[Indigo]<Yggdrasil>YggPrincOpsCM>FileStreamImpl.mesa!3

FileStreamImpl.mesa

MBrown on September 17, 1983 8:39 pm

Rovner on August 15, 1983 1:02 pm

Levin on September 22, 1983 3:22 pm

Birrell on August 23, 1983 3:14 pm

Schroeder on November 28, 1983 12:36 pm

Russ Atkinson (RRA) February 4, 1985 3:09:10 pm PST

Bob Hagmann February 13, 1989 10:12:32 am PST

Doug Wyatt, December 12, 1986 5:19:17 pm PST

Please maintain change log at end of file.

DIRECTORY

Basics USING [bytesPerWord, LongNumber, BITAND, LowHalf, UnsafeBlock],

BasicTime USING [GMT],

FileStream USING [FinalizationProc, SetLength],

FileStreamPrivate USING[ Data, FSDataHandle, BufferNodeHandle, BufferNode, DoFinalization, FileDataHandle, FileData, NodeStatus, ProcHandle, ProcHandleFromAccessRights, SetupBuffer, StartRequest ],

FS USING [ByteCount, BytesForPages, Close, Error, ErrorDesc, ExtendFileProc, GetInfo, GetName, InitialPosition, Lock, nullOpenFile, OpenFile, Read, SetByteCountAndCreatedTime, SetPageCount, StreamBufferParms, StreamOptions, Write],

FSBackdoor USING [ProduceError],

FSLock USING [RecordREF, RemoveREF ],

IO USING [Close, CreateStream, CreateStreamProcs, EndOfStream, Error, GetChar, GetIndex, GetLength, SetIndex, STREAM, StreamProcs],

IOUtils USING [closedStreamProcs, StoreData],

PrincOps USING [ByteBltBlock],

PrincOpsUtils USING [ByteBlt],

Process USING [GetPriority, Pause, Priority, priorityForeground, SetPriority],

Rope USING [ROPE],

RuntimeError USING [BoundsFault],

SafeStorage USING [EnableFinalization, FinalizationQueue, FQNext],

VM USING [ AddressForPageNumber, Allocate, BytesForPages, CantAllocate, Free, Interval, MakeChanged, MakeUnchanged, PageNumber, SwapIn];

FileStreamImpl: CEDAR MONITOR

LOCKS fileData.lockRecord USING fileData: FileDataHandle

IMPORTS Basics, FileStream, FileStreamPrivate, FS, FSBackdoor, FSLock, PrincOpsUtils, IO, IOUtils, Process, RuntimeError, SafeStorage, VM

EXPORTS FileStreamPrivate, FileStream = BEGIN OPEN Basics, Rope;

STREAM: TYPE = IO.STREAM;

ByteCount: TYPE = INT;

ByteNumber: TYPE = ByteCount; -- index rather than count

PageNumber: TYPE = VM.PageNumber;

PageCount: TYPE = VM.PageNumber;

bytesPerFilePage: CARDINAL = FS.BytesForPages[1];

maxVMPagesPerBuffer: INT = 65536/VM.BytesForPages[pages: 1] ;

clearLowBits: CARDINAL = CARDINAL.LAST-(bytesPerFilePage-1);

minFileExtend: INT = 10*bytesPerFilePage;

Data: TYPE = FileStreamPrivate.Data;

FSDataHandle: TYPE = FileStreamPrivate.FSDataHandle;

BufferNode: TYPE = FileStreamPrivate.BufferNode;

BufferNodeHandle: TYPE = FileStreamPrivate.BufferNodeHandle;

FileDataHandle: TYPE = FileStreamPrivate.FileDataHandle;

FileData: TYPE = FileStreamPrivate.FileData;

ProcHandle: TYPE = FileStreamPrivate.ProcHandle;

This code does not protect itself from parallel use of a stream by concurrent

processes. It assumes that the processes will synchronize at a higher level.

Parallel use of different streams for the same open file is expected, but the

read/write stream must be opened by StreamOpen or StreamFromOpenFile,

and read stream by StreamFromOpenStream.

Get and Put

CleanupAfterPut: ENTRY PROC [fileData: FileDataHandle, selfData: FSDataHandle] = INLINE {

Restores dataBytesInBuffer and fileLength if they are messed up by a PutChar or

PutBlock past the end of file.

Same logic in SetLengthUnderMonitor.

only call this routine with a write stream

currentNode: BufferNodeHandle = selfData.currentNode;

IF currentNode.didPut THEN {

currentNode.bufferDirty ← TRUE;

IF selfData.index > currentNode.dataBytesInBuffer THEN {

currentNode.dataBytesInBuffer ← selfData.index;

fileData.fileLength ← currentNode.firstFileByteInBuffer + selfData.index;

};

currentNode.didPut ← FALSE;

};

EstablishFileLength: ENTRY PROC[fileData: FileDataHandle ]

RETURNS [fileLength: INT] = INLINE {

Paw through write stream info to find the file length. The new length is

the true length modulo some uncertainity whether a put was done in parallel

during the execution of this routine. The file length returned is

as least as big as the file was when the monitor was acquired. This

is fine because the notion of EOF or file length for a reader of a file that is

in the process of being written is somewhat vague. A higher level protocol

should keep this straight in the client (why are you reading bytes that

might not be there yet?). Fix up dataBytesInBuffer if needed.

(This is mostly intended to allow the read stream to look at the file size seen

by the write stream.)

ENABLE UNWIND => NULL;

writeData: FSDataHandle ← fileData.writeStreamData;

IF writeData = NIL OR writeData.streamIsClosed THEN {

writeData ← NIL ;

RETURN[fileData.fileLength] ;

}

ELSE {

writeNode: BufferNodeHandle = writeData.currentNode ;

writeNode.dataBytesInBuffer ← MAX[ writeData.index, writeNode.dataBytesInBuffer] ;

fileLength ← MAX[

fileData.fileLength,

writeNode.firstFileByteInBuffer + writeNode.dataBytesInBuffer ] ;

fileData.fileLength ← fileLength;

};

writeData ← NIL ;

};

convertFStoIOError: PROC [self: STREAM, error: FS.ErrorDesc] = INLINE {

selfData: FSDataHandle ← NARROW[self.streamData];

IF selfData # NIL THEN {

selfData.FSErrorDesc ← error ;

IF selfData.ConvertFStoIOErrors THEN IO.Error[$Failure, self];

}

ELSE {

IO.Error[$Failure, self];

};

GetChar: PUBLIC PROC [self: STREAM] RETURNS [CHAR] = {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

node: BufferNodeHandle ← selfData.currentNode ;

c: CHAR;

fileLength: INT ;

IF bufferBad[node.status] THEN IO.Error[$Failure, self];

IF selfData.index >= node.dataBytesInBuffer THEN {

Suspect that end-of-buffer or end-of-file has been reached.

This may be false! However, the test is cheap and usually false.

fileLength ← EstablishFileLength[fileData: selfData.fileData ] ;

File length may be wrong if writer is using the same buffer

as the reader, so get a good file length. This is not cheap: we

have to get a monitor lock and maybe look inside the write stream.

Note that we use the local variable fileLength and not

selfData.fileData.fileLength.

IF fileLength <= selfData.index+node.firstFileByteInBuffer

THEN ERROR IO.EndOfStream[self];

We are not at EOF. If we are at EOB, then get the next buffer.

Not EOF and not EOB can occur if the writer has put some

char's into the buffer and this was not reflected in dataBytesInBuffer

until we did the EstablishFileLength call.

IF selfData.index = node.bufferBytes THEN node ← AdvanceBuffer[selfData] };

TRUSTED{c ← node.buffer[selfData.index]};

selfData.index ← selfData.index + 1;

selfData ← NIL ;

RETURN[c] ;

};

PutChar: PUBLIC PROC [self: STREAM, char: CHAR] = {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

node: BufferNodeHandle ← selfData.currentNode ;

IF selfData.index = node.bufferBytes THEN node ← AdvanceBuffer[selfData];

IF bufferBad[node.status] THEN IO.Error[$Failure, self];

TRUSTED{node.buffer[selfData.index] ← char};

selfData.index ← selfData.index + 1;

node.didPut ← TRUE ;

selfData ← NIL ;

};

Change use to IOUtils.AddNAT when the 0+0 gives NAT.LAST bug is fixed

AddNAT: PROC [a, b: NAT] RETURNS [NAT] = INLINE {

RETURN [MIN[CARDINAL[a]+CARDINAL[b], NAT.LAST]];

};

GetBlock: PUBLIC PROC [self: STREAM, block: REF TEXT, startIndex: NAT,

count: NAT]

RETURNS [nBytesRead: NAT] = TRUSTED {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

textBlock: PrincOps.ByteBltBlock;

countRemaining: NAT;

stopIndexPlusOne: NAT = MIN [block.maxLength, AddNAT[startIndex, count]];

IF bufferBad[selfData.currentNode.status] THEN IO.Error[$Failure, self];

textBlock ← [

blockPointer: LOOPHOLE[block, LONG POINTER] + TEXT[0].SIZE,

startIndex: startIndex,

stopIndexPlusOne: stopIndexPlusOne];

countRemaining ←

IF startIndex > stopIndexPlusOne THEN 0 ELSE stopIndexPlusOne-startIndex;

nBytesRead ← 0;

WHILE countRemaining # 0 DO

bufferBlock: PrincOps.ByteBltBlock ← [

blockPointer: selfData.currentNode.buffer,

startIndex: selfData.index,

stopIndexPlusOne: selfData.currentNode.dataBytesInBuffer];

countTransferred: CARDINAL ← 0;

IF bufferBlock.startIndex < bufferBlock.stopIndexPlusOne THEN

countTransferred ← PrincOpsUtils.ByteBlt[from: bufferBlock, to: textBlock];

selfData.index ← selfData.index + countTransferred;

nBytesRead ← nBytesRead + countTransferred;

IF (countRemaining ← countRemaining - countTransferred) = 0 THEN EXIT;

Bytes may be added concurrently with this get. EstablishFileLength gives

a true file length (which may be different from what it was when we

started this iteration) to see if there is more data to blt.

