|
|
#include "StdAfx.h"
#include "ADMTScript.h"
#include "NameCracker.h"
#include <LM.h>
#include <NtDsApi.h>
#pragma comment(lib, "NtDsApi.lib")
#include <DsGetDC.h>
#include "Error.h"
#include "AdsiHelpers.h"
using namespace NAMECRACKER;
namespace NAMECRACKER {
const _TCHAR CANONICAL_DELIMITER = _T('/'); const _TCHAR RDN_DELIMITER = _T('='); const _TCHAR SAM_DELIMITER = _T('\\'); const _TCHAR SAM_INVALID_CHARACTERS[] = _T("\"*+,./:;<=>?[\\]|"); const _TCHAR EXCLUDE_SAM_INVALID_CHARACTERS[] = _T("\"+,./:;<=>?[\\]|");
}
//---------------------------------------------------------------------------
// Name Cracker Class
//---------------------------------------------------------------------------
CNameCracker::CNameCracker() { }
CNameCracker::~CNameCracker() { }
void CNameCracker::CrackNames(const StringVector& vecNames) { // separate the names into canonical names,
// SAM account names and relative distinguished names
StringVector vecCanonicalNames; StringVector vecSamAccountNames; StringVector vecRelativeDistinguishedNames;
Separate(vecNames, vecCanonicalNames, vecSamAccountNames, vecRelativeDistinguishedNames);
// then crack canonical names
CrackCanonicalNames(vecCanonicalNames, vecRelativeDistinguishedNames);
// then crack relative distinguished names
CrackRelativeDistinguishedNames(vecRelativeDistinguishedNames, vecSamAccountNames);
// then crack SAM account names
CrackSamAccountNames(vecSamAccountNames, m_vecUnResolvedNames); }
void CNameCracker::SiftExcludeNames(const StringSet& setExcludeNames, const StringSet& setNamingAttributes, StringSet& setExcludeRDNs, StringSet& setExcludeSamAccountNames) const { //
// For each specified exclude name pattern.
//
for (StringSet::const_iterator it = setExcludeNames.begin(); it != setExcludeNames.end(); it++) { const _bstr_t& strPattern = *it;
PCTSTR pszPattern = strPattern;
if (pszPattern) { //
// If the exclude pattern contains an RDN delimiter character
// then assume RDN exclude pattern otherwise assume sAMAccountName
// exclude pattern.
//
tstring str = pszPattern;
UINT uDelimiter = str.find_first_of(RDN_DELIMITER);
if (uDelimiter == 0) { //
// The RDN delimiter character must follow a valid naming attribute
// therefore if the delimiter occurs at the beginning of the pattern
// then an error must be generated.
//
AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszPattern); } else if (uDelimiter != tstring::npos) { //
// Verify exclude pattern contains a valid naming attribute. The naming attribute
// must match the naming attribute(s) for the class(es) of objects being migrated.
//
tstring strNamingAttribute = str.substr(0, uDelimiter);
bool bValidNamingAttribute = false; StringSet::const_iterator itNA;
for (itNA = setNamingAttributes.begin(); itNA != setNamingAttributes.end(); itNA++) { if (_tcsicmp(strNamingAttribute.c_str(), *itNA) == 0) { bValidNamingAttribute = true; break; } }
if (bValidNamingAttribute == false) { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszPattern); }
setExcludeRDNs.insert(strPattern); } else { setExcludeSamAccountNames.insert(strPattern); } } } }
void CNameCracker::Separate( const StringVector& vecNames, StringVector& vecCanonicalNames, StringVector& vecSamAccountNames, StringVector& vecRelativeDistinguishedNames ) { // for each name in vector...
for (StringVector::const_iterator it = vecNames.begin(); it != vecNames.end(); it++) { const tstring& strName = *it;
// if non empty name...
if (strName.empty() == false) { LPCTSTR pszName = strName.c_str();
// then if name contains a solidus '/' character assume canonical name
// else if name contains a ? '=' character assume relative distinguished name
// else assume SAM account name
if (_tcschr(pszName, CANONICAL_DELIMITER)) { // assuming canonical name
vecCanonicalNames.push_back(strName); } else if (_tcschr(pszName, RDN_DELIMITER)) { // assuming relative distinguished name
vecRelativeDistinguishedNames.push_back(strName); } else { // assuming SAM account name
vecSamAccountNames.push_back(strName); } } } }
void CNameCracker::CrackCanonicalNames(const StringVector& vecCanonicalNames, StringVector& vecUnResolvedNames) { //
// for each name generate a complete canonical name
//
CNameVector vecNames; tstring strCanonical;
for (StringVector::const_iterator it = vecCanonicalNames.begin(); it != vecCanonicalNames.end(); it++) { const tstring& strName = *it;
// if first character is the solidus '/' character...
