inetd - Solaris Management Facility delegated restarter for inet services
SYNOPSIS
inetd [configuration-file] start | stop | refresh
svc:/network/inetd:default
DESCRIPTION
inetd is the delegated restarter for internet services for the Service Management Facility (SMF). Its basic responsibilities are to manage service states in response to administrative requests, system failures, and service failures; and, when appropriate, to listen for
network requests for services.
Services are no longer managed by editing the inetd configuration file, inetd.conf(4). Instead, you use inetconv(1M) to convert the configuration file content into SMF format services, then manage these services using inetadm(1M) and svcadm(1M). Once a service has been converted by inetconv, any changes to the legacy data in the inetd config
file will not become effective. However, inetd does alert the administrator when it notices change in the configuration file. See the start description under the "inetd Methods" section for further information.
Also note that the current inetd cannot be run from outside the SMF. This means it cannot be run from the command line, as was supported by the previous inetd. If you attempt to do this, a message is sent to stderr displaying mappings between the options
supported by the previous inetd to the SMF version of inetd.
inetd listens for connections on behalf of all services that are in either the online or degraded state. A service enters one of these states when the service is enabled by the user and inetd manages to listen on
its behalf. A listen attempt can fail if another server (whether standalone or a third-party internet service) is already listening on the same port. When this occurs, inetd logs this condition and continues trying to bind to the port at configured intervals a configured number
of times. See the property bind_fail_max under "Service Properties," below, for more details.
The configuration of all inetd's managed SMF services is read when it is started. It is reread when inetd is refreshed, which occurs in response to an SMF request, or when it receives a SIGHUP signal. See the refresh description under "inetd
Methods" for the behavior on configuration refresh.
You can use the inetadm(1M) or svccfg(1M) utilities
to make configuration changes to Internet services within the SMF repository. inetadm has the advantage over svccfg in that it provides an Internet/RPC service context.
Service States
As part of its service management duties, inetd implements a state machine for each of its managed services. The states in this machine are made up of the smf(5) set of states. The semantics of these states are as follows:
uninitialized
inetd has yet to process this service.
online
The service is handling new network requests and might have existing connections active.
degraded
The service has entered this state because it was able to listen and process requests for some, but not all, of the protocols specified for the service, having exhausted its listen retries. Existing network connections might be active.
offline
Connections might be active, but no new requests are being handled. This is a transient state. A service might be offline for any of the following reasons:
o
The service's dependencies are unmet. When its dependencies become met the service's state will be re-evaluated.
o
The service has exceeded its configured connection rate limit, max_con_rate. The service's state is re-evaluated when its connection offline timer, con_rate_offline, expires.
o
The service has reached its allowed number of active connections, max_copies. The service's state is re-evaluated when the number of active connections drops below max_copies.
o
inetd failed to listen on behalf of the service on all its protocols. As mentioned above, inetd retries up to a configured maximum number of times, at configured intervals.The service's state is re-evaluated when either a listen attempt is
successful or the retry limit is reached.
disabled
The service has been turned off by an administrator, is not accepting new connections, and has none active. Administrator intervention is required to exit this state.
maintenance
A service is in this state because it is either malfunctioning and needs adminstrator attention or because an administrator has requested it.
Events constituting malfunctioning include: inetd's inability to listen on behalf on any of the service's protocols before exceeding the service's bind retry limit, non-start methods returning with non-success return values, and the service exceeding its failure rate.
You request the maintenance state to perform maintenance on the service, such as applying a patch. No new requests are handled in this state, but existing connections might be active. Administrator intervention is required to exit this state.
Use inetadm(1M) to obtain the current state of a managed service.
Service Methods
As part of certain state transitions inetd will execute, if supplied, one of a set of methods provided by the service. The set of supported methods are:
inetd_start
Executed to handle a request for an online or degraded service. Since there is no separate state to distinguish a service with active connections, this method is not executed as part of a state transition.
inetd_offline
Executed when a service is taken from the online or degraded state to the offline state. For a wait-type service that at the time of execution is performing its own listening, this method should result
in it ceasing listening. This method will be executed before the disable method in the case an online/degraded service is disabled. This method is required to be implemented for a wait-type service.
inetd_online
Executed when a service transitions from the offline state to the online state. This method allows a service author to carry out some preparation prior to a service starting to handle requests.
inetd_disable
Executed when a service transitions from the offline state to the disabled state. It should result in any active connections for a service being terminated.
inetd_refresh
Executed when both of the following conditions are met:
o
inetd is refreshed, by means of the framework or a SIGHUP, or a request comes in to refresh the service, and
o
the service is currently in the online state and there are no configuration changes that would result in the service needing to be taken offline and brought back again.
The only compulsory method is the inetd_start method. In the absence of any of the others, inetd runs no method but behaves as if one was run successfully.
