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 = '2.0';
   our @ISA = qw(Exporter);

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

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

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

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

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

=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.

=cut

sub new {
   my ($klass) = @_;

   _new $klass, (rand 65536), (rand 65536), (rand 65536)
}

our @IDLE_CB;

sub DESTROY {
   undef $IDLE_CB[ &id ];
   &_free;
}

sub on_idle {
   $IDLE_CB[ &id ] = $_[1];
}

our $THR_RES_W = AE::io $THR_RES_FH, 0, sub {
   sysread $THR_RES_FH, my $buf, 8;

   for my $id (unpack "S*", $buf) {
      _stop_id $id;
      ($IDLE_CB[$id] || sub { })->();
   }
};

for(1..10) {
my $p = new AnyEvent::FastPing;#d#
$p->interval (0);
$p->max_rtt (0.5);
#$p->add_range (v127.0.0.1, v127.255.255.254, 0);
$p->add_range (v1.0.0.1, v1.255.255.254, 0);
$p->on_idle (my $cv = AE::cv);
my $cnt;
$p->on_recv (sub {
      $cnt++;
});
$p->start;
$cv->recv;
warn $cnt;
}

=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 @_;
}

=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:	1.11
Committed:	Mon Jan 31 05:35:48 2011 UTC (13 years, 3 months ago) by root
Branch:	MAIN
Changes since 1.10:	+49 -27 lines
Log Message:	halfway?
#	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	1.11	our $VERSION = '2.0';
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__, $VERSION);
39			}
40
41	root	1.11	our ($THR_RES_FD, $ICMP4_FD, $ICMP6_FD);
42	root	1.1
43			our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen";
44
45	root	1.11	our $ICMP4_FH; our $ICMP4_W = $ICMP4_FD >= 0 && (open $ICMP4_FH, "<&=$ICMP4_FD") && AE::io $ICMP4_FH, 0, \&_recv_icmp4;
46			our $ICMP6_FH; our $ICMP6_W = $ICMP6_FD >= 0 && (open $ICMP6_FH, "<&=$ICMP6_FD") && AE::io $ICMP6_FH, 0, \&_recv_icmp6;
47	root	1.1
48			=item AnyEvent::FastPing::ipv4_supported
49
50			Returns true if IPv4 is supported in this module and on this system.
51
52			=item AnyEvent::FastPing::ipv6_supported
53
54			Returns true if IPv6 is supported in this module and on this system.
55
56			=item AnyEvent::FastPing::icmp4_pktsize
57
58			Returns the number of bytes each IPv4 ping packet has.
59
60			=item AnyEvent::FastPing::icmp6_pktsize
61
62			Returns the number of bytes each IPv4 ping packet has.
63
64	root	1.11	=cut
65
66			sub new {
67			my ($klass) = @_;
68
69			_new $klass, (rand 65536), (rand 65536), (rand 65536)
70			}
71
72			our @IDLE_CB;
73
74			sub DESTROY {
75			undef $IDLE_CB[ &id ];
76			&_free;
77			}
78
79			sub on_idle {
80			$IDLE_CB[ &id ] = $_[1];
81			}
82
83			our $THR_RES_W = AE::io $THR_RES_FH, 0, sub {
84			sysread $THR_RES_FH, my $buf, 8;
85
86			for my $id (unpack "S*", $buf) {
87			_stop_id $id;
88			($IDLE_CB[$id] \|\| sub { })->();
89			}
90			};
91
92			for(1..10) {
93			my $p = new AnyEvent::FastPing;#d#
94			$p->interval (0);
95			$p->max_rtt (0.5);
96			#$p->add_range (v127.0.0.1, v127.255.255.254, 0);
97			$p->add_range (v1.0.0.1, v1.255.255.254, 0);
98			$p->on_idle (my $cv = AE::cv);
99			my $cnt;
100			$p->on_recv (sub {
101			$cnt++;
102			});
103			$p->start;
104			$cv->recv;
105			warn $cnt;
106			}
107
108	root	1.1	=item AnyEvent::FastPing::icmp_ping [ranges...], $send_interval, $payload, \&callback
109
110			Ping the given IPv4 address ranges. Each range is an arrayref of the
111			form C<[lo, hi, interval]>, where C<lo> and C<hi> are octet strings with
112			either 4 octets (for IPv4 addresses) or 16 octets (for IPV6 addresses),
113			representing the lowest and highest address to ping (you can convert a
114			dotted-quad IPv4 address to this format by using C<inet_aton $address>. The
115			range C<interval> is the minimum time in seconds between pings to the
116			given range. If omitted, defaults to C<$send_interval>.
117
118			The C<$send_interval> is the minimum interval between sending any two
119			packets and is a way to make an overall rate limit. If omitted, pings will
120	root	1.5	be sent as fast as possible.
121	root	1.1
