linux-manager
A Debian package that configures a Linux host or virtual machine as a server managed by the
Ubuntu Linux Manager. Every component is installed; the router (router.enabled) and the VPN
(vpn.enabled) are switched on in the configuration, and the rest is the core. Without the
router the server needs one network interface, and the VPN clients reach the server's networks
through NAT from its address. The package provides:
- site-to-site IKEv2 IPSec, with the router on, with pre-shared key or certificate authentication (strongSwan swanctl)
- IKEv2 access using EAP-TLS
- a two-level certificate authority: a root, which may be kept offline, and issuing CAs for the VPN clients, for site-to-site and for general use, with certificate templates, published certificates and CRLs, and an optional ACME server
- OpenVPN access with a certificate and a password per user, the users kept in a local
LDAP directory and managed with
linux-manager-useror the web service - optional WireGuard endpoint
- routing and NAT for a protected subnet, TCP MSS clamping, and UFW firewall rules
- firewall rules for the ports Docker publishes, when Docker is installed
- metrics of the server kept in PostgreSQL, served for an OpenTelemetry Collector and pushed over OTLP
- static web sites served by Caddy, uploaded through the web service or WebDAV, and an optional primary name server (PowerDNS) of the server's domains, with a secondary and records for the sites
Nothing in the package is specific to a cloud provider. It is built on the external and
internal NIC names, and a mode setting says how much of the rest you are supplying and
how much is read from the system. Detection happens only in the mode that asks for it, so
a router never quietly guesses which interface to NAT out of unless you told it to.
Platform-specific additions live in separate modules and are selected explicitly.
Installing the package configures nothing by default. The shipped platform is none,
which means the files are laid down and the machine is left alone until you say
otherwise.
Installing
The package can be installed by hand or as a step in automated provisioning. Both are supported equally, and both end up with the same two things: a configuration file and a service that applies it.
/etc/linux-manager/linux-manager.json the configuration
systemctl restart linux-manager-setup apply it
If the host runs Docker, the recommended order is Docker first and the package second. The package secures the ports Docker publishes when it finds the Docker service; the steps for a host that already runs Docker, and for installing Docker later, are under Docker.
By hand
apt-get install linux-manager # answer "none" to the platform question
$EDITOR /etc/linux-manager/linux-manager.json
systemctl restart linux-manager-setup
Answering none to the platform question defers everything: no further questions are
asked and nothing on the machine is configured. Set platform in the file when you are
ready.
Automated
Preseed the answers and the server comes up configured, with no follow-up command. Note
that platform must be something other than none, and mode says how much you are
supplying:
debconf-set-selections <<'EOF'
linux-manager linux-manager/platform select generic
linux-manager linux-manager/mode select manual
linux-manager linux-manager/external_interface string eth0
linux-manager linux-manager/internal_interface string eth1
linux-manager linux-manager/int_addr string 10.1.1.4
linux-manager linux-manager/int_gateway_ip string 10.1.1.1
linux-manager linux-manager/local_fqdn string router.example.com
linux-manager linux-manager/local_cidrs string 10.0.0.0/24
linux-manager linux-manager/remote_addrs string peer.example.net
linux-manager linux-manager/remote_id string peer.example.net
linux-manager linux-manager/remote_cidrs string 192.168.0.0/24
linux-manager linux-manager/psk password s3cr3t
EOF
DEBIAN_FRONTEND=noninteractive apt-get install -y linux-manager
With mode interfaces the int_addr and int_gateway_ip lines are dropped and read
from the internal NIC instead; with mode auto the interface names go too.
Alternatively write /etc/linux-manager/linux-manager.json directly before or after installing.
A configuration-management tool should do that and then restart linux-manager-setup.
See examples/ for a cloud-init template and a shell installer, and examples/azure/ / examples/gcp/ for full Terraform examples that deploy a router on Azure or GCP.
Configuration
/etc/linux-manager/linux-manager.json is JSON, mode 0600, and is created on first install. It
is managed with ucf, so it is yours to edit: a later dpkg-reconfigure applies the answers
given and asks before touching anything you altered by hand, and an upgrade asks no questions
and leaves the file as it is. Settings the file leaves out take their defaults when it is read, so a
file written by cloud-init or by hand needs only the settings that differ. A complete example is
installed at /usr/share/doc/linux-manager/linux-manager.json.example.
The file is an object of sections, each an object of settings. A section or setting that is left out takes its default, and an empty string or empty list means unset. Every value must have the type shown below; an unknown section or setting is an error.
| Section | Setting | Type | Default | Meaning |
|---|---|---|---|---|
general |
platform |
string | none |
none, generic, azure or gcp. none defers configuration entirely |
general |
mode |
string | manual |
How much you supply: manual, interfaces or auto |
router |
enabled |
boolean | false |
The router component: routing between the two NICs and site-to-site. Off, the server needs only interfaces.external, and site_to_site must be empty |
vpn |
enabled |
boolean | false |
The VPN component: IKEv2, OpenVPN and WireGuard. Off, p2s.enabled, openvpn.enabled and wireguard.enabled must be false. With the router off, the VPN clients reach local.cidrs through NAT from the server's address |
interfaces |
external |
string | Name of the NIC facing the untrusted network | |
interfaces |
internal |
string | Name of the NIC facing the protected network | |
wan |
local_fqdn |
string | This server's name or IP address: its IKE identity in fqdn mode, the name resolved in public_ip mode, the name of HTTPS, IKEv2 and OpenVPN, and the primary name server of its domains |
|
wan |
public_address |
string | The server's public IP address, which its own interfaces do not show behind a cloud's NAT: the address of the records of the sites with dns_address public |
|
local |
cidrs |
list | Local subnets routed via the internal side, masqueraded to the external one, and offered to IKEv2 and OpenVPN clients | |
local |
int_addr |
string | This host's address on the internal network | |
local |
int_gateway_ip |
string | Next hop on the internal side | |
site_to_site |
object | The site-to-site connections, each a name and an object of the settings below | ||
p2s |
enabled |
boolean | false |
IKEv2 access |
p2s |
address_pool |
string | CIDR assigned to IKEv2 clients | |
p2s |
dns_domains |
list | The domains IKEv2 clients resolve with the server's resolver (INTERNAL_DNS_DOMAIN, RFC 8598, and the match domains of the Apple VPN profile), at most 16 | |
p2s |
dns_listen |
string | internal |
The addresses of the resolver the IKEv2, OpenVPN and WireGuard clients are given: internal, external or both |
wireguard |
enabled |
boolean | false |
WireGuard endpoint |
wireguard |
address |
string | The server's own address on wg0 with the prefix of the peers' network, for example 192.168.200.1/24; the network and the broadcast address of a network larger than /31 are refused |
|
wireguard |
listen_port |
integer | 51820 |
UDP port |
openvpn |
enabled |
boolean | false |
OpenVPN access for users |
openvpn |
address_pool |
string | IPv4 CIDR assigned to OpenVPN clients | |
openvpn |
port |
integer | 1194 |
Port of the OpenVPN server on the external interface |
openvpn |
protocol |
string | udp |
udp or tcp |
docker |
mode |
string | auto |
auto secures Docker when its service is installed, on requires it, off removes the rules |
docker |
exposed_ports |
list | TCP ports and port ranges (8000:8100) reachable from the external interface; at most 15, a range counting as two |
|
web |
enabled |
boolean | true |
The status web service |
web |
listen |
string | both |
The interfaces the status web service listens on: internal, external or both |
web |
port |
integer | 7081 |
TCP port of the status web service |
web |
database |
string | A PostgreSQL connection string for the refresh tokens; without it, the linux-manager database on this host that the web service creates |
|
web |
trusted_proxies |
list | the address the call came in on | Addresses whose X-Forwarded-For header names the client for the /auth rate limit and the Client IP of the status |
web |
status_content |
string | basic |
What / and /status show: basic, standard or full (see Status web service) |
web |
shell |
boolean | false |
/shell opens a login shell for a member of admins with console login (see The shell) |
web |
shell_idle_minutes |
integer | 15 |
Minutes without a keystroke after which a shell is closed, 1 to 1440 |
web |
log_files |
list | Absolute paths of the log files /logs shows, at most 32 (see The /logs API) |
|
web |
shell_auth_timeout |
integer | 15 |
Minutes after the last use of a shell in which a new shell opens without signing in again, 0 (always sign in) to 1440 |
ssh |
ldap_login |
boolean | false |
sshd accepts the directory password of the console accounts in the directory (see Console login and SSH) |
https |
enabled |
boolean | true |
Caddy serves the status web service and the Manager over HTTPS for wan.local_fqdn |
https |
email |
string | The e-mail address of the Let's Encrypt account | |
https |
web_port |
integer | 7444 |
HTTPS port of the status web service |
https |
manager_port |
integer | 7443 |
HTTPS port of the Manager |
https |
acme_staging |
boolean | false |
Caddy takes its certificates from Let's Encrypt's staging environment alone, whose certificates no browser trusts and whose rate limits are higher, for routers rebuilt often (see HTTPS) |
https |
caddy_source |
string | ubuntu |
Where Caddy is installed from: ubuntu, the Ubuntu archive, or upstream, Caddy's own repository (see HTTPS) |
oidc |
enabled |
boolean | false |
The OpenID Provider at the status web service's HTTPS site, which needs web.enabled, https.enabled and wan.local_fqdn (see OpenID Provider) |
sites |
enabled |
boolean | false |
Caddy serves the web sites (see Web sites) |
sites |
upload_max_mb |
integer | 512 |
The largest upload of a site's file or zip archive, in MiB, 1 to 10240 |
sites |
webdav |
boolean | false |
WebDAV of the sites' folders, which needs https.enabled and wan.local_fqdn (see WebDAV) |
sites |
webdav_port |
integer | 7445 |
HTTPS port of WebDAV |
dns |
enabled |
boolean | false |
The primary name server of the server's domains, wan.local_fqdn, which must then be a DNS name (see Name server) |
dns |
secondary |
string | The secondary name server of every domain, a host name other than wan.local_fqdn; it may transfer the domains |
|
dns |
listen |
string | external |
The interfaces the name server answers on: internal, external or both |
dns |
ttl |
integer | 300 |
TTL of the records, 60 to 86400 seconds |
pki |
root_name |
string | Linux Manager Root CA |
Common name of the root CA, when the server creates it |
pki |
root_days |
integer | 7300 |
Days the root CA's certificate is valid |
pki |
ca_days |
integer | 3650 |
Days an issuing CA's certificate is valid |
pki |
key |
string | rsa-4096 |
Key of the root and the issuing CAs: rsa-2048, rsa-3072, rsa-4096, ec-p256 or ec-p384 |
pki |
crl_days |
integer | 7 |
Days the CRLs of the issuing CAs are valid |
pki |
root_crl_days |
integer | 180 |
Days the root CRL is valid, when the server signs it |
pki |
publish |
list | http |
Where the CA certificates and CRLs are published: http, http-80, https, https-default (see Certificates) |
pki |
p2s_name |
string | Linux Manager VPN CA |
Common name of the P2S issuing CA |
pki |
s2s_name |
string | Linux Manager Site-to-Site CA |
Common name of the site-to-site issuing CA |
pki |
issuing_cas |
object | The general-purpose issuing CAs, each a name and {name, templates, acme} (see Certificates) |
|
pki |
templates |
object | Certificate templates, each a name and the options it changes or defines (see Certificates) | |
pki |
extended_key_usages |
object | Custom extended key usages templates may give, each a name and its OID in dotted form (see Certificates) | |
pki |
root_key_warning |
boolean | true |
linux-manager-ca status and /ca warn while the root key is on the server; false silences that warning |
metrics |
enabled |
boolean | true |
The metrics service: collection, history and /metrics |
metrics |
listen |
string | internal |
The interfaces /metrics is served on: internal, external or both |
metrics |
port |
integer | 9464 |
TCP port of /metrics |
metrics |
https_port |
integer | 9465 |
HTTPS port of /metrics, served by Caddy while https.enabled is true |
metrics |
service_account |
string | metrics |
The service account /metrics takes, which linux-manager-setup adds and keeps |
metrics |
otlp_endpoint |
string | An OTLP/HTTP URL the metrics are pushed to every minute, for example https://collector.example.com:4318/v1/metrics |
|
metrics |
otlp_username |
string | The user name of HTTP Basic authentication at the OTLP endpoint | |
metrics |
otlp_password |
string | Its password | |
metrics |
keep_details |
string | 24 hours |
How long the per-minute metrics are kept |
metrics |
keep_hourly |
string | 1 month |
How long the hourly summaries are kept |
metrics |
keep_daily |
string | 3 months |
How long the daily summaries are kept |
metrics |
keep_monthly |
string | 1 year |
How long the monthly summaries are kept |
metrics |
keep_yearly |
string | 10 years |
How long the yearly summaries are kept |
backup |
enabled |
boolean | false |
Scheduled backups (see Backups) |
backup |
schedule |
string | daily |
hourly, daily or weekly |
backup |
target |
string | local |
local, an archive in backup.path, or pbs, a Proxmox Backup Server |
backup |
path |
string | /var/backups/linux-manager |
The folder of local archives, an absolute path |
backup |
keep_hourly |
integer | 24 |
Hourly local archives kept, with the hourly schedule |
backup |
keep_daily |
integer | 7 |
Daily local archives kept, with the hourly and the daily schedule |
backup |
keep_weekly |
integer | 4 |
Weekly local archives kept |
backup |
keep_monthly |
integer | 12 |
Monthly local archives kept |
backup |
keep_yearly |
integer | 0 |
Yearly local archives kept |
backup |
metrics |
boolean | false |
The backup holds the database of the metrics and the refresh tokens |
backup |
rclone_remote |
string | remote:path of /etc/linux-manager/rclone.conf the local archives are copied to |
|
backup |
rclone_mode |
string | sync |
sync mirrors the archives, copy deletes nothing in the cloud |
backup |
pbs_repository |
string | user@realm!token@host[:port]:datastore |
|
backup |
pbs_password |
string | The secret of the API token | |
backup |
pbs_fingerprint |
string | The SHA-256 fingerprint of the server's certificate, when the system's CAs do not verify it | |
backup |
pbs_namespace |
string | The namespace of the datastore, such as routers/azure |
|
backup |
pbs_keyfile |
string | An encryption key of proxmox-backup-client key create, an absolute path |
Site-to-site connections
site_to_site maps a connection name to its settings. A name starts with a lowercase letter
and has at most 31 lowercase letters, digits and hyphens; it names the strongSwan connection,
its child and its secret (ike-psk-<name>). The install answers fill the connection
remote-site; more are added in the file, with linux-manager-config-edit, or through /site-to-site
of the status web service, which the Manager uses.
| Setting | Type | Default | Meaning |
|---|---|---|---|
addrs |
list | Remote gateway addresses or FQDNs | |
id |
string | Remote IKE identity, an FQDN or an IPv4 or IPv6 address | |
cidrs |
list | Remote subnets reachable through the tunnel | |
local_cidrs |
list | Local subnets advertised into the tunnel | |
local_id_mode |
string | fqdn |
The server's IKE identity: fqdn is wan.local_fqdn, public_ip the address it resolves to, internal_ip the internal address |
auth |
string | psk |
How both sides authenticate: psk, a pre-shared key, or certificate |
psk |
string | Pre-shared key, with auth psk |
|
psk_file |
string | Path to a file holding the key; wins over psk |
|
remote_ca |
string | With auth certificate, the PEM certificate of the CA that issued the peer's certificate; without it the server's site-to-site CA |
A connection is set up once all of addrs, id, cidrs, local_cidrs and its local
identity are known, with psk the key, and with certificate the server's site-to-site
certificate; all such connections are written to /etc/swanctl/conf.d/site-to-site.conf.
Traffic from a connection's local_cidrs to its cidrs is not masqueraded.
With auth certificate, the server authenticates with its site-to-site certificate, which the
site-to-site issuing CA signs for wan.local_fqdn and for the public or the internal address
that a connection with auth certificate uses as its identity (local_id_mode public_ip or
internal_ip); it is issued again when that set of names changes. The peer's certificate must name id and be issued
by the server's site-to-site CA, or by the CA of remote_ca. A peer with no PKI of its own gets
its certificate from the server (template site-to-site, see Certificates), with the root and
the site-to-site CA certificates as its trust anchors.
A minimal site-to-site configuration in auto mode:
{
"general": { "platform": "generic", "mode": "auto" },
"wan": { "local_fqdn": "router.example.com" },
"local": { "cidrs": ["10.0.0.0/24"] },
"site_to_site": {
"remote-site": {
"addrs": ["peer.example.net"],
"id": "peer.example.net",
"cidrs": ["192.168.0.0/24"],
"local_cidrs": ["10.0.0.0/24"],
"psk": "s3cr3t"
}
}
}
The server's IKE identity (wan.local_fqdn in fqdn mode) and the remote one (id) are
an FQDN or an IPv4 or IPv6 address. They are written to the strongSwan configuration as given;
strongSwan reads an address as an IP identity and a name as an FQDN, so a leading @ is not needed
and is refused. IKEv2 and OpenVPN access need a name in wan.local_fqdn.
WireGuard peers are VPN users' devices: a VPN user of the server's directory may have up to 16
peers, each with its public key and its IPs, IPv4 addresses or networks, kept in the user's
entry (linuxManagerWireGuardPeer) and set with PUT /users/<name> (wireguard_peers) or on the
WireGuard tab of the Manager's Remote access management. A key is used by one peer only,
the IPs of two peers may not overlap, and no peer's IPs cover the server's own address of
wireguard.address. The package owns the [Interface] section of /etc/wireguard/wg0.conf and,
between the lines # BEGIN and # END WireGuard peers of the users of the server's directory, a
[Peer] section for each peer of a user who is not disabled; setup rewrites them when the
settings or the peers change, and a change of a user's peers, disabling, enabling or removing a
user with peers restarts setup. Every other [Peer] section, added by hand, is carried across
untouched. While the directory cannot be reached, the users' peers in the file stay as they are.
Every feature is independent. Leaving a group empty simply means that feature is not configured, and its generated files are removed.
