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