snmpd.conf - configuration file for the Net-SNMP SNMP agent

snmpd.conf is the configuration file which defines how the Net-SNMP SNMP agent operates. These files may contain any of the directives found in the DIRECTIVES section below. If this file is not found (or does not contain any access control directives), the agent will run but will not respond to any requests.

First, make sure you have read the snmp_config(5) manual page that describes how the Net-SNMP configuration files operate, where they are located and how they all work together.

Also, you might consider looking into the snmpconf application (perl script) which can help you build an snmpd.conf file by prompting you for information. You should try it. Really. Go ahead. Right now. Run:

snmpconf -g basic_setup

to get you started. See the snmpconf(1) manual page for more information.

The Net-SNMP SNMP agent reports much of its information through queries to the 1.3.6.1.4.1.2021 section of the MIB tree. Every MIB in this section has the following table entries in it.

.1 -- index

This is the table's index numbers for each of the DIRECTIVES listed below.

.2 -- name

The name of the given table entry. This should be unique, but is not required to be.

.100 -- errorFlag

This is a flag returning either the integer value 1 or 0 if an error is detected for this table entry.

.101 -- errorMsg

This is a DISPLAY-STRING describing any error triggering the errorFlag above.

.102 -- errorFix

If this entry is set to the integer value of 1 AND the errorFlag defined above is indeed a 1, a program or script will get executed with the table entry name from above as the argument. The program to be executed is configured in the net-snmp-config.h file at compile time.

Directives

Errors

To enable AgentX support in the SNMP master agent, put the following line in your snmpd.conf file:

master agentx

See README.agentx for further details.

AgentXSocket addr

This defines the address the master agent listens at. The default is /var/agentx/master. By default the Unix Domain socket is accessible only to subagents which have the same userid as the agent. Another possibility is localhost:705

AgentXTimeout addr

Defines the timeout period for an AgentX request. Default is 1 second.

AgentXRetries addr

Defines the number of retries for an AgentX request. Default is 5 retries.

You can also put the following in your subagent.conf file (where "subagent" is the name you used in your init_snmp("subagent") api call:

agentPingInterval NUM

This makes the subagent try and reconnect every NUM seconds to the master if it ever becomes (or starts) disconnected.

To enable and SMUX based sub-agent, such as gated, use the smuxpeer configuration entry

smuxpeer OID PASS

For gated a sensible entry might be smuxpeer .1.3.6.1.4.1.4.1.3 secret

If the agent is built with support for the UCD-DLMOD-MIB it is capable of loading agent MIB modules dynamically at startup through the dlmod directive and during runtime through use of the UCD-DLMOD-MIB. The following directive loads the shared object module file PATH which uses the module name prefix NAME.

dlmod NAME PATH

snmpd supports the View-Based Access Control Model (VACM) as defined in RFC 2575. To this end, it recognizes the following keywords in the configuration file: com2sec, group, access, and view as well as some easier-to-use wrapper directives: rocommunity, rwcommunity, rouser, rwuser. If IPv6 support has been enabled, the rocommunity6 and rwcommunity6 tokens will also be available. This section defines how to configure the snmpd program to accept various types and levels of access.

rouser USER [noauth|auth|priv [OID]]

rwuser USER [noauth|auth|priv [OID]]

Creates a SNMPv3 USM user in the VACM access configuration tables. It is more efficient (and powerful) to use the combined group, access, and view directives instead, but these wrapper directives are much simpler.

The minimum level of authentication and privacy the user must use is specified by the first token (which defaults to "auth"). The OID parameter restricts access for that user to everything below the given OID.

rocommunity COMMUNITY [SOURCE [OID]]

rwcommunity COMMUNITY [SOURCE [OID]]

These create read-only and read-write communities that can be used to access the agent. They are a quick wrapper around the more complex and powerful com2sec, group, access, and view directive lines. They are not as efficient either, as groups aren't created so the tables are possibly larger. In other words: don't use these if you have complex situations to set up. If your setup is simple or you don't mind a small performance hit, use these directives.

The format of the SOURCE is token is described in the com2sec directive section below. The OID token restricts access for that community to everything below that given OID.

rocommunity6 COMMUNITY [SOURCE [OID]]

rwcommunity6 COMMUNITY [SOURCE [OID]]

They are the alternative directives to the rocommunity, rwcommunity for the transport domain UDPIPv6. They are only valid in specifing UDPIPv6 as transport domain.

