mount_fusefs —"> Man pages sections > man8 > mount_fusefs


MOUNT_FUSEFS(8) System Manager's Manual MOUNT_FUSEFS(8)


mount_fusefsmount a Fuse file system daemon


mount_fusefs [-A] [-S] [-v] [-D fuse_daemon] [-O daemon_opts] [-s special] [-m node] [-h] [-V] [-o option ...] special node [fuse_daemon ...]


The most basic way of usage is as follows: before calling mount_fusefs, one starts a fuse daemon on the given special (in practice, the daemon gets one automatically, which can then be indentified via fstat(1)), and that special file can then be mounted by mount_fusefs.
However, the procedure of spawning a daemon can be automated so that it's also performed by mount_fusefs. If the command invoking a given fuse_daemon is appended to the list of arguments, mount_fusefs will call then fuse_daemon via that command, in a way that fuse_daemon will be instructed to attach itself to special. From that on mounting goes as in the simple case. (See also DAEMON MOUNTS.)
The special argument can be utilized by mount_fusefs in the following ways:
The options are as follows:
-A, --reject-allow_other
Prohibit the allow_other mount flag. Intended for to be used in scripts and the sudoers(5) file.
-S, --safe
Run in safe mode, that is, reject invoking a filesystem daemon.
Be verbose.
-D, --daemon daemon
Call the specified daemon.
-O, --daemon_opts opts
Add opts to the daemon's command line.
-s, --special special
Use special as special.
-m, --mountpath node
Mount on node.
-h, --help
Show help.
-V, --version
Show version information.
Mount options are specified via -o. The following options are available (and also their negated versions, by prefixing them with “no”):
Enable traditional (file mode based) permission checking in kernel.
Do not apply STRICT ACCESS POLICY. Only root can use this option.
Limit size of read requests with n.
Refuse shared mounting of the daemon. This is the default behaviour, to allow sharing, use expicitly -o noprivate.
Don't refuse unmounting if there are secondary mounts.
Prefix absolute symlinks with mountpoint.
Besides the above mount options, there is a set of pseudo-mount options which are supported by the Fuse library. One can list these by passing -h to a Fuse daemon. Most of these options have effect only on the behaviour of the daemon (that is, their scope is limited to userspace). However, there are some which do require in-kernel support. Currently the following ones are supported by the kernel:
Bypass the buffer cache system.
By default, cached buffers of a given file are flushed at each open(2). This option disables this behaviour.


Usually users don't need to use mount_fusefs directly, as the Fuse library enables Fuse daemons to invoke mount_fusefs by themselves. That is,
fuse_daemon mountpoint
has the same effect as
mount_fusefs auto mountpoint fuse_daemon
This is the recommended way of usage, unless you want to go beyond basic usage (eg, run daemon on a low privilege level, but mount it as root).


The strict access policy for Fuse filesystems lets one to use the filesystem only if the filesystem daemon has the same credentials (uid, real uid, gid, real gid) as the user.
This is applied for Fuse mounts by default, and only root can mount without the strict access policy (cf. the allow_other mount option).
The reason is to shield users from the daemon “spying” on their I/O activities.
Users might opt for willingly relax strict access policy (as far they are concerned) by doing their own secondary mount (cf. SHARED MOUNTS).


A Fuse daemon can be shared, ie. mounted multiple times. When doing the first (primary) mount, the spawner and the mounter of the daemon must have the same uid, or the mounter should be the superuser.
After the primary mount is in place, secondary mounts can be done by anyone (unless this feature is disabled by private). The behaviour of a secondary mount is analogous to that of symbolic links: they redirect all filesystem operations to the primary mount.
Doing a secondary mount is like signing an agreement: by this action, the mounter agrees that the Fuse daemon can trace her I/O activities. From that on, she is not banned from using the filesystem (either via her own mount or via the primary mount), regardless whether allow_other is used or not.
The device name of a secondary mount is the device name of the corresponding primary mount, followed by a '#' character and the index of the secondary mount, like /dev/fuse0#3.


System administratos might want to use a custom mount policy (ie., one going beyond the vfs.usermount sysctl). The primary tool for such purposes is sudo(8). However, given that mount_fusefs is capable of invoking an arbitrary program, one must be careful about this. mount_fusefs is designed in a way such that it makes that easy. For this purpose, there are options which disable certain risky features (cf. -S and -A), and command line parsing is done in a flexible way: mixing options and non-options allowed, but processing them stops at the third non-option argument (after the first two has been utilized as device and mountpoint). The rest of the command line specifies the daemon and its arguments. (Alternatively, the daemon, the special and the mount path can be specified using the respective options.) Note that mount_fusefs ignores the environment variable POSIXLY_CORRECT and always behaves as described.
In general, to be as scripting / sudoers(5) friendly as possible, no information has a fixed position in the command line, but once a given piece of information is provided, subsequent arguments/options cannot override it (maybe with the exception of some non-critical ones).


Setting this has the same effect as the -S option.
Setting this has the same effect as the -v option.
If this is set, mount_fusefs will ignore uknown mount options.
Adjust behaviour to the needs of the FUSE library. Currently it has effect on help output.
Although the following variables don't have effect on mount_fusefs itself, they affect the behaviour of fuse daemons:
Device to get attached to. If not set, the multiplexer path /dev/fuse is used.
File desciptor of an opened Fuse device to use. Overrides FUSE_DEV_NAME.
If this is set, the library won't attempt to mount the filesystem, even if a mountpoint argument is supplied.


Fuse devices by which the kernel and Fuse daemons can communicate.
The multiplexer path. An open(2) performed on it automatically gets passed to a free Fuse device by the kernel (which might be just created for this puprose).


Mounting the example filesystem of the Fuse distribution (from its directory): either
./fusexmp /mnt/fuse
mount_fusefs auto /mnt/fuse ./fusexmp
Doing the same in two steps, using /dev/fuse0:
FUSE_DEV_NAME=/dev/fuse0 ./fusexmp &&
mount_fusefs /dev/fuse0 /mnt/fuse
A script wrapper for fusexmp which ensures that mount_fusefs doesn't call any external utility and also provides a hacky (non race-free) automatic device selection:
#!/bin/sh -e
n=`ls /dev/fuse* | awk 'END{ print FNR }'`
FUSE_DEV_NAME=/dev/fuse$n fusexmp
mount_fusefs -S /dev/fuse$n /mnt/fuse “$@”
A better (race-free) script wrapper:
#!/bin/sh -e
exec 3<>/dev/fuse
FUSE_DEV_FD=3 fusexmp
mount_fusefs -S 3 /mnt/fuse “$@”


fstat(1), mount(8), umount(8), sudo(8)


Secondary mounts are to be unmounted via their device name. If they attempted to be unmounted via their filesystem root path, the unmount request will be forwarded to the primary mount path. In general, unmounting by device name is less error-prone than by mount path (although the latter will also work under normal circumstances).
If the daemon is specified via the -D and -O options, it will be invoked via system(3), and the daemon's command line will be also appended a “&” sygill, so that we don't have to wait for its termination. That is, you'd better use a simple command line when invoking the daemon via these options.


mount_fusefs appears as the part of the FreeBSD implementation of the Fuse userspace filesystem framework (see This user interface is FreeBSD specific.


special is treated as a multiplexer if and only if it's literally the same as auto or /dev/fuse. Other paths which are equivalent with /dev/fuse (eg., /../dev/fuse) are not.
January 13, 2006 Debian Sid