snmpd.examples - Man Page

The default agent behaviour (listing on the standard SNMP UDP port on all interfaces) is equivalent to the directive:

or simply

The agent can be configured to only accept requests sent to the local loopback interface (again listening on the SNMP UDP port), using:

or

It can be configured to accept both UDP and TCP requests (over both IPv4 and IPv6), using:

Other combinations are also valid.

The agent can be configured to relinquish any privileged access once it has opened the initial listening ports.  Given a suitable "snmp" group (defined in /etc/group), this could be done using the directives:

agentuser  nobody
agentgroup snmp

A similar effect could be achieved using numeric UID and/or GID values:

agentuser  #10
agentgroup #10

Rather than being generated pseudo-randomly, the engine ID for the agent could be calculated based on the MAC address of the second network interface (eth1), using the directives:

or it could be calculated from the (first) IP address, using:

or it could be specified explicitly, using:

The following directives will create three users, all using exactly the same authentication and encryption settings:

createUser me     MD5 "single pass phrase"
createUser myself MD5 "single pass phrase" AES
createUser andI   MD5 "single pass phrase" AES "single pass phrase"

Note that this defines three distinct users, who could be granted different levels of access.  Changing the passphrase for any one of these would not affect the other two.

Separate pass phrases can be specified for authentication and encryption:

Remember that these createUser directives should be defined in the /var/lib/net-snmp/snmpd.conf file, rather than the usual location.

The SNMPv3 users defined above can be granted access to the full MIB tree using the directives:

rouser me
rwuser onering

Or selective access to individual subtrees using:

rouser myself   .1.3.6.1.2
rwuser andI     system

Note that a combination repeating the same user, such as:

rouser onering
rwuser onering

should not be used. This would configure the user onering with read-only access (and ignore the rwuser entry altogether). The same holds for the community-based directives.

The directives:

rocommunity public
rwcommunity private

would define the commonly-expected read and write community strings for SNMPv1 and SNMPv2c requests.  This behaviour is not configured by default, and would need to be set up explicitly.

Note:

It would also be a very good idea to change private to something a little less predictable!

A slightly less vulnerable configuration might restrict what information could be retrieved:

or the management systems that settings could be manipulated from:

or a combination of the two.

This last pair of settings are equivalent to the full VACM definitions:

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

#                  sec.model  sec.name
group  worldGroup  v1         public
group  worldGroup  v2c        public
group  myGroup     v1         mynet
group  myGroup     v2c        mynet

#              incl/excl   subtree     [mask]
view   all     included    .1
view   sysView included    system

#              context model level   prefix  read    write  notify (unused)
access  worldGroup  ""  any  noauth  exact   system  none   none
access  myGroup     ""  any  noauth  exact   all     all    none

There are several points to note in this example:

The group directives must be repeated for  both SNMPv1 and SNMPv2c requests.

The com2sec security name is distinct from the community string that is mapped to it. They can be the same ("public") or different ("mynet"/"private") - but what appears in the group directive is the security name, regardless of the original community string.

Both of the view directives are defining simple OID subtrees, so neither of these require an explicit mask. The same holds for the "combined subtree2 view defined below. In fact, a mask field is only needed when defining row slices across a table (or similar views), and can almost always be omitted.

In general, it is advisible not to mix traditional and VACM-based access configuration settings, as these can sometimes interfere with each other in unexpected ways.  Choose a particular style of access configuration, and stick to it.

A similar configuration could also be configured as follows:

view   sys2View included    system
view   sys2View included    .1.3.6.1.2.1.25.1

authcommunity read       public  default      -v sys2View
authcommunity read,write private 10.10.10.0/8

This mechanism allows multi-subtree (or other non-simple) views to be used with the one-line rocommunity style of configuration.

It would also support configuring "write-only" access, should this be required.

