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