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 use IO::Poll qw(POLLIN POLLHUP);
  17
  18 use PVE::Exception qw(raise_perm_exc);
  19 use PVE::Storage;
  20 use PVE::SafeSyslog;
  21 use PVE::INotify;
  22 use PVE::JSONSchema qw(get_standard_option);
  23 use PVE::Tools qw(
  24     run_command
  25     dir_glob_foreach
  26     file_get_contents
  27     file_set_contents
  28     AT_FDCWD
  29     O_PATH
  30     $IPV4RE
  31     $IPV6RE
  32 );
  33 use PVE::CpuSet;
  34 use PVE::Network;
  35 use PVE::AccessControl;
  36 use PVE::ProcFSTools;
  37 use PVE::RESTEnvironment;
  38 use PVE::Syscall qw(:fsmount);
  39 use PVE::LXC::Config;
  40 use PVE::GuestHelpers qw(safe_string_ne safe_num_ne safe_boolean_ne);
  41 use PVE::LXC::Tools;
  42 use PVE::LXC::CGroup;
  43 use PVE::LXC::Monitor;
  44 use PVE::CGroup;
  45
  46 use Time::HiRes qw (gettimeofday);
  47 my $have_sdn;
  48 eval {
  49     require PVE::Network::SDN::Zones;
  50     $have_sdn = 1;
  51 };
  52
  53 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
  54
  55 my $nodename = PVE::INotify::nodename();
  56
  57 my $cpuinfo= PVE::ProcFSTools::read_cpuinfo();
  58
  59 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
  60
  61 sub config_list {
  62     my $vmlist = PVE::Cluster::get_vmlist();
  63     my $res = {};
  64     return $res if !$vmlist || !$vmlist->{ids};
  65     my $ids = $vmlist->{ids};
  66
  67     foreach my $vmid (keys %$ids) {
  68         next if !$vmid; # skip CT0
  69         my $d = $ids->{$vmid};
  70         next if !$d->{node} || $d->{node} ne $nodename;
  71         next if !$d->{type} || $d->{type} ne 'lxc';
  72         $res->{$vmid} = { type => 'lxc', vmid => $vmid };
  73     }
  74     return $res;
  75 }
  76
  77 # container status helpers
  78
  79 sub list_active_containers {
  80
  81     my $filename = "/proc/net/unix";
  82
  83     # similar test is used by lcxcontainers.c: list_active_containers
  84     my $res = {};
  85
  86     my $fh = IO::File->new ($filename, "r");
  87     return $res if !$fh;
  88
  89     while (defined(my $line = <$fh>)) {
  90         if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
  91             my $path = $1;
  92             if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
  93                 $res->{$1} = 1;
  94             }
  95         }
  96     }
  97
  98     close($fh);
  99
 100     return $res;
 101 }
 102
 103 # warning: this is slow
 104 sub check_running {
 105     my ($vmid) = @_;
 106
 107     my $active_hash = list_active_containers();
 108
 109     return 1 if defined($active_hash->{$vmid});
 110
 111     return undef;
 112 }
 113
 114 sub get_container_disk_usage {
 115     my ($vmid, $pid) = @_;
 116
 117     return PVE::Tools::df("/proc/$pid/root/", 1);
 118 }
 119
 120 my $last_proc_vmid_stat;
 121
 122 our $vmstatus_return_properties = {
 123     vmid => get_standard_option('pve-vmid'),
 124     status => {
 125         description => "LXC Container status.",
 126         type => 'string',
 127         enum => ['stopped', 'running'],
 128     },
 129     maxmem => {
 130         description => "Maximum memory in bytes.",
 131         type => 'integer',
 132         optional => 1,
 133         renderer => 'bytes',
 134     },
 135     maxswap => {
 136         description => "Maximum SWAP memory in bytes.",
 137         type => 'integer',
 138         optional => 1,
 139         renderer => 'bytes',
 140     },
 141     maxdisk => {
 142         description => "Root disk size in bytes.",
 143         type => 'integer',
 144         optional => 1,
 145         renderer => 'bytes',
 146     },
 147     name => {
 148         description => "Container name.",
 149         type => 'string',
 150         optional => 1,
 151     },
 152     uptime => {
 153         description => "Uptime.",
 154         type => 'integer',
 155         optional => 1,
 156         renderer => 'duration',
 157     },
 158     cpus => {
 159         description => "Maximum usable CPUs.",
 160         type => 'number',
 161         optional => 1,
 162     },
 163     lock => {
 164         description => "The current config lock, if any.",
 165         type => 'string',
 166         optional => 1,
 167     },
 168     tags => {
 169         description => "The current configured tags, if any.",
 170         type => 'string',
 171         optional => 1,
 172     }
 173 };
 174
 175 sub vmstatus {
 176     my ($opt_vmid) = @_;
 177
 178     my $list = $opt_vmid ? { $opt_vmid => { type => 'lxc', vmid => int($opt_vmid) }} : config_list();
 179
 180     my $active_hash = list_active_containers();
 181
 182     my $cpucount = $cpuinfo->{cpus} || 1;
 183
 184     my $cdtime = gettimeofday;
 185
 186     my $uptime = (PVE::ProcFSTools::read_proc_uptime(1))[0];
 187     my $clock_ticks = POSIX::sysconf(&POSIX::_SC_CLK_TCK);
 188
 189     my $unprivileged = {};
 190
 191     foreach my $vmid (keys %$list) {
 192         my $d = $list->{$vmid};
 193
 194         eval { $d->{pid} = int(find_lxc_pid($vmid)) if defined($active_hash->{$vmid}); };
 195         warn $@ if $@; # ignore errors (consider them stopped)
 196
 197         $d->{status} = $active_hash->{$vmid} ? 'running' : 'stopped';
 198
 199         my $cfspath = PVE::LXC::Config->cfs_config_path($vmid);
 200         my $conf = PVE::Cluster::cfs_read_file($cfspath) || {};
 201
 202         $unprivileged->{$vmid} = $conf->{unprivileged};
 203
 204         $d->{name} = $conf->{'hostname'} || "CT$vmid";
 205         $d->{name} =~ s/[\s]//g;
 206
 207         $d->{cpus} = $conf->{cores} || $conf->{cpulimit};
 208         $d->{cpus} = $cpucount if !$d->{cpus};
 209
 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             $d->{disk} = int($res->{used});
 215             $d->{maxdisk} = int($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_volume('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         $d->{template} = 1 if PVE::LXC::Config->is_template($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         my $cgroups = PVE::LXC::CGroup->new($vmid);
 257
 258         if (defined(my $mem = $cgroups->get_memory_stat())) {
 259             $d->{mem} = int($mem->{mem});
 260             $d->{swap} = int($mem->{swap});
 261         } else {
 262             $d->{mem} = 0;
 263             $d->{swap} = 0;
 264         }
 265
 266         if (defined(my $blkio = $cgroups->get_io_stats())) {
 267             $d->{diskread} = int($blkio->{diskread});
 268             $d->{diskwrite} = int($blkio->{diskwrite});
 269         } else {
 270             $d->{diskread} = 0;
 271             $d->{diskwrite} = 0;
 272         }
 273
 274         if (defined(my $cpu = $cgroups->get_cpu_stat())) {
 275             # Total time (in milliseconds) used up by the cpu.
 276             my $used_ms = $cpu->{utime} + $cpu->{stime};
 277
 278             my $old = $last_proc_vmid_stat->{$vmid};
 279             if (!$old) {
 280                 $last_proc_vmid_stat->{$vmid} = {
 281                     time => $cdtime,
 282                     used => $used_ms,
 283                     cpu => 0,
 284                 };
 285                 next;
 286             }
 287
 288             my $delta_ms = ($cdtime - $old->{time}) * $cpucount * 1000.0;
 289             if ($delta_ms > 1000.0) {
 290                 my $delta_used_ms = $used_ms - $old->{used};
 291                 $d->{cpu} = (($delta_used_ms / $delta_ms) * $cpucount) / $d->{cpus};
 292                 $last_proc_vmid_stat->{$vmid} = {
 293                     time => $cdtime,
 294                     used => $used_ms,
 295                     cpu => $d->{cpu},
 296                 };
 297             } else {
 298                 $d->{cpu} = $old->{cpu};
 299             }
 300         } else {
 301             $d->{cpu} = 0;
 302         }
 303     }
 304
 305     my $netdev = PVE::ProcFSTools::read_proc_net_dev();
 306
 307     foreach my $dev (keys %$netdev) {
 308         next if $dev !~ m/^veth([1-9]\d*)i/;
 309         my $vmid = $1;
 310         my $d = $list->{$vmid};
 311
 312         next if !$d;
 313
 314         $d->{netout} += $netdev->{$dev}->{receive};
 315         $d->{netin} += $netdev->{$dev}->{transmit};
 316
 317     }
 318
 319     return $list;
 320 }
 321
 322 sub find_lxc_console_pids {
 323
 324     my $res = {};
 325
 326     PVE::Tools::dir_glob_foreach('/proc', '\d+', sub {
 327         my ($pid) = @_;
 328
 329         my $cmdline = PVE::Tools::file_read_firstline("/proc/$pid/cmdline");
 330         return if !$cmdline;
 331
 332         my @args = split(/\0/, $cmdline);
 333
 334         # search for lxc-console -n <vmid>
 335         return if scalar(@args) != 3;
 336         return if $args[1] ne '-n';
 337         return if $args[2] !~ m/^\d+$/;
 338         return if $args[0] !~ m|^(/usr/bin/)?lxc-console$|;
 339
 340         my $vmid = $args[2];
 341
 342         push @{$res->{$vmid}}, $pid;
 343     });
 344
 345     return $res;
 346 }
 347
 348 sub find_lxc_pid {
 349     my ($vmid) = @_;
 350
 351     my $pid = undef;
 352     my $parser = sub {
 353         my $line = shift;
 354         $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
 355     };
 356     PVE::Tools::run_command(['lxc-info', '-n', $vmid, '-p'], outfunc => $parser);
 357
 358     die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
 359
 360     return $pid;
 361 }
 362
 363 sub open_pid_fd($) {
 364     my ($pid) = @_;
 365     sysopen(my $fd, "/proc/$pid", O_RDONLY | O_DIRECTORY)
 366         or die "failed to open /proc/$pid pid fd\n";
 367     return $fd;
 368 }
 369
 370 sub open_lxc_pid {
 371     my ($vmid) = @_;
 372
 373     # Find the pid and open:
 374     my $pid = find_lxc_pid($vmid);
 375     my $fd = open_pid_fd($pid);
 376
 377     # Verify:
 378     my $pid2 = find_lxc_pid($vmid);
 379
 380     return () if $pid != $pid2;
 381     return ($pid, $fd);
 382 }
 383
 384 sub open_ppid {
 385     my ($pid) = @_;
 386
 387     # Find the parent pid via proc and open it:
 388     my $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 389     my $ppid = $stat->{ppid} // die "failed to get parent pid\n";
 390
 391     my $fd = open_pid_fd($ppid);
 392
 393     # Verify:
 394     $stat = PVE::ProcFSTools::read_proc_pid_stat($pid);
 395     my $ppid2 = $stat->{ppid} // die "failed to get parent pid for verification\n";
 396
 397     return () if $ppid != $ppid2;
 398     return ($ppid, $fd);
 399 }
 400
 401 # Note: we cannot use Net:IP, because that only allows strict
 402 # CIDR networks
 403 sub parse_ipv4_cidr {
 404     my ($cidr, $noerr) = @_;
 405
 406     if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) &&  ($2 <= 32)) {
 407         return { address => $1, netmask => $PVE::Network::ipv4_reverse_mask->[$2] };
 408     }
 409
 410     return undef if $noerr;
 411
 412     die "unable to parse ipv4 address/mask\n";
 413 }
 414
 415 # With seccomp trap to userspace we now have the ability to optionally forward
 416 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
 417 #
 418 # This means that there are cases where we need to create an extra seccomp
 419 # profile for the container to load.