The format of the SOURCE is token is described in the com2sec6 directive section below. The OID token restricts access for that community to everything below that given OID.

com2sec [-Cn CONTEXT] NAME SOURCE COMMUNITY

This directive specifies the mapping from a source/community pair to a security name. SOURCE can be a hostname, a subnet, or the word "default". A subnet can be specified as IP/MASK or IP/BITS. For example, given a directive "com2sec myLocal 10.10.10.0/24 public" then this would match requests from IP addresses 10.10.10.0 through to 10.10.10.255, but not one from 10.10.11.1 or similar. The first source/community combination that matches the incoming packet is selected. If a CONTEXT is specified, the community name gets mapped to the SNMPv3 CONTEXT of the same name, otherwise the default context ("") is used.

com2sec6 [-Cn CONTEXT] NAME SOURCE COMMUNITY

This directive is the IPv6 version of com2sec. A subnet can be specified as IPv6/IPv6MASK or IPv6/BITS. It is only valid in specifing UDPIPv6 as transport domain.

group NAME MODEL SECURITY

This directive defines the mapping from securitymodel/securityname to group. MODEL is one of v1, v2c, or usm.

access NAME CONTEXT MODEL LEVEL PREFX READ WRITE NOTIFY

The access directive maps from group/security model/security level to a view. MODEL is one of any, v1, v2c, or usm. LEVEL is one of noauth, auth, or priv. PREFX specifies how CONTEXT should be matched against the context of the incoming pdu, either exact or prefix. READ, WRITE and NOTIFY specifies the view to be used for the corresponding access. For v1 or v2c access, LEVEL will be noauth, and CONTEXT will be empty.

view NAME TYPE SUBTREE [MASK]

This defines the named view. TYPE is either included or excluded. MASK is a list of hex octets, separated by '.' or ':'. The MASK defaults to "ff" if not specified.

The reason for the mask is, that it allows you to control access to one row in a table, in a relatively simple way. As an example, as an ISP you might consider giving each customer access to his or her own interface:

view cust1 included interfaces.ifTable.ifEntry.ifIndex.1 ff.a0
view cust2 included interfaces.ifTable.ifEntry.ifIndex.2 ff.a0

(interfaces.ifTable.ifEntry.ifIndex.1 == .1.3.6.1.2.1.2.2.1.1.1,
ff.a0 == 11111111.10100000. which nicely covers up and including
the row index, but lets the user vary the field of the row)
VACM Examples:

#        sec.name  source          community
com2sec  local     localhost       private
com2sec  mynet     10.10.10.0/24   public
com2sec  public    default         public
com2sec6 mynet     fec0::/64       public

# sec.model sec.name group mygroup v1 mynet group mygroup v2c mynet group mygroup usm mynet group local v1 local group local v2c local group local usm local group public v1 public group public v2c public group public usm public

# incl/excl subtree mask view all included .1 80 view system included system fe view mib2 included .iso.org.dod.internet.mgmt.mib-2 fc

# context sec.model sec.level prefix read write notify access mygroup "" any noauth exact mib2 none none access public "" any noauth exact system none none access local "" any noauth exact all all all

Default VACM model

The default configuration of the agent, as shipped, is functionally equivalent to the following entries:

com2sec	public	default	public
group	public	v1	public
group	public	v2c	public
group	public	usm	public
view 	all	included	.1
access	public	""	any	noauth	exact	all	none	none

engineID STRING

The snmpd agent needs to be configured with an engineID to be able to respond to SNMPv3 messages. With this configuration file line, the engineID will be configured from STRING. If this directive is omitted, the default value of the engineID is built from 2 fairly random elements: a random number and the current time in seconds.

createUser [-e ENGINEID] username (MD5|SHA) authpassphrase [DES] [privpassphrase]

MD5 and SHA are the authentication types to use. The only privacy protocol currently supported is DES. If the privacy passphrase is not specified, it is assumed to be the same as the authentication passphrase. Note that the users created will be useless unless they are also added to the VACM access control tables described above.

SHA authentication and DES privacy require OpenSSL to be installed and the agent to be built with OpenSSL support. MD5 authentication may be used without OpenSSL.

Warning: the minimum pass phrase length is 8 characters.

SNMPv3 users can be created at runtime using the snmpusm(1) command.

Instead of figuring out how to use this directive and where to put it (see below), just run "net-snmp-config --create-snmpv3-user" instead, which will add one of these lines to the right place.

