src/PVE/LXC.pm

   1 package PVE::LXC;
   2
   3 use strict;
   4 use warnings;
   5
   6 use POSIX qw(EINTR);
   7
   8 use Socket;
   9
  10 use File::Path;
  11 use File::Spec;
  12 use Cwd qw();
  13 use Fcntl qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY);
  14 use Errno qw(ELOOP ENOTDIR EROFS ECONNREFUSED ENOSYS EEXIST);
  15 use IO::Socket::UNIX;
  16
  17 use PVE::Exception qw(raise_perm_exc);
  18 use PVE::Storage;
  19 use PVE::SafeSyslog;
  20 use PVE::INotify;
  21 use PVE::JSONSchema qw(get_standard_option);
  22 use PVE::Tools qw(
  23     dir_glob_foreach file_get_contents file_set_contents lock_file
  24     lock_file_full AT_FDCWD O_PATH $IPV4RE $IPV6RE
  25 );
  26 use PVE::CpuSet;
  27 use PVE::Network;
  28 use PVE::AccessControl;
  29 use PVE::ProcFSTools;
  30 use PVE::Syscall qw(:fsmount);
  31 use PVE::LXC::Config;
  32 use PVE::GuestHelpers;
  33 use PVE::LXC::Tools;
  34
  35 use Time::HiRes qw (gettimeofday);
  36
  37 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
  38
  39 my $nodename = PVE::INotify::nodename();
  40
  41 my $cpuinfo= PVE::ProcFSTools::read_cpuinfo();
  42
  43 sub config_list {
  44     my $vmlist = PVE::Cluster::get_vmlist();
  45     my $res = {};
  46     return $res if !$vmlist || !$vmlist->{ids};
  47     my $ids = $vmlist->{ids};
  48
  49     foreach my $vmid (keys %$ids) {
  50         next if !$vmid; # skip CT0
  51         my $d = $ids->{$vmid};
  52         next if !$d->{node} || $d->{node} ne $nodename;
  53         next if !$d->{type} || $d->{type} ne 'lxc';
  54         $res->{$vmid} = { type => 'lxc', vmid => $vmid };
  55     }
  56     return $res;
  57 }
  58
  59 # container status helpers
  60
  61 sub list_active_containers {
  62
  63     my $filename = "/proc/net/unix";
  64
  65     # similar test is used by lcxcontainers.c: list_active_containers
  66     my $res = {};
  67
  68     my $fh = IO::File->new ($filename, "r");
  69     return $res if !$fh;
  70
  71     while (defined(my $line = <$fh>)) {
  72         if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
  73             my $path = $1;
  74             if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
  75                 $res->{$1} = 1;
  76             }
  77         }
  78     }
  79
  80     close($fh);
  81
  82     return $res;
  83 }
  84
  85 # warning: this is slow
  86 sub check_running {
  87     my ($vmid) = @_;
  88
  89     my $active_hash = list_active_containers();
  90
  91     return 1 if defined($active_hash->{$vmid});
  92
  93     return undef;
  94 }
  95
  96 sub get_container_disk_usage {
  97     my ($vmid, $pid) = @_;
  98
  99     return PVE::Tools::df("/proc/$pid/root/", 1);
 100 }
 101
 102 my $last_proc_vmid_stat;
 103
 104 my $parse_cpuacct_stat = sub {
 105     my ($vmid, $unprivileged) = @_;
 106
 107     my $raw = read_cgroup_value('cpuacct', $vmid, $unprivileged, 'cpuacct.stat', 1);
 108
 109     my $stat = {};
 110
 111     if ($raw =~ m/^user (\d+)\nsystem (\d+)\n/) {
 112
 113         $stat->{utime} = $1;
 114         $stat->{stime} = $2;
 115
 116     }
 117
 118     return $stat;
 119 };
 120
 121 our $vmstatus_return_properties = {
 122     vmid => get_standard_option('pve-vmid'),
 123     status => {
 124         description => "LXC Container status.",
 125         type => 'string',
 126         enum => ['stopped', 'running'],
 127     },
 128     maxmem => {
 129         description => "Maximum memory in bytes.",
 130         type => 'integer',
 131         optional => 1,
 132         renderer => 'bytes',
 133     },
 134     maxswap => {
 135         description => "Maximum SWAP memory in bytes.",
 136         type => 'integer',
 137         optional => 1,
 138         renderer => 'bytes',
 139     },
 140     maxdisk => {
 141         description => "Root disk size in bytes.",
 142         type => 'integer',
 143         optional => 1,
 144         renderer => 'bytes',
 145     },
 146     name => {
 147         description => "Container name.",
 148         type => 'string',
 149         optional => 1,
 150     },
 151     uptime => {
 152         description => "Uptime.",
 153         type => 'integer',
 154         optional => 1,
 155         renderer => 'duration',
 156     },
 157     cpus => {
 158         description => "Maximum usable CPUs.",
 159         type => 'number',
 160         optional => 1,
 161     },
 162     lock => {
 163         description => "The current config lock, if any.",
 164         type => 'string',
 165         optional => 1,
 166     },
 167     tags => {
 168         description => "The current configured tags, if any.",
 169         type => 'string',
 170         optional => 1,
 171     }
 172 };
 173
 174 sub vmstatus {
 175     my ($opt_vmid) = @_;
 176
 177     my $list = $opt_vmid ? { $opt_vmid => { type => 'lxc', vmid => $opt_vmid }} : config_list();
 178
 179     my $active_hash = list_active_containers();
 180
 181     my $cpucount = $cpuinfo->{cpus} || 1;
 182
 183     my $cdtime = gettimeofday;
 184
 185     my $uptime = (PVE::ProcFSTools::read_proc_uptime(1))[0];
 186     my $clock_ticks = POSIX::sysconf(&POSIX::_SC_CLK_TCK);
 187
 188     my $unprivileged = {};
 189
 190     foreach my $vmid (keys %$list) {
 191         my $d = $list->{$vmid};
 192
 193         eval { $d->{pid} = find_lxc_pid($vmid) if defined($active_hash->{$vmid}); };
 194         warn $@ if $@; # ignore errors (consider them stopped)
 195
 196         $d->{status} = $active_hash->{$vmid} ? 'running' : 'stopped';
 197
 198         my $cfspath = PVE::LXC::Config->cfs_config_path($vmid);
 199         my $conf = PVE::Cluster::cfs_read_file($cfspath) || {};
 200
 201         $unprivileged->{$vmid} = $conf->{unprivileged};
 202
 203         $d->{name} = $conf->{'hostname'} || "CT$vmid";
 204         $d->{name} =~ s/[\s]//g;
 205
 206         $d->{cpus} = $conf->{cores} || $conf->{cpulimit};
 207         $d->{cpus} = $cpucount if !$d->{cpus};
 208
 209         $d->{lock} = $conf->{lock} || '';
 210         $d->{tags} = $conf->{tags} if defined($conf->{tags});
 211
 212         if ($d->{pid}) {
 213             my $res = get_container_disk_usage($vmid, $d->{pid});
 214             $d->{disk} = $res->{used};
 215             $d->{maxdisk} = $res->{total};
 216         } else {
 217             $d->{disk} = 0;
 218             # use 4GB by default ??
 219             if (my $rootfs = $conf->{rootfs}) {
 220                 my $rootinfo = PVE::LXC::Config->parse_ct_rootfs($rootfs);
 221                 $d->{maxdisk} = $rootinfo->{size} || (4*1024*1024*1024);
 222             } else {
 223                 $d->{maxdisk} = 4*1024*1024*1024;
 224             }
 225         }
 226
 227         $d->{mem} = 0;
 228         $d->{swap} = 0;
 229         $d->{maxmem} = ($conf->{memory}||512)*1024*1024;
 230         $d->{maxswap} = ($conf->{swap}//0)*1024*1024;
 231
 232         $d->{uptime} = 0;
 233         $d->{cpu} = 0;
 234
 235         $d->{netout} = 0;
 236         $d->{netin} = 0;
 237
 238         $d->{diskread} = 0;
 239         $d->{diskwrite} = 0;
 240
 241         $d->{template} = PVE::LXC::Config->is_template($conf);
 242         $d->{lock} = $conf->{lock} if $conf->{lock};
 243     }
 244
 245     foreach my $vmid (keys %$list) {
 246         my $d = $list->{$vmid};
 247         my $pid = $d->{pid};
 248
 249         next if !$pid; # skip stopped CTs
 250
 251         my $proc_pid_stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 252         $d->{uptime} = int(($uptime - $proc_pid_stat->{starttime}) / $clock_ticks); # the method lxcfs uses
 253
 254         my $unpriv = $unprivileged->{$vmid};
 255
 256         if (-d '/sys/fs/cgroup/memory') {
 257             my $memory_stat = read_cgroup_list('memory', $vmid, $unpriv, 'memory.stat');
 258             my $mem_usage_in_bytes = read_cgroup_value('memory', $vmid, $unpriv, 'memory.usage_in_bytes');
 259
 260             $d->{mem} = $mem_usage_in_bytes - $memory_stat->{total_cache};
 261             $d->{swap} = read_cgroup_value('memory', $vmid, $unpriv, 'memory.memsw.usage_in_bytes') - $mem_usage_in_bytes;
 262         } else {
 263             $d->{mem} = 0;
 264             $d->{swap} = 0;
 265         }
 266
 267         if (-d '/sys/fs/cgroup/blkio') {
 268             my $blkio_bytes = read_cgroup_value('blkio', $vmid, 0, 'blkio.throttle.io_service_bytes', 1); # don't check if unpriv
 269             my @bytes = split(/\n/, $blkio_bytes);
 270             foreach my $byte (@bytes) {
 271                 if (my ($key, $value) = $byte =~ /(Read|Write)\s+(\d+)/) {
 272                     $d->{diskread} += $2 if $key eq 'Read';
 273                     $d->{diskwrite} += $2 if $key eq 'Write';
 274                 }
 275             }
 276         } else {
 277             $d->{diskread} = 0;
 278             $d->{diskwrite} = 0;
 279         }
 280
 281         if (-d '/sys/fs/cgroup/cpuacct') {
 282             my $pstat = $parse_cpuacct_stat->($vmid, $unpriv);
 283
 284             my $used = $pstat->{utime} + $pstat->{stime};
 285
 286             my $old = $last_proc_vmid_stat->{$vmid};
 287             if (!$old) {
 288                 $last_proc_vmid_stat->{$vmid} = {
 289                     time => $cdtime,
 290                     used => $used,
 291                     cpu => 0,
 292                 };
 293                 next;
 294             }
 295
 296             my $dtime = ($cdtime -  $old->{time}) * $cpucount * $cpuinfo->{user_hz};
 297
 298             if ($dtime > 1000) {
 299                 my $dutime = $used -  $old->{used};
 300
 301                 $d->{cpu} = (($dutime/$dtime)* $cpucount) / $d->{cpus};
 302                 $last_proc_vmid_stat->{$vmid} = {
 303                     time => $cdtime,
 304                     used => $used,
 305                     cpu => $d->{cpu},
 306                 };
 307             } else {
 308                 $d->{cpu} = $old->{cpu};
 309             }
 310         } else {
 311             $d->{cpu} = 0;
 312         }
 313     }
 314
 315     my $netdev = PVE::ProcFSTools::read_proc_net_dev();
 316
 317     foreach my $dev (keys %$netdev) {
 318         next if $dev !~ m/^veth([1-9]\d*)i/;
 319         my $vmid = $1;
 320         my $d = $list->{$vmid};
 321
 322         next if !$d;
 323
 324         $d->{netout} += $netdev->{$dev}->{receive};
 325         $d->{netin} += $netdev->{$dev}->{transmit};