IF EstablishFileLength[fileData: selfData.fileData ] <=

selfData.index + selfData.currentNode.firstFileByteInBuffer

THEN EXIT;

textBlock.startIndex ← textBlock.startIndex + countTransferred;

The below IF is needed for the same reason we called

EstablishFileLength above.

IF selfData.index = selfData.currentNode.bufferBytes THEN [] ← AdvanceBuffer[selfData];

ENDLOOP;

IF nBytesRead # 0 THEN block.length ← startIndex + nBytesRead;

selfData ← NIL ;

RETURN[nBytesRead] };

PutBlock: PUBLIC PROC [self: STREAM, block: REF READONLY TEXT, startIndex: NAT,

count: NAT] = TRUSTED {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

Fail if startIndex<0 or stopIndexPlusOne<0.

textBlock: PrincOps.ByteBltBlock;

countRemaining: NAT;

stopIndexPlusOne: NAT ← AddNAT[startIndex, count];

IF bufferBad[selfData.currentNode.status] THEN IO.Error[$Failure, self];

IF stopIndexPlusOne > block.maxLength THEN stopIndexPlusOne ← block.length;

textBlock ← [

blockPointer: LOOPHOLE[block, LONG POINTER] + TEXT[0].SIZE,

startIndex: startIndex,

stopIndexPlusOne: stopIndexPlusOne];

countRemaining ←

IF startIndex > stopIndexPlusOne THEN 0 ELSE stopIndexPlusOne-startIndex;

WHILE countRemaining # 0 DO

bufferBlock: PrincOps.ByteBltBlock ← [

blockPointer: selfData.currentNode.buffer,

startIndex: selfData.index,

stopIndexPlusOne: selfData.currentNode.bufferBytes]; -- allow put past current eof.

countTransferred: CARDINAL ← PrincOpsUtils.ByteBlt[from: textBlock, to: bufferBlock];

selfData.index ← selfData.index + countTransferred;

selfData.currentNode.didPut ← TRUE;

IF (countRemaining ← countRemaining - countTransferred) = 0 THEN EXIT;

textBlock.startIndex ← textBlock.startIndex + countTransferred;

[] ← AdvanceBuffer[selfData];

ENDLOOP;

selfData ← NIL ;

};

maxWordsMoved: INT = (LAST[CARDINAL] / bytesPerWord) - 1;

maxBytesMoved: INT = maxWordsMoved * bytesPerWord;

maxStopIndexPlusOne: INT = maxBytesMoved ;

all designed to make the max number of bytes transferred an integral number of

words, which is good

UnsafeGetBlock: PUBLIC UNSAFE PROC [self: STREAM, block: UnsafeBlock]

RETURNS [nBytesRead: INT] = UNCHECKED {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

textBlock: PrincOps.ByteBltBlock;

stopIndexPlusOne: INT;

IF bufferBad[selfData.currentNode.status] THEN IO.Error[$Failure, self];

IF block.startIndex < 0 OR block.count < 0 THEN ERROR RuntimeError.BoundsFault;

IF block.count = 0 THEN {

selfData ← NIL ;

RETURN [0];

};

IF block.startIndex > maxBytesMoved THEN {

scale block.startIndex into [0 .. bytesPerWord)

wordOffset: INT = block.startIndex / bytesPerWord;

block.base ← block.base + wordOffset;

block.startIndex ← block.startIndex - wordOffset*bytesPerWord;

};

stopIndexPlusOne ← block.startIndex + block.count;

nBytesRead ← 0;

Transfer at most maxBytesMoved bytes from the stream to block^.

Assert block.startIndex IN [0 .. maxStopIndexPlusOne), < stopIndexPlusOne

countRemaining: CARDINAL;

textBlock ← [

blockPointer: block.base,

startIndex: block.startIndex,

stopIndexPlusOne: MIN[maxStopIndexPlusOne, stopIndexPlusOne]];

countRemaining ← textBlock.stopIndexPlusOne - textBlock.startIndex;

Assert countRemaining > 0

The following loop transfers from the stream to textBlock^ until textBlock^ is full

or end of file is reached.

bufferBlock: PrincOps.ByteBltBlock ← [

blockPointer: selfData.currentNode.buffer,

startIndex: selfData.index,

stopIndexPlusOne: selfData.currentNode.dataBytesInBuffer];

countTransferred: CARDINAL ← 0;

IF bufferBlock.startIndex < bufferBlock.stopIndexPlusOne THEN

countTransferred ← PrincOpsUtils.ByteBlt[from: bufferBlock, to: textBlock];

selfData.index ← selfData.index + countTransferred;

nBytesRead ← nBytesRead + countTransferred;

IF (countRemaining ← countRemaining - countTransferred) = 0 THEN EXIT;

IF EstablishFileLength[fileData: selfData.fileData ] <=

selfData.index + selfData.currentNode.firstFileByteInBuffer THEN {

selfData ← NIL ;

GOTO return;

};

textBlock.startIndex ← textBlock.startIndex + countTransferred;

IF selfData.index = selfData.currentNode.bufferBytes THEN [] ← AdvanceBuffer[selfData];

ENDLOOP;

IF textBlock.stopIndexPlusOne = stopIndexPlusOne THEN {

selfData ← NIL ;

GOTO return;

};

Assert textBlock.stopIndexPlusOne = maxStopIndexPlusOne

block.base ← block.base + maxWordsMoved;

block.startIndex ← 0;

stopIndexPlusOne ← stopIndexPlusOne - maxBytesMoved;

ENDLOOP;

EXITS

return => RETURN [nBytesRead]

};

UnsafePutBlock: PUBLIC PROC [self: STREAM, block: UnsafeBlock] = TRUSTED {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

selfData: FSDataHandle ← NARROW[self.streamData];

textBlock: PrincOps.ByteBltBlock;

stopIndexPlusOne: INT;

IF bufferBad[selfData.currentNode.status] THEN IO.Error[$Failure, self];

IF block.startIndex < 0 OR block.count < 0 THEN ERROR RuntimeError.BoundsFault;

IF block.startIndex > maxBytesMoved THEN {

scale block.startIndex into [0 .. bytesPerWord)

wordOffset: INT = block.startIndex / bytesPerWord;

block.base ← block.base + wordOffset;

block.startIndex ← block.startIndex - wordOffset*bytesPerWord;

};

stopIndexPlusOne ← block.startIndex + block.count;

Transfer at most maxBytesMoved bytes from block^ to the stream.

Assert block.startIndex IN [0 .. maxStopIndexPlusOne), < stopIndexPlusOne

countRemaining: CARDINAL;

textBlock ← [

blockPointer: block.base,

startIndex: block.startIndex,

stopIndexPlusOne: MIN[maxStopIndexPlusOne, stopIndexPlusOne]];

countRemaining ← textBlock.stopIndexPlusOne - textBlock.startIndex;

Assert countRemaining > 0

The following loop transfers textBlock^ to the stream.

bufferBlock: PrincOps.ByteBltBlock ← [

blockPointer: selfData.currentNode.buffer,

startIndex: selfData.index,

stopIndexPlusOne: selfData.currentNode.bufferBytes]; -- allow put past current eof.

countTransferred: CARDINAL ← PrincOpsUtils.ByteBlt[from: textBlock, to: bufferBlock];

selfData.index ← selfData.index + countTransferred;

selfData.currentNode.didPut ← TRUE;

IF (countRemaining ← countRemaining - countTransferred) = 0 THEN EXIT;

textBlock.startIndex ← textBlock.startIndex + countTransferred;

[] �vanceBuffer[selfData];

ENDLOOP;

IF textBlock.stopIndexPlusOne = stopIndexPlusOne THEN EXIT;

Assert textBlock.stopIndexPlusOne = maxStopIndexPlusOne

block.base ← block.base + maxWordsMoved;

block.startIndex ← 0;

stopIndexPlusOne ← stopIndexPlusOne - maxBytesMoved;

ENDLOOP ;

selfData ← NIL ;

};

AdvanceBuffer: PROC [fsData: FSDataHandle] RETURNS [node: BufferNodeHandle]= {

On entry, index = dataBytesInBuffer = bufferBytes. Exit with same position in

file, but index < dataBytesInBuffer or EOF.