This directive should be placed into the /var/lib/snmp/snmpd.conf file instead of the other normal locations. The reason is that the information is read from the file and then the line is removed (eliminating the storage of the master password for that user) and replaced with the key that is derived from it. This key is a localized key, so that if it is stolen it can not be used to access other agents. If the password is stolen, however, it can be.

If you need to localize the user to a particular EngineID (this is useful mostly in the similar snmptrapd.conf file), you can use the -e argument to specify an EngineID as a hex value (EG, "0x01020304").

If you want to generate either your master or localized keys directly, replace the given password with a hexstring (preceeded by a "0x") and precede the hex string by a -m or -l token (respectively). EGs:

[these keys are *not* secure but are easy to visually parse for
counting purposes.  Please generate random keys instead of using
these examples]

createUser myuser SHA -l 0x0001020304050607080900010203040506070809 AES -l 0x00010203040506070809000102030405 createUser myuser SHA -m 0x0001020304050607080900010203040506070809 AES -m 0x0001020304050607080900010203040506070809

Due to the way localization happens, localized privacy keys are expected to be the length needed by the algorithm (128 bits for all supported algorithms). Master encryption keys, though, need to be the length required by the authentication algorithm not the length required by the encrypting algorithm (MD5: 16 bytes, SHA: 20 bytes).

syslocation STRING

syscontact STRING

sysname STRING

Sets the system location, system contact or system name for the agent. This information is reported in the 'system' group the mibII tree. Ordinarily these objects (sysLocation.0, sysContact.0 and sysName.0) are read-write. However, specifying the value for one of these objects by giving the appropriate token makes the corresponding object read-only, and attempts to set the value of the object will result in a notWritable error response.

sysservices NUMBER

Sets the value of the system.sysServices.0 object. For a host, a good value is 72.

sysdescr STRING

sysobjectid OID

Sets the system description or object ID for the agent. Although these values are not SNMP-writable, it is conceivable that a network administrator may wish to configure them to something other than the default values.

agentaddress [<transport-specifier>:]<transport-address>[,...]

Makes the agent list on the specified comma-separated list of listening addresses instead of the default behaviour, which is to listen on UDP port 161 on all IPv4 interfaces. See the section LISTENING ADDRESSES in the snmpd(8) manual page for more information about the format of listening addresses. For example, specifying agentaddress 161,tcp:161,localhost:9161 will make the agent listen on UDP port 161 on all IPv4 interfaces, TCP port 161 on all IPv4 interfaces and UDP port 9161 only on the interface associated with the localhost address.

agentgroup groupid

Change to this gid after opening port. The groupid may refer to a group by name or a number if the group number starts with '#'. For example, specifying agentgroup snmp will cause the agent to run as the snmp group or agentgroup #10 will cause the agent to run as the group with groupid 10.

agentuser uid

Change to this uid after opening port. The userid may refer to a user by name or a number if the user number starts with '#'. For example, specifying agentuser snmp will cause the agent to run as the snmp user or agentuser #10 will cause the agent to run as the user with userid 10.

interface NAME TYPE SPEED

For interfaces where the agent fails to guess correctly on the type and speed, this directive can supply additional information. TYPE is a type value as given in the IANAifType-MIB.

ignoredisk STRING

When scanning for available disk devices the agent might block in trying to open all possible disk devices. This might lead to a timeout when walking the device tree. Sometimes the next walk will run without timeout, sometimes it will timeout every time you try it.

If you experience such behaviour you might add this directive and give all device names not to be checked (i.e. opened). You might have more than one such directive in your configuration file stating all devices not to be opened. You might also specify those devices using wildcards similar to the syntax you can use in a bourne shell (see examples below).

Note: For a list of devices scanned for every system please consult the sources (host/hr_disk.c) and check for the Add_HR_Disk_entry() calls relevant for your type of OS.

Examples:

ignoredisk /dev/rdsk/c0t2d0

This directive prevents the device /dev/rdsk/c0t2d0 from being scanned.

ignoredisk /dev/rdsk/c0t[!6]d0

This directive prevents all devices /dev/rdsk/c0tXd0 except .../c0t6d0 from being scanned. For most systems similar is the following directive:

ignoredisk /dev/rdsk/c0t[0-57-9a-f]d0

ignoredisk /dev/rdsk/c1*

This directive prevents all devices whose device names start with /dev/rdsk/c1 from being scanned.

ignoredisk /dev/rdsk/c?t0d0

This directive prevents all devices /dev/rdsk/cXt0d0 ('X' might be any char) from being scanned.