Service Properties
Configuration for SMF-managed services is stored in the SMF repository. The configuration is made up of the basic configuration of a service, the configuration for each of the service's methods, and the default configuration applicable to all inetd-managed services.
For details on viewing and modifying the configuration of a service and the defaults, refer to inetadm(1M).
The basic configuration of a service is stored in a property group named inetd in the service. The properties comprising the basic configuration are as follows:
bind_addr
The address of the network interface to which the service should be bound. An empty string value causes the service to accept connections on any network interface.
bind_fail_interval
The time interval in seconds between a failed bind attempt and a retry. The values 0 and -1 specify that no retries are attempted and the first failure is handled the same as exceeding bind_fail_max.
bind_fail_max
The maximum number of times inetd retries binding to a service's associated port before giving up. The value -1 specifies that no retry limit is imposed. If none of the service's protocols were bound to before any imposed limit is reached,
the service goes to the maintenance state; otherwise, if not all of the protocols were bound to, the service goes to the degraded state.
con_rate_offline
The time in seconds a service will remain offline if it exceeds its configured maximum connection rate, max_con_rate. The values 0 and -1 specify that connection rate limiting is disabled.
connection_backlog
The backlog queue size. Represents a limit on the number of incoming client requests that can be queued at the listening endpoints for servers.
endpoint_type
The type of the socket used by the service or the value tli to signify a TLI-based service. Valid socket type values are: stream, dgram, raw, seqpacket.
failrate_cnt
The count portion of the service's failure rate limit. The failure rate limit applies to wait-type services and is reached when count instances of the service are started within a given time. Exceeding the rate results in the service
being transitioned to the maintenance state. This is different from the behavior of the previous inetd, which continued to retry every 10 minutes, indefinitely. The failrate_cnt check accounts for badly behaving servers that fail before
consuming the service request and which would otherwise be continually restarted, taxing system resources. Failure rate is equivalent to the -r option of the previous inetd. The values 0 and -1 specify that this feature is
disabled.
failrate_interval
The time portion in seconds of the service's failure rate. The values 0 and -1 specify that the failure rate limit feature is disabled.
inherit_env
If true, pass inetd's environment on to the service's start method. Regardless of this setting, inetd will set the variables SMF_FMRI, SMF_METHOD, and SMF_RESTARTER in the start method's
environment, as well as any environment variables set in the method context. These variables are described in smf_method(5).
isrpc
If true, this is an RPC service.
max_con_rate
The maximum allowed connection rate, in connections per second, for a nowait-type service. The values 0 and -1 specify that that connection rate limiting is disabled.
max_copies
The maximum number of copies of a nowait service that can run concurrently. The values 0 and -1 specify that copies limiting is disabled.
name
Can be set to one of the following values:
o
a service name understood by getservbyname(3SOCKET);
o
if isrpc is set to true, a service name understood by getrpcbyname(3NSL);
o
if isrpc is set to true, a valid RPC program number.
proto
In the case of socket-based services, this is a list of protocols supported by the service. Valid protocols are: tcp, tcp6, tcp6only, udp, udp6, and udp6only. In the
case of TLI services, this is a list of netids recognized by getnetconfigent(3NSL) supported by the service, plus the values tcp6only and udp6only. RPC/TLI services also support nettypes in this list, and inetd first tries to interpret the list member as a nettype for these service types. The values tcp6only and udp6only are new to inetd; these
values request that inetd listen only for and pass on true IPv6 requests (not IPv4 mapped ones). See "Configuring Protocols for Sockets-Based Services," below.
rpc_low_version
Lowest supported RPC version. Required when isrpc is set to true.
rpc_high_version
Highest supported RPC version. Required when isrpc is set to true.
tcp_trace
If true, and this is a nowait-type service, inetd logs the client's IP address and TCP port number, along with the name of the service, for each incoming connection, using the syslog(3C) facility. inetd uses the syslog facility code daemon and notice priority level. See syslog.conf(4) for a description of syslog codes and severity levels. This logging is separate from the logging done by the TCP wrappers facility.
tcp_trace is equivalent to the previous inetd's -t option (and the /etc/default/inetd property ENABLE_CONNECTION_LOGGING).
tcp_wrappers
If true, enable TCP wrappers access control. This applies only to services with endpoint_type set to streams and wait set to false. The syslog facility code daemon
is used to log allowed connections (using the notice severity level) and denied traffic (using the warning severity level). See syslog.conf(4) for a description of syslog codes and severity levels. The stability level of the TCP wrappers facility and its configuration files is External. As the TCP wrappers facility is not controlled by Sun, intra-release incompatibilities are not uncommon. See attributes(5).