Handles implicit file extension.

Called from GetChar, PutChar, GetBlock, PutBlock, UnsafeGetBlock, UnsafePutBlock.

fileData: FileDataHandle = fsData.fileData;

firstByteOfNextPage: INT = fsData.currentNode.firstFileByteInBuffer +

fsData.currentNode.bufferBytes;

changeSize: BOOL ← FALSE;

IF firstByteOfNextPage = maxLength THEN ERROR IO.Error[$Failure, NIL];

IF fsData.isWriteStream THEN CleanupAfterPut[fileData: fileData, selfData: fsData];

IF firstByteOfNextPage >= fileData.byteSize THEN {

newSize: INT ← 0;

IF fileData.extendFileProc # NIL THEN

newSize ← fileData.extendFileProc[firstByteOfNextPage] ;

fileData.byteSize ← IF newSize # 0 THEN MAX[newSize, firstByteOfNextPage]

ELSE fileData.byteSize + MAX[minFileExtend,

((fileData.byteSize/10)/bytesPerFilePage)*bytesPerFilePage];

SetFileSize[fileData.fileHandle, fileData.byteSize] };

node ← SetupBuffer[fileData: fileData, fsData: fsData, fileByte: firstByteOfNextPage];

fsData.index ← LowHalf[firstByteOfNextPage-fsData.currentNode.firstFileByteInBuffer];

fsData ← NIL ;

};

EndOf: PUBLIC PROC [self: STREAM] RETURNS[BOOL] = {

selfData: FSDataHandle ← NARROW[self.streamData];

node: BufferNodeHandle ← selfData.currentNode ;

do cheap test to see if not at EOF

IF selfData.index >= node.dataBytesInBuffer THEN {

Cheap test inconclusive. Find real file length.

fileLength: INT = EstablishFileLength[fileData: selfData.fileData];

IF fileLength <= selfData.index+node.firstFileByteInBuffer

THEN RETURN[TRUE];

};

selfData ← NIL ;

RETURN[FALSE];

};

CharsAvail: PUBLIC PROC [self: STREAM, wait: BOOL] RETURNS [INT] = {

RETURN[INT.LAST] };

GetIndex: PUBLIC PROC [self: STREAM] RETURNS [index: INT] = {

selfData: FSDataHandle ← NARROW[self.streamData];

index ← selfData.currentNode.firstFileByteInBuffer + selfData.index ;

selfData ← NIL ;

};

SetIndex: PUBLIC PROC [self: STREAM, index: INT] = {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

fsData: FSDataHandle ← NARROW[self.streamData];

currentNode: BufferNodeHandle ← fsData.currentNode ;

firstBufferByte: INT ← currentNode.firstFileByteInBuffer;

fileData: FileDataHandle = fsData.fileData ;

fileLength: INT ;

IF index < 0 THEN ERROR IO.Error[BadIndex, self];

Make sure dataBytesInBuffer and fileLength are correct by calling

CleanupAfterPut or EstablishFileLength

IF fsData.isWriteStream THEN {

CleanupAfterPut[fileData: fileData, selfData: fsData];

fileLength ← fileData.fileLength; }

ELSE fileLength ← EstablishFileLength[fileData: fileData ];

IF index > fileLength THEN ERROR IO.EndOfStream[self];

ensure that page containing byte "index" is in the buffer

IF index NOT IN [firstBufferByte .. firstBufferByte+currentNode.bufferBytes)

THEN {

firstBufferByte ← index - (index MOD currentNode.bufferBytes);

currentNode ← SetupBuffer[fileData: fileData, fsData: fsData, fileByte: firstBufferByte];

};

fsData.index ← index - firstBufferByte;

fsData ← NIL ;

};

Reset: PUBLIC PROC [self: STREAM] = {

SetIndex[self, GetLength[self]] };

Flush: PUBLIC PROC [self: STREAM] = {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

fsData: FSDataHandle ← NARROW[self.streamData];

IF fsData.isWriteStream THEN ForceOut[ fsData: fsData ];

fsData ← NIL ;

};

Close: PUBLIC PROC [self: STREAM, abort: BOOL] = {

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

fsData: FSDataHandle ← NARROW[self.streamData];

IF ~abort THEN ForceOut[ fsData: fsData ! UNWIND => {

I would like to do a CONTINUE, but that is a "GOTO" and thus will stop the UNWIND here.

There is some bug that confuses the UNWIND machinery that trashes fsData from the local frame. We recompute it here.

fsData: FSDataHandle ← NARROW[self.streamData];

CloseFileDataForStream[fileData: fsData.fileData, fsData: fsData, forceCleanUp: TRUE];

fsData ← NIL ;

self.streamData ← NIL ;

self.streamProcs ← IOUtils.closedStreamProcs;

};

];

CloseFileDataForStream[fileData: fsData.fileData, fsData: fsData];

fsData ← NIL ;

self.streamData ← NIL ;

self.streamProcs ← IOUtils.closedStreamProcs;

};

Procs that are called via the property list mechanism.

GetLength: PUBLIC PROC [self: STREAM] RETURNS [length: INT] = {

selfData: FSDataHandle ← NARROW[self.streamData];

IF selfData.streamIsClosed THEN ERROR IO.Error[StreamClosed, self];

length ← EstablishFileLength[fileData: selfData.fileData ] ;

selfData ← NIL ;

};

clearHighBits: CARDINAL = (bytesPerFilePage-1);

maxLength: INT = INT.LAST - bytesPerFilePage;

SetLength: PUBLIC PROC [self: STREAM, length: INT] = {

Note: do not reduce the size of a shortened file until stream closed.

ENABLE FS.Error => {

convertFStoIOError [self, error];

};

fsData: FSDataHandle ← NARROW[self.streamData];

IF fsData.streamIsClosed THEN ERROR IO.Error[StreamClosed, self];

IF length NOT IN [0 .. maxLength] THEN ERROR IO.Error[BadIndex, self];

SetLengthUnderMonitor[fileData: fsData.fileData, length: length];

IF fsData.index+fsData.currentNode.firstFileByteInBuffer > length THEN {

If old index was past EOF, then move it to EOF.

We leave the cloned read stream alone: if it does not do a setPosition then

it will get an EOF on its next read.

fsData.index ← 0 ;

SetIndex[self: self, index: length];

};

fsData ← NIL ;

};

RoundUpToPages: PROC [bytes: INT] RETURNS [INT] = INLINE {

bytes ← bytes + (bytesPerFilePage-1);

LOOPHOLE[bytes, LongNumber[pair]].lo ←

BITAND[LOOPHOLE[bytes, LongNumber[pair]].lo, clearLowBits];

RETURN[bytes];

};

PagesForRoundUpBytes: PROC [bytes: INT] RETURNS [INT] = INLINE {

RETURN[RoundUpToPages[bytes]/bytesPerFilePage];

};

SetLengthUnderMonitor: ENTRY PROC [fileData: FileDataHandle, length: INT] = {

ENABLE UNWIND => NULL;

newFileBytes: INT = RoundUpToPages[length];

oldFileLength: INT ;

nowNode: BufferNodeHandle ← fileData.firstBufferNode;

writeData: FSDataHandle ← fileData.writeStreamData ;

IF writeData # NIL AND writeData.currentNode.didPut THEN {

CleanupAfterPut logic is copied here, but we cannot call CleanupAfterPut

because it is an ENTRY

currentNode: BufferNodeHandle = writeData.currentNode;

currentNode.bufferDirty ← TRUE;

currentNode.didPut ← FALSE;

IF writeData.index > currentNode.dataBytesInBuffer THEN {

currentNode.dataBytesInBuffer ← writeData.index;

fileData.fileLength ← currentNode.firstFileByteInBuffer + writeData.index };

currentNode.didPut ← FALSE };

oldFileLength ← fileData.fileLength ;

fileData.fileLength ← length;

IF length < fileData.validBytesOnDisk THEN fileData.validBytesOnDisk ← length ;

grow file if needed

IF length > fileData.byteSize THEN {

fileData.byteSize ← newFileBytes;

SetFileSize[fileData.fileHandle, fileData.byteSize];

};

Look through nodes and adjust those past EOF or with EOF in buffer

UNTIL nowNode = NIL DO

IF (nowNode.status # invalid) THEN {

IF (nowNode.firstFileByteInBuffer+nowNode.bufferBytes > length) THEN {

IF nowNode.firstFileByteInBuffer >= length THEN {

All of the buffer is past EOF.