 326
 327     }
 328
 329     return $list;
 330 }
 331
 332 sub read_cgroup_list($$$$) {
 333     my ($group, $vmid, $unprivileged, $name) = @_;
 334
 335     my $content = read_cgroup_value($group, $vmid, $unprivileged, $name, 1);
 336
 337     return { split(/\s+/, $content) };
 338 }
 339
 340 sub read_cgroup_value($$$$$) {
 341     my ($group, $vmid, $unprivileged, $name, $full) = @_;
 342
 343     my $nsdir = $unprivileged ? '' : 'ns/';
 344     my $path = "/sys/fs/cgroup/$group/lxc/$vmid/${nsdir}$name";
 345
 346     return PVE::Tools::file_get_contents($path) if $full;
 347
 348     return PVE::Tools::file_read_firstline($path);
 349 }
 350
 351 sub write_cgroup_value {
 352    my ($group, $vmid, $name, $value) = @_;
 353
 354    my $path = "/sys/fs/cgroup/$group/lxc/$vmid/$name";
 355    PVE::ProcFSTools::write_proc_entry($path, $value) if -e $path;
 356
 357 }
 358
 359 sub find_lxc_console_pids {
 360
 361     my $res = {};
 362
 363     PVE::Tools::dir_glob_foreach('/proc', '\d+', sub {
 364         my ($pid) = @_;
 365
 366         my $cmdline = PVE::Tools::file_read_firstline("/proc/$pid/cmdline");
 367         return if !$cmdline;
 368
 369         my @args = split(/\0/, $cmdline);
 370
 371         # search for lxc-console -n <vmid>
 372         return if scalar(@args) != 3;
 373         return if $args[1] ne '-n';
 374         return if $args[2] !~ m/^\d+$/;
 375         return if $args[0] !~ m|^(/usr/bin/)?lxc-console$|;
 376
 377         my $vmid = $args[2];
 378
 379         push @{$res->{$vmid}}, $pid;
 380     });
 381
 382     return $res;
 383 }
 384
 385 sub find_lxc_pid {
 386     my ($vmid) = @_;
 387
 388     my $pid = undef;
 389     my $parser = sub {
 390         my $line = shift;
 391         $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
 392     };
 393     PVE::Tools::run_command(['lxc-info', '-n', $vmid, '-p'], outfunc => $parser);
 394
 395     die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
 396
 397     return $pid;
 398 }
 399
 400 sub open_pid_fd($) {
 401     my ($pid) = @_;
 402     sysopen(my $fd, "/proc/$pid", O_RDONLY | O_DIRECTORY)
 403         or die "failed to open /proc/$pid pid fd\n";
 404     return $fd;
 405 }
 406
 407 sub open_lxc_pid {
 408     my ($vmid) = @_;
 409
 410     # Find the pid and open:
 411     my $pid = find_lxc_pid($vmid);
 412     my $fd = open_pid_fd($pid);
 413
 414     # Verify:
 415     my $pid2 = find_lxc_pid($vmid);
 416
 417     return () if $pid != $pid2;
 418     return ($pid, $fd);
 419 }
 420
 421 sub open_ppid {
 422     my ($pid) = @_;
 423
 424     # Find the parent pid via proc and open it:
 425     my $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 426     my $ppid = $stat->{ppid} // die "failed to get parent pid\n";
 427
 428     my $fd = open_pid_fd($ppid);
 429
 430     # Verify:
 431     $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 432     my $ppid2 = $stat->{ppid} // die "failed to get parent pid for verification\n";
 433
 434     return () if $ppid != $ppid2;
 435     return ($ppid, $fd);
 436 }
 437
 438 # Note: we cannot use Net:IP, because that only allows strict
 439 # CIDR networks
 440 sub parse_ipv4_cidr {
 441     my ($cidr, $noerr) = @_;
 442
 443     if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) &&  ($2 <= 32)) {
 444         return { address => $1, netmask => $PVE::Network::ipv4_reverse_mask->[$2] };
 445     }
 446
 447     return undef if $noerr;
 448
 449     die "unable to parse ipv4 address/mask\n";
 450 }
 451
 452 sub get_cgroup_subsystems {
 453         my $v1 = {};
 454         my $v2 = 0;
 455         my $data = PVE::Tools::file_get_contents('/proc/self/cgroup');
 456         while ($data =~ /^\d+:([^:\n]*):.*$/gm) {
 457                 my $type = $1;
 458                 if (length($type)) {
 459                         $v1->{$_} = 1 foreach split(/,/, $type);
 460                 } else {
 461                         $v2 = 1;
 462                 }
 463         }
 464         return wantarray ? ($v1, $v2) : $v1;
 465 }
 466
 467 # With seccomp trap to userspace we now have the ability to optionally forward
 468 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
 469 #
 470 # This means that there are cases where we need to create an extra seccomp
 471 # profile for the container to load.
 472 #
 473 # This returns a configuration snippet added to the raw lxc config.
 474 sub make_seccomp_config {
 475     my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
 476     # User-configured profile has precedence, note that the user's entry would
 477     # be written 'after' this line anyway...
 478     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.seccomp.profile')) {
 479         # Warn the user if this conflicts with a feature:
 480         my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
 481         warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
 482             if length($warn) > 0;
 483         return '';
 484     }
 485
 486     # Privileged containers keep using the default (which is already part of
 487     # the files included via lxc.include, so we don't need to write it out,
 488     # that way it stays admin-configurable via /usr/share/lxc/config/... as
 489     # well)
 490     return '' if !$unprivileged;
 491
 492     my $rules = {
 493         keyctl => ['errno 38'],
 494     };
 495
 496     my $raw_conf = '';
 497
 498     # Unprivileged containers will get keyctl() disabled by default as a
 499     # workaround for systemd-networkd behavior. But we have an option to
 500     # explicitly enable it:
 501     if ($features->{keyctl}) {
 502         delete $rules->{keyctl};
 503     }
 504
 505     # By default, unprivileged containers cannot use `mknod` at all.
 506     # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
 507     # but for now this is experimental, so it has to be enabled via a feature
 508     # flag.
 509     # Note that we only handle block and char devices (like lxd), the rest we
 510     # leave up to the kernel. We may in the future remove this if seccomp gets
 511     # a way to tell the kernel to "continue" a syscall.
 512     if ($features->{mknod}) {
 513         my ($ok, $kernel) = PVE::ProcFSTools::check_kernel_release(5, 3);
 514         if (!$ok) {
 515             die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
 516         }
 517
 518         $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
 519         $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
 520
 521         $rules->{mknod} = [
 522             # condition: (mode & S_IFMT) == S_IFCHR
 523             'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
 524             # condition: (mode & S_IFMT) == S_IFBLK
 525             'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
 526         ];
 527         $rules->{mknodat} = [
 528             # condition: (mode & S_IFMT) == S_IFCHR
 529             'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
 530             # condition: (mode & S_IFMT) == S_IFBLK
 531             'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
 532         ];
 533     }
 534
 535     # Now build the custom seccomp rule text...
 536     my $extra_rules = join("\n", map {
 537         my $syscall = $_;
 538         map { "$syscall $_" } $rules->{$syscall}->@*
 539     } sort keys %$rules) . "\n";
 540
 541     return $raw_conf if $extra_rules eq "\n";
 542
 543     # We still have the "most common" config readily available, so don't write
 544     # out that one:
 545     if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
 546         # we have no extra $raw_conf and use the same we had in pve 6.1:
 547         return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
 548     }
 549
 550     # Write the rule file to the container's config path:
 551     my $rule_file = "$conf_dir/rules.seccomp";
 552     my $rule_data = file_get_contents("$LXC_CONFIG_PATH/common.seccomp")
 553         . $extra_rules;
 554     file_set_contents($rule_file, $rule_data);
 555     $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
 556
 557     return $raw_conf;
 558 }
 559
 560 # Since lxc-3.0.2 we can have lxc generate a profile for the container
 561 # automatically. The default should be equivalent to the old
 562 # `lxc-container-default-cgns` profile.
 563 #
 564 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
 565 # additional lines into the profile. We can use that to allow mounting specific
 566 # file systems.
 567 sub make_apparmor_config {
 568     my ($conf, $unprivileged, $features) = @_;
 569
 570     # user-configured profile has precedence, but first we go through our own
 571     # code to figure out whether we should warn the user:
 572
 573     my $raw = "lxc.apparmor.profile = generated\n";
 574     my @profile_uses;
 575
 576     if ($features->{fuse}) {
 577         # For the informational warning:
 578         push @profile_uses, 'features:fuse';
 579     }
 580
 581     # There's lxc.apparmor.allow_nesting now, which will add the necessary
 582     # apparmor lines, create an apparmor namespace for the container, but also
 583     # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
 584     # lxcfs mounted over them, because that would prevent the container from
 585     # mounting new instances of them for nested containers.
 586     if ($features->{nesting}) {
 587         push @profile_uses, 'features:nesting';
 588         $raw .= "lxc.apparmor.allow_nesting = 1\n"
 589     } else {
 590         # In the default profile in /etc/apparmor.d we patch this in because
 591         # otherwise a container can for example run `chown` on /sys, breaking
 592         # access to it for non-CAP_DAC_OVERRIDE tools on the host:
 593         $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
 594         $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
 595         # Preferably we could use the 'remount' flag but this does not sit well
 596         # with apparmor_parser currently:
 597         #  mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
 598     }
 599
 600     if (my $mount = $features->{mount}) {
 601         push @profile_uses, 'features:mount';
 602         foreach my $fs (PVE::Tools::split_list($mount)) {
 603             $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
 604         }
 605     }
 606
 607     # More to come?
