Re: [PATCH proxmox-backup v3 18/30] sync: push: add helper for loading known chunks from previous snapshot

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 14.04.26 um 14:59 schrieb Christian Ebner:
> diff --git a/src/server/push.rs b/src/server/push.rs
> @@ -808,6 +808,41 @@ pub(crate) async fn push_group(
>
> +async fn load_previous_snapshot_known_chunks(
> +    ...
> +) {
> +    if let Some(manifest) = upload_options.previous_manifest.as_ref() {
> +        if let Some((_id, crypt_config)) = &params.crypt_config {
> +            if let Ok(Some(fingerprint)) = manifest.fingerprint() {
> +                if *fingerprint.bytes() == crypt_config.fingerprint() {
> +                    // needs encryption during push, cannot reuse chunks from previous manifest
> +                    return;
> +                }
> +            }
> +        }

I think this has an issue when transitioning from unencrypted to encrypted
push. The guard returns early only when the previous manifest has a
matching fingerprint, but when params.crypt_config is Some and the
previous manifest has no fingerprint (was unencrypted), the `if let
Ok(Some(...))` doesn't match and we fall through to loading plaintext
chunk digests into known_chunks, if I'm not mistaken.

In push_index, those plaintext digests won't be in encrypted_mapping (the
chunk was never read/encrypted), so the else branch uses the plaintext
digest as-is for MergedChunkInfo::Known. The remote target then ends up
referencing old unencrypted chunks while the manifest says it's encrypted.

Simplest fix might be to just skip known chunk loading entirely when
crypt_config is set - the optimization can't really help when we need to
re-encrypt anyway, since chunk digests change with encryption.