diff --git a/Makefile b/Makefile
index f928d0af260633aa3af3224d9d493d8922426f18..de0a2b6dc7d893dbb2a02d3df5aa6740573f61f4 100644
--- a/Makefile
+++ b/Makefile
@@ -1,7 +1,7 @@
.DEFAULT_GOAL = all
#########################################################################
-# Copyright © 2018 Thomas Bellman, Linköping, Sweden
+# Copyright © 2018-2019 Thomas Bellman, Linköping, Sweden
# Licensed under the GNU LGPL v3+; see the README file for more information.
@@ -31,7 +31,7 @@ include Version.inc
#########################################################################
# Internal variables, not intended to be overridden by users
-_PYPLUGINS = check_ospf_nbr
+_PYPLUGINS = check_ospf_nbr check_ping_multiaddr
_PYLIBS = trh_nagioslib
_MIBFILES = OSPFV3-MIB.txt
_RPM_SPECFILE = $(PKGNAME)-$(VERSION).spec
diff --git a/README b/README
index 1045067855e6e08c83c727dc0f0b9fe513d026c8..f3213811fbb4123300a4d8532851fa5b17254c7f 100644
--- a/README
+++ b/README
@@ -1,7 +1,7 @@
-*- mode: text; fill-column: 70; -*-
This is a collection of Nagios plugins.
-Copyright © 2018 Thomas Bellman, and any other authors.
+Copyright © 2018-2019 Thomas Bellman, and any other authors.
===== LICENSING =====
@@ -28,12 +28,13 @@ SNMP MIB files are likely to be licensed under different terms.
===== WHAT IS THIS =====
This is a collection of Nagios plugins, with a focus on network
-monitoring. Currently only a single one, though:
+monitoring. Currently only two, though:
- check_ospf_nbr Check that an OSPF neighbour of a router is
present and in state FULL. Both OSPF v2 (for
IPv4) and OSPF v3 (for IPv6) is supported.
-
+ - check_ping_multiaddr Check that multiple IP addresses (both IPv4
+ and IPv6, at the same time) respond to ping.
===== INSTALLATION =====
@@ -78,6 +79,7 @@ modifications of the spec file
You will need at least the Net-SNMP (http://www.net-snmp.org/)
Python bindings to run these plugins. You will also need several
SNMP MIB files installed, e.g. the ones that come with Net-SNMP.
+The check_ping_multiaddr plugin requires fping(1) to be installed.
GNU Make is assumed for building.
diff --git a/check_ping_multiaddr.py b/check_ping_multiaddr.py
new file mode 100755
index 0000000000000000000000000000000000000000..44564751f3d9579da015f75d9277a3da9a9404e1
--- /dev/null
+++ b/check_ping_multiaddr.py
@@ -0,0 +1,277 @@
+#!/usr/bin/env python2
+# -*- coding: utf-8; indent-tabs-mode: nil -*-
+
+# Copyright © 2019 Thomas Bellman, Linköping, Sweden
+# Licensed under the GNU LGPL v3+; see the README file for more information.
+
+
+# Explicitly assign to __doc__ to avoid doc string being optimized out.
+__doc__ = """\
+Check that one or more IP addresses responds to ping.
+
+The intent of this Nagios plugin is to check that a host is alive and
+reachable on all of its addresses, in particular both IPv4 and IPv6
+addresses, which can be tested at the same time.
+
+Addresses can be specified either as numeric IP addresses, or as host
+names that will be resolved at runtime. By default, only the first
+address the resolver library returns for each host name will be used,
+but the -A/--all flag will cause all of the addresses returned by the
+resolver to be used. You need to give at least one of --ipv4 or --ipv6,
+but both can be specified at the same time.
+
+Note that Nagios does not allow you to declare multiple addresses for
+a host. To use this in a check_command for a host, you need to either
+let %prog resolve addresses at runtime (thus depending on
+DNS), or you need to generate a unique command declaration for each
+host; something like:
+
+ define host {
+ host_name smurf
+ address 198.51.100.23
+ check_command check-host-alive::smurf
+ }
+ define command {
+ command_name check-host-alive::smurf
+ command_line $USER1$/check_ping_multiaddr -A46 -- 198.51.100.23 2001:db8:4711:17::1:23
+ }
+"""
+
+__version__ = '<#VERSION#>'
+
+
+import sys
+import re
+import os
+import optparse
+import ipaddr
+import socket
+import subprocess
+
+import trh_nagioslib
+
+
+
+class ProgramFailure(Exception):
+ def __init__(self, status, msg):
+ Exception.__init__(self)
+ self.status = status
+ self.msg = msg
+
+
+class Options(optparse.OptionParser):
+
+ def __init__(self):
+ global __doc__, __version__
+ optparse.OptionParser.__init__(
+ self,
+ usage="%prog {-4|-6} [options] -- address ...",
+ formatter=trh_nagioslib.NoReflowHelpFormatter(),
+ version=__version__,
+ description=__doc__)
+ self.add_option(
+ '-4', '--ipv4', action='store_true', default=False,
+ help=("Use IPv4 [default: %default]."
