TUT:Writing a MIB Module

From Net-SNMP Wiki

Jump to: navigation, search

This page is the first in a series to guide you through the creation of custom extensions to the core Net-SNMP daemon, usually called "agent". In this article we will describe how to build your extension into the daemon. The two following articles will describe how to put your changes into a shared object (see Writing a Dynamically Loadable Object) and a separate daemon (see Writing a Subagent).

Extensions (or "MIB modules") usually make use of the API provided by Net-SNMP and are therefore written in C. For instructions on writing extensions in Perl, please read Extending SNMPd using Perl.

Contents

[edit]

The "Big Picture"

The main task of custom extensions is to register a callback function which handles requests for a specific OID. Version 5.0 of the Net-SNMP agent tries to make this task as easy as possible by providing so called helper modules. Each request is handled in a chain in Net-SNMP. You can inject helper modules into this chain, so that helper modules are called before control reaches the actual callback function. This can, for example, be used to install a cache in front of the actual data acquisition, which may be costly.

We will now walk you through the process of developing increasingly more complex MIB objects, beginning with a scalar, then a table, persistent data, notifications, and so on.

[edit]

MIBs For Dummies and mib2c

Writing modules for MIBs is a long and sometimes troublesome process, much like writing a parser. And just like writing a parser, you don't have to do everything by hand: MIBs can already be processed by computers pretty well, so there is no need to start from square one every time. The tool to convert an existing MIB to some C code is called mib2c and is part of the Net-SNMP distribution. There is a tutorial on using mib2c and mib2c.mfd.conf to generate a code template for a table. (TODO: Move that page to this wiki.)

[edit]

A simple scalar attached to a variable

The first example, scalar_int.c, only consists of an initialization function, init_scalar_int, which registers the module-global integer example1 with the SNMP agent (aka. "daemon"). The prototype and name of the initialization function must follow this schema: 

void init_module_name (void);

In the example code, the module's name is "scalar_int", hence the name of the initialization function.

Within the initialization function, the API function netsnmp_register_int_instance is called to actually register the integer at a specified OID. The prototype of that API function is: 

int netsnmp_register_int_instance (const char *name,
    oid *reg_oid, size_t reg_oid_len, int *it,
    Netsnmp_Node_Handler *subhandler);

The first argument, name, is a the short description of the object being registered. The second argument is the OID to assign to this object — the length of the OID is stored in the third argument. It is recommended to use the OID_LENGTH macro which will only work with statically allocated OIDs (arrays work, pointers don't).

For the sake of simplicity the default subhandler is used in the example code, therefore NULL is passed as fifth argument. The default handler allows the integer to be read with a GET request as well as set with a SET request.

Simple integer registration functions, including read-only registration and the default handlers, are declared in agent/instance.h.

[edit]

A simple scalar with the value returned from code

Here is some example code about writing a MIB Module that describes how to implement a generic instance (not tied to a variable, like the above example) as well as how to delay answering queries for a bit (in the example, an alarm is set to add a delay to the answer).

[edit]

A table of data, stored within the agent

Here is a more complex example that implements a table, where the table data is completely contained within the agent.

[edit]

Sending SNMP notifications (traps and informs) from inside the agent.

Here is an example of how to use the agent's internal notification sending API to send a notification to all of the agent's trap receivers.

[edit]

The older (and thus backward compatible) ucd-snmp 4.X API

Of course, you can continue to write code using the older API set. A tutorial on it can be found here.

[edit]

The MIB Module API

The new API is documented more completely here. Additionally, see the complete list of (mostly) documented available helpers and other information here.

[edit]

Compiling in your new MIB module

There are a few ways to get your new MIB module loaded and accessible via SNMP requests. We'll discuss all three ways separately. To make this easy to test the procedures outlined below, we've provided three simple mib modules which implement the three simple scalars in the NET-SNMP-TUTORIAL-MIB MIB. To see how MIBs can be properly used by the tools, please see the mib-options tutorial.

  1. Compile it into the master agent.

Lets assume you're going to compile in a new mib module. For our example, lets use the example mib module and it's header file. To do this, you would put the nstAgentModuleObject.h and nstAgentModuleObject.c files into the net-snmp source code directory. You do this by copying them into a agent/mibgroup/nstAgentModuleObject.h and agent/mibgroup/nstAgentModuleObject.c file.

Next, you have to configure the package to find them and compile them into the agent. To do this, you run the configure script giving it the extra module names you want it to load: 

 % ./configure --with-mib-modules="nstAgentModuleObject"

If you had multiple modules to include (like a second "XXX" module, for example), you can separate them with spaces inside the quotes (e.g., --with-mib-modules="nstAgentModuleObject XXX").

Note that nstAgentModuleObject is the prefix and the configure script will actually look for a nstAgentModuleObject.h and a nstAgentModuleObject.c file. You must have a .h file and you can not get it to work with just a .c file.

Build your new agent with your new code in it by running make: 

 % make

Finally, install the whole lot by running make install: 

 % make install

You can test out the functionality by starting the snmpd agent: 

 % /usr/local/sbin/snmpd

And then running snmpget and snmpset on the scalar object [proper authentication information not shown in this command]: 

 % snmpget localhost NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0
 NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0 = INTEGER: 1
 
 % snmpset localhost NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0 = 5
 NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0 = INTEGER: 5
 
 % snmpget localhost NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0
 NET-SNMP-TUTORIAL-MIB::nstAgentModuleObject.0 = INTEGER: 5

You can also compile your code into a "subagent" which then attaches itself to the master agent using the AgentX subagent protocol. Our libraries provide support to make this easy to do and this is discussed in greater detail in a later section.

Finally, you can also compile your code into pluggable shared object and tell the snmpd agent to load it. This is also discussed in greater detail in a later section .

[edit]

Set Processing

To process an SNMP-SET, the agent must use a series of calls to the mib module code to ensure that processing of all sets in the incoming packet can be successful. This gives you or other mib modules the chance to bail out early on in the transaction sequence and thus stop all of the transactions in the set from happening. This is important for continuity. However, it makes set code processing a bit more complex. Let's examine a simple state diagram that the master agent uses at each step of the way:

set-actions.jpg

In a perfect operation with no failures, we take the vertical path on the left. If any of the mib modules being acted upon returns an error of any kind, we will branch to the right to one of the failure states where you must clean up and possibly undo what you did in previous steps.

[edit]

Tutorial Sections

Retrieved from "http://www.net-snmp.org/wiki/index.php/TUT:Writing_a_MIB_Module"
Personal tools