/*
|
* CDDL HEADER START
|
*
|
* The contents of this file are subject to the terms of the
|
* Common Development and Distribution License, Version 1.0 only
|
* (the "License"). You may not use this file except in compliance
|
* with the License.
|
*
|
* You can obtain a copy of the license at legal-notices/CDDLv1_0.txt
|
* or http://forgerock.org/license/CDDLv1.0.html.
|
* See the License for the specific language governing permissions
|
* and limitations under the License.
|
*
|
* When distributing Covered Code, include this CDDL HEADER in each
|
* file and include the License file at legal-notices/CDDLv1_0.txt.
|
* If applicable, add the following below this CDDL HEADER, with the
|
* fields enclosed by brackets "[]" replaced with your own identifying
|
* information:
|
* Portions Copyright [yyyy] [name of copyright owner]
|
*
|
* CDDL HEADER END
|
*
|
*
|
* Copyright 2006-2010 Sun Microsystems, Inc.
|
* Portions Copyright 2011-2013 ForgeRock AS
|
*/
|
package org.opends.server.backends;
|
|
|
|
import java.io.File;
|
import java.io.FileInputStream;
|
import java.io.FileOutputStream;
|
import java.io.InputStream;
|
import java.io.IOException;
|
import java.io.OutputStream;
|
import java.security.MessageDigest;
|
import java.util.ArrayList;
|
import java.util.Arrays;
|
import java.util.Date;
|
import java.util.HashMap;
|
import java.util.HashSet;
|
import java.util.LinkedHashMap;
|
import java.util.LinkedHashSet;
|
import java.util.LinkedList;
|
import java.util.List;
|
import java.util.Map;
|
import java.util.Set;
|
import java.util.TreeSet;
|
import java.util.concurrent.ConcurrentHashMap;
|
import java.util.zip.Deflater;
|
import java.util.zip.ZipEntry;
|
import java.util.zip.ZipInputStream;
|
import java.util.zip.ZipOutputStream;
|
import javax.crypto.Mac;
|
|
import org.opends.messages.Message;
|
import org.opends.server.admin.Configuration;
|
import org.opends.server.admin.std.server.SchemaBackendCfg;
|
import org.opends.server.admin.server.ConfigurationChangeListener;
|
import org.opends.server.api.AlertGenerator;
|
import org.opends.server.api.Backend;
|
import org.opends.server.api.ClientConnection;
|
import org.opends.server.api.MatchingRule;
|
import org.opends.server.config.ConfigEntry;
|
import org.opends.server.config.ConfigException;
|
import org.opends.server.core.AddOperation;
|
import org.opends.server.core.DeleteOperation;
|
import org.opends.server.core.DirectoryServer;
|
import org.opends.server.core.ModifyOperation;
|
import org.opends.server.core.ModifyDNOperation;
|
import org.opends.server.core.SchemaConfigManager;
|
import org.opends.server.core.SearchOperation;
|
import org.opends.server.loggers.ErrorLogger;
|
import org.opends.server.loggers.debug.DebugTracer;
|
import org.opends.server.schema.AttributeTypeSyntax;
|
import org.opends.server.schema.DITContentRuleSyntax;
|
import org.opends.server.schema.DITStructureRuleSyntax;
|
import org.opends.server.schema.GeneralizedTimeSyntax;
|
import org.opends.server.schema.LDAPSyntaxDescriptionSyntax;
|
import org.opends.server.schema.MatchingRuleUseSyntax;
|
import org.opends.server.schema.NameFormSyntax;
|
import org.opends.server.schema.ObjectClassSyntax;
|
import org.opends.server.types.*;
|
import org.opends.server.util.DynamicConstants;
|
import org.opends.server.util.LDIFException;
|
import org.opends.server.util.LDIFReader;
|
import org.opends.server.util.LDIFWriter;
|
import org.opends.server.util.Validator;
|
|
import static org.opends.messages.BackendMessages.*;
|
import static org.opends.messages.ConfigMessages.*;
|
import static org.opends.messages.SchemaMessages.*;
|
import static org.opends.server.config.ConfigConstants.*;
|
import static org.opends.server.loggers.debug.DebugLogger.*;
|
import static org.opends.server.loggers.ErrorLogger.*;
|
import static org.opends.server.schema.SchemaConstants.*;
|
import static org.opends.server.util.ServerConstants.*;
|
import static org.opends.server.util.StaticUtils.*;
|
|
|
/**
|
* This class defines a backend to hold the Directory Server schema information.
|
* It is a kind of meta-backend in that it doesn't actually hold any data but
|
* rather dynamically generates the schema entry whenever it is requested.
|
*/
|
public class SchemaBackend
|
extends Backend
|
implements ConfigurationChangeListener<SchemaBackendCfg>, AlertGenerator
|
{
|
/**
|
* The tracer object for the debug logger.
|
*/
|
private static final DebugTracer TRACER = getTracer();
|
|
|
/**
|
* The fully-qualified name of this class.
|
*/
|
private static final String CLASS_NAME =
|
"org.opends.server.backends.SchemaBackend";
|
|
|
private static final String CONFIG_SCHEMA_ELEMENTS_FILE = "02-config.ldif";
|
private static final String CORE_SCHEMA_ELEMENTS_FILE = "00-core.ldif";
|
|
|
|
/**
|
* The set of user-defined attributes that will be included in the schema
|
* entry.
|
*/
|
private ArrayList<Attribute> userDefinedAttributes;
|
|
/**
|
* The attribute type that will be used to include the defined attribute
|
* types.
|
*/
|
private AttributeType attributeTypesType;
|
|
/**
|
* The attribute type that will be used to hold the schema creation timestamp.
|
*/
|
private AttributeType createTimestampType;
|
|
/** The attribute type that will be used to hold the schema creator's name. */
|
private AttributeType creatorsNameType;
|
|
/**
|
* The attribute type that will be used to include the defined DIT content
|
* rules.
|
*/
|
private AttributeType ditContentRulesType;
|
|
/**
|
* The attribute type that will be used to include the defined DIT structure
|
* rules.
|
*/
|
private AttributeType ditStructureRulesType;
|
|
/**
|
* The attribute type that will be used to include the defined attribute
|
* syntaxes.
|
*/
|
private AttributeType ldapSyntaxesType;
|
|
/**
|
* The attribute type that will be used to include the defined matching rules.
|
*/
|
private AttributeType matchingRulesType;
|
|
/**
|
* The attribute type that will be used to include the defined matching rule
|
* uses.
|
*/
|
private AttributeType matchingRuleUsesType;
|
|
/** The attribute that will be used to hold the schema modifier's name. */
|
private AttributeType modifiersNameType;
|
|
/**
|
* The attribute type that will be used to hold the schema modification
|
* timestamp.
|
*/
|
private AttributeType modifyTimestampType;
|
|
/**
|
* The attribute type that will be used to include the defined object classes.
|
*/
|
private AttributeType objectClassesType;
|
|
/** The attribute type that will be used to include the defined name forms. */
|
private AttributeType nameFormsType;
|
|
/**
|
* The value containing DN of the user we'll say created the configuration.
|
*/
|
private AttributeValue creatorsName;
|
|
/**
|
* The value containing the DN of the last user to modify the configuration.
|
*/
|
private AttributeValue modifiersName;
|
|
/** The timestamp that will be used for the schema creation time. */
|
private AttributeValue createTimestamp;
|
|
/**
|
* The timestamp that will be used for the latest schema modification time.
|
*/
|
private AttributeValue modifyTimestamp;
|
|
/**
|
* Indicates whether the attributes of the schema entry should always be
|
* treated as user attributes even if they are defined as operational.
|
*/
|
private boolean showAllAttributes;
|
|
/** The DN of the configuration entry for this backend. */
|
private DN configEntryDN;
|
|
/** The current configuration state. */
|
private SchemaBackendCfg currentConfig;
|
|
/** The set of base DNs for this backend. */
|
private DN[] baseDNs;
|
|
/** The set of objectclasses that will be used in the schema entry. */
|
private HashMap<ObjectClass,String> schemaObjectClasses;
|
|
/** The set of supported controls for this backend. */
|
private Set<String> supportedControls;
|
|
/** The set of supported features for this backend. */
|
private Set<String> supportedFeatures;
|
|
/** The time that the schema was last modified. */
|
private long modifyTime;
|
|
/**
|
* Regular expression used to strip minimum upper bound value from syntax
|
* Attribute Type Description. The value looks like: {count}.
|
*/
|
private String stripMinUpperBoundRegEx = "\\{\\d+\\}";
|
|
|
|
/**
|
* Creates a new backend with the provided information. All backend
|
* implementations must implement a default constructor that use
|
* <CODE>super()</CODE> to invoke this constructor.
|
*/
|
public SchemaBackend()
|
{
|
super();
|
|
// Perform all initialization in initializeBackend.
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void configureBackend(Configuration config)
|
throws ConfigException
|
{
|
// Make sure that a configuration entry was provided. If not, then we will
|
// not be able to complete initialization.
|
if (config == null)
|
{
|
Message message = ERR_SCHEMA_CONFIG_ENTRY_NULL.get();
|
throw new ConfigException(message);
|
}
|
|
Validator.ensureTrue(config instanceof SchemaBackendCfg);
|
SchemaBackendCfg cfg = (SchemaBackendCfg)config;
|
ConfigEntry configEntry = DirectoryServer.getConfigEntry(cfg.dn());
|
|
configEntryDN = configEntry.getDN();
|
|
// Get all of the attribute types that we will use for schema elements.
|
attributeTypesType =
|
DirectoryServer.getAttributeType(ATTR_ATTRIBUTE_TYPES_LC, true);
|
objectClassesType =
|
DirectoryServer.getAttributeType(ATTR_OBJECTCLASSES_LC, true);
|
matchingRulesType =
|
DirectoryServer.getAttributeType(ATTR_MATCHING_RULES_LC, true);
|
ldapSyntaxesType =
|
DirectoryServer.getAttributeType(ATTR_LDAP_SYNTAXES_LC, true);
|
ditContentRulesType =
|
DirectoryServer.getAttributeType(ATTR_DIT_CONTENT_RULES_LC, true);
|
ditStructureRulesType =
|
DirectoryServer.getAttributeType(ATTR_DIT_STRUCTURE_RULES_LC, true);
|
matchingRuleUsesType =
|
DirectoryServer.getAttributeType(ATTR_MATCHING_RULE_USE_LC, true);
|
nameFormsType = DirectoryServer.getAttributeType(ATTR_NAME_FORMS_LC, true);
|
|
// Initialize the lastmod attributes.
|
creatorsNameType =
|
DirectoryServer.getAttributeType(OP_ATTR_CREATORS_NAME_LC, true);
|
createTimestampType =
|
DirectoryServer.getAttributeType(OP_ATTR_CREATE_TIMESTAMP_LC, true);
|
modifiersNameType =
|
DirectoryServer.getAttributeType(OP_ATTR_MODIFIERS_NAME_LC, true);
|
modifyTimestampType =
|
DirectoryServer.getAttributeType(OP_ATTR_MODIFY_TIMESTAMP_LC, true);
|
|
// Construct the set of objectclasses to include in the schema entry.
|
schemaObjectClasses = new LinkedHashMap<ObjectClass,String>(3);
|
schemaObjectClasses.put(DirectoryServer.getTopObjectClass(), OC_TOP);
|
|
ObjectClass subentryOC = DirectoryServer.getObjectClass(OC_LDAP_SUBENTRY_LC,
|
true);
|
schemaObjectClasses.put(subentryOC, OC_LDAP_SUBENTRY);
|
|
ObjectClass subschemaOC = DirectoryServer.getObjectClass(OC_SUBSCHEMA,
|
true);
|
schemaObjectClasses.put(subschemaOC, OC_SUBSCHEMA);
|
|
|
// Define empty sets for the supported controls and features.
|
supportedControls = new HashSet<String>(0);
|
supportedFeatures = new HashSet<String>(0);
|
|
|
configEntryDN = configEntry.getDN();
|
|
DN[] newBaseDNs = new DN[cfg.getBaseDN().size()];
|
cfg.getBaseDN().toArray(newBaseDNs);
|
this.baseDNs = newBaseDNs;
|
|
creatorsName = AttributeValues.create(
|
creatorsNameType, newBaseDNs[0].toString());
|
modifiersName = AttributeValues.create(
|
modifiersNameType, newBaseDNs[0].toString());
|
|
long createTime = DirectoryServer.getSchema().getOldestModificationTime();
|
createTimestamp =
|
GeneralizedTimeSyntax.createGeneralizedTimeValue(createTime);
|
|
long newModifyTime =
|
DirectoryServer.getSchema().getYoungestModificationTime();
|
modifyTimestamp =
|
GeneralizedTimeSyntax.createGeneralizedTimeValue(newModifyTime);
|
|
|
// Get the set of user-defined attributes for the configuration entry. Any
|
// attributes that we don't recognize will be included directly in the
|
// schema entry.
|
userDefinedAttributes = new ArrayList<Attribute>();
|
for (List<Attribute> attrs :
|
configEntry.getEntry().getUserAttributes().values())
|
{
|
for (Attribute a : attrs)
|
{
|
if (! isSchemaConfigAttribute(a))
|
{
|
userDefinedAttributes.add(a);
|
}
|
}
|
}
|
for (List<Attribute> attrs :
|
configEntry.getEntry().getOperationalAttributes().values())
|
{
|
for (Attribute a : attrs)
|
{
|
if (! isSchemaConfigAttribute(a))
|
{
|
userDefinedAttributes.add(a);
|
}
|
}
|
}
|
|
|
// Determine whether to show all attributes.
|
showAllAttributes = cfg.isShowAllAttributes();
|
|
|
currentConfig = cfg;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void initializeBackend()
|
throws ConfigException, InitializationException
|
{
|
// Register each of the suffixes with the Directory Server. Also, register
|
// the first one as the schema base.
|
DirectoryServer.setSchemaDN(baseDNs[0]);
|
for (DN baseDN : baseDNs) {
|
try {
|
DirectoryServer.registerBaseDN(baseDN, this, true);
|
} catch (Exception e) {
|
if (debugEnabled()) {
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_BACKEND_CANNOT_REGISTER_BASEDN.get(
|
baseDN.toString(), getExceptionMessage(e));
|
throw new InitializationException(message, e);
|
}
|
}
|
|
|
// Identify any differences that may exist between the concatenated schema
|
// file from the last online modification and the current schema files. If
|
// there are any differences, then they should be from making changes to the
|
// schema files with the server offline.
|
try
|
{
|
// First, generate lists of elements from the current schema.
|
Set<String> newATs = new LinkedHashSet<String>();
|
Set<String> newOCs = new LinkedHashSet<String>();
|
Set<String> newNFs = new LinkedHashSet<String>();
|
Set<String> newDCRs = new LinkedHashSet<String>();
|
Set<String> newDSRs = new LinkedHashSet<String>();
|
Set<String> newMRUs = new LinkedHashSet<String>();
|
Set<String> newLSDs = new LinkedHashSet<String>();
|
Schema.genConcatenatedSchema(newATs, newOCs, newNFs, newDCRs, newDSRs,
|
newMRUs,newLSDs);
|
|
// Next, generate lists of elements from the previous concatenated schema.
|
// If there isn't a previous concatenated schema, then use the base
|
// schema for the current revision.
|
String concatFilePath;
|
File configFile = new File(DirectoryServer.getConfigFile());
|
File configDirectory = configFile.getParentFile();
|
File upgradeDirectory = new File(configDirectory, "upgrade");
|
File concatFile = new File(upgradeDirectory,
|
SCHEMA_CONCAT_FILE_NAME);
|
if (concatFile.exists())
|
{
|
concatFilePath = concatFile.getAbsolutePath();
|
}
|
else
|
{
|
concatFile = new File(upgradeDirectory,
|
SCHEMA_BASE_FILE_NAME_WITHOUT_REVISION +
|
DynamicConstants.REVISION_NUMBER);
|
if (concatFile.exists())
|
{
|
concatFilePath = concatFile.getAbsolutePath();
|
}
|
else
|
{
|
String runningUnitTestsStr =
|
System.getProperty(PROPERTY_RUNNING_UNIT_TESTS);
|
if ((runningUnitTestsStr != null) &&
|
runningUnitTestsStr.equalsIgnoreCase("true"))
|
{
|
Schema.writeConcatenatedSchema();
|
concatFile = new File(upgradeDirectory, SCHEMA_CONCAT_FILE_NAME);
|
concatFilePath = concatFile.getAbsolutePath();
|
}
|
else
|
{
|
Message message = ERR_SCHEMA_CANNOT_FIND_CONCAT_FILE.
|
get(upgradeDirectory.getAbsolutePath(), SCHEMA_CONCAT_FILE_NAME,
|
concatFile.getName());
|
throw new InitializationException(message);
|
}
|
}
|
}
|
|
Set<String> oldATs = new LinkedHashSet<String>();
|
Set<String> oldOCs = new LinkedHashSet<String>();
|
Set<String> oldNFs = new LinkedHashSet<String>();
|
Set<String> oldDCRs = new LinkedHashSet<String>();
|
Set<String> oldDSRs = new LinkedHashSet<String>();
|
Set<String> oldMRUs = new LinkedHashSet<String>();
|
Set<String> oldLSDs = new LinkedHashSet<String>();
|
Schema.readConcatenatedSchema(concatFilePath, oldATs, oldOCs, oldNFs,
|
oldDCRs, oldDSRs, oldMRUs,oldLSDs);
|
|
// Create a list of modifications and add any differences between the old
|
// and new schema into them.