 608
 609     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.apparmor.profile')) {
 610         if (length(my $used = join(', ', @profile_uses))) {
 611             warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
 612         }
 613         return '';
 614     }
 615
 616     return $raw;
 617 }
 618
 619 sub update_lxc_config {
 620     my ($vmid, $conf) = @_;
 621
 622     my $dir = "/var/lib/lxc/$vmid";
 623
 624     if ($conf->{template}) {
 625
 626         unlink "$dir/config";
 627
 628         return;
 629     }
 630
 631     my $raw = '';
 632
 633     die "missing 'arch' - internal error" if !$conf->{arch};
 634     $raw .= "lxc.arch = $conf->{arch}\n";
 635
 636     my $custom_idmap = PVE::LXC::Config->has_lxc_entry($conf, 'lxc.idmap');
 637     my $unprivileged = $conf->{unprivileged} || $custom_idmap;
 638
 639     my $ostype = $conf->{ostype} || die "missing 'ostype' - internal error";
 640
 641     my $cfgpath = '/usr/share/lxc/config';
 642     my $inc = "$cfgpath/$ostype.common.conf";
 643     $inc ="$cfgpath/common.conf" if !-f $inc;
 644     $raw .= "lxc.include = $inc\n";
 645     if ($unprivileged) {
 646         $inc = "$cfgpath/$ostype.userns.conf";
 647         $inc = "$cfgpath/userns.conf" if !-f $inc;
 648         $raw .= "lxc.include = $inc\n";
 649     }
 650
 651     my $features = PVE::LXC::Config->parse_features($conf->{features});
 652
 653     $raw .= make_seccomp_config($conf, $vmid, $dir, $unprivileged, $features);
 654     $raw .= make_apparmor_config($conf, $unprivileged, $features);
 655     if ($features->{fuse}) {
 656         $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
 657         $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
 658     }
 659
 660     # WARNING: DO NOT REMOVE this without making sure that loop device nodes
 661     # cannot be exposed to the container with r/w access (cgroup perms).
 662     # When this is enabled mounts will still remain in the monitor's namespace
 663     # after the container unmounted them and thus will not detach from their
 664     # files while the container is running!
 665     $raw .= "lxc.monitor.unshare = 1\n";
 666
 667     my $cgv1 = get_cgroup_subsystems();
 668
 669     # Should we read them from /etc/subuid?
 670     if ($unprivileged && !$custom_idmap) {
 671         $raw .= "lxc.idmap = u 0 100000 65536\n";
 672         $raw .= "lxc.idmap = g 0 100000 65536\n";
 673     }
 674
 675     if (!PVE::LXC::Config->has_dev_console($conf)) {
 676         $raw .= "lxc.console.path = none\n";
 677         $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n" if $cgv1->{devices};
 678     }
 679
 680     my $ttycount = PVE::LXC::Config->get_tty_count($conf);
 681     $raw .= "lxc.tty.max = $ttycount\n";
 682
 683     # some init scripts expect a linux terminal (turnkey).
 684     $raw .= "lxc.environment = TERM=linux\n";
 685
 686     my $utsname = $conf->{hostname} || "CT$vmid";
 687     $raw .= "lxc.uts.name = $utsname\n";
 688
 689     if ($cgv1->{memory}) {
 690         my $memory = $conf->{memory} || 512;
 691         my $swap = $conf->{swap} // 0;
 692
 693         my $lxcmem = int($memory*1024*1024);
 694         $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
 695
 696         my $lxcswap = int(($memory + $swap)*1024*1024);
 697         $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
 698     }
 699
 700     if ($cgv1->{cpu}) {
 701         if (my $cpulimit = $conf->{cpulimit}) {
 702             $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
 703             my $value = int(100000*$cpulimit);
 704             $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
 705         }
 706
 707         my $shares = $conf->{cpuunits} || 1024;
 708         $raw .= "lxc.cgroup.cpu.shares = $shares\n";
 709     }
 710
 711     die "missing 'rootfs' configuration\n"
 712         if !defined($conf->{rootfs});
 713
 714     my $mountpoint = PVE::LXC::Config->parse_ct_rootfs($conf->{rootfs});
 715
 716     $raw .= "lxc.rootfs.path = $dir/rootfs\n";
 717
 718     foreach my $k (sort keys %$conf) {
 719         next if $k !~ m/^net(\d+)$/;
 720         my $ind = $1;
 721         my $d = PVE::LXC::Config->parse_lxc_network($conf->{$k});
 722         $raw .= "lxc.net.$ind.type = veth\n";
 723         $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
 724         $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr});
 725         $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name});
 726         $raw .= "lxc.net.$ind.mtu = $d->{mtu}\n" if defined($d->{mtu});
 727     }
 728
 729     if ($cgv1->{cpuset}) {
 730         my $had_cpuset = 0;
 731         if (my $lxcconf = $conf->{lxc}) {
 732             foreach my $entry (@$lxcconf) {
 733                 my ($k, $v) = @$entry;
 734                 $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus';
 735                 $raw .= "$k = $v\n";
 736             }
 737         }
 738
 739         my $cores = $conf->{cores};
 740         if (!$had_cpuset && $cores) {
 741             my $cpuset = eval { PVE::CpuSet->new_from_cgroup('lxc', 'effective_cpus') };
 742             $cpuset = PVE::CpuSet->new_from_cgroup('', 'effective_cpus') if !$cpuset;
 743             my @members = $cpuset->members();
 744             while (scalar(@members) > $cores) {
 745                 my $randidx = int(rand(scalar(@members)));
 746                 $cpuset->delete($members[$randidx]);
 747                 splice(@members, $randidx, 1); # keep track of the changes
 748             }
 749             $raw .= "lxc.cgroup.cpuset.cpus = ".$cpuset->short_string()."\n";
 750         }
 751     }
 752
 753     File::Path::mkpath("$dir/rootfs");
 754
 755     PVE::Tools::file_set_contents("$dir/config", $raw);
 756 }
 757
 758 # verify and cleanup nameserver list (replace \0 with ' ')
 759 sub verify_nameserver_list {
 760     my ($nameserver_list) = @_;
 761
 762     my @list = ();
 763     foreach my $server (PVE::Tools::split_list($nameserver_list)) {
 764         PVE::JSONSchema::pve_verify_ip($server);
 765         push @list, $server;
 766     }
 767
 768     return join(' ', @list);
 769 }
 770
 771 sub verify_searchdomain_list {
 772     my ($searchdomain_list) = @_;
 773
 774     my @list = ();
 775     foreach my $server (PVE::Tools::split_list($searchdomain_list)) {
 776         # todo: should we add checks for valid dns domains?
 777         push @list, $server;
 778     }
 779
 780     return join(' ', @list);
 781 }
 782
 783 sub get_console_command {
 784     my ($vmid, $conf, $escapechar) = @_;
 785
 786     # '-1' as $escapechar disables keyboard escape sequence
 787     # any other passed char (a-z) will result in <Ctrl+$escapechar q>
 788
 789     my $cmode = PVE::LXC::Config->get_cmode($conf);
 790
 791     my $cmd = [];
 792     if ($cmode eq 'console') {
 793         push @$cmd, 'lxc-console', '-n',  $vmid, '-t', 0;
 794         push @$cmd, '-e', $escapechar if $escapechar;
 795     } elsif ($cmode eq 'tty') {
 796         push @$cmd, 'lxc-console', '-n',  $vmid;
 797         push @$cmd, '-e', $escapechar if $escapechar;
 798     } elsif ($cmode eq 'shell') {
 799         push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
 800     } else {
 801         die "internal error";
 802     }
 803
 804     return $cmd;
 805 }
 806
 807 sub get_primary_ips {
 808     my ($conf) = @_;
 809
 810     # return data from net0
 811
 812     return undef if !defined($conf->{net0});
 813     my $net = PVE::LXC::Config->parse_lxc_network($conf->{net0});
 814
 815     my $ipv4 = $net->{ip};
 816     if ($ipv4) {
 817         if ($ipv4 =~ /^(dhcp|manual)$/) {
 818             $ipv4 = undef
 819         } else {
 820             $ipv4 =~ s!/\d+$!!;
 821         }
 822     }
 823     my $ipv6 = $net->{ip6};
 824     if ($ipv6) {
 825         if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
 826             $ipv6 = undef;
 827         } else {
 828             $ipv6 =~ s!/\d+$!!;
 829         }
 830     }
 831
 832     return ($ipv4, $ipv6);
 833 }
 834
 835 sub delete_mountpoint_volume {
 836     my ($storage_cfg, $vmid, $volume) = @_;
 837
 838     return if PVE::LXC::Config->classify_mountpoint($volume) ne 'volume';
 839
 840     my ($vtype, $name, $owner) = PVE::Storage::parse_volname($storage_cfg, $volume);
 841
 842     if ($vmid == $owner) {
 843         PVE::Storage::vdisk_free($storage_cfg, $volume);
 844     } else {
 845         warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
 846     }
 847 }
 848
 849 sub destroy_lxc_container {
 850     my ($storage_cfg, $vmid, $conf, $replacement_conf) = @_;
 851
 852     PVE::LXC::Config->foreach_mountpoint($conf, sub {
 853         my ($ms, $mountpoint) = @_;
 854         delete_mountpoint_volume($storage_cfg, $vmid, $mountpoint->{volume});
 855     });
 856
 857     rmdir "/var/lib/lxc/$vmid/rootfs";
 858     unlink "/var/lib/lxc/$vmid/config";
 859     rmdir "/var/lib/lxc/$vmid";
 860     if (defined $replacement_conf) {
 861         PVE::LXC::Config->write_config($vmid, $replacement_conf);
 862     } else {
 863         PVE::LXC::Config->destroy_config($vmid);
 864     }
 865 }
 866
 867 sub vm_stop_cleanup {
 868     my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
 869
 870     return if $keepActive;
 871
 872     eval {
 873         my $vollist = PVE::LXC::Config->get_vm_volumes($conf);
 874         PVE::Storage::deactivate_volumes($storage_cfg, $vollist);
 875     };
 876     warn $@ if $@; # avoid errors - just warn
 877 }
 878
 879 my $safe_num_ne = sub {
 880     my ($a, $b) = @_;
 881
 882     return 0 if !defined($a) && !defined($b);
 883     return 1 if !defined($a);
 884     return 1 if !defined($b);
 885
 886     return $a != $b;
 887 };
 888
 889 my $safe_string_ne = sub {
 890     my ($a, $b) = @_;
 891
 892     return 0 if !defined($a) && !defined($b);
 893     return 1 if !defined($a);
 894     return 1 if !defined($b);
 895
 896     return $a ne $b;
 897 };
 898
 899 sub update_net {
 900     my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
 901
 902     if ($newnet->{type} ne 'veth') {
 903         # for when there are physical interfaces
 904         die "cannot update interface of type $newnet->{type}";
 905     }
 906
 907     my $veth = "veth${vmid}i${netid}";
 908     my $eth = $newnet->{name};
 909
 910     if (my $oldnetcfg = $conf->{$opt}) {
 911         my $oldnet = PVE::LXC::Config->parse_lxc_network($oldnetcfg);
 912
 913         if (&$safe_string_ne($oldnet->{hwaddr}, $newnet->{hwaddr}) ||
 914             &$safe_string_ne($oldnet->{name}, $newnet->{name})) {
 915
 916             PVE::Network::veth_delete($veth);
 917             delete $conf->{$opt};
 918             PVE::LXC::Config->write_config($vmid, $conf);
 919
 920             hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 921
 922         } else {
 923             if (&$safe_string_ne($oldnet->{bridge}, $newnet->{bridge}) ||
 924                 &$safe_num_ne($oldnet->{tag}, $newnet->{tag}) ||
 925                 &$safe_num_ne($oldnet->{firewall}, $newnet->{firewall})) {
 926
 927                 if ($oldnet->{bridge}) {
 928                     PVE::Network::tap_unplug($veth);
 929                     foreach (qw(bridge tag firewall)) {
 930                         delete $oldnet->{$_};
 931                     }
 932                     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 933                     PVE::LXC::Config->write_config($vmid, $conf);
 934                 }
 935
 936                 PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 937                 # This includes the rate:
 938                 foreach (qw(bridge tag firewall rate)) {
 939                     $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
 940                 }
 941             } elsif (&$safe_string_ne($oldnet->{rate}, $newnet->{rate})) {
 942                 # Rate can be applied on its own but any change above needs to
 943                 # include the rate in tap_plug since OVS resets everything.
