/* * Copyright (c) 1999, 2013, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. Oracle designates this * particular file as subject to the "Classpath" exception as provided * by Oracle in the LICENSE file that accompanied this code. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. */ package com.sun.jndi.ldap; import java.util.Enumeration; import java.util.Vector; import java.util.Locale; import javax.naming.*; import javax.naming.directory.Attributes; import javax.naming.directory.Attribute; import javax.naming.directory.BasicAttributes; /** * LdapName implements compound names for LDAP v3 as * specified by RFC 2253. *

* RFC 2253 has a few ambiguities and outright inconsistencies. These * are resolved as follows: *

*

* String names passed to LdapName or returned by it * use the full 16-bit Unicode character set. They may also contain * characters encoded into UTF-8 with each octet represented by a * three-character substring such as "\\B4". * They may not, however, contain characters encoded into UTF-8 with * each octet represented by a single character in the string: the * meaning would be ambiguous. *

* LdapName will properly parse all valid names, but * does not attempt to detect all possible violations when parsing * invalid names. It's "generous". *

* When names are tested for equality, attribute types and binary * values are case-insensitive, and string values are by default * case-insensitive. * String values with different but equivalent usage of quoting, * escaping, or UTF8-hex-encoding are considered equal. The order of * components in multi-valued RDNs (such as "ou=Sales+cn=Bob") is not * significant. * * @author Scott Seligman */ public final class LdapName implements Name { private transient String unparsed; // if non-null, the DN in unparsed form private transient Vector rdns; // parsed name components private transient boolean valuesCaseSensitive = false; /** * Constructs an LDAP name from the given DN. * * @param name An LDAP DN. To JNDI, a compound name. * * @throws InvalidNameException if a syntax violation is detected. */ public LdapName(String name) throws InvalidNameException { unparsed = name; parse(); } /* * Constructs an LDAP name given its parsed components and, optionally * (if "name" is not null), the unparsed DN. */ @SuppressWarnings("unchecked") // clone() private LdapName(String name, Vector rdns) { unparsed = name; this.rdns = (Vector)rdns.clone(); } /* * Constructs an LDAP name given its parsed components (the elements * of "rdns" in the range [beg,end)) and, optionally * (if "name" is not null), the unparsed DN. */ private LdapName(String name, Vector rdns, int beg, int end) { unparsed = name; this.rdns = new Vector<>(); for (int i = beg; i < end; i++) { this.rdns.addElement(rdns.elementAt(i)); } } public Object clone() { return new LdapName(unparsed, rdns); } public String toString() { if (unparsed != null) { return unparsed; } StringBuffer buf = new StringBuffer(); for (int i = rdns.size() - 1; i >= 0; i--) { if (i < rdns.size() - 1) { buf.append(','); } Rdn rdn = rdns.elementAt(i); buf.append(rdn); } unparsed = new String(buf); return unparsed; } public boolean equals(Object obj) { return ((obj instanceof LdapName) && (compareTo(obj) == 0)); } public int compareTo(Object obj) { LdapName that = (LdapName)obj; if ((obj == this) || // check possible shortcuts (unparsed != null && unparsed.equals(that.unparsed))) { return 0; } // Compare RDNs one by one, lexicographically. int minSize = Math.min(rdns.size(), that.rdns.size()); for (int i = 0 ; i < minSize; i++) { // Compare a single pair of RDNs. Rdn rdn1 = rdns.elementAt(i); Rdn rdn2 = that.rdns.elementAt(i); int diff = rdn1.compareTo(rdn2); if (diff != 0) { return diff; } } return (rdns.size() - that.rdns.size()); // longer DN wins } public int hashCode() { // Sum up the hash codes of the components. int hash = 0; // For each RDN... for (int i = 0; i < rdns.size(); i++) { Rdn rdn = rdns.elementAt(i); hash += rdn.hashCode(); } return hash; } public int size() { return rdns.size(); } public boolean isEmpty() { return rdns.isEmpty(); } public Enumeration getAll() { final Enumeration enum_ = rdns.elements(); return new Enumeration() { public boolean hasMoreElements() { return enum_.hasMoreElements(); } public String nextElement() { return enum_.nextElement().toString(); } }; } public String get(int pos) { return rdns.elementAt(pos).toString(); } public Name getPrefix(int pos) { return new LdapName(null, rdns, 0, pos); } public Name getSuffix(int pos) { return new LdapName(null, rdns, pos, rdns.size()); } public boolean startsWith(Name n) { int len1 = rdns.size(); int len2 = n.size(); return (len1 >= len2 && matches(0, len2, n)); } public boolean endsWith(Name n) { int len1 = rdns.size(); int len2 = n.size(); return (len1 >= len2 && matches(len1 - len2, len1, n)); } /** * Controls whether string-values are treated as case-sensitive * when the string values within names are compared. The default * behavior is case-insensitive comparison. */ public void setValuesCaseSensitive(boolean caseSensitive) { toString(); rdns = null; // clear any cached information try { parse(); } catch (InvalidNameException e) { // shouldn't happen throw new IllegalStateException("Cannot parse name: " + unparsed); } valuesCaseSensitive = caseSensitive; } /* * Helper method for startsWith() and endsWith(). * Returns true if components [beg,end) match the components of "n". * If "n" is not an LdapName, each of its components is parsed as * the string form of an RDN. * The following must hold: end - beg == n.size(). */ private boolean matches(int beg, int end, Name n) { for (int i = beg; i < end; i++) { Rdn rdn; if (n instanceof LdapName) { LdapName ln = (LdapName)n; rdn = ln.rdns.elementAt(i - beg); } else { String rdnString = n.get(i - beg); try { rdn = (new DnParser(rdnString, valuesCaseSensitive)).getRdn(); } catch (InvalidNameException e) { return false; } } if (!rdn.equals(rdns.elementAt(i))) { return false; } } return true; } public Name addAll(Name suffix) throws InvalidNameException { return addAll(size(), suffix); } /* * If "suffix" is not an LdapName, each of its components is parsed as * the string form of an RDN. */ public Name addAll(int pos, Name suffix) throws InvalidNameException { if (suffix instanceof LdapName) { LdapName s = (LdapName)suffix; for (int i = 0; i < s.rdns.size(); i++) { rdns.insertElementAt(s.rdns.elementAt(i), pos++); } } else { Enumeration comps = suffix.getAll(); while (comps.hasMoreElements()) { DnParser p = new DnParser(comps.nextElement(), valuesCaseSensitive); rdns.insertElementAt(p.getRdn(), pos++); } } unparsed = null; // no longer valid return this; } public Name add(String comp) throws InvalidNameException { return add(size(), comp); } public Name add(int pos, String comp) throws InvalidNameException { Rdn rdn = (new DnParser(comp, valuesCaseSensitive)).getRdn(); rdns.insertElementAt(rdn, pos); unparsed = null; // no longer valid return this; } public Object remove(int pos) throws InvalidNameException { String comp = get(pos); rdns.removeElementAt(pos); unparsed = null; // no longer valid return comp; } private void parse() throws InvalidNameException { rdns = (new DnParser(unparsed, valuesCaseSensitive)).getDn(); } /* * Best guess as to what RFC 2253 means by "whitespace". */ private static boolean isWhitespace(char c) { return (c == ' ' || c == '\r'); } /** * Given the value of an attribute, returns a string suitable * for inclusion in a DN. If the value is a string, this is * accomplished by using backslash (\) to escape the following * characters: *