What mode decides
mode states how much of the network configuration you are supplying, and therefore how
much the package reads from the system:
mode |
You provide | The package works out |
|---|---|---|
manual |
both interface names, int_addr, int_gateway_ip |
nothing |
interfaces |
both interface names | int_addr and int_gateway_ip, from those interfaces |
auto |
nothing | the interface names too: the external one from the routing table, the internal one as the only other NIC with an address |
manual is the only mode in which no detection code runs at all. auto is the
convenient one and can pick the wrong interface, which is a trade you make deliberately
by choosing it rather than something the package does behind your back.
Detection ignores virtual interfaces, those the kernel lists under /sys/devices/virtual:
loopback, bridges such as docker0, veth pairs and tunnels. Only the NICs of the machine
count, so Docker does not change the result. If the host has more than two NICs with an
address, auto stops and says so; use mode interfaces and name them.
The addresses strongSwan binds on always come from the external NIC, in every mode, so
they are never configured twice and cannot drift out of step with the host. If a named
interface does not exist the service stops and says so. In interfaces and auto mode
the internal gateway is taken from the routing table, falling back to the first host
address of the connected subnet with a warning when the routing table has nothing to say.
Docker
Docker publishes container ports in its own firewall rules, which bypass UFW and can expose
ports on the external interface. Docker evaluates the DOCKER-USER chain before its own
rules, so the package filters there. With docker.mode set to auto, the package secures
Docker when docker.service is installed, and does nothing otherwise.
On a host that already runs Docker
Installing Docker first is the recommended order. After installing the package:
-
Set
general.platformin/etc/linux-manager/linux-manager.json. While it isnone, nothing is configured, the Docker rules included. -
List the TCP ports the running containers must expose to the internet, for example
"docker": { "exposed_ports": ["443"] }. The default is none: once the rules are applied, every published port that is not listed is unreachable from the external interface, and remains reachable through the IPsec tunnel. -
Apply the configuration with
systemctl restart linux-manager-setup. Docker is running, so this also restartslinux-manager-docker, which installs the rules. -
Check the result:
iptables -S DOCKER-USER # "-j LINUX-MANAGER-DOCKER" is the first rule iptables -S LINUX-MANAGER-DOCKER # the rules of the external interface
When Docker is installed after the package
The Docker rules are not applied when Docker starts for the first time, so run
systemctl restart linux-manager-setup after installing Docker, then check as above. A reboot or a
restart of Docker applies them as well.
The rules sit in their own chain, LINUX-MANAGER-DOCKER, entered by one jump at the top of
DOCKER-USER; other rules in DOCKER-USER are left alone. For traffic that arrives on the
external interface, in this order:
- IPsec protected traffic is accepted.
- TCP to the ports in
docker.exposed_portsis accepted; none by default. - Packets of established and related connections are accepted.
- Everything else is dropped.
Traffic from every other interface, including the internal one and the Docker bridges, is not affected. The rules cover IPv4 only.
linux-manager-setup starts after docker.service, so Docker is up when it looks for it; the
ordering has no effect on a host without Docker. The linux-manager-docker service applies the
rules after docker.service starts and removes
them when Docker stops, so restarting Docker restores them. systemctl restart linux-manager-setup applies a changed configuration while Docker runs. With docker.mode set to
on, the service fails when Docker is not installed or its daemon is not running; with
off it removes the rules.
Editing the configuration
linux-manager-config-edit, run as root, shows and changes /etc/linux-manager/linux-manager.json:
linux-manager-config-edit show
linux-manager-config-edit set web.status_content standard
linux-manager-config-edit set web.port 7082
linux-manager-config-edit set web.trusted_proxies 127.0.0.1,10.0.0.4
linux-manager-config-edit set site_to_site.branch.addrs branch.example.net
linux-manager-config-edit remove site_to_site.branch
linux-manager-config-edit set --restart oidc.enabled true
show prints the file as indented JSON and does not change it. set takes a setting as
section.option, or as site_to_site.NAME.option for a site-to-site connection, which it
creates with the defaults when it does not exist, and converts the value to the setting's type:
a whole number, true or false, a list separated by commas (an empty value is an empty list),
or a string; an unknown setting is refused. remove site_to_site.NAME removes a connection. It
checks the whole configuration as linux-manager-setup does, and only then writes the file,
indented, with mode 0600, and reminds that linux-manager-setup applies it; with --restart, set
and remove restart linux-manager-setup and wait until it has applied the change, and exit with
status 1 when it fails. The status web service
writes the file the same way, and a lock keeps the two from overwriting each other's change. A
refused request exits with status 2, any other failure with 1.
Bash completion
Bash completes the commands, options and arguments of linux-manager-user, linux-manager-group,
linux-manager-role, linux-manager-token, linux-manager-config-edit, linux-manager-backup, linux-manager-ca, linux-manager-caddy
and linux-manager-site with the Tab key. The package installs a file per command in
/usr/share/bash-completion/completions, which the bash-completion package, recommended by
linux-manager, loads on the first Tab after the command. For root, who can read them, the
completions also offer the users and groups of the directory, the settings and site-to-site
connections of linux-manager.json with true and false for a boolean setting, the issuing CAs and
certificate templates, and the web sites; for other users only the fixed words.
sudo -i
linux-manager-config-edit set oidc.<Tab>
linux-manager-user passwd <Tab>
Applying changes
systemctl restart linux-manager-setup
The service also runs at boot. With platform none it does nothing and says so.
Otherwise it regenerates the configuration files, resolves the interfaces and their
addresses as mode directs, applies routes, generates WireGuard keys and injects
firewall rules. It provisions the user directory in slapd while OpenVPN or IKEv2
access is configured. It then starts strongswan, wg-quick@wg0 and
openvpn-server@linux-manager if the corresponding feature is configured, reloading or
restarting them only when their inputs changed, and
restarts the remote access clients' resolver only when its configuration changed. Turning a
feature off in the configuration stops its service and removes its files. Every step is
idempotent, so restarting with no changes does nothing.
To regenerate the configuration files and PKI without starting, stopping or reloading anything:
/usr/lib/linux-manager/configure
This does still write to the system: it renders into /etc/swanctl/,
/etc/systemd/resolved.conf.d/ and /etc/wireguard/, creates what the PKI is missing and
adds the root CA to the host trust store. What it leaves alone is routes, firewall rules and
services - that is what systemctl restart linux-manager-setup adds.
Certificates
The server has a two-level certificate authority, kept by linux-manager-setup and managed with
linux-manager-ca, linux-manager-user and /ca of the status web service, which the Manager
uses. It is held in ou=pki,dc=linux-manager,dc=local of the server's directory, next to the users:
cn=root the root CA; without linuxManagerPrivateKey while the root is offline
cn=<name> an issuing CA; with linuxManagerCertificateRequest while it waits for the offline root
cn=<name>-<time> an issuing CA taken out of use, with linuxManagerRetired
serialNumber=<serial>,cn=<CA> a certificate the CA issued
A CA entry (class linuxManagerAuthority) has its certificate (cACertificate;binary), key, CRL
(certificateRevocationList;binary) and CRL number. A certificate entry (class
linuxManagerCertificate) has the certificate (userCertificate;binary), its kind (ca, router,
user or external), template, names, expiry and revocation, and for the server's and the users'
certificates the key. A user's certificate is also the userCertificate of the user's entry. No
account of the directory reads a key (linuxManagerPrivateKey); only root does, through the local
socket. slapcat writes the whole PKI into every backup with the users.
ldapsearch -Q -Y EXTERNAL -H ldapi:/// -b ou=pki,dc=linux-manager,dc=local -s one cn
The CAs
The root signs the issuing CAs only (path length 1), and the issuing CAs sign the certificates (path length 0):
p2s, namedpki.p2s_name, signs the server's VPN server certificate, which IKEv2 and OpenVPN present, and the users' certificates. IKEv2 and OpenVPN accept client certificates of this CA only.s2s, namedpki.s2s_name, signs the server's site-to-site certificate and the certificates of its peers.- The general-purpose CAs of
pki.issuing_cas, each a name and{"name": common name, "templates": [...], "acme": false}. The name starts with a lowercase letter, has at most 31 lowercase letters, digits and hyphens, and is notroot,p2sors2s;templatesare the templates the CA issues.
Once a platform is chosen, linux-manager-setup creates what is missing: the root when it has
neither certificate nor key, each issuing CA, the server's certificates once wan.local_fqdn is
set, and the CRLs. The names, lifetimes and key type of pki apply when something is created. An
issuing CA removed from pki.issuing_cas is retired: its entry is renamed cn=<name>-<time>,
loses its key, and its certificate is revoked on the root CRL. The server renews its own certificates when less than a
third of their lifetime is left, when the names they need change, or when their CA changes; a
user's certificate is renewed by removing and adding the user.
The package copies what each consumer needs: the root, p2s and s2s into
/etc/swanctl/x509ca/ as root.pem, p2s.pem and s2s.pem (strongSwan reads one certificate
per file), the remote CA of each site-to-site connection that has one as s2s-<name>.pem, the
server's certificates into /etc/swanctl/x509/ and their keys into /etc/swanctl/private/, the
CRLs into /etc/swanctl/x509crl/, the root and p2s with their CRLs into
/etc/openvpn/server/linux-manager-ca.pem and linux-manager-crl.pem, and the root into the system
trust store.
linux-manager-ca status shows the root, the issuing CAs and what needs attention:
linux-manager-ca status
linux-manager-ca list --ca p2s
Templates
A template says what a certificate holds:
| Option | Type | Default | Meaning |
|---|---|---|---|
key |
string | rsa-3072 |
rsa-2048, rsa-3072, rsa-4096, ec-p256 or ec-p384, for a key the server makes |
days |
integer | 365 |
Days the certificate is valid, at most until its CA's certificate expires |
key_usage |
list | digital_signature |
Of digital_signature, content_commitment, key_encipherment, data_encipherment, key_agreement |
extended_key_usage |
list | Of the built-in server_auth, client_auth, code_signing, email_protection, time_stamping, ocsp_signing, ipsec_ike, ike_intermediate, and the custom usages of pki.extended_key_usages |
|
policies |
list | Certificate policy OIDs in dotted form, written into a Certificate Policies extension | |
subject |
string | name |
The common name: dns, a DNS name, which is also a subject alternative name; name, any name; email, an e-mail address |
sans |
list | The subject alternative names allowed: of dns, ip, email |
The built-in templates:
| Template | Key usage | Extended key usage | Subject | Names | Issued by |
|---|---|---|---|---|---|
vpn-server |
digital signature, key encipherment | server auth, IKE intermediate | dns |
dns, ip |
p2s, for the server only |
vpn-user |
digital signature | client auth | name |
dns, email |
p2s, for the users only |
site-to-site |
digital signature, key encipherment | server auth, client auth | dns |
dns, ip |
s2s |
server |
digital signature, key encipherment | server auth | dns |
dns, ip |
general-purpose CAs |
client |
digital signature | client auth | name |
dns, email |
general-purpose CAs |
pki.templates changes options of the built-in templates, which cannot be removed, and adds
templates of its own, named like the issuing CAs:
"pki": {
"issuing_cas": { "servers": { "name": "Example Server CA", "templates": ["server", "web", "smartcard"] } },
"extended_key_usages": { "smartcard_logon": "1.3.6.1.4.1.311.20.2.2" },
"templates": {
"server": { "days": 397 },
"web": { "key": "ec-p256", "days": 90, "key_usage": ["digital_signature"],
"extended_key_usage": ["server_auth"], "subject": "dns", "sans": ["dns"] },
"smartcard": { "key_usage": ["digital_signature", "key_encipherment"],
"extended_key_usage": ["client_auth", "smartcard_logon"], "subject": "name",
"sans": ["email"], "policies": ["1.3.6.1.4.1.32473.1.1"] }
}
}
A custom extended key usage is first added to pki.extended_key_usages, a name of at most 31
lowercase letters, digits, hyphens and underscores that is not a built-in one, and its OID; a
template then gives it by its name. A usage a template gives cannot be removed. policies puts
the OIDs of the issuance policies a certificate is issued under into its Certificate Policies
extension, without qualifiers.
Issuing and revoking
linux-manager-ca issue --ca servers --template server --csr web.csr > web.pem
linux-manager-ca issue --ca servers --template server --subject web.example.com --san 10.0.0.5 --pfx web.pfx
linux-manager-ca revoke 5F0A12C3D4 --reason key_compromise
With a CSR, the certificate is for the CSR's key, common name and subject alternative names,
checked against the template, and the certificate and the chain (issuing CA, root) are printed.
Without one, the server makes a key as the template says and writes a PKCS#12 file with the key,
the certificate and the chain, protected by the password read from the first line of standard
input; the server keeps the certificate and not the key. vpn-server and vpn-user are issued by
the server only.
revoke takes the serial number list shows and a reason: unspecified, key_compromise,
affiliation_changed, superseded, cessation_of_operation or privilege_withdrawn. A user's
certificate is revoked by removing the user, and an issuing CA by removing it from
pki.issuing_cas; a revoked router certificate is issued again by linux-manager-setup.
Published certificates and CRLs
Every certificate names its issuer's certificate (AIA) and CRL (CRL distribution point) at each
endpoint of pki.publish, in that order. They are served, without authentication, as DER at
/pki/<name>.crt and /pki/<name>.crl, the root as root, from copies in
/var/lib/linux-manager/pki-publish:
| Endpoint | URL | Served by |
|---|---|---|
http |
http://<wan.local_fqdn>:<web.port>/pki/ |
the status web service; needs web.enabled |
http-80 |
http://<wan.local_fqdn>/pki/ |
Caddy, on port 80, open on both interfaces; needs https.enabled |
https |
https://<wan.local_fqdn>:<https.web_port>/pki/ |
Caddy; needs https.enabled |
https-default |
https://<wan.local_fqdn>/pki/ |
Caddy, on port 443, open on both interfaces; needs https.enabled |
The URLs are written into a certificate when it is signed, so a change of pki.publish or of the
names applies to the certificates signed afterwards. An empty pki.publish leaves the extensions
out.
The CRLs of the issuing CAs are valid for pki.crl_days, and the root CRL, when the server signs
it, for pki.root_crl_days. A CRL is signed again at every revocation and, once half of its
validity has passed, by linux-manager-pki.timer, which runs linux-manager-ca maintain daily; it also
has linux-manager-setup renew the server's certificates when they are due. OpenVPN checks the CRL
of every CA of a client's chain, the root's included, and refuses every client while one has
expired.
Offline root
Keeping the root key off the server is recommended, not required; while it is on the server,
linux-manager-ca status says so, unless pki.root_key_warning is false. Export it, encrypted with a
passphrase read from the first line of standard input, keep the file in a safe place, and remove
the key from the server:
linux-manager-ca root export-key --out /root/linux-manager-root-key.pem
linux-manager-ca root remove-key
While the root key is not on the server, a new issuing CA gets its key and a CSR and waits;
linux-manager-ca status says so. Print the CSR and the extensions to sign it with, sign it with the
root on the offline machine, and import the certificate:
linux-manager-ca ca csr servers # the CSR, then the [ issuing_ca ] extensions
# on the offline machine, with the CSR in servers.csr and the extensions in issuing_ca.cnf:
openssl x509 -req -in servers.csr -CA root-cert.pem -CAkey linux-manager-root-key.pem -CAcreateserial \
-days 3650 -sha256 -extfile issuing_ca.cnf -extensions issuing_ca -out servers.pem
linux-manager-ca ca import servers servers.pem
The server cannot sign the root CRL without the key. Sign it where the key is, or put the key back for a moment, and take the CRL before the current one expires:
linux-manager-ca root import-key /root/linux-manager-root-key.pem # the passphrase on standard input
linux-manager-ca maintain
linux-manager-ca root remove-key
# or, with a CRL signed elsewhere:
linux-manager-ca root import-crl root.crl
An existing root CA of your own is used by placing its certificate, and its key if the server may
use it, as /etc/linux-manager/pki/root/cert.pem and key.pem before linux-manager-setup first runs
with a platform, as cloud-init can. Setup imports them into the directory and removes the files.
Without the key, the issuing CAs wait for their certificates as above.
ACME server
A general-purpose CA with "acme": true backs an ACME server of Caddy, while https.enabled, at
https://<wan.local_fqdn>:<https.web_port>/acme/<name>/directory, so servers get and renew
certificates with any ACME client:
certbot certonly --standalone -d web.example.com \
--server https://router.example.com:7444/acme/servers/directory
Caddy proves the client's control of the name as Let's Encrypt does, so it must reach the client.
The certificates are valid for 12 hours, carry Caddy's extensions, are not listed by
linux-manager-ca list, cannot be revoked, and no template applies to them. Caddy reads the CA's
certificate and key from /etc/caddy/linux-manager-pki/<name>/, readable by its group.
OpenVPN and users
With openvpn.enabled, an address_pool and wan.local_fqdn set, linux-manager-setup runs
openvpn-server@linux-manager with /etc/openvpn/server/linux-manager.conf. The server uses the
server's VPN server certificate, the root and the P2S CA and their CRLs (see Certificates),
pushes routes to local.cidrs and the server's addresses of p2s.dns_listen as DNS servers, and
opens openvpn.port on the external interface.
A client needs both a certificate of the P2S CA and the password of the user the certificate names:
- the certificate is checked against the root and the P2S CA and their CRLs (
crl-verify, read again when the file changes), and/usr/lib/linux-manager/openvpn-issuer(tls-verify) refuses a certificate another issuing CA signed; - the password is checked by
openvpn-auth-ldap, which binds to the user directory as the user; a user enrolled for TOTP types the password followed by the current 6-digit code; /usr/lib/linux-manager/openvpn-same-userrefuses a login whose user name is not the common name of the certificate.
