AnyEvent-FastPing/FastPing.pm

=head1 NAME

AnyEvent::FastPing - quickly ping a large number of hosts

=head1 SYNOPSIS

 use AnyEvent::FastPing;

=head1 DESCRIPTION

This module was written for a single purpose only: sending ICMP ECHO
REQUEST packets as quickly as possible to a large number of hosts
(thousands to millions).

It employs a sending thread and is fully event-driven (using AnyEvent), so
you have to run an event model supported by AnyEvent to use this module.

=head1 FUNCTIONS

=over 4

=cut

package AnyEvent::FastPing;

use common::sense;

use AnyEvent;

BEGIN {
   our $VERSION = '1.14';
   our @ISA = qw(Exporter);

   require Exporter;
   #Exporter::export_ok_tags (keys %EXPORT_TAGS);

   require XSLoader;
   XSLoader::load (__PACKAGE__, $VERSION);
}

our ($THR_REQ_FD, $THR_RES_FD, $ICMP4_FD, $ICMP6_FD);

our $THR_REQ_FH; open $THR_REQ_FH, ">&=$THR_REQ_FD" or die "FATAL: cannot fdopen";
our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen";

our $THR_REQ_W;
our $THR_RES_W = AE::io $THR_RES_FH, 0, sub {
   my $sv = _read_res
      or return;

   $sv->();
};

our $THR_REQ_BUF;

sub _send_req($) {
   $THR_REQ_BUF .= $_[0];

   $THR_REQ_W ||= AE::io $THR_REQ_FH, 1, sub {
      my $len = syswrite $THR_REQ_FH, $THR_REQ_BUF;
      substr $THR_REQ_BUF, 0, $len, "";

      undef $THR_REQ_W unless length $THR_REQ_BUF;
   };
}

=item AnyEvent::FastPing::ipv4_supported

Returns true if IPv4 is supported in this module and on this system.

=item AnyEvent::FastPing::ipv6_supported

Returns true if IPv6 is supported in this module and on this system.

=item AnyEvent::FastPing::icmp4_pktsize

Returns the number of bytes each IPv4 ping packet has.

=item AnyEvent::FastPing::icmp6_pktsize

Returns the number of bytes each IPv4 ping packet has.

=item AnyEvent::FastPing::icmp_ping [ranges...], $send_interval, $payload, \&callback

Ping the given IPv4 address ranges. Each range is an arrayref of the
form C<[lo, hi, interval]>, where C<lo> and C<hi> are octet strings with
either 4 octets (for IPv4 addresses) or 16 octets (for IPV6 addresses),
representing the lowest and highest address to ping (you can convert a
dotted-quad IPv4 address to this format by using C<inet_aton $address>. The
range C<interval> is the minimum time in seconds between pings to the
given range. If omitted, defaults to C<$send_interval>.

The C<$send_interval> is the minimum interval between sending any two
packets and is a way to make an overall rate limit. If omitted, pings will
be sent as fast as possible.

The C<$payload> is a 32 bit unsigned integer given as the ICMP ECHO
REQUEST ident and sequence numbers (in unspecified order :).

The request will be queued and all requests will be served by a background
thread in order. When all ranges have been pinged, the C<callback> will be
called.

Algorithm: Each range has an associated "next time to send packet"
time. The algorithm loops as long as there are ranges with hosts to be
pinged and always serves the range with the most urgent packet send
time. It will at most send one packet every C<$send_interval> seconds.

This will ensure that pings to the same range are nicely interleaved with
other ranges - this can help reduce per-subnet bandwidth while maintaining
an overall high packet rate.

The algorithm to send each packet is O(log n) on the number of ranges, so
even a large number of ranges (many thousands) is managable.

No storage is allocated per address.

Performance: On my 2 GHz Opteron system with a pretty average nvidia
gigabit network card I can ping around 60k to 200k adresses per second,
depending on routing decisions.

Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and
11.0.0.1-11.0.255.255 with at most 1000 packets/s. Do not, however, exceed
1000 packets/s overall:

   my $done = AnyEvent->condvar;

   AnyEvent::FastPing::icmp_ping
      [
         [v10.0.0.1, v10.0.0.15, .01],
         [v11.0.0.1, v11.0.255.255, .001],
      ],
      .001, 0x12345678,
      sub {
         warn "all ranges pinged\n";
         $done->broadcast;
      }
   ;

   $done->wait;

=cut

sub icmp_ping($$$&) {
   _send_req _req_icmp_ping @_;
}

our $ICMP4_FH;
our $ICMP4_W = (open $ICMP4_FH, "<&=$ICMP4_FD") && AE::io $ICMP4_FH, 0, \&_recv_icmp4;
our $ICMP6_FH;
our $ICMP6_W = (open $ICMP6_FH, "<&=$ICMP6_FD") && AE::io $ICMP6_FH, 0, \&_recv_icmp6;

=item AnyEvent::FastPing::register_cb \&cb

Register a callback that is called for every received ping reply
(regardless of whether a ping is still in process or not and regardless of
whether the reply is actually a reply to a ping sent earlier).

The code reference gets a single parameter - an arrayref with an
entry for each received packet (replies are being batched for greater
efficiency). Each packet is represented by an arrayref with three members:
the source address (an octet string of either 4 (IPv4) or 16 (IPv6) octets
length), the payload as passed to C<icmp_ping> and the round trip time in
seconds.

Example: register a callback which simply dumps the received data. Since
the coderef is created on the fly via sub, it would be hard to unregister
this callback again :)

   AnyEvent::FastPing::register_cb sub {
      for (@{$_[0]}) {
         printf "%s %d %g\n",
            (4 == length $_->[0] ? inet_ntoa $_->[0] : Socket6::inet_ntop (&AF_INET6, $_->[0])),
            $_->[2],
            $_->[1];
      }
   };

Example: a single ping reply with payload of 1 from C<::1> gets passed
like this:

   [ [
     "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\1",
     "0.000280141830444336",
     1
   ] ]

Example: ping replies for C<127.0.0.1> and C<127.0.0.2>, with a payload of
C<0x12345678>:

   [
      [
        "\177\0\0\1",
        "0.00015711784362793",
        305419896
      ],
      [
        "\177\0\0\2",
        "0.00090184211731",
        305419896
      ]
   ]

=item AnyEvent::FastPing::unregister_cb \&cb

Unregister the callback again (make sure you pass the same codereference
as to C<register_cb>).

=cut

our @CB;

sub register_cb($) {
   push @CB, $_[0];
}

sub unregister_cb($) {
   @CB = grep $_ != $_[0], @CB;
}

1;

=back

=head1 AUTHOR

   Marc Lehmann <schmorp@schmorp.de>
   http://home.schmorp.de/

=head1 LICENSE

   This software is distributed under the GENERAL PUBLIC LICENSE, version 2
   or any later version or, at your option, the Artistic License.

