]> git.proxmox.com Git - pve-common.git/blob - src/PVE/Network.pm
8cfac0dbe1db369860a660b97d4ede2d76ddafaf
[pve-common.git] / src / PVE / Network.pm
1 package PVE::Network;
2
3 use strict;
4 use warnings;
5 use PVE::Tools qw(run_command);
6 use PVE::ProcFSTools;
7 use PVE::INotify;
8 use File::Basename;
9 use IO::Socket::IP;
10 use POSIX qw(ECONNREFUSED);
11
12 use Net::IP;
13
14 # host network related utility functions
15
16 our $ipv4_reverse_mask = [
17 '0.0.0.0',
18 '128.0.0.0',
19 '192.0.0.0',
20 '224.0.0.0',
21 '240.0.0.0',
22 '248.0.0.0',
23 '252.0.0.0',
24 '254.0.0.0',
25 '255.0.0.0',
26 '255.128.0.0',
27 '255.192.0.0',
28 '255.224.0.0',
29 '255.240.0.0',
30 '255.248.0.0',
31 '255.252.0.0',
32 '255.254.0.0',
33 '255.255.0.0',
34 '255.255.128.0',
35 '255.255.192.0',
36 '255.255.224.0',
37 '255.255.240.0',
38 '255.255.248.0',
39 '255.255.252.0',
40 '255.255.254.0',
41 '255.255.255.0',
42 '255.255.255.128',
43 '255.255.255.192',
44 '255.255.255.224',
45 '255.255.255.240',
46 '255.255.255.248',
47 '255.255.255.252',
48 '255.255.255.254',
49 '255.255.255.255',
50 ];
51
52 our $ipv4_mask_hash_localnet = {
53 '255.255.0.0' => 16,
54 '255.255.128.0' => 17,
55 '255.255.192.0' => 18,
56 '255.255.224.0' => 19,
57 '255.255.240.0' => 20,
58 '255.255.248.0' => 21,
59 '255.255.252.0' => 22,
60 '255.255.254.0' => 23,
61 '255.255.255.0' => 24,
62 '255.255.255.128' => 25,
63 '255.255.255.192' => 26,
64 '255.255.255.224' => 27,
65 '255.255.255.240' => 28,
66 '255.255.255.248' => 29,
67 '255.255.255.252' => 30,
68 };
69
70 sub setup_tc_rate_limit {
71 my ($iface, $rate, $burst, $debug) = @_;
72
73 system("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1");
74 system("/sbin/tc filter del dev $iface parent ffff: protocol all pref 50 u32 >/dev/null 2>&1");
75 system("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1");
76 system("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1");
77
78 return if !$rate;
79
80 # tbf does not work for unknown reason
81 #$TC qdisc add dev $DEV root tbf rate $RATE latency 100ms burst $BURST
82 # so we use htb instead
83 run_command("/sbin/tc qdisc add dev $iface root handle 1: htb default 1");
84 run_command("/sbin/tc class add dev $iface parent 1: classid 1:1 " .
85 "htb rate ${rate}bps burst ${burst}b");
86
87 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
88 run_command("/sbin/tc filter add dev $iface parent ffff: " .
89 "protocol all prio 50 u32 match u32 0 0 " .
90 "police rate ${rate}bps burst ${burst}b mtu 64kb " .
