module Puppet
Type.newtype(:exec) do
include Puppet::Util::Execution
require 'timeout'
@doc = "Executes external commands.
Any command in an `exec` resource **must** be able to run multiple times
without causing harm --- that is, it must be *idempotent*. There are three
main ways for an exec to be idempotent:
* The command itself is already idempotent. (For example, `apt-get update`.)
* The exec has an `onlyif`, `unless`, or `creates` attribute, which prevents
Puppet from running the command unless some condition is met.
* The exec has `refreshonly => true`, which only allows Puppet to run the
command when some other resource is changed. (See the notes on refreshing
below.)
A caution: There's a widespread tendency to use collections of execs to
manage resources that aren't covered by an existing resource type. This
works fine for simple tasks, but once your exec pile gets complex enough
that you really have to think to understand what's happening, you should
consider developing a custom resource type instead, as it will be much
more predictable and maintainable.
**Refresh:** `exec` resources can respond to refresh events (via
`notify`, `subscribe`, or the `~>` arrow). The refresh behavior of execs
is non-standard, and can be affected by the `refresh` and
`refreshonly` attributes:
* If `refreshonly` is set to true, the exec will _only_ run when it receives an
event. This is the most reliable way to use refresh with execs.
* If the exec already would have run and receives an event, it will run its
command **up to two times.** (If an `onlyif`, `unless`, or `creates` condition
is no longer met after the first run, the second run will not occur.)
* If the exec already would have run, has a `refresh` command, and receives an
event, it will run its normal command, then run its `refresh` command
(as long as any `onlyif`, `unless`, or `creates` conditions are still met
after the normal command finishes).
* If the exec would **not** have run (due to an `onlyif`, `unless`, or `creates`
attribute) and receives an event, it still will not run.
* If the exec has `noop => true`, would otherwise have run, and receives
an event from a non-noop resource, it will run once (or run its `refresh`
command instead, if it has one).
In short: If there's a possibility of your exec receiving refresh events,
it becomes doubly important to make sure the run conditions are restricted.
**Autorequires:** If Puppet is managing an exec's cwd or the executable
file used in an exec's command, the exec resource will autorequire those
files. If Puppet is managing the user that an exec should run as, the
exec resource will autorequire that user."
# Create a new check mechanism. It's basically just a parameter that
# provides one extra 'check' method.
def self.newcheck(name, options = {}, &block)
@checks ||= {}
check = newparam(name, options, &block)
@checks[name] = check
end
def self.checks
@checks.keys
end
newproperty(:returns, :array_matching => :all, :event => :executed_command) do |property|
include Puppet::Util::Execution
munge do |value|
value.to_s
end
def event_name
:executed_command
end
defaultto "0"
attr_reader :output
desc "The expected exit code(s). An error will be returned if the
executed command has some other exit code. Defaults to 0. Can be
specified as an array of acceptable exit codes or a single value.
On POSIX systems, exit codes are always integers between 0 and 255.
On Windows, **most** exit codes should be integers between 0
and 2147483647.
Larger exit codes on Windows can behave inconsistently across different
tools. The Win32 APIs define exit codes as 32-bit unsigned integers, but
both the cmd.exe shell and the .NET runtime cast them to signed
integers. This means some tools will report negative numbers for exit
codes above 2147483647. (For example, cmd.exe reports 4294967295 as -1.)
Since Puppet uses the plain Win32 APIs, it will report the very large
number instead of the negative number, which might not be what you
expect if you got the exit code from a cmd.exe session.
Microsoft recommends against using negative/very large exit codes, and
you should avoid them when possible. To convert a negative exit code to
the positive one Puppet will use, add it to 4294967296."
