[Mon-commit] mon/mon.d dns.monitor,1.2,1.3

[David N. <vi...@us...>]

Update of /cvsroot/mon/mon/mon.d
In directory sc8-pr-cvs7.sourceforge.net:/tmp/cvs-serv18267

Modified Files:
	dns.monitor 
Log Message:
pulled in various CMU updates, including multiple master support,
configurable timeouts, and better errors in some cases


Index: dns.monitor
===================================================================
RCS file: /cvsroot/mon/mon/mon.d/dns.monitor,v
retrieving revision 1.2
retrieving revision 1.3
diff -C2 -d -r1.2 -r1.3
*** dns.monitor	15 Nov 2004 14:45:19 -0000	1.2
--- dns.monitor	5 Sep 2006 12:52:37 -0000	1.3
***************
*** 2,6 ****
  #
  # Copyright (C) 1998 David Eckelkamp
! # Copyright (C) 2002 David Nolan
  #
  #    This program is free software; you can redistribute it and/or modify
--- 2,6 ----
  #
  # Copyright (C) 1998 David Eckelkamp
! # Copyright (C) 2002-2006 Carnegie Mellon University
  #
  #    This program is free software; you can redistribute it and/or modify
***************
*** 26,34 ****
  =head1 SYNOPSIS
  
! B<dns.monitor> I<-zone zone [-zone zone ...]> I<-master master> I<[-serial_threshold num]> I<[-tcp]> I<server [server ...]>
  
- or
  
! B<dns.monitor> I<-caching_only> I<[-tcp]> I<-query record[:type] [-query record[:type] ...]> I<server [server ...]>
  
  =head1 DESCRIPTION
--- 26,69 ----
  =head1 SYNOPSIS
  
! B<dns.monitor> 
  
  
! =over 12
! 
! ( [ I<-zone zone [-zone zone ...]> 
! 
! 
! =over 4
! 
! 
! I<-master server [-master server ...]>
! 
! I<[-serial_threshold num]> 
! 
! I<[-failsingle]> ]
! 
! =back
! 
! | [ I<-caching_only> 
! 
! =over 4
! 
! I<-query record[:type[:value]] [-query record[:type[:value]] ...]> ] )
! 
! =back
! 
! I<[-tcp]> 
! 
! I<[-retry num]> 
! 
! I<[-retransmit num]> 
! 
! I<[-timeout num]> 
! 
! I<[-debug num]>
! 
! I<server [server ...]>
! 
! =back
  
  =head1 DESCRIPTION
***************
*** 40,68 ****
  server and one or more slave servers.  The I<zone> argument is the
  zone to check. There can be multiple I<zone> arguments. The I<master>
! argument is the master server for the I<zone>.  It will be queried for
! the base information.  Then each I<server> will be queried to verify
! that it has the correct answers.  If the I<serial_threshold> argument
! is provided, the slave servers must return a zone whose serial number
! is no more than the threshold from the serial number of the zone on
! the master.  (Zone serial numbers may not be identical during zone
! propagation, or on Dynamic DNS zones which may be updated hundreds or
! thousands of times an hour) It is assumed that each I<server> is
! supposed to be authoritative for the I<zone>. The I<-tcp> option
! will cause lookups to be done via TCP instead of the default UDP.
  
! In caching mode, specified via the I<-caching_only> switch, B<dns.monitor> 
! will perform a set of DNS queries to one or more servers.  The I<query> 
! argument is the query to perform.  The query may have an optional query type 
! specified as I<:type> on the end of the query.  I.e your.zone.com:MX will
! cause B<dns.monitor> to fetch the MX records for your.zone.com.  There can be 
! multiple I<query> arguments.  Each I<server> will be contacted to verify 
! that it returns a valid response to the query.  If you wish to use 
! B<dns.monitor> to verify that a caching DNS server is actually fetching fresh 
! data from other servers successfully, it is recommended that the DNS records 
! you query should have very short TTLs. 
  
! The exit code of B<dns.monitor> will be the highest number of servers which
! failed on a single zone/query, 0 if no problems occurred, or -1 if an error 
! with the script arguments was detected.
  
