I need to transport multiple very large files over an unstable and untrusted network, and the file contents are outputted as a data stream. I wanted to use OpenSSL for streaming authenticated encryption, but they purposefully don’t support that and are preachy about it.
Well, it turns out that XZ has checksumming built-in! It even has different algorithms (CRC32, CRC64, and SHA256). It’s part of the same file, within/before the encryption, and automatically verified by the decompression tool. I’m already using XZ for compression before encryption, so this is just super convenient and useful. Also, it seems like XZ supports threaded decompression now, when it didn’t before. Thanks XZ devs!
untrusted network
What stops the network operator from modifying the data and the checksum? Do you transfer the checksum out of band?
Wouldn’t be possible afaict, the encryption masks the xz archive which contains the checksum metadata. If the data is modified, decryption & extraction will simply fail.
The data will be undecipherable to a mitm anyway since it’s encrypted, the only real risk imo would be someone modifying the encrypted data in transit to attempt a zero day targeting the decryption process… chances of which are probably really low lol
I’m not a cryptographer (so maybe this is wrong), but my understanding is that although it’s possible to modify the cipher text, how those changes modify the plaintext are very difficult (or impossible) to predict. That can still be an attack vector if the attacker knows the structure of the plaintext (or just want to break something), but since the checksum is also encrypted, the chances that both the original file and checksum could be kept consistent after cipher text modification is basically zero.
A checksum and a digital signature aren’t the same thing. If you have a data block and a checksum of the data block, the data block can be modified and a new checksum can be computed to reflect the modifications. Instead of a checksum would be a digital signature using an asymmetric key. The data block would be modified but the signature of that block can’t be recomputed without the key used to sign it, which is not part of the transfer.
The data block would be modified but the signature of that block can’t be recomputed without the key used to sign it
Isn’t that also true of an encrypted checksum, though? For some plaintext block q there is a checksum r, but the attacker can only see and modify the encrypted q (Q) and encrypted r (R). How any change to Q would modify q (and R to r) can’t be known without knowing the encryption key, but the attacker would need to know that in order to keep q and r consistent.