 420 #
 421 # This returns a configuration snippet added to the raw lxc config.
 422 sub make_seccomp_config {
 423     my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
 424     # User-configured profile has precedence, note that the user's entry would
 425     # be written 'after' this line anyway...
 426     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.seccomp.profile')) {
 427         # Warn the user if this conflicts with a feature:
 428         my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
 429         warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
 430             if length($warn) > 0;
 431         return '';
 432     }
 433
 434     # Privileged containers keep using the default (which is already part of
 435     # the files included via lxc.include, so we don't need to write it out,
 436     # that way it stays admin-configurable via /usr/share/lxc/config/... as
 437     # well)
 438     return '' if !$unprivileged;
 439
 440     my $rules = {
 441         keyctl => ['errno 38'],
 442
 443         # Disable btrfs ioctrls since they don't work particularly well in user namespaces.
 444         # Particularly, without the mount option to enable rmdir removing snapshots, user
 445         # namespaces can create snapshots but neither `show` or `delete` them, which is quite
 446         # horrible, so for now, just disable this entirely:
 447         #
 448         # BTRFS_IOCTL_MAGIC 0x94, _IOC type shift is 8,
 449         # so `(req & 0xFF00) == 0x9400` is a btrfs ioctl and gets an EPERM
 450         ioctl  => ['errno 1 [1,0x9400,SCMP_CMP_MASKED_EQ,0xff00]'],
 451     };
 452
 453     my $raw_conf = '';
 454
 455     # Unprivileged containers will get keyctl() disabled by default as a
 456     # workaround for systemd-networkd behavior. But we have an option to
 457     # explicitly enable it:
 458     if ($features->{keyctl}) {
 459         delete $rules->{keyctl};
 460     }
 461
 462     # By default, unprivileged containers cannot use `mknod` at all.
 463     # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
 464     # but for now this is experimental, so it has to be enabled via a feature
 465     # flag.
 466     # Note that we only handle block and char devices (like lxd), the rest we
 467     # leave up to the kernel. We may in the future remove this if seccomp gets
 468     # a way to tell the kernel to "continue" a syscall.
 469     if ($features->{mknod}) {
 470         my ($ok, $kernel) = PVE::ProcFSTools::check_kernel_release(5, 3);
 471         if (!$ok) {
 472             die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
 473         }
 474
 475         $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
 476         $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
 477
 478         $rules->{mknod} = [
 479             # condition: (mode & S_IFMT) == S_IFCHR
 480             'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
 481             # condition: (mode & S_IFMT) == S_IFBLK
 482             'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
 483         ];
 484         $rules->{mknodat} = [
 485             # condition: (mode & S_IFMT) == S_IFCHR
 486             'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
 487             # condition: (mode & S_IFMT) == S_IFBLK
 488             'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
 489         ];
 490     }
 491
 492     # Now build the custom seccomp rule text...
 493     my $extra_rules = join("\n", map {
 494         my $syscall = $_;
 495         map { "$syscall $_" } $rules->{$syscall}->@*
 496     } sort keys %$rules) . "\n";
 497
 498     return $raw_conf if $extra_rules eq "\n";
 499
 500     # We still have the "most common" config readily available, so don't write
 501     # out that one:
 502     if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
 503         # we have no extra $raw_conf and use the same we had in pve 6.1:
 504         return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
 505     }
 506
 507     # Write the rule file to the container's config path:
 508     my $rule_file = "$conf_dir/rules.seccomp";
 509     my $rule_data = file_get_contents("$LXC_CONFIG_PATH/common.seccomp")
 510         . $extra_rules;
 511     file_set_contents($rule_file, $rule_data);
 512     $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
 513
 514     return $raw_conf;
 515 }
 516
 517 # Since lxc-3.0.2 we can have lxc generate a profile for the container
 518 # automatically. The default should be equivalent to the old
 519 # `lxc-container-default-cgns` profile.
 520 #
 521 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
 522 # additional lines into the profile. We can use that to allow mounting specific
 523 # file systems.
 524 sub make_apparmor_config {
 525     my ($conf, $unprivileged, $features) = @_;
 526
 527     # user-configured profile has precedence, but first we go through our own
 528     # code to figure out whether we should warn the user:
 529
 530     my $raw = "lxc.apparmor.profile = generated\n";
 531     my @profile_uses;
 532
 533     if ($features->{fuse}) {
 534         # For the informational warning:
 535         push @profile_uses, 'features:fuse';
 536     }
 537
 538     # There's lxc.apparmor.allow_nesting now, which will add the necessary
 539     # apparmor lines, create an apparmor namespace for the container, but also
 540     # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
 541     # lxcfs mounted over them, because that would prevent the container from
 542     # mounting new instances of them for nested containers.
 543     if ($features->{nesting}) {
 544         push @profile_uses, 'features:nesting';
 545         $raw .= "lxc.apparmor.allow_nesting = 1\n"
 546     } else {
 547         # In the default profile in /etc/apparmor.d we patch this in because
 548         # otherwise a container can for example run `chown` on /sys, breaking
 549         # access to it for non-CAP_DAC_OVERRIDE tools on the host:
 550         $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
 551         $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
 552         # Preferably we could use the 'remount' flag but this does not sit well
 553         # with apparmor_parser currently:
 554         #  mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
 555     }
 556
 557     if (my $mount = $features->{mount}) {
 558         push @profile_uses, 'features:mount';
 559         foreach my $fs (PVE::Tools::split_list($mount)) {
 560             $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
 561         }
 562     }
 563
 564     # More to come?
 565
 566     if (PVE::LXC::Config->has_lxc_entry($conf, 'lxc.apparmor.profile')) {
 567         if (length(my $used = join(', ', @profile_uses))) {
 568             warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
 569         }
 570         return '';
 571     }
 572
 573     return $raw;
 574 }
 575
 576 sub update_lxc_config {
 577     my ($vmid, $conf) = @_;
 578
 579     my $dir = "/var/lib/lxc/$vmid";
 580
 581     if ($conf->{template}) {
 582
 583         unlink "$dir/config";
 584
 585         return;
 586     }
 587
 588     my ($lxc_major, $lxc_minor) = get_lxc_version();
 589
 590     my $raw = '';
 591
 592     if ($lxc_major >= 4) {
 593         # Explicitly don't use relative directories, which is the default, but
 594         # note that we do this mostly because they are only applied for *some*
 595         # cgroups. Our pve-container@.service now starts lxc-start with `-F`,
 596         # so we also don't need to worry about the new monitor cgroup to
 597         # confuse systemd.
 598         $raw .= "lxc.cgroup.relative = 0\n";
 599
 600         # To make things easier, let's keep our previous cgroup layout and
 601         # simply move the monitor outside:
 602         $raw .= "lxc.cgroup.dir.monitor = lxc.monitor/$vmid\n";
 603         # cgroup namespace separation for stronger limits:
 604         $raw .= "lxc.cgroup.dir.container = lxc/$vmid\n";
 605         $raw .= "lxc.cgroup.dir.container.inner = ns\n";
 606     }
 607
 608     die "missing 'arch' - internal error" if !$conf->{arch};
 609     $raw .= "lxc.arch = $conf->{arch}\n";
 610
 611     my $custom_idmap = PVE::LXC::Config->has_lxc_entry($conf, 'lxc.idmap');
 612     my $unprivileged = $conf->{unprivileged} || $custom_idmap;
 613
 614     my $ostype = $conf->{ostype} || die "missing 'ostype' - internal error";
 615
 616     File::Path::mkpath($dir);
 617
 618     my $cfgpath = '/usr/share/lxc/config';
 619     my $inc = "$cfgpath/$ostype.common.conf";
 620     $inc ="$cfgpath/common.conf" if !-f $inc;
 621     $raw .= "lxc.include = $inc\n";
 622     if ($unprivileged) {
 623         $inc = "$cfgpath/$ostype.userns.conf";
 624         $inc = "$cfgpath/userns.conf" if !-f $inc;
 625         $raw .= "lxc.include = $inc\n";
 626     }
 627
 628     my $features = PVE::LXC::Config->parse_features($conf->{features});
 629
 630     $raw .= make_seccomp_config($conf, $vmid, $dir, $unprivileged, $features);
 631     $raw .= make_apparmor_config($conf, $unprivileged, $features);
 632     if ($features->{fuse}) {
 633         $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
 634         $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
 635     }
 636
 637     if ($unprivileged && !$features->{force_rw_sys}) {
 638         # unpriv. CT default to sys:rw, but that doesn't always plays well with
 639         # systemd, e.g., systemd-networkd https://systemd.io/CONTAINER_INTERFACE/
 640         $raw .= "lxc.mount.auto = sys:mixed\n";
 641     }
 642
 643     # WARNING: DO NOT REMOVE this without making sure that loop device nodes
 644     # cannot be exposed to the container with r/w access (cgroup perms).
 645     # When this is enabled mounts will still remain in the monitor's namespace
 646     # after the container unmounted them and thus will not detach from their
 647     # files while the container is running!
 648     $raw .= "lxc.monitor.unshare = 1\n";
 649
 650     my ($cgv1, $cgv2) = PVE::CGroup::get_cgroup_controllers();
 651
 652     # Should we read them from /etc/subuid?
 653     if ($unprivileged && !$custom_idmap) {
 654         $raw .= "lxc.idmap = u 0 100000 65536\n";
 655         $raw .= "lxc.idmap = g 0 100000 65536\n";
 656     }
 657
 658     if (!PVE::LXC::Config->has_dev_console($conf)) {
 659         $raw .= "lxc.console.path = none\n";
 660         if ($cgv1->{devices}) {
 661             $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n";
 662         } elsif (defined($cgv2)) {
 663             $raw .= "lxc.cgroup2.devices.deny = c 5:1 rwm\n";
 664         }
 665     }
 666
 667     my $ttycount = PVE::LXC::Config->get_tty_count($conf);
 668     $raw .= "lxc.tty.max = $ttycount\n";
 669
 670     # some init scripts expect a linux terminal (turnkey).
 671     $raw .= "lxc.environment = TERM=linux\n";
 672
 673     my $utsname = $conf->{hostname} || "CT$vmid";
 674     $raw .= "lxc.uts.name = $utsname\n";
 675
 676     if ($cgv1->{memory}) {
 677         my $memory = $conf->{memory} || 512;
 678         my $swap = $conf->{swap} // 0;
 679
 680         my $lxcmem = int($memory*1024*1024);
 681         $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
 682
 683         my $lxcswap = int(($memory + $swap)*1024*1024);
 684         $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
 685     } elsif ($cgv2->{memory}) {
 686         my $memory = $conf->{memory} || 512;
 687         my $swap = $conf->{swap} // 0;
 688
 689         my $lxcmem = int($memory*1024*1024);
 690         $raw .= "lxc.cgroup2.memory.max = $lxcmem\n";
 691
 692         my $lxcswap = int($swap*1024*1024);
 693         $raw .= "lxc.cgroup2.memory.swap.max = $lxcswap\n";
 694     }
 695
 696     if ($cgv1->{cpu}) {
 697         if (my $cpulimit = $conf->{cpulimit}) {
 698             $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
 699             my $value = int(100000*$cpulimit);
 700             $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
 701         }
 702
 703         my $shares = PVE::CGroup::clamp_cpu_shares($conf->{cpuunits});
 704         $raw .= "lxc.cgroup.cpu.shares = $shares\n";
 705     } elsif ($cgv2->{cpu}) {
 706         # See PVE::CGroup
 707         if (my $cpulimit = $conf->{cpulimit}) {
 708             my $value = int(100000*$cpulimit);
 709             $raw .= "lxc.cgroup2.cpu.max = $value 100000\n";
 710         }
 711
 712         if (defined(my $shares = $conf->{cpuunits})) {
 713             $shares = PVE::CGroup::clamp_cpu_shares($shares);
 714             $raw .= "lxc.cgroup2.cpu.weight = $shares\n";
 715         }
 716     }
 717
 718     die "missing 'rootfs' configuration\n"
 719         if !defined($conf->{rootfs});
 720
 721     my $mountpoint = PVE::LXC::Config->parse_volume('rootfs', $conf->{rootfs});
 722
 723     $raw .= "lxc.rootfs.path = $dir/rootfs\n";
 724
 725     foreach my $k (sort keys %$conf) {
 726         next if $k !~ m/^net(\d+)$/;
 727         my $ind = $1;
 728         my $d = PVE::LXC::Config->parse_lxc_network($conf->{$k});
 729         $raw .= "lxc.net.$ind.type = veth\n";
 730         $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
 731         $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr});
 732         $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name});
 733
 734         # Keep container from starting with invalid mtu configuration
 735         if (my $mtu = $d->{mtu}) {
 736             my $bridge_mtu = PVE::Network::read_bridge_mtu($d->{bridge});
 737             die "$k: MTU size '$mtu' is bigger than bridge MTU '$bridge_mtu'\n"
 738                 if ($mtu > $bridge_mtu);
 739
 740             $raw .= "lxc.net.$ind.mtu = $mtu\n";
 741         }
 742
 743         # Starting with lxc 4.0, we do not patch lxc to execute our up-scripts.
