]>
Commit | Line | Data |
---|---|---|
b9436cda DM |
1 | package PVE::Network; |
2 | ||
3 | use strict; | |
c36f332e | 4 | use warnings; |
f27d5e6b | 5 | |
b9436cda | 6 | use PVE::INotify; |
f27d5e6b TL |
7 | use PVE::ProcFSTools; |
8 | use PVE::Tools qw(run_command lock_file); | |
9 | ||
b9436cda | 10 | use File::Basename; |
b6bff92e | 11 | use IO::Socket::IP; |
d7cafe51 | 12 | use JSON; |
bf52d27b | 13 | use Net::IP; |
8286ef53 | 14 | use NetAddr::IP qw(:lower); |
f27d5e6b TL |
15 | use POSIX qw(ECONNREFUSED); |
16 | use Socket qw(NI_NUMERICHOST NI_NUMERICSERV); | |
bf52d27b | 17 | |
b9436cda DM |
18 | # host network related utility functions |
19 | ||
19819404 | 20 | our $PHYSICAL_NIC_RE = qr/(?:eth\d+|en[^:.]+|ib[^:.]+)/; |
3dabe28a | 21 | |
61aa94e4 WB |
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 $ipv4_mask_hash_localnet = { | |
19e609fd WB |
59 | '255.0.0.0' => 8, |
60 | '255.128.0.0' => 9, | |
61 | '255.192.0.0' => 10, | |
62 | '255.224.0.0' => 11, | |
63 | '255.240.0.0' => 12, | |
64 | '255.248.0.0' => 13, | |
65 | '255.252.0.0' => 14, | |
66 | '255.254.0.0' => 15, | |
61aa94e4 WB |
67 | '255.255.0.0' => 16, |
68 | '255.255.128.0' => 17, | |
69 | '255.255.192.0' => 18, | |
70 | '255.255.224.0' => 19, | |
71 | '255.255.240.0' => 20, | |
72 | '255.255.248.0' => 21, | |
73 | '255.255.252.0' => 22, | |
74 | '255.255.254.0' => 23, | |
75 | '255.255.255.0' => 24, | |
76 | '255.255.255.128' => 25, | |
77 | '255.255.255.192' => 26, | |
78 | '255.255.255.224' => 27, | |
79 | '255.255.255.240' => 28, | |
80 | '255.255.255.248' => 29, | |
81 | '255.255.255.252' => 30, | |
e43faad9 WB |
82 | '255.255.255.254' => 31, |
83 | '255.255.255.255' => 32, | |
61aa94e4 WB |
84 | }; |
85 | ||
74d1b045 | 86 | sub setup_tc_rate_limit { |
6256f2c3 | 87 | my ($iface, $rate, $burst) = @_; |
74d1b045 | 88 | |
2d6b3a90 FG |
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 $iface root >/dev/null 2>&1"); }; | |
74d1b045 | 94 | |
d6f2623b | 95 | return if !$rate; |
957753df | 96 | |
74d1b045 DM |
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 | ||
5d35df41 W |
104 | run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress"); |
105 | run_command("/sbin/tc filter add dev $iface parent ffff: " . | |
1b915170 | 106 | "prio 50 basic " . |
5d35df41 | 107 | "police rate ${rate}bps burst ${burst}b mtu 64kb " . |
edbdf0b2 | 108 | "drop"); |
74d1b045 DM |
109 | } |
110 | ||
ec9ada18 AD |
111 | sub tap_rate_limit { |
112 | my ($iface, $rate) = @_; | |
113 | ||
ad066ae2 | 114 | $rate = int($rate*1024*1024) if $rate; |
ec9ada18 AD |
115 | my $burst = 1024*1024; |
116 | ||
6256f2c3 | 117 | setup_tc_rate_limit($iface, $rate, $burst); |
ec9ada18 | 118 | } |
74d1b045 | 119 | |
1b6ad61c | 120 | sub read_bridge_mtu { |
605bb891 DM |
121 | my ($bridge) = @_; |
122 | ||
123 | my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu"); | |
124 | die "bridge '$bridge' does not exist\n" if !$mtu; | |
125 | # avoid insecure dependency; | |
126 | die "unable to parse mtu value" if $mtu !~ /^(\d+)$/; | |
127 | $mtu = int($1); | |
128 | ||
129 | return $mtu; | |
130 | }; | |
131 | ||
32cb7d27 | 132 | my $parse_tap_device_name = sub { |
6c80e6d6 | 133 | my ($iface, $noerr) = @_; |
605bb891 DM |
134 | |
135 | my ($vmid, $devid); | |
136 | ||
137 | if ($iface =~ m/^tap(\d+)i(\d+)$/) { | |
138 | $vmid = $1; | |
139 | $devid = $2; | |
32cb7d27 | 140 | } elsif ($iface =~ m/^veth(\d+)i(\d+)$/) { |
605bb891 DM |
141 | $vmid = $1; |
142 | $devid = $2; | |
143 | } else { | |
6c80e6d6 DM |
144 | return undef if $noerr; |
145 | die "can't create firewall bridge for random interface name '$iface'\n"; | |
605bb891 DM |
146 | } |
147 | ||
148 | return ($vmid, $devid); | |
149 | }; | |
150 | ||
70ab4434 | 151 | my $compute_fwbr_names = sub { |
