src/storage/blobfs/README.md - fuchsia - Git at Google

 # Blobfs: A content-addressable filesystem for Fuchsia

 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](/docs/concepts/filesystems/minfs.md) handles this case).

 The structure of blobfs is described in the [main
 documentation](/docs/concepts/filesystems/blobfs.md). This page documents some internal details.

 ## Version notes

 Blobfs follows the [storage versioning scheme](/src/storage/docs/versioning.md). 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:

   * 8.1, 9.1: The initial minor version.
   * 8.2, 9.2: Introduced a backup superblock when running under FVM.
   * 8.3, 9.3: Removed support for ZSTD seekable compression.

 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".
	# Blobfs: A content-addressable filesystem for Fuchsia

	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](/docs/concepts/filesystems/minfs.md) handles this case).

	The structure of blobfs is described in the [main
	documentation](/docs/concepts/filesystems/blobfs.md). This page documents some internal details.

	## Version notes

	Blobfs follows the [storage versioning scheme](/src/storage/docs/versioning.md). 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:

	* 8.1, 9.1: The initial minor version.
	* 8.2, 9.2: Introduced a backup superblock when running under FVM.
	* 8.3, 9.3: Removed support for ZSTD seekable compression.

	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".