* If the value is a byte array, it is converted to hex * notation (such as "#CEB1DF80"). */ public static String escapeAttributeValue(Object val) { return TypeAndValue.escapeValue(val); } /** * Given an attribute value formatted according to RFC 2253, * returns the unformatted value. Returns a string value as * a string, and a binary value as a byte array. */ public static Object unescapeAttributeValue(String val) { return TypeAndValue.unescapeValue(val); } /** * Serializes only the unparsed DN, for compactness and to avoid * any implementation dependency. * * @serialData The DN string and a boolean indicating whether * the values are case sensitive. */ private void writeObject(java.io.ObjectOutputStream s) throws java.io.IOException { s.writeObject(toString()); s.writeBoolean(valuesCaseSensitive); } private void readObject(java.io.ObjectInputStream s) throws java.io.IOException, ClassNotFoundException { unparsed = (String)s.readObject(); valuesCaseSensitive = s.readBoolean(); try { parse(); } catch (InvalidNameException e) { // shouldn't happen throw new java.io.StreamCorruptedException( "Invalid name: " + unparsed); } } static final long serialVersionUID = -1595520034788997356L; /* * DnParser implements a recursive descent parser for a single DN. */ static class DnParser { private final String name; // DN being parsed private final char[] chars; // characters in LDAP name being parsed private final int len; // length of "chars" private int cur = 0; // index of first unconsumed char in "chars" private boolean valuesCaseSensitive; /* * Given an LDAP DN in string form, returns a parser for it. */ DnParser(String name, boolean valuesCaseSensitive) throws InvalidNameException { this.name = name; len = name.length(); chars = name.toCharArray(); this.valuesCaseSensitive = valuesCaseSensitive; } /* * Parses the DN, returning a Vector of its RDNs. */ Vector getDn() throws InvalidNameException { cur = 0; Vector rdns = new Vector<>(len / 3 + 10); // leave room for growth if (len == 0) { return rdns; } rdns.addElement(parseRdn()); while (cur < len) { if (chars[cur] == ',' || chars[cur] == ';') { ++cur; rdns.insertElementAt(parseRdn(), 0); } else { throw new InvalidNameException("Invalid name: " + name); } } return rdns; } /* * Parses the DN, if it is known to contain a single RDN. */ Rdn getRdn() throws InvalidNameException { Rdn rdn = parseRdn(); if (cur < len) { throw new InvalidNameException("Invalid RDN: " + name); } return rdn; } /* * Parses the next RDN and returns it. Throws an exception if * none is found. Leading and trailing whitespace is consumed. */ private Rdn parseRdn() throws InvalidNameException { Rdn rdn = new Rdn(); while (cur < len) { consumeWhitespace(); String attrType = parseAttrType(); consumeWhitespace(); if (cur >= len || chars[cur] != '=') { throw new InvalidNameException("Invalid name: " + name); } ++cur; // consume '=' consumeWhitespace(); String value = parseAttrValue(); consumeWhitespace(); rdn.add(new TypeAndValue(attrType, value, valuesCaseSensitive)); if (cur >= len || chars[cur] != '+') { break; } ++cur; // consume '+' } return rdn; } /* * Returns the attribute type that begins at the next unconsumed * char. No leading whitespace is expected. * This routine is more generous than RFC 2253. It accepts * attribute types composed of any nonempty combination of Unicode * letters, Unicode digits, '.', '-', and internal space characters. */ private String parseAttrType() throws InvalidNameException { final int beg = cur; while (cur < len) { char c = chars[cur]; if (Character.isLetterOrDigit(c) || c == '.' || c == '-' || c == ' ') { ++cur; } else { break; } } // Back out any trailing spaces. while ((cur > beg) && (chars[cur - 1] == ' ')) { --cur; } if (beg == cur) { throw new InvalidNameException("Invalid name: " + name); } return new String(chars, beg, cur - beg); } /* * Returns the attribute value that begins at the next unconsumed * char. No leading whitespace is expected. */ private String parseAttrValue() throws InvalidNameException { if (cur < len && chars[cur] == '#') { return parseBinaryAttrValue(); } else if (cur < len && chars[cur] == '"') { return parseQuotedAttrValue(); } else { return parseStringAttrValue(); } } private String parseBinaryAttrValue() throws InvalidNameException { final int beg = cur; ++cur; // consume '#' while (cur < len && Character.isLetterOrDigit(chars[cur])) { ++cur; } return new String(chars, beg, cur - beg); } private String parseQuotedAttrValue() throws InvalidNameException { final int beg = cur; ++cur; // consume '"' while ((cur < len) && chars[cur] != '"') { if (chars[cur] == '\\') { ++cur; // consume backslash, then what follows } ++cur; } if (cur >= len) { // no closing quote throw new InvalidNameException("Invalid name: " + name); } ++cur ; // consume closing quote return new String(chars, beg, cur - beg); } private String parseStringAttrValue() throws InvalidNameException { final int beg = cur; int esc = -1; // index of the most recently escaped character while ((cur < len) && !atTerminator()) { if (chars[cur] == '\\') { ++cur; // consume backslash, then what follows esc = cur; } ++cur; } if (cur > len) { // 'twas backslash followed by nothing throw new InvalidNameException("Invalid name: " + name); } // Trim off (unescaped) trailing whitespace. int end; for (end = cur; end > beg; end--) { if (!isWhitespace(chars[end - 1]) || (esc == end - 1)) { break; } } return new String(chars, beg, end - beg); } private void consumeWhitespace() { while ((cur < len) && isWhitespace(chars[cur])) { ++cur; } } /* * Returns true if next unconsumed character is one that terminates * a string attribute value. */ private boolean atTerminator() { return (cur < len && (chars[cur] == ',' || chars[cur] == ';' || chars[cur] == '+')); } } /* * Class Rdn represents a set of TypeAndValue. */ static class Rdn { /* * A vector of the TypeAndValue elements of this Rdn. * It is sorted to facilitate set operations. */ private final Vector tvs = new Vector<>(); void add(TypeAndValue tv) { // Set i to index of first element greater than tv, or to // tvs.size() if there is none. int i; for (i = 0; i < tvs.size(); i++) { int diff = tv.compareTo(tvs.elementAt(i)); if (diff == 0) { return; // tv is a duplicate: ignore it } else if (diff < 0) { break; } } tvs.insertElementAt(tv, i); } public String toString() { StringBuffer buf = new StringBuffer(); for (int i = 0; i < tvs.size(); i++) { if (i > 0) { buf.append('+'); } buf.append(tvs.elementAt(i)); } return new String(buf); } public boolean equals(Object obj) { return ((obj instanceof Rdn) && (compareTo(obj) == 0)); } // Compare TypeAndValue components one by one, lexicographically. public int compareTo(Object obj) { Rdn that = (Rdn)obj; int minSize = Math.min(tvs.size(), that.tvs.size()); for (int i = 0; i < minSize; i++) { // Compare a single pair of type/value pairs. TypeAndValue tv = tvs.elementAt(i); int diff = tv.compareTo(that.tvs.elementAt(i)); if (diff != 0) { return diff; } } return (tvs.size() - that.tvs.size()); // longer RDN wins } public int hashCode() { // Sum up the hash codes of the components. int hash = 0; // For each type/value pair... for (int i = 0; i < tvs.size(); i++) { hash += tvs.elementAt(i).hashCode(); } return hash; } Attributes toAttributes() { Attributes attrs = new BasicAttributes(true); TypeAndValue tv; Attribute attr; for (int i = 0; i < tvs.size(); i++) { tv = tvs.elementAt(i); if ((attr = attrs.get(tv.getType())) == null) { attrs.put(tv.getType(), tv.getUnescapedValue()); } else { attr.add(tv.getUnescapedValue()); } } return attrs; } } /* * Class TypeAndValue represents an attribute type and its * corresponding value. */ static class TypeAndValue { private final String type; private final String value; // value, escaped or quoted private final boolean binary; private final boolean valueCaseSensitive; // If non-null, a canonical representation of the value suitable // for comparison using String.compareTo(). private String comparable = null; TypeAndValue(String