7   Backend


The backend daemon is the central component in the Clixon architecture. It consists of a main module and a number of dynamically loaded plugins. The backend has four APIs:

An XML file read at startup, possibly amended with -o options.
Internal interface / IPC
A NETCONF socket to frontend clients. This is by default a UNIX domain socket but can also be an IPv4 or IPv6 TCP socket but with limited functionality.
XML (or JSON) files storing configuration. The three main datastores are candidate, running and startup. A user edits the candidate datastore, commits the changes to running which triggers callbacks in plugins.
Backend plugins configure the base system with application-specific APIs. These API:s depend on how the underlying system is configured, examples include configuration files or a socket.

Note that a user typically does not access the datastores directly, it is possible to read, but write operations should not be done, since the backend daemon in most cases uses a datastore cache.

7.1   Command-line options

The backend have the following command-line options:
-h Help
-D <level> Debug level
-f <file> Clixon config file
-E <dir> Extra configuration directory
-l <option> Log on (s)yslog, std(e)rr, std(o)ut, (n)one or (f)ile. Syslog is default. If foreground, then syslog and stderr is default.
-C <format> Dump configuration options on stdout after loading. Format is one of xml|json|text
-d <dir> Specify backend plugin directory
-p <dir> Add Yang directory path (see CLICON_YANG_DIR)
-b <dir> Specify datastore directory
-F Run in foreground, do not run as daemon
-z Kill other config daemon and exit
-a <family> Internal backend socket family: UNIX|IPv4|IPv6
-u <path|addr> Internal socket domain path or IP addr (see -a)
-P <file> Process ID filename
-1 Run once and then quit (do not wait for events)
-s <mode> Specify backend startup mode: none|startup|running|init)
-c <file> Load extra XML configuration file, but do not commit.
-q Quit startup directly after upgrading and print result on stdout
-U <user> Run backend daemon as this user AND drop privileges permanently
-g <group> Client membership required to this group (default: clicon)
-y <file> Load yang spec file (override yang main module)
-o <option=value>
 Give configuration option overriding config file (see clixon-config.yang)

7.1.1   Logging and debugging

In case of debugging, the backend can be run in the foreground and with debug flags:

clixon_backend -FD 1

Note that debug levels can be combined as described in Section debugging.

Logging is by default on syslog. Alternatively, logging can be made on a file using the -l option:

clixon_backend -lf<file>

When run in foreground, logging is by default done on both syslog and stderr.

In a debugging mode, it can be useful to run in once-only mode, where the backend quits directly after starting up, instead of waiting for events:

clixon_backend -F1D 1

It may be useful to see all config options after load, taking into account default values, config-dirs and option overriding. This is normally dumped in debug level 1.

But you can also make an explicit dump of all config options on stdout using the -C option:

clixon_backend -1C xml

7.2   Startup

The backend can perform startup in four different modes. The difference is how the running state is handled, i.e., what state the system is in when you start the daemon and how loading the configuration affects it:

Do not touch running state. Typically after crash when running state and db are synched.
Initialize running state. Start with a completely clean running state.
Commit running db configuration into running state. Typically after reboot if a persistent running db exists.
Commit startup configuration into running state. After reboot when no persistent running db exists.

Use the -s option to select startup mode, example:

clixon_backend -s running

You may also add a default method in the configuration file:

<clixon-config xmlns="http://clicon.org/config">

When loading the startup/tmp configuration, the following actions are performed by the system:

  • Check syntax errors,
  • Upgrade callbacks.
  • Validation of the XML against the current Yang models
  • If errors are detected, enter failsafe mode.

The following config option is related to startup:

Enable process-control of restconf daemon, ie start/stop restconf daemon internally using fork/exec. Disable if you start the restconf daemon by other means.

7.3   IPC Socket

Frontends:           +----------+
CLI,          IPC    | backend  |
netconf      <--->   | daemon   |
restconf             |          |

The Clixon backend creates a socket that the frontend clients can connect to. Communication is made over this IPC socket using internal Netconf. The following config options are related to the internal socket:

Address family for communicating with clixon_backend. One of: UNIX, IPv4, or IPv6. Can also be set with -a command-line option. Default is UNIX which denotes a UNIX domain socket.
If the address family of the socket is AF_UNIX: Unix socket for communicating with clixon_backend. If the family is AF_INET it denottes the IPv4 address;
Inet socket port for communicating with clixon_backend (only IPv4|IPv6). Default is port 4535.
Group membership to access clixon_backend UNIX socket. Default is clicon. This is not available for IP sockets.

