]> git.proxmox.com Git - pve-common.git/blame - src/PVE/Network.pm
net: tap_plug() needs to apply rate limiting
[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 $ipv4_mask_hash_localnet = {
59 '255.255.0.0' => 16,
60 '255.255.128.0' => 17,
61 '255.255.192.0' => 18,
62 '255.255.224.0' => 19,
63 '255.255.240.0' => 20,
64 '255.255.248.0' => 21,
65 '255.255.252.0' => 22,
66 '255.255.254.0' => 23,
67 '255.255.255.0' => 24,
68 '255.255.255.128' => 25,
69 '255.255.255.192' => 26,
70 '255.255.255.224' => 27,
71 '255.255.255.240' => 28,
72 '255.255.255.248' => 29,
73 '255.255.255.252' => 30,
74};
75
74d1b045
DM
76sub setup_tc_rate_limit {
77 my ($iface, $rate, $burst, $debug) = @_;
78
957753df 79 system("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1");
5d35df41 80 system("/sbin/tc filter del dev $iface parent ffff: protocol all pref 50 u32 >/dev/null 2>&1");
edde1d46 81 system("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1");
74d1b045
DM
82 system("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1");
83
d6f2623b 84 return if !$rate;
957753df 85
74d1b045
DM
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
5d35df41
W
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 ${burst}b mtu 64kb " .
97 "drop flowid :1");
98
74d1b045
DM
99 if ($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 $iface parent ffff:");
104 }
105}
106
ec9ada18
AD
107sub tap_rate_limit {
108 my ($iface, $rate) = @_;
109
110 my $debug = 0;
ad066ae2 111 $rate = int($rate*1024*1024) if $rate;
ec9ada18
AD
112 my $burst = 1024*1024;
113
114 setup_tc_rate_limit($iface, $rate, $burst, $debug);
115}
74d1b045 116
605bb891
DM
117my $read_bridge_mtu = sub {
118 my ($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
32cb7d27 129my $parse_tap_device_name = sub {
6c80e6d6 130 my ($iface, $noerr) = @_;
605bb891
DM
131
132 my ($vmid, $devid);
133
134 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
135 $vmid = $1;
136 $devid = $2;
32cb7d27 137 } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) {
605bb891
DM
138 $vmid = $1;
139 $devid = $2;
140 } else {
6c80e6d6
DM
141 return undef if $noerr;
142 die "can't create firewall bridge for random interface name '$iface'\n";
605bb891
DM
143 }
144
145 return ($vmid, $devid);
146};
147
70ab4434 148my $compute_fwbr_names = sub {
605bb891
DM
149 my ($vmid, $devid) = @_;
150
151 my $fwbr = "fwbr${vmid}i${devid}";
f193aa74 152 # Note: the firewall use 'fwln+' to filter traffic to VMs
7d78a966
AD
153 my $vethfw = "fwln${vmid}i${devid}";
154 my $vethfwpeer = "fwpr${vmid}p${devid}";
155 my $ovsintport = "fwln${vmid}o${devid}";
605bb891 156
70ab4434 157 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
605bb891
DM
158};
159
160my $cond_create_bridge = sub {
161 my ($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
169my $bridge_add_interface = sub {
b0b34ffd 170 my ($bridge, $iface, $tag, $trunks) = @_;
605bb891
DM
171
172 system("/sbin/brctl addif $bridge $iface") == 0 ||
173 die "can't add interface 'iface' to bridge '$bridge'\n";
4d25f4aa
AD
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 ||
b0b34ffd 183 die "unable to add default vlan tags to interface $iface\n" if !$trunks;
4d25f4aa 184 }
b0b34ffd 185