+ " At least one of --ipv4 and --ipv6 must be given."))
+ self.add_option(
+ '-6', '--ipv6', action='store_true', default=False,
+ help=("Use IPv6 [default: %default]."
+ " At least one of --ipv4 and --ipv6 must be given."))
+ self.add_option(
+ '-A', '--all-addresses', action='store_true', default=False,
+ help=("Check all addresses each hostname resolves to."
+ " By default, only the first address for each host is"
+ " checked."))
+ self.add_option(
+ '-r', '--retries', action='store', type='int', default=9,
+ help=("Number of ICMP ECHO retries to send [default: %default]."
+ " Must be in the range 1 <= RETRIES <= 20."))
+ self.add_option(
+ '-d', '--debug', action='count', default=0,
+ help=("Increase debug level [default: %default]."))
+
+ def get_version(self):
+ progname = self.get_prog_name()
+ pkgname = "<#PKGNAME#>"
+ version = self.version
+ vinfo = "%s (%s) version %s" % (progname, pkgname, version)
+ return vinfo
+
+ def check_values(self, values, args):
+ if len(args) < 1:
+ self.error("At least one IP address is required")
+ if not values.ipv4 and not values.ipv6:
+ self.error("At least one IP version must be specified (-4, -6)")
+ if values.retries < 1 or values.retries > 20:
+ self.error(
+ "Retries (-r) must be in range 1 <= RETRIES <= 20")
+ return values,args
+
+ def exit(self, status=0, msg=None):
+ if msg:
+ sys.stderr.write(msg)
+ # Exit with EX_USAGE, unless status==0 (which happens for --help)
+ raise ProgramFailure(status=(status and os.EX_USAGE), msg=msg)
+
+
+OPTIONS,_ = Options().parse_args(['-A', '-6', '-4', '--', 'localhost'])
+
+
+def fail(status, fmt, *args):
+ progname = os.path.basename(sys.argv[0] or "check_ospf_nbr")
+ msg = progname + ": " + fmt % args + "\n"
+ sys.stderr.write(msg)
+ raise ProgramFailure(status=status, msg=msg)
+
+
+def chatter(level, fmt, *args, **kwargs):
+ if level <= OPTIONS.debug:
+ msg = fmt % (kwargs or args)
+ sys.stderr.write("#" + " " * level + msg + "\n")
+
+
+def eai_errno_to_symbol(errno):
+ for symbol in dir(socket):
+ if symbol.startswith('EAI_') and getattr(socket, symbol) == errno:
+ return symbol
+ return
+
+
+
+def collect_addresses(hosts, all_addresses, do_v4, do_v6):
+ if do_v4 and do_v6:
+ ipfamily = socket.AF_UNSPEC ; ipversion = None
+ elif do_v4:
+ ipfamily = socket.AF_INET ; ipversion = 4
+ elif do_v6:
+ ipfamily = socket.AF_INET6 ; ipversion = 6
+ else:
+ raise ValueError("Neither IPv4 nor IPv6 selected")
+
+ lookupflags = 0
+ lookupflags |= getattr(socket, 'AI_IDN', 0)
+
+ addresses = { 4: set(), 6: set() }
+ for host in hosts:
+ # Try it as a numerical IP address first
+ try:
+ addr = ipaddr.IPAddress(host, ipversion)
+ except ValueError:
+ pass
+ else:
+ addresses[addr.version].add(addr)
+ continue
+
+ # And if that failed, try resolving the name
+ try:
+ ipres = socket.getaddrinfo(host, None, ipfamily, 0, 0, lookupflags)
+ except socket.gaierror as e:
+ fail(os.EX_NOHOST, "%s, %s", e.strerror, host)
+ for ai in ipres:
+ ipfam = ai[0] ; ip = ai[4][0]
+ addr = ipaddr.IPAddress(ip)
+ addresses[addr.version].add(addr)
+ if not all_addresses:
+ break
+
+ return addresses
+
+
+
+__fping_parser_re = re.compile(r"^([0-9a-f.:]+) is ([a-z]+)$")
+
+def parse_fping_output(output, expected_addrs):
+ alive = set()
+ unreachable = set()
+ for line in filter(bool, output):
+ match = __fping_parser_re.match(line)
+ if not match:
+ continue
+ ip = ipaddr.IPAddress(match.group(1))
+ status = match.group(2)
+ if status == 'alive':
+ alive.add(ip)
+ elif status == 'unreachable':
+ unreachable.add(ip)
+ else:
+ raise RuntimeError(
+ "Unexpected status line from fping, " + repr(line))
+ not_reported = expected_addrs - alive - unreachable
+ unexpected = (alive | unreachable) - expected_addrs