605bb891 DM |
152 | my ($vmid, $devid) = @_; |
153 | ||
154 | my $fwbr = "fwbr${vmid}i${devid}"; | |
f193aa74 | 155 | # Note: the firewall use 'fwln+' to filter traffic to VMs |
7d78a966 AD |
156 | my $vethfw = "fwln${vmid}i${devid}"; |
157 | my $vethfwpeer = "fwpr${vmid}p${devid}"; | |
158 | my $ovsintport = "fwln${vmid}o${devid}"; | |
605bb891 | 159 | |
70ab4434 | 160 | return ($fwbr, $vethfw, $vethfwpeer, $ovsintport); |
605bb891 DM |
161 | }; |
162 | ||
e9b54cc6 WB |
163 | sub iface_delete($) { |
164 | my ($iface) = @_; | |
165 | run_command(['/sbin/ip', 'link', 'delete', 'dev', $iface], noerr => 1) | |
166 | == 0 or die "failed to delete interface '$iface'\n"; | |
167 | } | |
168 | ||
169 | sub iface_create($$@) { | |
170 | my ($iface, $type, @args) = @_; | |
171 | run_command(['/sbin/ip', 'link', 'add', $iface, 'type', $type, @args], noerr => 1) | |
172 | == 0 or die "failed to create interface '$iface'\n"; | |
173 | } | |
174 | ||
175 | sub iface_set($@) { | |
176 | my ($iface, @opts) = @_; | |
177 | run_command(['/sbin/ip', 'link', 'set', $iface, @opts], noerr => 1) | |
178 | == 0 or die "failed to set interface options for '$iface' (".join(' ', @opts).")\n"; | |
179 | } | |
180 | ||
181 | # helper for nicer error messages: | |
182 | sub iface_set_master($$) { | |
183 | my ($iface, $master) = @_; | |
184 | if (defined($master)) { | |
185 | eval { iface_set($iface, 'master', $master) }; | |
186 | die "can't enslave '$iface' to '$master'\n" if $@; | |
187 | } else { | |
188 | eval { iface_set($iface, 'nomaster') }; | |
189 | die "can't unenslave '$iface'\n" if $@; | |
190 | } | |
191 | } | |
192 | ||
605bb891 DM |
193 | my $cond_create_bridge = sub { |
194 | my ($bridge) = @_; | |
195 | ||
196 | if (! -d "/sys/class/net/$bridge") { | |
e9b54cc6 | 197 | iface_create($bridge, 'bridge'); |
86b84237 | 198 | disable_ipv6($bridge); |
605bb891 DM |
199 | } |
200 | }; | |
201 | ||
f3ccd9b4 WB |
202 | sub disable_ipv6 { |
203 | my ($iface) = @_; | |
204 | return if !-d '/proc/sys/net/ipv6'; # ipv6 might be completely disabled | |
205 | my $file = "/proc/sys/net/ipv6/conf/$iface/disable_ipv6"; | |
206 | open(my $fh, '>', $file) or die "failed to open $file for writing: $!\n"; | |
207 | print {$fh} "1\n" or die "failed to disable link-local ipv6 for $iface\n"; | |
208 | close($fh); | |
209 | } | |
210 | ||
354ec8de AD |
211 | my $bridge_disable_interface_learning = sub { |
212 | my ($iface) = @_; | |
213 | ||
214 | PVE::ProcFSTools::write_proc_entry("/sys/class/net/$iface/brport/unicast_flood", "0"); | |
215 | PVE::ProcFSTools::write_proc_entry("/sys/class/net/$iface/brport/learning", "0"); | |
216 | ||
217 | }; | |
218 | ||
605bb891 | 219 | my $bridge_add_interface = sub { |
b0b34ffd | 220 | my ($bridge, $iface, $tag, $trunks) = @_; |
605bb891 | 221 | |
f3ccd9b4 WB |
222 | # drop link local address (it can't be used when on a bridge anyway) |
223 | disable_ipv6($iface); | |
e9b54cc6 | 224 | iface_set_master($iface, $bridge); |
4d25f4aa AD |
225 | |
226 | my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering"); | |
227 | ||
228 | if ($vlan_aware) { | |
aa91ae3d AD |
229 | |
230 | eval { run_command(['/sbin/bridge', 'vlan', 'del', 'dev', $iface, 'vid', '1-4094']) }; | |
231 | die "failed to remove default vlan tags of $iface - $@\n" if $@; | |
232 | ||
233 | if ($trunks) { | |
234 | my @trunks_array = split /;/, $trunks; | |
235 | foreach my $trunk (@trunks_array) { | |
236 | eval { run_command(['/sbin/bridge', 'vlan', 'add', 'dev', $iface, 'vid', $trunk]) }; | |
237 | die "unable to add vlan $trunk to interface $iface - $@\n" if $@; | |
238 | } | |
239 | } elsif (!$tag) { | |
240 | eval { run_command(['/sbin/bridge', 'vlan', 'add', 'dev', $iface, 'vid', '2-4094']) }; | |
