Leaked source code of windows server 2003
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

1317 lines
32 KiB

/*++
Copyright (c) 1998 Microsoft Corporation
Module Name:
utilities.cpp
Abstract:
SIS Groveler utility functions
Authors:
Cedric Krumbein, 1998
Environment:
User Mode
Revision History:
--*/
#include "all.hxx"
//
// Remove a trailing slash if it exists from the given name
//
void TrimTrailingChar(
PWCHAR name,
WCHAR ch)
{
int slen = wcslen(name);
if ((slen > 0) && (name[slen-1] == ch))
name[slen-1] = 0;
}
/*****************************************************************************/
// GetPerformanceTime() converts the time interval
// measured using QueryPerformanceCounter() into milliseconds.
PerfTime GetPerformanceTime()
{
LARGE_INTEGER count;
QueryPerformanceCounter(&count);
return (PerfTime)count.QuadPart;
}
/*****************************************************************************/
// PerformanceTimeToMSec() converts the time interval measured
// using QueryPerformanceCounter() into milliseconds.
// PerformanceTimeToUSec() converts it into microseconds.
static DOUBLE frequency = 0.0;
DWORD PerformanceTimeToMSec(PerfTime timeInterval)
{
if (frequency == 0.0) {
LARGE_INTEGER intFreq;
QueryPerformanceFrequency(&intFreq);
frequency = (DOUBLE)intFreq.QuadPart;
}
return (DWORD)((DOUBLE)timeInterval * 1000.0 / frequency);
}
LONGLONG PerformanceTimeToUSec(PerfTime timeInterval)
{
if (frequency == 0.0) {
LARGE_INTEGER intFreq;
QueryPerformanceFrequency(&intFreq);
frequency = (DOUBLE)intFreq.QuadPart;
}
return (LONGLONG)((DOUBLE)timeInterval * 1000000.0 / frequency);
}
/*****************************************************************************/
// GetTime() returns the current file time.
DWORDLONG GetTime()
{
SYSTEMTIME systemTime;
FILETIME fileTime;
ULARGE_INTEGER time;
BOOL success;
GetSystemTime(&systemTime);
success = SystemTimeToFileTime(&systemTime, &fileTime);
ASSERT_ERROR(success);
time.HighPart = fileTime.dwHighDateTime;
time.LowPart = fileTime.dwLowDateTime;
return time.QuadPart;
}
/*****************************************************************************/
// PrintTime() converts the supplied file time into a printable string.
TCHAR *PrintTime(
TCHAR *string,
DWORDLONG time)
{
FILETIME fileTime;
SYSTEMTIME systemTime;
DWORD strLen;
BOOL success;
fileTime.dwHighDateTime = ((ULARGE_INTEGER *)&time)->HighPart;
fileTime.dwLowDateTime = ((ULARGE_INTEGER *)&time)->LowPart;
success = FileTimeToSystemTime(&fileTime, &systemTime);
ASSERT_ERROR(success);
strLen = _stprintf(string, _T("%02hu/%02hu/%02hu %02hu:%02hu:%02hu.%03hu"),
systemTime.wYear % 100,
systemTime.wMonth,
systemTime.wDay,
systemTime.wHour,
systemTime.wMinute,
systemTime.wSecond,
systemTime.wMilliseconds);
ASSERT(strLen == 21);
return string;
}
/*****************************************************************************/
// GetParentName() extracts the parent directory
// name out of a full-path file name.
BOOL GetParentName(
const TCHAR *fileName,
TFileName *parentName)
{
DWORD hi, lo;
ASSERT(fileName != NULL);
ASSERT(parentName != NULL);
if (fileName[0] == _T('\\'))
lo = 1;
else if (_istalpha(fileName[0])
&& fileName[1] == _T(':')
&& fileName[2] == _T('\\'))
lo = 3;
else
return FALSE;
hi = _tcslen(fileName) - 1;
if (hi < lo)
hi = lo;
else
for (; hi > lo; hi--)
if (fileName[hi] == _T('\\'))
break;
parentName->assign(fileName, hi);
return TRUE;
}
/*****************************************************************************/
// GetFileID gets the file's ID given its name.