|
List<Modification> mods = new LinkedList<Modification>();
|
Schema.compareConcatenatedSchema(oldATs, newATs, attributeTypesType,
|
mods);
|
Schema.compareConcatenatedSchema(oldOCs, newOCs, objectClassesType, mods);
|
Schema.compareConcatenatedSchema(oldNFs, newNFs, nameFormsType, mods);
|
Schema.compareConcatenatedSchema(oldDCRs, newDCRs, ditContentRulesType,
|
mods);
|
Schema.compareConcatenatedSchema(oldDSRs, newDSRs, ditStructureRulesType,
|
mods);
|
Schema.compareConcatenatedSchema(oldMRUs, newMRUs, matchingRuleUsesType,
|
mods);
|
Schema.compareConcatenatedSchema(oldLSDs, newLSDs, ldapSyntaxesType,
|
mods);
|
if (! mods.isEmpty())
|
{
|
// TODO : Raise an alert notification.
|
|
DirectoryServer.setOfflineSchemaChanges(mods);
|
|
// Write a new concatenated schema file with the most recent information
|
// so we don't re-find these same changes on the next startup.
|
Schema.writeConcatenatedSchema();
|
}
|
}
|
catch (InitializationException ie)
|
{
|
throw ie;
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_ERROR_DETERMINING_SCHEMA_CHANGES.get(
|
getExceptionMessage(e));
|
ErrorLogger.logError(message);
|
}
|
|
|
// Register with the Directory Server as a configurable component.
|
currentConfig.addSchemaChangeListener(this);
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void finalizeBackend()
|
{
|
currentConfig.removeSchemaChangeListener(this);
|
|
for (DN baseDN : baseDNs)
|
{
|
try
|
{
|
DirectoryServer.deregisterBaseDN(baseDN);
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Indicates whether the provided attribute is one that is used in the
|
* configuration of this backend.
|
*
|
* @param attribute The attribute for which to make the determination.
|
*
|
* @return <CODE>true</CODE> if the provided attribute is one that is used in
|
* the configuration of this backend, <CODE>false</CODE> if not.
|
*/
|
private boolean isSchemaConfigAttribute(Attribute attribute)
|
{
|
AttributeType attrType = attribute.getAttributeType();
|
return attrType.hasName(ATTR_SCHEMA_ENTRY_DN.toLowerCase()) ||
|
attrType.hasName(ATTR_BACKEND_ENABLED.toLowerCase()) ||
|
attrType.hasName(ATTR_BACKEND_CLASS.toLowerCase()) ||
|
attrType.hasName(ATTR_BACKEND_ID.toLowerCase()) ||
|
attrType.hasName(ATTR_BACKEND_BASE_DN.toLowerCase()) ||
|
attrType.hasName(ATTR_BACKEND_WRITABILITY_MODE.toLowerCase()) ||
|
attrType.hasName(ATTR_SCHEMA_SHOW_ALL_ATTRIBUTES.toLowerCase()) ||
|
attrType.hasName(ATTR_COMMON_NAME) ||
|
attrType.hasName(OP_ATTR_CREATORS_NAME_LC) ||
|
attrType.hasName(OP_ATTR_CREATE_TIMESTAMP_LC) ||
|
attrType.hasName(OP_ATTR_MODIFIERS_NAME_LC) ||
|
attrType.hasName(OP_ATTR_MODIFY_TIMESTAMP_LC);
|
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public DN[] getBaseDNs()
|
{
|
return baseDNs;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public long getEntryCount()
|
{
|
// There is always only a single entry in this backend.
|
return 1;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean isLocal()
|
{
|
// For the purposes of this method, this is a local backend.
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean isIndexed(AttributeType attributeType, IndexType indexType)
|
{
|
// All searches in this backend will always be considered indexed.
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public ConditionResult hasSubordinates(DN entryDN)
|
throws DirectoryException
|
{
|
return ConditionResult.FALSE;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public long numSubordinates(DN entryDN, boolean subtree)
|
throws DirectoryException
|
{
|
return 0L;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public Entry getEntry(DN entryDN)
|
throws DirectoryException
|
{
|
// If the requested entry was one of the schema entries, then create and
|
// return it.
|
DN[] dnArray = baseDNs;
|
for (DN baseDN : dnArray)
|
{
|
if (entryDN.equals(baseDN))
|
{
|
return getSchemaEntry(entryDN, false, true);
|
}
|
}
|
|
|
// There is never anything below the schema entries, so we will return null.
|
return null;
|
}
|
|
|
/**
|
* Generates and returns a schema entry for the Directory Server.
|
*
|
* @param entryDN The DN to use for the generated entry.
|
* @param includeSchemaFile A boolean indicating if the X-SCHEMA-FILE
|
* extension should be used when generating
|
* the entry.
|
*
|
* @return The schema entry that was generated.
|
*/
|
public Entry getSchemaEntry(DN entryDN, boolean includeSchemaFile)
|
{
|
return getSchemaEntry(entryDN, includeSchemaFile, false);
|
}
|
|
/**
|
* Generates and returns a schema entry for the Directory Server.
|
*
|
* @param entryDN The DN to use for the generated entry.
|
* @param includeSchemaFile A boolean indicating if the X-SCHEMA-FILE
|
* extension should be used when generating
|
* the entry.
|
* @param ignoreShowAllOption A boolean indicating if the showAllAttributes
|
* parameter should be ignored or not. It must
|
* only considered for Search operation, and
|
* definitely ignored for Modify operations, i.e.
|
* when calling through getEntry().
|
*
|
* @return The schema entry that was generated.
|
*/
|
private Entry getSchemaEntry(DN entryDN, boolean includeSchemaFile,
|
boolean ignoreShowAllOption)
|
{
|
Map<AttributeType, List<Attribute>> userAttrs =
|
new LinkedHashMap<AttributeType, List<Attribute>>();
|
|
Map<AttributeType, List<Attribute>> operationalAttrs =
|
new LinkedHashMap<AttributeType, List<Attribute>>();
|
|
// Add the RDN attribute(s) for the provided entry.
|
RDN rdn = entryDN.getRDN();
|
if (rdn != null)
|
{
|
int numAVAs = rdn.getNumValues();
|
for (int i = 0; i < numAVAs; i++)
|
{
|
AttributeType attrType = rdn.getAttributeType(i);
|
Attribute a = Attributes.create(attrType, rdn
|
.getAttributeValue(i));
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
attrList.add(a);
|
|
if (attrType.isOperational())
|
{
|
operationalAttrs.put(attrType, attrList);
|
}
|
else
|
{
|
userAttrs.put(attrType, attrList);
|
}
|
}
|
}
|
|
Schema schema = DirectoryServer.getSchema();
|
|
// Add the "attributeTypes" attribute.
|
Set<AttributeValue> valueSet = DirectoryServer.getAttributeTypeSet();
|
|
// Add the file name to the description of the attribute type if
|
// this was requested by the caller.
|
if (includeSchemaFile)
|
{
|
Set<AttributeValue> newValueSet =
|
new LinkedHashSet<AttributeValue>(valueSet.size());
|
|
for (AttributeValue value : valueSet)
|
{
|
try
|
{
|
// Build a new attribute from this value,
|
// get the File name from this attribute, build a new
|
// attribute including this file name.
|
AttributeType attrType = AttributeTypeSyntax.decodeAttributeType(
|
value.getValue(), schema, false);
|
attrType = DirectoryServer.getAttributeType(attrType.getOID());
|
|
newValueSet.add(
|
AttributeValues.create(attributeTypesType,
|
attrType.getDefinitionWithFileName()));
|
}
|
catch (DirectoryException e)
|
{
|
newValueSet.add(value);
|
}
|
}
|
valueSet = newValueSet;
|
}
|
|
AttributeBuilder builder = new AttributeBuilder(attributeTypesType,
|
ATTR_ATTRIBUTE_TYPES);
|
builder.setInitialCapacity(valueSet.size());
|
if (AttributeTypeSyntax.isStripSyntaxMinimumUpperBound())
|
{
|
for (AttributeValue v : valueSet)
|
{
|
// If it exists, strip the minimum upper bound value from the
|
// attribute value.
|
if (v.toString().indexOf('{') != -1)
|
{
|
// Create an attribute value from the stripped string and
|
// add it to the valueset.
|
String strippedStr = v.toString().replaceFirst(
|
stripMinUpperBoundRegEx, "");
|
ByteString s = ByteString.valueOf(strippedStr);
|
AttributeValue strippedVal = AttributeValues.create(s, s);
|
builder.add(strippedVal);
|
}
|
else
|
{
|
builder.add(v);
|
}
|
}
|
}
|
else
|
{
|
builder.addAll(valueSet);
|
}
|
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
if (attributeTypesType.isOperational() &&
|
(ignoreShowAllOption || (!showAllAttributes)))
|
{
|
operationalAttrs.put(attributeTypesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(attributeTypesType, attrList);
|
}
|
|
// Add the "objectClasses" attribute.
|
valueSet = DirectoryServer.getObjectClassSet();
|
|
// Add the file name to the description if this was requested by
|
// the caller.
|
if (includeSchemaFile)
|
{
|
Set<AttributeValue> newValueSet =
|
new LinkedHashSet<AttributeValue>(valueSet.size());
|
|
for (AttributeValue value : valueSet)
|
{
|
try
|
{
|
// Build a new attribute from this value, get the File name
|
// from this attribute, build a new attribute including this
|
// file name.
|
ObjectClass oc = ObjectClassSyntax.decodeObjectClass(
|
value.getValue(), schema, false);
|
oc = DirectoryServer.getObjectClass(oc.getOID());
|
newValueSet.add(AttributeValues.create(
|
objectClassesType, oc.getDefinitionWithFileName()));
|
}
|
catch (DirectoryException e)
|
{
|
newValueSet.add(value);
|
}
|
}
|
valueSet = newValueSet;
|
}
|
|
builder = new AttributeBuilder(objectClassesType, ATTR_OBJECTCLASSES);
|
builder.addAll(valueSet);
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (objectClassesType.isOperational() &&
|
(ignoreShowAllOption || (!showAllAttributes)))
|
{
|
operationalAttrs.put(objectClassesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(objectClassesType, attrList);
|
}
|
|
// Add the "matchingRules" attribute.
|
builder = new AttributeBuilder(matchingRulesType, ATTR_MATCHING_RULES);
|
builder.addAll(DirectoryServer.getMatchingRuleSet());
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (matchingRulesType.isOperational() &&
|
(ignoreShowAllOption || (!showAllAttributes)))
|
{
|
operationalAttrs.put(matchingRulesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(matchingRulesType, attrList);
|
}
|
|
// Add the "ldapSyntaxes" attribute.
|
builder = new AttributeBuilder(ldapSyntaxesType, ATTR_LDAP_SYNTAXES);
|
builder.addAll(DirectoryServer.getAttributeSyntaxSet());
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
// Note that we intentionally ignore showAllAttributes for
|
// attribute syntaxes, name forms, matching rule uses, DIT content
|
// rules, and DIT structure rules because those attributes aren't
|
// allowed in the subschema objectclass, and treating them as user
|
// attributes would cause schema updates to fail. This means that
|
// you'll always have to explicitly request these attributes in
|
// order to be able to see them.
|
if (ldapSyntaxesType.isOperational())
|
{
|
operationalAttrs.put(ldapSyntaxesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(ldapSyntaxesType, attrList);
|
}
|
|
// If there are any name forms defined, then add them.
|
valueSet = DirectoryServer.getNameFormSet();
|
if (!valueSet.isEmpty())
|
{
|
builder = new AttributeBuilder(nameFormsType, ATTR_NAME_FORMS);
|
builder.addAll(valueSet);
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (nameFormsType.isOperational())
|
{
|
operationalAttrs.put(nameFormsType, attrList);
|
}
|
else
|
{
|
userAttrs.put(nameFormsType, attrList);
|
}
|
}
|
|
// If there are any DIT content rules defined, then add them.
|
valueSet = DirectoryServer.getDITContentRuleSet();
|
if (!valueSet.isEmpty())
|
{
|
builder = new AttributeBuilder(ditContentRulesType,
|
ATTR_DIT_CONTENT_RULES);
|
builder.addAll(valueSet);
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (ditContentRulesType.isOperational())
|
{
|
operationalAttrs.put(ditContentRulesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(ditContentRulesType, attrList);
|
}
|
}
|
|
// If there are any DIT structure rules defined, then add them.
|
valueSet = DirectoryServer.getDITStructureRuleSet();
|
if (!valueSet.isEmpty())
|
{
|
builder = new AttributeBuilder(ditStructureRulesType,
|
ATTR_DIT_STRUCTURE_RULES);
|
builder.addAll(valueSet);
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (ditStructureRulesType.isOperational())
|
{
|
operationalAttrs.put(ditStructureRulesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(ditStructureRulesType, attrList);
|
}
|
}
|
|
// If there are any matching rule uses defined, then add them.
|
valueSet = DirectoryServer.getMatchingRuleUseSet();
|
if (!valueSet.isEmpty())
|
{
|
builder = new AttributeBuilder(matchingRuleUsesType,
|
ATTR_MATCHING_RULE_USE);
|
builder.addAll(valueSet);
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(builder.toAttribute());
|
|
if (matchingRuleUsesType.isOperational())
|
{
|
operationalAttrs.put(matchingRuleUsesType, attrList);
|
}
|
else
|
{
|
userAttrs.put(matchingRuleUsesType, attrList);
|
}
|
}
|
|
// Add the lastmod attributes.
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(Attributes.create(creatorsNameType, creatorsName));
|
operationalAttrs.put(creatorsNameType, attrList);
|
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(Attributes.create(createTimestampType, createTimestamp));
|
operationalAttrs.put(createTimestampType, attrList);
|
|
if (DirectoryServer.getSchema().getYoungestModificationTime() != modifyTime)
|
{
|
synchronized (this)
|
{
|
modifyTime = DirectoryServer.getSchema().getYoungestModificationTime();
|
modifyTimestamp = GeneralizedTimeSyntax
|
.createGeneralizedTimeValue(modifyTime);
|
}
|
}
|
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(Attributes.create(modifiersNameType, modifiersName));
|
operationalAttrs.put(modifiersNameType, attrList);
|
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(Attributes.create(modifyTimestampType, modifyTimestamp));
|
operationalAttrs.put(modifyTimestampType, attrList);
|
|
// Add the extra attributes.
|
Map<String, Attribute> attributes = DirectoryServer.getSchema()
|
.getExtraAttributes();
|
for (Attribute attribute : attributes.values())
|
{
|
attrList = new ArrayList<Attribute>(1);
|
attrList.add(attribute);
|
operationalAttrs.put(attribute.getAttributeType(), attrList);
|
}
|
|
// Add all the user-defined attributes.
|
for (Attribute a : userDefinedAttributes)
|
{
|
AttributeType type = a.getAttributeType();
|
|
if (type.isOperational())
|
{
|
List<Attribute> attrs = operationalAttrs.get(type);
|
if (attrs == null)
|
{
|
attrs = new ArrayList<Attribute>();
|
attrs.add(a);
|
operationalAttrs.put(type, attrs);
|
}
|
else
|
{
|
attrs.add(a);
|
}
|
}
|
else
|
{
|
List<Attribute> attrs = userAttrs.get(type);
|
if (attrs == null)
|
{
|
attrs = new ArrayList<Attribute>();
|
attrs.add(a);
|
userAttrs.put(type, attrs);
|
}
|
else
|
{
|
attrs.add(a);
|
}
|
}
|
}
|
|
// Construct and return the entry.
|
Entry e = new Entry(entryDN, schemaObjectClasses, userAttrs,
|
operationalAttrs);
|
e.processVirtualAttributes();
|
return e;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean entryExists(DN entryDN)
|
throws DirectoryException
|
{
|
// The specified DN must be one of the specified schema DNs.
|
DN[] baseArray = baseDNs;
|
for (DN baseDN : baseArray)
|
{
|
if (entryDN.equals(baseDN))
|
{
|
return true;
|
}
|
}
|
|
return false;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void addEntry(Entry entry, AddOperation addOperation)
|
throws DirectoryException
|
{
|
Message message =
|
ERR_SCHEMA_ADD_NOT_SUPPORTED.get(String.valueOf(entry.getDN()));
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void deleteEntry(DN entryDN, DeleteOperation deleteOperation)
|
throws DirectoryException
|
{
|
Message message =
|
ERR_SCHEMA_DELETE_NOT_SUPPORTED.get(String.valueOf(entryDN));
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void replaceEntry(Entry oldEntry, Entry newEntry,
|
ModifyOperation modifyOperation) throws DirectoryException
|
{
|
// Make sure that the authenticated user has the necessary UPDATE_SCHEMA
|
// privilege.
|
ClientConnection clientConnection = modifyOperation.getClientConnection();
|
if (! clientConnection.hasPrivilege(Privilege.UPDATE_SCHEMA,
|
modifyOperation))
|
{
|
Message message = ERR_SCHEMA_MODIFY_INSUFFICIENT_PRIVILEGES.get();
|
throw new DirectoryException(ResultCode.INSUFFICIENT_ACCESS_RIGHTS,
|
message);
|
}
|
|
|
ArrayList<Modification> mods =
|
new ArrayList<Modification>(modifyOperation.getModifications());
|
if (mods.isEmpty())
|
{
|
// There aren't any modifications, so we don't need to do anything.
|
return;
|
}
|
|
Schema newSchema = DirectoryServer.getSchema().duplicate();
|
TreeSet<String> modifiedSchemaFiles = new TreeSet<String>();
|
|
int pos = -1;
|
for (Modification m : mods)
|
{
|
pos++;
|
|
// Determine the type of modification to perform. We will support add and
|
// delete operations in the schema, and we will also support the ability
|
// to add a schema element that already exists and treat it as a
|
// replacement of that existing element.
|
Attribute a = m.getAttribute();
|
AttributeType at = a.getAttributeType();
|
switch (m.getModificationType())
|
{
|
case ADD:
|
if (at.equals(attributeTypesType))
|
{
|
for (AttributeValue v : a)
|
{
|
AttributeType type;
|
try
|
{
|
type = AttributeTypeSyntax.decodeAttributeType(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_ATTRTYPE.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addAttributeType(type, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(objectClassesType))
|
{
|
for (AttributeValue v : a)
|
{
|
ObjectClass oc;
|
try
|
{
|
oc = ObjectClassSyntax.decodeObjectClass(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_OBJECTCLASS.
