13 use Fcntl
qw(O_RDONLY O_WRONLY O_NOFOLLOW O_DIRECTORY);
14 use Errno
qw(ELOOP ENOTDIR EROFS ECONNREFUSED ENOSYS EEXIST);
17 use PVE
::Exception
qw(raise_perm_exc);
21 use PVE
::JSONSchema
qw(get_standard_option);
34 use PVE
::AccessControl
;
36 use PVE
::Syscall
qw(:fsmount);
38 use PVE
::GuestHelpers
qw(safe_string_ne safe_num_ne safe_boolean_ne);
41 use PVE
::LXC
::Monitor
;
44 use Time
::HiRes qw
(gettimeofday
);
47 require PVE
::Network
::SDN
::Zones
;
51 my $LXC_CONFIG_PATH = '/usr/share/lxc/config';
53 my $nodename = PVE
::INotify
::nodename
();
55 my $cpuinfo= PVE
::ProcFSTools
::read_cpuinfo
();
58 my $vmlist = PVE
::Cluster
::get_vmlist
();
60 return $res if !$vmlist || !$vmlist->{ids
};
61 my $ids = $vmlist->{ids
};
63 foreach my $vmid (keys %$ids) {
64 next if !$vmid; # skip CT0
65 my $d = $ids->{$vmid};
66 next if !$d->{node
} || $d->{node
} ne $nodename;
67 next if !$d->{type
} || $d->{type
} ne 'lxc';
68 $res->{$vmid} = { type
=> 'lxc', vmid
=> $vmid };
73 # container status helpers
75 sub list_active_containers
{
77 my $filename = "/proc/net/unix";
79 # similar test is used by lcxcontainers.c: list_active_containers
82 my $fh = IO
::File-
>new ($filename, "r");
85 while (defined(my $line = <$fh>)) {
86 if ($line =~ m/^[a-f0-9]+:\s+\S+\s+\S+\s+\S+\s+\S+\s+\S+\s+\d+\s+(\S+)$/) {
88 if ($path =~ m!^@/var/lib/lxc/(\d+)/command$!) {
99 # warning: this is slow
103 my $active_hash = list_active_containers
();
105 return 1 if defined($active_hash->{$vmid});
110 sub get_container_disk_usage
{
111 my ($vmid, $pid) = @_;
113 return PVE
::Tools
::df
("/proc/$pid/root/", 1);
116 my $last_proc_vmid_stat;
118 our $vmstatus_return_properties = {
119 vmid
=> get_standard_option
('pve-vmid'),
121 description
=> "LXC Container status.",
123 enum
=> ['stopped', 'running'],
126 description
=> "Maximum memory in bytes.",
132 description
=> "Maximum SWAP memory in bytes.",
138 description
=> "Root disk size in bytes.",
144 description
=> "Container name.",
149 description
=> "Uptime.",
152 renderer
=> 'duration',
155 description
=> "Maximum usable CPUs.",
160 description
=> "The current config lock, if any.",
165 description
=> "The current configured tags, if any.",
174 my $list = $opt_vmid ?
{ $opt_vmid => { type
=> 'lxc', vmid
=> $opt_vmid }} : config_list
();
176 my $active_hash = list_active_containers
();
178 my $cpucount = $cpuinfo->{cpus
} || 1;
180 my $cdtime = gettimeofday
;
182 my $uptime = (PVE
::ProcFSTools
::read_proc_uptime
(1))[0];
183 my $clock_ticks = POSIX
::sysconf
(&POSIX
::_SC_CLK_TCK
);
185 my $unprivileged = {};
187 foreach my $vmid (keys %$list) {
188 my $d = $list->{$vmid};
190 eval { $d->{pid
} = find_lxc_pid
($vmid) if defined($active_hash->{$vmid}); };
191 warn $@ if $@; # ignore errors (consider them stopped)
193 $d->{status
} = $active_hash->{$vmid} ?
'running' : 'stopped';
195 my $cfspath = PVE
::LXC
::Config-
>cfs_config_path($vmid);
196 my $conf = PVE
::Cluster
::cfs_read_file
($cfspath) || {};
198 $unprivileged->{$vmid} = $conf->{unprivileged
};
200 $d->{name
} = $conf->{'hostname'} || "CT$vmid";
201 $d->{name
} =~ s/[\s]//g;
203 $d->{cpus
} = $conf->{cores
} || $conf->{cpulimit
};
204 $d->{cpus
} = $cpucount if !$d->{cpus
};
206 $d->{lock} = $conf->{lock} || '';
207 $d->{tags
} = $conf->{tags
} if defined($conf->{tags
});
210 my $res = get_container_disk_usage
($vmid, $d->{pid
});
211 $d->{disk
} = $res->{used
};
212 $d->{maxdisk
} = $res->{total
};
215 # use 4GB by default ??
216 if (my $rootfs = $conf->{rootfs
}) {
217 my $rootinfo = PVE
::LXC
::Config-
>parse_volume('rootfs', $rootfs);
218 $d->{maxdisk
} = $rootinfo->{size
} || (4*1024*1024*1024);
220 $d->{maxdisk
} = 4*1024*1024*1024;
226 $d->{maxmem
} = ($conf->{memory
}||512)*1024*1024;
227 $d->{maxswap
} = ($conf->{swap
}//0)*1024*1024;
238 $d->{template
} = PVE
::LXC
::Config-
>is_template($conf);
239 $d->{lock} = $conf->{lock} if $conf->{lock};
242 foreach my $vmid (keys %$list) {
243 my $d = $list->{$vmid};
246 next if !$pid; # skip stopped CTs
248 my $proc_pid_stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
249 $d->{uptime
} = int(($uptime - $proc_pid_stat->{starttime
}) / $clock_ticks); # the method lxcfs uses
251 my $unpriv = $unprivileged->{$vmid};
253 my $cgroups = PVE
::LXC
::CGroup-
>new($vmid);
255 if (defined(my $mem = $cgroups->get_memory_stat())) {
256 $d->{mem
} = $mem->{mem
};
257 $d->{swap
} = $mem->{swap
};
263 if (defined(my $blkio = $cgroups->get_io_stats())) {
264 $d->{diskread
} = $blkio->{diskread
};
265 $d->{diskwrite
} = $blkio->{diskwrite
};
271 if (defined(my $cpu = $cgroups->get_cpu_stat())) {
272 # Total time (in milliseconds) used up by the cpu.
273 my $used_ms = $cpu->{utime} + $cpu->{stime
};
275 my $old = $last_proc_vmid_stat->{$vmid};
277 $last_proc_vmid_stat->{$vmid} = {
285 my $delta_ms = ($cdtime - $old->{time}) * $cpucount * 1000.0;
286 if ($delta_ms > 1000.0) {
287 my $delta_used_ms = $used_ms - $old->{used
};
288 $d->{cpu
} = (($delta_used_ms / $delta_ms) * $cpucount) / $d->{cpus
};
289 $last_proc_vmid_stat->{$vmid} = {
295 $d->{cpu
} = $old->{cpu
};
302 my $netdev = PVE
::ProcFSTools
::read_proc_net_dev
();
304 foreach my $dev (keys %$netdev) {
305 next if $dev !~ m/^veth([1-9]\d*)i/;
307 my $d = $list->{$vmid};
311 $d->{netout
} += $netdev->{$dev}->{receive
};
312 $d->{netin
} += $netdev->{$dev}->{transmit
};
319 sub find_lxc_console_pids
{
323 PVE
::Tools
::dir_glob_foreach
('/proc', '\d+', sub {
326 my $cmdline = PVE
::Tools
::file_read_firstline
("/proc/$pid/cmdline");
329 my @args = split(/\0/, $cmdline);
331 # search for lxc-console -n <vmid>
332 return if scalar(@args) != 3;
333 return if $args[1] ne '-n';
334 return if $args[2] !~ m/^\d+$/;
335 return if $args[0] !~ m
|^(/usr/bin
/)?lxc-console
$|;
339 push @{$res->{$vmid}}, $pid;
351 $pid = $1 if $line =~ m/^PID:\s+(\d+)$/;
353 PVE
::Tools
::run_command
(['lxc-info', '-n', $vmid, '-p'], outfunc
=> $parser);
355 die "unable to get PID for CT $vmid (not running?)\n" if !$pid;
362 sysopen(my $fd, "/proc/$pid", O_RDONLY
| O_DIRECTORY
)
363 or die "failed to open /proc/$pid pid fd\n";
370 # Find the pid and open:
371 my $pid = find_lxc_pid
($vmid);
372 my $fd = open_pid_fd
($pid);
375 my $pid2 = find_lxc_pid
($vmid);
377 return () if $pid != $pid2;
384 # Find the parent pid via proc and open it:
385 my $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
386 my $ppid = $stat->{ppid
} // die "failed to get parent pid\n";
388 my $fd = open_pid_fd
($ppid);
391 $stat = PVE
::ProcFSTools
::read_proc_pid_stat
($pid);
392 my $ppid2 = $stat->{ppid
} // die "failed to get parent pid for verification\n";
394 return () if $ppid != $ppid2;
398 # Note: we cannot use Net:IP, because that only allows strict
400 sub parse_ipv4_cidr
{
401 my ($cidr, $noerr) = @_;
403 if ($cidr =~ m!^($IPV4RE)(?:/(\d+))$! && ($2 > 7) && ($2 <= 32)) {
404 return { address
=> $1, netmask
=> $PVE::Network
::ipv4_reverse_mask-
>[$2] };
407 return undef if $noerr;
409 die "unable to parse ipv4 address/mask\n";
412 # Deprecated. Use `PVE::CGroup::get_cgroup_controllers()` instead.
413 sub get_cgroup_subsystems
{
414 PVE
::CGroup
::get_v1_controllers
();
417 # With seccomp trap to userspace we now have the ability to optionally forward
418 # certain syscalls to the "host" to handle (via our pve-lxc-syscalld daemon).
420 # This means that there are cases where we need to create an extra seccomp
421 # profile for the container to load.
423 # This returns a configuration snippet added to the raw lxc config.
