| 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 |
This module was written for a single purpose only: sending ICMP ECHO |
| 12 |
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 |
use common::sense; |
| 27 |
|
| 28 |
use AnyEvent; |
| 29 |
|
| 30 |
BEGIN { |
| 31 |
our $VERSION = '2.0'; |
| 32 |
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_RES_FD, $ICMP4_FD, $ICMP6_FD); |
| 42 |
|
| 43 |
our $THR_RES_FH; open $THR_RES_FH, "<&=$THR_RES_FD" or die "FATAL: cannot fdopen"; |
| 44 |
|
| 45 |
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 |
|
| 48 |
=item AnyEvent::FastPing::ipv4_supported |
| 49 |
|
| 50 |
Returns true iff IPv4 is supported in this module and on this system. |
| 51 |
|
| 52 |
=item AnyEvent::FastPing::ipv6_supported |
| 53 |
|
| 54 |
Returns true iff IPv6 is supported in this module and on this system. |
| 55 |
|
| 56 |
=item AnyEvent::FastPing::icmp4_pktsize |
| 57 |
|
| 58 |
Returns the number of octets per IPv4 ping packet (the whole IP packet |
| 59 |
including headers, excluding lower-level headers or trailers such as |
| 60 |
Ethernet). |
| 61 |
|
| 62 |
Can be used to calculate e.g. octets/s from rate ... |
| 63 |
|
| 64 |
my $octets_per_second = $packets_per_second * AnyEvent::FastPing::icmp4_pktsize; |
| 65 |
|
| 66 |
... or convert kilobit/second to packet rate ... |
| 67 |
|
| 68 |
my $packets_per_second = $kilobit_per_second |
| 69 |
* (1000 / 8 / AnyEvent::FastPing::icmp4_pktsize); |
| 70 |
|
| 71 |
etc. |
| 72 |
|
| 73 |
=item AnyEvent::FastPing::icmp6_pktsize |
| 74 |
|
| 75 |
Like AnyEvent::FastPing::icmp4_pktsize, but for IPv6. |
| 76 |
|
| 77 |
=back |
| 78 |
|
| 79 |
=head1 THE AnyEvent::FastPing CLASS |
| 80 |
|
| 81 |
The AnyEvent::FastPing class represents a single "pinger". A "pinger" |
| 82 |
comes with its own thread to send packets in the background, a rate-limit |
| 83 |
machinery and separate idle/receive callbacks. |
| 84 |
|
| 85 |
The recommended workflow (there are others) is this: 1. create a new |
| 86 |
AnyEvent::FastPing object 2. configure the address lists and ranges to |
| 87 |
ping, also configure an idle callback and optionally a receive callback |
| 88 |
3. C<start> the pinger. |
| 89 |
|
| 90 |
When the pinger has finished pinging all the configured addresses it will |
| 91 |
call the idle callback. |
| 92 |
|
| 93 |
The pinging process works like this: every range has a minimum interval |
| 94 |
between sends, which is used to limit the rate at which hosts in that |
| 95 |
range are being pinged. Distinct ranges are independent of each other, |
| 96 |
which is why there is a per-pinger "global" minimum interval as well. |
| 97 |
|
| 98 |
The pinger sends pings as fats as possible, while both obeying the pinger |
| 99 |
rate limit as well as range limits. |
| 100 |
|
| 101 |
When a range is exhausted, it is removed. When all ranges are exhausted, |
| 102 |
the pinger waits another C<max_rtt> seconds and then exits, causing the |
| 103 |
idle callback to trigger. |
| 104 |
|
| 105 |
Performance: On my 2 GHz Opteron system with a pretty average nvidia |
| 106 |
gigabit network card I can ping around 60k to 200k adresses per second, |
| 107 |
depending on routing decisions. |
| 108 |
|
| 109 |
Example: ping 10.0.0.1-10.0.0.15 with at most 100 packets/s, and |
| 110 |
11.0.0.1-11.0.255.255 with at most 1000 packets/s. Also ping the IPv6 |
| 111 |
loopback address 5 times as fast as possible. Do not, however, exceed 1000 |
| 112 |
packets/s overall. Dump each received reply: |
| 113 |
|
| 114 |
use AnyEvent::Socket; |
| 115 |
use AnyEvent::FastPing; |
| 116 |
|
| 117 |
my $done = AnyEvent->condvar; |
| 118 |
|
| 119 |
my $pinger = new AnyEvent::FastPing; |
| 120 |
|
| 121 |
$pinger->interval (1/1000); |
| 122 |
$pinger->max_rtt (0.1); # reasonably fast/reliable network |
| 123 |
|
| 124 |
$pinger->add_range (v10.0.0.1, v10.0.0.15, 1/100); |
| 125 |
$pinger->add_range (v11.0.0.1, v11.0.255.255, 1/1000); |
| 126 |
$pinger->add_hosts ([ (v0.0.0.0.0.0.0.0.0.0.0.0.0.0.0.1) x 5 ]); |
| 127 |
|
| 128 |
$pinger->on_recv (sub { |
| 129 |
for (@{ $_[0] }) { |
| 130 |
printf "%s %g\n", (AnyEvent::Socket::format_address $_->[0]), $_->[1]; |
| 131 |
} |
| 132 |
}); |
| 133 |
|
| 134 |
$pinger->on_idle (sub { |
| 135 |
print "done\n"; |
| 136 |