The user directory
Once a platform is chosen, linux-manager-setup keeps a directory of users in the local slapd,
whether or not OpenVPN or IKEv2 access is enabled, so users can be managed before either
is switched on:
- its own database,
dc=linux-manager,dc=localin/var/lib/ldap/linux-manager, next to the one the slapd package creates; root manages it through the local socket, with no administrator password; ou=usersholds one entry per user, named byuid: aninetOrgPersonfor a VPN user, anaccountfor a Manager-only account (see Managing users);- the package's own schema,
cn=linux-manager,cn=schema,cn=config, under the OID1.3.6.1.4.1.67380.10.1.2, listed inOID-REGISTRY.txt: the auxiliary classlinuxManagerAccount, which Manager-only and service accounts have, and the attributelinuxManagerScope, authorization scopes, of whichservicemarks a service account;displayNameanddescriptionare the display name and the description of a service account, andlinuxManagerRedirectUri,linuxManagerPostLogoutRedirectUriandlinuxManagerAllowedGroupthe settings of a service account that is a client of the OpenID Provider; ou=groupsholds onegroupOfNamesper group, named bycn, with its members' DNs inmemberand an optionaldescription;cn=user-manager,ou=servicesis the account that changes users and groups; it may writeou=usersandou=groupsand set passwords, never read them, and its password is in/etc/linux-manager/ldap/user-manager.secret(root only);- passwords are stored as ARGON2 hashes, and a disabled user is locked with the password policy
(
pwdAccountLockedTime); - the password policy locks an account for 15 minutes after 5 failed binds within 15 minutes,
whether they come from OpenVPN or from
/auth; a successful bind clears the count, andlinux-manager-user unlockends a lockout; - slapd's
otpoverlay is configured on the database; a user enrolled for TOTP is also anoathTOTPUserandoathTOTPTokenwhoseoathTOTPTokennames the entry itself, with the secret inoathSecret, which the user manager account may set and no one can read, andoathTOTPParamsnamingcn=totp,ou=policies(6 digits, HMAC-SHA1, 30 second steps, one step of tolerance). A bind of an enrolled user takes the password followed by a current code; - slapd listens on its local socket only (
SLAPD_SERVICESin/etc/default/slapd).
The directory is kept on purge.
Managing users
linux-manager-user manages the users. A user is a directory entry of one of three kinds, fixed when
it is added:
- a VPN user, an
inetOrgPersonwith a first name, a last name and an e-mail address (givenName,snandmail;cnis the first and the last name), and a client certificate with the user name as its common name, the entry'suserCertificate, whose key the PKI keeps; - a Manager-only account, an
account,simpleSecurityObjectandlinuxManagerAccountwith the name, the password and an optional e-mail address, for notifications: no names and no certificate, so no VPN access. It can sign in to/authand, as a member ofadmins, manage the server.addwithout the first and the last name adds one; its names cannot be set later; - a service account, a Manager-only account whose
linuxManagerScopeisservice, which another system signs in with to use a router service, such as an OpenTelemetry Collector reading/metrics; it cannot sign in to/auth. Besides its name, it has a display name and a description, a service account with redirect URIs or grants is a client of the OpenID Provider, and every one is a resource of it, with its scopes and roles (see OpenID Provider).add --serviceadds one, and the web service manages them at/service-accounts, not at/users. Internal service accounts are those the server creates and manages for its own services, display name and description included; at present that is the metrics service's, the accountmetrics.service_accountnames while metrics are enabled (see Metrics), whichlinux-manager-setupnames Metrics. The API only lists them:/service-accountsrefuses every other request on one, and/groupsrefuses to put one in or take one out of a group, with 403.
Passwords and TOTP secrets are read from the first line of standard input. The answer is text: a
table of the users, the client profile, a TOTP secret or a short message, and a failure is a
message on standard error.
--json, before the command, gives the answer as JSON on standard output, and a failure as
{"error": ...} there too; the web service uses it. -h or --help shows the commands, and
after a command its options.
echo 's3cr3t-pass' | linux-manager-user add alice --first-name Alice --last-name Smith --email alice@example.com
echo 'm4nager-pass' | linux-manager-user add ops
echo 's3rvice-pass' | linux-manager-user add backup --service
linux-manager-user update alice --email a.smith@example.com
linux-manager-user list
linux-manager-user profile alice > alice.ovpn
linux-manager-user --json list
echo 'n3w-pass' | linux-manager-user passwd alice
linux-manager-user disable alice
linux-manager-user enable alice
linux-manager-user unlock alice
linux-manager-user remove alice
linux-manager-user otp-new alice
echo 'JBSWY3DPEHPK3PXPJBSWY3DPEHPK3PXP' | linux-manager-user otp-enrol alice 123456
linux-manager-user otp-remove alice
| Command | Effect |
|---|---|
list |
The users with their kind, names and e-mail addresses, whether each is enabled or locked, has a certificate and is enrolled for TOTP, the number of passkeys and the console login |
add <name> --first-name F --last-name L --email E |
Adds a VPN user and issues the certificate |
add <name> [--email E] |
Adds a Manager-only account, with the e-mail address when given; one name without the other two options is refused |
add <name> --service [--display-name N] [--description D] [--email E] [--redirect-uri U ...] [--post-logout-redirect-uri U ...] [--allowed-group G ...] [--grant G ...] [--public-client yes|no] [--client-scope R:S ...] [--resource-scope S ...] |
Adds a service account, with its display name, description, e-mail address and OpenID Provider settings when given; the names are refused |
update <name> [--first-name F] [--last-name L] [--email E] |
Changes the names or the e-mail address |
update <name> [--display-name N] [--description D] [--email E] [--redirect-uri U ...] [--post-logout-redirect-uri U ...] [--allowed-group G ...] [--grant G ...] [--public-client yes|no] [--client-scope R:S ...] [--resource-scope S ...] |
Changes the display name, the description, the e-mail address or the OpenID Provider settings of a service account; '' removes one; a repeated option gives all values of its kind |
passwd <name> |
Sets the password |
disable <name>, enable <name> |
Locks the user out of OpenVPN and /auth, and lets the user in again |
unlock <name> |
Ends a lockout after failed logins and clears the failure count; a disabled user stays disabled |
remove <name> |
Takes the user out of its groups, revokes the certificate, renews the CRL and removes the user; refused while the user is the last member of a group |
profile <name> |
The OpenVPN client profile, with the CA, the certificate and the key inline |
otp-new <name> |
A new TOTP secret and its otpauth:// URI, issued by wan.local_fqdn, for an authenticator app; nothing is stored |
otp-enrol <name> <code> |
Stores the secret read from standard input when the code is a current code of it, replacing a secret the user has |
otp-remove <name> |
Removes the user's TOTP secret and ends the sessions of the user's own POSIX account |
console <name> (--posix | --account LOCAL | --off) |
Switches the console login of a member of admins enrolled for TOTP on, as the user's own POSIX account or a local account, or off (see Console login and SSH) |
passkeys-remove <name> |
Removes all the user's passkeys (see Passkeys) |
A user name is 1 to 32 lowercase letters, digits and hyphens, starting with a letter, and a
service account may instead be named by a GUID in lowercase, such as
3f2b8c1e-9a4d-4e6f-8b2a-1c5d7e9f0a3b; a
password has 8 to 128 characters; a first or last name has 1 to 64 printable characters; an
e-mail address has one @, no spaces and a dot in its domain; a TOTP secret is 32 base32
characters, as otp-new makes it, and a code 6 digits. Once enrolled, a user logs in to OpenVPN
and /auth with the password followed by a current code. In the answers manager_only, service
and internal tell the kind, and a Manager-only account has its names as null; update changes only its e-mail
address, and --email '' removes it. Changing a user's password, disabling, unlocking or removing the user also revokes its
refresh tokens of /auth and its codes, tokens and sessions of the OpenID Provider. The exit status is 0 on success, 2 for an invalid request, 3 when the
user does not exist, 4 when the user exists and 1 for any other failure.
When the package is installed with a Ubuntu Linux Manager administrator (the debconf questions
linux-manager/admin_username and admin_password), it leaves the answers in
/etc/linux-manager/manager-admin.json (root only). linux-manager-setup then creates that user as a
Manager-only account, adds it to the group admins, which it creates when missing, and removes
the file.
The certificate is also accepted by IKEv2 (EAP-TLS) access, which checks
certificates only, so disable does not affect it; remove revokes the certificate for both.
Apple VPN profile
While p2s.enabled is true and wan.local_fqdn is set, /users/<name>/apple-profile and a
user's own /account/apple-profile give an unsigned Apple configuration profile (.mobileconfig)
that sets up IKEv2 access in the built-in VPN client of macOS, iOS and iPadOS. It holds:
- the P2S CA and the root CA, the root installed as a trusted root;
- the user's certificate and key as a PKCS#12 encrypted with the password of the request, with 3DES and SHA-1, the only encryption Apple's clients read; the device asks for the password when the profile is installed, and the server does not keep it;
- an IKEv2 VPN to
wan.local_fqdn, with the user name as the local identity, the certificate over EAP-TLS, and AES-256, SHA2-256 and group 14 for IKE and ESP, as the server's IKEv2 connection (road-warrior) takes them; - with
p2s.dns_domains, the server's resolver for those domains and their subdomains (SupplementalMatchDomains). macOS and iOS use the DNS server of a VPN that does not carry all traffic only for its match domains.
The password is 8 to 128 characters. A user asking for the own profile also gives the current password and, when enrolled, a code.
Console login and SSH
A member of admins enrolled for TOTP may have console login: the Manager's Shell page
(see The shell), and, for the user's own POSIX account, SSH and sudo. Only root on the server
switches it, with linux-manager-user console or a token of linux-manager-token, as for the group
admins. Switching it on is refused without the enrolment, and a user whose enrolment is removed
is refused the shell, SSH and sudo until enrolling again; the user's open shells end within a
minute, and removing the enrolment ends the sessions of the user's own POSIX account:
linux-manager-user console alice --posix
linux-manager-user console bob --account azureuser
linux-manager-user console alice --off
linux-manager-config-edit set ssh.ldap_login true
- Own POSIX account (
--posix): the user's entry becomes aposixAccountof the directory with the user's name, a uid from 20000 up, the local grouplinux-manager-consoleas its group,/home/<name>and/bin/bash.nslcdmakes it known to the system over slapd's local socket and checks its password by binding as the user, so the user types the password followed by a current code, failed logins count towards the lockout, and a disabled user cannot sign in. The home is created at the first login and kept when the user is removed. Members oflinux-manager-consolemay run any command with sudo, with the same password. A local account of the same name is refused. - Local account (
--account): the console runs as an existing local account with a uid from 1000 to 19999, such as the VM's admin user; its SSH keys and sudo rights stay as they are. - Off (
--off): the system no longer knows the user's own POSIX account, and its sessions end; its uid is kept for the next time. Leavingadmins, or removing or renaming the group, switches the console login of the members off.
While ssh.ldap_login is true, setup lets sshd accept the password of the members of
linux-manager-console (/etc/ssh/sshd_config.d/linux-manager.conf); other accounts keep sshd's
settings. A password change is made in the Manager or with linux-manager-user passwd, not
with passwd.
Passkeys
A user who is not a service account may sign in to the Manager and to the OpenID Provider
with a passkey instead of the password and code. Passkeys are for those sign-ins only: OpenVPN,
the shell, SSH and sudo keep the password and code. A passkey belongs to wan.local_fqdn, so the
browser offers it on the HTTPS addresses on that name, and passkeys need the name set; a passkey made
before the name changes is no longer offered. The authenticator checks the user (a PIN or a
fingerprint), and the passkey is kept by it or by the user's password manager.
A user adds a passkey on the Manager's Account page with the current password and code,
and removes it there. Each passkey is a value of linuxManagerPasskey on the user's entry, with its
public key, label and the time it was added; a user has at most 10. An administrator removes all
passkeys of a user on the Users page, through /users, or with:
linux-manager-user passkeys-remove alice
A disabled or locked user is refused a passkey sign-in too.
Managing groups
linux-manager-group manages the groups. A group has a name, with the rules of a user name, an
optional description of at most 256 printable characters, and at least one member, each a user
of the directory. The group admins grants the management of the server, and the allowed groups
of a client of the OpenID Provider admit their members to it, and a role of a resource assigned to
a group is a role of its members; other groups record membership only. Renaming a group renames it
in the clients that allow it and in its roles, a group a client allows cannot be removed, and
removing a group takes it out of its roles. The output, --json, -h and --help work as for linux-manager-user.
linux-manager-group add ops --member alice --member bob --description 'Operations team'
linux-manager-group list
linux-manager-group update ops --name netops
linux-manager-group update netops --description ''
linux-manager-group member-add netops carol
linux-manager-group member-remove netops bob
linux-manager-group remove netops
| Command | Effect |
|---|---|
list |
The groups with their members and descriptions |
add <name> --member U [--member U ...] [--description D] |
Adds the group with its members |
update <name> [--name N] [--description D] |
Renames the group or sets its description; '' removes it |
remove <name> |
Removes the group; its members are kept; refused while a client of the OpenID Provider allows it |
member-add <name> <user> |
Adds the user to the group |
member-remove <name> <user> |
Takes the user out of the group; refused for the last member |
The exit status is 0 on success, 2 for an invalid request, 3 when the group or the user does not exist, 4 when the group exists and 1 for any other failure.
Platform modules
Platform-specific values do not belong in the core, so they are Python modules in
/usr/lib/linux-manager/linuxmanager_platforms/, imported at run time and selected by
general.platform. The platform is never probed: the caller already knows where it is
deploying.
none is not a module. It is the shipped default and means "configure nothing at all",
which is how a deferred install is expressed. generic is the module to choose when the
router should be configured but the platform adds nothing.
The shipped modules currently add nothing. IKEv2, OpenVPN and WireGuard clients are given the
server's addresses of p2s.dns_listen as their DNS servers, by default the internal address,
the WireGuard clients in the configuration the Manager makes for a device (DNS =),
where linux-manager-resolver.service, dnsmasq with /etc/linux-manager/dnsmasq.conf, answers and
the firewall opens port 53 for the client pools and the WireGuard network, so no provider resolver has to be advertised or
routed through the tunnel on any platform. dnsmasq forwards to systemd-resolved on
127.0.0.53, which keeps only its loopback stubs, and forwards no name without a dot, so the
server's own host name, _gateway and localhost are not answered to the clients. The clients reach
the internal address when it is inside local.cidrs, which they route through the tunnel. Port 53
of the other addresses stays free for the name server (see Name server).
The modules exist as the documented place for platform specifics if that changes.
A module may define any of the following, and takes the default for anything it omits:
| Name | Type | Effect |
|---|---|---|
EXTRA_P2S_DNS |
str |
added to the dns list of the IKEv2 address pool |
EXTRA_LOCAL_TS |
str |
added to local_ts on both connections |
apply(ctx) |
function | called once the system is configured, to apply platform-specific state; must be idempotent |
ctx is the template context, so a module reads what it needs from it -
ctx['wan_iface'], ctx['int_iface'], ctx['local_cidrs'] and the rest.
Adding a platform is adding one file. Modules are discovered by listing the package, so nothing in the core needs editing:
# /usr/lib/linux-manager/linuxmanager_platforms/example.py
"""Example provider."""
EXTRA_P2S_DNS = '192.0.2.53'
EXTRA_LOCAL_TS = '192.0.2.53/32'
Set general.platform to example and restart the service. The debconf question offers only the
modules shipped with the package; the configuration file accepts any module installed.
Status web service
The linux-manager-web service shows the state of the server on the addresses of the interfaces
web.listen names, the internal, the external or both (the default), and on 127.0.0.1, where
Caddy calls it (see HTTPS), with web.port (7081 by default). Its pages / and /status need no authentication and show no
configuration: no key, identity, subnet, or address but the client's. They show as much as
web.status_content says, read at each request:
status_content |
/ and /status show |
|---|---|
basic (default) |
The overall status (ok or degraded), the time and the Client IP |
standard |
Also the linux-manager version and the number of active services and of services in any other state, without their names |
full |
Also each service and its state, Docker, how many site-to-site connections are set up and how many of them are established (site_to_site: connections and established, null when swanctl cannot be asked), IKEv2, OpenVPN and WireGuard |
The Client IP is the address the request came from: the last entry of X-Forwarded-For for a
request from one of web.trusted_proxies, by default the address the request came in on, so
behind Caddy or a Ubuntu Linux Manager on the server it is the browser's. /status names the content in
content and the Client IP in client_ip; / keeps the time and the link to /status.
The service first makes its signing key, the OpenID Provider's key and the refresh tokens' database, retrying every 15 seconds, for example while PostgreSQL is not running yet, and logs each failure; then it listens.
The service is /usr/lib/linux-manager/web, which starts it, and the package
/usr/lib/linux-manager/linuxmanager_web/, which answers the requests:
handler: reads each request, checks its token, and passes it to the module of its path.common: what the modules share.- One module per family of paths:
status,auth,identities,connectivity,backup,pki,logs,sites,dns,metricsandshell.
Bootstrap diagnostics
Until a run of linux-manager-setup has succeeded, every run starts linux-manager-bootstrap, which
serves diagnostics as plain text on port 17081 of every address, and the firewall opens 17081
on the internal interface. The first successful run stops
the bootstrap service and closes its port, and writes /var/lib/linux-manager/setup.done, after
which it never runs again; a failed run leaves it running, so the reason stays readable until
a later run succeeds. The unit is never enabled.
