]>
Commit | Line | Data |
---|---|---|
f76a2828 DM |
1 | package PVE::LXC; |
2 | ||
3 | use strict; | |
4 | use warnings; | |
67afe46e | 5 | |
d14a9a1b | 6 | use POSIX qw(EINTR); |
f76a2828 | 7 | |
34fdb3d7 WB |
8 | use Socket; |
9 | ||
f76a2828 | 10 | use File::Path; |
2cfae16e WB |
11 | use File::Spec; |
12 | use Cwd qw(); | |
3cc56749 | 13 | use Fcntl qw(O_RDONLY); |
f76a2828 | 14 | |
f1ba1a4b | 15 | use PVE::Exception qw(raise_perm_exc); |
c65e0a6d | 16 | use PVE::Storage; |
f76a2828 DM |
17 | use PVE::SafeSyslog; |
18 | use PVE::INotify; | |
1bfe8728 | 19 | use PVE::Tools qw($IPV6RE $IPV4RE dir_glob_foreach lock_file lock_file_full); |
68fba17b | 20 | use PVE::Network; |
52389a07 | 21 | use PVE::AccessControl; |
228a5a1d | 22 | use PVE::ProcFSTools; |
67afe46e | 23 | use PVE::LXC::Config; |
688afc63 | 24 | use Time::HiRes qw (gettimeofday); |
f76a2828 DM |
25 | |
26 | use Data::Dumper; | |
27 | ||
27916659 DM |
28 | my $nodename = PVE::INotify::nodename(); |
29 | ||
688afc63 WL |
30 | my $cpuinfo= PVE::ProcFSTools::read_cpuinfo(); |
31 | ||
f9897acd | 32 | our $COMMON_TAR_FLAGS = [ '--sparse', '--numeric-owner', '--acls', |
fc4e132e WB |
33 | '--xattrs', |
34 | '--xattrs-include=user.*', | |
4132377b WB |
35 | '--xattrs-include=security.capability', |
36 | '--warning=no-xattr-write' ]; | |
fc4e132e | 37 | |
f76a2828 DM |
38 | sub config_list { |
39 | my $vmlist = PVE::Cluster::get_vmlist(); | |
40 | my $res = {}; | |
41 | return $res if !$vmlist || !$vmlist->{ids}; | |
42 | my $ids = $vmlist->{ids}; | |
43 | ||
44 | foreach my $vmid (keys %$ids) { | |
45 | next if !$vmid; # skip CT0 | |
46 | my $d = $ids->{$vmid}; | |
47 | next if !$d->{node} || $d->{node} ne $nodename; | |
48 | next if !$d->{type} || $d->{type} ne 'lxc'; | |
49 | $res->{$vmid}->{type} = 'lxc'; | |
50 | } | |
51 | return $res; | |
52 | } | |
53 | ||
5b4657d0 DM |
54 | sub destroy_config { |
55 | my ($vmid) = @_; | |
56 | ||
67afe46e | 57 | unlink PVE::LXC::Config->config_file($vmid, $nodename); |
3cc56749 FG |
58 | } |
59 | ||
822de0c3 DM |
60 | # container status helpers |
61 | ||
62 | sub list_active_containers { | |
cbb03fea | 63 | |
822de0c3 DM |
64 | my $filename = "/proc/net/unix"; |
65 | ||
66 | # similar test is used by lcxcontainers.c: list_active_containers | |
67 | my $res = {}; | |
cbb03fea | 68 | |
822de0c3 DM |
69 | my $fh = IO::File->new ($filename, "r"); |
70 | return $res if !$fh; | |
71 | ||
72 | while (defined(my $line = <$fh>)) { | |
28df2cde | 73 | if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) { |
822de0c3 | 74 | my $path = $1; |
27916659 | 75 | if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) { |
822de0c3 DM |
76 | $res->{$1} = 1; |
77 | } | |
78 | } | |
79 | } | |
80 | ||
81 | close($fh); | |
cbb03fea | 82 | |
822de0c3 DM |
83 | return $res; |
84 | } | |
f76a2828 | 85 | |
5c752bbf DM |
86 | # warning: this is slow |
87 | sub check_running { | |
88 | my ($vmid) = @_; | |
89 | ||
90 | my $active_hash = list_active_containers(); | |
91 | ||
92 | return 1 if defined($active_hash->{$vmid}); | |
cbb03fea | 93 | |
5c752bbf DM |
94 | return undef; |
95 | } | |
96 | ||
10fc3ba5 | 97 | sub get_container_disk_usage { |
73e03cb7 | 98 | my ($vmid, $pid) = @_; |
10fc3ba5 | 99 | |
73e03cb7 | 100 | return PVE::Tools::df("/proc/$pid/root/", 1); |
10fc3ba5 DM |
101 | } |
102 | ||
688afc63 WL |
103 | my $last_proc_vmid_stat; |
104 | ||
105 | my $parse_cpuacct_stat = sub { | |
106 | my ($vmid) = @_; | |
107 | ||
108 | my $raw = read_cgroup_value('cpuacct', $vmid, 'cpuacct.stat', 1); | |
109 | ||
110 | my $stat = {}; | |
111 | ||
112 | if ($raw =~ m/^user (\d+)\nsystem (\d+)\n/) { | |
113 | ||
114 | $stat->{utime} = $1; | |
115 | $stat->{stime} = $2; | |
116 | ||
117 | } | |
118 | ||
119 | return $stat; | |
120 | }; | |
121 | ||
f76a2828 DM |
122 | sub vmstatus { |
123 | my ($opt_vmid) = @_; | |
124 | ||
125 | my $list = $opt_vmid ? { $opt_vmid => { type => 'lxc' }} : config_list(); | |
126 | ||
822de0c3 | 127 | my $active_hash = list_active_containers(); |
cbb03fea | 128 | |
688afc63 WL |
129 | my $cpucount = $cpuinfo->{cpus} || 1; |
130 | ||
131 | my $cdtime = gettimeofday; | |
132 | ||
133 | my $uptime = (PVE::ProcFSTools::read_proc_uptime(1))[0]; | |
134 | ||
f76a2828 | 135 | foreach my $vmid (keys %$list) { |
f76a2828 | 136 | my $d = $list->{$vmid}; |
10fc3ba5 | 137 | |
d5588ee3 DM |
138 | eval { $d->{pid} = find_lxc_pid($vmid) if defined($active_hash->{$vmid}); }; |
139 | warn $@ if $@; # ignore errors (consider them stopped) | |
cbb03fea | 140 | |
d5588ee3 | 141 | $d->{status} = $d->{pid} ? 'running' : 'stopped'; |
f76a2828 | 142 | |
67afe46e | 143 | my $cfspath = PVE::LXC::Config->cfs_config_path($vmid); |
238a56cb | 144 | my $conf = PVE::Cluster::cfs_read_file($cfspath) || {}; |
cbb03fea | 145 | |
27916659 | 146 | $d->{name} = $conf->{'hostname'} || "CT$vmid"; |
238a56cb | 147 | $d->{name} =~ s/[\s]//g; |
cbb03fea | 148 | |
9db5687d | 149 | $d->{cpus} = $conf->{cpulimit} || $cpucount; |
44da0641 | 150 | |
d0226204 WB |
151 | $d->{lock} = $conf->{lock} || ''; |
152 | ||
d5588ee3 DM |
153 | if ($d->{pid}) { |
154 | my $res = get_container_disk_usage($vmid, $d->{pid}); | |
27916659 DM |
155 | $d->{disk} = $res->{used}; |
156 | $d->{maxdisk} = $res->{total}; | |
157 | } else { | |
158 | $d->{disk} = 0; | |
159 | # use 4GB by default ?? | |
160 | if (my $rootfs = $conf->{rootfs}) { | |
1b4cf758 | 161 | my $rootinfo = PVE::LXC::Config->parse_ct_rootfs($rootfs); |
27916659 DM |
162 | $d->{maxdisk} = int(($rootinfo->{size} || 4)*1024*1024)*1024; |
163 | } else { | |
164 | $d->{maxdisk} = 4*1024*1024*1024; | |
10fc3ba5 | 165 | } |
238a56cb | 166 | } |
cbb03fea | 167 | |
238a56cb DM |
168 | $d->{mem} = 0; |
169 | $d->{swap} = 0; | |
95df9a12 DM |
170 | $d->{maxmem} = ($conf->{memory}||512)*1024*1024; |
171 | $d->{maxswap} = ($conf->{swap}//0)*1024*1024; | |
e901d418 | 172 | |
238a56cb DM |
173 | $d->{uptime} = 0; |
174 | $d->{cpu} = 0; | |
e901d418 | 175 | |
238a56cb DM |
176 | $d->{netout} = 0; |
177 | $d->{netin} = 0; | |
f76a2828 | 178 | |
238a56cb DM |
179 | $d->{diskread} = 0; |
180 | $d->{diskwrite} = 0; | |
bb1ac2de | 181 | |
67afe46e | 182 | $d->{template} = PVE::LXC::Config->is_template($conf); |
f76a2828 | 183 | } |
cbb03fea | 184 | |
238a56cb DM |
185 | foreach my $vmid (keys %$list) { |
186 | my $d = $list->{$vmid}; | |
d5588ee3 DM |
187 | my $pid = $d->{pid}; |
188 | ||
189 | next if !$pid; # skip stopped CTs | |