424 sub make_seccomp_config
{
425 my ($conf, $vmid, $conf_dir, $unprivileged, $features) = @_;
426 # User-configured profile has precedence, note that the user's entry would
427 # be written 'after' this line anyway...
428 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.seccomp.profile')) {
429 # Warn the user if this conflicts with a feature:
430 my $warn = join(', ', grep { $features->{$_} } qw(keyctl mknod));
431 warn "explicitly configured lxc.seccomp.profile overrides the following settings: $warn\n"
432 if length($warn) > 0;
436 # Privileged containers keep using the default (which is already part of
437 # the files included via lxc.include, so we don't need to write it out,
438 # that way it stays admin-configurable via /usr/share/lxc/config/... as
440 return '' if !$unprivileged;
443 keyctl
=> ['errno 38'],
448 # Unprivileged containers will get keyctl() disabled by default as a
449 # workaround for systemd-networkd behavior. But we have an option to
450 # explicitly enable it:
451 if ($features->{keyctl
}) {
452 delete $rules->{keyctl
};
455 # By default, unprivileged containers cannot use `mknod` at all.
456 # Since lxc 3.2, we can use seccomp's trap to userspace feature for this,
457 # but for now this is experimental, so it has to be enabled via a feature
459 # Note that we only handle block and char devices (like lxd), the rest we
460 # leave up to the kernel. We may in the future remove this if seccomp gets
461 # a way to tell the kernel to "continue" a syscall.
462 if ($features->{mknod
}) {
463 my ($ok, $kernel) = PVE
::ProcFSTools
::check_kernel_release
(5, 3);
465 die "'mknod' feature requested, but kernel too old (found $kernel, required >= 5.3)\n";
468 $raw_conf .= "lxc.seccomp.notify.proxy = unix:/run/pve/lxc-syscalld.sock\n";
469 $raw_conf .= "lxc.seccomp.notify.cookie = $vmid\n";
472 # condition: (mode & S_IFMT) == S_IFCHR
473 'notify [1,8192,SCMP_CMP_MASKED_EQ,61440]',
474 # condition: (mode & S_IFMT) == S_IFBLK
475 'notify [1,24576,SCMP_CMP_MASKED_EQ,61440]',
477 $rules->{mknodat
} = [
478 # condition: (mode & S_IFMT) == S_IFCHR
479 'notify [2,8192,SCMP_CMP_MASKED_EQ,61440]',
480 # condition: (mode & S_IFMT) == S_IFBLK
481 'notify [2,24576,SCMP_CMP_MASKED_EQ,61440]',
485 # Now build the custom seccomp rule text...
486 my $extra_rules = join("\n", map {
488 map { "$syscall $_" } $rules->{$syscall}->@*
489 } sort keys %$rules) . "\n";
491 return $raw_conf if $extra_rules eq "\n";
493 # We still have the "most common" config readily available, so don't write
495 if ($raw_conf eq '' && $extra_rules eq "keyctl errno 38\n") {
496 # we have no extra $raw_conf and use the same we had in pve 6.1:
497 return "lxc.seccomp.profile = $LXC_CONFIG_PATH/pve-userns.seccomp\n";
500 # Write the rule file to the container's config path:
501 my $rule_file = "$conf_dir/rules.seccomp";
502 my $rule_data = file_get_contents
("$LXC_CONFIG_PATH/common.seccomp")
504 file_set_contents
($rule_file, $rule_data);
505 $raw_conf .= "lxc.seccomp.profile = $rule_file\n";
510 # Since lxc-3.0.2 we can have lxc generate a profile for the container
511 # automatically. The default should be equivalent to the old
512 # `lxc-container-default-cgns` profile.
514 # Additionally this also added `lxc.apparmor.raw` which can be used to inject
515 # additional lines into the profile. We can use that to allow mounting specific
517 sub make_apparmor_config
{
518 my ($conf, $unprivileged, $features) = @_;
520 # user-configured profile has precedence, but first we go through our own
521 # code to figure out whether we should warn the user:
523 my $raw = "lxc.apparmor.profile = generated\n";
526 if ($features->{fuse
}) {
527 # For the informational warning:
528 push @profile_uses, 'features:fuse';
531 # There's lxc.apparmor.allow_nesting now, which will add the necessary
532 # apparmor lines, create an apparmor namespace for the container, but also
533 # adds proc and sysfs mounts to /dev/.lxc/{proc,sys}. These do not have
534 # lxcfs mounted over them, because that would prevent the container from
535 # mounting new instances of them for nested containers.
536 if ($features->{nesting
}) {
537 push @profile_uses, 'features:nesting';
538 $raw .= "lxc.apparmor.allow_nesting = 1\n"
540 # In the default profile in /etc/apparmor.d we patch this in because
541 # otherwise a container can for example run `chown` on /sys, breaking
542 # access to it for non-CAP_DAC_OVERRIDE tools on the host:
543 $raw .= "lxc.apparmor.raw = deny mount -> /proc/,\n";
544 $raw .= "lxc.apparmor.raw = deny mount -> /sys/,\n";
545 # Preferably we could use the 'remount' flag but this does not sit well
546 # with apparmor_parser currently:
547 # mount options=(rw, nosuid, nodev, noexec, remount) -> /sys/,
550 if (my $mount = $features->{mount
}) {
551 push @profile_uses, 'features:mount';
552 foreach my $fs (PVE
::Tools
::split_list
($mount)) {
553 $raw .= "lxc.apparmor.raw = mount fstype=$fs,\n";
559 if (PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.apparmor.profile')) {
560 if (length(my $used = join(', ', @profile_uses))) {
561 warn "explicitly configured lxc.apparmor.profile overrides the following settings: $used\n";
569 sub update_lxc_config
{
570 my ($vmid, $conf) = @_;
572 my $dir = "/var/lib/lxc/$vmid";
574 if ($conf->{template
}) {
576 unlink "$dir/config";
581 my ($lxc_major, $lxc_minor) = get_lxc_version
();
585 if ($lxc_major >= 4) {
586 # Explicitly don't use relative directories, which is the default, but
587 # note that we do this mostly because they are only applied for *some*
588 # cgroups. Our pve-container@.service now starts lxc-start with `-F`,
589 # so we also don't need to worry about the new monitor cgroup to
591 $raw .= "lxc.cgroup.relative = 0\n";
593 # To make things easier, let's keep our previous cgroup layout and
594 # simply move the monitor outside:
595 $raw .= "lxc.cgroup.dir.monitor = lxc.monitor/$vmid\n";
596 # cgroup namespace separation for stronger limits:
597 $raw .= "lxc.cgroup.dir.container = lxc/$vmid\n";
598 $raw .= "lxc.cgroup.dir.container.inner = ns\n";
601 die "missing 'arch' - internal error" if !$conf->{arch
};
602 $raw .= "lxc.arch = $conf->{arch}\n";
604 my $custom_idmap = PVE
::LXC
::Config-
>has_lxc_entry($conf, 'lxc.idmap');
605 my $unprivileged = $conf->{unprivileged
} || $custom_idmap;
607 my $ostype = $conf->{ostype
} || die "missing 'ostype' - internal error";
609 my $cfgpath = '/usr/share/lxc/config';
610 my $inc = "$cfgpath/$ostype.common.conf";
611 $inc ="$cfgpath/common.conf" if !-f
$inc;
612 $raw .= "lxc.include = $inc\n";
614 $inc = "$cfgpath/$ostype.userns.conf";
615 $inc = "$cfgpath/userns.conf" if !-f
$inc;
616 $raw .= "lxc.include = $inc\n";
619 my $features = PVE
::LXC
::Config-
>parse_features($conf->{features
});
621 $raw .= make_seccomp_config
($conf, $vmid, $dir, $unprivileged, $features);
622 $raw .= make_apparmor_config
($conf, $unprivileged, $features);
623 if ($features->{fuse
}) {
624 $raw .= "lxc.apparmor.raw = mount fstype=fuse,\n";
625 $raw .= "lxc.mount.entry = /dev/fuse dev/fuse none bind,create=file 0 0\n";
628 if ($unprivileged && !$features->{force_rw_sys
}) {
629 # unpriv. CT default to sys:rw, but that doesn't always plays well with
630 # systemd, e.g., systemd-networkd https://systemd.io/CONTAINER_INTERFACE/
631 $raw .= "lxc.mount.auto = sys:mixed\n";
634 # WARNING: DO NOT REMOVE this without making sure that loop device nodes
635 # cannot be exposed to the container with r/w access (cgroup perms).
636 # When this is enabled mounts will still remain in the monitor's namespace
637 # after the container unmounted them and thus will not detach from their
638 # files while the container is running!
639 $raw .= "lxc.monitor.unshare = 1\n";
641 my $cgv1 = get_cgroup_subsystems
();
643 # Should we read them from /etc/subuid?
644 if ($unprivileged && !$custom_idmap) {
645 $raw .= "lxc.idmap = u 0 100000 65536\n";
646 $raw .= "lxc.idmap = g 0 100000 65536\n";
649 if (!PVE
::LXC
::Config-
>has_dev_console($conf)) {
650 $raw .= "lxc.console.path = none\n";
651 $raw .= "lxc.cgroup.devices.deny = c 5:1 rwm\n" if $cgv1->{devices
};
654 my $ttycount = PVE
::LXC
::Config-
>get_tty_count($conf);
655 $raw .= "lxc.tty.max = $ttycount\n";
657 # some init scripts expect a linux terminal (turnkey).