  =head1 AUTHOR
--- 75,116 ----
  server and one or more slave servers.  The I<zone> argument is the
  zone to check. There can be multiple I<zone> arguments. The I<master>
! argument is the master server for the I<zone>.  There can be multiple
! I<master> arguments.  The master server(s) will be queried for the
! base information.  If the I<serial_threshold> argument is provided,
! the serials collected from the I<master> servers are checked to be
! within I<serial_threshold>.  The greatest serial of all of the
! I<master> servers is chosen for comparison.  Then each I<server> will
! be queried to verify that it has the correct answers.  If the
! I<serial_threshold> argument is provided, the slave servers must
! return a zone whose serial number is no more than the threshold from
! the serial number of the zone on the master.  (Zone serial numbers may
! not be identical during zone propagation, or on Dynamic DNS zones
! which may be updated hundreds or thousands of times an hour) It is
! assumed that each I<server> is supposed to be authoritative for the
! I<zone>. The I<-tcp> option will cause lookups to be done via TCP
! instead of the default UDP.
  
! In caching mode, specified via the I<-caching_only> switch,
! B<dns.monitor> will perform a set of DNS queries to one or more
! servers.  The I<query> argument is the query to perform.  The query
! may have an optional query type specified as I<:type> on the end of
! the query.  I.e your.zone.com:MX will cause B<dns.monitor> to fetch
! the MX records for your.zone.com.  There can be multiple I<query>
! arguments.  The query type may also have an optional result specified
! as I<:value> on the end of the query (type must also be specified).
! Each I<server> will be contacted to verify that it returns a valid
! response to the query.  If a query result is specified B<dns.monitor>
! will return an error is the DNS query returns an answer which differs
! from the supplied result.  If you wish to use B<dns.monitor> to verify
! that a caching DNS server is actually fetching fresh data from other
! servers successfully, it is recommended that the DNS records you query
! should have very short TTLs.
  
! The exit code of B<dns.monitor> will be the highest number of servers
! which failed on a single zone/query, 0 if no problems occurred, or -1
! if an error with the script arguments was detected.  If all of the
! I<master> servers fail, the return code will be 252.  If using the
! I<failsingle> option and any I<master> server fails, the return code
! will be 251.
  
  =head1 AUTHOR
***************
*** 70,78 ****
  The script was originally written by David Eckelkamp <da...@tr...>
  
! The script was modified to support Caching DNS servers and frequently updating Zone serials by David Nolan <vi...@cm...>
  
  =cut
      
! #use strict;
  use Getopt::Long;
  use English;
--- 118,129 ----
  The script was originally written by David Eckelkamp <da...@tr...>
  
! The script was modified to support Caching DNS servers, configurable
! retry/timeout parameters, multiple DNS Master servers, and
! configurable Zone serials by David Nolan <vi...@cm...> and Jason
! Carr <jc...@an...> from Carnegie Mellon University.
  
  =cut
      
! use strict;
  use Getopt::Long;
  use English;
***************
*** 81,100 ****
  use Net::DNS::Packet;
  use Net::DNS::RR;
  
  my($Program) = basename($0);
  my(@Zones)   = ();
  my(@Queries)   = ();
! my($Master)  = undef;
  my($SerialThreshold) = (0);
  my($CachingServer) = (0);
  my($UseTCP) = (0);
! my(%OptVars) = ("master" => \$Master,
  		"zone" => \@Zones,
                  "serial_threshold" => \$SerialThreshold,
  	        "caching_only" => \$CachingServer,
  		"query" => \@Queries,
! 		"tcp" => \$UseTCP);
  
! if (!GetOptions(\%OptVars, "master=s", "zone=s@", "serial_threshold=s", "caching_only", "tcp", "query=s@")) {
      print STDERR "Problems with Options, sorry\n";
      exit -1;
--- 132,163 ----
  use Net::DNS::Packet;
  use Net::DNS::RR;
+ use Data::Dumper;
  
  my($Program) = basename($0);
  my(@Zones)   = ();
  my(@Queries)   = ();
! my(@Master)  = ();
  my($SerialThreshold) = (0);
  my($CachingServer) = (0);
  my($UseTCP) = (0);
! my ($retries, $retrans, $timeout) = ( 2, 5, undef );
! my $debug = 0;
! my $failsingle = 0;
! 
! my(%OptVars) = (
! 		"master" => \@Master,
  		"zone" => \@Zones,
                  "serial_threshold" => \$SerialThreshold,
  	        "caching_only" => \$CachingServer,
  		"query" => \@Queries,
! 		"retry" => \$retries,
! 		"retransmit" => \$retrans,
! 		"timeout" => \$timeout,
! 		"tcp" => \$UseTCP,
! 		"debug" => \$debug,
! 		"failsingle" => \$failsingle
! 	       );
  
