tahoestaticfs 0.1.1

Fuse filesystem on Tahoe-LAFS

Tahoestaticfs is a Fuse filesystem that enables read and write access to files stored on a Tahoe-LAFS grid.

Tahoestaticfs is designed for static, unchanging data, and caches file and directory metadata aggressively. Optionally, also file data can be cached.

Compared to mounting Tahoe-LAFS directories via sshfs (as of Tahoe-LAFS 1.9.2), Tahoestaticfs achieves better latency. This is true also for accessing data not in cache (e.g. ls -laR /mnt/tahoe with cold cache), due to the more aggressive directory metadata caching. Moreover, tahoestaticfs also supports random access reads to stored files, without requiring to download the full files.

Usage

tahoestaticfs [options] [mountpoint]

Tahoe-LAFS directory mounted as a filesystem, with local
caching. Cached data is encrypted with a key derived from the
directory capability mounted.

Dircap of the root directory is read from stdin on startup. In scripts, do::

    awk '/^root:/ {print $2}' < ~/.tahoe/private/aliases \\
        | tahoestaticfs ...

Cache can be invalidated by `touch <mountpoint>/.tahoestaticfs-invalidate`,
or by removing files in the cache directory.

Options:
    --version              show program's version number and exit
    -h, --help             show this help message and exit
    -o opt,[opt...]        mount options
    -c CACHE, --cache=CACHE
                           Cache directory
    -u NODE_URL, --node-url=NODE_URL
                           Tahoe gateway node URL
    -D, --cache-data       Cache also file data
    -S CACHE_SIZE, --cache-size=CACHE_SIZE
                           Target cache size
    -w WRITE_LIFETIME, --write-cache-lifetime=WRITE_LIFETIME
                           Cache lifetime for write operations (seconds).
                           Default: 10 sec
    -r READ_LIFETIME, --read-cache-lifetime=READ_LIFETIME
                           Cache lifetime for read operations (seconds).
                           Default: 10 sec

For example:

awk '/^root:/ {print $2}' < ~/.tahoe/private/aliases \\
    tahoestaticfs -c /var/cache/tahoefscache -D -S 5G -u http://127.0.0.1:8090 /mnt/tahoestatic

Warning

Do not do this:

echo URI:DIR2:... | tahoestaticfs

That makes the root capability visible to everyone. Instead, store the root capability in a file with appropriate permissions, for example reading it from your Tahoe-LAFS aliases file as shown above.

Caching and concurrency

Cached data and metadata becomes invalidated if it is older than the corresponding cache lifetimes. The cache lifetimes can be specified separately for metadata/data read and write operations.

The cache will stay in a consistent state with the Tahoe grid if modifications to the data are only performed via a single mounted filesystem, even if the cache lifetimes are infinite.

If modifications are done directly to files on the grid e.g. from other machines accessing the Tahoe storage, Tahoestaticfs filesystem will reflect the state of the files in Tahoe storage only after waiting one cache lifetime after the last modification.

Note that this is not an absolute guarantee, as Tahoe-LAFS itself does not guarantee full concurrency.

Nonzero cache lifetimes are in general necessary to obtain filesystem metadata performance necessary for typical workloads.

The cache can be invalidated manually, via touch <mountpoint>/.tahoestaticfs-invalidate. To invalidate only the cache of a subtree, do touch <mountpoint>/<somedir>/.tahoestaticfs-invalidate.

Encryption

Tahoestaticfs encrypts cached data and metadata retrieved from network before storing it on disk.

The purpose of the encryption is to hinder an attacker who, (i) has read-only access to the cache files, and (ii) attempts to determine either the root dircap or file data or metadata.

Each file is divided to 131072-byte chunks, and encryption is done using AES-CBC-256 separately for each chunk, to enable random access. The 16-byte IV for each chunk is random, and changed every time the chunk is written on disk.

A 256-bit master encryption key is derived from the rootcap mounted, combined with a randomly chosen 32-byte salt via PBKDF2. The iteration count is determined so that it takes around one second on the system in question, but is at least 10000. The salt is stored on-disk as-is.

The AES encryption keys are file-specific, and obtained via:

prk = HKDF-SHA256-Extract(salt2, master-key)
data_key | fn_key = HKDF-SHA256-Expand(prk, pathname, 96)

The salt2 is a second 32-byte randomly generated salt stored as-is on-disk. The 32-byte data_key is used as the AES-CBC encryption key. The 64-byte fn_key is used to generate filenames via HMAC-SHA512(fn_key, pathname).