f76a2828 | 190 | |
88a8696b TL |
191 | my $ctime = (stat("/proc/$pid"))[10]; # 10 = ctime |
192 | $d->{uptime} = time - $ctime; # the method lxcfs uses | |
22a77285 | 193 | |
238a56cb DM |
194 | $d->{mem} = read_cgroup_value('memory', $vmid, 'memory.usage_in_bytes'); |
195 | $d->{swap} = read_cgroup_value('memory', $vmid, 'memory.memsw.usage_in_bytes') - $d->{mem}; | |
b5289322 AD |
196 | |
197 | my $blkio_bytes = read_cgroup_value('blkio', $vmid, 'blkio.throttle.io_service_bytes', 1); | |
1e647c7c | 198 | my @bytes = split(/\n/, $blkio_bytes); |
b5289322 | 199 | foreach my $byte (@bytes) { |
1e647c7c DM |
200 | if (my ($key, $value) = $byte =~ /(Read|Write)\s+(\d+)/) { |
201 | $d->{diskread} = $2 if $key eq 'Read'; | |
202 | $d->{diskwrite} = $2 if $key eq 'Write'; | |
203 | } | |
b5289322 | 204 | } |
688afc63 WL |
205 | |
206 | my $pstat = &$parse_cpuacct_stat($vmid); | |
207 | ||
208 | my $used = $pstat->{utime} + $pstat->{stime}; | |
209 | ||
210 | my $old = $last_proc_vmid_stat->{$vmid}; | |
211 | if (!$old) { | |
212 | $last_proc_vmid_stat->{$vmid} = { | |
213 | time => $cdtime, | |
214 | used => $used, | |
215 | cpu => 0, | |
216 | }; | |
217 | next; | |
218 | } | |
219 | ||
220 | my $dtime = ($cdtime - $old->{time}) * $cpucount * $cpuinfo->{user_hz}; | |
221 | ||
222 | if ($dtime > 1000) { | |
223 | my $dutime = $used - $old->{used}; | |
224 | ||
225 | $d->{cpu} = (($dutime/$dtime)* $cpucount) / $d->{cpus}; | |
226 | $last_proc_vmid_stat->{$vmid} = { | |
227 | time => $cdtime, | |
228 | used => $used, | |
229 | cpu => $d->{cpu}, | |
230 | }; | |
231 | } else { | |
232 | $d->{cpu} = $old->{cpu}; | |
233 | } | |
238a56cb | 234 | } |
cbb03fea | 235 | |
68b8f4d1 WL |
236 | my $netdev = PVE::ProcFSTools::read_proc_net_dev(); |
237 | ||
238 | foreach my $dev (keys %$netdev) { | |
239 | next if $dev !~ m/^veth([1-9]\d*)i/; | |
240 | my $vmid = $1; | |
241 | my $d = $list->{$vmid}; | |
242 | ||
243 | next if !$d; | |
244 | ||
245 | $d->{netout} += $netdev->{$dev}->{receive}; | |
246 | $d->{netin} += $netdev->{$dev}->{transmit}; | |
247 | ||
248 | } | |
249 | ||
f76a2828 DM |
250 | return $list; |
251 | } | |
252 | ||
238a56cb DM |
253 | sub read_cgroup_value { |
254 | my ($group, $vmid, $name, $full) = @_; | |
255 | ||
256 | my $path = "/sys/fs/cgroup/$group/lxc/$vmid/$name"; | |
257 | ||
258 | return PVE::Tools::file_get_contents($path) if $full; | |
259 | ||
260 | return PVE::Tools::file_read_firstline($path); | |
261 | } | |
262 | ||
bf0b8c43 AD |
263 | sub write_cgroup_value { |
264 | my ($group, $vmid, $name, $value) = @_; | |
265 | ||
266 | my $path = "/sys/fs/cgroup/$group/lxc/$vmid/$name"; | |
267 | PVE::ProcFSTools::write_proc_entry($path, $value) if -e $path; | |
268 | ||
269 | } | |
270 | ||
52f1d76b DM |
271 | sub find_lxc_console_pids { |
272 | ||
273 | my $res = {}; | |
274 | ||
275 | PVE::Tools::dir_glob_foreach('/proc', '\d+', sub { | |
276 | my ($pid) = @_; | |
277 | ||
278 | my $cmdline = PVE::Tools::file_read_firstline("/proc/$pid/cmdline"); | |
279 | return if !$cmdline; | |
280 | ||
281 | my @args = split(/\0/, $cmdline); | |
282 | ||
c31ad455 | 283 | # search for lxc-console -n <vmid> |
cbb03fea | 284 | return if scalar(@args) != 3; |
52f1d76b DM |
285 | return if $args[1] ne '-n'; |
286 | return if $args[2] !~ m/^\d+$/; | |
287 | return if $args[0] !~ m|^(/usr/bin/)?lxc-console$|; | |
cbb03fea | 288 | |
52f1d76b | 289 | my $vmid = $args[2]; |
cbb03fea | 290 | |
52f1d76b DM |
291 | push @{$res->{$vmid}}, $pid; |
292 | }); | |
293 | ||
294 | return $res; | |
295 | } | |
296 | ||
bedeaaf1 AD |
297 | sub find_lxc_pid { |
298 | my ($vmid) = @_; | |
299 | ||
300 | my $pid = undef; | |
301 | my $parser = sub { | |
302 | my $line = shift; | |
8b25977f | 303 | $pid = $1 if $line =~ m/^PID:\s+(\d+)$/; |
bedeaaf1 | 304 | }; |
c39aa40a | 305 | PVE::Tools::run_command(['lxc-info', '-n', $vmid, '-p'], outfunc => $parser); |
bedeaaf1 | 306 | |
8b25977f | 307 | die "unable to get PID for CT $vmid (not running?)\n" if !$pid; |
cbb03fea | 308 | |
8b25977f | 309 | return $pid; |
bedeaaf1 AD |
310 | } |
311 | ||
cbb03fea | 312 | # Note: we cannot use Net:IP, because that only allows strict |
55fa4e09 DM |
313 | # CIDR networks |
314 | sub parse_ipv4_cidr { | |
315 | my ($cidr, $noerr) = @_; | |
316 | ||
f7a7b413 WB |
317 | if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) && ($2 <= 32)) { |
318 | return { address => $1, netmask => $PVE::Network::ipv4_reverse_mask->[$2] }; | |
55fa4e09 | 319 | } |
cbb03fea | 320 | |
55fa4e09 | 321 | return undef if $noerr; |
cbb03fea | 322 | |
55fa4e09 DM |
323 | die "unable to parse ipv4 address/mask\n"; |
324 | } | |
93285df8 | 325 | |
e22af68f | 326 | |
27916659 | 327 | sub update_lxc_config { |
c628ffa1 | 328 | my ($storage_cfg, $vmid, $conf) = @_; |
b80dd50a | 329 | |
bb1ac2de DM |
330 | my $dir = "/var/lib/lxc/$vmid"; |
331 | ||
332 | if ($conf->{template}) { | |
333 | ||
334 | unlink "$dir/config"; | |
335 | ||
336 | return; | |
337 | } | |
338 | ||
27916659 | 339 | my $raw = ''; |
b80dd50a | 340 | |
27916659 DM |
341 | die "missing 'arch' - internal error" if !$conf->{arch}; |
342 | $raw .= "lxc.arch = $conf->{arch}\n"; | |
b80dd50a | 343 | |
425b62cb WB |
344 | my $unprivileged = $conf->{unprivileged}; |
345 | my $custom_idmap = grep { $_->[0] eq 'lxc.id_map' } @{$conf->{lxc}}; | |
346 | ||
27916659 | 347 | my $ostype = $conf->{ostype} || die "missing 'ostype' - internal error"; |
238b7e3e | 348 | if ($ostype =~ /^(?:debian | ubuntu | centos | fedora | opensuse | archlinux | alpine | unmanaged)$/x) { |
c34f7efe WB |
349 | my $inc ="/usr/share/lxc/config/$ostype.common.conf"; |
350 | $inc ="/usr/share/lxc/config/common.conf" if !-f $inc; | |
351 | $raw .= "lxc.include = $inc\n"; | |
425b62cb | 352 | if ($unprivileged || $custom_idmap) { |
c34f7efe WB |
353 | $inc = "/usr/share/lxc/config/$ostype.userns.conf"; |
354 | $inc = "/usr/share/lxc/config/userns.conf" if !-f $inc; | |
355 | $raw .= "lxc.include = $inc\n" | |
425b62cb | 356 | } |
27916659 | 357 | } else { |
9a7a910b | 358 | die "implement me (ostype $ostype)"; |
27916659 | 359 | } |
b80dd50a | 360 | |
50df544c WB |
361 | # WARNING: DO NOT REMOVE this without making sure that loop device nodes |
362 | # cannot be exposed to the container with r/w access (cgroup perms). | |
363 | # When this is enabled mounts will still remain in the monitor's namespace | |
