/* * 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 2006-2009 Sun Microsystems, Inc. * Portions Copyright 2011-2015 ForgeRock AS */ package org.opends.server.schema; import static org.opends.messages.SchemaMessages.*; import static org.opends.server.schema.SchemaConstants.*; import static org.opends.server.util.StaticUtils.*; import java.util.LinkedHashMap; import java.util.LinkedHashSet; import java.util.LinkedList; import java.util.List; import org.forgerock.i18n.LocalizableMessage; import org.forgerock.i18n.LocalizableMessageDescriptor.Arg2; import org.forgerock.opendj.ldap.ByteSequence; import org.forgerock.opendj.ldap.ResultCode; import org.forgerock.opendj.ldap.schema.ObjectClassType; import org.forgerock.opendj.ldap.schema.Syntax; import org.opends.server.admin.std.server.AttributeSyntaxCfg; import org.opends.server.api.AttributeSyntax; import org.opends.server.core.DirectoryServer; import org.opends.server.types.AttributeType; import org.opends.server.types.DITContentRule; import org.opends.server.types.DirectoryException; import org.opends.server.types.ObjectClass; import org.opends.server.types.Schema; /** * This class implements the DIT content rule description syntax, which is used * to hold DIT content rule definitions in the server schema. The format of * this syntax is defined in RFC 2252. */ public class DITContentRuleSyntax extends AttributeSyntax<AttributeSyntaxCfg> { /** * Creates a new instance of this syntax. Note that the only thing that * should be done here is to invoke the default constructor for the * superclass. All initialization should be performed in the * <CODE>initializeSyntax</CODE> method. */ public DITContentRuleSyntax() { super(); } /** {@inheritDoc} */ @Override public Syntax getSDKSyntax(org.forgerock.opendj.ldap.schema.Schema schema) { return schema.getSyntax(SchemaConstants.SYNTAX_DIT_CONTENT_RULE_OID); } /** {@inheritDoc} */ @Override public String getName() { return SYNTAX_DIT_CONTENT_RULE_NAME; } /** {@inheritDoc} */ @Override public String getOID() { return SYNTAX_DIT_CONTENT_RULE_OID; } /** {@inheritDoc} */ @Override public String getDescription() { return SYNTAX_DIT_CONTENT_RULE_DESCRIPTION; } /** * Decodes the contents of the provided ASN.1 octet string as a DIT content * rule definition according to the rules of this syntax. Note that the * provided octet string value does not need to be normalized (and in fact, it * should not be in order to allow the desired capitalization to be * preserved). * * @param value The ASN.1 octet string containing the value * to decode (it does not need to be * normalized). * @param schema The schema to use to resolve references to * other schema elements. * @param allowUnknownElements Indicates whether to allow values that * reference a name form and/or superior rules * which are not defined in the server schema. * This should only be true when called by * {@code valueIsAcceptable}. * * @return The decoded DIT content rule definition. * * @throws DirectoryException If the provided value cannot be decoded as an * DIT content rule definition. */ public static DITContentRule decodeDITContentRule(ByteSequence value, Schema schema, boolean allowUnknownElements) throws DirectoryException { // Get string representations of the provided value using the provided form // and with all lowercase characters. String valueStr = value.toString(); String lowerStr = toLowerCase(valueStr); // We'll do this a character at a time. First, skip over any leading // whitespace. int pos = 0; int length = valueStr.length(); while (pos < length && valueStr.charAt(pos) == ' ') { pos++; } if (pos >= length) { // This means that the value was empty or contained only whitespace. That // is illegal. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_EMPTY_VALUE.get(); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // The next character must be an open parenthesis. If it is not, then that // is an error. char c = valueStr.charAt(pos++); if (c != '(') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_EXPECTED_OPEN_PARENTHESIS.get(valueStr, pos-1, c); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Skip over any spaces immediately following the opening parenthesis. while (pos < length && ((c = valueStr.charAt(pos)) == ' ')) { pos++; } if (pos >= length) { // This means that the end of the value was reached before we could find // the OID. Ths is illegal. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // The next set of characters must be the OID. Strictly speaking, this // should only be a numeric OID, but we'll also allow for the // "ocname-oid" case as well. Look at the first character to figure out // which we will be using. int oidStartPos = pos; if (isDigit(c)) { // This must be a numeric OID. In that case, we will accept only digits // and periods, but not consecutive periods. boolean lastWasPeriod = false; while (pos < length && ((c = valueStr.charAt(pos++)) != ' ')) { if (c == '.') { if (lastWasPeriod) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_DOUBLE_PERIOD_IN_NUMERIC_OID. get(valueStr, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } lastWasPeriod = true; } else if (! isDigit(c)) { // This must have been an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR_IN_NUMERIC_OID.