]> git.proxmox.com Git - pve-common.git/blame - src/PVE/Network.pm
Consider /31 and /32 valid subnet masks.
[pve-common.git] / src / PVE / Network.pm
CommitLineData
b9436cda
DM
1package PVE::Network;
2
3use strict;
c36f332e 4use warnings;
74d1b045 5use PVE::Tools qw(run_command);
b9436cda
DM
6use PVE::ProcFSTools;
7use PVE::INotify;
8use File::Basename;
b6bff92e
WB
9use IO::Socket::IP;
10use POSIX qw(ECONNREFUSED);
b9436cda 11
bf52d27b
WB
12use Net::IP;
13
12a235d6
WB
14use Socket qw(IPPROTO_IP);
15
16use constant IFF_UP => 1;
17use constant IFNAMSIZ => 16;
f029c1d0 18use constant SIOCGIFFLAGS => 0x8913;
12a235d6 19
b9436cda
DM
20# host network related utility functions
21
61aa94e4
WB
22our $ipv4_reverse_mask = [
23 '0.0.0.0',
24 '128.0.0.0',
25 '192.0.0.0',
26 '224.0.0.0',
27 '240.0.0.0',
28 '248.0.0.0',
29 '252.0.0.0',
30 '254.0.0.0',
31 '255.0.0.0',
32 '255.128.0.0',
33 '255.192.0.0',
34 '255.224.0.0',
35 '255.240.0.0',
36 '255.248.0.0',
37 '255.252.0.0',
38 '255.254.0.0',
39 '255.255.0.0',
40 '255.255.128.0',
41 '255.255.192.0',
42 '255.255.224.0',
43 '255.255.240.0',
44 '255.255.248.0',
45 '255.255.252.0',
46 '255.255.254.0',
47 '255.255.255.0',
48 '255.255.255.128',
49 '255.255.255.192',
50 '255.255.255.224',
51 '255.255.255.240',
52 '255.255.255.248',
53 '255.255.255.252',
54 '255.255.255.254',
55 '255.255.255.255',
56];
57
58our $ipv4_mask_hash_localnet = {
59 '255.255.0.0' => 16,
60 '255.255.128.0' => 17,
61 '255.255.192.0' => 18,
62 '255.255.224.0' => 19,
63 '255.255.240.0' => 20,
64 '255.255.248.0' => 21,
65 '255.255.252.0' => 22,
66 '255.255.254.0' => 23,
67 '255.255.255.0' => 24,
68 '255.255.255.128' => 25,
69 '255.255.255.192' => 26,
70 '255.255.255.224' => 27,
71 '255.255.255.240' => 28,
72 '255.255.255.248' => 29,
73 '255.255.255.252' => 30,
e43faad9
WB
74 '255.255.255.254' => 31,
75 '255.255.255.255' => 32,
61aa94e4
WB
76};
77
74d1b045
DM
78sub setup_tc_rate_limit {
79 my ($iface, $rate, $burst, $debug) = @_;
80
2d6b3a90
FG
81 # these are allowed / expected to fail, e.g. when there is no previous rate limit to remove
82 eval { run_command("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1"); };
83 eval { run_command("/sbin/tc filter del dev $iface parent ffff: protocol all pref 50 u32 >/dev/null 2>&1"); };
84 eval { run_command("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1"); };
85 eval { run_command("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1"); };
74d1b045 86
d6f2623b 87 return if !$rate;
957753df 88
74d1b045
DM
89 # tbf does not work for unknown reason
90 #$TC qdisc add dev $DEV root tbf rate $RATE latency 100ms burst $BURST
91 # so we use htb instead
92 run_command("/sbin/tc qdisc add dev $iface root handle 1: htb default 1");
93 run_command("/sbin/tc class add dev $iface parent 1: classid 1:1 " .
94 "htb rate ${rate}bps burst ${burst}b");
95
5d35df41
W
96 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
97 run_command("/sbin/tc filter add dev $iface parent ffff: " .
1b915170 98 "prio 50 basic " .
5d35df41
W
99 "police rate ${rate}bps burst ${burst}b mtu 64kb " .
