fix length decoding

the same issues as with decoding integers happen with the NIST521p curve
as it's large enough that the length encoding of some fields needs
to use multi-byte encoding

backport of a655d6f

Author: tomato42

ecdsa/der.py

@@ -148,13 +148,17 @@ def encode_length(l):
     return int2byte(0x80|llen) + s
 
 def read_length(string):
+    if not string:
+        raise UnexpectedDER("Empty string can't encode valid length value")
     num = string[0] if isinstance(string[0], integer_types) else ord(string[0])
     if not (num & 0x80):
         # short form
         return (num & 0x7f), 1
     # else long-form: b0&0x7f is number of additional base256 length bytes,
     # big-endian
     llen = num & 0x7f
+    if not llen:
+        raise UnexpectedDER("Invalid length encoding, length byte is 0")
     if llen > len(string)-1:
         raise UnexpectedDER("ran out of length bytes")
     return int(binascii.hexlify(string[1:1+llen]), 16), 1+llen

ecdsa/test_malformed_sigs.py

@@ -17,7 +17,7 @@
 # few thousand slow test cases; execute the most interesting only
 
 #for curve in curves:
-for curve in [NIST256p]:
+for curve in [NIST521p]:
     #for hash_alg in ["md5", "sha1", "sha224", "sha256", "sha384", "sha512"]:
     for hash_alg in ["sha256"]:
         key = SigningKey.generate(curve)