Mosquitto Go Auth is an authentication and authorization plugin for the Mosquitto MQTT broker. The name is terrible, I know, but it's too late to change it. And, you know: naming, cache invalidation, off-by-one errors and whatnot.

• The plugin is up to date and is compatible with the recent 2.0 Mosquitto version.
• Bug reports will be attended as they appear and will take priority over any work in progress.
• Reviewing ongoing PRs is my next priority.
• Feature requests are the lowest priority. Unless they are a super easy win in importance and implementation effort, I'll accept contributions and review PRs before considering implementing them myself.

This is an authentication and authorization plugin for mosquitto, a well known open source MQTT broker. It's written (almost) entirely in Go: it uses cgo to expose mosquitto's auth plugin needed functions, but internally just calls Go to get everything done.

It is greatly inspired in jpmens' mosquitto-auth-plug.

These are the backends that this plugin implements right now:

• Files
• PostgreSQL
• JWT (with local DB or remote API)
• HTTP
• Redis
• Mysql
• SQLite3
• MongoDB
• Custom (experimental)
• gRPC
• Javascript interpreter

Every backend offers user, superuser and acl checks, and include proper tests.

Please open an issue with the feature or enhancement tag to request new backends or additions to existing ones.

• Requirements
• Build
• Configuration
  • General options
  • Cache
  • Hashing
  • Log level
  • Prefixes
  • Backend options
  • Registering checks
• Files
  • Passwords file
  • ACL file
  • Testing Files
• PostgreSQL
  • Testing Postgres
• Mysql
  • Testing Mysql
• SQLite3
  • Testing SQLite3
• JWT
  • Remote mode
  • Local mode
  • JS mode
  • Testing JWT
• HTTP
  • Response mode
  • Params mode
  • Testing HTTP
• Redis
  • Testing Redis
• MongoDB
  • Testing MongoDB
• Custom (experimental)
  • Testing Custom
• gRPC
  • Service
  • Testing gRPC
• Javascript
  • Testing Javascript
• Using with LoRa Server
• Docker
  • Prebuilt images
  • Building images
• License

This package uses Go modules to manage dependencies. As it interacts with mosquitto, it makes use of cgo. Also, it (optionally) uses Redis for cache purposes.

Before building, you need to build mosquitto. For completeness, we'll build it with websockets, tls and srv support.

First, install dependencies (tested on Debian 9 and later, Linux Mint 18, 19 and 20):

sudo apt-get install libwebsockets8 libwebsockets-dev libc-ares2 libc-ares-dev openssl uuid uuid-dev

Download mosquitto and extract it (change versions accordingly):

wget https://mosquitto.org/files/source/mosquitto-1.6.10.tar.gz
tar xzvf mosquitto-1.6.10.tar.gz
cd mosquitto-1.6.10

Modify config.mk, setting websockets support. Then build mosquitto, add a mosquitto user and set ownership for /var/log/mosquitto and /var/lib/mosquitto/ (default log and persistence locations).

make
sudo make install
sudo groupadd mosquitto
sudo useradd -s /sbin/nologin mosquitto -g mosquitto -d /var/lib/mosquitto
sudo mkdir -p /var/log/mosquitto/ /var/lib/mosquitto/
sudo chown -R mosquitto:mosquitto /var/log/mosquitto/
sudo chown -R mosquitto:mosquitto /var/lib/mosquitto/

Finally, you may create a service for mosquitto. Create the file /etc/systemd/system/mosquitto.service with these annotations:

[Unit]
Description=Mosquitto MQTT v3.1/v3.1.1 server
Wants=network.target
Documentation=https://mosquitto.org/documentation/

[Service]
Type=simple
User=mosquitto
Group=mosquitto
ExecStart=/usr/local/sbin/mosquitto -c /etc/mosquitto/mosquitto.conf
Restart=on-failure
SyslogIdentifier=Mosquitto

[Install]
WantedBy=multi-user.target

If you are running another distro or need more details on building mosquitto, please check the offical mosquitto docs.

Only Linux (tested in Debian, Ubuntu and Mint ùs) and MacOS are supported.

Before attempting to build the plugin, make sure you have go installed on the system. The minimum required GO version for the current release is 1.13.8. To check which version (if any) of Go is installed on the system, simply run the following:

go version

If Go is not installed or the installed version is older than 1.13.8, please update it. You can retrieve and install the latest version of Go from the official Go download website:

# Update the following as per your system configuration
export GO_VERSION=1.16.4
export GO_OS=linux
export GO_ARCH=amd64

wget https://dl.google.com/go/go${GO_VERSION}.${GO_OS}-${GO_ARCH}.tar.gz -O golang.tar.gz
sudo tar -C /usr/local -xzf golang.tar.gz
export PATH=$PATH:/usr/local/go/bin
rm golang.tar.gz

# Prints the Go version
go version

This will fetch all the go dependecies and then build go-auth.so shared object:

make

This assumes that mosquitto.h, mosquitto_plugin.h and mosquitto_broker.h are located at /usr/local/include, which is true for a manually built mosquitto version in debian based systems (and probably others too), or manually built or installed through brew (brew install mosquitto) mosquitto version in MacOS.

If this doesn't work for your distribution or OS version, please check Makefile CFLAGS and LDFLAGS and adjust accordingly. File an issue or open a PR if you wish to contribute correct flags for your system.

Important notice: RPi support has been tested only until versions 1.4.x. The introduction of new plugin functions in Mosquitto may result in some issue compiling versions 1.5.x and later. Please reach me with any solutions you may find when resolving said issues.

To build on a Raspberry Pi (tested with Pi 3 B), you'll need to have Go installed first. You can install latest version (last tested was 1.10.1, change it to suit your needs) with something like this:

wget https://storage.googleapis.com/golang/go1.10.1.linux-armv6l.tar.gz
sudo tar -C /usr/local -xzf go1.10.1.linux-armv6l.tar.gz

Add Go to your path at .profile:

export PATH=$PATH:/usr/local/go/bin:~/go/bin

Source the file (source ~/.profile) and check Go was correctly installed (go version).

