[
https://issues.apache.org/jira/browse/CAMEL-24440?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18107256#comment-18107256
]
Andrea Cosentino commented on CAMEL-24440:
------------------------------------------
Three of the four sub-issues are fixed; the fourth is format-breaking and is
left out deliberately.
*Fixed.*
1. *Per-message initialization vector when inlining.* {{marshal}} threw
{{Inlining cannot be performed, as no initialization vector was specified}}
whenever {{shouldInlineInitializationVector}} was set without a static vector -
which is what pushed routes into configuring one and reusing it for every
message. A fresh vector is now generated per message when none is supplied.
Format-compatible: the vector is written into the message, so readers take it
from the stream unchanged. An explicitly supplied vector (configuration or the
{{CamelCryptoInitVector}} header) is still honoured.
2. *Uniform authentication failure.* Verified on JDK 21 that
{{CipherInputStream}} does surface bad padding, as {{IOException}} wrapping
{{BadPaddingException}} - I checked rather than assuming, since older JDKs
swallowed it. The two outcomes were plainly distinguishable, which the red test
shows directly:
{noformat}
expected: <Expected mac did not match actual mac
expected:5cc61463cee1bb52d9111feb9a3e25c33862fa10
actual:bc9749c9a62fb11c7cd38ac55e1b24617da71ff6>
but was: <Given final block not properly padded. Such issues can arise if a bad
key is used during decryption.>
{noformat}
Both now report {{Message authentication failed}}.
3. *MAC values out of the message.* Visible in the output above - the failure
text carried both MACs in hex, and the computed one is {{HMAC_k}} over the
plaintext just produced, so it hands the caller a value they could not
otherwise compute. Removed. ({{MessageDigest.isEqual}} was already in place for
the comparison itself.)
Also bounded the inlined vector length, which is read from the message and used
to size an allocation directly - {{new byte[ivLength]}} with {{ivLength}} up to
{{2^31}}. That was reported separately in the same method, and it was a three
line fix in code I was already changing.
*Not fixed: separating the HMAC key from the cipher key.*
{{HMACAccumulator}} derives the MAC key from the same {{key.getEncoded()}} as
the cipher. Deriving them separately - HKDF or equivalent - changes the MAC
written into the message, so data produced by any earlier version could no
longer be read, and data produced by a fixed version could not be read by an
unfixed one. That needs either an opt-in flag or a version marker in the output
format, and it is the one change here that cannot be backported. Worth its own
issue rather than being folded in.
Also unchanged: the composition is still MAC-then-encrypt, and {{unmarshal}}
still streams plaintext to the output before {{validate()}} runs. Validating
first means buffering the whole plaintext, which would break streaming for
large payloads - a real trade-off, and a separate decision from the uniform
error reporting that removes the distinguisher.
Module suite 75 green; the three new behaviours are covered and verified red
without the fix.
_Claude Code on behalf of oscerd_
> camel-crypto - CryptoDataFormat validates the MAC after decrypting, reuses
> the cipher key for the HMAC, and has no per-message IV
> ---------------------------------------------------------------------------------------------------------------------------------
>
> Key: CAMEL-24440
> URL: https://issues.apache.org/jira/browse/CAMEL-24440
> Project: Camel
> Issue Type: Bug
> Components: camel-crypto
> Reporter: Andrea Cosentino
> Assignee: Andrea Cosentino
> Priority: Major
> Fix For: 4.23.0
>
>
> Four related issues in {{CryptoDataFormat}}:
> 1. *MAC checked after decryption.* {{unmarshal()}} runs the body through
> {{CipherInputStream}} to completion before {{HMACAccumulator.validate()}} is
> called. Tampered input therefore surfaces as two distinguishable failures -
> {{IOException}} (bad padding, from the cipher) versus
> {{IllegalStateException}} (bad MAC) - rather than one uniform authentication
> failure.
> 2. *Key reuse.* {{HMACAccumulator}} is constructed with the HMAC key derived
> from the same {{key.getEncoded()}} used for the cipher. The same key material
> is used for both confidentiality and integrity.
> 3. *MAC values in the failure message.* {{HMACAccumulator.validate()}} builds
> its exception message with both the expected and the computed MAC in hex.
> 4. *No per-message IV.* {{getInitializationVector()}} returns the header
> value or the statically configured {{initVector}}; nothing ever generates a
> fresh one. {{marshal()}} even throws when {{inlineInitVector=true}} and no
> static IV was configured, which steers users towards a fixed IV for exactly
> the feature that exists to carry a per-message one. The javadoc notes the IV
> need not be secret but never that it must be unique.
> Proposal: derive separate cipher and MAC keys from the configured key (HKDF
> or equivalent); validate the MAC before releasing plaintext, or move to an
> AEAD mode; report a single uniform authentication failure without embedding
> MAC values; and generate a per-message {{SecureRandom}} IV when
> {{inlineInitVector}} is enabled instead of requiring a static one. Changing
> key derivation or IV handling breaks compatibility with data produced by
> earlier versions, so it needs an upgrade-guide entry and probably a version
> marker in the output format.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