|
get(v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addObjectClass(oc, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(nameFormsType))
|
{
|
for (AttributeValue v : a)
|
{
|
NameForm nf;
|
try
|
{
|
nf = NameFormSyntax.decodeNameForm(v.getValue(), newSchema,
|
false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_NAME_FORM.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addNameForm(nf, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ditContentRulesType))
|
{
|
for (AttributeValue v : a)
|
{
|
DITContentRule dcr;
|
try
|
{
|
dcr = DITContentRuleSyntax.decodeDITContentRule(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_DCR.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addDITContentRule(dcr, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ditStructureRulesType))
|
{
|
for (AttributeValue v : a)
|
{
|
DITStructureRule dsr;
|
try
|
{
|
dsr = DITStructureRuleSyntax.decodeDITStructureRule(
|
v.getValue(), newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_DSR.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addDITStructureRule(dsr, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(matchingRuleUsesType))
|
{
|
for (AttributeValue v : a)
|
{
|
MatchingRuleUse mru;
|
try
|
{
|
mru = MatchingRuleUseSyntax.decodeMatchingRuleUse(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_MR_USE.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
addMatchingRuleUse(mru, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ldapSyntaxesType))
|
{
|
for (AttributeValue v : a)
|
{
|
LDAPSyntaxDescription lsd;
|
try
|
{
|
lsd = LDAPSyntaxDescriptionSyntax.decodeLDAPSyntax(
|
v.getValue(), newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message =
|
ERR_SCHEMA_MODIFY_CANNOT_DECODE_LDAP_SYNTAX.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
addLdapSyntaxDescription(lsd, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_UNSUPPORTED_ATTRIBUTE_TYPE.get(a.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
|
break;
|
|
|
case DELETE:
|
if (a.isEmpty())
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_DELETE_NO_VALUES.get(a.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
|
if (at.equals(attributeTypesType))
|
{
|
for (AttributeValue v : a)
|
{
|
AttributeType type;
|
try
|
{
|
type = AttributeTypeSyntax.decodeAttributeType(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_ATTRTYPE.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeAttributeType(type, newSchema, mods, pos,
|
modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(objectClassesType))
|
{
|
for (AttributeValue v : a)
|
{
|
ObjectClass oc;
|
try
|
{
|
oc = ObjectClassSyntax.decodeObjectClass(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_OBJECTCLASS.
|
get(v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeObjectClass(oc, newSchema, mods, pos, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(nameFormsType))
|
{
|
for (AttributeValue v : a)
|
{
|
NameForm nf;
|
try
|
{
|
nf = NameFormSyntax.decodeNameForm(v.getValue(), newSchema,
|
false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_NAME_FORM.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeNameForm(nf, newSchema, mods, pos, modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ditContentRulesType))
|
{
|
for (AttributeValue v : a)
|
{
|
DITContentRule dcr;
|
try
|
{
|
dcr = DITContentRuleSyntax.decodeDITContentRule(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_DCR.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeDITContentRule(dcr, newSchema, mods, pos,
|
modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ditStructureRulesType))
|
{
|
for (AttributeValue v : a)
|
{
|
DITStructureRule dsr;
|
try
|
{
|
dsr = DITStructureRuleSyntax.decodeDITStructureRule(
|
v.getValue(), newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_DSR.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeDITStructureRule(dsr, newSchema, mods, pos,
|
modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(matchingRuleUsesType))
|
{
|
for (AttributeValue v : a)
|
{
|
MatchingRuleUse mru;
|
try
|
{
|
mru = MatchingRuleUseSyntax.decodeMatchingRuleUse(v.getValue(),
|
newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_MR_USE.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
|
removeMatchingRuleUse(mru, newSchema, mods, pos,
|
modifiedSchemaFiles);
|
}
|
}
|
else if (at.equals(ldapSyntaxesType))
|
{
|
for (AttributeValue v : a)
|
{
|
LDAPSyntaxDescription lsd;
|
try
|
{
|
lsd = LDAPSyntaxDescriptionSyntax.decodeLDAPSyntax(
|
v.getValue(), newSchema, false);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message =
|
ERR_SCHEMA_MODIFY_CANNOT_DECODE_LDAP_SYNTAX.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message, de);
|
}
|
removeLdapSyntaxDescription(lsd, newSchema, modifiedSchemaFiles);
|
}
|
}
|
else
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_UNSUPPORTED_ATTRIBUTE_TYPE.get(a.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
|
break;
|
|
|
case REPLACE:
|
if ((!m.isInternal()) &&
|
(!modifyOperation.isSynchronizationOperation()))
|
{
|
Message message = ERR_SCHEMA_INVALID_MODIFICATION_TYPE.get(
|
String.valueOf(m.getModificationType()));
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
else
|
{
|
// If this is not a Schema attribute, we put it
|
// in the extraAttribute map. This in fact acts as a replace.
|
if (SchemaConfigManager.isSchemaAttribute(a))
|
{
|
Message message = ERR_SCHEMA_INVALID_REPLACE_MODIFICATION.get(
|
a.getNameWithOptions());
|
ErrorLogger.logError(message);
|
}
|
else
|
{
|
newSchema.addExtraAttribute(at.getNameOrOID(), a);
|
modifiedSchemaFiles.add(FILE_USER_SCHEMA_ELEMENTS);
|
}
|
}
|
break;
|
|
default:
|
Message message = ERR_SCHEMA_INVALID_MODIFICATION_TYPE.get(
|
String.valueOf(m.getModificationType()));
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
}
|
|
|
// If we've gotten here, then everything looks OK, re-write all the
|
// modified Schema Files.
|
updateSchemaFiles(newSchema, modifiedSchemaFiles);
|
|
// Finally set DirectoryServer to use the new Schema.
|
DirectoryServer.setSchema(newSchema);
|
|
|
DN authzDN = modifyOperation.getAuthorizationDN();
|
if (authzDN == null)
|
{
|
authzDN = DN.nullDN();
|
}
|
|
modifiersName = AttributeValues.create(
|
modifiersNameType, authzDN.toString());
|
modifyTimestamp = GeneralizedTimeSyntax.createGeneralizedTimeValue(
|
System.currentTimeMillis());
|
}
|
|
|
|
/**
|
* Re-write all schema files using the provided new Schema and list of
|
* modified files.
|
*
|
* @param newSchema The new schema that should be used.
|
*
|
* @param modifiedSchemaFiles The list of files that should be modified.
|
*
|
* @throws DirectoryException When the new file cannot be written.
|
*/
|
private void updateSchemaFiles(
|
Schema newSchema, TreeSet<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// We'll re-write all
|
// impacted schema files by first creating them in a temporary location
|
// and then replacing the existing schema files with the new versions.
|
// If all that goes successfully, then activate the new schema.
|
HashMap<String,File> tempSchemaFiles = new HashMap<String,File>();
|
try
|
{
|
for (String schemaFile : modifiedSchemaFiles)
|
{
|
File tempSchemaFile = writeTempSchemaFile(newSchema, schemaFile);
|
tempSchemaFiles.put(schemaFile, tempSchemaFile);
|
}
|
|
installSchemaFiles(tempSchemaFiles);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
throw de;
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message =
|
ERR_SCHEMA_MODIFY_CANNOT_WRITE_NEW_SCHEMA.get(getExceptionMessage(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
finally
|
{
|
cleanUpTempSchemaFiles(tempSchemaFiles);
|
}
|
|
|
// Create a single file with all of the concatenated schema information
|
// that we can use on startup to detect whether the schema files have been
|
// edited with the server offline.
|
Schema.writeConcatenatedSchema();
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided attribute type to
|
* the given schema, replacing an existing type if necessary, and ensuring all
|
* other metadata is properly updated.
|
*
|
* @param attributeType The attribute type to add or replace in the
|
* server schema.
|
* @param schema The schema to which the attribute type should
|
* be added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided attribute type to the server
|
* schema.
|
*/
|
private void addAttributeType(AttributeType attributeType, Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified attribute type already exists. We'll check
|
// the OID and all of the names, which means that it's possible there could
|
// be more than one match (although if there is, then we'll refuse the
|
// operation).
|
AttributeType existingType =
|
schema.getAttributeType(attributeType.getOID());
|
for (String name : attributeType.getNormalizedNames())
|
{
|
AttributeType t = schema.getAttributeType(name);
|
if (t == null)
|
{
|
continue;
|
}
|
else if (existingType == null)
|
{
|
existingType = t;
|
}
|
else if (existingType != t)
|
{
|
// NOTE: We really do want to use "!=" instead of "! t.equals()"
|
// because we want to check whether it's the same object instance, not
|
// just a logical equivalent.
|
Message message = ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_ATTRTYPE.
|
get(attributeType.getNameOrOID(), existingType.getNameOrOID(),
|
t.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the new attribute type doesn't reference an undefined
|
// or OBSOLETE superior attribute type.
|
AttributeType superiorType = attributeType.getSuperiorType();
|
if (superiorType != null)
|
{
|
if (! schema.hasAttributeType(superiorType.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_UNDEFINED_SUPERIOR_ATTRIBUTE_TYPE.
|
get(attributeType.getNameOrOID(), superiorType.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (superiorType.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_OBSOLETE_SUPERIOR_ATTRIBUTE_TYPE.
|
get(attributeType.getNameOrOID(), superiorType.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// Make sure that none of the associated matching rules are marked OBSOLETE.
|
MatchingRule mr = attributeType.getEqualityMatchingRule();
|
if ((mr != null) && mr.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_ATTRTYPE_OBSOLETE_MR.get(
|
attributeType.getNameOrOID(), mr.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
mr = attributeType.getOrderingMatchingRule();
|
if ((mr != null) && mr.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_ATTRTYPE_OBSOLETE_MR.get(
|
attributeType.getNameOrOID(), mr.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
mr = attributeType.getSubstringMatchingRule();
|
if ((mr != null) && mr.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_ATTRTYPE_OBSOLETE_MR.get(
|
attributeType.getNameOrOID(), mr.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
mr = attributeType.getApproximateMatchingRule();
|
if ((mr != null) && mr.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_ATTRTYPE_OBSOLETE_MR.get(
|
attributeType.getNameOrOID(), mr.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
|
// If there is no existing type, then we're adding a new attribute.
|
// Otherwise, we're replacing an existing one.
|
if (existingType == null)
|
{
|
schema.registerAttributeType(attributeType, false);
|
String schemaFile = attributeType.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
attributeType.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterAttributeType(existingType);
|
schema.registerAttributeType(attributeType, false);
|
schema.rebuildDependentElements(existingType);
|
|
if ((attributeType.getSchemaFile() == null) ||
|
(attributeType.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingType.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
attributeType.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = attributeType.getSchemaFile();
|
String oldSchemaFile = existingType.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided attribute type from
|
* the server schema, ensuring all other metadata is properly updated. Note
|
* that this method will first check to see whether the same attribute type
|
* will be later added to the server schema with an updated definition, and if
|
* so then the removal will be ignored because the later add will be handled
|
* as a replace. If the attribute type will not be replaced with a new
|
* definition, then this method will ensure that there are no other schema
|
* elements that depend on the attribute type before allowing it to be
|
* removed.
|
*
|
* @param attributeType The attribute type to remove from the server
|
* schema.
|
* @param schema The schema from which the attribute type
|
* should be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided attribute type from the server
|
* schema.
|
*/
|
private void removeAttributeType(AttributeType attributeType, Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified attribute type is actually defined in the server
|
// schema. If not, then fail.
|
AttributeType removeType = schema.getAttributeType(attributeType.getOID());
|
if ((removeType == null) || (! removeType.equals(attributeType)))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_ATTRIBUTE_TYPE.get(
|
attributeType.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// See if there is another modification later to add the attribute type back
|
// into the schema. If so, then it's a replace and we should ignore the
|
// remove because adding it back will handle the replace.
|
for (int i=currentPosition+1; i < modifications.size(); i++)
|
{
|
Modification m = modifications.get(i);
|
Attribute a = m.getAttribute();
|
|
if ((m.getModificationType() != ModificationType.ADD) ||
|
(! a.getAttributeType().equals(attributeTypesType)))
|
{
|
continue;
|
}
|
|
for (AttributeValue v : a)
|
{
|
AttributeType at;
|
try
|
{
|
at = AttributeTypeSyntax.decodeAttributeType(v.getValue(), schema,
|
true);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_ATTRTYPE.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message,
|
de);
|
}
|
|
if (attributeType.getOID().equals(at.getOID()))
|
{
|
// We found a match where the attribute type is added back later, so
|
// we don't need to do anything else here.
|
return;
|
}
|
}
|
}
|
|
|
// Make sure that the attribute type isn't used as the superior type for
|
// any other attributes.
|
for (AttributeType at : schema.getAttributeTypes().values())
|
{
|
AttributeType superiorType = at.getSuperiorType();
|
if ((superiorType != null) && superiorType.equals(removeType))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_AT_SUPERIOR_TYPE.get(
|
removeType.getNameOrOID(), superiorType.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the attribute type isn't used as a required or optional
|
// attribute type in any objectclass.
|
for (ObjectClass oc : schema.getObjectClasses().values())
|
{
|
if (oc.getRequiredAttributes().contains(removeType) ||
|
oc.getOptionalAttributes().contains(removeType))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_AT_IN_OC.get(
|
removeType.getNameOrOID(), oc.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the attribute type isn't used as a required or optional
|
// attribute type in any name form.
|
for (List<NameForm> mappedForms :
|
schema.getNameFormsByObjectClass().values())
|
{
|
for(NameForm nf : mappedForms)
|
{
|
if (nf.getRequiredAttributes().contains(removeType) ||
|
nf.getOptionalAttributes().contains(removeType))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_AT_IN_NF.get(
|
removeType.getNameOrOID(), nf.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
}
|
}
|
|
|
// Make sure that the attribute type isn't used as a required, optional, or
|
// prohibited attribute type in any DIT content rule.
|
for (DITContentRule dcr : schema.getDITContentRules().values())
|
{
|
if (dcr.getRequiredAttributes().contains(removeType) ||
|
dcr.getOptionalAttributes().contains(removeType) ||
|
dcr.getProhibitedAttributes().contains(removeType))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_AT_IN_DCR.get(
|
removeType.getNameOrOID(), dcr.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the attribute type isn't referenced by any matching rule
|
// use.
|
for (MatchingRuleUse mru : schema.getMatchingRuleUses().values())
|
{
|
if (mru.getAttributes().contains(removeType))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_AT_IN_MR_USE.get(
|
removeType.getNameOrOID(), mru.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// If we've gotten here, then it's OK to remove the attribute type from
|
// the schema.
|
schema.deregisterAttributeType(removeType);
|
String schemaFile = removeType.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided objectclass to the
|
* given schema, replacing an existing class if necessary, and ensuring
|
* all other metadata is properly updated.
|
*
|
* @param objectClass The objectclass to add or replace in the
|
* server schema.
|
* @param schema The schema to which the objectclass should be
|
* added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided objectclass to the server schema.
|
*/
|
private void addObjectClass(ObjectClass objectClass, Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified objectclass already exists. We'll check the
|
// OID and all of the names, which means that it's possible there could be
|
// more than one match (although if there is, then we'll refuse the
|
// operation).
|
ObjectClass existingClass =
|
schema.getObjectClass(objectClass.getOID());
|
for (String name : objectClass.getNormalizedNames())
|
{
|
ObjectClass oc = schema.getObjectClass(name);
|
if (oc == null)
|
{
|
continue;
|
}
|
else if (existingClass == null)
|
{
|
existingClass = oc;
|
}
|
else if (existingClass != oc)
|
{
|
// NOTE: We really do want to use "!=" instead of "! t.equals()"
|
// because we want to check whether it's the same object instance, not
|
// just a logical equivalent.
|
Message message =
|
ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_OBJECTCLASS
|
.get(objectClass.getNameOrOID(),
|
existingClass.getNameOrOID(),
|
oc.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the new objectclass doesn't reference an undefined
|
// superior class, or an undefined required or optional attribute type,
|
// and that none of them are OBSOLETE.
|
for(ObjectClass superiorClass : objectClass.getSuperiorClasses())
|
{
|
if (! schema.hasObjectClass(superiorClass.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_UNDEFINED_SUPERIOR_OBJECTCLASS.get(
|
objectClass.getNameOrOID(), superiorClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (superiorClass.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_OBSOLETE_SUPERIOR_OBJECTCLASS.get(
|
objectClass.getNameOrOID(), superiorClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
for (AttributeType at : objectClass.getRequiredAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_OC_UNDEFINED_REQUIRED_ATTR.get(
|
objectClass.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_OC_OBSOLETE_REQUIRED_ATTR.get(
|
objectClass.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
for (AttributeType at : objectClass.getOptionalAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_OC_UNDEFINED_OPTIONAL_ATTR.get(
|
objectClass.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_OC_OBSOLETE_OPTIONAL_ATTR.get(
|
objectClass.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// If there is no existing class, then we're adding a new objectclass.
|
// Otherwise, we're replacing an existing one.
|
if (existingClass == null)
|
{
|
schema.registerObjectClass(objectClass, false);
|
String schemaFile = objectClass.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
objectClass.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterObjectClass(existingClass);
|
schema.registerObjectClass(objectClass, false);
|
schema.rebuildDependentElements(existingClass);
|
|
if ((objectClass.getSchemaFile() == null) ||
|
(objectClass.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingClass.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
objectClass.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = objectClass.getSchemaFile();
|
String oldSchemaFile = existingClass.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided objectclass from the
|
* server schema, ensuring all other metadata is properly updated. Note that
|
* this method will first check to see whether the same objectclass will be
|
* later added to the server schema with an updated definition, and if so then
|
* the removal will be ignored because the later add will be handled as a
|
* replace. If the objectclass will not be replaced with a new definition,
|
* then this method will ensure that there are no other schema elements that
|
* depend on the objectclass before allowing it to be removed.
|
*
|
* @param objectClass The objectclass to remove from the server
|
* schema.
|
* @param schema The schema from which the objectclass should
|
* be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided objectclass from the server
|
* schema.
|
*/
|
private void removeObjectClass(ObjectClass objectClass, Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified objectclass is actually defined in the server
|
// schema. If not, then fail.
|
ObjectClass removeClass = schema.getObjectClass(objectClass.getOID());
|
if ((removeClass == null) || (! removeClass.equals(objectClass)))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_OBJECTCLASS.get(
|
objectClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// See if there is another modification later to add the objectclass back
|
// into the schema. If so, then it's a replace and we should ignore the
|
// remove because adding it back will handle the replace.
|
for (int i=currentPosition+1; i < modifications.size(); i++)
|
{
|
Modification m = modifications.get(i);
|
Attribute a = m.getAttribute();
|
|
if ((m.getModificationType() != ModificationType.ADD) ||
|
(! a.getAttributeType().equals(objectClassesType)))
|
{
|
continue;
|
}
|
|
for (AttributeValue v : a)
|
{
|
ObjectClass oc;
|
try
|
{
|
oc = ObjectClassSyntax.decodeObjectClass(v.getValue(), schema, true);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_OBJECTCLASS.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message,
|
de);
|
}
|
|
if (objectClass.getOID().equals(oc.getOID()))
|
{
|
// We found a match where the objectClass is added back later, so we
|
// don't need to do anything else here.
|
return;
|
}
|
}
|
}
|
|
|
// Make sure that the objectclass isn't used as the superior class for any
|
// other objectclass.
|
for (ObjectClass oc : schema.getObjectClasses().values())
|
{
|
for(ObjectClass superiorClass : oc.getSuperiorClasses())
|
{
|
if (superiorClass.equals(removeClass))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_OC_SUPERIOR_CLASS.get(
|
removeClass.getNameOrOID(), superiorClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
}
|
}
|
|
|
// Make sure that the objectclass isn't used as the structural class for
|
// any name form.
|
List<NameForm> mappedForms = schema.getNameForm(removeClass);
|
if (mappedForms != null)
|
{
|
StringBuilder buffer = new StringBuilder();
|
for(NameForm nf : mappedForms)
|
{
|
buffer.append(nf.getNameOrOID());
|
buffer.append("\t");
|
}
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_OC_IN_NF.get(
|
removeClass.getNameOrOID(), buffer.toString());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// Make sure that the objectclass isn't used as a structural or auxiliary
|
// class for any DIT content rule.
|
for (DITContentRule dcr : schema.getDITContentRules().values())
|
{
|
if (dcr.getStructuralClass().equals(removeClass) ||
|
dcr.getAuxiliaryClasses().contains(removeClass))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_OC_IN_DCR.get(
|
removeClass.getNameOrOID(), dcr.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// If we've gotten here, then it's OK to remove the objectclass from the
|
// schema.
|
schema.deregisterObjectClass(removeClass);
|
String schemaFile = removeClass.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided name form to the
|
* the given schema, replacing an existing name form if necessary, and
|
* ensuring all other metadata is properly updated.
|
*
|
* @param nameForm The name form to add or replace in the server
|
* schema.
|
* @param schema The schema to which the name form should be
|
* added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided name form to the server schema.
|
*/
|
private void addNameForm(NameForm nameForm, Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified name form already exists. We'll check the
|
// OID and all of the names, which means that it's possible there could be
|
// more than one match (although if there is, then we'll refuse the
|
// operation).
|
NameForm existingNF =
|
schema.getNameForm(nameForm.getOID());
|
for (String name : nameForm.getNames().keySet())
|
{
|
NameForm nf = schema.getNameForm(name);
|
if (nf == null)
|
{
|
continue;
|
}
|
else if (existingNF == null)
|
{
|
existingNF = nf;
|
}
|
else if (existingNF != nf)
|
{
|
// NOTE: We really do want to use "!=" instead of "! t.equals()"
|
// because we want to check whether it's the same object instance, not
|
// just a logical equivalent.
|
Message message =
|
ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_NAME_FORM
|
.get(nameForm.getNameOrOID(), existingNF.getNameOrOID(),
|
nf.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// Make sure that the new name form doesn't reference an undefined
|
// structural class, or an undefined required or optional attribute type, or
|
// that any of them are marked OBSOLETE.
|
ObjectClass structuralClass = nameForm.getStructuralClass();
|
if (! schema.hasObjectClass(structuralClass.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_UNDEFINED_STRUCTURAL_OC.get(
|
nameForm.getNameOrOID(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
if (structuralClass.getObjectClassType() != ObjectClassType.STRUCTURAL)
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_OC_NOT_STRUCTURAL.get(
|
nameForm.getNameOrOID(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
if (structuralClass.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_OC_OBSOLETE.get(
|
nameForm.getNameOrOID(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
for (AttributeType at : nameForm.getRequiredAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_UNDEFINED_REQUIRED_ATTR.get(
|
nameForm.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_OBSOLETE_REQUIRED_ATTR.get(
|
nameForm.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
for (AttributeType at : nameForm.getOptionalAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_UNDEFINED_OPTIONAL_ATTR.get(
|
nameForm.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_NF_OBSOLETE_OPTIONAL_ATTR.get(
|
nameForm.getNameOrOID(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// If there is no existing class, then we're adding a new name form.
|
// Otherwise, we're replacing an existing one.
|
if (existingNF == null)
|
{
|
schema.registerNameForm(nameForm, false);
|
String schemaFile = nameForm.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
nameForm.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterNameForm(existingNF);
|
schema.registerNameForm(nameForm, false);
|
schema.rebuildDependentElements(existingNF);
|
|
if ((nameForm.getSchemaFile() == null) ||
|
(nameForm.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingNF.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
nameForm.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = nameForm.getSchemaFile();
|
String oldSchemaFile = existingNF.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided name form from the
|
* server schema, ensuring all other metadata is properly updated. Note that
|
* this method will first check to see whether the same name form will be
|
* later added to the server schema with an updated definition, and if so then
|
* the removal will be ignored because the later add will be handled as a
|
* replace. If the name form will not be replaced with a new definition, then
|
* this method will ensure that there are no other schema elements that depend
|
* on the name form before allowing it to be removed.
|
*
|
* @param nameForm The name form to remove from the server
|
* schema.
|
* @param schema The schema from which the name form should be
|
* be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided name form from the server schema.
|
*/
|
private void removeNameForm(NameForm nameForm, Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified name form is actually defined in the server schema.
|
// If not, then fail.
|
NameForm removeNF = schema.getNameForm(nameForm.getOID());
|
if ((removeNF == null) || (! removeNF.equals(nameForm)))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_NAME_FORM.get(
|
nameForm.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// See if there is another modification later to add the name form back
|
// into the schema. If so, then it's a replace and we should ignore the
|
// remove because adding it back will handle the replace.
|
for (int i=currentPosition+1; i < modifications.size(); i++)
|
{
|
Modification m = modifications.get(i);
|
Attribute a = m.getAttribute();
|
|
if ((m.getModificationType() != ModificationType.ADD) ||
|
(! a.getAttributeType().equals(nameFormsType)))
|
{
|
continue;
|
}
|
|
for (AttributeValue v : a)
|
{
|
NameForm nf;
|
try
|
{
|
nf = NameFormSyntax.decodeNameForm(v.getValue(), schema, true);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_NAME_FORM.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message,
|
de);
|
}
|
|
if (nameForm.getOID().equals(nf.getOID()))
|
{
|
// We found a match where the name form is added back later, so we
|
// don't need to do anything else here.
|
return;
|
}
|
}
|
}
|
|
|
// Make sure that the name form isn't referenced by any DIT structure
|
// rule.
|
DITStructureRule dsr = schema.getDITStructureRule(removeNF);
|
if (dsr != null)
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NF_IN_DSR.get(
|
removeNF.getNameOrOID(), dsr.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// If we've gotten here, then it's OK to remove the name form from the
|
// schema.
|
schema.deregisterNameForm(removeNF);
|
String schemaFile = removeNF.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided DIT content rule to
|
* the given schema, replacing an existing rule if necessary, and ensuring
|
* all other metadata is properly updated.
|
*
|
* @param ditContentRule The DIT content rule to add or replace in the
|
* server schema.
|
* @param schema The schema to which the DIT content rule
|
* should be added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided DIT content rule to the server
|
* schema.
|
*/
|
private void addDITContentRule(DITContentRule ditContentRule, Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified DIT content rule already exists. We'll check
|
// all of the names, which means that it's possible there could be more than
|
// one match (although if there is, then we'll refuse the operation).
|
DITContentRule existingDCR = null;
|
for (DITContentRule dcr : schema.getDITContentRules().values())
|
{
|
for (String name : ditContentRule.getNames().keySet())
|
{
|
if (dcr.hasName(name))
|
{
|
if (existingDCR == null)
|
{
|
existingDCR = dcr;
|
break;
|
}
|
else
|
{
|
Message message = ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_DCR.
|
get(ditContentRule.getName(), existingDCR.getName(),
|
dcr.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
}
|
}
|
}
|
|
|
// Get the structural class for the new DIT content rule and see if there's
|
// already an existing rule that is associated with that class. If there
|
// is, then it will only be acceptable if it's the DIT content rule that we
|
// are replacing (in which case we really do want to use the "!=" operator).
|
ObjectClass structuralClass = ditContentRule.getStructuralClass();
|
DITContentRule existingRuleForClass =
|
schema.getDITContentRule(structuralClass);
|
if ((existingRuleForClass != null) && (existingRuleForClass != existingDCR))
|
{
|
Message message = ERR_SCHEMA_MODIFY_STRUCTURAL_OC_CONFLICT_FOR_ADD_DCR.
|
get(ditContentRule.getName(), structuralClass.getNameOrOID(),
|
existingRuleForClass.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// Make sure that the new DIT content rule doesn't reference an undefined
|
// structural or auxiliaryclass, or an undefined required, optional, or
|
// prohibited attribute type.
|
if (! schema.hasObjectClass(structuralClass.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_UNDEFINED_STRUCTURAL_OC.get(
|
ditContentRule.getName(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (structuralClass.getObjectClassType() != ObjectClassType.STRUCTURAL)
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OC_NOT_STRUCTURAL.get(
|
ditContentRule.getName(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (structuralClass.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_STRUCTURAL_OC_OBSOLETE.get(
|
ditContentRule.getName(), structuralClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
for (ObjectClass oc : ditContentRule.getAuxiliaryClasses())
|
{
|
if (! schema.hasObjectClass(oc.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_UNDEFINED_AUXILIARY_OC.get(
|
ditContentRule.getName(), oc.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
if (oc.getObjectClassType() != ObjectClassType.AUXILIARY)
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OC_NOT_AUXILIARY.get(
|
ditContentRule.getName(), oc.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
if (oc.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OBSOLETE_AUXILIARY_OC.get(
|
ditContentRule.getName(), oc.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
for (AttributeType at : ditContentRule.getRequiredAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_UNDEFINED_REQUIRED_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OBSOLETE_REQUIRED_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
for (AttributeType at : ditContentRule.getOptionalAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_UNDEFINED_OPTIONAL_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OBSOLETE_OPTIONAL_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
for (AttributeType at : ditContentRule.getProhibitedAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_UNDEFINED_PROHIBITED_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DCR_OBSOLETE_PROHIBITED_ATTR.get(
|
ditContentRule.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// If there is no existing rule, then we're adding a new DIT content rule.
|
// Otherwise, we're replacing an existing one.
|
if (existingDCR == null)
|
{
|
schema.registerDITContentRule(ditContentRule, false);
|
String schemaFile = ditContentRule.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
ditContentRule.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterDITContentRule(existingDCR);
|
schema.registerDITContentRule(ditContentRule, false);
|
schema.rebuildDependentElements(existingDCR);
|
|
if ((ditContentRule.getSchemaFile() == null) ||
|
(ditContentRule.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingDCR.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
ditContentRule.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = ditContentRule.getSchemaFile();
|
String oldSchemaFile = existingDCR.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided DIT content rule
|
* from the server schema, ensuring all other metadata is properly updated.
|
* Note that this method will first check to see whether the same rule will be
|
* later added to the server schema with an updated definition, and if so then
|
* the removal will be ignored because the later add will be handled as a
|
* replace. If the DIT content rule will not be replaced with a new
|
* definition, then this method will ensure that there are no other schema
|
* elements that depend on the rule before allowing it to be removed.
|
*
|
* @param ditContentRule The DIT content rule to remove from the server
|
* schema.
|
* @param schema The schema from which the DIT content rule
|
* should be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided DIT content rule from the server
|
* schema.
|
*/
|
private void removeDITContentRule(DITContentRule ditContentRule,
|
Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified DIT content rule is actually defined in the server
|
// schema. If not, then fail.
|
DITContentRule removeDCR =
|
schema.getDITContentRule(ditContentRule.getStructuralClass());
|
if ((removeDCR == null) || (! removeDCR.equals(ditContentRule)))
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_DCR.get(ditContentRule.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// Since DIT content rules don't have any dependencies, then we don't need
|
// to worry about the difference between a remove or a replace. We can
|
// just remove the DIT content rule now, and if it is added back later then
|
// there still won't be any conflict.
|
schema.deregisterDITContentRule(removeDCR);
|
String schemaFile = removeDCR.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided DIT structure rule
|
* to the given schema, replacing an existing rule if necessary, and ensuring
|
* all other metadata is properly updated.
|
*
|
* @param ditStructureRule The DIT structure rule to add or replace in
|
* the server schema.
|
* @param schema The schema to which the DIT structure rule
|
* should be added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided DIT structure rule to the server
|
* schema.
|
*/
|
private void addDITStructureRule(DITStructureRule ditStructureRule,
|
Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified DIT structure rule already exists. We'll
|
// check the rule ID and all of the names, which means that it's possible
|
// there could be more than one match (although if there is, then we'll
|
// refuse the operation).
|
DITStructureRule existingDSR =
|
schema.getDITStructureRule(ditStructureRule.getRuleID());
|
//Boolean to check if the new rule is in use or not.
|
boolean inUse = false;
|
for (DITStructureRule dsr : schema.getDITStructureRulesByID().values())
|
{
|
for (String name : ditStructureRule.getNames().keySet())
|
{
|
if (dsr.hasName(name))
|
{
|
// We really do want to use the "!=" operator here because it's
|
// acceptable if we find match for the same object instance.
|
if ((existingDSR != null) && (existingDSR != dsr))
|
{
|
Message message = ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_DSR.
|
get(ditStructureRule.getNameOrRuleID(),
|
existingDSR.getNameOrRuleID(), dsr.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
inUse = true;
|
}
|
}
|
}
|
|
if(existingDSR != null && !inUse)
|
{
|
//We have an existing DSR with the same rule id but we couldn't find
|
//any existing rules sharing this name. It means that it is a
|
//new rule with a conflicting rule id.Raise an Exception as the
|
//rule id should be unique.
|
Message message = ERR_SCHEMA_MODIFY_RULEID_CONFLICTS_FOR_ADD_DSR.
|
get(ditStructureRule.getNameOrRuleID(),
|
existingDSR.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
|
// Get the name form for the new DIT structure rule and see if there's
|
// already an existing rule that is associated with that name form. If
|
// there is, then it will only be acceptable if it's the DIT structure rule
|
// that we are replacing (in which case we really do want to use the "!="
|
// operator).
|
NameForm nameForm = ditStructureRule.getNameForm();
|
DITStructureRule existingRuleForNameForm =
|
schema.getDITStructureRule(nameForm);
|
if ((existingRuleForNameForm != null) &&
|
(existingRuleForNameForm != existingDSR))
|
{
|
Message message = ERR_SCHEMA_MODIFY_NAME_FORM_CONFLICT_FOR_ADD_DSR.
|
get(ditStructureRule.getNameOrRuleID(), nameForm.getNameOrOID(),
|
existingRuleForNameForm.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// Make sure that the new DIT structure rule doesn't reference an undefined
|
// name form or superior DIT structure rule.
|
if (! schema.hasNameForm(nameForm.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_DSR_UNDEFINED_NAME_FORM.get(
|
ditStructureRule.getNameOrRuleID(), nameForm.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
if (nameForm.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DSR_OBSOLETE_NAME_FORM.get(
|
ditStructureRule.getNameOrRuleID(), nameForm.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
|
// If there are any superior rules, then make sure none of them are marked
|
// OBSOLETE.
|
for (DITStructureRule dsr : ditStructureRule.getSuperiorRules())
|
{
|
if (dsr.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_DSR_OBSOLETE_SUPERIOR_RULE.get(
|
ditStructureRule.getNameOrRuleID(), dsr.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// If there is no existing rule, then we're adding a new DIT structure rule.
|
// Otherwise, we're replacing an existing one.
|
if (existingDSR == null)
|
{
|
schema.registerDITStructureRule(ditStructureRule, false);
|
String schemaFile = ditStructureRule.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
ditStructureRule.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterDITStructureRule(existingDSR);
|
schema.registerDITStructureRule(ditStructureRule, false);
|
schema.rebuildDependentElements(existingDSR);
|
|
if ((ditStructureRule.getSchemaFile() == null) ||
|
(ditStructureRule.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingDSR.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
ditStructureRule.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = ditStructureRule.getSchemaFile();
|
String oldSchemaFile = existingDSR.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided DIT structure rule
|
* from the server schema, ensuring all other metadata is properly updated.
|
* Note that this method will first check to see whether the same rule will be
|
* later added to the server schema with an updated definition, and if so then
|
* the removal will be ignored because the later add will be handled as a
|
* replace. If the DIT structure rule will not be replaced with a new
|
* definition, then this method will ensure that there are no other schema
|
* elements that depend on the rule before allowing it to be removed.
|
*
|
* @param ditStructureRule The DIT structure rule to remove from the
|
* server schema.
|
* @param schema The schema from which the DIT structure rule
|
* should be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided DIT structure rule from the
|
* server schema.