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } else { lastWasPeriod = false; } } } else { // This must be a "fake" OID. In this case, we will only accept // alphabetic characters, numeric digits, and the hyphen. while (pos < length && ((c = valueStr.charAt(pos++)) != ' ')) { if (isAlpha(c) || isDigit(c) || c == '-' || (c == '_' && DirectoryServer.allowAttributeNameExceptions())) { // This is fine. It is an acceptable character. } else { // This must have been an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR_IN_STRING_OID.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } // If we're at the end of the value, then it isn't a valid DIT content rule // description. Otherwise, parse out the OID. if (pos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } String oid = lowerStr.substring(oidStartPos, pos-1); // Get the objectclass with the specified OID. If it does not exist or is // not structural, then fail. ObjectClass structuralClass = schema.getObjectClass(oid); if (structuralClass == null) { if (allowUnknownElements) { structuralClass = DirectoryServer.getDefaultObjectClass(oid); } else { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_UNKNOWN_STRUCTURAL_CLASS.get(valueStr, oid); throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message); } } else if (structuralClass.getObjectClassType() != ObjectClassType.STRUCTURAL) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_STRUCTURAL_CLASS_NOT_STRUCTURAL. get(valueStr, oid, structuralClass.getNameOrOID(), structuralClass.getObjectClassType()); throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message); } // Skip over the space(s) after the OID. while (pos < length && ((c = valueStr.charAt(pos)) == ' ')) { pos++; } if (pos >= length) { // This means that the end of the value was reached before we could find // the OID. Ths is illegal. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // At this point, we should have a pretty specific syntax that describes // what may come next, but some of the components are optional and it would // be pretty easy to put something in the wrong order, so we will be very // flexible about what we can accept. Just look at the next token, figure // out what it is and how to treat what comes after it, then repeat until // we get to the end of the value. But before we start, set default values // for everything else we might need to know. LinkedHashMap<String,String> names = new LinkedHashMap<>(); String description = null; boolean isObsolete = false; LinkedHashSet<ObjectClass> auxiliaryClasses = new LinkedHashSet<>(); LinkedHashSet<AttributeType> requiredAttributes = new LinkedHashSet<>(); LinkedHashSet<AttributeType> optionalAttributes = new LinkedHashSet<>(); LinkedHashSet<AttributeType> prohibitedAttributes = new LinkedHashSet<>(); LinkedHashMap<String,List<String>> extraProperties = new LinkedHashMap<>(); while (true) { StringBuilder tokenNameBuffer = new StringBuilder(); pos = readTokenName(valueStr, tokenNameBuffer, pos); String tokenName = tokenNameBuffer.toString(); String lowerTokenName = toLowerCase(tokenName); if (tokenName.equals(")")) { // We must be at the end of the value. If not, then that's a problem. if (pos < length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_UNEXPECTED_CLOSE_PARENTHESIS. get(valueStr, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } break; } else if (lowerTokenName.equals("name")) { // This specifies the set of names for the DIT content rule. It may be // a single name in single quotes, or it may be an open parenthesis // followed by one or more names in single quotes separated by spaces. c = valueStr.charAt(pos++); if (c == '\'') { StringBuilder userBuffer = new StringBuilder(); StringBuilder lowerBuffer = new StringBuilder(); pos = readQuotedString(valueStr, lowerStr, userBuffer, lowerBuffer, pos-1); names.put(lowerBuffer.toString(), userBuffer.toString()); } else if (c == '(') { StringBuilder userBuffer = new StringBuilder(); StringBuilder lowerBuffer = new StringBuilder(); pos = readQuotedString(valueStr, lowerStr, userBuffer, lowerBuffer, pos); names.put(lowerBuffer.toString(), userBuffer.toString()); while (true) { if (valueStr.charAt(pos) == ')') { // Skip over any spaces after the parenthesis. pos++; while (pos < length && ((c = valueStr.charAt(pos)) == ' ')) { pos++; } break; } else { userBuffer = new StringBuilder(); lowerBuffer = new StringBuilder(); pos = readQuotedString(valueStr, lowerStr, userBuffer, lowerBuffer, pos); names.put(lowerBuffer.toString(), userBuffer.toString()); } } } else { // This is an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } else if (lowerTokenName.equals("desc")) { // This specifies the description for the DIT content rule. It is an // arbitrary string of characters enclosed in single quotes. StringBuilder descriptionBuffer = new StringBuilder(); pos = readQuotedString(valueStr, descriptionBuffer, pos); description = descriptionBuffer.toString(); } else if (lowerTokenName.equals("obsolete")) { // This indicates whether the DIT content rule should be considered // obsolete. We do not need to do any more parsing for this token. isObsolete = true; } else if (lowerTokenName.equals("aux")) { LinkedList<ObjectClass> ocs = new LinkedList<>(); // This specifies the set of required auxiliary objectclasses for this // DIT content rule. It may be a single name or OID (not in quotes), or // it may be an open parenthesis followed by one or more names separated // by spaces and the dollar sign character, followed by a closing // parenthesis. c = valueStr.charAt(pos++); if (c == '(') { while (true) { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos); ObjectClass oc = schema.getObjectClass(woidBuffer.toString()); if (oc == null) { // This isn't good because it is an unknown auxiliary class. if (allowUnknownElements) { oc = DirectoryServer.getDefaultAuxiliaryObjectClass( woidBuffer.toString()); } else { throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, ERR_ATTR_SYNTAX_DCR_UNKNOWN_AUXILIARY_CLASS.get( valueStr, woidBuffer)); } } else if (oc.getObjectClassType() != ObjectClassType.AUXILIARY) { // This isn't good because it isn't an auxiliary class. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_AUXILIARY_CLASS_NOT_AUXILIARY. get(valueStr, woidBuffer, oc.getObjectClassType()); throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message); } ocs.add(oc); // The next character must be either a dollar sign or a closing // parenthesis. c = valueStr.charAt(pos++); if (c == ')') { // This denotes the end of the list. break; } else if (c != '$') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos-1); ObjectClass oc = schema.getObjectClass(woidBuffer.toString()); if (oc == null) { // This isn't good because it is an unknown auxiliary class. if (allowUnknownElements) { oc = DirectoryServer.getDefaultAuxiliaryObjectClass( woidBuffer.toString()); } else { throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, ERR_ATTR_SYNTAX_DCR_UNKNOWN_AUXILIARY_CLASS.get(valueStr, woidBuffer)); } } else if (oc.getObjectClassType() != ObjectClassType.AUXILIARY) { // This isn't good because it isn't an auxiliary class. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_AUXILIARY_CLASS_NOT_AUXILIARY. get(valueStr, woidBuffer, oc.getObjectClassType()); throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message); } ocs.add(oc); } auxiliaryClasses.addAll(ocs); } else if (lowerTokenName.equals("must")) { LinkedList<AttributeType> attrs = new LinkedList<>(); // This specifies the set of required attributes for the DIT content // rule. It may be a single name or OID (not in quotes), or it may be // an open parenthesis followed by one or more names separated by spaces // and the dollar sign character, followed by a closing parenthesis. c = valueStr.charAt(pos++); if (c == '(') { while (true) { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_REQUIRED_ATTR)); // The next character must be either a dollar sign or a closing parenthesis. c = valueStr.charAt(pos++); if (c == ')') { // This denotes the end of the list. break; } else if (c != '$') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos-1); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_REQUIRED_ATTR)); } requiredAttributes.addAll(attrs); } else if (lowerTokenName.equals("may")) { LinkedList<AttributeType> attrs = new LinkedList<>(); // This specifies the set of optional attributes for the DIT content // rule. It may be a single name or OID (not in quotes), or it may be // an open parenthesis followed by one or more names separated by spaces // and the dollar sign character, followed by a closing parenthesis. c = valueStr.charAt(pos++); if (c == '(') { while (true) { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_OPTIONAL_ATTR)); // The next character must be either a dollar sign or a closing parenthesis. c = valueStr.charAt(pos++); if (c == ')') { // This denotes the end of the list. break; } else if (c != '$') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos-1); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_OPTIONAL_ATTR)); } optionalAttributes.addAll(attrs); } else if (lowerTokenName.equals("not")) { LinkedList<AttributeType> attrs = new LinkedList<>(); // This specifies the set of prohibited attributes for the DIT content // rule. It may be a single name or OID (not in quotes), or it may be // an open parenthesis followed