658 $raw .= "lxc.environment = TERM=linux\n";
660 my $utsname = $conf->{hostname
} || "CT$vmid";
661 $raw .= "lxc.uts.name = $utsname\n";
663 if ($cgv1->{memory
}) {
664 my $memory = $conf->{memory
} || 512;
665 my $swap = $conf->{swap
} // 0;
667 my $lxcmem = int($memory*1024*1024);
668 $raw .= "lxc.cgroup.memory.limit_in_bytes = $lxcmem\n";
670 my $lxcswap = int(($memory + $swap)*1024*1024);
671 $raw .= "lxc.cgroup.memory.memsw.limit_in_bytes = $lxcswap\n";
675 if (my $cpulimit = $conf->{cpulimit
}) {
676 $raw .= "lxc.cgroup.cpu.cfs_period_us = 100000\n";
677 my $value = int(100000*$cpulimit);
678 $raw .= "lxc.cgroup.cpu.cfs_quota_us = $value\n";
681 my $shares = $conf->{cpuunits
} || 1024;
682 $raw .= "lxc.cgroup.cpu.shares = $shares\n";
685 die "missing 'rootfs' configuration\n"
686 if !defined($conf->{rootfs
});
688 my $mountpoint = PVE
::LXC
::Config-
>parse_volume('rootfs', $conf->{rootfs
});
690 $raw .= "lxc.rootfs.path = $dir/rootfs\n";
692 foreach my $k (sort keys %$conf) {
693 next if $k !~ m/^net(\d+)$/;
695 my $d = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$k});
696 $raw .= "lxc.net.$ind.type = veth\n";
697 $raw .= "lxc.net.$ind.veth.pair = veth${vmid}i${ind}\n";
698 $raw .= "lxc.net.$ind.hwaddr = $d->{hwaddr}\n" if defined($d->{hwaddr
});
699 $raw .= "lxc.net.$ind.name = $d->{name}\n" if defined($d->{name
});
700 $raw .= "lxc.net.$ind.mtu = $d->{mtu}\n" if defined($d->{mtu
});
702 # Starting with lxc 4.0, we do not patch lxc to execute our up-scripts.
703 if ($lxc_major >= 4) {
704 $raw .= "lxc.net.$ind.script.up = /usr/share/lxc/lxcnetaddbr\n";
709 if (my $lxcconf = $conf->{lxc
}) {
710 foreach my $entry (@$lxcconf) {
711 my ($k, $v) = @$entry;
712 $had_cpuset = 1 if $k eq 'lxc.cgroup.cpuset.cpus' || $k eq 'lxc.cgroup2.cpuset.cpus';
718 my ($cpuset_cgroup, $cpuset_version) = eval { PVE
::CGroup
::cpuset_controller_path
() };
719 if (defined($cpuset_cgroup)) {
720 $cpuset = eval { PVE
::CpuSet-
>new_from_path("$cpuset_cgroup/lxc", 1) }
721 || PVE
::CpuSet-
>new_from_path($cpuset_cgroup, 1);
723 my $cores = $conf->{cores
};
724 if (!$had_cpuset && $cores && $cpuset) {
725 my @members = $cpuset->members();
726 while (scalar(@members) > $cores) {
727 my $randidx = int(rand(scalar(@members)));
728 $cpuset->delete($members[$randidx]);
729 splice(@members, $randidx, 1); # keep track of the changes
731 my $ver = $cpuset_version == 1 ?
'' : '2';
732 $raw .= "lxc.cgroup$ver.cpuset.cpus = ".$cpuset->short_string()."\n";
735 File
::Path
::mkpath
("$dir/rootfs");
737 PVE
::Tools
::file_set_contents
("$dir/config", $raw);
740 # verify and cleanup nameserver list (replace \0 with ' ')
741 sub verify_nameserver_list
{
742 my ($nameserver_list) = @_;
745 foreach my $server (PVE
::Tools
::split_list
($nameserver_list)) {
746 PVE
::LXC
::Config
::verify_ip_with_ll_iface
($server);
750 return join(' ', @list);
753 sub verify_searchdomain_list
{
754 my ($searchdomain_list) = @_;
757 foreach my $server (PVE
::Tools
::split_list
($searchdomain_list)) {
758 # todo: should we add checks for valid dns domains?
762 return join(' ', @list);
765 sub get_console_command
{
766 my ($vmid, $conf, $escapechar) = @_;
768 # '-1' as $escapechar disables keyboard escape sequence
769 # any other passed char (a-z) will result in <Ctrl+$escapechar q>
771 my $cmode = PVE
::LXC
::Config-
>get_cmode($conf);
774 if ($cmode eq 'console') {
775 push @$cmd, 'lxc-console', '-n', $vmid, '-t', 0;
776 push @$cmd, '-e', $escapechar if $escapechar;
777 } elsif ($cmode eq 'tty') {
778 push @$cmd, 'lxc-console', '-n', $vmid;
779 push @$cmd, '-e', $escapechar if $escapechar;
780 } elsif ($cmode eq 'shell') {
781 push @$cmd, 'lxc-attach', '--clear-env', '-n', $vmid;
783 die "internal error";
789 sub get_primary_ips
{
792 # return data from net0
794 return undef if !defined($conf->{net0
});
795 my $net = PVE
::LXC
::Config-
>parse_lxc_network($conf->{net0
});
797 my $ipv4 = $net->{ip
};
799 if ($ipv4 =~ /^(dhcp|manual)$/) {
805 my $ipv6 = $net->{ip6
};
807 if ($ipv6 =~ /^(auto|dhcp|manual)$/) {
814 return ($ipv4, $ipv6);
817 sub delete_mountpoint_volume
{
818 my ($storage_cfg, $vmid, $volume) = @_;
820 return if PVE
::LXC
::Config-
>classify_mountpoint($volume) ne 'volume';
822 my ($vtype, $name, $owner) = PVE
::Storage
::parse_volname
($storage_cfg, $volume);
824 if ($vmid == $owner) {
825 PVE
::Storage
::vdisk_free
($storage_cfg, $volume);
827 warn "ignore deletion of '$volume', CT $vmid isn't the owner!\n";
831 sub destroy_lxc_container
{
832 my ($storage_cfg, $vmid, $conf, $replacement_conf) = @_;
834 PVE
::LXC
::Config-
>foreach_volume_full($conf, {include_unused
=> 1}, sub {
835 my ($ms, $mountpoint) = @_;
836 delete_mountpoint_volume
($storage_cfg, $vmid, $mountpoint->{volume
});
839 rmdir "/var/lib/lxc/$vmid/rootfs";
840 unlink "/var/lib/lxc/$vmid/config";
841 rmdir "/var/lib/lxc/$vmid";
842 if (defined $replacement_conf) {
843 PVE
::LXC
::Config-
>write_config($vmid, $replacement_conf);
845 PVE
::LXC
::Config-
>destroy_config($vmid);
849 sub vm_stop_cleanup
{
850 my ($storage_cfg, $vmid, $conf, $keepActive) = @_;
852 return if $keepActive;
855 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
856 PVE
::Storage
::deactivate_volumes
($storage_cfg, $vollist);
858 warn $@ if $@; # avoid errors - just warn
862 my ($vmid, $conf, $opt, $newnet, $netid, $rootdir) = @_;
864 if ($newnet->{type
} ne 'veth') {
865 # for when there are physical interfaces
866 die "cannot update interface of type $newnet->{type}";
869 my $veth = "veth${vmid}i${netid}";
870 my $eth = $newnet->{name
};
872 if (my $oldnetcfg = $conf->{$opt}) {
873 my $oldnet = PVE
::LXC
::Config-
>parse_lxc_network($oldnetcfg);
875 if (safe_string_ne
($oldnet->{hwaddr
}, $newnet->{hwaddr
}) ||
876 safe_string_ne
($oldnet->{name
}, $newnet->{name
})) {
878 PVE
::Network
::veth_delete
($veth);
879 delete $conf->{$opt};
880 PVE
::LXC
::Config-
>write_config($vmid, $conf);
882 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
885 if (safe_string_ne
($oldnet->{bridge
}, $newnet->{bridge
}) ||
886 safe_num_ne
($oldnet->{tag
}, $newnet->{tag
}) ||
887 safe_num_ne
($oldnet->{firewall
}, $newnet->{firewall
})) {
889 if ($oldnet->{bridge
}) {
890 PVE
::Network
::tap_unplug
($veth);
891 foreach (qw(bridge tag firewall)) {
892 delete $oldnet->{$_};
894 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
895 PVE
::LXC
::Config-
>write_config($vmid, $conf);
899 PVE
::Network
::SDN
::Zones
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
901 PVE
::Network
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
904 # This includes the rate:
905 foreach (qw(bridge tag firewall rate)) {
906 $oldnet->{$_} = $newnet->{$_} if $newnet->{$_};
908 } elsif (safe_string_ne
($oldnet->{rate
}, $newnet->{rate
})) {
909 # Rate can be applied on its own but any change above needs to
910 # include the rate in tap_plug since OVS resets everything.
911 PVE
::Network
::tap_rate_limit
($veth, $newnet->{rate
});
912 $oldnet->{rate
} = $newnet->{rate
}
914 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($oldnet);
915 PVE
::LXC
::Config-
>write_config($vmid, $conf);
918 hotplug_net
($vmid, $conf, $opt, $newnet, $netid);
921 update_ipconfig
($vmid, $conf, $opt, $eth, $newnet, $rootdir);
925 my ($vmid, $conf, $opt, $newnet, $netid) = @_;
927 my $veth = "veth${vmid}i${netid}";
928 my $vethpeer = $veth . "p";
929 my $eth = $newnet->{name
};
932 PVE
::Network
::SDN
::Zones
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
933 PVE
::Network
::SDN
::Zones
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
935 PVE
::Network
::veth_create
($veth, $vethpeer, $newnet->{bridge
}, $newnet->{hwaddr
});
936 PVE
::Network
::tap_plug
($veth, $newnet->{bridge
}, $newnet->{tag
}, $newnet->{firewall
}, $newnet->{trunks
}, $newnet->{rate
});
939 # attach peer in container
940 my $cmd = ['lxc-device', '-n', $vmid, 'add', $vethpeer, "$eth" ];
941 PVE
::Tools
::run_command
($cmd);
943 # link up peer in container
944 $cmd = ['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', '/sbin/ip', 'link', 'set', $eth ,'up' ];
945 PVE
::Tools
::run_command
($cmd);
947 my $done = { type
=> 'veth' };
948 foreach (qw(bridge tag firewall hwaddr name)) {
949 $done->{$_} = $newnet->{$_} if $newnet->{$_};
951 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($done);
953 PVE
::LXC
::Config-
>write_config($vmid, $conf);
956 sub update_ipconfig
{
957 my ($vmid, $conf, $opt, $eth, $newnet, $rootdir) = @_;
959 my $lxc_setup = PVE
::LXC
::Setup-
>new($conf, $rootdir);
961 my $optdata = PVE
::LXC
::Config-
>parse_lxc_network($conf->{$opt});
966 PVE
::Tools
::run_command
(['lxc-attach', '-n', $vmid, '-s', 'NETWORK', '--', @_], %$cmdargs);
968 my $ipcmd = sub { &$nscmd({}, '/sbin/ip', @_) };
970 my $change_ip_config = sub {
971 my ($ipversion) = @_;
973 my $family_opt = "-$ipversion";
974 my $suffix = $ipversion == 4 ?