241 | die "unable to add default vlan tags to interface $iface - $@\n" if $@; | |
242 | } | |
243 | ||
244 | $tag = 1 if !$tag; | |
245 | eval { run_command(['/sbin/bridge', 'vlan', 'add', 'dev', $iface, 'vid', $tag, 'pvid', 'untagged']) }; | |
246 | die "unable to add vlan $tag to interface $iface - $@\n" if $@; | |
4d25f4aa | 247 | } |
605bb891 DM |
248 | }; |
249 | ||
70ab4434 | 250 | my $ovs_bridge_add_port = sub { |
b0b34ffd AD |
251 | my ($bridge, $iface, $tag, $internal, $trunks) = @_; |
252 | ||
253 | $trunks =~ s/;/,/g if $trunks; | |
70ab4434 | 254 | |
89ea13ef FG |
255 | my $cmd = ['/usr/bin/ovs-vsctl']; |
256 | # first command | |
257 | push @$cmd, '--', 'add-port', $bridge, $iface; | |
258 | push @$cmd, "tag=$tag" if $tag; | |
259 | push @$cmd, "trunks=". join(',', $trunks) if $trunks; | |
260 | push @$cmd, "vlan_mode=native-untagged" if $tag && $trunks; | |
261 | ||
262 | if ($internal) { | |
263 | # second command | |
264 | push @$cmd, '--', 'set', 'Interface', $iface, 'type=internal'; | |
265 | } | |
266 | ||
267 | eval { run_command($cmd) }; | |
268 | die "can't add ovs port '$iface' - $@\n" if $@; | |
b0b34ffd | 269 | |
f3ccd9b4 | 270 | disable_ipv6($iface); |
70ab4434 DM |
271 | }; |
272 | ||
605bb891 DM |
273 | my $activate_interface = sub { |
274 | my ($iface) = @_; | |
275 | ||
89ea13ef FG |
276 | eval { run_command(['/sbin/ip', 'link', 'set', $iface, 'up']) }; |
277 | die "can't activate interface '$iface' - $@\n" if $@; | |
605bb891 DM |
278 | }; |
279 | ||
354ec8de AD |
280 | sub add_bridge_fdb { |
281 | my ($iface, $mac) = @_; | |
282 | ||
283 | my $learning = PVE::Tools::file_read_firstline("/sys/class/net/$iface/brport/learning"); | |
284 | return if $learning; | |
285 | ||
286 | my ($vmid, $devid) = &$parse_tap_device_name($iface, 1); | |
287 | return if !defined($vmid); | |
288 | ||
b8638604 | 289 | run_command(['/sbin/bridge', 'fdb', 'append', $mac, 'dev', $iface, 'master', 'static']); |
354ec8de AD |
290 | |
291 | my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid); | |
292 | ||
293 | if (-d "/sys/class/net/$vethfwpeer") { | |
b8638604 | 294 | run_command(['/sbin/bridge', 'fdb', 'append', $mac, 'dev', $vethfwpeer, 'master', 'static']); |
354ec8de AD |
295 | } |
296 | ||
297 | } | |
298 | ||
299 | sub del_bridge_fdb { | |
300 | my ($iface, $mac) = @_; | |
301 | ||
302 | my $learning = PVE::Tools::file_read_firstline("/sys/class/net/$iface/brport/learning"); | |
303 | return if $learning; | |
304 | ||
305 | my ($vmid, $devid) = &$parse_tap_device_name($iface, 1); | |
306 | return if !defined($vmid); | |
307 | ||
b8638604 | 308 | run_command(['/sbin/bridge', 'fdb', 'del', $mac, 'dev', $iface, 'master', 'static']); |
354ec8de AD |
309 | |
310 | my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid); | |
311 | ||
312 | if (-d "/sys/class/net/$vethfwpeer") { | |
b8638604 | 313 | run_command(['/sbin/bridge', 'fdb', 'del', $mac, 'dev', $vethfwpeer, 'master', 'static']); |
354ec8de AD |
314 | } |
315 | } | |
316 | ||
3aa99c70 AD |
317 | sub tap_create { |
318 | my ($iface, $bridge) = @_; | |
319 | ||
320 | die "unable to get bridge setting\n" if !$bridge; | |
321 | ||
1b6ad61c | 322 | my $bridgemtu = read_bridge_mtu($bridge); |
3aa99c70 | 323 | |
9bbc4e17 | 324 | eval { |
f3ccd9b4 | 325 | disable_ipv6($iface); |
b8638604 | 326 | run_command(['/sbin/ip', 'link', 'set', $iface, 'up', 'promisc', 'on', 'mtu', $bridgemtu]); |
098795e0 DM |
327 | }; |
328 | die "interface activation failed\n" if $@; | |
3aa99c70 AD |
329 | } |
330 | ||
35efc4eb AD |
331 | sub veth_create { |
332 | my ($veth, $vethpeer, $bridge, $mac) = @_; | |
333 | ||
334 | die "unable to get bridge setting\n" if !$bridge; | |
335 | ||
1b6ad61c | 336 | my $bridgemtu = read_bridge_mtu($bridge); |
35efc4eb AD |
