]> git.proxmox.com Git - pve-common.git/blob - data/PVE/Network.pm
ff4aa730fbef1c37df24206f0f5ba7b2d7c52a7c
[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) = @_;
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 die "wrong interface name $iface";
81 }
82
83 return ($vmid, $devid);
84 };
85
86 my $compute_fwbr_names = sub {
87 my ($vmid, $devid) = @_;
88
89 my $fwbr = "fwbr${vmid}i${devid}";
90 # Note: the firewall use 'fwln+' to filter traffic to VMs
91 my $vethfw = "fwln${vmid}i${devid}";
92 my $vethfwpeer = "fwpr${vmid}p${devid}";
93 my $ovsintport = "fwln${vmid}o${devid}";
94
95 return ($fwbr, $vethfw, $vethfwpeer, $ovsintport);
96 };
97
98 my $cond_create_bridge = sub {
99 my ($bridge) = @_;
100
101 if (! -d "/sys/class/net/$bridge") {
102 system("/sbin/brctl addbr $bridge") == 0 ||
103 die "can't add bridge '$bridge'\n";
104 }
105 };
106
107 my $bridge_add_interface = sub {
108 my ($bridge, $iface) = @_;
109
110 system("/sbin/brctl addif $bridge $iface") == 0 ||
111 die "can't add interface 'iface' to bridge '$bridge'\n";
112 };
113
114 my $ovs_bridge_add_port = sub {
115 my ($bridge, $iface, $tag, $internal) = @_;
116
117 my $cmd = "/usr/bin/ovs-vsctl add-port $bridge $iface";
118 $cmd .= " tag=$tag" if $tag;
119 $cmd .= " -- set Interface $iface type=internal" if $internal;
120 system($cmd) == 0 ||
121 die "can't add ovs port '$iface'\n";
122 };
123
124 my $activate_interface = sub {
125 my ($iface) = @_;
126
127 system("/sbin/ip link set $iface up") == 0 ||
128 die "can't activate interface '$iface'\n";
129 };
130
131 sub tap_create {
132 my ($iface, $bridge) = @_;
133
134 die "unable to get bridge setting\n" if !$bridge;
135
136 my $bridgemtu = &$read_bridge_mtu($bridge);
137
138 eval {
139 PVE::Tools::run_command("/sbin/ifconfig $iface 0.0.0.0 promisc up mtu $bridgemtu");
140 };
141 die "interface activation failed\n" if $@;
142 }
143
144 my $create_firewall_bridge_linux = sub {
145 my ($iface, $bridge) = @_;
146
147 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
148 my ($fwbr, $vethfw, $vethfwpeer) = &$compute_fwbr_names($vmid, $devid);
149
150 my $bridgemtu = &$read_bridge_mtu($bridge);
151
152 &$cond_create_bridge($fwbr);
153 &$activate_interface($fwbr);
154
155 copy_bridge_config($bridge, $fwbr);
156 # create veth pair
157 if (! -d "/sys/class/net/$vethfw") {
158 system("/sbin/ip link add name $vethfw type veth peer name $vethfwpeer mtu $bridgemtu") == 0 ||
159 die "can't create interface $vethfw\n";
160 }
161
162 # up vethpair
163 &$activate_interface($vethfw);
164 &$activate_interface($vethfwpeer);
165
166 &$bridge_add_interface($fwbr, $vethfw);
167 &$bridge_add_interface($bridge, $vethfwpeer);
168
169 return $fwbr;
170 };
171
172 my $create_firewall_bridge_ovs = sub {
173 my ($iface, $bridge, $tag) = @_;
174
175 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
176 my ($fwbr, undef, undef, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
177
178 my $bridgemtu = &$read_bridge_mtu($bridge);
179
180 &$cond_create_bridge($fwbr);
181 &$activate_interface($fwbr);
182
183 &$bridge_add_interface($fwbr, $iface);
184
185 &$ovs_bridge_add_port($bridge, $ovsintport, $tag, 1);
186
187 # set the same mtu for ovs int port
188 PVE::Tools::run_command("/sbin/ifconfig $ovsintport mtu $bridgemtu");
189
190 &$bridge_add_interface($fwbr, $ovsintport);
191 };
192
193 my $cleanup_firewall_bridge = sub {
194 my ($iface) = @_;
195
196 my ($vmid, $devid) = &$parse_tap_devive_name($iface);
197 my ($fwbr, $vethfw, $vethfwpeer, $ovsintport) = &$compute_fwbr_names($vmid, $devid);
198
199 # cleanup old port config from any openvswitch bridge
200 if (-d "/sys/class/net/$ovsintport") {
201 run_command("/usr/bin/ovs-vsctl del-port $ovsintport", outfunc => sub {}, errfunc => sub {});
202 }
203
204 # delete old vethfw interface
205 if (-d "/sys/class/net/$vethfw") {
206 run_command("/sbin/ip link delete dev $vethfw", outfunc => sub {}, errfunc => sub {});
207 }
208
209 # cleanup fwbr bridge
210 if (-d "/sys/class/net/$fwbr") {
211 run_command("/sbin/ip link set dev $fwbr down", outfunc => sub {}, errfunc => sub {});