+
+ return (alive, unreachable, not_reported, unexpected)
+
+
+def ping_addresses(addresses):
+ fpingcmds = {
+ 4: ['fping'],
+ 6: ['fping6'],
+ }
+ fpingflags = [
+ # These settings, with default 9 retries, gives ca 5 seconds timeout
+ # for unreachable addresses
+ '-i10', # -i10 is the fastest fping allows without being root
+ '-t250',
+ '-B1.125',
+ '-r%d' % (OPTIONS.retries,),
+ ]
+ all_output = []
+ for ipver,addrs in addresses.items():
+ if not addrs:
+ continue
+ cmd = fpingcmds[ipver] + fpingflags + map(str, addrs)
+ chatter(1, "Running %r", cmd)
+ p = subprocess.Popen(
+ cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE)
+ output,errors = p.communicate()
+ output = output.split("\n"); errors = errors.split("\n")
+ chatter(3, "Received output %r", output)
+ chatter(3, "Received errors %r", errors)
+ all_output += output
+
+ alive, unreachable, not_reported, unexpected = parse_fping_output(
+ all_output, set().union(*addresses.values()))
+
+ return (alive, unreachable, not_reported, unexpected)
+
+
+
+def main(argv):
+ global OPTIONS
+ OPTIONS, arg_addresses = Options().parse_args(argv[1:])
+
+ addresses = collect_addresses(
+ arg_addresses, OPTIONS.all_addresses, OPTIONS.ipv4, OPTIONS.ipv6)
+
+ for ipver,addrs in addresses.items():
+ chatter(2, "IPv%d addresses: %s", ipver, " ".join(map(str, addrs)))
+
+ (alive, unreachable, not_reported, unexpected) = \
+ ping_addresses(addresses)
+
+ ping_statuses = {
+ 'OK': [ '%s is alive' % (ip,) for ip in alive ],
+ 'WARNING': [ '%s was reported despite not being pinged' % (ip,)
+ for ip in unexpected ],
+ 'CRITICAL': [ '%s is unreachable' % (ip,) for ip in unreachable ],
+ 'UNKNOWN': [ '%s has no information' % (ip,) for ip in not_reported ],
+ }
+ lvl,message = trh_nagioslib.nagios_report(ping_statuses)
+ sys.stdout.write(message)
+ return lvl
+
+
+
+if __name__ == '__main__':
+ try:
+ code = main(sys.argv)
+ sys.exit(code)
+ except ProgramFailure as failure:
+ sys.exit(failure.status)
+ except Exception:
+ # An exception would normally cause Python to exit with code == 1,
+ # but that would be a WARNING for Nagios. Avoid that.
+ (exc_type, exc_value, exc_traceback) = sys.exc_info()
+ import traceback
+ traceback.print_exception(exc_type, exc_value, exc_traceback)
+ sys.exit(os.EX_SOFTWARE)
diff --git a/pkg/rpm.spec.in b/pkg/rpm.spec.in
index 1b48a58663edebf970af9980681871da31c27d4e..d8e8576c43539afcb44a6876de31bf8a240ae0c3 100644
--- a/pkg/rpm.spec.in
+++ b/pkg/rpm.spec.in
@@ -27,6 +27,7 @@ Requires: net-snmp-python
Requires: python-ipaddr
# net-snmp-libs and libsmi contain needed MIB files
Requires: net-snmp-libs, libsmi
+Requires: fping
%description
diff --git a/trh_nagioslib.py b/trh_nagioslib.py
index 8eb5661ae84ca9dcae04a2e8db7879ce11f6a77f..ba64ef14fffc7dca22cbd155d9524a70bc9df197 100644
--- a/trh_nagioslib.py
+++ b/trh_nagioslib.py
@@ -13,6 +13,8 @@ import re
import collections
import netsnmp
import subprocess
+import optparse
+import textwrap
# Cache of mappings from short names to fully qualified names.
@@ -251,3 +253,21 @@ def nagios_report(statuses):
message = "UNKNOWN: No status report\n"
return max_level, message
+
+
+
+class NoReflowHelpFormatter(optparse.IndentedHelpFormatter):
+ """A HelpFormatter for optparse that does not re-wrap/reflow text.
+
+ Intended for command descriptions that are already properly
+ pre-formatted.
+ """
+ def format_description(self, description):
+ if not description:
+ return ""
+ desc_width = min(70, self.width - self.current_indent)
+ indent = " " * self.current_indent
+ summary,body = (description.strip().split("\n", 1) + [""])[:2]
+ body = textwrap.dedent(body)
+ description = summary + "\n" + body + "\n"
+ return description