91 "drop flowid :1");
92
93 if ($debug) {
94 print "DEBUG tc settings\n";
95 system("/sbin/tc qdisc ls dev $iface");
96 system("/sbin/tc class ls dev $iface");
97 system("/sbin/tc filter ls dev $iface parent ffff:");
98 }
99 }
100
101 sub tap_rate_limit {
102 my ($iface, $rate) = @_;
103
104 my $debug = 0;
105 $rate = int($rate*1024*1024);
106 my $burst = 1024*1024;
107
108 setup_tc_rate_limit($iface, $rate, $burst, $debug);
109 }
110
111 my $read_bridge_mtu = sub {
112 my ($bridge) = @_;
113
114 my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu");
115 die "bridge '$bridge' does not exist\n" if !$mtu;
116 # avoid insecure dependency;
117 die "unable to parse mtu value" if $mtu !~ /^(\d+)$/;
118 $mtu = int($1);
119
120 return $mtu;
121 };
122
123 my $parse_tap_device_name = sub {
124 my ($iface, $noerr) = @_;
125
126 my ($vmid, $devid);
127
128 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
129 $vmid = $1;
130 $devid = $2;
131 } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) {
132 $vmid = $1;
133 $devid = $2;
134 } else {
135 return undef if $noerr;
136 die "can't create firewall bridge for random interface name '$iface'\n";
137 }
138
139 return ($vmid, $devid);
140 };
141
142 my $compute_fwbr_names = sub {
143 my ($vmid, $devid) = @_;
144
145 my $fwbr = "fwbr${vmid}i${devid}";
146 # Note: the firewall use 'fwln+' to filter traffic to VMs
147 my $vethfw = "fwln${vmid}i${devid}";
148 my $vethfwpeer = "fwpr${vmid}p${devid}";
149 my $ovsintport = "fwln${vmid}o${devid}";
150
151 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
152 };
153
154 my $cond_create_bridge = sub {
155 my ($bridge) = @_;
156
157 if (! -d "/sys/class/net/$bridge") {
158 system("/sbin/brctl addbr $bridge") == 0 ||
159 die "can't add bridge '$bridge'\n";
160 }
161 };
162
163 my $bridge_add_interface = sub {
164 my ($bridge, $iface, $tag, $trunks) = @_;
165
166 system("/sbin/brctl addif $bridge $iface") == 0 ||
167 die "can't add interface 'iface' to bridge '$bridge'\n";
168
169 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
170
171 if ($vlan_aware) {
172 if ($tag) {
173 system("/sbin/bridge vlan add dev $iface vid $tag pvid untagged") == 0 ||
174 die "unable to add vlan $tag to interface $iface\n";
175 } else {
176 system("/sbin/bridge vlan add dev $iface vid 2-4094") == 0 ||
177 die "unable to add default vlan tags to interface $iface\n" if !$trunks;
178 }
179
180 if ($trunks) {
181 my @trunks_array = split /;/, $trunks;
182 foreach my $trunk (@trunks_array) {
183 system("/sbin/bridge vlan add dev $iface vid $trunk") == 0 ||
184 die "unable to add vlan $trunk to interface $iface\n";
185 }
186 }
187 }
188 };
189
190 my $ovs_bridge_add_port = sub {
191 my ($bridge, $iface, $tag, $internal, $trunks) = @_;
192
193 $trunks =~ s/;/,/g if $trunks;
194
195 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
196 $cmd .= " tag=$tag" if $tag;
197 $cmd .= " trunks=". join(',', $trunks) if $trunks;
198 $cmd .= " vlan_mode=native-untagged" if $tag && $trunks;
199
200 $cmd .= " -- set Interface $iface type=internal" if $internal;
201 system($cmd) == 0 ||
202 die "can't add ovs port '$iface'\n";
203 };
204
205 my $activate_interface = sub {
206 my ($iface) = @_;
207
208 system("/sbin/ip link set $iface up") == 0 ||
209 die "can't activate interface '$iface'\n";
210 };
211
212 sub tap_create {
213 my ($iface, $bridge) = @_;
214
215 die "unable to get bridge setting\n" if !$bridge;
216
217 my $bridgemtu = &$read_bridge_mtu($bridge);
218
219 eval {
220 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
221 };
222 die "interface activation failed\n" if $@;
223 }
224
225 sub veth_create {
226 my ($veth, $vethpeer, $bridge, $mac) = @_;
227
228 die "unable to get bridge setting\n" if !$bridge;
229
230 my $bridgemtu = &$read_bridge_mtu($bridge);
231
232 # create veth pair
233 if (! -d "/sys/class/net/$veth") {
234 my $cmd = "/sbin/ip link add name $veth type veth peer name $vethpeer mtu $bridgemtu";
235 $cmd .= " addr $mac" if $mac;
236 system($cmd) == 0 || die "can't create interface $veth\n";
237 }
238
239 # up vethpair
240 &$activate_interface($veth);
241 &$activate_interface($vethpeer);
242 }
243
244 sub veth_delete {
245 my ($veth) = @_;
246
247 if (-d "/sys/class/net/$veth") {
248 run_command("/sbin/ip link delete dev $veth", outfunc => sub {}, errfunc => sub {});
249 }
250
251 }
252
253 my $create_firewall_bridge_linux = sub {
254 my ($iface, $bridge, $tag, $trunks) = @_;
255
256 my ($vmid, $devid) = &$parse_tap_device_name($iface);
