]> git.proxmox.com Git - pve-common.git/blob - src/PVE/Network.pm
ce4305a3390454f71d3141f14146965079c1a0c2
[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 $ipv4_mask_hash_localnet = {
59 '255.0.0.0' => 8,
60 '255.128.0.0' => 9,
61 '255.192.0.0' => 10,
62 '255.224.0.0' => 11,
63 '255.240.0.0' => 12,
64 '255.248.0.0' => 13,
65 '255.252.0.0' => 14,
66 '255.254.0.0' => 15,
67 '255.255.0.0' => 16,
68 '255.255.128.0' => 17,
69 '255.255.192.0' => 18,
70 '255.255.224.0' => 19,
71 '255.255.240.0' => 20,
72 '255.255.248.0' => 21,
73 '255.255.252.0' => 22,
74 '255.255.254.0' => 23,
75 '255.255.255.0' => 24,
76 '255.255.255.128' => 25,
77 '255.255.255.192' => 26,
78 '255.255.255.224' => 27,
79 '255.255.255.240' => 28,
80 '255.255.255.248' => 29,
81 '255.255.255.252' => 30,
82 '255.255.255.254' => 31,
83 '255.255.255.255' => 32,
84 };
85
86 sub setup_tc_rate_limit {
87 my ($iface, $rate, $burst, $debug) = @_;
88
89 # these are allowed / expected to fail, e.g. when there is no previous rate limit to remove
90 eval { run_command("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1"); };
91 eval { run_command("/sbin/tc filter del dev $iface parent ffff: protocol all pref 50 u32 >/dev/null 2>&1"); };
92 eval { run_command("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1"); };
93 eval { run_command("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1"); };
94
95 return if !$rate;
96
97 # tbf does not work for unknown reason
98 #$TC qdisc add dev $DEV root tbf rate $RATE latency 100ms burst $BURST
99 # so we use htb instead
100 run_command("/sbin/tc qdisc add dev $iface root handle 1: htb default 1");
101 run_command("/sbin/tc class add dev $iface parent 1: classid 1:1 " .
102 "htb rate ${rate}bps burst ${burst}b");
103
104 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
105 run_command("/sbin/tc filter add dev $iface parent ffff: " .
106 "prio 50 basic " .
107 "police rate ${rate}bps burst ${burst}b mtu 64kb " .
108 "drop flowid :1");
109
110 if ($debug) {
111 print "DEBUG tc settings\n";
112 system("/sbin/tc qdisc ls dev $iface");
113 system("/sbin/tc class ls dev $iface");
114 system("/sbin/tc filter ls dev $iface parent ffff:");
115 }
116 }
117
118 sub tap_rate_limit {
119 my ($iface, $rate) = @_;
120
121 my $debug = 0;
122 $rate = int($rate*1024*1024) if $rate;
123 my $burst = 1024*1024;
124
125 setup_tc_rate_limit($iface, $rate, $burst, $debug);
126 }
127
128 my $read_bridge_mtu = sub {
129 my ($bridge) = @_;
130
131 my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu");
132 die "bridge '$bridge' does not exist\n" if !$mtu;
133 # avoid insecure dependency;
134 die "unable to parse mtu value" if $mtu !~ /^(\d+)$/;
135 $mtu = int($1);
136
137 return $mtu;
138 };
139
140 my $parse_tap_device_name = sub {
141 my ($iface, $noerr) = @_;
142
143 my ($vmid, $devid);
144
145 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
146 $vmid = $1;
147 $devid = $2;
148 } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) {
149 $vmid = $1;
150 $devid = $2;
151 } else {
152 return undef if $noerr;
153 die "can't create firewall bridge for random interface name '$iface'\n";
154 }
155
156 return ($vmid, $devid);
157 };
158
159 my $compute_fwbr_names = sub {
160 my ($vmid, $devid) = @_;
161
162 my $fwbr = "fwbr${vmid}i${devid}";
163 # Note: the firewall use 'fwln+' to filter traffic to VMs
164 my $vethfw = "fwln${vmid}i${devid}";
165 my $vethfwpeer = "fwpr${vmid}p${devid}";
166 my $ovsintport = "fwln${vmid}o${devid}";
167
168 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
169 };
170
171 my $cond_create_bridge = sub {
172 my ($bridge) = @_;
173
174 if (! -d "/sys/class/net/$bridge") {
175 system("/sbin/brctl addbr $bridge") == 0 ||
176 die "can't add bridge '$bridge'\n";