The page, read again every 10 seconds, shows the state, the step and the failure of the last
setup run (/run/linux-manager/setup.json, which setup writes at every step), the state of the
server's services, the failed units, the addresses and routes, cloud-init status, and the end of
the logs of linux-manager-setup, linux-manager-web and linux-manager-ui:
curl http://10.0.0.4:17081/
| Page | Content |
|---|---|
/ |
The status for people: plain text for curl and wget, a page styled with Tailwind CSS for a browser, in the browser's light or dark scheme, that reloads itself every minute (the style sheet is loaded from a CDN, so the browser needs internet access; the text is readable without it) |
/status |
The same status as JSON |
/auth |
Signing in with a user's password and TOTP code, refreshing and signing out |
/config |
The configuration without its secret options, and what the server resolved and runs, as JSON; needs a token |
/users |
The users of OpenVPN and IKEv2 access; needs a token |
/service-accounts |
The service accounts; needs a token |
/groups |
The groups of users; needs a token |
/site-to-site |
The site-to-site connections: list, add, change and remove; needs a token |
/account |
The account of the signed-in user: its details, its password and its TOTP enrolment; needs a user's token |
/logs |
The last lines of the journal of the server's units and of the files of web.log_files; needs a token |
/sites |
The web sites and their files; needs a token |
/dns |
The name server's settings and records; needs a token |
/.well-known/openid-configuration, /oidc/... |
The OpenID Provider, on 127.0.0.1 only, where Caddy passes the requests (see OpenID Provider) |
The status holds the state of linux-manager-setup, slapd and each PostgreSQL cluster unit
(postgresql@<version>-<cluster>, or postgresql as no cluster without one), of
linux-manager-ui, of strongswan when site-to-site or
IKEv2 access is configured, of wg-quick@wg0 when WireGuard is, and of
openvpn-server@linux-manager when OpenVPN is; the number of connected
OpenVPN clients, from the server's status file, which OpenVPN rewrites every 5 seconds; the package version; whether
Docker is installed, and when it is, the state of docker.service and of linux-manager-docker
(the Docker rules), which do not change status; as site_to_site, the number of site-to-site
connections that are set up (connections) and of those established (established, null when
swanctl cannot be asked), without their names; the number of connected IKEv2 clients; and
the WireGuard peers with the age of their last handshake and their traffic. status is ok when
all these services are active, and degraded otherwise.
curl http://10.0.0.4:7081/
curl http://10.0.0.4:7081/status
Tokens and signing in
/config, /metrics/history, /metrics/settings, /https/settings, /users, /service-accounts, /groups, /site-to-site, /backup, /remote-access, /ca, /logs, /sites, /dns, /shell and /account need a JWT in an Authorization: Bearer header, signed with
HS256 by the key the web service makes on its first run, /etc/linux-manager/web.key (root only).
The key never leaves the server; while it does not exist these paths and /auth are not served
(404). A missing, expired or wrongly signed token, or one of another algorithm, gets 401 with
WWW-Authenticate: Bearer. There are two kinds of tokens:
- an access token of
/authnames a user insuband is valid for 5 minutes; every user may call/account, and for the other paths the user must be a member of the groupadmins, otherwise the request gets 403; - a token of
linux-manager-token, run as root on the server, has the claimadminand is valid for the lifetime given, without a maximum. It names no user, so/accountrefuses it with 403; it may call everything else, and only such a token may change the groupadmins: create it, rename a group to it, change or remove it, add or take out its members, or remove a user who is a member of it.
The first member of admins is added on the server:
linux-manager-group add admins --member alice --description 'Router administrators'
linux-manager-token --hours 1
linux-manager-token --json --days 30
POST /auth signs a user in. Request and answer bodies are JSON:
| Request | Body | Answer |
|---|---|---|
POST /auth |
{"name", "password", "code"}; code only for a user enrolled for TOTP |
the tokens and "name" |
POST /auth/passkey |
{} |
{"id", "options"}: a pending sign-in, valid 2 minutes, and the options of navigator.credentials.get in their JSON form |
POST /auth/passkey/complete |
{"id", "credential"}, credential the browser's answer as its toJSON() gives it |
the tokens and "name", the user whose passkey signed |
POST /auth/refresh |
{"refresh_token"} |
new tokens; the refresh token given is used up |
POST /auth/logout |
{"refresh_token"} |
{"revoked": true}, or false for a token not known |
The tokens are {"access_token", "token_type": "Bearer", "expires_in": 300, "refresh_token", "refresh_expires_in": 28800}. A wrong name, password or code, and a locked or disabled user,
get 401 with the same message; a refresh token is refused once its user is disabled, locked or
removed. Changing a user's password, disabling, unlocking or removing the user revokes the
user's refresh tokens. Five failed logins lock the account for 15 minutes (see the user
directory). A passkey that does not sign, is not the user's, or of a disabled, locked or service
account gets 401 too (see Passkeys). /auth, /auth/passkey/complete and /auth/refresh take
at most 10 calls a minute from one client address,
and answer 429 with Retry-After beyond that; for a call from one of web.trusted_proxies, by
default the address the call came in on, the client address is the last entry of
X-Forwarded-For.
Refresh tokens are kept, as their SHA-256, in PostgreSQL: by default in the database and role
linux-manager on the server, which the web service creates on its first run with a password in
/etc/linux-manager/database.secret (root only), reached over TCP on localhost; web.database
names another database. Tokens and passwords travel over plain HTTP on the internal network.
curl -H 'Content-Type: application/json' -d '{"name": "alice", "password": "s3cr3t-pass", "code": "123456"}' \
http://10.0.0.4:7081/auth
curl -H "Authorization: Bearer $TOKEN" http://10.0.0.4:7081/config
The /status/full page
/status/full, with the same token as /config, shows the whole status document, as /status
does with web.status_content full, whatever that setting says, so the Manager shows
every service and tunnel while the public status page shows less.
The shell
While web.shell is true, /shell gives a member of admins whose console login is on and
who is enrolled for TOTP (see Console login and SSH) a login shell on the server, as the user's
console account, for the Manager's Shell page. It is off by default and switched on and
off on the server only:
linux-manager-user console alice --posix
linux-manager-config-edit set web.shell true
| Request | Body | Answer |
|---|---|---|
POST /shell/passkey |
{} |
{"id", "options"}, options those of navigator.credentials.get for the user's own passkeys |
POST /shell |
{"password", "code"}, or {"passkey": {"id", "credential"}}; {} within web.shell_auth_timeout |
{"ticket", "expires_in": 30} |
GET /shell?ticket=<ticket> |
a WebSocket handshake | the terminal |
POST /shell needs an access token of /auth, not one of linux-manager-token, and signs its user
in again, with the password and the TOTP code, or with a passkey: POST /shell/passkey begins a
proof for one of the user's own passkeys with user verification, which the browser answers within
two minutes, and POST /shell takes its id and the browser's answer as credential. A passkey
with user verification is two factors in one, so it takes the place of the password and the code.
A wrong password or code gets 403 and counts as a failed login, and a passkey that is not accepted
gets 403. POST /shell gets 403 too while web.shell is false, the user's console login is off
or the user is not enrolled for TOTP: also with a passkey, the PAM login session of the shell,
pam_authz_search of nslcd for a directory account, and sudo need the enrolment. Without a
password or a passkey it gives a ticket while the user's last sign-in of
POST /shell or keystroke in a shell is less than web.shell_auth_timeout minutes ago, and
{"ticket": null} otherwise. The ticket opens one
shell for the same user within 30 seconds. Over the WebSocket, binary messages of the client
are keystrokes, a text message {"cols", "rows"} sets the window size, and binary messages of
the server are the terminal's output. The shell is a PAM login session of the console account,
run by systemd-run in a transient unit linux-manager-shell-<id>; sudo works as for SSH, with the
password and the TOTP code. It is closed after web.shell_idle_minutes without a keystroke, when
the shell exits, when the connection closes, or within a minute of the user's console login being
switched off or the TOTP enrolment removed; the server then sends a close frame with the reason. The
journal of linux-manager-web records who opened each shell, from which address, in which unit,
when it ended and why, and each refused ticket or shell with the reason.
The editor of the shell
In the shell, linux-manager-editor FILE..., or its short name lme, opens files in the editor of
the Manager's Shell page and returns at once:
lme /etc/linux-manager/linux-manager.json ~/notes.txt
- How the files reach the page: each shell has its session in
LINUX_MANAGER_SHELL.lmenames each file, as an absolute path, with that session on the socket/run/linux-manager/editor.sock. The web service accepts it only from a process that runs as the shell's console account, and sends the page a text message{"editor": "open", "session", "id", "path"}over the shell's WebSocket. - Reading and saving: the page uses the routes below, which need the user's access token. The
file is read and written as the console account by
/usr/lib/linux-manager/editor-file, run withsystemd-run --uid --pipe, so the account's permissions apply. A save writes the file in place, keeping its owner and mode, and creates a file that does not exist. - Limits: only regular files of UTF-8 text up to 1 MiB, and at most 20 open files per shell.
- Lifetime: the files are open while the shell lives.
lmeoutside the Manager's shell exits with status 2.
| Request | Body | Answer |
|---|---|---|
GET /shell/editor/<session>/<id> |
{"path", "content", "version"}, version the SHA-256 of the content |
|
PUT /shell/editor/<session>/<id> |
{"content", "version"} |
{"path", "version"}; 409 when the file no longer has version |
DELETE /shell/editor/<session>/<id> |
{"path", "closed": true} |
The /config page
/config shows what / and /status leave out, for a client such as Cloud Manager:
| Key | Content |
|---|---|
configuration |
linux-manager.json with the defaults applied, without its secret options at any level (psk, database, and any option named like a secret, password, token or key) |
network |
The external and internal interface, their addresses, the internal gateway, local.cidrs, site_to_site: each connection with name, local_identity, remote_identity, local_cidrs and remote_cidrs, interfaces: each interface but loopback with name, state, mac, ipv4, its IPv4 addresses, and ipv6, its global IPv6 addresses, as address/prefix, null when ip fails, and the IPv4 routes as ip -j -4 route prints them |
ipsec |
connections and sas: one object per connection and per IKE SA, read from swanctl --list-conns --pretty and swanctl --list-sas --pretty, with the connection's name under name and swanctl's keys and string values |
wireguard |
The public key and port of wg0, and each peer's public key, endpoint, allowed IPs, handshake age and traffic; no private or preshared key |
docker |
docker.exposed_ports and the rules of the LINUX-MANAGER-DOCKER chain |
postgresql |
clusters: each PostgreSQL cluster of pg_lsclusters with its version, name, port, status and data directory, and for one that is online its settings: server_version, listen_addresses, port, max_connections, shared_buffers, effective_cache_size, work_mem, password_encryption, ssl and timezone; null when pg_lsclusters fails |
caddy |
The installed Caddy: version, source, upstream once Caddy's repository is configured and ubuntu otherwise, and service, the state of caddy.service; null when Caddy is not installed |
oidc |
The OpenID Provider: enabled, and issuer, its issuer URL, null while it is disabled |
The /postgresql page
/postgresql, with the same token as /config, shows the databases of the PostgreSQL clusters on the
router:
- Content:
clusterslists each cluster with its version, name, port and status, itsdatabases(name, owner, encoding and size in bytes,pg_database_size) and their totalsize. - A cluster that is not online:
databasesandsizearenull. - Collection: the web service reads it with
psqlas thepostgresuser at each request.
The /logs API
/logs, with the same token as /config, shows the last lines of the journal and of log files,
read at each request:
| Request | Answer |
|---|---|
GET /logs |
{"units", "files", "max_lines"}: the units whose journal it shows, the files of web.log_files and the most lines one request answers, 1000 |
GET /logs/journal?unit=<unit>&lines=<n> |
{"lines": [...]}: the last lines of the unit's journal, or of the whole journal without unit, with ISO times |
GET /logs/file?path=<path>&lines=<n> |
{"lines": [...]}: the last lines of a file web.log_files lists |
The units are the package's own (linux-manager-setup, -web, -metrics, -pki, -backup,
-caddy, -resolver, -webdav), linux-manager-ui, caddy, slapd, nslcd, postgresql,
strongswan, wg-quick@wg0, openvpn-server@linux-manager, ssh and pdns. lines is 1 to 1000, 200 without it. A
file is read from at most its last MiB, so very long lines give fewer lines. A unit or a file not
in the lists is 404, as is a listed file that does not exist.
linux-manager-config-edit set web.log_files /var/log/syslog,/var/log/auth.log
The /sites API
/sites, with the same token as /config, manages the web sites, with the module
linux-manager-site uses too, /usr/lib/linux-manager/linuxmanager_sites.py:
| Request | Answer |
|---|---|
GET /sites |
{"settings", "sites"}: the sites section and the sites, each with its settings but private_key, with private_key_set, modified and directory |
PUT /sites |
Changes any of the settings of the sites section |
POST /sites |
Adds a site: {"name", ...} with any of its settings; an existing name is 409 |
GET /sites/<name> |
The site |
PUT /sites/<name> |
Changes any of its settings; an empty or absent private_key keeps the key |
DELETE /sites/<name>?content=true |
Removes the site, with content=true also its folder |
GET /sites/<name>/files/<path> |
A folder: {"path", "entries": [{"name", "folder", "size", "modified"}]}; a file: the file, as an attachment |
PUT /sites/<name>/files/<path> |
Writes the body to the file, creating its folders; with ?unzip=true the body is a zip archive extracted into the folder of the path: {"path", "files"} |
POST /sites/<name>/files/<path> |
{"folder": true} creates the folder; {"rename": "<path>"} moves the file or folder |
DELETE /sites/<name>/files/<path> |
Removes the file or folder |
The paths of /files/ are relative to the site's folder and may not leave it; a PUT needs a
Content-Length. The files of a site may also be used by a member of its editors group. Invalid
settings are 400, a missing site 404.
curl -X PUT -H "Authorization: Bearer $TOKEN" --data-binary @site.zip \
'https://router.example.com:7444/sites/www/files/?unzip=true'
The /dns API
/dns, with the same token as /config:
| Request | Answer |
|---|---|
GET /dns |
{"settings", "nameserver", "public_address", "installed", "running", "domains", "problem"}: the dns section, wan.local_fqdn, the primary name server, wan.public_address, whether PowerDNS is installed and runs, the domains, each {"name", "ttl", "manual", "records"} with its manual records and the records it serves, and why they cannot be read, null when they can |
PUT /dns |
Changes any of the settings of the dns section, and with public_address wan.public_address; answers {"settings", "public_address"} |
POST /dns/domains |
Adds a domain: {"name", "ttl"?}; an existing name is 409 |
PUT /dns/domains/<domain> |
Changes ttl of the domain |
DELETE /dns/domains/<domain> |
Removes the domain with its manual records |
POST /dns/domains/<domain>/records |
Adds a manual record: {"name", "type", "content", "ttl"?}; name is relative to the domain, @ for the domain itself, or absolute with a final dot, and ttl is the domain's without it |
DELETE /dns/domains/<domain>/records?name=&type=&content= |
Removes the manual records of the name and type, only the one of content when it is given |
A TXT content without quotes is quoted; MX is <preference> <host>, SRV <priority> <weight> <port> <host> and CAA <flags> <tag> "<value>". A CNAME must be its name's only record.
The /users API
/users manages users with the same token as /config, with the module linux-manager-user uses
too, /usr/lib/linux-manager/linuxmanager_users.py, so both apply the same rules. Request
and answer bodies are JSON; a request body is at most 65536 bytes.
| Request | Body | Answer |
|---|---|---|
GET /users |
{"users": [{"name", "manager_only", "service", "internal", "enabled", "locked", "certificate", "otp", "console", "wireguard_peers", "passkeys", "first_name", "last_name", "email"}, ...], "wireguard": {"enabled", "address"}}; console is null or {"enabled", "account", "local", "uid"}, wireguard_peers a list of {"public_key", "ips"}, passkeys a list of {"id", "label", "created"}, and wireguard the endpoint's settings, whose network the peers' addresses come from |
|
POST /users |
{"name", "password", "first_name", "last_name", "email"} for a VPN user; {"name", "password", "email"?} for a Manager-only account |
201 and the user |
PUT /users/<name> |
any of {"first_name", "last_name", "email", "wireguard_peers"}; wireguard_peers, [{"public_key", "ips"}], replaces a VPN user's WireGuard peers (see Configuration); for a Manager-only account only {"email"}, "" removing it |
the user |
PUT /users/<name>/password |
{"password": ...} |
the user |
POST /users/<name>/disable |
the user | |
POST /users/<name>/enable |
the user | |
POST /users/<name>/unlock |
the user | |
DELETE /users/<name> |
{"name", "removed": true} |
|
GET /users/<name>/profile |
{"name", "profile"}, the OpenVPN client profile |
|
POST /users/<name>/apple-profile |
{"password"} |
{"name", "profile"}, the Apple configuration profile of IKEv2 access (see Apple VPN profile) |
POST /users/<name>/otp |
{"name", "secret", "uri"}, a new TOTP secret; nothing is stored |
|
PUT /users/<name>/otp |
{"secret", "code"} |
the user, once the code matches the secret |
DELETE /users/<name>/otp |
the user | |
DELETE /users/<name>/passkeys |
the user, without passkeys | |
PUT /users/<name>/console |
{"enabled", "account"?}; only with a token of linux-manager-token |
the user |
A failure is {"error": ...} with 400 for an invalid request, 404 for a user that does not
exist or is a service account, 409 for a user that exists and 500 otherwise. /users does not
list service accounts; /service-accounts serves them. /users works whether or not OpenVPN or
IKEv2 access is enabled; the profile needs OpenVPN. Passwords, private keys and TOTP
secrets cross the internal network in plain HTTP, like the token.
curl -H "Authorization: Bearer $TOKEN" -H 'Content-Type: application/json' \
-d '{"name": "alice", "password": "s3cr3t-pass", "first_name": "Alice", "last_name": "Smith", "email": "alice@example.com"}' \
http://10.0.0.4:7081/users
The /service-accounts API
/service-accounts manages the service accounts with the same token and the same module as
/users.
| Request | Body | Answer |
|---|---|---|
GET /service-accounts |
{"service_accounts": [{"name", "display_name", "description", "internal", "enabled", "locked", "email", "redirect_uris", "post_logout_redirect_uris", "allowed_groups", "grants", "public_client", "client_scopes", "resource_scopes", "roles", ...}, ...], "oidc": {"enabled", "issuer"}}, each as /users shows a user, and the issuer of the OpenID Provider, null while it is disabled |
|
POST /service-accounts |
{"name", "password", "display_name"?, "description"?, "email"?, "redirect_uris"?, "post_logout_redirect_uris"?, "allowed_groups"?, "grants"?, "public_client"?, "client_scopes"?, "resource_scopes"?, "roles"?} |
201 and the account |
PUT /service-accounts/<name> |
any of {"display_name", "description", "email", "redirect_uris", "post_logout_redirect_uris", "allowed_groups", "grants", "public_client", "client_scopes", "resource_scopes", "roles"}; "" removes one of the first three, [] the values of the lists, false the public client |
the account |
PUT /service-accounts/<name>/password |
{"password": ...} |
the account |
POST /service-accounts/<name>/disable |
the account | |
POST /service-accounts/<name>/enable |
the account | |
POST /service-accounts/<name>/unlock |
the account | |
DELETE /service-accounts/<name> |
{"name", "removed": true} |
A display name has at most 64 printable characters and a description at most 256. The settings of the OpenID Provider (see OpenID Provider) are:
- URIs and allowed groups: lists of at most 16 values: URIs of
https, or ofhttpforlocalhost,127.0.0.1or::1, without a fragment and of at most 1024 characters, and names of existing groups; grants:client_credentialsanddevice_code;public_client,trueorfalse;client_scopes: at most 16 scopes of resources,<resource>:<scope>;resource_scopes: at most 16 scope names of lowercase letters, digits, dots, underscores and hyphens;roles:[{"name", "description", "groups", "accounts"}], which replaces the account's roles: role names of letters, digits, dots, underscores and hyphens, and existing groups and accounts.