'' : $ipversion;
978 my $newip = $newnet->{$ip};
979 my $newgw = $newnet->{$gw};
980 my $oldip = $optdata->{$ip};
981 my $oldgw = $optdata->{$gw};
983 my $change_ip = safe_string_ne
($oldip, $newip);
984 my $change_gw = safe_string_ne
($oldgw, $newgw);
986 return if !$change_ip && !$change_gw;
988 # step 1: add new IP, if this fails we cancel
989 my $is_real_ip = ($newip && $newip !~ /^(?:auto|dhcp|manual)$/);
990 if ($change_ip && $is_real_ip) {
991 eval { &$ipcmd($family_opt, 'addr', 'add', $newip, 'dev', $eth); };
998 # step 2: replace gateway
999 # If this fails we delete the added IP and cancel.
1000 # If it succeeds we save the config and delete the old IP, ignoring
1001 # errors. The config is then saved.
1002 # Note: 'ip route replace' can add
1006 if ($is_real_ip && !PVE
::Network
::is_ip_in_cidr
($newgw, $newip, $ipversion)) {
1007 &$ipcmd($family_opt, 'route', 'add', $newgw, 'dev', $eth);
1009 &$ipcmd($family_opt, 'route', 'replace', 'default', 'via', $newgw);
1013 # the route was not replaced, the old IP is still available
1014 # rollback (delete new IP) and cancel
1016 eval { &$ipcmd($family_opt, 'addr', 'del', $newip, 'dev', $eth); };
1017 warn $@ if $@; # no need to die here
1022 eval { &$ipcmd($family_opt, 'route', 'del', 'default'); };
1023 # if the route was not deleted, the guest might have deleted it manually
1027 if ($oldgw && $oldip && !PVE
::Network
::is_ip_in_cidr
($oldgw, $oldip)) {
1028 eval { &$ipcmd($family_opt, 'route', 'del', $oldgw, 'dev', $eth); };
1029 # warn if the route was deleted manually
1034 # from this point on we save the configuration
1035 # step 3: delete old IP ignoring errors
1036 if ($change_ip && $oldip && $oldip !~ /^(?:auto|dhcp)$/) {
1037 # We need to enable promote_secondaries, otherwise our newly added
1038 # address will be removed along with the old one.
1041 if ($ipversion == 4) {
1042 &$nscmd({ outfunc
=> sub { $promote = int(shift) } },
1043 'cat', "/proc/sys/net/ipv4/conf/$eth/promote_secondaries");
1044 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=1");
1046 &$ipcmd($family_opt, 'addr', 'del', $oldip, 'dev', $eth);
1048 warn $@ if $@; # no need to die here
1050 if ($ipversion == 4) {
1051 &$nscmd({}, 'sysctl', "net.ipv4.conf.$eth.promote_secondaries=$promote");
1055 foreach my $property ($ip, $gw) {
1056 if ($newnet->{$property}) {
1057 $optdata->{$property} = $newnet->{$property};
1059 delete $optdata->{$property};
1062 $conf->{$opt} = PVE
::LXC
::Config-
>print_lxc_network($optdata);
1063 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1064 $lxc_setup->setup_network($conf);
1067 &$change_ip_config(4);
1068 &$change_ip_config(6);
1072 my $open_namespace = sub {
1073 my ($vmid, $pid, $kind) = @_;
1074 sysopen my $fd, "/proc/$pid/ns/$kind", O_RDONLY
1075 or die "failed to open $kind namespace of container $vmid: $!\n";
1079 my $enter_namespace = sub {
1080 my ($vmid, $pid, $kind, $type) = @_;
1081 my $fd = $open_namespace->($vmid, $pid, $kind);
1082 PVE
::Tools
::setns
(fileno($fd), $type)
1083 or die "failed to enter $kind namespace of container $vmid: $!\n";
1087 my $get_container_namespace = sub {
1088 my ($vmid, $pid, $kind) = @_;
1091 if (!defined($pid)) {
1092 # Pin the pid while we're grabbing its stuff from /proc
1093 ($pid, $pidfd) = open_lxc_pid
($vmid)
1094 or die "failed to open pidfd of container $vmid\'s init process\n";
1097 return $open_namespace->($vmid, $pid, $kind);
1100 my $do_syncfs = sub {
1101 my ($vmid, $pid, $socket) = @_;
1103 &$enter_namespace($vmid, $pid, 'mnt', PVE
::Tools
::CLONE_NEWNS
);
1105 # Tell the parent process to start reading our /proc/mounts
1106 print {$socket} "go\n";
1109 # Receive /proc/self/mounts
1110 my $mountdata = do { local $/ = undef; <$socket> };
1126 # Now sync all mountpoints...
1127 my $mounts = PVE
::ProcFSTools
::parse_mounts
($mountdata);
1128 foreach my $mp (@$mounts) {
1129 my ($what, $dir, $fs) = @$mp;
1130 next if $nosyncfs{$fs};
1131 eval { PVE
::Tools
::sync_mountpoint
($dir); };
1136 sub sync_container_namespace
{
1138 my $pid = find_lxc_pid
($vmid);
1140 # SOCK_DGRAM is nicer for barriers but cannot be slurped
1141 socketpair my $pfd, my $cfd, AF_UNIX
, SOCK_STREAM
, PF_UNSPEC
1142 or die "failed to create socketpair: $!\n";
1145 die "fork failed: $!\n" if !defined($child);
1150 &$do_syncfs($vmid, $pid, $cfd);
1160 die "failed to enter container namespace\n" if $go ne "go\n";
1162 open my $mounts, '<', "/proc/$child/mounts"
1163 or die "failed to open container's /proc/mounts: $!\n";
1164 my $mountdata = do { local $/ = undef; <$mounts> };
1166 print {$pfd} $mountdata;
1169 while (waitpid($child, 0) != $child) {}
1170 die "failed to sync container namespace\n" if $? != 0;
1173 sub template_create
{
1174 my ($vmid, $conf) = @_;
1176 my $storecfg = PVE
::Storage
::config
();
1178 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1179 my ($ms, $mountpoint) = @_;
1181 my $volid = $mountpoint->{volume
};
1183 die "Template feature is not available for '$volid'\n"
1184 if !PVE
::Storage
::volume_has_feature
($storecfg, 'template', $volid);
1187 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1188 my ($ms, $mountpoint) = @_;
1190 my $volid = $mountpoint->{volume
};
1192 PVE
::Storage
::activate_volumes
($storecfg, [$volid]);
1194 my $template_volid = PVE
::Storage
::vdisk_create_base
($storecfg, $volid);
1195 $mountpoint->{volume
} = $template_volid;
1196 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq "rootfs");
1199 PVE
::LXC
::Config-
>write_config($vmid, $conf);
1202 sub check_ct_modify_config_perm
{
1203 my ($rpcenv, $authuser, $vmid, $pool, $newconf, $delete) = @_;
1205 return 1 if $authuser eq 'root@pam';
1208 my ($opt, $delete) = @_;
1209 if ($opt eq 'cores' || $opt eq 'cpuunits' || $opt eq 'cpulimit') {
1210 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.CPU']);
1211 } elsif ($opt eq 'rootfs' || $opt =~ /^mp\d+$/) {
1212 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Disk']);
1214 my $data = PVE
::LXC
::Config-
>parse_volume($opt, $newconf->{$opt});
1215 raise_perm_exc
("mount point type $data->{type} is only allowed for root\@pam")
1216 if $data->{type
} ne 'volume';
1217 } elsif ($opt eq 'memory' || $opt eq 'swap') {
1218 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Memory']);
1219 } elsif ($opt =~ m/^net\d+$/ || $opt eq 'nameserver' ||
1220 $opt eq 'searchdomain' || $opt eq 'hostname') {
1221 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Network']);
1222 } elsif ($opt eq 'features') {
1223 # For now this is restricted to root@pam
1224 raise_perm_exc
("changing feature flags is only allowed for root\@pam");
1225 } elsif ($opt eq 'hookscript') {
1226 # For now this is restricted to root@pam
1227 raise_perm_exc
("changing the hookscript is only allowed for root\@pam");
1229 $rpcenv->check_vm_perm($authuser, $vmid, $pool, ['VM.Config.Options']);
1233 foreach my $opt (keys %$newconf) {
1236 foreach my $opt (@$delete) {
1244 my ($vmid, $storage_cfg, $conf, $noerr) = @_;
1246 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1247 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1251 PVE
::LXC
::Config-
>foreach_volume_full($conf, {'reverse' => 1}, sub {
1252 my ($ms, $mountpoint) = @_;
1254 my $volid = $mountpoint->{volume
};
1255 my $mount = $mountpoint->{mp
};
1257 return if !$volid || !$mount;
1259 my $mount_path = "$rootdir/$mount";
1260 $mount_path =~ s!/+!/!g;
1262 return if !PVE
::ProcFSTools
::is_mounted
($mount_path);
1265 PVE
::Tools
::run_command
(['umount', '-d', $mount_path]);
1277 return $res; # tell caller if (some) umounts failed for the noerr case
1281 my ($vmid, $storage_cfg, $conf, $ignore_ro) = @_;
1283 my $rootdir = "/var/lib/lxc/$vmid/rootfs";
1284 File
::Path
::make_path
($rootdir);
1286 my $volid_list = PVE
::LXC
::Config-
>get_vm_volumes($conf);
1287 PVE
::Storage
::activate_volumes
($storage_cfg, $volid_list);
1289 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
1292 PVE
::LXC
::Config-
>foreach_volume($conf, sub {
1293 my ($ms, $mountpoint) = @_;
1295 $mountpoint->{ro
} = 0 if $ignore_ro;
1297 mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, undef, $rootuid, $rootgid);
1301 warn "mounting container failed\n";
1302 umount_all
($vmid, $storage_cfg, $conf, 1);
1310 sub mountpoint_mount_path
{
1311 my ($mountpoint, $storage_cfg, $snapname) = @_;
1313 return mountpoint_mount
($mountpoint, undef, $storage_cfg, $snapname);
1321 if ($line =~ m
@^(/dev/loop\d
+):@) {
1325 my $cmd = ['losetup', '--associated', $path];
1326 PVE
::Tools
::run_command
($cmd, outfunc
=> $parser);
1330 # Run a function with a file attached to a loop device.