DWORDLONG GetFileID(
const TCHAR *volName,
const TCHAR *fileName)
{
HANDLE fileHandle;
TCHAR *fname = NULL;
BOOL freeFname = FALSE;
BY_HANDLE_FILE_INFORMATION fileInfo;
ULARGE_INTEGER fileID;
ASSERT(fileName != NULL && fileName[0] != _T('\0'));
if (volName) {
DWORD bfsz = wcslen(volName) + wcslen(fileName) + 1;
fname = new WCHAR[bfsz];
(void)StringCchCopy(fname,bfsz,volName);
(void)StringCchCat(fname,bfsz,fileName);
freeFname = TRUE;
} else {
fname = (TCHAR *)fileName;
}
__try {
fileHandle = CreateFile(
fname,
0,
FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE,
NULL,
OPEN_EXISTING,
FILE_FLAG_BACKUP_SEMANTICS,
NULL);
if (fileHandle == INVALID_HANDLE_VALUE) {
fileID.QuadPart = 0;
__leave;
}
if (GetFileInformationByHandle(fileHandle, &fileInfo)) {
fileID.HighPart = fileInfo.nFileIndexHigh;
fileID.LowPart = fileInfo.nFileIndexLow;
} else {
fileID.QuadPart = 0;
}
CloseHandle(fileHandle);
} __finally {
if (freeFname) {
delete[] fname;
}
}
return fileID.QuadPart;
}
/*****************************************************************************/
// GetFileName gets the file's name given either
// an open handle to the file or the file's ID.
BOOL GetFileName(
HANDLE fileHandle,
TFileName *tFileName)
{
IO_STATUS_BLOCK ioStatusBlock;
NTSTATUS ntStatus;
for (int i = 2; i > 0; --i) {
if (tFileName->nameLenMax < 8) // sanity check
tFileName->resize();
ntStatus = NtQueryInformationFile(
fileHandle,
&ioStatusBlock,
tFileName->nameInfo,
tFileName->nameInfoSize,
FileNameInformation);
if (ntStatus != STATUS_BUFFER_OVERFLOW)
break;
ASSERT(tFileName->nameInfo->FileNameLength > tFileName->nameInfoSize - sizeof(ULONG));
tFileName->resize(tFileName->nameInfo->FileNameLength / sizeof(WCHAR) + 1);
}
if (ntStatus != STATUS_SUCCESS)
return FALSE;
tFileName->nameLen = tFileName->nameInfo->FileNameLength / sizeof(WCHAR);
tFileName->name[tFileName->nameLen] = _T('\0');
return TRUE;
}
BOOL GetFileName(
HANDLE volumeHandle,
DWORDLONG fileID,
TFileName *tFileName)
{
UNICODE_STRING fileIDString;
OBJECT_ATTRIBUTES objectAttributes;
IO_STATUS_BLOCK ioStatusBlock;
HANDLE fileHandle;
NTSTATUS ntStatus;
BOOL success;
fileIDString.Length = sizeof(DWORDLONG);
fileIDString.MaximumLength = sizeof(DWORDLONG);
fileIDString.Buffer = (WCHAR *)&fileID;
objectAttributes.Length = sizeof(OBJECT_ATTRIBUTES);
objectAttributes.RootDirectory = volumeHandle;
objectAttributes.ObjectName = &fileIDString;
objectAttributes.Attributes = OBJ_CASE_INSENSITIVE;
objectAttributes.SecurityDescriptor = NULL;
objectAttributes.SecurityQualityOfService = NULL;
ntStatus = NtCreateFile(
&fileHandle,
GENERIC_READ,
&objectAttributes,
&ioStatusBlock,
NULL,
0,
FILE_SHARE_VALID_FLAGS,
FILE_OPEN,
FILE_OPEN_BY_FILE_ID |
FILE_OPEN_REPARSE_POINT |
FILE_NO_INTERMEDIATE_BUFFERING,
NULL,
0);
if (ntStatus != STATUS_SUCCESS)
return FALSE;
success = GetFileName(fileHandle, tFileName);
NtClose(fileHandle);
return success;
}
/*****************************************************************************/
// GetCSIndex() returns the SIS reparse point's common store
// index. The file handle must point to an open reparse point.
