Fix buffer underflows in cipher light weight API input/output streams and beef up testing.

Buffer underflows could occur when:
- decrypting data > internal buffer size in output stream (input stream was fixed in prior commit)
- packet mode AE cipher (e.g. CCM) is used with a data size > internal buffer size (since all output is buffered)

Buffer is now sized appropriately to every cipher operation immediately prior to it (using getUpdateOutputSize/getOutputSize as appropriate) in both streams.

Tests now run over boundaries of various block/buffer sizes to try to expose issues (0, 64 bit block, 128 bit block, 1K, 2K, 4K).

1 files changed, 26 insertions, 0 deletions

diff --git a/core/src/main/java/org/bouncycastle/crypto/io/CipherIOException.java b/core/src/main/java/org/bouncycastle/crypto/io/CipherIOException.java
new file mode 100644
index 00000000..beeb60bc
--- /dev/null
+++ b/core/src/main/java/org/bouncycastle/crypto/io/CipherIOException.java
@@ -0,0 +1,26 @@
+package org.bouncycastle.crypto.io;
+
+import java.io.IOException;
+
+/**
+ * {@link IOException} wrapper around an exception indicating a problem with the use of a cipher.
+ */
+public class CipherIOException
+    extends IOException
+{
+    private static final long serialVersionUID = 1L;
+
+    private final Throwable cause;
+
+    public CipherIOException(String message, Throwable cause)
+    {
+        super(message);
+
+        this.cause = cause;
+    }
+
+    public Throwable getCause()
+    {
+        return cause;
+    }
+}
\ No newline at end of file