Set dataBytesInBuffer to 0 so that gets will find themselves at EOF,

and clean the node to avoid redundant write

nowNode.dataBytesInBuffer ← 0;

nowNode.bufferDirty ← FALSE ;

nowNode.didPut ← FALSE ;

}

ELSE {

EOF is in (or just past) this buffer

nowNode.dataBytesInBuffer ← length - nowNode.firstFileByteInBuffer ;

IF nowNode.didPut THEN {

nowNode.didPut ← FALSE ;

nowNode.bufferDirty ← TRUE ;

};

}

ELSE {

all of node is in the file

nowNode.dataBytesInBuffer ← nowNode.bufferBytes ;

};

nowNode ← nowNode.nextBufferNode ;

ENDLOOP;

writeData ← NIL ;

};

EraseChar: PUBLIC PROC [self: STREAM, char: CHAR] = {

index: INT = GetIndex[self];

IF index = 0 THEN ERROR IO.Error[IllegalBackup, self];

SetIndex[self, index-1];

IF GetChar[self] # char THEN {PutChar[self, '\\]; PutChar[self, char]}

ELSE SetIndex[self, index-1] };

Backup: PUBLIC PROC [self: STREAM, char: CHAR] = {

selfData: FSDataHandle ← NARROW[self.streamData];

index: INT;

IF selfData.streamIsClosed THEN ERROR IO.Error[StreamClosed, self];

index ← GetIndex[self];

IF index = 0 THEN ERROR IO.Error[IllegalBackup, self];

SetIndex[self, index-1];

IF GetChar[self] # char THEN ERROR IO.Error[IllegalBackup, self];

SetIndex[self, index-1];

selfData ← NIL ;

};

CloseFileDataForStream: ENTRY PROC [fileData: FileDataHandle, fsData: FSDataHandle, forceCleanUp: BOOL ← FALSE] = {

Processing for "Close" that must be done under the monitor.

forceCleanUp being TRUE is when Close has caught UNWIND, and we should do all we can to blow away the data structures.

ENABLE UNWIND => BROADCAST fileData.somethingHappened;

needDeleted: INT ← IF fileData.numberOfStreams = 0

THEN INT.LAST

ELSE fileData.streamBufferParms.nBuffers ;

lastNode: BufferNodeHandle ← NIL ;

node: BufferNodeHandle ;

IF (node ← fsData.readAheadNode) # NIL THEN {

make sure the read-ahead has completed

WHILE node.useCount = 1 AND node.status # valid AND

node.status # invalid AND node.status # needsSequentialRead DO

WAIT fileData.somethingHappened ;

ENDLOOP;

node.useCount ← node.useCount-1 ;

};

fsData.readAheadNode ← NIL;

IF ~forceCleanUp AND fsData.isWriteStream AND fileData.accessRights = $write AND

fileData.streamOptions[truncatePagesOnClose] THEN {

SetFileSize[fileData.fileHandle, fileData.fileLength] ;

};

We are now committed to closing the stream. Our last non-internal operation has completed (the SetFileSize).

IF (node ← fsData.currentNode) # NIL THEN node.useCount ← node.useCount-1 ;

fsData.currentNode ← NIL ;

look for up to two buffers to free if another stream is around

else, free all buffers

node ← fileData.firstBufferNode ;

UNTIL node = NIL OR needDeleted = 0 DO

IF node.useCount = 0 THEN {

TRUSTED{VM.Free[node.bufferInterval]};

IF fileData.firstBufferNode = node

THEN fileData.firstBufferNode ← node.nextBufferNode

ELSE lastNode.nextBufferNode ← node.nextBufferNode ;

needDeleted ← needDeleted - 1 ;

}

ELSE {

lastNode ← node ;

};

node ← node.nextBufferNode ;

ENDLOOP;

fsData.streamIsClosed ← TRUE;

IF (fileData.numberOfStreams ← fileData.numberOfStreams - 1) = 0 THEN {

fileDataTemp: FS.OpenFile ← fileData.fileHandle ;

fileData.fileHandle ← FS.nullOpenFile ;

IF fileData.streamOptions[closeFSOpenFileOnClose] THEN {

fileDataTemp.Close[! FS.Error => IF forceCleanUp THEN CONTINUE;];

-- let finialization close the file is we fail.

};

BROADCAST fileData.somethingHappened ;

fsData ← NIL ;

};

Insure that all buffers, except the one corresponding to the currentNode,

have been written to disk.

Normal cases are too return immediately when no writes are outstanding,

or to wait until one finishes.

FinishWrites: ENTRY PROC [fileData: FileDataHandle, fsData: FSDataHandle,

currentNode: BufferNodeHandle, parallelWritesOK: BOOL ← FALSE] = {

ENABLE UNWIND => NULL;

nowNode: BufferNodeHandle ← fileData.firstBufferNode;

IF fileData.accessRights = read THEN RETURN ;

IF currentNode = NIL THEN { -- this is only true when called via ForceOut

or when SetUpNodes can't get a node

WHILE fileData.writeCount # 0 DO

WAIT fileData.somethingHappened ;

ENDLOOP;

};

UNTIL nowNode = NIL DO

IF nowNode # currentNode THEN {

WHILE (nowNode.status = parallelReadActive) OR (nowNode.status = needsParallelRead) OR ( ~parallelWritesOK AND ((nowNode.status = parallelWriteActive) OR (nowNode.status = needsParallelWrite))) DO

WAIT fileData.somethingHappened;

ENDLOOP;

IF (nowNode.status = needsSequentialWrite OR (currentNode = NIL AND nowNode.status # invalid)) THEN {

What has happened is that an asynchronous write has failed,

or we are trying to flush all dirty pages in ForceOut.

The parallel process has given up, and we are about to re-do

the write under the monitor to get the signal and the stack

correct so that the client sees a correct view of the error.

nowNode.status ← sequentialWriteActive ;

TRUSTED{WriteFilePages[f: fsData.fileData.fileHandle, node: nowNode ! UNWIND => nowNode.status ← valid] };

nowNode.bufferDirty ← FALSE ;

nowNode.status ← valid ;

};

nowNode ← nowNode.nextBufferNode ;

ENDLOOP;

fsData ← NIL ;

};

nodeNowAvailable: ENTRY PROC [fileData: FileDataHandle] RETURNS [BOOL]= {

ENABLE UNWIND => NULL;

nowNode: BufferNodeHandle ;

nowNode ← fileData.firstBufferNode;

UNTIL nowNode = NIL DO

IF ( nowNode.status = valid OR nowNode.status = invalid) AND

nowNode.useCount = 0 THEN RETURN [TRUE];

IF nowNode.status = needsSequentialWrite THEN RETURN[FALSE];

nowNode ← nowNode.nextBufferNode ;

ENDLOOP;

WAIT fileData.somethingHappened;

ENDLOOP;

};

FinishRead: ENTRY PROC [fileData: FileDataHandle,

node: BufferNodeHandle, bufferSize: INT] = INLINE {

ENABLE UNWIND => NULL;

node.status ← valid;

node.dataBytesInBuffer ← bufferSize ;

BROADCAST fileData.somethingHappened;

};

FinishBadRead: ENTRY PROC [fileData: FileDataHandle, node: BufferNodeHandle] =

INLINE {

ENABLE UNWIND => NULL;

node.status ← invalid;

node.dataBytesInBuffer ← 0 ;

BROADCAST fileData.somethingHappened;

};

a pre-read has failed. Mark it needsSequentialRead if a stream needs it

to continue, or just ignore it if it was a real preread.

FinishBadPreRead: ENTRY PROC [fileData: FileDataHandle, node: BufferNodeHandle] =

INLINE {

ENABLE UNWIND => NULL;

IF (fileData.firstReadStream # NIL AND fileData.firstReadStream.currentNode = node) OR

(fileData.writeStreamData # NIL AND fileData.writeStreamData.currentNode = node) THEN {

node.status ← needsSequentialRead;

node.dataBytesInBuffer ← 0 ;

BROADCAST fileData.somethingHappened;

}

ELSE {

IF fileData.firstReadStream # NIL AND fileData.firstReadStream.readAheadNode = node THEN {

fileData.firstReadStream.readAheadNode ← NIL ;

node.useCount ← 0 ;

};

IF fileData.writeStreamData # NIL AND fileData.writeStreamData.readAheadNode = node THEN {

fileData.writeStreamData.readAheadNode ← NIL ;

node.useCount ← 0 ;

};

node.status ← invalid;

BROADCAST fileData.somethingHappened;

};

markNodeNotWritten: ENTRY PROC [fileData: FileDataHandle, node: BufferNodeHandle] =

INLINE {

ENABLE UNWIND => NULL;

node.status ← needsSequentialWrite ;

node.bufferDirty ← TRUE ;

fileData.writeCount ← fileData.writeCount - 1;

BROADCAST fileData.somethingHappened;

};

markNodeWritten: ENTRY PROC [fileData: FileDataHandle, node: BufferNodeHandle] =

INLINE {

ENABLE UNWIND => NULL;

node.status ← valid ;

fileData.writeCount ← fileData.writeCount - 1;

BROADCAST fileData.somethingHappened;

};

bumpWriteCount: ENTRY PROC [fileData: FileDataHandle] = INLINE {

ENABLE UNWIND => NULL;

fileData.writeCount ← fileData.writeCount + 1;

};

WaitForOneBufferNotWriting: ENTRY PROC [fileData: FileDataHandle] = INLINE {

ENABLE UNWIND => NULL;

WHILE fileData.writeCount >= fileData.streamBufferParms.nBuffers DO

WAIT fileData.somethingHappened;

ENDLOOP;

};

WaitForParallelWriteToComplete: ENTRY PROC [fileData: FileDataHandle,

node: BufferNodeHandle] = INLINE {

ENABLE UNWIND => NULL;