337 | |
338 | # create veth pair | |
339 | if (! -d "/sys/class/net/$veth") { | |
89ea13ef FG |
340 | my $cmd = ['/sbin/ip', 'link', 'add']; |
341 | # veth device + MTU | |
342 | push @$cmd, 'name', $veth; | |
343 | push @$cmd, 'mtu', $bridgemtu; | |
344 | push @$cmd, 'type', 'veth'; | |
345 | # peer device + MTU | |
346 | push @$cmd, 'peer', 'name', $vethpeer, 'mtu', $bridgemtu; | |
347 | ||
348 | push @$cmd, 'addr', $mac if $mac; | |
349 | ||
350 | eval { run_command($cmd) }; | |
351 | die "can't create interface $veth - $@\n" if $@; | |
35efc4eb AD |
352 | } |
353 | ||
354 | # up vethpair | |
f3ccd9b4 WB |
355 | disable_ipv6($veth); |
356 | disable_ipv6($vethpeer); | |
35efc4eb AD |
357 | &$activate_interface($veth); |
358 | &$activate_interface($vethpeer); | |
359 | } | |
360 | ||
f3f0bc3a AD |
361 | sub veth_delete { |
362 | my ($veth) = @_; | |
363 | ||
364 | if (-d "/sys/class/net/$veth") { | |
e9b54cc6 | 365 | iface_delete($veth); |
f3f0bc3a | 366 | } |
e0a862e2 | 367 | eval { tap_unplug($veth) }; |
f3f0bc3a | 368 | } |
35efc4eb | 369 | |
605bb891 | 370 | my $create_firewall_bridge_linux = sub { |
93cc2aa9 | 371 | my ($iface, $bridge, $tag, $trunks, $no_learning) = @_; |
605bb891 | 372 | |
32cb7d27 | 373 | my ($vmid, $devid) = &$parse_tap_device_name($iface); |
70ab4434 | 374 | my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid); |
605bb891 | 375 | |
605bb891 DM |
376 | &$cond_create_bridge($fwbr); |
377 | &$activate_interface($fwbr); | |
378 | ||
379 | copy_bridge_config($bridge, $fwbr); | |
35efc4eb | 380 | veth_create($vethfw, $vethfwpeer, $bridge); |
605bb891 | 381 | |
b0b34ffd | 382 | &$bridge_add_interface($bridge, $vethfwpeer, $tag, $trunks); |
93cc2aa9 | 383 | &$bridge_disable_interface_learning($vethfwpeer) if $no_learning; |
354ec8de | 384 | &$bridge_add_interface($fwbr, $vethfw); |
605bb891 | 385 | |
4d25f4aa | 386 | &$bridge_add_interface($fwbr, $iface); |
605bb891 DM |
387 | }; |
388 | ||
70ab4434 | 389 | my $create_firewall_bridge_ovs = sub { |
93cc2aa9 | 390 | my ($iface, $bridge, $tag, $trunks, $no_learning) = @_; |
70ab4434 | 391 | |
32cb7d27 | 392 | my ($vmid, $devid) = &$parse_tap_device_name($iface); |
70ab4434 DM |
393 | my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid); |
394 | ||
1b6ad61c | 395 | my $bridgemtu = read_bridge_mtu($bridge); |
70ab4434 DM |
396 | |
397 | &$cond_create_bridge($fwbr); | |
398 | &$activate_interface($fwbr); | |
399 | ||
400 | &$bridge_add_interface($fwbr, $iface); | |
401 | ||
b0b34ffd | 402 | &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1, $trunks); |
ac3a04b8 | 403 | &$activate_interface($ovsintport); |
70ab4434 DM |
404 | |
405 | # set the same mtu for ovs int port | |
b8638604 | 406 | run_command(['/sbin/ip', 'link', 'set', $ovsintport, 'mtu', $bridgemtu]); |
9bbc4e17 | 407 | |
70ab4434 | 408 | &$bridge_add_interface($fwbr, $ovsintport); |
93cc2aa9 | 409 | &$bridge_disable_interface_learning($ovsintport) if $no_learning; |
70ab4434 DM |
410 | }; |
411 | ||
412 | my $cleanup_firewall_bridge = sub { | |
605bb891 DM |
413 | my ($iface) = @_; |
414 | ||
32cb7d27 | 415 | my ($vmid, $devid) = &$parse_tap_device_name($iface, 1); |
9bbc4e17 | 416 | return if !defined($vmid); |
70ab4434 DM |
417 | my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid); |
418 | ||
419 | # cleanup old port config from any openvswitch bridge | |
420 | if (-d "/sys/class/net/$ovsintport") { | |
421 | run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {}); | |
422 | } | |
605bb891 DM |
423 | |
424 | # delete old vethfw interface | |
f3f0bc3a | 425 | veth_delete($vethfw); |
605bb891 DM |
426 | |
427 | # cleanup fwbr bridge | |
428 | if (-d "/sys/class/net/$fwbr") { | |
e9b54cc6 | 429 | iface_delete($fwbr); |
605bb891 DM |
430 | } |