! if (!GetOptions(\%OptVars, "master=s@", "zone=s@", "serial_threshold=s", "caching_only", "tcp", "query=s@", "retry=i", "retransmit=i", "timeout=i", "debug", "failsingle")) {
      print STDERR "Problems with Options, sorry\n";
      exit -1;
***************
*** 102,114 ****
  if ( $#ARGV < 0 ) {
      print STDERR "$Program: at least one server must be specified\n";
      exit -1;
  }
  if (!$CachingServer) {
!     if (!defined($Master)) {
          print STDERR "$Program: The zone master server must be specified\n";
          exit -1;
      }
      if ( !defined(@Zones) ) {
          print STDERR "$Program: At least one zone must be specified\n";
          exit -1;
      }
--- 165,180 ----
  if ( $#ARGV < 0 ) {
      print STDERR "$Program: at least one server must be specified\n";
+     usage();
      exit -1;
  }
  if (!$CachingServer) {
!     if (!defined(@Master)) {
          print STDERR "$Program: The zone master server must be specified\n";
+ 	usage();
          exit -1;
      }
      if ( !defined(@Zones) ) {
          print STDERR "$Program: At least one zone must be specified\n";
+ 	usage();
          exit -1;
      }
***************
*** 116,119 ****
--- 182,186 ----
      if ( !defined(@Queries) ) {
          print STDERR "$Program: At least one query must be specified\n";
+ 	usage();
          exit -1;
      }
***************
*** 125,131 ****
      my($bad_servers, $reason, $failcount, @FailedZones, @FailedServers, @Reasons);
      my($zone, $line, $i);
-     $maxfailcount = 0;
      foreach $zone (@Zones) {
! 	($bad_servers, $reason, $failcount) = dns_verify($zone, $Master, @ARGV);
  	if (defined($bad_servers)) {
  	    $err_cnt = $failcount if ($failcount > $err_cnt);
--- 192,197 ----
      my($bad_servers, $reason, $failcount, @FailedZones, @FailedServers, @Reasons);
      my($zone, $line, $i);
      foreach $zone (@Zones) {
! 	($bad_servers, $reason, $failcount) = dns_verify($zone, \@Master, \@ARGV);
  	if (defined($bad_servers)) {
  	    $err_cnt = $failcount if ($failcount > $err_cnt);
***************
*** 160,171 ****
      my($err_cnt) = 0;
      my($bad_servers, $reason, $failcount, @FailedQuerys, @FailedServers, @Reasons);
!     my($query, $type, $line, $i);
      foreach (@Queries) {
! 	($query, $type) = split /:/;
  	$type = 'A' if ($type eq "");
! 	($bad_servers, $reason, $failcount) = dns_test($query, $type, @ARGV);
  	if (defined($bad_servers)) {
  	    $err_cnt = $failcount if ($failcount > $err_cnt);
! 	    push(@FailedQuerys, "$query $type");
  	    push(@FailedServers, $bad_servers);
  	    push(@Reasons, $reason);
--- 226,238 ----
      my($err_cnt) = 0;
      my($bad_servers, $reason, $failcount, @FailedQuerys, @FailedServers, @Reasons);
!     my($query, $type, $line, $i, $target);
      foreach (@Queries) {
! 	($query, $type, $target) = split /:/;
  	$type = 'A' if ($type eq "");
! 	($bad_servers, $reason, $failcount) = dns_test($query, $type, $target, @ARGV);
  	if (defined($bad_servers)) {
  	    $err_cnt = $failcount if ($failcount > $err_cnt);
!  	    push(@FailedQuerys, "$query $type") if (!$target);
!  	    push(@FailedQuerys, "$query $type == $target $type") if ($target);
  	    push(@FailedServers, $bad_servers);
  	    push(@Reasons, $reason);
***************
*** 197,201 ****
  
      
! # dns_verity($zone, $master, $server, ...)
  