You might use more than one such wildcard expression in any such directive.

storageUseNFS NUMBER

Setting storageUseNFS to 1 causes all NFS and NFS-like file systems to be marked as 'Network Disks' in the hrStorageTable. This is according to RFC 2790. Not setting storageUseNFS or setting it to 2 causes NFS and NFS-like file systems to be marked as 'Fixed Disks' as it has been in previous versions of the ucd-snmp SNMP agent.

authtrapenable NUMBER

Setting authtrapenable to 1 enables generation of authentication failure traps. The default value is disabled(2). Ordinarily the corresponding object (snmpEnableAuthenTraps.0) is read-write, but setting its value via this token makes the object read-only, and attempts to set the value of the object will result in a notWritable error response.

override [-rw] OID TYPE VALUE

This directive allows you to override a particular OID with a different value (and possibly a different type of value). The -rw flag will allow snmp SETs to modify it's value as well. (note that if you're overriding original functionality, that functionality will be entirely lost. Thus SETS will do nothing more than modify the internal overridden value and will not perform any of the original functionality intended to be provided by the MIB object. It's an emulation only.) An example:

override sysDescr.0 octet_str "my own sysDescr"

That line will set the sysDescr.0 value to "my own sysDescr" as well as make it modifiable with SNMP SETs as well (which is actually illegal according to the MIB specifications).

Note that care must be taken when using this. For example, if you try to override a property of the 3rd interface in the ifTable with a new value and later the numbering within the ifTable changes it's index ordering you'll end up with problems and your modified value won't appear in the right place in the table.

Valid TYPEs are: integer, uinteger, octet_str, object_id, counter, null (for gauge's, use "uinteger"; for bit strings, use "octet_str"). Note that setting an object to "null" effectively delete's it as being accessible. No VALUE needs to be given if the object type is null.

More types should be available in the future.

trapcommunity STRING

This defines the default community string to be used when sending traps. Note that this command must be used prior to any of the following three commands that are intended use this community string.

trapsink HOST [COMMUNITY [PORT]]

trap2sink HOST [COMMUNITY [PORT]]

informsink HOST [COMMUNITY [PORT]]

These commands define the hosts to receive traps (and/or inform notifications). The daemon sends a Cold Start trap when it starts up. If enabled, it also sends traps on authentication failures. Multiple trapsink, trap2sink and informsink lines may be specified to specify multiple destinations. Use trap2sink to send SNMPv2 traps and informsink to send inform notifications. If COMMUNITY is not specified, the string from a preceding trapcommunity directive will be used. If PORT is not specified, the well known SNMP trap port (162) will be used.

trapsess [SNMPCMD_ARGS] HOST

This is a more generic trap configuration token that allows any type of trap destination to be specified with any version of SNMP. See the snmpcmd(1) manual page for further details on the arguments that can be passed as SNMPCMD ARGS . In addition to the arguments listed there, the special argument -Ci specifies that you want inform notifications to be used instead of unacknowledged traps (this requires that you also specify a version number of v2c or v3 as well).

proxy [-Cn CONTEXTNAME] [SNMPCMD ARGS] HOST OID [REMOTEOID]

This token specifies that any incoming requests under OID should be proxied on to another HOST instead. If a CONTEXTNAME is specified, it assigns the proxied tree to a particular context name within the local agent. This is the proper way to query multiple agents through a single proxy. Assign each remote agent to a different context name. Then you can use "snmpwalk -n contextname1" to walk one remote proxied agent and "snmpwalk -n contextname2" to walk another, assuming you are using SNMPv3 to talk to the proxy (snmpv1 and snmpv2c context mappings aren't currently supported but might be in the future). Optionally, relocate the local OID tree to the new location at the REMOTEOID. To authenticate to HOST you should use the appropriate set of SNMPCMD ARGS. See the snmpcmd(1) manual page for details.