431 | }; | |
432 | ||
f0c190ee | 433 | sub tap_plug { |
93cc2aa9 | 434 | my ($iface, $bridge, $tag, $firewall, $trunks, $rate, $opts) = @_; |
f0c190ee | 435 | |
93cc2aa9 TL |
436 | $opts = {} if !defined($opts); |
437 | ||
c1978f2e | 438 | my $no_learning = defined($opts->{learning}) && !$opts->{learning}; # default to learning on |
93cc2aa9 TL |
439 | |
440 | # cleanup old port config from any openvswitch bridge | |
441 | eval { | |
442 | run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}); | |
443 | }; | |
4cbabd40 | 444 | |
098795e0 | 445 | if (-d "/sys/class/net/$bridge/bridge") { |
70ab4434 | 446 | &$cleanup_firewall_bridge($iface); # remove stale devices |
605bb891 | 447 | |
4d25f4aa | 448 | my $vlan_aware = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/bridge/vlan_filtering"); |
098795e0 | 449 | |
4d25f4aa | 450 | if (!$vlan_aware) { |
b0b34ffd | 451 | die "vlan aware feature need to be enabled to use trunks" if $trunks; |
4d25f4aa AD |
452 | my $newbridge = activate_bridge_vlan($bridge, $tag); |
453 | copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge; | |
ff042056 | 454 | $bridge = $newbridge; |
4d25f4aa AD |
455 | $tag = undef; |
456 | } | |
457 | ||
458 | if ($firewall) { | |
93cc2aa9 | 459 | &$create_firewall_bridge_linux($iface, $bridge, $tag, $trunks, $no_learning); |
4d25f4aa | 460 | } else { |
b0b34ffd | 461 | &$bridge_add_interface($bridge, $iface, $tag, $trunks); |
4d25f4aa | 462 | } |
93cc2aa9 | 463 | $bridge_disable_interface_learning->($iface) if $no_learning; |
605bb891 | 464 | |
098795e0 | 465 | } else { |
70ab4434 DM |
466 | &$cleanup_firewall_bridge($iface); # remove stale devices |
467 | ||
468 | if ($firewall) { | |
93cc2aa9 | 469 | &$create_firewall_bridge_ovs($iface, $bridge, $tag, $trunks, $no_learning); |
70ab4434 | 470 | } else { |
b0b34ffd | 471 | &$ovs_bridge_add_port($bridge, $iface, $tag, undef, $trunks); |
70ab4434 | 472 | } |
4cbabd40 | 473 | } |
bce2a5b3 WB |
474 | |
475 | tap_rate_limit($iface, $rate); | |
f0c190ee AD |
476 | } |
477 | ||
a84b65c0 | 478 | sub tap_unplug { |
2db1cc0d | 479 | my ($iface) = @_; |
a84b65c0 | 480 | |
2db1cc0d DM |
481 | my $path= "/sys/class/net/$iface/brport/bridge"; |
482 | if (-l $path) { | |
483 | my $bridge = basename(readlink($path)); | |
484 | #avoid insecure dependency; | |
485 | ($bridge) = $bridge =~ /(\S+)/; | |
4cbabd40 | 486 | |
e9b54cc6 | 487 | iface_set_master($iface, undef); |
4cbabd40 | 488 | } |
9bbc4e17 | 489 | |
70ab4434 | 490 | &$cleanup_firewall_bridge($iface); |
dd44486e WB |
491 | #cleanup old port config from any openvswitch bridge |
492 | eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) }; | |
a84b65c0 AD |
493 | } |
494 | ||
b9436cda DM |
495 | sub copy_bridge_config { |
496 | my ($br0, $br1) = @_; | |
497 | ||
498 | return if $br0 eq $br1; | |
499 | ||
b8638604 TL |
500 | my $br_configs = [ |
501 | 'ageing_time', 'stp_state', 'priority', 'forward_delay', | |
502 | 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier', | |
503 | ]; | |
b9436cda DM |
504 | |
505 | foreach my $sysname (@$br_configs) { | |
506 | eval { | |
507 | my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname"); | |
508 | my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname"); | |
509 | if ($v0 ne $v1) { | |
aec04803 | 510 | PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0); |
b9436cda DM |
511 | } |
512 | }; | |
513 | warn $@ if $@; | |
514 | } | |
515 | } | |
516 | ||
70d89745 PRG |
517 | sub activate_bridge_vlan_slave { |
518 | my ($bridgevlan, $iface, $tag) = @_; | |
b9436cda | 519 | my $ifacevlan = "${iface}.$tag"; |
9bbc4e17 | 520 | |
b9436cda DM |
521 | # create vlan on $iface is not already exist |
522 | if (! -d "/sys/class/net/$ifacevlan") { | |
89ea13ef FG |
523 | eval { |
