ESOCK E32 Sockets server

8. Nov 2003.

============

¨ ESock implements the functionality required by a CPM, e.g. meeting rendezvous at 

  start-up and supporting the messages required by the framework (see R0). Also

  it doesnt shut down automatically but rather on request from the RootServer.

¨ ESock now supports MBufs for data transfer to protocol SAP implementations 

  over the PRT interface (CServProviderBase). The descriptor based Write() and 

  GetData() functions are deprecated.

¨ ESock now understands the BindFrom parameter in .ESK files. Allows a protocol 

  configuration to specify its binding below another protocol. This avoids the need 

  for it appearing in the upper protocol's configuration file, making such protocols 

  more modular. 

¨ New feature: Datagram Continuation, the ability to read a datagram in parts. With 

  PRT v1 when an application reads a datagram using RSocket::recv() ESock calls GetData() 

  upon the SAP for the exact amount of data requested by the client. If the datagram is 

  bigger than the requested amount the PRT will truncate it and return the  requested amount 

  of octets. It is up to the client to know whether the data was truncated. This leads to some 

  clients having to maintain absurdly large transfer buffers to cope with out-lier cases. 

  With PRTs implementing PRT1.5, ESock calls GetData() with a carefully oversized length 

  parameter, with the intent that the SAP delivers the entire datagram RMBufChain. A client 

  request to read the datagram which includes a TSockXfrLength parameter, eg  

  RSocket::Recv(TDes8& aDesc,TUint flags,TRequestStatus& aStatus,TSockXfrLength& aLen), 

  then completes with the remaining unread octets of the datagram in the TSockXfrLength. 

  The client can then read the remaining octets by further reads with the  'KSockReadContinuation' 

  flag OR'd into the flags field. Remaining octets are discarded upon the next read without the 

  KSockReadContinuation flag set.

  Notes:

    (1) This only works with PRT1.5 protocols.

    (2) Clients who do not set the KSockReadContinuation flag experience the original behaviour that overflows are discarded

    (3) Clients who need to treat datagrams as a "stream" can always set KSockReadContinuation as in the absence of remaining data a new GetData is performed.

¨ The MBuf Manager has been moved from ESock/MBufMgr to MBufMgr. All classes previously living 

  in the NIFMAN component, implementing extra functionality for MBufs have been moved to the 

  new MBufMgr component. Components needing this should link to the mbufmgr.dll library.

¨ ESock has been split into a client DLL (esock.dll) and a server DLL (esocksvr.dll).

¨ ESock is now using the Secure IPC framework. E.g. CServer2, RMessage2 and TIpcArgs.

29. Nov 2001 (alfredh)

======================

Defect fixes:

KUN-54UMKA "CSocket::Ioctl() panics when calling second Ioctl if first one leaves"

Problem fixed in ssock/ss_sock.cpp - now setting flag 'ioctl block' *after* a

successful memory allocation. This defect was found using SmsStack reg tests.

Verified on WINS and ARM4 (assabet board)

Also increased the stacksize for tcesock.exe to 0x10000 so I could run on assabet.

Version 2.1 (651)

==============

Merged in the changes to the test codes the projects now output logs 

	- TCESOCK  logs to \epoc32\wins\c\esoctest.txt under wins and on c:\esoctest.txt under thumb

	- TIPC logs to \epoc32\wins\c\ipctest.txt under wins and on c:\ipctest.txt under thumb

	- TSHUT logs to \epoc32\wins\c\tshutest.txt under wins and on c:\tshutest.txt under thumb

Version 2.0 (650)

=================

First release onto mainline share. Identical to version 555 on ER5u share.

Version 1.0 (555)

=================

- Previous fix for (EDNPGAL-4B4GEJ in ER5u) was broken and removed in ESOCK 554. The

correct fix has now been added to this release (ESOCK 555)

Version 1.0 (554)

=================

- defect fix for EDNDHAR-4BXLU6 (Web hits a SocketServer USER 47 Panic on receiving data)

in ER5u Defects. Involved removing the so-called defect fix that was put in ESOCK 552, for

"Large TCPIP packets seem to be lost".

Version 1.0 (553)

=================

-EPOC Connect was broken by 552 by the omission of the construction of a hard-coded 

ESOCK.INI for the WINC build. This has been rectified.

Version 1.0 (552)

=================

- Minor code changes to eliminate MSVC6 warnings. 

- defect fix for "Large TCPIP packets Seem to be Lost" (EDNPGAL-4B4GEJ in ER5u).

	This fix works for WINS but does not fix the defect on ARM. This is still under

	investigation.

- defect fix for "Need to remove all references to EPOC registry" - (EDNDMAY-49CE4E in ER5u

defects). Removed sections of code which used it (all in WINC build).