7.4   Backend files

The following config options control files related to the backend:

Location of backend .so plugins. Load all .so plugins in this dir as backend plugins in alphabetical order
Regexp of matching backend plugins in CLICON_BACKEND_DIR. default: *.so
Process-id file of backend daemon

7.5   Backend plugins

This section describes backend-specific plugins, see Plugin section for a general description of plugins.

7.5.1   Clixon_plugin_init

Apart from the generic plugin callbacks (init, start, etc), the following callbacks are specific to the backend:

Called just before server daemonizes(forks). Not called if in foreground.
Called after the server has daemonized and before privileges are dropped.
Provide state data XML from a plugin
Reset system status
General-purpose upgrade called once when loading the startup datastore
Transaction callbacks are invoked for two reasons: validation requests or commits. These callbacks are further described in transactions section.

7.5.2   Registered callbacks

The callback also supports three forms of registered callbacks:

  • rpc_callback_register() - for user-defined RPC callbacks, see Plugin section.
  • action_callback_register() - for user-defined Action callbacks.
  • upgrade_callback_register() - for upgrading, see Upgrade section.
  • clixon_pagination_cb_register() - for pagination, as described in Pagination section.

7.6   Transactions

Clixon follows NETCONF in its validate and commit semantics. Using the CLI or another frontend, you edit the candidate configuration, which is first validated for consistency and then committed to the running configuration.

A clixon developer writes commit functions to incrementally update a system state based on configuration changes. Writing transsaction callbacks is a core function of the clixon system.

The NETCONF validation and commit operation is implemented in clixon by a transaction mechanism, which ensures that user-written plugin callbacks are invoked atomically and revert on error.


Clixon currently runs in a single thread and all transactions run from begin, to end without interruption. However, it is strongly recommended that the callbaks be thread-safe for future compatibility.


Clixon can only guarantee a consistent system state if you follow the guidelines below. Behavior that is not documented here is not guaranteed and subject to change without notice.

The phases of a transaction are:


The begin callback indicates that a transaction is starting, but that the system level validation has not begun yet. Not much has happened at this point, and you should not rely on the transaction data or the XML ADD/DEL/CHANGE flags.

The intent of this callback is to notify the plugin that a transaction is beginning, and that it should allocate any resources necessary to handle the coming transaction.


The validate callback is triggered by a client requesting a validation of the current change set. Before this callback, clixon has completed the system level YANG validation. All transaction data is valid at this point (including XML flags). In this callback, the plugin further validates the transaction data and returns a success/failure indication. It must not change the state of the system in any way. Successful completion should guarantee that the commit will be without error. See the commit information below on why it is important to try. If a failure status is returned to clixon then the transaction is cancelled, and the transaction abort callback will be called from clixon.

You should place as much of the validation in the YANG model specification as possible. This does two things; it allows the user of the YANG model to understand the constraints clearly, and it is more efficient when the YANG validation catches problems before a transaction cycle happens.


This callback is used to indicate to the plugin that the validation was successful and indicates that the commit is coming next. The transaction data will be valid when this callback is called. The complete callback must not change the state of the system.

The usefulness of this callback is debatable, and it is possible that this callback will be removed in a future version.


This callback is called when the client requests a commit. The transaction data is valid, and both the system and plugin validations of the transaction data has completed without error.

The callback should perform the actions required to change the system state as indicated by the transaction target database.

Ideally all possible errors should be detected in the validate callback, but this clearly is a dream. A commit failure is a major event and leaves the system in an inconsistent state. In the event of an error, the callback must return an error status. Clixon will initiate its error processing to roll back the system to its state prior to the beginning of the transaction.

Transaction when commit done. After this, the source db is copied to target and default values removed.

The end callback is called when a transaction has successfully completed.

You should not depend upon transaction data being valid at this point and no changes to the system state may occur (changes at this point prevent the possibility of error handling).

Any resources allocated in the begin callback should be released at this point. When the callback returns, the transaction is complete.


When the revert callback is called the plugin must revert the system to the state prior (atomicity) to the beginning of the transaction. The transaction data is valid and is the same as in the commit callback. The code in the revert callback must assure that the system state matches the source database before returning.

A revert callback will be followed by an abort callback.


If a validate or commit operation fails, the terminal action is to call the abort callback rather than the end callback. The common purpose of this callback is to release any resources allocated in the begin callback.

If an abort callback occurs after a commit, then the revert callback will be called prior to the abort. The plugin developer may do the actual reversion of the commit in the revert or abort callback, or split the duties as desired, but upon completion of the abort callback, it must be guaranteed that the system state will be as before the begin callback.