Examples:

# assigns the entire mib tree on remotehost1 to the context of the
# same name:
proxy -Cn remotehost1 -v 1 -c public remotehost1 .1.3
# ditto, but for remotehost 2
proxy -Cn remotehost2 -v 1 -c public remotehost2 .1.3
# proxies only the ucdavis enterprises tree to the remote host using snmpv1
proxy -v 1 -c public remotehost .1.3.6.1.4.1.2021
# uses v3 to access remotehost and converts the remote .1.3.6.1.2.1.1
# oid to local .1.3.6.1.3.10 oid (another way to access mulitple hosts
# without using contexts)
proxy -v 3 -l noAuthNoPriv -u user remotehost .1.3.6.1.3.10 .1.3.6.1.2.1.1

pass MIBOID EXEC

(If you're writing perl scripts, please see the embedded perl support information later in this manual page). Passes entire control of MIBOID to the EXEC program. The EXEC program is called in one of the following three ways:

EXEC -g MIBOID

EXEC -n MIBOID

These call lines match to SNMP get and getnext requests. It is expected that the EXEC program will take the arguments passed to it and return the appropriate response through it's stdout.

The first line of stdout should be the MIB OID of the returning value. The second line should be the TYPE of value returned, where TYPE is one of the text strings: string, integer, unsigned, objectid, timeticks, ipaddress, counter, or gauge. The third line of stdout should be the VALUE corresponding with the returned TYPE.

For instance, if a script was to return the value integer value "42" when a request for .1.3.6.1.4.100 was requested, the script should return the following 3 lines:

.1.3.6.1.4.100

integer

42

To indicate that the script is unable to comply with the request due to an end-of-mib condition or an invalid request, simple exit and return no output to stdout at all. An SNMP error will be generated corresponding to the SNMP noSuchName response.

EXEC -s MIBOID TYPE VALUE

For SNMP set requests, the above call method is used. The TYPE passed to the EXEC program is one of the text strings: integer, counter, gauge, timeticks, ipaddress, objid, or string, indicating the type of value passed in the next argument.

Return nothing to stdout, and the set will assumed to have been successful. Otherwise, return one of the following error strings to signal an error: not-writable, or wrong-type and the appropriate error response will be generated instead.

Note:

By default, the only community allowed to write (ie snmpset) to your script will be the "private" community,or community #2 if defined differently by the "community" token discussed above. Which communities are allowed write access are controlled by the RWRITE definition in the snmplib/snmp_impl.h source file.

Example (in snmpd.conf):

pass .1.3.6.1.4.1.2021.255 /path/to/local/passtest

pass_persist MIBOID EXEC

Passes entire control of MIBOID to the EXEC program. Similar to pass, but the EXEC program continues to run after the initial request is answered. Also, pass and pass_persist block till they return.

Upon initialization, EXEC is passed the string "PING\n" in stdin, and it should respond by printing "PONG\n" to stdout.

For get and getnext requests, EXEC program is passed two lines, the command (get or getnext) and the MIB OID. It should return three lines, the MIB OID, the TYPE of value returned, the VALUE corresponding with the returned TYPE.

For example, if the value for .1.3.6.1.4.100 was requested, the following 2 lines would be passed in to stdin:

get

.1.3.6.1.4.100

To return the value, say, 42, the script would write to stdout:

.1.3.6.1.4.100

integer

42

To indicate that the script is unable to comply with the request due to an end-of-mib condition or an invalid request, print "NONE\n" to stdout.

Example (in snmpd.conf):

pass_persist .1.3.6.1.4.1.2021.255 /path/to/local/pass_persisttest

Warning: though embedded perl is working, not much functionality has been implemented yet and thus writing mib module pieces for the agent within perl is not trivial at this point. It should get better in future releases.

The net-snmp package has ability to call perl scripts directly inside the agent through embedded perl technology (similar to mod_perl for the apache web server). This must be turned on at compile time by passing --enable-embedded-perl to the configure script when the package is built. To see if your package was built with embedded perl, run "net-snmp-config --configure-options" and see if that flag was used.

If compiled in, it defines the following snmpd.conf configuration directives:

disablePerl true

This will turn off perl support entirely. If the embedded perl support stops working due to a change in perl, etc, this will stop any calls to the perl core.

perlInitFile FILE

Use FILE as the initialization file. This file is normally /usr/share/snmp/snmp_perl.pl but this token can override that default. This file performs any in-perl initialization that is needed before the rest of the perl directives (below) are called. It is sourced once just before the first perl directive is parsed. See the default file for an example of the initialization it performs.

perl EXPRESSION

Calls perl to evaluate an expression. Normally you'd want to do something like register a function to call when an OID is requested, but you can really do anything perl related you want. For example:

perl print "hello world\n";

is the most basic hello world example.