The internal service accounts, internal true, are only listed: every other request on one, and
a role assigned to one, gets 403, as the server manages them, their display name and description
included. A password change, disabling, unlocking and removing revoke the tokens the account was
issued. A failure is otherwise as for /users, with 404 also for a name that is not a service
account.
The /account API
/account serves the user the access token names, whether or not the user is a member of
admins, with the same module as /users:
| Request | Body | Answer |
|---|---|---|
GET /account |
the user as /users shows it, and "admin", whether the user is a member of admins |
|
PUT /account/password |
{"current_password", "current_code", "password"} |
the user |
POST /account/otp |
{"name", "secret", "uri"}, a new TOTP secret; nothing is stored |
|
PUT /account/otp |
{"current_password", "current_code", "secret", "code"} |
the user, once the code matches the secret |
POST /account/passkeys |
{"current_password", "current_code"} |
{"id", "options"}: a pending registration, valid 2 minutes, and the options of navigator.credentials.create in their JSON form |
PUT /account/passkeys |
{"id", "credential", "label"?}, credential the browser's answer as its toJSON() gives it; label at most 64 characters, Passkey without one |
the user |
DELETE /account/passkeys/<id> |
the user | |
POST /account/profile |
{"current_password", "current_code"} |
{"name", "profile"}, the user's OpenVPN client profile |
POST /account/apple-profile |
{"current_password", "current_code", "password"} |
{"name", "profile"}, the user's Apple configuration profile of IKEv2 access (see Apple VPN profile) |
current_code is needed only for a user enrolled for TOTP; removing a passkey needs neither. A wrong current password or code gets
403 and counts as a failed login; slapd accepts each code once, so a code used to sign in is not
accepted again, and a change right after signing in waits for the next code. A user enrols or
replaces the enrolment; removing it is left to /users and linux-manager-user. Changing the
password revokes the user's refresh tokens. Other failures are those of /users.
The /groups API
/groups manages groups with the same token and the same module as /users.
| Request | Body | Answer |
|---|---|---|
GET /groups |
{"groups": [{"name", "description", "members": [...]}, ...]} |
|
POST /groups |
{"name", "members": [...]}, and "description" |
201 and the group |
PUT /groups/<name> |
any of {"name", "description"}; "" removes the description |
the group |
DELETE /groups/<name> |
{"name", "removed": true} |
|
POST /groups/<name>/members |
{"user": ...} |
the group |
DELETE /groups/<name>/members/<user> |
the group |
A failure is {"error": ...} with 400 for an invalid request, among them taking out the last
member, 404 for a group or user that does not exist, 409 for a group that exists and 500
otherwise. DELETE /users/<name> is refused with 400 while the user is the last member of a
group.
The /site-to-site API
/site-to-site manages the site-to-site connections of linux-manager.json with the same token as
/users.
| Request | Body | Answer |
|---|---|---|
GET /site-to-site |
{"connections": [{"name", "addrs", "id", "cidrs", "local_cidrs", "local_id_mode", "auth", "psk_file", "remote_ca", "psk_set", "established"}, ...]} |
|
POST /site-to-site |
{"name", "addrs", "id", "cidrs", "local_cidrs", "local_id_mode", "auth", "psk", "remote_ca"}; psk is required with auth psk, the default |
201 and the connection |
PUT /site-to-site/<name> |
any of the settings of POST but name; an empty or absent psk keeps the key, a psk given replaces psk_file |
the connection |
DELETE /site-to-site/<name> |
{"name", "removed": true} |
A connection is shown without its key: psk_set says whether psk holds one, and established
whether its tunnel is up, null when swanctl cannot be asked. Each change is made under the lock
linux-manager-config-edit takes, checked as linux-manager-setup checks the configuration, written to
linux-manager.json, and applied by restarting linux-manager-setup without waiting for it. A failure
is {"error": ...} with 400 for an invalid request or configuration, with the reason, 404 for a
connection that does not exist, 409 for a name that exists and 500 otherwise. Keys cross the
internal network in plain HTTP, like the token.
The /backup API
/backup shows and changes the backup settings and starts a backup, with the same token as
/users.
| Request | Body | Answer |
|---|---|---|
GET /backup |
{"settings", "state", "archives", "next", "client", "rclone"} |
|
PUT /backup |
any of the settings of the backup section; an empty or absent pbs_password keeps the secret |
{"settings"} |
POST /backup/run |
202 and {"started": true}; 409 while a backup runs |
The answer of GET /backup holds:
settings: thebackupsection withoutpbs_password, withpbs_password_set.state: the last run,nullbefore the first. Itsresultisrunning,ok,warning(the backup was made but the cloud copy failed),failedorinterrupted(the run ended without recording its result). It also hasstarted,finished,target,archiveorsnapshot,sizein bytes,removed(the archives retention deleted),errorandcloud({"remote", "mode", "result", "error"}).archives: the local archives of this host, newest first, as{"name", "time", "size"}.next: the next run as systemd shows it,nullwhile backups are off.client: whetherproxmox-backup-clientis installed.rclone:{"installed", "config", "remotes"}, the remotes being the sections ofrclone.conf.
A change is made, checked and applied as for /site-to-site.
The /metrics/settings API
/metrics/settings shows and changes the settings of the metrics service, with the same token as
/users.
| Request | Body | Answer |
|---|---|---|
GET /metrics/settings |
{"settings"}: the metrics section without otlp_password, with otlp_password_set |
|
PUT /metrics/settings |
any of the settings of the metrics section; an empty or absent otlp_password keeps the secret, and an empty otlp_username removes it |
{"settings"} |
A change is made, checked and applied as for /site-to-site.
The /https/settings API
/https/settings shows the settings of HTTPS and changes the CA of Caddy's certificates, with the
same token as /users.
| Request | Body | Answer |
|---|---|---|
GET /https/settings |
{"settings"}: the https section |
|
PUT /https/settings |
{"acme_staging"}; any other setting of the https section is 400, as it is changed on the server with linux-manager-config-edit |
{"settings"} |
A change is made, checked and applied as for /site-to-site.
The /remote-access API
/remote-access shows and changes the remote access systems, IKEv2 (p2s),
OpenVPN (openvpn) and WireGuard (wireguard), with the same token as /users.
| Request | Body | Answer |
|---|---|---|
GET /remote-access |
{"p2s", "openvpn", "wireguard", "running", "local_fqdn", "wireguard_server"}; wireguard_server is {"public_key", "endpoint", "allowed_ips", "dns"}: the server's WireGuard public key (null before setup made it), <wan.local_fqdn>:<wireguard.listen_port> (null without a name), the networks a peer reaches through the tunnel, the server's local CIDRs and the WireGuard network, and the addresses of the remote access clients' resolver ([] while it does not run) |
|
PUT /remote-access/<system> |
any of the settings of the section p2s, openvpn or wireguard |
the section |
running tells for each system whether setup runs it. IKEv2 and OpenVPN run once enabled with an
address pool, wan.local_fqdn set and the server's VPN server certificate issued by the P2S CA;
WireGuard once enabled with an address. A system enabled without its address pool or address is
refused. A change is made, checked and applied as for /site-to-site. The server's firewall
follows the settings; a cloud firewall in front of the server, such as an Azure network security
group, must allow the OpenVPN and WireGuard ports itself.
The /ca API
/ca manages the PKI (see Certificates) with the same token as /users, with the module
linux-manager-ca uses too, /usr/lib/linux-manager/linuxmanager_pki.py.
| Request | Body | Answer |
|---|---|---|
GET /ca |
{"settings", "templates", "extended_key_usages", "status", "certificates"} |
|
PUT /ca |
any of the settings of the pki section |
{"settings", "templates"} |
POST /ca/certificates |
{"ca", "template", "csr"} |
201 and {"serial", "certificate", "chain"}, PEM |
POST /ca/certificates |
{"ca", "template", "subject", "sans", "password"} |
201 and {"serial", "pfx"}, a PKCS#12 file in base64 |
POST /ca/certificates/<serial>/revoke |
{"reason"}, optional |
the certificate, with reissue for a router certificate |
GET /ca/bundle/<name> |
{"name", "pem"}: the certificates a client trusts the CA with, the root alone for root, the issuing CA followed by the root for an issuing CA |
|
GET /ca/issuing/<name>/csr |
{"name", "csr", "extensions"} of an issuing CA waiting for the offline root |
|
PUT /ca/issuing/<name>/certificate |
{"certificate"}, PEM, signed by the root |
{"name", "subject", "expires"} |
PUT /ca/root/crl |
{"crl"}, PEM, signed by the root |
{"next_update"} |
The answer of GET /ca holds:
settings: thepkisection.templates: every template, the built-in ones included, with all its options.extended_key_usages: every extended key usage templates may give, name ->{"oid", "builtin"}.status:root(subject,expires,key, whether the root key is on the server,crl_next_updateandcrl_source,signedoruploaded),authorities, one per issuing CA (name,common_name,state:active,pendingormissing,builtin,templates,acme,subject,expires,key,crl_next_update,acme_url),retiredandwarnings.certificates: the certificates issued, newest first, each{"serial", "ca", "kind", "template", "subject", "sans", "created", "expires", "revoked", "reason", "state"};kindisca,router,userorexternal, andstatevalid,expiredorrevoked.
A change of the settings is made, checked and applied as for /site-to-site; an imported issuing
CA certificate and a revoked router certificate restart linux-manager-setup too. A failure is
{"error": ...} with 400 for an invalid request, 404 for something that does not exist, 409 when
the state does not allow it, as for a revoked certificate or an issuing CA that has its
certificate, and 500 otherwise. The root key never crosses the API; linux-manager-ca exports,
removes and imports it.
The service starts on its own, without waiting for linux-manager-setup, and keeps running. Every
30 seconds it reads web.listen, web.port and web.enabled from linux-manager.json and works
out the addresses of the interfaces web.listen names as general.mode directs; it listens on
each, moves when an address or the port changes, and stops listening while web.enabled is
false or no address is known. A change of web.listen takes effect within 30 seconds, or when
the service restarts. Each request reads the signing key and linux-manager.json, and the features
and the facts for /config from /run/linux-manager/web.json (readable by root only), which
linux-manager-setup writes; before that file exists the page shows no features.
linux-manager-setup opens web.port on the internal and the external interface in the firewall.
HTTPS
Caddy serves the status web service on https.web_port (7444 by
default), the Manager on https.manager_port (7443), and
while metrics are enabled the metrics service on metrics.https_port (9465), all for wan.local_fqdn, while https.enabled is true and wan.local_fqdn is set. It obtains
and renews the certificate from Let's Encrypt with the HTTP challenge, so wan.local_fqdn must
resolve to the server's public address and port 80 must be reachable from the internet; the
certificate and the ACME account are kept in /var/lib/caddy. Port 80 answers the challenge only;
it does not redirect to HTTPS. Caddy's default CAs are Let's Encrypt and ZeroSSL, which it takes
when Let's Encrypt refuses, such as for its rate limit of 5 certificates a week for the same name.
While https.acme_staging is true, Caddy's only CA is Let's Encrypt's staging environment, for
routers that are rebuilt often. Its rate limits are far higher, but no browser or client trusts
its certificates, so they are for tests only. The setting applies to every certificate Caddy
obtains from an ACME CA, those of the web sites with tls acme included.
linux-manager-config-edit set --restart https.acme_staging true
The Manager switches it on the Overview, and PUT /https/settings changes it (see The
/https/settings API).
linux-manager-setup writes Caddy's configuration to /etc/caddy/linux-manager.Caddyfile (root, group
caddy, the group Caddy runs as, mode 0640), which the
package's drop-in /usr/lib/systemd/system/caddy.service.d/linux-manager.conf makes caddy.service
read; caddy.service does not start before the file exists, and /etc/caddy/Caddyfile is not
used. Setup starts Caddy, reloads it when the file changes, and stops it and removes the file when
HTTPS is off and no web site is served (see Web sites). In the firewall it opens port 80 on the external interface and the HTTPS ports on the
internal and the external interface, and ports 80 and 443 on both when pki.publish has http-80
and https-default, on which Caddy serves only the CA files (see Certificates). Caddy calls both services on 127.0.0.1, where they listen
whatever web.listen and the manager's listen say, and passes the client's address in
X-Forwarded-For, which the manager trusts from 127.0.0.1 and the web service by default, as
the address the call came in on; a web.trusted_proxies that is set must list 127.0.0.1. The plain HTTP ports stay as
web.listen and the manager's listen set them. docker.exposed_ports may not take port 80 or
the HTTPS ports while https.enabled is true.
curl https://router.example.com:7444/status
Installing Caddy
The package does not depend on Caddy; linux-manager-setup has it installed, while HTTPS is on,
from the source https.caddy_source names:
ubuntu, the default: thecaddypackage of the Ubuntu archive.upstream: Caddy's own repository,https://dl.cloudsmith.io/public/caddy/stable/deb/debian, whose source the package writes to/etc/apt/sources.list.d/caddy-stable.sourceswith its key, downloaded over HTTPS, in/etc/apt/keyrings/caddy-stable.asc.
apt runs in linux-manager-caddy.service, which waits for dpkg's lock, so setup run by a package
installation does not wait for itself; the unit runs linux-manager-caddy apply and then setup
again, which serves with the new Caddy. linux-manager-caddy status shows the configured source, the
source in effect and the versions.
A router installed with ubuntu is switched to upstream's Caddy with:
linux-manager-caddy use-upstream
It sets https.caddy_source to upstream, configures the repository and replaces Caddy, printing
each step. Upstream versions sort above Ubuntu's, so once the repository is configured apt keeps
upstream's Caddy, and ubuntu has no effect; setup says so. Going back is a downgrade nothing
does on its own: remove the source and the key, run apt-get update, and install Ubuntu's version
with apt-get install --allow-downgrades caddy=<version>, the version apt-cache policy caddy
lists for the archive.
OpenID Provider
While oidc.enabled is true, the status web service is an OpenID Provider (OpenID Connect
Core 1.0) for the users of the directory. Its issuer is the web service's HTTPS site,
https://<wan.local_fqdn>:<https.web_port>, without the port when it is 443; the web service
answers it only on 127.0.0.1, where Caddy passes the requests, so the provider is never reached
over plain HTTP. The setting needs web.enabled, https.enabled and wan.local_fqdn.
linux-manager-config-edit set --restart oidc.enabled true
curl https://router.example.com:7444/.well-known/openid-configuration
| Endpoint | Purpose |
|---|---|
GET /.well-known/openid-configuration |
The discovery document |
GET /oidc/jwks |
The public key ID tokens and access tokens are signed with |
GET, POST /oidc/authorize |
The authorization code flow; the sign-in page while the browser has no session |
POST /oidc/token |
Exchanges a code, a refresh token, a device code or a client's credentials for tokens |
POST /oidc/device_authorization |
Begins the device authorization grant: a device code and a user code |
GET, POST /oidc/device |
The page where a user enters a device's user code, signs in and allows the device |
GET, POST /oidc/userinfo |
The claims of the user of an access token |
GET /oidc/logout |
Ends the session (RP-initiated logout) |
Clients
A client is a service account with a redirect URI or a grant. Its name is the client_id and its
password the client_secret, sent with HTTP Basic authentication (client_secret_basic) or in
the body of the token request (client_secret_post). The password is checked with a bind, so 5
wrong secrets lock the client for 15 minutes. A public client sends only its client_id (none).
Its settings are:
- redirect URIs: where the provider sends the browser back with a code; they give the client
the authorization code grant. A URI uses
https, orhttpforlocalhost,127.0.0.1or::1, and is matched exactly, but the port of anhttpone, which a native program chooses when it starts (RFC 8252); - post-logout redirect URIs: where
/oidc/logoutmay send the browser afterwards; - allowed groups: only their members may sign in to the client; without any, every user who is not a service account may;
- grants:
client_credentialsanddevice_code, besides the authorization code; - public client: the client has no secret, as a native program or a CLI, which cannot keep
one; it cannot have
client_credentials; - client scopes: the scopes of resources the client may ask for,
<resource>:<scope>.
echo 'cl1ent-s3cret' | linux-manager-user add 3f2b8c1e-9a4d-4e6f-8b2a-1c5d7e9f0a3b --service \
--display-name Grafana --redirect-uri https://grafana.example.com/login/generic_oauth \
--post-logout-redirect-uri https://grafana.example.com/login --allowed-group ops
linux-manager-user update 3f2b8c1e-9a4d-4e6f-8b2a-1c5d7e9f0a3b --allowed-group ''
Grants
| Grant | Who signs in | Tokens |
|---|---|---|
authorization_code |
A user, in the browser, with PKCE (S256) |
Access, refresh and ID token |
urn:ietf:params:oauth:grant-type:device_code |
A user, on another device | Access, refresh and ID token |
client_credentials |
The service account itself, with its secret | Access token |
refresh_token |
A new access and refresh token |
The implicit and hybrid flows, the password grant and dynamic registration are refused. A registered client is trusted, so a user is not asked for consent, except for a device.
Signing in
- The page: a user signs in on the provider's page with the name, the password and, when enrolled, the TOTP code, or with a passkey (see Passkeys).
- Lockout and rate limit: failed sign-ins count towards the lockout.