if (strName[0] == CANONICAL_DELIMITER) { // then generate complete canonical name
strCanonical = m_strDnsName + strName; } else { // otherwise if already complete canonical name for this domain...
if (_tcsnicmp(m_strDnsName.c_str(), strName.c_str(), m_strDnsName.length()) == 0) { // then add complete canonical name
strCanonical = strName; } else { // otherwise prefix DNS domain name with solidus and add
strCanonical = m_strDnsName + CANONICAL_DELIMITER + strName; } }
vecNames.push_back(SName(strName.c_str(), strCanonical.c_str())); }
//
// crack canonical names
//
CrackNames(CANONICAL_NAME, vecNames);
for (size_t i = 0; i < vecNames.size(); i++) { const SName& name = vecNames[i];
if (name.strResolved.empty() == false) { m_setResolvedNames.insert(name.strResolved); } else { vecUnResolvedNames.push_back(name.strPartial); } } }
void CNameCracker::CrackSamAccountNames(const StringVector& vecSamAccountNames, StringVector& vecUnResolvedNames) { //
// for each name generate a NT4 account name
//
CNameVector vecNames; tstring strNT4Account;
for (StringVector::const_iterator it = vecSamAccountNames.begin(); it != vecSamAccountNames.end(); it++) { const tstring& strName = *it;
// if first character is the reverse solidus '\' character...
if (strName[0] == SAM_DELIMITER) { // then generate downlevel name
strNT4Account = m_strFlatName + strName; } else { // otherwise if already downlevel name for this domain...
if (_tcsnicmp(m_strFlatName.c_str(), strName.c_str(), m_strFlatName.length()) == 0) { // then add downlevel name
strNT4Account = strName; } else { // otherwise prefix flat domain name with reverse solidus and add
strNT4Account = m_strFlatName + SAM_DELIMITER + strName; } }
vecNames.push_back(SName(strName.c_str(), strNT4Account.c_str())); }
//
// crack names
//
CrackNames(NT4_ACCOUNT_NAME, vecNames);
for (size_t i = 0; i < vecNames.size(); i++) { const SName& name = vecNames[i];
if (name.strResolved.empty() == false) { m_setResolvedNames.insert(name.strResolved); } else { vecUnResolvedNames.push_back(name.strPartial); } } }
void CNameCracker::CrackRelativeDistinguishedNames(const StringVector& vecRelativeDistinguishedNames, StringVector& vecUnResolvedNames) { CADsContainer adscContainer(m_spDefaultContainer);
CADsPathName adspnPath(adscContainer.GetADsPath());
for (StringVector::const_iterator it = vecRelativeDistinguishedNames.begin(); it != vecRelativeDistinguishedNames.end(); it++) { adspnPath.AddLeafElement(it->c_str());
bool bFound = false;
try { CADs adsObject((LPCTSTR)adspnPath.Retrieve(ADS_FORMAT_X500));
m_setResolvedNames.insert(tstring(_bstr_t(adsObject.Get(ATTRIBUTE_DISTINGUISHED_NAME))));
bFound = true; } catch (_com_error& ce) { #ifdef _DEBUG
_TCHAR sz[2048]; _stprintf(sz, _T("'%s' : %s : 0x%08lX\n"), it->c_str(), ce.ErrorMessage(), ce.Error()); OutputDebugString(sz); #endif
bFound = false; } catch (...) { bFound = false; }
if (!bFound) { vecUnResolvedNames.push_back(*it); }
adspnPath.RemoveLeafElement(); } }
void CNameCracker::CrackNames(NAME_FORMAT eFormat, CNameVector& vecNames) { HANDLE hDs = NULL; LPTSTR apszNames = NULL; PDS_NAME_RESULT pdnrResult = NULL;
try { if (vecNames.size() > 0) { DWORD dwError = DsBind(m_strDomainController.c_str(), NULL, &hDs);
if (dwError == NO_ERROR) { DWORD dwCount = vecNames.size(); LPCTSTR* apszNames = new LPCTSTR[dwCount];
if (apszNames != NULL) { for (DWORD dwIndex = 0; dwIndex < dwCount; dwIndex++) { apszNames[dwIndex] = vecNames[dwIndex].strComplete.c_str(); }
dwError = DsCrackNames( hDs, DS_NAME_NO_FLAGS, (eFormat == CANONICAL_NAME) ? DS_CANONICAL_NAME : (eFormat == NT4_ACCOUNT_NAME) ? DS_NT4_ACCOUNT_NAME : DS_UNKNOWN_NAME, DS_FQDN_1779_NAME, dwCount, const_cast<LPTSTR*>(apszNames), &pdnrResult );
if (dwError == NO_ERROR) { DWORD c = pdnrResult->cItems;
for (DWORD i = 0; i < c; i++) { DS_NAME_RESULT_ITEM& dnriItem = pdnrResult->rItems[i];
if (dnriItem.status == DS_NAME_NO_ERROR) { vecNames[i].strResolved = dnriItem.pName; } }
DsFreeNameResult(pdnrResult); } else { _com_issue_error(HRESULT_FROM_WIN32(dwError)); }
delete [] apszNames; } else { _com_issue_error(E_OUTOFMEMORY); }
DsUnBind(&hDs); } else { _com_issue_error(HRESULT_FROM_WIN32(dwError)); } } } catch (...) { if (pdnrResult) { DsFreeNameResult(pdnrResult); }
delete [] apszNames;
if (hDs) { DsUnBind(&hDs); }
throw; } }
namespace {
// SplitCanonicalName Method
//
// Given 'a.company.com/Sales/West/Name' this method splits the complete
// canonical name into its component parts Domain='a.company.com',
// Path='/Sales/West/', Name='Name'.
//
// Given 'Sales/West/Name' this method splits the partial canonical name
// into its component parts Domain='', Path='/Sales/West/', Name='Name'.
//
// Given 'Name' this method splits the partial canonical name into its
// component parts Domain='', Path='/', Name='Name'.
void SplitCanonicalName(LPCTSTR pszName, _bstr_t& strDomain, _bstr_t& strPath, _bstr_t& strName) { strDomain = (LPCTSTR)NULL; strPath = (LPCTSTR)NULL; strName = (LPCTSTR)NULL;
if (pszName) { tstring str = pszName;
UINT posA = 0; UINT posB = tstring::npos;
do { posA = str.find_first_of(_T('/'), posA ? posA + 1 : posA); } while ((posA != 0) && (posA != tstring::npos) && (str[posA - 1] == _T('\\')));
do { posB = str.find_last_of(_T('/'), (posB != tstring::npos) ? posB - 1 : posB); } while ((posB != 0) && (posB != tstring::npos) && (str[posB - 1] == _T('\\')));
strDomain = str.substr(0, posA).c_str(); strPath = str.substr(posA, posB - posA).c_str(); strName = str.substr(posB).c_str(); } }
void SplitPath(LPCTSTR pszPath, _bstr_t& strPath, _bstr_t& strName) { strPath = (LPCTSTR)NULL; strName = (LPCTSTR)NULL;
if (pszPath) { tstring str = pszPath;
UINT pos = str.find_first_of(_T('\\'));
if (pos != tstring::npos) { strName = pszPath; } else { UINT posA = str.find_first_of(_T('/'));
if (posA == tstring::npos) { strName = (_T("/") + str).c_str(); } else { UINT posB = str.find_last_of(_T('/'));
strPath = str.substr(posA, posB - posA).c_str(); strName = str.substr(posB).c_str(); } } } }
}
//---------------------------------------------------------------------------
// Ignore Case String Less
//---------------------------------------------------------------------------
struct SIgnoreCaseStringLess : public std::binary_function<tstring, tstring, bool> { bool operator()(const tstring& x, const tstring& y) const { bool bLess;
LPCTSTR pszX = x.c_str(); LPCTSTR pszY = y.c_str();
if (pszX == pszY) { bLess = false; } else if (pszX == NULL) { bLess = true; } else if (pszY == NULL) { bLess = false; } else { bLess = _tcsicmp(pszX, pszY) < 0; }
return bLess; } };
//---------------------------------------------------------------------------
// CDomainMap Implementation
//---------------------------------------------------------------------------
class CDomainMap : public std::map<_bstr_t, StringSet, IgnoreCaseStringLess> { public:
CDomainMap() { }
void Initialize(const StringSet& setNames) { _bstr_t strDefaultDns(_T("/")); _bstr_t strDefaultFlat(_T("\\"));
for (StringSet::const_iterator it = setNames.begin(); it != setNames.end(); it++) { tstring strName = *it;
// if not an empty name...
if (strName.empty() == false) { // if name contains a canonical name delimiter...