 944                 PVE::Network::tap_rate_limit($veth, $newnet->{rate});
 945                 $oldnet->{rate} = $newnet->{rate}
 946             }
 947             $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 948             PVE::LXC::Config->write_config($vmid, $conf);
 949         }
 950     } else {
 951         hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 952     }
 953
 954     update_ipconfig($vmid, $conf, $opt, $eth, $newnet, $rootdir);
 955 }
 956
 957 sub hotplug_net {
 958     my ($vmid, $conf, $opt, $newnet, $netid) = @_;
 959
 960     my $veth = "veth${vmid}i${netid}";
 961     my $vethpeer = $veth . "p";
 962     my $eth = $newnet->{name};
 963
 964     PVE::Network::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr});
 965     PVE::Network::tap_plug($veth, $newnet->{bridge}, $newnet->{tag}, $newnet->{firewall}, $newnet->{trunks}, $newnet->{rate});
 966
 967     # attach peer in container
 968     my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
 969     PVE::Tools::run_command($cmd);
 970
 971     # link up peer in container
 972     $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up'  ];
 973     PVE::Tools::run_command($cmd);
 974
 975     my $done = { type => 'veth' };
 976     foreach (qw(bridge tag firewall hwaddr name)) {
 977         $done->{$_} = $newnet->{$_} if $newnet->{$_};
 978     }
 979     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($done);
 980
 981     PVE::LXC::Config->write_config($vmid, $conf);
 982 }
 983
 984 sub update_ipconfig {
 985     my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
 986
 987     my $lxc_setup = PVE::LXC::Setup->new($conf, $rootdir);
 988
 989     my $optdata = PVE::LXC::Config->parse_lxc_network($conf->{$opt});
 990     my $deleted = [];
 991     my $added = [];
 992     my $nscmd = sub {
 993         my $cmdargs = shift;
 994         PVE::Tools::run_command(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
 995     };
 996     my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
 997
 998     my $change_ip_config = sub {
 999         my ($ipversion) = @_;
1000
1001         my $family_opt = "-$ipversion";
1002         my $suffix = $ipversion == 4 ? '' : $ipversion;
1003         my $gw= "gw$suffix";
1004         my $ip= "ip$suffix";
1005
1006         my $newip = $newnet->{$ip};
1007         my $newgw = $newnet->{$gw};
1008         my $oldip = $optdata->{$ip};
1009         my $oldgw = $optdata->{$gw};
1010
1011         my $change_ip = &$safe_string_ne($oldip, $newip);
1012         my $change_gw = &$safe_string_ne($oldgw, $newgw);
1013
1014         return if !$change_ip && !$change_gw;
1015
1016         # step 1: add new IP, if this fails we cancel
1017         my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
1018         if ($change_ip && $is_real_ip) {
1019             eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
1020             if (my $err = $@) {
1021                 warn $err;
1022                 return;
1023             }
1024         }
1025
1026         # step 2: replace gateway
1027         #   If this fails we delete the added IP and cancel.
1028         #   If it succeeds we save the config and delete the old IP, ignoring
1029         #   errors. The config is then saved.
1030         # Note: 'ip route replace' can add
1031         if ($change_gw) {
1032             if ($newgw) {
1033                 eval {
1034                     if ($is_real_ip && !PVE::Network::is_ip_in_cidr($newgw, $newip, $ipversion)) {
1035                         &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1036                     }
1037                     &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1038                 };
1039                 if (my $err = $@) {
1040                     warn $err;
1041                     # the route was not replaced, the old IP is still available
1042                     # rollback (delete new IP) and cancel
1043                     if ($change_ip) {
1044                         eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1045                         warn $@ if $@; # no need to die here
1046                     }
1047                     return;
1048                 }
1049             } else {
1050                 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1051                 # if the route was not deleted, the guest might have deleted it manually
1052                 # warn and continue
1053                 warn $@ if $@;
1054             }
1055             if ($oldgw && $oldip && !PVE::Network::is_ip_in_cidr($oldgw, $oldip)) {
1056                 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1057                 # warn if the route was deleted manually
1058                 warn $@ if $@;
1059             }
1060         }
1061
1062         # from this point on we save the configuration
1063         # step 3: delete old IP ignoring errors
1064         if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1065             # We need to enable promote_secondaries, otherwise our newly added
1066             # address will be removed along with the old one.
1067             my $promote = 0;
1068             eval {
1069                 if ($ipversion == 4) {
1070                     &$nscmd({ outfunc => sub { $promote = int(shift) } },
1071                             'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1072                     &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1073                 }
1074                 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1075             };
1076             warn $@ if $@; # no need to die here
1077
1078             if ($ipversion == 4) {
1079                 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1080             }
1081         }
1082
1083         foreach my $property ($ip, $gw) {
1084             if ($newnet->{$property}) {
1085                 $optdata->{$property} = $newnet->{$property};
1086             } else {
1087                 delete $optdata->{$property};
1088             }
1089         }
1090         $conf->{$opt} = PVE::LXC::Config->print_lxc_network($optdata);
1091         PVE::LXC::Config->write_config($vmid, $conf);
1092         $lxc_setup->setup_network($conf);
1093     };
1094
1095     &$change_ip_config(4);
1096     &$change_ip_config(6);
1097
1098 }
1099
1100 my $open_namespace = sub {
1101     my ($vmid, $pid, $kind) = @_;
1102     sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1103         or die "failed to open $kind namespace of container $vmid: $!\n";
1104     return $fd;
1105 };
1106
1107 my $enter_namespace = sub {
1108     my ($vmid, $pid, $kind, $type) = @_;
1109     my $fd = $open_namespace->($vmid, $pid, $kind);
1110     PVE::Tools::setns(fileno($fd), $type)
1111         or die "failed to enter $kind namespace of container $vmid: $!\n";
1112     close $fd;
1113 };
1114
1115 my $get_container_namespace = sub {
1116     my ($vmid, $pid, $kind) = @_;
1117
1118     my $pidfd;
1119     if (!defined($pid)) {
1120         # Pin the pid while we're grabbing its stuff from /proc
1121         ($pid, $pidfd) = open_lxc_pid($vmid)
1122             or die "failed to open pidfd of container $vmid\'s init process\n";
1123     }
1124
1125     return $open_namespace->($vmid, $pid, $kind);
1126 };
1127
1128 my $do_syncfs = sub {
1129     my ($vmid, $pid, $socket) = @_;
1130
1131     &$enter_namespace($vmid, $pid, 'mnt', PVE::Tools::CLONE_NEWNS);
1132
1133     # Tell the parent process to start reading our /proc/mounts
1134     print {$socket} "go\n";
1135     $socket->flush();
1136
1137     # Receive /proc/self/mounts
1138     my $mountdata = do { local $/ = undef; <$socket> };
1139     close $socket;
1140
1141     # Now sync all mountpoints...