1331 # The loop device is always detached afterwards (or set to autoclear).
1332 # Returns the loop device.
1333 sub run_with_loopdev
{
1334 my ($func, $file, $readonly) = @_;
1335 my $device = query_loopdev
($file);
1336 # Try to reuse an existing device
1338 # We assume that whoever setup the loop device is responsible for
1346 if ($line =~ m
@^(/dev/loop\d
+)$@) {
1356 push @$losetup_cmd, '-r' if $readonly;
1357 PVE
::Tools
::run_command
($losetup_cmd, outfunc
=> $parser);
1358 die "failed to setup loop device for $file\n" if !$device;
1359 eval { &$func($device); };
1361 PVE
::Tools
::run_command
(['losetup', '-d', $device]);
1366 # In scalar mode: returns a file handle to the deepest directory node.
1367 # In list context: returns a list of:
1368 # * the deepest directory node
1369 # * the 2nd deepest directory (parent of the above)
1370 # * directory name of the last directory
1371 # So that the path $2/$3 should lead to $1 afterwards.
1372 sub walk_tree_nofollow
($$$;$$) {
1373 my ($start, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1375 sysopen(my $fd, $start, O_PATH
| O_DIRECTORY
)
1376 or die "failed to open start directory $start: $!\n";
1378 return walk_tree_nofollow_fd
($start, $fd, $subdir, $mkdir, $rootuid, $rootgid);
1382 sub walk_tree_nofollow_fd
($$$$;$$) {
1383 my ($start_dirname, $start_fd, $subdir, $mkdir, $rootuid, $rootgid) = @_;
1385 # splitdir() returns '' for empty components including the leading /
1386 my @comps = grep { length($_)>0 } File
::Spec-
>splitdir($subdir);
1389 my $dir = $start_dirname;
1390 my $last_component = undef;
1392 foreach my $component (@comps) {
1393 $dir .= "/$component";
1394 my $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1397 # failed, check for symlinks and try to create the path
1398 die "symlink encountered at: $dir\n" if $! == ELOOP
|| $! == ENOTDIR
;
1399 die "cannot open directory $dir: $!\n" if !$mkdir;
1401 # We don't check for errors on mkdirat() here and just try to
1402 # openat() again, since at least one error (EEXIST) is an
1403 # expected possibility if multiple containers start
1404 # simultaneously. If someone else injects a symlink now then
1405 # the subsequent openat() will fail due to O_NOFOLLOW anyway.
1406 PVE
::Tools
::mkdirat
(fileno($fd), $component, 0755);
1408 $next = PVE
::Tools
::openat
(fileno($fd), $component, O_NOFOLLOW
| O_DIRECTORY
);
1409 die "failed to create path: $dir: $!\n" if !$next;
1411 PVE
::Tools
::fchownat
(fileno($next), '', $rootuid, $rootgid, PVE
::Tools
::AT_EMPTY_PATH
)
1412 if defined($rootuid) && defined($rootgid);
1415 close $second if defined($last_component) && $second != $start_fd;
1416 $last_component = $component;
1421 return ($fd, defined($last_component) && $second, $last_component) if wantarray;
1422 close $second if defined($last_component) && $second != $start_fd;
1426 # To guard against symlink attack races against other currently running
1427 # containers with shared recursive bind mount hierarchies we prepare a
1428 # directory handle for the directory we're mounting over to verify the
1429 # mountpoint afterwards.
1430 sub __bindmount_prepare
{
1431 my ($hostroot, $dir) = @_;
1432 my $srcdh = walk_tree_nofollow
($hostroot, $dir, 0);
1436 # Assuming we mount to rootfs/a/b/c, verify with the directory handle to 'b'
1437 # ($parentfd) that 'b/c' (openat($parentfd, 'c')) really leads to the directory
1438 # we intended to bind mount.
1439 sub __bindmount_verify
{
1440 my ($srcdh, $parentfd, $last_dir, $ro) = @_;
1443 # Open the mount point path coming from the parent directory since the
1444 # filehandle we would have gotten as first result of walk_tree_nofollow
1445 # earlier is still a handle to the underlying directory instead of the
1447 $destdh = PVE
::Tools
::openat
(fileno($parentfd), $last_dir, PVE
::Tools
::O_PATH
| O_NOFOLLOW
| O_DIRECTORY
);
1448 die "failed to open mount point: $!\n" if !$destdh;
1451 # no separate function because 99% of the time it's the wrong thing to use.
1452 if (syscall(PVE
::Syscall
::faccessat
, fileno($destdh), $dot, &POSIX
::W_OK
, 0) != -1) {
1453 die "failed to mark bind mount read only\n";
1455 die "read-only check failed: $!\n" if $! != EROFS
;
1458 # For the rootfs we don't have a parentfd so we open the path directly.
1459 # Note that this means bindmounting any prefix of the host's
1460 # /var/lib/lxc/$vmid path into another container is considered a grave
1462 sysopen $destdh, $last_dir, O_PATH
| O_DIRECTORY
;
1463 die "failed to open mount point: $!\n" if !$destdh;
1466 my ($srcdev, $srcinode) = stat($srcdh);
1467 my ($dstdev, $dstinode) = stat($destdh);
1471 return ($srcdev == $dstdev && $srcinode == $dstinode);
1474 # Perform the actual bind mounting:
1475 sub __bindmount_do
{
1476 my ($dir, $dest, $ro, @extra_opts) = @_;
1477 PVE
::Tools
::run_command
(['mount', '-o', 'bind', @extra_opts, $dir, $dest]);
1479 eval { PVE
::Tools
::run_command
(['mount', '-o', 'bind,remount,ro', $dest]); };
1481 warn "bindmount error\n";
1482 # don't leave writable bind-mounts behind...
1483 PVE
::Tools
::run_command
(['umount', $dest]);
1490 my ($dir, $parentfd, $last_dir, $dest, $ro, @extra_opts) = @_;
1492 my $srcdh = __bindmount_prepare
('/', $dir);
1494 __bindmount_do
($dir, $dest, $ro, @extra_opts);
1496 if (!__bindmount_verify
($srcdh, $parentfd, $last_dir, $ro)) {
1497 PVE
::Tools
::run_command
(['umount', $dest]);
1498 die "detected mount path change at: $dir\n";
1502 # Cleanup $rootdir a bit (double and trailing slashes), build the mount path
1503 # from $rootdir and $mount and walk the path from $rootdir to the final
1504 # directory to check for symlinks.
1505 sub __mount_prepare_rootdir
{
1506 my ($rootdir, $mount, $rootuid, $rootgid) = @_;
1507 $rootdir =~ s!/+!/!g;
1508 $rootdir =~ s!/+$!!;
1509 my $mount_path = "$rootdir/$mount";
1510 my ($mpfd, $parentfd, $last_dir) = walk_tree_nofollow
($rootdir, $mount, 1, $rootuid, $rootgid);
1511 return ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir);
1514 # use $rootdir = undef to just return the corresponding mount path
1515 sub mountpoint_mount
{
1516 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1517 return __mountpoint_mount
($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, undef);
1520 sub mountpoint_stage
{
1521 my ($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid) = @_;
1522 my ($path, $loop, $dev) =
1523 __mountpoint_mount
($mountpoint, $stage_dir, $storage_cfg, $snapname, $rootuid, $rootgid, 1);
1525 if (!defined($path)) {
1526 return undef if $! == ENOSYS
;
1527 die "failed to mount subvolume: $!\n";
1530 # We clone the mount point and leave it there in order to keep them connected to eg. loop
1531 # devices in case we're hotplugging (which would allow contaienrs to unmount the new mount
1534 my $fd = PVE
::Tools
::open_tree
(&AT_FDCWD
, $stage_dir, &OPEN_TREE_CLOEXEC
| &OPEN_TREE_CLONE
)
1535 or die "open_tree() on mount point failed: $!\n";
1537 return wantarray ?
($path, $loop, $dev, $fd) : $fd;
1540 sub mountpoint_insert_staged
{
1541 my ($mount_fd, $rootdir_fd, $mp_dir, $opt, $rootuid, $rootgid) = @_;
1543 if (!defined($rootdir_fd)) {
1544 sysopen($rootdir_fd, '.', O_PATH
| O_DIRECTORY
)
1545 or die "failed to open '.': $!\n";
1548 my $dest_fd = walk_tree_nofollow_fd
('/', $rootdir_fd, $mp_dir, 1, $rootuid, $rootgid);
1550 PVE
::Tools
::move_mount
(
1555 &MOVE_MOUNT_F_EMPTY_PATH
| &MOVE_MOUNT_T_EMPTY_PATH
,
1556 ) or die "failed to move '$opt' into container hierarchy: $!\n";
1559 # Use $stage_mount, $rootdir is treated as a temporary path to "stage" the file system. The user
1560 # can then open a file descriptor to it which can be used with the `move_mount` syscall.
1561 # Note that if the kernel does not support the new mount API, this will not perform any action
1562 # and return `undef` with $! = ENOSYS.