Now get requirements and build as usual (just have some more patience).

There seems to be missing packages in some Raspbian versions, so you should try to apt update before installing dependencies. Alternatively, you cand build openssl like this:

git clone git:https://git.openssl.org/openssl.git
cd openssl
./config
make
make test
sudo make install

For websockets support, you'll have to build libwebsockets, which needs cmake. So something like this should do the trick:

sudo apt-get install cmake
git clone https://github.com/warmcat/libwebsockets.git
cd libwebsockets
mkdir build
cd build
cmake ..
make
make install

The plugin is configured in Mosquitto's configuration file (typically mosquitto.conf). You may define all options there, or include e.g. a conf.d dir for plugin configuration:

include_dir /etc/mosquitto/conf.d

Create some conf file (e.g., go-auth.conf) at your preferred location, e.g. /etc/mosquitto/conf.d/, and register the plugin's shared object path and desired backends with:

auth_plugin /etc/mosquitto/conf.d/go-auth.so

auth_opt_backends files, postgres, jwt

Set all other plugin options below in the same file.

There are 2 types of caches supported: an in memory one using go-cache, or a Redis backed one.

Set cache option to true to use a cache (defaults to false when missing) and cache_type to set the type of the cache. By default the plugin will use go-cache unless explicitly told to use Redis. Set cache_reset to flush the cache on mosquitto startup (hydrating go-cache on startup is not yet supported).

Update v1.2: Set cache_refresh to refresh expiration each time a record is found in the cache (defaults to false). Before v1.2 cache was always refreshed upon check. In order to prevent security issues, where an attacker would frequently check on a topic to keep their granted status, even when revoked in the underlying backend, this has been turned into an option that defaults to no refreshing.

Finally, set expiration times in seconds for authentication (auth) and authorization (acl) caches:

auth_opt_cache true
auth_opt_cache_type redis
auth_opt_cache_reset true
auth_opt_cache_refresh true

auth_opt_auth_cache_seconds 30
auth_opt_acl_cache_seconds 30
auth_opt_auth_jitter_seconds 3
auth_opt_acl_jitter_seconds 3

auth_jitter_seconds and acl_jitter_seconds options allow to randomize cache expiration time by a given offset The value used for expiring a cache record would then be cache_seconds +/- jitter_seconds. With above values (30 seconds for cache and 3 seconds for jitter), effective expiration would yield any value between 27 and 33 seconds. Setting a jitter value is useful to reduce lookups storms that could occur every auth/acl_cache_seconds if lots of clients connected at the same time, e.g. after a server restart when all clients may reconnect immediately creating lots of entries expiring at the same time. You may omit or set jitter options to 0 to disable this feature.

If cache_reset is set to false or omitted, cache won't be flushed upon service start.

When using Redis, the following defaults will be used if no values are given. Also, these are the available options for cache:

auth_opt_cache_host localhost
auth_opt_cache_port 6379
auth_opt_cache_password pwd
auth_opt_cache_db 3

If you want to use a Redis cluster as your cache, you may omit previous Redis options and instead need to set auth_opt_cache_mode to cluster and provide the different addresses as a list of comma separated host:port strings with the auth_opt_cache_addresses options:

auth_opt_cache_mode cluster
auth_opt_cache_addresses host1:port1,host2:port2,host3:port3

Notice that if cache_mode is not provided or isn't equal to cluster, cache will default to use a single instance with the common options. If instead the mode is set to cluster but no addresses are given, the plugin will default to not use a cache.

There are 3 options for password hashing available: PBKDF2 (default), Bcrypt and Argon2ID. Every backend that needs one -that's all but grpc, http and custom- gets a hasher and whether it uses specific options or general ones depends on the auth opts passed.

Provided options define what hasher each backend will use:

• If there are general hashing options available but no backend ones, then every backend will use those general ones for its hasher.
• If there are no options available in general and none for a given backend either, that backend will use defaults (see hashing/hashing.go for default values).
• If there are options for a given backend but no general ones, the backend will use its own hasher and any backend that doesn't register a hasher will use defaults.

You may set the desired general hasher with this option, passing either pbkdf2, bcrypt or argon2id values. When not set, the option will default to pbkdf2.

auth_opt_hasher pbkdf2

Each hasher has specific options. Notice that when using the pw utility, these values must match those used to generate the password.

auth_opt_hasher_salt_size 16           # salt bytes length
auth_opt_hasher_iterations 100000      # number of iterations
auth_opt_hasher_keylen 64              # key length
auth_opt_hasher_algorithm sha512       # hashing algorithm, either sha512 (default) or sha256
auth_opt_hasher_salt_encoding          # salt encoding, either base64 (default) or utf-8

auth_opt_hasher_cost 10                # key expansion iteration count

auth_opt_hasher_salt_size 16           # salt bytes length
auth_opt_hasher_iterations 3           # number of iterations
auth_opt_hasher_keylen 64              # key length
auth_opt_hasher_memory 4096            # amount of memory (in kibibytes) to use
auth_opt_hasher_parallelism 2          # degree of parallelism (i.e. number of threads)

These options may be defined for each backend that needs a hasher by prepending the backend's name to the option, e.g. for setting argon2id as Postgres' hasher:

auth_opt_pg_hasher argon2id
auth_opt_pg_hasher_salt_size 16           # salt bytes length
auth_opt_pg_hasher_iterations 3           # number of iterations
auth_opt_pg_hasher_keylen 64              # key length
auth_opt_pg_hasher_memory 4096            # amount of memory (in kibibytes) to use
auth_opt_pg_hasher_parallelism            # degree of parallelism (i.e. number of threads)

You can set the log level with the log_level option. Valid values are: debug, info, warn, error, fatal and panic. If not set, default value is info.

auth_opt_log_level debug

Log destination may be set with log_dest option. Valid values are stderr (default), stdout and file. In the latter case the log_file option needs to be set, e.g.:

auth_opt_log_dest file
auth_opt_log_file /var/log/mosquitto/mosquitto.log

If log_dest or log_file are invalid, or if there's an error opening the file (e.g. no permissions), logging will default to stderr.