undef $pinger; |
| 137 |
}); |
| 138 |
|
| 139 |
$pinger->start; |
| 140 |
$done->wait; |
| 141 |
|
| 142 |
=head2 METHODS |
| 143 |
|
| 144 |
=over 4 |
| 145 |
|
| 146 |
=item $pinger = new AnyEvent::FastPing |
| 147 |
|
| 148 |
Creates a new pinger - right now there can be at most C<65536> pingers in |
| 149 |
a process, although that limit might change to somethind drastically lower |
| 150 |
- you should be stringy with your pinger objects. |
| 151 |
|
| 152 |
=cut |
| 153 |
|
| 154 |
sub new { |
| 155 |
my ($klass) = @_; |
| 156 |
|
| 157 |
_new $klass, (rand 65536), (rand 65536), (rand 65536) |
| 158 |
} |
| 159 |
|
| 160 |
our @IDLE_CB; |
| 161 |
|
| 162 |
sub DESTROY { |
| 163 |
undef $IDLE_CB[ &id ]; |
| 164 |
&_free; |
| 165 |
} |
| 166 |
|
| 167 |
=item $pinger->on_recv ($callback->([[$host, $rtt], ...])) |
| 168 |
|
| 169 |
Registeres a callback to be called for ping replies. If no callback has |
| 170 |
been registered than ping replies will be ignored, otherwise this module |
| 171 |
calculates the round trip time, in seconds, for each reply and calls this |
| 172 |
callback. |
| 173 |
|
| 174 |
The callback receives a single argument, which is an array reference |
| 175 |
with an entry for each reply packet (the replies will be batched for |
| 176 |
efficiency). Each member in the array reference is again an array |
| 177 |
reference with exactly two members: the binary host addresss (4 octets for |
| 178 |
IPv4, 16 for IPv6) and the approximate round trip time, in seconds. |
| 179 |
|
| 180 |
The replies will be passed to the callback as soon as they arrive, and |
| 181 |
this callback can be called many times with batches of replies. |
| 182 |
|
| 183 |
The receive callback will be called whenever a suitable reply arrives, |
| 184 |
whether generated by this pinger or not, whether this pinger is started |
| 185 |
or not. The packets will have a unique 64 bit ID to distinguish them from |
| 186 |
other pinger objects and other generators, but this doesn't help against |
| 187 |
malicious replies. |
| 188 |
|
| 189 |
Note that very high packet rates can overwhelm your process, causing |
| 190 |
replies to be dropped (configure your kernel with long receive queues for |
| 191 |
raw sockets if this is a problem). |
| 192 |
|
| 193 |
Example: register a callback which simply dumps the received data. |
| 194 |
|
| 195 |
use AnyEvent::Socket; |
| 196 |
|
| 197 |
$pinger->on_recv (sub { |
| 198 |
for (@{ $_[0] }) { |
| 199 |
printf "%s %g\n", (AnyEvent::Socket::format_address $_->[0]), $_->[1]; |
| 200 |
} |
| 201 |
}; |
| 202 |
|
| 203 |
Example: a single ping reply with payload of 1 from C<::1> gets passed |
| 204 |
like this: |
| 205 |
|
| 206 |
[ |
| 207 |
[ "\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\1", 0.000280141830444336 ] |
| 208 |
] |
| 209 |
|
| 210 |
Example: ping replies for C<127.0.0.1> and C<127.0.0.2>: |
| 211 |
|
| 212 |
[ |
| 213 |
[ "\177\0\0\1", 0.00015711784362793 ], |
| 214 |
[ "\177\0\0\2", 0.00090184211731 ] |
| 215 |
] |
| 216 |
|
| 217 |
=item $pinger->on_idle ($callback->()) |
| 218 |
|
| 219 |
Registers a callback to be called when the pinger becomes I<idle>, that |
| 220 |
is, it has been started, has exhausted all ping ranges and waited for |
| 221 |
the C<max_rtt> time. An idle pinger is also stopped, so the callback can |
| 222 |
instantly add new ranges, if it so desires. |
| 223 |
|
| 224 |
=cut |
| 225 |
|
| 226 |
sub on_idle { |
| 227 |
$IDLE_CB[ &id ] = $_[1]; |
| 228 |
} |
| 229 |
|
| 230 |
our $THR_RES_W = AE::io $THR_RES_FH, 0, sub { |
| 231 |
sysread $THR_RES_FH, my $buf, 8; |
| 232 |
|
| 233 |
for my $id (unpack "S*", $buf) { |
| 234 |
_stop_id $id; |
| 235 |
($IDLE_CB[$id] || sub { })->(); |
| 236 |
} |
| 237 |
}; |
| 238 |
|
| 239 |
=item $pinger->interval ($seconds) |
| 240 |
|
| 241 |
Configures the minimum interval between packet sends for this pinger - the |
| 242 |
pinger will not send packets faster than this rate (or atcually 1 / rate), |
| 243 |
even if individual ranges have a lower interval. |
| 244 |
|
| 245 |
A value of C<0> selects the fastests possible speed (currently no faster |
| 246 |
than 1_000_000 packets/s). |
| 247 |
|
| 248 |
=item $pinger->max_rtt ($seconds) |
| 249 |
|
| 250 |
If your idle callback were called instantly after all ranges were |