# Make output a bit prettier
def change_to_s(currentvalue, newvalue)
_("executed successfully")
end
# First verify that all of our checks pass.
def retrieve
# We need to return :notrun to trigger evaluation; when that isn't
# true, we *LIE* about what happened and return a "success" for the
# value, which causes us to be treated as in_sync?, which means we
# don't actually execute anything. I think. --daniel 2011-03-10
if @resource.check_all_attributes
return :notrun
else
return self.should
end
end
# Actually execute the command.
def sync
event = :executed_command
tries = self.resource[:tries]
try_sleep = self.resource[:try_sleep]
begin
tries.times do |try|
# Only add debug messages for tries > 1 to reduce log spam.
debug("Exec try #{try+1}/#{tries}") if tries > 1
@output, @status = provider.run(self.resource[:command])
break if self.should.include?(@status.exitstatus.to_s)
if try_sleep > 0 and tries > 1
debug("Sleeping for #{try_sleep} seconds between tries")
sleep try_sleep
end
end
rescue Timeout::Error
self.fail Puppet::Error, _("Command exceeded timeout"), $!
end
if log = @resource[:logoutput]
case log
when :true
log = @resource[:loglevel]
when :on_failure
unless self.should.include?(@status.exitstatus.to_s)
log = @resource[:loglevel]
else
log = :false
end
end
unless log == :false
@output.split(/\n/).each { |line|
self.send(log, line)
}
end
end
unless self.should.include?(@status.exitstatus.to_s)
if @resource.parameter(:command).sensitive
# Don't print sensitive commands in the clear
self.fail(_("[command redacted] returned %{status} instead of one of [%{expected}]") % { status: @status.exitstatus, expected: self.should.join(",") })
else
self.fail(_("'%{cmd}' returned %{status} instead of one of [%{expected}]") % { cmd: self.resource[:command], status: @status.exitstatus, expected: self.should.join(",") })
end
end
event
end
end
newparam(:command) do
isnamevar
desc "The actual command to execute. Must either be fully qualified
or a search path for the command must be provided. If the command
succeeds, any output produced will be logged at the instance's
normal log level (usually `notice`), but if the command fails
(meaning its return code does not match the specified code) then
any output is logged at the `err` log level."
validate do |command|
raise ArgumentError, _("Command must be a String, got value of class %{klass}") % { klass: command.class } unless command.is_a? String
end
end
newparam(:path) do
desc "The search path used for command execution.
Commands must be fully qualified if no path is specified. Paths
can be specified as an array or as a '#{File::PATH_SEPARATOR}' separated list."
# Support both arrays and colon-separated fields.
def value=(*values)
@value = values.flatten.collect { |val|
val.split(File::PATH_SEPARATOR)
}.flatten
end
end
newparam(:user) do
desc "The user to run the command as. Note that if you
use this then any error output is not currently captured. This
is because of a bug within Ruby. If you are using Puppet to
create this user, the exec will automatically require the user,
as long as it is specified by name.
Please note that the $HOME environment variable is not automatically set
when using this attribute."
validate do |user|
if Puppet.features.microsoft_windows?
self.fail _("Unable to execute commands as other users on Windows")
elsif !Puppet.features.root? && resource.current_username() != user
self.fail _("Only root can execute commands as other users")
end
end
end
newparam(:group) do
desc "The group to run the command as. This seems to work quite
haphazardly on different platforms -- it is a platform issue
not a Ruby or Puppet one, since the same variety exists when
running commands as different users in the shell."
# Validation is handled by the SUIDManager class.
end
newparam(:cwd, :parent => Puppet::Parameter::Path) do
desc "The directory from which to run the command. If
this directory does not exist, the command will fail."
end
newparam(:logoutput) do
desc "Whether to log command output in addition to logging the
exit code. Defaults to `on_failure`, which only logs the output
when the command has an exit code that does not match any value
specified by the `returns` attribute. As with any resource type,
the log level can be controlled with the `loglevel` metaparameter."
defaultto :on_failure
newvalues(:true, :false, :on_failure)
end
newparam(:refresh) do
desc "An alternate command to run when the `exec` receives a refresh event
from another resource. By default, Puppet runs the main command again.
For more details, see the notes about refresh behavior above, in the
description for this resource type.