177 }
178 };
179
180 my $bridge_add_interface = sub {
181 my ($bridge, $iface, $tag, $trunks) = @_;
182
183 system("/sbin/brctl addif $bridge $iface") == 0 ||
184 die "can't add interface 'iface' to bridge '$bridge'\n";
185
186 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
187
188 if ($vlan_aware) {
189 if ($tag) {
190 system("/sbin/bridge vlan add dev $iface vid $tag pvid untagged") == 0 ||
191 die "unable to add vlan $tag to interface $iface\n";
192
193 warn "Caution: Setting VLAN ID 1 on a VLAN aware bridge may be dangerous\n" if $tag == 1;
194 } else {
195 system("/sbin/bridge vlan add dev $iface vid 2-4094") == 0 ||
196 die "unable to add default vlan tags to interface $iface\n" if !$trunks;
197 }
198
199 if ($trunks) {
200 my @trunks_array = split /;/, $trunks;
201 foreach my $trunk (@trunks_array) {
202 system("/sbin/bridge vlan add dev $iface vid $trunk") == 0 ||
203 die "unable to add vlan $trunk to interface $iface\n";
204 }
205 }
206 }
207 };
208
209 my $ovs_bridge_add_port = sub {
210 my ($bridge, $iface, $tag, $internal, $trunks) = @_;
211
212 $trunks =~ s/;/,/g if $trunks;
213
214 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
215 $cmd .= " tag=$tag" if $tag;
216 $cmd .= " trunks=". join(',', $trunks) if $trunks;
217 $cmd .= " vlan_mode=native-untagged" if $tag && $trunks;
218
219 $cmd .= " -- set Interface $iface type=internal" if $internal;
220 system($cmd) == 0 ||
221 die "can't add ovs port '$iface'\n";
222 };
223
224 my $activate_interface = sub {
225 my ($iface) = @_;
226
227 system("/sbin/ip link set $iface up") == 0 ||
228 die "can't activate interface '$iface'\n";
229 };
230
231 sub tap_create {
232 my ($iface, $bridge) = @_;
233
234 die "unable to get bridge setting\n" if !$bridge;
235
236 my $bridgemtu = &$read_bridge_mtu($bridge);
237
238 eval {
239 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
240 };
241 die "interface activation failed\n" if $@;
242 }
243
244 sub veth_create {
245 my ($veth, $vethpeer, $bridge, $mac) = @_;
246
247 die "unable to get bridge setting\n" if !$bridge;
248
249 my $bridgemtu = &$read_bridge_mtu($bridge);
250
251 # create veth pair
252 if (! -d "/sys/class/net/$veth") {
253 my $cmd = "/sbin/ip link add name $veth type veth peer name $vethpeer mtu $bridgemtu";
254 $cmd .= " addr $mac" if $mac;
255 system($cmd) == 0 || die "can't create interface $veth\n";
256 }
257
258 # up vethpair
259 &$activate_interface($veth);
260 &$activate_interface($vethpeer);
261 }
262
263 sub veth_delete {
264 my ($veth) = @_;
265
266 if (-d "/sys/class/net/$veth") {
267 run_command("/sbin/ip link delete dev $veth", outfunc => sub {}, errfunc => sub {});
268 }
269
270 }
271
272 my $create_firewall_bridge_linux = sub {
273 my ($iface, $bridge, $tag, $trunks) = @_;
274
275 my ($vmid, $devid) = &$parse_tap_device_name($iface);
276 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
277
278 &$cond_create_bridge($fwbr);
279 &$activate_interface($fwbr);
280
281 copy_bridge_config($bridge, $fwbr);
282 veth_create($vethfw, $vethfwpeer, $bridge);
283
284 &$bridge_add_interface($fwbr, $vethfw);
285 &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks);
286
287 &$bridge_add_interface($fwbr, $iface);
288 };
289
290 my $create_firewall_bridge_ovs = sub {
291 my ($iface, $bridge, $tag, $trunks) = @_;
292
293 my ($vmid, $devid) = &$parse_tap_device_name($iface);
294 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
295
296 my $bridgemtu = &$read_bridge_mtu($bridge);
297
298 &$cond_create_bridge($fwbr);
299 &$activate_interface($fwbr);
300
301 &$bridge_add_interface($fwbr, $iface);
302
303 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks);
304 &$activate_interface($ovsintport);
305
306 # set the same mtu for ovs int port