524 | my $cmd = ['/sbin/ip', 'link', 'add']; | |
525 | push @$cmd, 'link', $iface; | |
526 | push @$cmd, 'name', $ifacevlan; | |
527 | push @$cmd, 'type', 'vlan', 'id', $tag; | |
528 | run_command($cmd); | |
529 | }; | |
530 | die "can't add vlan tag $tag to interface $iface - $@\n" if $@; | |
b9436cda | 531 | |
86b84237 WB |
532 | # remove ipv6 link-local address before activation |
533 | disable_ipv6($ifacevlan); | |
534 | } | |
f3ccd9b4 | 535 | |
b9436cda | 536 | # be sure to have the $ifacevlan up |
605bb891 | 537 | &$activate_interface($ifacevlan); |
b9436cda DM |
538 | |
539 | # test if $vlaniface is already enslaved in another bridge | |
540 | my $path= "/sys/class/net/$ifacevlan/brport/bridge"; | |
541 | if (-l $path) { | |
542 | my $tbridge = basename(readlink($path)); | |
70d89745 | 543 | if ($tbridge ne $bridgevlan) { |
b9436cda | 544 | die "interface $ifacevlan already exist in bridge $tbridge\n"; |
eee4b32a PRG |
545 | } else { |
546 | # Port already attached to bridge: do nothing. | |
547 | return; | |
b9436cda DM |
548 | } |
549 | } | |
550 | ||
70d89745 | 551 | # add $ifacevlan to the bridge |
605bb891 | 552 | &$bridge_add_interface($bridgevlan, $ifacevlan); |
70d89745 PRG |
553 | } |
554 | ||
555 | sub activate_bridge_vlan { | |
556 | my ($bridge, $tag_param) = @_; | |
557 | ||
558 | die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge"; | |
559 | ||
560 | return $bridge if !defined($tag_param); # no vlan, simply return | |
561 | ||
562 | my $tag = int($tag_param); | |
563 | ||
564 | die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094; | |
565 | ||
566 | my $bridgevlan = "${bridge}v$tag"; | |
567 | ||
c9030d97 PRG |
568 | my @ifaces = (); |
569 | my $dir = "/sys/class/net/$bridge/brif"; | |
899f8c4a | 570 | PVE::Tools::dir_glob_foreach($dir, '(((eth|bond)\d+|en[^.]+)(\.\d+)?)', sub { |
5ffa7628 | 571 | push @ifaces, $_[0]; |
c9030d97 PRG |
572 | }); |
573 | ||
5ffa7628 | 574 | die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0; |
c9030d97 | 575 | |
a712bf6e WB |
576 | lock_network(sub { |
577 | # add bridgevlan if it doesn't already exist | |
578 | if (! -d "/sys/class/net/$bridgevlan") { | |
e9b54cc6 | 579 | iface_create($bridgevlan, 'bridge'); |
a712bf6e | 580 | } |
b9436cda | 581 | |
a712bf6e WB |
582 | # for each physical interface (eth or bridge) bind them to bridge vlan |
583 | foreach my $iface (@ifaces) { | |
584 | activate_bridge_vlan_slave($bridgevlan, $iface, $tag); | |
585 | } | |
70d89745 | 586 | |
a712bf6e | 587 | #fixme: set other bridge flags |
b9436cda | 588 | |
f3ccd9b4 WB |
589 | # remove ipv6 link-local address before activation |
590 | disable_ipv6($bridgevlan); | |
a712bf6e | 591 | # be sure to have the bridge up |
f3ccd9b4 | 592 | &$activate_interface($bridgevlan); |
a712bf6e | 593 | }); |
b9436cda DM |
594 | return $bridgevlan; |
595 | } | |
596 | ||
b6bff92e WB |
597 | sub tcp_ping { |
598 | my ($host, $port, $timeout) = @_; | |
599 | ||
600 | my $refused = 1; | |
601 | ||
602 | $timeout = 3 if !$timeout; # sane default | |
603 | if (!$port) { | |
604 | # Net::Ping defaults to the echo port | |
605 | $port = 7; | |
606 | } else { | |
607 | # Net::Ping's port_number() implies service_check(1) | |
608 | $refused = 0; | |
609 | } | |
610 | ||
611 | my ($sock, $result); | |
612 | eval { | |
613 | $result = PVE::Tools::run_with_timeout($timeout, sub { | |
614 | $sock = IO::Socket::IP->new(PeerHost => $host, PeerPort => $port, Type => SOCK_STREAM); | |
615 | $result = $refused if $! == ECONNREFUSED; | |
616 | }); | |
617 | }; | |
618 | if ($sock) { | |
619 | $sock->close(); | |
620 | $result = 1; | |
621 | } | |
622 | return $result; | |
623 | } | |
624 | ||
bf52d27b WB |
625 | sub IP_from_cidr { |
626 | my ($cidr, $version) = @_; | |