Do not, I repeat, do not set log_level to debug in production, it may leak sensitive information. Reason? When debugging it's quite useful to log actual passwords, hashes, etc. to check which backend or hasher is failing to do its job. This should be used only when debugging locally, I can't stress enough how log level should never, ever be set to debug in production.

You've been warned.

By default, if backend had an error (and no other backend granted access), an error is returned to Mosquitto.

It's possible to enable retry, which will immediately retry all configured backends. This could be useful if the backend may be behind a load-balancer (like HTTP backend) and one instance may fail:

auth_opt_retry_count 2

The above example will do up to 2 retries (3 calls in total considering the original one) if the responsible backend had an error or was down while performing the check.

Though the plugin may have multiple backends enabled, there's a way to specify which backend must be used for a given user: prefixes. When enabled, prefixes allow to check if the username contains a predefined prefix in the form prefix_username and use the configured backend for that prefix. Options to enable and set prefixes are the following:

auth_opt_check_prefix true
auth_opt_prefixes filesprefix, pgprefix, jwtprefix

Prefixes must meet the declared backends order and number. If amounts don't match, the plugin will default to prefixes disabled.

Underscores (_) are not allowed in the prefixes, as a username's prefix will be checked against the first underscore's index. Of course, if a username has no underscore or valid prefix, it'll be checked against all backends.

By default superuser checks are supported and enabled in all backends but Files (see details below). They may be turned off per backend by either setting individual disable options or not providing necessary options such as queries for DB backends, or for all of them by setting this global option to true:

auth_opt_disable_superuser true

Any other value or missing option will have superuser enabled.

Mosquitto 1.5 introduced a new ACL access value, MOSQ_ACL_SUBSCRIBE, which is similar to the classic MOSQ_ACL_READ value but not quite the same:

 *  MOSQ_ACL_SUBSCRIBE when a client is asking to subscribe to a topic string.
 *                     This differs from MOSQ_ACL_READ in that it allows you to
 *                     deny access to topic strings rather than by pattern. For
 *                     example, you may use MOSQ_ACL_SUBSCRIBE to deny
 *                     subscriptions to '#', but allow all topics in
 *                     MOSQ_ACL_READ. This allows clients to subscribe to any
 *                     topic they want, but not discover what topics are in use
 *                     on the server.
 *  MOSQ_ACL_READ      when a message is about to be sent to a client (i.e. whether
 *                     it can read that topic or not).

The main difference is that subscribe is checked at first, when a client connects and tells the broker it wants to subscribe to some topic, while read is checked when an actual message is being published to that topic, which makes it particular. So in practice you could deny general subscriptions such as # by returning false from the acl check when you receive MOSQ_ACL_SUBSCRIBE, but allow any particular one by returning true on MOSQ_ACL_READ. Please take this into consideration when designing your ACL records on every backend.

Also, these are the current available values from mosquitto:

#define MOSQ_ACL_NONE 0x00
#define MOSQ_ACL_READ 0x01
#define MOSQ_ACL_WRITE 0x02
#define MOSQ_ACL_SUBSCRIBE 0x04

If you're using prior versions then MOSQ_ACL_SUBSCRIBE is not available and you don't need to worry about it.

Any other options with a leading auth_opt_ are handed to the plugin and used by the backends. Individual backends have their options described in the sections below.

As of now every backend has proper but really ugly tests in place: they expect services running for each backend, and are also pretty outdated and cumbersome to work with in general. This issue captures these concerns and a basic plan to refactor tests: #67.

You may run all tests (see Testing X for each backend's testing requirements) like this:

make test

Backends may register which checks they'll run, enabling the option to only check user auth through some backends, for example an HTTP one, while delegating ACL checks to another backend, e.g. Files. By default, when the option is not present, all checks for that backend will be enabled (unless superuser is globally disabled in the case of superuser checks). For user and acl checks, at least one backend needs to be registered, either explicitly or by default.

You may register which checks a backend will perform with the option auth_opt_backend_register followed by comma separated values of the registered checks, e.g.:

auth_opt_http_register user
auth_opt_files_register user, acl
auth_opt_redis_register superuser

Possible values for checks are user, superuser and acl. Any other value will result in an error on plugin initialization.

The files backend implements the regular password and acl checks as described in mosquitto. Passwords should be in PBKDF2, Bcrypt or Argon2ID format (for other backends too), see Hashing for more details about different hashing strategies. Hashes may be generated using the pw utility (built by default when running make) included in the plugin (or one of your own). Passwords may also be tested using the pw-test package.

Usage of pw:

Usage of ./pw:
  -a string
    	algorithm: sha256 or sha512 (default "sha512")
  -c int
    	bcrypt ost param (default 10)
  -e string
    	salt encoding (default "base64")
  -h string
    	hasher: pbkdf2, argon2 or bcrypt (default "pbkdf2")
  -i int
    	hash iterations: defaults to 100000 for pbkdf2, please set to a reasonable value for argon2 (default 100000)
  -l int
    	key length, recommended values are 32 for sha256 and 64 for sha512
  -m int
    	memory for argon2 hash (default 4096)
  -p string
    	password
  -pl int
    	parallelism for argon2 (default 2)
  -s int
    	salt size (default 16)

For this backend passwords and acls file paths must be given:

auth_opt_files_password_path /path/to/password_file
auth_opt_files_acl_path /path/to/acl_file

The following are correctly formatted examples of password and acl files:

test1:PBKDF2$sha512$100000$2WQHK5rjNN+oOT+TZAsWAw==$TDf4Y6J+9BdnjucFQ0ZUWlTwzncTjOOeE00W4Qm8lfPQyPCZACCjgfdK353jdGFwJjAf6vPAYaba9+z4GWK7Gg==
test2:PBKDF2$sha512$100000$o513B9FfaKTL6xalU+UUwA==$mAUtjVg1aHkDpudOnLKUQs8ddGtKKyu+xi07tftd5umPKQKnJeXf1X7RpoL/Gj/ZRdpuBu5GWZ+NZ2rYyAsi1g==

user test1
topic write test/topic/1
topic read test/topic/2

user test2
topic read test/topic/+

user test3
topic read test/#

pattern read test/%u
pattern read test/%c

The ACLs file follows mosquitto's regular syntax: mosquitto(5).