 744         if ($lxc_major >= 4) {
 745             $raw .= "lxc.net.$ind.script.up = /usr/share/lxc/lxcnetaddbr\n";
 746         }
 747     }
 748
 749     my $had_cpuset = 0;
 750     if (my $lxcconf = $conf->{lxc}) {
 751         foreach my $entry (@$lxcconf) {
 752             my ($k, $v) = @$entry;
 753             $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus' || $k eq 'lxc.cgroup2.cpuset.cpus';
 754             $raw .= "$k = $v\n";
 755         }
 756     }
 757
 758     my $cpuset;
 759     my ($cpuset_cgroup, $cpuset_version) = eval { PVE::CGroup::cpuset_controller_path() };
 760     if (defined($cpuset_cgroup)) {
 761         $cpuset = eval { PVE::CpuSet->new_from_path("$cpuset_cgroup/lxc", 1) }
 762             || PVE::CpuSet->new_from_path($cpuset_cgroup, 1);
 763     }
 764     my $cores = $conf->{cores};
 765     if (!$had_cpuset && $cores && $cpuset) {
 766         my @members = $cpuset->members();
 767         while (scalar(@members) > $cores) {
 768             my $randidx = int(rand(scalar(@members)));
 769             $cpuset->delete($members[$randidx]);
 770             splice(@members, $randidx, 1); # keep track of the changes
 771         }
 772         my $ver = $cpuset_version == 1 ? '' : '2';
 773         $raw .= "lxc.cgroup$ver.cpuset.cpus = ".$cpuset->short_string()."\n";
 774     }
 775
 776     File::Path::mkpath("$dir/rootfs");
 777
 778     PVE::Tools::file_set_contents("$dir/config", $raw);
 779 }
 780
 781 # verify and cleanup nameserver list (replace \0 with ' ')
 782 sub verify_nameserver_list {
 783     my ($nameserver_list) = @_;
 784
 785     my @list = ();
 786     foreach my $server (PVE::Tools::split_list($nameserver_list)) {
 787         PVE::LXC::Config::verify_ip_with_ll_iface($server);
 788         push @list, $server;
 789     }
 790
 791     return join(' ', @list);
 792 }
 793
 794 sub verify_searchdomain_list {
 795     my ($searchdomain_list) = @_;
 796
 797     my @list = ();
 798     foreach my $server (PVE::Tools::split_list($searchdomain_list)) {
 799         # todo: should we add checks for valid dns domains?
 800         push @list, $server;
 801     }
 802
 803     return join(' ', @list);
 804 }
 805
 806 sub get_console_command {
 807     my ($vmid, $conf, $escapechar) = @_;
 808
 809     # '-1' as $escapechar disables keyboard escape sequence
 810     # any other passed char (a-z) will result in <Ctrl+$escapechar q>
 811
 812     my $cmode = PVE::LXC::Config->get_cmode($conf);
 813
 814     my $cmd = [];
 815     if ($cmode eq 'console') {
 816         push @$cmd, 'lxc-console', '-n',  $vmid, '-t', 0;
 817         push @$cmd, '-e', $escapechar if $escapechar;
 818     } elsif ($cmode eq 'tty') {
 819         push @$cmd, 'lxc-console', '-n',  $vmid;
 820         push @$cmd, '-e', $escapechar if $escapechar;
 821     } elsif ($cmode eq 'shell') {
 822         push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
 823     } else {
 824         die "internal error";
 825     }
 826
 827     return $cmd;
 828 }
 829
 830 sub get_primary_ips {
 831     my ($conf) = @_;
 832
 833     # return data from net0
 834
 835     return undef if !defined($conf->{net0});
 836     my $net = PVE::LXC::Config->parse_lxc_network($conf->{net0});
 837
 838     my $ipv4 = $net->{ip};
 839     if ($ipv4) {
 840         if ($ipv4 =~ /^(dhcp|manual)$/) {
 841             $ipv4 = undef
 842         } else {
 843             $ipv4 =~ s!/\d+$!!;
 844         }
 845     }
 846     my $ipv6 = $net->{ip6};
 847     if ($ipv6) {
 848         if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
 849             $ipv6 = undef;
 850         } else {
 851             $ipv6 =~ s!/\d+$!!;
 852         }
 853     }
 854
 855     return ($ipv4, $ipv6);
 856 }
 857
 858 sub delete_mountpoint_volume {
 859     my ($storage_cfg, $vmid, $volume) = @_;
 860
 861     return if PVE::LXC::Config->classify_mountpoint($volume) ne 'volume';
 862
 863     my ($vtype, $name, $owner) = PVE::Storage::parse_volname($storage_cfg, $volume);
 864
 865     if ($vmid == $owner) {
 866         PVE::Storage::vdisk_free($storage_cfg, $volume);
 867     } else {
 868         warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
 869     }
 870 }
 871
 872 sub destroy_lxc_container {
 873     my ($storage_cfg, $vmid, $conf, $replacement_conf, $purge_unreferenced) = @_;
 874
 875     my $volids = {};
 876     my $remove_volume = sub {
 877         my ($ms, $mountpoint) = @_;
 878
 879         my $volume = $mountpoint->{volume};
 880
 881         return if $volids->{$volume};
 882         $volids->{$volume} = 1;
 883
 884         delete_mountpoint_volume($storage_cfg, $vmid, $volume);
 885     };
 886     PVE::LXC::Config->foreach_volume_full($conf, {include_unused => 1}, $remove_volume);
 887
 888     PVE::LXC::Config->foreach_volume_full($conf->{pending}, {include_unused => 1}, $remove_volume);
 889
 890     if ($purge_unreferenced) { # also remove unreferenced disk
 891         my $vmdisks = PVE::Storage::vdisk_list($storage_cfg, undef, $vmid, undef, 'rootdir');
 892         PVE::Storage::foreach_volid($vmdisks, sub {
 893             my ($volid, $sid, $volname, $d) = @_;
 894             eval { PVE::Storage::vdisk_free($storage_cfg, $volid) };
 895             warn $@ if $@;
 896         });
 897     }
 898
 899     rmdir "/var/lib/lxc/$vmid/rootfs";
 900     unlink "/var/lib/lxc/$vmid/config";
 901     rmdir "/var/lib/lxc/$vmid";
 902     if (defined $replacement_conf) {
 903         PVE::LXC::Config->write_config($vmid, $replacement_conf);
 904     } else {
 905         PVE::LXC::Config->destroy_config($vmid);
 906     }
 907 }
 908
 909 sub vm_stop_cleanup {
 910     my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
 911
 912     return if $keepActive;
 913
 914     eval {
 915         my $vollist = PVE::LXC::Config->get_vm_volumes($conf);
 916         PVE::Storage::deactivate_volumes($storage_cfg, $vollist);
 917     };
 918     warn $@ if $@; # avoid errors - just warn
 919 }
 920
 921 sub net_tap_plug : prototype($$) {
 922     my ($iface, $net) = @_;
 923     my ($bridge, $tag, $firewall, $trunks, $rate, $hwaddr) =
 924         $net->@{'bridge', 'tag', 'firewall', 'trunks', 'rate', 'hwaddr'};
 925
 926     if ($have_sdn) {
 927         PVE::Network::SDN::Zones::tap_plug($iface, $bridge, $tag, $firewall, $trunks, $rate);
 928         PVE::Network::SDN::Zones::add_bridge_fdb($iface, $hwaddr, $bridge, $firewall);
 929     } else {
 930         PVE::Network::tap_plug($iface, $bridge, $tag, $firewall, $trunks, $rate, { mac => $hwaddr });
 931     }
 932 }
 933
 934 sub update_net {
 935     my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
 936
 937     if ($newnet->{type} ne 'veth') {
 938         # for when there are physical interfaces
 939         die "cannot update interface of type $newnet->{type}";
 940     }
 941
 942     my $veth = "veth${vmid}i${netid}";
 943     my $eth = $newnet->{name};
 944
 945     if (my $oldnetcfg = $conf->{$opt}) {
 946         my $oldnet = PVE::LXC::Config->parse_lxc_network($oldnetcfg);
 947
 948         if (safe_string_ne($oldnet->{hwaddr}, $newnet->{hwaddr}) ||
 949             safe_string_ne($oldnet->{name}, $newnet->{name})) {
 950
 951             PVE::Network::veth_delete($veth);
 952             delete $conf->{$opt};
 953             PVE::LXC::Config->write_config($vmid, $conf);
 954
 955             hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 956
 957         } else {
 958             if (safe_string_ne($oldnet->{bridge}, $newnet->{bridge}) ||
 959                 safe_num_ne($oldnet->{tag}, $newnet->{tag}) ||
 960                 safe_num_ne($oldnet->{firewall}, $newnet->{firewall})
 961             ) {
 962
 963                 if ($oldnet->{bridge}) {
 964                     PVE::Network::tap_unplug($veth);
 965                     foreach (qw(bridge tag firewall)) {
 966                         delete $oldnet->{$_};
 967                     }
 968                     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 969                     PVE::LXC::Config->write_config($vmid, $conf);
 970                 }
 971
 972                 PVE::LXC::net_tap_plug($veth, $newnet);
 973
 974                 # This includes the rate:
 975                 foreach (qw(bridge tag firewall rate)) {
 976                     $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
 977                 }
 978             } elsif (safe_string_ne($oldnet->{rate}, $newnet->{rate})) {
 979                 # Rate can be applied on its own but any change above needs to
 980                 # include the rate in tap_plug since OVS resets everything.