POST /oidc/login,/oidc/deviceand/oidc/device_authorizationtake at most 10 calls a minute from one client address, as/authdoes. - Who may not sign in: a disabled, locked or removed user, a service account, and a user outside the client's allowed groups. The codes and tokens already issued to such a user stop working.
- The session: the sign-in starts a cookie of the issuer's
/oidc/path, valid for 8 hours, so later requests of any client go on without the page.prompt=loginandmax_ageask for a new sign-in, andprompt=noneanswerslogin_requiredwithout a session.
Device authorization (RFC 8628)
A client with device_code, such as a CLI on a host without a browser, signs a user in like this:
- It posts
client_id(and its secret, unless it is public),scopeand optionallyresourceto/oidc/device_authorization. The answer holdsdevice_code,user_code(8 letters, shown asXXXX-XXXX),verification_uri,verification_uri_complete,expires_in(600 seconds) andinterval(5 seconds). - It shows the user the code and the URI, then polls
/oidc/tokenwithgrant_type=urn:ietf:params:oauth:grant-type:device_codeanddevice_code. Until the user decides, the answer isauthorization_pending; a poll sooner than the interval answersslow_downand adds 5 seconds to it. - The user opens the URI on any device, enters the code, signs in, and allows or denies the
client, the resource and the scopes shown. The next poll answers the tokens, or
access_denied; an expired code answersexpired_token.
curl -d client_id=backup-cli -d 'scope=openid profile' \
https://router.example.com:7444/oidc/device_authorization
Client credentials
A confidential client with client_credentials gets an access token as its own service account,
without a user. The scopes are those of a resource, and the profile scopes give the service
account's claims; openid is refused, and no refresh token is issued.
curl -u backup-job:'j0b-s3cret' -d grant_type=client_credentials -d resource=s3 -d scope=s3:read \
https://router.example.com:7444/oidc/token
Resources, roles and scopes
Every service account is also a resource, an application that checks the server's tokens, as an API registration of Entra ID:
- Scopes: the resource scopes the account defines, such as
read, which clients ask for as<resource>:<scope>(s3:read) when their client scopes allow them. - Roles: named roles, such as
Reader, each assigned to groups, users and service accounts;linux-manager-rolemanages them.
echo 's3-s3cret' | linux-manager-user add s3 --service --resource-scope read --resource-scope write
linux-manager-role add s3 Reader --description 'Reads every bucket'
linux-manager-role member-add s3 Reader --group backup
linux-manager-user update backup-job --grant client_credentials --client-scope s3:read
A request is for a resource when it names it with resource (RFC 8707), its name or
<issuer>/<name>, or asks for its scopes; more than one resource, or one that is not a
resource, is invalid_target. A scope the client may not ask for is invalid_scope. Without a
resource, the tokens are for the client itself. The resource is kept with the code, the device code
and the refresh token, and a token request naming another one is invalid_target.
| Command | What it does |
|---|---|
linux-manager-role list [<resource>] |
The roles with the groups and accounts they are assigned to |
linux-manager-role add <resource> <role> [--description D] |
Add a role |
linux-manager-role remove <resource> <role> |
Remove a role |
linux-manager-role member-add <resource> <role> --group G|--account A |
Assign the role |
linux-manager-role member-remove <resource> <role> --group G|--account A |
Take the role away |
A scope a client may ask for cannot be removed from its resource, nor the resource while a client may ask for its scopes. Removing a service account removes its roles and takes it out of the others; removing or renaming a group does the same in the roles.
Tokens
| Scope | Claims |
|---|---|
openid (required but by client_credentials, which refuses it) |
sub, the entryUUID of the user's entry |
profile |
preferred_username (the user name), name, given_name, family_name |
email |
email |
groups |
groups, the names of the user's groups |
<resource>:<scope> |
None; the scope is in the access token's scope |
- ID tokens: RS256 JWTs for the client (
aud), valid for 5 minutes, with the claims of the scopes,auth_time,amr(pwd,pwdandotp, orhwkanduserfor a passkey), androles, the user's roles of the client itself./oidc/userinfoanswers the same claims. - Access tokens: RS256 JWTs of RFC 9068 (
typat+jwt), valid for 5 minutes, which a resource verifies with the JWKS. Their claims:iss,sub,iat,expandjti;aud, the resource, or the client without one, andclient_id, the client;scope, the scopes granted;roles, the roles of the resource the user or the service account holds, directly or by a group;- the claims of the scopes, and of a user
auth_timeandamr.
- Roles are read at every issue, refresh included, so a role taken away is gone from the next token.
- The key: tokens are signed with an RSA key of 3072 bits that the web service makes on its
first run,
/etc/linux-manager/oidc.key(root only), and named in the JWKS by its RFC 7638 thumbprint. - Lifetimes: a code is valid for 1 minute and used once, a device code for 10 minutes, and a refresh token for 8 hours and used once: each refresh answers a new one.
- Storage and revocation: the codes, device codes, tokens and sessions are kept, as their
SHA-256, in the database of the refresh tokens of
/auth. Changing the password of a user or a service account, disabling, unlocking or removing it ends its sessions and revokes its codes and tokens.
Web sites
While sites.enabled is true, Caddy serves static web sites from folders on the server, each
/var/lib/linux-manager/sites/<name>. The sites are rows of the database linux-manager-sites, which
the web service creates on its first run with its own role and the password in
/etc/linux-manager/sites-database.secret; they are managed with linux-manager-site, /sites and the
Ubuntu Linux Manager, each change applied by restarting linux-manager-setup. A site has:
| Setting | Meaning |
|---|---|
name |
1 to 31 lowercase letters, digits and hyphens, starting with a letter; also the name of its folder |
enabled |
Whether Caddy serves it |
hosts |
The host names (or IP addresses) Caddy answers on the port, the names TLS clients send as SNI; none answers any name |
port |
The TCP port, 443 by default |
listen |
internal, external (the default) or both: the interfaces whose firewall opens the port |
tls |
none (plain HTTP), acme (Let's Encrypt, the default), ca (certificates of an issuing CA of pki.issuing_cas, named by ca) or file (certificate and private_key, PEM) |
index_file |
The file served for a folder, index.html by default |
spa |
Whether a path without a file is served the site's index_file, for single-page applications |
editors |
The group whose members may change the site's files besides the admins |
dns |
Who serves the records of its host names: user (the default), a DNS provider outside the server, or router, the server's name server (see Name server) |
dns_address |
With dns router, the server's address in the records: public (the default, wan.public_address), internal or external |
A site may not take a port of the server's own services (SSH, web.port, the HTTPS ports, the
metrics ports, the Manager's, sites.webdav_port, 7446, 53, the bootstrap ports, a TCP
openvpn.port and docker.exposed_ports). It may use wan.local_fqdn as a host name, on a port
of its own, such as 443, which the server's services leave free; its record is then kept where the
domain of wan.local_fqdn is, so dns router takes neither it, nor wildcards or addresses, and
public needs wan.public_address. Enabled sites on one port need host
names of their own, the same listen and the same kind of TLS (plain HTTP or HTTPS). A site of
acme needs host names, without wildcards or addresses, that resolve to the server, and port 80,
which setup opens on the external interface. The certificate of file must be the key's and name
the hosts; it is written to /etc/caddy/linux-manager-sites/<name> (root, group caddy, 0640). An
issuing CA of ca is given to Caddy as the ACME servers' CAs are, in /etc/caddy/linux-manager-pki;
its certificates are trusted by clients that trust the server's root. Caddy runs while HTTPS is
ready or a site is served; while the sites cannot be read, setup leaves the Caddyfile as it is.
linux-manager-config-edit set --restart sites.enabled true
linux-manager-site add www --host www.example.com --host example.com --spa
linux-manager-site add intranet --host intranet.example.lan --port 8443 --listen internal --tls ca --ca web \
--editors web-team
linux-manager-site update www --no-enabled
linux-manager-site remove intranet --content
linux-manager-site remove keeps the folder without --content. Renaming a group renames it in the
sites; a group a site has cannot be removed.
Site files
The files of a site are read and written through the web service (see The /sites API) by the
admins and the members of the site's editors group, and through WebDAV. An upload is at most
sites.upload_max_mb MiB (512 by default, at most 10240); a zip archive is extracted into a
folder, and one with an absolute path, a .., a link, more than 10000 members or more content than
the limit is refused before anything is written. Files are written with mode 0644 and folders with
0755, so Caddy reads them.
WebDAV
While sites.webdav is true and HTTPS is ready, linux-manager-setup runs
linux-manager-webdav.service, rclone serve webdav on 127.0.0.1:7446, which Caddy serves at
https://<wan.local_fqdn>:<sites.webdav_port> (7445 by default) and the firewall opens on both
interfaces. A user signs in with the name and password of the server's directory, followed by the
TOTP code for an enrolled user; an admin sees a folder for every site, any other user the folders of
the sites whose editors group the user is a member of, and no folder means no access. rclone writes
an upload to its cache in /var/cache/linux-manager-webdav and to the site's folder a few seconds
later. rclone is installed from the Ubuntu archive by linux-manager-install@webdav.service when it is
missing.
linux-manager-config-edit set --restart sites.webdav true
Name server
While dns.enabled is true, linux-manager-setup runs PowerDNS (pdns.service) as the
primary name server of the server's domains, on port 53 of the addresses of dns.listen
(internal, external, the default, or both), which the firewall opens for UDP and TCP. The
domains and records are in the database linux-manager-dns (password in /etc/linux-manager/dns-database.secret),
read by PowerDNS's gpgsql backend; setup writes the settings
/etc/powerdns/pdns.d/linux-manager.conf (root, group pdns, 0640) and installs pdns-server and
pdns-backend-pgsql from the Ubuntu archive with linux-manager-install@dns.service when they are
missing, without the recommended bind backend.
| Setting | Meaning |
|---|---|
wan.local_fqdn |
The primary name server of every domain: the SOA's primary and the first NS |
dns.secondary |
The secondary name server of every domain, the second NS; the addresses it resolves to are notified of every change and may transfer the domains (AXFR) |
dns.ttl |
The TTL of the records and the SOA's negative TTL of a domain without its own, 60 to 86400 seconds, 300 by default |
A domain the server hosts, for example example.com, with wan.local_fqdn router.other.net,
dns.secondary ns2.provider.net and a site of www.example.com and example.com with dns
router and dns_address public (wan.public_address 203.0.113.10):
example.com. SOA router.other.net. hostmaster.example.com. <serial> 10800 3600 604800 300
example.com. NS router.other.net.
example.com. NS ns2.provider.net.
example.com. A 203.0.113.10
www.example.com. A 203.0.113.10
The domains are added with /dns or the Manager, each with:
- TTL: the TTL of its records,
dns.ttlwhen not set; - Manual records (A, AAAA, CNAME, TXT, MX, SRV, CAA), served too, but not those of a name and type the domain's own records have.
The host names of an enabled site with dns router get an A or AAAA record of its dns_address
in the closest domain they are in; a site with dns user gets none, and a host name in no domain
of the server gets none either. wan.local_fqdn gets no record in the domains: its A record is kept
by the DNS of its own domain, wherever that is. A domain inside another domain of the server is
delegated from it with NS records. Setup writes the domains at each run, and after each change
through /dns; the SOA's serial grows only when the records change. While IKEv2, OpenVPN or
WireGuard is enabled, dnsmasq answers their clients on port 53 of the addresses of
p2s.dns_listen, by default the internal address, so dns.listen must not share an interface with
it. The parent zone of each domain must delegate it to wan.local_fqdn and dns.secondary, with
glue records for those of them inside the domain. While dns.secondary is not set, the domains have
wan.local_fqdn as their only name server, and setup warns.
linux-manager-config-edit set dns.secondary ns2.provider.net
linux-manager-config-edit set wan.public_address 203.0.113.10
linux-manager-config-edit set --restart dns.enabled true
Metrics
While metrics.enabled is true, linux-manager-setup runs linux-manager-metrics, which every minute
takes a sample of the server with psutil:
- the CPU utilization;
- the memory used and in total;
- the free and the total space of
/; - the bytes received and sent per second on the external and the internal interface;
- the 1, 5 and 15 minute load averages;
- the VPN clients of each enabled VPN system: IKEv2 and OpenVPN clients, and WireGuard peers with a handshake in the last 3 minutes;
- for each site-to-site connection that is set up: whether it is up (1 or 0), the bytes and the packets per second received and sent through its child SAs since the previous sample, and the seconds its IKE SA has been up. The child SAs' counters start again at a rekey; a counter lower than in the previous sample counts from zero.
History
The samples are kept in the server's PostgreSQL database, the one of the refresh tokens
(web.database), in four detail tables, one row per minute, and per interface, kind of client or
site-to-site connection, and each is summarized hourly, daily, monthly and yearly. A summary row
has the start of its period, in UTC, and the average and the maximum of each value; the average
of a connection's up is the part of the period it was up:
| Level | Tables | Kept |
|---|---|---|
| Details | metrics_system, metrics_network, metrics_vpn, metrics_site_to_site |
metrics.keep_details |
| Hourly | metrics_system_hourly, metrics_network_hourly, metrics_vpn_hourly, metrics_site_to_site_hourly |
metrics.keep_hourly |
| Daily | metrics_system_daily, metrics_network_daily, metrics_vpn_daily, metrics_site_to_site_daily |
metrics.keep_daily |
| Monthly | metrics_system_monthly, metrics_network_monthly, metrics_vpn_monthly, metrics_site_to_site_monthly |
metrics.keep_monthly |
| Yearly | metrics_system_yearly, metrics_network_yearly, metrics_vpn_yearly, metrics_site_to_site_yearly |
metrics.keep_yearly |
metrics_site_to_site has the columns connection, up, rx, tx, packets_rx, packets_tx
and uptime.
How it works:
-
Summaries: a period is summarized once it is complete, from the level below. A stopped service catches up on the periods it missed, as long as their rows are still kept.
-
Garbage collection: runs on start and once a day. It deletes the rows older than their retention, a count and a unit such as
24 hours,3 monthsor10 years. -
History API:
GET /metrics/history?range=hour|day|month|yearof the status web service, protected like/config, answers the views of the Manager's Metrics tab. Each point has the system values,networkby interface,clientsby kind andsite_to_siteby connection, each value as itsavgandmax:Range Bucket Source hour1 minute details day10 minutes details month6 hours hourly year1 month monthly -
Site-to-site availability:
site_to_site_upof the same answer divides the view into slots of its own,slot, and gives for each slot and connection the percentage of the samples in which the connection was up, 0 to 100, aspointsof{"time", "connections": {name: percent}}:Range Slot Source hour5 minutes details day1 hour details month1 day hourly year1 month monthly A sample is taken every minute, so an interruption shorter than a minute may be missed; a slot without samples has no value, and the year view, like the others of its level, has the complete months only.
Scraping
GET /metrics of the metrics service serves the last sample in the Prometheus text format, written by prometheus_client.
- Metric names: the OpenTelemetry semantic convention names in their Prometheus form:
system_cpu_utilization;system_memory_usage_bytesandsystem_memory_limit_bytes;system_filesystem_usage_bytesandsystem_filesystem_utilization;system_network_io_bytes_total;system_cpu_load_average_1m,_5mand_15m;linux_manager_clients, bykind:p2s,openvpnandwireguard;- by
connection, for each site-to-site connection:linux_manager_site_to_site_up,linux_manager_site_to_site_uptime_seconds, and, bydirectiontoo,linux_manager_site_to_site_io_bytes_per_secondandlinux_manager_site_to_site_packets_per_second. The rates are gauges, since the child SAs' counters start again at each rekey.
- Listening: on the interfaces
metrics.listennames,internalby default, atmetrics.port, and on127.0.0.1. While HTTPS is on, Caddy serves it atmetrics.https_portforwan.local_fqdn.linux-manager-setupopens both ports on the internal and the external interface. - Authentication: HTTP Basic as the service account
metrics.service_accountnames,metricsby default. The password is checked with a bind to the user directory, so 5 failed attempts lock the account as for any user. - The account:
linux-manager-setupowns it while metrics are enabled.- It adds the account with a random password, which it writes to
/etc/linux-manager/metrics.secret(root only). Give that password to the collector. - When the file is missing or no longer holds the account's password, setup sets a new random password and ends any lockout. Delete the file and restart
linux-manager-setupto change the password. - A disabled account is left as it is.
- A user of that name who is not a service account is a setup failure.
- It adds the account with a random password, which it writes to
An OpenTelemetry Collector scrapes it with its prometheus receiver:
receivers:
prometheus:
config:
scrape_configs:
- job_name: linux-manager
scrape_interval: 60s
scheme: https
basic_auth:
username: metrics
password_file: /etc/otelcol/linux-manager-metrics.secret # a copy of /etc/linux-manager/metrics.secret
static_configs:
- targets: ['router.example.com:9465']
Pushing
When metrics.otlp_endpoint is set, every sample is also pushed there as an OTLP/HTTP JSON request.
- Content: the semantic convention names (
system.cpu.utilization,system.network.ioand so on),linux_manager.clients, the gaugeslinux_manager.site_to_site.up,.uptime,.ioand.packetswith the attributeconnection, and the resource attributesservice.namelinux-managerandhost.name. - Authentication: HTTP Basic with
metrics.otlp_usernameandmetrics.otlp_passwordwhen they are set. - Failures: a failed push is logged and not repeated; the next sample is pushed a minute later.
Backups
While backup.enabled is true, linux-manager-setup runs linux-manager-backup.timer on backup.schedule. The timer starts linux-manager-backup.service, which runs linux-manager-backup run. The schedule is written to /etc/systemd/system/linux-manager-backup.timer.d/schedule.conf:
hourlyruns at the start of each hour;dailyruns at midnight;weeklyruns at midnight between Sunday and Monday.
A run missed while the server was off is made at the next boot.