There's no special superuser check for this backend since granting a user all permissions on # works in the same way. Furthermore, if this is the only backend registered, then providing no ACLs file path will default to grant all permissions for authenticated users when doing ACL checks (but then, why use a plugin if you can just use Mosquitto's static file checks, right?): if, instead, no ACLs file path is provided but there are more backends registered, this backend will default to deny any permissions for any user (again, back to basics).

Proper test files are provided in the repo (see test-files dir) and are needed in order to test this backend.

The postgres backend allows to specify queries for user, superuser and acl checks to be tested against your schema.

The following auth_opt_ options are supported:

Depending on the sslmode given, sslcert, sslkey and sslrootcert will be used. Options for sslmode are:

disable - No SSL
require - Always SSL (skip verification)
verify-ca - Always SSL (verify that the certificate presented by the server was signed by a trusted CA)
verify-full - Always SSL (verify that the certification presented by the server was signed by a trusted CA and the server host name matches the one in the certificate)

Queries work pretty much the same as in jpmen's plugin, so here's his discription (with some little changes) about them:

The SQL query for looking up a user's password hash is mandatory. The query
MUST return a single row only (any other number of rows is considered to be
"user not found"), and it MUST return a single column only with the PBKDF2
password hash. A single `'$1'` in the query string is replaced by the
username attempting to access the broker.

SELECT pass FROM account WHERE username = $1 limit 1


The SQL query for checking whether a user is a _superuser_ - and thus
circumventing ACL checks - is optional. If it is specified, the query MUST
return a single row with a single value: 0 is false and 1 is true.
A single `'$1`' in the query string is replaced by the
username attempting to access the broker. The following example uses the
same `users` table, but it could just as well reference a distinct table
or view.

SELECT COUNT(*) FROM account WHERE username = $1 AND super = 1

The SQL query for checking ACLs is optional, but if it is specified, the
`postgres` backend can try to limit access to particular topics or topic branches
depending on the value of a database table. The query MAY return zero or more
rows for a particular user, each returning EXACTLY one column containing a
topic (wildcards are supported). A single `'$1`' in the query string is
replaced by the username attempting to access the broker, and a single `'$2`' is
replaced with the integer value `1` signifying a read-only access attempt
(SUB) or `2` signifying a read-write access attempt (PUB).

In the following example, the table has a column `rw` containing 1 for
readonly topics, 2 for writeonly topics and 3 for readwrite topics:

SELECT topic FROM acl WHERE (username = $1) AND rw = $2

When option pg_superquery is not present, Superuser check will always return false, hence there'll be no superusers.

When option pg_aclquery is not present, AclCheck will always return true, hence all authenticated users will be authorized to pub/sub to any topic.

Example configuration:

auth_opt_pg_host localhost
auth_opt_pg_port 5432
auth_opt_pg_dbname appserver
auth_opt_pg_user appserver
auth_opt_pg_password appserver
auth_opt_pg_connect_tries 5
auth_opt_pg_userquery select password_hash from "user" where username = $1 and is_active = true limit 1
auth_opt_pg_superquery select count(*) from "user" where username = $1 and is_admin = true
auth_opt_pg_aclquery select distinct 'application/' || a.id || '/#' from "user" u inner join organization_user ou on ou.user_id = u.id inner join organization o on o.id = ou.organization_id inner join application a on a.organization_id = o.id where u.username = $1 and $2 = $2

DB connect tries: on startup, depending on pg_connect_tries option, the plugin will try to connect and ping the DB a max number of times or forever every 2 seconds. By default it will try to reconnect forever to maintain backwards compatibility and avoid issues when mosquitto starts before the DB service does, but you may choose to ping a max amount of times by setting any positive number. If given 0, the DB will try to connect only once, which would be the same as setting the option to 1.

For instructions on how to set a backend specific hasher or use the general one, see Hashing.

In order to test the postgres backend, a simple DB with name, user and password "go_auth_test" is expected.

User, database and test DB tables may be created with these commands:

create user go_auth_test with login password 'go_auth_test';
create database go_auth_test with owner go_auth_test;

create table test_user(
id bigserial primary key,
username character varying (100) not null,
password_hash character varying (200) not null,
is_admin boolean not null);

create table test_acl(
id bigserial primary key,
test_user_id bigint not null references test_user on delete cascade,
topic character varying (200) not null,
rw int not null);

The mysql backend works almost exactly as the postgres one, except for a few configurations and that options start with mysql_ instead of pg_. One change has to do with the connection protocol, either a Unix socket or tcp (options are unix or tcp). If unix socket is the selected protocol, then a socket path must be given:

auth_opt_mysql_protocol unix
auth_opt_mysql_socket /path/to/socket

The default protocol when the option is missing will be tcp, even if a socket path is given.

Another change has to do with sslmode options, with options being true, false, skip-verify or custom. When custom mode is given, sslcert, sslkey and sslrootcert paths are expected. If the option is not set or one or more required paths are missing, it will default to false.

Also, default host localhost and port 3306 will be used if none are given.

To allow native passwords, set the option to true:

auth_opt_mysql_allow_native_passwords true

Supported options for mysql are:

Finally, placeholders for mysql differ from those of postgres, changing from $1, $2, etc., to simply ?. These are some example queries for mysql:

User query:

SELECT pass FROM account WHERE username = ? limit 1

Superuser query:

SELECT COUNT(*) FROM account WHERE username = ? AND super = 1

Acl query:

SELECT topic FROM acl WHERE (username = ?) AND rw = ?

DB connect tries: on startup, depending on mysql_connect_tries option, the plugin will try to connect and ping the DB a max number of times or forever every 2 seconds. By default it will try to reconnect forever to maintain backwards compatibility and avoid issues when mosquitto starts before the DB service does, but you may choose to ping a max amount of times by setting any positive number. If given 0, the DB will try to connect only once, which would be the same as setting the option to 1.