307 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
308
309 &$bridge_add_interface($fwbr, $ovsintport);
310 };
311
312 my $cleanup_firewall_bridge = sub {
313 my ($iface) = @_;
314
315 my ($vmid, $devid) = &$parse_tap_device_name($iface, 1);
316 return if !defined($vmid);
317 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
318
319 # cleanup old port config from any openvswitch bridge
320 if (-d "/sys/class/net/$ovsintport") {
321 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
322 }
323
324 # delete old vethfw interface
325 veth_delete($vethfw);
326
327 # cleanup fwbr bridge
328 if (-d "/sys/class/net/$fwbr") {
329 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
330 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
331 }
332 };
333
334 sub tap_plug {
335 my ($iface, $bridge, $tag, $firewall, $trunks, $rate) = @_;
336
337 #cleanup old port config from any openvswitch bridge
338 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
339
340 if (-d "/sys/class/net/$bridge/bridge") {
341 &$cleanup_firewall_bridge($iface); # remove stale devices
342
343 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
344
345 if (!$vlan_aware) {
346 die "vlan aware feature need to be enabled to use trunks" if $trunks;
347 my $newbridge = activate_bridge_vlan($bridge, $tag);
348 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
349 $bridge = $newbridge;
350 $tag = undef;
351 }
352
353 if ($firewall) {
354 &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks);
355 } else {
356 &$bridge_add_interface($bridge, $iface, $tag, $trunks);
357 }
358
359 } else {
360 &$cleanup_firewall_bridge($iface); # remove stale devices
361
362 if ($firewall) {
363 &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks);
364 } else {
365 &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks);
366 }
367 }
368
369 tap_rate_limit($iface, $rate);
370 }
371
372 sub tap_unplug {
373 my ($iface) = @_;
374
375 my $path= "/sys/class/net/$iface/brport/bridge";
376 if (-l $path) {
377 my $bridge = basename(readlink($path));
378 #avoid insecure dependency;
379 ($bridge) = $bridge =~ /(\S+)/;
380
381 system("/sbin/brctl delif $bridge $iface") == 0 ||
382 die "can't del interface '$iface' from bridge '$bridge'\n";
383
384 }
385
386 &$cleanup_firewall_bridge($iface);
387 #cleanup old port config from any openvswitch bridge
388 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
389 }
390
391 sub copy_bridge_config {
392 my ($br0, $br1) = @_;
393
394 return if $br0 eq $br1;
395
396 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
397 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
398
399 foreach my $sysname (@$br_configs) {
400 eval {
401 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
402 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
403 if ($v0 ne $v1) {
404 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
405 }
406 };
407 warn $@ if $@;
408 }
409 }
410
411 sub activate_bridge_vlan_slave {
412 my ($bridgevlan, $iface, $tag) = @_;
413 my $ifacevlan = "${iface}.$tag";
414
415 # create vlan on $iface is not already exist
416 if (! -d "/sys/class/net/$ifacevlan") {
417 system("/sbin/ip link add link $iface name ${iface}.${tag} type vlan id $tag") == 0 ||
418 die "can't add vlan tag $tag to interface $iface\n";
419 }
420
421 # be sure to have the $ifacevlan up
422 &$activate_interface($ifacevlan);
423
424 # test if $vlaniface is already enslaved in another bridge
425 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
426 if (-l $path) {
427 my $tbridge = basename(readlink($path));
428 if ($tbridge ne $bridgevlan) {
429 die "interface $ifacevlan already exist in bridge $tbridge\n";
430 } else {
431 # Port already attached to bridge: do nothing.