- defect fix for "Need to remove calls to deprecated functions" (EDNDMAY-49CDVT in ER5u defects).

  Removed two lines which call RThread::SetNotifyIfDie() and RThread::SetMark().

Version 1.0 (550)

=================

- Split Nifman from Esock.

- Removed all narrow-specific stuff from MNT.CMD and EBLD.BAT.

- Removed narrow def files.

- Did some general tidying up of header files.

Version 1.0(510)

================

- Fixed (another) defect in Nifman code for restarting the thread

  (ER5 Defect Database EDNKPIA-465MLC)

Version 1.0(509)

================

- Changed the Nifman code for restarting the thread

Version 1.0(508)

================

- Re systemised the ESock thread

Version 1.0(507)

================

- Added EBLD.BAT

- Change in CS_CLI.CPP RSocket::GetOpt

Version 1.0(506)

================

- Boilerplated source files

- Changed esock to run as non-system thread

- edit mnt to use evalid.bat for validation

Version 1.0(505)

================

-defect fix

Transfer socket failed with an socket openned in passive mode

defect reference: SW1-322

Version 1.0(504)

================

- Fixed a defect which meant that debug protocols would not run

with a release ESOCK.

Version 1.0(503)

================

- debug/Rel stuff

- Unicode UID, added a ROM directory

Version 1.0(502)

================

- VC5

Version 1.0(501)

================

- Fixed the handling of UNICODE ini files. 

- This was tested on a RACK.

Version 1.0(500)

================

- Finished MAKMAKE stuff including testcode

- Mad proofed the Nif interface

- First release for ER5

Version 1.0(093)

================

- Changed ESOCK tom use MAKMAKE

Version 1.0(92)

================

- Changed Nifman to depend on the CommDB header file rather than using the

  definitions for datbase fields in nifvar.h

- Changed MS CallBack codes to be >1000 rather than -ve as -ve values cannot

  be stored in the database

Version 1.0(91)

================

- Fixed a defect in the MBuf code, didn't ulink when removing a buffer

  from a chain.

- Changed the Nif Interface for reading descriptors

Version 1.0(90)

================

- Modifications for callback. BC across the ESOCK i/f but not the NifMan

  interface, so a new agent is required. I.e. a kipper (Netdial2) which

  supports callback.

Version 1.0(072)

================

- Fixed WINC build because make files did not have the __WINS__ macro in them

- Added test code for WINC to tcesock.exe

Version 1.0(071)

================

- MARM debug build is now BC with MARM release as per E32 125.

- Updated obey files.

- Incrementeted build number to 071.

- Made changes to support PLP over TCPIP, i.e. allowing INI file to be 

  loaded from the MS Windows in WINC builds.

- Work around in place of E32 WINC/WINS/WIndows 95 TLS defect by using Windows 

  TLS instead of EPOC32 TLS

Version 1.0 (400)

=================

- R4 release. Standard set of R4 dependencies.

- Standardised MNT verbs. Can now build project on a clean drive using 

  BLDFROMNOTHING.

Version 1.0(070)

================

ESOCK.DLL is BC with Protea v.100 and beyond (nonunicode)

NIFMAN.DLL is BC with build 066 (nonunicode)

UNICODE ESOCK.DLL and UNICODE NIFMAN.DLL not BC with previous releases

Not SC with previous releases due to changes in TDes8 v.s TDes declarations

General changes

- Updated all build scripts to build and get UNICODE components

- Ran all test code in UNICODE on WINS and MARM rom on a rack D

Changes to NIFMAN.DLL

- Introduced freeze files for UNICODE WINS and MARM

- MNifIfNotify::Authenticate() now takes TDes arguments as oppose to TDes8

- Changed NIFMAN::SetSocketState() to return KErrUnsupported if it is given an 

  unrecognised

- CNifAgentBase::Authenticate() now takes TDes arguments as oppose to TDes8

- Changed INI file reading code to interpret ini files with leading FFFE bytes as UNICODE.

ESOCK.DLL

- Frozen def file changed for UNICODE WINS and MARM

- RServiceResolver::GetByName() & GetByNumber now take a TDesC argument as oppose to TDes8

- THostName definition changed to TBuf<0x40> rather than TBuf8<0x40>

- TServiceName definition changed to TBuf<0x20> rather than TBuf8<0x20>

- Added wide IPC Read/Write functions to server side

- Changed server side buffer to THostName rather than TBuf8<30> for SetHostName call

  There is a potential defect here which limits host name size of only 30 characters

- Changed INI file reading code to interpret ini files with leading FFFE bytes as UNICODE.