For instructions on how to set a backend specific hasher or use the general one, see Hashing.

In order to test the mysql backend, a simple DB with name, user and password "go_auth_test" is expected.

User, database and test DB tables may be created with these commands:

create user 'go_auth_test'@'localhost' identified by 'go_auth_test';
create database go_auth_test;
grant all privileges on go_auth_test.* to 'go_auth_test'@'localhost';

create table test_user(
id mediumint not null auto_increment,
username varchar(100) not null,
password_hash varchar(200) not null,
is_admin boolean not null,
primary key(id)
);

create table test_acl(
id mediumint not null auto_increment,
test_user_id mediumint not null,
topic varchar(200) not null,
rw int not null,
primary key(id),
foreign key(test_user_id) references test_user(id)
ON DELETE CASCADE
ON UPDATE CASCADE
);

The sqlite backend works in the same way as postgres and mysql do, except that being a light weight db, it has fewer configuration options. The following auth_opt_ options are supported:

SQLite3 allows to connect to an in-memory db, or a single file one, so source maybe memory (not :memory:) or the path to a file db.

Example configuration:

sqlite_source /home/user/db/mosquitto_auth.db

Query parameters placeholders may be ? or $1, $2, etc.

sqlite_userquery SELECT pass FROM account WHERE username = ? limit 1

sqlite_superquery SELECT COUNT(*) FROM account WHERE username = ? AND super = 1

sqlite_aclquery SELECT topic FROM acl WHERE (username = ?) AND rw >= ?

DB connect tries: on startup, depending on sqlite_connect_tries option, the plugin will try to connect and ping the DB a max number of times or forever every 2 seconds. By default it will try to reconnect forever to maintain backwards compatibility and avoid issues when mosquitto starts before the DB service does, but you may choose to ping a max amount of times by setting any positive number. If given 0, the DB will try to connect only once, which would be the same as setting the option to 1.

For instructions on how to set a backend specific hasher or use the general one, see Hashing.

There are no requirements, as the tests create (and later delete) the DB and tables, or just use a temporary in memory one.

The jwt backend is for auth with a JWT remote API, a local DB, a JavaScript VM interpreter or an ACL file. Global otions for JWT are:

The following auth_opt_ options are supported by the jwt backend when remote is set to true:

URIs (like jwt_getuser_uri) are expected to be in the form /path. For example, if jwt_with_tls is false, jwt_host is localhost, jwt_port 3000 and jwt_getuser_uri is /user, mosquitto will send a POST request to https://localhost:3000/user to get a response to check against. How data is sent (either json encoded or as form values) and received (as a simple http status code, a json encoded response or plain text), is given by options jwt_response_mode and jwt_params_mode.

If the option jwt_superuser_uri is not set then superuser checks are disabled for this mode.

For all URIs, the backend will send a request with the Authorization header set to Bearer token, where token should be a correct JWT token and corresponds to the username received from Mosquitto.

When jwt_parse_token is set, the backend will parse the token using jwt_secret and extract the username from either the claim's Subject (sub field), or from the username field when jwt_userfield is set to Username. This username will be sent along other params in all requests, and the Authorization header will be set to Bearer token as usual.

Notice that failing to provide jwt_secret or passing a wrong one will result in an error when parsing the token and the request will not be made. Set these options only if you intend to keep the plugin synced with your JWT service and wish for the former to pre-parse the token.

When response mode is set to json, the backend expects the URIs to return a status code (if not 2XX, unauthorized) and a json response, consisting of two fields:

Ok: bool Error: string

If either the status is different from 2XX or Ok is false, auth will fail (not authenticated/authorized). In the latter case, an Error message stating why it failed will be included.

When response mode is set to status, the backend expects the URIs to return a simple status code (if not 2XX, unauthorized).

When response mode is set to text, the backend expects the URIs to return a status code (if not 2XX, unauthorized) and a plain text response of simple "ok" when authenticated/authorized, and any other message (possibly an error message explaining failure to authenticate/authorize) when not.

When params mode is set to json, the backend will send a json encoded string with the relevant data. For example, for acl check, this will get sent:

{ "topic": "mock/topic", "clientid": "mock_client", "acc": 1 //1 is read, 2 is write, 3 is readwrite, 4 is subscribe }

When set to form, it will send params like a regular html form post, so acc will be a string instead of an int.

Important: Please note that when using JWT, username and password are not needed, so for user and superuser check the backend will send an empty string or empty form values. On the other hand, all three cases will set the "authorization" header with the jwt token, which mosquitto will pass to the plugin as the regular "username" param.

Update: The username is expected to be set at the Subject field of the JWT claims (it was expected at Username earlier).

To clarify this, here's an example for connecting from a javascript frontend using the Paho MQTT js client (notice how the jwt token is set in userName and password has any string as it will not get checked):

initMqttClient(applicationID, mode, devEUI) {
    const hostname = window && window.location && window.location.hostname;
    let wsbroker = hostname;  //mqtt websocket enabled broker
    let wsport = 1884; // port for above
    let date = new Date();
    let clientid = this.getRand() + "_" + date.getTime();
    console.log("Trying to connect to mqtt with hostname: " + hostname + " and clientid " + clientid);
    let mqttClient = new window.Paho.MQTT.Client(wsbroker, wsport,
        clientid);

    mqttClient.onConnectionLost = function (responseObject) {
      console.log("connection lost: " + responseObject.errorMessage);
    };
    mqttClient.onMessageArrived = function (message) {
      console.log(message.destinationName, ' -- ', message.payloadString);
    };

    let that = this;

    let sslOption = true;
    if(hostname == "localhost") {
      sslOption = false;
    }

    let options = {
      timeout: 3,
      userName: this.getToken(),
      password: "any",
      useSSL: sslOption,
      keepAliveInterval: 3600,
      reconnect: true,
      onSuccess: function () {
        console.log("mqtt connected");
        // Connection succeeded; subscribe to our topic, you can add multile lines of these

        let topic = 'application/' + applicationID + '/device/' + devEUI + '/data';
        console.log("Subscribing to topic " + topic);
        mqttClient.subscribe(topic, {qos: 0});