364 | # after the container unmounted them and thus will not detach from their | |
365 | # files while the container is running! | |
c16b8890 | 366 | $raw .= "lxc.monitor.unshare = 1\n"; |
58cc92a9 | 367 | |
425b62cb WB |
368 | # Should we read them from /etc/subuid? |
369 | if ($unprivileged && !$custom_idmap) { | |
370 | $raw .= "lxc.id_map = u 0 100000 65536\n"; | |
371 | $raw .= "lxc.id_map = g 0 100000 65536\n"; | |
372 | } | |
373 | ||
d250604f | 374 | if (!PVE::LXC::Config->has_dev_console($conf)) { |
eeaea429 DM |
375 | $raw .= "lxc.console = none\n"; |
376 | $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n"; | |
377 | } | |
4f958489 | 378 | |
1b4cf758 | 379 | my $ttycount = PVE::LXC::Config->get_tty_count($conf); |
27916659 | 380 | $raw .= "lxc.tty = $ttycount\n"; |
cbb03fea | 381 | |
c31ad455 | 382 | # some init scripts expect a linux terminal (turnkey). |
a691a5a3 DM |
383 | $raw .= "lxc.environment = TERM=linux\n"; |
384 | ||
27916659 DM |
385 | my $utsname = $conf->{hostname} || "CT$vmid"; |
386 | $raw .= "lxc.utsname = $utsname\n"; | |
cbb03fea | 387 | |
27916659 DM |
388 | my $memory = $conf->{memory} || 512; |
389 | my $swap = $conf->{swap} // 0; | |
390 | ||
391 | my $lxcmem = int($memory*1024*1024); | |
392 | $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n"; | |
a12a36e0 | 393 | |
27916659 DM |
394 | my $lxcswap = int(($memory + $swap)*1024*1024); |
395 | $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n"; | |
396 | ||
397 | if (my $cpulimit = $conf->{cpulimit}) { | |
398 | $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n"; | |
399 | my $value = int(100000*$cpulimit); | |
400 | $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n"; | |
a12a36e0 WL |
401 | } |
402 | ||
27916659 DM |
403 | my $shares = $conf->{cpuunits} || 1024; |
404 | $raw .= "lxc.cgroup.cpu.shares = $shares\n"; | |
405 | ||
1b4cf758 | 406 | my $mountpoint = PVE::LXC::Config->parse_ct_rootfs($conf->{rootfs}); |
a3076d81 | 407 | |
c9a5774b | 408 | $raw .= "lxc.rootfs = $dir/rootfs\n"; |
27916659 DM |
409 | |
410 | my $netcount = 0; | |
411 | foreach my $k (keys %$conf) { | |
412 | next if $k !~ m/^net(\d+)$/; | |
413 | my $ind = $1; | |
1b4cf758 | 414 | my $d = PVE::LXC::Config->parse_lxc_network($conf->{$k}); |
27916659 DM |
415 | $netcount++; |
416 | $raw .= "lxc.network.type = veth\n"; | |
18862537 | 417 | $raw .= "lxc.network.veth.pair = veth${vmid}i${ind}\n"; |
27916659 DM |
418 | $raw .= "lxc.network.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr}); |
419 | $raw .= "lxc.network.name = $d->{name}\n" if defined($d->{name}); | |
420 | $raw .= "lxc.network.mtu = $d->{mtu}\n" if defined($d->{mtu}); | |
a12a36e0 WL |
421 | } |
422 | ||
e576f689 DM |
423 | if (my $lxcconf = $conf->{lxc}) { |
424 | foreach my $entry (@$lxcconf) { | |
425 | my ($k, $v) = @$entry; | |
426 | $netcount++ if $k eq 'lxc.network.type'; | |
427 | $raw .= "$k = $v\n"; | |
428 | } | |
429 | } | |
27916659 | 430 | |
e576f689 DM |
431 | $raw .= "lxc.network.type = empty\n" if !$netcount; |
432 | ||
27916659 DM |
433 | File::Path::mkpath("$dir/rootfs"); |
434 | ||
435 | PVE::Tools::file_set_contents("$dir/config", $raw); | |
b80dd50a DM |
436 | } |
437 | ||
117636e5 DM |
438 | # verify and cleanup nameserver list (replace \0 with ' ') |
439 | sub verify_nameserver_list { | |
440 | my ($nameserver_list) = @_; | |
441 | ||
442 | my @list = (); | |
443 | foreach my $server (PVE::Tools::split_list($nameserver_list)) { | |
444 | PVE::JSONSchema::pve_verify_ip($server); | |
445 | push @list, $server; | |
446 | } | |
447 | ||
448 | return join(' ', @list); | |
449 | } | |
450 | ||
451 | sub verify_searchdomain_list { | |
452 | my ($searchdomain_list) = @_; | |
453 | ||
454 | my @list = (); | |
455 | foreach my $server (PVE::Tools::split_list($searchdomain_list)) { | |
456 | # todo: should we add checks for valid dns domains? | |
457 | push @list, $server; | |
458 | } | |
459 | ||
460 | return join(' ', @list); | |
461 | } | |
462 | ||
aca816ad DM |
463 | sub get_console_command { |
464 | my ($vmid, $conf) = @_; | |
465 | ||
1b4cf758 | 466 | my $cmode = PVE::LXC::Config->get_cmode($conf); |
aca816ad DM |
467 | |
468 | if ($cmode eq 'console') { | |
469 | return ['lxc-console', '-n', $vmid, '-t', 0]; | |
470 | } elsif ($cmode eq 'tty') { | |
471 | return ['lxc-console', '-n', $vmid]; | |
472 | } elsif ($cmode eq 'shell') { | |
473 | return ['lxc-attach', '--clear-env', '-n', $vmid]; | |
474 | } else { | |
475 | die "internal error"; | |
476 | } | |
477 | } | |
478 | ||
c325b32f DM |
479 | sub get_primary_ips { |
480 | my ($conf) = @_; | |
481 | ||
482 | # return data from net0 | |
cbb03fea | 483 | |
27916659 | 484 | return undef if !defined($conf->{net0}); |
1b4cf758 | 485 | my $net = PVE::LXC::Config->parse_lxc_network($conf->{net0}); |
c325b32f DM |
486 | |
487 | my $ipv4 = $net->{ip}; | |
db78a181 WB |
488 | if ($ipv4) { |
489 | if ($ipv4 =~ /^(dhcp|manual)$/) { | |
490 | $ipv4 = undef | |
491 | } else { | |
492 | $ipv4 =~ s!/\d+$!!; | |
493 | } | |
494 | } | |
65e5eaa3 | 495 | my $ipv6 = $net->{ip6}; |
db78a181 | 496 | if ($ipv6) { |
5f291c7d | 497 | if ($ipv6 =~ /^(auto|dhcp|manual)$/) { |
db78a181 WB |
498 | $ipv6 = undef; |
499 | } else { | |
500 | $ipv6 =~ s!/\d+$!!; | |
501 | } | |
502 | } | |
cbb03fea | 503 | |
c325b32f DM |
504 | return ($ipv4, $ipv6); |
505 | } | |
148d1cb4 | 506 | |
b407293b WB |
507 | sub delete_mountpoint_volume { |
508 | my ($storage_cfg, $vmid, $volume) = @_; | |
509 | ||
d250604f | 510 | return if PVE::LXC::Config->classify_mountpoint($volume) ne 'volume'; |
b407293b WB |
511 | |
512 | my ($vtype, $name, $owner) = PVE::Storage::parse_volname($storage_cfg, $volume); | |
513 | PVE::Storage::vdisk_free($storage_cfg, $volume) if $vmid == $owner; | |
514 | } | |
ef241384 | 515 | |
27916659 | 516 | sub destroy_lxc_container { |
148d1cb4 DM |
517 | my ($storage_cfg, $vmid, $conf) = @_; |
518 | ||
d250604f | 519 | PVE::LXC::Config->foreach_mountpoint($conf, sub { |
db8989e1 | 520 | my ($ms, $mountpoint) = @_; |
b407293b | 521 | delete_mountpoint_volume($storage_cfg, $vmid, $mountpoint->{volume}); |
db8989e1 WB |
522 | }); |
523 | ||
27916659 DM |
524 | rmdir "/var/lib/lxc/$vmid/rootfs"; |
525 | unlink "/var/lib/lxc/$vmid/config"; | |
526 | rmdir "/var/lib/lxc/$vmid"; | |
527 | destroy_config($vmid); | |
528 | ||
529 | #my $cmd = ['lxc-destroy', '-n', $vmid ]; | |
530 | #PVE::Tools::run_command($cmd); | |
148d1cb4 | 531 | } |