BOOL GetCSIndex(
HANDLE fileHandle,
CSID *csIndex)
{
IO_STATUS_BLOCK ioStatusBlock;
BYTE buffer[MAXIMUM_REPARSE_DATA_BUFFER_SIZE];
REPARSE_DATA_BUFFER *reparseBuffer;
SI_REPARSE_BUFFER *sisReparseBuffer;
ASSERT(fileHandle != NULL);
ASSERT(csIndex != NULL);
if (NtFsControlFile(
fileHandle,
NULL,
NULL,
NULL,
&ioStatusBlock,
FSCTL_GET_REPARSE_POINT,
NULL,
0,
buffer,
MAXIMUM_REPARSE_DATA_BUFFER_SIZE) != STATUS_SUCCESS) {
memset(csIndex, 0, sizeof(CSID));
return FALSE;
}
reparseBuffer = (REPARSE_DATA_BUFFER *)buffer;
if (reparseBuffer->ReparseTag != IO_REPARSE_TAG_SIS) {
memset(csIndex, 0, sizeof(CSID));
return FALSE;
}
sisReparseBuffer = (SI_REPARSE_BUFFER *)
reparseBuffer->GenericReparseBuffer.DataBuffer;
if (sisReparseBuffer->ReparsePointFormatVersion != SIS_REPARSE_BUFFER_FORMAT_VERSION) {
memset(csIndex, 0, sizeof(CSID));
return FALSE;
}
*csIndex = sisReparseBuffer->CSid;
return TRUE;
}
/*****************************************************************************/
// GetCSName() converts the common store
// index into a dynamically allocated string.
TCHAR *GetCSName(CSID *csIndex)
{
TCHAR *rpcStr;
RPC_STATUS rpcStatus;
ASSERT(csIndex != NULL);
rpcStatus = UuidToString(csIndex, (unsigned short **)&rpcStr);
if (rpcStatus != RPC_S_OK) {
ASSERT(rpcStr == NULL);
return NULL;
}
ASSERT(rpcStr != NULL);
return rpcStr;
}
/*****************************************************************************/
// FreeCSName frees the string allocated by GetCSName().
VOID FreeCSName(TCHAR *rpcStr)
{
RPC_STATUS rpcStatus;
ASSERT(rpcStr != NULL);
rpcStatus = RpcStringFree((unsigned short **)&rpcStr);
ASSERT(rpcStatus == RPC_S_OK);
}
/*****************************************************************************/
// Checksum() generates a checksum on the data supplied in the buffer.
// The checksum function used is selected at compile-time; currently
// the 131-hash and the "Bill 32" hash functions are implemented.
