A zstd frame begins with magic 28 B5 2F FD and a frame header (window descriptor, dictionary ID, content size, checksum flag), followed by blocks (raw, RLE or compressed) and an optional XXH64 checksum. Compressed blocks use Huffman-coded literals and FSE-coded sequences. Skippable frames allow metadata.
Where it is used
Linux packages and kernel images, filesystem compression (Btrfs, ZFS, SquashFS).
Backups and databases.
History
Developed by Yann Collet at Facebook (2015-2016); standardized in RFC 8878.
Status today
Rapidly growing adoption; actively maintained.
Structures read by the implementation
Offsets are in bytes from the start of the buffer named in each heading. Field names are taken from the implementation and are not an official specification. Fixed-width integers are shown as u/i (unsigned/signed) followed by bit width and byte order.
Structures and constants were extracted from: Carving/Plugins/Zstd/Zstd.cs, IO/Compression/ZstdDecoder.cs, Modules/FileSystems/ZstdArchiveModule.cs. Prose sections are a summary from public sources; verify details before relying on them.