68fba17b | 532 | |
ef241384 | 533 | sub vm_stop_cleanup { |
5fa890f0 | 534 | my ($storage_cfg, $vmid, $conf, $keepActive) = @_; |
ef241384 DM |
535 | |
536 | eval { | |
537 | if (!$keepActive) { | |
bf9d912c | 538 | |
d250604f | 539 | my $vollist = PVE::LXC::Config->get_vm_volumes($conf); |
a8b6b8a7 | 540 | PVE::Storage::deactivate_volumes($storage_cfg, $vollist); |
ef241384 DM |
541 | } |
542 | }; | |
543 | warn $@ if $@; # avoid errors - just warn | |
544 | } | |
545 | ||
93cdbbfb AD |
546 | my $safe_num_ne = sub { |
547 | my ($a, $b) = @_; | |
548 | ||
549 | return 0 if !defined($a) && !defined($b); | |
550 | return 1 if !defined($a); | |
551 | return 1 if !defined($b); | |
552 | ||
553 | return $a != $b; | |
554 | }; | |
555 | ||
556 | my $safe_string_ne = sub { | |
557 | my ($a, $b) = @_; | |
558 | ||
559 | return 0 if !defined($a) && !defined($b); | |
560 | return 1 if !defined($a); | |
561 | return 1 if !defined($b); | |
562 | ||
563 | return $a ne $b; | |
564 | }; | |
565 | ||
566 | sub update_net { | |
bedeaaf1 | 567 | my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_; |
93cdbbfb | 568 | |
18862537 WB |
569 | if ($newnet->{type} ne 'veth') { |
570 | # for when there are physical interfaces | |
571 | die "cannot update interface of type $newnet->{type}"; | |
572 | } | |
573 | ||
574 | my $veth = "veth${vmid}i${netid}"; | |
93cdbbfb AD |
575 | my $eth = $newnet->{name}; |
576 | ||
18862537 | 577 | if (my $oldnetcfg = $conf->{$opt}) { |
1b4cf758 | 578 | my $oldnet = PVE::LXC::Config->parse_lxc_network($oldnetcfg); |
18862537 WB |
579 | |
580 | if (&$safe_string_ne($oldnet->{hwaddr}, $newnet->{hwaddr}) || | |
581 | &$safe_string_ne($oldnet->{name}, $newnet->{name})) { | |
93cdbbfb | 582 | |
18862537 | 583 | PVE::Network::veth_delete($veth); |
bedeaaf1 | 584 | delete $conf->{$opt}; |
67afe46e | 585 | PVE::LXC::Config->write_config($vmid, $conf); |
93cdbbfb | 586 | |
18862537 | 587 | hotplug_net($vmid, $conf, $opt, $newnet, $netid); |
bedeaaf1 | 588 | |
380962c7 WB |
589 | } else { |
590 | if (&$safe_string_ne($oldnet->{bridge}, $newnet->{bridge}) || | |
591 | &$safe_num_ne($oldnet->{tag}, $newnet->{tag}) || | |
592 | &$safe_num_ne($oldnet->{firewall}, $newnet->{firewall})) { | |
bedeaaf1 | 593 | |
18862537 | 594 | if ($oldnet->{bridge}) { |
bedeaaf1 | 595 | PVE::Network::tap_unplug($veth); |
18862537 WB |
596 | foreach (qw(bridge tag firewall)) { |
597 | delete $oldnet->{$_}; | |
598 | } | |
1b4cf758 | 599 | $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet); |
67afe46e | 600 | PVE::LXC::Config->write_config($vmid, $conf); |
bedeaaf1 | 601 | } |
93cdbbfb | 602 | |
380962c7 WB |
603 | PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate}); |
604 | # This includes the rate: | |
605 | foreach (qw(bridge tag firewall rate)) { | |
18862537 WB |
606 | $oldnet->{$_} = $newnet->{$_} if $newnet->{$_}; |
607 | } | |
380962c7 WB |
608 | } elsif (&$safe_string_ne($oldnet->{rate}, $newnet->{rate})) { |
609 | # Rate can be applied on its own but any change above needs to | |
610 | # include the rate in tap_plug since OVS resets everything. | |
611 | PVE::Network::tap_rate_limit($veth, $newnet->{rate}); | |
612 | $oldnet->{rate} = $newnet->{rate} | |
613 | } | |
614 | $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet); | |
615 | PVE::LXC::Config->write_config($vmid, $conf); | |
93cdbbfb AD |
616 | } |
617 | } else { | |
18862537 | 618 | hotplug_net($vmid, $conf, $opt, $newnet, $netid); |
93cdbbfb AD |
619 | } |
620 | ||
bedeaaf1 | 621 | update_ipconfig($vmid, $conf, $opt, $eth, $newnet, $rootdir); |
93cdbbfb AD |
622 | } |
623 | ||
624 | sub hotplug_net { | |
18862537 | 625 | my ($vmid, $conf, $opt, $newnet, $netid) = @_; |
93cdbbfb | 626 | |
18862537 | 627 | my $veth = "veth${vmid}i${netid}"; |
cbb03fea | 628 | my $vethpeer = $veth . "p"; |
93cdbbfb AD |
629 | my $eth = $newnet->{name}; |
630 | ||
631 | PVE::Network::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr}); | |
380962c7 | 632 | PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate}); |
93cdbbfb | 633 | |
cbb03fea | 634 | # attach peer in container |
93cdbbfb AD |
635 | my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ]; |
636 | PVE::Tools::run_command($cmd); | |
637 | ||
cbb03fea | 638 | # link up peer in container |
93cdbbfb AD |
639 | $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up' ]; |
640 | PVE::Tools::run_command($cmd); | |
bedeaaf1 | 641 | |
18862537 WB |
642 | my $done = { type => 'veth' }; |
643 | foreach (qw(bridge tag firewall hwaddr name)) { | |
644 | $done->{$_} = $newnet->{$_} if $newnet->{$_}; | |
645 | } | |
1b4cf758 | 646 | $conf->{$opt} = PVE::LXC::Config->print_lxc_network($done); |
bedeaaf1 | 647 | |
67afe46e | 648 | PVE::LXC::Config->write_config($vmid, $conf); |
93cdbbfb AD |
649 | } |
650 | ||
68a05bb3 | 651 | sub update_ipconfig { |
bedeaaf1 AD |
652 | my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_; |
653 | ||
f2104b80 | 654 | my $lxc_setup = PVE::LXC::Setup->new($conf, $rootdir); |
bedeaaf1 | 655 | |
1b4cf758 | 656 | my $optdata = PVE::LXC::Config->parse_lxc_network($conf->{$opt}); |
84e0c123 WB |
657 | my $deleted = []; |
658 | my $added = []; | |
8d723477 WB |
659 | my $nscmd = sub { |
660 | my $cmdargs = shift; | |
661 | PVE::Tools::run_command(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs); | |
84e0c123 | 662 | }; |
8d723477 | 663 | my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) }; |
2bfd1615 | 664 | |
84e0c123 | 665 | my $change_ip_config = sub { |
f39002a6 DM |
666 | my ($ipversion) = @_; |
667 | ||
668 | my $family_opt = "-$ipversion"; | |
669 | my $suffix = $ipversion == 4 ? '' : $ipversion; | |
84e0c123 WB |
670 | my $gw= "gw$suffix"; |
671 | my $ip= "ip$suffix"; | |
bedeaaf1 | 672 | |
6178b0dd WB |
673 | my $newip = $newnet->{$ip}; |
674 | my $newgw = $newnet->{$gw}; | |
675 | my $oldip = $optdata->{$ip}; | |
676 | ||
677 | my $change_ip = &$safe_string_ne($oldip, $newip); | |
678 | my $change_gw = &$safe_string_ne($optdata->{$gw}, $newgw); | |
bedeaaf1 | 679 | |
84e0c123 | 680 | return if !$change_ip && !$change_gw; |
68a05bb3 | 681 | |
84e0c123 | 682 | # step 1: add new IP, if this fails we cancel |
292aff54 WB |
683 | my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/); |
684 | if ($change_ip && $is_real_ip) { | |
8d723477 | 685 | eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); }; |
84e0c123 WB |
686 | if (my $err = $@) { |