1563 sub __mountpoint_mount
{
1564 my ($mountpoint, $rootdir, $storage_cfg, $snapname, $rootuid, $rootgid, $stage_mount) = @_;
1566 if (defined($stage_mount) && !PVE
::LXC
::Tools
::can_use_new_mount_api
()) {
1571 # When staging mount points we always mount to $rootdir directly (iow. as if `mp=/`).
1572 # This is required since __mount_prepare_rootdir() will return handles to the parent directory
1573 # which we use in __bindmount_verify()!
1574 my $mount = $stage_mount ?
'/': $mountpoint->{mp
};
1576 my $volid = $mountpoint->{volume
};
1577 my $type = $mountpoint->{type
};
1578 my $quota = !$snapname && !$mountpoint->{ro
} && $mountpoint->{quota
};
1581 return if !$volid || !$mount;
1586 my ($mpfd, $parentfd, $last_dir);
1588 if (defined($rootdir)) {
1589 ($rootdir, $mount_path, $mpfd, $parentfd, $last_dir) =
1590 __mount_prepare_rootdir
($rootdir, $mount, $rootuid, $rootgid);
1593 if (defined($stage_mount)) {
1594 $mount_path = $rootdir;
1597 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1599 die "unknown snapshot path for '$volid'" if !$storage && defined($snapname);
1603 if (my $mountopts = $mountpoint->{mountoptions
}) {
1604 my @opts = split(/;/, $mountpoint->{mountoptions
});
1605 push @$optlist, grep { PVE
::LXC
::Config
::is_valid_mount_option
($_) } @opts;
1608 my $acl = $mountpoint->{acl
};
1609 if (defined($acl)) {
1610 push @$optlist, ($acl ?
'acl' : 'noacl');
1613 my $optstring = join(',', @$optlist);
1614 my $readonly = $mountpoint->{ro
};
1617 @extra_opts = ('-o', $optstring) if $optstring;
1621 my $scfg = PVE
::Storage
::storage_config
($storage_cfg, $storage);
1623 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid, $snapname);
1625 $path = PVE
::Storage
::path
($storage_cfg, $volid, $snapname) if !defined($path);
1627 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1628 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1630 $format = 'iso' if $vtype eq 'iso'; # allow to handle iso files
1632 if ($format eq 'subvol') {
1634 my (undef, $name) = PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1635 if (defined($snapname)) {
1636 $name .= "\@$snapname";
1637 if ($scfg->{type
} eq 'zfspool') {
1638 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, '-t', 'zfs', "$scfg->{pool}/$name", $mount_path]);
1640 die "cannot mount subvol snapshots for storage type '$scfg->{type}'\n";
1643 if (defined($acl) && $scfg->{type
} eq 'zfspool') {
1644 my $acltype = ($acl ?
'acltype=posixacl' : 'acltype=noacl');
1645 PVE
::Tools
::run_command
(['zfs', 'set', $acltype, "$scfg->{pool}/$name"]);
1647 bindmount
($path, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts);
1648 warn "cannot enable quota control for bind mounted subvolumes\n" if $quota;
1651 return wantarray ?
($path, 0, undef) : $path;
1652 } elsif ($format eq 'raw' || $format eq 'iso') {
1653 # NOTE: 'mount' performs canonicalization without the '-c' switch, which for
1654 # device-mapper devices is special-cased to use the /dev/mapper symlinks.
1655 # Our autodev hook expects the /dev/dm-* device currently
1656 # and will create the /dev/mapper symlink accordingly
1657 $path = Cwd
::realpath
($path);
1658 die "failed to get device path\n" if !$path;
1659 ($path) = ($path =~ /^(.*)$/s); #untaint
1663 if ($format eq 'iso') {
1664 PVE
::Tools
::run_command
(['mount', '-o', 'ro', @extra_opts, $path, $mount_path]);
1665 } elsif ($isBase || defined($snapname)) {
1666 PVE
::Tools
::run_command
(['mount', '-o', 'ro,noload', @extra_opts, $path, $mount_path]);
1669 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0';
1671 push @extra_opts, '-o', 'ro' if $readonly;
1672 PVE
::Tools
::run_command
(['mount', @extra_opts, $path, $mount_path]);
1676 my $use_loopdev = 0;
1677 if ($scfg->{path
}) {
1678 $mounted_dev = run_with_loopdev
($domount, $path, $readonly);
1680 } elsif ($scfg->{type
} eq 'drbd' || $scfg->{type
} eq 'lvm' ||
1681 $scfg->{type
} eq 'rbd' || $scfg->{type
} eq 'lvmthin') {
1682 $mounted_dev = $path;
1685 die "unsupported storage type '$scfg->{type}'\n";
1687 return wantarray ?
($path, $use_loopdev, $mounted_dev) : $path;
1689 die "unsupported image format '$format'\n";
1691 } elsif ($type eq 'device') {
1692 push @extra_opts, '-o', 'ro' if $readonly;
1693 push @extra_opts, '-o', 'usrjquota=aquota.user,grpjquota=aquota.group,jqfmt=vfsv0' if $quota;
1694 # See the NOTE above about devicemapper canonicalization
1695 my ($devpath) = (Cwd
::realpath
($volid) =~ /^(.*)$/s); # realpath() taints
1696 PVE
::Tools
::run_command
(['mount', @extra_opts, $volid, $mount_path]) if $mount_path;
1697 return wantarray ?
($volid, 0, $devpath) : $volid;
1698 } elsif ($type eq 'bind') {
1699 die "directory '$volid' does not exist\n" if ! -d
$volid;
1700 bindmount
($volid, $parentfd, $last_dir//$rootdir, $mount_path, $readonly, @extra_opts) if $mount_path;
1701 warn "cannot enable quota control for bind mounts\n" if $quota;
1702 return wantarray ?
($volid, 0, undef) : $volid;
1705 die "unsupported storage";
1708 sub mountpoint_hotplug
($$$) {
1709 my ($vmid, $conf, $opt, $mp, $storage_cfg) = @_;
1711 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
1713 # We do the rest in a fork with an unshared mount namespace, because:
1714 # -) change our papparmor profile to that of /usr/bin/lxc-start
1715 # -) we're now going to 'stage' # the mountpoint, then grab it, then move into the
1716 # container's namespace, then mount it.
1718 PVE
::Tools
::run_fork
(sub {
1719 # Pin the container pid longer, we also need to get its monitor/parent:
1720 my ($ct_pid, $ct_pidfd) = open_lxc_pid
($vmid)
1721 or die "failed to open pidfd of container $vmid\'s init process\n";
1723 my ($monitor_pid, $monitor_pidfd) = open_ppid
($ct_pid)
1724 or die "failed to open pidfd of container $vmid\'s monitor process\n";
1726 my $ct_mnt_ns = $get_container_namespace->($vmid, $ct_pid, 'mnt');
1727 my $monitor_mnt_ns = $get_container_namespace->($vmid, $monitor_pid, 'mnt');
1729 # Grab a file descriptor to our apparmor label file so we can change into the 'lxc-start'
1730 # profile to lower our privileges to the same level we have in the start hook:
1731 sysopen(my $aa_fd, "/proc/self/attr/current", O_WRONLY
)
1732 or die "failed to open '/proc/self/attr/current' for writing: $!\n";
1733 # But switch namespaces first, to make sure the namespace switches aren't blocked by
1736 # Change into the monitor's mount namespace. We "pin" the mount into the monitor's
1737 # namespace for it to remain active there since the container will be able to unmount
1738 # hotplugged mount points and thereby potentially free up loop devices, which is a security
1740 PVE
::Tools
::setns
(fileno($monitor_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1742 or die "failed to change root directory within the monitor's mount namespace: $!\n";
1744 my $dir = get_staging_mount_path
($opt);
1746 # Now switch our apparmor profile before mounting:
1747 my $data = 'changeprofile /usr/bin/lxc-start';
1748 if (syswrite($aa_fd, $data, length($data)) != length($data)) {
1749 die "failed to change apparmor profile: $!\n";
1751 # Check errors on close as well:
1753 or die "failed to change apparmor profile (close() failed): $!\n";
1755 my $mount_fd = mountpoint_stage
($mp, $dir, $storage_cfg, undef, $rootuid, $rootgid);
1757 PVE
::Tools
::setns
(fileno($ct_mnt_ns), PVE
::Tools
::CLONE_NEWNS
);
1759 or die "failed to change root directory within the container's mount namespace: $!\n";
1761 mountpoint_insert_staged
($mount_fd, undef, $mp->{mp
}, $opt, $rootuid, $rootgid);
1765 # Create a directory in the mountpoint staging tempfs.
1766 sub get_staging_mount_path
($) {
1769 my $target = get_staging_tempfs
() . "/$opt";
1770 if (!mkdir($target) && $! != EEXIST
) {
1771 die "failed to create directory $target: $!\n";
1777 # Mount /run/pve/mountpoints as tmpfs
1778 sub get_staging_tempfs
() {
1779 # We choose a path in /var/lib/lxc/ here because the lxc-start apparmor profile restricts most
1781 my $target = '/var/lib/lxc/.pve-staged-mounts';
1782 if (!mkdir($target)) {
1783 return $target if $! == EEXIST
;
1784 die "failed to create directory $target: $!\n";
1787 PVE
::Tools
::mount
("none", $target, 'tmpfs', 0, "size=8k,mode=755")
1788 or die "failed to mount $target as tmpfs: $!\n";
1794 my ($dev, $rootuid, $rootgid) = @_;
1802 "root_owner=$rootuid:$rootgid",
1807 # a hack to print only the relevant stuff, i.e., the one which could help on repair
1808 if ($line =~ /^(Creating filesystem|Filesystem UUID|Superblock backups|\s+\d+, \d)/) {
1814 print STDERR
"$line\n" if $line && $line !~ /^mke2fs \d\.\d/;
1820 my ($storage_cfg, $volid, $rootuid, $rootgid) = @_;
1822 if ($volid =~ m!^/dev/.+!) {
1827 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1829 die "cannot format volume '$volid' with no storage\n" if !$storage;
1831 PVE
::Storage
::activate_volumes
($storage_cfg, [$volid]);
1833 my $path = PVE
::Storage
::map_volume
($storage_cfg, $volid);
1835 $path = PVE
::Storage
::path
($storage_cfg, $volid) if !defined($path);
1837 my ($vtype, undef, undef, undef, undef, $isBase, $format) =
1838 PVE
::Storage
::parse_volname
($storage_cfg, $volid);
1840 die "cannot format volume '$volid' (format == $format)\n"
1841 if $format ne 'raw';
1843 mkfs
($path, $rootuid, $rootgid);
1847 my ($storecfg, $vollist) = @_;
1849 foreach my $volid (@$vollist) {
1850 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
1856 my ($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid) = @_;
1858 my $needs_chown = 0;
1861 my $scfg = PVE
::Storage
::storage_config
($storecfg, $storage);
1862 # fixme: use better naming ct-$vmid-disk-X.raw?