627 | ||
628 | return if $cidr !~ m!^(\S+?)/(\S+)$!; | |
629 | my ($ip, $prefix) = ($1, $2); | |
630 | ||
631 | my $ipobj = Net::IP->new($ip, $version); | |
632 | return if !$ipobj; | |
633 | ||
634 | $version = $ipobj->version(); | |
635 | ||
636 | my $binmask = Net::IP::ip_get_mask($prefix, $version); | |
637 | return if !$binmask; | |
638 | ||
639 | my $masked_binip = $ipobj->binip() & $binmask; | |
640 | my $masked_ip = Net::IP::ip_bintoip($masked_binip, $version); | |
641 | return Net::IP->new("$masked_ip/$prefix"); | |
642 | } | |
643 | ||
644 | sub is_ip_in_cidr { | |
645 | my ($ip, $cidr, $version) = @_; | |
646 | ||
647 | my $cidr_obj = IP_from_cidr($cidr, $version); | |
648 | return undef if !$cidr_obj; | |
649 | ||
650 | my $ip_obj = Net::IP->new($ip, $version); | |
651 | return undef if !$ip_obj; | |
652 | ||
123c3104 FE |
653 | my $overlap = $cidr_obj->overlaps($ip_obj); |
654 | ||
b0e3bcc1 FE |
655 | return if !defined($overlap); |
656 | ||
123c3104 | 657 | return $overlap == $Net::IP::IP_B_IN_A_OVERLAP || $overlap == $Net::IP::IP_IDENTICAL; |
bf52d27b WB |
658 | } |
659 | ||
d7cafe51 TL |
660 | # get all currently configured addresses that have a global scope, i.e., are reachable from the |
661 | # outside of the host and thus are neither loopback nor link-local ones | |
662 | # returns an array ref of: { addr => "IP", cidr => "IP/PREFIXLEN", family => "inet|inet6" } | |
663 | sub get_reachable_networks { | |
664 | my $raw = ''; | |
665 | run_command([qw(ip -j addr show up scope global)], outfunc => sub { $raw .= shift }); | |
4e405958 | 666 | my $decoded = decode_json($raw); |
d7cafe51 | 667 | |
4e405958 TL |
668 | my $addrs = []; # filter/transform first so that we can sort correctly more easily below |
669 | for my $e ($decoded->@*) { | |
670 | next if !$e->{addr_info} || grep { $_ eq 'LOOPBACK' } $e->{flags}->@*; | |
671 | push $addrs->@*, grep { scalar(keys $_->%*) } $e->{addr_info}->@* | |
672 | } | |
d7cafe51 | 673 | my $res = []; |
4e405958 TL |
674 | for my $info (sort { $a->{family} cmp $b->{family} || $a->{local} cmp $b->{local} } $addrs->@*) { |
675 | push $res->@*, { | |
676 | addr => $info->{local}, | |
677 | cidr => "$info->{local}/$info->{prefixlen}", | |
678 | family => $info->{family}, | |
679 | }; | |
d7cafe51 TL |
680 | } |
681 | ||
682 | return $res; | |
683 | } | |
beb9820f | 684 | |
ac487a88 TL |
685 | # get one or all local IPs that are not loopback ones, able to pick up the following ones (in order) |
686 | # - the hostname primary resolves too, follows gai.conf (admin controlled) and will be prioritised | |
687 | # - all configured in the interfaces configuration | |
688 | # - all currently networks known to the kernel in the current (root) namespace | |
689 | # returns a single address if no parameter is passed, and all found, grouped by type, if `all => 1` | |
690 | # is passed. | |
691 | sub get_local_ip { | |
692 | my (%param) = @_; | |
693 | ||
694 | my $nodename = PVE::INotify::nodename(); | |
695 | my $resolved_host = eval { get_ip_from_hostname($nodename) }; | |
696 | ||
697 | return $resolved_host if defined($resolved_host) && !$param{all}; | |
698 | ||
699 | my $all = { v4 => {}, v6 => {} }; # hash to avoid duplicates and group by type | |
700 | ||
701 | my $ifaces = PVE::INotify::read_file('interfaces', 1)->{data}->{ifaces}; | |
702 | for my $if (values $ifaces->%*) { | |
703 | next if $if->{type} eq 'loopback' || (!defined($if->{address}) && !defined($if->{address6})); | |
704 | my ($v4, $v6) = ($if->{address}, $if->{address6}); | |
705 | ||
706 | return ($v4 // $v6) if !$param{all}; # prefer v4, admin can override $resolved_host via hosts/gai.conf | |
707 | ||
708 | $all->{v4}->{$v4} = 1 if defined($v4); | |
709 | $all->{v6}->{$v6} = 1 if defined($v6); | |