type, String value, boolean valueCaseSensitive) { this.type = type; this.value = value; binary = value.startsWith("#"); this.valueCaseSensitive = valueCaseSensitive; } public String toString() { return (type + "=" + value); } public int compareTo(Object obj) { // NB: Any change here affecting equality must be // reflected in hashCode(). TypeAndValue that = (TypeAndValue)obj; int diff = type.compareToIgnoreCase(that.type); if (diff != 0) { return diff; } if (value.equals(that.value)) { // try shortcut return 0; } return getValueComparable().compareTo(that.getValueComparable()); } public boolean equals(Object obj) { // NB: Any change here must be reflected in hashCode(). if (!(obj instanceof TypeAndValue)) { return false; } TypeAndValue that = (TypeAndValue)obj; return (type.equalsIgnoreCase(that.type) && (value.equals(that.value) || getValueComparable().equals(that.getValueComparable()))); } public int hashCode() { // If two objects are equal, their hash codes must match. return (type.toUpperCase(Locale.ENGLISH).hashCode() + getValueComparable().hashCode()); } /* * Returns the type. */ String getType() { return type; } /* * Returns the unescaped value. */ Object getUnescapedValue() { return unescapeValue(value); } /* * Returns a canonical representation of "value" suitable for * comparison using String.compareTo(). If "value" is a string, * it is returned with escapes and quotes stripped away, and * hex-encoded UTF-8 converted to 16-bit Unicode chars. * If value's case is to be ignored, it is returned in uppercase. * If "value" is binary, it is returned in uppercase but * otherwise unmodified. */ private String getValueComparable() { if (comparable != null) { return comparable; // return cached result } // cache result if (binary) { comparable = value.toUpperCase(Locale.ENGLISH); } else { comparable = (String)unescapeValue(value); if (!valueCaseSensitive) { // ignore case comparable = comparable.toUpperCase(Locale.ENGLISH); } } return comparable; } /* * Given the value of an attribute, returns a string suitable * for inclusion in a DN. */ static String escapeValue(Object val) { return (val instanceof byte[]) ? escapeBinaryValue((byte[])val) : escapeStringValue((String)val); } /* * Given the value of a string-valued attribute, returns a * string suitable for inclusion in a DN. This is accomplished by * using backslash (\) to escape the following characters: * leading and trailing whitespace * , = + < > # ; " \ */ private static String escapeStringValue(String val) { final String escapees = ",=+<>#;\"\\"; char[] chars = val.toCharArray(); StringBuffer buf = new StringBuffer(2 * val.length()); // Find leading and trailing whitespace. int lead; // index of first char that is not leading whitespace for (lead = 0; lead < chars.length; lead++) { if (!isWhitespace(chars[lead])) { break; } } int trail; // index of last char that is not trailing whitespace for (trail = chars.length - 1; trail >= 0; trail--) { if (!isWhitespace(chars[trail])) { break; } } for (int i = 0; i < chars.length; i++) { char c = chars[i]; if ((i < lead) || (i > trail) || (escapees.indexOf(c) >= 0)) { buf.append('\\'); } buf.append(c); } return new String(buf); } /* * Given the value of a binary attribute, returns a string * suitable for inclusion in a DN (such as "#CEB1DF80"). */ private static String escapeBinaryValue(byte[] val) { StringBuffer buf = new StringBuffer(1 + 2 * val.length); buf.append("#"); for (int i = 0; i < val.length; i++) { byte b = val[i]; buf.append(Character.forDigit(0xF & (b >>> 4), 16)); buf.append(Character.forDigit(0xF & b, 16)); } return (new String(buf)).toUpperCase(Locale.ENGLISH); } /* * Given an attribute value formatted according to RFC 2253, * returns the unformatted value. Escapes and quotes are * stripped away, and hex-encoded UTF-8 is converted to 16-bit * Unicode chars. Returns a string value as a String, and a * binary value as a byte