 981                 PVE::Network::tap_rate_limit($veth, $newnet->{rate});
 982                 $oldnet->{rate} = $newnet->{rate}
 983             }
 984             $conf->{$opt} = PVE::LXC::Config->print_lxc_network($oldnet);
 985             PVE::LXC::Config->write_config($vmid, $conf);
 986         }
 987     } else {
 988         hotplug_net($vmid, $conf, $opt, $newnet, $netid);
 989     }
 990
 991     update_ipconfig($vmid, $conf, $opt, $eth, $newnet, $rootdir);
 992 }
 993
 994 sub hotplug_net {
 995     my ($vmid, $conf, $opt, $newnet, $netid) = @_;
 996
 997     my $veth = "veth${vmid}i${netid}";
 998     my $vethpeer = $veth . "p";
 999     my $eth = $newnet->{name};
1000
1001     if ($have_sdn) {
1002         PVE::Network::SDN::Zones::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr});
1003     } else {
1004         PVE::Network::veth_create($veth, $vethpeer, $newnet->{bridge}, $newnet->{hwaddr});
1005     }
1006
1007     PVE::LXC::net_tap_plug($veth, $newnet);
1008
1009     # attach peer in container
1010     my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
1011     PVE::Tools::run_command($cmd);
1012
1013     # link up peer in container
1014     $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up'  ];
1015     PVE::Tools::run_command($cmd);
1016
1017     my $done = { type => 'veth' };
1018     foreach (qw(bridge tag firewall hwaddr name)) {
1019         $done->{$_} = $newnet->{$_} if $newnet->{$_};
1020     }
1021     $conf->{$opt} = PVE::LXC::Config->print_lxc_network($done);
1022
1023     PVE::LXC::Config->write_config($vmid, $conf);
1024 }
1025
1026 sub update_ipconfig {
1027     my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
1028
1029     my $lxc_setup = PVE::LXC::Setup->new($conf, $rootdir);
1030
1031     my $optdata = PVE::LXC::Config->parse_lxc_network($conf->{$opt});
1032     my $deleted = [];
1033     my $added = [];
1034     my $nscmd = sub {
1035         my $cmdargs = shift;
1036         PVE::Tools::run_command(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
1037     };
1038     my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
1039
1040     my $change_ip_config = sub {
1041         my ($ipversion) = @_;
1042
1043         my $family_opt = "-$ipversion";
1044         my $suffix = $ipversion == 4 ? '' : $ipversion;
1045         my $gw= "gw$suffix";
1046         my $ip= "ip$suffix";
1047
1048         my $newip = $newnet->{$ip};
1049         my $newgw = $newnet->{$gw};
1050         my $oldip = $optdata->{$ip};
1051         my $oldgw = $optdata->{$gw};
1052
1053         my $change_ip = safe_string_ne($oldip, $newip);
1054         my $change_gw = safe_string_ne($oldgw, $newgw);
1055
1056         return if !$change_ip && !$change_gw;
1057
1058         # step 1: add new IP, if this fails we cancel
1059         my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
1060         if ($change_ip && $is_real_ip) {
1061             eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
1062             if (my $err = $@) {
1063                 warn $err;
1064                 return;
1065             }
1066         }
1067
1068         # step 2: replace gateway
1069         #   If this fails we delete the added IP and cancel.
1070         #   If it succeeds we save the config and delete the old IP, ignoring
1071         #   errors. The config is then saved.
1072         # Note: 'ip route replace' can add
1073         if ($change_gw) {
1074             if ($newgw) {
1075                 eval {
1076                     if ($is_real_ip && !PVE::Network::is_ip_in_cidr($newgw, $newip, $ipversion)) {
1077                         &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1078                     }
1079                     &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1080                 };
1081                 if (my $err = $@) {
1082                     warn $err;
1083                     # the route was not replaced, the old IP is still available
1084                     # rollback (delete new IP) and cancel
1085                     if ($change_ip) {
1086                         eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1087                         warn $@ if $@; # no need to die here
1088                     }
1089                     return;
1090                 }
1091             } else {
1092                 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1093                 # if the route was not deleted, the guest might have deleted it manually
1094                 # warn and continue
1095                 warn $@ if $@;
1096             }
1097             if ($oldgw && $oldip && !PVE::Network::is_ip_in_cidr($oldgw, $oldip)) {
1098                 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1099                 # warn if the route was deleted manually
1100                 warn $@ if $@;
1101             }
1102         }
1103
1104         # from this point on we save the configuration
1105         # step 3: delete old IP ignoring errors
1106         if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1107             # We need to enable promote_secondaries, otherwise our newly added
1108             # address will be removed along with the old one.
1109             my $promote = 0;
1110             eval {
1111                 if ($ipversion == 4) {
1112                     &$nscmd({ outfunc => sub { $promote = int(shift) } },
1113                             'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1114                     &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1115                 }
1116                 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1117             };
1118             warn $@ if $@; # no need to die here
1119
1120             if ($ipversion == 4) {
1121                 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1122             }
1123         }
1124
1125         foreach my $property ($ip, $gw) {
1126             if ($newnet->{$property}) {
1127                 $optdata->{$property} = $newnet->{$property};
1128             } else {
1129                 delete $optdata->{$property};
1130             }
1131         }
1132         $conf->{$opt} = PVE::LXC::Config->print_lxc_network($optdata);
1133         PVE::LXC::Config->write_config($vmid, $conf);
1134         $lxc_setup->setup_network($conf);
1135     };
1136
1137     &$change_ip_config(4);
1138     &$change_ip_config(6);
1139
1140 }
1141
1142 my $open_namespace = sub {
1143     my ($vmid, $pid, $kind) = @_;
1144     sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1145         or die "failed to open $kind namespace of container $vmid: $!\n";
1146     return $fd;
1147 };
1148
1149 my $enter_namespace = sub {
1150     my ($vmid, $pid, $kind, $type) = @_;
1151     my $fd = $open_namespace->($vmid, $pid, $kind);
1152     PVE::Tools::setns(fileno($fd), $type)
1153         or die "failed to enter $kind namespace of container $vmid: $!\n";
1154     close $fd;
1155 };
1156
1157 my $get_container_namespace = sub {
1158     my ($vmid, $pid, $kind) = @_;
1159
1160     my $pidfd;
1161     if (!defined($pid)) {
1162         # Pin the pid while we're grabbing its stuff from /proc
1163         ($pid, $pidfd) = open_lxc_pid($vmid)
1164             or die "failed to open pidfd of container $vmid\'s init process\n";
1165     }
1166
1167     return $open_namespace->($vmid, $pid, $kind);
1168 };
1169
1170 my $do_syncfs = sub {
1171     my ($vmid, $pid, $socket) = @_;
1172
1173     &$enter_namespace($vmid, $pid, 'mnt', PVE::Tools::CLONE_NEWNS);
1174
1175     # Tell the parent process to start reading our /proc/mounts
1176     print {$socket} "go\n";
1177     $socket->flush();
1178
1179     # Receive /proc/self/mounts
1180     my $mountdata = do { local $/ = undef; <$socket> };
1181     close $socket;
1182
1183     my %nosyncfs = (
1184         cgroup => 1,
1185         cgroup2 => 1,
1186         devtmpfs => 1,
1187         devpts => 1,
1188         'fuse.lxcfs' => 1,
1189         fusectl => 1,
1190         mqueue => 1,
1191         proc => 1,
1192         sysfs => 1,
1193         tmpfs => 1,
1194     );
1195
1196     # Now sync all mountpoints...
1197     my $mounts = PVE::ProcFSTools::parse_mounts($mountdata);
1198     foreach my $mp (@$mounts) {
1199         my ($what, $dir, $fs) = @$mp;
1200         next if $nosyncfs{$fs};
1201         eval { PVE::Tools::sync_mountpoint($dir); };
1202         warn $@ if $@;
1203     }
1204 };
1205
1206 sub sync_container_namespace {
1207     my ($vmid) = @_;
1208     my $pid = find_lxc_pid($vmid);
1209
1210     # SOCK_DGRAM is nicer for barriers but cannot be slurped
1211     socketpair my $pfd, my $cfd, AF_UNIX, SOCK_STREAM, PF_UNSPEC
1212         or die "failed to create socketpair: $!\n";
1213
1214     my $child = fork();
1215     die "fork failed: $!\n" if !defined($child);
1216
1217     if (!$child) {
1218         eval {
1219             close $pfd;
1220             &$do_syncfs($vmid, $pid, $cfd);
1221         };
1222         if (my $err = $@) {
1223             warn $err;
1224             POSIX::_exit(1);
1225         }
1226         POSIX::_exit(0);
1227     }
1228     close $cfd;
1229     my $go = <$pfd>;
1230     die "failed to enter container namespace\n" if $go ne "go\n";
1231
1232     open my $mounts, '<', "/proc/$child/mounts"
1233         or die "failed to open container's /proc/mounts: $!\n";
1234     my $mountdata = do { local $/ = undef; <$mounts> };
1235     close $mounts;
1236     print {$pfd} $mountdata;
1237     close $pfd;
1238
1239     while (waitpid($child, 0) != $child) {}
1240     die "failed to sync container namespace\n" if $? != 0;
1241 }
1242
1243 sub template_create {
1244     my ($vmid, $conf) = @_;
1245
1246     my $storecfg = PVE::Storage::config();
1247
1248     PVE::LXC::Config->foreach_volume($conf, sub {
1249         my ($ms, $mountpoint) = @_;
1250
1251         my $volid = $mountpoint->{volume};
1252
1253         die "Template feature is not available for '$volid'\n"
1254             if !PVE::Storage::volume_has_feature($storecfg, 'template', $volid);
1255     });
1256
1257     PVE::LXC::Config->foreach_volume($conf, sub {
1258         my ($ms, $mountpoint) = @_;
1259
1260         my $volid = $mountpoint->{volume};
1261
1262         PVE::Storage::activate_volumes($storecfg, [$volid]);
1263
1264         my $template_volid = PVE::Storage::vdisk_create_base($storecfg, $volid);
1265         $mountpoint->{volume} = $template_volid;
1266         $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1267     });
1268
1269     PVE::LXC::Config->write_config($vmid, $conf);
1270 }
1271
1272 sub check_ct_modify_config_perm {
1273     my ($rpcenv, $authuser, $vmid, $pool, $oldconf, $newconf, $delete, $unprivileged) = @_;
1274
1275     return 1 if $authuser eq 'root@pam';
1276     my $storage_cfg = PVE::Storage::config();
1277
1278     my $check = sub {
1279         my ($opt, $delete) = @_;
1280         if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1281             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1282         } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1283             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1284             return if $delete;
1285             my $data = PVE::LXC::Config->parse_volume($opt, $newconf->{$opt});
1286             raise_perm_exc("mount point type $data->{type} is only allowed for root\@pam")
1287                 if $data->{type} ne 'volume';
1288             my $volid = $data->{volume};
1289             if ($volid =~ $NEW_DISK_RE) {
1290                 my $sid = $1;
1291                 $rpcenv->check($authuser, "/storage/$sid", ['Datastore.AllocateSpace']);
1292             } else {
1293                 PVE::Storage::check_volume_access(
1294                     $rpcenv,
1295                     $authuser,
1296                     $storage_cfg,
1297                     $vmid,
1298                     $volid,
1299                     'rootdir',
1300                 );
1301             }
1302         } elsif ($opt eq 'memory' || $opt eq 'swap') {
1303             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1304         } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1305                  $opt eq 'searchdomain' || $opt eq 'hostname') {
1306             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1307         } elsif ($opt eq 'features') {
1308             raise_perm_exc("changing feature flags for privileged container is only allowed for root\@pam")
1309                 if !$unprivileged;
1310
1311             my $nesting_changed = 0;
1312             my $other_changed = 0;
1313             if (!$delete) {
1314                 my $features = PVE::LXC::Config->parse_features($newconf->{$opt});
1315                 if (defined($oldconf) && $oldconf->{$opt}) {
1316                     # existing container with features
1317                     my $old_features = PVE::LXC::Config->parse_features($oldconf->{$opt});
1318                     for my $feature ((keys %$old_features, keys %$features)) {
1319                         my $old = $old_features->{$feature} // '';
1320                         my $new = $features->{$feature} // '';
1321                         if ($old ne $new) {