      },
      onFailure: function (message) {
        console.log("Connection failed: " + message.errorMessage);
      }
    };

    mqttClient.connect(options);
    return mqttClient;
  }

When set to local mode, the backend will try to validate JWT tokens against a DB backend, either postgres or mysql, given by the jwt_db option. Options for the DB connection are the almost the same as the ones given in the Postgres and Mysql backends but prefixed with jwt_, e.g.:

auth_opt_jwt_pg_host localhost

The difference is that a specific jwt_userquery returning a count must be given since JWT backend won't use the password passed along by mosquitto, but instead should only use the username derived from the JWT token, e.g.:

auth_opt_jwt_userquery select count(*) from test_user where username = $1 limit 1

Thus, the following specific JWT local auth_opt_ options are supported:

Notice that general jwt_secret is mandatory when using this mode. jwt_userfield is still optional and serves as a mean to extract the username from either the claim's Subject (sub field), or from the username field when jwt_userfield is set to Username

As mentioned, only the userquery must not be prefixed by the underlying DB, and now expects an integer result instead of a password hash, as the JWT token needs no password checking. An example of a different query using either DB is given for the user query.

For postgres:

auth_opt_jwt_userquery select count(*) from "user" where username = $1 and is_active = true limit 1

For mysql:

auth_opt_jwt_userquery select count(*) from "user" where username = ? and is_active = true limit 1

Important note:

Since local JWT follows the underlying DB backend's way of working, both of these hold true:

• When option jwt_superquery is not present, Superuser check will always return false, hence there'll be no superusers.
• When option jwt_aclquery is not present, AclCheck will always return true, hence all authenticated users will be authorized to pub/sub to any topic.

When set to js JWT will act in JS mode, which allows to run a JavaScript interpreter VM to conduct checks. Options for this mode are:

This mode expects the user to define JS scripts that return a boolean result to the check in question.

The backend will pass mosquitto provided arguments along, that is token for both user and superuser check; token, topic, clientid and acc for ACL checks.

Optionally, username will be passed as an argument when auth_opt_jwt_parse_token option is set. As with remote mode, this will need auth_opt_jwt_secret to be set and correct, and auth_opt_jwt_userfield to be optionally set.

This is a valid, albeit pretty useless, example script for ACL checks (see test-files/jwt dir for test scripts):

function checkAcl(token, topic, clientid, acc) {
    if(token != "correct") {
        return false;
    }

    if(topic != "test/topic") {
        return false;
    }

    if(clientid != "id") {
        return false;
    }

    if(acc != 1) {
        return false;
    }

    return true;
}

checkAcl(token, topic, clientid, acc);

With auth_opt_jwt_parse_token the signature would be function checkAcl(token, topic, clientid, acc, username) instead.

Finally, this mode uses otto under the hood to run the scripts. Please check their documentation for supported features and known limitations.

When set to files JWT will run in Files mode, which allows to check user ACLs from a given file. These ACLs follow the exact same syntax and semantics as those from the Files backend.

Options for this mode are:

Notice there's no passwords file option since usernames come from parsing the JWT token and no password check is required. Thus, you should be careful about general ACL rules and prefer to explicitly set rules for each valid user.

If this shows to be a pain, I'm open to add a file that sets valid users, i.e. like the passwords file for regular Files backend but without actual passwords.

If you run into the case where you want to grant some general access but only to valid registered users, and find that duplicating rules for each of them in ACLs file is really a pain, please open an issue for discussion.

Since JWT needs not to check passwords, there's no need to configure a hasher.

If prefixes are enabled the client should prefix their JWT tokens with the prefix defined in the auth options: the plugin will strip the prefix from the value forwarded by Mosquitto so that the token is a valid JWT one. If the client fails to do so, this backend will still work, but since no prefix is recognized, this might incur in the overhead of potentially checking against some or all of the other backends before checking against the expected JWT one.

This backend expects the same test DBs from the Postgres and Mysql test suites.

The http backend is very similar to the JWT one, but instead of a jwt token it uses simple username/password to check for user auth, and username for superuser and acls.

The following auth_opt_ options are supported:

When response mode is set to json, the backend expects the URIs to return a status code (if not 2XX, unauthorized) and a json response, consisting of two fields:

Ok: bool Error: string

If either the status is different from 2XX or Ok is false, auth will fail (not authenticated/authorized). In the latter case, an Error message stating why it failed will be included.

When response mode is set to status, the backend expects the URIs to return a simple status code (if not 2XX, unauthorized).

When response mode is set to text, the backend expects the URIs to return a status code (if not 2XX, unauthorized) and a plain text response of simple "ok" when authenticated/authorized, and any other message (possibly an error message explaining failure to authenticate/authorize) when not.

When params mode is set to json, the backend will send a json encoded string with the relevant data. For example, for user authentication, this will get sent:

{ "username": "user", "password": "pass", "clientid": "clientid" }

When set to form, it will send params like a regular html form post.

This backend has no special requirements as the http servers are specially mocked to test different scenarios.

The redis backend allows to check user, superuser and acls in a defined format. As with the files and different DB backends, passwords hash must be stored and can be created with the pw utility.

For user check, Redis must contain the KEY username and the password hash as value.

For superuser check, a user will be a superuser if there exists a KEY username:su and it returns a string value "true".

Acls may be defined as user specific or for any user, and as subscribe only (MOSQ_ACL_SUBSCRIBE), read only (MOSQ_ACL_READ), write only (MOSQ_ACL_WRITE) or readwrite (MOSQ_ACL_READ | MOSQ_ACL_WRITE, not MOSQ_ACL_SUBSCRIBE) rules.

For user specific rules, SETS with KEYS "username:sacls", "username:racls", "username:wacls" and "username:rwacls", and topics (supports single level or whole hierarchy wildcards, + and #) as MEMBERS of the SETS are expected for subscribe, read, write and readwrite topics. username must be replaced with the specific username for each user containing acls.

For common rules, SETS with KEYS "common:sacls", "common:racls", "common:wacls" and "common:rwacls", and topics (supports single level or whole hierarchy wildcards, + and #) as MEMBERS of the SETS are expected for read, write and readwrite topics.