687 | warn $err; | |
688 | return; | |
689 | } | |
bedeaaf1 | 690 | } |
bedeaaf1 | 691 | |
84e0c123 WB |
692 | # step 2: replace gateway |
693 | # If this fails we delete the added IP and cancel. | |
694 | # If it succeeds we save the config and delete the old IP, ignoring | |
695 | # errors. The config is then saved. | |
696 | # Note: 'ip route replace' can add | |
697 | if ($change_gw) { | |
6178b0dd | 698 | if ($newgw) { |
292aff54 WB |
699 | eval { |
700 | if ($is_real_ip && !PVE::Network::is_ip_in_cidr($newgw, $newip, $ipversion)) { | |
701 | &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth); | |
702 | } | |
703 | &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw); | |
704 | }; | |
84e0c123 WB |
705 | if (my $err = $@) { |
706 | warn $err; | |
707 | # the route was not replaced, the old IP is still available | |
708 | # rollback (delete new IP) and cancel | |
709 | if ($change_ip) { | |
8d723477 | 710 | eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); }; |
84e0c123 WB |
711 | warn $@ if $@; # no need to die here |
712 | } | |
713 | return; | |
714 | } | |
715 | } else { | |
8d723477 | 716 | eval { &$ipcmd($family_opt, 'route', 'del', 'default'); }; |
84e0c123 WB |
717 | # if the route was not deleted, the guest might have deleted it manually |
718 | # warn and continue | |
719 | warn $@ if $@; | |
720 | } | |
2bfd1615 | 721 | } |
2bfd1615 | 722 | |
6178b0dd | 723 | # from this point on we save the configuration |
84e0c123 | 724 | # step 3: delete old IP ignoring errors |
6178b0dd | 725 | if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) { |
8d723477 WB |
726 | # We need to enable promote_secondaries, otherwise our newly added |
727 | # address will be removed along with the old one. | |
728 | my $promote = 0; | |
729 | eval { | |
730 | if ($ipversion == 4) { | |
731 | &$nscmd({ outfunc => sub { $promote = int(shift) } }, | |
732 | 'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries"); | |
733 | &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1"); | |
734 | } | |
735 | &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth); | |
736 | }; | |
84e0c123 | 737 | warn $@ if $@; # no need to die here |
8d723477 WB |
738 | |
739 | if ($ipversion == 4) { | |
740 | &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote"); | |
741 | } | |
bedeaaf1 AD |
742 | } |
743 | ||
84e0c123 WB |
744 | foreach my $property ($ip, $gw) { |
745 | if ($newnet->{$property}) { | |
746 | $optdata->{$property} = $newnet->{$property}; | |
747 | } else { | |
748 | delete $optdata->{$property}; | |
749 | } | |
bedeaaf1 | 750 | } |
1b4cf758 | 751 | $conf->{$opt} = PVE::LXC::Config->print_lxc_network($optdata); |
67afe46e | 752 | PVE::LXC::Config->write_config($vmid, $conf); |
84e0c123 WB |
753 | $lxc_setup->setup_network($conf); |
754 | }; | |
bedeaaf1 | 755 | |
f39002a6 DM |
756 | &$change_ip_config(4); |
757 | &$change_ip_config(6); | |
489e960d WL |
758 | |
759 | } | |
760 | ||
34fdb3d7 WB |
761 | my $enter_namespace = sub { |
762 | my ($vmid, $pid, $which, $type) = @_; | |
763 | sysopen my $fd, "/proc/$pid/ns/$which", O_RDONLY | |
764 | or die "failed to open $which namespace of container $vmid: $!\n"; | |
765 | PVE::Tools::setns(fileno($fd), $type) | |
766 | or die "failed to enter $which namespace of container $vmid: $!\n"; | |
767 | close $fd; | |
768 | }; | |
769 | ||
770 | my $do_syncfs = sub { | |
771 | my ($vmid, $pid, $socket) = @_; | |
772 | ||
773 | &$enter_namespace($vmid, $pid, 'mnt', PVE::Tools::CLONE_NEWNS); | |
774 | ||
775 | # Tell the parent process to start reading our /proc/mounts | |
776 | print {$socket} "go\n"; | |
777 | $socket->flush(); | |
778 | ||
779 | # Receive /proc/self/mounts | |
780 | my $mountdata = do { local $/ = undef; <$socket> }; | |
781 | close $socket; | |
782 | ||
783 | # Now sync all mountpoints... | |
784 | my $mounts = PVE::ProcFSTools::parse_mounts($mountdata); | |
785 | foreach my $mp (@$mounts) { | |
786 | my ($what, $dir, $fs) = @$mp; | |
787 | next if $fs eq 'fuse.lxcfs'; | |
788 | eval { PVE::Tools::sync_mountpoint($dir); }; | |
789 | warn $@ if $@; | |
790 | } | |
791 | }; | |
792 | ||
793 | sub sync_container_namespace { | |
794 | my ($vmid) = @_; | |
795 | my $pid = find_lxc_pid($vmid); | |
796 | ||
797 | # SOCK_DGRAM is nicer for barriers but cannot be slurped | |
798 | socketpair my $pfd, my $cfd, AF_UNIX, SOCK_STREAM, PF_UNSPEC | |
799 | or die "failed to create socketpair: $!\n"; | |
800 | ||
801 | my $child = fork(); | |
802 | die "fork failed: $!\n" if !defined($child); | |
803 | ||
804 | if (!$child) { | |
805 | eval { | |
806 | close $pfd; | |
807 | &$do_syncfs($vmid, $pid, $cfd); | |
808 | }; | |
809 | if (my $err = $@) { | |
810 | warn $err; | |
811 | POSIX::_exit(1); | |
812 | } | |
813 | POSIX::_exit(0); | |
814 | } | |
815 | close $cfd; | |
816 | my $go = <$pfd>; | |
817 | die "failed to enter container namespace\n" if $go ne "go\n"; | |
818 | ||
819 | open my $mounts, '<', "/proc/$child/mounts" | |
820 | or die "failed to open container's /proc/mounts: $!\n"; | |
821 | my $mountdata = do { local $/ = undef; <$mounts> }; | |
822 | close $mounts; | |
823 | print {$pfd} $mountdata; | |
824 | close $pfd; | |
825 | ||
826 | while (waitpid($child, 0) != $child) {} | |
827 | die "failed to sync container namespace\n" if $? != 0; | |
828 | } | |
829 | ||
bb1ac2de DM |
830 | sub template_create { |
831 | my ($vmid, $conf) = @_; | |
832 | ||
833 | my $storecfg = PVE::Storage::config(); | |
834 | ||
1b4cf758 | 835 | my $rootinfo = PVE::LXC::Config->parse_ct_rootfs($conf->{rootfs}); |
bb1ac2de DM |
836 | my $volid = $rootinfo->{volume}; |
837 | ||
838 | die "Template feature is not available for '$volid'\n" | |
839 | if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid); | |
840 | ||
841 | PVE::Storage::activate_volumes($storecfg, [$volid]); | |
842 | ||
843 | my $template_volid = PVE::Storage::vdisk_create_base($storecfg, $volid); | |
844 | $rootinfo->{volume} = $template_volid; | |
1b4cf758 | 845 | $conf->{rootfs} = PVE::LXC::Config->print_ct_mountpoint($rootinfo, 1); |
bb1ac2de | 846 | |
67afe46e | 847 | PVE::LXC::Config->write_config($vmid, $conf); |
bb1ac2de DM |
848 | } |
849 | ||
52389a07 | 850 | sub check_ct_modify_config_perm { |
f1ba1a4b | 851 | my ($rpcenv, $authuser, $vmid, $pool, $newconf, $delete) = @_; |
52389a07 | 852 | |
c81f19d1 | 853 | return 1 if $authuser eq 'root@pam'; |
52389a07 | 854 | |
f1ba1a4b WB |
