]> git.proxmox.com Git - pve-common.git/blob - src/PVE/Network.pm
PVE::Network: add some common constants
[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
10 # host network related utility functions
11
12 our $ipv4_reverse_mask = [
13 '0.0.0.0',
14 '128.0.0.0',
15 '192.0.0.0',
16 '224.0.0.0',
17 '240.0.0.0',
18 '248.0.0.0',
19 '252.0.0.0',
20 '254.0.0.0',
21 '255.0.0.0',
22 '255.128.0.0',
23 '255.192.0.0',
24 '255.224.0.0',
25 '255.240.0.0',
26 '255.248.0.0',
27 '255.252.0.0',
28 '255.254.0.0',
29 '255.255.0.0',
30 '255.255.128.0',
31 '255.255.192.0',
32 '255.255.224.0',
33 '255.255.240.0',
34 '255.255.248.0',
35 '255.255.252.0',
36 '255.255.254.0',
37 '255.255.255.0',
38 '255.255.255.128',
39 '255.255.255.192',
40 '255.255.255.224',
41 '255.255.255.240',
42 '255.255.255.248',
43 '255.255.255.252',
44 '255.255.255.254',
45 '255.255.255.255',
46 ];
47
48 our $ipv4_mask_hash_localnet = {
49 '255.255.0.0' => 16,
50 '255.255.128.0' => 17,
51 '255.255.192.0' => 18,
52 '255.255.224.0' => 19,
53 '255.255.240.0' => 20,
54 '255.255.248.0' => 21,
55 '255.255.252.0' => 22,
56 '255.255.254.0' => 23,
57 '255.255.255.0' => 24,
58 '255.255.255.128' => 25,
59 '255.255.255.192' => 26,
60 '255.255.255.224' => 27,
61 '255.255.255.240' => 28,
62 '255.255.255.248' => 29,
63 '255.255.255.252' => 30,
64 };
65
66 sub setup_tc_rate_limit {
67 my ($iface, $rate, $burst, $debug) = @_;
68
69 system("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1");
70 system("/sbin/tc filter del dev $iface parent ffff: protocol ip prio 50 estimator 1sec 8sec >/dev/null 2>&1");
71 system("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1");
72 system("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1");
73
74 return if !$rate;
75
76 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
77
78 # this does not work wit virtio - don't know why (setting "mtu 64kb" does not help)
79 #run_command("/sbin/tc filter add dev $iface parent ffff: protocol ip prio 50 u32 match ip src 0.0.0.0/0 police rate ${rate}bps burst ${burst}b drop flowid :1");
80 # so we use avrate instead
81 run_command("/sbin/tc filter add dev $iface parent ffff: " .
82 "protocol ip prio 50 estimator 1sec 8sec " .
83 "u32 match ip src 0.0.0.0/0 police avrate ${rate}bps drop flowid :1");
84
85 # tbf does not work for unknown reason
86 #$TC qdisc add dev $DEV root tbf rate $RATE latency 100ms burst $BURST
87 # so we use htb instead
88 run_command("/sbin/tc qdisc add dev $iface root handle 1: htb default 1");
89 run_command("/sbin/tc class add dev $iface parent 1: classid 1:1 " .