=cut

Revision:	1.9
Committed:	Sun Jan 16 11:48:15 2011 UTC (13 years, 4 months ago) by root
Branch:	MAIN
CVS Tags:	rel-1_14
Changes since 1.8:	+8 -9 lines
Log Message:	AE
#	User	Rev	Content
1	root	1.1	=head1 NAME
2
3			AnyEvent::FastPing - quickly ping a large number of hosts
4
5			=head1 SYNOPSIS
6
7			use AnyEvent::FastPing;
8
9			=head1 DESCRIPTION
10
11	root	1.2	This module was written for a single purpose only: sending ICMP ECHO
12	root	1.1	REQUEST packets as quickly as possible to a large number of hosts
13			(thousands to millions).
14
15			It employs a sending thread and is fully event-driven (using AnyEvent), so
16			you have to run an event model supported by AnyEvent to use this module.
17
18			=head1 FUNCTIONS
19
20			=over 4
21
22			=cut
23
24			package AnyEvent::FastPing;
25
26	root	1.9	use common::sense;
27	root	1.1
28			use AnyEvent;
29
30			BEGIN {
31	root	1.9	our $VERSION = '1.14';
32	root	1.1	our @ISA = qw(Exporter);
33
34			require Exporter;
35			#Exporter::export_ok_tags (keys %EXPORT_TAGS);
36
37			require XSLoader;
38			XSLoader::load (__PACKAGE__, $VERSION);
39			}
40
41			our ($THR_REQ_FD, $THR_RES_FD, $ICMP4_FD, $ICMP6_FD);
42
43			our $THR_REQ_FH; open $THR_REQ_FH, ">&=$THR_REQ_FD" or die "FATAL: cannot fdopen";
44			our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen";
45
46			our $THR_REQ_W;
47	root	1.9	our $THR_RES_W = AE::io $THR_RES_FH, 0, sub {
48	root	1.1	my $sv = _read_res
49			or return;
50
51			$sv->();
52	root	1.9	};
53	root	1.1
54			our $THR_REQ_BUF;
55
56			sub _send_req($) {
57			$THR_REQ_BUF .= $_[0];
58
59	root	1.9	$THR_REQ_W \|\|= AE::io $THR_REQ_FH, 1, sub {
60	root	1.1	my $len = syswrite $THR_REQ_FH, $THR_REQ_BUF;
61			substr $THR_REQ_BUF, 0, $len, "";
62
63			undef $THR_REQ_W unless length $THR_REQ_BUF;
64	root	1.9	};
65	root	1.1	}
66
67			=item AnyEvent::FastPing::ipv4_supported
68
69			Returns true if IPv4 is supported in this module and on this system.
70
71			=item AnyEvent::FastPing::ipv6_supported
72
73			Returns true if IPv6 is supported in this module and on this system.
74
75			=item AnyEvent::FastPing::icmp4_pktsize
76
77			Returns the number of bytes each IPv4 ping packet has.
78
79			=item AnyEvent::FastPing::icmp6_pktsize
80
81			Returns the number of bytes each IPv4 ping packet has.
82
83			=item AnyEvent::FastPing::icmp_ping [ranges...], $send_interval, $payload, \&callback
84
85			Ping the given IPv4 address ranges. Each range is an arrayref of the
86			form C<[lo, hi, interval]>, where C<lo> and C<hi> are octet strings with
87			either 4 octets (for IPv4 addresses) or 16 octets (for IPV6 addresses),
88			representing the lowest and highest address to ping (you can convert a
89			dotted-quad IPv4 address to this format by using C<inet_aton $address>. The
90			range C<interval> is the minimum time in seconds between pings to the
91			given range. If omitted, defaults to C<$send_interval>.
92
93			The C<$send_interval> is the minimum interval between sending any two
94			packets and is a way to make an overall rate limit. If omitted, pings will
95	root	1.5	be sent as fast as possible.
96	root	1.1
97			The C<$payload> is a 32 bit unsigned integer given as the ICMP ECHO
98			REQUEST ident and sequence numbers (in unspecified order :).
99
100			The request will be queued and all requests will be served by a background
101			thread in order. When all ranges have been pinged, the C<callback> will be
102			called.
103
104			Algorithm: Each range has an associated "next time to send packet"
105			time. The algorithm loops as long as there are ranges with hosts to be
106			pinged and always serves the range with the most urgent packet send
107			time. It will at most send one packet every C<$send_interval> seconds.
108
