FUSE driver for HFS+ filesystems, based on NetBSD's kernel driver with modifications.
hfsfuse embeds and extends NetBSD's HFS+ kernel driver into a portable library for use with FUSE and other userspace tools. hfsfuse was created for use on FreeBSD and other Unix-like systems that lack a native HFS+ driver, but can also be used on Linux and macOS as an alternative to their kernel drivers.
hfsfuse also includes a standalone tool, hfsdump, to inspect the contents of an HFS+ volume without FUSE.
This driver is read-only and cannot write to or alter the target filesystem.
Supported
- Journaled and non-journaled HFS+
- Unicode normalization for pathnames via utf8proc
- Hard links, including directory hard links (i.e. Time Machine backups)
- Resource fork, Finder info, and creation/backup time access via extended attributes
- birthtime (with compatible FUSE)
- User-defined extended attributes
- HFS+ compression with zlib and lzfse
Not supported
- HFS without the "+", aka "Mac OS Standard" volumes. For these, try hfsutils.
- Writing
With the FUSE headers and library for your platform installed, running make install (gmake on *BSD) from the project root will build and install hfsfuse and hfsdump and should be sufficient for most use cases. See below for more details or skip to usage.
hfsfuse aims to be widely portable across Unix-like systems. Build requirements include GNU Make, a C11 compiler with a GCC-compatible frontend, and a POSIX-compatible shell and utilities.
hfsfuse's supporting libraries and standalone hfsdump tool require only a POSIX-2008 compatible libc, and can also be built natively on Windows with either Mingw-w64 or msys2.
The FUSE driver requires a version 2 compatible FUSE library, and is known to work with the following implementations:
hfsfuse optionally uses these additional libraries to enable certain functionality:
- utf8proc for working with non-ASCII pathnames
- ublio for read caching, which may improve performance
- zlib and lzfse for reading files with HFS+ compression
utf8proc and ublio are both bundled with hfsfuse and built by default. hfsfuse can be configured to use already-installed versions of these if available, or may be built without them entirely if the respective functionality is not needed (see Configuring).
hfsfuse is configured by passing options directly to make, and separate configure and build steps are not needed. make showconfig can be used to print available make options and their current values.
For repeated builds using the same options, or to more easily edit config values, make config can optionally be used to generate a config.mak file which will be used by future invocations.
To configure hfsfuse's optional utf8proc and ublio dependencies, use WITH_DEP=(none/local/system). The default behavior with no arguments is to use the bundled versions of these and is the same as using
make WITH_UBILIO=local WITH_UTF8PROC=local
To ease portability, the Makefile will attempt to detect certain features of the host libc in an autoconf-like way, and creates a series of defines for these labeled HAVE_FEATURENAME. To override and skip checks for a given feature, these may be provided directly to make or overridden in config.mak.
The default make and make install targets build and install hfsfuse and hfsdump. hfsdump can also be built standalone with make hfsdump, in which case FUSE is not needed.
hfsfuse's supporting libraries can be built and installed independently using make lib and make install-lib. Applications can use these to read from HFS+ volumes by including hfsuser.h and linking with libhfsuser, libhfs, and ublio/utf8proc if configured.
Some version information is generated from the git repository. For distributions outside of revision control of tree builds, run make version within the repository first or provide your own version.h.
hfsfuse <opts> <device> <mountpoint>
Where <opts> are any series of arguments to be passed along to FUSE. Use hfsfuse -h for general options or hfsfuse -H for a list of all switches supported by FUSE.
hfsfuse-specific options are shown below
usage: hfsfuse [-hHv] [-o options] device mountpoint
general options:
-o opt,[opt...] mount options
-h --help this help
-H --fullhelp list all FUSE options
-v --version
HFS options:
--force force mount volumes with dirty journal
-o rsrc_only only mount the resource forks of files
-o noallow_other restrict filesystem access to mounting user
-o cache_size=N size of lookup cache (1024)
-o blksize=N set a custom read size/alignment in bytes
you should only set this if you are sure it is being misdetected
-o rsrc_ext=suffix special suffix for filenames which can be used to access their resource fork
or alternatively their data fork if mounted in rsrc_only mode
-o default_file_mode=N octal filesystem permissions for Mac OS Classic files (755)
-o default_dir_mode=N octal filesystem permissions for Mac OS Classic directories (777)
-o default_uid=N unix user ID for Mac OS Classic files (0)
-o default_gid=N unix group ID for Mac OS Classic files (0)
-o disable_symlinks treat symbolic links as regular files. may be used to view extended attributes
of these on systems that don't support symlink xattrs
-o noublio disable ublio read layer
-o ublio_items=N number of ublio cache entries, 0 for no caching (64)
-o ublio_grace=N reclaim cache entries only after N requests (32)
Note for Haiku users: under Haiku, FUSE applications cannot be invoked directly. Instead, make install will install hfsfuse as a userlandfs add-on, which can be loaed with:
/system/servers/userlandfs_server hfsfuse
Afterwards filesystems may be mounted like so:
mount -t userlandfs -o "hfsfuse <opts> <device>" <mountpoint>
hfsdump <device> <command> <node>
command may be either stat, read, or xattr: stat prints the record structure, while read copies the node's contents to standard out (or lists if node is a directory).