100 "drop flowid :1");
101
74d1b045
DM
102 if ($debug) {
103 print "DEBUG tc settings\n";
104 system("/sbin/tc qdisc ls dev $iface");
105 system("/sbin/tc class ls dev $iface");
106 system("/sbin/tc filter ls dev $iface parent ffff:");
107 }
108}
109
ec9ada18
AD
110sub tap_rate_limit {
111 my ($iface, $rate) = @_;
112
113 my $debug = 0;
ad066ae2 114 $rate = int($rate*1024*1024) if $rate;
ec9ada18
AD
115 my $burst = 1024*1024;
116
117 setup_tc_rate_limit($iface, $rate, $burst, $debug);
118}
74d1b045 119
605bb891
DM
120my $read_bridge_mtu = sub {
121 my ($bridge) = @_;
122
123 my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu");
124 die "bridge '$bridge' does not exist\n" if !$mtu;
125 # avoid insecure dependency;
126 die "unable to parse mtu value" if $mtu !~ /^(\d+)$/;
127 $mtu = int($1);
128
129 return $mtu;
130};
131
32cb7d27 132my $parse_tap_device_name = sub {
6c80e6d6 133 my ($iface, $noerr) = @_;
605bb891
DM
134
135 my ($vmid, $devid);
136
137 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
138 $vmid = $1;
139 $devid = $2;
32cb7d27 140 } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) {
605bb891
DM
141 $vmid = $1;
142 $devid = $2;
143 } else {
6c80e6d6
DM
144 return undef if $noerr;
145 die "can't create firewall bridge for random interface name '$iface'\n";
605bb891
DM
146 }
147
148 return ($vmid, $devid);
149};
150
70ab4434 151my $compute_fwbr_names = sub {
605bb891
DM
152 my ($vmid, $devid) = @_;
153
154 my $fwbr = "fwbr${vmid}i${devid}";
f193aa74 155 # Note: the firewall use 'fwln+' to filter traffic to VMs
7d78a966
AD
156 my $vethfw = "fwln${vmid}i${devid}";
157 my $vethfwpeer = "fwpr${vmid}p${devid}";
158 my $ovsintport = "fwln${vmid}o${devid}";
605bb891 159
70ab4434 160 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
605bb891
DM
161};
162
163my $cond_create_bridge = sub {
164 my ($bridge) = @_;
165
166 if (! -d "/sys/class/net/$bridge") {
167 system("/sbin/brctl addbr $bridge") == 0 ||
168 die "can't add bridge '$bridge'\n";
169 }
170};
171
172my $bridge_add_interface = sub {
b0b34ffd 173 my ($bridge, $iface, $tag, $trunks) = @_;
605bb891
DM
174
175 system("/sbin/brctl addif $bridge $iface") == 0 ||
176 die "can't add interface 'iface' to bridge '$bridge'\n";
4d25f4aa
AD
177
178 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
179
180 if ($vlan_aware) {
181 if ($tag) {
182 system("/sbin/bridge vlan add dev $iface vid $tag pvid untagged") == 0 ||
183 die "unable to add vlan $tag to interface $iface\n";
5d662b31
DC
184
185 warn "Caution: Setting VLAN ID 1 on a VLAN aware bridge may be dangerous\n" if $tag == 1;
4d25f4aa
AD
186 } else {
187 system("/sbin/bridge vlan add dev $iface vid 2-4094") == 0 ||
b0b34ffd 188 die "unable to add default vlan tags to interface $iface\n" if !$trunks;
4d25f4aa 189 }
b0b34ffd 190
846337ad
WB
191 if ($trunks) {
192 my @trunks_array = split /;/, $trunks;
193 foreach my $trunk (@trunks_array) {
194 system("/sbin/bridge vlan add dev $iface vid $trunk") == 0 ||
195 die "unable to add vlan $trunk to interface $iface\n";
196 }
b0b34ffd 197 }
4d25f4aa 198 }
605bb891
DM
199};
200
70ab4434 201my $ovs_bridge_add_port = sub {
b0b34ffd
AD
202 my ($bridge, $iface, $tag, $internal, $trunks) = @_;
203
204 $trunks =~ s/;/,/g if $trunks;
70ab4434
DM
205
206 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
207 $cmd .= " tag=$tag" if $tag;
b0b34ffd
AD
208 $cmd .= " trunks=". join(',', $trunks) if $trunks;
209 $cmd .= " vlan_mode=native-untagged" if $tag && $trunks;
210
70ab4434
DM
