Decode attribute content differently from text node content (#255)

As described in issue #254 `&para` is a full complete character reference when decoding text node content, but not when decoding attribute content which causes problems for URL attribute values like /test?param1=foo&param2=bar As shown via JS test code in that issue, a small set of next characters prevent a character reference name match from being considered complete. This commit: - modifies the decode functions to take an extra parameter `boolean inAttribute`, and modifies the Trie traversal loops to not store a longest match so far based on that parameter and some next character tests - modifies the HTML lexer to pass that attribute appropriately - for backwards compat, leaves the old APIs in place but `@deprecated` - adds unit tests for the decode functions - adds a unit test for the specific input from the issue This change should make us more conformant with observed browser behaviour so is not expected to cause compatibility problems for existing users. Fixes #254

src/main/java/org/owasp/html/Encoding.java

@@ -41,8 +41,21 @@ public final class Encoding {
  *
  * @param s text/html
  * @return text/plain
+ * @deprecated specify whether s is in an attribute value
  */
  public static String decodeHtml(String s) {
+ return decodeHtml(s, false);
+ }
+
+ /**
+ * Decodes HTML entities to produce a string containing only valid
+ * Unicode scalar values.
+ *
+ * @param s text/html
+ * @param inAttribute is s in an attribute value?
+ * @return text/plain
+ */
+ public static String decodeHtml(String s, boolean inAttribute) {
  int firstAmp = s.indexOf('&');
  int safeLimit = longestPrefixOfGoodCodeunits(s);
  if ((firstAmp & safeLimit) < 0) { return s; }
@@ -55,7 +68,7 @@ public static String decodeHtml(String s) {
  int amp = firstAmp;
  while (amp >= 0) {
  sb.append(s, pos, amp);
- int end = HtmlEntities.appendDecodedEntity(s, amp, n, sb);
+ int end = HtmlEntities.appendDecodedEntity(s, amp, n, inAttribute, sb);
  pos = end;
  amp = s.indexOf('&', end);
  }

src/main/java/org/owasp/html/HtmlEntities.java

@@ -2307,9 +2307,26 @@ final class HtmlEntities {
  * in {@code html}.
  * @param sb string builder to append to.
  * @return The offset after the end of the decoded sequence in {@code html}.
+ * @deprecated specify whether html is in an attribute value.
  */
  public static int appendDecodedEntity(
- String html, int offset, int limit, StringBuilder sb) {
+ String html, int offset, int limit, StringBuilder sb) {
+ return appendDecodedEntity(html, offset, limit, false, sb);
+ }
+
+ /**
+ * Decodes any HTML entity at the given location and appends it to a string
+ * builder. This handles both named and numeric entities.
+ *
+ * @param html HTML text.
+ * @param offset the position of the sequence to decode in {@code html}.
+ * @param limit the last position that could be part of the sequence to decode
+ * in {@code html}.
+ * @param sb string builder to append to.
+ * @return The offset after the end of the decoded sequence in {@code html}.
+ */
+ public static int appendDecodedEntity(
+ String html, int offset, int limit, boolean inAttribute, StringBuilder sb) {
  char ch = html.charAt(offset);
  if ('&' != ch) {
  sb.append(ch);
@@ -2422,19 +2439,20 @@ public static int appendDecodedEntity(
  char nameChar = html.charAt(i);
  t = t.lookup(nameChar);
  if (t == null) { break; }
- if (t.isTerminal()) {
+ if (t.isTerminal() && mayComplete(inAttribute, html, i, limit)) {
  longestDecode = t;
  tail = i + 1;
  }
  }
+ // Try again, case insensitively.
  if (longestDecode == null) {
  t = ENTITY_TRIE;
  for (int i = offset + 1; i < limit; ++i) {
  char nameChar = html.charAt(i);
  if ('Z' >= nameChar && nameChar >= 'A') { nameChar |= 32; }
  t = t.lookup(nameChar);
  if (t == null) { break; }
- if (t.isTerminal()) {
+ if (t.isTerminal() && mayComplete(inAttribute, html, i, limit)) {
  longestDecode = t;
  tail = i + 1;
  }
@@ -2456,11 +2474,35 @@ public static int appendDecodedEntity(
 
  private static boolean isHtmlIdContinueChar(char ch) {
  int chLower = ch | 32;
- return ('0' <= chLower && chLower <= '9')
+ return ('0' <= ch && ch <= '9')
  || ('a' <= chLower && chLower <= 'z')
  || ('-' == ch);
  }
 
+ /** True if the character at i in html may complete a named character reference */
+ private static boolean mayComplete(boolean inAttribute, String html, int i, int limit) {
+ if (inAttribute && html.charAt(i) != ';' && i + 1 < limit) {
+ // See if the next character blocks treating this as a full match.
+ // This avoids problems like "&para" being treated as a decoding in
+ // <a href="?foo&param=1">
+ if (continuesCharacterReferenceName(html.charAt(i + 1))) {
+ return false;
+ }
+ }
+ return true;
+ }
+
+ /**
+ * @see <a href="https://github.com/OWASP/java-html-sanitizer/issues/254#issuecomment-1080864368"
+ * >comments in issue 254</a>
+ */
+ private static boolean continuesCharacterReferenceName(char ch) {
+ int chLower = ch | 32;
+ return ('0' <= ch && ch <= '9')
+ || ('a' <= chLower && chLower <= 'z')
+ || (ch == '=');
+ }
+
 // /** A possible entity name like "amp" or "gt". */
 // public static boolean isEntityName(String name) {
 // Trie t = ENTITY_TRIE;

src/main/java/org/owasp/html/HtmlSanitizer.java

@@ -144,7 +144,7 @@ public static void sanitize(
  switch (token.type) {
  case TEXT:
  receiver.text(
- Encoding.decodeHtml(htmlContent.substring(token.start, token.end)));
+ Encoding.decodeHtml(htmlContent.substring(token.start, token.end), false));
  break;
  case UNESCAPED:
  receiver.text(Encoding.stripBannedCodeunits(
@@ -177,8 +177,9 @@ public static void sanitize(
  htmlContent.substring(tagBodyToken.start, tagBodyToken.end)));
  break;
  case ATTRVALUE:
- attrs.add(Encoding.decodeHtml(stripQuotes(
- htmlContent.substring(tagBodyToken.start, tagBodyToken.end))));
+ String attributeContentRaw =
+ stripQuotes(htmlContent.substring(tagBodyToken.start, tagBodyToken.end));
+ attrs.add(Encoding.decodeHtml(attributeContentRaw, true));
  attrsReadyForName = true;
  break;
  case TAGEND: