]> git.proxmox.com Git - pve-common.git/blob - data/PVE/Network.pm
00639f62331d8de1440f3ec25c9c4051e8bf9d06
[pve-common.git] / data / 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 sub setup_tc_rate_limit {
13 my ($iface, $rate, $burst, $debug) = @_;
14
15 system("/sbin/tc class del dev $iface parent 1: classid 1:1 >/dev/null 2>&1");
16 system("/sbin/tc filter del dev $iface parent ffff: protocol ip prio 50 estimator 1sec 8sec >/dev/null 2>&1");
17 system("/sbin/tc qdisc del dev $iface ingress >/dev/null 2>&1");
18 system("/sbin/tc qdisc del dev $iface root >/dev/null 2>&1");
19
20 return if !$rate;
21
22 run_command("/sbin/tc qdisc add dev $iface handle ffff: ingress");
23
24 # this does not work wit virtio - don't know why (setting "mtu 64kb" does not help)
25 #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");
26 # so we use avrate instead
27 run_command("/sbin/tc filter add dev $iface parent ffff: " .
28 "protocol ip prio 50 estimator 1sec 8sec " .
29 "u32 match ip src 0.0.0.0/0 police avrate ${rate}bps drop flowid :1");
30
31 # tbf does not work for unknown reason
32 #$TC qdisc add dev $DEV root tbf rate $RATE latency 100ms burst $BURST
33 # so we use htb instead
34 run_command("/sbin/tc qdisc add dev $iface root handle 1: htb default 1");
35 run_command("/sbin/tc class add dev $iface parent 1: classid 1:1 " .
36 "htb rate ${rate}bps burst ${burst}b");
37
38 if ($debug) {
39 print "DEBUG tc settings\n";
40 system("/sbin/tc qdisc ls dev $iface");
41 system("/sbin/tc class ls dev $iface");
42 system("/sbin/tc filter ls dev $iface parent ffff:");
43 }
44 }
45
46 sub tap_rate_limit {
47 my ($iface, $rate) = @_;
48
49 my $debug = 0;
50 $rate = int($rate*1024*1024);
51 my $burst = 1024*1024;
52
53 setup_tc_rate_limit($iface, $rate, $burst, $debug);
54 }
55
56 my $read_bridge_mtu = sub {
57 my ($bridge) = @_;
58
59 my $mtu = PVE::Tools::file_read_firstline("/sys/class/net/$bridge/mtu");
60 die "bridge '$bridge' does not exist\n" if !$mtu;
61 # avoid insecure dependency;
62 die "unable to parse mtu value" if $mtu !~ /^(\d+)$/;
63 $mtu = int($1);
64
65 return $mtu;
66 };
67
68 my $parse_tap_devive_name = sub {
69 my ($iface, $noerr) = @_;
70
71 my ($vmid, $devid);
72
73 if ($iface =~ m/^tap(\d+)i(\d+)$/) {
74 $vmid = $1;
75 $devid = $2;
76 } elsif ($iface =~ m/^veth(\d+)\.(\d+)$/) {
77 $vmid = $1;
78 $devid = $2;
79 } else {
80 return undef if $noerr;
81 die "can't create firewall bridge for random interface name '$iface'\n";
82 }
83
84 return ($vmid, $devid);
85 };
86
87 my $compute_fwbr_names = sub {
88 my ($vmid, $devid) = @_;
89
90 my $fwbr = "fwbr${vmid}i${devid}";
91 # Note: the firewall use 'fwln+' to filter traffic to VMs
92 my $vethfw = "fwln${vmid}i${devid}";
93 my $vethfwpeer = "fwpr${vmid}p${devid}";
94 my $ovsintport = "fwln${vmid}o${devid}";
95
96 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
97 };
98
99 my $cond_create_bridge = sub {
100 my ($bridge) = @_;
101
102 if (! -d "/sys/class/net/$bridge") {
103 system("/sbin/brctl addbr $bridge") == 0 ||
104 die "can't add bridge '$bridge'\n";
105 }
106 };
107
108 my $bridge_add_interface = sub {
109 my ($bridge, $iface) = @_;
110
111 system("/sbin/brctl addif $bridge $iface") == 0 ||
112 die "can't add interface 'iface' to bridge '$bridge'\n";
113 };
114
115 my $ovs_bridge_add_port = sub {
116 my ($bridge, $iface, $tag, $internal) = @_;
117
118 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
119 $cmd .= " tag=$tag" if $tag;
120 $cmd .= " -- set Interface $iface type=internal" if $internal;
121 system($cmd) == 0 ||
122 die "can't add ovs port '$iface'\n";
123 };
124
125 my $activate_interface = sub {
126 my ($iface) = @_;
127
128 system("/sbin/ip link set $iface up") == 0 ||
129 die "can't activate interface '$iface'\n";
130 };
131
132 sub tap_create {
133 my ($iface, $bridge) = @_;
134
135 die "unable to get bridge setting\n" if !$bridge;
136
137 my $bridgemtu = &$read_bridge_mtu($bridge);
138
139 eval {
140 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
141 };
142 die "interface activation failed\n" if $@;
143 }
144
145 my $create_firewall_bridge_linux = sub {
146 my ($iface, $bridge) = @_;
147
148 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
149 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
150
151 my $bridgemtu = &$read_bridge_mtu($bridge);
152
153 &$cond_create_bridge($fwbr);
154 &$activate_interface($fwbr);
155
156 copy_bridge_config($bridge, $fwbr);
157 # create veth pair
158 if (! -d "/sys/class/net/$vethfw") {
159 system("/sbin/ip link add name $vethfw type veth peer name $vethfwpeer mtu $bridgemtu") == 0 ||
160 die "can't create interface $vethfw\n";
161 }
162
163 # up vethpair
164 &$activate_interface($vethfw);
165 &$activate_interface($vethfwpeer);
166
167 &$bridge_add_interface($fwbr, $vethfw);
168 &$bridge_add_interface($bridge, $vethfwpeer);
169
170 return $fwbr;
171 };
172
173 my $create_firewall_bridge_ovs = sub {
174 my ($iface, $bridge, $tag) = @_;
175
176 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