855 | my $check = sub { |
856 | my ($opt, $delete) = @_; | |
52389a07 DM |
857 | if ($opt eq 'cpus' || $opt eq 'cpuunits' || $opt eq 'cpulimit') { |
858 | $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']); | |
e59a61ed | 859 | } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) { |
52389a07 | 860 | $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']); |
f1ba1a4b | 861 | return if $delete; |
1b4cf758 FG |
862 | my $data = $opt eq 'rootfs' ? PVE::LXC::Config->parse_ct_rootfs($newconf->{$opt}) |
863 | : PVE::LXC::Config->parse_ct_mountpoint($newconf->{$opt}); | |
f1ba1a4b | 864 | raise_perm_exc("mountpoint type $data->{type}") if $data->{type} ne 'volume'; |
52389a07 DM |
865 | } elsif ($opt eq 'memory' || $opt eq 'swap') { |
866 | $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']); | |
867 | } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' || | |
868 | $opt eq 'searchdomain' || $opt eq 'hostname') { | |
869 | $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']); | |
870 | } else { | |
871 | $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']); | |
872 | } | |
f1ba1a4b WB |
873 | }; |
874 | ||
875 | foreach my $opt (keys %$newconf) { | |
876 | &$check($opt, 0); | |
877 | } | |
878 | foreach my $opt (@$delete) { | |
879 | &$check($opt, 1); | |
52389a07 DM |
880 | } |
881 | ||
882 | return 1; | |
883 | } | |
884 | ||
9622e848 | 885 | sub umount_all { |
da629848 | 886 | my ($vmid, $storage_cfg, $conf, $noerr) = @_; |
9622e848 DM |
887 | |
888 | my $rootdir = "/var/lib/lxc/$vmid/rootfs"; | |
d250604f | 889 | my $volid_list = PVE::LXC::Config->get_vm_volumes($conf); |
9622e848 | 890 | |
d250604f | 891 | PVE::LXC::Config->foreach_mountpoint_reverse($conf, sub { |
9622e848 DM |
892 | my ($ms, $mountpoint) = @_; |
893 | ||
894 | my $volid = $mountpoint->{volume}; | |
895 | my $mount = $mountpoint->{mp}; | |
896 | ||
897 | return if !$volid || !$mount; | |
898 | ||
d18f96b4 | 899 | my $mount_path = "$rootdir/$mount"; |
f845a93d | 900 | $mount_path =~ s!/+!/!g; |
9622e848 | 901 | |
228a5a1d WL |
902 | return if !PVE::ProcFSTools::is_mounted($mount_path); |
903 | ||
9622e848 | 904 | eval { |
d18f96b4 | 905 | PVE::Tools::run_command(['umount', '-d', $mount_path]); |
9622e848 DM |
906 | }; |
907 | if (my $err = $@) { | |
908 | if ($noerr) { | |
909 | warn $err; | |
910 | } else { | |
911 | die $err; | |
912 | } | |
913 | } | |
914 | }); | |
9622e848 DM |
915 | } |
916 | ||
917 | sub mount_all { | |
7b49dfe0 | 918 | my ($vmid, $storage_cfg, $conf) = @_; |
9622e848 DM |
919 | |
920 | my $rootdir = "/var/lib/lxc/$vmid/rootfs"; | |
1adc7e53 | 921 | File::Path::make_path($rootdir); |
9622e848 | 922 | |
d250604f | 923 | my $volid_list = PVE::LXC::Config->get_vm_volumes($conf); |
9622e848 DM |
924 | PVE::Storage::activate_volumes($storage_cfg, $volid_list); |
925 | ||
926 | eval { | |
d250604f | 927 | PVE::LXC::Config->foreach_mountpoint($conf, sub { |
9622e848 DM |
928 | my ($ms, $mountpoint) = @_; |
929 | ||
da629848 | 930 | mountpoint_mount($mountpoint, $rootdir, $storage_cfg); |
9622e848 DM |
931 | }); |
932 | }; | |
933 | if (my $err = $@) { | |
e2007ac2 | 934 | warn "mounting container failed\n"; |
9622e848 | 935 | umount_all($vmid, $storage_cfg, $conf, 1); |
e2007ac2 | 936 | die $err; |
9622e848 DM |
937 | } |
938 | ||
da629848 | 939 | return $rootdir; |
9622e848 DM |
940 | } |
941 | ||
942 | ||
b15c75fc | 943 | sub mountpoint_mount_path { |
da629848 | 944 | my ($mountpoint, $storage_cfg, $snapname) = @_; |
b15c75fc | 945 | |
da629848 | 946 | return mountpoint_mount($mountpoint, undef, $storage_cfg, $snapname); |
b15c75fc | 947 | } |
cc6b0307 | 948 | |
2cfae16e WB |
949 | my $check_mount_path = sub { |
950 | my ($path) = @_; | |
951 | $path = File::Spec->canonpath($path); | |
952 | my $real = Cwd::realpath($path); | |
953 | if ($real ne $path) { | |
954 | die "mount path modified by symlink: $path != $real"; | |
955 | } | |
956 | }; | |
957 | ||
21f292ff WB |
958 | sub query_loopdev { |
959 | my ($path) = @_; | |
960 | my $found; | |
961 | my $parser = sub { | |
962 | my $line = shift; | |
963 | if ($line =~ m@^(/dev/loop\d+):@) { | |
964 | $found = $1; | |
965 | } | |
966 | }; | |
967 | my $cmd = ['losetup', '--associated', $path]; | |
968 | PVE::Tools::run_command($cmd, outfunc => $parser); | |
969 | return $found; | |
970 | } | |
971 | ||
50df544c WB |
972 | # Run a function with a file attached to a loop device. |
973 | # The loop device is always detached afterwards (or set to autoclear). | |
974 | # Returns the loop device. | |
975 | sub run_with_loopdev { | |
976 | my ($func, $file) = @_; | |
54d11e5c WB |
977 | my $device = query_loopdev($file); |
978 | # Try to reuse an existing device | |
979 | if ($device) { | |
980 | # We assume that whoever setup the loop device is responsible for | |
981 | # detaching it. | |
982 | &$func($device); | |
983 | return $device; | |
984 | } | |
985 | ||
50df544c WB |
986 | my $parser = sub { |
987 | my $line = shift; | |
988 | if ($line =~ m@^(/dev/loop\d+)$@) { | |
989 | $device = $1; | |
990 | } | |
991 | }; | |
992 | PVE::Tools::run_command(['losetup', '--show', '-f', $file], outfunc => $parser); | |
993 | die "failed to setup loop device for $file\n" if !$device; | |
994 | eval { &$func($device); }; | |
995 | my $err = $@; | |
996 | PVE::Tools::run_command(['losetup', '-d', $device]); | |
997 | die $err if $err; | |
998 | return $device; | |
999 | } | |
1000 | ||
c2744c97 WB |
1001 | sub bindmount { |
1002 | my ($dir, $dest, $ro, @extra_opts) = @_; | |
1003 | PVE::Tools::run_command(['mount', '-o', 'bind', @extra_opts, $dir, $dest]); | |
1004 | if ($ro) { | |
1005 | eval { PVE::Tools::run_command(['mount', '-o', 'bind,remount,ro', $dest]); }; | |
1006 | if (my $err = $@) { | |
1007 | warn "bindmount error\n"; | |
1008 | # don't leave writable bind-mounts behind... | |
1009 | PVE::Tools::run_command(['umount', $dest]); | |
1010 | die $err; | |
1011 | } | |
1012 | } | |
1013 | } | |
1014 | ||
b15c75fc | 1015 | # use $rootdir = undef to just return the corresponding mount path |
cc6b0307 | 1016 | sub mountpoint_mount { |
da629848 | 1017 | my ($mountpoint, $rootdir, $storage_cfg, $snapname) = @_; |
cc6b0307 AD |
1018 | |
1019 | my $volid = $mountpoint->{volume}; | |
1020 | my $mount = $mountpoint->{mp}; | |
7c921c80 | 1021 | my $type = $mountpoint->{type}; |
50df544c WB |
1022 | my $quota = !$snapname && !$mountpoint->{ro} && $mountpoint->{quota}; |