90 "htb rate ${rate}bps burst ${burst}b");
91
92 if ($debug) {
93 print "DEBUG tc settings\n";
94 system("/sbin/tc qdisc ls dev $iface");
95 system("/sbin/tc class ls dev $iface");
96 system("/sbin/tc filter ls dev $iface parent ffff:");
97 }
98 }
99
100 sub tap_rate_limit {
101 my ($iface, $rate) = @_;
102
103 my $debug = 0;
104 $rate = int($rate*1024*1024);
105 my $burst = 1024*1024;
106
107 setup_tc_rate_limit($iface, $rate, $burst, $debug);
108 }
109
110 my $read_bridge_mtu = sub {
111 my ($bridge) = @_;
112
113 my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu");
114 die "bridge '$bridge' does not exist\n" if !$mtu;
115 # avoid insecure dependency;
116 die "unable to parse mtu value" if $mtu !~ /^(\d+)$/;
117 $mtu = int($1);
118
119 return $mtu;
120 };
121
122 my $parse_tap_devive_name = sub {
123 my ($iface, $noerr) = @_;
124
125 my ($vmid, $devid);
126
127 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
128 $vmid = $1;
129 $devid = $2;
130 } elsif ($iface =~ m/^veth(\d+)\.(\d+)$/) {
131 $vmid = $1;
132 $devid = $2;
133 } else {
134 return undef if $noerr;
135 die "can't create firewall bridge for random interface name '$iface'\n";
136 }
137
138 return ($vmid, $devid);
139 };
140
141 my $compute_fwbr_names = sub {
142 my ($vmid, $devid) = @_;
143
144 my $fwbr = "fwbr${vmid}i${devid}";
145 # Note: the firewall use 'fwln+' to filter traffic to VMs
146 my $vethfw = "fwln${vmid}i${devid}";
147 my $vethfwpeer = "fwpr${vmid}p${devid}";
148 my $ovsintport = "fwln${vmid}o${devid}";
149
150 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
151 };
152
153 my $cond_create_bridge = sub {
154 my ($bridge) = @_;
155
156 if (! -d "/sys/class/net/$bridge") {
157 system("/sbin/brctl addbr $bridge") == 0 ||
158 die "can't add bridge '$bridge'\n";
159 }
160 };
161
162 my $bridge_add_interface = sub {
163 my ($bridge, $iface) = @_;
164
165 system("/sbin/brctl addif $bridge $iface") == 0 ||
166 die "can't add interface 'iface' to bridge '$bridge'\n";
167 };
168
169 my $ovs_bridge_add_port = sub {
170 my ($bridge, $iface, $tag, $internal) = @_;
171
172 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
173 $cmd .= " tag=$tag" if $tag;
174 $cmd .= " -- set Interface $iface type=internal" if $internal;
175 system($cmd) == 0 ||
176 die "can't add ovs port '$iface'\n";
177 };
178
179 my $activate_interface = sub {
180 my ($iface) = @_;
181
182 system("/sbin/ip link set $iface up") == 0 ||
183 die "can't activate interface '$iface'\n";
184 };
185
186 sub tap_create {
187 my ($iface, $bridge) = @_;
188
189 die "unable to get bridge setting\n" if !$bridge;
190
191 my $bridgemtu = &$read_bridge_mtu($bridge);
192
193 eval {
194 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
195 };
196 die "interface activation failed\n" if $@;
197 }
198
199 sub veth_create {
200 my ($veth, $vethpeer, $bridge, $mac) = @_;
201
202 die "unable to get bridge setting\n" if !$bridge;
203
204 my $bridgemtu = &$read_bridge_mtu($bridge);
205
206 # create veth pair
207 if (! -d "/sys/class/net/$veth") {
208 my $cmd = "/sbin/ip link add name $veth type veth peer name $vethpeer mtu $bridgemtu";
209 $cmd .= " addr $mac" if $mac;
210 system($cmd) == 0 || die "can't create interface $veth\n";
211 }
212
213 # up vethpair
214 &$activate_interface($veth);
215 &$activate_interface($vethpeer);
216 }
217
218 sub veth_delete {
219 my ($veth) = @_;
220
221 if (-d "/sys/class/net/$veth") {
222 run_command("/sbin/ip link delete dev $veth", outfunc => sub {}, errfunc => sub {});
223 }
224
225 }
226
227 my $create_firewall_bridge_linux = sub {
228 my ($iface, $bridge) = @_;
229
230 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
231 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
232
233 &$cond_create_bridge($fwbr);
234 &$activate_interface($fwbr);
235
236 copy_bridge_config($bridge, $fwbr);
237 veth_create($vethfw, $vethfwpeer, $bridge);
238
239 &$bridge_add_interface($fwbr, $vethfw);
240 &$bridge_add_interface($bridge, $vethfwpeer);
241
242 return $fwbr;
243 };
244
245 my $create_firewall_bridge_ovs = sub {
246 my ($iface, $bridge, $tag) = @_;
247
248 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
249 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
250
251 my $bridgemtu = &$read_bridge_mtu($bridge);
252
253 &$cond_create_bridge($fwbr);
254 &$activate_interface($fwbr);
255
256 &$bridge_add_interface($fwbr, $iface);
257
258 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1);
259 &$activate_interface($ovsintport);
260
261 # set the same mtu for ovs int port