WHILE node.status = needsParallelWrite OR node.status = parallelWriteActive DO

WAIT fileData.somethingHappened;

ENDLOOP;

};

ProcessNode: PUBLIC PROC [ fileData: FileDataHandle, node: BufferNodeHandle ] = {

IF node.status = needsParallelRead THEN {

node.status ← parallelReadActive ;

ReadAhead [node: node, fileData: fileData];

RETURN ;

};

IF node.status = needsParallelWrite THEN {

node.status ← parallelWriteActive ;

parallelWriteBuffer [ node: node, fileData: fileData ] ;

RETURN ;

};

ERROR ;

};

parallelWriteBuffer: PROC [node: BufferNodeHandle, fileData: FileDataHandle] = {

ENABLE FS.Error => {

By catching and ignoring FS errors, we insure that the write will

later be done in the process of the client so that signals will

look correct.

markNodeNotWritten[fileData: fileData, node: node];

GOTO done;

};

TRUSTED{WriteFilePages[f: fileData.fileHandle, node: node] };

markNodeWritten[fileData: fileData, node: node];

EXITS

done => RETURN

};

ReadAhead: PROC [node: BufferNodeHandle, fileData: FileDataHandle] = {

ENABLE FS.Error => {

on FS errors, invalidate the pre-read

FinishBadPreRead[fileData: fileData, node: node];

GOTO done;

};

bytesToRead: INT ;

fileByte: INT = node.firstFileByteInBuffer ;

IF (bytesToRead ←

MIN[fileData.fileLength - fileByte, node.bufferBytes]) > 0 THEN

TRUSTED{ReadFilePages[f: fileData.fileHandle, from: fileByte,

numPages: PagesForRoundUpBytes[bytesToRead], to: node.buffer,

interval: node.bufferInterval]};

FinishRead[fileData: fileData, node: node, bufferSize: bytesToRead];

EXITS

done => RETURN

};

SetupBuffer: PUBLIC PROC [fileData: FileDataHandle,

fsData: FSDataHandle, fileByte: INT] RETURNS [currentNode: BufferNodeHandle] = {

For write streams, didPut = FALSE if on entry (someone else called CleanupAfterPut).

Arranges buffer so that fileByte (must be buffer-aligned) is the first byte in it.

If buffer is dirty, writes it to file.

Maintains invariants of dataBytesInBuffer, bufferBytes, and

firstFileByteInBuffer in the face of all this. DOES NOT update index.

Called from AdvanceBuffer, SetIndex, SetLength,

StreamFromOpenStream and StreamFromOpenFile.

node: BufferNodeHandle ← fsData.currentNode ;

readAheadNode: BufferNodeHandle;

currentNodeStatus: FileStreamPrivate.NodeStatus;

success: BOOL ← FALSE ;

IF node = NIL THEN node ← fileData.firstBufferNode ;

write buffer if needed

IF node.bufferDirty AND fsData.isWriteStream THEN {

See if there are buffers that must be written sequentially.

FinishWrites[fileData: fileData, fsData: fsData, currentNode: node, parallelWritesOK: TRUE];

Extend file if we are about to write over it.

IF node.dataBytesInBuffer + node.firstFileByteInBuffer > fileData.byteSize THEN {

fileData.byteSize ← node.dataBytesInBuffer + node.firstFileByteInBuffer ;

SetFileSize[fileData.fileHandle, fileData.byteSize] ;

};

IF fileData.validBytesOnDisk < node.dataBytesInBuffer + node.firstFileByteInBuffer THEN

fileData.validBytesOnDisk ← node.dataBytesInBuffer + node.firstFileByteInBuffer ;

WaitForParallelWriteToComplete[fileData: fileData, node: node];

node.status ← needsParallelWrite ;

node.bufferDirty ← FALSE ;

bumpWriteCount[ fileData: fileData];

FileStreamPrivate.StartRequest [ fileData: fileData, node: node ] ;

WaitForOneBufferNotWriting[fileData: fileData];

};

WHILE success # TRUE DO

[success, currentNode, readAheadNode] ←

SetUpNodes[fileData: fileData, fsData: fsData, fileByte: fileByte];

Copy the status out of the node before the tests. Since the status

can change at any time, it would be possible to have none of the

arms of the SELECT executed when there was a pre-read to do.

(This may not be necessary anymore since I changed the way the

SELECT is done, but it doesn't hurt).

IF success THEN {

currentNodeStatus ← currentNode.status ;

SELECT currentNodeStatus FROM

invalid,needsParallelRead,parallelReadActive,needsSequentialRead,sequentialReadActive=>{

IF readAheadNode = NIL THEN {

makeNodeValid[fileData: fileData, node: currentNode ];

}

ELSE {

myPriority: Process.Priority ;

myPriority ← Process.GetPriority[];

Process.SetPriority[Process.priorityForeground];

FileStreamPrivate.StartRequest [ fileData: fileData, node: readAheadNode ];

makeNodeValid[fileData: fileData, node: currentNode ];

Process.SetPriority[myPriority];

};

valid,needsParallelWrite,parallelWriteActive,needsSequentialWrite,sequentialWriteActive=>{

IF readAheadNode # NIL THEN {

FileStreamPrivate.StartRequest [ fileData: fileData, node: readAheadNode ];

};

ENDCASE ;

fsData ← NIL ;

} ELSE {

wait around and try to get a node that is ok to re-use.

Note that just because one becomes free does not mean that SetUpNodes

will find it next time around the loop (another stream may have grabbed it).

If the buffers get into an error state, then nodeNowAvailable returns FALSE

and we use FinishWrites to clean up.

IF NOT nodeNowAvailable[fileData: fileData] THEN

FinishWrites[fileData: fileData, fsData: fsData, currentNode: NIL];

};

ENDLOOP;

};

makeNodeValid: PROC [fileData: FileDataHandle, node: BufferNodeHandle] = {

bytesToRead: INT ;

WHILE node.status # valid DO

IF doTheRead[fileData: fileData, node: node] THEN {

bytesToRead ← MIN[fileData.fileLength - node.firstFileByteInBuffer, node.bufferBytes];

IF fileData.validBytesOnDisk <= node.firstFileByteInBuffer THEN {

Avoid read: the data is trash on disk. We are extending the file anyway.

FinishRead[fileData: fileData, node: node, bufferSize: bytesToRead];

}

ELSE {

IF bytesToRead > 0 THEN TRUSTED{

ReadFilePages[f: fileData.fileHandle,

from: node.firstFileByteInBuffer,

numPages: PagesForRoundUpBytes[bytesToRead],

to: node.buffer,

interval: node.bufferInterval

! FS.Error => {

FinishBadRead[ fileData: fileData, node: node];

};

];

};

FinishRead[fileData: fileData, node: node, bufferSize: bytesToRead];

};

ENDLOOP;

};

doTheRead: ENTRY PROC [fileData: FileDataHandle, node: BufferNodeHandle]

RETURNS [BOOL] = INLINE {

ENABLE UNWIND => NULL;

IF node.status = invalid OR node.status = needsSequentialRead THEN {

node.status ← sequentialReadActive ;

RETURN [TRUE];

};

IF node.status = valid THEN RETURN [ FALSE ] ELSE WAIT fileData.somethingHappened ;

RETURN [ FALSE ] ;

};

SetUpNodes: ENTRY PROC [fileData: FileDataHandle, fsData: FSDataHandle, fileByte: INT]

RETURNS [success: BOOL ← TRUE, currentNode: BufferNodeHandle ← NIL,

nextNode: BufferNodeHandle ← NIL] = {

This procedure runs under the monitor.

It looks for buffers for the current and next nodes.

The node returned in currentNode is the one to use as current. If it is

marked "active", then the caller must fill it before use.

The node nextNode is returned as NIL if no preread is needed. If non-NIL,

the caller should arrange to preread into this buffer.