What a backup holds
A backup holds the state that linux-manager-setup cannot make again:
| Item | Content |
|---|---|
/etc/linux-manager |
linux-manager.json, the secrets, the Manager's settings and rclone.conf |
/etc/wireguard |
The WireGuard keys and the peers |
/etc/ufw, /etc/default/ufw |
The firewall rules |
files/<path> |
Each psk_file outside /etc/linux-manager |
ldap/linux-manager.ldif |
The user directory, written by slapcat: users, groups, service accounts, password hashes and TOTP secrets, and the PKI: the CAs and their keys, the root key while it is on the server, and the certificates issued, with the keys of the server's and the users' certificates |
postgresql/linux-manager.dump |
With backup.metrics, the database of the metrics and the refresh tokens, written by pg_dump -Fc |
/var/lib/linux-manager/sites |
The folders of the web sites |
postgresql/linux-manager-sites.dump, postgresql/linux-manager-dns.dump |
The databases of the web sites and of the name server, once they exist |
manifest.json |
The host, the time, the version of the package, the schedule and the items |
The files and the directory are copied while the locks of the configuration and of the users are held, so no change of the API or the command-line tools falls between them; pg_dump reads a snapshot of its own. The copy is staged in /var/lib/linux-manager/backup/staging, which is removed after each run. The backup holds every secret of the server, so keep it as safe as the server itself.
Local archives
With backup.target local, a backup is a tar archive compressed with zstd, linux-manager-<host>-<UTC time>.tar.zst, in backup.path (mode 0600, the folder 0700). The folder may be a mount, such as an NFS share or an Azure Files share, which exports the archives.
After each backup, the archives of this host that the retention settings no longer keep are deleted. The rules are those of Proxmox Backup Server's prune:
- The settings apply in order:
keep_hourly,keep_daily,keep_weekly(ISO weeks, Monday to Sunday),keep_monthly,keep_yearly. - Each keeps the newest archive of each of its last N periods that have archives, in local time.
- A period that has an archive a former setting kept is passed over.
- A setting finer than the schedule does not apply:
keep_hourlyonly withhourly,keep_dailywithhourlyanddaily. At least one setting that applies must be above 0.
Cloud copy with rclone
With backup.rclone_remote set, each local backup is followed by an rclone run that copies the archives of this host to the remote. rclone comes from the rclone package, which the package suggests:
sudo apt-get install rclone
sudo rclone config --config /etc/linux-manager/rclone.conf
sudo chmod 0600 /etc/linux-manager/rclone.conf
sudo linux-manager-config-edit set backup.rclone_remote azure:backups/router
sudo systemctl restart linux-manager-setup
The two modes:
sync(the default) makes the remote mirror the archives of this host, so an archive retention deletes is deleted remotely too.copyuploads new archives and deletes nothing, so retention in the cloud is left to its own rules, such as a lifecycle policy.
Other files in the remote path are left alone in both modes. A failed cloud copy leaves the local backup in place, and the run's result is warning.
Proxmox Backup Server
With backup.target pbs, the staged copy is sent with proxmox-backup-client as the archive router.pxar of the snapshot host/<host>/<time>, in backup.pbs_namespace when it is set. The server keeps the snapshots by its own prune jobs; the server prunes nothing.
The static client comes from Proxmox's client repository, which the package suggests but does not set up. Following Proxmox's installation guide, as /etc/apt/sources.list.d/pbs-client.sources:
Types: deb
URIs: http://download.proxmox.com/debian/pbs-client
Suites: trixie
Components: main
Signed-By: /usr/share/keyrings/proxmox-archive-keyring.gpg
Get the key https://enterprise.proxmox.com/debian/proxmox-archive-keyring-trixie.gpg into /usr/share/keyrings/proxmox-archive-keyring.gpg, check it against the SHA-256 the installation guide gives, and install proxmox-backup-client-static. Until it is installed, a backup to PBS fails with a message that names the package; nothing else depends on it.
On the server, make an API token for the server and give the token itself the DatastoreBackup role on /datastore/<store>, or on the namespace path. backup.pbs_repository names the token, such as backup@pbs!router@pbs.example.com:store, and backup.pbs_password holds its secret. Set backup.pbs_fingerprint when the server's certificate is self-signed. With backup.pbs_keyfile, the snapshot is encrypted with a key of proxmox-backup-client key create --kdf none <file>. Keep a copy of that key away from the server, since a restore needs it.
Running, listing and restoring
sudo linux-manager-backup run # a backup now, whether or not backups are enabled
sudo linux-manager-backup list # the local archives, or the snapshots on the server
sudo linux-manager-backup restore /var/backups/linux-manager/linux-manager-router-20261025T030000Z.tar.zst
sudo linux-manager-backup restore --pbs host/router/2026-10-25T03:00:00Z
Each run is recorded in /var/lib/linux-manager/backup.json, which /backup and the Manager show. Only one run or restore takes place at a time.
restore is only on the command line. It shows what the backup holds and asks for yes, which --yes gives. It then:
- stops the web service, the metrics service, WebDAV, PowerDNS and slapd;
- puts back the files;
- loads the directory with
slapaddinto an empty/var/lib/ldap/linux-manager; - sets the database roles' passwords to the restored password files, and restores each database the backup holds with
pg_restore --clean; - restarts
linux-manager-setup, which regenerates everything else, followed by WireGuard, OpenVPN, strongSwan and the Manager, and reloads ufw.
What it replaces is kept in /var/lib/linux-manager/backup/before-restore/<UTC time>. To restore onto a new VM, install the package first and let linux-manager-setup run once. The restored configuration names the interfaces of the old VM, so change them with linux-manager-config-edit if they differ.
Ubuntu Linux Manager
The Manager is the package's web application: it shows the state of the server and manages
the users of its OpenVPN and IKEv2 access. Users sign in with their user name, password
and, when enrolled for one-time passwords, the code, or with a passkey they added on their
account page; only members of the server's group admins
can use the tabs, as the server decides. Every user has an account page at /user, which is the
only page of a user who is not a member of admins. Administrators switch between Router (/:
the Overview, Metrics, Logs, Backup and System tabs), Identities (/user-manager: the Users, Groups and Service
Accounts tabs), Connectivity (/site-to-site-manager: the Status, Network and Connections tabs), Remote
Access (/remote-access-manager: the IKEv2, OpenVPN and WireGuard tabs), Certificates
(/ca-manager), Web Sites (/sites-manager: the Sites, Files and DNS tabs) and
Docker (/docker-manager) with the page menu at the
right end of the header, a button naming the page shown. The button with the signed-in user's
name, left of Sign out, opens the account page, and the terminal icon left of it the Shell
(/shell); pressed on its page, each returns to the page it was opened from.
-
Help (the question mark in the header between the theme switcher and the page menu, for every signed-in user): a panel on the right, under which the page moves left, with the help topic of the page and tab shown. The topics are the Markdown files of
docs/(index.mdand one per page or tab), served by the manager at/docsand rendered in the browser with marked and DOMPurify, in the style of github-markdown-css in the light or dark scheme. A link to another topic opens it in the panel; Back returns to the previous one and the house icon toindex.md. Topics link to SVG and PNG pictures by paths relative to their folder. The browser keeps whether the panel is open. -
Settings (the gear in the header, hidden on the sign-in and the account page): how often the Overview and the Metrics tab read the server on their own, kept in the browser's localStorage, so each browser has its own. Each tab reads manually, or every 5, 10 or 30 seconds, 1, 5 or 15 minutes; the Overview tab every 10 seconds and the Metrics tab every minute by default. Manual reads when the tab opens and on Refresh. A change applies at once, also in the browser's other tabs; Defaults sets both back.
-
Account (
/user): the Account and the Passkeys panel side by side, with One-time passwords and VPN profiles below. The user's name, kind, names, e-mail address, whether they are an administrator and whether they are enrolled for one-time passwords; Change password in the Account panel, which opens a dialog and signs the user out after the change; enrolling an authenticator app or replacing it; and the user's passkeys, each with its label and when it was added, with Add a passkey and a remove icon. Each change but the removal of a passkey asks for the current password and, for an enrolled user, a code not already used to sign in. Removing the enrolment is left to an administrator. A VPN user downloads the own OpenVPN profile (.ovpn) and the Apple VPN profile of the server's IKEv2 access (.mobileconfig), each with the current password and the code when enrolled, the Apple profile also with a password that protects the key in it. -
Passkeys: Sign in with a passkey on the sign-in page uses a passkey the browser or the user's password manager offers for the server's
wan.local_fqdn, without the user name, the password or the code. Passkeys work on the Manager's HTTPS address on that name only, not on another name or in development onlocalhost, and in browsers with the JSON methods of WebAuthn Level 3. They sign in to the Manager, and open a shell on the Shell page in place of the password and code. -
Certificates (
/ca-manager, in the page menu of administrators): the server's two-level CA, from its/ca, in tabs:- Root: the root CA's subject, expiry, whether its key is on the server, marked with a warning sign, or offline, and its CRL, with Download root CA, the root's certificate alone, and Upload CRL for a root CRL signed offline; the general information: where certificates and CRLs are published, the CA key type, the lifetimes, and the CRL validity; the certificates of the issuing CAs, each active one with a download icon for its CA bundle, and the retired ones.
- Templates: the certificate templates with their key, lifetime, subject, usages, alternative names, certificate policies and the CAs that issue them; Add template, Edit on each, Remove on one of the server's own and Reset on a changed built-in one, which brings back its defaults. A template's extended key usages are chosen from those of the next tab, and its certificate policies are OIDs typed in.
- Extended key usages: the usages templates may give, with their OIDs and the templates that give them; the built-in ones are shown and not changed, and custom ones are added with a name and an OID, and changed or removed.
- P2S Issuing CA and S2S Issuing CA: the CA's common name, state (active, or pending while it waits for the offline root), expiry, CRL and templates, and the certificates it issued. An active CA gives its CA bundle, the issuing CA followed by the root, with Download CA bundle; a pending CA gives its CSR with the openssl extensions to sign it with, and takes the certificate signed by the root.
- Issuing CAs: the general-purpose CAs with their state, expiry, CRL, templates and ACME directory URL; Add issuing CA, and Edit and Remove on each: its name, fixed once it exists, its common name, the templates it issues and its ACME server; an active CA's download icon gives its CA bundle. A removed CA is retired by the server, which revokes its certificate on the root CRL. Below, their certificates, filtered by CA. - Certificates: each with its kind, template, names, expiry and state; a certificate of another system or of the server is revoked after a confirmation with a reason.
- Issue (above the tabs): from a CSR, saving the certificate with its chain, or with a key the server makes, saving a PFX file protected by a password; the server keeps no copy of the key.
- Settings (above the tabs): the names, lifetimes, key type, CRL validity and publishing endpoints.
- Warnings (above the tabs): what the server says needs attention, such as a CRL about to expire.
The root key is exported, removed and imported on the server with
linux-manager-ca, never through the Manager. -
Overview: the server's overall status, ok or degraded, on the title bar of the Status panel; at the top of that panel, a table of
linux-managerwith the installed version and, when the repository has a newer one, that version; Check update is always in its top right corner, and Upgrade is added beside it when an update is found. Then IKEv2, OpenVPN and WireGuard, read from the server's authenticated/status/full, and from its/configthe status web service and its port, whether HTTPS is on and for which name, with a chip while Caddy takes Let's Encrypt's staging CA and a settings button that switches it (https.acme_staging, through the server's/https/settings), and Caddy, its HTTPS proxy: its version, whether it comes from Ubuntu or upstream (switched on the server withlinux-manager-caddy use-upstream) with the configuredhttps.caddy_sourcewhen that differs, and the state ofcaddy.service. The Services panel shows the state of each of the server's systemd services and of Caddy, with a chip on its title bar telling whether the server's services are active. The Addresses panel links, opening in a new tab, to the status page over HTTPS and over plain HTTP on the addressesweb.listennames, and to the Manager over HTTPS, and shows the issuer and the discovery URL of the server's OpenID Provider, each with an icon that copies it, or off while the provider is disabled. Further panels show each PostgreSQL cluster of the router with its main settings and its databases with their sizes, and one Configuration panel with a group of rows under a heading for each section the other pages do not show, the status web service (web) and HTTPS (https) among them, read every 10 seconds by default (see Settings). Nothing is read, and Refresh is off, while a check for updates, an upgrade or a reboot runs; the tab reads again when it ends.- Panel controls: each panel folds by its title bar, and every panel but Status hides by its [X]; Panels shows them again.
- Edit layout: moves panels by dragging and switches each between one column and both.
- Layout storage: kept in the browser.
-
Metrics: charts of the server's CPU, memory, free disk space in bytes and in percent, the bytes per second of each interface, the VPN clients and the load averages, and, with a line per site-to-site connection, Site-to-site up, Site-to-site traffic and Site-to-site packets. They cover the last hour, day, month or year, each bucket as its average or its maximum. Site-to-site up is the percentage of time each connection was up, 0 to 100, in slots of 5 minutes for an hour, 1 hour for a day, 1 day for a month and 1 month for a year, as the server computes it; Average and Maximum do not change it. The charts are read every minute by default (see Settings) from the server's
/metrics/history. The charts fold, hide, move and change width as the Server page's panels do, with a layout of their own. A click on a line's legend item hides or shows the line, for example the 1 and 5 minute load averages to see the 15 minute one; the browser keeps the hidden lines of each chart. Settings edits the server'smetricssection through its/metrics/settings: whether the metrics service runs, the interfaces and the ports of/metrics, its service account, the OTLP endpoint with its user name and password, and how long each level of the history is kept. An empty password keeps the current one. -
Logs: the last lines of the server's logs, from its
/logs. The Journal sub-tab shows the systemd journal of a unit the server names, or of all units, and the Files sub-tab a log file of the server'sweb.log_files, set on the server withlinux-manager-config-edit. 100, 200, 500 or 1000 lines, up to the server's limit of one read, are read when the unit, the file or the number changes and on Refresh. While Follow is on, the log is read every 2 seconds and kept scrolled to its last line, astail -fshows it. -
Users: the users with their names and e-mail addresses, without the service accounts, all of them, the VPN users or the Manager-only accounts, searched by user name, names and e-mail address, with adding (user name, first name, last name, e-mail address and password, or, with Manager access only, a Manager-only account with the user name, the password and an optional e-mail address: no certificate and no VPN access), editing the names and the e-mail address of VPN users and the e-mail address of Manager-only accounts, setting the password, disabling, enabling, unlocking after failed logins, removing and downloading the OpenVPN client profile, offered only while the server runs OpenVPN, and the Apple VPN profile (
.mobileconfig, IKEv2), offered only while the server runs IKEv2 access, whose dialog takes the password that protects the user's key in it, and whether each is enrolled for TOTP one-time passwords. Enrolling shows a new secret as a QR code and as text for an authenticator app, and stores it once a code of the app is entered; the enrolment can be removed. The Passkeys column shows how many passkeys each user has, and the remove icon removes all of them after a confirmation. The Console column shows the console login of each member ofadmins: off, the user's own LDAP POSIX account with its uid, or the local account it runs as, marked as refused while the user is not enrolled for TOTP; root switches it on the server withlinux-manager-user console, the Manager does not. -
Groups: the groups with their descriptions and members, searched by name, description and member, with adding a group, editing its name, description and members in one dialog, and removing it. A group has at least one member; the members offered are the users and the service accounts but the internal ones.
-
Service Accounts: the accounts other systems sign in with to use the server's services, such as an OpenTelemetry Collector reading
/metrics, from the server's/service-accounts; they cannot sign in to the Manager. Each is shown with its display name, description, kind, e-mail address, OpenID Provider settings and state, searched by name, display name, description and e-mail address, with adding one (name, a new GUID unless a user name replaces it, display name, description, an optional e-mail address, the OpenID Provider settings and a password), editing the display name, the description, the e-mail address and the OpenID Provider settings, setting the password, disabling, enabling, unlocking after failed logins and removing after a confirmation. An account with redirect URIs or grants is a client of the server's OpenID Provider, its name the client ID and its password the secret; its settings are the redirect URIs and the post-logout redirect URIs, one per line, the allowed groups, chosen from the server's groups, whose members alone may sign in to it, the client credentials and device authorization grants, whether it is a public client without a secret, and the scopes of resources it may ask for. Every account is also a resource of the provider, with its scopes and its roles, each assigned to groups, users and service accounts, which the tokens for it carry. While the provider is enabled, the title of the list has the number of its clients. Internal accounts, which the server creates and manages for its own services with all their properties, display name and description included (at present the metrics service's), are marked Internal and only listed: the server refuses every change of them through its API, so they have no actions here. -
Connectivity, Connections tab: the server's IPsec site-to-site connections with their remote peer, remote identity, remote and local CIDRs, local identity mode, authentication (whether the pre-shared key is set, read from a file or missing, or, with certificates, whether the peer's CA is the server's site-to-site CA or a remote CA) and whether the tunnel is established; with adding a connection, editing it and removing it after a confirmation. With a pre-shared key, the key is typed as a password, required while the connection has none and kept when left empty, and is never shown; with certificates, the PEM certificate of a remote CA is optional. The server applies a change in the background by running its setup again; the list is read again after each change. The connection the server was installed with is
remote-site; Cloud Manager names the one it deploys withrouter_site_to_site_name,remote-siteby default. -
Connectivity, Status tab: each connection the server set up, with the local and the remote identity and the subnets it resolved, and how long the connection has been up, and their IPsec security associations, child SAs, connections and connection children, from the server's
/config. The other IPsec connections, IKEv2 access, are on the IKEv2 tab of Remote Access. -
Connectivity, Network tab: the Interfaces panel lists every interface but loopback by name with its type,
externalwhen it holds an IPv4 default route andinternalotherwise, its state, IPv4 and IPv6 addresses and MAC address; the Router panel, shown while the router is on, the internal gateway and the local CIDRs; and the IPv4 routes, from the server's/config; the panels fold, hide, move and change width, and the browser keeps their layout. -
Remote Access (
/remote-access-manager): a tab each for IKEv2, OpenVPN and WireGuard access, from the server's/remote-accessand/status/full: the settings, whether the server runs the system, why an enabled one does not run yet, and the connected clients; for IKEv2 the IPsec security associations, child SAs, connections and connection children of IKEv2 access, all but the site-to-site ones, from the server's/config; for WireGuard each connection with its user, found by the peer's key (a short key for a peer added towg0.confby hand), its IPs, endpoint, handshake and traffic, from the server's/configand/users. Edit enables or disables the system and sets its address pool, for WireGuard the address ofwg0, for IKEv2 the DNS domains its clients resolve with the server's DNS, one per line, and the port, with OpenVPN's protocol; the server applies a change in the background. The WireGuard tab also lists the peers of the VPN users, each a device with its user, public key and IPs: a peer is removed with its button after a confirmation, and added with Add peer, whose dialog takes the user, the device's public key and the IPs, which start as the next free address of thewg0network. The server keeps the peers with their users and checks that keys are unique and IPs do not overlap. Its Server section shows the server's public key, with a copy button, the endpoint and the allowed IPs a device needs; Show config opens a device's configuration, ready to copy, with placeholders for the device's private key and IPs and, while it runs, the server's resolver as the device's DNS server. The eye of a peer opens its device's configuration the same way, with the peer's IPs as its address. In both dialogs a switch leaves out the[Interface]andPrivateKeylines, which WireGuard for Windows fills in for a new empty tunnel, so the rest pastes below them; the browser keeps its position with its other settings. Each peer's configuration downloads as<user>-<n>.confwith them filled in and a placeholder for the device's private key, which the router never has. IKEv2 and OpenVPN also needwan.local_fqdnand the server's VPN server certificate. A cloud firewall in front of the server, such as an Azure network security group, must allow the OpenVPN and WireGuard ports itself. -
Web Sites (
/sites-manager): the static web sites the server's Caddy serves, from its/sitesand/dns, in tabs:- Sites: the Settings panel shows whether the sites are served, the largest upload and WebDAV with its URL and a copy icon, edited with Settings. The list shows each site with its URLs, linked, its port, the interfaces whose firewall opens it, its TLS (none, Let's Encrypt, an issuing CA of the server or a certificate and key) and its editors group, searched by name, host name and group; Add and Edit take the name, fixed once added, the host names, one per line, the port, the interfaces, the TLS with the issuing CA or the PEM certificate chain and key, pasted or read from files, the key kept when left empty, the index file, the single-page fallback and the editors group. Enable, Disable and Remove, which removes the site's folder too when asked; the folder icon opens the site on the Files tab.