257 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
258
259 &$cond_create_bridge($fwbr);
260 &$activate_interface($fwbr);
261
262 copy_bridge_config($bridge, $fwbr);
263 veth_create($vethfw, $vethfwpeer, $bridge);
264
265 &$bridge_add_interface($fwbr, $vethfw);
266 &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks);
267
268 &$bridge_add_interface($fwbr, $iface);
269 };
270
271 my $create_firewall_bridge_ovs = sub {
272 my ($iface, $bridge, $tag, $trunks) = @_;
273
274 my ($vmid, $devid) = &$parse_tap_device_name($iface);
275 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
276
277 my $bridgemtu = &$read_bridge_mtu($bridge);
278
279 &$cond_create_bridge($fwbr);
280 &$activate_interface($fwbr);
281
282 &$bridge_add_interface($fwbr, $iface);
283
284 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks);
285 &$activate_interface($ovsintport);
286
287 # set the same mtu for ovs int port
288 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
289
290 &$bridge_add_interface($fwbr, $ovsintport);
291 };
292
293 my $cleanup_firewall_bridge = sub {
294 my ($iface) = @_;
295
296 my ($vmid, $devid) = &$parse_tap_device_name($iface, 1);
297 return if !defined($vmid);
298 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
299
300 # cleanup old port config from any openvswitch bridge
301 if (-d "/sys/class/net/$ovsintport") {
302 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
303 }
304
305 # delete old vethfw interface
306 veth_delete($vethfw);
307
308 # cleanup fwbr bridge
309 if (-d "/sys/class/net/$fwbr") {
310 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
311 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
312 }
313 };
314
315 sub tap_plug {
316 my ($iface, $bridge, $tag, $firewall, $trunks) = @_;
317
318 #cleanup old port config from any openvswitch bridge
319 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
320
321 if (-d "/sys/class/net/$bridge/bridge") {
322 &$cleanup_firewall_bridge($iface); # remove stale devices
323
324 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
325
326 if (!$vlan_aware) {
327 die "vlan aware feature need to be enabled to use trunks" if $trunks;
328 my $newbridge = activate_bridge_vlan($bridge, $tag);
329 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
330 $bridge = $newbridge;
331 $tag = undef;
332 }
333
334 if ($firewall) {
335 &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks);
336 } else {
337 &$bridge_add_interface($bridge, $iface, $tag, $trunks);
338 }
339
340 } else {
341 &$cleanup_firewall_bridge($iface); # remove stale devices
342
343 if ($firewall) {
344 &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks);
345 } else {
346 &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks);
347 }
348 }
349 }
350
351 sub tap_unplug {
352 my ($iface) = @_;
353
354 my $path= "/sys/class/net/$iface/brport/bridge";
355 if (-l $path) {
356 my $bridge = basename(readlink($path));
357 #avoid insecure dependency;
358 ($bridge) = $bridge =~ /(\S+)/;
359
360 system("/sbin/brctl delif $bridge $iface") == 0 ||
361 die "can't del interface '$iface' from bridge '$bridge'\n";
362
363 }
364
365 &$cleanup_firewall_bridge($iface);
366 }
367
368 sub copy_bridge_config {
369 my ($br0, $br1) = @_;
370
371 return if $br0 eq $br1;
372
373 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
374 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
375
376 foreach my $sysname (@$br_configs) {
377 eval {
378 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
379 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
380 if ($v0 ne $v1) {
381 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
382 }
383 };
384 warn $@ if $@;
385 }
386 }
387
388 sub activate_bridge_vlan_slave {
389 my ($bridgevlan, $iface, $tag) = @_;
390 my $ifacevlan = "${iface}.$tag";
391
392 # create vlan on $iface is not already exist
393 if (! -d "/sys/class/net/$ifacevlan") {
394 system("/sbin/ip link add link $iface name ${iface}.${tag} type vlan id $tag") == 0 ||
395 die "can't add vlan tag $tag to interface $iface\n";
396 }
397
398 # be sure to have the $ifacevlan up
399 &$activate_interface($ifacevlan);
400
401 # test if $vlaniface is already enslaved in another bridge
402 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
403 if (-l $path) {
404 my $tbridge = basename(readlink($path));
405 if ($tbridge ne $bridgevlan) {
406 die "interface $ifacevlan already exist in bridge $tbridge\n";
407 } else {
408 # Port already attached to bridge: do nothing.