ENABLE UNWIND => NULL;

nowNode: BufferNodeHandle ← fileData.firstBufferNode;

availableNode: BufferNodeHandle ← NIL ;

availableNodeLRUCount: INT ← 1000000;

maxLRUCount: INT ← 0 ;

node: BufferNodeHandle ;

oldCurrentNode: BufferNodeHandle ← fsData.currentNode ;

oldFirstByteInBuffer: INT = IF fsData.currentNode = NIL THEN -1

ELSE fsData.currentNode.firstFileByteInBuffer;

bufferBytes: INT = nowNode.bufferBytes ;

IF (node ← fsData.currentNode) # NIL THEN node.useCount ← node.useCount-1 ;

IF (node ← fsData.readAheadNode) # NIL THEN node.useCount ← node.useCount-1 ;

fsData.currentNode ← NIL ;

fsData.readAheadNode ← NIL;

UNTIL nowNode = NIL DO

firstByte: INT ← nowNode.firstFileByteInBuffer;

IF nowNode.LRUCount > maxLRUCount THEN maxLRUCount ← nowNode.LRUCount ;

SELECT TRUE FROM

buffer already has correct position in file

firstByte = fileByte => {

currentNode ← nowNode ;

fsData.currentNode ← nowNode ;

nowNode.useCount ← nowNode.useCount+1 ;

};

buffer is next after current

firstByte = fileByte+bufferBytes => {

IF fileData.streamBufferParms.nBuffers > 1 THEN {

nowNode.useCount ← nowNode.useCount+1 ;

fsData.readAheadNode ← nowNode ;

}

ELSE {

IF( nowNode.status = valid OR nowNode.status = invalid)

AND nowNode.useCount = 0 AND

nowNode.LRUCount <= availableNodeLRUCount THEN {

availableNodeLRUCount ← nowNode.LRUCount ;

availableNode ← nowNode ;

};

buffer not "near" stream pointer (tested in above two cases)

and it is not active, and it is not near the other stream (if it exits)

( nowNode.status = valid OR nowNode.status = invalid) AND

nowNode.useCount = 0 AND nowNode.LRUCount <= availableNodeLRUCount => {

availableNodeLRUCount ← nowNode.LRUCount ;

availableNode ← nowNode ;

};

ENDCASE; -- SELECT TRUE

nowNode ← nowNode.nextBufferNode ;

ENDLOOP;

IF currentNode = NIL THEN {

IF availableNode = NIL THEN RETURN[FALSE];

currentNode ← availableNode ;

currentNode.LRUCount ← maxLRUCount + 1 ;

currentNode.useCount ← currentNode.useCount+1 ;

currentNode.status ← invalid ;

currentNode.firstFileByteInBuffer ← fileByte ;

availableNode ← NIL ;

fsData.currentNode ← currentNode ;

};

preread if a sequence has been established.

IF fileData.streamBufferParms.nBuffers > 1 AND

(oldFirstByteInBuffer+bufferBytes = fileByte) AND

(fsData.lastFirstByteInBuffer+bufferBytes = oldFirstByteInBuffer) AND

(fileData.fileLength > fileByte+bufferBytes) AND

(fileData.validBytesOnDisk > fsData.lastFirstByteInBuffer+bufferBytes) THEN {

IF fsData.readAheadNode # NIL THEN {

a node already points to the right place in the file

IF fsData.readAheadNode.status = invalid THEN {

nextNode ← fsData.readAheadNode ;

nextNode.status ← needsParallelRead ;

};

}

ELSE {

IF availableNode = NIL THEN {

availableNodeLRUCount ← 1000000;

nowNode ← fileData.firstBufferNode;

UNTIL nowNode = NIL DO

IF ( nowNode.status = valid OR nowNode.status = invalid) AND

nowNode.useCount = 0 AND

nowNode.LRUCount <= availableNodeLRUCount THEN

availableNode ← nowNode ;

nowNode ← nowNode.nextBufferNode ;

ENDLOOP;

};

IF availableNode # NIL THEN {

nextNode ← availableNode ;

nextNode.status ← needsParallelRead ;

nextNode.LRUCount ← maxLRUCount + 1 ;

nextNode.useCount ← nextNode.useCount+1 ;

nextNode.firstFileByteInBuffer ← fileByte + bufferBytes ;

fsData.readAheadNode ← nextNode ;

};

fsData.lastFirstByteInBuffer ← oldFirstByteInBuffer ;

fsData ← NIL ;

};

ForceOut: PROC [fsData: FSDataHandle] = {

Called from Flush for write streams, or Close for any stream

This is the only proc that sets byte length, and only proc that finishes trans.

fileData: FileDataHandle = fsData.fileData;

node: BufferNodeHandle ← fsData.currentNode ;

IF fsData.isWriteStream THEN CleanupAfterPut[fileData: fileData, selfData: fsData];

IF node.dataBytesInBuffer + node.firstFileByteInBuffer > fileData.byteSize THEN {

fileData.byteSize ← node.dataBytesInBuffer + node.firstFileByteInBuffer ;

SetFileSize[fileData.fileHandle, fileData.byteSize] ;

};

IF fileData.validBytesOnDisk < node.dataBytesInBuffer + node.firstFileByteInBuffer THEN

fileData.validBytesOnDisk ← node.dataBytesInBuffer + node.firstFileByteInBuffer ;

This call does the writes under the monitor lock.

This should be true for read streams since you want to stop the writer from

dirting buffers.

For write streams, you could get by without the lock provided you were

extremely careful about a ForceOut on the read stream, a StreamFromOpenStream

(it will allocate more buffers). The easy way out is to use the monitor.

FinishWrites[ fileData: fileData, fsData: fsData, currentNode: NIL];

IF fsData.isWriteStream AND fileData.accessRights = $write AND

fileData.fileLength # fileData.byteLength THEN {

fileData.byteLength ← fileData.fileLength;

fileData.fileHandle.SetByteCountAndCreatedTime[fileData.byteLength] };

fsData ← NIL ;

};

SaveStreamError: PUBLIC PROCEDURE [self: STREAM, error: FS.ErrorDesc] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => fsData.FSErrorDesc ← error ;

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

ErrorFromStream: PUBLIC PROCEDURE [self: STREAM] RETURNS [FS.ErrorDesc] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => RETURN [fsData.FSErrorDesc];

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

SetStreamClassData: PUBLIC PROCEDURE [self: STREAM, data: REF ANY] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => fsData.StreamClassData ← data ;

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

GetStreamClassData: PUBLIC PROCEDURE [self: STREAM] RETURNS [data: REF ANY] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => RETURN [fsData.StreamClassData] ;

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

SetFinalizationProc: PUBLIC PROCEDURE [self: STREAM, proc: FileStream.FinalizationProc] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => fsData.FinalizationProc ← proc ;

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

GetFinalizationProc: PUBLIC PROCEDURE [self: STREAM] RETURNS

[proc: FileStream.FinalizationProc] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => RETURN [fsData.FinalizationProc] ;

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

bufferBad: PROCEDURE [status: FileStreamPrivate.NodeStatus] RETURNS [bad: BOOL] = INLINE {

SELECT status FROM

valid, needsParallelWrite, parallelWriteActive, needsSequentialWrite, sequentialWriteActive => RETURN[FALSE];

ENDCASE => RETURN[TRUE];

};

Talking to FS

ReadFilePages: PROC [f: FS.OpenFile, from: ByteNumber, numPages: INT, to: LONG POINTER, interval: VM.Interval] = INLINE {

p: PageNumber = from/bytesPerFilePage;

TRUSTED{f.Read[from: p, nPages: numPages, to: to]};

VM.MakeUnchanged[interval];

};

WriteFilePages: PROC [f: FS.OpenFile, node: BufferNodeHandle] = {

numPages: INT = PagesForRoundUpBytes[node.dataBytesInBuffer];

interval: VM.Interval = node.bufferInterval;

vmPage: VM.PageNumber ;

firstPage: VM.PageNumber ← 0 ;

lastPage: VM.PageNumber ← 0 ;

firstPage ← interval.page;

lastPage ← interval.page+numPages-1;

IF node.firstFileByteInBuffer < 0 THEN RETURN; -- not a valid buffer to write

FOR vmPage IN [interval.page..interval.page+numPages) DO

IF VM.State[vmPage].dataState # unchanged THEN {

lastPage ← vmPage ;

IF firstPage = 0 THEN firstPage ← vmPage ;

};

ENDLOOP;

IF firstPage # 0 THEN {

ENABLE UNWIND => VM.MakeChanged[interval];

from: LONG POINTER ;

TRUSTED {from ← VM.AddressForPageNumber[firstPage]} ;

VM.MakeUnchanged[interval];

f.Write[from: from, nPages: lastPage-firstPage+1,

to: (node.firstFileByteInBuffer +

VM.BytesForPages[firstPage-interval.page])/bytesPerFilePage];

};

SetFileSize: PROC [f: FS.OpenFile, byteSize: ByteCount] = {

f.SetPageCount[pages: (byteSize+bytesPerFilePage-1)/bytesPerFilePage];

};

GetFileLock: PROC [f: FS.OpenFile] RETURNS [FS.Lock] = {

RETURN [f.GetInfo[].lock]

};

ProcHandleFromAccessRights: PUBLIC PROC [accessRights: FS.Lock]

RETURNS [ procs: FileStreamPrivate.ProcHandle] = {

SELECT accessRights FROM

read => RETURN [nucleusFileIOReadProcs];

write => RETURN [nucleusFileIOAllProcs];

ENDCASE => RETURN[NIL];

};

Stream creation

StreamFromOpenFile: PUBLIC PROC [openFile: FS.OpenFile, accessRights: FS.Lock, initialPosition: FS.InitialPosition, streamOptions: FS.StreamOptions, streamBufferParms: FS.StreamBufferParms, extendFileProc: FS.ExtendFileProc] RETURNS [stream: STREAM] = {

no monitors are needed in this code since, until this proc returns,

no other code can refer to the streams

pageAllocation: PageCount;

byteLength: ByteCount;

fileName: ROPE = openFile.GetName[].fullFName;

fsData: FSDataHandle;

fsDataFile: FileDataHandle ;

node: BufferNodeHandle ;

Index must always be less than 64K, so we have to clip off a page from the max.