109			This will ensure that pings to the same range are nicely interleaved with
110			other ranges - this can help reduce per-subnet bandwidth while maintaining
111			an overall high packet rate.
112
113			The algorithm to send each packet is O(log n) on the number of ranges, so
114			even a large number of ranges (many thousands) is managable.
115
116			No storage is allocated per address.
117
118			Performance: On my 2 GHz Opteron system with a pretty average nvidia
119			gigabit network card I can ping around 60k to 200k adresses per second,
120			depending on routing decisions.
121
122			Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and
123			11.0.0.1-11.0.255.255 with at most 1000 packets/s. Do not, however, exceed
124			1000 packets/s overall:
125
126			my $done = AnyEvent->condvar;
127
128			AnyEvent::FastPing::icmp_ping
129			[
130			[v10.0.0.1, v10.0.0.15, .01],
131			[v11.0.0.1, v11.0.255.255, .001],
132			],
133			.001, 0x12345678,
134			sub {
135			warn "all ranges pinged\n";
136			$done->broadcast;
137			}
138			;
139
140			$done->wait;
141
142			=cut
143
144			sub icmp_ping($$$&) {
145			_send_req _req_icmp_ping @_;
146			}
147
148			our $ICMP4_FH;
149	root	1.9	our $ICMP4_W = (open $ICMP4_FH, "<&=$ICMP4_FD") && AE::io $ICMP4_FH, 0, \&_recv_icmp4;
150	root	1.1	our $ICMP6_FH;
151	root	1.9	our $ICMP6_W = (open $ICMP6_FH, "<&=$ICMP6_FD") && AE::io $ICMP6_FH, 0, \&_recv_icmp6;
152	root	1.1
153			=item AnyEvent::FastPing::register_cb \&cb
154
155			Register a callback that is called for every received ping reply
156			(regardless of whether a ping is still in process or not and regardless of
157			whether the reply is actually a reply to a ping sent earlier).
158
159			The code reference gets a single parameter - an arrayref with an
160	root	1.6	entry for each received packet (replies are being batched for greater
161	root	1.1	efficiency). Each packet is represented by an arrayref with three members:
162			the source address (an octet string of either 4 (IPv4) or 16 (IPv6) octets
163			length), the payload as passed to C<icmp_ping> and the round trip time in
164			seconds.
165
166	root	1.8	Example: register a callback which simply dumps the received data. Since
167			the coderef is created on the fly via sub, it would be hard to unregister
168			this callback again :)
169
170			AnyEvent::FastPing::register_cb sub {
171			for (@{$_[0]}) {
172			printf "%s %d %g\n",
173			(4 == length $_->[0] ? inet_ntoa $_->[0] : Socket6::inet_ntop (&AF_INET6, $_->[0])),
174			$_->[2],
175			$_->[1];
176			}
177			};
178
179	root	1.1	Example: a single ping reply with payload of 1 from C<::1> gets passed
180			like this:
181
182			[ [
183			"\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\1",
184			"0.000280141830444336",
185			1
186			] ]
187
188			Example: ping replies for C<127.0.0.1> and C<127.0.0.2>, with a payload of
189			C<0x12345678>:
190
191			[
192			[
193			"\177\0\0\1",
194			"0.00015711784362793",
195			305419896
196			],
197			[
198			"\177\0\0\2",
199			"0.00090184211731",
200			305419896
201			]
202			]
203
204			=item AnyEvent::FastPing::unregister_cb \&cb
205
206			Unregister the callback again (make sure you pass the same codereference
207			as to C<register_cb>).
208
209			=cut
210
211			our @CB;
212
213	root	1.8	sub register_cb($) {
214	root	1.1	push @CB, $_[0];
215			}
216
217			sub unregister_cb($) {
218			@CB = grep $_ != $_[0], @CB;
219			}
220
221			1;
222
223			=back
224
225			=head1 AUTHOR
226
227	root	1.4	Marc Lehmann <schmorp@schmorp.de>
228			http://home.schmorp.de/
229	root	1.1
230	root	1.4	=head1 LICENSE
231	root	1.1
232	root	1.4	This software is distributed under the GENERAL PUBLIC LICENSE, version 2
233			or any later version or, at your option, the Artistic License.
234	root	1.1
235			=cut
236