1142     my $mounts = PVE::ProcFSTools::parse_mounts($mountdata);
1143     foreach my $mp (@$mounts) {
1144         my ($what, $dir, $fs) = @$mp;
1145         next if $fs eq 'fuse.lxcfs';
1146         eval { PVE::Tools::sync_mountpoint($dir); };
1147         warn $@ if $@;
1148     }
1149 };
1150
1151 sub sync_container_namespace {
1152     my ($vmid) = @_;
1153     my $pid = find_lxc_pid($vmid);
1154
1155     # SOCK_DGRAM is nicer for barriers but cannot be slurped
1156     socketpair my $pfd, my $cfd, AF_UNIX, SOCK_STREAM, PF_UNSPEC
1157         or die "failed to create socketpair: $!\n";
1158
1159     my $child = fork();
1160     die "fork failed: $!\n" if !defined($child);
1161
1162     if (!$child) {
1163         eval {
1164             close $pfd;
1165             &$do_syncfs($vmid, $pid, $cfd);
1166         };
1167         if (my $err = $@) {
1168             warn $err;
1169             POSIX::_exit(1);
1170         }
1171         POSIX::_exit(0);
1172     }
1173     close $cfd;
1174     my $go = <$pfd>;
1175     die "failed to enter container namespace\n" if $go ne "go\n";
1176
1177     open my $mounts, '<', "/proc/$child/mounts"
1178         or die "failed to open container's /proc/mounts: $!\n";
1179     my $mountdata = do { local $/ = undef; <$mounts> };
1180     close $mounts;
1181     print {$pfd} $mountdata;
1182     close $pfd;
1183
1184     while (waitpid($child, 0) != $child) {}
1185     die "failed to sync container namespace\n" if $? != 0;
1186 }
1187
1188 sub template_create {
1189     my ($vmid, $conf) = @_;
1190
1191     my $storecfg = PVE::Storage::config();
1192
1193     PVE::LXC::Config->foreach_mountpoint($conf, sub {
1194         my ($ms, $mountpoint) = @_;
1195
1196         my $volid = $mountpoint->{volume};
1197
1198         die "Template feature is not available for '$volid'\n"
1199             if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
1200     });
1201
1202     PVE::LXC::Config->foreach_mountpoint($conf, sub {
1203         my ($ms, $mountpoint) = @_;
1204
1205         my $volid = $mountpoint->{volume};
1206
1207         PVE::Storage::activate_volumes($storecfg, [$volid]);
1208
1209         my $template_volid = PVE::Storage::vdisk_create_base($storecfg, $volid);
1210         $mountpoint->{volume} = $template_volid;
1211         $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1212     });
1213
1214     PVE::LXC::Config->write_config($vmid, $conf);
1215 }
1216
1217 sub check_ct_modify_config_perm {
1218     my ($rpcenv, $authuser, $vmid, $pool, $newconf, $delete) = @_;
1219
1220     return 1 if $authuser eq 'root@pam';
1221
1222     my $check = sub {
1223         my ($opt, $delete) = @_;
1224         if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1225             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1226         } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1227             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1228             return if $delete;
1229             my $data = $opt eq 'rootfs' ? PVE::LXC::Config->parse_ct_rootfs($newconf->{$opt})
1230                                         : PVE::LXC::Config->parse_ct_mountpoint($newconf->{$opt});
1231             raise_perm_exc("mount point type $data->{type} is only allowed for root\@pam")
1232                 if $data->{type} ne 'volume';
1233         } elsif ($opt eq 'memory' || $opt eq 'swap') {
1234             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1235         } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1236                  $opt eq 'searchdomain' || $opt eq 'hostname') {
1237             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1238         } elsif ($opt eq 'features') {
1239             # For now this is restricted to root@pam
1240             raise_perm_exc("changing feature flags is only allowed for root\@pam");
1241         } elsif ($opt eq 'hookscript') {
1242             # For now this is restricted to root@pam
1243             raise_perm_exc("changing the hookscript is only allowed for root\@pam");
1244         } else {
1245             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1246         }
1247     };
1248
1249     foreach my $opt (keys %$newconf) {
1250         &$check($opt, 0);
1251     }
1252     foreach my $opt (@$delete) {
1253         &$check($opt, 1);
1254     }
1255
1256     return 1;
1257 }
1258
1259 sub umount_all {
1260     my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1261
1262     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1263     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1264
1265     my $res = 1;
1266
1267     PVE::LXC::Config->foreach_mountpoint_reverse($conf, sub {
1268         my ($ms, $mountpoint) = @_;
1269
1270         my $volid = $mountpoint->{volume};
1271         my $mount = $mountpoint->{mp};
1272
1273         return if !$volid || !$mount;
1274
1275         my $mount_path = "$rootdir/$mount";
1276         $mount_path =~ s!/+!/!g;
1277
1278         return if !PVE::ProcFSTools::is_mounted($mount_path);
1279
1280         eval {
1281             PVE::Tools::run_command(['umount', '-d', $mount_path]);
1282         };
1283         if (my $err = $@) {
1284             if ($noerr) {
1285                 $res = 0;
1286                 warn $err;
1287             } else {
1288                 die $err;
1289             }
1290         }
1291     });
1292
1293     return $res; # tell caller if (some) umounts failed for the noerr case
1294 }
1295
1296 sub mount_all {
1297     my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1298
1299     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1300     File::Path::make_path($rootdir);
1301
1302     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1303     PVE::Storage::activate_volumes($storage_cfg, $volid_list);
1304
1305     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
1306
1307     eval {
1308         PVE::LXC::Config->foreach_mountpoint($conf, sub {
1309             my ($ms, $mountpoint) = @_;
1310
1311             $mountpoint->{ro} = 0 if $ignore_ro;
1312
1313             mountpoint_mount($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1314         });
1315     };
1316     if (my $err = $@) {
1317         warn "mounting container failed\n";
1318         umount_all($vmid, $storage_cfg, $conf, 1);
1319         die $err;
1320     }
1321
1322     return $rootdir;
1323 }
1324
1325
1326 sub mountpoint_mount_path {
1327     my ($mountpoint, $storage_cfg, $snapname) = @_;
1328
1329     return mountpoint_mount($mountpoint, undef, $storage_cfg, $snapname);
1330 }
1331
1332 sub query_loopdev {
1333     my ($path) = @_;
1334     my $found;
1335     my $parser = sub {
1336         my $line = shift;
1337         if ($line =~ m@^(/dev/loop\d+):@) {
1338             $found = $1;
1339         }
1340     };
1341     my $cmd = ['losetup', '--associated', $path];
1342     PVE::Tools::run_command($cmd, outfunc => $parser);
1343     return $found;
1344 }
1345
1346 # Run a function with a file attached to a loop device.
1347 # The loop device is always detached afterwards (or set to autoclear).
1348 # Returns the loop device.
1349 sub run_with_loopdev {
1350     my ($func, $file, $readonly) = @_;
1351     my $device = query_loopdev($file);
1352     # Try to reuse an existing device
1353     if ($device) {
1354         # We assume that whoever setup the loop device is responsible for
1355         # detaching it.
1356         &$func($device);
1357         return $device;
1358     }
1359
1360     my $parser = sub {
1361         my $line = shift;
1362         if ($line =~ m@^(/dev/loop\d+)$@) {
1363             $device = $1;
1364         }
1365     };
1366     my $losetup_cmd = [
1367         'losetup',
1368         '--show',
1369         '-f',
1370         $file,
1371     ];
1372     push @$losetup_cmd, '-r' if $readonly;
1373     PVE::Tools::run_command($losetup_cmd, outfunc => $parser);
1374     die "failed to setup loop device for $file\n" if !$device;
1375     eval { &$func($device); };
1376     my $err = $@;
1377     PVE::Tools::run_command(['losetup', '-d', $device]);
1378     die $err if $err;
1379     return $device;
1380 }
1381
1382 # In scalar mode: returns a file handle to the deepest directory node.
1383 # In list context: returns a list of:
1384 #   * the deepest directory node
1385 #   * the 2nd deepest directory (parent of the above)
1386 #   * directory name of the last directory
1387 # So that the path $2/$3 should lead to $1 afterwards.
1388 sub walk_tree_nofollow($$$;$$) {
1389     my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1390
1391     sysopen(my $fd, $start, O_PATH | O_DIRECTORY)
1392         or die "failed to open start directory $start: $!\n";
1393
1394     return walk_tree_nofollow_fd($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1395 }
1396
1397
1398 sub walk_tree_nofollow_fd($$$$;$$) {
1399     my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1400
1401     # splitdir() returns '' for empty components including the leading /
1402     my @comps = grep { length($_)>0 } File::Spec->splitdir($subdir);
1403
1404     my $fd = $start_fd;
1405     my $dir = $start_dirname;
1406     my $last_component = undef;
1407     my $second = $fd;
1408     foreach my $component (@comps) {
1409         $dir .= "/$component";
1410         my $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1411
1412         if (!$next) {
1413             # failed, check for symlinks and try to create the path
1414             die "symlink encountered at: $dir\n" if $! == ELOOP || $! == ENOTDIR;
1415             die "cannot open directory $dir: $!\n" if !$mkdir;
1416
1417             # We don't check for errors on mkdirat() here and just try to
1418             # openat() again, since at least one error (EEXIST) is an
1419             # expected possibility if multiple containers start
1420             # simultaneously. If someone else injects a symlink now then
1421             # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1422             PVE::Tools::mkdirat(fileno($fd), $component, 0755);
1423
1424             $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1425             die "failed to create path: $dir: $!\n" if !$next;
1426
1427             PVE::Tools::fchownat(fileno($next), '', $rootuid, $rootgid, PVE::Tools::AT_EMPTY_PATH)
1428                 if defined($rootuid) && defined($rootgid);
1429         }
1430
1431         close $second if defined($last_component) && $second != $start_fd;
1432         $last_component = $component;
1433         $second = $fd;
1434         $fd = $next;
1435     }
1436
1437     return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1438     close $second if defined($last_component) && $second != $start_fd;
1439     return $fd;
1440 }
1441
1442 # To guard against symlink attack races against other currently running
1443 # containers with shared recursive bind mount hierarchies we prepare a
1444 # directory handle for the directory we're mounting over to verify the
1445 # mountpoint afterwards.
1446 sub __bindmount_prepare {
1447     my ($hostroot, $dir) = @_;
1448     my $srcdh = walk_tree_nofollow($hostroot, $dir, 0);
1449     return $srcdh;
1450 }
1451
1452 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1453 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1454 # we intended to bind mount.
1455 sub __bindmount_verify {
1456     my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1457     my $destdh;
1458     if ($parentfd) {
1459         # Open the mount point path coming from the parent directory since the
1460         # filehandle we would have gotten as first result of walk_tree_nofollow
1461         # earlier is still a handle to the underlying directory instead of the
1462         # mounted path.
1463         $destdh = PVE::Tools::openat(fileno($parentfd), $last_dir, PVE::Tools::O_PATH | O_NOFOLLOW | O_DIRECTORY);
1464         die "failed to open mount point: $!\n" if !$destdh;
1465         if ($ro) {
1466             my $dot = '.';
1467             # no separate function because 99% of the time it's the wrong thing to use.
1468             if (syscall(PVE::Syscall::faccessat, fileno($destdh), $dot, &POSIX::W_OK, 0) != -1) {
1469                 die "failed to mark bind mount read only\n";
1470             }
1471             die "read-only check failed: $!\n" if $! != EROFS;
1472         }
1473     } else {
1474         # For the rootfs we don't have a parentfd so we open the path directly.
1475         # Note that this means bindmounting any prefix of the host's
1476         # /var/lib/lxc/$vmid path into another container is considered a grave
1477         # security error.
1478         sysopen $destdh, $last_dir, O_PATH | O_DIRECTORY;
1479         die "failed to open mount point: $!\n" if !$destdh;
1480     }
1481
1482     my ($srcdev, $srcinode) = stat($srcdh);
1483     my ($dstdev, $dstinode) = stat($destdh);
1484     close $srcdh;
1485     close $destdh;
1486
1487     return ($srcdev == $dstdev && $srcinode == $dstinode);
1488 }
1489
1490 # Perform the actual bind mounting:
1491 sub __bindmount_do {
1492     my ($dir, $dest, $ro, @extra_opts) = @_;
1493     PVE::Tools::run_command(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1494     if ($ro) {
1495         eval { PVE::Tools::run_command(['mount', '-o', 'bind,remount,ro', $dest]); };
1496         if (my $err = $@) {
1497             warn "bindmount error\n";
1498             # don't leave writable bind-mounts behind...
