t2sz

It compresses a file into a seekable zstd splitting the file into multiple frames.

If the file is a tar archive it compresses each file in the archive into an independent frame, hence the name: tar 2 seekable zstd.

It operates in two modes. Tar archive mode and raw mode.

By default it runs in tar archive mode for files ending with .tar, unless -r is specified.

For all other files it runs in raw mode.

In tar archive mode it compresses the archive keeping each file in a different frame, unless -s or -S is used.

This allows fast seeking and extraction of a single file without decompressing the whole archive.

When -s SIZE is used in tar mode, if the size of the file being compressed into a block is less than SIZE then another one will be added in the same block, and so on until the sum of the sizes of all files packed together is at least SIZE. A file will be never split as SIZE is just a minimum value.

When -s SIZE is used in raw mode then it defines exactly the input block size and bigger inputs will be split in blocks of this size accordingly. If there isn't enough input data the last block will be smaller.

When -S SIZE is used, files bigger than SIZE will be split in blocks of SIZE length. It is available only in tar mode and ignored in raw mode.

The compressed archive can be decompressed with any Zstandard tool, including zstd.

To take advantage of seeking see the following projects:

• C/C++ library: libzstd-seek
• Python library: indexed_zstd
• FUSE mount: ratarmount

Build

Linux

You'll need libzstd-dev

sudo apt install libzstd-dev

git clone https://github.com/martinellimarco/t2sz
mkdir t2sz/build
cd t2sz/build
cmake .. -DCMAKE_BUILD_TYPE="Release"
make

Install with

sudo make install

Or if you want a debian package you can run

cpack

then install it with

sudo dpkg -i t2sz*.deb

macOS

You'll need zstd from Homebrew:

brew install zstd

git clone https://github.com/martinellimarco/t2sz
mkdir t2sz/build
cd t2sz/build
cmake .. -DCMAKE_BUILD_TYPE="Release"
make

Install with

sudo make install

Windows (cross-compilation)

On macOS or Linux, cross-compile static Windows executables using Docker:

docker run --rm -v "$(pwd)":/src ubuntu:24.04 bash /src/windows/win-cross-build.sh both

This produces t2sz.exe for both amd64 and arm64 in build-windows-{amd64,arm64}-release/. Pre-built executables are also available on the release page.

Usage

Usage: t2sz [OPTIONS...] [TAR ARCHIVE | -]

Use '-' as the input filename to read from standard input.

Examples:
        t2sz -s 10M any.file                        Compress any.file to any.file.zst, each input block will be of 10M
        t2sz archive.tar                            Compress archive.tar to archive.tar.zst
        t2sz -o output.tar.zst archive.tar          Compress archive.tar to output.tar.zst
        t2sz -o - archive.tar                       Compress archive.tar to standard output
        t2sz -r -o out.zst -                        Compress stdin (raw mode) to out.zst
        t2sz -o out.tar.zst -                       Compress tar from stdin to out.tar.zst
        t2sz -r -o - -                              Compress stdin to stdout (raw mode)

Options:
        -l [1..22]         Set compression level, from 1 (lower) to 22 (highest). Default is 3.
        -o FILENAME        Output file name. Use '-' to write to standard output.
                           When reading from stdin ('-') and -o is omitted, output defaults to stdout.
        -s SIZE            In raw mode: the exact size of each input block, except the last one.
                           In tar mode: the minimum size of an input block, in bytes.
                                        A block is composed by one or more whole files.
                                        A file is never split unless -S is used.
                                        If not specified one block will contain exactly one file, no matter the file size.
                                        Each block is compressed to a zstd frame but if the archive has a lot of small files
                                        having a file per block doesn't compress very well. With this you can set a trade off.
                           The greater is SIZE the smaller will be the archive at the expense of the seek speed.
                           SIZE may be followed by the following multiplicative suffixes:
                               k/K/KiB = 1024
                               M/MiB = $1024^{2}$
                               G/GiB = $1024^{3}$
                               kB/KB = 1000
                               MB = $1000^{2}$
                               GB = $1000^{3}$
        -S SIZE            In raw mode: it is ignored.
                           In tar mode: the maximum size of an input block, in bytes.
                           Unlike -s this option may split big files in smaller chunks.
                           Remember that each block is compressed independently and a small value here will result in a bigger archive.
                           -S can be used together with -s but MUST be greater or equal to its value.
                           If -S and -s are equal the input block will be of exactly that size, if there is enough input data.
                           Like -s SIZE may be followed by one of the multiplicative suffixes described above.
        -T [1..N]          Number of thread to spawn. It improves compression speed but cost more memory. Default is single thread.
                           It requires libzstd >= 1.5.0 or an older version compiler with ZSTD_MULTITHREAD.
                           If `-s` or `-S` are too small it is possible that a lower number of threads will be used.
        -r                 Raw mode or non-tar mode. Treat tar archives as regular files, without any special handling.
        -j                 Do not generate a seek table.
        -v                 Verbose. List the elements in the tar archive and their size.
        -f                 Overwrite output without prompting.
        -h                 Print this help.
        -V                 Print the version.

License

See LICENSE

Release

Debian-based

Download the latest stable source code or .deb from the release page. This is the recommended version.

Arch-based

Check out t2sz on AUR or pamac build t2sz

Homebrew

On Linux and macOS you can install with Homebrew:

brew install t2sz

Snap

For your convenience you can install the latest release from the snap store but beware that it is distributed in strict mode and it can access only your home directory by default.

You can add access to removable devices such as those stored in /media with sudo snap connect t2sz:removable-media.

If you want to give it access to every file you can install it with --devmode.

Testing

The project includes a comprehensive test suite covering round-trip correctness (compress → decompress → SHA-256 verify), memory safety (AddressSanitizer + UBSanitizer), and code coverage (LLVM).

See TESTING.md for prerequisites, build configurations, and the full list of test cases.

Fuzzing

The project includes libFuzzer harnesses for fuzz-testing the tar header parser and the streaming compression paths.

See FUZZING.md for build instructions and usage.