1322                             if ($feature eq 'nesting') {
1323                                 $nesting_changed = 1;
1324                                 next;
1325                             } else {
1326                                 $other_changed = 1;
1327                                 last;
1328                             }
1329                         }
1330                     }
1331                 } else {
1332                     # new container or no features defined
1333                     if (scalar(keys %$features) == 1 && $features->{nesting}) {
1334                         $nesting_changed = 1;
1335                     } elsif (scalar(keys %$features) > 0) {
1336                         $other_changed = 1;
1337                     }
1338                 }
1339             } else {
1340                 my $features = PVE::LXC::Config->parse_features($oldconf->{$opt});
1341                 if (scalar(keys %$features) == 1 && $features->{nesting}) {
1342                     $nesting_changed = 1;
1343                 } elsif (scalar(keys %$features) > 0) {
1344                     $other_changed = 1;
1345                 }
1346             }
1347             raise_perm_exc("changing feature flags (except nesting) is only allowed for root\@pam")
1348                 if $other_changed;
1349             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Allocate'])
1350                 if $nesting_changed;
1351         } elsif ($opt eq 'hookscript') {
1352             # For now this is restricted to root@pam
1353             raise_perm_exc("changing the hookscript is only allowed for root\@pam");
1354         } elsif ($opt eq 'tags') {
1355             my $old = $oldconf->{$opt};
1356             my $new = $delete ? '' : $newconf->{$opt};
1357             PVE::GuestHelpers::assert_tag_permissions($vmid, $old, $new, $rpcenv, $authuser);
1358         } else {
1359             $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1360         }
1361     };
1362
1363     foreach my $opt (keys %$newconf) {
1364         &$check($opt, 0);
1365     }
1366     foreach my $opt (@$delete) {
1367         &$check($opt, 1);
1368     }
1369
1370     return 1;
1371 }
1372
1373 sub umount_all {
1374     my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1375
1376     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1377     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1378
1379     my $res = 1;
1380
1381     PVE::LXC::Config->foreach_volume_full($conf, {'reverse' => 1}, sub {
1382         my ($ms, $mountpoint) = @_;
1383
1384         my $volid = $mountpoint->{volume};
1385         my $mount = $mountpoint->{mp};
1386
1387         return if !$volid || !$mount;
1388
1389         my $mount_path = "$rootdir/$mount";
1390         $mount_path =~ s!/+!/!g;
1391
1392         return if !PVE::ProcFSTools::is_mounted($mount_path);
1393
1394         eval {
1395             PVE::Tools::run_command(['umount', '-d', $mount_path]);
1396         };
1397         if (my $err = $@) {
1398             if ($noerr) {
1399                 $res = 0;
1400                 warn $err;
1401             } else {
1402                 die $err;
1403             }
1404         }
1405     });
1406
1407     return $res; # tell caller if (some) umounts failed for the noerr case
1408 }
1409
1410 sub mount_all {
1411     my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1412
1413     my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1414     File::Path::make_path($rootdir);
1415
1416     my $volid_list = PVE::LXC::Config->get_vm_volumes($conf);
1417     PVE::Storage::activate_volumes($storage_cfg, $volid_list);
1418
1419     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
1420
1421     eval {
1422         PVE::LXC::Config->foreach_volume($conf, sub {
1423             my ($ms, $mountpoint) = @_;
1424
1425             $mountpoint->{ro} = 0 if $ignore_ro;
1426
1427             mountpoint_mount($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1428         });
1429     };
1430     if (my $err = $@) {
1431         warn "mounting container failed\n";
1432         umount_all($vmid, $storage_cfg, $conf, 1);
1433         die $err;
1434     }
1435
1436     return $rootdir;
1437 }
1438
1439
1440 sub mountpoint_mount_path {
1441     my ($mountpoint, $storage_cfg, $snapname) = @_;
1442
1443     return mountpoint_mount($mountpoint, undef, $storage_cfg, $snapname);
1444 }
1445
1446 sub query_loopdev {
1447     my ($path) = @_;
1448     my $found;
1449     my $parser = sub {
1450         my $line = shift;
1451         if ($line =~ m@^(/dev/loop\d+):@) {
1452             $found = $1;
1453         }
1454     };
1455     my $cmd = ['losetup', '--associated', $path];
1456     PVE::Tools::run_command($cmd, outfunc => $parser);
1457     return $found;
1458 }
1459
1460 # Run a function with a file attached to a loop device.
1461 # The loop device is always detached afterwards (or set to autoclear).
1462 # Returns the loop device.
1463 sub run_with_loopdev {
1464     my ($func, $file, $readonly) = @_;
1465     my $device = query_loopdev($file);
1466     # Try to reuse an existing device
1467     if ($device) {
1468         # We assume that whoever setup the loop device is responsible for
1469         # detaching it.
1470         &$func($device);
1471         return $device;
1472     }
1473
1474     my $parser = sub {
1475         my $line = shift;
1476         if ($line =~ m@^(/dev/loop\d+)$@) {
1477             $device = $1;
1478         }
1479     };
1480     my $losetup_cmd = [
1481         'losetup',
1482         '--show',
1483         '-f',
1484         $file,
1485     ];
1486     push @$losetup_cmd, '-r' if $readonly;
1487     PVE::Tools::run_command($losetup_cmd, outfunc => $parser);
1488     die "failed to setup loop device for $file\n" if !$device;
1489     eval { &$func($device); };
1490     my $err = $@;
1491     PVE::Tools::run_command(['losetup', '-d', $device]);
1492     die $err if $err;
1493     return $device;
1494 }
1495
1496 # In scalar mode: returns a file handle to the deepest directory node.
1497 # In list context: returns a list of:
1498 #   * the deepest directory node
1499 #   * the 2nd deepest directory (parent of the above)
1500 #   * directory name of the last directory
1501 # So that the path $2/$3 should lead to $1 afterwards.
1502 sub walk_tree_nofollow($$$;$$) {
1503     my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1504
1505     sysopen(my $fd, $start, O_PATH | O_DIRECTORY)
1506         or die "failed to open start directory $start: $!\n";
1507
1508     return walk_tree_nofollow_fd($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1509 }
1510
1511
1512 sub walk_tree_nofollow_fd($$$$;$$) {
1513     my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1514
1515     # splitdir() returns '' for empty components including the leading /
1516     my @comps = grep { length($_)>0 } File::Spec->splitdir($subdir);
1517
1518     my $fd = $start_fd;
1519     my $dir = $start_dirname;
1520     my $last_component = undef;
1521     my $second = $fd;
1522     foreach my $component (@comps) {
1523         $dir .= "/$component";
1524         my $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1525
1526         if (!$next) {
1527             # failed, check for symlinks and try to create the path
1528             die "symlink encountered at: $dir\n" if $! == ELOOP || $! == ENOTDIR;
1529             die "cannot open directory $dir: $!\n" if !$mkdir;
1530
1531             # We don't check for errors on mkdirat() here and just try to
1532             # openat() again, since at least one error (EEXIST) is an
1533             # expected possibility if multiple containers start
1534             # simultaneously. If someone else injects a symlink now then
1535             # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1536             PVE::Tools::mkdirat(fileno($fd), $component, 0755);
1537
1538             $next = PVE::Tools::openat(fileno($fd), $component, O_NOFOLLOW | O_DIRECTORY);
1539             die "failed to create path: $dir: $!\n" if !$next;
1540
1541             PVE::Tools::fchownat(fileno($next), '', $rootuid, $rootgid, PVE::Tools::AT_EMPTY_PATH)
1542                 if defined($rootuid) && defined($rootgid);
1543         }
1544
1545         close $second if defined($last_component) && $second != $start_fd;
1546         $last_component = $component;
1547         $second = $fd;
1548         $fd = $next;
1549     }
1550
1551     return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1552     close $second if defined($last_component) && $second != $start_fd;
1553     return $fd;
1554 }
1555
1556 # To guard against symlink attack races against other currently running
1557 # containers with shared recursive bind mount hierarchies we prepare a
1558 # directory handle for the directory we're mounting over to verify the
1559 # mountpoint afterwards.
1560 sub __bindmount_prepare {
1561     my ($hostroot, $dir) = @_;
1562     my $srcdh = walk_tree_nofollow($hostroot, $dir, 0);
1563     return $srcdh;
1564 }
1565
1566 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1567 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1568 # we intended to bind mount.
1569 sub __bindmount_verify {
1570     my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1571     my $destdh;
1572     if ($parentfd) {
1573         # Open the mount point path coming from the parent directory since the
1574         # filehandle we would have gotten as first result of walk_tree_nofollow
1575         # earlier is still a handle to the underlying directory instead of the
1576         # mounted path.
1577         $destdh = PVE::Tools::openat(fileno($parentfd), $last_dir, PVE::Tools::O_PATH | O_NOFOLLOW | O_DIRECTORY);
1578         die "failed to open mount point: $!\n" if !$destdh;
1579         if ($ro) {
1580             my $dot = '.';
1581             # no separate function because 99% of the time it's the wrong thing to use.
1582             if (syscall(PVE::Syscall::faccessat, fileno($destdh), $dot, &POSIX::W_OK, 0) != -1) {
1583                 die "failed to mark bind mount read only\n";
1584             }
1585             die "read-only check failed: $!\n" if $! != EROFS;
1586         }
1587     } else {
1588         # For the rootfs we don't have a parentfd so we open the path directly.
1589         # Note that this means bindmounting any prefix of the host's
1590         # /var/lib/lxc/$vmid path into another container is considered a grave
1591         # security error.
1592         sysopen $destdh, $last_dir, O_PATH | O_DIRECTORY;
1593         die "failed to open mount point: $!\n" if !$destdh;
1594     }
1595
1596     my ($srcdev, $srcinode) = stat($srcdh);
1597     my ($dstdev, $dstinode) = stat($destdh);
1598     close $srcdh;
1599     close $destdh;
1600
1601     return ($srcdev == $dstdev && $srcinode == $dstinode);
1602 }
1603
1604 # Perform the actual bind mounting:
1605 sub __bindmount_do {
1606     my ($dir, $dest, $ro, @extra_opts) = @_;
1607     PVE::Tools::run_command(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1608     if ($ro) {
1609         eval { PVE::Tools::run_command(['mount', '-o', 'bind,remount,ro', $dest]); };
1610         if (my $err = $@) {
1611             warn "bindmount error\n";
1612             # don't leave writable bind-mounts behind...
1613             PVE::Tools::run_command(['umount', $dest]);
1614             die $err;
1615         }
1616     }
1617 }
1618
1619 sub bindmount {
1620     my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1621
1622     my $srcdh = __bindmount_prepare('/', $dir);
1623
1624     __bindmount_do($dir, $dest, $ro, @extra_opts);
1625
1626     if (!__bindmount_verify($srcdh, $parentfd, $last_dir, $ro)) {
1627         PVE::Tools::run_command(['umount', $dest]);
1628         die "detected mount path change at: $dir\n";
1629     }
1630 }
1631
1632 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1633 # from $rootdir and $mount and walk the path from $rootdir to the final
1634 # directory to check for symlinks.
1635 sub __mount_prepare_rootdir {
1636     my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1637     $rootdir =~ s!/+!/!g;
1638     $rootdir =~ s!/+$!!;
1639     my $mount_path = "$rootdir/$mount";
1640     my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow($rootdir, $mount, 1, $rootuid, $rootgid);
1641     return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1642 }
1643
1644 # use $rootdir = undef to just return the corresponding mount path
1645 sub mountpoint_mount {
1646     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1647     return __mountpoint_mount($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1648 }
1649
1650 sub mountpoint_stage {
1651     my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1652     my ($path, $loop, $dev) =
1653         __mountpoint_mount($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1654
1655     if (!defined($path)) {
1656         return undef if $! == ENOSYS;
1657         die "failed to mount subvolume: $!\n";
1658     }
1659
1660     # We clone the mount point and leave it there in order to keep them connected to eg. loop
1661     # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1662     # point).