1499             PVE::Tools::run_command(['umount', $dest]);
1500             die $err;
1501         }
1502     }
1503 }
1504
1505 sub bindmount {
1506     my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1507
1508     my $srcdh = __bindmount_prepare('/', $dir);
1509
1510     __bindmount_do($dir, $dest, $ro, @extra_opts);
1511
1512     if (!__bindmount_verify($srcdh, $parentfd, $last_dir, $ro)) {
1513         PVE::Tools::run_command(['umount', $dest]);
1514         die "detected mount path change at: $dir\n";
1515     }
1516 }
1517
1518 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1519 # from $rootdir and $mount and walk the path from $rootdir to the final
1520 # directory to check for symlinks.
1521 sub __mount_prepare_rootdir {
1522     my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1523     $rootdir =~ s!/+!/!g;
1524     $rootdir =~ s!/+$!!;
1525     my $mount_path = "$rootdir/$mount";
1526     my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow($rootdir, $mount, 1, $rootuid, $rootgid);
1527     return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1528 }
1529
1530 # use $rootdir = undef to just return the corresponding mount path
1531 sub mountpoint_mount {
1532     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1533     return __mountpoint_mount($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1534 }
1535
1536 sub mountpoint_stage {
1537     my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1538     my ($path, $loop, $dev) =
1539         __mountpoint_mount($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1540
1541     if (!defined($path)) {
1542         return undef if $! == ENOSYS;
1543         die "failed to mount subvolume: $!\n";
1544     }
1545
1546     # We clone the mount point and leave it there in order to keep them connected to eg. loop
1547     # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1548     # point).
1549     my $err;
1550     my $fd = PVE::Tools::open_tree(&AT_FDCWD, $stage_dir, &OPEN_TREE_CLOEXEC | &OPEN_TREE_CLONE)
1551         or die "open_tree() on mount point failed: $!\n";
1552
1553     return wantarray ? ($path, $loop, $dev, $fd) : $fd;
1554 }
1555
1556 sub mountpoint_insert_staged {
1557     my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1558
1559     if (!defined($rootdir_fd)) {
1560         sysopen($rootdir_fd, '.', O_PATH | O_DIRECTORY)
1561             or die "failed to open '.': $!\n";
1562     }
1563
1564     my $dest_fd = walk_tree_nofollow_fd('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1565
1566     PVE::Tools::move_mount(
1567         fileno($mount_fd),
1568         '',
1569         fileno($dest_fd),
1570         '',
1571         &MOVE_MOUNT_F_EMPTY_PATH | &MOVE_MOUNT_T_EMPTY_PATH,
1572     ) or die "failed to move '$opt' into container hierarchy: $!\n";
1573 }
1574
1575 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1576 #   can then open a file descriptor to it which can be used with the `move_mount` syscall.
1577 #   Note that if the kernel does not support the new mount API, this will not perform any action
1578 #   and return `undef` with $! = ENOSYS.
1579 sub __mountpoint_mount {
1580     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1581
1582     if (defined($stage_mount) && !PVE::LXC::Tools::can_use_new_mount_api()) {
1583         $! = ENOSYS;
1584         return undef;
1585     }
1586
1587     # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1588     # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1589     # which we use in __bindmount_verify()!
1590     my $mount = $stage_mount ? '/': $mountpoint->{mp};
1591
1592     my $volid = $mountpoint->{volume};
1593     my $type = $mountpoint->{type};
1594     my $quota = !$snapname && !$mountpoint->{ro} && $mountpoint->{quota};
1595     my $mounted_dev;
1596
1597     return if !$volid || !$mount;
1598
1599     $mount =~ s!/+!/!g;
1600
1601     my $mount_path;
1602     my ($mpfd, $parentfd, $last_dir);
1603
1604     if (defined($rootdir)) {
1605         ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1606             __mount_prepare_rootdir($rootdir, $mount, $rootuid, $rootgid);
1607     }
1608
1609     if (defined($stage_mount)) {
1610         $mount_path = $rootdir;
1611     }
1612
1613     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1614
1615     die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1616
1617     my $optlist = [];
1618
1619     if (my $mountopts = $mountpoint->{mountoptions}) {
1620         my @opts = split(/;/, $mountpoint->{mountoptions});
1621         push @$optlist, grep { PVE::LXC::Config::is_valid_mount_option($_) } @opts;
1622     }
1623
1624     my $acl = $mountpoint->{acl};
1625     if (defined($acl)) {
1626         push @$optlist, ($acl ? 'acl' : 'noacl');
1627     }
1628
1629     my $optstring = join(',', @$optlist);
1630     my $readonly = $mountpoint->{ro};
1631
1632     my @extra_opts;
1633     @extra_opts = ('-o', $optstring) if $optstring;
1634
1635     if ($storage) {
1636
1637         my $scfg = PVE::Storage::storage_config($storage_cfg, $storage);
1638
1639         my $path = PVE::Storage::map_volume($storage_cfg, $volid, $snapname);
1640
1641         $path = PVE::Storage::path($storage_cfg, $volid, $snapname) if !defined($path);
1642
1643         my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1644             PVE::Storage::parse_volname($storage_cfg, $volid);
1645
1646         $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1647
1648         if ($format eq 'subvol') {
1649             if ($mount_path) {
1650                 if ($snapname) {
1651                     if ($scfg->{type} eq 'zfspool') {
1652                         my $path_arg = $path;
1653                         $path_arg =~ s!^/+!!;
1654                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', $path_arg, $mount_path]);
1655                     } else {
1656                         die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1657                     }
1658                 } else {
1659                     if (defined($acl) && $scfg->{type} eq 'zfspool') {
1660                         my $acltype = ($acl ? 'acltype=posixacl' : 'acltype=noacl');
1661                         my (undef, $name) = PVE::Storage::parse_volname($storage_cfg, $volid);
1662                         $name .= "\@$snapname" if defined($snapname);
1663                         PVE::Tools::run_command(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1664                     }
1665                     bindmount($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1666                     warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1667                 }
1668             }
1669             return wantarray ? ($path, 0, undef) : $path;
1670         } elsif ($format eq 'raw' || $format eq 'iso') {
1671             # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1672             # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1673             # Our autodev hook expects the /dev/dm-* device currently
1674             # and will create the /dev/mapper symlink accordingly
1675             $path = Cwd::realpath($path);
1676             die "failed to get device path\n" if !$path;
1677             ($path) = ($path =~ /^(.*)$/s); #untaint
1678             my $domount = sub {
1679                 my ($path) = @_;
1680                 if ($mount_path) {
1681                     if ($format eq 'iso') {
1682                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1683                     } elsif ($isBase || defined($snapname)) {
1684                         PVE::Tools::run_command(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1685                     } else {
1686                         if ($quota) {
1687                             push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1688                         }
1689                         push @extra_opts, '-o', 'ro' if $readonly;
1690                         PVE::Tools::run_command(['mount', @extra_opts, $path, $mount_path]);
1691                     }
1692                 }
1693             };
1694             my $use_loopdev = 0;
1695             if ($scfg->{path}) {
1696                 $mounted_dev = run_with_loopdev($domount, $path, $readonly);
1697                 $use_loopdev = 1;
1698             } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' ||
1699                      $scfg->{type} eq 'rbd' || $scfg->{type} eq 'lvmthin') {
1700                 $mounted_dev = $path;
1701                 &$domount($path);
1702             } else {
1703                 die "unsupported storage type '$scfg->{type}'\n";
1704             }
1705             return wantarray ? ($path, $use_loopdev, $mounted_dev) : $path;
1706         } else {
1707             die "unsupported image format '$format'\n";
1708         }
1709     } elsif ($type eq 'device') {
1710         push @extra_opts, '-o', 'ro' if $readonly;
1711         push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1712         # See the NOTE above about devicemapper canonicalization
1713         my ($devpath) = (Cwd::realpath($volid) =~ /^(.*)$/s); # realpath() taints
1714         PVE::Tools::run_command(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1715         return wantarray ? ($volid, 0, $devpath) : $volid;
1716     } elsif ($type eq 'bind') {
1717         die "directory '$volid' does not exist\n" if ! -d $volid;
1718         bindmount($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1719         warn "cannot enable quota control for bind mounts\n" if $quota;
1720         return wantarray ? ($volid, 0, undef) : $volid;
1721     }
1722
1723     die "unsupported storage";
1724 }
1725
1726 sub mountpoint_hotplug($$$) {
1727     my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1728
1729     my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
1730
1731     # We do the rest in a fork with an unshared mount namespace, because:
1732     #  -) change our papparmor profile to that of /usr/bin/lxc-start
1733     #  -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1734     #     container's namespace, then mount it.
1735
1736     PVE::Tools::run_fork(sub {
1737         # Pin the container pid longer, we also need to get its monitor/parent:
1738         my ($ct_pid, $ct_pidfd) = open_lxc_pid($vmid)
1739             or die "failed to open pidfd of container $vmid\'s init process\n";
1740
1741         my ($monitor_pid, $monitor_pidfd) = open_ppid($ct_pid)
1742             or die "failed to open pidfd of container $vmid\'s monitor process\n";
1743
1744         my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1745         my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1746
1747         # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1748         # profile to lower our privileges to the same level we have in the start hook:
1749         sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY)
1750             or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1751         # But switch namespaces first, to make sure the namespace switches aren't blocked by
1752         # apparmor.
1753
1754         # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1755         # namespace for it to remain active there since the container will be able to unmount
1756         # hotplugged mount points and thereby potentially free up loop devices, which is a security
1757         # concern.
1758         PVE::Tools::setns(fileno($monitor_mnt_ns), PVE::Tools::CLONE_NEWNS);
1759         chdir('/')
1760             or die "failed to change root directory within the monitor's mount namespace: $!\n";
1761
1762         my $dir = get_staging_mount_path($opt);
1763
1764         # Now switch our apparmor profile before mounting:
1765         my $data = 'changeprofile /usr/bin/lxc-start';
1766         if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1767             die "failed to change apparmor profile: $!\n";
1768         }
1769         # Check errors on close as well:
1770         close($aa_fd)
1771             or die "failed to change apparmor profile (close() failed): $!\n";
1772
1773         my $mount_fd = mountpoint_stage($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1774
1775         PVE::Tools::setns(fileno($ct_mnt_ns), PVE::Tools::CLONE_NEWNS);
1776         chdir('/')
1777             or die "failed to change root directory within the container's mount namespace: $!\n";
1778
1779         mountpoint_insert_staged($mount_fd, undef, $mp->{mp}, $opt, $rootuid, $rootgid);
1780     });
1781 }
1782
1783 # Create a directory in the mountpoint staging tempfs.