  # This subroutine takes 3 or more arguments. The first argument is the name of
--- 264,268 ----
  
      
! # dns_verify($zone, \@master, \@Servers)
  
  # This subroutine takes 3 or more arguments. The first argument is the name of
***************
*** 218,225 ****
  sub dns_verify {
      # First verify that we have enough arguments.
!     my($Zone, $Master, @Servers) = @_;
      my($result) = undef;
      my(@failed, $res, $soa_req, $Serial, $error_cnt, $server);
  
      # Query the $Master for the SOA of $Zone and get the serial number.
      $res = new Net::DNS::Resolver;
--- 285,297 ----
  sub dns_verify {
      # First verify that we have enough arguments.
!     my($Zone) = shift;
!     my(@Master) = @{shift()};
!     my(@Servers) = @{shift()};
      my($result) = undef;
      my(@failed, $res, $soa_req, $Serial, $error_cnt, $server);
  
+     my(%serials) = ();
+     my(%errors) = ();
+ 
      # Query the $Master for the SOA of $Zone and get the serial number.
      $res = new Net::DNS::Resolver;
***************
*** 227,258 ****
      $res->defnames(0);		# don't append default zone
      $res->recurse(0);		# no recursion
!     $res->retry(2);		# 2 retries before failure
!     $res->nameservers($Master);
!     $soa_req = $res->query($Zone, "SOA");
!     if (!defined($soa_req) || ($soa_req->header->ancount <= 0)) {
! 	return($Master,
! 	       sprintf("SOA query for $Zone from $Master failed %s\n",
! 		       $res->errorstring));
      }
!     unless ($soa_req->header->aa) {
! 	return($Master, 
! 	       sprintf("$Master is not authoritative for $Zone\n"));
      }
!     unless ($soa_req->header->ancount == 1) {
! 	return($Master,
! 	       sprintf("Too many answers for SOA query to %s for %s\n",
! 		       $Master, 
! 		       $Zone));
      }
!     unless (($soa_req->answer)[0]->type eq "SOA") {
! 	return($Master,
! 	       printf("Query for SOA for %s from %s failed: " .
! 		      "return type = %s\n",
! 		      $Zone,
! 		      $Master, 
! 		      ($soa_req->answer)[0]->type),
! 	       scalar @Servers);
      }
!     $Serial = ($soa_req->answer)[0]->serial;
  