#define HASH131
// #define BILL32HASH
Signature Checksum(
const VOID *buffer,
DWORD bufferLen,
DWORDLONG offset,
Signature firstWord)
{
Signature *bufferPtr,
word,
signature;
DWORD numWords,
numBytes,
rotate;
ASSERT(buffer != NULL);
bufferPtr = (Signature *)buffer;
numWords = bufferLen / sizeof(Signature);
numBytes = bufferLen % sizeof(Signature);
signature = firstWord;
#ifdef BILL32HASH
rotate = (DWORD)(offset / sizeof(Signature) % (sizeof(Signature)*8-1));
#endif
while (numWords-- > 0) {
word = *bufferPtr++;
#ifdef HASH131
signature = signature * 131 + word;
#endif
#ifdef BILL32HASH
signature ^= ROTATE_RIGHT(word, rotate);
rotate = (rotate+1) % (sizeof(Signature)*8-1);
#endif
}
if (numBytes > 0) {
word = 0;
memcpy(&word, bufferPtr, numBytes);
#ifdef HASH131
signature = signature * 131 + word;
#endif
#ifdef BILL32HASH
signature ^= ROTATE_RIGHT(word, rotate);
#endif
}
return signature;
}
/*****************************************************************************/
/************************ Table class private methods ************************/
/*****************************************************************************/
DWORD Table::Hash(
const VOID *key,
DWORD keyLen) const
{
USHORT *keyPtr;
DWORD hashValue;
if (keyLen == 0)
return 0;
ASSERT(key != NULL);
if (keyLen <= sizeof(DWORD)) {
hashValue = 0;
memcpy(&hashValue, key, keyLen);
return hashValue;
}
keyPtr = (USHORT *)key;
hashValue = 0;
while (keyLen >= sizeof(USHORT)) {
hashValue = hashValue*37 + (DWORD)*keyPtr++;
keyLen -= sizeof(USHORT);
}
if (keyLen > 0)
hashValue = hashValue*37 + (DWORD)*(BYTE *)keyPtr;
hashValue *= TABLE_RANDOM_CONSTANT;
if ((LONG)hashValue < 0)
hashValue = (DWORD)-(LONG)hashValue;
hashValue %= TABLE_RANDOM_PRIME;
return hashValue;
}
/*****************************************************************************/
DWORD Table::BucketNum(DWORD hashValue) const
{
DWORD bucketNum;
ASSERT(expandIndex < 1U << level);
ASSERT(numBuckets == (1U << level) + expandIndex);
bucketNum = hashValue & ~(~0U << level);
if (bucketNum < expandIndex)
bucketNum = hashValue & ~(~0U << (level+1));
ASSERT(bucketNum < numBuckets);
return bucketNum;
}
/*****************************************************************************/
VOID Table::Expand()
{
TableEntry **oldSlotAddr,
**newSlotAddr,
*oldChain,
*newChain,
*entry;
TableSegment **newDirectory,
*newSegment;
DWORD oldNewMask;
#if DBG
TableEntry *prevChain;
DWORD mask;
#endif
// Increase the directory size if necessary.
ASSERT(directory != NULL);
ASSERT(dirSize >= TABLE_SEGMENT_SIZE);
ASSERT(dirSize % TABLE_SEGMENT_SIZE == 0);
if (numBuckets >= dirSize * TABLE_SEGMENT_SIZE) {
newDirectory = new TableSegment * [dirSize + TABLE_DIR_SIZE];
ASSERT(newDirectory != NULL);
memcpy(newDirectory, directory, sizeof(TableSegment *) * dirSize);
memset(newDirectory+dirSize, 0, sizeof(TableSegment *) * TABLE_DIR_SIZE);
dirSize += TABLE_DIR_SIZE;
delete directory;
directory = newDirectory;
}
// Find the old bucket to be expanded.
ASSERT(expandIndex >> TABLE_SEGMENT_BITS < dirSize);
oldSlotAddr = &directory[expandIndex >> TABLE_SEGMENT_BITS]
->slot[expandIndex & TABLE_SEGMENT_MASK];
ASSERT(oldSlotAddr != NULL);
// Find the new bucket, and create a new segment if necessary.
ASSERT(numBuckets >> TABLE_SEGMENT_BITS < dirSize);
newSegment = directory[numBuckets >> TABLE_SEGMENT_BITS];
if (newSegment == NULL) {
newSegment = new TableSegment;
ASSERT(newSegment != NULL);
memset(newSegment, 0, sizeof(TableSegment));
directory[numBuckets >> TABLE_SEGMENT_BITS] = newSegment;
}
newSlotAddr = &newSegment->slot[numBuckets & TABLE_SEGMENT_MASK];
ASSERT(*newSlotAddr == NULL);
// Relocate entries from the old to the new bucket.
oldNewMask = 1U << level;
oldChain = NULL;
newChain = NULL;
entry = *oldSlotAddr;
#if DBG
prevChain = NULL;
mask = ~(~0U << (level+1));
#endif
while (entry != NULL) {
ASSERT((entry->hashValue & ~(~0U << level)) == expandIndex);
ASSERT( entry->prevChain == prevChain);
// This entry moves to the new bucket.
