI_DST_MAXSIZE, "dst_maxsize");
+/**
+* Max transmitted ICMP errors or echo replies within ca. 8 second timeslot.
+*
+* - Default #KTcpipIni_IcmpLimit
+*/
+_LIT(TCPIP_INI_ICMP_LIMIT, "icmp_limit");
+/**
+* Route Information option.
+*
+* The type code for RouteInformation has not been defined by the
+* IANA. This option allows enabling the option by specifying some
+* non-zero type, which identifies the option in Router Avertisement
+* message. (Experimental)
+*
+* - Default: 0 (not enabled)
+*/
+_LIT(TCPIP_INI_RA_OPT_ROUTE, "ra_opt_route");
+/**
+* Recursive DNS Server Option (RDNSS)
+*
+* The type code for RouteInformation has not been defined by the
+* IANA. This option allows enabling the option by specifying some
+* non-zero type, which identifies the option in Router Avertisement
+* message. (Experimental)
+*
+* - Default: 0 (not enabled)
+*/
+_LIT(TCPIP_INI_RA_OPT_RDNSS, "ra_opt_rdnss");
+/** @} */
+
+/**
+* @name TCP Parameters
+* @{
+*/
+/** The section name [tcp] */
+_LIT(TCPIP_INI_TCP, "tcp");
+/**
+* TCP maximum segment size.
+* Upper limit is bounded by the network interface MTU.
+*
+* - Default: tcp_mss= 65535
+* - Range: [0..65535]
+*/
+_LIT(TCPIP_INI_TCP_MSS, "tcp_mss");
+/**
+* TCP receive buffer size.
+* - Default: tcp_recv_buf= 8192
+* - Range: [1024..65535]
+*/
+_LIT(TCPIP_INI_TCP_RECV_BUF, "tcp_recv_buf");
+
+#ifdef SYMBIAN_ADAPTIVE_TCP_RECEIVE_WINDOW
+/**
+* TCP receive maximum window size.
+* - Default: tcp_recv_max_buf= 262144
+* - Range: [131070-1073741823]
+*/
+_LIT(TCPIP_INI_TCP_RECV_MAX_WND, "tcp_recv_max_wnd");
+#endif //SYMBIAN_ADAPTIVE_TCP_RECEIVE_WINDOW
+
+/**
+* TCP send buffer size.
+* - Default: tcp_send_buf= 8192
+* - Range: [1024..65535]
+*/
+_LIT(TCPIP_INI_TCP_SEND_BUF, "tcp_send_buf");
+/**
+* Minimum retransmission timeout (ms).
+* - Default: tcp_min_rto= 1000
+* - Range: 1000 ms
+*/
+_LIT(TCPIP_INI_TCP_MIN_RTO, "tcp_min_rto");
+/**
+* Maximum retransmission timeout (ms).
+* - Default: tcp_max_rto= 60000
+* - Range: 60000 ms
+*/
+_LIT(TCPIP_INI_TCP_MAX_RTO, "tcp_max_rto");
+/**
+* Initial retransmission timeout (ms).
+* - Default: tcp_initial_rto= 3000
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_INITIAL_RTO, "tcp_initial_rto");
+/**
+* SRTT smoothing.
+*
+* alpha = 1 / tcp_srtt_smooth, as defined in RFC2988.
+* - Default: tcp_srtt_smooth= 8
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_SRTT_SMOOTH, "tcp_srtt_smooth");
+/**
+* MDEV smoothing.
+* beta = 1 / tcp_rttvar_smooth, as defined in RFC2988.
+* - Default: tcp_rttvar_smooth= 4
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_RTTVAR_SMOOTH, "tcp_rttvar_smooth");
+/**
+* G value in RTO calculation (RFC2988).
+* Constant representing clock granularity.
+* - Default: machine specific.
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_RTO_G, "tcp_rto_g");
+/**
+* K value in RTO calculation (RFC2988).