|
*/
|
private void removeDITStructureRule(DITStructureRule ditStructureRule,
|
Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified DIT structure rule is actually defined in the server
|
// schema. If not, then fail.
|
DITStructureRule removeDSR =
|
schema.getDITStructureRule(ditStructureRule.getRuleID());
|
if ((removeDSR == null) || (! removeDSR.equals(ditStructureRule)))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_DSR.get(
|
ditStructureRule.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// See if there is another modification later to add the DIT structure rule
|
// back into the schema. If so, then it's a replace and we should ignore
|
// the remove because adding it back will handle the replace.
|
for (int i=currentPosition+1; i < modifications.size(); i++)
|
{
|
Modification m = modifications.get(i);
|
Attribute a = m.getAttribute();
|
|
if ((m.getModificationType() != ModificationType.ADD) ||
|
(! a.getAttributeType().equals(ditStructureRulesType)))
|
{
|
continue;
|
}
|
|
for (AttributeValue v : a)
|
{
|
DITStructureRule dsr;
|
try
|
{
|
dsr = DITStructureRuleSyntax.decodeDITStructureRule(
|
v.getValue(), schema, true);
|
}
|
catch (DirectoryException de)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, de);
|
}
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_DECODE_DSR.get(
|
v.getValue().toString(), de.getMessageObject());
|
throw new DirectoryException(
|
ResultCode.INVALID_ATTRIBUTE_SYNTAX, message,
|
de);
|
}
|
|
if (ditStructureRule.getRuleID() == dsr.getRuleID())
|
{
|
// We found a match where the DIT structure rule is added back later,
|
// so we don't need to do anything else here.
|
return;
|
}
|
}
|
}
|
|
|
// Make sure that the DIT structure rule isn't the superior for any other
|
// DIT structure rule.
|
for (DITStructureRule dsr : schema.getDITStructureRulesByID().values())
|
{
|
if (dsr.getSuperiorRules().contains(removeDSR))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_DSR_SUPERIOR_RULE.get(
|
removeDSR.getNameOrRuleID(), dsr.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
}
|
|
|
// If we've gotten here, then it's OK to remove the DIT structure rule from
|
// the schema.
|
schema.deregisterDITStructureRule(removeDSR);
|
String schemaFile = removeDSR.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided matching rule use
|
* to the given schema, replacing an existing use if necessary, and ensuring
|
* all other metadata is properly updated.
|
*
|
* @param matchingRuleUse The matching rule use to add or replace in the
|
* server schema.
|
* @param schema The schema to which the matching rule use
|
* should be added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided matching rule use to the server
|
* schema.
|
*/
|
private void addMatchingRuleUse(MatchingRuleUse matchingRuleUse,
|
Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// First, see if the specified matching rule use already exists. We'll
|
// check all of the names, which means that it's possible that there could
|
// be more than one match (although if there is, then we'll refuse the
|
// operation).
|
MatchingRuleUse existingMRU = null;
|
for (MatchingRuleUse mru : schema.getMatchingRuleUses().values())
|
{
|
for (String name : matchingRuleUse.getNames().keySet())
|
{
|
if (mru.hasName(name))
|
{
|
if (existingMRU == null)
|
{
|
existingMRU = mru;
|
break;
|
}
|
else
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_MULTIPLE_CONFLICTS_FOR_ADD_MR_USE.get(
|
matchingRuleUse.getName(),
|
existingMRU.getName(),
|
mru.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM,
|
message);
|
}
|
}
|
}
|
}
|
|
|
// Get the matching rule for the new matching rule use and see if there's
|
// already an existing matching rule use that is associated with that
|
// matching rule. If there is, then it will only be acceptable if it's the
|
// matching rule use that we are replacing (in which case we really do want
|
// to use the "!=" operator).
|
MatchingRule matchingRule = matchingRuleUse.getMatchingRule();
|
MatchingRuleUse existingMRUForRule =
|
schema.getMatchingRuleUse(matchingRule);
|
if ((existingMRUForRule != null) && (existingMRUForRule != existingMRU))
|
{
|
Message message = ERR_SCHEMA_MODIFY_MR_CONFLICT_FOR_ADD_MR_USE.
|
get(matchingRuleUse.getName(), matchingRule.getNameOrOID(),
|
existingMRUForRule.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (matchingRule.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_MRU_OBSOLETE_MR.get(
|
matchingRuleUse.getName(), matchingRule.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
|
|
// Make sure that the new matching rule use doesn't reference an undefined
|
// attribute type.
|
for (AttributeType at : matchingRuleUse.getAttributes())
|
{
|
if (! schema.hasAttributeType(at.getOID()))
|
{
|
Message message = ERR_SCHEMA_MODIFY_MRU_UNDEFINED_ATTR.get(
|
matchingRuleUse.getName(), at.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
else if (at.isObsolete())
|
{
|
Message message = ERR_SCHEMA_MODIFY_MRU_OBSOLETE_ATTR.get(
|
matchingRuleUse.getName(), matchingRule.getNameOrOID());
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message);
|
}
|
}
|
|
|
// If there is no existing matching rule use, then we're adding a new one.
|
// Otherwise, we're replacing an existing matching rule use.
|
if (existingMRU == null)
|
{
|
schema.registerMatchingRuleUse(matchingRuleUse, false);
|
String schemaFile = matchingRuleUse.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
matchingRuleUse.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterMatchingRuleUse(existingMRU);
|
schema.registerMatchingRuleUse(matchingRuleUse, false);
|
schema.rebuildDependentElements(existingMRU);
|
|
if ((matchingRuleUse.getSchemaFile() == null) ||
|
(matchingRuleUse.getSchemaFile().length() == 0))
|
{
|
String schemaFile = existingMRU.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
matchingRuleUse.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = matchingRuleUse.getSchemaFile();
|
String oldSchemaFile = existingMRU.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required to remove the provided matching rule use
|
* from the server schema, ensuring all other metadata is properly updated.
|
* Note that this method will first check to see whether the same matching
|
* rule use will be later added to the server schema with an updated
|
* definition, and if so then the removal will be ignored because the later
|
* add will be handled as a replace. If the matching rule use will not be
|
* replaced with a new definition, then this method will ensure that there are
|
* no other schema elements that depend on the matching rule use before
|
* allowing it to be removed.
|
*
|
* @param matchingRuleUse The matching rule use to remove from the
|
* server schema.
|
* @param schema The schema from which the matching rule use
|
* should be removed.
|
* @param modifications The full set of modifications to be processed
|
* against the server schema.
|
* @param currentPosition The position of the modification currently
|
* being performed.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to remove
|
* the provided matching rule use from the server
|
* schema.
|
*/
|
private void removeMatchingRuleUse(MatchingRuleUse matchingRuleUse,
|
Schema schema,
|
ArrayList<Modification> modifications,
|
int currentPosition,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
// See if the specified DIT content rule is actually defined in the server
|
// schema. If not, then fail.
|
MatchingRuleUse removeMRU =
|
schema.getMatchingRuleUse(matchingRuleUse.getMatchingRule());
|
if ((removeMRU == null) || (! removeMRU.equals(matchingRuleUse)))
|
{
|
Message message = ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_MR_USE.get(
|
matchingRuleUse.getName());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
// Since matching rule uses don't have any dependencies, then we don't need
|
// to worry about the difference between a remove or a replace. We can
|
// just remove the DIT content rule now, and if it is added back later then
|
// there still won't be any conflict.
|
schema.deregisterMatchingRuleUse(removeMRU);
|
String schemaFile = removeMRU.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Handles all processing required for adding the provided ldap syntax
|
* description to the given schema, replacing an existing ldap syntax
|
* description if necessary, and ensuring all other metadata is properly
|
* updated.
|
*
|
* @param ldapSyntaxDesc The ldap syntax description to add or replace in
|
* the server schema.
|
* @param schema The schema to which the name form should be
|
* added.
|
* @param modifiedSchemaFiles The names of the schema files containing
|
* schema elements that have been updated as part
|
* of the schema modification.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to add
|
* the provided ldap syntax description to the
|
* server schema.
|
*/
|
private void addLdapSyntaxDescription(LDAPSyntaxDescription ldapSyntaxDesc,
|
Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
//Check if there is an existing syntax with this oid.
|
String oid = ldapSyntaxDesc.getLdapSyntaxDescriptionSyntax().getOID();
|
|
// We allow only unimplemented syntaxes to be substituted.
|
if(schema.getSyntax(oid) !=null)
|
{
|
Message message = ERR_ATTR_SYNTAX_INVALID_LDAP_SYNTAX.get(
|
ldapSyntaxDesc.getDefinition(),oid);
|
throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION,
|
message);
|
}
|
|
LDAPSyntaxDescription existingLSD =
|
schema.getLdapSyntaxDescription(oid);
|
|
// If there is no existing lsd, then we're adding a new ldapsyntax.
|
// Otherwise, we're replacing an existing one.
|
if (existingLSD == null)
|
{
|
schema.registerLdapSyntaxDescription(ldapSyntaxDesc, false);
|
String schemaFile = ldapSyntaxDesc.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
ldapSyntaxDesc.setSchemaFile(schemaFile);
|
}
|
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
schema.deregisterLdapSyntaxDescription(existingLSD);
|
schema.registerLdapSyntaxDescription(ldapSyntaxDesc, false);
|
schema.rebuildDependentElements(existingLSD);
|
|
if ((ldapSyntaxDesc.getSchemaFile() == null) ||
|
(ldapSyntaxDesc.getSchemaFile().length() == 0))
|
{
|
String schemaFile = ldapSyntaxDesc.getSchemaFile();
|
if ((schemaFile == null) || (schemaFile.length() == 0))
|
{
|
schemaFile = FILE_USER_SCHEMA_ELEMENTS;
|
}
|
|
ldapSyntaxDesc.setSchemaFile(schemaFile);
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
else
|
{
|
String newSchemaFile = ldapSyntaxDesc.getSchemaFile();
|
String oldSchemaFile = existingLSD.getSchemaFile();
|
if ((oldSchemaFile == null) || oldSchemaFile.equals(newSchemaFile))
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
}
|
else
|
{
|
modifiedSchemaFiles.add(newSchemaFile);
|
modifiedSchemaFiles.add(oldSchemaFile);
|
}
|
}
|
}
|
}
|
|
|
|
//Gets rid of the ldap syntax description.
|
private void removeLdapSyntaxDescription(LDAPSyntaxDescription ldapSyntaxDesc,
|
Schema schema,
|
Set<String> modifiedSchemaFiles)
|
throws DirectoryException
|
{
|
//See if the specified ldap syntax description is actually defined in the
|
//server schema. If not, then fail. Note that we are checking only the
|
//real part of the ldapsyntaxes attribute. A virtual value is not searched
|
// and hence never deleted.
|
String oid = ldapSyntaxDesc.getLdapSyntaxDescriptionSyntax().getOID();
|
LDAPSyntaxDescription removeLSD = schema.getLdapSyntaxDescription(oid);
|
|
if ((removeLSD == null) || (! removeLSD.equals(ldapSyntaxDesc)))
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_REMOVE_NO_SUCH_LSD.get(oid);
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
schema.deregisterLdapSyntaxDescription(removeLSD);
|
String schemaFile = removeLSD.getSchemaFile();
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
|
|
|
/**
|
* Creates an empty entry that may be used as the basis for a new schema file.
|
*
|
* @return An empty entry that may be used as the basis for a new schema
|
* file.
|
*/
|
private Entry createEmptySchemaEntry()
|
{
|
Map<ObjectClass,String> objectClasses =
|
new LinkedHashMap<ObjectClass,String>();
|
objectClasses.put(DirectoryServer.getTopObjectClass(), OC_TOP);
|
objectClasses.put(DirectoryServer.getObjectClass(OC_LDAP_SUBENTRY_LC, true),
|
OC_LDAP_SUBENTRY);
|
objectClasses.put(DirectoryServer.getObjectClass(OC_SUBSCHEMA, true),
|
OC_SUBSCHEMA);
|
|
Map<AttributeType,List<Attribute>> userAttributes =
|
new LinkedHashMap<AttributeType,List<Attribute>>();
|
|
Map<AttributeType,List<Attribute>> operationalAttributes =
|
new LinkedHashMap<AttributeType,List<Attribute>>();
|
|
DN dn = DirectoryServer.getSchemaDN();
|
RDN rdn = dn.getRDN();
|
for (int i=0; i < rdn.getNumValues(); i++)
|
{
|
AttributeType type = rdn.getAttributeType(i);
|
List<Attribute> attrList = new LinkedList<Attribute>();
|
attrList.add(Attributes.create(type, rdn.getAttributeValue(i)));
|
if (type.isOperational())
|
{
|
operationalAttributes.put(type, attrList);
|
}
|
else
|
{
|
userAttributes.put(type, attrList);
|
}
|
}
|
|
return new Entry(dn, objectClasses, userAttributes, operationalAttributes);
|
}
|
|
|
|
|
/**
|
* Writes a temporary version of the specified schema file.
|
*
|
* @param schema The schema from which to take the definitions to be
|
* written.
|
* @param schemaFile The name of the schema file to be written.
|
*
|
* @throws DirectoryException If an unexpected problem occurs while
|
* identifying the schema definitions to include
|
* in the schema file.
|
*
|
* @throws IOException If an unexpected error occurs while attempting to
|
* write the temporary schema file.
|
*
|
* @throws LDIFException If an unexpected problem occurs while generating
|
* the LDIF representation of the schema entry.
|
*/
|
private File writeTempSchemaFile(Schema schema, String schemaFile)
|
throws DirectoryException, IOException, LDIFException
|
{
|
// Start with an empty schema entry.
|
Entry schemaEntry = createEmptySchemaEntry();
|
|
/**
|
* Add all of the ldap syntax descriptions to the schema entry. We do
|
* this only for the real part of the ldapsyntaxes attribute. The real part
|
* is read and write to/from the schema files.
|
*/
|
Set<AttributeValue> values = new LinkedHashSet<AttributeValue>();
|
for (LDAPSyntaxDescription ldapSyntax :
|
schema.getLdapSyntaxDescriptions().values())
|
{
|
if(schemaFile.equals(ldapSyntax.getSchemaFile()))
|
{
|
values.add(AttributeValues.create(ldapSyntaxesType,
|
ldapSyntax.getDefinition()));
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(ldapSyntaxesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(ldapSyntaxesType, attrList);
|
}
|
|
// Add all of the appropriate attribute types to the schema entry. We need
|
// to be careful of the ordering to ensure that any superior types in the
|
// same file are written before the subordinate types.
|
Set<AttributeType> addedTypes = new HashSet<AttributeType>();
|
values = new LinkedHashSet<AttributeValue>();
|
for (AttributeType at : schema.getAttributeTypes().values())
|
{
|
if (schemaFile.equals(at.getSchemaFile()))
|
{
|
addAttrTypeToSchemaFile(schema, schemaFile, at, values, addedTypes, 0);
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(attributeTypesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(attributeTypesType, attrList);
|
}
|
|
|
// Add all of the appropriate objectclasses to the schema entry. We need
|
// to be careful of the ordering to ensure that any superior classes in the
|
// same file are written before the subordinate classes.
|
Set<ObjectClass> addedClasses = new HashSet<ObjectClass>();
|
values = new LinkedHashSet<AttributeValue>();
|
for (ObjectClass oc : schema.getObjectClasses().values())
|
{
|
if (schemaFile.equals(oc.getSchemaFile()))
|
{
|
addObjectClassToSchemaFile(schema, schemaFile, oc, values, addedClasses,
|
0);
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(objectClassesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(objectClassesType, attrList);
|
}
|
|
|
// Add all of the appropriate name forms to the schema entry. Since there
|
// is no hierarchical relationship between name forms, we don't need to
|
// worry about ordering.
|
values = new LinkedHashSet<AttributeValue>();
|
for (List<NameForm> forms : schema.getNameFormsByObjectClass().values())
|
{
|
for(NameForm nf : forms)
|
{
|
if (schemaFile.equals(nf.getSchemaFile()))
|
{
|
values.add(AttributeValues.create(
|
nameFormsType, nf.getDefinition()));
|
}
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(nameFormsType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(nameFormsType, attrList);
|
}
|
|
|
// Add all of the appropriate DIT content rules to the schema entry. Since
|
// there is no hierarchical relationship between DIT content rules, we don't
|
// need to worry about ordering.
|
values = new LinkedHashSet<AttributeValue>();
|
for (DITContentRule dcr : schema.getDITContentRules().values())
|
{
|
if (schemaFile.equals(dcr.getSchemaFile()))
|
{
|
values.add(AttributeValues.create(ditContentRulesType,
|
dcr.getDefinition()));
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(ditContentRulesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(ditContentRulesType, attrList);
|
}
|
|
|
// Add all of the appropriate DIT structure rules to the schema entry. We
|
// need to be careful of the ordering to ensure that any superior rules in
|
// the same file are written before the subordinate rules.
|
Set<DITStructureRule> addedDSRs = new HashSet<DITStructureRule>();
|
values = new LinkedHashSet<AttributeValue>();
|
for (DITStructureRule dsr : schema.getDITStructureRulesByID().values())
|
{
|
if (schemaFile.equals(dsr.getSchemaFile()))
|
{
|
addDITStructureRuleToSchemaFile(schema, schemaFile, dsr, values,
|
addedDSRs, 0);
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(ditStructureRulesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(ditStructureRulesType, attrList);
|
}
|
|
|
// Add all of the appropriate matching rule uses to the schema entry. Since
|
// there is no hierarchical relationship between matching rule uses, we
|
// don't need to worry about ordering.