Finally, options for Redis are not mandatory and are the following:

auth_opt_redis_host localhost
auth_opt_redis_port 6379
auth_opt_redis_db dbname
auth_opt_redis_password pwd
auth_opt_redis_disable_superuser true
auth_opt_redis_mode cluster
auth_opt_redis_addresses host1:port1,host2:port2,host3:port3

When not present, host defaults to "localhost", port to 6379, db to 2 and no password is set.

If you want to use a Redis Cluster as your backend, you need to set auth_opt_redis_mode to cluster and provide the different addresses as a list of comma separated host:port strings with the auth_opt_redis_addresses options. If auth_opt_redis_mode is set to another value or not set, Redis defaults to single instance behaviour. If it is correctly set but no addresses are given, the backend will fail to initialize.

For instructions on how to set a backend specific hasher or use the general one, see Hashing.

In order to test the Redis backend, the plugin needs to be able to connect to a redis server located at localhost, on port 6379, without using password and that a database named 2 exists (to avoid messing with the commonly used 0 and 1).

All these requirements are met with a fresh installation of Redis without any custom configurations (at least when building or installing from the distro's repos in Debian based systems, and probably in other distros too).

After testing, db 2 will be flushed.

If you wish to test Redis auth, you may set the requirepass option at your redis.conf to match the password given in the test case:

requirepass go_auth_test

To test a Redis Cluster the plugin expects that there's a cluster with 3 masters at localhost:7000, localhost:7001 and localhost:7002. The easiest way to achieve this is just running some dockerized cluster such as https://github.com/Grokzen/docker-redis-cluster, which I used to test that the cluster mode is working, but building a local cluster should work just fine. I know that this test is pretty bad, and so are the general testing expectations. I'm looking to replace the whole suite with a proper dockerized environment that can also run automatic tests on pushes to ensure any changes are safe, but that will take some time.

The mongo backend, as the redis one, defines some formats to checks user, superuser and acls. Two collections are defined, one for users and the other for common acls.

In the first case, a user consists of a "username" string, a "password" string (as always, PBKDF2 hash), a "superuser" boolean, and an "acls" array of rules. These rules consis of a "topic" string and an int "acc", where 1 means read only, 2 means write only, 3 means readwrite and 4 means subscribe (see ACL access values section for more details).

Example user:

	{ "_id" : ObjectId("5a4e760f708ba1a1601fa40f"),
		"username" : "test",
		"password" : "PBKDF2$sha512$100000$os24lcPr9cJt2QDVWssblQ==$BK1BQ2wbwU1zNxv3Ml3wLuu5//hPop3/LvaPYjjCwdBvnpwusnukJPpcXQzyyjOlZdieXTx6sXAcX4WnZRZZnw==",
		"superuser" : true,
		"acls" : [
			{ "topic" : "test/topic/1", "acc" : 1 },
            { "topic" : "test/topic/1", "acc" : 4 },
			{ "topic" : "single/topic/+", "acc" : 1},
			{ "topic" : "hierarchy/#", "acc" : 1 },
			{ "topic" : "write/test", "acc" : 2 },
			{ "topic" : "test/readwrite/1", "acc" : 3 }
		]
	}

Common acls are just like user ones, but live in their own collection and are applicable to any user. Pattern matching against username or clientid acls should be included here.

Example acls:

	{ "_id" : ObjectId("5a4e760f708ba1a1601fa411"), "topic" : "pattern/%u", "acc" : 1 }
	{ "_id" : ObjectId("5a4e760f708ba1a1601fa413"), "topic" : "pattern/%c", "acc" : 1 }

Options for mongo are not mandatory and are the following:

auth_opt_mongo_host localhost
auth_opt_mongo_port 6379
auth_opt_mongo_dbname dbname
auth_opt_mongo_username user
auth_opt_mongo_password pwd
auth_opt_mongo_users users_collection_name
auth_opt_mongo_acls acls_collection_name
auth_opt_mongo_disable_superuser true
auth_opt_mongo_with_tls true
auth_opt_mongo_insecure_skip_verify false

The last two set names for the collections to be used for the given database.

When not set, these options default to:

host:            "localhost"
port:            "27017"
username:        ""
password:        ""
dbame:           "mosquitto"
users: 			 "users"
acls:  			 "acls"
with_tls:        "false"
insecure_skip_verify: "false"

If you experience any problem connecting to a replica set, please refer to this issue.

For instructions on how to set a backend specific hasher or use the general one, see Hashing.

Much like redis, to test this backend the plugin needs to be able to connect to a mongodb server located at localhost, on port 27017, without using username or password.

All this requirements are met with a fresh installation of MongoDB without any custom configurations (at least when building or installing from the distro's repos in Debian based systems, and probably in other distros too).

As with sqlite, this backend constructs the collections and inserts relevant data, which are whiped out after testing is done, so no user actions are required.

If you wish to test Mongo's auth, you'll need to run mongo with the --auth flag, have a user go_auth_test with password go_auth_test with the dbOwner role over the mosquitto_test DB and uncomment these lines from mongo_test.go:

	//authOpts["mongo_username"] = "go_auth_test"
	//authOpts["mongo_password"] = "go_auth_test"

Using the plugin package from Go, this project allows to write your own custom backend, compile it as a shared object and link to it from mosquitto-go-auth. Check Go pluing docs for more details.

In order to create your own plugin, you need to declare a main package that exposes the following functions (and uses the logrus package for logging):

package main

import (
	log "github.com/sirupsen/logrus"
)

func Init(authOpts map[string]string, logLevel log.Level) error {
	//Initialize your plugin with the necessary options
	return nil
}

func GetUser(username, password, clientid string) (bool, error) {
	return false, nil
}

func GetSuperuser(username string) (bool, error) {
	return false, nil
}

func CheckAcl(username, topic, clientid string, acc int) (bool, error) {
	return false, nil
}

func GetName() string {
	return "Your plugin name"
}

func Halt() {
	//Do whatever cleanup is needed.
}

Init should initialize anything that your plugin needs from the options passed in authOpts. These options may be given through the configuration as any other one, following the auth_opt_whatever_else pattern.