212 run_command("/sbin/brctl delbr $fwbr", outfunc => sub {}, errfunc => sub {});
213 }
214 };
215
216 sub tap_plug {
217 my ($iface, $bridge, $tag, $firewall) = @_;
218
219 #cleanup old port config from any openvswitch bridge
220 eval {run_command("/usr/bin/ovs-vsctl del-port $iface", outfunc => sub {}, errfunc => sub {}) };
221
222 if (-d "/sys/class/net/$bridge/bridge") {
223 &$cleanup_firewall_bridge($iface); # remove stale devices
224
225 my $newbridge = activate_bridge_vlan($bridge, $tag);
226 copy_bridge_config($bridge, $newbridge) if $bridge ne $newbridge;
227
228 $newbridge = &$create_firewall_bridge_linux($iface, $newbridge) if $firewall;
229
230 &$bridge_add_interface($newbridge, $iface);
231 } else {
232 &$cleanup_firewall_bridge($iface); # remove stale devices
233
234 if ($firewall) {
235 &$create_firewall_bridge_ovs($iface, $bridge, $tag);
236 } else {
237 &$ovs_bridge_add_port($bridge, $iface, $tag);
238 }
239 }
240 }
241
242 sub tap_unplug {
243 my ($iface) = @_;
244
245 my $path= "/sys/class/net/$iface/brport/bridge";
246 if (-l $path) {
247 my $bridge = basename(readlink($path));
248 #avoid insecure dependency;
249 ($bridge) = $bridge =~ /(\S+)/;
250
251 system("/sbin/brctl delif $bridge $iface") == 0 ||
252 die "can't del interface '$iface' from bridge '$bridge'\n";
253
254 }
255
256 &$cleanup_firewall_bridge($iface);
257 }
258
259 sub copy_bridge_config {
260 my ($br0, $br1) = @_;
261
262 return if $br0 eq $br1;
263
264 my $br_configs = [ 'ageing_time', 'stp_state', 'priority', 'forward_delay',
265 'hello_time', 'max_age', 'multicast_snooping', 'multicast_querier'];
266
267 foreach my $sysname (@$br_configs) {
268 eval {
269 my $v0 = PVE::Tools::file_read_firstline("/sys/class/net/$br0/bridge/$sysname");
270 my $v1 = PVE::Tools::file_read_firstline("/sys/class/net/$br1/bridge/$sysname");
271 if ($v0 ne $v1) {
272 PVE::ProcFSTools::write_proc_entry("/sys/class/net/$br1/bridge/$sysname", $v0);
273 }
274 };
275 warn $@ if $@;
276 }
277 }
278
279 sub activate_bridge_vlan_slave {
280 my ($bridgevlan, $iface, $tag) = @_;
281 my $ifacevlan = "${iface}.$tag";
282
283 # create vlan on $iface is not already exist
284 if (! -d "/sys/class/net/$ifacevlan") {
285 system("/sbin/vconfig add $iface $tag") == 0 ||
286 die "can't add vlan tag $tag to interface $iface\n";
287 }
288
289 # be sure to have the $ifacevlan up
290 &$activate_interface($ifacevlan);
291
292 # test if $vlaniface is already enslaved in another bridge
293 my $path= "/sys/class/net/$ifacevlan/brport/bridge";
294 if (-l $path) {
295 my $tbridge = basename(readlink($path));
296 if ($tbridge ne $bridgevlan) {
297 die "interface $ifacevlan already exist in bridge $tbridge\n";
298 } else {
299 # Port already attached to bridge: do nothing.
300 return;
301 }
302 }
303
304 # add $ifacevlan to the bridge
305 &$bridge_add_interface($bridgevlan, $ifacevlan);
306 }
307
308 sub activate_bridge_vlan {
309 my ($bridge, $tag_param) = @_;
310
311 die "bridge '$bridge' is not active\n" if ! -d "/sys/class/net/$bridge";
312
313 return $bridge if !defined($tag_param); # no vlan, simply return
314
315 my $tag = int($tag_param);
316
317 die "got strange vlan tag '$tag_param'\n" if $tag < 1 || $tag > 4094;
318
319 my $bridgevlan = "${bridge}v$tag";
320
321 my @ifaces = ();
322 my $dir = "/sys/class/net/$bridge/brif";
323 PVE::Tools::dir_glob_foreach($dir, '((eth|bond)\d+)', sub {
324 push @ifaces, $_[0];
325 });
326
327 die "no physical interface on bridge '$bridge'\n" if scalar(@ifaces) == 0;
328
329 # add bridgevlan if it doesn't already exist
330 if (! -d "/sys/class/net/$bridgevlan") {
331 system("/sbin/brctl addbr $bridgevlan") == 0 ||
332 die "can't add bridge $bridgevlan\n";
333 }
334
335 # for each physical interface (eth or bridge) bind them to bridge vlan
336 foreach my $iface (@ifaces) {
337 activate_bridge_vlan_slave($bridgevlan, $iface, $tag);
338 }
339
340 #fixme: set other bridge flags
341
342 # be sure to have the bridge up
343 system("/sbin/ip link set $bridgevlan up") == 0 ||
344 die "can't up bridge $bridgevlan\n";
345
346 return $bridgevlan;
347 }
348
349 1;