+* K constant, RTO = SRTT + max(K * RTTVAR, G) (RFC2988)
+* - Default: tcp_rto_k= 4
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_RTO_K, "tcp_rto_k");
+/**
+* Maximum retransmit burst size
+* - Default: tcp_max_burst= 2
+* - Range: [1..]
+*/
+_LIT(TCPIP_INI_TCP_MAX_BURST, "tcp_max_burst");
+/**
+* Delayed acknowledgement delay time (ms).
+* - Default: tcp_ack_delay= 200
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_ACK_DELAY, "tcp_ack_delay");
+/**
+* Number of retransmissions per connect() attempt.
+* - Default: tcp_syn_retries= 5
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_SYN_RETRIES, "tcp_syn_retries");
+/**
+* Number of retransmissions before black hole diagnostics.
+* Maximum retries before MTU reduction.
+* - Default: tcp_retries1= 3
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_RETRIES1, "tcp_retries1");
+/**
+* Maximum number of retransmissions.
+* - Default: tcp_retries2= 12
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_RETRIES2, "tcp_retries2");
+/**
+* MSL2 timeout (ms).
+* TIME-WAIT & FIN-WAIT timeout.
+* - Default: tcp_msl2= 60000
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_MSL2, "tcp_msl2");
+/**
+* Large initial window (RFC2414).
+* Set initial cwnd according to RFC2414
+* - Default: .tcp_rfc2414= 1
+* - Range: 1=enable, 0=disable.
+*/
+_LIT(TCPIP_INI_TCP_RFC2414, "tcp_rfc2414");
+/**
+* Initial congestion window size (segments) when RFC2414 is disabled.
+* - Default: tcp_initial_cwnd= 2
+* - Range: [1..]
+*/
+_LIT(TCPIP_INI_TCP_INITIAL_CWND, "tcp_initial_cwnd");
+/**
+* TCP timestamps (RFC1323).
+* - Default: tcp_timestamps= 1
+* - Range: 1=enable, 0=disable
+*/
+_LIT(TCPIP_INI_TCP_TIMESTAMPS, "tcp_timestamps");
+/**
+* Enable selective acknowledgements (RFC2018).
+* - Default: tcp_sack= 1
+* - Range: 1=enable, 0=disable
+*/
+_LIT(TCPIP_INI_TCP_SACK, "tcp_sack");
+/**
+* Apply TIME-WAIT state also to local connections.
+* - Default: tcp_local_timewait= 0
+* - Range: 1=enable, 0=disable
+*/
+_LIT(TCPIP_INI_TCP_LOCAL_TIMEWAIT, "tcp_local_timewait");
+/**
+* Typical degree of packet reordering in the network.
+* - Default: tcp_reordering= 3
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_REORDERING, "tcp_reordering");
+/**
+* Variant of the nagle algorithm.
+* - Default: tcp_strict_nagle= 0
+* - Range: 0 = modern, 1 = strict
+*/
+_LIT(TCPIP_INI_TCP_STRICT_NAGLE, "tcp_strict_nagle");
+/**
+* Limited transmit window size in segments (RFC3042).
+* - Default: tcp_ltx_window= 2
+* - Range: 0=disable, [1..]
+*/
+_LIT(TCPIP_INI_TCP_LTX_WINDOW, "tcp_ltx_window");
+/**
+* Zero window probe style.
+* - Default: tcp_probe_style= 0
+* - Range:
+* - 0 = probe with a single byte (standard)
+* - 1 = probe with a full segment (BSD)
+* - 2 = probe with an out-of-window acknowledgement (experimental)
+*/
+_LIT(TCPIP_INI_TCP_PROBE_STYLE, "tcp_probe_style");
+/**
+* Acknowledge pushed segments immediately.
+*
+* - Default: ?
+* - Range: ?
+*/
+_LIT(TCPIP_INI_TCP_PUSH_ACK, "tcp_push_ack");
+/**
+* F-RTO, also disables timestamp spike detection.