1663     my $err;
1664     my $fd = PVE::Tools::open_tree(&AT_FDCWD, $stage_dir, &OPEN_TREE_CLOEXEC | &OPEN_TREE_CLONE)
1665         or die "open_tree() on mount point failed: $!\n";
1666
1667     return wantarray ? ($path, $loop, $dev, $fd) : $fd;
1668 }
1669
1670 sub mountpoint_insert_staged {
1671     my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1672
1673     if (!defined($rootdir_fd)) {
1674         sysopen($rootdir_fd, '.', O_PATH | O_DIRECTORY)
1675             or die "failed to open '.': $!\n";
1676     }
1677
1678     my $dest_fd = walk_tree_nofollow_fd('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1679
1680     PVE::Tools::move_mount(
1681         fileno($mount_fd),
1682         '',
1683         fileno($dest_fd),
1684         '',
1685         &MOVE_MOUNT_F_EMPTY_PATH | &MOVE_MOUNT_T_EMPTY_PATH,
1686     ) or die "failed to move '$opt' into container hierarchy: $!\n";
1687 }
1688
1689 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1690 #   can then open a file descriptor to it which can be used with the `move_mount` syscall.
1691 #   Note that if the kernel does not support the new mount API, this will not perform any action
1692 #   and return `undef` with $! = ENOSYS.
1693 sub __mountpoint_mount {
1694     my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1695
1696     if (defined($stage_mount) && !PVE::LXC::Tools::can_use_new_mount_api()) {
1697         $! = ENOSYS;
1698         return undef;
1699     }
1700
1701     # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1702     # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1703     # which we use in __bindmount_verify()!
1704     my $mount = $stage_mount ? '/': $mountpoint->{mp};
1705
1706     my $volid = $mountpoint->{volume};
1707     my $type = $mountpoint->{type};
1708     my $quota = !$snapname && !$mountpoint->{ro} && $mountpoint->{quota};
1709     my $mounted_dev;
1710
1711     return if !$volid || !$mount;
1712
1713     $mount =~ s!/+!/!g;
1714
1715     my $mount_path;
1716     my ($mpfd, $parentfd, $last_dir);
1717
1718     if (defined($rootdir)) {
1719         ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1720             __mount_prepare_rootdir($rootdir, $mount, $rootuid, $rootgid);
1721     }
1722
1723     if (defined($stage_mount)) {
1724         $mount_path = $rootdir;
1725     }
1726
1727     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1728
1729     die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1730
1731     my $optlist = [];
1732
1733     if (my $mountopts = $mountpoint->{mountoptions}) {
1734         my @opts = split(/;/, $mountpoint->{mountoptions});
1735         push @$optlist, grep { PVE::LXC::Config::is_valid_mount_option($_) } @opts;
1736     }
1737
1738     my $acl = $mountpoint->{acl};
1739     if (defined($acl)) {
1740         push @$optlist, ($acl ? 'acl' : 'noacl');
1741     }
1742
1743     my $optstring = join(',', @$optlist);
1744     my $readonly = $mountpoint->{ro};
1745
1746     my @extra_opts;
1747     @extra_opts = ('-o', $optstring) if $optstring;
1748
1749     if ($storage) {
1750
1751         my $scfg = PVE::Storage::storage_config($storage_cfg, $storage);
1752
1753         my $path = PVE::Storage::map_volume($storage_cfg, $volid, $snapname);
1754
1755         $path = PVE::Storage::path($storage_cfg, $volid, $snapname) if !defined($path);
1756
1757         my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1758             PVE::Storage::parse_volname($storage_cfg, $volid);
1759
1760         $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1761
1762         if ($format eq 'subvol') {
1763             if ($mount_path) {
1764                 my (undef, $name) = PVE::Storage::parse_volname($storage_cfg, $volid);
1765                 if (defined($snapname)) {
1766                     $name .= "\@$snapname";
1767                     if ($scfg->{type} eq 'zfspool') {
1768                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', "$scfg->{pool}/$name", $mount_path]);
1769                     } else {
1770                         die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1771                     }
1772                 } else {
1773                     if (defined($acl) && $scfg->{type} eq 'zfspool') {
1774                         my $acltype = ($acl ? 'acltype=posixacl' : 'acltype=noacl');
1775                         PVE::Tools::run_command(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1776                     }
1777                     bindmount($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1778                     warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1779                 }
1780             }
1781             return wantarray ? ($path, 0, undef) : $path;
1782         } elsif ($format eq 'raw' || $format eq 'iso') {
1783             # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1784             # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1785             # Our autodev hook expects the /dev/dm-* device currently
1786             # and will create the /dev/mapper symlink accordingly
1787             $path = Cwd::realpath($path);
1788             die "failed to get device path\n" if !$path;
1789             ($path) = ($path =~ /^(.*)$/s); #untaint
1790             my $domount = sub {
1791                 my ($path) = @_;
1792                 if ($mount_path) {
1793                     if ($format eq 'iso') {
1794                         PVE::Tools::run_command(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1795                     } elsif ($isBase || defined($snapname)) {
1796                         PVE::Tools::run_command(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1797                     } else {
1798                         if ($quota) {
1799                             push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1800                         }
1801                         push @extra_opts, '-o', 'ro' if $readonly;
1802                         PVE::Tools::run_command(['mount', @extra_opts, $path, $mount_path]);
1803                     }
1804                 }
1805             };
1806             my $use_loopdev = 0;
1807             if ($scfg->{content}->{rootdir}) {
1808                 if ($scfg->{path}) {
1809                     $mounted_dev = run_with_loopdev($domount, $path, $readonly);
1810                     $use_loopdev = 1;
1811                 } else {
1812                     $mounted_dev = $path;
1813                     &$domount($path);
1814                 }
1815             } else {
1816                 die "storage '$storage' does not support containers\n";
1817             }
1818             return wantarray ? ($path, $use_loopdev, $mounted_dev) : $path;
1819         } else {
1820             die "unsupported image format '$format'\n";
1821         }
1822     } elsif ($type eq 'device') {
1823         push @extra_opts, '-o', 'ro' if $readonly;
1824         push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1825         # See the NOTE above about devicemapper canonicalization
1826         my ($devpath) = (Cwd::realpath($volid) =~ /^(.*)$/s); # realpath() taints
1827         PVE::Tools::run_command(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1828         return wantarray ? ($volid, 0, $devpath) : $volid;
1829     } elsif ($type eq 'bind') {
1830         die "directory '$volid' does not exist\n" if ! -d $volid;
1831         bindmount($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1832         warn "cannot enable quota control for bind mounts\n" if $quota;
1833         return wantarray ? ($volid, 0, undef) : $volid;
1834     }
1835
1836     die "unsupported storage";
1837 }
1838
1839 sub mountpoint_hotplug :prototype($$$$$) {
1840     my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1841
1842     my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
1843
1844     # We do the rest in a fork with an unshared mount namespace, because:
1845     #  -) change our papparmor profile to that of /usr/bin/lxc-start
1846     #  -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1847     #     container's namespace, then mount it.
1848
1849     PVE::Tools::run_fork(sub {
1850         # Pin the container pid longer, we also need to get its monitor/parent:
1851         my ($ct_pid, $ct_pidfd) = open_lxc_pid($vmid)
1852             or die "failed to open pidfd of container $vmid\'s init process\n";
1853
1854         my ($monitor_pid, $monitor_pidfd) = open_ppid($ct_pid)
1855             or die "failed to open pidfd of container $vmid\'s monitor process\n";
1856
1857         my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1858         my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1859
1860         # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1861         # profile to lower our privileges to the same level we have in the start hook:
1862         sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY)
1863             or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1864         # But switch namespaces first, to make sure the namespace switches aren't blocked by
1865         # apparmor.
1866
1867         # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1868         # namespace for it to remain active there since the container will be able to unmount
1869         # hotplugged mount points and thereby potentially free up loop devices, which is a security
1870         # concern.
1871         PVE::Tools::setns(fileno($monitor_mnt_ns), PVE::Tools::CLONE_NEWNS);
1872         chdir('/')
1873             or die "failed to change root directory within the monitor's mount namespace: $!\n";
1874
1875         my $dir = get_staging_mount_path($opt);
1876
1877         # Now switch our apparmor profile before mounting:
1878         my $data = 'changeprofile /usr/bin/lxc-start';
1879         if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1880             die "failed to change apparmor profile: $!\n";
1881         }
1882         # Check errors on close as well:
1883         close($aa_fd)
1884             or die "failed to change apparmor profile (close() failed): $!\n";
1885
1886         my $mount_fd = mountpoint_stage($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1887
1888         PVE::Tools::setns(fileno($ct_mnt_ns), PVE::Tools::CLONE_NEWNS);
1889         chdir('/')
1890             or die "failed to change root directory within the container's mount namespace: $!\n";
1891
1892         mountpoint_insert_staged($mount_fd, undef, $mp->{mp}, $opt, $rootuid, $rootgid);
1893     });
1894 }
1895
1896 # Create a directory in the mountpoint staging tempfs.
1897 sub get_staging_mount_path($) {
1898     my ($opt) = @_;
1899
1900     my $target = get_staging_tempfs() . "/$opt";
1901     if (!mkdir($target) && $! != EEXIST) {
1902         die "failed to create directory $target: $!\n";
1903     }
1904
1905     return $target;
1906 }
1907
1908 # Mount /run/pve/mountpoints as tmpfs
1909 sub get_staging_tempfs() {
1910     # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1911     # mounts to that.
1912     my $target = '/var/lib/lxc/.pve-staged-mounts';
1913     if (!mkdir($target)) {
1914         return $target if $! == EEXIST;
1915         die "failed to create directory $target: $!\n";
1916     }
1917
1918     PVE::Tools::mount("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1919         or die "failed to mount $target as tmpfs: $!\n";
1920
1921     return $target;
1922 }
1923
1924 sub mkfs {
1925     my ($dev, $rootuid, $rootgid) = @_;
1926
1927     run_command(
1928         [
1929             'mkfs.ext4',
1930             '-O',
1931             'mmp',
1932             '-E',
1933             "root_owner=$rootuid:$rootgid",
1934             $dev,
1935         ],
1936         outfunc => sub {
1937             my $line = shift;
1938             # a hack to print only the relevant stuff, i.e., the one which could help on repair
1939             if ($line =~ /^(Creating filesystem|Filesystem UUID|Superblock backups|\s+\d+, \d)/) {
1940                 print "$line\n";
1941             }
1942         },
1943         errfunc => sub {
1944             my $line = shift;
1945             print STDERR "$line\n" if $line && $line !~ /^mke2fs \d\.\d/;
1946         }
1947     );
1948 }
1949
1950 sub format_disk {
1951     my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1952
1953     if ($volid =~ m!^/dev/.+!) {
1954         mkfs($volid);
1955         return;
1956     }
1957
1958     my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
1959
1960     die "cannot format volume '$volid' with no storage\n" if !$storage;
1961
1962     PVE::Storage::activate_volumes($storage_cfg, [$volid]);
1963
1964     my $path = PVE::Storage::map_volume($storage_cfg, $volid);
1965
1966     $path = PVE::Storage::path($storage_cfg, $volid) if !defined($path);
1967
1968     my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1969         PVE::Storage::parse_volname($storage_cfg, $volid);
1970
1971     die "cannot format volume '$volid' (format == $format)\n"
1972         if $format ne 'raw';
1973
1974     mkfs($path, $rootuid, $rootgid);
1975 }
1976
1977 sub destroy_disks {
1978     my ($storecfg, $vollist) = @_;
1979
1980     foreach my $volid (@$vollist) {
1981         eval { PVE::Storage::vdisk_free($storecfg, $volid); };
1982         warn $@ if $@;
1983     }
1984 }
1985
1986 sub alloc_disk {
1987     my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
1988
1989     my $needs_chown = 0;
1990     my $volid;
1991
1992     my $scfg = PVE::Storage::storage_config($storecfg, $storage);
1993     # fixme: use better naming ct-$vmid-disk-X.raw?
