Blobfs is a filesystem for Fuchsia that stores all files according to their hashes. This is used for all binaries and other program-related data for Fuchsia. It is not used for storing dynamic data and it has no hierarchy (minfs handles this case).
The structure of blobfs is described in the main documentation. This page documents some internal details.
Blobfs follows the storage versioning scheme. However, there were some versions before this was adopted that do not follow this scheme.
In major version 8, blobfs used “padded” Merkle trees. In this format, the Merkle tree was padded out to the nearest block boundary and immediately preceeded the blob data on the block device.
In major version 9, blobfs switched to “compact” Merkle trees where the Merkle tree data immediately follows the blob data and neither the offset nor the size is block-padded. This saves space. This change was made when the minor version was treated as a monotonically increasing “revision” and was not reset. The format was controlled by a build flag so both major version 8 and major version 9 were in use at the same time. No migration from version 8 to version 9 was done.
Across versions 8 and 9 are the following minor versions:
In major version 10, a flag was added to the superblock to indicate if the blobs will be written out in the “padded” or “compact” Merkle tree format, and a flag is added to each inode to indicate whether it is currently in the “padded” or “compact” formats. Both major version 8 and major version 9 were migrated directly to major version 10 setting their superblock flag to preserve the format being written prior to the migration. This was done to enable the migration from the “padded” format to the “compact” format without changing the format generated by an existing device. A later release is expected to change the written format to always be “compact”.