+*
+* If enabled, the TCP sender tries to probe with new previously
+* unsent data segments after an RTO whether the RTO was spurious.
+* If the RTO is detected spurious, the sender continues by sending
+* new data instead of retransmitting.
+* - Default: tcp_frto= 0
+* - Range: 1 = enable, 0 = disable
+*/
+_LIT(TCPIP_INI_TCP_FRTO, "tcp_frto");
+/**
+* Send DSACKs for out-of-order segments.
+* If enabled, send DSACKs in acknowledgements when receiving
+* duplicate segments (RFC 2883). The data sender behaviour is
+* not modified.
+* - Default: tcp_dsack= 1
+* - Range: 1 = enable, 0 = disable
+*/
+_LIT(TCPIP_INI_TCP_DSACK, "tcp_dsack");
+/**
+* Interval between keepalive probes.
+* Delay for sending a TCP Keep-Alive probe after the connection
+* has become idle (seconds). This is also the interval between
+* succesful, replied keepalive probes.
+* - Default: tcp_keepalive_intv= 7200 (= 2h)
+* - Range: [0..]
+*/
+_LIT(TCPIP_INI_TCP_KEEPALIVE_INTV, "tcp_keepalive_intv");
+/**
+* Number of unreplied probes before quitting.
+* Number of unreplied TCP Keep-Alive probes to send
+* before the sender gives up and terminates the connection.
+* - Default: tcp_num_keepalives= 8
+* - Range: [1..]
+*/
+_LIT(TCPIP_INI_TCP_NUM_KEEPALIVES, "tcp_num_keepalives");
+/**
+* Interval between keepalive rexmits
+* Interval between consecutive retransmissions of a Keep-Alive
+* probe if there has been no response from the other end (seconds).
+* - Default: tcp_keepalive_rxmt= 75
+* - Range: [0..1800]
+*/
+_LIT(TCPIP_INI_TCP_KEEPALIVE_RXMT, "tcp_keepalive_rxmt");
+
+/**
+* Number of FIN retransmissions before socket is excluded from route flow count.
+* After the FIN persistency limit is exceeded, link layer and TCP/IP stack is given permission
+* to go down if no one else uses them.
+* Value 0 means that socket is included in the interface flow count until it is really closed,
+* i.e. in TIME_WAIT state.
+* - Default: tcp_fin_persistency= 2
+* - Range: [0..tcp_retries2]
+*/
+_LIT(TCPIP_INI_TCP_FIN_PERSISTENCY, "tcp_fin_persistency");
+
+/**
+Determines how to set congestion control parameters after spurious RTO.
+1 = Eifel response (revert ssthresh and cwnd).
+2 = Half congestion window and ssthresh.
+3 = Set cwnd to 1 but revert ssthresh (Early versions of DCLOR).
+
+- Default: tcp_spurious_rto_response= 1
+- Range: [1..3]
+*/
+_LIT(TCPIP_INI_TCP_SPURIOUS_RTO_RESPONSE, "tcp_spurious_rto_response");
+
+/**
+* ECN.
+* Toggles Explicit Congestion Notification [RFC3168] on and off.
+* Two alternatives of ECN are available: when parameter value is 1
+* the host uses ECN capable bit ECT(1). When value is 2, the host
+* uses ECN capable bit ECT(0). Of these two, value 2 is recommended,
+* because some older network nodes may not recognize ECT(1) bit.
+* - Default: tcp_ecn= 0
+* - Range: 0= disabled, 1 & 2 = enabled)
+*/
+_LIT(TCPIP_INI_TCP_ECN, "tcp_ecn");
+
+/**
+TCP Option alignment toggle.
+If set to 1, TCP options are aligned on 32-bit boundaries using the NOP option.
+- Default: tcp_alignopt= 0
+- Range: 0, 1
+*/
+_LIT(TCPIP_INI_TCP_ALIGNOPT, "tcp_alignopt");
+
+/**
+* TCP Window scaling option.