846337ad
WB
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 }
b0b34ffd 192 }
4d25f4aa 193 }
605bb891
DM
194};
195
70ab4434 196my $ovs_bridge_add_port = sub {
b0b34ffd
AD
197 my ($bridge, $iface, $tag, $internal, $trunks) = @_;
198
199 $trunks =~ s/;/,/g if $trunks;
70ab4434
DM
200
201 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
202 $cmd .= " tag=$tag" if $tag;
b0b34ffd
AD
203 $cmd .= " trunks=". join(',', $trunks) if $trunks;
204 $cmd .= " vlan_mode=native-untagged" if $tag && $trunks;
205
70ab4434
DM
206 $cmd .= " -- set Interface $iface type=internal" if $internal;
207 system($cmd) == 0 ||
208 die "can't add ovs port '$iface'\n";
209};
210
605bb891
DM
211my $activate_interface = sub {
212 my ($iface) = @_;
213
214 system("/sbin/ip link set $iface up") == 0 ||
215 die "can't activate interface '$iface'\n";
216};
217
3aa99c70
AD
218sub tap_create {
219 my ($iface, $bridge) = @_;
220
221 die "unable to get bridge setting\n" if !$bridge;
222
605bb891 223 my $bridgemtu = &$read_bridge_mtu($bridge);
3aa99c70 224
098795e0
DM
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 $@;
3aa99c70
AD
229}
230
35efc4eb
AD
231sub veth_create {
232 my ($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($vethpeer);
248}
249
f3f0bc3a
AD
250sub veth_delete {
251 my ($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}
35efc4eb 258
605bb891 259my $create_firewall_bridge_linux = sub {
b0b34ffd 260 my ($iface, $bridge, $tag, $trunks) = @_;
605bb891 261
32cb7d27 262 my ($vmid, $devid) = &$parse_tap_device_name($iface);
70ab4434 263 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
605bb891 264
605bb891
DM
265 &$cond_create_bridge($fwbr);
266 &$activate_interface($fwbr);
267
268 copy_bridge_config($bridge, $fwbr);
35efc4eb 269 veth_create($vethfw, $vethfwpeer, $bridge);
605bb891 270
7d78a966 271 &$bridge_add_interface($fwbr, $vethfw);
b0b34ffd 272 &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks);
605bb891 273
4d25f4aa 274 &$bridge_add_interface($fwbr, $iface);
605bb891
DM
275};
276
70ab4434 277my $create_firewall_bridge_ovs = sub {
b0b34ffd 278 my ($iface, $bridge, $tag, $trunks) = @_;
70ab4434 279
32cb7d27 280 my ($vmid, $devid) = &$parse_tap_device_name($iface);
70ab4434
DM
281 my ($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, $iface);
289
b0b34ffd 290 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks);
ac3a04b8 291 &$activate_interface($ovsintport);
70ab4434
DM
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
299my $cleanup_firewall_bridge = sub {
605bb891
DM
300 my ($iface) = @_;
301
32cb7d27 302 my ($vmid, $devid) = &$parse_tap_device_name($iface, 1);
6c80e6d6 303 return if !defined($vmid);
70ab4434
DM
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 $ovsintport", outfunc => sub {}, errfunc => sub {});
309 }
605bb891
DM
310
311 # delete old vethfw interface
f3f0bc3a 312 veth_delete($vethfw);
605bb891
DM
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 $fwbr", outfunc => sub {}, errfunc => sub {});
318 }
319};
320
f0c190ee 321sub tap_plug {
bce2a5b3 322 my ($iface, $bridge, $tag, $firewall, $trunks, $rate) = @_;
f0c190ee 323
4cbabd40
AD
324 #cleanup old port config from any openvswitch bridge