1023 | my $mounted_dev; | |
b15c75fc | 1024 | |
cc6b0307 AD |
1025 | return if !$volid || !$mount; |
1026 | ||
b15c75fc DM |
1027 | my $mount_path; |
1028 | ||
1029 | if (defined($rootdir)) { | |
1030 | $rootdir =~ s!/+$!!; | |
1031 | $mount_path = "$rootdir/$mount"; | |
f845a93d | 1032 | $mount_path =~ s!/+!/!g; |
2cfae16e | 1033 | &$check_mount_path($mount_path); |
b15c75fc | 1034 | File::Path::mkpath($mount_path); |
116ce06f | 1035 | } |
b15c75fc DM |
1036 | |
1037 | my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1); | |
cc6b0307 | 1038 | |
b15c75fc | 1039 | die "unknown snapshot path for '$volid'" if !$storage && defined($snapname); |
cc6b0307 | 1040 | |
471dd315 WB |
1041 | my $optstring = ''; |
1042 | if (defined($mountpoint->{acl})) { | |
1043 | $optstring .= ($mountpoint->{acl} ? 'acl' : 'noacl'); | |
1044 | } | |
c2744c97 | 1045 | my $readonly = $mountpoint->{ro}; |
471dd315 | 1046 | |
c019c56a | 1047 | my @extra_opts = ('-o', $optstring) if $optstring; |
471dd315 | 1048 | |
b15c75fc DM |
1049 | if ($storage) { |
1050 | ||
1051 | my $scfg = PVE::Storage::storage_config($storage_cfg, $storage); | |
1052 | my $path = PVE::Storage::path($storage_cfg, $volid, $snapname); | |
1053 | ||
1054 | my ($vtype, undef, undef, undef, undef, $isBase, $format) = | |
1055 | PVE::Storage::parse_volname($storage_cfg, $volid); | |
1056 | ||
c87b9dd8 DM |
1057 | $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files |
1058 | ||
b15c75fc | 1059 | if ($format eq 'subvol') { |
30de33be DM |
1060 | if ($mount_path) { |
1061 | if ($snapname) { | |
e84f7f5d DM |
1062 | if ($scfg->{type} eq 'zfspool') { |
1063 | my $path_arg = $path; | |
1064 | $path_arg =~ s!^/+!!; | |
471dd315 | 1065 | PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', $path_arg, $mount_path]); |
e84f7f5d | 1066 | } else { |
30de33be DM |
1067 | die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n"; |
1068 | } | |
e84f7f5d | 1069 | } else { |
c2744c97 | 1070 | bindmount($path, $mount_path, $readonly, @extra_opts); |
50df544c | 1071 | warn "cannot enable quota control for bind mounted subvolumes\n" if $quota; |
30de33be | 1072 | } |
b15c75fc | 1073 | } |
5f6280cf | 1074 | return wantarray ? ($path, 0, undef) : $path; |
c87b9dd8 | 1075 | } elsif ($format eq 'raw' || $format eq 'iso') { |
ada088e6 WB |
1076 | # NOTE: 'mount' performs canonicalization without the '-c' switch, which for |
1077 | # device-mapper devices is special-cased to use the /dev/mapper symlinks. | |
1078 | # Our autodev hook expects the /dev/dm-* device currently | |
1079 | # and will create the /dev/mapper symlink accordingly | |
1080 | ($path) = (Cwd::realpath($path) =~ /^(.*)$/s); # realpath() taints | |
50df544c WB |
1081 | my $domount = sub { |
1082 | my ($path) = @_; | |
1083 | if ($mount_path) { | |
1084 | if ($format eq 'iso') { | |
1085 | PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]); | |
1086 | } elsif ($isBase || defined($snapname)) { | |
1087 | PVE::Tools::run_command(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]); | |
1088 | } else { | |
1089 | if ($quota) { | |
1090 | push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0'; | |
1091 | } | |
c2744c97 | 1092 | push @extra_opts, '-o', 'ro' if $readonly; |
50df544c WB |
1093 | PVE::Tools::run_command(['mount', @extra_opts, $path, $mount_path]); |
1094 | } | |
1095 | } | |
1096 | }; | |
30de33be | 1097 | my $use_loopdev = 0; |
b15c75fc | 1098 | if ($scfg->{path}) { |
50df544c | 1099 | $mounted_dev = run_with_loopdev($domount, $path); |
30de33be | 1100 | $use_loopdev = 1; |
2e879877 DM |
1101 | } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' || |
1102 | $scfg->{type} eq 'rbd' || $scfg->{type} eq 'lvmthin') { | |
50df544c WB |
1103 | $mounted_dev = $path; |
1104 | &$domount($path); | |
b15c75fc DM |
1105 | } else { |
1106 | die "unsupported storage type '$scfg->{type}'\n"; | |
1107 | } | |
50df544c | 1108 | return wantarray ? ($path, $use_loopdev, $mounted_dev) : $path; |
b15c75fc DM |
1109 | } else { |
1110 | die "unsupported image format '$format'\n"; | |
1111 | } | |
7c921c80 | 1112 | } elsif ($type eq 'device') { |
c2744c97 | 1113 | push @extra_opts, '-o', 'ro' if $readonly; |
471dd315 | 1114 | PVE::Tools::run_command(['mount', @extra_opts, $volid, $mount_path]) if $mount_path; |
50df544c | 1115 | return wantarray ? ($volid, 0, $volid) : $volid; |
e2007ac2 DM |
1116 | } elsif ($type eq 'bind') { |
1117 | die "directory '$volid' does not exist\n" if ! -d $volid; | |
2cfae16e | 1118 | &$check_mount_path($volid); |
c2744c97 | 1119 | bindmount($volid, $mount_path, $readonly, @extra_opts) if $mount_path; |
50df544c WB |
1120 | warn "cannot enable quota control for bind mounts\n" if $quota; |
1121 | return wantarray ? ($volid, 0, undef) : $volid; | |
b15c75fc DM |
1122 | } |
1123 | ||
1124 | die "unsupported storage"; | |
cc6b0307 AD |
1125 | } |
1126 | ||
6c871c36 | 1127 | sub mkfs { |
d216e891 | 1128 | my ($dev, $rootuid, $rootgid) = @_; |
6c871c36 | 1129 | |
d216e891 WB |
1130 | PVE::Tools::run_command(['mkfs.ext4', '-O', 'mmp', |
1131 | '-E', "root_owner=$rootuid:$rootgid", | |
1132 | $dev]); | |
6c871c36 DM |
1133 | } |
1134 | ||
1135 | sub format_disk { | |
d216e891 | 1136 | my ($storage_cfg, $volid, $rootuid, $rootgid) = @_; |
6c871c36 DM |
1137 | |
1138 | if ($volid =~ m!^/dev/.+!) { | |
1139 | mkfs($volid); | |
1140 | return; | |
1141 | } | |
1142 | ||
1143 | my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1); | |
1144 | ||
1145 | die "cannot format volume '$volid' with no storage\n" if !$storage; | |
1146 | ||
08ca136d DM |
1147 | PVE::Storage::activate_volumes($storage_cfg, [$volid]); |
1148 | ||
6c871c36 DM |
1149 | my $path = PVE::Storage::path($storage_cfg, $volid); |
1150 | ||
1151 | my ($vtype, undef, undef, undef, undef, $isBase, $format) = | |
1152 | PVE::Storage::parse_volname($storage_cfg, $volid); | |
1153 | ||
1154 | die "cannot format volume '$volid' (format == $format)\n" | |
1155 | if $format ne 'raw'; | |
1156 | ||
d216e891 | 1157 | mkfs($path, $rootuid, $rootgid); |
6c871c36 DM |
1158 | } |
1159 | ||
1160 | sub destroy_disks { | |
1161 | my ($storecfg, $vollist) = @_; | |
1162 | ||
1163 | foreach my $volid (@$vollist) { | |
1164 | eval { PVE::Storage::vdisk_free($storecfg, $volid); }; | |
1165 | warn $@ if $@; | |
1166 | } | |
1167 | } | |
1168 | ||
1169 | sub create_disks { | |
1170 | my ($storecfg, $vmid, $settings, $conf) = @_; | |