1866 if ($scfg->{type
} eq 'dir' || $scfg->{type
} eq 'nfs' || $scfg->{type
} eq 'cifs' ) {
1868 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw',
1872 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol',
1876 } elsif ($scfg->{type
} eq 'zfspool') {
1878 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'subvol',
1881 } elsif ($scfg->{type
} eq 'drbd' || $scfg->{type
} eq 'lvm' || $scfg->{type
} eq 'lvmthin') {
1883 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1886 } elsif ($scfg->{type
} eq 'rbd') {
1888 $volid = PVE
::Storage
::vdisk_alloc
($storecfg, $storage, $vmid, 'raw', undef, $size_kb);
1891 die "unable to create containers on storage type '$scfg->{type}'\n";
1893 format_disk
($storecfg, $volid, $rootuid, $rootgid) if $do_format;
1896 # in case formatting got interrupted:
1897 if (defined($volid)) {
1898 eval { PVE
::Storage
::vdisk_free
($storecfg, $volid); };
1904 return ($volid, $needs_chown);
1907 our $NEW_DISK_RE = qr/^([^:\s]+):(\d+(\.\d+)?)$/;
1909 my ($storecfg, $vmid, $settings, $conf, $pending) = @_;
1914 my (undef, $rootuid, $rootgid) = PVE
::LXC
::parse_id_maps
($conf);
1915 my $chown_vollist = [];
1917 PVE
::LXC
::Config-
>foreach_volume($settings, sub {
1918 my ($ms, $mountpoint) = @_;
1920 my $volid = $mountpoint->{volume
};
1921 my $mp = $mountpoint->{mp
};
1923 my ($storage, $volname) = PVE
::Storage
::parse_volume_id
($volid, 1);
1925 if ($storage && ($volid =~ $NEW_DISK_RE)) {
1926 my ($storeid, $size_gb) = ($1, $2);
1928 my $size_kb = int(${size_gb
}*1024) * 1024;
1930 my $needs_chown = 0;
1931 ($volid, $needs_chown) = alloc_disk
($storecfg, $vmid, $storage, $size_kb, $rootuid, $rootgid);
1932 push @$chown_vollist, $volid if $needs_chown;
1933 push @$vollist, $volid;
1934 $mountpoint->{volume
} = $volid;
1935 $mountpoint->{size
} = $size_kb * 1024;
1937 $conf->{pending
}->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1939 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1942 # use specified/existing volid/dir/device
1943 $conf->{$ms} = PVE
::LXC
::Config-
>print_ct_mountpoint($mountpoint, $ms eq 'rootfs');
1947 PVE
::Storage
::activate_volumes
($storecfg, $chown_vollist, undef);
1948 foreach my $volid (@$chown_vollist) {
1949 my $path = PVE
::Storage
::path
($storecfg, $volid, undef);
1950 chown($rootuid, $rootgid, $path);
1952 PVE
::Storage
::deactivate_volumes
($storecfg, $chown_vollist, undef);
1954 # free allocated images on error
1956 destroy_disks
($storecfg, $vollist);
1962 sub update_disksize
{
1963 my ($vmid, $conf, $all_volumes) = @_;
1966 my $prefix = "CT $vmid:";
1968 my $update_mp = sub {
1969 my ($key, $mp, @param) = @_;
1970 my $size = $all_volumes->{$mp->{volume
}}->{size
} // 0;
1972 if (!defined($mp->{size
}) || $size != $mp->{size
}) {
1974 print "$prefix updated volume size of '$mp->{volume}' in config.\n";
1975 $mp->{size
} = $size;
1976 my $nomp = 1 if ($key eq 'rootfs');
1977 $conf->{$key} = PVE
::LXC
::Config-
>print_ct_mountpoint($mp, $nomp);
1981 PVE
::LXC
::Config-
>foreach_volume($conf, $update_mp);
1987 my ($vmid, $conf, $all_volumes) = @_;
1990 my $prefix = "CT $vmid:";
1992 # Note: it is allowed to define multiple storage entries with the same path
1993 # (alias), so we need to check both 'volid' and real 'path' (two different
1994 # volid can point to the same path).
1996 # used and unused disks
2000 foreach my $opt (keys %$conf) {
2001 next if ($opt !~ m/^unused\d+$/);
2002 my $vol = $all_volumes->{$conf->{$opt}};
2003 $refpath->{$vol->{path
}} = $vol->{volid
};
2006 foreach my $key (keys %$all_volumes) {
2007 my $vol = $all_volumes->{$key};
2008 my $in_use = PVE
::LXC
::Config-
>is_volume_in_use($conf, $vol->{volid
});
2009 my $path = $vol->{path
};
2012 $refpath->{$path} = $key;
2013 delete $orphans->{$path};
2015 if ((!$orphans->{$path}) && (!$refpath->{$path})) {
2016 $orphans->{$path} = $key;
2021 for my $key (keys %$orphans) {
2022 my $disk = $orphans->{$key};
2023 my $unused = PVE
::LXC
::Config-
>add_unused_volume($conf, $disk);
2027 print "$prefix add unreferenced volume '$disk' as '$unused' to config.\n";
2035 my ($cfg, $vmid) = @_;
2037 my $info = PVE
::Storage
::vdisk_list
($cfg, undef, $vmid);
2039 my $all_volumes = {};
2040 foreach my $storeid (keys %$info) {
2041 foreach my $item (@{$info->{$storeid}}) {
2042 my $volid = $item->{volid
};
2043 next if !($volid && $item->{size
});
2044 $item->{path
} = PVE
::Storage
::path
($cfg, $volid);
2045 $all_volumes->{$volid} = $item;
2049 return $all_volumes;
2053 my ($vmid, $nolock, $dryrun) = @_;
2055 my $cfg = PVE
::Storage
::config
();
2057 # FIXME: Remove once our RBD plugin can handle CT and VM on a single storage
2058 # see: https://pve.proxmox.com/pipermail/pve-devel/2018-July/032900.html
2059 foreach my $stor (keys %{$cfg->{ids
}}) {
2060 delete($cfg->{ids
}->{$stor}) if !$cfg->{ids
}->{$stor}->{content
}->{rootdir
};
2063 print "rescan volumes...\n";
2064 my $all_volumes = scan_volids
($cfg, $vmid);
2066 my $updatefn = sub {
2070 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2072 PVE
::LXC
::Config-
>check_lock($conf);
2075 foreach my $volid (keys %$all_volumes) {
2076 my $info = $all_volumes->{$volid};
2077 $vm_volids->{$volid} = $info if $info->{vmid
} == $vmid;
2080 my $upu = update_unused
($vmid, $conf, $vm_volids);
2081 my $upd = update_disksize
($vmid, $conf, $vm_volids);
2082 $changes = $upu || $upd;
2084 PVE
::LXC
::Config-
>write_config($vmid, $conf) if $changes && !$dryrun;
2087 if (defined($vmid)) {
2091 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2094 my $vmlist = config_list
();
2095 foreach my $vmid (keys %$vmlist) {
2099 PVE
::LXC
::Config-
>lock_config($vmid, $updatefn, $vmid);
2106 # bash completion helper
2108 sub complete_os_templates
{
2109 my ($cmdname, $pname, $cvalue) = @_;
2111 my $cfg = PVE
::Storage
::config
();
2115 if ($cvalue =~ m/^([^:]+):/) {
2119 my $vtype = $cmdname eq 'restore' ?
'backup' : 'vztmpl';
2120 my $data = PVE
::Storage
::template_list
($cfg, $storeid, $vtype);
2123 foreach my $id (keys %$data) {
2124 foreach my $item (@{$data->{$id}}) {
2125 push @$res, $item->{volid
} if defined($item->{volid
});
2132 my $complete_ctid_full = sub {
2135 my $idlist = vmstatus
();
2137 my $active_hash = list_active_containers
();
2141 foreach my $id (keys %$idlist) {
2142 my $d = $idlist->{$id};
2143 if (defined($running)) {
2144 next if $d->{template
};
2145 next if $running && !$active_hash->{$id};
2146 next if !$running && $active_hash->{$id};
2155 return &$complete_ctid_full();
2158 sub complete_ctid_stopped
{
2159 return &$complete_ctid_full(0);
2162 sub complete_ctid_running
{
2163 return &$complete_ctid_full(1);
2173 my $lxc = $conf->{lxc
};
2174 foreach my $entry (@$lxc) {
2175 my ($key, $value) = @$entry;
2176 # FIXME: remove the 'id_map' variant when lxc-3.0 arrives
2177 next if $key ne 'lxc.idmap' && $key ne 'lxc.id_map';
2178 if ($value =~ /^([ug])\s+(\d+)\s+(\d+)\s+(\d+)\s*$/) {
2179 my ($type, $ct, $host, $length) = ($1, $2, $3, $4);
2180 push @$id_map, [$type, $ct, $host, $length];
2182 $rootuid = $host if $type eq 'u';
2183 $rootgid = $host if $type eq 'g';
2186 die "failed to parse idmap: $value\n";
2190 if (!@$id_map && $conf->{unprivileged
}) {
2191 # Should we read them from /etc/subuid?