- Changed Changed CHostResolvProvdBase and CServiceResolvProvdBase functions to 

  use TDes instead of TDes8 declarations

- Had to increase the stack size to pass the nifman tests, so it is now KDefaultStackSize*sizeof(TText)

- Fixed SW1-836 RSocket::Write does not pass flags through on stream sends because of type 'o'

- Fixed SW1-851 Double clean up stack push for protocol family reference

- Added KSocketInternalWriteBit & KSocketInternalReadBit for protected Send/Recv flags

Version 1.0(069)

================

No source code changes but released with documentation under version control

Version 1.0(068)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 066

-Changed version numbers returned by RNif class to be teh same as ESOCK

-Added functionality to Nif::SetSocketState() to set the ESOCK socket internal

buffer size.  ENifBuffers512 sets the buffers to 512 bytes, ENifBuffers1024 to 1024

and ENifBuffers2048 to 2048.

-Incremented build number

Version 1.0(067)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 066

-Fixed SW1-203 'Possible panic if immediate restart after disconnect'

-Fixed SW1-204 'Stage transition from Uninitialised to Unconnected not picked up by RNifMonitor'

-Added iState==EDisconnecting __ASSERT_debug to NIFMAN DisconnectComplete() upcall

Version 1.0(066)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 063 with exception of added functions

Changes to NIFMAN

-Added derived class RNifMonitor.  RNifMonitor::Open() opens a session to the NIFMAN

server. It is similar to a RNif session however the corresponding agent is not

loaded nor is it referenced.  The only supported commands for RNifMonitor are:

Progress(), LastProgressError(), ProgressNotification(), CancelProgressNotification().

Other RNif functions will return KErrNotSupported.  These changes are a response

to SW1-195.

-Changed TPtrC(0,0) default arguments to TPtr() as per SW1-194

-Changed KNifOptLevel constant declaration from 401 to 0x191 to be consistent with other

hex declarations

-Added constants KAgentUninitialised=0 and KAgentUnconnected=1 and removed 

KMinAgentProgress=0.  This is all about allowing NIFMAN to give consistant updates

through the RNifMonitor class.  KAgentUninitalised is the default stage whenever

a monitor session connects and there is no agent.  As soon as an agent appears

the stage will change to KAgentUnconnected.  When the agent eventually unloads

the stage will change back to KAgentUinitialised.  These stages are consistient with

NTRAS and NETDIAL agents, plus it means the stage will be updated correctly in the 

final instance.

-Moved the last progress error variable storage from the agent reference into the

sessions.  This means a session must be active to get a last error.  Calling

LastProgressError() will now clear the last error from the particular session.

-Made various changes on the NIFMAN server side to support RNifMonitor class.  The

main changes involved ensuring the monitor sessions are updated when agents are born

/killed.

-Changed RNif client side code to make better code reuse when opening sessions.

-Added test code for RNifMonitor class and progress update changes.

Changes to ESOCK

-Incremented build numbers

Version 1.0(065)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 063

Changes to NIFMAN

-A new variable has been added "Timeout\StopIfNoClient".  If this variable is true or

the variable is not found then the interface will be stopped immediately if is is not

yet established and there are no network users, (i.e. internet protocol using socket

server sessions)

-A new variable has been added "Timeout\\ReconnectIfNoClient".  If this variable is not

found or is false then the reconnect dialog will not be requested if there

are no network users

Changes to ESOCK

-Incremented build numbers

Version 1.0(064)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 060 with exception of added functions

Changes to NIFMAN

-EsWatch thread now creates a reference to the nifman library in its

  entry point.  This avoids exceptions because esock dynamically loads nifman

-EsWatch now looks at the thread exit type in addition to the logon

  completion code

Changes to ESOCK

-Incremented build numbers

Version 1.0(063)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 060 with exception of added functions

Changes to NIFMAN

-Added following functions to get progress information and to stop network interfaces

  Nif::ProgressL(TNifProgress& aProgress, const TDesC& aName)

  Nif::Stop(TAny* aId, CNifIfBase* aIf)

  Nif::ProgressL(TNifProgress& aProgress, TAny* aId, CNifIfBase* aIf)

-Added constant KNifOptInterfaceProgress=3, for protocols to implement as a GetOpt

-Added test code as appropriate

Changes to ESOCK

-Incremented build number

Version 1.0(062)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 060

Changes to NIFMAN

-Fixed SW1-137 'Network timeouts don't allow zero seconds'

 If 0 is returned from the database the timeout is set to the minumum 1

 If any negative number is returned by the database the timeout is infinite

 If the database does not support the timeout values the timeout is infinite

-Fixed SW1-81 'PC.PRT can panic the Socket Server'

 If the port number is out of range KErrTooBig is returned

Changes to ESOCK

-Incremented build number

Version 1.0(061)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 060

Changes to NIFMAN

-Changed function name CNifFactory::ConstructL to CNifFactory::InitL, which means its

 still BC.  In any case the only project using this function is ETHER802

-Added enumeration ENifSocketCreated for Nif::SetSocketState.  This is BC because

 the enumeration maps directly onto appropriate ESOCK internal socket state.