Note that this alternate command runs with the same `provider`, `path`,
`user`, and `group` as the main command. If the `path` isn't set, you
must fully qualify the command's name."
validate do |command|
provider.validatecmd(command)
end
end
newparam(:environment) do
desc "An array of any additional environment variables you want to set for a
command. e.g.: `[ 'HOME=/root', 'MAIL=root@example.com']`
Note that if you use this to set PATH, it will override the `path`
attribute. Multiple environment variables should be specified as an
array."
validate do |values|
values = [values] unless values.is_a? Array
values.each do |value|
unless value =~ /\w+=/
raise ArgumentError, _("Invalid environment setting '%{value}'") % { value: value }
end
end
end
end
newparam(:umask, :required_feature => :umask) do
desc "Sets the umask to be used while executing this command"
munge do |value|
if value =~ /^0?[0-7]{1,4}$/
return value.to_i(8)
else
raise Puppet::Error, _("The umask specification is invalid: %{value}") % { value: value.inspect }
end
end
end
newparam(:timeout) do
desc "The maximum time the command should take. If the command takes
longer than the timeout, the command is considered to have failed
and will be stopped. The timeout is specified in seconds. The default
timeout is 300 seconds and you can set it to 0 to disable the timeout."
munge do |value|
value = value.shift if value.is_a?(Array)
begin
value = Float(value)
rescue ArgumentError
raise ArgumentError, _("The timeout must be a number."), $!.backtrace
end
[value, 0.0].max
end
defaultto 300
end
newparam(:tries) do
desc "The number of times execution of the command should be tried.
Defaults to '1'. This many attempts will be made to execute
the command until an acceptable return code is returned.
Note that the timeout parameter applies to each try rather than
to the complete set of tries."
munge do |value|
if value.is_a?(String)
unless value =~ /^[\d]+$/
raise ArgumentError, _("Tries must be an integer")
end
value = Integer(value)
end
raise ArgumentError, _("Tries must be an integer >= 1") if value < 1
value
end
defaultto 1
end
newparam(:try_sleep) do
desc "The time to sleep in seconds between 'tries'."
munge do |value|
if value.is_a?(String)
unless value =~ /^[-\d.]+$/
raise ArgumentError, _("try_sleep must be a number")
end
value = Float(value)
end
raise ArgumentError, _("try_sleep cannot be a negative number") if value < 0
value
end
defaultto 0
end
newcheck(:refreshonly) do
desc <<-'EOT'
The command should only be run as a
refresh mechanism for when a dependent object is changed. It only
makes sense to use this option when this command depends on some
other object; it is useful for triggering an action:
# Pull down the main aliases file
file { '/etc/aliases':
source => 'puppet://server/module/aliases',
}
# Rebuild the database, but only when the file changes
exec { newaliases:
path => ['/usr/bin', '/usr/sbin'],
subscribe => File['/etc/aliases'],
refreshonly => true,
}
Note that only `subscribe` and `notify` can trigger actions, not `require`,
so it only makes sense to use `refreshonly` with `subscribe` or `notify`.
EOT
newvalues(:true, :false)
# We always fail this test, because we're only supposed to run
# on refresh.
def check(value)
# We have to invert the values.
if value == :true
false
else
true
end
end
end
newcheck(:creates, :parent => Puppet::Parameter::Path) do
desc <<-'EOT'
A file to look for before running the command. The command will
only run if the file **doesn't exist.**
This parameter doesn't cause Puppet to create a file; it is only
useful if **the command itself** creates a file.
exec { 'tar -xf /Volumes/nfs02/important.tar':
cwd => '/var/tmp',
creates => '/var/tmp/myfile',
path => ['/usr/bin', '/usr/sbin',],
}
In this example, `myfile` is assumed to be a file inside
`important.tar`. If it is ever deleted, the exec will bring it
back by re-extracting the tarball. If `important.tar` does **not**
actually contain `myfile`, the exec will keep running every time
Puppet runs.
EOT
accept_arrays
# If the file exists, return false (i.e., don't run the command),
# else return true
def check(value)
! Puppet::FileSystem.exist?(value)
end
end
newcheck(:unless) do
desc <<-'EOT'
A test command that checks the state of the target system and restricts
when the `exec` can run. If present, Puppet runs this test command
first, then runs the main command unless the test has an exit code of 0
(success). For example:
exec { '/bin/echo root >> /usr/lib/cron/cron.allow':
path => '/usr/bin:/usr/sbin:/bin',
unless => 'grep root /usr/lib/cron/cron.allow 2>/dev/null',
}
This would add `root` to the cron.allow file (on Solaris) unless
`grep` determines it's already there.
Note that this test command runs with the same `provider`, `path`,
`user`, and `group` as the main command. If the `path` isn't set, you
must fully qualify the command's name.
