Bug 66425: Avoid exceptions found via poi-fuzz

Processing formats uses regular expressions. Very complex formats
can recurse very deeply and thus can cause StackOVerflows depending
on the used stack-size.

In order to handle this a bit more gracefully, we now catch this
and report a better exception with details about the parsed 
format and potential mitigation.

Fixes https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=66137

git-svn-id: https://svn.apache.org/repos/asf/poi/trunk@1919342 13f79535-47bb-0310-9956-ffa450edef68

2 files changed, 31 insertions, 11 deletions

diff --git a/poi/src/main/java/org/apache/poi/ss/format/CellFormat.java b/poi/src/main/java/org/apache/poi/ss/format/CellFormat.java
index c95caca273..9a420540a5 100644
--- a/poi/src/main/java/org/apache/poi/ss/format/CellFormat.java
+++ b/poi/src/main/java/org/apache/poi/ss/format/CellFormat.java
@@ -28,7 +28,6 @@ import java.util.regex.Pattern;
 
 import javax.swing.JLabel;
 
-import org.apache.logging.log4j.Level;
 import org.apache.logging.log4j.LogManager;
 import org.apache.logging.log4j.Logger;
 import org.apache.poi.ss.usermodel.Cell;
@@ -182,19 +181,28 @@ public class CellFormat {
         Matcher m = ONE_PART.matcher(format);
         List<CellFormatPart> parts = new ArrayList<>();
 
-        while (m.find()) {
-            try {
-                String valueDesc = m.group();
+        try {
+            while (m.find()) {
+                try {
+                    String valueDesc = m.group();
 
-                // Strip out the semicolon if it's there
-                if (valueDesc.endsWith(";"))
-                    valueDesc = valueDesc.substring(0, valueDesc.length() - 1);
+                    // Strip out the semicolon if it's there
+                    if (valueDesc.endsWith(";"))
+                        valueDesc = valueDesc.substring(0, valueDesc.length() - 1);
 
-                parts.add(new CellFormatPart(locale, valueDesc));
-            } catch (RuntimeException e) {
-                LOG.warn("Invalid format: {}", CellFormatter.quote(m.group()), e);
-                parts.add(null);
+                    parts.add(new CellFormatPart(locale, valueDesc));
+                } catch (RuntimeException e) {
+                    LOG.warn("Invalid format: {}", CellFormatter.quote(m.group()), e);
+                    parts.add(null);
+                }
             }
+        } catch (StackOverflowError e) {
+            // very complex formats can cause the regex-parsing to exceed the available stack
+            // we want to handle this more gracefully by catching it and reporting a bit more
+            // details in the error message
+            throw new IllegalStateException("The provided format is too complex: " + format +
+                    ", you can try to increase Java Stack size via commandline argument '-Xss' " +
+                    "to allow handling this format");
         }
 
         formatPartCount = parts.size();
diff --git a/poi/src/test/java/org/apache/poi/hssf/extractor/TestExcelExtractor.java b/poi/src/test/java/org/apache/poi/hssf/extractor/TestExcelExtractor.java
index 495ceb9837..8e27789f97 100644
--- a/poi/src/test/java/org/apache/poi/hssf/extractor/TestExcelExtractor.java
+++ b/poi/src/test/java/org/apache/poi/hssf/extractor/TestExcelExtractor.java
@@ -382,4 +382,16 @@ final class TestExcelExtractor {
             assertContains(txt, "Macro2");
         }
     }
+
+    @Test
+    void testStackOverflowInRegex() throws IOException {
+        try (ExcelExtractor extractor = createExtractor("clusterfuzz-testcase-minimized-POIHSSFFuzzer-4657005060816896.xls")) {
+            extractor.getText();
+        } catch (IllegalStateException e) {
+            // we either get a StackOverflow or a parsing error depending on the stack-size of the current JVM,
+            // so we expect both here
+            assertTrue(e.getMessage().contains("Provided formula is too complex") ||
+                    e.getMessage().contains("Did not have a ExtendedFormatRecord"));
+        }
+    }
 }