UINT posDelimiter = strName.find(CANONICAL_DELIMITER);
if (posDelimiter != tstring::npos) { // then assume canonical name
if (posDelimiter == 0) { // then generate complete canonical name
Insert(strDefaultDns, *it); } else { // otherwise if path component before delimiter contains
// a period
UINT posDot = strName.find(_T('.'));
if (posDot < posDelimiter) { // then assume a complete canonical name with DNS domain name prefix
Insert(strName.substr(0, posDelimiter).c_str(), *it); } else { // otherwise assume domain name has not been specified
Insert(strDefaultDns, *it); } } } else { // otherwise if name contains a NT account name delimiter
// character and no invalid SAM account name characters...
UINT posDelimiter = strName.find(SAM_DELIMITER);
if (posDelimiter != tstring::npos) { if (strName.find_first_of(SAM_INVALID_CHARACTERS, posDelimiter + 1) == tstring::npos) { if (posDelimiter == 0) { Insert(strDefaultFlat, *it); } else { // then assume SAM account name
Insert(strName.substr(0, posDelimiter).c_str(), strName.substr(posDelimiter).c_str()); } } else { // otherwise assume relative distinguished name
Insert(strDefaultDns, *it); } } else { Insert(strDefaultDns, *it); } } } } }
protected:
void Insert(_bstr_t strDomain, _bstr_t strName) { iterator it = find(strDomain);
if (it == end()) { std::pair<iterator, bool> pair = insert(value_type(strDomain, StringSet())); it = pair.first; }
it->second.insert(strName); } };
//---------------------------------------------------------------------------
// CDomainToPathMap Implementation
//---------------------------------------------------------------------------
// Initialize Method
void CDomainToPathMap::Initialize(LPCTSTR pszDefaultDomainDns, LPCTSTR pszDefaultDomainFlat, const StringSet& setNames) { CDomainMap map;
map.Initialize(setNames);
for (CDomainMap::const_iterator itDomain = map.begin(); itDomain != map.end(); itDomain++) { _bstr_t strDomainName = itDomain->first;
LPCTSTR pszDomainName = strDomainName;
if (pszDomainName && ((*pszDomainName == _T('/')) || (*pszDomainName == _T('\\')))) { strDomainName = (pszDefaultDomainDns && (_tcslen(pszDefaultDomainDns) > 0)) ? pszDefaultDomainDns : pszDefaultDomainFlat; } else { if (GetValidDomainName(strDomainName) == false) { strDomainName = (pszDefaultDomainDns && (_tcslen(pszDefaultDomainDns) > 0)) ? pszDefaultDomainDns : pszDefaultDomainFlat; } }
iterator it = find(strDomainName);
if (it == end()) { std::pair<iterator, bool> pair = insert(value_type(strDomainName, StringSet())); it = pair.first; }
StringSet& setNames = it->second;
const StringSet& set = itDomain->second;
for (StringSet::const_iterator itSet = set.begin(); itSet != set.end(); itSet++) { setNames.insert(*itSet); } } }
// GetValidDomainName Method
bool CDomainToPathMap::GetValidDomainName(_bstr_t& strDomainName) { bool bValid = false;
PDOMAIN_CONTROLLER_INFO pdci;
// attempt to retrieve DNS name of domain controller supporting active directory service
DWORD dwError = DsGetDcName(NULL, strDomainName, NULL, NULL, DS_RETURN_DNS_NAME, &pdci);
// if domain controller not found, attempt to retrieve flat name of domain controller
if (dwError == ERROR_NO_SUCH_DOMAIN) { dwError = DsGetDcName(NULL, strDomainName, NULL, NULL, DS_RETURN_FLAT_NAME, &pdci); }
// if domain controller found then save name otherwise generate error
if (dwError == NO_ERROR) { strDomainName = pdci->DomainName;
NetApiBufferFree(pdci);
bValid = true; }
return bValid; }
//
// CNameToPathMap Implementation
//
CNameToPathMap::CNameToPathMap() { }
