I would like to transparently provide a Flatpak app with a fully sandboxed ~/ directory located at an arbitrary location of my choosing.

  • That is, I want to sandbox the entirety of ~/ such that writing to arbitrary subdirectories works as expected, and map it to a location on an external disk (mounted at /mnt/<disk-id>) instead of ~/.var/app/<app-id>.
  • Ideally I'd like to relocate all of the per-user data for a particular app, but I'm trying to limit the scope of this question.
  • I know that I can grant access to arbitrary locations via --filesystem=<path> and that I can map individual subdirectories under user ~/ to the corresponding sandbox directory via --persist=<path>. However, neither of these options would appear to satisfy my use case.

The Flatpak docs (sandbox permissions and flatpak-override) seem to indicate that the sandbox uses a tempfs and only bind mounts specific directories into it. In particular, the flatpak-override --persist docs seem to imply that individual subdirectories must be mapped on a case by case basis in order for data to be persisted. Contradicting this, the sandbox overview seems to be saying that ~/.var/app/<app-id> is simply bind mounted to sandbox ~/. The latter appears to be what actually happens. When I used a sandboxed app to create a directory within ~/ that had not been mapped into the sandbox, the corresponding directory was created under ~/.var/app/<app-id> and persisted across app restarts.

Given the apparent behavior, would it be safe to simply bind mount the desired location to ~/.var/app/<app-id>? I'm hesitant to blindly attempt this because I know that Flatpak is engaging in a great deal of file system shenanigans under the hood in order to create a robust sandbox. In particular, it pivot_roots the sandbox filesystem, would be bind mounting the location I had bind mounted to (do I need an rbind here?), could be performing further bind mounts in subdirectories (will this play well with my bind mount?), and is likely taking other precautions (now or in the future) which I'm worried might break any indirections done by me in strange and creative ways.


You must log in to answer this question.