710 | } | |
711 | ||
97809c69 | 712 | my $live = eval { get_reachable_networks() } // []; |
ac487a88 TL |
713 | for my $info ($live->@*) { |
714 | my $addr = $info->{addr}; | |
715 | ||
716 | return $addr if !$param{all}; | |
717 | ||
718 | if ($info->{family} eq 'inet') { | |
719 | $all->{v4}->{$addr} = 1; | |
720 | } else { | |
721 | $all->{v6}->{$addr} = 1; | |
722 | } | |
723 | } | |
724 | ||
725 | return undef if !$param{all}; # getting here means no early return above triggered -> no IPs | |
726 | ||
727 | my $res = []; # order gai.conf controlled first, then group v4 and v6, simply lexically sorted | |
728 | if ($resolved_host) { | |
729 | push $res->@*, $resolved_host; | |
730 | delete $all->{v4}->{$resolved_host}; | |
731 | delete $all->{v6}->{$resolved_host}; | |
732 | } | |
733 | push $res->@*, sort { $a cmp $b } keys $all->{v4}->%*; | |
734 | push $res->@*, sort { $a cmp $b } keys $all->{v6}->%*; | |
735 | ||
736 | return $res; | |
737 | } | |
738 | ||
beb9820f TL |
739 | sub get_local_ip_from_cidr { |
740 | my ($cidr) = @_; | |
741 | ||
1e55a6cd | 742 | my $IPs = {}; |
ef737f0b | 743 | my $i = 1; |
b15e50dd TL |
744 | run_command(['/sbin/ip', 'address', 'show', 'to', $cidr, 'up'], outfunc => sub { |
745 | if ($_[0] =~ m!^\s*inet(?:6)?\s+($PVE::Tools::IPRE)(?:/\d+|\s+peer\s+)!) { | |
ef737f0b | 746 | $IPs->{$1} = $i++ if !exists($IPs->{$1}); |
beb9820f | 747 | } |
b15e50dd | 748 | }); |
beb9820f | 749 | |
ef737f0b | 750 | return [ sort { $IPs->{$a} <=> $IPs->{$b} } keys %{$IPs} ]; |
beb9820f TL |
751 | } |
752 | ||
87aa00de TL |
753 | sub addr_to_ip { |
754 | my ($addr) = @_; | |
755 | my ($err, $host, $port) = Socket::getnameinfo($addr, NI_NUMERICHOST | NI_NUMERICSERV); | |
756 | die "failed to get numerical host address: $err\n" if $err; | |
757 | return ($host, $port) if wantarray; | |
758 | return $host; | |
759 | } | |
760 | ||
761 | sub get_ip_from_hostname { | |
762 | my ($hostname, $noerr) = @_; | |
763 | ||
5bd1e56b | 764 | my @res = eval { PVE::Tools::getaddrinfo_all($hostname) }; |
87aa00de | 765 | if ($@) { |
4ed6974a | 766 | die "hostname lookup '$hostname' failed - $@" if !$noerr; |
87aa00de TL |
767 | return undef; |
768 | } | |
769 | ||
5bd1e56b | 770 | for my $ai (@res) { |
29dde5f4 TL |
771 | my $ip = addr_to_ip($ai->{addr}); |
772 | if ($ip !~ m/^127\.|^::1$/) { | |
773 | return wantarray ? ($ip, $ai->{family}) : $ip; | |
5bd1e56b TL |
774 | } |
775 | } | |
29dde5f4 TL |
776 | # NOTE: we only get here if no WAN/LAN IP was found, so this is now the error path! |
777 | die "address lookup for '$hostname' did not find any IP address\n" if !$noerr; | |
778 | return undef; | |
87aa00de TL |
779 | } |
780 | ||
a712bf6e WB |
781 | sub lock_network { |
782 | my ($code, @param) = @_; | |
783 | my $res = lock_file('/var/lock/pve-network.lck', 10, $code, @param); | |
784 | die $@ if $@; | |
785 | return $res; | |
786 | } | |
787 | ||
8286ef53 FE |
788 | # the canonical form of the given IP, i.e. dotted quad for IPv4 and RFC 5952 for IPv6 |
789 | sub canonical_ip { | |
790 | my ($ip) = @_; | |
791 | ||
792 | my $ip_obj = NetAddr::IP->new($ip) or die "invalid IP string '$ip'\n"; | |
793 | ||
794 | return $ip_obj->canon(); | |
795 | } | |
796 | ||
8f75194c FE |
797 | # List of unique, canonical IPs in the provided list. |
798 | # Keeps the original order, filtering later duplicates. | |
799 | sub unique_ips { | |
800 | my ($ips) = @_; | |
801 | ||
802 | my $res = []; | |
803 | my $seen = {}; | |
804 | ||
805 | for my $ip (@{$ips}) { | |
806 | $ip = canonical_ip($ip); | |
807 | ||
808 | next if $seen->{$ip}; | |
809 | ||
810 | $seen->{$ip} = 1; | |
811 | push @{$res}, $ip; | |
812 | } | |
813 | ||
814 | return $res; | |
815 | } | |
816 | ||
b9436cda | 817 | 1; |