IF streamBufferParms.vmPagesPerBuffer = 128 THEN

streamBufferParms.vmPagesPerBuffer ← 127;

IF accessRights = $write AND GetFileLock[openFile] # $write THEN

FSBackdoor.ProduceError[wrongLock, fileName];

[pages: pageAllocation, bytes: byteLength] ← openFile.GetInfo[];

fsData ← NEW[Data ← []];

fsDataFile ← NEW[FileData ← [

fileName: fileName,

accessRights: accessRights,

fileLength: byteLength,

fileHandle: openFile,

streamBufferParms: streamBufferParms,

extendFileProc: extendFileProc,

streamOptions: streamOptions,

byteLength: byteLength,

byteSize: pageAllocation*bytesPerFilePage,

validBytesOnDisk: byteLength]

];

IF fsDataFile.byteLength > fsDataFile.byteSize THEN ERROR;

fsData.fileData ← fsDataFile ;

fsDataFile.firstBufferNode ← node ←

CreateBufferSpace[streamBufferParms.vmPagesPerBuffer, accessRights];

FOR i: INT IN [2..streamBufferParms.nBuffers] DO

node.nextBufferNode ←

CreateBufferSpace[streamBufferParms.vmPagesPerBuffer, accessRights];

node ← node.nextBufferNode ;

ENDLOOP;

stream ← IO.CreateStream[FileStreamPrivate.ProcHandleFromAccessRights[accessRights], fsData];

IOUtils.StoreData[self: stream, key: $Name, data: fsDataFile.fileName];

IF accessRights = $write

THEN {

fsDataFile.writeStreamData ← fsData ;

fsData.isWriteStream ← TRUE ;

IF fsDataFile.byteSize = 0 THEN {

fsDataFile.byteSize ← NewByteSize[fsDataFile.byteSize];

SetFileSize[fsDataFile.fileHandle, fsDataFile.byteSize];

};

}

ELSE {

fsDataFile.firstReadStream ← fsData;

};

IF initialPosition = start

THEN {

[] ← FileStreamPrivate.SetupBuffer[fileData: fsDataFile, fsData: fsData, fileByte: 0]

}

ELSE {

initialPosition = end

node ← FileStreamPrivate.SetupBuffer[fileData: fsDataFile,

fsData: fsData, fileByte: PageContainingLastByte[fsDataFile.fileLength]];

fsData.index ← node.dataBytesInBuffer;

};

IF streamOptions[tiogaRead] AND byteLength > 0 THEN {

isTioga: BOOL; len: INT;

[yes: isTioga, len: len] ← IsThisThingATiogaFile[stream];

IF isTioga THEN {

IF accessRights = $read

THEN {

make length look changed by sneaky call to SetLength (not in stream procs).

since stream is opened for read only, this call won't change the length in the file.

FileStream.SetLength[stream, len];

fsDataFile.tiogaReader ← TRUE

}

ELSE {

you can't incrementally update a Tioga file with IO!

stream.Close[];

FSBackdoor.ProduceError[cantUpdateTiogaFile, fileName];

}

};

IF FileStreamPrivate.DoFinalization THEN {

FSLock.RecordREF[fsData];

SafeStorage.EnableFinalization[fsData];

};

fsData.ConvertFStoIOErrors ← TRUE;

fsData ← NIL ;

RETURN[stream];

};--StreamFromOpenFile

PageContainingLastByte: PROC [byteLen: INT] RETURNS [INT] = INLINE {

IF byteLen = 0 THEN RETURN[0] ELSE {

byteLen ← byteLen - 1;

LOOPHOLE[byteLen, LongNumber[pair]].lo ←

BITAND[LOOPHOLE[byteLen, LongNumber[pair]].lo, clearLowBits];

RETURN[byteLen]

};

OpenFileFromStream: PUBLIC PROC [self: STREAM] RETURNS [FS.OpenFile] = {

WITH self.streamData SELECT FROM

fsData: FSDataHandle => RETURN [fsData.fileData.fileHandle];

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

StreamFromOpenStream: PUBLIC PROC [self: STREAM] RETURNS [stream: STREAM] = {

newData: FSDataHandle ;

filePos: INT ;

WITH self.streamData SELECT FROM

selfData: FSDataHandle => {

fileData: FileDataHandle = selfData.fileData;

IF NOT selfData.isWriteStream OR fileData.firstReadStream # NIL THEN

FSBackdoor.ProduceError [code: notImplemented,

explanation: "self is not a write stream, or there already is a read stream"];

newData ← NEW[Data ← [ ] ];

newData.fileData ← fileData ;

stream ← IO.CreateStream[FileStreamPrivate.ProcHandleFromAccessRights[$read],

newData];

IOUtils.StoreData[self: stream, key: $Name, data: fileData.fileName];

fileData.firstReadStream ← newData;

filePos ← SetUpClonedStream[fileData: fileData, fsData: selfData];

[] ← FileStreamPrivate.SetupBuffer[fileData: fileData, fsData: newData,

fileByte: selfData.currentNode.firstFileByteInBuffer] ;

newData.index ← selfData.index ;

IF FileStreamPrivate.DoFinalization THEN {

FSLock.RecordREF[newData];

SafeStorage.EnableFinalization[newData];

};

newData.ConvertFStoIOErrors ← TRUE;

newData ← NIL ;

};

ENDCASE => ERROR IO.Error[NotImplementedForThisStream, self];

};

SetUpClonedStream: ENTRY PROC [fileData: FileDataHandle, fsData: FSDataHandle] RETURNS [filePos: INT]= {

ENABLE UNWIND => NULL;

node: BufferNodeHandle ← fileData.firstBufferNode;

Find last node

UNTIL node.nextBufferNode = NIL DO

node ← node.nextBufferNode;

ENDLOOP ;

Allocate some more nodes.

FOR i:INT IN [1..fileData.streamBufferParms.nBuffers] DO

node.nextBufferNode ←

CreateBufferSpace[fileData.streamBufferParms.vmPagesPerBuffer, read];

node ← node.nextBufferNode;

ENDLOOP;

fileData.numberOfStreams ← fileData.numberOfStreams + 1 ;

filePos ← fsData.index + fsData.currentNode.firstFileByteInBuffer ;

fsData ← NIL ;

};

Buffer management

CreateBufferSpace: PROC [vmPagesPerBuffer: INT [1 .. 128], accessRights: FS.Lock] RETURNS [BufferNodeHandle] = {

vmPages: INT ← MIN[vmPagesPerBuffer, maxVMPagesPerBuffer] ;

newBuffer: BufferNodeHandle ← NEW[BufferNode];

allocateCounter: INT ← 0;

newBuffer.bufferInterval ← VM.Allocate[count: vmPages

! VM.CantAllocate => {

We cannot accept anything but the right size interval.

The program assumes that all buffers are the same size.

Process.Pause[4];

IF (allocateCounter ← allocateCounter + 1) < 100 THEN RETRY;

};

];

TRUSTED {

newBuffer.buffer ← VM.AddressForPageNumber[newBuffer.bufferInterval.page];

IF accessRights = write THEN VM.SwapIn[newBuffer.bufferInterval];

};

newBuffer.bufferBytes ← VM.BytesForPages[pages: vmPages];

RETURN[newBuffer]

};

NewByteSize: PROC [byteCount: ByteCount] RETURNS [ByteCount] = {

RETURN [byteCount+5120];

};

Tioga

IsThisThingATiogaFile: PROC [h: STREAM] RETURNS [yes: BOOL, len: INT] = {

pos, length: INT;

{ -- block so EXITS code can use pos, len, and length.

controlHeaderId: ARRAY [0..fileIdSize) OF CHAR = [235C,312C];

controlTrailerId: ARRAY [0..fileIdSize) OF CHAR = [205C,227C];

commentHeaderId: ARRAY [0..fileIdSize) OF CHAR = [0C,0C];

fileIdSize: NAT = 2;

numTrailerLengths: NAT = 3; -- <file-props-length> <data-length> <file-length>

endSize: NAT = fileIdSize+numTrailerLengths*4; -- trailer plus three lengths

ReadLen: PROC [h: STREAM] RETURNS [INT] = {

start: PACKED ARRAY [0..3] OF CHARACTER;

start[0] ← h.GetChar[]; start[1] ← h.GetChar[];

start[2] ← h.GetChar[]; start[3] ← h.GetChar[];

RETURN [LOOPHOLE[start]] };

commentStart, commentLen, propsLen, controlLen, controlEnd: INT;

pos ← h.GetIndex[]; -- save position to restore later

length ← h.GetLength[]; -- length including any trailer stuff

controlEnd ← length-endSize; -- where the trailer info starts

IF controlEnd <= 0 THEN GOTO fail; -- too small

h.SetIndex[controlEnd]; -- set up to read the trailer

FOR i:NAT IN [0..fileIdSize) DO -- read the controlTrailerId

IF h.GetChar[] # controlTrailerId[i] THEN GOTO fail;

ENDLOOP;

IF (propsLen ← ReadLen[h]) NOT IN [0..controlEnd) THEN GOTO fail;

IF (commentStart ← ReadLen[h]) NOT IN [0..controlEnd) THEN GOTO fail;

IF ReadLen[h] # length THEN GOTO fail;

IF commentStart > 0

THEN { -- may have padded text with a null

h.SetIndex[commentStart-1];

len ← IF h.GetChar[]=0C THEN commentStart-1 ELSE commentStart }

ELSE h.SetIndex[len ← commentStart];