array. */ static Object unescapeValue(String val) { char[] chars = val.toCharArray(); int beg = 0; int end = chars.length; // Trim off leading and trailing whitespace. while ((beg < end) && isWhitespace(chars[beg])) { ++beg; } while ((beg < end) && isWhitespace(chars[end - 1])) { --end; } // Add back the trailing whitespace with a preceding '\' // (escaped or unescaped) that was taken off in the above // loop. Whether or not to retain this whitespace is // decided below. if (end != chars.length && (beg < end) && chars[end - 1] == '\\') { end++; } if (beg >= end) { return ""; } if (chars[beg] == '#') { // Value is binary (eg: "#CEB1DF80"). return decodeHexPairs(chars, ++beg, end); } // Trim off quotes. if ((chars[beg] == '\"') && (chars[end - 1] == '\"')) { ++beg; --end; } StringBuffer buf = new StringBuffer(end - beg); int esc = -1; // index of the last escaped character for (int i = beg; i < end; i++) { if ((chars[i] == '\\') && (i + 1 < end)) { if (!Character.isLetterOrDigit(chars[i + 1])) { ++i; // skip backslash buf.append(chars[i]); // snarf escaped char esc = i; } else { // Convert hex-encoded UTF-8 to 16-bit chars. byte[] utf8 = getUtf8Octets(chars, i, end); if (utf8.length > 0) { try { buf.append(new String(utf8, "UTF8")); } catch (java.io.UnsupportedEncodingException e) { // shouldn't happen } i += utf8.length * 3 - 1; } else { throw new IllegalArgumentException( "Not a valid attribute string value:" + val +", improper usage of backslash"); } } } else { buf.append(chars[i]); // snarf unescaped char } } // Get rid of the unescaped trailing whitespace with the // preceding '\' character that was previously added back. int len = buf.length(); if (isWhitespace(buf.charAt(len - 1)) && esc != (end - 1)) { buf.setLength(len - 1); } return new String(buf); } /* * Given an array of chars (with starting and ending indexes into it) * representing bytes encoded as hex-pairs (such as "CEB1DF80"), * returns a byte array containing the decoded bytes. */ private static byte[] decodeHexPairs(char[] chars, int beg, int end) { byte[] bytes = new byte[(end - beg) / 2]; for (int i = 0; beg + 1 < end; i++) { int hi = Character.digit(chars[beg], 16); int lo = Character.digit(chars[beg + 1], 16); if (hi < 0 || lo < 0) { break; } bytes[i] = (byte)((hi<<4) + lo); beg += 2; } if (beg != end) { throw new IllegalArgumentException( "Illegal attribute value: #" + new String(chars)); } return bytes; } /* * Given an array of chars (with starting and ending indexes into it), * finds the largest prefix consisting of hex-encoded UTF-8 octets, * and returns a byte array containing the corresponding UTF-8 octets. * * Hex-encoded UTF-8 octets look like this: * \03\B1\DF\80 */ private static byte[] getUtf8Octets(char[] chars, int beg, int end) { byte[] utf8 = new byte[(end - beg) / 3]; // allow enough room int len = 0; // index of first unused byte in utf8 while ((beg + 2 < end) && (chars[beg++] == '\\')) { int hi = Character.digit(chars[beg++], 16); int lo = Character.digit(chars[beg++], 16); if (hi < 0 || lo < 0) { break; } utf8[len++] = (byte)((hi<<4) + lo); } if (len == utf8.length) { return utf8; } else { byte[] res = new byte[len]; System.arraycopy(utf8, 0, res, 0, len); return res; } } } /* * For testing. */ /* public static void main(String[] args) { try { if (args.length == 1) { // parse and print components LdapName n = new LdapName(args[0]); Enumeration rdns = n.rdns.elements(); while (rdns.hasMoreElements()) { Rdn rdn = (Rdn)rdns.nextElement(); for (int i = 0; i < rdn.tvs.size(); i++) { System.out.print("[" + rdn.tvs.elementAt(i) + "]"); } System.out.println(); } } else { // compare two names LdapName n1 = new LdapName(args[0]); LdapName n2 = new LdapName(args[1]); n1.unparsed = null; n2.unparsed = null; boolean eq = n1.equals(n2); System.out.println("[" + n1 + (eq ? "] == [" : "] != [") + n2 + "]"); } } catch (Exception e) { e.printStackTrace(); } } */ }