if ((entry->hashValue & oldNewMask) != 0) {
if (newChain == NULL) {
*newSlotAddr = entry;
entry->prevChain = NULL;
} else {
newChain->nextChain = entry;
entry ->prevChain = newChain;
}
newChain = entry;
ASSERT((entry->hashValue & mask) == numBuckets);
}
// This entry stays in the old bucket.
else {
if (oldChain == NULL) {
*oldSlotAddr = entry;
entry->prevChain = NULL;
} else {
oldChain->nextChain = entry;
entry ->prevChain = oldChain;
}
oldChain = entry;
ASSERT((entry->hashValue & mask) == expandIndex);
}
#if DBG
prevChain = entry;
#endif
entry = entry->nextChain;
}
// Finish off each bucket chain.
if (oldChain == NULL)
*oldSlotAddr = NULL;
else
oldChain->nextChain = NULL;
if (newChain == NULL)
*newSlotAddr = NULL;
else
newChain->nextChain = NULL;
// Adjust the expand index and level, and increment the number of buckets.
if (++expandIndex == 1U << level) {
level++;
expandIndex = 0;
}
numBuckets++;
ASSERT(expandIndex < 1U << level);
ASSERT(numBuckets == (1U << level) + expandIndex);
}
/*****************************************************************************/
VOID Table::Contract()
{
TableEntry **targetSlotAddr,
**victimSlotAddr,
*firstVictimEntry,
*prevChain,
*entry;
TableSegment **newDirectory;
#if DBG
DWORD mask;
#endif
// Adjust the expand index and level, and decrement the number of buckets.
ASSERT(expandIndex < 1U << level);
ASSERT(numBuckets == (1U << level) + expandIndex);
if (expandIndex > 0)
expandIndex--;
else
expandIndex = (1U << --level) - 1;
numBuckets--;
ASSERT(expandIndex < 1U << level);
ASSERT(numBuckets == (1U << level) + expandIndex);
// Find the target and victim buckets.
ASSERT(directory != NULL);
ASSERT(dirSize >= TABLE_SEGMENT_SIZE);
ASSERT(dirSize % TABLE_SEGMENT_SIZE == 0);
targetSlotAddr = &directory[expandIndex >> TABLE_SEGMENT_BITS]
->slot[expandIndex & TABLE_SEGMENT_MASK];
victimSlotAddr = &directory[numBuckets >> TABLE_SEGMENT_BITS]
->slot[numBuckets & TABLE_SEGMENT_MASK];
ASSERT(targetSlotAddr != NULL);
ASSERT(victimSlotAddr != NULL);
// If the victim buffer isn't empty, ...
if ((firstVictimEntry = *victimSlotAddr) != NULL) {
#if DBG
mask = ~(~0U << (level+1));
#endif
ASSERT((firstVictimEntry->hashValue & mask) == numBuckets);
ASSERT( firstVictimEntry->prevChain == NULL);
// ... find the end of the target bucket chain, ...
entry = *targetSlotAddr;
prevChain = NULL;
while (entry != NULL) {
ASSERT((entry->hashValue & mask) == expandIndex);
ASSERT( entry->prevChain == prevChain);
prevChain = entry;
entry = entry->nextChain;
}
// ... then add the victim bucket chain to the end of the target bucket chain.
if (prevChain == NULL)
*targetSlotAddr = firstVictimEntry;
else {
prevChain->nextChain = firstVictimEntry;
firstVictimEntry->prevChain = prevChain;
}
}
// Delete the victim bucket, and delete the victim segment if no buckets remain.
if ((numBuckets & TABLE_SEGMENT_MASK) == 0) {
delete directory[numBuckets >> TABLE_SEGMENT_BITS];
directory[numBuckets >> TABLE_SEGMENT_BITS] = NULL;
} else
*victimSlotAddr = NULL;
// Reduce the size of the directory if necessary.