211 $cmd .= " -- set Interface $iface type=internal" if $internal;
212 system($cmd) == 0 ||
213 die "can't add ovs port '$iface'\n";
214};
215
605bb891
DM
216my $activate_interface = sub {
217 my ($iface) = @_;
218
219 system("/sbin/ip link set $iface up") == 0 ||
220 die "can't activate interface '$iface'\n";
221};
222
3aa99c70
AD
223sub tap_create {
224 my ($iface, $bridge) = @_;
225
226 die "unable to get bridge setting\n" if !$bridge;
227
605bb891 228 my $bridgemtu = &$read_bridge_mtu($bridge);
3aa99c70 229
098795e0
DM
230 eval {
231 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
232 };
233 die "interface activation failed\n" if $@;
3aa99c70
AD
234}
235
35efc4eb
AD
236sub veth_create {
237 my ($veth, $vethpeer, $bridge, $mac) = @_;
238
239 die "unable to get bridge setting\n" if !$bridge;
240
241 my $bridgemtu = &$read_bridge_mtu($bridge);
242
243 # create veth pair
244 if (! -d "/sys/class/net/$veth") {
245 my $cmd = "/sbin/ip link add name $veth type veth peer name $vethpeer mtu $bridgemtu";
246 $cmd .= " addr $mac" if $mac;
247 system($cmd) == 0 || die "can't create interface $veth\n";
248 }
249
250 # up vethpair
251 &$activate_interface($veth);
252 &$activate_interface($vethpeer);
253}
254
f3f0bc3a
AD
255sub veth_delete {
256 my ($veth) = @_;
257
258 if (-d "/sys/class/net/$veth") {
259 run_command("/sbin/ip link delete dev $veth", outfunc => sub {}, errfunc => sub {});
260 }
261
262}
35efc4eb 263
605bb891 264my $create_firewall_bridge_linux = sub {
b0b34ffd 265 my ($iface, $bridge, $tag, $trunks) = @_;
605bb891 266
32cb7d27 267 my ($vmid, $devid) = &$parse_tap_device_name($iface);
70ab4434 268 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
605bb891 269
605bb891
DM
270 &$cond_create_bridge($fwbr);
271 &$activate_interface($fwbr);
272
273 copy_bridge_config($bridge, $fwbr);
35efc4eb 274 veth_create($vethfw, $vethfwpeer, $bridge);
605bb891 275
7d78a966 276 &$bridge_add_interface($fwbr, $vethfw);
b0b34ffd 277 &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks);
605bb891 278
4d25f4aa 279 &$bridge_add_interface($fwbr, $iface);
605bb891
DM
280};
281
70ab4434 282my $create_firewall_bridge_ovs = sub {
b0b34ffd 283 my ($iface, $bridge, $tag, $trunks) = @_;
70ab4434 284
32cb7d27 285 my ($vmid, $devid) = &$parse_tap_device_name($iface);
70ab4434
DM
286 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
287
288 my $bridgemtu = &$read_bridge_mtu($bridge);
289
290 &$cond_create_bridge($fwbr);
291 &$activate_interface($fwbr);
292
293 &$bridge_add_interface($fwbr, $iface);
294
b0b34ffd 295 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks);
ac3a04b8 296 &$activate_interface($ovsintport);
70ab4434
DM
297
298 # set the same mtu for ovs int port
299 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
300
301 &$bridge_add_interface($fwbr, $ovsintport);
302};
303
304my $cleanup_firewall_bridge = sub {
605bb891
DM
305 my ($iface) = @_;
306
32cb7d27 307 my ($vmid, $devid) = &$parse_tap_device_name($iface, 1);
6c80e6d6 308 return if !defined($vmid);
70ab4434
DM
309 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
310
311 # cleanup old port config from any openvswitch bridge
312 if (-d "/sys/class/net/$ovsintport") {
313 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
314 }
605bb891
DM
315
316 # delete old vethfw interface
f3f0bc3a 317 veth_delete($vethfw);
605bb891
DM
318
319 # cleanup fwbr bridge
320 if (-d "/sys/class/net/$fwbr") {
321 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
322 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
323 }
324};