1994
1995     eval {
1996         my $do_format = 0;
1997         if ($scfg->{content}->{rootdir} && $scfg->{path}) {
1998             if ($size_kb > 0 && !($scfg->{type} eq 'btrfs' && $scfg->{quotas})) {
1999                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
2000                 $do_format = 1;
2001             } else {
2002                 $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol', undef, $size_kb);
2003                 $needs_chown = 1;
2004             }
2005         } elsif ($scfg->{type} eq 'zfspool') {
2006             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'subvol', undef, $size_kb);
2007             $needs_chown = 1;
2008         } elsif ($scfg->{content}->{rootdir}) {
2009             $volid = PVE::Storage::vdisk_alloc($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
2010             $do_format = 1;
2011         } else {
2012             die "content type 'rootdir' is not available or configured on storage '$storage'\n";
2013         }
2014         format_disk($storecfg, $volid, $rootuid, $rootgid) if $do_format;
2015     };
2016     if (my $err = $@) {
2017         # in case formatting got interrupted:
2018         if (defined($volid)) {
2019             eval { PVE::Storage::vdisk_free($storecfg, $volid); };
2020             warn $@ if $@;
2021         }
2022         die $err;
2023     }
2024
2025     return ($volid, $needs_chown);
2026 }
2027
2028 sub create_disks {
2029     my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
2030
2031     my $vollist = [];
2032
2033     eval {
2034         my (undef, $rootuid, $rootgid) = PVE::LXC::parse_id_maps($conf);
2035         my $chown_vollist = [];
2036
2037         PVE::LXC::Config->foreach_volume($settings, sub {
2038             my ($ms, $mountpoint) = @_;
2039
2040             my $volid = $mountpoint->{volume};
2041             my $mp = $mountpoint->{mp};
2042
2043             my ($storage, $volname) = PVE::Storage::parse_volume_id($volid, 1);
2044
2045             if ($storage && ($volid =~ $NEW_DISK_RE)) {
2046                 my ($storeid, $size_gb) = ($1, $2);
2047
2048                 my $size_kb = int(${size_gb}*1024) * 1024;
2049
2050                 my $needs_chown = 0;
2051                 ($volid, $needs_chown) = alloc_disk($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
2052                 push @$chown_vollist, $volid if $needs_chown;
2053                 push @$vollist, $volid;
2054                 $mountpoint->{volume} = $volid;
2055                 $mountpoint->{size} = $size_kb * 1024;
2056                 if ($pending) {
2057                     $conf->{pending}->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2058                 } else {
2059                     $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2060                 }
2061             } else {
2062                 # use specified/existing volid/dir/device
2063                 $conf->{$ms} = PVE::LXC::Config->print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
2064             }
2065         });
2066
2067         PVE::Storage::activate_volumes($storecfg, $chown_vollist, undef);
2068         foreach my $volid (@$chown_vollist) {
2069             my $path = PVE::Storage::path($storecfg, $volid, undef);
2070             chown($rootuid, $rootgid, $path);
2071         }
2072         PVE::Storage::deactivate_volumes($storecfg, $chown_vollist, undef);
2073     };
2074     # free allocated images on error
2075     if (my $err = $@) {
2076         destroy_disks($storecfg, $vollist);
2077         die $err;
2078     }
2079     return $vollist;
2080 }
2081
2082 sub update_disksize {
2083     my ($vmid, $conf, $all_volumes) = @_;
2084
2085     my $changes;
2086     my $prefix = "CT $vmid:";
2087
2088     my $update_mp = sub {
2089         my ($key, $mp, @param) = @_;
2090         my $size = $all_volumes->{$mp->{volume}}->{size} // 0;
2091
2092         if (!defined($mp->{size}) || $size != $mp->{size}) {
2093             $changes = 1;
2094             print "$prefix updated volume size of '$mp->{volume}' in config.\n";
2095             $mp->{size} = $size;
2096             my $nomp = 1 if ($key eq 'rootfs');
2097             $conf->{$key} = PVE::LXC::Config->print_ct_mountpoint($mp, $nomp);
2098         }
2099     };
2100
2101     PVE::LXC::Config->foreach_volume($conf, $update_mp);
2102
2103     return $changes;
2104 }
2105
2106 sub update_unused {
2107     my ($vmid, $conf, $all_volumes) = @_;
2108
2109     my $changes;
2110     my $prefix = "CT $vmid:";
2111
2112     # Note: it is allowed to define multiple storage entries with the same path
2113     # (alias), so we need to check both 'volid' and real 'path' (two different
2114     # volid can point to the same path).
2115
2116     # used and unused disks
2117     my $refpath = {};
2118     my $orphans = {};
2119
2120     foreach my $opt (keys %$conf) {
2121         next if ($opt !~ m/^unused\d+$/);
2122         my $vol = $all_volumes->{$conf->{$opt}};
2123         $refpath->{$vol->{path}} = $vol->{volid};
2124     }
2125
2126     foreach my $key (keys %$all_volumes) {
2127         my $vol = $all_volumes->{$key};
2128         my $in_use = PVE::LXC::Config->is_volume_in_use($conf, $vol->{volid});
2129         my $path = $vol->{path};
2130
2131         if ($in_use) {
2132             $refpath->{$path} = $key;
2133             delete $orphans->{$path};
2134         } else {
2135             if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2136                 $orphans->{$path} = $key;
2137             }
2138         }
2139     }
2140
2141     for my $key (keys %$orphans) {
2142         my $disk = $orphans->{$key};
2143         my $unused = PVE::LXC::Config->add_unused_volume($conf, $disk);
2144
2145         if ($unused) {
2146             $changes = 1;
2147             print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2148         }
2149     }
2150
2151     return $changes;
2152 }
2153
2154 sub scan_volids {
2155     my ($cfg, $vmid) = @_;
2156
2157     my $info = PVE::Storage::vdisk_list($cfg, undef, $vmid, undef, 'rootdir');
2158
2159     my $all_volumes = {};
2160     foreach my $storeid (keys %$info) {
2161         foreach my $item (@{$info->{$storeid}}) {
2162             my $volid = $item->{volid};
2163             next if !($volid && $item->{size});
2164             $item->{path} = PVE::Storage::path($cfg, $volid);
2165             $all_volumes->{$volid} = $item;
2166         }
2167     }
2168
2169     return $all_volumes;
2170 }
2171
2172 sub rescan {
2173     my ($vmid, $nolock, $dryrun) = @_;
2174
2175     my $cfg = PVE::Storage::config();
2176
2177     print "rescan volumes...\n";
2178     my $all_volumes = scan_volids($cfg, $vmid);
2179
2180     my $updatefn =  sub {
2181         my ($vmid) = @_;
2182
2183         my $changes;
2184         my $conf = PVE::LXC::Config->load_config($vmid);
2185
2186         PVE::LXC::Config->check_lock($conf);
2187
2188         my $vm_volids = {};
2189         foreach my $volid (keys %$all_volumes) {
2190             my $info = $all_volumes->{$volid};
2191             $vm_volids->{$volid} = $info if $info->{vmid} == $vmid;
2192         }
2193
2194         my $upu = update_unused($vmid, $conf, $vm_volids);
2195         my $upd = update_disksize($vmid, $conf, $vm_volids);
2196         $changes = $upu || $upd;
2197
2198         PVE::LXC::Config->write_config($vmid, $conf) if $changes && !$dryrun;
2199     };
2200
2201     if (defined($vmid)) {
2202         if ($nolock) {
2203             &$updatefn($vmid);
2204         } else {
2205             PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2206         }
2207     } else {
2208         my $vmlist = config_list();
2209         foreach my $vmid (keys %$vmlist) {
2210             if ($nolock) {
2211                 &$updatefn($vmid);
2212             } else {
2213                 PVE::LXC::Config->lock_config($vmid, $updatefn, $vmid);
2214             }
2215         }
2216     }
2217 }
2218
2219
2220 # bash completion helper
2221
2222 sub complete_os_templates {
2223     my ($cmdname, $pname, $cvalue) = @_;
2224
2225     my $cfg = PVE::Storage::config();
2226
2227     my $storeid;
2228
2229     if ($cvalue =~ m/^([^:]+):/) {
2230         $storeid = $1;
2231     }
2232
2233     my $vtype = $cmdname eq 'restore' ? 'backup' : 'vztmpl';
2234     my $data = PVE::Storage::template_list($cfg, $storeid, $vtype);
2235
2236     my $res = [];
2237     foreach my $id (keys %$data) {
2238         foreach my $item (@{$data->{$id}}) {
2239             push @$res, $item->{volid} if defined($item->{volid});
2240         }
2241     }
2242
2243     return $res;
2244 }
2245
2246 my $complete_ctid_full = sub {
2247     my ($running) = @_;
2248
2249     my $idlist = vmstatus();
2250
2251     my $active_hash = list_active_containers();
2252
2253     my $res = [];
2254
2255     foreach my $id (keys %$idlist) {
2256         my $d = $idlist->{$id};
2257         if (defined($running)) {
2258             next if $d->{template};
2259             next if $running && !$active_hash->{$id};
2260             next if !$running && $active_hash->{$id};
2261         }
2262         push @$res, $id;
2263
2264     }
2265     return $res;
2266 };
2267
2268 sub complete_ctid {
2269     return &$complete_ctid_full();
2270 }
2271
2272 sub complete_ctid_stopped {
2273     return &$complete_ctid_full(0);
2274 }
2275
2276 sub complete_ctid_running {
2277     return &$complete_ctid_full(1);
2278 }
2279
2280 sub parse_id_maps {
2281     my ($conf) = @_;
2282
2283     my $id_map = [];
2284     my $rootuid = 0;
2285     my $rootgid = 0;
2286
2287     my $lxc = $conf->{lxc};
2288     foreach my $entry (@$lxc) {
2289         my ($key, $value) = @$entry;
2290         # FIXME: remove the 'id_map' variant when lxc-3.0 arrives
2291         next if $key ne 'lxc.idmap' && $key ne 'lxc.id_map';
2292         if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2293             my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2294             push @$id_map, [$type, $ct, $host, $length];
2295             if ($ct == 0) {
2296                 $rootuid = $host if $type eq 'u';
2297                 $rootgid = $host if $type eq 'g';
2298             }
2299         } else {
2300             die "failed to parse idmap: $value\n";
2301         }
2302     }
2303
2304     if (!@$id_map && $conf->{unprivileged}) {
2305         # Should we read them from /etc/subuid?