This parameter can also take an array of commands. For example:
unless => ['test -f /tmp/file1', 'test -f /tmp/file2'],
This `exec` would only run if every command in the array has a
non-zero exit code.
EOT
validate do |cmds|
cmds = [cmds] unless cmds.is_a? Array
cmds.each do |command|
provider.validatecmd(command)
end
end
# Return true if the command does not return 0.
def check(value)
begin
output, status = provider.run(value, true)
rescue Timeout::Error
err _("Check %{value} exceeded timeout") % { value: value.inspect }
return false
end
output.split(/\n/).each { |line|
self.debug(line)
}
status.exitstatus != 0
end
end
newcheck(:onlyif) do
desc <<-'EOT'
A test command that checks the state of the target system and restricts
when the `exec` can run. If present, Puppet runs this test command
first, and only runs the main command if the test has an exit code of 0
(success). For example:
exec { 'logrotate':
path => '/usr/bin:/usr/sbin:/bin',
provider => shell,
onlyif => 'test `du /var/log/messages | cut -f1` -gt 100000',
}
This would run `logrotate` only if that test returns true.
Note that this test command runs with the same `provider`, `path`,
`user`, and `group` as the main command. If the `path` isn't set, you
must fully qualify the command's name.
This parameter can also take an array of commands. For example:
onlyif => ['test -f /tmp/file1', 'test -f /tmp/file2'],
This `exec` would only run if every command in the array has an
exit code of 0 (success).
EOT
validate do |cmds|
cmds = [cmds] unless cmds.is_a? Array
cmds.each do |command|
provider.validatecmd(command)
end
end
# Return true if the command returns 0.
def check(value)
begin
output, status = provider.run(value, true)
rescue Timeout::Error
err _("Check %{value} exceeded timeout") % { value: value.inspect }
return false
end
output.split(/\n/).each { |line|
self.debug(line)
}
status.exitstatus == 0
end
end
# Exec names are not isomorphic with the objects.
@isomorphic = false
validate do
provider.validatecmd(self[:command])
end
# FIXME exec should autorequire any exec that 'creates' our cwd
autorequire(:file) do
reqs = []
# Stick the cwd in there if we have it
reqs << self[:cwd] if self[:cwd]
file_regex = Puppet.features.microsoft_windows? ? %r{^([a-zA-Z]:[\\/]\S+)} : %r{^(/\S+)}
self[:command].scan(file_regex) { |str|
reqs << str
}
self[:command].scan(/^"([^"]+)"/) { |str|
reqs << str
}
[:onlyif, :unless].each { |param|
next unless tmp = self[param]
tmp = [tmp] unless tmp.is_a? Array
tmp.each do |line|
# And search the command line for files, adding any we
# find. This will also catch the command itself if it's
# fully qualified. It might not be a bad idea to add
# unqualified files, but, well, that's a bit more annoying
# to do.
reqs += line.scan(file_regex)
end
}
# For some reason, the += isn't causing a flattening
reqs.flatten!
reqs
end
autorequire(:user) do
# Autorequire users if they are specified by name
if user = self[:user] and user !~ /^\d+$/
user
end
end
def self.instances
[]
end
# Verify that we pass all of the checks. The argument determines whether
# we skip the :refreshonly check, which is necessary because we now check
# within refresh
def check_all_attributes(refreshing = false)
self.class.checks.each { |check|
next if refreshing and check == :refreshonly
if @parameters.include?(check)
val = @parameters[check].value
val = [val] unless val.is_a? Array
val.each do |value|
return false unless @parameters[check].check(value)
end
end
}
true
end
def output
if self.property(:returns).nil?
return nil
else
return self.property(:returns).output
end
end
# Run the command, or optionally run a separately-specified command.
def refresh
if self.check_all_attributes(true)
if cmd = self[:refresh]
provider.run(cmd)
else
self.property(:returns).sync
end
end
end
def current_username
Etc.getpwuid(Process.uid).name
end
private
def set_sensitive_parameters(sensitive_parameters)
# Respect sensitive commands
if sensitive_parameters.include?(:command)
sensitive_parameters.delete(:command)
parameter(:command).sensitive = true
end
super(sensitive_parameters)
end
end
end