The full contents of the 'system' group (with the exception of sysUpTime) can be explicitly configured using:

# Override 'uname -a' and hardcoded system OID - inherently read-only values
sysDescr     Universal Turing Machine mk I
sysObjectID  .1.3.6.1.4.1.8072.3.2.1066

# Override default values from 'configure' - makes these objects read-only
sysContact   Alan.Turing@pre-cs.man.ac.uk
sysName      tortoise.turing.com
sysLocation  An idea in the mind of AT

# Standard end-host behaviour
sysServices  72

The list of devices probed for potential inclusion in the hrDiskStorageTable (and hrDeviceTable) can be amended using any of the following directives:

which prevents the device /dev/rdsk/c0t2d0 from being scanned,

ignoredisk /dev/rdsk/c0t[!6]d0
ignoredisk /dev/rdsk/c0t[0-57-9a-f]d0

either of which prevents all devices /dev/rdsk/c0tXd0 (except .../c0t6d0) from being scanned,

which prevents all devices whose device names start with /dev/rdsk/c1 from being scanned, or

which prevents all devices /dev/rdsk/cXt0d0 (where 'X' is any single character) from being scanned.

The list of services running on a system can be monitored (and provision made for correcting any problems), using:

# At least one web server process must be running at all times
proc    httpd
procfix httpd  /etc/rc.d/init.d/httpd restart

# There should never be more than 10 mail processes running
#    (more implies a probable mail storm, so shut down the mail system)
proc    sendmail   10
procfix sendmail  /etc/rc.d/init.d/sendmail stop

# There should be a single network management agent running
#   ("There can be only one")
proc    snmpd    1  1

Also see the "DisMan Event MIB" section later on.

The state of disk storage can be monitored using:

includeAllDisks 10%
disk /var 20%
disk /usr  3%
#  Keep 100 MB free for crash dumps
disk /mnt/crash  100000

A simple check for an overloaded system might be:

A more refined check (to allow brief periods of heavy use, but recognise sustained medium-heavy load) might be:

TODO

file FILE [MAXSIZE]
logmatch NAME PATH CYCLETIME REGEX

Configuring the agent to report invalid access attempts might be done by:

authtrapenable 1
trapcommunity  public
trap2sink      localhost

Alternatively, the second and third directives could be combined (and an acknowledgement requested) using:

A configuration with repeated sink destinations, such as:

trapsink       localhost
trap2sink      localhost
informsink     localhost

should NOT be used, as this will cause multiple copies of each trap to be sent to the same trap receiver.

TODO - discuss SNMPv3 traps

trapsess  snmpv3 options  localhost:162

TODO - mention trapd access configuration

The simplest configuration for active self-monitoring of the agent, by the agent, for the agent, is probably:

# Set up the credentials to retrieve monitored values
createUser    _internal MD5 "the first sign of madness"
iquerySecName _internal
rouser        _internal

# Active the standard monitoring entries
defaultMonitors         yes
linkUpDownNotifications yes

# If there's a problem, then tell someone!
trap2sink localhost

The first block sets up a suitable user for retrieving the information to by monitored, while the following pair of directives activates various built-in monitoring entries.

Note that the DisMan directives are not themselves sufficient to actively report problems - there also needs to be a suitable destination configured to actually send the resulting notifications to.

A more detailed monitor example is given by:

This defines an explicit boolean monitor entry, looking for any process using more than 10MB of active memory.  Such processes will be reported using the (standard) DisMan trap mteTriggerFired, but adding an extra (wildcarded) varbind hrSWRunName.

This entry also specifies an explicit user (me, as defined earlier) for retrieving the monitored values, and building the trap.

Objects that could potentially fluctuate around the specified level are better monitored using a threshold monitor entry:

This will send a mteTriggerRising trap whenever the incoming traffic rises above (roughly) 500 kB/s on any network interface, and a corresponding mteTriggerFalling trap when it falls below 100 kB/s again.

Note that this monitors the deltas between successive samples (-D) rather than the actual sample values themselves.  The same effect could be obtained using:

The linkUpDownNotifications directive above is broadly equivalent to:

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

This defines the traps to be sent (using notificationEvent), and explicitly references the relevant notification in the corresponding monitor entry (rather than using the default DisMan traps).