1171 | ||
1172 | my $vollist = []; | |
1173 | ||
1174 | eval { | |
d216e891 WB |
1175 | my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf); |
1176 | my $chown_vollist = []; | |
1177 | ||
d250604f | 1178 | PVE::LXC::Config->foreach_mountpoint($settings, sub { |
6c871c36 DM |
1179 | my ($ms, $mountpoint) = @_; |
1180 | ||
1181 | my $volid = $mountpoint->{volume}; | |
1182 | my $mp = $mountpoint->{mp}; | |
1183 | ||
1184 | my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1); | |
1185 | ||
e2007ac2 | 1186 | if ($storage && ($volid =~ m/^([^:\s]+):(\d+(\.\d+)?)$/)) { |
8ed5ff9d | 1187 | my ($storeid, $size_gb) = ($1, $2); |
6c871c36 | 1188 | |
8ed5ff9d | 1189 | my $size_kb = int(${size_gb}*1024) * 1024; |
6c871c36 DM |
1190 | |
1191 | my $scfg = PVE::Storage::storage_config($storecfg, $storage); | |
1192 | # fixme: use better naming ct-$vmid-disk-X.raw? | |
1193 | ||
1194 | if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs') { | |
8ed5ff9d | 1195 | if ($size_kb > 0) { |
6c871c36 | 1196 | $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', |
8ed5ff9d | 1197 | undef, $size_kb); |
d216e891 | 1198 | format_disk($storecfg, $volid, $rootuid, $rootgid); |
6c871c36 DM |
1199 | } else { |
1200 | $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol', | |
1201 | undef, 0); | |
d216e891 | 1202 | push @$chown_vollist, $volid; |
6c871c36 DM |
1203 | } |
1204 | } elsif ($scfg->{type} eq 'zfspool') { | |
1205 | ||
1206 | $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol', | |
8ed5ff9d | 1207 | undef, $size_kb); |
d216e891 | 1208 | push @$chown_vollist, $volid; |
2e879877 | 1209 | } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' || $scfg->{type} eq 'lvmthin') { |
6c871c36 | 1210 | |
8ed5ff9d | 1211 | $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb); |
d216e891 | 1212 | format_disk($storecfg, $volid, $rootuid, $rootgid); |
6c871c36 DM |
1213 | |
1214 | } elsif ($scfg->{type} eq 'rbd') { | |
1215 | ||
1216 | die "krbd option must be enabled on storage type '$scfg->{type}'\n" if !$scfg->{krbd}; | |
8ed5ff9d | 1217 | $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb); |
d216e891 | 1218 | format_disk($storecfg, $volid, $rootuid, $rootgid); |
6c871c36 DM |
1219 | } else { |
1220 | die "unable to create containers on storage type '$scfg->{type}'\n"; | |
1221 | } | |
1222 | push @$vollist, $volid; | |
71c780b9 WB |
1223 | $mountpoint->{volume} = $volid; |
1224 | $mountpoint->{size} = $size_kb * 1024; | |
1b4cf758 | 1225 | $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs'); |
6c871c36 | 1226 | } else { |
e2007ac2 | 1227 | # use specified/existing volid/dir/device |
1b4cf758 | 1228 | $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs'); |
6c871c36 DM |
1229 | } |
1230 | }); | |
d216e891 WB |
1231 | |
1232 | PVE::Storage::activate_volumes($storecfg, $chown_vollist, undef); | |
1233 | foreach my $volid (@$chown_vollist) { | |
1234 | my $path = PVE::Storage::path($storecfg, $volid, undef); | |
1235 | chown($rootuid, $rootgid, $path); | |
1236 | } | |
1237 | PVE::Storage::deactivate_volumes($storecfg, $chown_vollist, undef); | |
6c871c36 DM |
1238 | }; |
1239 | # free allocated images on error | |
1240 | if (my $err = $@) { | |
1241 | destroy_disks($storecfg, $vollist); | |
1242 | die $err; | |
1243 | } | |
1244 | return $vollist; | |
1245 | } | |
1246 | ||
68e8f3c5 DM |
1247 | # bash completion helper |
1248 | ||
1249 | sub complete_os_templates { | |
1250 | my ($cmdname, $pname, $cvalue) = @_; | |
1251 | ||
1252 | my $cfg = PVE::Storage::config(); | |
1253 | ||
9e9bc3a6 | 1254 | my $storeid; |
68e8f3c5 DM |
1255 | |
1256 | if ($cvalue =~ m/^([^:]+):/) { | |
1257 | $storeid = $1; | |
1258 | } | |
1259 | ||
1260 | my $vtype = $cmdname eq 'restore' ? 'backup' : 'vztmpl'; | |
1261 | my $data = PVE::Storage::template_list($cfg, $storeid, $vtype); | |
1262 | ||
1263 | my $res = []; | |
1264 | foreach my $id (keys %$data) { | |
1265 | foreach my $item (@{$data->{$id}}) { | |
1266 | push @$res, $item->{volid} if defined($item->{volid}); | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | return $res; | |
1271 | } | |
1272 | ||
68e8f3c5 DM |
1273 | my $complete_ctid_full = sub { |
1274 | my ($running) = @_; | |
1275 | ||
1276 | my $idlist = vmstatus(); | |
1277 | ||
1278 | my $active_hash = list_active_containers(); | |
1279 | ||
1280 | my $res = []; | |
1281 | ||
1282 | foreach my $id (keys %$idlist) { | |
1283 | my $d = $idlist->{$id}; | |
1284 | if (defined($running)) { | |
1285 | next if $d->{template}; | |
1286 | next if $running && !$active_hash->{$id}; | |
1287 | next if !$running && $active_hash->{$id}; | |
1288 | } | |
1289 | push @$res, $id; | |
1290 | ||
1291 | } | |
1292 | return $res; | |
1293 | }; | |
1294 | ||
1295 | sub complete_ctid { | |
1296 | return &$complete_ctid_full(); | |
1297 | } | |
1298 | ||
1299 | sub complete_ctid_stopped { | |
1300 | return &$complete_ctid_full(0); | |
1301 | } | |
1302 | ||
1303 | sub complete_ctid_running { | |
1304 | return &$complete_ctid_full(1); | |
1305 | } | |
1306 | ||
c6a605f9 WB |
1307 | sub parse_id_maps { |
1308 | my ($conf) = @_; | |
1309 | ||
1310 | my $id_map = []; | |
1311 | my $rootuid = 0; | |
1312 | my $rootgid = 0; | |
1313 | ||
1314 | my $lxc = $conf->{lxc}; | |
1315 | foreach my $entry (@$lxc) { | |
1316 | my ($key, $value) = @$entry; | |
1317 | next if $key ne 'lxc.id_map'; | |
1318 | if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) { | |
1319 | my ($type, $ct, $host, $length) = ($1, $2, $3, $4); | |
1320 | push @$id_map, [$type, $ct, $host, $length]; | |
1321 | if ($ct == 0) { | |
1322 | $rootuid = $host if $type eq 'u'; | |
1323 | $rootgid = $host if $type eq 'g'; | |
1324 | } | |
1325 | } else { | |
1326 | die "failed to parse id_map: $value\n"; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | if (!@$id_map && $conf->{unprivileged}) { | |
1331 | # Should we read them from /etc/subuid? | |
1332 | $id_map = [ ['u', '0', '100000', '65536'], | |
1333 | ['g', '0', '100000', '65536'] ]; | |
1334 | $rootuid = $rootgid = 100000; | |
1335 | } | |
1336 | ||
1337 | return ($id_map, $rootuid, $rootgid); | |
1338 | } | |
1339 | ||
01dce99b WB |
1340 | sub userns_command { |
1341 | my ($id_map) = @_; | |
1342 | if (@$id_map) { | |
1343 | return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--']; | |
1344 | } | |
1345 | return []; | |
1346 | } | |
1347 | ||
846a66b0 | 1348 | |
f76a2828 | 1349 | 1; |