|
- Handling OnePassSig packets in an unbuffered manner introduces a
layering violation: when we encounter a OnePassSig packet, we want
to push a HashedReader on the BufferedReader stack, but that is
popped when the readers associated with the OnePassSig are popped!
Thus, we need to introduce a bit of ugliness (OnePassSig::parse
needs to pop its readers and then push the HashedReader at the
high level). This is unfortunate, but it appears to be necessary.
- Hashing literal data packets is ugly! Only the content of a
literal data packet is hashed; neither the packet's headers, the
packet's meta-data nor the length information is included in the
hash. This, in particular, adds some ugliness to the
BufferedReaderPartialBodyFilter implementation: it needs to
disable hashing when decoding a literal data packet.
- This implementation has several limitations:
- We only handle a single level of Signature nesting. That is, we
don't support things like having two OnePassSig packets, both of
which have their 'last' bit set (meaning that the outer
signature is over the inner Signature and the content, not just
the content). If the parser encounters such a message, it
silently ignores the nesting. In practice, this functionality
is rarely used: GnuPG neither produces such messages nor does it
correctly handle them.
- Each OnePassSig packet pushes another HashedReader on the
BufferedReader stack. This can cause a stack overflow if there
are too many OnePassSig packets. Instead, there should only be
a single HashedReader per level of nesting, and the amount of
nesting must be limited.
- If there are multiple OnePassSig packets at a nesting level
using the same hash algorithm, we don't reuse hashes, because
Nettle doesn't currently support cloning hashes.
|