if (numBuckets <= (dirSize - TABLE_DIR_SIZE) * TABLE_SEGMENT_SIZE
&& dirSize > TABLE_DIR_SIZE) {
dirSize -= TABLE_DIR_SIZE;
newDirectory = new TableSegment * [dirSize];
ASSERT(newDirectory != NULL);
memcpy(newDirectory, directory, sizeof(TableSegment *) * dirSize);
delete directory;
directory = newDirectory;
}
}
/*****************************************************************************/
/************************ Table class public methods *************************/
/*****************************************************************************/
Table::Table()
{
firstEntry = NULL;
lastEntry = NULL;
numEntries = 0;
numBuckets = TABLE_SEGMENT_SIZE;
expandIndex = 0;
level = TABLE_SEGMENT_BITS;
dirSize = TABLE_DIR_SIZE;
directory = new TableSegment * [dirSize];
ASSERT(directory != NULL);
memset(directory, 0, sizeof(TableSegment *) * dirSize);
directory[0] = new TableSegment;
ASSERT(directory[0] != NULL);
memset(directory[0], 0, sizeof(TableSegment));
}
/*****************************************************************************/
Table::~Table()
{
TableEntry *entry,
*prevEntry;
DWORD numSegments,
segmentNum,
count;
entry = firstEntry;
prevEntry = NULL;
count = 0;
while (entry != NULL) {
ASSERT(entry->prevEntry == prevEntry);
prevEntry = entry;
entry = entry->nextEntry;
delete prevEntry->data;
delete prevEntry;
count++;
}
ASSERT(count == numEntries);
numSegments = numBuckets >> TABLE_SEGMENT_BITS;
ASSERT(directory != NULL);
ASSERT(dirSize >= TABLE_SEGMENT_SIZE);
ASSERT(dirSize % TABLE_SEGMENT_SIZE == 0);
ASSERT(numSegments <= dirSize);
for (segmentNum = 0; segmentNum < numSegments; segmentNum++) {
ASSERT(directory[segmentNum] != NULL);
delete directory[segmentNum];
}
delete directory;
}
/*****************************************************************************/
BOOL Table::Put(
VOID *data,
DWORD keyLen)
{
TableEntry **slotAddr,
*prevChain,
*entry;
DWORD hashValue,
bucketNum;
#if DBG
DWORD mask;
#endif
ASSERT(data != NULL);
ASSERT(keyLen > 0);
// Find the bucket for this data.
hashValue = Hash(data, keyLen);
bucketNum = BucketNum(hashValue);
#if DBG
mask = ~(~0U << (bucketNum < expandIndex || bucketNum >= 1U << level
? level+1 : level));
#endif
ASSERT(directory != NULL);
slotAddr = &directory[bucketNum >> TABLE_SEGMENT_BITS]
->slot[bucketNum & TABLE_SEGMENT_MASK];
ASSERT(slotAddr != NULL);
entry = *slotAddr;
prevChain = NULL;
// Look at each entry in the bucket to determine if the data is
// already present. If a matching entry is found, return FALSE.
while (entry != NULL) {
ASSERT((entry->hashValue & mask) == bucketNum);
ASSERT( entry->prevChain == prevChain);
if (hashValue == entry->hashValue
&& keyLen == entry->keyLen
&& memcmp(data, entry->data, keyLen) == 0)
return FALSE;
prevChain = entry;
entry = entry->nextChain;
}
// No entry with matching data was found in this bucket.
// Create a new entry and add it to the end of the bucket chain.
entry = new TableEntry;
ASSERT(entry != NULL);
if (prevChain == NULL) {
*slotAddr = entry;
entry->prevChain = NULL;
} else {
prevChain->nextChain = entry;
entry ->prevChain = prevChain;
}
entry->nextChain = NULL;
// Add the entry to the end of the doubly-linked list.
if (lastEntry == NULL) {
ASSERT(firstEntry == NULL);
ASSERT(numEntries == 0);
firstEntry = entry;
entry->prevEntry = NULL;
} else {
ASSERT(firstEntry != NULL);
ASSERT(numEntries > 0);
lastEntry->nextEntry = entry;
entry ->prevEntry = lastEntry;
}
entry->nextEntry = NULL;
lastEntry = entry;
numEntries++;
// Fill out the entry.