1784 sub get_staging_mount_path($) {
1785     my ($opt) = @_;
1786
1787     my $target = get_staging_tempfs() . "/$opt";
1788     if (!mkdir($target) && $! != EEXIST) {
1789         die "failed to create directory $target: $!\n";
1790     }
1791
1792     return $target;
1793 }
1794
1795 # Mount /run/pve/mountpoints as tmpfs
1796 sub get_staging_tempfs() {
1797     # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1798     # mounts to that.
1799     my $target = '/var/lib/lxc/.pve-staged-mounts';
1800     if (!mkdir($target)) {
1801         return $target if $! == EEXIST;
1802         die "failed to create directory $target: $!\n";
1803     }
1804
1805     PVE::Tools::mount("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1806         or die "failed to mount $target as tmpfs: $!\n";
1807
1808     return $target;
1809 }
1810
1811 sub mkfs {
1812     my ($dev, $rootuid, $rootgid) = @_;
1813
1814     PVE::Tools::run_command(['mkfs.ext4', '-O', 'mmp',
1815                              '-E', "root_owner=$rootuid:$rootgid",
1816                              $dev]);
1817 }
1818
1819 sub format_disk {
1820     my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1821
1822     if ($volid =~ m!^/dev/.+!) {
1823         mkfs($volid);
1824         return;
1825     }
1826
1827     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1828
1829     die "cannot format volume '$volid' with no storage\n" if !$storage;
1830
1831     PVE::Storage::activate_volumes($storage_cfg, [$volid]);
1832
1833     my $path = PVE::Storage::map_volume($storage_cfg, $volid);
1834
1835     $path = PVE::Storage::path($storage_cfg, $volid) if !defined($path);
1836
1837     my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1838         PVE::Storage::parse_volname($storage_cfg, $volid);
1839
1840     die "cannot format volume '$volid' (format == $format)\n"
1841         if $format ne 'raw';
1842
1843     mkfs($path, $rootuid, $rootgid);
1844 }
1845
1846 sub destroy_disks {
1847     my ($storecfg, $vollist) = @_;
1848
1849     foreach my $volid (@$vollist) {
1850         eval { PVE::Storage::vdisk_free($storecfg, $volid); };
1851         warn $@ if $@;
1852     }
1853 }
1854
1855 sub alloc_disk {
1856     my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
1857
1858     my $needs_chown = 0;
1859     my $volid;
1860
1861     my $scfg = PVE::Storage::storage_config($storecfg, $storage);
1862     # fixme: use better naming ct-$vmid-disk-X.raw?
1863
1864     eval {
1865         my $do_format = 0;
1866         if ($scfg->{type} eq 'dir' || $scfg->{type} eq 'nfs' || $scfg->{type} eq 'cifs' ) {
1867             if ($size_kb > 0) {
1868                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw',
1869                                                    undef, $size_kb);
1870                 $do_format = 1;
1871             } else {
1872                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol',
1873                                                    undef, 0);
1874                 $needs_chown = 1;
1875             }
1876         } elsif ($scfg->{type} eq 'zfspool') {
1877
1878             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol',
1879                                                undef, $size_kb);
1880             $needs_chown = 1;
1881         } elsif ($scfg->{type} eq 'drbd' || $scfg->{type} eq 'lvm' || $scfg->{type} eq 'lvmthin') {
1882
1883             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1884             $do_format = 1;
1885
1886         } elsif ($scfg->{type} eq 'rbd') {
1887
1888             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1889             $do_format = 1;
1890         } else {
1891             die "unable to create containers on storage type '$scfg->{type}'\n";
1892         }
1893         format_disk($storecfg, $volid, $rootuid, $rootgid) if $do_format;
1894     };
1895     if (my $err = $@) {
1896         # in case formatting got interrupted:
1897         if (defined($volid)) {
1898             eval { PVE::Storage::vdisk_free($storecfg, $volid); };
1899             warn $@ if $@;
1900         }
1901         die $err;
1902     }
1903
1904     return ($volid, $needs_chown);
1905 }
1906
1907 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
1908 sub create_disks {
1909     my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
1910
1911     my $vollist = [];
1912
1913     eval {
1914         my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
1915         my $chown_vollist = [];
1916
1917         PVE::LXC::Config->foreach_mountpoint($settings, sub {
1918             my ($ms, $mountpoint) = @_;
1919
1920             my $volid = $mountpoint->{volume};
1921             my $mp = $mountpoint->{mp};
1922
1923             my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1924
1925             if ($storage && ($volid =~ $NEW_DISK_RE)) {
1926                 my ($storeid, $size_gb) = ($1, $2);
1927
1928                 my $size_kb = int(${size_gb}*1024) * 1024;
1929
1930                 my $needs_chown = 0;
1931                 ($volid, $needs_chown) = alloc_disk($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
1932                 push @$chown_vollist, $volid if $needs_chown;
1933                 push @$vollist, $volid;
1934                 $mountpoint->{volume} = $volid;
1935                 $mountpoint->{size} = $size_kb * 1024;
1936                 if ($pending) {
1937                     $conf->{pending}->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1938                 } else {
1939                     $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1940                 }
1941             } else {
1942                 # use specified/existing volid/dir/device
1943                 $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1944             }
1945         });
1946
1947         PVE::Storage::activate_volumes($storecfg, $chown_vollist, undef);
1948         foreach my $volid (@$chown_vollist) {
1949             my $path = PVE::Storage::path($storecfg, $volid, undef);
1950             chown($rootuid, $rootgid, $path);
1951         }
1952         PVE::Storage::deactivate_volumes($storecfg, $chown_vollist, undef);
1953     };
1954     # free allocated images on error
1955     if (my $err = $@) {
1956         destroy_disks($storecfg, $vollist);
1957         die $err;
1958     }
1959     return $vollist;
1960 }
1961
1962 sub update_disksize {
1963     my ($vmid, $conf, $all_volumes) = @_;
1964
1965     my $changes;
1966     my $prefix = "CT $vmid:";
1967
1968     my $update_mp = sub {
1969         my ($key, $mp, @param) = @_;
1970         my $size = $all_volumes->{$mp->{volume}}->{size} // 0;
1971
1972         if (!defined($mp->{size}) || $size != $mp->{size}) {
1973             $changes = 1;
1974             print "$prefix updated volume size of '$mp->{volume}' in config.\n";
1975             $mp->{size} = $size;
1976             my $nomp = 1 if ($key eq 'rootfs');
1977             $conf->{$key} = PVE::LXC::Config->print_ct_mountpoint($mp, $nomp);
1978         }
1979     };
1980
1981     PVE::LXC::Config->foreach_mountpoint($conf, $update_mp);
1982
1983     return $changes;
1984 }
1985
1986 sub update_unused {
1987     my ($vmid, $conf, $all_volumes) = @_;
1988
1989     my $changes;
1990     my $prefix = "CT $vmid:";
1991
1992     # Note: it is allowed to define multiple storage entries with the same path
1993     # (alias), so we need to check both 'volid' and real 'path' (two different
1994     # volid can point to the same path).
1995
1996     # used and unused disks
1997     my $refpath = {};
1998     my $orphans = {};
1999
2000     foreach my $opt (keys %$conf) {
2001         next if ($opt !~ m/^unused\d+$/);
2002         my $vol = $all_volumes->{$conf->{$opt}};
2003         $refpath->{$vol->{path}} = $vol->{volid};
2004     }
2005
2006     foreach my $key (keys %$all_volumes) {
2007         my $vol = $all_volumes->{$key};
2008         my $in_use = PVE::LXC::Config->is_volume_in_use($conf, $vol->{volid});
2009         my $path = $vol->{path};
2010
2011         if ($in_use) {
2012             $refpath->{$path} = $key;
2013             delete $orphans->{$path};
2014         } else {
2015             if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2016                 $orphans->{$path} = $key;
2017             }
2018         }
2019     }
2020
2021     for my $key (keys %$orphans) {
2022         my $disk = $orphans->{$key};
2023         my $unused = PVE::LXC::Config->add_unused_volume($conf, $disk);
2024
2025         if ($unused) {
2026             $changes = 1;
2027             print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2028         }
2029     }
2030
2031     return $changes;
2032 }
2033
2034 sub scan_volids {
2035     my ($cfg, $vmid) = @_;
2036
2037     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid);
2038
2039     my $all_volumes = {};
2040     foreach my $storeid (keys %$info) {
2041         foreach my $item (@{$info->{$storeid}}) {
2042             my $volid = $item->{volid};
2043             next if !($volid && $item->{size});
2044             $item->{path} = PVE::Storage::path($cfg, $volid);
2045             $all_volumes->{$volid} = $item;
2046         }
2047     }
2048
2049     return $all_volumes;
2050 }
2051
2052 sub rescan {
2053     my ($vmid, $nolock, $dryrun) = @_;
2054
2055     my $cfg = PVE::Storage::config();
2056
2057     # FIXME: Remove once our RBD plugin can handle CT and VM on a single storage
2058     # see: https://pve.proxmox.com/pipermail/pve-devel/2018-July/032900.html
2059     foreach my $stor (keys %{$cfg->{ids}}) {
2060         delete($cfg->{ids}->{$stor}) if !$cfg->{ids}->{$stor}->{content}->{rootdir};
2061     }
2062
2063     print "rescan volumes...\n";
2064     my $all_volumes = scan_volids($cfg, $vmid);
2065
2066     my $updatefn =  sub {
2067         my ($vmid) = @_;
2068
2069         my $changes;
2070         my $conf = PVE::LXC::Config->load_config($vmid);
2071
2072         PVE::LXC::Config->check_lock($conf);
2073
2074         my $vm_volids = {};
2075         foreach my $volid (keys %$all_volumes) {
2076             my $info = $all_volumes->{$volid};
2077             $vm_volids->{$volid} = $info if $info->{vmid} == $vmid;
2078         }
2079
2080         my $upu = update_unused($vmid, $conf, $vm_volids);
2081         my $upd = update_disksize($vmid, $conf, $vm_volids);
2082         $changes = $upu || $upd;
2083
2084         PVE::LXC::Config->write_config($vmid, $conf) if $changes && !$dryrun;
2085     };
2086
2087     if (defined($vmid)) {
2088         if ($nolock) {
2089             &$updatefn($vmid);
2090         } else {
2091             PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2092         }
2093     } else {
2094         my $vmlist = config_list();
2095         foreach my $vmid (keys %$vmlist) {
2096             if ($nolock) {
2097                 &$updatefn($vmid);
2098             } else {
2099                 PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2100             }
2101         }
2102     }
2103 }
2104
2105
2106 # bash completion helper
2107
2108 sub complete_os_templates {
2109     my ($cmdname, $pname, $cvalue) = @_;
2110
2111     my $cfg = PVE::Storage::config();
2112
2113     my $storeid;
2114
2115     if ($cvalue =~ m/^([^:]+):/) {
2116         $storeid = $1;
2117     }
2118
2119     my $vtype = $cmdname eq 'restore' ? 'backup' : 'vztmpl';
2120     my $data = PVE::Storage::template_list($cfg, $storeid, $vtype);
2121
2122     my $res = [];
2123     foreach my $id (keys %$data) {
2124         foreach my $item (@{$data->{$id}}) {
2125             push @$res, $item->{volid} if defined($item->{volid});
2126         }
2127     }
2128
2129     return $res;
2130 }
2131
2132 my $complete_ctid_full = sub {
2133     my ($running) = @_;
2134
2135     my $idlist = vmstatus();
2136
2137     my $active_hash = list_active_containers();
2138
2139     my $res = [];
2140
2141     foreach my $id (keys %$idlist) {
2142         my $d = $idlist->{$id};
2143         if (defined($running)) {
2144             next if $d->{template};
2145             next if $running && !$active_hash->{$id};
2146             next if !$running && $active_hash->{$id};
2147         }
2148         push @$res, $id;
2149
2150     }
2151     return $res;
2152 };
2153
2154 sub complete_ctid {
2155     return &$complete_ctid_full();
2156 }
2157
2158 sub complete_ctid_stopped {
2159     return &$complete_ctid_full(0);
2160 }
2161
2162 sub complete_ctid_running {
2163     return &$complete_ctid_full(1);
2164 }
2165
2166 sub parse_id_maps {
2167     my ($conf) = @_;
2168
2169     my $id_map = [];
2170     my $rootuid = 0;
2171     my $rootgid = 0;
2172
2173     my $lxc = $conf->{lxc};
2174     foreach my $entry (@$lxc) {
2175         my ($key, $value) = @$entry;
2176         # FIXME: remove the 'id_map' variant when lxc-3.0 arrives
2177         next if $key ne 'lxc.idmap' && $key ne 'lxc.id_map';
2178         if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2179             my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2180             push @$id_map, [$type, $ct, $host, $length];
2181             if ($ct == 0) {
2182                 $rootuid = $host if $type eq 'u';
2183                 $rootgid = $host if $type eq 'g';
2184             }
2185         } else {
2186             die "failed to parse idmap: $value\n";
2187         }
2188     }
2189
2190     if (!@$id_map && $conf->{unprivileged}) {
2191         # Should we read them from /etc/subuid?