262 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
263
264 &$bridge_add_interface($fwbr, $ovsintport);
265 };
266
267 my $cleanup_firewall_bridge = sub {
268 my ($iface) = @_;
269
270 my ($vmid, $devid) = &$parse_tap_devive_name($iface, 1);
271 return if !defined($vmid);
272 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
273
274 # cleanup old port config from any openvswitch bridge
275 if (-d "/sys/class/net/$ovsintport") {
276 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
277 }
278
279 # delete old vethfw interface
280 veth_delete($vethfw);
281
282 # cleanup fwbr bridge
283 if (-d "/sys/class/net/$fwbr") {
284 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
285 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
286 }
287 };
288
289 sub tap_plug {
290 my ($iface, $bridge, $tag, $firewall) = @_;
291
292 #cleanup old port config from any openvswitch bridge
293 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
294
295 if (-d "/sys/class/net/$bridge/bridge") {
296 &$cleanup_firewall_bridge($iface); # remove stale devices
297
298 my $newbridge = activate_bridge_vlan($bridge, $tag);
299 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
300
301 $newbridge = &$create_firewall_bridge_linux($iface, $newbridge) if $firewall;
302
303 &$bridge_add_interface($newbridge, $iface);
304 } else {
305 &$cleanup_firewall_bridge($iface); # remove stale devices
306
307 if ($firewall) {
308 &$create_firewall_bridge_ovs($iface, $bridge, $tag);
309 } else {
310 &$ovs_bridge_add_port($bridge, $iface, $tag);
311 }
312 }
313 }
314
315 sub tap_unplug {
316 my ($iface) = @_;
317
318 my $path= "/sys/class/net/$iface/brport/bridge";
319 if (-l $path) {
320 my $bridge = basename(readlink($path));
321 #avoid insecure dependency;
322 ($bridge) = $bridge =~ /(\S+)/;
323
324 system("/sbin/brctl delif $bridge $iface") == 0 ||
325 die "can't del interface '$iface' from bridge '$bridge'\n";
326
327 }
328
329 &$cleanup_firewall_bridge($iface);
330 }
331
332 sub copy_bridge_config {
333 my ($br0, $br1) = @_;
334
335 return if $br0 eq $br1;
336
337 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
338 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
339
340 foreach my $sysname (@$br_configs) {
341 eval {
342 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
343 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
344 if ($v0 ne $v1) {
345 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
346 }
347 };
348 warn $@ if $@;
349 }
350 }
351
352 sub activate_bridge_vlan_slave {
353 my ($bridgevlan, $iface, $tag) = @_;
354 my $ifacevlan = "${iface}.$tag";
355
356 # create vlan on $iface is not already exist
357 if (! -d "/sys/class/net/$ifacevlan") {
358 system("/sbin/vconfig add $iface $tag") == 0 ||
359 die "can't add vlan tag $tag to interface $iface\n";
360 }
361
362 # be sure to have the $ifacevlan up
363 &$activate_interface($ifacevlan);
364
365 # test if $vlaniface is already enslaved in another bridge
366 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
367 if (-l $path) {
368 my $tbridge = basename(readlink($path));
369 if ($tbridge ne $bridgevlan) {
370 die "interface $ifacevlan already exist in bridge $tbridge\n";
371 } else {
372 # Port already attached to bridge: do nothing.
373 return;
374 }
375 }
376
377 # add $ifacevlan to the bridge
378 &$bridge_add_interface($bridgevlan, $ifacevlan);
379 }
380
381 sub activate_bridge_vlan {
382 my ($bridge, $tag_param) = @_;
383
384 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
385
386 return $bridge if !defined($tag_param); # no vlan, simply return
387
388 my $tag = int($tag_param);
389
390 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
391
392 my $bridgevlan = "${bridge}v$tag";
393
394 my @ifaces = ();
395 my $dir = "/sys/class/net/$bridge/brif";
396 PVE::Tools::dir_glob_foreach($dir, '((eth|bond)\d+)', sub {
397 push @ifaces, $_[0];
398 });
399
400 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
401
402 # add bridgevlan if it doesn't already exist
403 if (! -d "/sys/class/net/$bridgevlan") {
404 system("/sbin/brctl addbr $bridgevlan") == 0 ||
405 die "can't add bridge $bridgevlan\n";
406 }
407
408 # for each physical interface (eth or bridge) bind them to bridge vlan
409 foreach my $iface (@ifaces) {
410 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
411 }
412
413 #fixme: set other bridge flags
414
415 # be sure to have the bridge up
416 system("/sbin/ip link set $bridgevlan up") == 0 ||
417 die "can't up bridge $bridgevlan\n";
418
419 return $bridgevlan;
420 }
421
422 1;