|
values = new LinkedHashSet<AttributeValue>();
|
for (MatchingRuleUse mru : schema.getMatchingRuleUses().values())
|
{
|
if (schemaFile.equals(mru.getSchemaFile()))
|
{
|
values.add(AttributeValues.create(matchingRuleUsesType,
|
mru.getDefinition()));
|
}
|
}
|
|
if (! values.isEmpty())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
AttributeBuilder builder = new AttributeBuilder(matchingRuleUsesType);
|
builder.addAll(values);
|
attrList.add(builder.toAttribute());
|
schemaEntry.putAttribute(matchingRuleUsesType, attrList);
|
}
|
|
|
if (schemaFile.equals(FILE_USER_SCHEMA_ELEMENTS))
|
{
|
Map<String, Attribute> attributes = schema.getExtraAttributes();
|
for (Attribute attribute : attributes.values())
|
{
|
ArrayList<Attribute> attrList = new ArrayList<Attribute>(1);
|
attrList.add(attribute);
|
schemaEntry.putAttribute(attribute.getAttributeType(), attrList);
|
}
|
}
|
|
// Create a temporary file to which we can write the schema entry.
|
File tempFile = File.createTempFile(schemaFile, "temp");
|
LDIFExportConfig exportConfig =
|
new LDIFExportConfig(tempFile.getAbsolutePath(),
|
ExistingFileBehavior.OVERWRITE);
|
LDIFWriter ldifWriter = new LDIFWriter(exportConfig);
|
ldifWriter.writeEntry(schemaEntry);
|
ldifWriter.close();
|
|
return tempFile;
|
}
|
|
|
|
/**
|
* Adds the definition for the specified attribute type to the provided set of
|
* attribute values, recursively adding superior types as appropriate.
|
*
|
* @param schema The schema containing the attribute type.
|
* @param schemaFile The schema file with which the attribute type is
|
* associated.
|
* @param attributeType The attribute type whose definition should be added
|
* to the value set.
|
* @param values The set of values for attribute type definitions
|
* already added.
|
* @param addedTypes The set of attribute types whose definitions have
|
* already been added to the set of values.
|
* @param depth A depth counter to use in an attempt to detect
|
* circular references.
|
*/
|
private void addAttrTypeToSchemaFile(Schema schema, String schemaFile,
|
AttributeType attributeType,
|
Set<AttributeValue> values,
|
Set<AttributeType> addedTypes,
|
int depth)
|
throws DirectoryException
|
{
|
if (depth > 20)
|
{
|
Message message = ERR_SCHEMA_MODIFY_CIRCULAR_REFERENCE_AT.get(
|
attributeType.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (addedTypes.contains(attributeType))
|
{
|
return;
|
}
|
|
AttributeType superiorType = attributeType.getSuperiorType();
|
if ((superiorType != null) &&
|
schemaFile.equals(superiorType.getSchemaFile()) &&
|
(! addedTypes.contains(superiorType)))
|
{
|
addAttrTypeToSchemaFile(schema, schemaFile, superiorType, values,
|
addedTypes, depth+1);
|
}
|
|
values.add(AttributeValues.create(attributeTypesType,
|
attributeType.getDefinition()));
|
addedTypes.add(attributeType);
|
}
|
|
|
|
/**
|
* Adds the definition for the specified objectclass to the provided set of
|
* attribute values, recursively adding superior classes as appropriate.
|
*
|
* @param schema The schema containing the objectclass.
|
* @param schemaFile The schema file with which the objectclass is
|
* associated.
|
* @param objectClass The objectclass whose definition should be added to
|
* the value set.
|
* @param values The set of values for objectclass definitions
|
* already added.
|
* @param addedClasses The set of objectclasses whose definitions have
|
* already been added to the set of values.
|
* @param depth A depth counter to use in an attempt to detect
|
* circular references.
|
*/
|
private void addObjectClassToSchemaFile(Schema schema, String schemaFile,
|
ObjectClass objectClass,
|
Set<AttributeValue> values,
|
Set<ObjectClass> addedClasses,
|
int depth)
|
throws DirectoryException
|
{
|
if (depth > 20)
|
{
|
Message message = ERR_SCHEMA_MODIFY_CIRCULAR_REFERENCE_OC.get(
|
objectClass.getNameOrOID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (addedClasses.contains(objectClass))
|
{
|
return;
|
}
|
|
for(ObjectClass superiorClass : objectClass.getSuperiorClasses())
|
{
|
if (schemaFile.equals(superiorClass.getSchemaFile()) &&
|
(! addedClasses.contains(superiorClass)))
|
{
|
addObjectClassToSchemaFile(schema, schemaFile, superiorClass, values,
|
addedClasses, depth+1);
|
}
|
}
|
values.add(AttributeValues.create(objectClassesType,
|
objectClass.getDefinition()));
|
addedClasses.add(objectClass);
|
}
|
|
|
|
/**
|
* Adds the definition for the specified DIT structure rule to the provided
|
* set of attribute values, recursively adding superior rules as appropriate.
|
*
|
* @param schema The schema containing the DIT structure rule.
|
* @param schemaFile The schema file with which the DIT structure rule
|
* is associated.
|
* @param ditStructureRule The DIT structure rule whose definition should be
|
* added to the value set.
|
* @param values The set of values for DIT structure rule
|
* definitions already added.
|
* @param addedDSRs The set of DIT structure rules whose definitions
|
* have already been added added to the set of
|
* values.
|
* @param depth A depth counter to use in an attempt to detect
|
* circular references.
|
*/
|
private void addDITStructureRuleToSchemaFile(Schema schema, String schemaFile,
|
DITStructureRule ditStructureRule,
|
Set<AttributeValue> values,
|
Set<DITStructureRule> addedDSRs, int depth)
|
throws DirectoryException
|
{
|
if (depth > 20)
|
{
|
Message message = ERR_SCHEMA_MODIFY_CIRCULAR_REFERENCE_DSR.get(
|
ditStructureRule.getNameOrRuleID());
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
if (addedDSRs.contains(ditStructureRule))
|
{
|
return;
|
}
|
|
for (DITStructureRule dsr : ditStructureRule.getSuperiorRules())
|
{
|
if (schemaFile.equals(dsr.getSchemaFile()) && (! addedDSRs.contains(dsr)))
|
{
|
addDITStructureRuleToSchemaFile(schema, schemaFile, dsr, values,
|
addedDSRs, depth+1);
|
}
|
}
|
|
values.add(AttributeValues.create(ditStructureRulesType,
|
ditStructureRule.getDefinition()));
|
addedDSRs.add(ditStructureRule);
|
}
|
|
|
|
/**
|
* Moves the specified temporary schema files in place of the active versions.
|
* If an error occurs in the process, then this method will attempt to restore
|
* the original schema files if possible.
|
*
|
* @param tempSchemaFiles The set of temporary schema files to be activated.
|
*
|
* @throws DirectoryException If a problem occurs while attempting to
|
* install the temporary schema files.
|
*/
|
private void installSchemaFiles(HashMap<String,File> tempSchemaFiles)
|
throws DirectoryException
|
{
|
// Create lists that will hold the three types of files we'll be dealing
|
// with (the temporary files that will be installed, the installed schema
|
// files, and the previously-installed schema files).
|
ArrayList<File> installedFileList = new ArrayList<File>();
|
ArrayList<File> tempFileList = new ArrayList<File>();
|
ArrayList<File> origFileList = new ArrayList<File>();
|
|
File schemaInstanceDir =
|
new File(SchemaConfigManager.getSchemaDirectoryPath());
|
|
for (String name : tempSchemaFiles.keySet())
|
{
|
installedFileList.add(new File(schemaInstanceDir, name));
|
tempFileList.add(tempSchemaFiles.get(name));
|
origFileList.add(new File(schemaInstanceDir, name + ".orig"));
|
}
|
|
|
// If there are any old ".orig" files laying around from a previous
|
// attempt, then try to clean them up.
|
for (File f : origFileList)
|
{
|
if (f.exists())
|
{
|
f.delete();
|
}
|
}
|
|
|
// Copy all of the currently-installed files with a ".orig" extension. If
|
// this fails, then try to clean up the copies.
|
try
|
{
|
for (int i=0; i < installedFileList.size(); i++)
|
{
|
File installedFile = installedFileList.get(i);
|
File origFile = origFileList.get(i);
|
|
if (installedFile.exists())
|
{
|
copyFile(installedFile, origFile);
|
}
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
boolean allCleaned = true;
|
for (File f : origFileList)
|
{
|
try
|
{
|
if (f.exists())
|
{
|
if (! f.delete())
|
{
|
allCleaned = false;
|
}
|
}
|
}
|
catch (Exception e2)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e2);
|
}
|
|
allCleaned = false;
|
}
|
}
|
|
if (allCleaned)
|
{
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_WRITE_ORIG_FILES_CLEANED.get(
|
getExceptionMessage(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
else
|
{
|
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_WRITE_ORIG_FILES_NOT_CLEANED
|
.get(getExceptionMessage(e));
|
|
DirectoryServer.sendAlertNotification(this,
|
ALERT_TYPE_CANNOT_COPY_SCHEMA_FILES,
|
message);
|
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// Try to copy all of the temporary files into place over the installed
|
// files. If this fails, then try to restore the originals.
|
try
|
{
|
for (int i=0; i < installedFileList.size(); i++)
|
{
|
File installedFile = installedFileList.get(i);
|
File tempFile = tempFileList.get(i);
|
copyFile(tempFile, installedFile);
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
for (File f : installedFileList)
|
{
|
try
|
{
|
if (f.exists())
|
{
|
f.delete();
|
}
|
}
|
catch (Exception e2)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e2);
|
}
|
}
|
}
|
|
boolean allRestored = true;
|
for (int i=0; i < installedFileList.size(); i++)
|
{
|
File installedFile = installedFileList.get(i);
|
File origFile = origFileList.get(i);
|
|
try
|
{
|
if (origFile.exists())
|
{
|
if (! origFile.renameTo(installedFile))
|
{
|
allRestored = false;
|
}
|
}
|
}
|
catch (Exception e2)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e2);
|
}
|
|
allRestored = false;
|
}
|
}
|
|
if (allRestored)
|
{
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_WRITE_NEW_FILES_RESTORED.get(
|
getExceptionMessage(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
else
|
{
|
Message message = ERR_SCHEMA_MODIFY_CANNOT_WRITE_NEW_FILES_NOT_RESTORED
|
.get(getExceptionMessage(e));
|
|
DirectoryServer.sendAlertNotification(this,
|
ALERT_TYPE_CANNOT_WRITE_NEW_SCHEMA_FILES,
|
message);
|
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// At this point, we're committed to the schema change, so we can't throw
|
// any more exceptions, but all we have left is to clean up the original and
|
// temporary files.
|
for (File f : origFileList)
|
{
|
try
|
{
|
if (f.exists())
|
{
|
f.delete();
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
|
for (File f : tempFileList)
|
{
|
try
|
{
|
if (f.exists())
|
{
|
f.delete();
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* Creates a copy of the specified file.
|
*
|
* @param from The source file to be copied.
|
* @param to The destination file to be created.
|
*
|
* @throws IOException If a problem occurs.
|
*/
|
private void copyFile(File from, File to)
|
throws IOException
|
{
|
byte[] buffer = new byte[4096];
|
FileInputStream inputStream = null;
|
FileOutputStream outputStream = null;
|
try
|
{
|
inputStream = new FileInputStream(from);
|
outputStream = new FileOutputStream(to, false);
|
|
int bytesRead = inputStream.read(buffer);
|
while (bytesRead > 0)
|
{
|
outputStream.write(buffer, 0, bytesRead);
|
bytesRead = inputStream.read(buffer);
|
}
|
}
|
finally
|
{
|
if (inputStream != null)
|
{
|
try
|
{
|
inputStream.close();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
|
if (outputStream != null)
|
{
|
outputStream.close();
|
}
|
}
|
}
|
|
|
|
/**
|
* Performs any necessary cleanup in an attempt to delete any temporary schema
|
* files that may have been left over after trying to install the new schema.
|
*
|
* @param tempSchemaFiles The set of temporary schema files that have been
|
* created and are candidates for cleanup.
|
*/
|
private void cleanUpTempSchemaFiles(HashMap<String,File> tempSchemaFiles)
|
{
|
if ((tempSchemaFiles == null) || tempSchemaFiles.isEmpty())
|
{
|
return;
|
}
|
|
for (File f : tempSchemaFiles.values())
|
{
|
try
|
{
|
if (f.exists())
|
{
|
f.delete();
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void renameEntry(DN currentDN, Entry entry,
|
ModifyDNOperation modifyDNOperation)
|
throws DirectoryException
|
{
|
Message message =
|
ERR_SCHEMA_MODIFY_DN_NOT_SUPPORTED.get(String.valueOf(currentDN));
|
throw new DirectoryException(ResultCode.UNWILLING_TO_PERFORM, message);
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void search(SearchOperation searchOperation)
|
throws DirectoryException
|
{
|
DN baseDN = searchOperation.getBaseDN();
|
|
boolean found = false;
|
DN[] dnArray = baseDNs;
|
DN matchedDN = null;
|
for (DN dn : dnArray)
|
{
|
if (dn.equals(baseDN))
|
{
|
found = true;
|
break;
|
}
|
else if (dn.isAncestorOf(baseDN))
|
{
|
matchedDN = dn;
|
break;
|
}
|
}
|
|
if (! found)
|
{
|
Message message = ERR_SCHEMA_INVALID_BASE.get(String.valueOf(baseDN));
|
throw new DirectoryException(ResultCode.NO_SUCH_OBJECT, message,
|
matchedDN, null);
|
}
|
|
|
// If it's a onelevel or subordinate subtree search, then we will never
|
// match anything since there isn't anything below the schema.
|
SearchScope scope = searchOperation.getScope();
|
if ((scope == SearchScope.SINGLE_LEVEL) ||
|
(scope == SearchScope.SUBORDINATE_SUBTREE))
|
{
|
return;
|
}
|
|
|
// Get the schema entry and see if it matches the filter. If so, then send
|
// it to the client.
|
Entry schemaEntry = getSchemaEntry(baseDN, false);
|
SearchFilter filter = searchOperation.getFilter();
|
if (filter.matchesEntry(schemaEntry))
|
{
|
searchOperation.returnEntry(schemaEntry, null);
|
}
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public Set<String> getSupportedControls()
|
{
|
return supportedControls;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public Set<String> getSupportedFeatures()
|
{
|
return supportedFeatures;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean supportsLDIFExport()
|
{
|
// We will only export the DSE entry itself.
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void exportLDIF(LDIFExportConfig exportConfig)
|
throws DirectoryException
|
{
|
// Create the LDIF writer.
|
LDIFWriter ldifWriter;
|
try
|
{
|
ldifWriter = new LDIFWriter(exportConfig);
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_UNABLE_TO_CREATE_LDIF_WRITER.get(
|
stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
|
// Write the root schema entry to it. Make sure to close the LDIF
|
// writer when we're done.
|
try
|
{
|
ldifWriter.writeEntry(getSchemaEntry(baseDNs[0], true, true));
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message =
|
ERR_SCHEMA_UNABLE_TO_EXPORT_BASE.get(stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
finally
|
{
|
try
|
{
|
ldifWriter.close();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
}
|
}
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean supportsLDIFImport()
|
{
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public LDIFImportResult importLDIF(LDIFImportConfig importConfig)
|
throws DirectoryException
|
{
|
LDIFReader reader;
|
try
|
{
|
reader = new LDIFReader(importConfig);
|
}
|
catch (Exception e)
|
{
|
Message message =
|
ERR_MEMORYBACKEND_CANNOT_CREATE_LDIF_READER.get(String.valueOf(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
try
|
{
|
while (true)
|
{
|
Entry e = null;
|
try
|
{
|
e = reader.readEntry();
|
if (e == null)
|
{
|
break;
|
}
|
}
|
catch (LDIFException le)
|
{
|
if (! le.canContinueReading())
|
{
|
Message message =
|
ERR_MEMORYBACKEND_ERROR_READING_LDIF.get(String.valueOf(e));
|
throw new DirectoryException(
|
DirectoryServer.getServerErrorResultCode(),
|
message, le);
|
}
|
else
|
{
|
continue;
|
}
|
}
|
|
importEntry(e);
|
}
|
|
return new LDIFImportResult(reader.getEntriesRead(),
|
reader.getEntriesRejected(),
|
reader.getEntriesIgnored());
|
}
|
catch (DirectoryException de)
|
{
|
throw de;
|
}
|
catch (Exception e)
|
{
|
Message message =
|
ERR_MEMORYBACKEND_ERROR_DURING_IMPORT.get(String.valueOf(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
finally
|
{
|
reader.close();
|
}
|
}
|
|
|
/**
|
* Import an entry in a new schema by :
|
* - duplicating the schema
|
* - iterating over each element of the newSchemaEntry and comparing
|
* with the existing schema
|
* - if the new schema element do not exist : add it
|
*
|
* FIXME : attributeTypes and objectClasses are the only elements
|
* currently taken into account.
|
*
|
* @param newSchemaEntry The entry to be imported.
|
*/
|
private void importEntry(Entry newSchemaEntry)
|
throws DirectoryException
|
{
|
Schema schema = DirectoryServer.getSchema();
|
Schema newSchema = DirectoryServer.getSchema().duplicate();
|
TreeSet<String> modifiedSchemaFiles = new TreeSet<String>();
|
|
// Get the attributeTypes attribute from the entry.
|
AttributeTypeSyntax attrTypeSyntax;
|
try
|
{
|
attrTypeSyntax = (AttributeTypeSyntax)
|
schema.getSyntax(SYNTAX_ATTRIBUTE_TYPE_OID);
|
if (attrTypeSyntax == null)
|
{
|
attrTypeSyntax = new AttributeTypeSyntax();
|
attrTypeSyntax.initializeSyntax(null);
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
attrTypeSyntax = new AttributeTypeSyntax();
|
}
|
|
AttributeType attributeAttrType =
|
schema.getAttributeType(ATTR_ATTRIBUTE_TYPES_LC);
|
if (attributeAttrType == null)
|
{
|
attributeAttrType =
|
DirectoryServer.getDefaultAttributeType(ATTR_ATTRIBUTE_TYPES,
|
attrTypeSyntax);
|
}
|
|
// loop on the attribute types in the entry just received
|
// and add them in the existing schema.