by one or more names separated by spaces // and the dollar sign character, followed by a closing parenthesis. c = valueStr.charAt(pos++); if (c == '(') { while (true) { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_PROHIBITED_ATTR)); // The next character must be either a dollar sign or a closing parenthesis. c = valueStr.charAt(pos++); if (c == ')') { // This denotes the end of the list. break; } else if (c != '$') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, pos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { StringBuilder woidBuffer = new StringBuilder(); pos = readWOID(lowerStr, woidBuffer, pos-1); attrs.add(getAttribute(schema, allowUnknownElements, valueStr, woidBuffer, ERR_ATTR_SYNTAX_DCR_UNKNOWN_PROHIBITED_ATTR)); } prohibitedAttributes.addAll(attrs); } else { // This must be a non-standard property and it must be followed by // either a single value in single quotes or an open parenthesis // followed by one or more values in single quotes separated by spaces // followed by a close parenthesis. LinkedList<String> valueList = new LinkedList<>(); pos = readExtraParameterValues(valueStr, valueList, pos); extraProperties.put(tokenName, valueList); } } // Make sure that none of the prohibited attributes is required by the // structural or any of the auxiliary classes. for (AttributeType t : prohibitedAttributes) { if (structuralClass.isRequired(t)) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_PROHIBITED_REQUIRED_BY_STRUCTURAL. get(valueStr, t.getNameOrOID(), structuralClass.getNameOrOID()); throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, message); } for (ObjectClass oc : auxiliaryClasses) { if (oc.isRequired(t)) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_PROHIBITED_REQUIRED_BY_AUXILIARY. get(valueStr, t.getNameOrOID(), oc.getNameOrOID()); throw new DirectoryException( ResultCode.CONSTRAINT_VIOLATION, message); } } } return new DITContentRule(value.toString(), structuralClass, names, description, auxiliaryClasses, requiredAttributes, optionalAttributes, prohibitedAttributes, isObsolete, extraProperties); } private static AttributeType getAttribute(Schema schema, boolean allowUnknownElements, String valueStr, StringBuilder woidBuffer, Arg2<Object, Object> msg) throws DirectoryException { String woidString = woidBuffer.toString(); AttributeType attr = schema.getAttributeType(woidString); if (attr == null) { // This isn't good because it means that the DIT content rule // refers to an attribute type that we don't know anything about. if (!allowUnknownElements) { throw new DirectoryException(ResultCode.CONSTRAINT_VIOLATION, msg.get(valueStr, woidString)); } attr = DirectoryServer.getAttributeTypeOrDefault(woidString); } return attr; } /** * Reads the next token name from the DIT content rule definition, skipping * over any leading or trailing spaces, and appends it to the provided buffer. * * @param valueStr The string representation of the DIT content rule * definition. * @param tokenName The buffer into which the token name will be written. * @param startPos The position in the provided string at which to start * reading the token name. * * @return The position of the first character that is not part of the token * name or one of the trailing spaces after it. * * @throws DirectoryException If a problem is encountered while reading the * token name. */ private static int readTokenName(String valueStr, StringBuilder tokenName, int startPos) throws DirectoryException { // Skip over any spaces at the beginning of the value. char c = '\u0000'; int length = valueStr.length(); while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Read until we find the next space. while (startPos < length && ((c = valueStr.charAt(startPos++)) != ' ')) { tokenName.append(c); } // Skip over any trailing spaces after the value. while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } // Return the position of the first non-space character after the token. return startPos; } /** * Reads the value of a string enclosed in single quotes, skipping over the * quotes and any leading or trailing spaces, and appending the string to the * provided buffer. * * @param valueStr The user-provided representation of the DIT content * rule definition. * @param valueBuffer The buffer into which the user-provided representation * of the value will be placed. * @param startPos The position in the provided string at which to start * reading the quoted string. * * @return The position of the first character that is not part of the quoted * string or one of the trailing spaces after it. * * @throws DirectoryException If a problem is encountered while reading the * quoted string. */ private static int readQuotedString(String valueStr, StringBuilder valueBuffer, int startPos) throws DirectoryException { // Skip over any spaces at the beginning of the value. char c = '\u0000'; int length = valueStr.length(); while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // The next character must be a single quote. if (c != '\'') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_EXPECTED_QUOTE_AT_POS.get(valueStr, startPos, c); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Read until we find the closing quote. startPos++; while (startPos < length && ((c = valueStr.charAt(startPos)) != '\'')) { valueBuffer.append(c); startPos++; } // Skip over any trailing spaces after the value. startPos++; while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } // If we're at the end of the value, then that's illegal. if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Return the position of the first non-space character after the token. return startPos; } /** * Reads the value of a string enclosed in single quotes, skipping over the * quotes and any leading or trailing spaces, and appending the string to the * provided buffer. * * @param valueStr The user-provided representation of the DIT content * rule definition. * @param lowerStr The all-lowercase representation of the DIT content * rule definition. * @param userBuffer The buffer into which the user-provided representation * of the value will be placed. * @param lowerBuffer The buffer into which the all-lowercase representation * of the value will be placed. * @param startPos The position in the provided string at which to start * reading the quoted string. * * @return The position of the first character that is not part of the quoted * string or one of the trailing spaces after it. * * @throws DirectoryException If a problem is encountered while reading the * quoted string. */ private static int readQuotedString(String valueStr, String lowerStr, StringBuilder userBuffer, StringBuilder lowerBuffer, int startPos) throws DirectoryException { // Skip over any spaces at the beginning of the value. char c = '\u0000'; int length = lowerStr.length(); while (startPos < length && ((c = lowerStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(lowerStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // The next character must be a single quote. if (c != '\'') { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_EXPECTED_QUOTE_AT_POS.get(valueStr, startPos, c); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Read until we find the closing quote. startPos++; while (startPos < length && ((c = lowerStr.charAt(startPos)) != '\'')) { lowerBuffer.append(c); userBuffer.append(valueStr.charAt(startPos)); startPos++; } // Skip over any trailing spaces after the value. startPos++; while (startPos < length && ((c = lowerStr.charAt(startPos)) == ' ')) { startPos++; } // If we're at the end of the value, then that's illegal. if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(lowerStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Return the position of the first non-space character after the token. return startPos; } /** * Reads an attributeType/objectclass description or numeric OID from the * provided string, skipping over any leading or trailing spaces, and * appending the value to the provided buffer. * * @param lowerStr The string from which the name or OID is to be read. * @param woidBuffer The buffer into which the name or OID should be * appended. * @param startPos The position at which to start reading. * * @return The position of the first character after the name or OID that is * not a space. * * @throws DirectoryException If a problem is encountered while reading the * name or OID. */ private static int readWOID(String lowerStr, StringBuilder woidBuffer, int startPos) throws DirectoryException { // Skip over any spaces at the beginning of the value. char c = '\u0000'; int length = lowerStr.length(); while (startPos < length && ((c = lowerStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(lowerStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // The next character must be either numeric (for an OID) or alphabetic (for // an attribute type/objectclass description). if (isDigit(c)) { // This must be a numeric OID. In that case, we will accept only digits // and periods, but not consecutive periods. boolean lastWasPeriod = false; while (startPos < length && ((c = lowerStr.charAt(startPos++)) != ' ')) { if (c == '.') { if (lastWasPeriod) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_DOUBLE_PERIOD_IN_NUMERIC_OID. get(lowerStr, startPos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } else { woidBuffer.append(c); lastWasPeriod = true; } } else if (! isDigit(c)) { // Technically, this must be an illegal character. However, it is // possible that someone just got sloppy and did not include a space // between the name/OID and a closing parenthesis. In that case, // we'll assume it's the end of the value. What's more, we'll have // to prematurely return to nasty side effects from stripping off // additional