325
f0c190ee 326sub tap_plug {
bce2a5b3 327 my ($iface, $bridge, $tag, $firewall, $trunks, $rate) = @_;
f0c190ee 328
4cbabd40
AD
329 #cleanup old port config from any openvswitch bridge
330 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
331
098795e0 332 if (-d "/sys/class/net/$bridge/bridge") {
70ab4434 333 &$cleanup_firewall_bridge($iface); # remove stale devices
605bb891 334
4d25f4aa 335 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
098795e0 336
4d25f4aa 337 if (!$vlan_aware) {
b0b34ffd 338 die "vlan aware feature need to be enabled to use trunks" if $trunks;
4d25f4aa
AD
339 my $newbridge = activate_bridge_vlan($bridge, $tag);
340 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
ff042056 341 $bridge = $newbridge;
4d25f4aa
AD
342 $tag = undef;
343 }
344
345 if ($firewall) {
b0b34ffd 346 &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks);
4d25f4aa 347 } else {
b0b34ffd 348 &$bridge_add_interface($bridge, $iface, $tag, $trunks);
4d25f4aa 349 }
605bb891 350
098795e0 351 } else {
70ab4434
DM
352 &$cleanup_firewall_bridge($iface); # remove stale devices
353
354 if ($firewall) {
b0b34ffd 355 &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks);
70ab4434 356 } else {
b0b34ffd 357 &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks);
70ab4434 358 }
4cbabd40 359 }
bce2a5b3
WB
360
361 tap_rate_limit($iface, $rate);
f0c190ee
AD
362}
363
a84b65c0 364sub tap_unplug {
2db1cc0d 365 my ($iface) = @_;
a84b65c0 366
2db1cc0d
DM
367 my $path= "/sys/class/net/$iface/brport/bridge";
368 if (-l $path) {
369 my $bridge = basename(readlink($path));
370 #avoid insecure dependency;
371 ($bridge) = $bridge =~ /(\S+)/;
4cbabd40 372
098795e0 373 system("/sbin/brctl delif $bridge $iface") == 0 ||
2db1cc0d 374 die "can't del interface '$iface' from bridge '$bridge'\n";
605bb891 375
4cbabd40 376 }
70ab4434
DM
377
378 &$cleanup_firewall_bridge($iface);
dd44486e
WB
379 #cleanup old port config from any openvswitch bridge
380 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
a84b65c0
AD
381}
382
b9436cda
DM
383sub copy_bridge_config {
384 my ($br0, $br1) = @_;
385
386 return if $br0 eq $br1;
387
388 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
ba4af65b 389 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
b9436cda
DM
390
391 foreach my $sysname (@$br_configs) {
392 eval {
393 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
394 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
395 if ($v0 ne $v1) {
aec04803 396 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
b9436cda
DM
397 }
398 };
399 warn $@ if $@;
400 }
401}
402
70d89745
PRG
403sub activate_bridge_vlan_slave {
404 my ($bridgevlan, $iface, $tag) = @_;
b9436cda 405 my $ifacevlan = "${iface}.$tag";
70d89745 406
b9436cda
DM
407 # create vlan on $iface is not already exist
408 if (! -d "/sys/class/net/$ifacevlan") {
6fc54cb2 409 system("/sbin/ip link add link $iface name ${iface}.${tag} type vlan id $tag") == 0 ||
02c9a6b4 410 die "can't add vlan tag $tag to interface $iface\n";
b9436cda
DM
411 }
412
413 # be sure to have the $ifacevlan up
605bb891 414 &$activate_interface($ifacevlan);
b9436cda
DM
415
416 # test if $vlaniface is already enslaved in another bridge
417 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
418 if (-l $path) {
419 my $tbridge = basename(readlink($path));
70d89745 420 if ($tbridge ne $bridgevlan) {
b9436cda 421 die "interface $ifacevlan already exist in bridge $tbridge\n";
eee4b32a
PRG
422 } else {
423 # Port already attached to bridge: do nothing.
