cyrus.conf

Cyrus configuration file

Description

cyrus.conf is the configuration file for the Cyrus master(8) process. It defines the startup procedures, services, events and daemons to be spawned, managed and tended to by master.

The /etc/cyrus.conf file consists of a series of entries divided into sections of the form

section {
    name arguments
        ...
        ...
        ...
}

where section is the name of the section, name is the name of the entry and arguments is the whitespace-separated list of arguments for the entry. The name may be any sequence of alphabetic and numeric characters, but may not contain punctuation such as '-' or '_'. In the SERVICES section, names must be unique.

Blank lines and lines beginning with ``#'' are ignored.

Section Descriptions

The paragraphs below detail the four sections (START, SERVICES, EVENTS, DAEMON) that can be placed in the /etc/cyrus.conf file. The arguments that are available for each entry within the section are described, and each argument's default value is shown.

An important distinction exists between SERVICES and DAEMON ; the former have sockets which master(8) will listen on (either IP or Unix domain) while the latter do not. Similarly, processes listed in START will be run to completion before any SERVICES are started, while those in DAEMON will be managed by master(8).

Note

If master(8) is started in debugging mode (-D) the behavior of DAEMON will be altered, as master(8) will no longer be backgrounded. Thus, processes started under DAEMON may not be terminated by master(8).

Arguments can appear in any order. Some arguments have no default value, these are listed with ``<no default>''. For string arguments, the value MUST be enclosed in double quotes.

START

This section lists the processes to run before any SERVICES are spawned. This section is typically used to initialize databases. Master itself will not startup until all tasks in START have completed, so put no blocking commands here.

cmd=<no default>

The command (with options) to spawn as a child process. This string argument is required.

Note

Prior to v3, non-service daemons like idled were started from START but would background themselves, thus not blocking. Post v3 these are better managed through the DAEMON section, under which master will provide life-cycle management (i.e. restarting dead processes).

SERVICES

This section is the heart of the /etc/cyrus.conf file. It lists the processes that should be spawned to handle client connections made on certain Internet/UNIX sockets.

babysit=0

Integer value - if non-zero, will make sure at least one process is pre-forked, and will set the maxforkrate to 10 if it's zero.

cmd=<no default>

The command (with options) to spawn as a child process. This string argument is required.

listen=<no default>

The UNIX or internet socket to listen on. This string field is required and takes one of the following forms:

path
[ host : ] port

where path is the explicit path to a UNIX socket, host is either the hostname or bracket-enclosed IP address of a network interface, and port is either a port number or service name (as listed in /etc/services).

If host is missing, 0.0.0.0 (all interfaces) is assumed. Use localhost or 127.0.0.1 to restrict access, i.e. when a proxy on the same host is front-ending Cyrus.

Note that on most systems UNIX socket paths are limited to around 100 characters. See your system documentation for specifics.

proto=tcp

The protocol used for this service (tcp, tcp4, tcp6, udp, udp4, udp6). This string argument is optional.

tcp4, udp4: These arguments are used to bind the service to IPv4 only.

tcp6, udp6: These arguments are used to bind the service to IPv6 only, if the operating system supports this.

tcp, udp: These arguments are used to bind to both IPv4 and IPv6 if possible.

prefork=0

The number of instances of this service to always have running and waiting for a connection (for faster initial response time). This integer value is optional. Note that if you are listening on multiple network types (i.e. ipv4 and ipv6) then one process will be forked for each address, causing twice as many processes as you might expect.

maxchild=-1

The maximum number of instances of this service to spawn. A value of -1 means unlimited. This integer value is optional.

maxfds=256

The maximum number of file descriptors to which to limit this process. This integer value is optional.

maxforkrate=0

Maximum number of processes to fork per second - the master will insert sleeps to ensure it doesn't fork faster than this on average.

EVENTS

This section lists processes that should be run at specific intervals, similar to cron jobs. This section is typically used to perform scheduled cleanup/maintenance.

cmd=<no default>

The command (with options) to spawn as a child process. This string argument is required.

period=0

The interval (in minutes) at which to run the command. This integer value is optional, but SHOULD be a positive integer > 10.

at=<hhmm>

The time (24-hour format) at which to run the command each day. If set to a valid time (0000-2359), period is automatically set to 1440. This string argument is optional.

DAEMON

This section lists long running daemons to start before any SERVICES are spawned. master(8) will ensure that these processes are running, restarting any process which dies or forks. All listed processes will be shutdown when master(8) is exiting.

cmd=<no default>

The command (with options) to spawn as a child process. This string argument is required.

wait=0

Switch: whether or not master(8) should wait for this daemon to successfully start before continuing to load.

If wait=n (the default), the daemon will be started asynchronously along with the service processes. The daemon process will not have file descriptor 3 open, and does not need to indicate its readiness.

If wait=y, the daemon MUST write "ok\r\n" to file descriptor 3 to indicate its readiness; if it does not do this, and master has been told to wait, master will continue to wait.... If it writes anything else to this descriptor, or closes it before writing "ok\r\n", master will exit with an error.

Daemons with wait=y will be started sequentially in the order they are listed in cyrus.conf, waiting for each to report readiness before the next is started.

Service processes, and wait=n daemons, are not started until after the wait=y daemons are all started and ready.

At shutdown, wait=y daemons will be terminated sequentially in the reverse order they were started, commencing after all other services and wait=n daemons have finished.

If a daemon that was started with wait=y exits unexpectedly, such that master restarts it, master will restart it asynchronously, without waiting for it to report its readiness. In this case, file descriptor 3 will not be open and the daemon should not try to write to it.

If master is told to reread its config with a SIGHUP, this signal will be passed on to wait=y daemons like any other service. If the daemon exits in response to the signal, master will restart it asynchronously, without waiting for it to report its readiness. In this case too, file descriptor 3 will not be open and the daemon should not try to write to it.

Examples

# example cyrus.conf

START {
    recover       cmd="ctl_cyrusdb -r"
}

SERVICES {
    imap          cmd="imapd" listen="imap" prefork=1
    imaps         cmd="imapd -s" listen="imaps" prefork=0
    lmtpunix      cmd="lmtpd" listen="/var/imap/socket/lmtp"
    lmtp          cmd="lmtpd" listen="localhost:lmtp"
}

EVENTS {
    checkpoint    cmd="ctl_cyrusdb -c" period=30
    delprune      cmd="cyr_expire -E 3" at=0400
    tlsprune      cmd="tls_prune" at=0400
}

DAEMON {
    idled         cmd="idled"
}

Access Control

When TCP Wrappers is used to control access to Cyrus services, the name of the service entry should be used as the process name in the hosts_access(5) table. For instance, in the example above, "imap", "imaps", "lmtpunix" and "lmtp" would be used as the process names. This allows a single daemon such as imapd to be run in different modes or configurations (i.e., SSL and non-SSL enabled) yet still have separate access control rules.