characters. if (c == ')') { return startPos-1; } // This must have been an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR_IN_NUMERIC_OID.get( lowerStr, c, startPos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } else { woidBuffer.append(c); lastWasPeriod = false; } } } else if (isAlpha(c)) { // This must be an attribute type/objectclass description. In this case, // we will only accept alphabetic characters, numeric digits, and the hyphen. while (startPos < length && ((c = lowerStr.charAt(startPos++)) != ' ')) { if (isAlpha(c) || isDigit(c) || c == '-' || (c == '_' && DirectoryServer.allowAttributeNameExceptions())) { woidBuffer.append(c); } else { // Technically, this must be an illegal character. However, it is // possible that someone just got sloppy and did not include a space // between the name/OID and a closing parenthesis. In that case, // we'll assume it's the end of the value. What's more, we'll have // to prematurely return to nasty side effects from stripping off // additional characters. if (c == ')') { return startPos-1; } // This must have been an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR_IN_STRING_OID.get( lowerStr, c, startPos-1); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(lowerStr, c, startPos); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Skip over any trailing spaces after the value. while (startPos < length && ((c = lowerStr.charAt(startPos)) == ' ')) { startPos++; } // If we're at the end of the value, then that's illegal. if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(lowerStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Return the position of the first non-space character after the token. return startPos; } /** * Reads the value for an "extra" parameter. It will handle a single unquoted * word (which is technically illegal, but we'll allow it), a single quoted * string, or an open parenthesis followed by a space-delimited set of quoted * strings or unquoted words followed by a close parenthesis. * * @param valueStr The string containing the information to be read. * @param valueList The list of "extra" parameter values read so far. * @param startPos The position in the value string at which to start * reading. * * @return The "extra" parameter value that was read. * * @throws DirectoryException If a problem occurs while attempting to read * the value. */ private static int readExtraParameterValues(String valueStr, List<String> valueList, int startPos) throws DirectoryException { // Skip over any leading spaces. int length = valueStr.length(); char c = '\u0000'; while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } // Look at the next character. If it is a quote, then parse until the next // quote and end. If it is an open parenthesis, then parse individual // values until the close parenthesis and end. Otherwise, parse until the // next space and end. if (c == '\'') { // Parse until the closing quote. StringBuilder valueBuffer = new StringBuilder(); startPos++; while (startPos < length && ((c = valueStr.charAt(startPos)) != '\'')) { valueBuffer.append(c); startPos++; } startPos++; valueList.add(valueBuffer.toString()); } else if (c == '(') { startPos++; // We're expecting a list of values. Quoted, space separated. while (true) { // Skip over any leading spaces; while (startPos < length && ((c = valueStr.charAt(startPos)) == ' ')) { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } if (c == ')') { // This is the end of the list. startPos++; break; } else if (c == '(') { // This is an illegal character. LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_ILLEGAL_CHAR.get(valueStr, c, startPos); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } else if (c == '\'') { // We have a quoted string StringBuilder valueBuffer = new StringBuilder(); startPos++; while (startPos < length && ((c = valueStr.charAt(startPos)) != '\'')) { valueBuffer.append(c); startPos++; } valueList.add(valueBuffer.toString()); startPos++; } else { //Consider unquoted string StringBuilder valueBuffer = new StringBuilder(); while (startPos < length && ((c = valueStr.charAt(startPos)) != ' ')) { valueBuffer.append(c); startPos++; } valueList.add(valueBuffer.toString()); } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException(ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } } } else { // Parse until the next space. StringBuilder valueBuffer = new StringBuilder(); while (startPos < length && ((c = valueStr.charAt(startPos)) != ' ')) { valueBuffer.append(c); startPos++; } valueList.add(valueBuffer.toString()); } // Skip over any trailing spaces. while (startPos < length && valueStr.charAt(startPos) == ' ') { startPos++; } if (startPos >= length) { LocalizableMessage message = ERR_ATTR_SYNTAX_DCR_TRUNCATED_VALUE.get(valueStr); throw new DirectoryException( ResultCode.INVALID_ATTRIBUTE_SYNTAX, message); } return startPos; } }