2192         $id_map = [ ['u', '0', '100000', '65536'],
2193                     ['g', '0', '100000', '65536'] ];
2194         $rootuid = $rootgid = 100000;
2195     }
2196
2197     return ($id_map, $rootuid, $rootgid);
2198 }
2199
2200 sub userns_command {
2201     my ($id_map) = @_;
2202     if (@$id_map) {
2203         return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2204     }
2205     return [];
2206 }
2207
2208 sub vm_start {
2209     my ($vmid, $conf, $skiplock) = @_;
2210
2211     # apply pending changes while starting
2212     if (scalar(keys %{$conf->{pending}})) {
2213         my $storecfg = PVE::Storage::config();
2214         PVE::LXC::Config->vmconfig_apply_pending($vmid, $conf, $storecfg);
2215         $conf = PVE::LXC::Config->load_config($vmid); # update/reload
2216     }
2217
2218     update_lxc_config($vmid, $conf);
2219
2220     my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2221
2222     if ($skiplock) {
2223         open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2224         close($fh);
2225     }
2226
2227     my $cmd = ['systemctl', 'start', "pve-container\@$vmid"];
2228
2229     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-start', 1);
2230     eval { PVE::Tools::run_command($cmd); };
2231     if (my $err = $@) {
2232         unlink $skiplock_flag_fn;
2233         die $err;
2234     }
2235     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'post-start');
2236
2237     return;
2238 }
2239
2240 # Helper to stop a container completely and make sure it has stopped completely.
2241 # This is necessary because we want the post-stop hook to have completed its
2242 # unmount-all step, but post-stop happens after lxc puts the container into the
2243 # STOPPED state.
2244 # $kill - if true it will always do an immediate hard-stop
2245 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2246 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2247 sub vm_stop {
2248     my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2249
2250     # Open the container's command socket.
2251     my $path = "\0/var/lib/lxc/$vmid/command";
2252     my $sock = IO::Socket::UNIX->new(
2253         Type => SOCK_STREAM(),
2254         Peer => $path,
2255     );
2256     if (!$sock) {
2257         return if $! == ECONNREFUSED; # The container is not running
2258         die "failed to open container ${vmid}'s command socket: $!\n";
2259     }
2260
2261     my $conf = PVE::LXC::Config->load_config($vmid);
2262     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-stop');
2263
2264     # Stop the container:
2265
2266     my $cmd = ['lxc-stop', '-n', $vmid];
2267
2268     if ($kill) {
2269         push @$cmd, '--kill'; # doesn't allow timeouts
2270     } else {
2271         # lxc-stop uses a default timeout
2272         push @$cmd, '--nokill' if !$kill_after_timeout;
2273
2274         if (defined($shutdown_timeout)) {
2275             push @$cmd, '--timeout', $shutdown_timeout;
2276             # Give run_command 5 extra seconds
2277             $shutdown_timeout += 5;
2278         }
2279     }
2280
2281     eval { PVE::Tools::run_command($cmd, timeout => $shutdown_timeout) };
2282     if (my $err = $@) {
2283         warn $@ if $@;
2284     }
2285
2286     my $result = <$sock>;
2287
2288     return if !defined $result; # monitor is gone and the ct has stopped.
2289     die "container did not stop\n";
2290 }
2291
2292 sub vm_reboot {
2293     my ($vmid, $timeout, $skiplock) = @_;
2294
2295     PVE::LXC::Config->lock_config($vmid, sub {
2296         return if !check_running($vmid);
2297
2298         vm_stop($vmid, 0, $timeout, 1); # kill if timeout exceeds
2299
2300         my $conf = PVE::LXC::Config->load_config($vmid);
2301         vm_start($vmid, $conf);
2302     });
2303 }
2304
2305 sub run_unshared {
2306     my ($code) = @_;
2307
2308     return PVE::Tools::run_fork(sub {
2309         # Unshare the mount namespace
2310         die "failed to unshare mount namespace: $!\n"
2311             if !PVE::Tools::unshare(PVE::Tools::CLONE_NEWNS);
2312         PVE::Tools::run_command(['mount', '--make-rslave', '/']);
2313         return $code->();
2314     });
2315 }
2316
2317 my $copy_volume = sub {
2318     my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid,  $rootgid) = @_;
2319
2320     my $src_mp = { volume => $src_volid, mp => '/', ro => 1 };
2321     $src_mp->{type} = PVE::LXC::Config->classify_mountpoint($src_volid);
2322
2323     my $dst_mp = { volume => $dst_volid, mp => '/', ro => 0 };
2324     $dst_mp->{type} = PVE::LXC::Config->classify_mountpoint($dst_volid);
2325
2326     my @mounted;
2327     eval {
2328         # mount and copy
2329         mkdir $src;
2330         mountpoint_mount($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2331         push @mounted, $src;
2332         mkdir $dest;
2333         mountpoint_mount($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2334         push @mounted, $dest;
2335
2336         $bwlimit //= 0;
2337
2338         PVE::Tools::run_command(['/usr/bin/rsync', '--stats', '-X', '-A', '--numeric-ids',
2339                                  '-aH', '--whole-file', '--sparse', '--one-file-system',
2340                                  "--bwlimit=$bwlimit", "$src/", $dest]);
2341     };
2342     my $err = $@;
2343
2344     # Wait for rsync's children to release dest so that
2345     # consequent file operations (umount, remove) are possible
2346     while ((system {"fuser"} "fuser",  "-s", $dest) == 0) {sleep 1};
2347
2348     foreach my $mount (reverse @mounted) {
2349         eval { PVE::Tools::run_command(['/bin/umount', $mount], errfunc => sub{})};
2350         warn "Can't umount $mount\n" if $@;
2351     }
2352
2353     # If this fails they're used as mount points in a concurrent operation
2354     # (which should not happen but there's also no real need to get rid of them).
2355     rmdir $dest;
2356     rmdir $src;
2357
2358     die $err if $err;
2359 };
2360
2361 # Should not be called after unsharing the mount namespace!
2362 sub copy_volume {
2363     my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2364
2365     die "cannot copy volumes of type $mp->{type}\n" if $mp->{type} ne 'volume';
2366     File::Path::make_path("/var/lib/lxc/$vmid");
2367     my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2368     my $src  = "/var/lib/lxc/$vmid/.copy-volume-2";
2369
2370     # get id's for unprivileged container
2371     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
2372
2373     # Allocate the disk before unsharing in order to make sure zfs subvolumes
2374     # are visible in this namespace, otherwise the host only sees the empty
2375     # (not-mounted) directory.
2376     my $new_volid;
2377     eval {
2378         # Make sure $mp contains a correct size.
2379         $mp->{size} = PVE::Storage::volume_size_info($storage_cfg, $mp->{volume});
2380         my $needs_chown;
2381         ($new_volid, $needs_chown) = alloc_disk($storage_cfg, $vmid, $storage, $mp->{size}/1024, $rootuid, $rootgid);
2382         if ($needs_chown) {
2383             PVE::Storage::activate_volumes($storage_cfg, [$new_volid], undef);
2384             my $path = PVE::Storage::path($storage_cfg, $new_volid, undef);
2385             chown($rootuid, $rootgid, $path);
2386         }
2387
2388         run_unshared(sub {
2389             $copy_volume->($mp->{volume}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2390         });
2391     };
2392     if (my $err = $@) {
2393         PVE::Storage::vdisk_free($storage_cfg, $new_volid)
2394             if defined($new_volid);
2395         die $err;
2396     }
2397
2398     return $new_volid;
2399 }
2400
2401 1;