If you want to register your custom plugin, you need to add plugin to the auth_opt_backends option, and the option auth_opt_plugin_path with the absolute path to your-plugin.so.

GetUser, GetSuperuser and CheckAcl should respond with simple true/false to authenticate/authorize a user or pub/sub.

GetName is used only for logging purposes, as in debug level which plugin authenticated/authorized a user or pub/sub is logged.

You can build your plugin with:

go build -buildmode=plugin

Check the plugin directory for dummy example and makefile.

As this option is custom written by yourself, there are no tests included in the project.

The grpc backend allows to check for user auth, superuser and acls against a gRPC service.

The following auth_opt_ options are supported:

The last one, grpc_fail_on_dial_error indicates if failing to dial the service on initialization should be treated as a fatal error, or it should only be logged and then an attempt to redial should be made on every user or ACL check until the connection may be established. Then the backend will assume it has a healthy client and let the underlying package manage automatic reconnections.

The gRPC server should implement the service defined at grpc/auth.proto, which looks like this:

syntax = "proto3";

package grpc;

import "google/protobuf/empty.proto";


// AuthService is the service providing the auth interface.
service AuthService {

    // GetUser tries to authenticate a user.
    rpc GetUser(GetUserRequest) returns (AuthResponse) {}

    // GetSuperuser checks if a user is a superuser.
    rpc GetSuperuser(GetSuperuserRequest) returns (AuthResponse) {}

    // CheckAcl checks user's authorization for the given topic.
    rpc CheckAcl(CheckAclRequest) returns (AuthResponse) {}

    // GetName retrieves the name of the backend.
    rpc GetName(google.protobuf.Empty) returns (NameResponse) {}

    // Halt signals the backend to halt.
    rpc Halt(google.protobuf.Empty) returns (google.protobuf.Empty) {}

}

message GetUserRequest {
    // Username.
    string username = 1;
    // Plain text password.
    string password = 2;
    // The client connection's id.
    string clientid = 3;
}

message GetSuperuserRequest {
    // Username.
    string username = 1;
}

message CheckAclRequest {
    // Username.
    string username = 1;
    // Topic to be checked for.
    string topic = 2;
    // The client connection's id.
    string clientid = 3;
    // Topic access.
    int32 acc = 4;
}

message AuthResponse {
    // If the user is authorized/authenticated.
    bool ok = 1;
}

message NameResponse {
    // The name of the gRPC backend.
    string name = 1;
}

This backend has no special requirements as a gRPC server is mocked to test different scenarios.

The javascript backend allows to run a JavaScript interpreter VM to conduct checks. Options for this mode are:

This backend expects the user to define JS scripts that return a boolean result to the check in question.

The backend will pass mosquitto provided arguments along, that is:

• username, password and clientid for user checks.
• username for superuser checks.
• username, topic, clientid and acc for ACL checks.

This is a valid, albeit pretty useless, example script for ACL checks (see test-files/jwt dir for test scripts):

function checkAcl(username, topic, clientid, acc) {
    if(username != "correct") {
        return false;
    }

    if(topic != "test/topic") {
        return false;
    }

    if(clientid != "id") {
        return false;
    }

    if(acc != 1) {
        return false;
    }

    return true;
}

checkAcl(username, topic, clientid, acc);

Notice the password will be passed to the script as given by mosquitto, leaving any hashing to the script.

This backend has no special requirements as javascript test files are provided to test different scenarios.

See the official MQTT authentication & authorization guide for isntructions on using the plugin with the LoRa Server project.

Please be aware that, since Docker isn't actively used by the maintainer of this project, support for issues regarding Docker, the provided images and building Docker images is very limited and usually driven by other contributors.

Only images for x86_64/AMD64 and ARMv7 have been tested. ARMv6 and ARM64 hardware was not available to the contributor creating the build workflow.

Prebuilt images are provided on Dockerhub under iegomez/mosquitto-go-auth. To run the latest image, use the following command and replace /conf with the location of your .conf files: docker run -it -p 1884:1884 -p 1883:1883 -v /conf:/etc/mosquitto iegomez/mosquitto-go-auth

This project provides a Dockerfile for building a Docker container that contains mosquitto and the mosquitto-go-auth plug-in.

Building containers is only supported on x86_64/AMD64 machines with multi-arch build support via Docker Buildx. This allows building containers for x86_64/AMD64, ARMv6, ARMv7 and ARM64 on a single x86_64/AMD64 machine. For further instructions regarding Buildx, please refer to its documentation ond Docker's website.

• clone this repository: git clone https://github.com/iegomez/mosquitto-go-auth.git
• change into the project folder cd mosquitto-go-auth
• build containers for your desired architectures: docker buildx build --platform linux/amd64,linux/arm64,linux/arm/v7 .

Since there are several issues with using alpine based images we are using debian:stable-slim for both our build and final image. The final image size is about 60 MB.

Documented issues:

• #14
• #15
• #20

The Dockerfile compiles mosquitto using the source code from the version specified by MOSQUITTO_VERSION.

Mosquitto released versions can be found at https://mosquitto.org/files/source/

Since tests require multiple backends (PostgreSQL, Mysql, Redis...), a Dockerfile.test provide and image with all required backend. To use it:

docker build -t mosquitto-go-auth.test -f Dockerfile.runtest .
docker run --rm -ti mosquitto-go-auth.test sh ./run-test-in-docker.sh

Or using local source (avoid the need to rebuild image):

docker run -v $(pwd):/app --rm -ti mosquitto-go-auth.test sh ./run-test-in-docker.sh

You may even specify the command to run after backends are started, which allow to run only some tests or even get a shell inside the containers:

docker run -v $(pwd):/app --rm -ti mosquitto-go-auth.test sh ./run-test-in-docker.sh make test-backends

docker run -v $(pwd):/app --rm -ti mosquitto-go-auth.test sh ./run-test-in-docker.sh bash

mosquitto-go-auth is distributed under the MIT license. See also LICENSE.