The init script by default initializes a $agent variable which is a pointer to a NetSnmp::agent object through which you can register callbacks when a section of the OID tree is hit:

perl use Data::Dumper;
perl sub myroutine  { print "got called: ",Dumper(@_),"\n"; }
perl $agent->register('mylink', '.1.3.6.1.8765', \&myroutine);

Sourcing an external file:

perl 'do /path/to/file.pl';

No better examples exist at the moment, unfortunately. Look for improved support in future releases. Comments on how this looks as an architecture are certainly appreciated now.

Warning: this implementation has not been extensively tested and is additionally not known to be entirely complete. The concepts defined here should function appropriately, however, but no promises are made at this time.

If your agent was compiled with support for the DISMAN-EVENT-MIB (you can enable this by running the net-snmp configure script with the --with-mib-modules=disman/event-mib argument) you have support for having the agent check its own data at regular intervals and to send out traps when certain conditions occur. Traps are sent when expressions are first noticed, not once per evaluation. Once a test expression fires a trap, the test will have to fail again before a new trap is sent. See the DISMAN-EVENT-MIB documentation for more details. This can be configured either using the MIB tables themselves or by using these special key words:

agentSecName NAME

The DISMAN-EVENT-MIB support requires a valid user name for which to scan your agent with. This can either be specified using the agentSecName token or by explicitly list one on the "monitor" lines described below using the -u switch. Either way, a "rouser" line (or equivalent access control settings) must also be specified with the same security name name. If you need an example, just do something like this:

agentSecName internal
rouser internal

And everything below should work just fine.

monitor [OPTIONS] NAME EXPRESSION

This token tells the agent to monitor itself for problems based on EXPRESSION. Expression is a simple expression based on an oid, a comparison operator (!=, ==, <, <=, >, >=) and an integer value (see the examples below). NAME is merely an arbitrary name of your choosing for administrative purposes only. OPTIONS include the following possibilities:

-t

Use a threshold monitor instead of a boolean monitor. This means that expression should be a low and high value. If the given OID passes beyond the high value, a rising alarm will triggered. A falling alarm will then be triggered after it falls below the low value.

-r FREQUENCY

Monitors the given expression every FREQUENCY seconds. The default is 600 (10 minutes).

-u SECNAME

Use the SECNAME security name for scanning the local host. Specifically, this SECNAME must then be given access control rights via something like the "rouser" snmpd.conf token for this expression to be valid at all. If not specified, it uses the default security name specified by the agentsecname snmpd.conf token. Either the -u flag or a valid agentsecname token must be specified (and that name must be given proper access control rights via a "rouser" token).

-o OID

Specifies additional object values to be delivered with in the resulting trap in addition to the normal trap objects. This is useful for obtaining other columns in the table for the row that triggered the expression. See the examples below for more details.

-e EVENTNAME

Specifies an event name that describes what to do when the trigger is fired. Currently, this must be the name of a notificationEvent event as described below.

The following example configuration checks the hrSWRunPerfTable table (listing running processes) for any process which is consuming > 10Mb of memory. It performs this check every 600 seconds (the default). For every process it finds exceeding the limit, it will end out exactly one notification. In addition to the normal hrSWRunPerfMem oid and value sent in the trap, the hrSWRunName object will also be sent. Note that the hrSWRunName object actually occurs in a different table, but since the indexes to the two tables are the same this works out alright.

rouser me
monitor -u me -o sysUpTime.0 -o hrSWRunName "high process memory" hrSWRunPerfMem > 10000

The above line would produce a trap which, when formated by snmptrapd, would look like:

2002-04-05 13:33:53 localhost.localdomain [udp:127.0.0.1:32931]:
        sysUpTimeInstance = Timeticks: (1629) 0:00:16.29        snmpTrapOID.0 = OID: mteTriggerFired    mteHotTrigger = high process memory     mteHotTargetName =      mteHotContextName =     mteHotOID = OID: hrSWRunPerfMem.1968    mteHotValue = 28564     hrSWRunName.1968 = "xemacs"

This shows my xemacs process using 28Mb of resident memory. Which, considering it's xemacs, is not that surprising.

Threshold example:

monitor -t -r 15 -o prNames -o prErrMessage "process table" prErrorFlag 0 1

notificationEvent NAME NOTIFICATION [[-w] OID_OBJECT ...]