432 return;
433 }
434 }
435
436 # add $ifacevlan to the bridge
437 &$bridge_add_interface($bridgevlan, $ifacevlan);
438 }
439
440 sub activate_bridge_vlan {
441 my ($bridge, $tag_param) = @_;
442
443 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
444
445 return $bridge if !defined($tag_param); # no vlan, simply return
446
447 my $tag = int($tag_param);
448
449 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
450
451 my $bridgevlan = "${bridge}v$tag";
452
453 my @ifaces = ();
454 my $dir = "/sys/class/net/$bridge/brif";
455 PVE::Tools::dir_glob_foreach($dir, '(((eth|bond)\d+|en[^.]+)(\.\d+)?)', sub {
456 push @ifaces, $_[0];
457 });
458
459 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
460
461 # add bridgevlan if it doesn't already exist
462 if (! -d "/sys/class/net/$bridgevlan") {
463 system("/sbin/brctl addbr $bridgevlan") == 0 ||
464 die "can't add bridge $bridgevlan\n";
465 }
466
467 # for each physical interface (eth or bridge) bind them to bridge vlan
468 foreach my $iface (@ifaces) {
469 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
470 }
471
472 #fixme: set other bridge flags
473
474 # be sure to have the bridge up
475 system("/sbin/ip link set $bridgevlan up") == 0 ||
476 die "can't up bridge $bridgevlan\n";
477
478 return $bridgevlan;
479 }
480
481 sub tcp_ping {
482 my ($host, $port, $timeout) = @_;
483
484 my $refused = 1;
485
486 $timeout = 3 if !$timeout; # sane default
487 if (!$port) {
488 # Net::Ping defaults to the echo port
489 $port = 7;
490 } else {
491 # Net::Ping's port_number() implies service_check(1)
492 $refused = 0;
493 }
494
495 my ($sock, $result);
496 eval {
497 $result = PVE::Tools::run_with_timeout($timeout, sub {
498 $sock = IO::Socket::IP->new(PeerHost => $host, PeerPort => $port, Type => SOCK_STREAM);
499 $result = $refused if $! == ECONNREFUSED;
500 });
501 };
502 if ($sock) {
503 $sock->close();
504 $result = 1;
505 }
506 return $result;
507 }
508
509 sub IP_from_cidr {
510 my ($cidr, $version) = @_;
511
512 return if $cidr !~ m!^(\S+?)/(\S+)$!;
513 my ($ip, $prefix) = ($1, $2);
514
515 my $ipobj = Net::IP->new($ip, $version);
516 return if !$ipobj;
517
518 $version = $ipobj->version();
519
520 my $binmask = Net::IP::ip_get_mask($prefix, $version);
521 return if !$binmask;
522
523 my $masked_binip = $ipobj->binip() & $binmask;
524 my $masked_ip = Net::IP::ip_bintoip($masked_binip, $version);
525 return Net::IP->new("$masked_ip/$prefix");
526 }
527
528 sub is_ip_in_cidr {
529 my ($ip, $cidr, $version) = @_;
530
531 my $cidr_obj = IP_from_cidr($cidr, $version);
532 return undef if !$cidr_obj;
533
534 my $ip_obj = Net::IP->new($ip, $version);
535 return undef if !$ip_obj;
536
537 return $cidr_obj->overlaps($ip_obj) == $Net::IP::IP_B_IN_A_OVERLAP;
538 }
539
540 # struct ifreq { // FOR SIOCGIFFLAGS:
541 # char ifrn_name[IFNAMSIZ]
542 # short ifru_flags
543 # };
544 my $STRUCT_IFREQ_SIOCGIFFLAGS = 'Z' . IFNAMSIZ . 's1';
545 sub get_active_interfaces {
546 # Use the interface name list from /proc/net/dev
547 open my $fh, '<', '/proc/net/dev'
548 or die "failed to open /proc/net/dev: $!\n";
549 # And filter by IFF_UP flag fetched via a PF_INET6 socket ioctl:
550 my $sock;
551 socket($sock, PF_INET6, SOCK_DGRAM, &IPPROTO_IP)
552 or socket($sock, PF_INET, SOCK_DGRAM, &IPPROTO_IP)
553 or return [];
554
555 my $ifaces = [];
556 while(defined(my $line = <$fh>)) {
557 next if $line !~ /^\s*([^:\s]+):/;
558 my $ifname = $1;
559 my $ifreq = pack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifname, 0);
560 if (!defined(ioctl($sock, SIOCGIFFLAGS, $ifreq))) {
561 warn "failed to get interface flags for: $ifname\n";
562 next;
563 }
564 my ($name, $flags) = unpack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifreq);
565 push @$ifaces, $ifname if ($flags & IFF_UP);
566 }
567 close $fh;
568 close $sock;
569 return $ifaces;
570 }
571
572 1;