entry->hashValue = hashValue;
entry->keyLen = keyLen;
entry->data = data;
// Expand the table if necessary.
if (numEntries > numBuckets * TABLE_MAX_LOAD) {
Expand();
ASSERT(numEntries <= numBuckets * TABLE_MAX_LOAD);
}
return TRUE;
}
/*****************************************************************************/
VOID *Table::Get(
const VOID *key,
DWORD keyLen,
BOOL erase)
{
TableEntry **slotAddr,
*entry,
*prevChain;
DWORD hashValue,
bucketNum;
VOID *dataPtr;
#if DBG
DWORD mask;
#endif
ASSERT(key != NULL);
ASSERT(keyLen > 0);
// Find the bucket for this data.
hashValue = Hash(key, keyLen);
bucketNum = BucketNum(hashValue);
#if DBG
mask = ~(~0U << (bucketNum < expandIndex || bucketNum >= 1U << level
? level+1 : level));
#endif
ASSERT(directory != NULL);
slotAddr = &directory[bucketNum >> TABLE_SEGMENT_BITS]
->slot[bucketNum & TABLE_SEGMENT_MASK];
ASSERT(slotAddr != NULL);
entry = *slotAddr;
prevChain = NULL;
// Look at each entry in the bucket.
while (entry != NULL) {
ASSERT((entry->hashValue & mask) == bucketNum);
ASSERT( entry->prevChain == prevChain);
if (hashValue == entry->hashValue
&& keyLen == entry->keyLen
&& memcmp(key, entry->data, keyLen) == 0) {
// The entry with matching data has been found.
dataPtr = entry->data;
ASSERT(dataPtr != NULL);
// If erasure is disabled, remove the entry from the doubly-linked list ...
if (erase) {
if (entry->prevEntry == NULL) {
ASSERT(firstEntry == entry);
firstEntry = entry->nextEntry;
} else
entry->prevEntry->nextEntry = entry->nextEntry;
if (entry->nextEntry == NULL) {
ASSERT(lastEntry == entry);
lastEntry = entry->prevEntry;
} else
entry->nextEntry->prevEntry = entry->prevEntry;
// ... and from the bucket chain, ...
if (prevChain == NULL)
*slotAddr = entry->nextChain;
else
prevChain->nextChain = entry->nextChain;
if (entry->nextChain != NULL) {
ASSERT(entry->nextChain->prevChain == entry);
entry->nextChain->prevChain = prevChain;
}
// ... then delete the entry.
delete entry;
// Decrement the number of entries, and contract the table if necessary.
numEntries--;
if (numBuckets > TABLE_SEGMENT_SIZE
&& numEntries < numBuckets * TABLE_MIN_LOAD) {
Contract();
ASSERT(numBuckets <= TABLE_SEGMENT_SIZE
|| numEntries >= numBuckets * TABLE_MIN_LOAD);
}
}
return dataPtr;
}
// No entry with matching data has yet been found.
// Continue following the bucket chain.
prevChain = entry;
entry = entry->nextChain;
}
// No entry with matching data was found in this bucket.
return NULL;
}
/*****************************************************************************/
VOID *Table::GetFirst(
DWORD *keyLen,
BOOL erase)
{
TableEntry **slotAddr,
*entry;
DWORD bucketNum;
VOID *dataPtr;
// If the table is empty, then simply return.
if (firstEntry == NULL) {
ASSERT(lastEntry == NULL);
ASSERT(numEntries == 0);
return NULL;
}
dataPtr = firstEntry->data;
ASSERT(dataPtr != NULL);
if (keyLen != NULL) {
*keyLen = firstEntry->keyLen;
ASSERT(firstEntry->keyLen > 0);
}
// If erasure is enabled, remove the first entry from the doubly-linked list ...
if (erase) {
entry = firstEntry;
firstEntry = entry->nextEntry;
if (firstEntry == NULL) {
ASSERT(numEntries == 1);
ASSERT(lastEntry == entry);
lastEntry = NULL;
} else {
ASSERT(numEntries > 1);
ASSERT(firstEntry->prevEntry == entry);
firstEntry->prevEntry = NULL;
}
// ... and from the bucket chain, ...