7.6.1   Transaction Examples

In a system with two plugins, for example, a transaction sequence looks like the following:

Backend   Plugin1    Plugin2
|          |          |
+--------->+--------->+ begin
|          |          |
+--------->+--------->+ validate
|          |          |
+--------->+--------->+ complete
|          |          |
+--------->+--------->+ commit
|          |          |
+--------->+--------->+ commit_done
|          |          |
+--------->+--------->+ end

If an error occurs in the commit call of plugin2, for example, the transaction is aborted and the commit reverted:

Backend   Plugin1    Plugin2
|          |          |
+--------->+--------->+ begin
|          |          |
+--------->+--------->+ validate
|          |          |
+--------->+---->X    + commit error
|          |          |
+--------->+          + revert
|          |          |
+--------->+--------->+ abort

7.6.2   Callbacks

In the the context of a callback, there are two XML trees:

The original XML tree, such as “running”
The new XML tree, such as “candidate”

There are three vectors pointing into the XML trees:

The delete vector consists of XML nodes in “src” that are removed in “target”
The add vector consists of nodes that exists in “target” and do not exist in “src”
The change vector consists of nodes that exists in both “src” and “target” but are different

All transaction callbacks are called with a transaction-data argument (td). The transaction data describes a system transition from a src to target state. The struct contains source and target XML tree (e.g. candidate/running) in the form of XML tree vectors (dvec, avec, cvec) and associated lengths. These contain the difference between src and target XML, ie “what has changed”. It is up to the validate callbacks to ensure that these changes are OK. It is up to the commit callbacks to enforce these changes in the configuration of the system.

The “td” parameter can be accessed with the following functions:

uint64_t transaction_id(transaction_data td)
Get the unique transaction-id
void   *transaction_arg(transaction_data td)
Get plugin/application specific callback argument
int     transaction_arg_set(transaction_data td, void *arg)
Set callback argument
cxobj  *transaction_src(transaction_data td)
Get source database xml tree
cxobj  *transaction_target(transaction_data td)
Get target database xml tree
cxobj **transaction_dvec(transaction_data td)
Get vector of xml nodes that are deleted src->target
size_t  transaction_dlen(transaction_data td)
Get length of delete xml vector
cxobj **transaction_avec(transaction_data td)
Get vector of xml nodes that are added src->target
size_t  transaction_alen(transaction_data td)
Get length of add xml vector
cxobj **transaction_scvec(transaction_data td)
Get vector of xml nodes that changed
cxobj **transaction_tcvec(transaction_data td)
Get vector of xml nodes that changed
size_t  transaction_clen(transaction_data td)
Get length of changed xml vector


A programmer can also use XML flags that are set in “src” and “target” XML trees to identify what has changed. The following flags are used to makr the trees:

All deleted XML nodes in “src” and all its descendants
All added XML nodes in “target” and all its descendants
All changed XML nodes in both “src” and “target” and all its descendants. Also all ancestors of all added, deleted and changed nodes.

For example, assume the tree (A B) is replaced with (B C), then the two trees are marked with the following flags:

          src            target
           o CHANGE        o CHANGE
           |               |
           o CHANGE        o CHANGE
          /|               |\
DELETE   A B               B C ADD

You can use functions, such as xpath_vec_flag() to query for changed nodes:

if (xpath_vec_flag(xcur, nsc, "//symbol/foo", XML_FLAG_ADD, &vec, &veclen) < 0)
for (i=0; i<veclen; i++){
   xn = vec[i];

7.7   Privileges

The backend process itself does not really require any specific access, but it may be an important topic for an application using clixon when the plugins are designed. A plugin may need to access privileged system resources (such as configure files).

The backend itself is usually started as root: sudo clixon_backend -s init, which means that the plugins also run as root (being part of the same process).

The backend can also be started as a non-root user. However, you may need to set some config options to allow user write access, for example as follows(there may be others):


7.7.1   Dropping privileges

You may want to start the backend as root and then drop privileges to a non-root user which is a common technique to limit exposure of exploits.

This can be done either by command line-options: sudo clicon_backend -s init -U clicon or (more generally) using configure options:


This will initialize resources as root and then permanently drop uid:s to the unprivileged user (clicon in the example abobe). It will also change ownership of several files to the user, including datastores and the clicon socket (if the socket is unix domain).

Note that the unprivileged user must exist on the system, see Install section.

7.7.2   Drop privileges temporary

If you drop privileges permanently, you need to access all privileged resources initially before the drop. For a plugin designer, this means that you need to access privileges system resources in the plugin_init or plugin_start callbacks. The transaction callbacks, for example, will be run in unprivileged mode.

An alternative is to drop privileges temporary and then be able to raise privileges when needed:


In this mode, a plugin callback (eg commit), can temporarily raise the privileges when accessing system resources, and the lower them when done.

An example C-code for raising privileges in a plugin is as follows:

uid_t euid = geteuid();
... make high privilege stuff...