325 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
326
098795e0 327 if (-d "/sys/class/net/$bridge/bridge") {
70ab4434 328 &$cleanup_firewall_bridge($iface); # remove stale devices
605bb891 329
4d25f4aa 330 my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering");
098795e0 331
4d25f4aa 332 if (!$vlan_aware) {
b0b34ffd 333 die "vlan aware feature need to be enabled to use trunks" if $trunks;
4d25f4aa
AD
334 my $newbridge = activate_bridge_vlan($bridge, $tag);
335 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
ff042056 336 $bridge = $newbridge;
4d25f4aa
AD
337 $tag = undef;
338 }
339
340 if ($firewall) {
b0b34ffd 341 &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks);
4d25f4aa 342 } else {
b0b34ffd 343 &$bridge_add_interface($bridge, $iface, $tag, $trunks);
4d25f4aa 344 }
605bb891 345
098795e0 346 } else {
70ab4434
DM
347 &$cleanup_firewall_bridge($iface); # remove stale devices
348
349 if ($firewall) {
b0b34ffd 350 &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks);
70ab4434 351 } else {
b0b34ffd 352 &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks);
70ab4434 353 }
4cbabd40 354 }
bce2a5b3
WB
355
356 tap_rate_limit($iface, $rate);
f0c190ee
AD
357}
358
a84b65c0 359sub tap_unplug {
2db1cc0d 360 my ($iface) = @_;
a84b65c0 361
2db1cc0d
DM
362 my $path= "/sys/class/net/$iface/brport/bridge";
363 if (-l $path) {
364 my $bridge = basename(readlink($path));
365 #avoid insecure dependency;
366 ($bridge) = $bridge =~ /(\S+)/;
4cbabd40 367
098795e0 368 system("/sbin/brctl delif $bridge $iface") == 0 ||
2db1cc0d 369 die "can't del interface '$iface' from bridge '$bridge'\n";
605bb891 370
4cbabd40 371 }
70ab4434
DM
372
373 &$cleanup_firewall_bridge($iface);
a84b65c0
AD
374}
375
b9436cda
DM
376sub copy_bridge_config {
377 my ($br0, $br1) = @_;
378
379 return if $br0 eq $br1;
380
381 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
ba4af65b 382 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
b9436cda
DM
383
384 foreach my $sysname (@$br_configs) {
385 eval {
386 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
387 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
388 if ($v0 ne $v1) {
aec04803 389 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
b9436cda
DM
390 }
391 };
392 warn $@ if $@;
393 }
394}
395
70d89745
PRG
396sub activate_bridge_vlan_slave {
397 my ($bridgevlan, $iface, $tag) = @_;
b9436cda 398 my $ifacevlan = "${iface}.$tag";
70d89745 399
b9436cda
DM
400 # create vlan on $iface is not already exist
401 if (! -d "/sys/class/net/$ifacevlan") {
6fc54cb2 402 system("/sbin/ip link add link $iface name ${iface}.${tag} type vlan id $tag") == 0 ||
02c9a6b4 403 die "can't add vlan tag $tag to interface $iface\n";
b9436cda
DM
404 }
405
406 # be sure to have the $ifacevlan up
605bb891 407 &$activate_interface($ifacevlan);
b9436cda
DM
408
409 # test if $vlaniface is already enslaved in another bridge
410 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
411 if (-l $path) {
412 my $tbridge = basename(readlink($path));
70d89745 413 if ($tbridge ne $bridgevlan) {
b9436cda 414 die "interface $ifacevlan already exist in bridge $tbridge\n";
eee4b32a
PRG
415 } else {
416 # Port already attached to bridge: do nothing.