424 return;
b9436cda
DM
425 }
426 }
427
70d89745 428 # add $ifacevlan to the bridge
605bb891 429 &$bridge_add_interface($bridgevlan, $ifacevlan);
70d89745
PRG
430}
431
432sub activate_bridge_vlan {
433 my ($bridge, $tag_param) = @_;
434
435 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
436
437 return $bridge if !defined($tag_param); # no vlan, simply return
438
439 my $tag = int($tag_param);
440
441 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
442
443 my $bridgevlan = "${bridge}v$tag";
444
c9030d97
PRG
445 my @ifaces = ();
446 my $dir = "/sys/class/net/$bridge/brif";
899f8c4a 447 PVE::Tools::dir_glob_foreach($dir, '(((eth|bond)\d+|en[^.]+)(\.\d+)?)', sub {
5ffa7628 448 push @ifaces, $_[0];
c9030d97
PRG
449 });
450
5ffa7628 451 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
c9030d97 452
b9436cda
DM
453 # add bridgevlan if it doesn't already exist
454 if (! -d "/sys/class/net/$bridgevlan") {
9e14b1b7 455 system("/sbin/brctl addbr $bridgevlan") == 0 ||
b9436cda
DM
456 die "can't add bridge $bridgevlan\n";
457 }
458
70d89745 459 # for each physical interface (eth or bridge) bind them to bridge vlan
c9030d97
PRG
460 foreach my $iface (@ifaces) {
461 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
462 }
70d89745 463
b9436cda
DM
464 #fixme: set other bridge flags
465
466 # be sure to have the bridge up
467 system("/sbin/ip link set $bridgevlan up") == 0 ||
468 die "can't up bridge $bridgevlan\n";
70d89745 469
b9436cda
DM
470 return $bridgevlan;
471}
472
b6bff92e
WB
473sub tcp_ping {
474 my ($host, $port, $timeout) = @_;
475
476 my $refused = 1;
477
478 $timeout = 3 if !$timeout; # sane default
479 if (!$port) {
480 # Net::Ping defaults to the echo port
481 $port = 7;
482 } else {
483 # Net::Ping's port_number() implies service_check(1)
484 $refused = 0;
485 }
486
487 my ($sock, $result);
488 eval {
489 $result = PVE::Tools::run_with_timeout($timeout, sub {
490 $sock = IO::Socket::IP->new(PeerHost => $host, PeerPort => $port, Type => SOCK_STREAM);
491 $result = $refused if $! == ECONNREFUSED;
492 });
493 };
494 if ($sock) {
495 $sock->close();
496 $result = 1;
497 }
498 return $result;
499}
500
bf52d27b
WB
501sub IP_from_cidr {
502 my ($cidr, $version) = @_;
503
504 return if $cidr !~ m!^(\S+?)/(\S+)$!;
505 my ($ip, $prefix) = ($1, $2);
506
507 my $ipobj = Net::IP->new($ip, $version);
508 return if !$ipobj;
509
510 $version = $ipobj->version();
511
512 my $binmask = Net::IP::ip_get_mask($prefix, $version);
513 return if !$binmask;
514
515 my $masked_binip = $ipobj->binip() & $binmask;
516 my $masked_ip = Net::IP::ip_bintoip($masked_binip, $version);
517 return Net::IP->new("$masked_ip/$prefix");
518}
519
520sub is_ip_in_cidr {
521 my ($ip, $cidr, $version) = @_;
522
523 my $cidr_obj = IP_from_cidr($cidr, $version);
524 return undef if !$cidr_obj;
525
526 my $ip_obj = Net::IP->new($ip, $version);
527 return undef if !$ip_obj;
528
529 return $cidr_obj->overlaps($ip_obj) == $Net::IP::IP_B_IN_A_OVERLAP;
530}
531
12a235d6
WB
532# struct ifreq { // FOR SIOCGIFFLAGS:
533# char ifrn_name[IFNAMSIZ]
534# short ifru_flags
535# };
536my $STRUCT_IFREQ_SIOCGIFFLAGS = 'Z' . IFNAMSIZ . 's1';
537sub get_active_interfaces {
538 # Use the interface name list from /proc/net/dev
539 open my $fh, '<', '/proc/net/dev'
540 or die "failed to open /proc/net/dev: $!\n";
541 # And filter by IFF_UP flag fetched via a PF_INET6 socket ioctl:
ab08ec79
WB
542 my $sock;
543 socket($sock, PF_INET6, SOCK_DGRAM, &IPPROTO_IP)
544 or socket($sock, PF_INET, SOCK_DGRAM, &IPPROTO_IP)
545 or return [];
12a235d6
WB
546
547 my $ifaces = [];
548 while(defined(my $line = <$fh>)) {
549 next if $line !~ /^\s*([^:\s]+):/;
550 my $ifname = $1;
c4534006 551 my $ifreq = pack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifname, 0);
f029c1d0 552 if (!defined(ioctl($sock, SIOCGIFFLAGS, $ifreq))) {
12a235d6
WB
553 warn "failed to get interface flags for: $ifname\n";
554 next;
555 }
556 my ($name, $flags) = unpack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifreq);
c4534006 557 push @$ifaces, $ifname if ($flags & IFF_UP);
12a235d6
WB
558 }
559 close $fh;
560 close $sock;
561 return $ifaces;
562}
563
b9436cda 5641;