- Files: the files of the chosen site, a folder at a time, with the path linked up to the
site's folder. Upload sends chosen files, or files dropped on the list, into the folder shown,
with a progress bar; while Extract zip archives is on, a
.zipfile is extracted there instead. New folder, Rename, Remove and Download. A file larger than the server's largest upload is not sent. - DNS: the server's primary name server: its settings with whether PowerDNS is installed and runs, the server's FQDN as the primary name server, edited with Settings (enabled, secondary name server, public IP address, interfaces, TTL); the domains, added with Add domain (domain, TTL), edited and removed with their icons; and for the selected domain the records it serves, and its manual records, added with Add record (name, type, content and TTL) and removed with their icon.
-
Docker (
/docker-manager): whether Docker is installed, the state ofdocker.serviceand of the server's Docker rules, the ports exposed on the external interface and the rules of theLINUX-MANAGER-DOCKERchain, from the server's/status/fulland/config. -
Shell (
/shell, the terminal icon in the header): a login shell on the server in xterm.js, the terminal of VS Code, as the signed-in user's console account, the user's own LDAP POSIX account or a local account. Opening it asks for the password and the code again, or Open with a passkey takes one of the user's passkeys, unlocked with the device's PIN or biometrics, unless the user used a shell within the server'sweb.shell_auth_timeoutminutes. A user without console login sees how root switches it on, and one not enrolled for one-time passwords that the shell needs the enrolment. The server offers the shell only while itsweb.shellistrue, set on the server withlinux-manager-config-edit, and closes it afterweb.shell_idle_minuteswithout a keystroke; the page then shows the reason and keeps the terminal readable.- Editor:
lme FILE...(linux-manager-editor) in the shell opens the files in a Monaco editor on the page, one tab per file, and returns at once. The server names each file in a text message of the shell's WebSocket, and the page reads and saves it through/api/shell/editor/<session>/<id>, as the console account. Saves have a body limit of 2 MiB plus 64 KiB, for a file of up to 1 MiB escaped in JSON; other calls keep 64 KiB. - Editor icon: while a file is open, an icon left of the shell icon switches the page between the terminal and the editor, and the shell keeps running. A button left of Save shows or hides the minimap, remembered by the browser. Save or Ctrl+S writes the file; a file changed on the server since it was read offers Reload or Overwrite. Closing a tab with unsaved changes asks first, and the icon goes when the last tab closes or the shell ends.
- Loading: Monaco is bundled with the page, without its language services, and loaded only when the first file opens.
- Editor:
-
Backup: the server's backups, from its
/backup:- Settings: whether scheduled backups run, the schedule (hourly, daily or weekly), the target and whether the metrics database is included. For a local folder it also shows the folder, the archives kept and the cloud copy with rclone; for a Proxmox Backup Server, the repository, whether the token secret is set, the fingerprint, the namespace and the encryption key file.
- Last run: its result, times, archive or snapshot, size, cloud copy and error, and the next run.
- Local archives: with their times and sizes, searched by name and time, and shown 10, 25, 50 or 100 at a time.
- Edit: changes the settings in one dialog. Only the retention settings that apply to the schedule are offered. The rclone remote is chosen from the remotes of the server's
/etc/linux-manager/rclone.conf, with a path and the mode, sync or copy. The token secret is typed as a password and kept when left empty. The server applies a change in the background. - Run now: starts a backup, and the tab reads every 3 seconds until it ends.
- Warnings: shown when the PBS client or rclone the settings need is not installed on the server.
Restoring is done on the server with
linux-manager-backup restore. The Overview tab has a Backup panel with the result of the last run, the next run and Run now. -
System: the server's operating system. The System updates panel lists the packages
apt-get upgradewould upgrade, each with its installed and its new version, from/api/system-updates; Check update is always in its top right corner, and Upgrade is added beside it when a package can be upgraded. Upgrade runsapt-get upgrade -yafter a confirmation and shows its output in a dialog, as Upgrade on the Overview tab does, with Reload page and Reboot when they apply.
How it works
linux-manager-ui.service runs /usr/lib/linux-manager/manager, a Python server that
calls the server with python3-requests and keeps sessions in PostgreSQL with python3-psycopg.
It serves the application from /usr/share/linux-manager/manager/web.
POST /api/sessionwith{"name", "password", "code"}signs in through the server's/auth, passing the browser's address inX-Forwarded-For; on success the manager keeps the user's access and refresh tokens in the tablesessionsand sets the cookielinux_manager_ui_session(HttpOnly,SameSite=Strict), which holds a random id of which only the SHA-256 is stored.GET /api/sessionanswers{"name"}of the signed-in user, andDELETE /api/sessionsigns out and revokes the refresh token.- Every other call to
/api/<path>needs a session and is passed to the server's status web service at/<path>with the session's access token, renewed through/auth/refreshwhen it expires within 30 seconds. The method, the body and the server's answer pass unchanged. A session ends when the server refuses to renew it; sessions are kept across restarts. GET,HEADandPUTof a web site's files,/api/sites/<name>/files/<path>, pass streamed: the body of aPUTgoes to the server as it arrives, with itsContent-TypeandContent-Length, which it needs, and without the limit of 65536 bytes, and the server's answer comes back with itsContent-Type,Content-LengthandContent-Disposition, so a download is not held in memory either. The server'ssites.upload_max_mblimits an upload.GET /docs/<file>, with a session, gives a file of the help from/usr/share/linux-manager/manager/docs, which the package installs from the repository'sdocs/: Markdown (text/markdown), SVG or PNG; any other file, or a path outside the folder, is 404. The answers carryX-Content-Type-Options: nosniffand aContent-Security-Policythat keeps an SVG opened on its own from running scripts.- A WebSocket handshake on
/api/shell, with a session and anOriginof the page's own host, is passed to the server's/shellwith the session's access token; after both handshakes the manager relays the WebSocket's bytes both ways until either side closes.POST /api/shell, which gives the ticket the handshake needs, andPOST /api/shell/passkey, which begins the passkey proof it may take, pass as every other call does. /api/infogives the version of the package and the URL of the server's status page, without a session. On the server, whilerouter_urlandterraform_dirare not set and the status web service listens on the address the call came in on, the URL has the name or address the browser used andweb.port, for examplehttp://router.example.com:7081, and for a page opened over the server's HTTPS proxyhttpsandhttps.web_port, for examplehttps://router.example.com:7444; otherwise it isrouter_url./api/updates, with a session, gives the installed and the candidate version oflinux-managerfromapt-cache policy, that is from the repository apt is configured with, as of apt's last update of its package lists (/var/lib/apt/periodic/update-success-stamp, answered aschecked), whether an update is available, and ascheckingwhether a check for updates runs. The Overview tab shows it at the top of its Status panel. Where apt cannot be asked, such as on a workstation, it answers"available": false.POST /api/updates, by a member ofadmins, checks for updates: it runsapt-get updatein the transient unitlinux-manager-ui-checkoutside the manager's unit, waiting up to 5 minutes for apt's lock, and touchesupdate-success-stampwhen it succeeds. The page's Check update button starts it and reads the versions again until it ends. It answers 202 when started, 409 while a check, an upgrade or a reboot runs, 403 for a user who is not an admin, and 503 where apt is missing./api/system-updates, with a session, gives the packagesapt-get upgradewould upgrade, from its simulation (apt-get -s upgrade), each with its installed and its candidate version, in the shape of/api/updates. The System tab shows it; Check update there isPOST /api/updates.POST /api/upgrade, by a member ofadmins(the server's/accountdecides), upgradeslinux-manager; the page's Upgrade button starts it.- Where it runs: the helper
/usr/lib/linux-manager/upgrade, in the transient unitlinux-manager-ui-upgrade(systemd-run) outside the manager's unit. The manager's own restart and the server web service's restart during the upgrade do not end it. - What it does: first
dpkg --configure -a, thenapt-get install --only-upgrade. It asks nothing, keeps changed configuration files, and waits up to 5 minutes for the dpkg lock. - Record: the state goes to
/var/lib/linux-manager/upgrade.jsonand the output toupgrade.log. - Answers: 202 when started, 409 while an upgrade or a reboot runs, 403 for a user who is not an admin, 503 where apt is missing.
- Where it runs: the helper
GET /api/upgrade, with a session, gives that state and the last 200 lines of the output.- States:
idle,running,succeeded,failed, orinterruptedwhen the upgrade ended without recording a result. - Reboot information: whether
/var/run/reboot-requiredexists, the packages that ask for it, and whether a reboot is about to happen.
- States:
POST /api/system-upgrade, by a member ofadmins, upgrades the system's packages, andGET /api/system-upgradegives its state, as/api/upgradedoes forlinux-manager.- What it does: the helper with
--system, in the transient unitlinux-manager-ui-system-upgrade: firstdpkg --configure -a, thenapt-get upgrade -y, with the same options. - Record:
/var/lib/linux-manager/system-upgrade.jsonandsystem-upgrade.log. - Answers: as those of
/api/upgrade; only one of the two upgrades runs at a time, and either answers 409 while the other runs.
- What it does: the helper with
POST /api/reboot, by a member ofadmins, reboots the server 5 seconds later (the transient timerlinux-manager-ui-reboot), and answers 409 while an upgrade runs. The page offers it when a reboot is required.
| Answer | When |
|---|---|
| 401 | Not signed in, or the session has ended |
| 403 | The server refuses the user, who is not a member of admins |
| 502 | The server cannot be reached |
| 503 | No router or no session database is configured, or the database fails |
| 413 | The request body is larger than 65536 bytes, but for the upload of a site's file |
| 411 | The upload of a site's file has no Content-Length |
The server's rate limit of sign-ins counts the browser's address when the manager's address is
in the server's web.trusted_proxies; by default that is the server's address the call came in
on, which is the address a manager on the server calls from, so it needs nothing more.
On the server, the server's Caddy serves the manager over HTTPS on https.manager_port of
linux-manager.json (7443 by default) for wan.local_fqdn, with a certificate from Let's Encrypt,
and calls it on 127.0.0.1, where the manager also listens. For a call from 127.0.0.1 with
X-Forwarded-For, the manager takes the browser's address from the last entry of that header.
The service needs nothing else to start and keeps running. Every 30 seconds it works out where to
listen and starts, moves or stops a server on each address to match, so a change of listen,
bind or port takes effect within 30 seconds, or when the service restarts. Each request
reads the settings again.
Before it listens, it creates its sessions database on the server when it is missing and connects to the sessions database, retrying every 15 seconds and logging each failure, for example while PostgreSQL is not running yet.
Settings
/etc/linux-manager/manager.json (all optional; see
/usr/share/doc/linux-manager/manager.json.example):
| Setting | Meaning | Default |
|---|---|---|
router_url |
The server's status web service | http://<first address of the web service>:<web.port>, the internal address when it listens on both |
database |
A PostgreSQL connection string for the sessions | The linux-manager-ui database on the server |
listen |
The interfaces to listen on when bind is not set: internal, external or both, and always 127.0.0.1 |
both |
bind |
The one address to listen on; the server's HTTPS proxy reaches the manager only when it is 127.0.0.1 |
none, listen applies |
port |
The port to listen on, HTTP | 7080 |
terraform_dir |
A Terraform root directory whose outputs give router_url when it is not set; dev.sh sets it |
none |
iac_tool |
The tool that reads the outputs of terraform_dir: terraform or tofu |
terraform |
From the outputs of <iac_tool> -chdir=<terraform_dir> output -json, read every 30 seconds at
most, router_url is the output router_web_url, the status web service's URL on the server's
private address with its port. It needs that tool, the
one the root directory was initialized with, and access to the state.
The addresses
of the interfaces listen names are worked out as general.mode directs, as the status web
service does, and linux-manager-setup opens port on the internal and the external interface;
restart linux-manager-setup after changing port. When database is not set, the manager creates the role and the database
linux-manager-ui on the server's PostgreSQL on its first run, with a password in
/etc/linux-manager/manager-database.secret (root only), and reaches it over TCP on localhost.
Make manager.json readable by root only when its database holds a password.
First administrator
The package asks two debconf questions on a fresh installation, linux-manager/admin_username
and linux-manager/admin_password. With both answered, it writes them to
/etc/linux-manager/manager-admin.json (root only) and clears them from debconf;
linux-manager-setup creates that user as a Manager-only account in the group admins, creating
the group when missing, and removes the file. Without a user name no account and no group is
created. They can be preseeded:
echo 'linux-manager linux-manager/admin_username string admin' | debconf-set-selections
echo 'linux-manager linux-manager/admin_password password s3cr3t-pass' | debconf-set-selections
Running the application on a workstation
dev.sh runs the manager and the Vite development server, which reloads on changes, on the
workstation; the application is at http://localhost:5180. Its first argument is the
config.json of the platform that deployed the server. Its module_root, relative to the
directory of that file, is the Terraform root module whose outputs give the server's address.
The second argument, optional, is the connection string of a PostgreSQL database for the
sessions; without it the sessions are kept in the database PGDATABASE names, or the one named
after the login, and libpq takes the server, port, user, password and SSL mode from the PG*
variables of the environment, such as those of the platform's pg.env.sh. dev.sh writes them
as terraform_dir and database, with the platform's iac_tool and bind 127.0.0.1, to dev.json in the repository root, which is not in Git, and
the manager reads that and listens on 127.0.0.1:7080:
./dev.sh /path/to/azure-platform/config.json
./dev.sh /path/to/azure-platform/config.json 'host=localhost dbname=manager user=manager password=secret'
It creates .venv with psycopg[binary] and requests when they are missing, and installs the npm dependencies
when they are missing or out of date. The manager also takes --config <file> in place of
/etc/linux-manager/manager.json.
The Vite servers pass /api to VITE_MANAGER (http://127.0.0.1:7080 by default), so the built
application runs against any manager, at http://localhost:4180:
cd web
npm install
npm run build
VITE_MANAGER=http://10.0.0.4:7080 npm run preview
Removal
apt-get remove stops linux-manager-docker, which removes the Docker rules. It terminates the
tunnels and takes wg-quick@wg0, openvpn-server@linux-manager, caddy, linux-manager-metrics,
linux-manager-webdav, linux-manager-backup.timer and linux-manager-pki.timer down, but leaves the
configuration in place. strongswan is left enabled, since it is a shared service that
may be in use by something else.
apt-get purge additionally removes the generated files and the firewall rules, restoring
/etc/ufw/before.rules to its original content, and hands linux-manager.json back to ucf
before deleting it so a later reinstall starts clean. It also removes the web service's
signing key and the OpenID Provider's, the password file of the metrics service account, the Manager's
settings and the record of its last upgrades, the database of the refresh
tokens and the metrics and those of the web sites, the name server and the Manager's sessions, their roles and their password
files when the package created them, Caddy's configuration /etc/caddy/linux-manager.Caddyfile and PowerDNS's
/etc/powerdns/pdns.d/linux-manager.conf; the certificate in
/var/lib/caddy belongs to the caddy package. It removes the backup timer's schedule, the
record of the last backup, the published CA files, the copies of the CAs in
/etc/caddy/linux-manager-pki, the sites' certificates in /etc/caddy/linux-manager-sites and OpenVPN's CA and CRL files.
Caddy, rclone and PowerDNS stay installed, and with Caddy the
source of its repository. Key material, users, content and backups are never deleted:
/etc/linux-manager/ldap, /etc/linux-manager/rclone.conf, the user directory in slapd with the PKI, the WireGuard
key pair, everything under /etc/swanctl, the folders of the web sites in /var/lib/linux-manager/sites, the archives in
backup.path and /var/lib/linux-manager/backup/before-restore are left alone.
Building
./build.sh
The build runs in a container from the repository root: a Node.js stage builds the
Manager's application in web/ with Vite, and an Ubuntu stage builds the package from
debian-package/, with the application and the help in docs/, and writes it to out/. The
container runtime is Apple's container on macOS and Docker on Linux, or CONTAINER_RUNTIME
when set, for example Docker on a remote engine:
CONTAINER_RUNTIME=docker DOCKER_CONTEXT=build-host ./build.sh
./publish.sh uploads the package to the component main of the distribution resolute
(Ubuntu 26.04, the only supported release) of the configured repository, skipping a
distribution that has the version already; DISTRIBUTIONS names others.