| 251 |
exhausted and you destroyed the object inside (which is common), then |
| 252 |
there would be no chance to receive some replies, as there would be no |
| 253 |
time fo the packet to travel over the network. |
| 254 |
|
| 255 |
This can be fixed by starting a timer in the idle callback, or more simply |
| 256 |
by selecting a suitable C<max_rtt> value, which should be the maximum time |
| 257 |
you allow a ping packet to travel to its destinazion and back. |
| 258 |
|
| 259 |
The pinger thread automatically waits for this amount of time before becoming idle. |
| 260 |
|
| 261 |
The default is currently C<0.5> seconds, which is usually plenty. |
| 262 |
|
| 263 |
=item $pinger->add_range ($lo, $hi[, $interval]) |
| 264 |
|
| 265 |
Ping the IPv4 (or IPv6, but see below) address range, starting at binary |
| 266 |
address C<$lo> and ending at C<$hi> (both C<$lo> and C<$hi> will be |
| 267 |
pinged), generating no more than one ping per C<$interval> seconds (or as |
| 268 |
fast as possible if omitted). |
| 269 |
|
| 270 |
You can convert IP addresses from text to binary form by |
| 271 |
using C<AnyEvent::Util::parse_address>, C<Socket::inet_aton>, |
| 272 |
C<Socket6::inet_pton> or any other method that you like :) |
| 273 |
|
| 274 |
The algorithm to select the next address is O(log n) on the number of |
| 275 |
ranges, so even a large number of ranges (many thousands) is managable. |
| 276 |
|
| 277 |
No storage is allocated per address. |
| 278 |
|
| 279 |
Note that, while IPv6 addresses are currently supported, the usefulness of |
| 280 |
this option is extremely limited and might be gone in future versions - if |
| 281 |
you want to ping a number of IPv6 hosts, better specify them individually |
| 282 |
using the C<add_hosts> method. |
| 283 |
|
| 284 |
=item $pinger->add_hosts ([$host...], $interval, $interleave) |
| 285 |
|
| 286 |
Similar to C<add_range>, but uses a list of single addresses instead. The |
| 287 |
list is specified as an array reference as first argument. Each entry in |
| 288 |
the array should be a binary host address, either IPv4 or IPv6. Currently, |
| 289 |
all entries in the list must be either IPv4 B<OR> IPv6, so you have to |
| 290 |
create two host ranges if you have mixed addresses. |
| 291 |
|
| 292 |
Minimum C<$interval> is the same as for C<add_range> and can be left out. |
| 293 |
|
| 294 |
C<$interlave> specifies an increment between addresses: often address |
| 295 |
lists are generated in a way that results in clustering - first all |
| 296 |
addresses from one subnet, then from the next, and so on. To avoid this, |
| 297 |
you can specify an interleave factor. If it is C<1> (the default), then |
| 298 |
every address is pinged in the order specified. If it is C<2>, then only |
| 299 |
every second address will be pinged in the first round, followed by a |
| 300 |
second round with the others. Higher factors will create C<$interleave> |
| 301 |
runs of addresses spaced C<$interleave> indices in the list. |
| 302 |
|
| 303 |
The special value C<0> selects a (hopefully) suitable interleave factor |
| 304 |
automatically - currently C<256> for lists with less than 65536 addresses, |
| 305 |
and the square root of the list length otherwise. |
| 306 |
|
| 307 |
=item $pinger->start |
| 308 |
|
| 309 |
Start the pinger, unless it is running already. While a pinger is running |
| 310 |
you must not modify the pinger. If you want to change a parameter, you |
| 311 |
have to C<stop> the pinger first. |
| 312 |
|
| 313 |
The pinger will automatically stop when destroyed. |
| 314 |
|
| 315 |
=item $pinger->stop |
| 316 |
|
| 317 |
Stop the pinger, if it is running. A pinger can be stopped at any time, |
| 318 |
after which it's current state is preserved - starting it again will |
| 319 |
continue where it left off. |
| 320 |
|
| 321 |
=cut |
| 322 |
|
| 323 |
1; |
| 324 |
|
| 325 |
=back |
| 326 |
|
| 327 |
=head1 AUTHOR |
| 328 |
|
| 329 |
Marc Lehmann <schmorp@schmorp.de> |
| 330 |
http://home.schmorp.de/ |
| 331 |
|
| 332 |
=head1 LICENSE |
| 333 |
|
| 334 |
This software is distributed under the GENERAL PUBLIC LICENSE, version 2 |
| 335 |
or any later version or, at your option, the Artistic License. |
| 336 |
|
| 337 |
=cut |
| 338 |
|