417 return;
b9436cda
DM
418 }
419 }
420
70d89745 421 # add $ifacevlan to the bridge
605bb891 422 &$bridge_add_interface($bridgevlan, $ifacevlan);
70d89745
PRG
423}
424
425sub activate_bridge_vlan {
426 my ($bridge, $tag_param) = @_;
427
428 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
429
430 return $bridge if !defined($tag_param); # no vlan, simply return
431
432 my $tag = int($tag_param);
433
434 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
435
436 my $bridgevlan = "${bridge}v$tag";
437
c9030d97
PRG
438 my @ifaces = ();
439 my $dir = "/sys/class/net/$bridge/brif";
ce1c3cc4 440 PVE::Tools::dir_glob_foreach($dir, '((eth|bond)\d+(\.\d+)?)', sub {
5ffa7628 441 push @ifaces, $_[0];
c9030d97
PRG
442 });
443
5ffa7628 444 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
c9030d97 445
b9436cda
DM
446 # add bridgevlan if it doesn't already exist
447 if (! -d "/sys/class/net/$bridgevlan") {
9e14b1b7 448 system("/sbin/brctl addbr $bridgevlan") == 0 ||
b9436cda
DM
449 die "can't add bridge $bridgevlan\n";
450 }
451
70d89745 452 # for each physical interface (eth or bridge) bind them to bridge vlan
c9030d97
PRG
453 foreach my $iface (@ifaces) {
454 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
455 }
70d89745 456
b9436cda
DM
457 #fixme: set other bridge flags
458
459 # be sure to have the bridge up
460 system("/sbin/ip link set $bridgevlan up") == 0 ||
461 die "can't up bridge $bridgevlan\n";
70d89745 462
b9436cda
DM
463 return $bridgevlan;
464}
465
b6bff92e
WB
466sub tcp_ping {
467 my ($host, $port, $timeout) = @_;
468
469 my $refused = 1;
470
471 $timeout = 3 if !$timeout; # sane default
472 if (!$port) {
473 # Net::Ping defaults to the echo port
474 $port = 7;
475 } else {
476 # Net::Ping's port_number() implies service_check(1)
477 $refused = 0;
478 }
479
480 my ($sock, $result);
481 eval {
482 $result = PVE::Tools::run_with_timeout($timeout, sub {
483 $sock = IO::Socket::IP->new(PeerHost => $host, PeerPort => $port, Type => SOCK_STREAM);
484 $result = $refused if $! == ECONNREFUSED;
485 });
486 };
487 if ($sock) {
488 $sock->close();
489 $result = 1;
490 }
491 return $result;
492}
493
bf52d27b
WB
494sub IP_from_cidr {
495 my ($cidr, $version) = @_;
496
497 return if $cidr !~ m!^(\S+?)/(\S+)$!;
498 my ($ip, $prefix) = ($1, $2);
499
500 my $ipobj = Net::IP->new($ip, $version);
501 return if !$ipobj;
502
503 $version = $ipobj->version();
504
505 my $binmask = Net::IP::ip_get_mask($prefix, $version);
506 return if !$binmask;
507
508 my $masked_binip = $ipobj->binip() & $binmask;
509 my $masked_ip = Net::IP::ip_bintoip($masked_binip, $version);
510 return Net::IP->new("$masked_ip/$prefix");
511}
512
513sub is_ip_in_cidr {
514 my ($ip, $cidr, $version) = @_;
515
516 my $cidr_obj = IP_from_cidr($cidr, $version);
517 return undef if !$cidr_obj;
518
519 my $ip_obj = Net::IP->new($ip, $version);
520 return undef if !$ip_obj;
521
522 return $cidr_obj->overlaps($ip_obj) == $Net::IP::IP_B_IN_A_OVERLAP;
523}
524
12a235d6
WB
525# struct ifreq { // FOR SIOCGIFFLAGS:
526# char ifrn_name[IFNAMSIZ]
527# short ifru_flags
528# };
529my $STRUCT_IFREQ_SIOCGIFFLAGS = 'Z' . IFNAMSIZ . 's1';
530sub get_active_interfaces {
531 # Use the interface name list from /proc/net/dev
532 open my $fh, '<', '/proc/net/dev'
533 or die "failed to open /proc/net/dev: $!\n";
534 # And filter by IFF_UP flag fetched via a PF_INET6 socket ioctl:
535 socket my $sock, PF_INET6, SOCK_DGRAM, &IPPROTO_IP
536 or die "failed to open socket\n";
537
538 my $ifaces = [];
539 while(defined(my $line = <$fh>)) {
540 next if $line !~ /^\s*([^:\s]+):/;
541 my $ifname = $1;
542 my $ifreq = pack($STRUCT_IFREQ_SIOCGIFFLAGS, $1, 0);
f029c1d0 543 if (!defined(ioctl($sock, SIOCGIFFLAGS, $ifreq))) {
12a235d6
WB
544 warn "failed to get interface flags for: $ifname\n";
545 next;
546 }
547 my ($name, $flags) = unpack($STRUCT_IFREQ_SIOCGIFFLAGS, $ifreq);
548 push @$ifaces, $1 if ($flags & IFF_UP);
549 }
550 close $fh;
551 close $sock;
552 return $ifaces;
553}
554
b9436cda 5551;