systemd automount sshfs: how to automatically mount a remote sshfs filesystem with systemd

mnt-mymountpoint.mount

# Change the relevant {{ PARTS OF THIS FILE }} for your remote address etc.
# Make sure this unit file is named similarly to your mountpoint; e.g., for /mnt/mymountpoint name this file mnt-mymountpoint.mount
# On Ubuntu:
# $ sudo cp mnt-mymountpoint.mount /lib/systemd/system/
# $ sudo systemctl enable mnt-mymountpoint.mount
# $ sudo systemctl start mnt-mymountpoint.mount
# On Fedora:
# $ sudo cp mnt-mymountpoint.mount /etc/systemd/system
# $ sudo systemctl enable mnt-mymountpoint.mount
# $ sudo systemctl start mnt-mymountpoint.mount

[Unit]
Description=Mount my remote filesystem over sshfs with fuse

[Install]
WantedBy=multi-user.target

[Mount]
What={{ USER }}@{{ HOST }}:{{ REMOTE DIR }}
Where={{ MOUNTPOINT like /mnt/mymountdir }}
Type=fuse.sshfs
# I recommend using your SSH key (no password authentication) with the following options so that you don't have to mount every time you boot
Options=_netdev,allow_other,IdentityFile=/home/{{ MY LOCAL USER WITH SSH KEY IN ITS HOME DIRECTORY }}/.ssh/id_rsa,reconnect,x-systemd.automount,uid=1000,gid=1000
# Change to your uid and gid as well according to the output of `cat /etc/group`

You're an absolute lifesaver @Kreijstal !

I am wondering though, are there serious security issues with sshing to a server using sudo? I can imagine that terminal control codes injected into a .bash_logout file or some such might be a potential hazard, though if they were that would be a fairly serious risk for any account. It's also kind of moot here given my understanding is if I run this using systemd.mount, no shell session is spawned.

Either way, thanks for saving the few strands of hair I haven't already pulled out in the process of getting this all to work, it's been a surprisingly bumpy ride.

EDIT: Welp, nevermind, it completely stopped working again.

EDIT: Welp, nevermind, it completely stopped working again.

rip, is the systemd logs the same error?

No idea why it worked before or what caused it to stop working, however now when I do it manually, it asks for a password, so seems there was something wrong with the identity file I was pointing it to.

So I generated a new key for root, and everything appears to be working fine again. No doubt posting this comment will mean that it immediately breaks however.

Assuming it doesn't, thanks again for you help, here's the systemd.mount for anybody interested, this allows you to use the same mount unit with as system and user, but mount as user, allowing you to trigger using an automount unit (which only the system session can do), assuming an identically named key file is in both /root/.ssh/ and $HOME/.ssh.

The "{ VAR }" syntax is for the reader to manually input values:

[Unit]
Description=Mount { SERVER_NAME } server over SSHFS

[Install]
WantedBy=remote-fs.target
WantedBy=multi-user.target

[Mount]
What={ REMOTE_USER }@{ REMOTE_HOST }:{ REMOTE_PATH }
Where={ LOCAL_PATH }
Type=fuse.sshfs
Options=_netdev,reconnect,allow_other,uid={ LOCAL_USER_ID },gid={ LOCAL_USER_GID },ServerAliveInterval=60,ServerAliveCountMax=6,rw,nosuid,default_permissions,idmap=user,follow_symlinks,IdentityFile=%h/.ssh/{ KEY_FILENAME }

EDIT: I should mention that one problem with this approach is that one cannot unmount the path as user without creating a bunch of ancillary files for each possible method of unmounting, e.g., umount command vs utilising dbus, etc. So I'm working on seeing if I can find a way around that, especially given some filemanagers such as dolphin will hang if you enter even the parent directory of the chosen mountpoint, presumably because it's preemptively fetching the file listing for the purposes of its treeview expansion.

On a minimal Debian 12 installation that uses the ifupdown package, it was a race with the mount units losing. Even after adding After=network-online.target the mount unit was run before DHCP had returned the address resulting in the terse read connection reset by peer in the journal. Disabling the interface in /etc/network/interfaces and enabling systemd-networkd.service resolved the race.

While this was in a QEMU virtual machine, this may be useful for a Debian installation that uses the ifupdown package.