409 return;
410 }
411 }
412
413 # add $ifacevlan to the bridge
414 &$bridge_add_interface($bridgevlan, $ifacevlan);
415 }
416
417 sub activate_bridge_vlan {
418 my ($bridge, $tag_param) = @_;
419
420 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
421
422 return $bridge if !defined($tag_param); # no vlan, simply return
423
424 my $tag = int($tag_param);
425
426 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
427
428 my $bridgevlan = "${bridge}v$tag";
429
430 my @ifaces = ();
431 my $dir = "/sys/class/net/$bridge/brif";
432 PVE::Tools::dir_glob_foreach($dir, '((eth|bond)\d+(\.\d+)?)', sub {
433 push @ifaces, $_[0];
434 });
435
436 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
437
438 # add bridgevlan if it doesn't already exist
439 if (! -d "/sys/class/net/$bridgevlan") {
440 system("/sbin/brctl addbr $bridgevlan") == 0 ||
441 die "can't add bridge $bridgevlan\n";
442 }
443
444 # for each physical interface (eth or bridge) bind them to bridge vlan
445 foreach my $iface (@ifaces) {
446 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
447 }
448
449 #fixme: set other bridge flags
450
451 # be sure to have the bridge up
452 system("/sbin/ip link set $bridgevlan up") == 0 ||
453 die "can't up bridge $bridgevlan\n";
454
455 return $bridgevlan;
456 }
457
458 sub tcp_ping {
459 my ($host, $port, $timeout) = @_;
460
461 my $refused = 1;
462
463 $timeout = 3 if !$timeout; # sane default
464 if (!$port) {
465 # Net::Ping defaults to the echo port
466 $port = 7;
467 } else {
468 # Net::Ping's port_number() implies service_check(1)
469 $refused = 0;
470 }
471
472 my ($sock, $result);
473 eval {
474 $result = PVE::Tools::run_with_timeout($timeout, sub {
475 $sock = IO::Socket::IP->new(PeerHost => $host, PeerPort => $port, Type => SOCK_STREAM);
476 $result = $refused if $! == ECONNREFUSED;
477 });
478 };
479 if ($sock) {
480 $sock->close();
481 $result = 1;
482 }
483 return $result;
484 }
485
486 sub IP_from_cidr {
487 my ($cidr, $version) = @_;
488
489 return if $cidr !~ m!^(\S+?)/(\S+)$!;
490 my ($ip, $prefix) = ($1, $2);
491
492 my $ipobj = Net::IP->new($ip, $version);
493 return if !$ipobj;
494
495 $version = $ipobj->version();
496
497 my $binmask = Net::IP::ip_get_mask($prefix, $version);
498 return if !$binmask;
499
500 my $masked_binip = $ipobj->binip() & $binmask;
501 my $masked_ip = Net::IP::ip_bintoip($masked_binip, $version);
502 return Net::IP->new("$masked_ip/$prefix");
503 }
504
505 sub is_ip_in_cidr {
506 my ($ip, $cidr, $version) = @_;
507
508 my $cidr_obj = IP_from_cidr($cidr, $version);
509 return undef if !$cidr_obj;
510
511 my $ip_obj = Net::IP->new($ip, $version);
512 return undef if !$ip_obj;
513
514 return $cidr_obj->overlaps($ip_obj) == $Net::IP::IP_B_IN_A_OVERLAP;
515 }
516
517 1;