]> git.proxmox.com Git - pve-common.git/blob - src/PVE/Network.pm
ab073505710150a5a24bea052e52070117fe99a6
[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 use Socket qw(IPPROTO_IP);
15
16 use constant IFF_UP => 1;
17 use constant IFNAMSIZ => 16;
18 use constant SIOCGIFFLAGS => 0x8913;
19
20 # host network related utility functions
21
22 our $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
58 our $ipv4_mask_hash_localnet = {
59 '255.255.0.0' => 16,
60 '255.255.128.0' => 17,
61 '255.255.192.0' => 18,
62 '255.255.224.0' => 19,
63 '255.255.240.0' => 20,
64 '255.255.248.0' => 21,
65 '255.255.252.0' => 22,
66 '255.255.254.0' => 23,
67 '255.255.255.0' => 24,
68 '255.255.255.128' => 25,
69 '255.255.255.192' => 26,
70 '255.255.255.224' => 27,
71 '255.255.255.240' => 28,
72 '255.255.255.248' => 29,
73 '255.255.255.252' => 30,
74 '255.255.255.254' => 31,
75 '255.255.255.255' => 32,
76 };
77
78 sub setup_tc_rate_limit {
79 my ($iface, $rate, $burst, $debug) = @_;
80
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 $iface root >/dev/null 2>&1"); };
86
87 return if !$rate;
88
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
96 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
97 run_command("/sbin/tc filter add dev $iface parent ffff: " .
98 "prio 50 basic " .
99 "police rate ${rate}bps burst ${burst}b mtu 64kb " .
100 "drop flowid :1");
101
102 if ($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 $iface parent ffff:");
107 }
108 }
109
110 sub tap_rate_limit {
111 my ($iface, $rate) = @_;
112
113 my $debug = 0;
114 $rate = int($rate*1024*1024) if $rate;
115 my $burst = 1024*1024;
116
117 setup_tc_rate_limit($iface, $rate, $burst, $debug);
118 }
119
120 my $read_bridge_mtu = sub {
121 my ($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
132 my $parse_tap_device_name = sub {
133 my ($iface, $noerr) = @_;
134
135 my ($vmid, $devid);
136
137 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
138 $vmid = $1;
139 $devid = $2;
140 } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) {
141 $vmid = $1;
142 $devid = $2;
143 } else {
144 return undef if $noerr;
145 die "can't create firewall bridge for random interface name '$iface'\n";
146 }
147
148 return ($vmid, $devid);
149 };
150
151 my $compute_fwbr_names = sub {
152 my ($vmid, $devid) = @_;
153
154 my $fwbr = "fwbr${vmid}i${devid}";
155 # Note: the firewall use 'fwln+' to filter traffic to VMs
156 my $vethfw = "fwln${vmid}i${devid}";
157 my $vethfwpeer = "fwpr${vmid}p${devid}";
158 my $ovsintport = "fwln${vmid}o${devid}";
159
160 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
161 };
162
163 my $cond_create_bridge = sub {
164 my ($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
172 my $bridge_add_interface = sub {
173 my ($bridge, $iface, $tag, $trunks) = @_;
174
175 system("/sbin/brctl addif $bridge $iface") == 0 ||
176 die "can't add interface 'iface' to bridge '$bridge'\n";
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";
184
185 warn "Caution: Setting VLAN ID 1 on a VLAN aware bridge may be dangerous\n" if $tag == 1;
186 } else {
187 system("/sbin/bridge vlan add dev $iface vid 2-4094") == 0 ||
188 die "unable to add default vlan tags to interface $iface\n" if !$trunks;
189 }
190
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 }
197 }
198 }
199 };
200
201 my $ovs_bridge_add_port = sub {
202 my ($bridge, $iface, $tag, $internal, $trunks) = @_;
203
204 $trunks =~ s/;/,/g if $trunks;
205
206 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
207 $cmd .= " tag=$tag" if $tag;
208 $cmd .= " trunks=". join(',', $trunks) if $trunks;
209 $cmd .= " vlan_mode=native-untagged" if $tag && $trunks;
210
211 $cmd .= " -- set Interface $iface type=internal" if $internal;
212 system($cmd) == 0 ||
213 die "can't add ovs port '$iface'\n";
214 };
215
216 my $activate_interface = sub {
217 my ($iface) = @_;
218
219 system("/sbin/ip link set $iface up") == 0 ||
220 die "can't activate interface '$iface'\n";
221 };
222
223 sub tap_create {
224 my ($iface, $bridge) = @_;
225
226 die "unable to get bridge setting\n" if !$bridge;
227
228 my $bridgemtu = &$read_bridge_mtu($bridge);
229
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 $@;
234 }
235
236 sub veth_create {
237 my ($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($vethpeer);
253 }
254
255 sub veth_delete {
256 my ($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 }
263
264 my $create_firewall_bridge_linux = sub {
265 my ($iface, $bridge, $tag, $trunks) = @_;
266
267 my ($vmid, $devid) = &$parse_tap_device_name($iface);
268 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
269
270 &$cond_create_bridge($fwbr);
271 &$activate_interface($fwbr);
272
273 copy_bridge_config($bridge, $fwbr);
274 veth_create($vethfw, $vethfwpeer, $bridge);
275
276 &$bridge_add_interface($fwbr, $vethfw);
277 &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks);
278
279 &$bridge_add_interface($fwbr, $iface);
280 };
281
282 my $create_firewall_bridge_ovs = sub {
283 my ($iface, $bridge, $tag, $trunks) = @_;
284
285 my ($vmid, $devid) = &$parse_tap_device_name($iface);
286 my ($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, $iface);
294
295 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks);