|
List<Attribute> attrList = newSchemaEntry.getAttribute(attributeAttrType);
|
Set<String> oidList = new HashSet<String>(1000);
|
if ((attrList != null) && (! attrList.isEmpty()))
|
{
|
for (Attribute a : attrList)
|
{
|
// Look for attributetypes that could have been added to the schema
|
// or modified in the schema
|
for (AttributeValue v : a)
|
{
|
// Parse the attribute type.
|
AttributeType attrType = AttributeTypeSyntax.decodeAttributeType(
|
v.getValue(), schema, false);
|
String schemaFile = attrType.getSchemaFile();
|
if ((schemaFile != null) &&
|
(schemaFile.equals(CONFIG_SCHEMA_ELEMENTS_FILE)))
|
{
|
// Don't import the file containing the definitions of the
|
// Schema elements used for configuration because these
|
// definitions may vary between versions of OpenDJ.
|
continue;
|
}
|
|
oidList.add(attrType.getOID());
|
try
|
{
|
// Register this attribute type in the new schema
|
// unless it is already defined with the same syntax.
|
AttributeType oldAttrType =
|
schema.getAttributeType(attrType.getOID());
|
if ((oldAttrType == null) ||
|
(!oldAttrType.toString().equals(attrType.toString())))
|
{
|
newSchema.registerAttributeType(attrType, true);
|
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
}
|
catch (DirectoryException de)
|
{
|
Message message =
|
NOTE_SCHEMA_IMPORT_FAILED.get(
|
attrType.toString(), de.getMessage());
|
logError(message);
|
}
|
catch (Exception e)
|
{
|
Message message =
|
NOTE_SCHEMA_IMPORT_FAILED.get(
|
attrType.toString(), e.getMessage());
|
logError(message);
|
}
|
}
|
}
|
}
|
|
// loop on all the attribute types in the current schema and delete
|
// them from the new schema if they are not in the imported schema entry.
|
ConcurrentHashMap<String, AttributeType> currentAttrTypes =
|
newSchema.getAttributeTypes();
|
|
for (AttributeType removeType : currentAttrTypes.values())
|
{
|
String schemaFile = removeType.getSchemaFile();
|
if ((schemaFile != null) &&
|
((schemaFile.equals(CONFIG_SCHEMA_ELEMENTS_FILE)) ||
|
(schemaFile.equals(CORE_SCHEMA_ELEMENTS_FILE))) )
|
{
|
// Don't import the file containing the definitions of the
|
// Schema elements used for configuration because these
|
// definitions may vary between versions of OpenDJ.
|
// Also never delete anything from the core schema file.
|
continue;
|
}
|
if (!oidList.contains(removeType.getOID()))
|
{
|
newSchema.deregisterAttributeType(removeType);
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
}
|
|
// loop on the objectClasses from the entry, search if they are
|
// already in the current schema, add them if not.
|
ObjectClassSyntax ocSyntax;
|
try
|
{
|
ocSyntax = (ObjectClassSyntax) schema.getSyntax(SYNTAX_OBJECTCLASS_OID);
|
if (ocSyntax == null)
|
{
|
ocSyntax = new ObjectClassSyntax();
|
ocSyntax.initializeSyntax(null);
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
ocSyntax = new ObjectClassSyntax();
|
}
|
|
AttributeType objectclassAttrType =
|
schema.getAttributeType(ATTR_OBJECTCLASSES_LC);
|
if (objectclassAttrType == null)
|
{
|
objectclassAttrType =
|
DirectoryServer.getDefaultAttributeType(ATTR_OBJECTCLASSES,
|
ocSyntax);
|
}
|
|
oidList.clear();
|
List<Attribute> ocList = newSchemaEntry.getAttribute(objectclassAttrType);
|
if ((ocList != null) && (! ocList.isEmpty()))
|
{
|
for (Attribute a : ocList)
|
{
|
for (AttributeValue v : a)
|
{
|
// It IS important here to allow the unknown elements that could
|
// appear in the new config schema.
|
ObjectClass newObjectClass = ObjectClassSyntax.decodeObjectClass(
|
v.getValue(), newSchema, true);
|
String schemaFile = newObjectClass.getSchemaFile();
|
if ((schemaFile != null) &&
|
(schemaFile.equals(CONFIG_SCHEMA_ELEMENTS_FILE)))
|
{
|
// Don't import the file containing the definitions of the
|
// Schema elements used for configuration because these
|
// definitions may vary between versions of OpenDJ.
|
continue;
|
}
|
|
// Now we know we are not in the config schema, let's check
|
// the unknown elements ... sadly but simply by redoing the
|
// whole decoding.
|
newObjectClass = ObjectClassSyntax.decodeObjectClass(
|
v.getValue(), newSchema, false);
|
oidList.add(newObjectClass.getOID());
|
try
|
{
|
// Register this ObjectClass in the new schema
|
// unless it is already defined with the same syntax.
|
ObjectClass oldObjectClass =
|
schema.getObjectClass(newObjectClass.getOID());
|
if ((oldObjectClass == null) ||
|
(!oldObjectClass.toString().equals(newObjectClass.toString())))
|
{
|
newSchema.registerObjectClass(newObjectClass, true);
|
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
}
|
catch (DirectoryException de)
|
{
|
Message message =
|
NOTE_SCHEMA_IMPORT_FAILED.get(
|
newObjectClass.toString(), de.getMessage());
|
logError(message);
|
}
|
catch (Exception e)
|
{
|
Message message =
|
NOTE_SCHEMA_IMPORT_FAILED.get(
|
newObjectClass.toString(), e.getMessage());
|
logError(message);
|
}
|
}
|
}
|
}
|
|
// loop on all the attribute types in the current schema and delete
|
// them from the new schema if they are not in the imported schema entry.
|
ConcurrentHashMap<String, ObjectClass> currentObjectClasses =
|
newSchema.getObjectClasses();
|
|
for (ObjectClass removeClass : currentObjectClasses.values())
|
{
|
String schemaFile = removeClass.getSchemaFile();
|
if ((schemaFile != null) &&
|
(schemaFile.equals(CONFIG_SCHEMA_ELEMENTS_FILE)))
|
{
|
// Don't import the file containing the definition of the
|
// Schema elements used for configuration because these
|
// definitions may vary between versions of OpenDJ.
|
continue;
|
}
|
if (!oidList.contains(removeClass.getOID()))
|
{
|
newSchema.deregisterObjectClass(removeClass);
|
|
if (schemaFile != null)
|
{
|
modifiedSchemaFiles.add(schemaFile);
|
}
|
}
|
}
|
|
// Finally, if there were some modifications, save the new schema
|
// in the Schema Files and update DirectoryServer.
|
if (!modifiedSchemaFiles.isEmpty())
|
{
|
updateSchemaFiles(newSchema, modifiedSchemaFiles);
|
DirectoryServer.setSchema(newSchema);
|
}
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean supportsBackup()
|
{
|
// We do support an online backup mechanism for the schema.
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean supportsBackup(BackupConfig backupConfig,
|
StringBuilder unsupportedReason)
|
{
|
// We should support online backup for the schema in any form. This
|
// implementation does not support incremental backups, but in this case
|
// even if we're asked to do an incremental we'll just do a full backup
|
// instead. So the answer to this should always be "true".
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void createBackup(BackupConfig backupConfig)
|
throws DirectoryException
|
{
|
// Get the properties to use for the backup. We don't care whether or not
|
// it's incremental, so there's no need to get that.
|
String backupID = backupConfig.getBackupID();
|
BackupDirectory backupDirectory = backupConfig.getBackupDirectory();
|
boolean compress = backupConfig.compressData();
|
boolean encrypt = backupConfig.encryptData();
|
boolean hash = backupConfig.hashData();
|
boolean signHash = backupConfig.signHash();
|
|
|
// Create a hash map that will hold the extra backup property information
|
// for this backup.
|
HashMap<String,String> backupProperties = new HashMap<String,String>();
|
|
|
// Get the crypto manager and use it to obtain references to the message
|
// digest and/or MAC to use for hashing and/or signing.
|
CryptoManager cryptoManager = DirectoryServer.getCryptoManager();
|
Mac mac = null;
|
MessageDigest digest = null;
|
String macKeyID = null;
|
|
if (hash)
|
{
|
if (signHash)
|
{
|
try
|
{
|
macKeyID = cryptoManager.getMacEngineKeyEntryID();
|
backupProperties.put(BACKUP_PROPERTY_MAC_KEY_ID, macKeyID);
|
|
mac = cryptoManager.getMacEngine(macKeyID);
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_BACKUP_CANNOT_GET_MAC.get(
|
macKeyID, stackTraceToSingleLineString(e));
|
throw new DirectoryException(
|
DirectoryServer.getServerErrorResultCode(), message,
|
e);
|
}
|
}
|
else
|
{
|
String digestAlgorithm =
|
cryptoManager.getPreferredMessageDigestAlgorithm();
|
backupProperties.put(BACKUP_PROPERTY_DIGEST_ALGORITHM, digestAlgorithm);
|
|
try
|
{
|
digest = cryptoManager.getPreferredMessageDigest();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_BACKUP_CANNOT_GET_DIGEST.get(
|
digestAlgorithm, stackTraceToSingleLineString(e));
|
throw new DirectoryException(
|
DirectoryServer.getServerErrorResultCode(), message,
|
e);
|
}
|
}
|
}
|
|
|
// Create an output stream that will be used to write the archive file. At
|
// its core, it will be a file output stream to put a file on the disk. If
|
// we are to encrypt the data, then that file output stream will be wrapped
|
// in a cipher output stream. The resulting output stream will then be
|
// wrapped by a zip output stream (which may or may not actually use
|
// compression).
|
String filename = null;
|
OutputStream outputStream;
|
try
|
{
|
filename = SCHEMA_BACKUP_BASE_FILENAME + backupID;
|
File archiveFile = new File(backupDirectory.getPath() + File.separator +
|
filename);
|
if (archiveFile.exists())
|
{
|
int i=1;
|
while (true)
|
{
|
archiveFile = new File(backupDirectory.getPath() + File.separator +
|
filename + "." + i);
|
if (archiveFile.exists())
|
{
|
i++;
|
}
|
else
|
{
|
filename = filename + "." + i;
|
break;
|
}
|
}
|
}
|
|
outputStream = new FileOutputStream(archiveFile, false);
|
backupProperties.put(BACKUP_PROPERTY_ARCHIVE_FILENAME, filename);
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_BACKUP_CANNOT_CREATE_ARCHIVE_FILE.
|
get(String.valueOf(filename), backupDirectory.getPath(),
|
getExceptionMessage(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
// If we should encrypt the data, then wrap the output stream in a cipher
|
// output stream.
|
if (encrypt)
|
{
|
try
|
{
|
outputStream
|
= cryptoManager.getCipherOutputStream(outputStream);
|
}
|
catch (CryptoManagerException e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_SCHEMA_BACKUP_CANNOT_GET_CIPHER.get(
|
stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// Wrap the file output stream in a zip output stream.
|
ZipOutputStream zipStream = new ZipOutputStream(outputStream);
|
|
Message message = ERR_SCHEMA_BACKUP_ZIP_COMMENT.get(
|
DynamicConstants.PRODUCT_NAME,
|
backupID);
|
zipStream.setComment(String.valueOf(message));
|
|
if (compress)
|
{
|
zipStream.setLevel(Deflater.DEFAULT_COMPRESSION);
|
}
|
else
|
{
|
zipStream.setLevel(Deflater.NO_COMPRESSION);
|
}
|
|
// Create a Comment Entry in the zip file
|
// This ensure the backup is never empty, even wher
|
// there is no schema file to backup.
|
|
String commentName = "schema.comment";
|
|
// We'll put the name in the hash, too.
|
if (hash)
|
{
|
if (signHash)
|
{
|
mac.update(getBytes(commentName));
|
} else
|
{
|
digest.update(getBytes(commentName));
|
}
|
}
|
try
|
{
|
ZipEntry zipEntry = new ZipEntry(commentName);
|
zipStream.putNextEntry(zipEntry);
|
zipStream.closeEntry();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
try
|
{
|
zipStream.close();
|
} catch (Exception e2)
|
{
|
}
|
|
message = ERR_SCHEMA_BACKUP_CANNOT_BACKUP_SCHEMA_FILE.get(commentName,
|
stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer
|
.getServerErrorResultCode(), message, e);
|
}
|
|
// Get the path to the directory in which the schema files reside and
|
// then get a list of all the files in that directory.
|
String schemaInstanceDirPath =
|
SchemaConfigManager.getSchemaDirectoryPath();
|
File schemaDir;
|
File[] schemaFiles = null;
|
|
try
|
{
|
schemaDir = new File(schemaInstanceDirPath);
|
schemaFiles = schemaDir.listFiles();
|
}
|
catch (Exception e)
|
{
|
// Can't locate or list Instance schema directory
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
message = ERR_SCHEMA_BACKUP_CANNOT_LIST_SCHEMA_FILES.get(
|
schemaInstanceDirPath, stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
// Iterate through the schema files and write them to the zip stream. If
|
// we're using a hash or MAC, then calculate that as well.
|
byte[] buffer = new byte[8192];
|
String parent = ".instance";
|
|
for (File schemaFile : schemaFiles)
|
{
|
if (backupConfig.isCancelled())
|
{
|
break;
|
}
|
|
if (!schemaFile.isFile())
|
{
|
// If there are any non-file items in the directory (e.g., one or more
|
// subdirectories), then we'll skip them.
|
continue;
|
}
|
|
String baseName = schemaFile.getName();
|
|
// We'll put the name in the hash, too.
|
if (hash)
|
{
|
if (signHash)
|
{
|
mac.update(getBytes(baseName + parent));
|
} else
|
{
|
digest.update(getBytes(baseName + parent));
|
}
|
}
|
|
InputStream inputStream = null;
|
try
|
{
|
ZipEntry zipEntry = new ZipEntry(baseName + parent);
|
zipStream.putNextEntry(zipEntry);
|
|
inputStream = new FileInputStream(schemaFile);
|
while (true)
|
{
|
int bytesRead = inputStream.read(buffer);
|
if (bytesRead < 0 || backupConfig.isCancelled())
|
{
|
break;
|
}
|
|
if (hash)
|
{
|
if (signHash)
|
{
|
mac.update(buffer, 0, bytesRead);
|
} else
|
{
|
digest.update(buffer, 0, bytesRead);
|
}
|
}
|
|
zipStream.write(buffer, 0, bytesRead);
|
}
|
|
zipStream.closeEntry();
|
inputStream.close();
|
} catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
try
|
{
|
if (inputStream != null)
|
{
|
inputStream.close();
|
}
|
} catch (Exception e2)
|
{
|
}
|
|
try
|
{
|
zipStream.close();
|
} catch (Exception e2)
|
{
|
}
|
|
message = ERR_SCHEMA_BACKUP_CANNOT_BACKUP_SCHEMA_FILE.get(baseName,
|
stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer
|
.getServerErrorResultCode(), message, e);
|
}
|
}
|
|
|
// We're done writing the file, so close the zip stream (which should also
|
// close the underlying stream).
|
try
|
{
|
zipStream.close();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
message = ERR_SCHEMA_BACKUP_CANNOT_CLOSE_ZIP_STREAM.get(
|
filename, backupDirectory.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
// Get the digest or MAC bytes if appropriate.
|
byte[] digestBytes = null;
|
byte[] macBytes = null;
|
if (hash)
|
{
|
if (signHash)
|
{
|
macBytes = mac.doFinal();
|
}
|
else
|
{
|
digestBytes = digest.digest();
|
}
|
}
|
|
|
// Create the backup info structure for this backup and add it to the backup
|
// directory.
|
// FIXME -- Should I use the date from when I started or finished?
|
BackupInfo backupInfo = new BackupInfo(backupDirectory, backupID,
|
new Date(), false, compress,
|
encrypt, digestBytes, macBytes,
|
null, backupProperties);
|
|
try
|
{
|
backupDirectory.addBackup(backupInfo);
|
backupDirectory.writeBackupDirectoryDescriptor();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
message = ERR_SCHEMA_BACKUP_CANNOT_UPDATE_BACKUP_DESCRIPTOR.get(
|
backupDirectory.getDescriptorPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void removeBackup(BackupDirectory backupDirectory,
|
String backupID)
|
throws DirectoryException
|
{
|
BackupInfo backupInfo = backupDirectory.getBackupInfo(backupID);
|
if (backupInfo == null)
|
{
|
Message message = ERR_BACKUP_MISSING_BACKUPID.get(
|
backupDirectory.getPath(), backupID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
HashMap<String,String> backupProperties = backupInfo.getBackupProperties();
|
|
String archiveFilename =
|
backupProperties.get(BACKUP_PROPERTY_ARCHIVE_FILENAME);
|
File archiveFile = new File(backupDirectory.getPath(), archiveFilename);
|
|
try
|
{
|
backupDirectory.removeBackup(backupID);
|
}
|
catch (ConfigException e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
e.getMessageObject());
|
}
|
|
try
|
{
|
backupDirectory.writeBackupDirectoryDescriptor();
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
Message message = ERR_BACKUP_CANNOT_UPDATE_BACKUP_DESCRIPTOR.get(
|
backupDirectory.getDescriptorPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
// Remove the archive file.
|
archiveFile.delete();
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public boolean supportsRestore()
|
{
|
// We will provide a restore, but only for offline operations.