CNameToPathMap::CNameToPathMap(StringSet& setNames) { Initialize(setNames); }
void CNameToPathMap::Initialize(StringSet& setNames) { _bstr_t strDomain; _bstr_t strPath; _bstr_t strName;
for (StringSet::iterator it = setNames.begin(); it != setNames.end(); it++) { // SplitPath(*it, strPath, strName);
SplitCanonicalName(*it, strDomain, strPath, strName);
Add(strName, strPath); } }
void CNameToPathMap::Add(_bstr_t& strName, _bstr_t& strPath) { iterator it = find(strName);
if (it == end()) { std::pair<iterator, bool> pair = insert(value_type(strName, StringSet()));
it = pair.first; }
it->second.insert(strPath); }
//
// IgnoreCaseStringLess Implementation
//
bool IgnoreCaseStringLess::operator()(const _bstr_t& x, const _bstr_t& y) const { bool bLess;
LPCTSTR pszThis = x; LPCTSTR pszThat = y;
if (pszThis == pszThat) { bLess = false; } else if (pszThis == NULL) { bLess = true; } else if (pszThat == NULL) { bLess = false; } else { bLess = _tcsicmp(pszThis, pszThat) < 0; }
return bLess; }
//
// CCompareStrings Implementation
//
CCompareStrings::CCompareStrings() { }
CCompareStrings::CCompareStrings(StringSet& setNames) { Initialize(setNames); }
void CCompareStrings::Initialize(StringSet& setNames) { for (StringSet::iterator it = setNames.begin(); it != setNames.end(); it++) { m_vecCompareStrings.push_back(CCompareString(*it)); } }
bool CCompareStrings::IsMatch(LPCTSTR pszName) { bool bIs = false;
CompareStringVector::iterator itBeg = m_vecCompareStrings.begin(); CompareStringVector::iterator itEnd = m_vecCompareStrings.end();
for (CompareStringVector::iterator it = itBeg; it != itEnd; it++) { if (it->IsMatch(pszName)) { bIs = true;
break; } }
return bIs; }
//
// CCompareString Implementation
//
CCompareStrings::CCompareString::CCompareString(LPCTSTR pszCompare) { if (pszCompare) { Initialize(pszCompare); } }
CCompareStrings::CCompareString::CCompareString(const CCompareString& r) : m_nType(r.m_nType), m_strCompare(r.m_strCompare) { }
void CCompareStrings::CCompareString::Initialize(LPCTSTR pszCompare) { if (pszCompare) { tstring str = pszCompare;
UINT uLength = str.length();
if (uLength == 0) { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszCompare); }
bool bBeg = (str[0] == _T('*')); bool bEnd = ((uLength > 1) && (str[uLength - 1] == _T('*'))) ? true : false;
if (bBeg && bEnd) { // contains
m_nType = 3; str = str.substr(1, uLength - 2); } else if (bBeg) { // ends with
m_nType = 2; str = str.substr(1, uLength - 1); } else if (bEnd) { // begins with
m_nType = 1; str = str.substr(0, uLength - 1); } else { // equals
m_nType = 0; }
if (str.length() > 0) { m_strCompare = str.c_str(); } else { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszCompare); } } else { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, _T("")); } }
bool CCompareStrings::CCompareString::IsMatch(LPCTSTR psz) { bool bIs = false;
if (psz) { switch (m_nType) { case 0: // equals
{ bIs = (_tcsicmp(psz, m_strCompare) == 0); break; } case 1: // begins with
{ bIs = (_tcsnicmp(psz, m_strCompare, m_strCompare.length()) == 0); break; } case 2: // ends with
{ UINT cchT = _tcslen(psz); UINT cchC = m_strCompare.length();
if (cchT >= cchC) { bIs = (_tcsnicmp(psz + cchT - cchC, m_strCompare, cchC) == 0); } break; } case 3: // contains
{ PTSTR pszT = NULL; PTSTR pszC = NULL;
try { pszT = new _TCHAR[_tcslen(psz) + 1]; pszC = new _TCHAR[m_strCompare.length() + 1];
if (pszT && pszC) { _tcscpy(pszT, psz); _tcscpy(pszC, m_strCompare);
_tcslwr(pszT); _tcslwr(pszC);