+* Controls TCP window scaling option [RFC 1323]. -1 means window scaling
+* is disabled. 0 means that window scaling capability is advertised in TCP
+* handshake and the advertised window scale factor is determined based on
+* the available receive window. Values 1 to 6 mean that window scaling is
+* advertised on TCP handshake, but the given scale factor is always used
+* in advertisement regardless of the receive window size.
+*
+* - Default: tcp_winscale= 0
+* - Range: [-1..6]
+*/
+_LIT(TCPIP_INI_TCP_WINSCALE, "tcp_winscale");
+/** @} */
+
+/**
+* @name UDP parameters
+* @{
+*/
+/** The section name [udp]. */
+_LIT(TCPIP_INI_UDP, "udp");
+/**
+* Causes UDP send to block on interface setup.
+* Block UDP send while waiting for network interface setup.
+* - Default: udp_wait_nif= 1
+* - Range: 1 = enable, 0=disable (1 is conformant with old Symbian IPv4 stack)
+*/
+_LIT(TCPIP_INI_UDP_WAIT_NIF, "udp_wait_nif");
+/**
+* Maximum inbound bytes queued by UDP.
+* UDP receive queue size (octets, including header). Received packets will be
+* discarded if receive queue is full. The queue will always hold at least one
+* packet, regardles of packet size.
+* - Default: udp_recv_buf= 8192
+* - Range: [0..655535]
+*/
+_LIT(TCPIP_INI_UDP_RECV_BUF, "udp_recv_buf");
+/** @} */
+
+/**
+* @name Generic network group options.
+* For sections [ip6], [ip], [icmp] and [icmp6].
+* The section name is the protocol name..
+* @{
+*/
+/**
+* Maximum inbound bytes queued.
+*
+* Maximum inbound bytes queued by ip6, ip, icmp or icmp6
+* socket (SAP).
+*
+* The test at arrival of new packet goes as follows:
+* - if already queued packets exceed the limit, packet is dropped
+* - if already queued packets do not exceed the limit, packet is
+* queued (even if doing so puts the queue over the limit)
+*
+* From above, some useful special cases follow:
+*
+* - Zero limits queue to single packet at time. Incoming
+* packets are dropped until application reads the queued
+* packet.
+*
+* - Negative value, nothing is queued and no data can be
+* received via this socket (SAP).
+*
+* Default value: 8192
+*/
+_LIT(TCPIP_INI_RECV_BUF, "recv_buf");
+/** @} */
+
+/**
+* @name Hostname section
+*
+* @{
+*/
+/** The section name [host] */
+_LIT(TCPIP_INI_HOST, "host");
+/**
+* The host name.
+* The default value returned by RHostResolver::GetHostName(),
+* if no SetHostName() has been issued.
+*
+* - Default: hostname= localhost.
+*/
+_LIT(TCPIP_INI_HOSTNAME, "hostname");
+/** @} */
+
+/**
+* @name Hook ordering options
+*
+* The ordering of outbound flowhooks, outbound and inbound posthooks
+* '*' indicates "position" of all hooks that don't match the specific
+* listed names.
+*
+* @note
+* The options are only examined when a protocol binds to the stack.
+* The control affects the bind only if the name matches exactly
+* (case sensitive). The options can include protocol names that are
+* not installed in the system (there is no harm, as it doesn't match
+* any protocol).
+* @{
+*/
+/** The section name [hook] */
+_LIT(TCPIP_INI_HOOK, "hook");
+/**
+* Controls ordering of the outbound flow hooks.
+* For example, hook_flow= ipsec,mip6,*,foobar says that ipsec
+* hook will be run first, then mip6, and foobar will be the last.
+* All other hooks (matched by "*" are run between mip6 and foobar.
+* If the "*" is missing, it is implicitly assumed to be at the end of the string.