FOR i:NAT IN [0..fileIdSize) DO -- read the commentHeaderId

IF h.GetChar[] # commentHeaderId[i] THEN GOTO fail;

ENDLOOP;

commentLen ← ReadLen[h]; -- the length of the comment section

IF commentStart+commentLen NOT IN [0..controlEnd) THEN GOTO fail;

h.SetIndex[commentStart+commentLen]; -- go to start of control info

FOR i:NAT IN [0..fileIdSize) DO -- check the controlHeaderId

IF h.GetChar[] # controlHeaderId[i] THEN GOTO fail;

ENDLOOP;

controlLen ← ReadLen[h]; -- the length of the control section

IF commentStart+commentLen+controlLen # length THEN GOTO fail;

GOTO succeed;

EXITS

fail => { h.SetIndex[pos]; RETURN [FALSE, length] };

succeed => { h.SetIndex[pos]; RETURN [TRUE, len] };

};

Procedure records (never modified)

nucleusFileIOReadProcs: PUBLIC FileStreamPrivate.ProcHandle = IO.CreateStreamProcs[

variety: $input, class: $File,

getChar: GetChar,

endOf: EndOf,

charsAvail: CharsAvail,

getBlock: GetBlock,

unsafeGetBlock: UnsafeGetBlock,

putChar: NIL, -- not implemented

putBlock: NIL, -- call PutChar

unsafePutBlock: NIL, -- call PutChar

flush: Flush,

reset: Reset,

close: Close,

getIndex: GetIndex,

setIndex: SetIndex,

backup: Backup,

getLength: GetLength

];

nucleusFileIOAllProcs: PUBLIC FileStreamPrivate.ProcHandle = IO.CreateStreamProcs[

variety: $inputOutput, class: $File,

getChar: GetChar,

endOf: EndOf,

charsAvail: CharsAvail,

getBlock: GetBlock,

unsafeGetBlock: UnsafeGetBlock,

putChar: PutChar,

putBlock: PutBlock,

unsafePutBlock: UnsafePutBlock,

flush: Flush,

reset: Reset,

close: Close,

getIndex: GetIndex,

setIndex: SetIndex,

backup: Backup,

getLength: GetLength,

setLength: SetLength,

eraseChar: EraseChar

];

Finalization code

fQ: SafeStorage.FinalizationQueue;

Finalize: PROC = BEGIN

fsData: FSDataHandle ← NARROW [ SafeStorage.FQNext[fQ] ];

streamIsClosed: BOOL ;

forceCleanUp: BOOL ← FALSE;

FSLock.RemoveREF[fsData];

streamIsClosed ← fsData.streamIsClosed ;

IF NOT fsData.streamIsClosed AND fsData.currentNode # NIL THEN {

ForceOut[ fsData: fsData

! FS.Error => {forceCleanUp ← TRUE; CONTINUE}];

CloseFileDataForStream[fileData: fsData.fileData, fsData: fsData, forceCleanUp: forceCleanUp

! FS.Error => CONTINUE];

};

IF fsData.isWriteStream THEN {

IF fsData.fileData.writeStreamData = fsData THEN

fsData.fileData.writeStreamData ← NIL ;

}

ELSE {

IF fsData.fileData.firstReadStream = fsData THEN

fsData.fileData.firstReadStream ← NIL ;

};

IF fsData.FinalizationProc # NIL THEN {

fsData.FinalizationProc [ openFile: fsData.fileData.fileHandle,

data: fsData.StreamClassData,

closed: streamIsClosed];

};

fsData ← NIL;

ENDLOOP;

END;

Initialization

start code: start up finalization stuff

IF FileStreamPrivate.DoFinalization THEN {

fQ ← SafeStorage.NewFQ[];

SafeStorage.EstablishFinalization[CODE[Data], 2, fQ];

TRUSTED { Process.Detach[FORK Finalize[]] };

};

END.

CHANGE LOG

Created by MBrown on June 22, 1983 10:08 am

By editing FileIOAlpineImpl.

Changed by MBrown on August 19, 1983 2:44 pm

Close FS file when closing stream (this should really be an option).

Changed by Birrell on August 23, 1983 3:14 pm

In SetFileSize: byteSize/bytesPerFilePage -> (byteSize+bytesPerFilePage-1)/bytesPerFilePage.

Changed by MBrown on August 25, 1983 1:18 pm

In SetIndex: fsData.byteSize < firstBufferByte+fsData.dataBytesInBuffer -> fsData.byteSize <= firstBufferByte+fsData.dataBytesInBuffer. Implemented GetFileLock (was stubbed waiting for FS). In StreamFromOpenFile, if stream open for write and file has no pages, extend it.

Changed by MBrown on September 17, 1983 8:41 pm

Conversion to new IO interface.

Changed by Hagmann on November 22, 1983 4:29 pm

Implement multiple-page stream buffer.

Implement coupled read and write streams on same open file.

Changed data structures in FileStreamPrivate, and fixed references in this module

to the new data structures. This meant changes to nearly every routine.

Close file during StreamOpen if an error occurs in StreamFromOpenFile.

Implement streamOptions and streamBufferParms features.

Added finalization.

Changed name from FileIOFSImpl.

Split out create code to make FileStreamCreateImpl smaller since compiler blows up in pass 3

Changed by Hagmann on November 28, 1983 12:00 pm

Fixed EndOf bug for multiple streams.

Added test for DoFinalization to enable FileStream testing without making a boot file

Changed by Hagmann on December 6, 1983 4:52 pm

Added code for process cache

Changed by Hagmann on December 27, 1983 4:52 pm

Fixed bug (reported by Plass, found by Nix) in unsafegetblock and unsafeputblock

for blocks > 32K.

Changed by Hagmann on December 29, 1983 12:54 pm

Fixed bug (reported and new code by Plass) in AddNAT

Changed by Hagmann on January 3, 1984 11:34 am

Added conditional conversion for FS to IO Errors in convertFStoIOError.

This makes FS errors that occur during stream open appear as FS errors, not IO errors

(fixes bug reported by Nix).

Changed by Hagmann on January 10, 1984 9:04 am

Fixed stream hung problem reported by Willie-Sue and Stewart. SetupBuffer was treating

buffers that are in some stage of writing was not valid. It would wait for them to become

valid (due to a parallel read), or do sequential reads if the parallel read failed. This caused

buffers that were writing in parallel that had a write error to wait forever (the node never

became valid or had a parallel read error).

Changed by Hagmann on January 20, 1984 5:11 pm

Added logic to use the changed/unchanged VM information for writing to disk

Changed by Hagmann on February 6, 1984 2:19:10 pm PST

Added check for read-only streams to ForceOut. This is to avoid a bug

is the changed/unchanged VM information after a rollback (some pages looked

changed for a read-only file).

Changed by Hagmann on May 9, 1984 9:12:21 am PDT

Added check to ensure that read aheads complete before the buffer is deallocated by VM

during a close. This fixes a problem encountered by Frank Crow while running the TSetter,

and narrowed down to FileStream by Willie-Sue.

Changed by December 18, 1984 5:40:06 pm PST

Added UNWIND in Close's call to ForceOut, and the forceCleanUp argument to CloseFileDataForStream. Reformatted file.

Bob Hagmann January 31, 1985 5:41:22 pm PST

remove use of FSLock

changes to: Finalize

Bob Hagmann May 14, 1985 11:09:42 am PDT

put back in use of FSLock

changes to: Finalize, Initialization

Bob Hagmann September 18, 1985 07:06:15 PDT

WriteFilePages and FinishWrites changes to avoid writing pages improperly after shortening the file

Bob Hagmann October 25, 1985 2:02:44 pm PDT

WriteFilePages to mark interval unchanged after write. Fixed subtile finalization bug

Bob Hagmann June 9, 1986 5:10:02 pm PDT

checked for node invalid on all get/put entries (previous error can get us into trouble)
merged in FileStreamCreateImpl

Change log for FileStreamCreateImpl:

Created by Hagmann on November 22, 1983 4:30 pm

By cutting this out of FSFileIOImpl.

Changed by Hagmann on November 28, 1983 12:01 pm

Added test for DoFinalization to enable FileStream testing without making a boot file

Changed by Hagmann on December 6, 1983 4:52 pm

Removed code for process cache

Changed by Hagmann on January 3, 1984 11:33 am

Added enable for ConvertFStoIOErrors.

Russ Atkinson (RRA) May 9, 1985 1:56:42 pm PDT

Added VM.SwapIn for write buffer pages to avoid bogus page faults

changes to: StreamFromOpenFile, SetUpClonedStream, CreateBufferSpace

Bob Hagmann May 14, 1985 11:10:58 am PDT

put back in use of FSLock

changes to: StreamFromOpenFile, StreamFromOpenStream

Bob Hagmann November 13, 1986 1:02:46 pm PST

modified to allow multiple parallel writes

Doug Wyatt, December 12, 1986 5:18:31 pm PST

changed Basics.LongNumber[num] to Basics.LongNumber[pair]

changes to: RoundUpToPages, PageContainingLastByte