Skip to content

Files and storage

Conclave enclaves can do file I/O but it's not mapped to the host filesystem. Instead, all filesystem activity is mapped to an in-memory file system. The goal of this support is to enable usage of libraries and applications that need to load data files from disk, to provide a simple 'scratch space' for files if you find them easier to work with, and to enable you to use file APIs to prepare data before it's encrypted for storage.

How to store a file on the real disk

If you have a byte stream you'd like to persist on the host, you must do this explicitly by using the mail-to-self pattern.

Create a mail using the regular APIs as covered in the tutorial and send it with a routing hint like self or any other string that your host code will recognise as meaning "this should be stored". At startup the host can read the mails it saved on behalf of the enclave and deliver them. Your enclave can then deserialize and save the contents to an in-memory hashmap, or if you'd rather do the decoding later just stash the bytes in the in-memory file system. If the enclave acknowledges a mail, that tells the host to delete it.

It's up to the host how to organise the actual files on disk. By keeping this logic outside the enclave the host/developers can change how storage works later, for instance, switching from local disk to S3 compatible blob storage, or using a database, or splitting files over multiple types of disk, doing replication etc. The attested logic won't change and no extra audit work will be generated.

Why not direct mapping to the host?

Some enclave platforms map file IO directly to the host file system. Conclave doesn't do this for these reasons:

  1. Data given to the host must be protected in various ways. It must be encrypted, authenticated and made robust against security patches that rotate keys. The enclave may want to verify it's been given the most recent stored data by the host, to block rewind attacks. Data may need to be padded to stop the host guessing what's inside it by looking at the message size. In more advanced scenarios you may want to store data in such a way that any enclave in a cluster can read it. All these features are provided already by the mail API, so it makes sense to re-use it.
  2. Sophisticated side channel attacks exist when the host can observe the pattern of accesses to a data store. With regular file IO, software outside your control (e.g. from libraries) may do reads, writes and seeks in patterns that leak important parts of the data. By receiving mails at startup and decrypting them into the in-memory file system you avoid exposing your file IO patterns to the host by exploiting the platform's built in protections for hiding memory access patterns.
  3. A common need for files is simply libraries that require them for configuration. In this case it makes sense to ship a pre-baked unencrypted filesystem as part of the enclave source code itself, where it can be audited and incorporated into the remote attestation.
  4. Log files are a common output of enclaves, but it doesn't make sense to seal them to the enclave itself. That would just slow things down for no reason. It makes more sense to buffer logs in memory and then from time to time either emit them straight to the host unencrypted, or send them somewhere else like the client that triggered the action that generated the logs, to a trusted administrator client, or even another enclave. Conclave's local calls and mail functionality make this easy.

How to use

A good way to use the file system is via the java.nio.file API. Here are some common tasks this API allows for:

  1. Getting InputStream and OutputStream objects for reading and writing as streams.
  2. Creating files, directories, moving files, renaming them etc.
  3. Getting a FileChannel that supports random access via the Files.newByteChannel API.
  4. Watching the filesystem for changes.
  5. Easily reading/writing text files from/to arrays of lines.
  6. Symbolic links are supported.
  7. Walking file trees are supported.
  8. Copying file trees in and out of zip files.

The JavaDocs provide detailed explanations of how to do all these things, but if you're not sure where to start take a look at the Files class which has many convenient utility methods.

It'll usually be easier to store structured data using conventional data structures, but if you want to you can also store data in these files. Because it's all in-memory, it'll be fast.


Currently the old API isn't fully wired up. In future releases all Java file APIs will be connected to the in-memory file system, along with native C-level stubs for POSIX APIs so native code can also benefit.