177 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
178
179 my $bridgemtu = &$read_bridge_mtu($bridge);
180
181 &$cond_create_bridge($fwbr);
182 &$activate_interface($fwbr);
183
184 &$bridge_add_interface($fwbr, $iface);
185
186 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1);
187 &$activate_interface($ovsintport);
188
189 # set the same mtu for ovs int port
190 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
191
192 &$bridge_add_interface($fwbr, $ovsintport);
193 };
194
195 my $cleanup_firewall_bridge = sub {
196 my ($iface) = @_;
197
198 my ($vmid, $devid) = &$parse_tap_devive_name($iface, 1);
199 return if !defined($vmid);
200 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
201
202 # cleanup old port config from any openvswitch bridge
203 if (-d "/sys/class/net/$ovsintport") {
204 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
205 }
206
207 # delete old vethfw interface
208 if (-d "/sys/class/net/$vethfw") {
209 run_command("/sbin/ip link delete dev $vethfw", outfunc => sub {}, errfunc => sub {});
210 }
211
212 # cleanup fwbr bridge
213 if (-d "/sys/class/net/$fwbr") {
214 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
215 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
216 }
217 };
218
219 sub tap_plug {
220 my ($iface, $bridge, $tag, $firewall) = @_;
221
222 #cleanup old port config from any openvswitch bridge
223 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
224
225 if (-d "/sys/class/net/$bridge/bridge") {
226 &$cleanup_firewall_bridge($iface); # remove stale devices
227
228 my $newbridge = activate_bridge_vlan($bridge, $tag);
229 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
230
231 $newbridge = &$create_firewall_bridge_linux($iface, $newbridge) if $firewall;
232
233 &$bridge_add_interface($newbridge, $iface);
234 } else {
235 &$cleanup_firewall_bridge($iface); # remove stale devices
236
237 if ($firewall) {
238 &$create_firewall_bridge_ovs($iface, $bridge, $tag);
239 } else {
240 &$ovs_bridge_add_port($bridge, $iface, $tag);
241 }
242 }
243 }
244
245 sub tap_unplug {
246 my ($iface) = @_;
247
248 my $path= "/sys/class/net/$iface/brport/bridge";
249 if (-l $path) {
250 my $bridge = basename(readlink($path));
251 #avoid insecure dependency;
252 ($bridge) = $bridge =~ /(\S+)/;
253
254 system("/sbin/brctl delif $bridge $iface") == 0 ||
255 die "can't del interface '$iface' from bridge '$bridge'\n";
256
257 }
258
259 &$cleanup_firewall_bridge($iface);
260 }
261
262 sub copy_bridge_config {
263 my ($br0, $br1) = @_;
264
265 return if $br0 eq $br1;
266
267 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
268 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
269
270 foreach my $sysname (@$br_configs) {
271 eval {
272 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
273 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
274 if ($v0 ne $v1) {
275 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
276 }
277 };
278 warn $@ if $@;
279 }
280 }
281
282 sub activate_bridge_vlan_slave {
283 my ($bridgevlan, $iface, $tag) = @_;
284 my $ifacevlan = "${iface}.$tag";
285
286 # create vlan on $iface is not already exist
287 if (! -d "/sys/class/net/$ifacevlan") {
288 system("/sbin/vconfig add $iface $tag") == 0 ||
289 die "can't add vlan tag $tag to interface $iface\n";
290 }
291
292 # be sure to have the $ifacevlan up
293 &$activate_interface($ifacevlan);
294
295 # test if $vlaniface is already enslaved in another bridge
296 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
297 if (-l $path) {
298 my $tbridge = basename(readlink($path));
299 if ($tbridge ne $bridgevlan) {
300 die "interface $ifacevlan already exist in bridge $tbridge\n";
301 } else {
302 # Port already attached to bridge: do nothing.
303 return;
304 }
305 }
306
307 # add $ifacevlan to the bridge
308 &$bridge_add_interface($bridgevlan, $ifacevlan);
309 }
310
311 sub activate_bridge_vlan {
312 my ($bridge, $tag_param) = @_;
313
314 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
315
316 return $bridge if !defined($tag_param); # no vlan, simply return
317
318 my $tag = int($tag_param);
319
320 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
321
322 my $bridgevlan = "${bridge}v$tag";
323
324 my @ifaces = ();
325 my $dir = "/sys/class/net/$bridge/brif";
326 PVE::Tools::dir_glob_foreach($dir, '((eth|bond)\d+)', sub {
327 push @ifaces, $_[0];
328 });
329
330 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
331
332 # add bridgevlan if it doesn't already exist
333 if (! -d "/sys/class/net/$bridgevlan") {
334 system("/sbin/brctl addbr $bridgevlan") == 0 ||
335 die "can't add bridge $bridgevlan\n";
336 }
337
338 # for each physical interface (eth or bridge) bind them to bridge vlan
339 foreach my $iface (@ifaces) {
340 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
341 }
342
343 #fixme: set other bridge flags
344
345 # be sure to have the bridge up
346 system("/sbin/ip link set $bridgevlan up") == 0 ||
347 die "can't up bridge $bridgevlan\n";
348
349 return $bridgevlan;
350 }
351
352 1;