2306         $id_map = [ ['u', '0', '100000', '65536'],
2307                     ['g', '0', '100000', '65536'] ];
2308         $rootuid = $rootgid = 100000;
2309     }
2310
2311     return ($id_map, $rootuid, $rootgid);
2312 }
2313
2314 sub userns_command {
2315     my ($id_map) = @_;
2316     if (@$id_map) {
2317         return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2318     }
2319     return [];
2320 }
2321
2322 my sub print_ct_stderr_log {
2323     my ($vmid) = @_;
2324     my $log = eval { file_get_contents("/run/pve/ct-$vmid.stderr") };
2325     return if !$log;
2326
2327     while ($log =~ /^\h*(lxc-start:?\s+$vmid:?\s*\S+\s*)?(.*?)\h*$/gm) {
2328         my $line = $2;
2329         print STDERR "$line\n";
2330     }
2331 }
2332 my sub print_ct_warn_log {
2333     my ($vmid) = @_;
2334     my $log_fn = "/run/pve/ct-$vmid.warnings";
2335     my $log = eval { file_get_contents($log_fn) };
2336     return if !$log;
2337
2338     while ($log =~ /^\h*\s*(.*?)\h*$/gm) {
2339         PVE::RESTEnvironment::log_warn($1);
2340     }
2341     unlink $log_fn or warn "could not unlink '$log_fn' - $!\n";
2342 }
2343
2344 my sub monitor_state_change($$) {
2345     my ($monitor_socket, $vmid) = @_;
2346     die "no monitor socket\n" if !defined($monitor_socket);
2347
2348     while (1) {
2349         my ($type, $name, $value) = PVE::LXC::Monitor::read_lxc_message($monitor_socket);
2350
2351         die "monitor socket: got EOF\n" if !defined($type);
2352
2353         next if $name ne "$vmid" || $type ne 'STATE';
2354
2355         if ($value eq PVE::LXC::Monitor::STATE_STARTING) {
2356             alarm(0); # don't timeout after seeing the starting state
2357         } elsif ($value eq PVE::LXC::Monitor::STATE_ABORTING ||
2358                  $value eq PVE::LXC::Monitor::STATE_STOPPING ||
2359                  $value eq PVE::LXC::Monitor::STATE_STOPPED) {
2360             return 0;
2361         } elsif ($value eq PVE::LXC::Monitor::STATE_RUNNING) {
2362             return 1;
2363         } else {
2364             warn "unexpected message from monitor socket - " .
2365                  "type: '$type' - value: '$value'\n";
2366         }
2367     }
2368 }
2369 my sub monitor_start($$) {
2370     my ($monitor_socket, $vmid) = @_;
2371
2372     my $success = eval {
2373         PVE::Tools::run_with_timeout(10, \&monitor_state_change, $monitor_socket, $vmid)
2374     };
2375     if (my $err = $@) {
2376         warn "problem with monitor socket, but continuing anyway: $err\n";
2377     } elsif (!$success) {
2378         print_ct_stderr_log($vmid);
2379         die "startup for container '$vmid' failed\n";
2380     }
2381 }
2382
2383 sub vm_start {
2384     my ($vmid, $conf, $skiplock, $debug) = @_;
2385
2386     # apply pending changes while starting
2387     if (scalar(keys %{$conf->{pending}})) {
2388         my $storecfg = PVE::Storage::config();
2389         PVE::LXC::Config->vmconfig_apply_pending($vmid, $conf, $storecfg);
2390         PVE::LXC::Config->write_config($vmid, $conf);
2391         $conf = PVE::LXC::Config->load_config($vmid); # update/reload
2392     }
2393
2394     update_lxc_config($vmid, $conf);
2395
2396     my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2397
2398     if ($skiplock) {
2399         open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2400         close($fh);
2401     }
2402
2403     my $storage_cfg = PVE::Storage::config();
2404     my $vollist = PVE::LXC::Config->get_vm_volumes($conf);
2405
2406     PVE::Storage::activate_volumes($storage_cfg, $vollist);
2407
2408     my $monitor_socket = eval { PVE::LXC::Monitor::get_monitor_socket() };
2409     warn $@ if $@;
2410
2411     unlink "/run/pve/ct-$vmid.stderr"; # systemd does not truncate log files
2412
2413     my $is_debug = $debug || (!defined($debug) && $conf->{debug});
2414     my $base_unit = $is_debug ? 'pve-container-debug' : 'pve-container';
2415
2416     my $cmd = ['systemctl', 'start', "$base_unit\@$vmid"];
2417
2418     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-start', 1);
2419     eval {
2420         run_command($cmd);
2421
2422         monitor_start($monitor_socket, $vmid) if defined($monitor_socket);
2423
2424         # if debug is requested, print the log it also when the start succeeded
2425         print_ct_stderr_log($vmid) if $is_debug;
2426
2427         print_ct_warn_log($vmid); # always print warn log, if any
2428     };
2429     if (my $err = $@) {
2430         unlink $skiplock_flag_fn;
2431         die $err;
2432     }
2433     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'post-start');
2434
2435     return;
2436 }
2437
2438 # Helper to stop a container completely and make sure it has stopped completely.
2439 # This is necessary because we want the post-stop hook to have completed its
2440 # unmount-all step, but post-stop happens after lxc puts the container into the
2441 # STOPPED state.
2442 # $kill - if true it will always do an immediate hard-stop
2443 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2444 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2445 sub vm_stop {
2446     my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2447
2448     # Open the container's command socket.
2449     my $path = "\0/var/lib/lxc/$vmid/command";
2450     my $sock = IO::Socket::UNIX->new(
2451         Type => SOCK_STREAM(),
2452         Peer => $path,
2453     );
2454     if (!$sock) {
2455         return if $! == ECONNREFUSED; # The container is not running
2456         die "failed to open container ${vmid}'s command socket: $!\n";
2457     }
2458
2459     my $conf = PVE::LXC::Config->load_config($vmid);
2460     PVE::GuestHelpers::exec_hookscript($conf, $vmid, 'pre-stop');
2461
2462     # Stop the container:
2463
2464     my $cmd = ['lxc-stop', '-n', $vmid];
2465
2466     if ($kill) {
2467         push @$cmd, '--kill'; # doesn't allow timeouts
2468     } else {
2469         # lxc-stop uses a default timeout
2470         push @$cmd, '--nokill' if !$kill_after_timeout;
2471
2472         if (defined($shutdown_timeout)) {
2473             push @$cmd, '--timeout', $shutdown_timeout;
2474             # Give run_command 5 extra seconds
2475             $shutdown_timeout += 5;
2476         }
2477     }
2478
2479     eval { run_command($cmd, timeout => $shutdown_timeout) };
2480
2481     # Wait until the command socket is closed.
2482     # In case the lxc-stop call failed, reading from the command socket may block forever,
2483     # so poll with another timeout to avoid freezing the shutdown task.
2484     if (my $err = $@) {
2485         warn $err if $err;
2486
2487         my $poll = IO::Poll->new();
2488         $poll->mask($sock => POLLIN | POLLHUP); # watch for input and EOF events
2489         $poll->poll($shutdown_timeout); # IO::Poll timeout is in seconds
2490         return if ($poll->events($sock) & POLLHUP);
2491     } else {
2492         my $result = <$sock>;
2493         return if !defined $result; # monitor is gone and the ct has stopped.
2494     }
2495
2496     die "container did not stop\n";
2497 }
2498
2499 sub vm_reboot {
2500     my ($vmid, $timeout, $skiplock) = @_;
2501
2502     PVE::LXC::Config->lock_config($vmid, sub {
2503         return if !check_running($vmid);
2504
2505         vm_stop($vmid, 0, $timeout, 1); # kill if timeout exceeds
2506
2507         my $conf = PVE::LXC::Config->load_config($vmid);
2508         vm_start($vmid, $conf);
2509     });
2510 }
2511
2512 sub run_unshared {
2513     my ($code) = @_;
2514
2515     return PVE::Tools::run_fork(sub {
2516         # Unshare the mount namespace
2517         die "failed to unshare mount namespace: $!\n"
2518             if !PVE::Tools::unshare(PVE::Tools::CLONE_NEWNS);
2519         run_command(['mount', '--make-rslave', '/']);
2520         return $code->();
2521     });
2522 }
2523
2524 my $copy_volume = sub {
2525     my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid,  $rootgid) = @_;
2526
2527     my $src_mp = { volume => $src_volid, mp => '/', ro => 1 };
2528     $src_mp->{type} = PVE::LXC::Config->classify_mountpoint($src_volid);
2529
2530     my $dst_mp = { volume => $dst_volid, mp => '/', ro => 0 };
2531     $dst_mp->{type} = PVE::LXC::Config->classify_mountpoint($dst_volid);
2532
2533     my @mounted;
2534     eval {
2535         # mount and copy
2536         mkdir $src;
2537         mountpoint_mount($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2538         push @mounted, $src;
2539         mkdir $dest;
2540         mountpoint_mount($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2541         push @mounted, $dest;
2542
2543         $bwlimit //= 0;
2544
2545         run_command([
2546             'rsync',
2547             '--stats',
2548             '-X',
2549             '-A',
2550             '--numeric-ids',
2551             '-aH',
2552             '--whole-file',
2553             '--sparse',
2554             '--one-file-system',
2555             "--bwlimit=$bwlimit",
2556             "$src/",
2557             $dest
2558         ]);
2559     };
2560     my $err = $@;
2561
2562     # Wait for rsync's children to release dest so that
2563     # consequent file operations (umount, remove) are possible
2564     while ((system {"fuser"} "fuser",  "-s", $dest) == 0) {sleep 1};
2565
2566     foreach my $mount (reverse @mounted) {
2567         eval { run_command(['/bin/umount', $mount], errfunc => sub{})};
2568         warn "Can't umount $mount\n" if $@;
2569     }
2570
2571     # If this fails they're used as mount points in a concurrent operation
2572     # (which should not happen but there's also no real need to get rid of them).
2573     rmdir $dest;
2574     rmdir $src;
2575
2576     die $err if $err;
2577 };
2578
2579 # Should not be called after unsharing the mount namespace!
2580 sub copy_volume {
2581     my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2582
2583     die "cannot copy volumes of type $mp->{type}\n" if $mp->{type} ne 'volume';
2584     File::Path::make_path("/var/lib/lxc/$vmid");
2585     my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2586     my $src  = "/var/lib/lxc/$vmid/.copy-volume-2";
2587
2588     # get id's for unprivileged container
2589     my (undef, $rootuid, $rootgid) = parse_id_maps($conf);
2590
2591     # Allocate the disk before unsharing in order to make sure zfs subvolumes
2592     # are visible in this namespace, otherwise the host only sees the empty
2593     # (not-mounted) directory.
2594     my $new_volid;
2595     eval {
2596         # Make sure $mp contains a correct size.
2597         $mp->{size} = PVE::Storage::volume_size_info($storage_cfg, $mp->{volume});
2598         my $needs_chown;
2599         ($new_volid, $needs_chown) = alloc_disk($storage_cfg, $vmid, $storage, $mp->{size}/1024, $rootuid, $rootgid);
2600         if ($needs_chown) {
2601             PVE::Storage::activate_volumes($storage_cfg, [$new_volid], undef);
2602             my $path = PVE::Storage::path($storage_cfg, $new_volid, undef);
2603             chown($rootuid, $rootgid, $path);
2604         }
2605
2606         run_unshared(sub {
2607             $copy_volume->($mp->{volume}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2608         });
2609     };
2610     if (my $err = $@) {
2611         PVE::Storage::vdisk_free($storage_cfg, $new_volid)
2612             if defined($new_volid);
2613         die $err;
2614     }
2615
2616     return $new_volid;
2617 }
2618
2619 sub get_lxc_version() {
2620     my $version;
2621     run_command([qw(lxc-start --version)], outfunc => sub {
2622         my ($line) = @_;
2623         # We only parse out major & minor version numbers.
2624         if ($line =~ /^(\d+)\.(\d+)(?:\D.*)?$/) {
2625             $version = [$1, $2];
2626         }
2627     });
2628
2629     die "failed to get lxc version\n" if !defined($version);
2630
2631     # return as a list:
2632     return $version->@*;
2633 }
2634
2635 sub freeze($) {
2636     my ($vmid) = @_;
2637     if (PVE::CGroup::cgroup_mode() == 2) {
2638         PVE::LXC::Command::freeze($vmid, 30);
2639     } else {
2640         PVE::LXC::CGroup->new($vmid)->freeze_thaw(1);
2641     }
2642 }
2643
2644 sub thaw($) {
2645     my ($vmid) = @_;
2646     if (PVE::CGroup::cgroup_mode() == 2) {
2647         PVE::LXC::Command::unfreeze($vmid, 30);
2648     } else {
2649         PVE::LXC::CGroup->new($vmid)->freeze_thaw(0);
2650     }
2651 }
2652
2653 1;