The defaultMonitors directive above is equivalent to a series of (boolean) monitor entries:

and will send a trap whenever any of these entries indicate a problem.

An alternative approach would be to automatically invoke the corresponding "fix" action:

setEvent   prFixIt  prErrFix = 1
monitor -e prFixIt "procTable" prErrorFlag   != 0

(and similarly for any of the other defaultMonitor entries).

The agent could be configured to reload its configuration once an hour, using:

Alternatively this could be configured to be run at specific times of day (perhaps following rotation of the logs):

The one-shot style of scheduling is rather less common, but the secret SNMP virus could be activated on the next occurance of Friday 13th using:

Old Style

exec [MIBOID] NAME PROG ARGS"
sh   [MIBOID] NAME PROG ARGS"
execfix NAME PROG ARGS"

New Style

extend [MIBOID] NAME PROG ARGS"
extendfix [MIBOID] NAME PROG ARGS"

One-Shot

Persistent

If embedded perl support is enabled in the agent, the default initialisation is equivalent to the directives:

disablePerl  false
perlInitFile /usr/share/snmp/snmp_perl.pl

The main mechanism for defining embedded perl scripts is the perl directive.  A very simple (if somewhat pointless) MIB handler could be registered using:

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

This relies on the $agent object, defined in the example snmp_perl.pl file.

A more realistic MIB handler might be:

XXX - WHAT ???

Alternatively, this code could be stored in an external file, and loaded using:

TODO

dlmod NAME PATH"

A configuration for acting as a simple proxy for two other SNMP agents (running on remote systems) might be:

com2sec -Cn rem1context  rem1user default  remotehost1
com2sec -Cn rem2context  rem2user default  remotehost2

proxy -Cn rem1context  -v 1 -c public  remotehost1  .1.3
proxy -Cn rem2context  -v 1 -c public  remotehost2  .1.3

(plus suitable access control entries).

The same proxy directives would also work with (incoming) SNMPv3 requests, which can specify a context directly. It would probably be more sensible to use contexts of remotehost1 and remotehost2 - the names above were chosen to indicate how these directives work together.

Note that the administrative settings for the proxied request are specified explicitly, and are independent of the settings from the incoming request.

An alternative use for the proxy directive is to pass part of the OID tree to another agent (either on a remote host or listening on a different port on the same system), while handling the rest internally:

This mechanism can be used to link together two separate SNMP agents.

A less usual approach is to map one subtree into a different area of the overall MIB tree (either locally or on a remote system):

# uses SNMPv3 to access the MIB tree .1.3.6.1.2.1.1 on 'remotehost'
# and maps this to the local tree .1.3.6.1.3.10
proxy -v 3 -l noAuthNoPriv -u user remotehost .1.3.6.1.3.10 .1.3.6.1.2.1.1

smuxsocket 127.0.0.1
smuxpeer .1.3.6.1.2.1.14 ospf_pass

The Net-SNMP agent could be configured to operate as an AgentX master agent (listening on a non-standard named socket, and running using the access privileges defined earlier), using:

master agentx
agentXSocket /tmp/agentx/master
agentXPerms  0660 0550 nobody snmp

A sub-agent wishing to connect to this master agent would need the same agentXSocket directive, or the equivalent code:

netsnmp_ds_set_string(NETSNMP_DS_APPLICATION_ID, NETSNMP_DS_AGENT_X_SOCKET, 
                      "/tmp/agentx/master");

A loopback networked AgentX configuration could be set up using:

agentXSocket   tcp:localhost:705
agentXTimeout  5
agentXRetries  2

on the master side, and:

agentXSocket   tcp:localhost:705
agentXTimeout  10
agentXRetries  1
agentXPingInterval 600

on the client.

Note that the timeout and retry settings can be asymmetric for the two directions, and the sub-agent can poll the master agent at regular intervals (600s = every 10 minutes), to ensure the connection is still working.