2192 $id_map = [ ['u', '0', '100000', '65536'],
2193 ['g', '0', '100000', '65536'] ];
2194 $rootuid = $rootgid = 100000;
2197 return ($id_map, $rootuid, $rootgid);
2200 sub userns_command
{
2203 return ['lxc-usernsexec', (map { ('-m', join(':', @$_)) } @$id_map), '--'];
2208 my sub print_ct_stderr_log
{
2210 my $log = eval { file_get_contents
("/run/pve/ct-$vmid.stderr") };
2213 while ($log =~ /^\h*(lxc-start:?\s+$vmid:?\s*\S+\s*)?(.*?)\h*$/gm) {
2215 print STDERR
"$line\n";
2219 my sub monitor_state_change
($$) {
2220 my ($monitor_socket, $vmid) = @_;
2221 die "no monitor socket\n" if !defined($monitor_socket);
2224 my ($type, $name, $value) = PVE
::LXC
::Monitor
::read_lxc_message
($monitor_socket);
2226 die "monitor socket: got EOF\n" if !defined($type);
2228 next if $name ne "$vmid" || $type ne 'STATE';
2230 if ($value eq PVE
::LXC
::Monitor
::STATE_STARTING
) {
2231 alarm(0); # don't timeout after seeing the starting state
2232 } elsif ($value eq PVE
::LXC
::Monitor
::STATE_ABORTING
||
2233 $value eq PVE
::LXC
::Monitor
::STATE_STOPPING
||
2234 $value eq PVE
::LXC
::Monitor
::STATE_STOPPED
) {
2236 } elsif ($value eq PVE
::LXC
::Monitor
::STATE_RUNNING
) {
2239 warn "unexpected message from monitor socket - " .
2240 "type: '$type' - value: '$value'\n";
2244 my sub monitor_start
($$) {
2245 my ($monitor_socket, $vmid) = @_;
2247 my $success = eval {
2248 PVE
::Tools
::run_with_timeout
(10, \
&monitor_state_change
, $monitor_socket, $vmid)
2251 warn "problem with monitor socket, but continuing anyway: $err\n";
2252 } elsif (!$success) {
2253 print_ct_stderr_log
($vmid);
2254 die "startup for container '$vmid' failed\n";
2259 my ($vmid, $conf, $skiplock, $debug) = @_;
2261 # apply pending changes while starting
2262 if (scalar(keys %{$conf->{pending
}})) {
2263 my $storecfg = PVE
::Storage
::config
();
2264 PVE
::LXC
::Config-
>vmconfig_apply_pending($vmid, $conf, $storecfg);
2265 PVE
::LXC
::Config-
>write_config($vmid, $conf);
2266 $conf = PVE
::LXC
::Config-
>load_config($vmid); # update/reload
2269 update_lxc_config
($vmid, $conf);
2271 my $skiplock_flag_fn = "/run/lxc/skiplock-$vmid";
2274 open(my $fh, '>', $skiplock_flag_fn) || die "failed to open $skiplock_flag_fn for writing: $!\n";
2278 my $storage_cfg = PVE
::Storage
::config
();
2279 my $vollist = PVE
::LXC
::Config-
>get_vm_volumes($conf);
2281 PVE
::Storage
::activate_volumes
($storage_cfg, $vollist);
2283 my $monitor_socket = eval { PVE
::LXC
::Monitor
::get_monitor_socket
() };
2286 unlink "/run/pve/ct-$vmid.stderr"; # systemd does not truncate log files
2288 my $is_debug = $debug || (!defined($debug) && $conf->{debug
});
2289 my $base_unit = $is_debug ?
'pve-container-debug' : 'pve-container';
2291 my $cmd = ['systemctl', 'start', "$base_unit\@$vmid"];
2293 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-start', 1);
2297 monitor_start
($monitor_socket, $vmid) if defined($monitor_socket);
2299 # if debug is requested, print the log it also when the start succeeded
2300 print_ct_stderr_log
($vmid) if $is_debug;
2303 unlink $skiplock_flag_fn;
2306 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'post-start');
2311 # Helper to stop a container completely and make sure it has stopped completely.
2312 # This is necessary because we want the post-stop hook to have completed its
2313 # unmount-all step, but post-stop happens after lxc puts the container into the
2315 # $kill - if true it will always do an immediate hard-stop
2316 # $shutdown_timeout - the timeout to wait for a gracefull shutdown
2317 # $kill_after_timeout - if true, send a hardstop if shutdown timed out
2319 my ($vmid, $kill, $shutdown_timeout, $kill_after_timeout) = @_;
2321 # Open the container's command socket.
2322 my $path = "\0/var/lib/lxc/$vmid/command";
2323 my $sock = IO
::Socket
::UNIX-
>new(
2324 Type
=> SOCK_STREAM
(),
2328 return if $! == ECONNREFUSED
; # The container is not running
2329 die "failed to open container ${vmid}'s command socket: $!\n";
2332 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2333 PVE
::GuestHelpers
::exec_hookscript
($conf, $vmid, 'pre-stop');
2335 # Stop the container:
2337 my $cmd = ['lxc-stop', '-n', $vmid];
2340 push @$cmd, '--kill'; # doesn't allow timeouts
2342 # lxc-stop uses a default timeout
2343 push @$cmd, '--nokill' if !$kill_after_timeout;
2345 if (defined($shutdown_timeout)) {
2346 push @$cmd, '--timeout', $shutdown_timeout;
2347 # Give run_command 5 extra seconds
2348 $shutdown_timeout += 5;
2352 eval { run_command
($cmd, timeout
=> $shutdown_timeout) };
2357 my $result = <$sock>;
2359 return if !defined $result; # monitor is gone and the ct has stopped.
2360 die "container did not stop\n";
2364 my ($vmid, $timeout, $skiplock) = @_;
2366 PVE
::LXC
::Config-
>lock_config($vmid, sub {
2367 return if !check_running
($vmid);
2369 vm_stop
($vmid, 0, $timeout, 1); # kill if timeout exceeds
2371 my $conf = PVE
::LXC
::Config-
>load_config($vmid);
2372 vm_start
($vmid, $conf);
2379 return PVE
::Tools
::run_fork
(sub {
2380 # Unshare the mount namespace
2381 die "failed to unshare mount namespace: $!\n"
2382 if !PVE
::Tools
::unshare
(PVE
::Tools
::CLONE_NEWNS
);
2383 run_command
(['mount', '--make-rslave', '/']);
2388 my $copy_volume = sub {
2389 my ($src_volid, $src, $dst_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid) = @_;
2391 my $src_mp = { volume
=> $src_volid, mp
=> '/', ro
=> 1 };
2392 $src_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($src_volid);
2394 my $dst_mp = { volume
=> $dst_volid, mp
=> '/', ro
=> 0 };
2395 $dst_mp->{type
} = PVE
::LXC
::Config-
>classify_mountpoint($dst_volid);
2401 mountpoint_mount
($src_mp, $src, $storage_cfg, $snapname, $rootuid, $rootgid);
2402 push @mounted, $src;
2404 mountpoint_mount
($dst_mp, $dest, $storage_cfg, undef, $rootuid, $rootgid);
2405 push @mounted, $dest;
2418 '--one-file-system',
2419 "--bwlimit=$bwlimit",
2426 # Wait for rsync's children to release dest so that
2427 # consequent file operations (umount, remove) are possible
2428 while ((system {"fuser"} "fuser", "-s", $dest) == 0) {sleep 1};
2430 foreach my $mount (reverse @mounted) {
2431 eval { run_command
(['/bin/umount', $mount], errfunc
=> sub{})};
2432 warn "Can't umount $mount\n" if $@;
2435 # If this fails they're used as mount points in a concurrent operation
2436 # (which should not happen but there's also no real need to get rid of them).
2443 # Should not be called after unsharing the mount namespace!
2445 my ($mp, $vmid, $storage, $storage_cfg, $conf, $snapname, $bwlimit) = @_;
2447 die "cannot copy volumes of type $mp->{type}\n" if $mp->{type
} ne 'volume';
2448 File
::Path
::make_path
("/var/lib/lxc/$vmid");
2449 my $dest = "/var/lib/lxc/$vmid/.copy-volume-1";
2450 my $src = "/var/lib/lxc/$vmid/.copy-volume-2";
2452 # get id's for unprivileged container
2453 my (undef, $rootuid, $rootgid) = parse_id_maps
($conf);
2455 # Allocate the disk before unsharing in order to make sure zfs subvolumes
2456 # are visible in this namespace, otherwise the host only sees the empty
2457 # (not-mounted) directory.
2460 # Make sure $mp contains a correct size.
2461 $mp->{size
} = PVE
::Storage
::volume_size_info
($storage_cfg, $mp->{volume
});
2463 ($new_volid, $needs_chown) = alloc_disk
($storage_cfg, $vmid, $storage, $mp->{size
}/1024, $rootuid, $rootgid);
2465 PVE
::Storage
::activate_volumes
($storage_cfg, [$new_volid], undef);
2466 my $path = PVE
::Storage
::path
($storage_cfg, $new_volid, undef);
2467 chown($rootuid, $rootgid, $path);
2471 $copy_volume->($mp->{volume
}, $src, $new_volid, $dest, $storage_cfg, $snapname, $bwlimit, $rootuid, $rootgid);
2475 PVE
::Storage
::vdisk_free
($storage_cfg, $new_volid)
2476 if defined($new_volid);
2483 sub get_lxc_version
() {
2485 run_command
([qw(lxc-start --version)], outfunc
=> sub {
2487 # We only parse out major & minor version numbers.
2488 if ($line =~ /^(\d+)\.(\d+)(?:\D.*)?$/) {
2489 $version = [$1, $2];
2493 die "failed to get lxc version\n" if !defined($version);
2496 return $version->@*;
2501 PVE
::LXC
::CGroup-
>new($vmid)->freeze_thaw(1);
2506 PVE
::LXC
::CGroup-
>new($vmid)->freeze_thaw(0);