You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
(Only compile tested locally)
An issue was raised in that a current Red Hat internal build system
was performing a signature just on the per-architecture manifest,
but the current proxy code is expecting a signature on the manifest
list.
To quote Miloslav from that bug:
> Podman signs both the per-platform items and the top level,
> and enforces signatures only on per-platform items. cosign,
> by default, signs only the top level (and has an option to
> sign everything), I’m not sure which one it enforces.
> I don’t immediately recall other platforms.
We believe the current proxy code is secure since
we always require signatures (if configured) on the manifest
list, and the manifest list covers the individual manifests
via sha256 digest.
However, we want to support signatures only being present
on the per-arch manifest too in order to be flexible.
Yet, we can't hard switch to requiring signatures on the
per-arch manifests as that would immediately break anyone
relying on the current behavior of validating the toplevel.
Change the logic here to:
- Verify signature on the manifest list, and cache the error (if any)
- Fetch the manifest
- Verify signature on the manifest
- Allow if *either* signature was accepted; conversely, only error
if signature validation failed on *both* the manifest list and
manifest
This also switches things to cache the manifest upfront instead
of doing it lazily on `GetManifest/GetConfig`; in practice
callers were always immediately requesting those anyways.
The use case of just fetching blobs exists (e.g. to resume
an interrupted fetch), but is relatively obscure and
in general I think it's good to re-verify signatures on
each operation.
Signed-off-by: Colin Walters <[email protected]>
0 commit comments