+*/
+_LIT(TCPIP_INI_HOOK_FLOW, "hook_flow");
+/**
+* Controls ordering of the inbound protocol hooks.
+* For example, hook_any= ipsec,*,foobar says that ipsec
+* hook will be run first, and foobar will be the last. This can
+* be useful, if foobar is a protocol that wants to handle only
+* packets that have been verified by the IPSEC.
+*/
+_LIT(TCPIP_INI_HOOK_INANY, "hook_inany");
+/**
+* Controls the ordering of outbound post-processing hooks.
+* For example, hook_outbound= *,probe,qos says that last two
+* post-processors before NIF are probe and qos in this order.
+* Any other post-processor will be run before them. If the "*" is missing,
+* it is implicitly assumed to be at the end of the string.
+*/
+_LIT(TCPIP_INI_HOOK_OUTBOUND, "hook_outbound");
+/**
+* Controls the ordering of inbound pre-processing hooks.
+* For example, hook_inbound= probe says that packet from the NIF
+* is given first to probe, and to other possible pre-processors after it.
+* If the "*" is missing, it is implicitly assumed to be at the end of the string.
+*/
+_LIT(TCPIP_INI_HOOK_INBOUND, "hook_inbound");
+/**
+* Controls the ordering of packet forwarding hooks.
+* Usage is similar to the above examples.
+*/
+_LIT(TCPIP_INI_HOOK_FORWARD, "hook_forward");
+/** @} */
+
+/**
+* @name Compile time default values
+*
+* Many TCPIP.INI file parameters will have a hard coded default,
+* if the value is not specified in the ini file. Some of the
+* default constants are defined below.
+*
+* The name of the constant is generated directly from the string
+* literal name according to the following template
+*
+* TCPIP_INI_PARAMETER_NAME => KTcpipIni_ParameterName
+*
+* @{
+*/
+/** frag_count= 10 */
+const TInt KTcpipIni_FragCount = 10; // [0..maxint]
+/** frag_total= 30000 */
+const TInt KTcpipIni_FragTotal = 30000; // [0..maxint]
+/** maxttl= 69 */
+const TInt KTcpipIni_Maxttl = 69; // [1..255]
+/** linklocalttl= 1 */
+const TInt KTcpipIni_LinkLocalttl = 1; // [-1,,255]
+/** maxdials= 1 */
+const TInt KTcpipIni_Maxdials = 1; // [0..maxint]
+/** forward= 0 */
+const TInt KTcpipIni_Forward = 0; // [0..1]
+/** ipv4linklocal= 0 */
+const TInt KTcpipIni_Ipv4Linklocal = 0; // [0..1]
+/** probeaddress= 0 */
+const TInt KTcpipIni_ProbeAddress = 0; // [0..1]
+/** nodefendid= 0 */
+const TInt KTcpipIni_NoDefendId = 0; // [0..1]
+/** noiferror= 0 */
+const TInt KTcpipIni_Noiferror = 0; // [0..1]
+/** maxholdtime= 60 */
+const TInt KTcpipIni_Maxholdtime = 60; // [0..maxint]
+/** icmp_limit= 10 */
+const TInt KTcpipIni_IcmpLimit = 30; // [0..maxint]
+
+/** timeoutpriority= 12.
+* Hardcoded for now. No tcpip.ini parameter yet. This
+* defines the CActive priority used by the interface
+* manager for it's internal CTimeoutManager.
+*/
+const TInt KTcpipIni_TimeoutPriority = 12;
+/** keepinterfaceup= 1 */
+const TInt KTcpipIni_KeepInterfaceUp = 1; // [0..1]
+/** shutdowndelay= 10 */
+const TInt KTcpipIni_ShutdownDelay = 10; // [0..maxint] (0=no delay)
+/** hostname= localhost */
+_LIT(KTcpipIni_Hostname, "localhost"); // default host name
+/** @} */
+
+/** @} */
+
+#endif