Philip Helger created WSS-735:
---------------------------------
Summary: Allow selecting the JCE provider for SwA attachment
decryption, to enable streaming of large attachments
Key: WSS-735
URL: https://issues.apache.org/jira/browse/WSS-735
Project: WSS4J
Issue Type: Bug
Components: WSS4J Core
Affects Versions: 4.0.2
Reporter: Philip Helger
Assignee: Colm O hEigeartaigh
Decrypting an AES-GCM SwA attachment needs heap proportional to the attachment
size with the default provider, even though
AttachmentUtils.setupAttachmentDecryptionStream is stream based. SunJCE
deliberately buffers all ciphertext until the tag was verified
(GaloisCounterMode.GCMDecrypt.doUpdate: "store internally until doFinal"), so
CipherInputStream returns no data before the end of the stream. For use cases
that require AES-GCM, such as the Peppol AS4 profile, this limits the
attachment size by the available heap.
A provider that decrypts incrementally avoids this, but it cannot be selected
for attachments, because the cipher is created without a provider:
* EncryptionUtils.decryptAttachment: Cipher.getInstance(jceAlgorithm)
* EncryptionUtils.decryptXopAttachment: Cipher.getInstance(jceAlgorithm)
* Encryptor (attachment encryption): Cipher.getInstance(jceAlgorithm)
Key transport already handles this through KeyUtils.getCipherInstance, which
falls back to JCEMapper.getProviderId(). The only workaround is the JVM-wide
jdk.security.provider.preferred security property, which also affects all other
users of AES/GCM in the JVM, possibly including TLS.
Proposal:
# Use KeyUtils.getCipherInstance at the three places above, so that
JCEMapper.setProviderId applies to them as well.
# Add an API to choose the provider per request, as JCEMapper.setProviderId is
global and also changes key transport and other cipher use in the same JVM. For
example, a new property on RequestData such as
setAttachmentDecryptionProvider(String) that is handed through to
EncryptionUtils. The exact shape is up to you.
Security note: An incremental provider releases plaintext before the tag was
verified, and a tag mismatch only shows up as an IOException at the end of the
stream. Callers must read to the end and must not use the data before that.
This should be mentioned in the Javadoc of the new option.
I'm happy to help with testing.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]