node is either an integer inode/CNID to lookup, or a full path from the root of the volume being inspected.
If the command and node are ommitted, hfsdump prints the volume header and exits.
/rsrc may be appended to the path of a read operation to dump the resource fork instead.
The xattr command lists extended attribute names for the given node. If an attribute name is provided following the node argument, its value will be printed to standard out.
hfsfuse exposes some nonstandard HFS+ attributes as extended attributes. These include:
- hfsfuse.record.date_created: The date created as an ISO-8601 timestamp. Identical to
st_birthtimeon macOS or FreeBSD. - hfsfuse.record.date_backedup: The backup time of a file as an ISO-8601 timestamp.
- com.apple.FinderInfo: The Finder info as binary data, presented the same as with the macOS native driver.
- com.apple.ResourceFork: The resource fork as binary data.
The resource fork may also be accessed during normal use by defining a special file suffix with the rsrc_ext option. When this is set, any lookup that ends in this suffix returns resource fork data for the corresponding file. For example, when mounting with -orsrc_ext=.rsrc, "image.psd.rsrc" can be used to access the resource fork for image.psd.
Of course, "image.psd.rsrc" may also exist independently, so this option can be set to anything suitable (:rsrc, -resource, etc) with the only condition being that it cannot include a path separator (as FUSE intercepts these).
Because of this, the more familiar /rsrc suffix used by previous releases of macOS is not supported in hfsfuse, but may still be used with hfsdump.
Finally, the entire volume may be mounted in resource-fork only mode using the rsrc_only option. In this mode, all entries on the filesystem are presented using the size and contents of their resource fork. Files with no resource fork will appear as empty, 0 size entries.
This option may be combined with the rsrc_ext option described above, in which case the special suffix will instead be used to access the regular data fork.
On Linux you may encounter the following error when inspecting xattrs: user.com.apple.ResourceFork: Argument list too long
This occurs when the resource fork is larger than the maximum allowed extended attribute size of 64kb. In this case you can still access the resource fork as described above by setting the rsrc_ext option or mounting in rsrc_only mode.
Under Haiku only, all extended attribute values are hex encoded to allow binary attribute data to be passed along from FUSE. Values may be read using a tool like xxd, e.g. to read the resource fork extended attribute for "file":
catattr -r user.com.apple.ResourceFork file | xxd -r -pExtended attributes are presented in the preferred namespace for the OS, typically user.. Alternate namespaces may be chosen when building hfsfuse by setting the XATTR_NAMESPACE make var.
This should include the trailing . as an empty value indicates no namespacing (such as on macOS.)
HFS+ symbolic links may contain extended attributes of their own separate from their link destination.
For systems such as Linux that do not support reading user-namespace extended attributes of symlinks, these can be viewed by using the -o disable_symlinks mount option, which will cause hfsfuse to display all symlinks as regular files.
HFS+ filesystems created on Mac OS Classic do not contain the typical set of Unix ownership and permission information for files and folders. For these hfsfuse provides the options default_file_mode, default_dir_mode, default_uid, and default_gid to specify fallback values if needed.
These defaults only apply to filesystem entries that are missing this information, they don't affect files or folders with existing permissions. They are applied before the FUSE uid, gid, and umask options or any fuse-idmap conversions and will still be subject to them.
Disk images can be mounted using dmg2img.
One-liner to extract the HFS+ partition in a DMG to an img:
dmg2img -p $(dmg2img -l image.dmg | grep Apple_HFS | cut -d' ' -f2 | cut -d: -f1) image.dmg image.img
hfsfuse <opts> /dev/md`mdconfig -f image.img` <mountpoint>
mnt=$(losetup -f)
losetup $mnt image.img
hfsfuse <opts> $mnt <mountpoint>
When sharing a disk between systems it's often convenient to establish a mapping between corresponding users/groups. FUSE offers uid and gid options to force ownership of all files on the mounted system to the provided id, but more involved mappings for multiple users or specific user and group combinations can be done using the idmap FUSE module.