296 &$activate_interface($ovsintport);
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
304 my $cleanup_firewall_bridge = sub {
305 my ($iface) = @_;
306
307 my ($vmid, $devid) = &$parse_tap_device_name($iface, 1);
308 return if !defined($vmid);
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 $ovsintport", outfunc => sub {}, errfunc => sub {});
314 }
315
316 # delete old vethfw interface
317 veth_delete($vethfw);
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 $fwbr", outfunc => sub {}, errfunc => sub {});
323 }
324 };
325
326 sub tap_plug {
327 my ($iface, $bridge, $tag, $firewall, $trunks, $rate) = @_;
328
329 #cleanup old port config from any openvswitch bridge
330 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
331
332 if (-d "/sys/class/net/$bridge/bridge") {
333 &$cleanup_firewall_bridge($iface); # remove stale devices
334
335 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
336
337 if (!$vlan_aware) {
338 die "vlan aware feature need to be enabled to use trunks" if $trunks;
339 my $newbridge = activate_bridge_vlan($bridge, $tag);
340 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
341 $bridge = $newbridge;
342 $tag = undef;
343 }
344
345 if ($firewall) {
346 &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks);
347 } else {
348 &$bridge_add_interface($bridge, $iface, $tag, $trunks);
349 }
350
351 } else {
352 &$cleanup_firewall_bridge($iface); # remove stale devices
353
354 if ($firewall) {
355 &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks);
356 } else {
357 &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks);
358 }
359 }
360
361 tap_rate_limit($iface, $rate);
362 }
363
364 sub tap_unplug {
365 my ($iface) = @_;
366
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+)/;
372
373 system("/sbin/brctl delif $bridge $iface") == 0 ||
374 die "can't del interface '$iface' from bridge '$bridge'\n";
375
376 }
377
378 &$cleanup_firewall_bridge($iface);
379 #cleanup old port config from any openvswitch bridge
380 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
381 }
382
383 sub copy_bridge_config {
384 my ($br0, $br1) = @_;
385
386 return if $br0 eq $br1;
387
388 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
389 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
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 $v1) {
396 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
397 }
398 };
399 warn $@ if $@;
400 }
401 }
402
403 sub activate_bridge_vlan_slave {
404 my ($bridgevlan, $iface, $tag) = @_;
405 my $ifacevlan = "${iface}.$tag";
406
407 # create vlan on $iface is not already exist
408 if (! -d "/sys/class/net/$ifacevlan") {
409 system("/sbin/ip link add link $iface name ${iface}.${tag} type vlan id $tag") == 0 ||
410 die "can't add vlan tag $tag to interface $iface\n";
411 }
412
413 # be sure to have the $ifacevlan up
414 &$activate_interface($ifacevlan);
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));
420 if ($tbridge ne $bridgevlan) {
421 die "interface $ifacevlan already exist in bridge $tbridge\n";
422 } else {
423 # Port already attached to bridge: do nothing.
424 return;
425 }
426 }
427
428 # add $ifacevlan to the bridge
429 &$bridge_add_interface($bridgevlan, $ifacevlan);
430 }
431
432 sub activate_bridge_vlan {
433 my ($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
445 my @ifaces = ();
446 my $dir = "/sys/class/net/$bridge/brif";
447 PVE::Tools::dir_glob_foreach($dir, '(((eth|bond)\d+|en[^.]+)(\.\d+)?)', sub {
448 push @ifaces, $_[0];
449 });
450
451 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
452
453 # add bridgevlan if it doesn't already exist
454 if (! -d "/sys/class/net/$bridgevlan") {
455 system("/sbin/brctl addbr $bridgevlan") == 0 ||
456 die "can't add bridge $bridgevlan\n";
457 }
458
459 # for each physical interface (eth or bridge) bind them to bridge vlan
460 foreach my $iface (@ifaces) {
461 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
462 }
463
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";
469
470 return $bridgevlan;
471 }
472
473 sub tcp_ping {
474 my ($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 $result;
499 }
500
501 sub IP_from_cidr {
502 my ($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/$prefix");
518 }
519
520 sub is_ip_in_cidr {
521 my ($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
532 # struct ifreq { // FOR SIOCGIFFLAGS:
533 # char ifrn_name[IFNAMSIZ]
534 # short ifru_flags
535 # };
536 my $STRUCT_IFREQ_SIOCGIFFLAGS = 'Z' . IFNAMSIZ . 's1';
537 sub 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:
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 [];
546
547 my $ifaces = [];
548 while(defined(my $line = <$fh>)) {
549 next if $line !~ /^\s*([^:\s]+):/;
550 my $ifname = $1;
551 my $ifreq = pack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifname, 0);
552 if (!defined(ioctl($sock, SIOCGIFFLAGS, $ifreq))) {
553 warn "failed to get interface flags for: $ifname\n";
554 next;
555 }
556 my ($name, $flags) = unpack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifreq);
557 push @$ifaces, $ifname if ($flags & IFF_UP);
558 }
559 close $fh;
560 close $sock;
561 return $ifaces;
562 }
563
564 1;