      # Now, foreach server given on the command line, get the serial number from
--- 299,380 ----
      $res->defnames(0);		# don't append default zone
      $res->recurse(0);		# no recursion
!     $res->retry($retries);		# retries before failure
!     $res->retrans($retrans);    # retransmission interval
!     $res->udp_timeout($timeout); # set udp timeout
!     $res->tcp_timeout($timeout); # set tcp timeout
! 
!     $error_cnt=0;
! 
!     # Loop through each master server
!     foreach my $qs (@Master) {
! 	$res->nameservers($qs);
! 	$soa_req = $res->query($Zone, "SOA");
! 	if (!defined($soa_req) || ($soa_req->header->ancount <= 0)) {
! 	    $error_cnt++;
! 	    $errors{$qs} = sprintf("SOA query for $Zone from $qs failed %s\n", $res->errorstring);
! 	    if ($res->errorstring eq 'NOERROR') {
! 		$errors{$qs} .= sprintf("    Empty answer received.  (No zone on server?)\n")
! 	    }
! 	    if ($failsingle) { return ($qs, $errors{$qs}, 251); }
! 	    next;
! 	}
! 	unless ($soa_req->header->aa) {
! 	    $error_cnt++;
! 	    $errors{$qs} = sprintf("$qs is not authoritative for $Zone\n");
! 	    if ($failsingle) { return ($qs, $errors{$qs}, 251); }
! 	    next;
! 	}
! 	unless ($soa_req->header->ancount == 1) {
! 	    $error_cnt++;
! 	    $errors{$qs} = sprintf("Too many answers for SOA query to %s for %s\n", $qs, $Zone);
! 	    if ($failsingle) { return ($qs, $errors{$qs}, 251); }
! 	    next;
! 	}
! 	unless (($soa_req->answer)[0]->type eq "SOA") {
! 	    $error_cnt++;
! 	    $errors{$qs} = printf("Query for SOA for %s from %s failed: " . "return type = %s\n", $Zone, $qs, ($soa_req->answer)[0]->type);
! 	    if ($failsingle) { return ($qs, $errors{$qs}, 251); }
! 	    next;
! 	}
! 
! 	$serials{$qs} = ($soa_req->answer)[0]->serial;
      }
! 
! 
!     if ($debug >= 2) {
! 	print Data::Dumper->Dump([\%serials], ['serials']);
      }
! 
! 	
!     if ($error_cnt == scalar @Master) {
! 	# all masters errored
! 	return("", values %errors, 251);
      }
! 	
!     my $maxvalue = undef;
!     my $minvalue = undef;
!     my $maxkey   = undef;
!     my $minkey   = undef;
! 
!     foreach my $key (keys %serials) {
! 	if ($serials{$key} > $maxvalue) {
! 	    $maxvalue = $serials{$key};
! 	    $maxkey = $key;
! 	}
! 
! 	if (($serials{$key} < $minvalue) || (!defined $minkey)) {
! 	    $minvalue = $serials{$key};
! 	    $minkey = $key;
! 	}
!     }	
! 	
!     if (abs($maxvalue - $minvalue) > $SerialThreshold) {
! 	return ($minkey, "\nQuery to $minkey about $Zone failed\n" .
! 		"Serial number = $minvalue, should have been $maxvalue\n", 252)
      }
! 	
!     $Serial = $maxvalue;
! 
!     return ("", "\nNo SOA Serial found for $Zone!?!?", 252) if (!$Serial);
  
      # Now, foreach server given on the command line, get the serial number from
***************
*** 264,268 ****
  	$res->defnames(0);		# don't append default zone
  	$res->recurse(0);		# no recursion
! 	$res->retry(2);		# 2 retries before failure
  	$res->nameservers($server);
  	$soa_req = $res->query($Zone, "SOA");
--- 386,394 ----
  	$res->defnames(0);		# don't append default zone
  	$res->recurse(0);		# no recursion
! 	$res->retry($retries);		
! 	$res->retrans($retrans);
! 	$res->udp_timeout($timeout);
! 	$res->tcp_timeout($timeout);
! 
  	$res->nameservers($server);
  	$soa_req = $res->query($Zone, "SOA");
***************
*** 272,281 ****
  	    $result .= sprintf("\nSOA query for $Zone from $server failed %s\n",
  			       $res->errorstring);
  	    next;
  	}
! 	unless(($soa_req->header->aa || $CachingServer) && $soa_req->header->ancount == 1 &&
! 	       ($soa_req->answer)[0]->type eq "SOA" &&
! 	        ((abs(($soa_req->answer)[0]->serial - $Serial))
!                   <= $SerialThreshold)) {
  	    $error_cnt++;
  	    push(@failed, $server);
--- 398,410 ----
  	    $result .= sprintf("\nSOA query for $Zone from $server failed %s\n",
  			       $res->errorstring);
+ 	    if ($res->errorstring eq 'NOERROR') {
+ 		$result .= sprintf("    Empty answer received.  (No zone on server?)\n");
+ 	    }
  	    next;
  	}
! 	unless($soa_req->header->aa 
! 	       && $soa_req->header->ancount == 1 
! 	       && ($soa_req->answer)[0]->type eq "SOA" 
! 	       && ((abs(($soa_req->answer)[0]->serial - $Serial)) <= $SerialThreshold)) {
  	    $error_cnt++;
  	    push(@failed, $server);
***************
*** 301,310 ****
  
  
! # dns_test($query, $type, $server, ...)
  
! # This subroutine takes 3 or more arguments. The first argument is the name of
  # the DNS record to query.  The second argument is the type of the DNS
! # query to perform. The third and rest of the arguments are taken as
! # names of caching DNS servers.  Each server will be queried for the 
  # given record and type
  
--- 430,440 ----
  
  
! # dns_test($query, $type, $target, $server, ...)
  