-Added StartC32() function into RNif client side if socket server does not exist

-A thread 'EsWatch' is now started by NIFMAN to monitor ths socket server thread

 if the thread suffers an abnormal exit then the C32Exe process is restarted.

-Speeded up the operation of Nif::CheckIniConfig() by only searching for ESOCK.INI

 once on the server side

-The mbuf reaper started by NIFMAN will no longer remove the first mbuf pool.

Changes to ESOCK

-Incremented build number

Version 1.0(060)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 059 with exception of added functions

Changes to NIFMAN

-Added EXPORT CNifFactory::ConstructL

-Added EXPORT Nif::SetSocketState()

-Incremented build number

-Fixed SW1-97 'Cannot start NetDial twice in quick succession on Series 5'

-Fixed alloc heaven in Nifman::CheckIniConfig function

Changes to ESOCK

-Moved __UHEAP_MARK & __UHEAP_MARKEND statements in thread creation for debug build

-Now explicitly deleting active scheduler & cleanup trap

-Changed void CResolverProvdBase::SetNotify() to an inline

Version 1.0(059)

================

ESOCK.DLL is BC with Protea v.100 and beyond

NIFMAN.DLL is BC with build 058, but strings requested from agents have changed 

making it incompatible with TCPIP 026.

Changes to NIFMAN

-Added following Functions to NIFMAN

    Nifman::CheckIniConfig() - External function will ensure the ESOCK.INI file has been parsed

	Nif::IsProtocolLoaded() - NIFMAN internal function to determine if named protocol is loaded

	Nif::CheckInstalledL() - NIFMAN internal function to check nifman is installed

	Nif::CheckInstallMBufManager() - internal function which also checks in ini file for MBuf max heap size

-Added CMBufPoolReaper class which goes off every 5 seconds to recover mbuf manager heap

-Standardised all NIFVAR.H variables to have correct prefixes

-Changed 'ShortTimeout' to 'Timeout\\Client'

-Changed 'LongTimeout' to 'Timeout\\Route'

Changes to ESOCK

-Incremented build number

-Added friend class CMBufPoolReaper to RMBufQ

-Added friend class CMBufPoolReaper & Nif to CMBufManager

-Removed inline C'tor from CMBufManager

-Changed CMBufManager data variables from protected back to private

-Added friend class Nif to CProtocolRef

-Changed CESockIniData data members from private to protected

-Fixed SW1-68 'OOM Access violation' as suggested

-Fixed SW1-69 'Unchecked return value' as suggested

-Fixed SW1-42 'Inconsistent asserts' as suggested

-Fixed SW1-43 'ESock Fault after calling connect' as suggested

-Removed bogus assert from CProtocolBase d'tor

Version 1.0(058)

================

ESOCK.DLL is BC with Protea V.100 and beyond

New Component

-Moved NIFMAN.DLL from TCPIP into ESOCK & redesigned NIFMAN.DLL to cope with

multiple agents/nifs

-Added 3 test modules IDUMMY.NIF, ADUMMY.AGT & TNIFMAN.EXE.

-Changed PDUMMY.PRT to include nifman tests

Note new nifman will only work with TCPIP 026 and beyond

Changes to ESOCK (i.e. changes affecting Protea ROM)

-Added inline accessor for CProtocolBase::iRefCount - 'CProtocolBase::RefCount() const'

-Added KUidEsockExtension = 0x10000387 to ES_PROT.H

-Incremented build number

-Added inline C'tor for CMBufManager

-Changed data variables access for CMBufManager from private to protected

-Fixed SW3-1616 'RMBufChain::TrimEnd() does not deal with an offset of zero correctly'

-Fixed SW1-183 'Possible orphaning of library handle under OOM when loading extension module'

-Fixed SW1-384 'Possible exception under OOM caused by prematurely unloaded library/partially constructed object'

-Commented out various __ASSERT_debug's SS_SOCK.CPP lines 1332 & 1579 to enable TCPIPs

    half close mechanism to work properly

-Added inline accessor to CSocketScheduler::iGlobals - 'CSocketScheduler::Globals()'

-Fixed SW1-383 'Possible OOM exception because 'new' should be 'new (ELeave)'

Other

-IPC.PRT is now included in the ESOCK releasables.

-Given CIpcProtocolHolder::GetNextFreePort TInt return type to eliminate warning

Version 1.0(057)

================

BC

-Fixed SW3-1332 'Potential OOM problem with Socket Server exclusive mode'

-Incremented build number to 057