122			The C<$payload> is a 32 bit unsigned integer given as the ICMP ECHO
123			REQUEST ident and sequence numbers (in unspecified order :).
124
125			The request will be queued and all requests will be served by a background
126			thread in order. When all ranges have been pinged, the C<callback> will be
127			called.
128
129			Algorithm: Each range has an associated "next time to send packet"
130			time. The algorithm loops as long as there are ranges with hosts to be
131			pinged and always serves the range with the most urgent packet send
132			time. It will at most send one packet every C<$send_interval> seconds.
133
134			This will ensure that pings to the same range are nicely interleaved with
135			other ranges - this can help reduce per-subnet bandwidth while maintaining
136			an overall high packet rate.
137
138			The algorithm to send each packet is O(log n) on the number of ranges, so
139			even a large number of ranges (many thousands) is managable.
140
141			No storage is allocated per address.
142
143			Performance: On my 2 GHz Opteron system with a pretty average nvidia
144			gigabit network card I can ping around 60k to 200k adresses per second,
145			depending on routing decisions.
146
147			Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and
148			11.0.0.1-11.0.255.255 with at most 1000 packets/s. Do not, however, exceed
149			1000 packets/s overall:
150
151			my $done = AnyEvent->condvar;
152
153			AnyEvent::FastPing::icmp_ping
154			[
155			[v10.0.0.1, v10.0.0.15, .01],
156			[v11.0.0.1, v11.0.255.255, .001],
157			],
158			.001, 0x12345678,
159			sub {
160			warn "all ranges pinged\n";
161			$done->broadcast;
162			}
163			;
164
165			$done->wait;
166
167			=cut
168
169			sub icmp_ping($$$&) {
170	root	1.11	# _send_req _req_icmp_ping @_;
171	root	1.1	}
172
173			=item AnyEvent::FastPing::register_cb \&cb
174
175			Register a callback that is called for every received ping reply
176			(regardless of whether a ping is still in process or not and regardless of
177			whether the reply is actually a reply to a ping sent earlier).
178
179			The code reference gets a single parameter - an arrayref with an
180	root	1.6	entry for each received packet (replies are being batched for greater
181	root	1.1	efficiency). Each packet is represented by an arrayref with three members:
182			the source address (an octet string of either 4 (IPv4) or 16 (IPv6) octets
183			length), the payload as passed to C<icmp_ping> and the round trip time in
184			seconds.
185
186	root	1.8	Example: register a callback which simply dumps the received data. Since
187			the coderef is created on the fly via sub, it would be hard to unregister
188			this callback again :)
189
190			AnyEvent::FastPing::register_cb sub {
191			for (@{$_[0]}) {
192			printf "%s %d %g\n",
193			(4 == length $_->[0] ? inet_ntoa $_->[0] : Socket6::inet_ntop (&AF_INET6, $_->[0])),
194			$_->[2],
195			$_->[1];
196			}
197			};
198
199	root	1.1	Example: a single ping reply with payload of 1 from C<::1> gets passed
200			like this:
201
202			[ [
203			"\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\1",
204			"0.000280141830444336",
205			1
206			] ]
207
208			Example: ping replies for C<127.0.0.1> and C<127.0.0.2>, with a payload of
209			C<0x12345678>:
210
211			[
212			[
213			"\177\0\0\1",
214			"0.00015711784362793",
215			305419896
216			],
217			[
218			"\177\0\0\2",
219			"0.00090184211731",
220			305419896
221			]
222			]
223
224			=item AnyEvent::FastPing::unregister_cb \&cb
225
226			Unregister the callback again (make sure you pass the same codereference
227			as to C<register_cb>).
228
229			=cut
230
231			our @CB;
232
233	root	1.8	sub register_cb($) {
234	root	1.1	push @CB, $_[0];
235			}
236
237			sub unregister_cb($) {
238			@CB = grep $_ != $_[0], @CB;
239			}
240
241			1;
242
243			=back
244
245			=head1 AUTHOR
246
247	root	1.4	Marc Lehmann <schmorp@schmorp.de>
248			http://home.schmorp.de/
249	root	1.1
250	root	1.4	=head1 LICENSE
251	root	1.1
252	root	1.4	This software is distributed under the GENERAL PUBLIC LICENSE, version 2
253			or any later version or, at your option, the Artistic License.
254	root	1.1
255			=cut
256