This will create a notification event, which can be fired by attaching
it to a monitor using a monitor's -e switch and an identical NAME
field.  The NOTIFICATION to be sent should be the OID of a
notification.  Additional objects can be included, and by default the
suffix of the row/object being monitored will be appended to the
object identifier unless it's told not to be a wild-card object by
prefixing it with the -w switch.  EG, if you're monitoring the ifTable
and you want your trap to include the ifDescr object for the row that
was discovered, don't add the -w switch and the .INDEX field will be
appended.  If the OID is fully qualified (EG, "sysContact.0") and no
instance suffix should be appended to it then add a -w switch before
it.  See the linkUpDownNotifications token below for example usage of
this token to send the linkUp and linkDown traps.
linkUpDownNotifications yes

This will make the DISMAN-EVENT-MIB support watch the ifTable to
determine when a network interface is taken up or down.  When this
happens, a linkUp or linkDown notification will be triggered.  This is
exactly equivalent to doing:

notificationEvent  linkUpTrap    linkUp   ifIndex ifAdminStatus ifOperStatus
notificationEvent  linkDownTrap  linkDown ifIndex ifAdminStatus ifOperStatus

monitor -r 60 -e linkUpTrap "Generate linkUp" ifOperStatus != 2 monitor -r 60 -e linkDownTrap "Generate linkDown" ifOperStatus == 2

defaultMonitors yes

By default, the agent and the DISMAN-EVENT-MIB support do nothing until configured. Typically people wish to watch a bunch of tables within the UCD-SNMP-MIB which are designed specifically for reporting problems. If the "defaultMonitors yes" line is put into the snmpd.conf file (which must be accompanied by an appropriate agentSecName line and a rouser line), the following monitoring conditions will be installed:

monitor	-o prNames -o prErrMessage "process table" prErrorFlag != 0
monitor	-o memErrorName -o memSwapErrorMsg "memory" memSwapError != 0
monitor	-o extNames -o extOutput "extTable" extResult != 0
monitor	-o dskPath -o dskErrorMsg "dskTable" dskErrorFlag != 0
monitor	-o laNames -o laErrMessage  "laTable" laErrorFlag != 0
monitor	-o fileName -o fileErrorMsg  "fileTable" fileErrorFlag != 0

If you're trying to figure out aspects of the various mib modules (possibly some that you've added yourself), the following may help you spit out some useful debugging information. First off, please read the snmpd manual page on the -D flag. Then the following configuration snmpd.conf token, combined with the -D flag, can produce useful output:

injectHandler HANDLER modulename

This will insert new handlers into the section of the mib tree referenced by "modulename". The types of handlers available for insertion are:

stash_cache  - Caches information returned from the lower level.  This
               greatly help the performance of the agent, at the cost
               of caching the data such that its no longer "live" for
               30 seconds (in this future, this will be configurable).
               Note that this means snmpd will use more memory as well
               while the information is cached.  Currently this only
               works for handlers registered using the table_iterator
               support, which is only a few mib tables.  To use it,
               you need to make sure to install it before the
               table_iterator point in the chain, so to do this:

injectHandler stash_cache NAME table_iterator

If you want a table to play with, try walking the nsModuleTable with and without this injected.

debug - Prints out lots of debugging information when the -Dhelper:debug flag is passed to the snmpd application.

read_only - Forces turning off write support for the given module.

serialize - If a module is failing to handle multiple requests properly (using the new 5.0 module API), this will force the module to only receive one request at a time.

bulk_to_next - If a module registers to handle getbulk support, but for some reason is failing to implement it properly, this module will convert all getbulk requests to getnext requests before the final module receives it.

Figuring out module names

To figure out which modules you can inject things into, snmpwalk the nsModuleTable which will give you a list of all named modules registered within the agent.

See the EXAMPLE.CONF file in the top level source directory for a more detailed example of how the above information is used in real examples.

The Net-SNMP agent can be forced to re-read its configuration files. It can be told to do so by one of two ways:

1.

An snmpset of integer(1) to UCD-SNMP-MIB::versionUpdateConfig.0 (.1.3.6.1.4.1.2021.100.11.0)

2.

A "kill -HUP" signal sent to the snmpd agent process.

/etc/snmp/snmpd.conf

snmpconf(1), snmpusm(1), snmp.conf(5), snmp_config(5), snmpd(8), EXAMPLE.conf, read_config(3).