no API to surface buffered record fragments

[jmayclin]

Problem:

There is no way to determine if s2n-tls has ingested a record fragment. This makes it difficult to work with connection serialization.

Solution:

At the very least, an API to determine if there is any buffered, ciphertext data would be useful.
s2n_peek returns the available plaintext to be read.
s2n_buffered_peek has very complicated behavior, but basically doesn't solve this.

[jmayclin]

Closely related, I think it is currently impossible to correctly use the serialization feature with recv_buffering enabled on the connection.

To make connection serialization compatible with recv_buffering, we would need to offer an s2n_regurgitate API or similar, which would be used to give unprocessed ciphertext back to the application.

The much neater solution would be to get out of the buffering/IO game entirely.

[jmayclin]

I'm growing increasingly convinced that we should change s2n_buffered_peek to be the solution to this. This would have this nice solution of

s2n_peek + s2n_buffered_peek would always return the exact amount of data buffered in the connection. Basically

• s2n_peek -> s2n_processed_record_peek
• s2n_buffered_peek -> s2n_unprocessed_record_peak