if (entry->prevChain == NULL) {
bucketNum = BucketNum(entry->hashValue);
ASSERT(directory != NULL);
slotAddr = &directory[bucketNum >> TABLE_SEGMENT_BITS]
->slot[bucketNum & TABLE_SEGMENT_MASK];
ASSERT( slotAddr != NULL);
ASSERT(*slotAddr == entry);
*slotAddr = entry->nextChain;
} else {
ASSERT(entry->prevChain->nextChain == entry);
entry->prevChain->nextChain = entry->nextChain;
}
if (entry->nextChain != NULL) {
ASSERT(entry->nextChain->prevChain == entry);
entry->nextChain->prevChain = entry->prevChain;
}
// ... then delete the entry.
delete entry;
// Decrement the number of entries, and contract the table if necessary.
numEntries--;
if (numBuckets > TABLE_SEGMENT_SIZE
&& numEntries < numBuckets * TABLE_MIN_LOAD) {
Contract();
ASSERT(numBuckets <= TABLE_SEGMENT_SIZE
|| numEntries >= numBuckets * TABLE_MIN_LOAD);
}
}
return dataPtr;
}
/*****************************************************************************/
DWORD Table::Number() const
{
return numEntries;
}
/*****************************************************************************/
/************************* FIFO class public methods *************************/
/*****************************************************************************/
FIFO::FIFO()
{
head = tail = NULL;
numEntries = 0;
}
/*****************************************************************************/
FIFO::~FIFO()
{
FIFOEntry *entry = head,
*oldEntry;
DWORD count = 0;
while ((oldEntry = entry) != NULL) {
entry = entry->next;
delete oldEntry->data;
delete oldEntry;
count++;
}
ASSERT(count == numEntries);
}
/*****************************************************************************/
VOID FIFO::Put(VOID *data)
{
FIFOEntry *newEntry;
ASSERT(data != NULL);
newEntry = new FIFOEntry;
ASSERT(newEntry != NULL);
newEntry->next = NULL;
newEntry->data = data;
if (tail != NULL)
tail->next = newEntry;
else
head = newEntry;
tail = newEntry;
numEntries++;
}
/*****************************************************************************/
VOID *FIFO::Get()
{
FIFOEntry *oldHead;
VOID *dataPtr;
if (head == NULL) {
ASSERT(tail == NULL);
ASSERT(numEntries == 0);
return NULL;
}
ASSERT(tail != NULL);
ASSERT(numEntries > 0);
dataPtr = head->data;
oldHead = head;
head = head->next;
delete oldHead;
if (head == NULL)
tail = NULL;
numEntries--;
return dataPtr;
}
/*****************************************************************************/
DWORD FIFO::Number() const
{
return numEntries;
}
/*****************************************************************************/
/************************* LIFO class public methods *************************/
/*****************************************************************************/
LIFO::LIFO()
{
top = NULL;
numEntries = 0;
}
/*****************************************************************************/
LIFO::~LIFO()
{
LIFOEntry *entry = top,
*oldEntry;
DWORD count = 0;
while ((oldEntry = entry) != NULL) {
entry = entry->next;
delete oldEntry->data;
delete oldEntry;
count++;
}
ASSERT(count == numEntries);
}
/*****************************************************************************/
VOID LIFO::Put(VOID *data)
{
LIFOEntry *newEntry;
ASSERT(data != NULL);
newEntry = new LIFOEntry;
ASSERT(newEntry != NULL);
newEntry->next = top;
newEntry->data = data;
top = newEntry;
numEntries++;
}
/*****************************************************************************/
VOID *LIFO::Get()
{
LIFOEntry *oldTop;
VOID *dataPtr;
if (top == NULL) {
ASSERT(numEntries == 0);
return NULL;
}
ASSERT(numEntries > 0);
dataPtr = top->data;
oldTop = top;
top = top->next;
delete oldTop;
numEntries--;
return dataPtr;
}
/*****************************************************************************/
DWORD LIFO::Number() const
{
return numEntries;
}