bIs = (_tcsstr(pszT, pszC) != NULL); } else { _com_issue_error(E_OUTOFMEMORY); } } catch (...) { delete [] pszC; delete [] pszT; throw; }
delete [] pszC; delete [] pszT; break; } } }
return bIs; }
//
// CCompareRDNs Implementation
//
CCompareRDNs::CCompareRDNs() { }
CCompareRDNs::CCompareRDNs(StringSet& setNames) { Initialize(setNames); }
void CCompareRDNs::Initialize(StringSet& setNames) { for (StringSet::iterator it = setNames.begin(); it != setNames.end(); it++) { m_vecCompare.push_back(CCompareRDN(*it)); } }
bool CCompareRDNs::IsMatch(LPCTSTR pszName) { bool bIs = false;
CompareVector::iterator itBeg = m_vecCompare.begin(); CompareVector::iterator itEnd = m_vecCompare.end();
for (CompareVector::iterator it = itBeg; it != itEnd; it++) { if (it->IsMatch(pszName)) { bIs = true;
break; } }
return bIs; }
//
// CCompareRDN Implementation
//
CCompareRDNs::CCompareRDN::CCompareRDN(LPCTSTR pszCompare) { if (pszCompare) { Initialize(pszCompare); } }
CCompareRDNs::CCompareRDN::CCompareRDN(const CCompareRDN& r) : m_nPatternType(r.m_nPatternType), m_strType(r.m_strType), m_strValue(r.m_strValue) { }
void CCompareRDNs::CCompareRDN::Initialize(LPCTSTR pszCompare) { if (pszCompare == NULL) { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, _T("")); }
tstring str = pszCompare;
UINT uDelimiter = str.find_first_of(_T('='));
//
// The pattern must include the RDN delimiter character.
//
if ((uDelimiter == 0) || (uDelimiter == tstring::npos)) { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszCompare); }
//
// Retrieve the naming attribute portion.
//
m_strType = str.substr(0, uDelimiter).c_str();
//
// Retrieve the naming attribute value portion.
//
tstring strValue = str.substr(uDelimiter + 1);
UINT uLength = strValue.length();
if (uLength == 0) { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszCompare); }
//
// Determine pattern type.
//
bool bBeg = (strValue[0] == _T('*')); bool bEnd = ((uLength > 1) && (strValue[uLength - 1] == _T('*'))) ? true : false;
if (bBeg && bEnd) { // contains
m_nPatternType = 3; strValue = strValue.substr(1, uLength - 2); } else if (bBeg) { // ends with
m_nPatternType = 2; strValue = strValue.substr(1, uLength - 1); } else if (bEnd) { // begins with
m_nPatternType = 1; strValue = strValue.substr(0, uLength - 1); } else { // equals
m_nPatternType = 0; }
if (strValue.length() > 0) { m_strValue = strValue.c_str(); } else { AdmtThrowError(GUID_NULL, GUID_NULL, E_INVALIDARG, IDS_E_INVALID_FILTER_STRING, pszCompare); } }
bool CCompareRDNs::CCompareRDN::IsMatch(LPCTSTR psz) { bool bIs = false;
if (psz) { PCTSTR pszDelimiter = _tcschr(psz, _T('='));
if (pszDelimiter) { if (_tcsnicmp(m_strType, psz, pszDelimiter - psz) == 0) { PCTSTR pszValue = pszDelimiter + 1;
switch (m_nPatternType) { case 0: // equals
{ bIs = (_tcsicmp(pszValue, m_strValue) == 0); break; } case 1: // begins with
{ bIs = (_tcsnicmp(pszValue, m_strValue, m_strValue.length()) == 0); break; } case 2: // ends with
{ UINT cchT = _tcslen(pszValue); UINT cchC = m_strValue.length();
if (cchT >= cchC) { bIs = (_tcsnicmp(pszValue + cchT - cchC, m_strValue, cchC) == 0); } break; } case 3: // contains
{ PTSTR pszT = NULL; PTSTR pszC = NULL;
try { pszT = new _TCHAR[_tcslen(pszValue) + 1]; pszC = new _TCHAR[m_strValue.length() + 1];
if (pszT && pszC) { _tcscpy(pszT, pszValue); _tcscpy(pszC, m_strValue);
_tcslwr(pszT); _tcslwr(pszC);
bIs = (_tcsstr(pszT, pszC) != NULL); } else { _com_issue_error(E_OUTOFMEMORY); } } catch (...) { delete [] pszC; delete [] pszT; throw; }
delete [] pszC; delete [] pszT; break; } } } } }
return bIs; }
|