For more information about configuring TCP wrappers, you can refer to the tcpd(1M) and hosts_access(4) man pages, which are delivered as part of the Solaris operating system at /usr/sfw/man. These pages are not part of the standard
Solaris man pages, available at /usr/man.
tcp_wrappers is equivalent to the previous inetd's /etc/default/inetd property ENABLE_TCPWRAPPERS.
wait
If true this is a wait-type service, otherwise it is a nowait-type service. A wait-type service has the following characteristics:
o
Its inetd_start method will take over listening duties on the service's bound endpoint when it is executed.
o
inetd will wait for it to exit after it is executed before it resumes listening duties.
Datagram servers must be configured as being of type wait, as they are always invoked with the original datagram endpoint that will participate in delivering the service bound to the specified service. They do not have separate "listening" and "accepting"
sockets. Connection-oriented services, such as TCP stream services can be designed to be either of type wait or nowait.
A number of the basic properties are optional for a service. In their absence, their values are taken from the set of default values present in the defaults property group in the inetd service. These properties, with their seed values, are listed below.
Note that these values are configurable through inetadm(1M).
Each method specified for a service will have its configuration stored in the SMF repository, within a property group of the same name as the method. The set of properties allowable for these methods includes those specified for the services managed by svc.startd(1M). (See svc.startd(1M) for further details.) Additionally, for the inetd_start method,
you can set the arg0 property.
The arg0 property allows external wrapper programs to be used with inetd services. Specifically, it allows the first argument, argv[0], of the service's start method to be something other than the path of the server program.
In the case where you want to use an external wrapper program and pass arguments to the service's daemon, the arguments should be incorporated as arguments to the wrapper program in the exec property. For example:
In addition to the special method tokens mentioned in smf_method(5), inetd also supports the :kill_process token
for wait-type services. This results in behavior identical to that if the :kill token were supplied, except that the kill signal is sent only to the parent process of the wait-type service's start method,
not to all members of its encompassing process contract (see process(4)).
Configuring Protocols for Sockets-Based Services
When configuring inetd for a sockets-based service, you have the choice, depending on what is supported by the service, of the alternatives described under the proto property, above. The following are guidelines for which proto values
to use:
o
For a service that supports only IPv4: tcp and udp
o
For a service that supports only IPv6: tcp6only and udp6only
o
For a service that supports both IPv4 and IPv6:
o
Obsolete and not recommended: tcp6 and udp6
o
Recommended: use two separate entries that differ only in the proto field. One entry has tcp and the other has tcp6only, or udp plus udp6only.
See EXAMPLES for an example of a configuration of a service that supports both IPv4 and IPv6.
inetd Methods
inetd provides the methods listed below for consumption by the master restarter, svc.startd(1M).
start
Causes inetd to start providing service. This results in inetd beginning to handle smf requests for its managed services and network requests for those services that are in either the online or degraded state.
In addition, inetd also checks if the inetd.conf(4)-format configuration file it is monitoring has changed since the last inetconv(1M) conversion was carried out. If it has, then a message telling the administrator to re-run inetconv to effect the changes made is logged in syslog.
stop
Causes inetd to stop providing service. At this point, inetd transitions each of its services that are not in either the maintenance or disabled states to the offline state, running any
appropriate methods in the process.
refresh
Results in a refresh being performed for each of its managed services and the inetd.conf(4) format configuration file being checked for change,
as in the start method. When a service is refreshed, its behavior depends on its current state:
o
if it is in the maintenance or disabled states, no action is performed because the configuration will be read and consumed when the service leaves the state;
o
if it is in the offline state, the configuration will be read and any changes consumed immediately;
o
if it is in the online or degraded state and the configuration has changed such that a re-binding is necessary to conform to it, then the service will be transitioned to the offline state and back again, using the new configuration
for the bind;
o
if it is in the online state and a re-binding is not necessary, then the inetd_refresh method of the service, if provided, will be run to allow onlinewait-type services to consume any other changes.
OPTIONS
No options are supported.
OPERANDS
configuration-file
Specifies an alternate location for the legacy service file (inetd.conf(4)).
start|stop|refresh
Specifies which of inetd's methods should be run.
EXAMPLES
Example 1 Configuring a Service that Supports Both IPv4 and IPv6
The following commands illustrate the existence of services that support both IPv4 and IPv6 and assign proto properties to those services.
The inetd daemon performs the same function as, but is implemented significantly differently from, the daemon of the same name in Solaris 9 and prior Solaris operating system releases. In the current Solaris release, inetd is part of the Solaris Management
Facility (see smf(5)) and will run only within that facility.
The /etc/default/inetd file has been deprecated. The functionality represented by the properties ENABLE_CONNECTION_LOGGING and ENABLE_TCP_WRAPPERS are now available as the tcp_trace and tcp_wrappers properties,
respectively. These properties are described above, under "Service Properties".