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override()
|
public void restoreBackup(RestoreConfig restoreConfig)
|
throws DirectoryException
|
{
|
// First, make sure that the requested backup exists.
|
BackupDirectory backupDirectory = restoreConfig.getBackupDirectory();
|
String backupPath = backupDirectory.getPath();
|
String backupID = restoreConfig.getBackupID();
|
BackupInfo backupInfo = backupDirectory.getBackupInfo(backupID);
|
if (backupInfo == null)
|
{
|
Message message =
|
ERR_SCHEMA_RESTORE_NO_SUCH_BACKUP.get(backupID, backupPath);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
|
// Read the backup info structure to determine the name of the file that
|
// contains the archive. Then make sure that file exists.
|
String backupFilename =
|
backupInfo.getBackupProperty(BACKUP_PROPERTY_ARCHIVE_FILENAME);
|
if (backupFilename == null)
|
{
|
Message message =
|
ERR_SCHEMA_RESTORE_NO_BACKUP_FILE.get(backupID, backupPath);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
File backupFile = new File(backupPath + File.separator + backupFilename);
|
try
|
{
|
if (! backupFile.exists())
|
{
|
Message message =
|
ERR_SCHEMA_RESTORE_NO_SUCH_FILE.get(backupID, backupFile.getPath());
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
}
|
catch (DirectoryException de)
|
{
|
throw de;
|
}
|
catch (Exception e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_CHECK_FOR_ARCHIVE.get(
|
backupID, backupFile.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
// If the backup is hashed, then we need to get the message digest to use
|
// to verify it.
|
byte[] unsignedHash = backupInfo.getUnsignedHash();
|
MessageDigest digest = null;
|
if (unsignedHash != null)
|
{
|
String digestAlgorithm =
|
backupInfo.getBackupProperty(BACKUP_PROPERTY_DIGEST_ALGORITHM);
|
if (digestAlgorithm == null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_UNKNOWN_DIGEST.get(backupID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
try
|
{
|
digest = DirectoryServer.getCryptoManager().getMessageDigest(
|
digestAlgorithm);
|
}
|
catch (Exception e)
|
{
|
Message message =
|
ERR_SCHEMA_RESTORE_CANNOT_GET_DIGEST.get(backupID, digestAlgorithm);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// If the backup is signed, then we need to get the MAC to use to verify it.
|
byte[] signedHash = backupInfo.getSignedHash();
|
Mac mac = null;
|
if (signedHash != null)
|
{
|
String macKeyID =
|
backupInfo.getBackupProperty(BACKUP_PROPERTY_MAC_KEY_ID);
|
if (macKeyID == null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_UNKNOWN_MAC.get(backupID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
|
try
|
{
|
mac = DirectoryServer.getCryptoManager().getMacEngine(macKeyID);
|
}
|
catch (Exception e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_GET_MAC.get(
|
backupID, macKeyID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// Create the input stream that will be used to read the backup file. At
|
// its core, it will be a file input stream.
|
InputStream inputStream;
|
try
|
{
|
inputStream = new FileInputStream(backupFile);
|
}
|
catch (Exception e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_OPEN_BACKUP_FILE.get(
|
backupID, backupFile.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
// If the backup is encrypted, then we need to wrap the file input stream
|
// in a cipher input stream.
|
if (backupInfo.isEncrypted())
|
{
|
try
|
{
|
inputStream = DirectoryServer.getCryptoManager()
|
.getCipherInputStream(inputStream);
|
}
|
catch (CryptoManagerException e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_GET_CIPHER.get(
|
backupFile.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
// Now wrap the resulting input stream in a zip stream so that we can read
|
// its contents. We don't need to worry about whether to use compression or
|
// not because it will be handled automatically.
|
ZipInputStream zipStream = new ZipInputStream(inputStream);
|
|
|
// Determine whether we should actually do the restore, or if we should just
|
// try to verify the archive. If we are not going to verify only, then
|
// move the current schema directory out of the way so we can keep it around
|
// to restore if a problem occurs.
|
String schemaInstanceDirPath =
|
SchemaConfigManager.getSchemaDirectoryPath();
|
|
File schemaInstanceDir = new File(schemaInstanceDirPath);
|
|
String backupInstanceDirPath = null;
|
File schemaBackupInstanceDir = null;
|
boolean verifyOnly = restoreConfig.verifyOnly();
|
if (! verifyOnly)
|
{
|
// Rename the current schema directory if it exists.
|
try
|
{
|
if (schemaInstanceDir.exists())
|
{
|
String schemaBackupInstanceDirPath = schemaInstanceDirPath + ".save";
|
backupInstanceDirPath = schemaBackupInstanceDirPath;
|
schemaBackupInstanceDir = new File(backupInstanceDirPath);
|
if (schemaBackupInstanceDir.exists())
|
{
|
int i=2;
|
while (true)
|
{
|
backupInstanceDirPath = schemaBackupInstanceDirPath + i;
|
schemaBackupInstanceDir = new File(backupInstanceDirPath);
|
if (schemaBackupInstanceDir.exists())
|
{
|
i++;
|
}
|
else
|
{
|
break;
|
}
|
}
|
}
|
|
schemaInstanceDir.renameTo(schemaBackupInstanceDir);
|
}
|
}
|
catch (Exception e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_RENAME_CURRENT_DIRECTORY.
|
get(backupID, schemaInstanceDirPath,
|
String.valueOf(backupInstanceDirPath),
|
stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
// Create a new directory to hold the restored schema files.
|
try
|
{
|
schemaInstanceDir.mkdirs();
|
}
|
catch (Exception e)
|
{
|
// Try to restore the previous schema directory if possible. This will
|
// probably fail in this case, but try anyway.
|
if (schemaBackupInstanceDir != null)
|
{
|
try
|
{
|
schemaBackupInstanceDir.renameTo(schemaInstanceDir);
|
Message message =
|
NOTE_SCHEMA_RESTORE_RESTORED_OLD_SCHEMA.get(
|
schemaInstanceDirPath);
|
logError(message);
|
}
|
catch (Exception e2)
|
{
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_RESTORE_OLD_SCHEMA.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
}
|
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_CREATE_SCHEMA_DIRECTORY.get(
|
backupID, schemaInstanceDirPath, stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// Read through the archive file an entry at a time. For each entry, update
|
// the digest or MAC if necessary, and if we're actually doing the restore,
|
// then write the files out into the schema directory.
|
byte[] buffer = new byte[8192];
|
while (true)
|
{
|
ZipEntry zipEntry;
|
try
|
{
|
zipEntry = zipStream.getNextEntry();
|
}
|
catch (Exception e)
|
{
|
// Tell the user where the previous schema was archived.
|
if (schemaBackupInstanceDir != null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_OLD_SCHEMA_SAVED.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_GET_ZIP_ENTRY.get(
|
backupID, backupFile.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
if (zipEntry == null)
|
{
|
break;
|
}
|
|
|
// Get the filename for the zip entry and update the digest or MAC as
|
// necessary.
|
String fileName = zipEntry.getName();
|
if (digest != null)
|
{
|
digest.update(getBytes(fileName));
|
}
|
if (mac != null)
|
{
|
mac.update(getBytes(fileName));
|
}
|
|
String baseDirPath = schemaInstanceDirPath;
|
Boolean restoreIt = true;
|
if (fileName.endsWith(".instance"))
|
{
|
fileName = fileName.substring(0,fileName.lastIndexOf(".instance"));
|
}
|
else
|
{
|
// Skip file.
|
// ".install" files are from old backups and should be ignored
|
restoreIt = false;
|
}
|
|
// If we're doing the restore, then create the output stream to write the
|
// file.
|
OutputStream outputStream = null;
|
if ((! verifyOnly) && restoreIt)
|
{
|
String filePath = baseDirPath + File.separator + fileName;
|
try
|
{
|
outputStream = new FileOutputStream(filePath);
|
}
|
catch (Exception e)
|
{
|
// Tell the user where the previous schema was archived.
|
if (schemaBackupInstanceDir != null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_OLD_SCHEMA_SAVED.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_CREATE_FILE.get(
|
backupID, filePath, stackTraceToSingleLineString(e));
|
throw new DirectoryException(
|
DirectoryServer.getServerErrorResultCode(), message,
|
e);
|
}
|
}
|
|
|
// Read the contents of the file and update the digest or MAC as
|
// necessary. If we're actually restoring it, then write it into the
|
// new schema directory.
|
try
|
{
|
while (true)
|
{
|
int bytesRead = zipStream.read(buffer);
|
if (bytesRead < 0)
|
{
|
// We've reached the end of the entry.
|
break;
|
}
|
|
|
// Update the digest or MAC if appropriate.
|
if (digest != null)
|
{
|
digest.update(buffer, 0, bytesRead);
|
}
|
|
if (mac != null)
|
{
|
mac.update(buffer, 0, bytesRead);
|
}
|
|
|
// Write the data to the output stream if appropriate.
|
if (outputStream != null)
|
{
|
outputStream.write(buffer, 0, bytesRead);
|
}
|
}
|
|
|
// We're at the end of the file so close the output stream if we're
|
// writing it.
|
if (outputStream != null)
|
{
|
outputStream.close();
|
}
|
}
|
catch (Exception e)
|
{
|
// Tell the user where the previous schema was archived.
|
if (schemaBackupInstanceDir != null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_OLD_SCHEMA_SAVED.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
|
Message message = ERR_SCHEMA_RESTORE_CANNOT_PROCESS_ARCHIVE_FILE.get(
|
backupID, fileName, stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
}
|
|
|
// Close the zip stream since we don't need it anymore.
|
try
|
{
|
zipStream.close();
|
}
|
catch (Exception e)
|
{
|
Message message = ERR_SCHEMA_RESTORE_ERROR_ON_ZIP_STREAM_CLOSE.get(
|
backupID, backupFile.getPath(), stackTraceToSingleLineString(e));
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message, e);
|
}
|
|
|
// At this point, we should be done with the contents of the ZIP file and
|
// the restore should be complete. If we were generating a digest or MAC,
|
// then make sure it checks out.
|
if (digest != null)
|
{
|
byte[] calculatedHash = digest.digest();
|
if (Arrays.equals(calculatedHash, unsignedHash))
|
{
|
Message message = NOTE_SCHEMA_RESTORE_UNSIGNED_HASH_VALID.get();
|
logError(message);
|
}
|
else
|
{
|
// Tell the user where the previous schema was archived.
|
if (schemaBackupInstanceDir != null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_OLD_SCHEMA_SAVED.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
|
Message message =
|
ERR_SCHEMA_RESTORE_UNSIGNED_HASH_INVALID.get(backupID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
}
|
|
if (mac != null)
|
{
|
byte[] calculatedSignature = mac.doFinal();
|
if (Arrays.equals(calculatedSignature, signedHash))
|
{
|
Message message = NOTE_SCHEMA_RESTORE_SIGNED_HASH_VALID.get();
|
logError(message);
|
}
|
else
|
{
|
// Tell the user where the previous schema was archived.
|
if (schemaBackupInstanceDir != null)
|
{
|
Message message = ERR_SCHEMA_RESTORE_OLD_SCHEMA_SAVED.get(
|
schemaBackupInstanceDir.getPath());
|
logError(message);
|
}
|
|
Message message = ERR_SCHEMA_RESTORE_SIGNED_HASH_INVALID.get(backupID);
|
throw new DirectoryException(DirectoryServer.getServerErrorResultCode(),
|
message);
|
}
|
}
|
|
|
// If we are just verifying the archive, then we're done.
|
if (verifyOnly)
|
{
|
Message message =
|
NOTE_SCHEMA_RESTORE_VERIFY_SUCCESSFUL.get(backupID, backupPath);
|
logError(message);
|
return;
|
}
|
|
|
// If we've gotten here, then the archive was restored successfully. Get
|
// rid of the temporary copy we made of the previous schema directory and
|
// exit.
|
if (schemaBackupInstanceDir != null)
|
{
|
recursiveDelete(schemaBackupInstanceDir);
|
}
|
|
Message message = NOTE_SCHEMA_RESTORE_SUCCESSFUL.get(backupID, backupPath);
|
logError(message);
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public boolean isConfigurationChangeAcceptable(
|
SchemaBackendCfg configEntry,
|
List<Message> unacceptableReasons)
|
{
|
return true;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public ConfigChangeResult applyConfigurationChange(
|
SchemaBackendCfg backendCfg)
|
{
|
ResultCode resultCode = ResultCode.SUCCESS;
|
boolean adminActionRequired = false;
|
ArrayList<Message> messages = new ArrayList<Message>();
|
|
|
// Check to see if we should apply a new set of base DNs.
|
Set<DN> newBaseDNs;
|
try
|
{
|
newBaseDNs = new HashSet<DN>(backendCfg.getSchemaEntryDN());
|
if (newBaseDNs.isEmpty())
|
{
|
newBaseDNs.add(DN.decode(DN_DEFAULT_SCHEMA_ROOT));
|
}
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
|
messages.add(ERR_SCHEMA_CANNOT_DETERMINE_BASE_DN.get(
|
String.valueOf(configEntryDN),
|
getExceptionMessage(e)));
|
resultCode = DirectoryServer.getServerErrorResultCode();
|
newBaseDNs = null;
|
}
|
|
|
// Check to see if we should change the behavior regarding whether to show
|
// all schema attributes.
|
boolean newShowAllAttributes = backendCfg.isShowAllAttributes();
|
|
|
// Check to see if there is a new set of user-defined attributes.
|
ArrayList<Attribute> newUserAttrs = new ArrayList<Attribute>();
|
try
|
{
|
ConfigEntry configEntry = DirectoryServer.getConfigEntry(configEntryDN);
|
for (List<Attribute> attrs :
|
configEntry.getEntry().getUserAttributes().values())
|
{
|
for (Attribute a : attrs)
|
{
|
if (! isSchemaConfigAttribute(a))
|
{
|
newUserAttrs.add(a);
|
}
|
}
|
}
|
for (List<Attribute> attrs :
|
configEntry.getEntry().getOperationalAttributes().values())
|
{
|
for (Attribute a : attrs)
|
{
|
if (! isSchemaConfigAttribute(a))
|
{
|
newUserAttrs.add(a);
|
}
|
}
|
}
|
}
|
catch (ConfigException e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
messages.add(ERR_CONFIG_BACKEND_ERROR_INTERACTING_WITH_BACKEND_ENTRY.get(
|
String.valueOf(configEntryDN),
|
stackTraceToSingleLineString(e)));
|
resultCode = DirectoryServer.getServerErrorResultCode();
|
}
|
|
|
if (resultCode == ResultCode.SUCCESS)
|
{
|
// Get an array containing the new base DNs to use.
|
DN[] dnArray = new DN[newBaseDNs.size()];
|
newBaseDNs.toArray(dnArray);
|
|
|
// Determine the set of DNs to add and delete. When this is done, the
|
// deleteBaseDNs will contain the set of DNs that should no longer be used
|
// and should be deregistered from the server, and the newBaseDNs set will
|
// just contain the set of DNs to add.
|
Set<DN> deleteBaseDNs = new HashSet<DN>(baseDNs.length);
|
for (DN baseDN : baseDNs)
|
{
|
if (! newBaseDNs.remove(baseDN))
|
{
|
deleteBaseDNs.add(baseDN);
|
}
|
}
|
|
for (DN dn : deleteBaseDNs)
|
{
|
try
|
{
|
DirectoryServer.deregisterBaseDN(dn);
|
messages.add(INFO_SCHEMA_DEREGISTERED_BASE_DN.get(
|
String.valueOf(dn)));
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
messages.add(ERR_SCHEMA_CANNOT_DEREGISTER_BASE_DN.get(
|
String.valueOf(dn),
|
getExceptionMessage(e)));
|
resultCode = DirectoryServer.getServerErrorResultCode();
|
}
|
}
|
|
baseDNs = dnArray;
|
for (DN dn : newBaseDNs)
|
{
|
try
|
{
|
DirectoryServer.registerBaseDN(dn, this, true);
|
messages.add(INFO_SCHEMA_REGISTERED_BASE_DN.get(String.valueOf(dn)));
|
}
|
catch (Exception e)
|
{
|
if (debugEnabled())
|
{
|
TRACER.debugCaught(DebugLogLevel.ERROR, e);
|
}
|
|
messages.add(ERR_SCHEMA_CANNOT_REGISTER_BASE_DN.get(
|
String.valueOf(dn),
|
getExceptionMessage(e)));
|
resultCode = DirectoryServer.getServerErrorResultCode();
|
}
|
}
|
|
|
showAllAttributes = newShowAllAttributes;
|
|
|
userDefinedAttributes = newUserAttrs;
|
Message message = INFO_SCHEMA_USING_NEW_USER_ATTRS.get();
|
messages.add(message);
|
}
|
|
|
currentConfig = backendCfg;
|
return new ConfigChangeResult(resultCode, adminActionRequired, messages);
|
}
|
|
|
|
/**
|
* Indicates whether to treat common schema attributes like user attributes
|
* rather than operational attributes.
|
*
|
* @return {@code true} if common attributes should be treated like user
|
* attributes, or {@code false} if not.
|
*/
|
boolean showAllAttributes()
|
{
|
return showAllAttributes;
|
}
|
|
|
|
/**
|
* Specifies whether to treat common schema attributes like user attributes
|
* rather than operational attributes.
|
*
|
* @param showAllAttributes Specifies whether to treat common schema
|
* attributes like user attributes rather than
|
* operational attributes.
|
*/
|
void setShowAllAttributes(boolean showAllAttributes)
|
{
|
this.showAllAttributes = showAllAttributes;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public DN getComponentEntryDN()
|
{
|
return configEntryDN;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public String getClassName()
|
{
|
return CLASS_NAME;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public Map<String, String> getAlerts()
|
{
|
Map<String, String> alerts = new LinkedHashMap<String, String>();
|
|
alerts.put(ALERT_TYPE_CANNOT_COPY_SCHEMA_FILES,
|
ALERT_DESCRIPTION_CANNOT_COPY_SCHEMA_FILES);
|
alerts.put(ALERT_TYPE_CANNOT_WRITE_NEW_SCHEMA_FILES,
|
ALERT_DESCRIPTION_CANNOT_WRITE_NEW_SCHEMA_FILES);
|
|
return alerts;
|
}
|
|
|
|
/**
|
* {@inheritDoc}
|
*/
|
@Override
|
public void preloadEntryCache() throws UnsupportedOperationException {
|
throw new UnsupportedOperationException("Operation not supported.");
|
}
|
}
|