Piotr Karwasz created COLLECTIONS-903:
-----------------------------------------
Summary: Validate the capacity read in AbstractHashedMap and
AbstractReferenceMap doReadObject() as the constructors do
Key: COLLECTIONS-903
URL: https://issues.apache.org/jira/browse/COLLECTIONS-903
Project: Commons Collections
Issue Type: Improvement
Components: Map
Reporter: Piotr Karwasz
The constructors of {{AbstractHashedMap}} normalize the requested capacity
through {{calculateNewCapacity()}} before allocating the table: it is rounded
up to a power of two and capped at {{MAXIMUM_CAPACITY}}.
Deserialization skips that step. Both {{doReadObject()}} methods allocate the
table directly from the value read from the stream:
{code:java}
// AbstractHashedMap.doReadObject()
final int capacity = in.readInt();
final int size = in.readInt();
init();
threshold = calculateThreshold(capacity, loadFactor);
data = new HashEntry[capacity];
{code}
{code:java}
// AbstractReferenceMap.doReadObject()
final int capacity = in.readInt();
init();
data = new HashEntry[capacity];
{code}
As a result, a stream can produce tables that the constructors would never
create. These include tables whose length is not a power of two (the index
computation assumes one), tables larger than {{MAXIMUM_CAPACITY}}, and a
{{NegativeArraySizeException}} instead of an {{InvalidObjectException}} for
negative values.
Following Oracle's Secure Coding Guidelines for Java SE, Guideline 8-3 ("View
deserialization the same as object construction"), {{doReadObject()}} should
apply the same checks as the constructors:
{code:java}
final int capacity = calculateNewCapacity(in.readInt());
{code}
In {{AbstractHashedMap}}, the stream also carries the number of entries, and
the table is rebuilt with {{put()}} anyway. The initial allocation can
therefore also be limited to what {{size}} and {{loadFactor}} require, with the
table growing as entries are actually read.
This brings deserialization in line with the constructors. It does not make
deserializing untrusted data supported. See
https://commons.apache.org/threat-model-0.1.0.html.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)