! # This subroutine takes 4 or more arguments. The first argument is the name of
  # the DNS record to query.  The second argument is the type of the DNS
! # query to perform. The third argument is the name of a second DNS record to query,
! # whose results should match the first query.  The fourth and rest of the arguments are 
! # taken as names of caching DNS servers.  Each server will be queried for the 
  # given record and type
  
***************
*** 317,323 ****
  sub dns_test {
      # First verify that we have enough arguments.
!     my($Query, $Master, @Servers) = @_;
      my($result) = undef;
!     my(@failed, $res, $soa_req, $Serial, $error_cnt, $server);
  
      # Now, foreach server given on the command line, 
--- 447,453 ----
  sub dns_test {
      # First verify that we have enough arguments.
!     my($Query, $type, $target, @Servers) = @_;
      my($result) = undef;
!     my(@failed, $res, $req, $treq, $Serial, $error_cnt, $server);
  
      # Now, foreach server given on the command line, 
***************
*** 325,347 ****
      $error_cnt = 0;
      foreach $server (@Servers) {
! 	$res = new Net::DNS::Resolver;
! 	$res->usevc(1) if ($UseTCP);
! 	$res->defnames(0);		# don't append default zone
! 	$res->recurse(0);		# no recursion
! 	$res->retry(2);		# 2 retries before failure
! 	$res->nameservers($server);
! 	$soa_req = $res->query($Query, $type);
! 	if (!defined($soa_req) || ($soa_req->header->ancount <= 0)) {
! 	    $error_cnt++;
! 	    push(@failed, $server);
! 	    $result .= sprintf("\n$type query for $Query from $server failed %s\n",
! 			       $res->errorstring);
! 	    next;
! 	}
!     }
      if ($error_cnt == 0) {
! 	return(undef, undef, undef);
      } else {
! 	return("@failed", $result, $error_cnt);
      }
  }
--- 455,519 ----
      $error_cnt = 0;
      foreach $server (@Servers) {
!         $res = new Net::DNS::Resolver;
!         $res->defnames(0);              # don't append default zone
!         $res->retry($retries);          # 2 retries before failure
!         $res->retrans($retrans);
!         $res->udp_timeout($timeout);
!         $res->tcp_timeout($timeout);
!         $res->nameservers($server);
!         $req = $res->query($Query, $type);
!         if (!defined($req) || ($req->header->ancount <= 0)) {
!             $error_cnt++;
!             push(@failed, $server);
!             $result .= sprintf("\n$type query for $Query from $server failed %s\n",
!                                $res->errorstring);
!             next;
!           } elsif ($target) {
!             $treq = $res->query($target, $type);
!             my $status = 0;
!             foreach my $qans ($req->answer) {
!               print STDERR $qans->string."\n" if ($debug);
!               print STDERR $qans->rdatastr."\n" if ($debug);
!               foreach my $tans ($treq->answer) {
!                 print STDERR "target\n" if ($debug);
!                 print STDERR $tans->string."\n" if ($debug);
!                 print STDERR $tans->rdatastr."\n" if ($debug);
!                 if ($tans->rdatastr eq $qans->rdatastr) {
!                   print STDERR "match found\n" if ($debug);
!                   $status = 1;
!                   last;
!                 }
!               }
!               last if ($status);
!             }
!             if (!$status) {
!               $error_cnt++;
!               push @failed, $server;
!               $result .= "Query $Query:$type failed to match $target\n";
!             }
!           }
!       }
      if ($error_cnt == 0) {
!         return(undef, undef, undef);
      } else {
!         return("@failed", $result, $error_cnt);
      }
  }
+ 
+ sub usage {
+   print STDERR <<END_USAGE;
+ Usage: dns.monitor -zone zone [-zone zone ...] 
+                    -master master 
+                    [-serial_threshold num] 
+                    server [server ...]
+    or: dns.monitor -caching_only 
+                    -query record[:type] [-query record[:type] ...] 
+                    server [server ...]
+ Optional Arguments for either mode:
+        -retry num
+        -retransmit num
+        -timeout num
+        -debug num
+        
+ END_USAGE
+ }